Datasets:

---

BioLaySumm

/

BioLaySumm2025-eLife

like 0

Follow

BioLaySumm Shared Task at ACL 3

The dorsal raphe nucleus is the predominant source of central serotonin , where neuronal activity regulates complex emotional behaviors . Action potential firing of serotonin dorsal raphe neurons is driven via α1-adrenergic receptors ( α1-AR ) activation . Despite this crucial role , the ion channels responsible for α1-AR-mediated depolarization are unknown . Here , we show in mouse brain slices that α1-AR-mediated excitatory synaptic transmission is mediated by the ionotropic glutamate receptor homolog cation channel , delta glutamate receptor 1 ( GluD1 ) . GluD1R-channels are constitutively active under basal conditions carrying tonic inward current and synaptic activation of α1-ARs augments tonic GluD1R-channel current . Further , loss of dorsal raphe GluD1R-channels produces an anxiogenic phenotype . Thus , GluD1R-channels are responsible for α1-AR-dependent induction of persistent pacemaker-type firing of dorsal raphe neurons and regulate dorsal raphe-related behavior . Given the widespread distribution of these channels , ion channel function of GluD1R as a regulator of neuronal excitability is proposed to be widespread in the nervous system . Recent reports estimate that 1 in 5 adults worldwide are affected by a mental health disorder , with anxiety and depression being the most common affecting more than 260 million people ( GBD 2017 Disease and Injury Incidence and Prevalence Collaborators , 2018 ) . Most current pharmacotherapies to treat these disorders target serotonin receptors or serotonin clearance . The dorsal raphe nucleus is the largest serotonergic nucleus in the brain and the predominant source of central serotonin ( 5-HT ) . In vivo , tonic noradrenergic input to the dorsal raphe that activates Gαq/11 protein-coupled α1-adrenergic receptors ( α1-ARs ) is required for 5-HT neurons to fire action potentials ( Baraban and Aghajanian , 1980; Baraban et al . , 1978 ) and release 5-HT ( Clement et al . , 1992 ) . In dorsal raphe brain slices , synaptic activation of α1-ARs produces a slow membrane depolarization lasting tens of seconds ( Yoshimura et al . , 1985 ) . Despite having a crucial role in regulating 5-HT neuron excitability , the ion channels responsible for the depolarization remain unknown . Throughout the central and peripheral nervous system , activation of Gαq/11 protein-coupled receptors ( GqPCRs ) , namely metabotropic glutamate mGluRs , muscarinic acetylcholine M1 ( mAChRs ) , or α1-ARs produces slow , noisy inward currents . Multiple mechanisms have been reported to underlie the inward current including: inhibition of K+ current ( including leak , Ca2+-activated , and Kv7/M-current ) ( Benson et al . , 1988; Halliwell and Adams , 1982; Madison et al . , 1987; Shen and North , 1992 ) , modulation of TTX-sensitive persistent Na+ current ( Yamada-Hanff and Bean , 2013 ) , and activation of transient potential receptor canonical ( TRPC ) ( Hartmann et al . , 2008; Kim et al . , 2003 ) , Na+-leak ( NALCN ) ( Lu et al . , 2009 ) , or delta glutamate receptor-channels ( Ady et al . , 2014; Benamer et al . , 2018 ) . The delta glutamate receptors , GluD1R and GluD2R , are mysterious members of the ionotropic glutamate receptor family in that they are not gated by glutamate ( Araki et al . , 1993; Lomeli et al . , 1993 ) . One theory is that they are strictly scaffolding proteins or synaptic organizers , rather than ion conducting channels . But wild-type channels have been reported to conduct in response to activation of mGluRs ( Ady et al . , 2014; Benamer et al . , 2018 ) . GluD1R ( Grid1 ) mRNA is expressed widely throughout the brain , with notably high levels in the dorsal raphe ( Hepp et al . , 2015; Konno et al . , 2014 ) . Here , we used a combination of in vitro patch-clamp electrophysiology and pharmacology with a CRISPR/Cas9 viral genetic strategy to determine that activation of α1-ARs in the dorsal raphe depolarizes neurons via GluD1R-channel conductance . We utilize the α1-AR-GluD1R-EPSC to explore conduction and biophysical properties of GluD1R-channels , to ultimately glean a greater understanding of GluD1R-channel gating . Lastly , we demonstrate that functional deletion of GluD1R-channels in the dorsal raphe produces an anxiogenic behavioral phenotype . Electrophysiological recordings were made from dorsal raphe neurons in acute brain slices from wild-type mice at 35° C in the presence of NMDAR , AMPAR , KainateR , GABA-AR , and 5-HT1AR antagonists . With cell-attached recordings , a train of 5 electrical stimuli ( 60 Hz ) , delivered to the brain slice via a monopolar stimulating electrode , produced firing in previously quiescent neurons , which was blocked by application of the α1-AR antagonist , prazosin ( 100 nM , Figure 1A ) . The excitation produced 20±5 action potentials that lasted 9 . 0±3 . 0 s , with a latency of 650 . 6±0 . 1 ms from onset of stimulation to the first action potential ( Figure 1B-E ) . In whole-cell recording using a potassium-based internal solution , the same train of electrical stimuli produced prolonged action potential firing ( Figure 1F ) . In voltage-clamp mode ( Vhold -65 mV ) , the same stimulation produced a slow and long-lasting ( 27 . 4±2 . 3 s , n=10 ) excitatory postsynaptic current ( EPSC , Figure 1F ) that was eliminated by the application of prazosin ( Figure 1G ) . Prazosin had no effect on basal whole-cell current ( -3 . 8±3 . 4 pA , p=0 . 232 , n=10 , data not shown ) indicating a lack of persistent inward current due to noradrenaline tone . On average , the duration of the α1-AR-EPSC was orders of magnitude longer than fast AMPAR channel-mediated EPSCs ( ~103 . 5× ) and ~18× longer than ‘slow’ 5-HT1A receptor-G protein-coupled inwardly rectifying potassium channel ( GIRK ) -dependent IPSCs ( Gantz et al . , 2015a; Figure 1H ) . To test whether α1-AR-EPSCs were dependent on G protein-signaling , recordings were made with an internal solution containing GDPβS-Li3 ( 1 . 8 mM ) in place of GTP . Disruption of G protein signaling with intracellular dialysis of GDPβS-Li3 eliminated the α1-AR-EPSC within 5-20 mins post-break-in ( p=0 . 004 , n=9 ) , whereas dialysis with LiCl alone had no effect on the amplitude of the α1-AR-EPSC ( p=0 . 625 , n=4 , Figure 1I ) . These findings demonstrate a cell-autonomous requirement of G protein signaling in the generation of the α1-AR-EPSC . Application of tetrodotoxin ( 1 μM ) reversibly abolished the α1-AR-EPSC , demonstrating a dependence on presynaptic action potentials ( Figure 1J ) . Disruption of the vesicular monoamine transporter with reserpine ( 1 μM ) or removal of external Ca2+ also eliminated the α1-AR-EPSC , indicating noradrenaline release is vesicular ( Figure 1K and L ) . Under our recording conditions , resistance of the membrane ( Rm ) significantly decreased during the α1-AR-EPSC , indicative of opening of ion channels ( Figure 2A ) . Membrane noise variance ( σ2 ) increased significantly during the EPSC compared to membrane noise under basal conditions ( Figure 2B and C ) . The α1-AR-EPSC σ2 – amplitude relationship was well fit by linear regression , suggestive of a consistent conductance state , yielding an estimate of a -1 . 16 pA unitary current ( Figure 2D ) . Voltage ramps from -120 to -10 mV ( 1 mV/10 ms ) before and during the α1-AR-EPSC ( Figure 2E ) showed that the current reversed polarity at -28 . 6±2 . 4 mV ( Figure 2E-G ) . Exogenous application of noradrenaline ( 30 μM , in the presence of α2-AR antagonist , idazoxan , 1 μM ) produced an inward current ( INA ) with a similar reversal potential ( -25 . 1±2 . 9 mV , Figure 2G ) . Replacing extracellular Na+ ( 126 mM ) with N-methyl D-glucamine ( NMDG ) completely abolished inward INA , suggesting Na+ as the prominent charge carrier ( Figure 2H ) . Increasing extracellular K+ from 2 . 5 to 6 . 5 or 10 . 5 mM , expected to shift Ek from -107 to -81 and -69 mV , respectively , had no effect on the amplitude of the α1-AR-EPSC at Vhold -65 mV ( Figure 2I ) nor -120 mV ( p=0 . 692 , n=11 , data not shown ) , but produced a significant depolarizing shift in Erev of the α1-AR-EPSC ( Figure 2J ) , suggesting the channel is also permeable to K+ , and may be 2-3× as permeable to K+ as Na+ . Removal of external MgCl2 had no significant effect on Erev ( -28 . 5±5 . 7 mV ) , nor on the amplitude of INA ( Figure 2K and L ) . Removal of external CaCl2 also had no effect on Erev ( -30 . 3±3 . 5 mV ) but significantly augmented inward INA , ( Figure 2K and L ) . Taken together , the data suggest that α1-AR-dependent current , whether produced by vesicular release of noradrenaline or exogenous noradrenaline application is carried through a mixed cation channel , with inward current carried predominantly by Na+ entry . Here , measurements of Erev assume voltage-independence of the channel and the signaling mechanism by which α1-AR signal to the channel . To test for voltage-dependence , we employed a two-pulse voltage-step protocol . Current was measured at Vhold -120 mV following a conditioning pre-pulse ( -120 to 30 mV , 150 ms ) before and after application of noradrenaline ( Figure 2—figure supplement 1A and B ) . INA was isolated by subtracting the current under basal conditions from the current during noradrenaline . Conductance ( GNA ) was calculated , using an Erev of -25 . 1 mV . Conditioning depolarizing pre-pulses incrementally reduced GNA and the increase in membrane noise induced by noradrenaline measured at Vhold -120 mV ( Figure 2—figure supplement 1C and D ) , demonstrating voltage-dependence of inward INA , such that depolarization reduced conductance . To assess involvement of GluD1R-channels in carrying the α1-AR-EPSC , we applied 1-Naphthyl acetyl spermine ( NASPM ) , a synthetic analogue of Joro spider toxin that is an open-channel blocker of some other Ca2+-permeable ionotropic glutamate receptors ( Blaschke et al . , 1993; Guzmán et al . , 2017; Koike et al . , 1997 ) and of GluDR-channels ( Benamer et al . , 2018; Kohda et al . , 2000 ) . Application of NASPM ( 100 μM , 6 min ) blocked the α1-AR-EPSC ( 96 . 0 ± 12 . 5% reduction ) , which recovered to baseline after a wash of >30 mins ( Figures 3A , B , E and I ) . NASPM also produced an apparent outward current ( INSP ) of 20 . 5 ± 3 . 7 pA with an Erev of −31 . 4 ± 4 . 8 mV ( Figures 3A , C , E and G ) and a reduction in membrane noise ( Figure 3A and D ) . After washout , INSP reversed with a similar time course of recovery of the α1-AR-EPSC ( Figure 3E ) . INSP was associated with an increase in Rm ( Figure 3F ) indicating a closure of channels . Replacing extracellular Na+ ( 126 mM ) with NMDG eliminated INSP ( Figure 3G ) . Thus , INSP was due to block of tonic Na+-dependent inward current . INSP was not dependent on prior electrical stimulation of the brain slice , as the magnitude of INSP was similar between stimulated and unstimulated brain slices ( Figure 3H ) . Given that NASPM is an open-channel blocker ( Koike et al . , 1997 ) , we tested whether electrically evoking an α1-AR-EPSC during the application of NASPM was required for block . After obtaining a steady α1-AR-EPSC baseline , NASPM was applied for 6 min without stimulating the brain slice . The α1-AR-EPSC was blocked when stimulation was reapplied ( Figure 3I ) , indicating that the channels underlying the α1-AR-EPSC were already blocked . Thus , the α1-AR-EPSC is mediated by channels that are at least transiently open at rest and may be the same channels underlie the apparent outward current induced by NASPM . GluDRs bind the amino acids D-serine and glycine , both of which partially reduce constitutively open mutant and wild-type GluDR channel current ( Ady et al . , 2014; Benamer et al . , 2018; Naur et al . , 2007; Yadav et al . , 2011 ) , likely by inducing a conformational change in the channel that resembles a desensitized state ( Hansen et al . , 2009 ) . Application of D-serine ( 10 mM , 13 . 5 min ) reduced the amplitude of the α1-AR-EPSC by 49 . 7 ± 9 . 6% ( Figure 4A and E ) , without affecting unitary channel current ( Figure 4B ) . Application of glycine ( 10 mM , 4 . 5 mins , in the presence of the glycine receptor antagonist , strychnine ( 10 μM ) , also reduced the amplitude of the α1-AR-EPSC by 70 . 9 ± 11 . 0% ( Figure 4C and E ) , without affecting unitary channel current ( Figure 4D ) . Lastly , we found that application of the glutamate receptor antagonist kynurenic acid ( 1 mM , 10 . 5 min ) reduced the α1-AR-EPSC amplitude by 65 . 6 ± 8 . 3% ( Figure 4E ) . Next , a viral genetic strategy was used to functionally delete GluD1R-channels by targeting the encoding gene , Grid1 , via CRISPR/Cas9 ( Figure 5—figure supplement 1A–C ) . In brief , one of two cocktails of AAV1 viruses were microinjected into the dorsal raphe of wild-type mice . The Grid1 cocktail that targeted GluD1R-channels included AAV1 viruses encoding Cas9 , and mouse Grid1 guide RNA with a nuclear envelope-embedded enhanced green fluorescent protein ( eGFP ) reporter . A separate cohort received a control cocktail of AAV1 viruses encoding Cas9 and eGFP reporter ( control ) . Brain slices were prepared after >4 weeks and the dorsal raphe was microdissected and frozen on dry ice to assess the mutation of Grid1 . Restriction enzyme site-digested PCR confirmed in vivo mutation of Grid1 at the expected site ( Figure 5—figure supplement 1D ) . In separate Grid1 and control cohorts , brain slices were prepared and whole-cell voltage-clamp recordings were made from transduced and non-transduced neurons visualized in brain slices by expression of eGFP . In eGFP+ neurons from control mice , electrical stimulation produced a decrease in Rm and an α1-AR-EPSC , and bath application of noradrenaline caused inward INA ( Figure 5 ) . However , in eGFP+ neurons from Grid1 mice , electrical stimulation did not change Rm ( Figure 5A ) and no α1-AR-EPSC was detected above baseline noise ( Figure 5B and C ) . In addition , inward INA was substantially smaller in eGFP+ neurons from Grid1 mice , as compared to eGFP+ neurons from control mice ( Figure 5D ) . In the same slices from Grid1 mice , eGFP- neurons still had an α1-AR-EPSC and inward INA ( Figure 5B and D ) . Lastly , bath application of NASPM produced an apparent outward current in eGFP+ neurons from control mice , but not from Grid1 mice ( Figure 5E ) . Taken together , these results demonstrate that conduction through GluD1R-channels is necessary for the α1-AR-EPSC and the NASPM-sensitive tonic inward current . To assay a functional role of GluD1R-channels in dorsal raphe-related behavior , wild-type mice received a microinjection into the dorsal raphe of either Grid1 or control virus cocktails . Behavioral assays were conducted >4 weeks post-injection , then the accuracy of the dorsal raphe injection and limited-spread of transduction was verified post-hoc by immunohistochemistry ( Figure 6A ) . Basal locomotion was assayed in a dark arena . There was no difference between the two groups in the total distance traveled ( Figure 6B ) nor in the velocity of movements between control and Grid1 mice ( p=0 . 772 , n = 18 and 16 , data not shown ) . Next , mice were tested on an elevated plus maze in a well-lit room , an experimental assay of rodent anxiety behavior ( Walf and Frye , 2007 ) known to involve both serotonergic and non-serotonergic neurons in the dorsal raphe ( Lawther et al . , 2015 ) . Grid1 mice spent less time in the open arms when compared to control mice ( Figure 6C–E ) . Control and Grid1 mice made a similar total number of entries to either open or enclosed arms ( control: 39 . 4 ± 2 . 0; Grid1: 36 . 0 ± 2 . 3 , p=0 . 697 ) , but Grid1 mice made proportionally fewer entries to the open arms ( Figure 6D ) . Time spent grooming or in stretched-attend postures were similar between control and Grid1 mice ( Figure 6F and G ) . Since movement in the elevated plus maze reflects conflict between innate drive to explore of a novel environment and natural avoidance of open spaces ( Walf and Frye , 2007 ) , we also examined exploratory behaviors . Grid1 mice spent less time lowering their head over the edge of the open arms than control mice ( head-dipping , Figure 6H ) , suggestive of decreased exploratory behavior . However , Grid1 mice spent a similar amount of time rearing in the enclosed arms compared to control mice ( Figure 6I ) suggesting innate exploratory drive in the enclosed arms was intact . Taken together , these results are indicative of heightened anxiety after functional deletion of GluD1R-channels in the dorsal raphe . In vivo , 5-HT neurons in the dorsal raphe require noradrenaline release and subsequent activation of α1-ARs to maintain persistent action potential firing ( Baraban and Aghajanian , 1980 ) . The activation of α1-ARs in the dorsal raphe by exogenous agonist was thought to depolarize neurons through net reduction of K+ conductance , transiently activating calcium-activated K+ current while persistently decreasing another K+ current , and by activation of an unidentified non-K+ conductance ( Pan et al . , 1994 ) . In a more recent study in the dorsal raphe , Brown et al . ( 2002 ) reported that activation of α1-ARs , induces Na+-dependent inward current with an Erev of −23 mV , similar to our findings . Our study identifies GluD1 receptor-channels as the ion channel that carries this mixed cation current , indicating that modulation of GluD1R-channels is a key constituent in driving persistent action potential firing of the 5-HT neurons . In principle , inward GluD1R-channel current may bring the membrane potential to threshold , but recruitment of other voltage-gated ion channels is expected to underlie the persistent pacemaker-like activity . Intriguingly , Brown et al . ( 2002 ) demonstrated that activation of Gq protein-coupled histamine H1 and orexin OX2 receptors also produced an inward current that was occluded by the α1-AR-dependent current . Whether these receptors , and other GqPCRs , augment GluD1R-channel current remains to be determined . Dysregulation of the 5-HT signaling neuropsychiatric disorders is well-established . Pharmacotherapies to boost serotonin signaling are common and often efficacious in some of these conditions . Genetic association studies have identified GRID1 as a susceptibility gene for psychiatric conditions , including schizophrenia , major depressive disorder , bipolar disorder , autism spectrum disorder , and alcohol dependence ( Edwards et al . , 2012; Fallin et al . , 2005; Griswold et al . , 2012 ) . Global Grid1 knock-out mice display abnormal social behaviors , including heightened aggression and decreased social interaction , as well as altered emotional behaviors ( Yadav et al . , 2012 ) that are analogous to features of neuropsychiatric conditions in humans . Our study found that functional deletion of GluD1R-channels , specifically in the dorsal raphe , produces a heightened anxiety-like response in the elevated plus maze without changing basal locomotion and exploratory behaviors in non-threatening environments . Previous studies have demonstrated that both 5-HT and non-5-HT/GABAergic dorsal raphe neurons are activated by aversive , anxiety or fear-producing stimuli ( Seo et al . , 2019; Silveira et al . , 1993 ) , with regional subpopulation specificity ( Grahn et al . , 1999; Grahn et al . , 2019; Lawther et al . , 2015 ) . Our viral strategy functionally deleted GluD1R-channels in a non-specific manner , targeting all dorsal raphe neurons , including 5-HT and GABAergic neurons . Given the rich diversity of dorsal raphe neuron subtypes and subdivisions within the 5-HT neurons ( Huang et al . , 2019; Luo et al . , 2015; Ren et al . , 2018 ) , future work will be needed to parse the behavioral role of GluD1R-channels with subnuclei/subpopulation specificity . GluDR have been characterized as scaffold proteins or synaptic organizers , regulating LTD , endocytosis and trafficking of AMPAR , formation of excitatory and inhibitory synapses , and spine density , independent of ion conduction through the pore ( Fossati et al . , 2019; Hirai et al . , 2003; Schmid and Hollmann , 2008; Tao et al . , 2018 ) . Similarly , NMDAR are known to signal through non-ionotropic or ‘metabotropic’ mechanisms where ion conduction is not required , to regulate LTD , AMPAR endocytosis , and spine morphology ( Dore et al . , 2016 ) . The ability of GluDR-channels to carry ionic current does not conflict with its known role as a synaptic organizer , but rather expands the similarities between NMDAR and GluDR . The largest obstacle in advancing the understanding of the ionotropic nature of GluDR is the lack of known agonist and inability to gate the intact channel . The majority of studies have been performed on constitutively open mutant or chimeric channels . In domain-swapped chimeric channels , agonist binding to the ligand-binding domain ( LBD ) of AMPAR or KainateR opens the GluDR-channel pore and generates a substantial current , but the LBD of GluDR on the pore region of AMPAR or KainateR-channels fails to generate current ( Orth et al . , 2013; Schmid et al . , 2009 ) . Two prior studies have demonstrated that in heterologous systems and brain slices , activation of metabotropic glutamate receptors ( mGluR ) produces an inward current carried by GluD1R- ( Benamer et al . , 2018 ) or GluD2R-channels ( Ady et al . , 2014 ) , concluding that mGluR activation triggers gating of GluDR channels . The congruous explanation of our results is that , in dorsal raphe neurons , GluD1R-channels are functional and open under basal conditions , carrying subthreshold , tonic Na+ current . Activation of α1-ARs , by exogenous agonist or synaptic release of noradrenaline modulates gating of GluD1R-channels and excites dorsal raphe neurons by increasing tonic GluD1R-channel inward current . In general , the kinetics of iGluR synaptic currents are controlled by the lifetime of the receptor-agonist complex and the rate of desensitization and deactivation . The presence of ambient levels of glutamate and glycine along with slow desensitization activate NMDAR to produce a tonic inward current ( Sah et al . , 1989 ) . Our results demonstrate that GluD1R are functional ion channels , but whether they function as ligand-gated receptor-channels that open in response to a chemical signal , is not yet determined . What remains to be understood are the conditions that permit GluD1R-channel opening and why their activation has been largely elusive in heterologous expression systems . Reminiscent of times before the discovery of glycine as a necessary co-agonist at NMDAR ( Johnson and Ascher , 1987; Kleckner and Dingledine , 1988 ) , it may be that an endogenous agonist needed for gating is present in brain slices . Alternatively , it is possible GluD1R-channels are gated by an intracellular factor or require expression of accessory or interacting protein ( Tomita , 2010 ) . Tonic activation of α1-ARs cannot explain the tonic inward current as α1-AR antagonism did not change basal whole-cell current . The mechanism by which α1-ARs increase GluD1R-channel current also remains to be described and may be distinct from the tonic activation . It is well-established that GqPCRs , especially mGluR and mAChR , bidirectionally change NMDAR and AMPAR ionic currents , producing the two major forms of synaptic plasticity , long-term potentiation ( LTP ) and long-term depression ( LTD ) , in part through a variety of distinct postsynaptic mechanisms ( Hunt and Castillo , 2012 ) . To our knowledge , the duration of the α1-AR-EPSC ( ~27 s ) is exceptional for any known synaptic current and more closely resembles the duration of short-term synaptic plasticity; for instance , endocannabinoid-mediated short-term depression ( Lu and Mackie , 2016 ) . Canonically , GqPCRs activate phospholipase C which hydrolyzes the integral membrane lipid phosphatidylinositol 4 , 5-bisphosphate ( PIP2 ) into inositol triphosphate ( IP3 ) and diacylglycerol . PIP2 stabilizes Kv7 channels such that PIP2 hydrolysis following mAChR activation accounts for inhibition of M-current ( Suh and Hille , 2002 ) . In contrast , PIP2 inhibits TRPV4 channels , such that GqPCR-dependent PIP2 depletion allows for TRPV4 channels to open ( Harraz et al . , 2018 ) . By the same signaling cascade , GqPCRs stimulate the production of the endocannabinoid , 2-AG , that can act directly on ion channels in the membrane ( Gantz and Bean , 2017 ) . Thus , one possibility is that α1-ARs modulate GluD1R-channels through membrane lipid signaling , involving PIP2 , diacylglycerol , or 2-AG , as it can take tens-of-seconds to minutes for ion channels to recover from modulation by membrane lipids ( Gantz and Bean , 2017; Suh and Hille , 2002 ) . Alternatively , there may be direct modulation of GluD1R-channels by G protein subunits or activation of protein kinase signaling cascades . The inclusion of the calcium-chelator BAPTA in the internal recording solution makes it unlikely that α1-ARs modulate GluD1R-channels via IP3 and calcium release from intracellular stores ( Hoesch et al . , 2004 ) . Largely , it remains to be seen whether these intracellular signaling cascades , many of which are known to affect NMDAR- and AMPAR-channels , modify GluDR-channels . In heterologous systems and constitutively open mutant GluDR-channels , the current reverses polarity around 0 mV ( Zuo et al . , 1997 ) , akin to AMPAR- and NMDAR-channels , while our results show Erev of ~ −30 mV . While slow voltage-ramps were employed to minimize space-clamp error , we cannot rule out that some of the difference may be attributed to space-clamp error in brain slices , especially since the magnitude of subtracted current is small relative to total membrane current at depolarized potentials . However , there are many reports of inward currents produced by activation of many different GqPCRs with reversal potentials between −40 and −23 mV ( Awad et al . , 2000; Brown et al . , 2002; Yamada-Hanff and Bean , 2013 ) under different recording conditions; a commonality that is unlikely to be accounted for by space-clamp error alone . Tail current analysis revealed voltage-dependence of INA , such that depolarization reduced conductance . These data may reflect block of GluD1R-channels by endogenous intracellular polyamines , as established for calcium-permeable AMPAR- and KainateR-channels ( Bowie and Mayer , 1995 ) . Another important consideration is that our measurements may be subject to voltage-dependence of the signaling pathway between α1-ARs and GluD1R-channels . Taken together , measurements here should be considered an estimate of GluD1R-channels , and more precisely as the current-voltage relationship of the α1-ARs-GluD1R-channel signaling complex . In summary , the α1-AR-mediated depolarization of dorsal raphe neurons that drives action potential firing in vivo is carried by the mixed cation channel , GluD1R . Thus in addition to their role as a scaffold protein , GluD1R are functional ion channels that critically regulate neuronal excitability . Many of the biophysical properties of the GluD1R-channel are like other members of the ionotropic glutamate receptor family . Given the widespread distribution of these receptors throughout the brain ( Hepp et al . , 2015 ) , ion channel function of GluD1R may be prevalent and relevant to neuronal excitability and circuit function in different parts of the throughout the nervous system . This study lays the foundation to investigate the ion channel function of GluD1R in excitatory GqPCR-dependent synaptic transmission and regulation of neuronal excitability , expanding upon the wealth of knowledge of pharmacology and regulatory elements established for NMDAR and AMPAR signaling . All studies were conducted in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory animals with the approval of the National Institute on Drug Abuse Animal Care and Use Committee . Wild-type C57BL/6J ( >3 months old ) mice of either sex were used . Mice were group-housed on a 12:12 hr reverse light cycle . The methods for brain slice preparation and electrophysiological recordings were almost identical to previous reports in the dorsal raphe ( Gantz et al . , 2015a ) and ventral midbrain ( Gantz et al . , 2015b ) . In brief , mice were deeply anesthetized with isoflurane and killed by decapitation . Brains were removed quickly and placed in warmed artificial cerebral spinal fluid ( modified Krebs’ buffer ) containing ( in mM ) : 126 NaCl , 2 . 5 KCl , 1 . 2 MgCl2 , 1 . 2 CaCl2 , 1 . 2 NaH2PO4 , 21 . 5 NaHCO3 , and 11 D-glucose with 5 μM MK-801 to reduce excitotoxicity and increase viability , bubbled with 95/5% O2/CO2 . In the same solution , coronal dorsal raphe slices ( 220 μm ) were obtained using a vibrating microtome ( Leica 1220S ) and incubated at 32°C > 30 min prior to recording . Slices were then mounted in a recording chamber and perfused ~3 mL/min with ~35°C modified Krebs’ buffer . Electrophysiological recordings were made with a Multiclamp 700B amplifier ( Molecular Devices ) , Digidata 1440A A/D converter ( Molecular Devices ) , and Clampex 10 . 4 software ( Molecular Devices ) with borosilicate glass electrodes ( King Precision Glass ) wrapped with Parafilm to reduce pipette capacitance ( Gantz and Bean , 2017 ) . Pipette resistances were 1 . 8–2 . 8 MΩ when filled with an internal solution containing , ( in mM ) 104 . 56 K-methylsulfate , 5 . 30 NaCl , 4 . 06 MgCl2 , 4 . 06 CaCl2 , 7 . 07 HEPES ( K ) , 3 . 25 BAPTA ( K4 ) , 0 . 26 GTP ( sodium salt ) , 4 . 87 ATP ( sodium salt ) , 4 . 59 creatine phosphate ( sodium salt ) , pH 7 . 32 with KOH , mOsm ~285 , for whole-cell patch-clamp recordings . Series resistance was monitored throughout the experiment . Transmitter release was evoked by trains of electrical stimuli delivered via a Krebs’ buffer-filled monopolar stimulating electrode positioned in the dorsal raphe , within 200 μm of the recorded neuron ( Gantz et al . , 2015a ) . Cell-attached recordings were made from quiescent neurons in slice , using pipettes filled with modified Krebs’ buffer . For experiments involving viral microinjections , transduced neurons were identified in the slice by visualization of eGFP . Reported voltages are corrected for a liquid junction potential of −8 mV between the internal solution and external solution . All drugs were applied by bath application . All experiments were conducted following incubation in an NMDAR channel blocker ( MK-801 , 5 μM , >1 hr ) , and then with AMPAR and KainateR ( NBQX , 3 μM ) , GABA-AR ( picrotoxin , 100 μM ) , and 5-HT1AR ( WAY-100635 , 300 nM ) antagonists in the external solution . In addition , a α2-adrenergic receptor antagonist ( idazoxan , 1 μM ) was added for experiments where noradrenaline was applied and a glycine receptor antagonist ( strychnine , 10 μM ) was added when glycine was applied . Unitary current was calculated from fluctuation analysis , as previously described ( Bean et al . , 1990 ) , assuming the macroscopic current arises from independent , identical channels with a low probability of opening , according probability theory; i = σ2/[I ( 1 p ) ] where i is unitary current , σ2 is the variance , I is mean current amplitude , and p is probability of opening . CRISPR SpCas9 gRNA target sites were identified in the mouse Grid1 gene ( NC_000080 . 6 ) using CRISPOR ( Haeussler et al . , 2016 ) . The seed sequence ( GAACCCTAGCCCTGACGGCG ) was chosen based on its relatively high specificity scores and the observation that it contains a Bgl I restriction enzyme site ( GCCNNNN^NGGC ) that overlaps with the Cas9 cleavage site . C57BL/6J mouse genomic DNA was isolated from tail biopsies or brain pieces containing microdissected dorsal raphe by digestion in DNA lysis buffer ( 50 mM KCl , 50 mM Tris-HCl ( pH 8 . 0 ) , 2 . 5 mM EDTA , 0 . 45% NP-40 , 0 . 45% Tween-20 , 0 . 5 ug/mL proteinase K ) for 3 hr at 55°C , and 1 hr at 65°C . Lysates were then used as templates to amplify a 654 basepair fragment including the 390F gRNA target site using Q5 HotStart Master mix ( New England Biolabs ) . A portion of the finished PCR reaction was treated with Bgl I restriction enzyme ( New England Biolabs ) for 60 min and processed on an AATI fragment analyzer . The AAV vector plasmid encoding SpCas9 ( Swiech et al . , 2015 ) ( pX551 ) expressed from the Mecp2 promoter was a gift from Feng Zhang ( Addgene plasmid # 60957 , AAV-Cas9 ) . The AAV packaging plasmid encoding a nuclear envelope-embedded eGFP reporter ( Addgene 131682 ) was constructed by amplifying the KASH domain from ( Addgene 60231 , a gift of Feng Zhang ) and fusing it ( in-frame ) to the end of coding region for eGFP in ( Addgene 60058 , pOTTC407 ) using ligation-independent cloning ( AAV-empty , Figure 5—figure supplement 1A ) . gRNA was cloned into a mU6 expression cassette and then moved into an AAV backbone expressing a nuclear envelope-embedded ( KASH-tagged ) eGFP reporter ( Addgene 131683 ) by PCR amplification and ligation-independent cloning ( AAV-Grid1 ) . Insert-containing clones were verified by sequencing and restriction fragment analysis prior to virus production . All AAV vectors were produced using triple transfection method as previously described ( Howard and Harvey , 2017 ) . All vectors were produced using serotype 1 capsid proteins and titered by droplet digital PCR . Mice were anesthetized with a cocktail of ketamine/xylazine , immobilized in a stereotaxic frame ( David Kopf Instruments ) , and received one midline injection of a 1:1 ( v/v ) cocktail of viruses AAV-Cas9 and AAV-empty or AAV-Grid1 for total volume 400 nL delivered over 4 min . The coordinates for injection were AP −4 . 4; ML 1 . 19 , 20° angle; DV −3 . 62 mm , with respect to bregma . Prior to surgery , mice were injected subcutaneously with warm saline ( 0 . 5 mL ) to replace fluid lost during surgery and given carprofen ( 5 mg/kg ) post-surgery for pain relief . Mice recovered for >4 weeks to allow expression . Behavioral assays were conducted during the dark cycle , using 3 separate cohorts of AAV-Cas9 and AAV-empty or AAV-Grid1-injected mice as biological replicates , 30–55 d post-injection . To measure basal locomotion , mice were placed in locomotor boxes ( VersaMax System , Omnitech Electronics , Inc ) in a dark room for 1 hr , following prior habituation to the locomotor boxes for >2 d ( 1 hr/d ) . VersaMax Analyzer software was used to determine the total distance traveled , time spent moving , and velocity of movement in the last 30 min of the session . The boxes were cleaned with 70% ethanol and allowed to dry between trials . The elevated-plus maze was used to assay anxiety-related behaviors ( Walf and Frye , 2007 ) . The apparatus ( Med associates , Inc ) was placed 30 cm above the floor and consisted of two plastic light gray open arms ( 30 × 5 cm ) and two black enclosed arms ( 30 × 5 cm ) extending from a central platform ( 5 × 5 × 5 cm ) at 90 degrees . Following habituation to the brightly lit room , mice were placed individually in the center of the maze , facing an open arm . Video tracking EthoVision XT software ( Noldus Information Technology ) was used to track mouse location , total distance traveled , velocity of movement , body elongation , and entries and time spent into the open and enclosed arms for each 5 min trial . Duration of head-dips , grooming , and enclosed-arm rearing were scored manually from videos played a 0 . 5x speed . Rearing in the enclosed arm was often associated with pressing one or both forepaws to the wall . Stretched-attend postures was defined by body elongation ( 70% threshold ) and movement velocity <1 cm/s . No mice fell or jumped from the maze and open-arm rearing was not observed . The maze was cleaned with 70% ethanol after every trial and allowed to dry before the next trial . Mice were excluded from analysis if there was limited or no expression in the dorsal raphe , or if expression spread rostrally to the ventral tegmental area or caudally to locus coeruleus . Following behavioral assays , mice were euthanized with Euthasol and transcardially perfused with PBS followed by ice-cold 4% paraformaldehyde in PBS ( pH 7 . 4 ) . Brains were fixed overnight at 4 C and then sliced coronally in 50 μm sections . Alternatively , mice were anesthetized with isoflurane and euthanized by decapitation . Brains were removed and slices were prepared as for brain slice electrophysiology ( 220 μm ) , then fixed in room temperature 4% paraformaldehyde in PBS for 1 hr . Slices were mounted with Fluoromount-G with DAPI ( Invitrogen ) aqueous mounting medium . Confocal images were collected on an Olympus microscope ( 4x , 0 . 16 NA ) and processed using Fiji . Data were analyzed using Clampfit 10 . 7 . Data are presented as representative traces , or in scatter plots where each point is an individual cell , and bar graphs with means ± SEM . In traces with electrical stimulation , stimulation artifacts have been blanked for clarity . Unless otherwise noted , n = number of distinct cells or mice as biological replicates . No sample was tested in the same experiment more than once ( technical replication ) . Erevs were determined by linear regression for each cell . Recordings in which current did not cross 0 pA were omitted from analysis . To minimize space-clamp errors , analysis of current during voltage ramps was limited to −10 mV where the currents were typically less than 500 pA . Ramp currents were averaged in 2 mV bins ( 20 ms ) . Data sets with n > 30 were tested for normality with a Shapiro-Wilk test . When possible ( within-group comparisons ) , significant differences were determined for two group comparisons by paired t-tests , Wilcoxon matched-pairs signed rank test , and in more than two group comparisons by nonparametric repeated-measures ANOVA ( Friedman test ) . Significant mean differences in between-group comparisons were determined for two group comparisons by Mann Whitney tests , and in more than two group comparisons by Kruskal-Wallis tests . ANOVAs were followed , when p<0 . 05 by Dunn’s multiple comparisons post hoc test . Linear trends were analyzed using a mixed model ANOVA . A difference of p<0 . 05 was considered significant . For behavioral assays , Grubbs test was used to identify outliers . Basal locomotion and time spent in stretched-attend posture from one Grid1 mouse each were found to be outliers and were excluded from group comparisons . Exact values are reported unless p<0 . 0001 or>0 . 999 . Statistical analysis was performed using GraphPad Prism 8 ( GraphPad Software , Inc ) .

Serotonin is a chemical that allows cells to communicate in the nervous system of many animals . It is also particularly important in the treatment of mental health disorders: a large number of antidepressants work by preventing nerve cells from clearing away serotonin , therefore increasing the overall level of the molecule in the brain . Yet , exactly how serotonin is released remains unclear . When a serotonin-producing cell is activated , a series of biochemical processes lead to the creation of an electric current that , ultimately , is required for the cell to release serotonin . This mechanism starts when the α1-adrenergic receptor , a protein at the surface of the cell , detects noradrenaline molecules . However , on its own , the α1-adrenergic receptor is unable to create an electric current: this requires ion channels , another type of protein which can let charged particles in and out of the cell . Here , Gantz et al . set out to determine the identity of the ion channel that allows noradrenaline signals to generate electrical activity in cells which can release serotonin . Electrical and chemical manipulation of mouse brain slices revealed that an ion channel called delta-glutamate 1 was active in serotonin-producing cells exposed to noradrenaline . In fact , applying toxins that specifically blocked the activity of this channel also prevented the cells from responding electrically to noradrenaline . Further experiments used mice whose serotonin-producing cells were genetically modified to turn off delta-glutamate 1 . In turn , these animals showed anxiety-like behaviors , which could be consistent with a drop in serotonin levels . This is in line with previous human studies showing that patients with depression and other mental health conditions have mutations in the gene for delta-glutamate 1 . Taken together , these results give an insight into the electrical activity of serotonin-producing cells . Further work is now required to examine how changes in the gene that codes for delta-glutamate 1 ultimately affect the release of serotonin . This could potentially help to understand if certain individuals may not be able to properly produce this chemical . As many antidepressants work by retaining serotonin that is already present in the brain , this knowledge could ultimately help patients who do not currently respond to treatment .

2020

Delta glutamate receptor conductance drives excitation of mouse dorsal raphe neurons

We have designed a membrane ‘staple’ , which consists of membrane-anchored repeats of the trans-aggregating FM domain that face the lumen of the secretory pathway . In the presence of the disaggregating drug these proteins transit the secretory pathway . When the drug is removed these proteins form electron-dense plaques which we term staples . Unexpectedly , when initially positioned within the cis-Golgi , staples remained at the cis face of the Golgi even after many hours . By contrast , soluble FM-aggregates transited the Golgi . Staples and soluble aggregates placed in cis-Golgi cisternae therefore have different fates . Whereas the membrane staples are located in the flattened , stacked central regions of the cisternae , the soluble aggregates are in the dilated rims . This suggests that while the cisternae are static on the time scale of protein traffic , the dilated rims are mobile and progress in the cis → trans direction via a mechanism that we term ‘Rim Progression’ . Anterograde transport ( cis → trans ) through the Golgi stack is a prerequisite for virtually all proteins that are ultimately secreted from the cell , targeted to the cell surface membrane , or localized in a plethora of internal membrane-bound compartments in plants and animals . Despite the fact that this fundamental process was first recognized over 50 years ago , there is still today no general agreement on how it works . The greatest distinction among the competing models for anterograde transport ( reviewed in Rothman , 2010 ) is whether the 4–6 cisternae typically comprising the Golgi stack are proposed to be static or mobile over the time scale of protein flow across the Golgi ( typically 5–20 min ) . In ‘static’ models , a transport process such as budding/fusing COPI transport vesicles or possibly tubules is required to create anterograde flow of cargo . In contrast , in ‘mobile’ models , the cisternae themselves continuously move as intact units in the cis → trans direction ( termed ‘cisternal progression’ ) , forming at the cis face and being consumed at the trans face , so that no inter-cisternal transport process is required in the anterograde direction . In order to allow the Golgi stack to retain its resident proteins ( such as glycosyltransferases ) in the face of continuous cisternal turnover , mobile/cisternal progression models also require concomitant retrograde transport ( trans → cis or Golgi → ER , or both ) of steady-state Golgi resident enzymes , termed ‘cisternal maturation’ ( Glick and Malhotra , 1998; Glick and Luini , 2011 ) . The strongest evidence for cisternal progression in Golgi stacks is that intrinsically large soluble cargoes such as immature collagen aggregates ( Bonfanti et al . , 1998 ) —which are far too large to be accommodated by COPI transport vesicles—can nonetheless be rapidly transported across the stack . On this basis it has been widely accepted by deductive reasoning , as distinct from direct visualization , that whole cisternae must indeed progress across the stack . This of course does not rule out that parallel processes of COPI vesicular transport ( Pelham and Rothman , 2000 ) and/or tubule-based transport ( Trucco et al . , 2004 ) could also occur . We sought to further test the cisternal progression paradigm by extending the deductive approach from soluble aggregates to aggregates fixed in the membrane , which should behave identically to soluble aggregates according to cisternal progression models . For this purpose , we designed reversible membrane ‘staples’ loosely modeled on the adhesion protein cadherin , which can adhere cells homo-typically . Cadherins consist of tandem repeats of an adhesive unit that binds itself preferentially in trans ( Pokutta and Weis , 2007 ) . A similar unit , which additionally is controllable with soluble ligands that can enter cells , is the FKBP mutant termed FM , which according to its crystal packing also interacts in trans ( Rollins et al . , 2000 ) , with the additional feature that this interaction is disrupted by ligands such as AP21998 even in living cells ( Rivera et al . , 2000 ) . This suggests that tandem membrane-anchored repeats of FM domains should mimic cadherin in producing a defined structure ( staple ) that will reversibly adhere intracellular membranes when it faces the cytoplasm , or in the present application produce intra-lumenal adhesions within Golgi cisternae when the FM repeats are lumenally-oriented . To our surprise , we found that luminal staples and compositionally analogous soluble aggregates placed in cis-Golgi cisternae had entirely different fates . Whereas the soluble aggregates were rapidly transported , the membrane staples remained in place . As the basic unit of a lumenal membrane staple we designed a chimeric protein , termed CD8lumenal , that contained a signal sequence , for its proper insertion and translocation across the ER membrane , fused to 1 ) a fluorescent protein ( GFP or DsRed ) to track the protein within the cells by confocal microscopy , 2 ) four repeats of a domain ( FM4 ) designed to self-aggregate in trans in the absence of the disaggregating drug ( AP21998 ) ( Rollins et al . , 2000 ) , and 3 ) the CD8 protein , known to be a trans-membrane protein ultimately targeted to the plasma membrane ( Hennecke and Cosson , 1993; Lavieu et al . , 2010 ) . This CD8lumenal protein should initially reside in the ER oriented with its FM4-containing aggregation domain facing the lumen , and so long as it remains non-aggregated ( in the presence of AP21998 ) should be effectively transported to the cell surface . On the other hand , if AP21998 is removed while in the ER—or at any later stage of transport through the Golgi—then intra-lumenal adhesions are expected to form as CD8lumenal aggregates homo-typically in trans , like a cadherin . These staples are expected to remain in the ER if they are planar and too large to be packaged into a small COPII vesicle . Various structural models suggest a planar adhesion plaque will form that closely links two sides of ER lamellae or tubules , with an intra-lumenal separation between the stapled membranes that from first principles could vary from 10 to 23 nm ( Figure 1A ) , depending on the preferred arrangement of trans-aggregating FM4 domains . In particular , however , if repeat FM domains interact like cadherins ( Miyaguchi , 2000; Pokutta and Weis , 2007 ) primarily via their most-membrane distant module ( i . e . , at their tips ) , then we would expect a membrane separation in the stapled state close to 25 nm ( model 1 in Figure 1A ) . 10 . 7554/eLife . 00558 . 003Figure 1 . Aggregated-CD8Lumenal as an ER staple . ( A ) Domains harbored by CD8lumenal . Predicted models of trans-interacting CD8lumenal preoteins that are expected to form lumenal staples . Predicted intra-lumenal space varies from 10 to 23 nm . ( B ) Electron-micrograph showing ER of HeLa cells expressing CD8lumenal in the absence of the disaggregating drug . ER-aggregated CD8lumenal , which form ER-staples , appear as electron-dense flat features ( red square ) . ( C ) Lumenal staples trigger lumenal constriction of ER tubules , consistent with model 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 00558 . 00310 . 7554/eLife . 00558 . 004Figure 1—figure supplement 1 . Cytosolic staples stack the ER . HeLa cells expressing CD8cytosolic were incubated without the disaggregating drug . Electron-micrograph showing that cytosolic staples trigger artificial stacking of ER membranes . DOI: http://dx . doi . org/10 . 7554/eLife . 00558 . 004 When CD8lumenal is expressed in HeLa cells in the absence of the disaggregating drug AP21998 , numerous , discrete electron-dense plaques are in fact readily observed by electron microscopy ( Figure 1B , red square ) . The staples trigger lumenal constrictions within ER tubules ( Figure 1C ) whose separation averages 25 ± 5 nm , suggesting that model 1 in which only the distal FM domains are typically bonded prevails in the ER ( Figure 1A ) most closely approximates the structure of the adhesion plaque . Remarkably , when we used cytosolic staples that harbor the FM domains at the cytosolic face of the membranes ( CD8cytosolic ) , ER membranes were artificially stacked , confirming the trans–nature of the staples interaction ( Figure 1—figure supplement 1 ) . Although we exclusively visualized staples resulting from trans-interactions ( two opposed membranes ) , we cannot rule out that cis-interactions ( on the same membrane ) may occurs when the physical constraints are favorable . At the confocal level , the staples showed a reticular pattern typical of ER . Only when AP21998 is added to disaggregate the staples in the ER is CD8lumenal transported to the cell surface ( Figure 2A ) . Importantly , when the staples are stuck in the ER they did not prevent the production or transport of co-expressed Golgi resident enzymes ( ST-RFP ) , showing that ER staples are not toxic ( Figure 2B ) . 10 . 7554/eLife . 00558 . 005Figure 2 . Disaggregated-CD8lumenal as an anterograde cargo . ( A ) Confocal micrograph showing HeLa cells expressing CD8lumenal in the presence or in the absence of the disaggregating drug . Without drug , ER-aggregated CD8lumenal remained at the ER ( ER staples ) , whereas with drug ( for 4 hr ) disaggregated CD8lumenal is transported to the plasma membrane ( PM ) , where it could be detected with an anti-GFP antibody in non-permeabilized cells . ( B ) Confocal micrograph showing HeLa cells co-expressing CD8lumenal with ST-RFP in the absence of the disaggregating drug . The ER staples do not alter Golgi targeting of ST-RFP . ( C ) Immunoblot showing that 4 hr treatment of HeLa cells with the disaggregating drug increases TritonX-100 solubility of CD8lumenal . Disaggregated-CD8lumenal shows reduced mobility on SDS-PAGE gel ( * upper band ) . ( D ) Immunoblot showing that Jacalin , a lectin that binds galactose residues , exclusively precipitates disaggregated CD8lumenal . DOI: http://dx . doi . org/10 . 7554/eLife . 00558 . 00510 . 7554/eLife . 00558 . 006Figure 2—figure supplement 1 . Lifetime of staples and disaggregated CD8lumenal . HeLa cells expressing CD8lumenal were radiolabeled with 35S methionine for 30 min in the presence or the absence of the disaggregating drug ( pulse ) . Then cells were washed and incubated at 20°C in media containing 3 mM methionine , and harvested at different time-points ( chase ) . For the study of Golgi staples , after the radiolabeling pulse , cells were shifted to 16°C for 1 hr in the presence of the drug , after which the drug was removed and cells were incubated at 20°C . CD8lumenal from the cell extracts were immuno-precipitated using anti-GFP antibody , processed for SDS-PAGE . Lifetime was estimated by measuring the densitometry ( using ImageJ ) on a film exposed overnight to the gel containing the radioactive samples . DOI: http://dx . doi . org/10 . 7554/eLife . 00558 . 006 We also employed biochemical tests of cell extracts to characterize the staples . When tested by SDS-PAGE , the disaggregating drug increased Triton-X100 solubility ( without sonication procedure ) of CD8lumenal in cell extracts by a factor 3 . 9 ± 0 . 9 ( Figure 2C ) . Note that this relatively modest response of CD8lumenal to the disaggregating-drug ( 44% Triton X100-solubility after 4 hr treatment ( Figure 2C ) can be attributed to the transient transfection procedure . A large portion of the transfected cells showed a very high level of CD8lumenal over-expression , and did not respond well to the disaggregating drug . As a consequence CD8lumenal , remained aggregated in the ER even during drug treatment . This also explains the apparent moderate efficiency of transport of disaggregated-CD8 when measured by bulk assay ( Figure 3B ) . Disaggregated CD8lumenal , expected to be O-glycosylated ( Nilsson et al . , 1989; Lavieu et al . , 2010 ) , appeared as an isoform with reduced SDS-PAGE motility ( * Figure 2C ) . As expected , Jacalin , a lectin that binds D-Galactose residues , exclusively precipitates this isoform ( Figure 2D ) . In addition , metabolic radiolabeling experiments showed that staples ( ∼5 . 5 hr half-life ) were slightly more stable that disaggregated CD8lumenal ( ∼2 . 5 hr half-life ) ( Figure 2—figure supplement 1 ) . Such stability is enough to guarantee the presence of the staples during numerous complete rounds of anterograde transport either from ER to Golgi or across the entire Golgi ( around 20 rounds if we assume that one round of transport takes 15 min ) . Note that staples persist in the ER and in the Golgi ( Figure 2—figure supplement 1 ) for the same time period ( 5–6 hr ) before degradation by a cellular machinery that is not yet known . 10 . 7554/eLife . 00558 . 007Figure 3 . Re-aggregated CD8lumenial as a Golgi staple , and its retention within the Golgi . ( A ) General procedure: 16°C temperature block and disaggregation/re-aggregation cycles are combined to form the staples in the cis-Golgi and assess the cisternal progression model . HeLa cells expressing CD8lumenal were incubated for 1 . 5 hr at 16°C in the presence of the disaggregating drug , and then the drug was removed for 30 min at 16°C to allow re-aggregation in the cis-Golgi . Temperature was shifted to 37°C for 2 hr in the absence ( 2 ) or in the presence ( 3 ) of the disaggregating drug . As a negative control , cells were incubated at the same temperatures but without any drug at any time ( 1 ) . ( B ) Immunoblot , after surface biotinylation , showing that disaggregated CD8lumenal ( 3 ) is targeted to the plasma membrane ( PM ) , whereas ER-aggregated-CD8lumenal ( 1 ) , and Golgi-re-aggregated-CD8lumenal ( 2 ) , and endogenous actin ( lower panel ) are not . Graph , normalized quantification of PM targeted CD8lumenal , with ( 3 ) set to 1 . Data represent the mean of three independent experiments , ( C ) Confocal micrograph confirming the ER localization of aggregated CD8lumenal , ( ER staples , 1 ) , the PM localization of disaggregated CD8lumenal ( 3 ) , and the Golgi retention of re-aggregated CD8lumenal ( Golgi staples , white arrow , 2 ) . All experiments were conducted in the presence of cycloheximide . DOI: http://dx . doi . org/10 . 7554/eLife . 00558 . 00710 . 7554/eLife . 00558 . 008Figure 3—figure supplement 1 . Glycosylation profile of CD8lumenal . HeLa cells expressing CD8lumenal were incubated with the disaggregating drug at 16°C for 2 hr prior to being shifted at 37°C for 2 hr in the presence ( 1 ) or absence ( 2 ) of the drug . As negative control , cells were incubated at the same temperatures but without drug at any time ( 3 ) . ( A ) Immunoblot showing that Jacalin , a lectin that binds Galatose residues , precipitates disaggregated CD8lumenal from cells incubated at 37°C ( accordingly to Figure 1D ) . Helix-Pomatia , a lectin that binds N-Acetyl-Galactosamine , precipitates disaggregated CD8lumenal from cells incubated at 16°C . DOI: http://dx . doi . org/10 . 7554/eLife . 00558 . 008 Together , these results showed that ER-aggregated CD8lumenal forms staples that remain within the ER during their lifetime . However , this is a reversible state because adding the disaggregating drug dissolved a portion of the staples and allowed disaggregated CD8lumenal to behave as a proper anterograde cargo that is glycosylated in the Golgi prior to being targeted to the cell surface membrane . Temperature blocks have allowed the selective interruption of the secretory pathway in mammalian cells . A classic example is the 20°C temperature block , which slows exit from the trans-most Golgi cisterna ( TGN ) more than it slows ER → Golgi transport or transport from cis → trans Golgi ( Matlin and Simons , 1983 ) . As a result secretory cargoes accumulate in the TGN , from which they can be synchronously released by raising the temperature to the physiological value of 37°C . Temperature blocks in the 15–18°C range slow export of secretory cargo from the ER , but slow anterograde transport in the Golgi stack even more , resulting in limited penetration of the Golgi stack with accumulation in the ER–Golgi intermediate compartment ( ERGIC ) and the cis-most Golgi cisterna or two ( Hobman et al . , 1992; Balch et al . , 1994; Volchuk et al . , 2000 ) . Following an established procedure ( Volchuk et al . , 2000 ) we employed a 16°C temperature block to load and accumulate some of the disaggregated CD8lumenal into the cis-Golgi of HeLa cells . Then , continuing at 16°C , we washout the disaggregating drug to allow for membrane staples reformation in the cis-Golgi . Finally , to explore the fate of these staples we raised the temperature to 37°C to allow anterograde transport to resume ( Figure 3A ) . Unless it is explicitly stated to the contrary , all of these experiments were performed in presence of cycloheximide to prevent protein synthesis . An independent , biochemical test confirmed that the unassembled CD8lumenal had reached the cis-Golgi but not the trans-Golgi at 16°C . Specifically , a portion of disaggregated CD8lumenal is precipitated by Helix Pomatia ( Figure 3—figure supplement 1 ) , a lectin that binds a N-Acetyl-Galactosamine residue that is only added within the cis-Golgi ( Roth et al . , 1994 ) . In contrast , the lectin Jacalin , which requires trans-Golgi localized modifications for binding ( Spicer and Schulte , 1992 ) , did not precipitate CD8lumenal at 16°C ( Figure 3—figure supplement 1 ) , although it did at 37°C ( Figure 2D ) . To explore the fate of the staples when transport is permitted , we first used a surface biotinylation method and immunofluorescence in intact cells to assess the surface exposure of CD8lumenal after release of the temperature block to allow anterograde transport to resume . Surprisingly , the CD8lumenal re-aggregated in the Golgi at 16°C did not reach the surface and was retained in the Golgi as judged by confocal microscopy ( Figure 3B , C , lane 2 ) . Golgi retention of the staples is due to controlled aggregation , because this block to transport was removed when the drug was added back , and CD8lumenal ( disaggregated ) now moves from the Golgi to the plasma membrane ( Figure 3B , C , lane 3 ) . Note that except when mentioned otherwise , we used as a loading control 10% of the pellet fraction ( insoluble fraction after sonication ) , which is almost free of soluble disaggregated-CD8lumenal ( including the glycosylated-form ) . Again we attributed the apparent low efficiency of CD8lumenal transport to transient transfection . Cells that respond to the drug show >50% transport efficiency to the PM after 2 hr chase ( Figure 3C , lane 3 ) , consistently with the efficiency of transport of regular CD8-GFP ( FM4-free ) reported previously ( Lavieu et al . , 2010 ) . Next , we used confocal microscopy to further independently delineate where the staples accumulated within the Golgi at 16°C and their fate during a chase at 20°C which allows anterograde transport to proceed to the trans-Golgi but not beyond . We confirmed that co-expressed cis- and trans-Golgi fluorescent markers ( Grasp65 and Golgin97 , respectively ) could be well discriminated by confocal microscopy and the degree of separation quantified with Pearson’s coefficient ( Figure 4A and graph ) . As a negative control we showed that ER-staples did not co-localize with either the cis- or trans-Golgi markers ( Figure 4B , 4 and graph ) . Using the same method we demonstrated that the initial localization of membrane staples Golgi staples formed at 16°C overlaps with the cis- but not the trans-Golgi marker . This co-localization with the cis- but not the trans-Golgi marker persists following warm-up to 20°C ( Figure 4B , 2 , 3 and graph ) , implying that the staples do not migrate in the cis → trans direction within the stack , even under conditions where this transport process is expected to resume . Again , the block to transport of the staples is entirely due to aggregation , because adding the drug to disaggregate the staples now allows transport and increased co-localization with the trans-Golgi marker following the warm-up to 20°C ( Figure 4B , 2 and graph ) . 10 . 7554/eLife . 00558 . 009Figure 4 . cis-Golgi retention of Golgi staples at the light microscopy level . ( A ) Confocal micrograph showing HeLa cells co-expressing DsRed-Golgin97 , a trans-Golgi marker , with GFP-Golgin97 ( upper panel ) or Grasp65-GFP , a cis-Golgi marker ( lower panel ) . Cis- and trans-Golgi are readily distinguishable at the confocal microscopy resolution . Red square , higher magnification . ( B ) HeLa cells co-expressing DsRed-CD8lumenal with GFP-Golgin97 ( 1 , 2 , 4 ) or Grasp65-GFP ( 3 ) were incubated at 16°C for 2 hr with the disaggregating drug , prior to shifting the temperature to 20°C for two additional hours in the presence ( 1 ) or in the absence ( 2 , 3 ) of the disaggregating drug . As a negative control , cells were incubated at the same temperature but without any drug at any time ( 4 ) . Confocal micrograph showed that Golgi-re-aggregated-CD8lumenal , presumably forming cis-Golgi lumenal staples , showed a better co-localization with Grasp65 ( 3 ) than with Golgin97 ( 2 ) . Disaggregated-CD8lumenal showed a stronger co-localization with Golgin97 than with Grasp65 ( 1 ) , whereas ER-aggregated-CD8lumenal is segregated from both Golgi markers ( 4 ) . Graph , Pearson’s coefficient on the entire field of view , illustrating the co-localization between the different markers . The data represent the mean of three experiments in which a total of >20 fields were counted . DOI: http://dx . doi . org/10 . 7554/eLife . 00558 . 009 Electron microscopy was used to further confirm the results from confocal microscopy . The staples were recognized in thin sections as electron dense plaques within Golgi cisternae ( arrows , Figure 5A–C ) spanning a gap of 27 ± 3 nm ( Figure 5D ) between opposing membrane surfaces , similar to their morphology within the ER . Staples formed at 16°C and then chased at 20°C for either 2 or 6 hr ( which according to confocal microscopy remained in the cis-Golgi ) were as would now be expected exclusively localized at one face of the Golgi , within the very first or more rarely the second and the third cisternae ( Figure 5A–C and graph Figure 7C for quantification ) . Immuno-electron microscopy established that this face is the cis-Golgi since the gold particles recognizing the GFP harbored by the staples co-localized with the gold particles that bind GM130 , a cis-Golgi marker ( Figure 5E ) . Confirming this , the staples were at the opposed end of the stack from clathrin , which is a trans-Golgi marker ( Figure 5F ) . 10 . 7554/eLife . 00558 . 010Figure 5 . cis-Golgi retention of Golgi staples at the EM level . HeLa cells expressing CD8lumenal were incubated at 16°C in the presence of the disaggregating drug for 2 hr , and then the drug was removed for 30 min , prior to shifting the temperature to 20°C for two to six additional hours . ( A ) – ( C ) Electron-micrograph showing that the staples ( white arrows ) remain at one face of the Golgi , within the first cistern ( A ) or in the first two cisternae ( B and C ) after 2 or 6 hr chase at 20°C , respectively . Red square , higher magnification . Scale bar , 200 nm . ( D ) Intra-lumenal space of stapled Golgi cisternae . Staples that form flat features are localized at the center of the cisternae . Their topological arrangements prevent them from localizing at the rim of the cisternae . ( E ) Electron-micrograph showing clathrin ( 10 nm gold particle ) , which label the trans-Golgi , and the staples ( 5 nm gold particles ) are at two opposite faces of the Golgi . ( F ) Gm130 ( 5 nm gold particles ) , a cis Golgi marker , is at the same face as the staples ( 10 nm gold particles ) . ( G and H ) HeLa cells expressing CD8lumenal were incubated at 20°C ( instead of 16°C ) in the presence of the disaggregating drug for 2 hr , and then the drug was removed for 30 min , prior to being incubated at 20°C for two additional hours . ( G ) Immunoblot , after surface biotinylation , showing that the staples formed in the medial/trans Golgi do not reach the PM . ( H ) Electron micrograph showing that the staples formed at 20°C are actually retained throughout the Golgi and not only at the TGN face . DOI: http://dx . doi . org/10 . 7554/eLife . 00558 . 01010 . 7554/eLife . 00558 . 011Figure 5—figure supplement 1 . Staples within the Golgi . Similar to Figure 5 Hela cells expressing CD8lumenal were incubated at 16°C in the presence of the disaggregating drug for 2 hr , and then the drug was removed for 30 min , prior to shifting the temperature to 20°C for six additional hours . Electron micrographs show the staple at one face of the Golgi ( arrow ) , which is opposite the swollen face of the Golgi ( typical of trans Golgi face at 20°C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00558 . 011 Consistently , lectin-precipitation experiment showed that staples formed at 16°C and chased at 20°C harbored a cis-glycan signature ( precipitated by Helix pomatis ) but not the trans-glycan signature ( not precipitated by Jacalin ) ( Figure 3—figure supplement 1 , lane 2 ) Altogether , the above experiments reveal that membrane staples formed in the first cisterna of the Golgi stack do not move across the stack , even after several hours . Interestingly , the staples accumulate at the central portion of the cis-most cisternae ( Figure 5A–C and Figure 5—figure supplement 1 ) being excluded from their rims . This could be the consequence of the fixed dimension of the staples , which are too small to fit across a dilated rim ( Figure 5D ) . So , the staples would be expected to preferentially form in the flattened central regions of a cisterna . Because large soluble aggregates such as collagens can move forward rapidly across the stack ( Bonfanti et al . , 1998 ) it was unexpected to observe that membrane staples remain fixed in cis-most cisternae . We therefore wanted to see if staples deposited in later Golgi cisternae might also be static . To deposit staples throughout the Golgi stack we allowed CD8lumenal to enter the Golgi at 20°C in the presence of drug to maintain it in the disaggregated state . Then , the drug was removed at 20°C . Surface biotinylation established that staples formed at 20°C remained at the Golgi even after the cells were warmed up to 37°C for 2 hr ( Figure 5G ) , similar to the cis-Golgi staples . Electron microscopy revealed that the staples formed at 20°C were present and retained throughout the entire Golgi stack , including the trans-most Golgi cisterna ( Figure 5H ) . This is not surprising because significant back-up of cargo in the stack at 20°C has been previously documented ( van Deurs et al . , 1988 ) . The fact that the staples are retained in place at every level of the Golgi stack further emphasizes the static nature of the cisterna on the time scale being studied ( 5–6 hr ) and also rules out the caveat that the apparently stable cis localization of staples at 16°C results from anterograde transport followed by efficient retrieval to the cis cisternae . To rigorously conclude that the staples are immobile during ongoing anterograde transport , we need to demonstrate positively that anterograde transport continues in the same stacks that retain the staples . To test this , we introduced a well-studied anterograde cargo into cells harboring staples , the VSV-encoded G protein . Specifically , we used a GFP-tagged version of a temperature-sensitive mutant G protein that is retained in the ER at 39°C , but which can exit the ER and be transported to the cell surface via the Golgi when the temperature is subsequently lowered to 32°C ( Presley et al . , 1997 ) . In these experiments , the staples were labeled with DsRed ( DsRed-CD8luminal ) and co-expressed with GFP-tagged VSV-G . We estimated by confocal microscopy that >95% of the cells were co-transfected . We used the 39°C temperature block to retain VSV-G-GFP in the ER , then a fraction of both VSV-G-GFP and disaggregated DsRed-CD8luminal ( +drug ) were partially released into the Golgi at 16°C , prior to re-aggregating CD8lumenal ( −drug ) at 16°C into cis-Golgi staples . Only then was the temperature raised to 32°C for 2 hr to test the transport of VSV-G-GFP ( the fraction localized into the stapled cis-Golgi and the fraction that was still into the ER ) . Surface biotinylation showed that the amount of VSV-G detected at the surface of cells that harbored staples in their Golgi was 91 ± 12% of the amount of VSV-G detected at the surface of cells harboring a normal Golgi ( Figure 6A ) . This showed that the stapling procedure did not interfere dramatically with the transport of classical anterograde cargo . 10 . 7554/eLife . 00558 . 012Figure 6 . Unaltered anterograde transport within short and long-term stapled-Golgi . HeLa cells co-expressing DsRed-CD8lumenal with VSVG-GFP were incubated overnight at 39°C in the absence of the disaggregating drug ( 1 ) prior to shifting the temperature to 16°C for 2 hr in the presence of the disaggregating drug ( 4 ) . Then the temperature was shifted to 32°C in the presence ( 2 ) or in the absence ( 3 ) of the drug . ( A ) Immunoblot , after surface biotinylation , showing that VSVG-GFP is targeted to the PM regardless of the re-aggregation status of CD8lumenal within the Golgi . Graph , normalized quantification of VSVG-GFP ( green bar ) and DsRed-CD8lumenal ( red bar ) . Data represent the mean of three independent experiments . ( B ) HeLa cells co-expressing VSVG-PAGFP and DsRed-CD8lumenal were incubated at 20°C in the presence of the disaggregating drug for 1 hr , prior to removing the drug for 4 hr to form staples throughout the Golgi . VSVG-PAGFP was photo-activated ( arrow ) within the stapled Golgi and VSVG-PAGFP fluorescent signal was measured over time at 32°C . HeLa cells co-expressing VSVG-PAGFP with ST-RFP were used as a control . Confocal micrographs illustrating the release of VSVG-PAGFP from long-term stapled-Golgi . Graph , normalized fluorescence quantification of VSVG-PAGFP over the time in stapled Golgi ( green ) and control Golgi ( red ) . Data represent the mean of at least three different experiments . ( C ) Saos-2 cells expressing GFP-CD8lumenal were incubated at 20°C in the presence ( 1 , 2 ) or in the absence ( 3 ) of the disaggregating drug . When required Golgi staples were formed for 2 hr at 20°C ( 2 ) , prior to shifting the temperature to 37°C in the presence of ascorbate to allow collagen-I release . Fresh media was added at the beginning of the chase . Immunoblot , after surface biotinylation and TCA precipitation of the media , showing that collagen-I is secreted whether or not staples are pre-positioned in the Golgi ( 2 ) or ER ( 3 ) . Cell fractions that are used as loading control correspond to the Triton X-100 soluble fraction after sonication ( 10% input lysate ) . Graph , normalized export of CD8lumenal ( green bar ) and Collagen-I ( red bar ) . Data represent the mean of two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 00558 . 01210 . 7554/eLife . 00558 . 013Figure 6—figure supplement 1 . Golgi staples do not inhibit the rate of secretion of Collagen-I and MMP2 . Saos-2 cells expressing GFP-CD8lumenal were incubated at 20°C for 1 hr in the presence of the disaggregating drug . When required Golgi staples were formed for 2 hr ( 2 ) , prior to shifting the temperature to 37°C in the presence of ascorbate to allow collagen-I release . The media of each sample ( one per time point ) was harvested at the indicated time and subjected to TCA precipitation . Cells corresponding to the 2 hr time point were submitted to surface biotinylation . Cell fractions that are used as loading control correspond to the Triton X-100 soluble fraction after sonication ( 10% lysate ) . Immunoblots show the rate of transport of MMP2 and collagen-I ( measured by densitometry , see graph ) , and the status ( Glycan-induced gel mobility reduction ( * ) and PM targeting ) of GFP-CD8lumenal . DOI: http://dx . doi . org/10 . 7554/eLife . 00558 . 01310 . 7554/eLife . 00558 . 014Figure 6—figure supplement 2 . Golgi staples do not prevent redistribution of Golgi membranes within ER mediated by BFA . HeLa cells co-expressing ManII-DsRed and GFP-CD8lumenal were incubated at 20°C in the presence of the disaggregating drug for 1 hr . Then , when required , the drug was removed to promote staples formation within the Golgi ( upper panel ) . BFA ( 10 μg/ml ) was added for 30 min , then cells were fixed and prepared for confocal microscopy . Confocal micrographs show redistribution of CD8lumenal and ManII-DsRed under BFA treatment regardless of the aggregation status of CD8lumenal . White arrows , residual staples . DOI: http://dx . doi . org/10 . 7554/eLife . 00558 . 01410 . 7554/eLife . 00558 . 015Figure 6—figure supplement 3 . Staples are laterally mobile and do not perturb lateral diffusion of Golgi resident enzymes . HeLa cells expressing GFP-CD8lumenal were incubated in the absence of the disaggregating to keep the protein aggregated within the ER . FRAP was performed at 20°C ( ER staples ) . As a negative control , cells were fixed 15 min with 4% PFA prior to performing FRAP . For Golgi localization of CD8lumenal , the disaggregating drug was added for 2 hr at 20°C to load the disaggregated cargo within the Golgi . FRAP was performed at 20°C in the presence of the drug ( Disaggregated CD8 ) . To monitor the lateral diffusion of the Golgi staples , drug was removed for 30 min prior to monitoring the fluorescence recovery of the Golgi staples at 20°C in the absence of the disaggregating drug ( Golgi staples ) . Finally , ManII-GFP was co-expressed with DsRed-CD8lumenal , staples were positioned within the Golgi as explained above , and FRAP of ManII-GFP was performed at 20°C in the absence of the disaggregating drug ( ManII-GFP [Stapled Golgi] ) . ROI used for FRAP analysis corresponded approximately to 1/3 of the Golgi . Graph shows the fluorescence intensity signal overtime . The data for each condition represents the mean of two or three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 00558 . 015 To be certain that the very Golgi harboring static staples continued to engage in productive anterograde transport , we measured the rate of efflux of VSV-G from Golgi areas harboring staples as compared to those that did not . We used photo-activatable GFP-tagged VSV-G ( PAGFP-VSV-G [Patterson and Lippincott-Schwartz , 2002] ) to activate a cohort of protein within a Golgi area to enable us to measure the release of VSV-G from that same Golgi complex . The rate of the release ( half-time ∼ 5 min ) of PA-GFP-VSV-G from even long-term ( 4 hr ) stapled Golgi was indistinguishable from that measured from a non-stapled Golgi identified with a red-tagged Golgi marker ( ST-RFP ) ( Figure 6B ) . This observation and the fact that stapled-Golgi remain properly stacked established that stapled-Golgi is fully functional for transport . Then we asked if Golgi-staples could interfere with the transport of endogenous large cargo . We used Saos-2 cells , which secrete endogenously produced collagen-I . After transfection with GFP-CD8lumenal , Saos-2 cells were incubated at 20°C in the presence of the disaggregating drug , followed by drug removal to allow the positioning of the staples within the Golgi ( Figure 6C upper panel , lane 1 , 2 show the glycosylated form of CD8lumenal ) . 2 hr after stapling the Golgi membranes , the culture media was removed and cells were incubated at 37°C in the presence of fresh media containing ascorbate to release collagen-I from the ER . After a 2 hr chase the amount of newly secreted collagen-I was determined by TCA precipitation/immunobloting procedure ( Figure 6C ) . Cells harboring stapled-Golgi for 2 hr showed 92 . 5 ± 13% of freshly secreted collagen-I when compared to cells harboring normal Golgi ( set to 100% ) . This suggests that , as reported above with for VSV-G , anterograde transport of Collagen-I is not altered when staples are positioned within the Golgi . To rigorously assess if Golgi staples had only minor effect on secretion , we further tested the rate of transport of collagen-I and MMP2 , two proteins secreted by Saos-2 cells . The kinetics of secretion for both proteins did not show any robust differences whether the Golgi apparatus was stapled or not ( Figure 6—figure supplement 1 ) . Again , this suggests that the amount of staples that is loaded in the Golgi membranes during the re-aggregation procedure does not poison the anterograde transport function of the Golgi . To further test if Golgi staples interfere with the behavior of Golgi membranes , we assessed the redistribution of Golgi resident enzymes into the ER under BFA treatment ( Lippincott-Schwartz et al . , 1989 ) . Redistribution of Golgi enzymes into the ER was observed whether Golgi staples were present or not ( Figure 6—figure supplement 2 ) . Note that consistently with the moderate disaggregation reported above ( 44% solubilization in Triton X100 ) , some aggregates remained in the ER ( arrows Figure 6—figure supplement 2 ) . These ER-remaining aggregates did not perturb the behavior of the Golgi resident enzymes ( Golgi targeting and BFA-redistribution ) . Importantly , FRAP experiments ( Figure 6—figure supplement 3 ) showed that staples ( localized either within the Golgi or within the ER ) readily diffuse within the cisternal membranes and do not perturb the diffusion of Golgi resident enzymes; in fact , Golgi-staples diffused only slightly more slowly than disaggregated CD8lumenal at 20°C ( Figure 6—figure supplement 3 ) . This would explain in a satisfying manner why the essential functions of the Golgi ( and ER ) such as transport and glycosylation are not perturbed by the presence of the staples , since these behave like any cargo or resident protein within the membrane , as further evidenced by the re-distribution of staples ( like any other Golgi cargo or resident enzyme ) to the ER as a result of BFA treatment . Retrograde transport is also a major aspect of Golgi function , and it would potentially be of interest to investigate if staples interfere with such function . While we cannot presently rule out this caveat , even if retrograde transport were blocked it would not diminish our conclusion that anterograde transport of both small and large cargo ( next section ) can occur at normal speeds without movement of cisternae in the stack . Previously , we found that soluble aggregates that are compositionally similar to the staples ( the main difference being that they were not membrane-anchored ) rapidly move forward through the Golgi , analogous to collagens ( Volchuk et al . , 2000 ) . We noted that these soluble aggregates were concentrated at the rims of the Golgi cisternae , analogous to collagens . Given the prima facia contradiction between our current observation of immobile staples and our prior observation of mobile soluble aggregates composed of virtually the same protein , we repeated the earlier studies and compared the fates of soluble aggregates and staples in the same cells . We first asked whether lumenally-expressed soluble FM4 ( FM4-hGH ) and membrane-bound FM4 proteins ( CD8lumenal ) co-aggregate in the cell . This would be expected because they share the same homotypic adhesion FM modules . As expected , when the two FM-containing proteins were present in the ER they were efficiently co-immunoprecipitated from cell extracts ( Figure 7—figure supplement 1 ) . However , when these same proteins were disaggregated and allowed to leave the ER and then re-aggregated for 1 hr in the cis-Golgi at 16°C , or even further in the Golgi stack at 20°C , the Golgi-modified forms of these same proteins were no longer co-immunoprecipitated ( Figure 7—figure supplement 1 ) . On the contrary homo-aggregates constituted of soluble cargo ( hGH ) were efficiently formed even after re-aggregation at 16°C or 20°C ( Figure 7—figure supplement 1 ) . This suggests that hetero-aggregates are not formed efficiently within the Golgi cisternae . Confocal microscopy independently established that the soluble and membrane-bound FM proteins segregated into separate aggregates within the same Golgi area ( Figure 7—figure supplement 2 ) and , electron microscopy confirmed that this separation occurs even within the same cisternal compartment . Even when both reside in the same cisterna , the electron-dense plaques comprising the staples are localized in the central regions of the cisterna while the soluble aggregates are concentrated at the dilated rims ( Figure 7E ) . 10 . 7554/eLife . 00558 . 016Figure 7 . Unaltered transport of soluble aggregates within stapled Golgi . HeLa cells co-expressing GFP-CD8lumenal and DsRed-hGH , both harboring the FM-aggregation domains , were incubated at 16°C for 2 hr in the presence of the disaggregating drug ( 3 , negative control ) prior to shifting the temperature to 37°C in the presence ( 1 ) or in the absence of the drug ( 2 ) . ( A ) Immunoblot , after surface biotinylation and TCA precipitation of the media , showing that DsRed-hGH is secreted regardless of its own re-aggregation status , or the re-aggregation status of CD8lumenal . Prior to being secreted disaggregated and re-aggregated DsRed-hGH are cleaved by furin . FL , full-length isoform of DsRed-hGH . FC , furin-cleaved isoform of DsRed-hGH . Graph , normalized quantification of PM targeted CD8lumenal ( green bar ) and normalized secreted DsRed-hGH ( red bar ) . Data represent the mean of three independent experiments . ( B ) Cells were incubated at 16°C for 2 hr in the presence of the disaggregating drug , then the temperature was shifted to 20°C for two additional hours in the absence of the drug . Electron micrograph showing Golgi staples ( green star ) that remains at the cis-Golgi , which is adjacent to ER membranes , whereas the soluble aggregates ( red star ) is localized to the opposite face of the Golgi . ( C ) Graph , distribution of staples and soluble aggregates within the Golgi stack: cells expressing CD8lumenal or hGH were subjected to the re-aggregation procedure at 16°C , prior to being incubated at 20°C for 2 hr and prepared for classical EM . The distribution of the aggregates ( soluble and staples ) within the Golgi stack is shown as a percentage of the total in Golgi areas . ( D ) Electron micrograph showing segregation between Golgi lumenal staple ( green star ) and soluble aggregates ( red star ) . Golgi staples form one flat feature in the center of one cisterna whereas the soluble aggregates form spherical aggregates that are localized at the rims of the cisternae throughout the Golgi . ( E ) HeLa cells expressing hGH or CD8lumenal were subjected to the re-aggregation procedure at 16°C for 2 hr prior to being processed for conventional EM . The upper panel shows the distribution of the soluble aggregates at the rim of the cistern , the lower panel illustrates the central positioning of the staple . Graph , lateral distribution coefficient ( r ) of staples and soluble aggregates within Golgi cisternae . The distances that separate the middle of the aggregates from each end of the cisterna ( d1 and d2 ) were measured , and ( r ) was calculated as follow r = ( d1 − d2 ) / ( d1 + d2 ) with d1 ≥ d2 . As a result , ( r ) → 0 for object at the center , whereas ( r ) → 1 for object at the rim . DOI: http://dx . doi . org/10 . 7554/eLife . 00558 . 01610 . 7554/eLife . 00558 . 017Figure 7—figure supplement 1 . Co-immunoprecipitation of ER-staples and ER-soluble aggregates . HeLa cells co-expressing GFP-CD8lumenal or GFP-hGH , with DsRed-hGH , both harboring the FM-aggregation domains , were incubated overnight in the presence of the disaggregating drug ( 1 ) or in the absence of the disaggregating drug to allow aggregation within the ER ( 2 ) . Cells were incubated for 2 hr with the disaggregating drug at 20°C ( 3 ) or 16°C ( 4 ) prior to removing the drug to trigger re-aggregation over 1 hr . Cell extracts were incubated with an anti-GFP antibody and Protein-A agarose beads . After elution , samples were subjected to SDS-PAGE . Immunoblots showing that hetero-interactions are only detected within the ER after longtime incubation , whereas homo-interactions are detected even within the Golgi at 16°C or 20°C . DOI: http://dx . doi . org/10 . 7554/eLife . 00558 . 01710 . 7554/eLife . 00558 . 018Figure 7—figure supplement 2 . Segregation of staples from soluble aggregates within the Golgi at the light level . HeLa cells co-expressing GFP-CD8lumenal and DsRed-hGH , both harboring the FM-aggregation domains , were incubated at 20°C for 2 hr in the presence of the disaggregating drug prior to shifting the temperature to 37°C in the absence of the drug . ( A ) Confocal micrograph showing the localization of both aggregated proteins within the Golgi . Graph , intensity profile showing the lack of co-localization . DOI: http://dx . doi . org/10 . 7554/eLife . 00558 . 018 What are the fates of soluble aggregates as compared to staples in the Golgi when anterograde transport is permitted to resume at 37°C ? Whereas the Golgi retained the staple protein it released the soluble aggregates to the secretory pathway ( Figure 7A ) . This biochemical result confirms yet again that hetero-aggregates ( composed of staples and soluble FM proteins ) were not formed efficiently within the Golgi , confirming by yet another approach that the two FM proteins aggregate separately within the Golgi cisternae ( even when using a 2:1 membrane:soluble protein ratio ) . Electron microscopy was also used to follow the fate of the two types of aggregates . When aggregates accumulated in the cis-Golgi are allowed to move anterograde through but not exit the Golgi following warm-up to 20°C , ≈75% of the staples , as expected , remain at the cis-face of the Golgi , whereas ≈75% of the soluble aggregates are transported to the opposite , trans-face of the same Golgi stacks ( Figure 7B and graph ) . We conclude that the anterograde transport process permits large soluble aggregates to be carried forward between cisternae that are immobile . Does the presence of staples perturb anterograde transport ? On the one hand , because staples freely diffuse like any other membrane constituent it is hard to imagine how a moderate amount of staples would broadly perturb Golgi transport or function such as glycosylation , and there is no positive evidence from our studies that they do so . On the other hand , because they span and link between two sides of the cisterna , it is easy to see why staples should remain in the thin centers of the cisternae , far from the dilated rims ( whose width is too great to accommodate the staples ) where transport vesicles bud , and also why staples would be sterically excluded from budding COPI or other vesicles if they did occasionally reach a rim , and thus would remain within the cisterna . With this in mind , our findings can be simply summarized as follows: 1 ) membrane staples localize to the central regions of Golgi cisternae where they segregate from large soluble aggregates which mainly concentrate at the dilated rims; and 2 ) membrane staples remain in place as anterograde transport occurs , whereas large soluble aggregates and small cargo are transported at normal rates . The obvious explanation of the above is that while the cisternae in the stack ( containing staples mainly in their central regions ) are static , the rims of cisternae—mainly containing the soluble aggregates—are mobile ( at least on the average ) in the anterograde direction , a process we term ‘rim progression’ . However , the unique nature of the artificial staples prevents us from definitively concluding that ‘pure maturation’ does not occur in region of the Golgi stacks that are depleted of staples . Although we established by using broad sets of assays that anterograde transport through stapled Golgi is not altered , we cannot rule out that staples introduced , non-obvious , minor side effects on Golgi . It would be very easy to mis-interpret rim progression as cisternal progression if the only marker being studied is in the rim ( Bonfanti et al . , 1998 ) . In retrospect , the principal evidence for cisternal progression in Golgi stacks was the mobility of large aggregates like collagens that are located to the rims ( Bonfanti et al . , 1998 ) . But in retrospect a stricter interpretation of that data would have been to conclude that the ‘at least some portions’ of the cisterna containing the collagen aggregates ( i . e . , the rims ) are mobile , without reaching any conclusion with respect to the rest of the cisterna . A striking feature of the collagen data is that while the majority of the aggregates are indeed rapidly transported , a sizable minority ( ∼20% [Bonfanti et al . , 1998] ) remains in the Golgi cisternae , mainly in the centers of the cisternae , consistent with our results and interpretation . We do not know how the rims of otherwise stable stacks progress , but the most likely possibility would be cycles of fission and fusion that are known to take place in the Golgi ( Pfeffer , 2010 ) . Pfeffer has considered how this could be organized in the cis → trans direction by known Rab cascades in the Cisternal Progenitor model , which could well be the basis for the transport of rims dilated with aggregates . We would imagine that after fission a transiently separated rim would nonetheless remain in register in the stack because its membrane would still be engaged in stacking/tethering interactions . Pulse-chase experiments ( Volchuk et al . , 2000 ) showed the appearance of detached rims closely-associated with the edges of the stack that transiently appeared during the transport of soluble aggregates ( which we termed mega-vesicles at the time ) . Retention of rims in transit would facilitate orderly rim progression . To the extent that the bulk of a Golgi stack is static on the time scale of anterograde transport , this reduces the quantitative demand on retrograde transport of Golgi resident proteins and the need for cisternal maturation . In budding yeast , where cisternae generally do not stack , the process of cisternal maturation is well documented , though formal proof that anterograde cargo is present in maturing cisternae is still lacking ( Losev et al . , 2006; Matsuura-Tokita et al . , 2006 ) . We do not wish to conclude that cisternal maturation does not occur in the stacked Golgi of complex plants and animals , or that it may not be important in re-assembly of the Golgi after cell division , only that it does not appear to be required on the time scale of biosynthetic protein transport in interphase in stacked Golgi . How are smaller cargoes that do not create dilated rims transported across the Golgi if the cisternae are static ? We propose that rim progression is dedicated to large soluble aggregates that due to physical constraints are forced into the rims away from adhesive centers of cisternae , and differs mechanistically from vesicle budding . COPI vesicles are known to bud extensively from the rims of Golgi cisternae at every level and contain–in addition to retrograde-directed cargo–anterograde-directed cargo such as insulin , albumin , and VSV-G ( Rothman , 2010 ) ; so this biochemically well-understood pathway ( Popoff et al . , 2011 ) would seem to be the principal contender . There may well be other , parallel pathways depending on physiological state , specifically budding/fusing tubules , which have been observed especially under conditions of protein over-expression ( Ladinsky et al . , 1999; Trucco et al . , 2004 ) . Our findings , along with other evidence ( Brugger et al . , 2000; Patterson et al . , 2008 ) , point to a broad division of the Golgi cisternae into two functionally distinct domains: the centers and the rims . The rims specialize in exit and entry into the stack and are dynamic , while the centers are static and adherent to form the stack and specialize in biosynthesis and processing . Much remains to be learned about how these domains segregate and organize the multitude of tasks associated with Golgi function . HeLa cells were maintained at 37°C in 5% CO2 in DMEM ( Gibco , Grand Island , NY , United States ) supplemented with 10% FBS ( Gibco ) . Saos-2 cells were maintained with the same media supplemented with ascorbate ( 25 μg/ml ) . HeLa cells were transfected using lipofectamine 2000 ( Invitrogen , Grand Island , NY , United States ) as recommended by the manufacturer . Saos-2 cells were transfected using electroporation ( Nepa21type II model from Nepa gene , Chiba , Japan ) . CD8lumenal ( ss-GFP-FM4-CD8 and ss-DsRed-FM4-CD8 ) were generated by sequential insertion of GFP ( or DsRed ) , and CD8 encoding sequences into a pC4 ss-FM4 backbone vector ( ARIAD ) , using XbaI/SpeI compatibility and BamH1 restriction site . CD8 was amplified from pC4-CD8-GFP plasmid ( Lavieu et al . , 2010 ) . A similar strategy was used to create ss-DsRed-FM4hGH and ss-GFP-FM4hGH . PAGFP-VSVG and tsVSVG-GFP were ordered from Addgene . J Rohrer provided ST-RFP . Grasp65 and Golgin97 encoding plasmids were a gift from Y Wang and S Munro , respectively . These experiments were performed as described previously ( Lavieu et al . , 2010 ) with slight modifications . HeLa cells were grown in six-well plates . All the disaggregation/re-aggregation experiments were performed in HBSS ( Gibco ) supplemented with 10% FBS ( Gibco ) and 100 μg/ml cycloheximide ( Sigma , St Louis , MO , United States ) . Disaggregating drug ( AP21998; ARIAD , Cambridge , MA , United States ) was used at 1 μM . 16°C , 20°C , 32°C and 37°C incubations were performed using temperature-controlled incubators . Briefly , subconfluent HeLa cells were biotinylated in 1 mg/ml EZ-Link Sulfo-NHS-SS Biotin at 4°C for 30 min and washed two times with PBS . After neutralization with 100 mM glycine pH7 at 4°C for 30 min , cells were washed , detached and collected by centrifugation . Pellets were resuspended in lysis buffer ( 1% Triton X-100 , 5 mM EDTA , 150 mM NaCl pH7 . 4 and 1% protease inhibitor ) for 1 hr at 4°C and sonicated ( 80W , 10 s pulse ) . The cell lysate was centrifuged , and the supernatant was collected . Samples were incubated with neutravidin agarose resin at 4°C overnight for Biotin-labeled protein precipitation . The neutravidin interacting proteins were eluted with SDS sample buffer from neutravidin pellets . Extracted proteins were first separated in SDS-polyacrylamide gels and then transferred onto nitrocellulose membranes for immunoblotting . After blocking with fat-free milk , the membranes were incubated with appropriate primary antibodies . Primary antibodies were detected by chemiluminescence using horseradish peroxidase-conjugated secondary antibodies . Fluorographs were quantitatively scanned using the NIH image software . When TCA precipitation was required , 0 . 1% FBS was used . Chase media was collected and precipitated overnight at 4°C with 10% TCA . After centrifugation TCA pellets were washed with acetone prior to being resuspended in loading buffer and analyzed by immunoblotting blot as described above . For lectin precipitation , HeLa cells extracts containing CD8lumenal were incubated at 4°C with Jacalin-immobilized lectin ( Sigma-Aldrich ) or Helix pomatia Gel-HPA-immobilized lectin ( EY Laboratories , San Mateo , CA , United States ) prior to being processed for immunoblotting as described above . For immunoblotting , we used monoclonal anti-GFP ( Roche , Brandford , CT , United States ) , polyclonal anti-mCherry ( Biovision , Milpitas , CA , United States ) , anti-Collagen-I ( SP1 . D8 from Developmental Studies Hybridoma Bank , Iowa City , IA , United States ) , and polyclonal anti-MMP2 ( Cell Signaling , Boston , MA , United States ) . HeLa cells were grown on glass coverslip in 24-well plates . For surface labeling , cells were incubated on ice for an hour with a monoclonal anti-GFP antibody ( Roche ) prior to being fixed for 10 min with 4% PFA and incubated with an Alexa Fluor 546-coupled antibody . Cells were then processed for confocal microscopy imaging , which was performed in multi-tracking mode on either a Zeiss LSM510 or a Zeiss LSM510 META . Images were analyzed using Zeiss LSM510 software or using ImageJ ( co-localization finder plugin ) . For photo-activation experiments we used a multicolor spinning-disk confocal based on an inverted Olympus microscope ( IX-71 ) and Perkin-Elmer Ultraview system with 5-laser and FRAP/photoactivation . VSVG-PAGFP was photo-activated within the Golgi region ( identified by the co-expressed DsRed-tagged protein ) at 405 nm ( 80% laser power ) for 5 s . VSVG-PA-GFP was imaged using 488 nm light ( 10% laser power ) .

Most plant and animal cells contain an organelle known as the Golgi apparatus , which consists of a series of four to six stacked cisternae . Almost all the proteins that are secreted from the cell , or targeted to its plasma membrane , transit through the Golgi . This process takes roughly 5–20 min . Although transport of proteins through the Golgi was first observed more than 50 years ago , it is still unclear exactly how this process occurs . One possibility is that proteins to be packaged move through the cisternae enclosed in vesicles , as if on a conveyor belt . Alternatively , the proteins themselves may remain stationary while the Golgi cisternae move over them . Now , Lavieu et al . provide evidence that the Golgi shows both mobile and static behaviour depending on the type and size of the cargo being processed . To distinguish between these two mechanisms , they created a new type of protein cargo—which they called a ‘staple’—that became fixed to the walls on each side of the cisternae and could not , therefore , move freely through the Golgi . They compared the processing of this protein to that of a more typical soluble protein cargo , which could move freely through the Golgi stack . Surprisingly , the Golgi processed these two types of cargo in very different ways . The staples remained embedded in the walls in the center of the cisternae , whereas the conventional soluble cargo was transported past the staples and collected at the edges of the cisternae , which are known as rims . These are wider than the center of the cisternae , and the staples are too narrow to span them . Lavieu et al . suggest that the Golgi cisternae can be divided into two functionally distinct domains: the centers of cisternae , which remain stationary , and the edges or rims , which can move . In addition to increasing our understanding of how proteins are prepared for transport inside cells , this new mechanism reconciles seemingly conflicting data by revealing that the Golgi can be both mobile and static .

2013

Stapled Golgi cisternae remain in place as cargo passes through the stack

Planar cell polarity ( PCP ) requires the asymmetric sorting of distinct signaling receptors to distal and proximal surfaces of polarized epithelial cells . We have examined the transport of one PCP signaling protein , Vangl2 , from the trans Golgi network ( TGN ) in mammalian cells . Using siRNA knockdown experiments , we find that the GTP-binding protein , Arfrp1 , and the clathrin adaptor complex 1 ( AP-1 ) are required for Vangl2 transport from the TGN . In contrast , TGN export of Frizzled 6 , which localizes to the opposing epithelial surface from Vangl2 , does not depend on Arfrp1 or AP-1 . Mutagenesis studies identified a YYXXF sorting signal in the C-terminal cytosolic domain of Vangl2 that is required for Vangl2 traffic and interaction with the μ subunit of AP-1 . We propose that Arfrp1 exposes a binding site on AP-1 that recognizes the Vangl2 sorting motif for capture into a transport vesicle destined for the proximal surface of a polarized epithelial cell . Planar cell polarity ( PCP ) governs the organization of epithelial cells along a plane parallel to the surface of the epithelium . This long range order orchestrates proper development and organ function . The establishment of PCP is regulated by a set of evolutionarily conserved signaling receptors . A key feature of these signaling receptors is that they are asymmetrically localized on the cell boundaries during PCP signaling ( Klein and Mlodzik , 2005 ) . The mechanisms that mediate the asymmetric localization of PCP signaling molecules remain unclear . One hypothesis is that interactions between PCP signaling molecules across cell junctions could stabilize their polarized localization to opposing cell boundaries ( Klein and Mlodzik , 2005; Chen et al . , 2008 ) . Proteins that organize epithelial cells include the atypical cadherin Fat , Dachsous and the Golgi resident protein Four-jointed in Drosophila which have been proposed to provide long range patterning cues to regulate PCP asymmetry ( Bayly and Axelrod , 2011 ) . Additional evidence suggests that intracellular trafficking may also contribute to the asymmetric localization of PCP signaling receptors ( Shimada et al . , 2006; Strutt and Strutt , 2008 ) . Coat-protein-mediated cargo protein sorting at the trans Golgi network ( TGN ) is an essential step of biosynthetic trafficking and regulates targeting of a variety of transmembrane cargoes to their final destinations ( Rodriguez-Boulan et al . , 2005 ) . Among the known vesicle coat proteins , clathrin adaptor complexes ( AP ) have been shown to mediate sorting of various transmembrane cargoes at the TGN by directly interacting with tyrosine- or dileucine-based sorting motifs localized within the cytosolic domain of a transmembrane cargo molecule ( Rodriguez-Boulan et al . , 2005; Burgos et al . , 2010 ) . Recently , AP-1 has been shown to functionally interact with a novel Golgi-export motif within the tertiary structure of Kir2 . 1 channel ( Ma et al . , 2011 ) . In addition to APs , a new type of coat protein complex , exomer , regulates the transport of Chs3p and Fus1p from the TGN to the plasma membrane in yeast ( Wang et al . , 2006; Barfield et al . , 2009 ) . Sorting of some soluble secretory cargo at the TGN requires the actin-severing protein ADF/cofilin and the Ca2+ATPase SPCA1 ( von Blume et al . , 2009 , 2011; Curwin et al . , 2012 ) . Assembly of coat protein complexes on membranes is initiated by Arf or Arf-like small GTPases that switch between GDP- and GTP-bound states . Upon GTP binding , Arf proteins expose an N-terminal myristoyl group attached to an amphipathic helix which mediates membrane recruitment and induces membrane curvature ( Lee et al . , 2004 , 2005; Bielli et al . , 2005; Beck et al . , 2008 ) . GTP-binding also causes a conformational change in the switch domain of Arf proteins which promotes the membrane recruitment of cytosolic effectors , including coat proteins and lipid modification enzymes ( Gillingham and Munro , 2007; Donaldson and Jackson , 2011 ) . Mammalian cells possess 6 Arf proteins and more than 20 Arf-like proteins . The intracellular roles of the majority of Arf proteins are poorly understood . A genome-wide RNA interference screen indicates that Arf1 and Arfrp1 are required for secretion of recombinant luciferase from Drosophila S2 cells ( Wendler et al . , 2010 ) . Arf1 regulates the membrane recruitment of various proteins including coats such as COPI , APs , GGAs and the lipid modification enzymes , phospholipase D and PtdIns 4-kinase ( Donaldson and Jackson , 2011 ) . Arfrp1 is essential for survival and has been shown to mediate the trafficking of VSVG , E-cadherin and the glucose transporters GLUT4 and GLUT2 as well as to regulate lipid droplet growth ( Shin et al . , 2005; Zahn et al . , 2008; Nishimoto-Morita et al . , 2009; Hesse et al . , 2010; Hommel et al . , 2010; Hesse et al . , 2012 ) but the molecular mechanisms underlying its intracellular function are unknown . Given the asymmetric distribution of PCP signaling molecules on the surface of epithelial cells , distinct sorting or coat protein complexes may be required for their traffic from the TGN . In this study , we focused on identifying the coat proteins that mediate TGN export of a conserved four-transmembrane PCP signaling receptor , Vangl2 . In Drosophila , mutation in Strabismus , the Drosophila homolog of Vangl2 , causes defects in the organization of wing hairs and induces defects in the orientation of eye ommatidia ( Taylor et al . , 1998; Wolff and Rubin , 1998 ) . In vertebrates , Vangl2 regulates convergent extension ( Torban et al . , 2004 ) . Mouse Vangl2 looptail mutants , which are defective in ER export , cause severe defects in neural tube closure and disrupt the orientation of stereociliary bundles in mouse cochlea ( Kibar et al . , 2001a , 2001b; Montcouquiol et al . , 2003; Merte et al . , 2010 ) . To explore the coat proteins that mediate TGN export of Vangl2 , we started by screening the effects on Vangl2 trafficking upon siRNA knockdown of selected Golgi-localized Arf proteins . Our analysis indicates that Arfrp1 regulates TGN export of Vangl2 . We find that AP-1 is an effector of Arfrp1 and that the two interact to regulate TGN export of Vangl2 . Interestingly , TGN export of one other PCP signaling receptor , Frizzled-6 , is independent of the Arfrp1/AP-1 machinery , suggesting that differential sorting machineries regulate the TGN export of Vangl2 and Frizzled 6 , which may contribute to their opposing localization on the epithelial cell surface . To identify the Arf proteins that regulate TGN export of Vangl2 , we performed an siRNA knockdown screen focusing on selected Golgi-localized Arf proteins in HeLa cells stably expressing HA-Vangl2 . The screen indicated that knockdown of Arf1 or Arfrp1 caused a juxtanuclear accumulation of Vangl2 whereas knockdown of other Golgi-localized Arfs did not affect Vangl2 trafficking . Arf1 , which shares a 34% sequence identity with Arfrp1 , plays a general role in regulating membrane recruitment of various vesicle coat proteins and lipid modification enzymes ( Donaldson and Jackson , 2011 ) . Arfrp1 is more specifically localized at the TGN and has been shown to regulate TGN-to-plasma membrane transport of E-cadherin and VSV-G ( Shin et al . , 2005; Zahn et al . , 2008; Nishimoto-Morita et al . , 2009 ) . However , what Arfrp1 does to promote traffic has not been explored . We thus focused on Arfrp1 and it’s role in the transport of PCP signaling proteins . The expression of Arfrp1 was efficiently reduced after siRNA treatment ( Figure 1G ) and knockdown of Arfrp1 caused a juxtanuclear accumulation of Vangl2 in a majority ( 65% ) of the cells compared to mock treated cells ( Figure 1A , D , H ) . Transport-arrested Vangl2 colocalized with the TGN marker , Golgin 97 ( Figure 1A–F ) but not the early endosomal marker EEA1 , the late endosomal marker Rab7 or the recycling endosomal marker Rab11 ( Figure 1—figure supplement 1 ) . Quantification of colocalization indicated that Vangl2 correlated more closely with the TGN marker , Golgin 97 , than with the cis-Golgi marker , GM130 ( Figure 1—figure supplement 2 ) . These results suggest that Arfrp1 regulates the export of Vangl2 from the TGN . 10 . 7554/eLife . 00160 . 003Figure 1 . Knockdown of Arfrp1 leads to accumulation of Vangl2 at the TGN . ( A ) – ( F ) HeLa cells stably expressing HA-Vangl2 were either mock transfected or transfected with siRNA against Arfrp1 . At day 3 after transfection , the cells were analyzed by indirect immunofluorescence . Size bar = 10 μM . ( G ) HeLa cell lysates from cells transfected with control siRNA or siRNA against Arfrp1 were analyzed by immunoblotting with anti-Arfrp1 antibody and , as a loading control , anti-GM130 antibody . ( H ) Quantification of the fraction of cells showing Golgi-accumulated Vangl2 in control or siRNA-treated HeLa cells stably expressing HA-Vangl2 ( N = 3; >100 cells counted for each experiment ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00160 . 00310 . 7554/eLife . 00160 . 004Figure 1—figure supplement 1 . Juxtanuclear accumulated Vangl2 in Arfrp1 knockdown cells is not colocalized with endosomal markers . HeLa cells were transfected with siRNA against Arfrp1 and re-transfected after 48 hr with a plasmid encoding HA-Vangl2 . After an additional 24 hr , cells were immunofluorescently labeled to evaluate coincident localization with HA-Vangl2 and EEA1 ( A ) – ( C ) , HA-Vangl2 and Rab7 ( D ) – ( F ) and HA-Vangl2 and Rab11 ( G ) – ( I ) . Size bar = 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 00160 . 00410 . 7554/eLife . 00160 . 005Figure 1—figure supplement 2 . Juxtanuclear accumulated Vangl2 in Arfrp1 knockdown cells colocalizes with Golgin 97 more than with GM130 . ( A ) – ( I ) HeLa cells were transfected with siRNA against Arfrp1 and re-transfected after 48 hr with plasmid encoding HA-Vangl2 . After an additional 24 hr , cells were immunofluorescently labeled to evaluate coincident localization with Golgin 97 and GM130 ( A–C ) , HA-Vangl2 and GM130 ( D–F ) and HA-Vangl2 and Golgin 97 ( G–I ) . Size bar = 10 μm . ( J ) Colocalization was quantified by analyzing the average value of the fraction of each marker's area that coincided with the other marker ( mean ± SD; >15 cells each ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00160 . 005 To elucidate the roles of Arfrp1 in TGN export of Vangl2 , we sought to identify the effectors of Arfrp1 using affinity chromatography . A similar approach documented the specific interaction between the BBsome , which functions as a coat complex that sorts membrane proteins to primary cilia , and Arl6 ( Jin et al . , 2010 ) . Bovine brain cytosol was incubated with purified GST-tagged Arfrp1 dominant negative ( T31N ) and dominant active ( Q79L ) mutant pre-loaded with GDP or GTPγS , respectively . After incubation , bound proteins were eluted and analyzed by SDS-PAGE and silver staining . A series of protein bands were recovered in the eluate of GTPγS-loaded GST-Arfrp1 ( Q79L ) immobilized on glutathione beads ( Figure 2A ) . One of the bands was identified by mass spectrometry as the γ subunit of the adaptor complex 1 ( AP-1 ) ( Figure 2A ) . Immunoblot analysis confirmed that both γ1-adaptin and μ1-adaptin preferentially interacted with the GTPγS-loaded Arfrp1 ( Q79L ) , whereas EEA1 , CRMP2 and dynamin II showed no binding or no GTP-dependent binding ( Figure 2B , C ) . Moreover , the δ subunit of AP-3 and the α subunit of AP-2 showed no detectable binding ( Figure 2C ) , suggesting the interactions between Arfrp1 and subunits of AP-1 are specific . 10 . 7554/eLife . 00160 . 006Figure 2 . Subunits of AP-1 preferentially interact with the GTP-bound Arfrp1 . ( A ) Bovine brain cytosol was incubated with purified GDP-loaded dominant negative form ( T31N ) or GTPγS-loaded dominant active form ( Q79L ) of GST-Arfrp1 . After incubation , the eluted fraction was resolved by SDS-PAGE and silver stained . Protein identification in the indicated gel slice performed by mass spectrometry revealed γ1-adaptin and serine/threonine-protein kinase ( A-Raf ) respectively . ( B ) , ( C ) . Bovine brain cytosol was incubated with purified GDP-loaded GST-Arfrp1 ( wt ) or GTPγS-loaded GST-Arfrp1 ( Q79L ) . After incubation , the entire sample of bound γ1-adaptin , μ1-adaptin and other indicated proteins was analyzed by immunoblot . DOI: http://dx . doi . org/10 . 7554/eLife . 00160 . 006 The results from the affinity isolation suggest that AP-1 is an effector of Arfrp1 , possibly cooperating to mediate TGN export of Vangl2 . Consistent with this hypothesis , the C-terminal cytosolic domain of Vangl2 contains a conserved basolateral-sorting motif ( YXXF ) which is known to interact with the AP complexes ( Bonifacino and Lippincott-Schwartz , 2003 ) ( red box , Figure 3A ) . Indeed , HA-Vangl2 is localized basolaterally in MDCK cells ( Kallay et al . , 2006 ) . To test whether this motif is important for the localization of Vangl2 , we generated a series of HA-Vangl2 mutant constructs and examined their localization . Strikingly , four Vangl2 mutants bearing mutations in the YXXF motif , including the single mutation ( F283A ) , showed no detectable surface pattern ( Figure 3E , H , K , Q ) . At high levels of expression , mutant Vangl2 was retained in the ER . However , at lower levels of expression , these Vangl2 mutant proteins accumulated in the juxtanuclear area which colocalized with the TGN marker , Golgin 97 ( Figure 3E–M , Q–S ) . A Vangl2 YXXF double mutant ( Y280A , F283A ) and Vangl2 looptail mutant ( D255E ) displayed quite distinctive localization to the TGN and ER , respectively ( Figure 3—figure supplement 1 ) . A single tyrosine mutant , Vangl2 Y280A , was only partially transport defective ( Figure 3N–P ) , whereas the double mutant Y279A Y280A resulted in a more complete arrest of mutant Vangl2 at the TGN ( Figure 3—figure supplement 2 ) . As a control , substituting alanine for both leucines adjacent to the YXXF motif ( green box , Figure 3A ) had no effect on Vangl2 localization ( Figure 3T–V ) . These results suggest that TGN export of Vangl2 depends on the conserved YYXXF motif in the C-terminal , cytosolic domain . 10 . 7554/eLife . 00160 . 007Figure 3 . TGN export of Vangl2 depends on the conserved YYXXF sorting motif in the C-terminal cytosolic domain . ( A ) Sequence alignment of Vangl1 and Vangl2 from different species indicates that Vangl2 C-terminal cytosolic domain contains a conserved YYXXF sorting motif . ( B ) – ( V ) COS7 cells were transiently transfected with plasmids encoding HA-Vangl2 wild type ( B–D ) or the indicated mutant constructs ( E–V ) . At day 1 after transfection , the cells were analyzed by indirect immunofluorescence using antibodies against HA tag and Golgin 97 . Note the contrast in panel N was adjusted to reveal the weak surface pattern of Vangl2 . Size Bar = 10 μM . DOI: http://dx . doi . org/10 . 7554/eLife . 00160 . 00710 . 7554/eLife . 00160 . 008Figure 3—figure supplement 1 . Vangl2 tyrosine mutants are not colocalized with the ER marker . COS7 cells were co-transfected with plasmids encoding the ER marker ( GFP-Bcl2-Cb5 ) and the indicated HA-Vangl2 mutant construct . At day 1 after transfection , colocalization between GFP-Bcl2-Cb5 and the indicated Vangl2 mutant construct was analyzed by immunofluorescence . Size bar = 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 00160 . 00810 . 7554/eLife . 00160 . 009Figure 3—figure supplement 2 . Vangl2 Y279A Y280A is blocked at the TGN . COS7 cells were transfected with HA-Vangl2 wild type ( A ) – ( C ) or HA-Vangl2 ( Y279A , Y280A ) ( D ) – ( F ) . After transfection for 24 hr , cells were analyzed by immunofluorescence using anti-HA and anti-Golgin 97 antibody . Size bar = 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 00160 . 009 The tyrosine-based sorting motif is known to interact with the μ subunit of the AP complexes ( Bonifacino and Lippincott-Schwartz , 2003 ) . To test whether μ1-adaptin interacts with Vangl2 via the YYXXF motif , we performed GST pull-down assays using purified GST-μ1 and lysates from COS7 cells transiently transfected with HA-Vangl2 wild-type or mutant constructs . HA-Vangl2 wild type specifically bound GST-μ1 ( Figure 4A ) . The interaction between Vangl2 and GST-μ1 was severely reduced when crucial residues of the YYXXF motif were mutated , whereas alanine substitutions of the adjacent dileucine amino acids had no effect ( Figure 4A ) . Yeast two-hybrid analysis confirmed that μ1-adaptin interacted with Vangl2 and mutation of the basolateral sorting motif , including the restrictive F283A substitution , inhibited this interaction ( Figure 4B ) . The less restrictive single tyrosine mutant , Vangl2 ( Y280A ) , interacted weakly with μ1-adaptin whereas mutating both tyrosine residues blocked interaction ( Figure 4B ) . To test whether the Vangl2 cytosolic domain directly interacts with μ1-adaptin , we purified GST-μ1 and MBP-tagged Vangl2 C-terminal domain proteins . The MBP-Vangl2 C-terminal domain bound GST-μ1 whereas mutation of the YYXXF motif blocked this interaction ( Figure 4C ) , consistent with a direct and signal-dependent interaction . The interaction pattern correlated well with the Vangl2 mutant localization analysis in transfected cells ( Figure 3B–V ) . 10 . 7554/eLife . 00160 . 010Figure 4 . μ1-adaptin directly interacts with Vangl2 C-terminal cytosolic domain in an YYXXF-motif dependent manner . ( A ) Cell lysates from COS7 cells transiently transfected with plasmids encoding HA-Vangl2 wild-type or the indicated Vangl2 mutant constructs were incubated with glutathione beads bearing similar amounts of GST or GST-μ1 . The entire sample of bound HA-Vangl2 was evaluated by immunoblotting with anti-HA antibody . ( B ) Yeast two-hybrid analyses recapitulated the results of the GST-pull down assay . Serial dilutions of the yeast colonies co-expressing the indicated constructs were dotted on the correspondent selective media . Pictures were taken after 3 days of growth . ( C ) Purified MBP-Vangl2 C-terminus wild type , or the indicated mutant constructs were incubated with glutathione beads bearing GST-μ1 . The entire sample of bound MBP-Vangl2 was evaluated by immunoblot . DOI: http://dx . doi . org/10 . 7554/eLife . 00160 . 010 To test whether AP-1 mediates TGN export of Vangl2 , we knocked down the expression of the μ and γ subunits of AP-1 in HeLa cells transiently transfected with HA-Vangl2 . Immunoblot analysis showed that the expression of μ1- and γ1-adaptins was significantly reduced after siRNA treatment ( Figure 5A ) . As before , we focused on cells expressing lower levels of HA-Vangl2 and observed an accumulation of HA-Vangl2 in the juxtanuclear area , colocalized with Golgin 97 , with weak or no detectable surface labeling in over 60% of the treated cells ( Figure 5E–J and quantification in Figure 5K ) . Around 20% of mock-treated cells displayed Golgi-localized Vangl2 ( Figure 5K ) but retained strong surface labeling . As a control , knockdown of μ3-adaptin , which did not bind Vangl2 ( not shown ) , or knockdown of δ3-adaptin had no significant effects on the localization of Vangl2 ( Figure 5K ) . The interaction data and knockdown analysis suggest that AP-1 directly mediates TGN export of Vangl2 . 10 . 7554/eLife . 00160 . 011Figure 5 . Knockdown of μ1-adaptin or γ1-adaptin accumulates Vangl2 at the TGN . ( A ) HeLa cells were mock transfected or transfected with siRNA against the indicated subunit of the AP-1 or AP-3 complex . At day 3 after transfection , total cell lysates were analyzed by immunoblotting with antibody against the indicated adaptin subunits or , as loading controls , p115 and tubulin . ( B ) – ( J ) HeLa cells were mock transfected ( B–D ) or transfected with siRNAs against μ1-adaptin ( E–G ) or γ1-adaptin ( H–J ) and re-transfected after 48 hr with plasmid encoding HA-Vangl2 . After an additional 24 hr , cells were analyzed by immunofluorescence . Size bar = 10 μM . ( K ) Quantification of the fraction of cells showing Golgi-accumulated Vangl2 ( N = 3; >150 cells expressing lower levels of Vangl2 counted for each experiment ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00160 . 011 In order to assess the role of Arfrp1 and AP-1 in the sorting of Vangl2 , we evaluated the interaction of pure components with synthetic membranes . First , we examined the recruitment of AP-1 to activated Arfrp1 using a liposome flotation assay . Purified Arfrp1-His associated with liposomes in the presence of GTPγS but not GDP ( Figure 6A ) . Using the same flotation assay , we observed AP-1 complex recruited to liposomes incubated with GTPγS-Arfrp1-His , but not to those incubated with GDP-Arfrp1-His ( Figure 6B , C ) . These results suggest that Arfrp1 binds AP-1 on the surface of liposomes in a concentration-dependent manner . 10 . 7554/eLife . 00160 . 012Figure 6 . Arfrp1 directly recruits purified AP-1 complex to liposomes and this process is stimulated by Vangl2 C-terminal cytosolic domain . ( A ) Purified Arfrp1-His was incubated with liposomes labeled with Texas Red-PE in the presence of GDP or GTPγS . After centrifugation , fractions were collected from the bottom to the top and analyzed by immunoblotting using anti-His antibody . ( B ) , ( C ) . Liposomes were sequentially incubated with Arfrp1-His at the indicated concentration in the presence of GDP or GTPγS , then with purified AP-1 complex . After centrifugation , the top fractions were collected , scanned to reveal fluorescence in the Texas Red channel as an indicator of the amount of liposomes and analyzed by immunoblotting using anti-His and anti-γ1 antibodies ( B ) and the levels of γ1-adaptin normalized to the amount of lipids were quantified ( C ) . ( D ) , ( E ) . Liposomes were sequentially incubated with Arfrp1-His alone or Vangl2 cytosolic domain alone or both , then with purified AP-1 complex . After centrifugation , the top fractions were collected , scanned to reveal fluorescence in the Texas Red channel , and analyzed by immunoblotting using anti-γ1 and anti-His antibodies ( D ) and the levels of γ1-adaptin normalized to the amount of lipids were quantified ( E , N =2 ) . ( F ) , ( G ) . Liposomes were sequentially incubated with Arfrp1-FLAG or Arf1-FLAG in the presence or absence of Vangl2 cytosolic domain , then with purified AP-1 complex . After centrifugation , the top fractions were collected , scanned to reveal fluorescence in the Texas Red channel and analyzed by immunoblotting using anti-γ1 and anti-FLAG antibodies ( F ) and the levels of γ1-adaptin normalized to the amount of lipids were quantified ( G , N = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00160 . 01210 . 7554/eLife . 00160 . 013Figure 6—figure supplement 1 . Sorting signal-dependent binding of Arfrp1 to Vangl2 in cell lysates; Vangl2 binds Arfrp1 more efficiently than Arf1 . ( A ) Cell lysates from COS7 cells transiently transfected with plasmids encoding HA-Vangl2 were incubated with glutathione beads bearing similar amounts of GTPγS-loaded GST-Arf1 or GST-Arfrp1 . After incubation , the entire sample of bound HA-Vangl2 was detected by immunoblot . ( B ) Cell lysates from COS7 cells transiently transfected with plasmids encoding Vangl2 wild type or the indicated Vangl2 mutant constructs were incubated with glutathione beads bearing similar amount of GTPγS-loaded Arfrp1 . The entire sample of bound HA-Vangl2 was evaluated by immunoblot . DOI: http://dx . doi . org/10 . 7554/eLife . 00160 . 013 Next , we sought to analyze whether Arfrp1 , in association with AP-1 , could recruit Vangl2 cytosolic domain to liposomes . We were unable to address the recruitment of Vangl2 directly because purified Vangl2 cytosolic domain bound liposomes by itself . As an alternative approach , we evaluated the influence of the Vangl2 cytosolic domain and Arfrp1-GTPγS on the membrane recruitment of AP-1 . As shown in Figure 6D , E , membrane recruitment of AP-1 was enhanced approximately threefold in the presence of both Vangl2 cytosolic domain and Arfrp1-GTPγS . Importantly , a Vangl2 sorting signal mutant , Y279A Y280A , failed to stimulate Arfrp1-mediated AP-1 recruitment . These results suggest that the Vangl2 sorting signal enhances AP-1 recruitment to membranes containing Arfrp1-GTP . Arf1 also mediates membrane recruitment of AP-1 . A peptide containing the mannose-6-phosphate receptor sorting signal stimulates Arf1-mediated membrane recruitment of AP-1 to liposomes ( Zhu et al . , 1998 , 1999; Lee et al . , 2008 ) . We evaluated the effect of the Vangl2 cytosolic domain on Arf1-mediated AP-1 recruitment using FLAG-tagged Arf1 and Arfrp1 purified from mammalian cells . In contrast to incubations containing Arfrp1-GTPγS , Vangl2 C-terminal domain did not stimulate AP-1 recruitment to liposomes in the presence of Arf1-GTPγS ( Figure 6F , G ) . This result suggests that the stimulation effect is specific for Arfrp1 and indicates that Arfrp1- but not Arf1- associated AP-1 provides a preferred binding site for the Vangl2 sorting signal . As expected , HA-Vangl2 from COS7 cell lysates interacted with GST-Arfrp1 but weakly with GST-Arf1 ( Figure 6—figure supplement 1A ) . The interaction between GST-Arfrp1 and HA-Vangl2 depended on the YYXXF motif ( Figure 6—figure supplement 1B ) suggesting that Arfrp1 interacts with Vangl2 indirectly through the AP-1 complex . Vangl2 and Frizzled-6 localize on opposing surfaces at cell–cell junctions in epithelial tissues . Because the TGN is a cargo sorting station , it is possible that Frizzled-6 and Vangl2 may use different vesicle sorting machineries to exit the TGN . Unlike Vangl2 , Frizzled 6 was inefficiently transported to the cell surface in transfected HeLa cells . However , when Frizzled-6 was co-expressed with Celsr1 , an atypical cadherin , both proteins co-localized at cell junctions ( Figure 7A–C ) ( Devenport and Fuchs , 2008 ) . Unlike Vangl2 , knockdown of Arfrp1 or μ1-adaptin had no detectable effects on the localization of Frizzled-6 and Celsr1 ( Figure 7D–I ) . Frizzled-6 and Celsr1 have no known tyrosine- or dileucine-based sorting motifs in their cytosolic domains . To test whether Arfrp1 or μ1-adaptin interact with Frizzled-6 or Celsr1 , we performed GST-pull down analysis as before . GST-Arfrp1 and GST-μ1 bound HA-Vangl2 but not GFP-Frizzled-6 or GFP-Celsr1 in cell lysates from COS7 cells co-transfected with HA-Vangl2 and GFP-Celsr1 ( Figure 7J ) or co-transfected with HA-Vangl2 and GFP-Frizzled 6 ( Figure 7K ) . These results suggest that sorting of Frizzled 6 and Celsr1 at the TGN is independent of the Arfrp1/AP-1 machinery . 10 . 7554/eLife . 00160 . 014Figure 7 . TGN export of Frizzled-6 and Celsr1 is independent of the Arfrp1/AP-1 machinery . ( A ) – ( I ) . HeLa cells were either mock transfected ( A–C ) or transfected with siRNA against Arfrp1 ( D–F ) or μ1-adaptin ( G–I ) and re-transfected after 48 hr with plasmids encoding GFP-Celsr1 and Myc-Frizzled 6 . After an additional 24 hr , cells were analyzed by immunofluorescence . Size bar = 10 μm . ( J ) , ( K ) . Cell lysates ( 250 μl ) containing 1 mg/ml proteins from COS7 cells co-transfected with HA-Vangl2 and GFP-Celsr1 ( J ) or HA-Vangl2 and GFP-Frizzled 6 ( K ) were incubated with glutathione beads bearing 1 μg of GST , GTPγS-loaded GST-Arfrp1 or GST-μ1 . The entire sample of bound HA-Vangl2 , GFP-Celsr1 or GFP-Frizzled 6 were detected by immunoblot . DOI: http://dx . doi . org/10 . 7554/eLife . 00160 . 014 In addition to Vangl2 , Arfrp1 is known to regulate TGN-to-plasma membrane trafficking of VSVG and E-cadherin ( Shin et al . , 2005; Zahn et al . , 2008; Nishimoto-Morita et al . , 2009 ) . Each of these cargo molecules contains a basolateral sorting motif in the C-terminal cytosolic domain . Sequence alignment of protein tyrosine kinase 7 ( PTK7 ) , another plasma-membrane localized regulator of planar cell polarity ( Lu et al . , 2004 ) , revealed a conserved tyrosine sorting motif ( YVDL ) in its predicted cytosolic domain ( Figure 8A ) . We used a C-terminal Myc-His-tagged PTK7 ( PTK7-Myc-His ) to examine the effect of Arfrp1 depletion on traffic from the TGN . COS7 cells were transfected with control siRNA or siRNA against Arfrp1 and re-transfected after 48 hr with plasmids encoding PTK7-Myc-His . These conditions achieved an siRNA-specific depletion of Arfrp1 ( Figure 8H ) . At steady state , around 50% of cells showed both surface- and Golgi-localized PTK7 in control cells and this localization pattern was not significantly changed in Arfrp1 knockdown cells . Given the high background of PTK7 delayed in the TGN in transfected COS7 cells , we adjusted the experimental conditions using a 20°C incubation followed by cycloheximide to synchronize a pool of newly-synthesized PTK7 in the TGN in control cells and Arfrp1 knockdown cells . After incubation at 20°C , a majority of cells ( 80% ) showed strong accumulation of PTK7 at the TGN . After cells were returned to 32°C , a significantly higher percentage accumulated PTK7 at the TGN when Arfrp1 was depleted than in cells treated with control siRNA ( Figure 8B–G and Figure 8I , 12 ± 8% vs 49 ± 6% ) . In contrast , the TGN localization of HA-Frizzled 6 was not enhanced by depletion of Arfrp1 ( Figure 8J ) . These results suggest that Arfrp1 also regulates TGN export of PTK7 . 10 . 7554/eLife . 00160 . 015Figure 8 . Arfrp1 regulates TGN export of PTK7 . ( A ) Sequence alignment of PTK7 from different species reveals a conserved tyrosine sorting motif in its predicted C-terminal cytosolic domain . ( B ) – ( G ) COS7 cells were transfected with control siRNA or siRNA against Arfrp1 and re-transfected after 48 hr with plasmids encoding PTK7-Myc-His . After an additional 24 hr , cells were incubated at 20°C in the presence of 30 μg/ml cyclohexmide for 4 hr then shifted to 32°C for 90 min . After incubation , cells were analyzed by immunofluorescence using antibodies against His and TGN46 . Size bar = 10 μm . ( H ) COS7 cell lysates from cells transfected with control siRNA or siRNA against Arfrp1 were analyzed by immunoblotting with anti-Arfrp1 antibody and , as a loading control , anti-GM130 antibody . ( I ) The fraction of cells showing TGN-accumulated PTK7 was quantified after incubation at 32°C ( mean ± SD; N = 3; over 150 cells were counted for each group ) . ( J ) Similar siRNA knockdown and temperature shift experiments were performed in COS7 cells transfected with HA-Frizzled 6 . The appearance of TGN-accumulated HA-Frizzled 6 was quantified in cells treated with control siRNA or siRNA against Arfrp1 after an incubation at 32°C ( mean ± SD; N = 2; over 100 cells were counted for each group ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00160 . 015 Protein kinase D ( PKD ) mediates membrane fission to generate TGN to cell surface transport carriers containing basolateral cargo molecules ( Yeaman et al . , 2004; Malhotra and Campelo , 2011 ) . Expression of the kinase dead form of glutathione S-transferase tagged PKD2 ( GST-PKD2-KD ) or PKD3 ( GST-PKD3-KD ) in COS7 cells resulted in the accumulation of HA-Vangl2 and HA-Frizzled 6 at the juxtanuclear area , colocalized with the TGN marker , TGN46 ( Figure 9 ) . Thus , although Vangl2 and Frizzled 6 display distinct requirements for Arfrp1 and AP-1 , they both depend on PKD for traffic from the TGN . We suggest that Vangl2 ( and PTK7 ) and Frizzled 6 are sorted by independent means into separate transport vesicles but that they share a common mechanism for membrane fission to form these carriers . 10 . 7554/eLife . 00160 . 016Figure 9 . TGN export of Vangl2 and Frizzled 6 is protein kinase D dependent . COS7 cells were co-transfected with GST-PKD2-KD and HA-Frizzled 6 ( A ) – ( D ) , GST-PKD3-KD and HA-Frizzled 6 ( E ) – ( H ) , GST-PKD2-KD and HA-Vangl2 ( I ) – ( L ) or GST-PKD3-KD and HA-Vangl2 ( M ) – ( P ) . Day 1 after transfection , cells were analyzed by immunofluorescence using anti-HA , anti-TGN46 and anti-GST antibodies . Size bar = 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 00160 . 016 The TGN is an essential sorting station where newly-synthesized cargo proteins and lipids are packaged for transport to various destinations at the cell surface , extracellular matrix and the endosome system . The variety of cargo molecules and the need to reach diverse destinations has complicated the assignment of a general sorting mechanism from the TGN . At least some cargo traffic from the TGN depends on vesicle coat proteins that bind distinct sorting peptide motifs on the cytosolic domain of membrane cargo proteins ( De Matteis and Luini , 2008; Barfield et al . , 2009 ) . Here we show that TGN export of Vangl2 , a PCP signaling receptor , depends on an unexpected complex of a TGN-localized Arf GTP-binding protein , Arfrp1 , and the Golgi-localized clathrin adaptor complex , AP-1 . siRNA knockdown of Arfrp1 or of subunits of AP-1 arrest Vangl2 traffic at the TGN as determined by co-localization of Vangl2 and the TGN marker , Golgin 97 . Further , we have identified a sorting signal within the C-terminal cytoplasmic domain of Vangl2 , YYXXF , the Phe residue of which is crucial for Vangl2 binding to AP-1 and traffic from the TGN to the cell surface . We propose a model wherein the interaction of Arfrp1-GTP and AP-1 exposes a sorting recognition determinant of the AP-1 µ subunit that binds the sorting motif on Vangl2 ( Figure 10A ) , and this binding in turn helps to stabilize AP-1 assembly on membranes . 10 . 7554/eLife . 00160 . 017Figure 10 . Proposed model . ( A ) Model depicting Arfrp1- and AP-1-mediated TGN sorting of Vangl2 . Arfrp1 is recruited to TGN membranes upon GTP binding , possibly mediated by a TGN localized GEF . Subsequently , GTP-bound Arfrp1 recruits AP-1 to TGN membranes . GTP-bound Arfrp1 also promotes an open conformation of AP-1 that directly interacts with the tyrosine motif on Vangl2 cytosolic domain , thereby enriching Vangl2 in budding vesicles . Binding of Vangl2 cytosolic domain to AP-1 , in turn , stabilizes the membrane association of AP-1 to allow sufficient time for AP-1 polymerization ( possibly with clathrin as a coat outer layer ) and vesicle budding . This model is consistent with reports showing that tyrosine sorting motifs promote membrane recruitment of AP-1 mediated by Arf1 ( Crottet et al . , 2002; Lee et al . , 2008 ) . ( B ) The asymmetrically localized PCP signaling molecules , including Vangl2 and Frizzled 6 , are sorted by different sorting machineries for export from the TGN . Differential TGN sorting and polarized trafficking of these signaling receptors may contribute to their asymmetric distribution and the laterally polarized organization of epithelial cells . DOI: http://dx . doi . org/10 . 7554/eLife . 00160 . 017 Mammalian cells possess two-dozen different Arf and Arf-like ( Arl ) proteins , only a few of which have been implicated in protein sorting or vesicle traffic ( Gillingham and Munro , 2007 ) . For example , Sar1 initiates COPII coat assembly at the ER ( Zanetti et al . , 2012 ) ; Arf1 is required for COPI-mediated vesicle budding and for the recruitment of clathrin AP-1 to the TGN and endosomes ( Gillingham and Munro , 2007 ) ; Arl6 binds the BBsome to segregate cell surface proteins into the membrane of the primary cilium ( Jin et al . , 2010 ) . Several of the Arfs and Arls are localized to the TGN of mammalian cells , and at least one , Arfrp1 , is required for TGN to cell surface traffic of E-cadherin and VSV-G ( Shin et al . , 2005; Zahn et al . , 2008; Nishimoto-Morita et al . , 2009 ) . Here we show that knockdown of Arfrp1 arrests the traffic of Vangl2 in a compartment that colocalizes with TGN markers but not with endosomal markers , confirming that Arfrp1 regulates trafficking from the late Golgi cisternae . Arfrp1 is essential at an early stage in mouse embryonic development ( Mueller et al . , 2002 ) , possibly because it plays a role in the traffic of crucial cell surface proteins . However , unlike Sar1 , which is required for traffic of all secretory cargo from the ER , Arfrp1 is not generally required for the transport of plasma membrane proteins from the TGN . For example , we show in this report that Frizzled 6 , another PCP signaling receptor that is displayed on the distal cell surface opposite to Vangl2 on the proximal surface of epithelial cells , does not depend on Arfrp1 for its transit from the TGN . Arfrp1 is proposed to regulate the membrane recruitment of Arl1 which in turn recruits GRIP domain-containing proteins to the TGN membrane ( Panic et al . , 2003; Zahn et al . , 2006 ) . However , at least one group reported that knockdown of Arfrp1 does not affect the localization of Arl1 and GRIP-domain containing proteins in mammalian cells ( Nishimoto-Morita et al . , 2009 ) . Moreover , Arfrp1 and Arl1 appear to play different roles in trafficking between TGN and endosomes ( Nishimoto-Morita et al . , 2009 ) . Our analysis indicates that knockdown of Arl1 does not affect the localization of Vangl2 suggesting that Arl1 and its associated GRIP-domain containing proteins are not involved in TGN sorting of Vangl2 . Using immobilized GDP- and GTP-mutant forms of Arfrp1 , we observed a GTP-selective interaction with the AP-1 complex in a crude cytosol fraction . Neither AP-2 nor AP-3 was detected in the proteins that bound to Arfrp1-GTP . We found that the µ and γ subunits of the AP-1 complex interact with Arfrp1 and that µ and γ subunits are required for the transit of Vangl2 from the TGN . Further , we observed that the residues of the YYXXF sorting motif required for traffic of Vangl2 are also crucial for the interaction of μ1-adaptin with Vangl2 . The YYXXF sorting motif fits the consensus sequence of the canonical YXXΦ motif which has been identified in plasma membrane proteins that traffic to the basolateral surface in polarized cells . PTK7 , another regulator of planar cell polarity also contains a conserved YXXΦ motif in its predicted C-terminal cytosolic domain and TGN export of PTK7 is also regulated by Arfrp1 . Mutation in the canonical YXXΦ motif causes mis-sorting of basolateral proteins to the apical domains ( Muth and Caplan , 2003 ) . Here we show that alanine substitution of both of the tyrosine residues or alanine substitution of the phenylalanine residue completely blocks TGN export of Vangl2 to a greater extent than when Arfrp1 is depleted . The surface-localized Vangl2 may be retained during the course of the Arfrp1 knockdown whereas YXXΦ mutant Vangl2 may not reach the cell surface during the course of the transfection . Alternatively , a partially redundant Arf or Arl may replace Arfrp1 to mediate inefficient traffic of Vangl2 from the TGN . Five different AP adaptor complexes have been identified in mammals , each serving a distinct role in traffic at the TGN , endosomes and cell surface ( Bonifacino and Lippincott-Schwartz , 2003; Hirst et al . , 2011 ) . The μ subunit of each adaptor complex preferentially binds distinct but overlapping sets of YXXΦ motifs based on the identity of X and Φ residues and the residues surrounding the tyrosine sorting motif ( Ohno et al . , 1998 ) . AP-1 regulates trafficking of mannose-6-P receptor from the TGN to endosomes ( Bonifacino and Lippincott-Schwartz , 2003 ) and mediates TGN export of potassium channels ( Ma et al . , 2011 ) . Given its central role in membrane traffic , deletion of various subunits of AP-1 leads to an embryonic lethal phenotype in the mouse ( Ohno , 2006 ) . In epithelial cells , some biosynthetic proteins traverse recycling endosomes en route to the basolateral membrane ( Fölsch et al . , 2009 ) . Correspondingly , epithelial cells possess two isoforms of AP-1 , a Golgi-localized AP-1A and a recycling endosome-localized AP-1B . AP-1A is proposed to mediate TGN export , thus this isoform may participate in Vangl2 traffic . We have not explored the post-Golgi pathway Vangl2 takes en route to the cell surface , thus it remains possible that Vangl2 invokes a recycling endosome in its itinerary to the proximal surface of an epithelial cell . Membrane recruitment of AP-1 is proposed to require Arfs and PI4P ( Wang et al . , 2003 ) . Another adaptor-like protein , GGA , has been shown to mediate membrane recruitment of PI4-kinase , which may then create a binding site for AP-1 ( Daboussi et al . , 2012 ) . We find that the sorting motif in the C-terminal domain of Vangl2 enhances AP-1 binding to Arfrp1-GTPγS on the surface of synthetic liposomes . Similarly , Arf1-GTP binding to AP-1 is promoted by a peptide containing the sorting signal on the cation-independent mannose 6-phosphate receptor ( Lee et al . , 2008 ) and recruitment of AP-1 to synthetic liposomes requires tyrosine sorting motifs ( Crottet et al . , 2002 ) . Structural analysis has suggested that adaptor complexes have open and closed cargo binding sites whose transition is implicated to be influenced by Arf-GTP binding ( Figure 10A ) ( Jackson et al . , 2010; Yu et al . , 2012 ) . Binding of a cargo-sorting motif to the open state may then stabilize coat assembly on membranes in preparation for transport vesicle budding . Our results build on this model of cargo capture to suggest that coat-adaptors may have more than two active conformations influenced by different Arf proteins . In the case of Vangl2 , we propose that Arfrp1 and the Vangl2 sorting motif favor an open conformation-exposed µ subunit of AP-1 that is not available in the complex of Arf1 and AP-1 . This combination may be responsible for the capture of cargo proteins destined for transport to the proximal cell surface domain in polarized epithelial cells . In a distinct example , a YKFFE sequence recognized by AP-4 directs the traffic of amyloid precursor protein ( APP ) from the TGN to early endosomes ( Burgos et al . , 2010 ) . This motif binds to a novel site on the surface of the μ4 subunit opposite the canonical tyrosine-signal-binding site . Key residues in this novel binding site are conserved in the μ subunits of other AP complexes ( Burgos et al . , 2010 ) . The YYXXF motif on Vangl2 could occupy an alternative site on μ1 , and Arfrp1-AP-1 , but not Arf1-AP-1 , may promote the exposure of this site . Currently , there is no direct structural evidence for this possibility . In contrast to these examples , Frizzled , which appears to be transported independent of Arfrp1 and AP-1 , may rely on another Arf and adaptor protein for traffic to the distal cell surface domain ( Figure 10B ) . Small interference siRNAs against Arfrp1 or against subunits of different adaptor complexes were purchased from Qiagen ( Valencia , CA ) , Thermo Scientific ( Rockford , IL ) or Ambion ( Grand Island , NY ) . The target sequence against Arfrp1 was CACCACCACCGTGGGCCTAAA . The target sequence against μ1-adaptin was AAGGCATCAAGTATCGGAAGA . The target sequence against γ1-adaptin was TAGCACAGGTTGCCACTAA . The target sequence against δ3-adaptin was CGCTGAAAATTCCTATGTT . The target sequence against μ3-adaptin was CCAAGGTACTAACATGGGA . Antibodies and dilutions for immunoblotting were: mouse anti-μ1a ( Abnova , Taipei , Taiwan , 1:2000 ) , mouse anti-arfrp1 ( Abnova , 1:500 ) , mouse anti-γ1 ( BD Transduction Laboratory , San Jose , CA , 1:2000 ) , rabbit anti-μ3 ( Proteintech Group , Chicago , IL , 1:2000 ) , mouse anti-δ3 ( Rockland , Gilbertsville , PA , 1:2000 ) , mouse anti-Golgin 97 ( Invitrogen , Grand Island , NY , 1:500 for immunofluorescence ( IF ) ) , rabbit anti-MBP ( New England Biolabs , Ipswich , MA , 1:4000 ) , mouse anti-GM130 ( BD Biosciences , San Jose , CA , 1:500 for IF ) , rabbit anti-HA ( Cell Signaling , Danvers , MA , 1:200 for IF , 1:2000 for immunoblotting ) , mouse anti-GFP ( Santa Cruz Biotechnology , Santa Cruz , CA , 1:2000 ) , mouse anti-Myc ( Cell Signaling , Danvers , MA , 1:2000 ) , mouse anti-EEA1 ( BD Biosciences , San Jose , CA , 1:2000 ) , mouse anti-tubulin ( Abcam , Cambridge , MA , 1:2500 ) , rabbit anti-CRMP2 ( antibody-online , Atlanta , GA , 1:3000 ) , mouse anti-His ( Qiagen , CA , 1:200 for IF and 1:2000 for immunoblotting ) , sheep anti-TGN46 ( AbD Serotec , UK , 1:200 ) , rabbit anti-Rab11 ( Invitrogen , Grand Island , NY , 1:200 ) , goat anti-Rab7 ( Santa Cruz Biotechnology , CA , 1:200 ) and goat-anti-dynamin II ( Affinity Bioreagent , Golden , CO , 1:2000 ) . HeLa cells , HeLa cells stably expressing HA-Vangl2 and COS7 cells were maintained in GIBCO Dulbecco's Modified Eagle Medium containing 10% Fetal Bovine Serum ( FBS ) , 10 mU/mL of penicillin and 0 . 1 mg/mL of streptomycin . Transfection of siRNA or DNA constructs into HeLa cells or COS7 cells was performed using lipofectamine 2000 as described in the manual provided by Invitrogen . For immunofluorescence , cells growing on coverslips were fixed in 4% PFA for 20 min then washed five times with 500 μl of PBS and incubated with permeabilization buffer ( PBS containing 0 . 1% TX-100 , 0 . 2 M Glycine , 2 . 5% FBS ) at RT for 30 min . Then cells were incubated with primary antibody and secondary antibody in permeabilization buffer for 30 min . Each antibody incubation was following by five times wash with PBS . Images were acquired with a Zeiss LSM 510 confocal microscope system or a Zeiss Axioobserver Z1 microscope system . Image J ( http://rsb . info . nih . gov/ij/ ) was used for colocalization analysis ( Guo et al . , 2008 ) . Briefly , the two images were adjusted to be the same average intensity of pixel value using divide function . A threshold was chosen manually to select the area stained with a Golgi marker . Subsequently , the numbers of above threshold pixels were determined for each Golgi marker ( A and B ) . Colocalized pixels were determined using the colocalization function with a fixed ratio of 0 . 75 ( C ) . Finally the value of colocalization was determined by the average value of C/A and C/B . COS7 cells were transfected with control siRNA or siRNA against Arfrp1 and re-transfected after 48 hr with plasmids encoding PTK7-Myc-His or HA-Frizzled 6 . After an additional 24 hr , cells were incubated in opti-MEM ( Invitrogen , Grand Island , NY ) containing 10% FBS and 30 μg/ml cycloheximide at 20°C for 4 hr to accumulate cargo proteins at the TGN . Cells were then shifted to 32°C for 90 min to restore transport from the TGN and analyzed by immunofluorescence ( Wakana et al . , 2012 ) . Glutathione transferase ( GST ) fusion protein purification was performed as described previously ( Guo et al . , 2008 ) . Briefly , full length constructs for μ1 , Arfrp1 wild type , and T31N and Q79L mutants were cloned in a pGEX-2T vector ( GE Healthcare Biosciences , NJ ) . The constructs were transformed in BL21 cells and individual colonies were grown to O . D . 0 . 6 in 500 ml of Luria broth ( LB ) at 37°C . Protein expression was induced with 0 . 5 mM isopropyl-1-thio-β-d-galactopyranoside ( IPTG ) for 5 hr at 25°C . Cells were centrifuged , washed with PBS and lysed in lysis buffer ( 50 mM Tris , pH 8 . 0 , 5 mM EDTA , 150 mM NaCl , 10% glycerol , 5 mM dithiothreitol , 0 . 5 mg/ml lysosome , proteinase inhibitor cocktail , complete , EDTA free , one tablet for 50 ml solution , Roche , Mannheim , Germany ) . After 30 min on ice , the cell lysates were adjusted to contain 0 . 5% TX-100 and sonicated four times for 30 s each time and centrifuged at 55k for 30 min in a Beckman TLA 100 . 3 rotor for the ultracentrifuge . The supernatant fraction was incubated with 250 μl glutathione-agarose beads for 2 hr at 4°C . After incubation , the beads were washed four times with PBS containing 1 mM DTT and 0 . 1% Tween 20 then two times with PBS . The beads were either used for a binding assay or mixed with elution buffer ( 50 mM Tris , pH 8 . 0 , 250 mM KCl , 1 mM DTT , 25 mM glutathione , pH 8 . 0 , proteinase inhibitor cocktail ) . MBP and His fusion protein purification were performed according to the protocol provided by Qiagen ( Valencia , CA ) or New England Biolabs ( Ipswich , MA ) respectively . The AP-1 complex was purified as described previously ( Lee et al . , 2008 ) . Cyanogen bromide-activated Sepharose-4B beads ( 6 mg , GE Healthcare Biosciences , NJ ) were incubated with 1ml 1 mM HCl on ice for 15 min , then the beads were washed four times with 1 ml coupling buffer ( 0 . 1 M NaHCO3 pH 8 . 3 , 0 . 5 M NaCl ) and incubated with 30 μg mouse antibody against γ1-adaptin ( 100/3 , Abcam , MA ) in 750 μl coupling buffer at 4°C overnight . After incubation , the beads were washed five times with 1 ml coupling buffer and then transferred to 0 . 1 M Tris–HCl buffer , pH 8 . 0 , and incubated on ice for 2 hr followed by washing with at least three cycles of buffer at alternative pHs ( coupling buffer followed by 0 . 1 M acetic acid/sodium acetate , pH 4 . 0 , 0 . 5 M NaCl ) . Beads were then incubated with 3 ml 8 mg/ml bovine brain cytosol prepared as described by Christoforidis and Zerial ( 2000 ) in HKM buffer ( 20 mM Hepes , pH 7 . 4 , 100 mM KCl , 5 mM MgCl2 ) at 4°C overnight . After incubation , the beads were washed four times with 1 ml HKM buffer and then eluted with 150 μl HKM buffer containing 0 . 3 mg/ml peptide corresponding to the hinge region of γ1-adaptin at 4°C for 5 hr . The eluted fraction was dialyzed against HKM buffer . A modified protocol from Jin et al . was performed to detect proteins that bind specifically to the GTP-bound Arfrp1 ( Jin et al . , 2010 ) . Briefly , GST fused to dominant negative or dominant active forms of Arfrp1 were purified from bacteria in a lysis buffer containing 5mM EDTA to extract Mg2+ and nucleotides . Proteins ( 50 μg ) on glutathione beads was incubated with nucleotide loading buffer ( 20 mM Hepes , pH 7 . 4 , 100 mM KCl , 5 mM MgCl2 , 500 μM GDP or GTPγS , proteinase inhibitor cocktail ) at room temperature for 1 hr . After incubation , the beads were mixed with bovine brain cytosol in binding buffer ( 20 mM Hepes , pH 7 . 4 , 100 mM KCl , 5 mM MgCl2 , 100 mM GDP or GTPγS , 0 . 1% TX100 , proteinase inhibitor cocktail ) at 4°C overnight . Beads were then mixed with washing buffer ( 20 mM Hepes , pH 7 . 4 , 500 mM KCl , 5 mM MgCl2 , 100 mM GDP or GTPγS , 0 . 1% TX100 , proteinase inhibitor cocktail ) and then with washing buffer without nucleotide and MgCl2 . Bound proteins were desorbed with elution buffer ( 20 mM Hepes , 500 mM KCl , 1 mM reversed GDP or GTPγS , 0 . 1% TX100 , proteinase inhibitor cocktail , 25 mM EDTA ) . Eluted fractions were concentrated in Amicon ultracentrifuge filters , and samples were electrophoresed on a 4–20% gradient gel which was stained with a silver staining kit ( Silver Quest , Invitrogen ) . Aliquots of the eluted fraction were also processed for immunoblot analysis . Binding assays to detect interactions between μ1-adaptin and various Vangl2 constructs were carried out with 4 μl of compact glutathione beads bearing 1 μg of GST-μ1 . The beads were incubated with 0 . 5 μg purified MBP-Vangl2 cytosolic domain wild type or mutant constructs in 400 μl binding buffer ( 100 mM KCl , 20 mM Hepes , pH 7 . 4 , 5 mM MgCl2 , 0 . 5% TX-100 ) containing 0 . 1 mg/ml BSA , or incubated with 150 μl 0 . 2mg/ml cell lysates from COS7 cells transiently transfected with HA-Vangl2 wild type or mutant constructs , in binding buffer at 4°C for 90 min . After incubation , the beads were washed with four times with 500 μl binding buffer and the bound material was analyzed by immunoblot . The yeast two-hybrid assay was carried out as described previously ( Ohno et al . , 1998 ) . The yeast strain ( PJ69-4A ) was cotransformed with mouse μ1A construct in pACT2 and Vangl2 cytosolic domain wild type or mutant constructs in pGBT9 . Colonies coexpressing both constructs were selected by their ability to grow on selective medium ( dropout without tryptophan and leucine ) . After selection for 3 days , individual colonies were inoculated in selective medium at 30°C overnight . The colonies were then resuspended in water and the cell concentration was adjusted to OD600 = 1 . 0 and serial dilutions were generated . Equal amount of cells for each serial dilution were spotted on selective medium and pictures were taken after 3 days of growth on the selective medium . Lipids and cholesterol , except Texas red PE , were purchased from Avanti ( Alabaster , Alabama ) . Texas red PE was purchased from Invitrogen ( Grand Island , NY ) . Lipids and cholesterol were mixed in chloroform in the following molar ratio ( Bacia et al . , 2011 ) : 51% 1 , 2 , dioleoyl-sn-glycero-3-phosphocholine ( DOPC ) , 22% 1 , 2 , dioleoyl-sn-glycero-3-phosphoethanolamine ( DOPE ) , 8% 1 , 2 , dioleoyl-sn-glycero-3-[phospho-l-serine]sodium salt ( DOPS ) , 5% 1 , 2 , dioleoyl-sn-glycero-3-phosphate ( monosodium salt ) ( DOPA ) , 8% l-α-phosphatidylinositol ( PI ) , 2 . 2% l-α-phosphatidylinositol-4-phosphate ( PI4P ) , 0 . 8% l-α-phosphatidylinositol-4 , 5-bisphosphate ( PI ( 4 , 5 ) P2 ) , 2% 1 , 2 , dipalmitoyl-sn-glycero-3- ( cytidine diphosphate ) ( CDP-DAG ) , 1% Texas red 1 , 2 , dihexadecanoyl-sn-glycero-3-phosphoetanolamine ( TX-PE ) and cholesterol ( 20% by weight ) . Chloroform was evaporated in a vacuum with an argon flow and rotation in a 37°C water bath . Liposomes were generated by rotating the dried lipid film in HKM buffer ( 20 mM Hepes , pH 7 . 4 , 100 mM KCl , 5 mM MgCl2 ) in a 37°C water bath for 2 hr . Liposomes were extruded to achieve homogeneity in size using the Mini-Extruder ( Avanti Polar Lipids , Inc . ) and Nuclepore track-etched membranes with 400-nm pores ( Whatman , Sanford , ME ) . Samples containing 1 . 5 μg of Arfrp1-His , Arfrp1-FLAG or Arf1-FLAG in the presence or absence of 1 μg MBP-Vangl2 cytosolic domain wild type or tyrosine mutant were incubated with 8 μl of 1 . 8 mg/ml liposomes in HKM buffer containing 100 μM nucleotides at room temperature for 45 min in 50 μl of reaction mixture . After incubation , 2 μg purified AP-1 was added and incubated for an additional 1 hr at RT . The reaction mixture was adjusted to 1 . 75 M sucrose and overlayed with 100 μl 0 . 75 M sucrose and 30 μl HKM buffer . The samples were centrifuged at 55 , 000 rpm in a TLS55 rotor in the Beckman-ultracentrifuge for 2 . 5 hr at 4°C . Fractions were collected from the bottom of the tube using a peristaltic pump ( RAININ , Columbus , OH ) and aliquots were analyzed by SDS-PAGE and immunoblot . Proteins were visualized and quantified using a Bio-Red GelDot imaging system . Flotation of liposomes after centrifugation was monitored by following Texas Red-PE fluorescence .

Most cells in multicellular organisms possess a property known as polarity that is reflected , in part , in the organization of the cell surface into distinct domains . One well-known axis in epithelial cells , such as those in the skin , divides the cell into an apical domain , which faces out , and a basal domain , which faces the underlying tissue . These cells rely on the distribution of structural components inside the cell , or within the cell membrane , to tell the difference between these two directions . Epithelial cells also possess a second type of polarity , planar cell polarity , that ensures that cells adjacent to each other in the plane parallel to the skin tissue are oriented correctly with respect to each other during development . This ensures , in turn , that hairs , scales , feathers and so on are all aligned . All eukaryotic cells sort and process proteins within an organelle called the Golgi apparatus , and proteins that are required at a specific destination within the cell , such as the cell surface membrane , carry specific molecular sorting signals that act as address labels to convey the protein into and within the secretory pathway . As one of these proteins moves through the Golgi apparatus , its sorting signals are recognized by coat proteins , such as clathrin , that subsequently form a vesicle around it . The assembly of this vesicle is initiated by an enzyme from the Arf family , but the enzyme must first undergo a conformational change ( by exchanging a molecule of GDP for one of GTP ) before formation can begin . The resulting vesicle can then be sent on its way to the address indicated by its Golgi-to-cell-surface sorting signal . These sorting signals also help to establish planar cell polarity in cells by ensuring that proteins called signaling receptors are distributed asymmetrically within the cell membrane . Guo et al . have now examined the mechanism behind the asymmetric sorting of two proteins that are involved in planar cell polarity: Vangl2 and Frizzled 6 . In an effort to understand why these proteins are localized to opposite surfaces of epithelial cells , Guo et al . used genetic techniques to reduce the expression of Golgi-localized Arf proteins in epithelial cell cultures . They found that knockdown of a protein called Arfrp1 caused Vangl2 to accumulate in the last station of the Golgi complex instead of being transported to the cell surface membrane . Then , using a technique called affinity chromatography , they demonstrated that a coat protein called the clathrin adaptor complex ( AP-1 ) had to be present for the formation of vesicles around Vangl2 . Moreover , disrupting AP-1 and Arfrp1 did not prevent Frizzled 6 being transported to the cell surface membrane . This suggests that cells use several distinct adaptor proteins and coat complexes to ensure that proteins from the Golgi apparatus go to specific locations on the cell surface and , thus , help to establish planar cell polarity .

2013

A novel GTP-binding protein–adaptor protein complex responsible for export of Vangl2 from the trans Golgi network

Genome-encoded microRNAs ( miRNAs ) provide a posttranscriptional regulatory layer that controls the differentiation and function of various cellular systems , including hematopoietic cells . miR-142 is one of the most prevalently expressed miRNAs within the hematopoietic lineage . To address the in vivo functions of miR-142 , we utilized a novel reporter and a loss-of-function mouse allele that we have recently generated . In this study , we show that miR-142 is broadly expressed in the adult hematopoietic system . Our data further reveal that miR-142 is critical for megakaryopoiesis . Genetic ablation of miR-142 caused impaired megakaryocyte maturation , inhibition of polyploidization , abnormal proplatelet formation , and thrombocytopenia . Finally , we characterized a network of miR-142-3p targets which collectively control actin filament homeostasis , thereby ensuring proper execution of actin-dependent proplatelet formation . Our study reveals a pivotal role for miR-142 activity in megakaryocyte maturation and function , and demonstrates a critical contribution of a single miRNA in orchestrating cytoskeletal dynamics and normal hemostasis . microRNAs ( miRNAs ) are single-stranded RNA molecules of 22 nucleotides in length , processed from endogenous hairpin transcripts . miRNAs provide cells with a sequence-based silencing mechanism through base-pairing of a minimal recognition sequence , called the miRNA ‘seed’ ( Bartel , 2009; Carthew and Sontheimer , 2009; Fabian et al . , 2010 ) . miRNAs control hematopoiesis and the function of both lymphoid and myeloid progeny ( Chen and Lodish , 2005; Lawrie , 2007; Garzon and Croce , 2008; Navarro and Lieberman , 2010 ) . For example , in vivo studies uncovered roles for miR-451 in erythropoiesis ( Patrick et al . , 2010; Rasmussen et al . , 2010; Yu et al . , 2010 ) , for miR-223 in granulopoiesis ( Johnnidis et al . , 2008 ) , for miR-150 in the commitment of multipotent myeloerythroid progenitors ( Lu et al . , 2008 ) , and for miR-155 in the mammalian immune system ( Rodriguez et al . , 2007; Thai et al . , 2007; Mann et al . , 2010; O'Connell et al . , 2010 , 2007 ) . Megakaryocytes ( MKs ) display a distinctive miRNA expression pattern ( Garzon et al . , 2006; Opalinska et al . , 2010; Xu et al . , 2012 ) . However , functional genetic studies dissecting the role of miRNA in megakaryopoiesis are still limited . In the present work , we focused on miR-142 , a hematopoietic-specific miRNA , which resides in a genomic locus that was previously associated with t ( 8;17 ) translocation in B-cell leukemia ( Gauwerky et al . , 1989 ) . Pioneering experimental evidence has suggested miR-142 involvement in lymphocyte differentiation ( Chen et al . , 2004 ) and recently , miR-142 was also shown to play a role in the specification of definitive hemangioblasts ( Lu et al . , 2013; Nimmo et al . , 2013 ) , and in lymphoid and myeloid lineages ( Gauwerky et al . , 1989; Chen et al . , 2004; Huang et al . , 2009; Belz , 2013; Lagrange et al . , 2013; Lu et al . , 2013; Nimmo et al . , 2013; Sonda et al . , 2013; Sun et al . , 2013; Zhou et al . , 2013 ) . Furthermore , miR-142 is involved in the compound immune response to North American eastern equine encephalitis virus ( Trobaugh et al . , 2014 ) and our work uncovered a key role for miR-142 in the maintenance of CD4+ dendritic cells ( Mildner et al . , 2013 ) . MK maturation is an intricate process that involves DNA replication in the absence of cytoplasmic division , termed endomitosis . Polyploid MKs also exhibit distinctive cytoskeletal rearrangements that enable the biosynthesis of platelets ( also known as thrombocytes ) . The elaborated actin cytoskeleton of MKs is uniquely organized in a way that allows cytoplasmic proplatelet protrusions to bend and bifurcate into multiple tips , from which platelets are subsequently released to the bloodstream . Dysregulation of the actin cytoskeleton impairs the generation of mature platelets ( Hartwig and Italiano , 2006; Bender et al . , 2010 ) . Accordingly , actin-modulating genes were shown to be crucial for MK maturation and have been implicated in the etiology of platelet-related disorders ( Villa et al . , 1995; Kajiwara et al . , 1999 ) . We characterize miR-142-deficient mice that display an array of hematological defects , including pronounced thrombocytopenia . We show that miR-142 controls multiple facets of MK differentiation including control of cell size , ploidy , and proplatelet elaboration . Furthermore , we demonstrate that miR-142-3p controls platelet biosynthesis by orchestrating the coordinated expression of several distinct nodes in a network of actin cytoskeleton regulators . Our study reveals a novel miRNA-dependent circuit that maintains cytoskeletal integrity , and suggests that a single miRNA may broadly regulate cell function by controlling a coherent set of effectors in a given pathway . A miR-142−/− allele was created by insertion of an exogenous gene trap sequence ∼50 bp upstream of the murine pre-miR-142 . This cassette disrupted normal transcription and drove the expression of a beta-galactosidase reporter gene ( Stanford et al . , 2001; Hansen et al . , 2008; Osokine et al . , 2008; Figure 1A ) . 10 . 7554/eLife . 01964 . 003Figure 1 . Pronounced thrombocytopenia in miR-142−/− mice . ( A ) Left panel: schematic representation of the gene trap cassette targeting the murine miR-142 locus . The WT and mutant loci are shown with the gene trap-targeting vector . pre-miR-142 is shown as a red box . LTR , long terminal repeats; SA , splice acceptor; betaGeo , beta-galactosidase-Neomycin resistance fusion protein; pA , polyA signal . Right panel: Genomic PCR confirmation of miR-142 trap insertion . ( B ) Quantitative real-time ( q ) PCR performed on cDNA derived from peripheral blood mononuclear cells reveals nullification of miR-142-3p and miR-142-5p expression in miR-142−/− animals . Representative results from one of two independent experiments are shown ( mean + SEM ) with three animals in each group . ***p<0 . 0005 . ( C ) Beta-galactosidase activity in ex vivo hematopoietic cell populations isolated from miR-142+/+ ( red ) , and miR-142+/− ( blue ) mice as determined by fluorescence-activated cell sorting ( FACS ) of Fluorescein Di-beta-D-Galactopyranoside-treated cells . Assayed cell types include T-cells ( CD4+ and CD8+ ) and B-cells ( immature , marginal and mature ) derived from the spleen , granulocytes , monocytes ( Gr1+ and Gr1− ) , natural killer ( NK ) cells , and megakaryocytes ( MKs ) derived from the BM . ( D and E ) Significant decrease in circulating red blood cells ( RBC , panel D ) and white blood cells ( WBC , panel E ) in 2-month-old miR-142−/− mice . Representative results from one of two independent experiments are shown ( mean + SEM ) with at least five animals in each group . *p<0 . 05; **p<0 . 005 . ( F ) Significant decrease in circulating platelet numbers in 2- and 12-month-old miR-142−/− mice . Representative results from one of two independent experiments are shown ( mean + SEM ) with at least five animals in each group . **p<0 . 005; ***p<0 . 0005 . ( G ) Enlarged mean platelet volume ( MPV ) in 2- and 12-month-old miR-142−/− mice . Representative results from one of two independent experiments are shown ( mean + SEM ) with at least five animals in each group . **p<0 . 005; ***p<0 . 0005 . DOI: http://dx . doi . org/10 . 7554/eLife . 01964 . 00310 . 7554/eLife . 01964 . 004Figure 1—source data 1 . Mendelian distribution of miR-142 intercrosses . Genotypes of E14 . 5 and P21 offspring from miR-142+/− intercrosses reveals partial perinatal or juvenile lethality in miR-142−/− mice . The actual and expected number of mice for each genotype at the indicated stages is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 01964 . 00410 . 7554/eLife . 01964 . 005Figure 1—figure supplement 1 . miR-142 hematopoietic-intrinsic expression is required for thrombopoiesis . Schematic diagram of the experimental design for re-introduction of miR-142−/− BM into lethally irradiated WT mouse recipient ( left panel ) , which results in reduced platelet counts relative to controls , 6 weeks after transfer ( right panel ) . Data are gained from four animals per group ( mean + SEM ) *p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 01964 . 005 To confirm miR-142 nullification , we collected circulating mononuclear cells from peripheral blood of homozygous mutant mice and wild-type ( WT ) littermates . miR-142-3p is the guide strand from the miR-142 hairpin , whereas the sister ‘passenger’ miR-142-5p strand is expressed in negligible levels ( Nimmo et al . , 2013; Figure 1B ) . The expression of both miR-142-3p and miR-142-5p was abolished in miR-142−/− circulating blood cells as exemplified by quantitative real-time PCR ( qPCR , Figure 1B ) . Genotyping the progeny of miR-142+/− intercrosses at embryonic day 14 . 5 ( E14 . 5 ) revealed the predicted Mendelian distribution of miR-142+/+ , miR-142+/− , and miR-142−/− embryos . However , postnatal survival at 3 weeks of age of miR-142 homozygous offsprings was lower than expected ( 18% instead of 25% ) , demonstrating that roughly a third of miR-142−/− mice died perinatally ( Figure 1—source data 1 ) . Surviving miR-142−/− mice did not display overt physical abnormalities , were fertile , and bred normally . To comprehensively characterize miR-142 expression pattern , we performed a fluorescence-based semi-quantitative detection of beta-galactosidase activity in viable hematopoietic cells . In accordance with previous reports ( Lagos-Quintana et al . , 2002; Chen et al . , 2004; Ramkissoon et al . , 2006; Merkerova et al . , 2008 ) , this study revealed pan-hematopoietic activity of the miR-142 promoter , which drove the expression of a LacZ transgene from the endogenous miR-142 locus in all lymphoid and myeloid lineages examined ( Figure 1C ) . To assess the impact of miR-142 loss in vivo , we performed complete blood counts that revealed reduced numbers of erythrocytes and white blood cells in miR-142−/− animals relative to WT littermates ( Figure 1D , E ) . Interestingly , at 2 months of age , miR-142−/− mice displayed a striking ∼50% decrease in platelet counts and a ∼10% increase in mean platelet volume ( MPV ) , relative to controls ( Figure 1F , G ) . Thus , miR-142−/− animals suffer from macrothrombocytopenia . The reduction in platelet numbers and the concordant increase in platelet size were even more pronounced when analyzed at 1 year of age ( Figure 1F , G ) . In addition , lethally irradiated WT mice , which were reconstituted with miR-142−/− bone marrow ( BM ) cells and analyzed 6 weeks following transfer , showed platelet paucity relative to controls that were reconstituted with WT BM cells ( Figure 1—figure supplement 1 ) . However , the decrease in platelet numbers in this model was less significant compared to that demonstrated in germline miR-142-deficient mice ( Figure 1F ) , plausibly due to contribution of WT cells . Taken together , these data demonstrate that hematopoietic-specific miR-142 activity is required for normal platelet production . We hypothesized that the diminished numbers of circulating platelets might stem from a defect in the development of MKs . First , we confirmed that miR-142-3p is the functional dominant ‘guide’ strand in MKs , and that both sister miR-142 species are nullified in miR-142−/− MKs ( Figure 2—figure supplement 1A ) . We also confirmed that the expression of Bzrap1 , a regulator of synaptic transmission ( Chardenot et al . , 2002; Mittelstaedt and Schoch , 2007 ) positioned 3 . 5 Kb downstream of miR-142 , is unchanged in miR-142−/− MKs ( Figure 2—figure supplement 1B ) . To gain insight into the impact of miR-142 nullification on early MK development , we performed a high-resolution flow cytometry assay for the characterization of myeloerythroid progenitors ( Pronk et al . , 2007; Pronk and Bryder , 2011 ) . The numbers of miR-142−/− bipotent MK-erythroid precursors ( PreMegEs ) were marginally increased relative to controls ( Figure 2A , B ) . In contrast , the direct descendants of PreMegEs , namely the unipotent MK-progenitors ( MkPs ) , which further give rise to mature MKs , were significantly increased in miR-142−/− animals , relative to WT controls ( Figure 2A , B ) . Intriguingly , the expression levels of regulatory markers of MK development remained largely unchanged in miR-142−/− PreMegEs and MkPs , relative to WT controls ( Figure 2C , D ) . 10 . 7554/eLife . 01964 . 006Figure 2 . Perturbed myeloerythroid development in the absence of miR-142 . ( A ) Diagram of gating strategy used to define the myeloerythroid progenitor populations ( top panel ) , and representative FACS profiles of mutant miR-142−/− and WT BM cells ( bottom panels ) . ( B ) Flow cytometric analysis of whole tibia BM-resident mega-erythroid progenitors ( PreMegE; lin−c-kit+CD150+CD105−CD41− ) and MK progenitors ( MkP; lin−c-kit+CD41+ ) , of 2-month-old animals . miR-142 deficiency results in increased MkP numbers and only modest , insignificant , changes in PreMegEs . Representative results from one of two independent experiments ( mean + SEM ) with at least three animals per group . *p<0 . 05 . ( C and D ) qPCR expression analysis of critical regulators of MK development: GATA binding protein 1 ( Gata1 ) , GATA binding protein 2 ( Gata2 ) , zinc finger protein , multitype 1 ( Zfpm1 ) , Kruppel-like factor 1 ( Klf1 ) , Friend leukemia integration 1 ( Fli1 ) , spleen focus forming virus proviral integration oncogene ( Spi1 ) , Runt-related transcription factor 1 ( Runx1 ) , and T cell acute lymphocytic leukemia 1 ( Tal1 ) in miR-142−/− PreMegEs ( C ) and MkPs ( D ) , relative to controls . Data normalized to Hprt expression and to the mRNA expression in WT controls and presented as mean + SEM . *p<0 . 05; **p<0 . 005 . ( E ) Left panel , representative FACS profiles of mutant miR-142−/− and WT BM cells . Right panel , gating CD41+/CD42+ cells out of total BM , reveals increased mutant miR-142−/− MK numbers relative to WT controls . Representative results from one of two independent experiments ( mean + SEM ) , at least three animals in each group . **p<0 . 005 . ( F ) Increased numbers of von Willebrand factor ( vWF ) -positive MKs per high power field ( hpf ) in miR-142−/− BMs , relative to WT controls . Representative results from one of two independent experiments ( mean + SEM ) , four cross-sections measured from each group . ***p<0 . 0005 . ( G ) Left panel , CFU–MK assays demonstrate increased miR-142−/− MK numbers per colony . Scale bars , 50 μm . Right panel , increased numbers of MkPs in miR-142−/− BM , revealed by CFU–MK colony formation assay . Representative results from one of two independent experiments ( mean + SEM ) , two biological samples in each group . *p<0 . 05 . ( H ) Schematic representation of the experimental design for competitive repopulation assay . ( I ) Representative FACS profiles for chimeric animals in competitive repopulation assay: [miR-142−/− ( CD45 . 2 ) /WT ( CD45 . 1 ) > WT ( CD45 . 1 ) ] . Flow cytometry performed 6 weeks after transplantation . ( J ) Quantification of CD45 . 2/CD45 . 1 ratios , calculated for each gate in three different animals . Dashed line indicates ratio of 1 . Values >1 indicate that miR-142−/− ( CD45 . 2 ) mutant cells out-compete WT ( CD45 . 1 ) cells , whereas values <1 reveal the advantage of WT ( CD45 . 1 ) cells . The CD45 . 2/CD45 . 1 ratio for B220-positive cells served as control . Representative results from one of two independent experiments ( mean + SEM ) , three animals in each group . *p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 01964 . 00610 . 7554/eLife . 01964 . 007Figure 2—figure supplement 1 . Gene expression in miR-142−/− MKs . ( A and B ) qPCR expression analyses confirm nullification of miR-142-3p and miR-142-5p ( A ) and similar Bzrap1 expression ( B ) in miR-142−/− MKs , relative to WT controls . Representative results from one of two independent experiments ( mean + SEM ) , three animals in each group . ***p<0 . 0005; ( NS ) not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 01964 . 00710 . 7554/eLife . 01964 . 008Figure 2—figure supplement 2 . Splenomegaly and increased numbers of splenic MKs in miR-142−/− mice . ( A ) Representative spleen sections were stained with hematoxylin and eosin ( H&E , top panel ) and anti-von Willebrand Factor ( anti vWF; bottom panel ) . Arrows indicate MKs . Scale bars , 100 μm . ( B ) A significant increase in the number of splenic megakaryocytes per high power field ( hpf ) is observed in miR-142−/− mice . Data are obtained from four animals in each group ( mean + SEM ) . ***p<0 . 0005 . ( C ) Splenomegaly is detected in miR-142−/− mice . Data are obtained from four animals in each group ( mean + SEM ) . ***p<0 . 0005 . DOI: http://dx . doi . org/10 . 7554/eLife . 01964 . 008 The observed expansion of MkPs in miR-142 mutants was further supported by a concordant elevation in the numbers of CD41+/CD42+ miR-142−/− MKs relative to WT controls ( Figure 2E ) . We next investigated the numbers of MKs in situ , by immunohistochemical staining of femoral BM sections against von Willebrand Factor ( vWF ) , which is specifically expressed in MKs from an early stage of differentiation ( Tomer , 2004 ) . miR-142−/− mice displayed a ∼50% increase in the numbers of vWF-positive MKs , relative to WT littermates ( Figure 2F ) . Noteworthy , miR-142−/− mice exhibited splenomegaly and a marked elevation in the number of splenic MKs , relative to control littermates , suggesting extramedullary thrombopoiesis ( Figure 2—figure supplement 2 ) . To further confirm the expansion of miR-142−/− MkPs , we employed a colony forming unit-megakaryocyte ( CFU-MK ) assay that quantifies the MkP numbers in the BM . We observed higher numbers of CFU-MK colonies of miR-142−/− BMs , relative to WT controls ( Figure 2G , right panel ) . Furthermore , each miR-142−/− colony typically harbored more cells than control colonies ( Figure 2G , left panel ) . To elucidate whether MkP expansion represented a cell-intrinsic phenomenon , we employed a competitive repopulation experiment ( Figure 2H ) . Thus , we injected CD45 . 2+/miR-142−/− and congenic CD45 . 1+/WT BM cells in equal numbers into lethally irradiated CD45 . 1+ recipient mice . The reconstituted BM populations were analyzed for 6 weeks following transplantation . Cells expressing the pan-B-cell marker , B220 , were equally represented by CD45 . 2+/miR-142−/− and CD45 . 1+/WT genotypes and served as engraftment controls . CD45 . 2+/miR-142−/− PreMegE levels showed a mild increase relative to CD45 . 1+/WT counterparts , confirming that this population is not appreciably affected by the loss of miR-142 ( Figure 2I , J ) . In contrast , CD45 . 2+/miR-142−/− MkPs were over-represented in chimeric BMs at a ratio of ∼5:1 relative to CD45 . 1+/WT MkPs ( Figure 2I , J ) . Thus , the MkP expansion observed in miR-142−/− BM is cell-autonomous . Taken together , these data suggest that miR-142 activity regulates the differentiation of the MK lineage in a cell-intrinsic manner . The observed elevation in miR-142−/− MK frequency was unexpected , because increased MK numbers are usually correlated with higher platelet counts ( Schafer , 2004 ) . Thus , since the pronounced thrombocytopenia in miR-142-deficient mice was not caused due to a lack of MKs , we hypothesized that it may result from a block in MK maturation . An initial clue that miR-142−/− MKs were premature , came from the observation that the average size of vWF-positive miR-142−/− MKs in the femoral BM was smaller than that of WT MKs . Indeed , miR-142−/− MKs showed a ∼25% reduction in sectional area , relative to WT counterparts ( Figure 3A , B ) . 10 . 7554/eLife . 01964 . 009Figure 3 . Impaired maturation of miR-142−/− MKs . ( A ) Representative BM sections of miR-142−/− vs WT controls , stained with anti-von Willebrand Factor ( anti vWF ) . In the bottom right corner of each section there is an enlarged image of a representative vWF-positive MK . Scale bars , 50 μm . ( B ) Reduced diameter of vWF-positive MKs in miR-142−/− BM relative to WT controls . Representative results from one of two independent experiments ( mean + SEM ) . Data collected from four cross-sections measured and >100 cells per group . ***p<0 . 0005 . ( C ) Representative brightfield ( BF , top panel ) and May–Grünwald Giemsa-stained ( MGG , bottom panel ) micrographs of FL-derived MK cultures , following enrichment by a BSA density gradient . Scale bars , 100 μm . ( D ) Size quantification of FL-derived MK , measured as pixel area and normalized to WT controls , reveals reduction of miR-142−/− MK cell area . Representative results from one of two independent experiments ( mean + SEM ) , >20 cells measured per group . ***p<0 . 0005 . ( E ) Representative FACS plot of DNA content analysis for FL-derived MKs stained for CD41 and DAPI ( left ) and quantification of ploidy in FL-derived MKs , presented as a percentage of cells out of total CD41+ cells ( right ) . AU , arbitrary units . Representative results from one of two independent experiments ( mean + SEM ) ≥4 animals in each group . *p<0 . 05; **p<0 . 005; ***p<0 . 0005 . ( F ) Representative FACS plot of DNA content analysis for BM-derived MKs stained for CD41 and DAPI ( left ) and quantification of ploidy in BM-derived MKs , presented as percentage of cells out of total CD41+ cells ( right ) . AU , arbitrary units . Representative results from one of two independent experiments are shown ( mean + SEM ) ≥4 animals in each group . *p<0 . 05; **p<0 . 005 . ( G ) Left panel: representative micrographs of proplatelet formation ( PPF ) in FL-derived WT or miR-142−/− MKs ( white arrows denote MKs extending proplatelets ) . Scale bars , 50 μm . Right panel: quantification revealed reduced PPF levels in miR-142−/− FL-derived MKs . Representative results from one of two independent experiments a ( mean ± SEM ) , three animals in each group , each animal represented by 7–9 experimental repeats in distinct wells and each dot is a representation of a single well . ***p<0 . 0005 . ( H ) Re-introduction of miR-142-3p using dsRNA mimetics was sufficient to restore WT PPF levels to miR-142-deficient differentiated MKs . Overexpression of miR-142-3p mimic did not result in significant increase in PPF levels in WT MKs . Each dot represents data from a single well . **p<0 . 005; ( NS ) not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 01964 . 00910 . 7554/eLife . 01964 . 010Figure 3—figure supplement 1 . miR-142-3p expression in differentiated MKs following transfection of dsRNA mimetics . qPCR expression analyses confirm overexpression of miR-142-3p in WT and miR-142−/− FL-derived MKs transfected with miR-142-3p mimic , relative to MKs transfected with control mimics . **p<0 . 005 . DOI: http://dx . doi . org/10 . 7554/eLife . 01964 . 010 We then differentiated MKs from E14 . 5 fetal liver ( FL ) , under defined ex vivo conditions , as previously described ( Shivdasani and Schulze , 2005 ) . After 4 days in culture , FL-derived miR-142−/− MKs exhibited smaller cell size , compared to control MKs ( Figure 3C , D ) . Polyploidization is an additional important feature of MK maturity , which is associated with effective platelet production ( Levine et al . , 1982; Mattia et al . , 2002; Ravid et al . , 2002; Lee et al . , 2009 ) . We therefore tested the number of MK nuclei by flow cytometry . CD41+ BM-derived miR-142−/− MKs , exhibited reduced overall ploidy ( Figure 3E ) . Furthermore , the fraction of mature ( ≥16N ) miR-142−/− MKs was significantly diminished , relative to WT controls , whereas the percentage of low ploidy immature MK forms ( ≤8N ) was higher in miR-142−/− BM ( Figure 3E ) . Similar data were gained by ex vivo differentiation of FL-derived MKs , whereby high ploidy number ( >32 ) was observed in only 4% of the miR-142−/− MK , relative to 12% in control MKs ( Figure 3F ) . Thus , miR-142 is essential for normal endomitosis and reduced miR-142−/− platelet numbers might result from accumulation of immature , low-polyploid MKs that are poor producers of platelets ( Mattia et al . , 2002 ) . Proplatelet formation ( PPF ) represents the final phase of MK maturation , culminating in platelet release into the bloodstream ( Machlus and Italiano , 2013 ) . To analyze whether miR-142 is involved in this process , we performed an ex vivo PPF study on FL-derived MKs . Remarkably , we observed a striking threefold reduction in miR-142−/− MKs that were extending proplatelets , relative to control MKs ( Figure 3G ) . We next re-introduced miR-142-3p into differentiated MKs , using dsRNA mimetics ( Figure 3—figure supplement 1 ) . The introduction of miR-142-3p was sufficient to recapitulate WT PPF levels in cells that are genetically miR-142-deficient ( Figure 3H ) . Conversely , overexpression of miR-142-3p mimics in WT MKs did not yield any significant increase in PPF levels . Thus , miR-142-3p activity is essential for proper MK maturation and its loss results in defective platelet biogenesis . Furthermore , since re-introduction of miR-142-3p into miR-142−/− differentiated MKs restored functional identity , we conclude that there might be a continuous requirement for miR-142-3p activity to maintain MK maturity . Dynamic rearrangement and organization of MK cytoskeletal structures is essential for normal megakaryopoiesis , endomitosis , and platelet production ( Hartwig and Italiano , 2006; Lordier et al . , 2008; Thon and Italiano , 2012 ) . We therefore sought to characterize by confocal microscopy filamentous actin ( F-actin ) and tubulin in FL-derived MKs that were allowed to adhere to fibronectin for 3 hr . This analysis revealed a markedly disturbed cytoskeletal organization in miR-142−/− MKs compared to WT counterparts ( Figure 4A ) and mutant cells displayed a more immature circular profile than WT counterparts ( Figure 4B ) . 10 . 7554/eLife . 01964 . 011Figure 4 . Disturbed actin cytoskeletal architecture and dynamics in the absence of miR-142 . ( A ) Representative micrographs of WT and miR-142−/− FL-derived MKs , cultured for 5 days with TPO and subsequently plated on fibronectin-coated cover-slips for 3 hr . F-actin ( phalloidin , red ) , tubulin ( green ) . The merged panels also depict DAPI in blue . Scale bars , 20 μm . ( B ) Circularity of FL-derived MKs ( on an arbitrary scale of 0–1 ) , measured using ImageJ software . miR-142−/− MKs were more circular than WT controls , reflecting immaturity and relative deficiency of proplatelet-like structures . Representative results from one of two independent experiments ( mean + SEM ) , >100 cells measured per group ***p<0 . 0005 . ( C ) Representative flow cytometry-based single-cell images of FL-derived MKs as obtained by ImagestreamX flow cytometer , stained with anti - CD61 antibody ( yellow ) , FITC-Phalloidin ( F-actin , green ) , Alexa594-DNaseI ( G-actin , red ) and DAPI ( blue ) . Scale bar , 10 μm . ( D–G ) Reduced F-actin/G-actin ratio in FL-derived MKs ( D ) , reduced cell area ( E ) , increased circularity ( F ) , and increased F-actin polarity ( G ) , in miR-142−/− MKs relative to WT controls revealed by Imagestream analysis . Four animals per group ( mean + SEM ) . *p<0 . 05; ***p<0 . 0005 . ( H ) Representative micrographs of WT and miR-142−/− MKs stained with Phalloidin–Rhodamine for detection of actin stress fibers after cytochalasin D ( CytoD ) washout . Left panel depicts MKs stained following 30 min of CytoD treatment . Middle and right panels depict MKs stained 1 hr and 2 hr after CytoD washout , respectively . Scale bars , 50 μm . ( I ) The fraction of WT MKs exhibiting stress fibers at 1 hr and 2 hr after CytoD washout is larger relative to miR-142−/− MKs . Representative results from one of two independent experiments ( mean + SEM ) , >50 cells counted in each group . ***p<0 . 0005 . DOI: http://dx . doi . org/10 . 7554/eLife . 01964 . 011 Accordingly , ImagestreamX flow cytometery revealed a lower ratio of filamentous to globular ( F/G ) actin in miR-142−/− CD61+ MKs than in control MKs ( Figure 4C , D ) . This analysis reaffirmed that miR-142−/− MKs are smaller in size , more circular and exhibited a more homogenous F-actin distribution than controls ( Figure 4E–G ) . Finally , we performed a cytochalasin D ( CytoD ) washout study , which followed the re-assembly of actin filaments after forced depolymerization ( Figure 4H ) . The number of miR-142−/− MKs that were able to create stress-fibers following treatment with CytoD was significantly lower , relative to control MKs ( Figure 4I ) . Collectively , these data reveal that miR-142 is pivotal for normal actin dynamics and architecture in MKs . To determine the molecular mechanism for miR-142-mediated control of megakaryocytic development , we performed a genome-wide study for the identification of differentially expressed genes in miR-142−/− vs WT differentiated MKs . This study revealed that roughly 800 mRNAs were significantly deregulated due to loss of miR-142 ( Figure 5A ) . Sylamer analysis ( van Dongen et al . , 2008 ) of microarray data , from WT and miR-142−/− FL-derived MKs , uncovered a highly significant enrichment for the 7- and 8-mer seeds of miR-142-3p among genes that were up-regulated in miR-142−/− MKs ( Blue lines; Figure 5B ) . However , such enrichment was not evident for any other miRNA including the sister miRNA , miR-142-5p ( Green lines; Figure 5B ) , further substantiating that miR-142-3p is the dominant functional miRNA from the pre-miR-142 hairpin . Accordingly , we depicted significant up-regulation of miR-142-3p targets within the set of TargetScan predicted targets ( Friedman et al . , 2009; Figure 5—figure supplement 1 ) . 10 . 7554/eLife . 01964 . 012Figure 5 . miR-142 Regulates a group of cytoskeletal regulatory genes during megakaryopoiesis . ( A ) A log2-scale scatter plot , presenting the expression of mRNAs from FL-derived WT MKs ( x axis ) , and miR-142−/− MKs ( y axis ) . The inset table depicts the number of genes that are significantly up- or down-regulated ( > 2 fold change , p<0 . 05 ) . ( B ) Enrichment landscape plot for all 876 7mer motifs complementary to canonical mouse miRNA seed regions , gained by Sylamer analysis ( van Dongen et al . , 2008 ) . Sorted 17 , 000 gene list ordered from mostly up-regulated to mostly down-regulated in the miR-142−/− MKs on the x axis reveal the enrichment of only two motifs , which are both corresponding to the expected impact of miR-142-3p ‘seed’ on MK transcriptome ( blue , 7mer–m8; light blue , 7mer–A1 ) . miR-142-5p seed motifs are not enriched ( green , 7mer-m8; light green , 7mer-A1 ) . Horizontal dotted lines represent a Bonferroni-corrected p value threshold of 0 . 05 . ( C ) A schematic representation of bioinformatic pipeline . Genes that were commonly and significantly up-regulated in three expression arrays of miR-142−/− hematopoietic cells ( CD24+ in vitro-derived dendritic cells [DCs] , CD24− DCs and MKs ) , were superimposed with miR-142-3p TargetScan ( TS ) -predicted target genes . The resultant list was subjected to GO analysis using the DAVID bioinformatic tool . The majority of enriched GO categories were annotated to cytoskeletal and actin-binding genes . ( D ) qPCR expression analysis of a novel set of miR-142-3p target mRNAs: Cofilin-2 ( Cfl2 ) , Wiskott–Aldrich syndrome-like ( Wasl ) , Biorientation of chromosomes in cell division 1 ( Bod1 ) , Twinfilin-1 ( Twf1 ) , Integrin alpha V ( Itgav ) and Glucocorticoid receptor DNA binding factor 1 ( Grlf1 ) genes in miR-142−/− MKs , relative to controls , normalized to Hprt expression and to the mRNA expression in WT controls . Data are presented as mean + SEM . *p<0 . 05; **p<0 . 005; ***p<0 . 0005 . ( E ) Reintroducing miR-142-3p using dsRNA mimetics was sufficient to restore Wasl ( left panel ) and Cfl2 ( right panel ) expression levels in miR-142-deficient differentiated MKs ( black bars ) . In addition , overexpression of the miR-142-3p mimic resulted in significant reduction of Wasl ( left panel ) and Cfl2 ( right panel ) expression levels in WT MKs ( white bars ) . *p<0 . 05; **p<0 . 005; ( NS ) not significant . ( F and G ) Western blot analysis of representative miR-142-3p target genes . Cell lysates from WT and miR-142−/− FL-derived MKs were subjected to SDS-polyacrylamide gel electrophoresis . WASL ( F ) and COFILIN-2 ( G ) were immunodetected and assessed by densitometry ( right panel in F and G ) . ATP-synthase and GAPDH are indicators of protein loading levels , respectively ( left panels in F and G ) . Representative results from one of two independent experiments ( mean + SEM ) , four biological samples in each group . *p<0 . 05; **p<0 . 005 . ( H–J ) Relative luciferase activity of reporters that harbor the 3′UTR of novel miRNA targets: Wasl ( H ) , Cfl2 ( I ) , and Grlf1 ( J ) . Luciferase reporter activity is repressed by transfection of miR-142 expression vector ( gray bars ) in HEK-293T cells , whereas reporters that harbor a mutated version of the 3′UTR are insensitive to miR-142 . Data normalized to the activity of firefly luciferase that is co-expressed from the dual reporter and to a negative control miRNA vector and presented as mean + SEM . *p<0 . 05; **p<0 . 005; ***p<0 . 0005 . DOI: http://dx . doi . org/10 . 7554/eLife . 01964 . 01210 . 7554/eLife . 01964 . 013Figure 5—source data 1 . GO analysis for differentially regulated genes ( >twofold ) in miR-142−/− MKs . DOI: http://dx . doi . org/10 . 7554/eLife . 01964 . 01310 . 7554/eLife . 01964 . 014Figure 5—source data 2 . Genes commonly up-regulated ( >1 . 5-fold ) in miR-142−/− MKs and DCs ( CD24+ and CD24− ) . TargetScan predicted targets of miR-142-3p miR-142-5p are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 01964 . 01410 . 7554/eLife . 01964 . 015Figure 5—figure supplement 1 . Expression distribution plot of miR-142 putative targets . miR-142-3p TargetScan-predicted targets ( blue right panel ) are over-represented among genes upregulated in miR-142−/− FL-derived MKs ( p<10E−16 ) . miR-142-5p TargetScan-predicted targets ( blue , left panel ) show same distribution as the background . DOI: http://dx . doi . org/10 . 7554/eLife . 01964 . 01510 . 7554/eLife . 01964 . 016Figure 5—figure supplement 2 . qPCR expression analysis of miR-142 putative targets in precursor cell populations . qPCR expression analysis of Wasl and Cfl2 in PreMegE ( left panel ) and MkP ( right panel ) precursor cell populations sorted from miR-142−/− and WT BMs . *p<0 . 05; ***p<0 . 0005 . DOI: http://dx . doi . org/10 . 7554/eLife . 01964 . 01610 . 7554/eLife . 01964 . 017Figure 5—figure supplement 3 . miR-142-3p directly regulates cytoskeletal genes . Schematic representations of 3′UTRs of cytoskeletal genes ( A ) Cfl2 , ( B ) Wasl , ( C ) Bod1 , Twf1 , ( E ) Itgav , ( F ) Grfl1 with miR-142-3p binding sites and the corresponding mutations made to test direct interactions with miR-142-3p . Conserved bases within the seed region are indicated in red . Not drawn to scale . DOI: http://dx . doi . org/10 . 7554/eLife . 01964 . 01710 . 7554/eLife . 01964 . 018Figure 5—figure supplement 4 . miR-142-3p directly regulates cytoskeletal genes . Relative luciferase activity of reporters that harbor the 3′UTR of novel miR-142-3p targets: Bod1 ( A ) , Twf1 ( B ) , and Itgav ( C ) . Luciferase reporter activity is repressed by introduction of miR-142 ( grey bars ) into HEK-293T cells , whereas mutated reporters become insensitive to miR-142 . Data are normalized to the activity of firefly luciferase that is co-expressed from the dual reporter and to a negative control miRNA vector . DOI: http://dx . doi . org/10 . 7554/eLife . 01964 . 018 Gene ontology ( GO ) analysis using the database for annotation , visualization , and integrated discovery ( DAVID ) algorithm ( Huang da et al . , 2009a , 2009b ) revealed perturbations of a variety of cellular processes in miR-142−/− MKs ( Figure 5—source data 1 ) . To enhance the power of our analysis , we next focused on targets that were commonly up-regulated in miR-142−/− MKs as well as in miR-142−/− CD24+ and CD24- in vitro-derived DCs that we previously characterized ( Mildner et al . , 2013 ) . Forty genes were found to be commonly up-regulated in all three miR-142−/− hematopoietic cell types ( Figure 5C , Figure 5—source data 2 ) . This list is highly enriched with direct miR-142-3p putative targets ( 12 out of 40 genes ) . Subsequent DAVID analysis performed on these genes uncovered a significant enrichment for actin- and cytoskeleton-related functional GO categories ( Figure 5C ) . Indeed , half of the dozen predicted miR-142-3p targets encode for pivotal actin cytoskeleton-associated proteins , including Cofilin-2 ( Cfl2 ) , Glucocorticoid receptor DNA binding factor 1 ( Grlf1 ) , Biorientation of chromosomes in cell division 1 ( Bod1 ) , Integrin alpha V ( Itgav ) , Twinfilin-1 ( Twf1 ) , and Wiskott–Aldrich syndrome-like ( Wasl ) . Intriguingly , the latter was recently suggested as a potential target of miR-142-3p in the process of actin-mediated mycobacterial infection ( Bettencourt et al . , 2013 ) . The up-regulation of these six miR-142-3p targets in miR-142−/− MKs , was further confirmed by qPCR ( Figure 5D ) . Among these targets , Cfl2 expression was also found to be up-regulated in miR-142−/− MkPs ( Figure 5—figure supplement 2 ) . Introducing of miR-142-3p into differentiated MKs , using dsRNA mimetics , was sufficient to normalize the expression levels of representative miR-142-3p targets , Wasl and Cfl2 in miR-142−/− MKs ( Figure 5E ) . Western blot analysis revealed up-regulation of the protein products of Wasl and Cfl2 in miR-142−/− MKs , further substantiating them as bona fide miR-142-3p targets ( Figure 5F–G ) . Using luciferase reporter assays , we showed that the mRNAs of Wasl , Cfl2 , Grlf1 , Itgav and Twf1 are directly targeted by miR-142-3p ( Figure 5H–J , Figure 5—figure supplements 3 and 4 ) . Furthermore , the quantitative contributions of individual binding sites within the target 3′ UTR was revealed by stepwise loss of miR-142 regulation , when miR-142-3p binding sites were sequentially mutated ( Figure 5H–J , Figure 5—figure supplements 3 and 4 ) . Taken together , we discovered that the expression of a compound set of actin cytoskeleton regulators is post-transcriptionally controlled by miR-142-3p . miR-142-deficient MKs displayed perturbed actin filament dynamics and diminished proplatelet formation . This is presumably due to de-repression of several actin cytoskeleton components , including Wasl , Cfl2 , Twf1 , Itgav or Grlf1 , which are all direct miR-142-3p targets . Thus , we hypothesized that knocking down these miR-142-3p targets may relieve the PPF defect in miR-142−/− MKs . For this purpose , we transduced FL-derived MKs with short hairpin RNA ( shRNA ) -expressing lentiviral vectors that effectively knocked-down targets with more than three miR-142-3p binding sites , namely Wasl , Cfl2 or Grlf1 ( shWasl , shCfl2 and shGrlf1 , respectively , Figure 6—figure supplement 1 ) . A lentivirus encoding a shRNA directed against RFP ( shRFP ) was employed as a control . Following transduction , differentiation of MKs was induced and PPF was quantified 3 days later . The PPF defect was clearly evident in miR-142−/− MKs transduced with shRFP ( Figure 6A , B ) . In contrast , miR-142−/− MKs that were transduced with shRNA directed against Cfl2 , Wasl , Grlf1 , displayed a prominent elevation of PPF levels , and approximated levels observed in control WT MKs transduced with shRFP ( Figure 6A , B ) . This strong elevation in PPF levels was limited to miR-142−/− MKs , whereas in WT MKs , knockdown of miR-142-3p targets did not significantly alter PPF levels , except for Wasl-targeting shRNA that exhibited a modest increase in PPF efficacy following transduction ( Figure 6—figure supplement 2A ) . Additional pairwise target comparison revealed compound relationships between miR-142-3p targets ( Figure 6C; Figure 6—figure supplement 2B ) . Remarkably , knockdown of both Cfl2 and Wasl , which carry out opposing functions within the actin regulatory network , did not yield any significant increase in PPF levels in miR-142-deficient MKs ( Figure 6C ) . Conversely , concomitant Cfl2 and Grlf1 knockdown , two proteins that are important for destabilizing actin polymers , enhanced PPF in miR-142 null MKs ( Figure 6C ) . Lastly , a pool of shRNA against miR-142-3p targets was able to fully restore PPF capacity to WT levels ( Figure 6A–C ) . These data demonstrate that the expression of a set of actin cytoskeleton regulators should be tightly orchestrated by miR-142-3p in order to effectively promote platelet biogenesis . 10 . 7554/eLife . 01964 . 019Figure 6 . miR-142-3p targets a battery of actin cytoskeleton regulators to facilitate proplatelet formation . ( A ) Representative micrographs of WT and miR-142−/− ( KO ) FL-derived MKs , transduced with the indicated shRNA vectors and cultured for 48 hr with TPO . White arrows denote MKs extending proplatelets . Scale bars , 50 μm . ( B and C ) miR-142−/− MKs transduced with shRNAs targeting individual ( B ) , paired ( C ) , or a combined set ( B and C ) of miR-142-3p targets restore PPF levels . WT MKs were transduced with a control shRNA-targeting RFP . Representative results from one of two independent experiments ( mean + SEM ) , five experimental repeats in each group ( white dots ) . *p<0 . 05; **p<0 . 005; ***p<0 . 0005; ( NS ) not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 01964 . 01910 . 7554/eLife . 01964 . 020Figure 6—figure supplement 1 . Knockdown validation of miR-142-3p targets . qPCR expression analysis of Cfl2 , Wasl , and Grlf1 genes in NIH-3T3 fibroblasts that were transduced with either lentivrial vectors encoding shRNAs directed against miR-142-3p targets ( shCfl2 , shWasl , shGrlf1; grey bars ) , or with lentivirus harboring control shRNA directed against RFP ( shRFP; black bars ) . Data are presented as mean + SEM , and normalized to Hprt expression and to the mRNA expression in shRFP-transduced cells . **p<0 . 005; ***p<0 . 0005 . DOI: http://dx . doi . org/10 . 7554/eLife . 01964 . 02010 . 7554/eLife . 01964 . 021Figure 6—figure supplement 2 . Knockdown of miR-142-3p targets in WT MKs has no effect on proplatelet formation . WT MKs transduced with shRNAs targeting miR-142-3p targets exhibited PPF levels slightly higher than WT MKs that were transduced with a control shRNA-targeting RFP . Representative results from one of two independent experiments ( mean + SEM ) , five biological samples in each group . **p<0 . 005 . DOI: http://dx . doi . org/10 . 7554/eLife . 01964 . 021 Hematopoietic-specific miR-142 has emerged in recent years as a critical regulator of various blood and lymphoid cell lineages ( Gauwerky et al . , 1989; Chen et al . , 2004; Huang et al . , 2009; Belz , 2013; Lagrange et al . , 2013; Lu et al . , 2013; Mildner et al . , 2013; Nimmo et al . , 2013; Sonda et al . , 2013; Sun et al . , 2013; Zhou et al . , 2013 ) . Our analysis unveils the critical functions of miR-142 in the MK lineage . Using a recently-established mouse model , we show that deletion of the miR-142 allele results in pronounced thrombocytopenia . Our in vivo studies and culture assays reveal that proper miR-142 function is essential , in a cell-intrinsic manner , for MK maturation , including control of cell size , polyploidization and proplatelet elaboration . Furthermore , MKs require sustained miR-142-3p expression , as re-introduction of synthetic miR-142-3p mimetics , even onto differentiated MKs , was sufficient to restore functional maturity . Mechanistically , we demonstrate that miR-142-3p mediates the repression of an interconnected set of actin cytoskeleton regulators . These collectively contribute to MK maturation and their dysregulation is responsible for incomplete maturation observed in miR-142-deficient MKs . Very often , the sets of target genes that are predicted to be regulated by a particular miRNAs do not coalesce into coherent networks with distinct biological functions ( Stark et al . , 2005 ) . However , our work provides a clear example of an individual miRNA that co-regulates a network of functionally-associated targets . Indeed , the ability of a miRNA to modulate the expression of multiple targets within the same pathway simultaneously was previously suggested in other cellular contexts . For instance , the bicistronic miR-143/145 miRNA gene cluster dictates smooth-muscle cell phenotypic switching by orchestrating the expression of a cadre of cytoskeletal remodeling regulators ( Xin et al . , 2009 ) . Furthermore , miR-125 family members control hematopoietic stem cell pool size by targeting a cohort of proapoptotic genes ( Guo et al . , 2010; Ooi et al . , 2010 ) . Therefore , a single miRNA that cumulatively targets several nodes within the same biological circuit may serve as an effective means to control cellular behavior . The actin cytoskeleton participates in a wide array of cellular functions , and the dynamic turnover between its F-actin and monomeric G-actin forms is regulated by a large number of actin-binding proteins . Within this cytoskeletal regulatory network , the set of genes targeted by miR-142-3p contains components with divergent functions . For example , Cfl2 and Twf1 participate in disassembly of actin filaments ( Bamburg , 1999; Moseley et al . , 2006 ) . Likewise , Grlf1 is a GTPase activating protein that has been implicated in disruption of the organized actin cytoskeleton ( McGlade et al . , 1993 ) . Wasl , on the other hand , promotes actin polymerization by catalyzing filament branching together with the Arp2/3 complex ( Rotty et al . , 2013 ) . Thus , miR-142 deficiency destabilizes feedback loops required for actin filament homeostasis , stress fiber formation , and actin remodeling . This in turn impairs proplatelet generation and plausibly other MK-intrinsic cellular functions , such as endomitosis . It is also likely that additional miR-142-3p targets participate in the regulation of MK differentiation . For example , the over-representation of miR-142-deficient MkPs might result from a dysregulation of a distinct set of targets that are not necessarily related to actin regulation . Repression of miR-142-3p actin-associated targets was found to be sufficient for restoring PPF levels in miR-142-deficient MKs . Interestingly , knocking down the same targets in WT MKs did not result in significant increase in PPF capacity over WT baseline levels . This might be related to a certain threshold in PPF levels that cannot be crossed , even when regulation of miR-142-3p is mimicked . In summary , our analysis suggests a cardinal role for miR-142 in maturation of MKs , and in particular in controlling a network of chief actin regulators to facilitate MK terminal differentiation . The data challenges the prevailing paradigm , that miRNAs exert only subtle effects , often elicited by specific stressors ( Ebert and Sharp , 2012; Mendell and Olson , 2012; Pelaez and Carthew , 2012 ) , by providing in vivo evidence that genetic manipulation of a single miRNA may have a significant impact on cellular commitment and differentiation . Several platelet disorders have been associated with mutations in genes involved in actin organization , including Wiskott-Aldrich syndrome protein ( WASP ) ( Massaad et al . , 2013 ) and Actinin alpha 1 ( ACTN1 ) ( Kunishima et al . , 2013 ) . Because miR-142 locus is involved in B-cell leukemogenesis ( Gauwerky et al . , 1989 ) and since miR-142 is necessary for CD4+ DCs development ( Mildner et al . , 2013 ) and MK differentiation , miR-142 may function as a broad hematopoietic pro-differentiation factor . Thus , changes in miR-142 levels or activity may lead to platelet disorders or to hematopoietic malignancies . Mice strains were housed and handled in accordance with protocols approved by the Institutional Animal Care and Use Committee of WIS . To generate miR-142−/− mice , a gene-trapped embryonic stem cell clone ( ES , C57BL/6J strain ) from TIGM ( College Station , TX ) , was chosen based on an insertion upstream of miR-142 hairpin . ES cells were microinjected into C57BL/6J host blastocysts . Chimeras and further transmission of the targeted allele through the male germline to heterozygous pedigree was confirmed by PCR analysis of genomic tail DNA . Homozygous and WT littermate mice were generated by additional intercrosses . For transplantation experiments , recipient mice were lethally irradiated , using a 10 . 5Gy cesium source . After 10 days of ciprofloxacin prophylaxis , approximately 5 × 106 cells were injected into the tail vein for repopulation of the hematopoietic system . About 100 μl whole blood was retro-orbitally drawn from age- and sex-matched miR-142−/− and WT littermates into glass capillary tubes that were pre-treated with 5 μl of 0 . 5M EDTA , to prevent coagulation . Complete blood count was performed on ADVIA 120 Hematology System ( Siemens Healthcare , Erlangen , Germany ) by American Medical Laboratories ( Herzliya , Israel ) . Femora and spleens of age- and sex-matched miR-142−/− and WT littermates were excised after euthanasia and fixed overnight in 4% paraformaldehyde . Femora were then decalcified in 14% EDTA for 2–5 days . Specimens were dehydrated in graded ethanols , washed and processed into paraffin blocks . Longitudinal paraffin sections were stained with hematoxylin and eosin ( H&E ) , May–Grünwald Giemsa or immunostained with anti-von Willebrand Factor ( Dako , Agilent technologies , Santa Clara , CA ) . An Axioplan light microscope ( Carl Zeiss , Oberkochen , Germany ) , equipped with an eyepiece graticule ( grid ) , was utilized for quantification of MK numbers , per 20X magnification field , in 5 μm BM or spleen sections . MKs size was measured as the maximal diameter in 50 consecutive MKs in the BM of the distal femur , using the cellSens digital imaging software on an Olympus BX51 microscope at 40X magnification . Mouse BM cells were harvested via flushing of the long bones with Dulbecco modified Eagle medium ( DMEM; Invitrogen , Life Technologies , Carlsbad , CA ) supplemented with 10% fetal bovine serum ( FBS; Invitrogen , Life Technologies , Carlsbad , CA ) , was followed by filtering through a 70-µm nylon mesh cell strainer , to remove bone debris . BM mononuclear cells were cultured in MethoCult M3231 medium supplemented with 50 ng/ml Thrombopoietin ( TPO; PeproTech , Rocky Hill , NJ ) , for 7 days according to the manufacturer's protocols ( StemCell Technologies , Vancouver , British Columbia , Canada ) . Colonies containing >3 MKs were counted as CFU–MKs . Duplicate assays were performed for each mouse . At least two mice were analyzed for each sample group . Mouse FLs , collected on E14 . 5 , were processed into single-cell suspension by successively passing through 18- 21- and 23-gauge needles . Cells were then cultured in DMEM supplemented with 10% FBS , 50 ng/ml murine TPO , 2 mM L-glutamine and penicillin–streptomycin . After 4–5 days , MKs were purified using a discontinuous gradient of bovine serum albumin ( BSA , 3% , 1 . 5% , and 0% , Sigma-Aldrich , St . Louis , MO ) . About 1–2 × 104 purified MKs were cultured per 6 . 4 mm diameter well in suspension in flat bottom 96-well plates . After 16 hr of incubation , the fraction of proplatelet-forming MKs per well was scored with a light microscope under label-blinded experimental designs . For Immunocytofluorescence , FL-derived MKs were allowed to adhere to fibronectin-coated cover slips for 3 hr . Cover slips were rinsed with PBS , fixed with 3 . 7% formaldehyde , and permeabilized with 0 . 1% Triton X–100 ( Sigma-Aldrich , St . Louis , MO ) . Cells were blocked with PBS , 2% BSA and incubated with Phalloidin–Rhodamine ( gift from Benny Geiger , Weizmann Institute of Science ) and anti-alpha–Tubulin antibody ( gift from Alexander Bershadsky , Weizmann Institute of Science ) for 1 hr . Following blocking , 4'6-Diamidino-2-phenylindole dihydrochloride ( DAPI ) was added for 5 min before slides were mounted with Immu-Mount ( Thermo Scientific ) . For actin dynamics experiments , MKs were treated after plating for 30 min with 1 mM cytochalasin D ( gift from Benny Shilo , Weizmann Institute of Science ) and subsequently the drug was washed-out by three medium changes . Cells were fixed with 4% paraformaldehyde at 1 hr or 2 hr after washout and then permeabilized with 0 . 2% Triton X-100 in PBS . Stress fibers were stained with FITC-conjugated Phalloidin ( Sigma-Aldrich , St . Louis , MO ) and fluorescent micrographs were captured with a Zeiss LSM510 Laser Scanning confocal microscope . Flow cytometric analysis was performed on a LSRII flow cytometer ( BD Biosciences , San Jose , CA ) with FlowJo Version 8 . 8 . 7 software ( TreeStar , Ashland , OR ) . BM from 6- to 8-weeks-old mouse femora and tibiae or FL-derived MKs were treated with red blood cell lysis buffer ( Ammonium-Chloride-Kalium , ACK , 0 . 15 M NH4Cl , 0 . 1 M KHCO3 , 1 mM EDTA in PBS ) . Cells were stained with PE-conjugated anti-CD41 antibody ( Abcam , Cambridge , England ) , FITC-conjugated anti-CD42b ( Emfret ) , or APC-conjugated anti-CD61 antibody . For DNA content analysis , cells were further fixed with 2% paraformaldehyde , stained with 1 µg/ml DAPI , 0 . 1% BSA ( Sigma-Aldrich , St . Louis , MO ) and 0 . 1% Saponin ( Sigma-Aldrich , St . Louis , MO ) . For detailed analysis , freshly obtained BM cells were stained with APC anti-CD150 , PE anti-CD41 ( Abcam , Cambridge , England ) , FITC anti-Sca-1 , Brilliant Violet anti-CD117 , PE-Cy7 anti-CD105 , Alexa700 anti-CD16/32 , biotin-labeled lineage cell detection cocktail ( CD4 , CD8α , B220 , Ter119 , Gr1 , CD11b ) , and streptavidin PerCP–Cy5 . 5 . All antibodies were from BioLegend or eBioscience , unless otherwise indicated . MK-erythroid bipotent progenitors , PreMegE , were gated by lin−c-kit+CD150+CD105−CD41− or lin−sca1−c-kit+CD16/32− , and MK progenitors , MkP , were gated by lin−sca1−c-kit+CD41+ ( Pronk et al . , 2007; Pronk and Bryder , 2011 ) . Sorted PreMegE and MkP were collected and RNA was extracted using RNeasy micro kit ( Qiagen , Venlo , Netherlands ) . Quantification of actin intensity and morphocytometry was performed with ImagestreamX flow cytometer and IDEAS 6 . 0 software ( Amnis Corp . , Seattle , WA ) . 2 × 104 FL-derived MKs were stained with APC-conjugated CD61 ( Biolegend , San Diego , CA ) , fixed using the Cytofix/Cytoperm kit ( BD Biosciences , San Jose , CA ) , and further stained with FITC-conjugated-Phalloidin , Alexa594-conjugated DNaseI ( Invitrogen , Life Technologies , Carlsbad , CA ) and DAPI . Images were compensated for fluorescent dye overlap by using single-stain controls . Analysis was done on in-focus single cell images as previously described ( George et al . , 2006 ) with single cell gating , using the area and aspect ratio features . Cell area was calculated in square microns from brightfield images . Circularity was calculated as average distance of object boundary from center , divided by the variation of this distance . Thus , shapes approximating circle exhibited low variation and gained higher values ( in arbitrary units ) . Actin polarity was calculated using the Delta Centroid XY feature , which calculates the distance ( in microns ) between image center ( brightfield ) and the intensity-weighted actin image center ( higher values indicate increased polarity ) . The F/G actin ratio was calculated by dividing the corresponding pixel intensities for each cell . HEK-293T cells were transfected by calcium phosphate with pLKO . 1 encoding shRNAs for knockdown of Wasl , Grlf1 , Cfl2 , or RFP ( TRC , Broad Institute of MIT and Harvard , Cambridge , MA ) , and lentivirus packaging plasmids ( pPAX2 pMD2 ) . Lentivirus supernatants were stored , and for knockdown efficacy assessment , lentiviral particles were added at multiplicity of infection ( MOI ) of 2 , onto 5 × 104 NIH-3T3 cells in 16-mm wells that were incubated with medium containing 8 μg/ml polybrene ( Sigma-Aldrich , St . Louis , MO ) . Selection of transduced cells was performed with puromycin ( 2 μg/ml ) that was added from the second day and until cells were harvested on day 5 . For MK transduction , approximately 2 × 105 FL cells were cultured up to 4 days in medium supplemented with 50 ng/ml TPO . On the fourth day , cells were purified using a BSA gradient and 1 × 104 cells were placed in 6 . 4-mm rounded wells in suspension with medium containing 8 μg/ml polybrene that was freshly supplemented with TPO . Lentiviral particles were added at MOI of 25 and transducted through centrifugation at 900×g , 32°C for 90 min . For mock transduction , an equivalent volume of medium was added . Cells were incubated at 37°C overnight , washed in PBS and split into five different 6 . 4-mm wells in a 96-well flat bottom suspension plate . Ultra-small core-shell maghemite nanoparticles consisting of a cerium [Ce ( III/IV ) ] cation-doped CAN-γ-Fe2O3 core and a coordinated branched polyethylenimine ( b-PEI25 ) shell ( 25 kDa ) have been prepared according to Israel et al . ( patent application PCT/IL2014/050064 ) . Typically , a CAN-γ-Fe2O3 NPs aqueous suspension ( 1 . 0 ml , Fe: 1 . 93 mg/ml–1 . 93 mg total Fe , 0 . 0346 mmol Fe , ICP-AES measurement ) was diluted in 25 . 0 ml double distilled water ( ddH2O ) . For the nanoparticle functional shell , we used b-PEI25 , which enables electrostatic binding of nucleosides and endosome destabilization by osmotic imbalance , leading to subsequent release of RNA into cell cytoplasm . Therefore , 10 . 13 mg of polycationic branched b-PEI25 , ( 10 . 0 mg/ml stock solution , 0 . 4053 µmol , Sigma-Aldrich , St . Louis , MO ) were added to CAN-γ-Fe2O3 NPs at a 1:5 . 25 ratio ( wt/wt ) . Mild b-PEI25 coordinated coating was accomplished by overnight orbital shaking at room temperature . The resulting crude core-shell b-PEI25-CAN-γ-Fe2O3 nanoparticles were washed three times in 10 ml ddH2O using an Ultra-15 Amicon centrifugal filter ( 100K , EMD-Millipore , Billerica , MA ) operated for 5 min at 4 , 000 rpm . Then , a size exclusion process was performed by centrifugation ( 8 , 000 rpm , 16 min , 18°C and 7 , 000 rpm , 10 min , 18°C ) afforded the corresponding cleaned b-PEI25-CAN-γ-Fe2O3 nanoparticles . Selected nanocomposite characterization of such functional nanoparticles disclosed respective average TEM/DLS NP diameters of 6 . 86 ± 1 . 55 and 82 . 90 nm ± 1 . 26 ( DLS , PDI: 0 . 195 ) . NP ξ potential ( ddH2O ) is +31 . 1 mV . TGA weight loss ( N2 atmosphere , 200–410°C temperature range ) is 73 . 62% . For MK transfection , 0 . 49 μg ( 100 nM ) of miR-142-3p mimics dsRNA oligonucleotides , or control sequence ( Integrated DNA Technologies , Inc . , Coralville , IA ) were mixed with b-PEI25-CAN-γ-Fe2O3 nanoparticles at a 0 . 315 Fe/dsRNA wt/wt ratio , incubated 15 min at RT , and then , transfected to 3 × 104 FL-derived MKs in 35-mm plates . mmu-miR-142-3p guide sequence is: U*G*UAGUGUUUCCUACUUUAUmGmGA . An extensively-modified passenger strand sequence is: 5′-C3 ( spacer ) /UmCCmAUmAAmAGmUAmGGmAAmACmACmUAmCA/3′-Cy5 . 5 ( dye ) . ‘m’ indicates a 2'O-Methyl RNA and ‘*’ indicates a phosphorothioate internucleotide linkage . Cells were then incubated at 37°C overnight , washed in PBS and further cultured in flat bottom 6 . 4-mm wells in the presence of TPO , for 48 hr after which PPF levels were scored , and RNA was extracted with RNeasy micro kit ( Qiagen , Venlo , Netherlands ) . Total RNA was isolated with Tri-Reagent ( MRC ) following manufacturer's instructions . RNA quality was assessed with ND–1000 Nanodrop ( Peqlab ) and on a 1 . 5% agarose gel prior to gene-expression profiling using the Mouse Genome Gene 1 . 0 ST Affymetrix Gene Chip according to the manufacturer's instructions . For real-time Quantitative ( q ) PCR , cDNA synthesis was carried out by using oligo d ( T ) primer ( Promega ) and SuperScript II reverse transcriptase ( Invitrogen , Life Technologies , Carlsbad , CA ) , following manufacturer's instructions . qPCR analysis of mRNA expression was performed on a LightCycler 480 Real-Time PCR System ( F . Hoffmann-La Roche Ltd , Basel , Switzerland ) , using KAPA SYBR FAST qPCR Kit ( Kapa Biosystems , Wilmington , MA ) . Efficiency of each primer pair was confirmed by serial dilutions of templates . For quantification of mature mmu–miR–142 forms , cDNA synthesis was carried out by the miScript Reverse Transcription Kit and qPCR reaction utilized miScript SYBR Green PCR Kit ( with miScript Universal Primer , Qiagen , Venlo , Netherlands ) . U6 and hypoxanthine phosphoribosyltransferase 1 ( Hprt ) were used as a reference for normalization of miRNA and mRNA levels , respectively . All primer sequences are provided in Supplementary file 1 . For protein quantification , FL-derived MKs were lysed in radioimmunoprecipitation ( RIPA ) buffer with protease and phosphatase inhibitors . Protein concentration was determined using protein assay ( Bio-Rad Laboratories Inc . Hercules , CA ) . 20 μg of proteins were separated by SDS polyacrylamide gel electrophoresis , electrotransferred onto 0 . 2-mm nitrocellulose membrane , blocked in TBS , 0 . 1% Tween20 and 5% dry milk for 1 hr and incubated overnight with primary Antibodies: anti-N-WASP/Wasl ( 4848s; 1:1000; Cell Signaling Technology ) , anti-Cofilin2 ( ab96678; 1:1000; Abcam , Cambridge , England ) , anti-GAPDH ( AM4300; 1:10 , 000; Ambion ) and anti-ATP-synthase ( MS507; 1:2000; MitoSciences , Eugene , OR ) . HRP-conjugated secondary antibody ( Jackson ImmunoResearch Laboratories , West Grove , PA ) was diluted in TBS , 0 . 1% Tween20 . Immunoreactive proteins were detected using ECL ( GE Healthcare , Little Chalfont , UK ) and imaged using ImageQuant Las4010 . Quantification of blots was performed using ImageJ imaging software . Microarray data may be found at the Gene Expression Omnibus ( GEO ) under accession number GSE52141 .

DNA carries all the information needed for life . This includes the codes required for making proteins , as well as instructions on when , where , and how much of these proteins need to be produced . There are a number of ways by which cells control protein manufacturing , one of which is based on small RNAs called microRNAs . Before proteins are assembled , the DNA molecule is copied into a temporary replica dubbed messenger RNA . microRNAs are able to recognize specific messenger RNA molecules and block protein production . microRNAs serve a very important regulatory role in our bodies and are involved in virtually all cellular processes , including the production of all classes of blood and immune cells . Platelets seal injuries and prevent excessive bleeding by creating a clot at the location of a wound . Platelets are produced in huge cellular factories called megakaryocytes , which need to have a flexible and dynamic internal skeleton or cytoskeleton to produce the platelets . Chapnik et al . focus on one specific microRNA gene , which is vital for the production and the function of several classes of blood and immune cells . Chapnik et al . created a mouse model that does not produce one specific microRNA—miR-142—and found that mutant mice produced fewer platelets than normal mice . Although one possible explanation for this is that the mutant mice also had fewer megakaryocytes than normal , Chapnik et al . unexpectedly found that the number of megakaryocytes was in fact higher . However , these megakaryocytes do not reach functional maturity , which is required for platelet production . Many of the megakaryocytes made by the mutant mice were also smaller than normal and had an unusual cytoskeleton . Using a genomic approach and molecular tools , Chapnik et al . show that miR-142 affects the production of several of the proteins responsible for the dynamic flexibility of the cytoskeleton in mature megakaryocytes . Therefore , a single microRNA can target multiple different proteins that coordinate the same pathway in the cells that are critical for clotting and hence for human health . miR-142 has also been suggested to have important functions in blood stem cells and in blood cancer ( leukemia ) . Therefore , the new mouse model could be used to investigate many other facets of the blood and immune system . Further research could also focus on whether the same cytoskeletal network is in charge of miR-142 activity in other blood cells , or if miR-142 silences different targets in different cells .

2014

miR-142 orchestrates a network of actin cytoskeleton regulators during megakaryopoiesis

The Nef protein of HIV-1 downregulates the cell surface co-receptor CD4 by hijacking the clathrin adaptor complex AP-2 . The structural basis for the hijacking of AP-2 by Nef is revealed by a 2 . 9 Å crystal structure of Nef bound to the α and σ2 subunits of AP-2 . Nef binds to AP-2 via its central loop ( residues 149–179 ) and its core . The determinants for Nef binding include residues that directly contact AP-2 and others that stabilize the binding-competent conformation of the central loop . Residues involved in both direct and indirect interactions are required for the binding of Nef to AP-2 and for downregulation of CD4 . These results lead to a model for the docking of the full AP-2 tetramer to membranes as bound to Nef , such that the cytosolic tail of CD4 is situated to interact with its binding site on Nef . The human immunodeficiency virus type 1 ( HIV-1 ) is a lentivirus that causes acquired immunodeficiency syndrome ( AIDS ) . HIV-1 has a small genome encoding the main structural proteins Gag , Pol and Env , the regulatory proteins Tat and Rev , and the accessory proteins Nef , Vif , Vpr , and Vpu ( Frankel and Young , 1998 ) . During viral maturation , Pol is proteolytically cleaved to generate three proteins with enzymatic activity: the viral protease , integrase , and reverse transcriptase . These enzymes are the main targets for chemotherapeutic agents currently in use for the prevention and treatment of AIDS . Combination therapies with these agents have dramatically reduced HIV-1 transmission as well as HIV-1-associated morbidity and mortality . However , concerns about the development of drug resistance in addition to their side effects have fueled a continued search for additional targets . The accessory protein Nef was recognized early on as a potential target for inhibition of the pathogenic effects of HIV-1 ( Coleman et al . , 2001; Foster and Garcia , 2008 ) . Although not essential for infection in cell culture , Nef enhances viral replication and disease progression in vivo . The pathogenic effects of Nef are underscored by the observation that patients infected with Nef-deficient strains of HIV-1 often do not develop AIDS for over 10 years even if untreated ( these patients are referred to as ‘long-term non-progressors’ or ‘slow progressors’ ) ( Deacon et al . , 1995; Kirchhoff et al . , 1995; Gorry et al . , 2007 ) . Inhibition of Nef thus holds the promise to have a similarly beneficial effect . To date , however , this potential has not been realized mainly because Nef has no enzymatic activity and its mechanisms of action are insufficiently understood . At the cellular level , Nef has been ascribed multiple functions , of which the best characterized and most critical for pathogenesis is the downregulation of CD4 from the surface of infected cells ( Guy et al . , 1987; Garcia and Miller , 1991; Carl et al . , 2000; Glushakova et al . , 2001 ) . CD4 is a transmembrane protein that acts as a co-receptor in both the host’s immune response and the initial binding of HIV-1 to their target cells ( Bowers et al . , 1997 ) . Nef-induced CD4 downregulation interferes with the immune system ( Skowronski et al . , 1993 ) , prevents superinfection ( Benson et al . , 1993 ) and promotes virion release ( Lama et al . , 1999; Ross et al . , 1999 ) , all of which contribute to enhanced HIV-1 propagation . HIV-1 Nef is a small , polymorphic protein of 200–215 amino acids having a myristoylated N-terminus . X-ray crystallography ( Lee et al . , 1996; Arold et al . , 1997; Horenkamp et al . , 2011; Jia et al . , 2012 ) and NMR ( Grzesiek et al . , 1996a , 1997 ) have shown that Nef has a folded core ( residues 55–65 and 84–203 ) , with flexible N-terminal ( residues 1–54 ) and C-terminal ( residues 204–206 ) segments , and a central flexible loop ( residues 149–179 ) ( residue numbers correspond to the NL4-3 strain of HIV-1 ) . CD4 downregulation depends on both Nef myristoylation ( Aiken et al . , 1994 ) and specific residues in the loop , including Leu164 and Leu165 ( Bresnahan et al . , 1998; Craig et al . , 1998; Greenberg et al . , 1998; Janvier et al . , 2003 ) , which are in a sequence context fitting the [DE]XXXL[LI] motif for dileucine-based sorting signals ( Bonifacino and Traub , 2003 ) , and the diacidic motif , Asp174-Asp175 ( Aiken et al . , 1996; Lindwasser et al . , 2008 ) . Myristoylation allows recruitment of Nef from the cytosol to the inner leaflet of the plasma membrane ( Yu and Felsted , 1992 ) while the loop engages the clathrin-associated adaptor protein 2 ( AP-2 ) complex ( Jin et al . , 2005; Chaudhuri et al . , 2007; Doray et al . , 2007; Lindwasser et al . , 2008; Mattera et al . , 2011; Jin et al . , 2013 ) . The Nef:AP-2 complex interacts with the cytosolic tail of CD4 , leading to the cooperative assembly of a tripartite Nef:AP-2:CD4 complex ( Chaudhuri et al . , 2009 ) . This complex nucleates the formation of clathrin-coated pits ( Foti et al . , 1997; Greenberg et al . , 1997; Burtey et al . , 2007 ) that mediate rapid internalization of CD4 , followed by its delivery to lysosomes via the multivesicular body pathway ( Aiken et al . , 1994; Rhee and Marsh , 1994; daSilva et al . , 2009 ) . Biochemical analyses have demonstrated that binding of Nef to AP-2 is direct and dependent on the dileucine and diacidic motifs , and other residues , in the Nef loop ( Chaudhuri et al . , 2007; Doray et al . , 2007; Lindwasser et al . , 2008; Chaudhuri et al . , 2009; Mattera et al . , 2011 ) . AP-2 is a heterotetramer composed of α , β2 , μ2 and σ2 subunits . The N-terminal ‘trunk’ domains of α and β2 together with the whole μ2 and σ2 subunits constitute the core of the complex , whereas the C-terminal ‘hinge’ and ‘ear’ domains of α and β2 form long projections that extend from the core ( Owen et al . , 2004 ) . The AP-2 core undergoes a large conformation change from a ‘locked’ to an ‘open’ conformation that allows it to bind sorting signals and to be recruited to membranes via interaction with the phosphatidylinositol lipid PI ( 4 , 5 ) P2 ( Jackson et al . , 2010 ) . Nef has been shown to bind to the α–σ2 ‘hemicomplex’ ( Chaudhuri et al . , 2007; Doray et al . , 2007 ) . The σ2 subunit ( with a small contribution from the α subunit ) , harbors a binding site for [DE]XXXL[LI]-type signals from host cell proteins ( Kelly et al . , 2008; Jackson et al . , 2010; Mattera et al . , 2011 ) . Mutational analyses have shown that this site is also required for Nef binding , most likely through recognition of the Nef dileucine motif ( Mattera et al . , 2011 ) . The α subunit has an additional site , comprising Lys298 and Arg341 , which is also required for Nef binding and CD4 downregulation ( Chaudhuri et al . , 2009 ) . Although it is tempting to hypothesize that these basic residues interact with the Nef diacidic motif , there is currently no direct evidence for such an interaction . Importantly , this second site on α is not known to participate in any host cell function , making it a possible target for selective interference . Despite progress in the identification of determinants of the Nef-AP-2 interaction , the conformation of the Nef loop when bound to AP-2 and the molecular details of the interaction are not known . To elucidate the structural basis for this interaction , we have solved the crystal structure at 2 . 9 Å resolution of Nef ( residues 54–203 ) in complex with the α ( residues 1–396 ) and σ2 ( full-length ) subunits of AP-2 . The structure reveals that the entire central loop is well ordered , and that most of it contacts the α−σ2 hemicomplex . The Nef core is directly involved in contacts as well as serves as a scaffold to position the central loop . The structure leads to a model for the docking of HIV-1 Nef onto the plasma membrane in conjunction with AP-2 , and suggests how the AP-2:Nef complex binds to the CD4 cytosolic tail in the membrane setting . In the absence of PI ( 4 , 5 ) P2-containing membranes , the AP-2 core is in a locked conformation that has low affinity for both physiological cargoes and Nef . Previously , we assayed a version of the AP-2 core in which the μ2 C-terminal domain was deleted so as to destabilize the locked conformation . This construct bound to HIV-1 Nef with Kd = 6 μM as judged by surface plasmon resonance ( Chaudhuri et al . , 2007 ) . However , the conformational lability introduced into this construct made it unsuitable for crystallization . We built on the finding that Nef interacts with the α–σ2 hemicomplex as judged by yeast three hybrid ( Y3H ) ( Chaudhuri et al . , 2007 ) and pulldown assays ( Doray et al . , 2007 ) . Hemicomplex constructs including the full α trunk domain are poorly stable , because a large amount of hydrophobic surface area is exposed on the C-terminal part of the trunk domain when the hemicomplex is excised from the intact AP-2 core . A truncated version of the homologous γ-ζ hemicomplex of COPI including the first 17 helices of the γ trunk domain was found to be suitable for crystallography ( Yu et al . , 2012 ) . We designed a construct comprising the first 19 helices ( residues 1–396 ) of the α trunk domain and co-expressed it with full-length σ2 ( Figure 1A , B ) . This portion of the α trunk includes all of the Nef-interacting residues of α that have been documented to date . This construct bound to the HIV-1 NL4-3 Nef ( 54–203 ) ( hereafter , ‘Nef’ ) with Kd = 1 . 8 μM and 1:1 stoichiometry , as determined by isothermal titration calorimetry ( ITC ) ( Figure 1C ) . The comparatively high affinity of the interaction and the congruence with previous results with the tetrameric construct led us to conclude that this hemicomplex included all the major determinants of the AP-2:Nef interaction . 10 . 7554/eLife . 01754 . 003Figure 1 . Nef binds with low micromolar affinity to the AP-2 α–σ2 hemicomplex . ( A ) Schematic representation of AP-2 α–σ2 and Nef protein constructs . AP-2 α ( 1–396 ) ( cyan ) and full-length σ2 ( magenta ) were generated as a stable subcomplex and the interaction with the indicated Nef construct ( 54–203 ) ( orange ) was analyzed . ( B ) SDS gel of recombinant AP-2 α–σ2 and Nef proteins . ( C ) Isothermal titration calorimetry of the titration of His-tagged Nef ( 54–203 ) to the AP-2 α–σ2 hemicomplex . The upper panel shows the differential heat released when Nef ( 0 . 6 mM ) was injected into AP-2 α–σ2 solution ( 40 μM ) in 2 . 1 μl aliquots . DOI: http://dx . doi . org/10 . 7554/eLife . 01754 . 003 Nef was co-crystallized with the α1–396 form of the AP-2 α–σ2 hemicomplex ( hereafter , ‘α–σ2’ ) . The structure was determined by molecular replacement at 2 . 9 Å resolution ( Figure 2A , B; Table 1 ) . The asymmetric unit contains four Nef:α–σ2 complexes , all in similar conformations , with small variations in the quality of the electron density . The description will focus on the B , C , and D chains , for which the Nef:α–σ2 interface is most clearly visualized . Nef buries 1170 Å2 in this interface , of which two-thirds is buried against σ2 and the remainder against α . The α–σ2 unit is essentially rigid , in a conformation identical to that seen in other structures of the AP-2 complex ( Collins et al . , 2002; Kelly et al . , 2008; Jackson et al . , 2010 ) . The Nef core ( excluding the central loop 149–179 ) contains the five α-helices ( H2 , H3 , H6 , H7 , H8 ) and five-stranded β-sheet visualized in other Nef crystal structures ( Lee et al . , 1996; Figure 2C ) . The core also manifests a poorly ordered N-terminal helix ( H1 ) spanning residues 55–65 , which was not visualized in all of the chains . The identity of this helix was provisionally assigned on the basis of the only other crystal structure in which this region was visualized ( Breuer et al . , 2011 ) . This helix was first identified by solution NMR and contains the primary binding site for CD4 ( Grzesiek et al . , 1996a ) . In contrast to most other crystal structures , the central loop from residues 149–179 was visualized in its entirety ( Figure 2A ) . The central loop contains two additional helices , one from residues 150–157 ( H4 ) , and the other a single turn from 167–170 ( H5 ) . The central loop interacts extensively both with the α and σ2 subunits ( Figure 2D ) , with the greatest contact surface involving σ2 . The core interacts primarily via a network of interactions between helix H3 and the α subunit . 10 . 7554/eLife . 01754 . 004Figure 2 . Crystal structure of the AP-2 α–σ2:Nef complex . ( A ) F0-Fc omit map of Nef loop ( 149–179 ) with the final model superimposed . The map is contoured at 2 . 0 σ . ( B ) Overall ribbon representation of AP-2 α ( cyan ) and AP-2 σ2 subunits ( magenta ) in complex with Nef ( orange ) . ( C ) Detailed ribbon model of Nef ( orange ) with the secondary structures indicated . ( D ) Ribbon model of the Nef central loop ( 149–179 ) , which includes helix H4 ( 150–157 ) , the acidic-dileucine motif ( 160ExxxLL165 ) , helix H5 ( 167–170 ) , and the C-terminal turn-rich segment ( 171–179 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01754 . 00410 . 7554/eLife . 01754 . 005Table 1 . Statistics of crystallographic data collection and refinementDOI: http://dx . doi . org/10 . 7554/eLife . 01754 . 005ConstructAP-2: σ2 ( 1–143 ) , α ( 1–396 ) ; Nef ( 54–203 ) Data collection X-ray sourceAPS 22-ID Wavelength ( Å ) 1 . 0000 Space groupP212121 Cell dimensionsa = 109 . 56 Å , b = 168 . 03 Å , c = 200 . 20 Å , α = β = γ = 90° Resolution ( Å ) ( last shell ) 50 . 00–2 . 90 ( 3 . 00–2 . 90 ) Unique reflections80 , 188 Rsym* ( % ) 18 . 4 ( 53 . 9 ) I/σ7 . 4 ( 1 . 9 ) Completeness95 . 8 ( 80 . 2 ) Redundancy5 . 2 ( 3 . 2 ) Refinement Rwork/Rfree ( % ) 21 . 9/26 . 7 Average B values ( Å2 ) 39 . 5 Number of protein atoms21 , 588 R . m . s . bond length deviation ( Å ) 0 . 015 R . m . s . bond angle deviation ( ° ) 1 . 16 Ramachandran Plot ( % ) Favored98 . 4 allowed1 . 2 outlier0 . 4*Rsym = ΣhΣi|Ii ( h ) −<I>|/ΣhΣiIi ( h ) , where I is the observed intensity and <I> is the average intensity of multiple observations of symmetry-related reflections . The landmarks within the central loop are helix H4 ( 150–157 ) , the acidic-dileucine motif ( 160–165 ) , helix H5 ( 167–170 ) , and finally a series of turns centered on Met173 ( 171–179 ) ( Figure 2D ) . The intramolecular interactions of H4 are with other sections of the central loop , explaining why this helix has not been observed in other structures of Nef . The Leu164-Leu165 pair of the dileucine motif anchors the loop in a pocket on σ2 just as seen for a dileucine peptide bound to the unlatched AP-2 tetramer ( Kelly et al . , 2008; Figure 3A ) . The dileucine peptide in the unlatched structure ( Kelly et al . , 2008 ) used for comparison is derived from CD4 , but it is important to emphasize that this binding mode is dependent on phosphorylation and is unrelated to Nef-dependent downregulation . The pocket walls are formed by hydrophobic residues of σ2 . Nef Glu160 of the motif binds to basic residues on both σ2 and α ( Figure 3B ) . Residues of H4 , notably Glu154 , make electrostatic interactions with a second basic patch on σ2 ( Figure 3C ) . H4 and the dileucine motif have little or no interaction with the Nef core , and their conformation seems to be specified by their interactions with α–σ2 . 10 . 7554/eLife . 01754 . 006Figure 3 . The AP-2 α–σ2:Nef interface . ( A ) Stick representation of the Nef dileucine motif ( Leu164 and Leu165 , orange ) interacting with AP-2 σ2 ( magenta ) , compared with a bound dileucine peptide ( blue , PDB id: 2JKR ) ( Kelly et al . , 2008 ) . ( B ) Nef Glu160 of the acidic-dileucine motif forms hydrogen bonds with AP-2 α R21 and σ2 R15 . The hydrogen bond is listed as a purple dashed line . ( C ) Nef Glu154 in helix H4 ( orange ) forms hydrogen bonds with AP-2 σ2 R10 and R61 ( magenta ) . ( D ) The C-terminal part of Nef loop ( 171–178 ) interacts with both AP-2 α and σ2 ( magenta ) . ( E ) The key Nef diacidic motif Asp174 and Asp175 forms intramolecular hydrogen bonds that stabilize the loop conformation . Hydrogen bonds occur between the side chain of Asp174 and the main-chain amide NH of Gln104 , and between the side-chains of Nef Asp175 and Arg134 . ( F ) A salt bridge between Asp108 of Nef helix H3 bridges the Nef core to a basic patch on α . DOI: http://dx . doi . org/10 . 7554/eLife . 01754 . 006 In contrast , helix H5 and the C-terminal turn segment are sandwiched between α–σ2 and the Nef core . Their structure clearly depends on the interactions with α–σ2 , since they are not otherwise visualized in this conformation . However , the Nef core also seems to have an important role in organizing this segment . H5 packs against the β-sheet of the core , and serves primarily to orient the hydrophobic loop with respect to the dileucine motif . The tight turns of the C-terminal part of the loop serve to project a number of charged and hydrophobic side chains into complementary interactions with both α and σ2 ( Figure 3D ) . This unusual sequence of turns is anchored at its ends by H5 and by the strand β5 of the core . The whole turn-rich section of the loop from 171–179 is anchored internally by a hydrogen bond between Nef Asp174 and the main-chain amide of Gln104 , and by a partially buried salt bridge between Nef Asp175 of the loop and Nef Arg134 of the core β sheet ( Figure 3E ) . The structure also reveals that residues of Nef helix H3 of the core directly contact AP-2 . In particular , Gln104 , Arg105 , and Asp108 bind to a basic patch on α ( Figure 3F ) . This polar interface adjoins the mixed polar and hydrophobic interface created by the C-terminal turn segment of the central loop . There is one other minor interaction with the core region , involving Nef Pro129 of the β2-β3 loop . The Pro side-chain forms van der Waals interactions with atoms of the α Arg341 side-chain . While the central loop clearly dominates the interactions overall , the Nef core interactions are also significant , and represents one of the completely unexpected findings from the structural analysis . The Nef-AP-2 interaction is so central to CD4 downregulation that it has inspired exhaustive mutational analyses ( Aiken et al . , 1996; Hua et al . , 1997; Craig et al . , 1998; Janvier et al . , 2003; Lindwasser et al . , 2008; Chaudhuri et al . , 2009; Mattera et al . , 2011; Jin et al . , 2013 ) . These results can now be mapped onto the structure ( Figure 4 ) . We performed additional mutagenesis to test for the functional importance of residues that were newly identified by the structure determination ( Figures 5 and 6 ) . Mutations in the regions of AP-2 that were already known to bind dileucine signals had the expected loss of interaction . These include σ2 Y62A and A63D ( Figure 5C ) . Mutations in regions unique to Nef binding , notably σ2 R60E ( Figure 5C ) and α E342K ( Figure 5D ) , also eliminated binding . Other nearby residues with more peripheral interactions , including σ2 N48A , H85A , and C99A and α V300A , Q301A , and N344A , had lesser mutational phenotypes , if any ( Figure 5C , D ) . The results of these analyses are represented in Figure 4 together with previously published data . The collective body of work and its structural mapping are summarized in Table 2 . The large majority of the mutational hits map to residues that directly participate in Nef-AP-2 contacts . The consistency validates both the previous mutational approach and the structural findings . 10 . 7554/eLife . 01754 . 007Figure 4 . Structural mapping of mutations that interfere with binding and CD4 downregulation . ( A ) The surface representation shows the contact between AP-2 α−σ2 and Nef . ( B ) AP-2 α−σ2 or Nef interfaces are rotated by 90° to expose the interaction surfaces directly to view . Interacting residues in AP-2 α are colored in yellow , residues in AP-2 σ2 are colored in pink , and residues in HIV-1 Nef are highlighted in light blue . DOI: http://dx . doi . org/10 . 7554/eLife . 01754 . 00710 . 7554/eLife . 01754 . 008Figure 5 . Structural interface mutants of Nef and AP-2 complexes prevent binding . Y3H analysis of HIV-1 Nef and AP-2 α–σ2 hemicomplexes with mutations of residues revealed in the crystal structure . ( A ) Diagram of the plasmids used in Y3H analysis . NL4-3 Nef or mouse Tyrosinase cytosolic tail was cloned into MCS1 of pBridge and expressed as a GAL4BD fusion protein . AP-2 σ2 or AP-1 σ1 was cloned into MCS2 of pBridge and expressed without Met . AP-2 α or AP-3 δ was cloned into MCS of pGADT7 and expressed as a GAL4AD fusion protein . ( B–D ) . The indicated combinations of double transformants were plated in media lacking Leu , Trp , Met and His ( −HIS ) , −HIS with 3-AT ( 1 mM or 5 mM ) or Leu , Trp and Met ( +HIS ) . mTyr , mouse Tyrosinase cytosolic domain . DOI: http://dx . doi . org/10 . 7554/eLife . 01754 . 00810 . 7554/eLife . 01754 . 009Figure 6 . Nef interface mutants do not downregulate CD4 . Nef Asp108 , Arg134 , and Glu177 are required for the Nef-induced CD4 downregulation . ( A ) HeLa cells were cotransfected with pCMV-CD4 and pIRES-eGFP-Nef wild-type or mutant plasmids for 24 hr . The cells were then stained with APC-conjugated anti-CD4 antibody and PE-conjugated anti-Transferrin receptor ( TfR ) antibody . GFP was used as an indicator for transfected cells . The D174A , D175A mutant Nef was used as a negative control ( Shaded curves in all plots ) . Data shown are representative of three independent experiments . ( B ) The graph shows the relative number of CD4 positive cells from Figure 6A ( mean ± SD; N = 3; asterisks: p<0 . 001 compared with wild-type Nef ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01754 . 00910 . 7554/eLife . 01754 . 010Table 2 . Functional importance of Nef-AP-2 interacting residuesDOI: http://dx . doi . org/10 . 7554/eLife . 01754 . 010Nef residueReferencesInteracting residuesInteractionsReferencesQ104This studyα K298* , K299Hydrogen bond ( Chaudhuri et al . , 2009 ) D108*This studyα K298* , K299Salt bridge ( Chaudhuri et al . , 2009 ) P129This studyα R341*Van der Waals ( Chaudhuri et al . , 2009 ) R134*This studyNef D175*Nef core-to-loop internal salt bridgeThis study; ( Mattera et al . , 2011 ) E154 ( Lindwasser et al . , 2008 ) σ2 R61 , R10Salt bridgeThis studyN157 ( Lindwasser et al . , 2008 ) σ2 A63Van der Waals ( Mattera et al . , 2011 ) E160 ( Lindwasser et al . , 2008 ) σ2 R15 , α R21Salt bridge ( Mattera et al . , 2011 ) N161 ( Lindwasser et al . , 2008 ) σ2 C99 , L101Van der WaalsThis study; ( Mattera et al . , 2011 ) S163 ( Lindwasser et al . , 2008 ) σ2 N97Weak hydrogen bondsL164* ( Janvier et al . , 2003; Lindwasser et al . , 2008 ) σ2 Y62 , A63 , L65 , F67 , V88 , L91 , V98 , L103HydrophobicThis study; ( Mattera et al . , 2011 ) L165* ( Janvier et al . , 2003; Lindwasser et al . , 2008 ) σ2 Y62 , H85 , V88 , E89 , N92HydrophobicThis study; ( Mattera et al . , 2011 ) S169 ( Lindwasser et al . , 2008 ) σ2 A63Van der Waals ( Mattera et al . , 2011 ) L170* ( Lindwasser et al . , 2008; Jin et al . , 2013 ) σ2 Y62HydrophobicThis studyH171This study; ( Lindwasser et al . , 2008 ) σ2 A63Hydrogen bond to main chain carbonylThis study; ( Mattera et al . , 2011 ) M173 ( Lindwasser et al . , 2008 ) α Q301 , V300; σ2 R60 , Y62 , H85Hydrophobic and nitrogen-sulfur hydrogen bondThis studyD174* ( Lindwasser et al . , 2008 ) Nef Q104Nef core-to-loop internal hydrogen bond to main chain amideThis studyD175* ( Lindwasser et al . , 2008 ) Nef R134*Nef core-to-loop internal salt bridgeThis study; ( Chaudhuri et al . , 2009 ) P176 ( Lindwasser et al . , 2008 ) α R341* , E342Van der WaalsThis study; ( Chaudhuri et al . , 2009 ) E177*This study; ( Lindwasser et al . , 2008 ) α R341*Salt bridgeThis study; ( Chaudhuri et al . , 2009 ) R178 ( Lindwasser et al . , 2008 ) α E342 , σ2 N48Strong hydrogen bond , weak salt bridgeThis studyBold: mutation inhibits binding . Italics: mutation has no effect on binding . Plain text: not tested . Only residues tested as single amino acid substitutions are included . *Mutation inhibits CD4 downregulation activity . All residues tested as single amino acid substitutions except for K298/R341 , which were tested together . Previous analyses had highlighted the importance of a basic patch on α comprising both Lys298-Lys299 and Arg341 . The present structure revealed that Nef Glu177 is the primary interaction partner for α Arg341 . The Nef E177K mutant manifested a reduction in both α–σ2 binding ( Figure 5B ) and CD4 downregulation ( Figure 6A , B ) , consistent with its structural role . The deepened insight obtained from the crystal structure shows that this patch is better conceived of as a polar , rather than basic , patch . For example , the side chain of α Glu342 has a close approach to Nef Arg178 . The finding that Nef Asp174 and Asp175 do not contact AP-2 directly was a surprise . These two residues are required for AP-2 binding ( Figure 5B ) and CD4 downregulation ( Lindwasser et al . , 2008 ) , and had been expected to interact with a basic patch on α . They are , in fact , in the general vicinity of a basic patch , close to α Lys298 and Lys299 . Nevertheless , Asp174 , while close to the interface , is not in direct contact with AP-2 . Its key role appears to be to anchor the turn section to the N-terminus of helix H3 . The partial positive charge at the helix N-terminus can form an interaction with Asp side-chains that is almost as energetically favorable as a salt bridge . In this case , a short hydrogen bond is formed with the main-chain amide group of Nef Gln104 . Asp175 has a similar role in conformational stabilization . Asp175 is partially buried in a contact with the Nef core , and forms a salt bridge with Arg134 . We hypothesized that the internal Nef Arg134-Asp175 internal salt bridge is important for stabilizing the C-terminal turn portion of the central loop in its AP-2 binding conformation . Indeed , mutation of Arg134 to Glu abrogated interaction of Nef with α–σ2 ( Figure 5B ) . Moreover , co-expression in HeLa cells of CD4 with Nef R134E followed by FACS analysis showed that this Nef mutant almost completely lost its ability to downregulate CD4 ( Figure 6A , B ) . These results phenocopy the D175A change ( Lindwasser et al . , 2008 ) and are consistent with a critical role for the salt bridge in stabilizing the conformation of the Nef loop required for AP-2 binding and CD4 downregulation . While most mutational studies of the Nef-AP-2 interaction focused on the Nef central loop , one face of helix H3 , including residues Gln104 and Asp108 , contacts AP-2 . The Nef mutant Q104A behaves like wild type in both binding to α–σ2 ( Figure 5B ) and CD4 downregulation ( Figure 6A , B ) . This mutation does not alter the main-chain , and so will not affect the ability of the main-chain of residue 104 to help anchor the central loop via Asp174 . The charge-reversal mutant D108K , however , eliminates α–σ2 binding as judged by Y3H ( Figure 5B ) and CD4 downregulation ( Figure 6A , B ) . This finding is consistent with the salt bridge seen between α Lys299 and Nef Asp108 in the structure . Another core residue , Pro129 , has limited van der Waals interactions with α , thus it was not surprising that its mutation has no effect on the interaction ( Figure 5B ) . These results corroborate the importance of core helix H3 in the AP-2 interaction and CD4 downregulation . AP complexes function as membrane-bound , tetrameric assemblies . A consensus view of the structure and membrane-docking mode of AP complexes has emerged from the structures of active conformations of the tetrameric AP-1 and AP-2 cores ( Jackson et al . , 2010; Canagarajah et al . , 2013; Ren et al . , 2013 ) . A model for the membrane-bound AP-2:Nef complex in the consensus docking geometry was generated by superposition of the α–σ2 hemicomplex on the full open tetramer ( PDB 2XA7 ) ( Jackson et al . , 2010 ) . A steric clash was observed with helix 1 of the β2 subunit . β2 helix 1 was previously shown to be conformationally labile in the unlatching ( partial activation ) of AP-2 ( Kelly et al . , 2008 ) , and it seems reasonable to expect that it pivots in the full AP-2:Nef complex to prevent a collision . The consensus docking of the AP-2:Nef complex places Nef proximal to the membrane such that the N-terminal helix H1 is aligned parallel and in contact with the membrane ( Figure 7A , B ) . The C-terminus of helix H2 and parts of the central loop also align such that they would contact the membrane surface . The N-terminus of H1 is exposed to solvent such that there would be nothing to impede contact of the myristoylated N-terminus with the membrane . Indeed , the exposed face of H1 conjoins with H2 and parts of the central loop of Nef and the membrane-proximal parts of AP-2 to form a bowl that is 30 × 30 Å across , with a clearance of ∼10 Å from the membrane ( Figure 7C ) . Strikingly , the exposed face of H1 contains residues Trp57 and Leu58 that have been implicated in direct binding to CD4 ( Grzesiek et al . , 1996a; Figure 7D ) . Other residues implicated in CD4 binding , including Leu97 , Arg106 , and Leu110 , also project into the bowl ( Grzesiek et al . , 1996a ) . The edge of the bowl includes residues from α , σ2 , and β2 ( Figure 7C , D ) . This suggests that multivalent interactions between Nef , CD4 , and AP-2 likely drive cooperativity in the formation of the ternary complex . 10 . 7554/eLife . 01754 . 011Figure 7 . Docking of the unlocked AP-2:Nef complex to the membrane . ( A ) The unlocked conformation of AP-2 core bound to myristoylated Nef ( orange ) . The AP-2 α–σ2:Nef structure was first aligned with the open conformation of AP-2 core structure ( 2XA7 ) , and then docked on the membrane . The second view ( B ) is shown by rotating the first by 90° . Schematics are shown to the right of ( A ) and ( B ) . ( C ) The surface of the AP-2:Nef complex as viewed from the membrane . Nef residues that are mapped by CD4 binding ( Grzesiek et al . , 1996b ) are colored in blue . ( D ) Stick representation of Nef residues ( blue ) that interact with the CD4 cytosolic tail . DOI: http://dx . doi . org/10 . 7554/eLife . 01754 . 011 The structure of the AP-2:Nef complex provides a framework to unify nearly two decades’ worth of research of the molecular basis for CD4 downregulation by Nef . The large majority of Nef residues that have been implicated in CD4 downregulation reside in the central loop . The structure shows that all of these residues are well ordered in the complex , in contrast to all previous structures of Nef–effector complexes . Nearly all of these Nef residues directly contact AP-2 . One notable exception is Asp175 , which had been anticipated to bind to a basic patch on the surface of the α subunit . The structure revealed an unexpected role for Asp175 in stabilizing the conformation of the central loop . The structure also shows that the Nef core has both a direct role in forming polar interactions with AP-2 and an indirect role in scaffolding the conformation of the central loop . The role of Asp174 and Asp175 amounts to the formation of an effector-specific polar core within Nef . This provides insight into the underlying reason for the unusual architecture of Nef , as a single domain protein with disproportionately large internal loops . This architecture gives Nef an exceptional degree of plasticity , allowing multiple functions to be encoded within a relatively small structure . The structure is beautifully consistent with the emerging consensus picture that all activated AP complexes seem to bind to membranes in the same conformation and the same geometry ( Jackson et al . , 2010; Canagarajah et al . , 2013; Ren et al . , 2013 ) . The membrane-docking geometry suggested by the unlocked states of AP-1 and AP-2 places Nef such that it is touching the membrane , with its N-terminal region membrane proximal . This is consistent with its essential N-terminal myristoylation . One question still to be resolved is the disposition of the first helix of the β2-adaptin trunk , which collides sterically with Nef in the modeled conformation . It seems straightforward that this helix could pivot out of the way , but this has yet to be directly tested . The general proximity of Nef , and its N-terminal region in particular , to the membrane is similar to the model proposed for the AP-1:Nef complex that functions in MHC-I downregulation ( Jia et al . , 2012 ) . However , the details of the molecular contacts between Nef and the membrane surface differ ( Figure 8 ) . This suggests that Nef has more than one way to interact with membrane surfaces . 10 . 7554/eLife . 01754 . 012Figure 8 . Nef uses different surfaces to bind to different regions of AP-1 and AP-2 . ( A ) Nef binds to different subunits of AP-2 ( top ) and AP-1 ( bottom ) , but docks onto the membrane in both cases . The MHC-I cytoplasmic domain ( CD ) :Nef complex binds to the μ1 C-terminal domain ( CTD ) of the AP-1 core . One copy of AP-1 μ1-CTD:MHC-I-CD:Nef complex ( Jia et al . , 2012 ) ( pdb: 4EN2 ) was aligned with the open conformation of AP-1 core structure ( Ren et al . , 2013 ) ( pdb: 4HMY ) , and then the AP-1 complex was docked onto the membrane in the same orientation as shown for AP-2 in Figure 7 and in the top panel . ( B ) Structural superposition of Nef ( blue ) as bound to the μ1 subunit of AP-1 upon Nef ( orange ) bound to the α–σ2 hemicomplex of AP-2 in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 01754 . 012 The structural model , taken together with previous mapping of the CD4 binding site on Nef ( Grzesiek et al . , 1996b ) , suggests how CD4 binds to the AP-2 Nef complex on membranes . Residues of Nef helices H1 and H3 that are implicated in CD4 binding form one edge of a cavern 10 Å high and 30 × 30 Å across , with the membrane serving as the floor . The Nef binding site on CD4 comprises an approximate 17-residue tract ( Preusser et al . , 2001 ) that begins ∼10 residues C-terminal to the end of the transmembrane helix . The entirety of the cavern is within 10 Å of the membrane surface . Thus localization of the CD4 tail within the cavern would be completely consistent with the proximity of the Nef binding site and transmembrane domain of CD4 . The N-terminal loop of Nef is required for high-affinity binding to the CD4 cytosolic tail ( Preusser et al . , 2001 ) . We therefore propose that the N-terminal loop of Nef folds around the CD4 tail within or at the edge of the cavern . The dissociation constants for CD4:Nef and AP-2:Nef are known separately , and both are approximately 1 μM . The affinity and kinetics of CD4 binding to the AP-2:Nef complex are unknown , but this model suggests that the binding would likely be tighter than for these isolated components , and the off rate correspondingly slower . This would contribute to delivery of CD4 to the ESCRT machinery and its ultimate degradation in the lysosome ( daSilva et al . , 2009 ) . The search for ways to combat the emergence of resistance to antiretrovirals , and to ultimately eradicate HIV , has led to an intense interest in targeting the interactions of HIV and host proteins ( Jaeger et al . , 2011 ) . Host proteins , unlike viral proteins , are not susceptible to rapid mutation nor are they under selective pressure to resist therapy . The ideal host protein target site would be essential for the viral replication cycle , yet dispensable for host function . Such a site would also be ‘groovy’ , that is to say , highly invaginated and capable of binding small molecules . The interaction surface complementary to the C-terminal turn segment of the Nef central loop would appear to fit this criterion . Indeed , a deep pocket in AP-2 directly below the binding site for Nef Arg178 appears to be under-utilized by Nef ( Figure 9 ) . Future analysis of the ternary AP-2:Nef:CD4 tail complex may reveal additional promising sites . 10 . 7554/eLife . 01754 . 013Figure 9 . A highly concave pocket specific for the Nef interaction . Nef is shown in a stick model and the highly concave AP-2 surface is shown in the vicinity of Nef Arg178 . This region has no known interactions with physiological cargoes . DOI: http://dx . doi . org/10 . 7554/eLife . 01754 . 013 Protein expression plasmids were constructed by restriction cloning . Rat α-adaptin ( 1–396 ) was subcloned as an N-terminal GST fusion together with rat σ2-adaptin into the pST39 polycistronic vector ( Tan , 2001 ) . A TEV protease cleavage site was introduced between the GST tag and α-adaptin . HIV-1 Nef ( 54–203 ) was subcloned into pHis2 ( Sheffield et al . , 1999 ) and expressed as a fusion with an N-terminal His6 tag and a TEV cleavage site . All plasmids were verified by DNA sequencing . The AP-2 α–σ2 hemicomplex was expressed in Rosetta2 cells ( Novagen ) and induced with 0 . 3 mM IPTG at 20°C overnight . The cells were lysed by sonication in PBS buffer , pH 7 . 4 , 10% glycerol , 5 mM β-mercaptoethanol ( BME ) , 5 mM EDTA , and a protease inhibitor cocktail ( Sigma , St . Louis , MO ) . The clarified supernatant was first purified on GST Sepharose 4B resin ( GE healthcare ) . After His6-TEV cleavage at 4°C overnight , the sample was diluted in SP buffer A: 30 mM HEPES pH 7 . 4 , 3 mM BME and then loaded onto a HiTrap SP HP 5 ml column ( GE healthcare , Piscataway , NJ ) . Elution from the SP column was performed with a 70 ml linear gradient from 0–500 mM NaCl in SP buffer A . After each fraction was analyzed by SDS gel , the fractions were pooled and passed through 1 ml of GST resin and a Ni-NTA column ( Qiagen , Valencia , CA ) to capture the GST and His6-TEV . This sample was further purified on a HiLoad 16/60 Superdex 200 column ( GE healthcare ) in 20 mM Tris pH 7 . 4 , 200 mM NaCl , and 0 . 3 mM TCEP . HIV-1 Nef constructs were expressed in BL21 ( DE3 ) star cells ( Invitrogen , Carlsbad , CA ) , and induced at 25°C overnight . The cell pellet was lysed by sonication and the lysate was loaded onto a Ni-NTA column in 50 mM Tris pH 7 . 4 , 300 mM NaCl , 20 mM imidazole , 3 mM BME , 10% glycerol and protease inhibitor cocktail . The protein was eluted with 0 . 1 M imidazole , followed by TEV cleavage at 4°C overnight . After passing through Ni-NTA column to capture the cleaved His6 tag , the sample was loaded to HiLoad 16/60 Superdex 75 column ( GE healthcare ) in the sample buffer . Both TEV cleaved α–σ2 hemicomplex and His6-tagged Nef ( 54-203 ) were purified by size-exclusion chromatography in the same ITC buffer of 20 mM Tris pH 7 . 4 , 200 mM NaCl , 0 . 3 mM TCEP . The sample cell contained 0 . 2 ml of 40 μM α–σ2 hemicomplex , and Nef ( 600 μM ) was added in 18 injections of 2 . 1 μl each . Data from Nef injections into buffer blanks were subtracted from sample data before analysis . Measurements were repeated three times and carried out on an itc200 instrument ( MicroCal , Northampton , MA ) . The data were processed using Origin software ( MicroCal ) . The binding constant ( Kd ) was fitted using a one-site model . The α–σ2 hemicomplex was mixed with Nef ( 54–203 ) at a molar ratio of 1:1 . 2 in 20 mM Tris pH 7 . 4 , 200 mM NaCl , 0 . 3 mM TCEP . Crystallization was carried out by sitting-drop vapor diffusion using an automated liquid-handling system ( Mosquito , TTP LabTech , UK ) at 288 K in 96-well plates . The optimized reservoir solution contained a mixture of 49 μl of Wizard I #29 ( 100 mM CHES pH 9 . 5 , 200 mM NaCl , 10% PEG 8000 , Emerald Bio , Bedford , MA ) and 21 μl of 70% glycerol , adjusted to 0 . 2 mM inositol hexakisphosphate . The ratio of protein/precipitant in the drop was set at 2:1 . The final crystal was obtained in 2–4 days by micro-seeding at 5 mg/ml α–σ2 hemicomplex . The crystals were soaked in the cryprotectant paratone-N ( Hampton research , Aliso Viejo , CA ) and frozen in liquid N2 . Native data were collected from a single frozen crystal using a MAR CCD detector at beamline 22-ID , Advanced Photon Source . All data were processed and scaled using HKL2000 ( HKL research , Charlottesville , VA ) . The crystal diffracted to 2 . 9 Å resolution , and belonged to space group P212121 with unit cell dimensions a = 109 . 56 Å , b = 168 . 03 Å , c = 200 . 20 Å , α = β = γ = 90° . A molecular replacement solution was found using partial structures derived from the locked AP-2 core ( PDB: 2VGL ) ( Collins et al . , 2002 ) and from Nef/Hck-SH3 ( PDB: 3REA ) ( Breuer et al . , 2011 ) as search models with Phaser ( McCoy et al . , 2007 ) . Model building and refinement were carried out using Coot ( Emsley et al . , 2010 ) and Phenix ( Adams et al . , 2010; Table 1 ) . Structural figures were generated with PyMol ( DeLano , 2002 ) . Y3H analysis was performed as previously described ( Chaudhuri et al . , 2007 , 2009 ) . NL4-3 Nef or mouse tyrosinase cytosolic tail DNAs were subcloned into the pBridge vector ( Clontech , CA ) along with rat σ1 or σ2 . Rat α and δ subunit DNAs were subcloned into the pGADT7 vector ( Clonetech , CA ) . All the point mutants used in this study were generated by site-directed mutagenesis , using the QuikChange II XL ( Agilent technologies , Santa Clara , CA ) . The canonical dileucine-containing tyrosinase tail construct was included as a positive control for the formation of a functional complex , and the σ1 subunit of AP-1 and the δ subunit of AP-3 were included as negative controls for self-activation . The mutations were verified by DNA sequencing . The Saccharomyces cerevisiae HF7c strain was cotransformed with the indicated pairs of pBridge and pGADT7 constructs , using EZ Yeast Transformation Kit ( MP biomedicals , Solon , OH ) . Double transformants were selected and grown on plates lacking Leu , Trp , and Met ( +HIS ) for 3 days , then the colonies from each transformant were normalized and plated on + HIS plates and plates lacking Leu , Trp , Met , and HIS ( −HIS ) with/without 3-AT ( 3-amino-1 , 2 , 4-triazole ) for 4 days . FACS analysis was performed as described before ( Chaudhuri et al . , 2009 ) . Wild-type or mutant NL4-3 Nef was subcloned into the pIRES2-eGFP vector ( Clontech , CA ) . HeLa cells were co-transfected with pCMV-human CD4 and pIRES2-eGFP Nef wild-type or each mutant for 24 hr . The cells were then collected and stained with APC-conjugated anti-CD4 antibody and PE-conjugated anti-Transferrin receptor ( TfR ) antibody . The fluorescence was measured on a FACScalibur flow cytometer and analyzed by using CellQuest software ( Becton Dickinson , Franklin Lakes , NJ ) . Only GFP positive cells were counted , and the inactive D174A , D175A Nef mutant was used as a negative control .

Infection by a pathogen , such as a bacterium or virus , activates both the innate immune response—which is immediate but not specific to the pathogen—and the adaptive immune response , which is stronger and specific to the pathogen . White blood cells called CD4+ T helper cells play an important role in the early stages of the adaptive immune response by helping to activate and regulate other white blood cells that go on to eradicate the pathogen . HIV-1 is a retrovirus that infects immune cells that have the CD4 receptor on their surface , including CD4+ T helper cells . As the number of worker CD4+ T helper cells falls , the adaptive immune response gradually weakens , and the HIV-1 infected individual becomes increasingly susceptible to infection and disease . An individual is said to develop AIDS when either their CD4+ T helper cell count falls below 200 cells per microliter or they begin to experience specific diseases associated with the HIV-1 infection . In an effort to prevent and treat AIDS , researchers have worked to understand the HIV-1 genome and have developed medicines that target the enzymatic activity of viral proteins involved in viral replication . When used in combination , these drugs have helped to reduce transmission of HIV-1 , and also to reduce deaths from the disease . However , worries about side effects and drug resistance mean that there is a need to develop new drugs . The HIV-1 genome codes for a number of accessory proteins , including a protein known as Nef that attacks the CD4+ T helper cells , removing the CD4 protein that gives the cells their name . This reduces the ability of the T cells to activate the immune system and allows the virus to spread . Nef acts by forming a complex with a protein called AP-2 in the T cells , and this complex then interacts with the CD4 proteins , causing them to be internalized and then destroyed inside the cells . Ren et al . have now worked out the structure of the Nef:AP-2 complex at the molecular level and identified the amino acid residues within the Nef protein that interact with the AP-2 protein . This allowed Ren et al . to propose a detailed model of the interaction between the complex and the CD4 protein , and how this leads to the protein being destroyed . This information could be used to develop drugs that work by blocking the amino residues on AP-2 that bind to Nef . Moreover , since these sites are not susceptible to rapid mutations , such drugs are less likely to encounter the problem of drug resistance .

2014

How HIV-1 Nef hijacks the AP-2 clathrin adaptor to downregulate CD4

Increased expression of Down Syndrome Cell Adhesion Molecule ( Dscam ) is implicated in the pathogenesis of brain disorders such as Down syndrome ( DS ) and fragile X syndrome ( FXS ) . Here , we show that the cellular defects caused by dysregulated Dscam levels can be ameliorated by genetic and pharmacological inhibition of Abelson kinase ( Abl ) both in Dscam-overexpressing neurons and in a Drosophila model of fragile X syndrome . This study offers Abl as a potential therapeutic target for treating brain disorders associated with dysregulated Dscam expression . Dscam levels are increased in the brains of human patients with DS and in mouse models of DS ( Saito et al . , 2000; Alves-Sampaio et al . , 2010 ) . Recent research also suggests that fragile X mental retardation protein ( FMRP ) binds directly to the mRNAs of Dscam from mouse brain ( Brown et al . , 2001; Darnell et al . , 2011 ) , and studies in Drosophila neurons further confirmed that FMRP suppresses Dscam translation ( Cvetkovska et al . , 2013; Kim et al . , 2013 ) . In the dendritic arborization ( da ) neurons in Drosophila larva , Dscam expression level is instructive for presynaptic terminal growth ( Kim et al . , 2013 ) . Consistent with this , increased Dscam in Drosophila FXS models results in enlarged presynaptic arbors ( Kim et al . , 2013 ) . These findings indicate the importance of proper Dscam levels in normal development and in the pathogenesis of brain disorders . Because of the link between increased Dscam expression and neuronal defects in DS and FXS models , targeting Dscam or its signaling mechanism might prove therapeutic for these disorders . Currently , neither methods for targeting Dscam proteins nor those for targeting the signaling pathway activated by dysregulated Dscam are available , impeding the development of such therapies . In fact , very little is known about how Dscam signaling is transduced in vivo . In Drosophila , Dscam has previously been shown to bind to Dock ( Schmucker et al . , 2000 ) , while in mammals it has been shown to associate with Uncoordinated-5C , Focal adhesion kinase ( FAK ) , Fyn kinase , and PAK1 ( Li and Guan , 2004; Purohit et al . , 2012 ) . In addition , studies suggest possible genetic interactions between Dscam and the Abelson tyrosine kinase ( Abl ) in neurite development in the central nervous system ( CNS ) of Drosophila embryos ( Andrews et al . , 2008; Yu et al . , 2009 ) . However , evidence demonstrating the requirement of these potential interactors for the defects that arise from increased Dscam expression is lacking . Moreover , whether pharmacologically targeting these molecules in vivo might alleviate the effects of increased Dscam expression is unknown . The evolutionarily conserved Abl kinase transduces extracellular cues into cytoskeletal rearrangements that affect cell motility and shape ( Bradley and Koleske , 2009 ) and is implicated in axonal development , including axon guidance and extension ( Wills et al . , 1999a; Wills et al . , 1999b; Wills et al . , 2002; Hsouna et al . , 2003; Lee et al . , 2004; Forsthoefel et al . , 2005 ) . Overexpression of Abl causes increased axon growth in the Drosophila CNS ( Leyssen et al . , 2005 ) , which is reminiscent of the effect caused by Dscam overexpression in C4da neurons ( Kim et al . , 2013 ) . In addition , previous studies in Drosophila have indicated that abl mutations have an additive effect with Dscam mutations , such that abl/Dscam double mutant embryos have more severe axon midline crossing defects than either abl or Dscam mutants alone ( Andrews et al . , 2008; Yu et al . , 2009 ) . However , the molecular nature of this interaction , that is , whether or not Dscam acts through Abl , and particularly whether inhibition of Abl mitigates neuronal defects caused by dysregulated Dscam , is unknown . Here we show that Dscam activates Abl through its cytoplasmic domain , which is required for the presynaptic arbor enlargement caused by dysregulated Dscam expression in vivo . Importantly , we demonstrate that the pharmacological inhibition of Abl ameliorates exuberant presynaptic arbor growth both in flies overexpressing Dscam and in a fly model of FXS . We took advantage of the Drosophila larval class IV dendritic arborization ( C4da ) neurons to delineate the molecular mechanism of Dscam signaling in presynaptic arbor development , because the presynaptic terminal growth of these neurons is highly sensitive to Dscam levels in a linear fashion ( Kim et al . , 2013 ) . For example , loss of Dscam causes C4da presynaptic terminals to fail to grow while increased Dscam levels lead to increased presynaptic terminal growth ( Kim et al . , 2013 ) . From tests of candidate genes that potentially mediate Dscam function , including FAK , Fyn , PAK , RhoA , and Abl , we identified Abl as a key molecule mediating Dscam's functions in presynaptic terminal growth . We first asked whether Abl is sufficient to promote presynaptic terminal growth in C4da neurons . Consistent with a previous study performed in Drosophila adult CNS neurons ( Leyssen et al . , 2005 ) , overexpression of Abl in C4da neurons caused significant overgrowth of the presynaptic terminals ( Figure 1A , B , E ) . Since Abl is known to have both kinase-dependent and kinase-independent functions ( Henkemeyer et al . , 1990; Schwartzberg et al . , 1991; Tybulewicz et al . , 1991 ) , we tested whether expression of a kinase-dead form of Abl , Abl-K417N ( Henkemeyer et al . , 1990; Wills et al . , 1999b ) , could promote presynaptic terminal growth . We found that C4da presynaptic terminals overexpressing Abl-K417N were indistinguishable from wild-type ( Figure 1D , E ) , indicating that Abl kinase activity is required . Consistent with the idea that Abl kinase activation is important , expression of a constitutively active form of Abl , BCR-Abl , led to extremely exuberant overgrowth ( Figure 1C , E ) . Taken together , these results suggest that Abl is sufficient to promote presynaptic terminal growth and that the extent to which Abl instructs presynaptic terminal growth is related to Abl kinase activation . 10 . 7554/eLife . 05196 . 003Figure 1 . Dscam requires Abl to promote presynaptic terminal growth . ( A–E ) Abl is sufficient to cause presynaptic terminal overgrowth in C4da neurons . Transgenes were expressed with a C4da neuron-specific Gal4 driver , ppk-Gal4 , and presynaptic terminals were visualized with a membrane monomeric RFP ( mCD8-mRFP ) transgene . Overexpression of Abl ( B ) leads to a modest increase in presynaptic terminal growth as compared to control ( A ) . Overexpression of the constitutively active BCR-Abl ( C ) leads to robustly increased presynaptic terminal growth , while overexpression of kinase-dead Abl-K417N ( D ) is indistinguishable from control . Quantification of the number of axon connectives is shown in ( E ) . Scale bar is 10 μm . ( F–K ) Abl is required in C4da neurons for Dscam to instruct presynaptic terminal growth . The arrowhead in each panel points to the location where an axon elaborates the presynaptic terminal arbor . The MARCM technique was used to generate and visualize single mutant C4da neurons . While overexpression of Dscam::GFP ( G ) in single C4da presynaptic terminals leads to increased length when compared to control ( F ) , overexpression of Dscam in abl1 mutant neurons ( H ) leads to presynaptic terminal lengths that are indistinguishable from abl1 mutant neurons ( I ) . Similarly , overexpression of Dscam in abl4 mutant neurons ( J ) does not significantly change presynaptic terminal length when compared to abl4 mutant neurons ( K ) . ( L–N ) Abl is required to instruct presynaptic terminal growth in dFMRP mutants . ( M ) Loss of dFMRP leads to increased presynaptic terminal growth , which has previously been shown to require Dscam . Loss of one copy of abl in dFMRPΔ50M mutant neurons ( N ) leads to presynaptic terminal lengths that are indistinguishable from control ( L ) . Scale bar is 10 μm . ( O and P ) Quantification of the presynaptic terminal length in C4da neurons of indicated genotypes . Sample number is shown in white within each bar . DOI: http://dx . doi . org/10 . 7554/eLife . 05196 . 00310 . 7554/eLife . 05196 . 004Figure 1—figure supplement 1 . Loss of abl does not affect Dscam::GFP expression level . ( A and B ) Loss of abl does not affect Dscam::GFP expression in C4da cell bodies . ( A ) Example images of C4da neuron cell bodies ( white arrowheads ) in control ( left ) or abl1 homozygous mutant ( right ) animals . Upper images show merged signals of mCD8::mRFP and Dscam::GFP , while lower images show Dscam::GFP alone . Scale bar is 10 µm . ( B ) Quantification of the relative intensity of Dscam::GFP fluorescence normalized to mCD8::mRFP . Sample number is shown inside each bar . ( C and D ) Loss of abl does not affect Dscam::GFP expression in C4da presynaptic terminals . The MARCM technique was used to generate and visualize single mutant C4da neurons . ( C ) Example images of C4da presynaptic terminals in control ( left ) and abl1 mutant clones . Upper images show merged signals of mCD8::mRFP and Dscam::GFP , lower images show Dscam::GFP alone . Scale bar is 10 µm . ( D ) Quantification of the relative intensity of Dscam::GFP normalized to mCD8::mRFP . Sample number is shown inside each bar . DOI: http://dx . doi . org/10 . 7554/eLife . 05196 . 00410 . 7554/eLife . 05196 . 005Figure 1—figure supplement 2 . Loss of abl does not affect C4da dendritic length or morphology . Representative images of control ( A ) and abl1 mutant C4da neuron clones ( B ) . The average total dendritic length is not significantly different between these two conditions ( C ) . Scale bar is 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05196 . 00510 . 7554/eLife . 05196 . 006Figure 1—figure supplement 3 . Single Dscam isoform-induced ectopic repulsion between class I and class III dendrites does not require abl . The dendritic field of the class I da neuron vpda ( traced in magenta ) normally overlaps extensively with that of the class III da neuron v'pda ( traced in cyan ) ( A ) . When a transgene expressing a single Dscam isoform is overexpressed in both neurons , their dendritic fields segregate ( B ) , exhibiting an ectopic repulsion . The expression of the same Dscam transgene in abl1 neurons also leads to ectopic repulsion ( C ) . Original background images show the pan-neuronal marker labeled with anti-horseradish-peroxidase antibody ( red ) and Dscam::GFP transgene expression ( green ) . ( D ) Quantification of the number of dendritic branch crossing . Sample number is shown in white inside each bar . Scale bar is 25 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05196 . 006 Since overexpression of Abl increases presynaptic terminal growth , similar to Dscam , we next tested whether Dscam requires Abl to instruct presynaptic terminal growth . For this , we used the mosaic analysis with a repressible cell marker ( MARCM ) technique to overexpress Dscam in abl1 mutant C4da neurons ( Lee and Luo , 2001 ) and assessed presynaptic terminal length . We found that although Dscam overexpression led to significantly ( 150% ) longer presynaptic terminals than wild-type clones ( Figure 1F , G , O ) , abl1 mutant clones that overexpressed Dscam did not differ in length from abl1 mutant clones ( Figure 1H , I , O ) . Presynaptic terminal length was also subtly but significantly shorter in abl1 mutant clones compared to wild-type controls ( Figure 1I , O ) . A different loss-of-function allele of abl , abl4 , exhibited similar effects on the presynaptic overgrowth caused by Dscam overexpression ( Figure 1J , K , O ) , confirming that loss of abl function is responsible for blocking the presynaptic phenotypes caused by increased Dscam levels . As a control , abl loss-of-function mutations did not affect the expression of the Dscam transgenes in the C4da cell body or presynaptic terminals ( Figure 1—figure supplement 1 ) . As a further proof-of-concept , we asked whether loss of abl could mitigate the effects of dysregulated Dscam levels without utilizing Dscam transgenes . FXS is caused by an absence of FMRP ( Kremer et al . , 1991 ) , and is modeled in Drosophila using loss-of-function mutants for the Drosophila homolog of FMR1 , dFMRP ( Zhang et al . , 2001; Dockendorff et al . , 2002 ) . It has previously been shown that FMRP binds to Dscam mRNA in both mammals and Drosophila ( Darnell et al . , 2011; Cvetkovska et al . , 2013; Kim et al . , 2013 ) and that dFMRP represses Dscam expression to control presynaptic terminal growth , so that dFMRP mutants exhibit increased presynaptic terminal length in C4da neurons ( Kim et al . , 2013 ) . Strikingly , loss of only a single copy of abl significantly rescued presynaptic terminal length to wild-type levels ( Figure 1L–N , P ) . These results suggest that Abl is required for Dscam to instruct presynaptic terminal growth . An important function of Dscam in neuronal development is to mediate self-avoidance between neurites of the same neuron ( Zipursky and Grueber , 2013 ) . Abl does not seem to be required by Dscam for either dendrite growth ( Figure 1—figure supplement 2 ) or for dendritic self-avoidance in C4da neurons . Loss of abl did not compromise the ectopic avoidance caused by overexpressing Dscam in distinct types of da neurons ( Hattori et al . , 2007; Hughes et al . , 2007; Matthews et al . , 2007 ) ( Figure 1—figure supplement 3 ) . This suggests a divergence in Dscam signaling for the development of presynaptic terminals and dendritic branches . Taken together , these results indicate that Abl is specifically required for Dscam-mediated presynaptic terminal growth . Next , we asked how Abl might mediate Dscam signaling . Abl can be activated by binding to specific proteins , such as the cytoplasmic domains of membrane receptors ( Bradley and Koleske , 2009 ) . In contrast to the exuberant presynaptic terminal overgrowth caused by Dscam overexpression in C4da neurons ( Figure 2A , middle ) , overexpressing a mutant form of Dscam that lacked most of the cytoplasmic domain ( DscamΔCyto ) did not cause presynaptic terminal overgrowth ( Figure 2A , bottom ) . DscamΔCyto was trafficked to the axon terminals and expressed at a similar level to full-length Dscam ( Figure 2—figure supplement 1 ) . These results suggest that the cytoplasmic domain is required for Dscam to instruct presynaptic terminal growth . 10 . 7554/eLife . 05196 . 007Figure 2 . Dscam binds to Abl through its cytoplasmic domain . ( A ) The cytoplasmic domain of Dscam is required for instructing presynaptic terminal growth . Overexpression of full-length Dscam under the control of ppk-Gal4 ( A , middle ) leads to exuberant presynaptic terminal overgrowth when compared to control ( A , top ) . However , overexpression of DscamΔCyto ( A , bottom ) fails to increase presynaptic terminal growth . Scale bar is 10 μm . ( B ) Dscam binds Abl via its cytoplasmic domain . S2 cells were co-transfected with Abl::Myc along with either Dscam::GFP , DscamΔCyto::GFP , or an empty vector . Dscam::GFP was immunoprecipitated with anti-GFP antibody and bound Abl::Myc was examined with anti-Myc antibody ( top ) . Immunoprecipitated Dscam::GFP and input Dscam::GFP was examined with anti-GFP ( bottom ) . ( C ) Abl colocalizes and redistributes with Dscam but not with DscamΔCyto in presynaptic terminals in vivo . When expressed alone , Abl::Myc shows a diffuse pattern ( bottom ) . When expressed along with Dscam::GFP ( top ) , Abl::Myc redistributes into punctate structures that colocalize with Dscam::GFP . When expressed along with DscamΔCyto::GFP ( middle ) , Abl::Myc does not redistribute , displaying a similar pattern to when Abl::Myc is expressed alone ( bottom ) . This is quantified using Manders' Correlation Coefficient . M1 presents a measure of the fraction of Abl::Myc that overlaps with Dscam ( ΔCyto ) ::GFP , while M2 presents a measure of the fraction of Dscam ( ΔCyto ) ::GFP that overlaps with Abl::Myc . Both M1 and M2 are significantly increased in Abl-Dscam coexpression when compared to Abl-DscamΔCyto coexpression . Scale bar is 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05196 . 00710 . 7554/eLife . 05196 . 008Figure 2—figure supplement 1 . DscamΔCyto::GFP is trafficked to presynaptic terminals at a similar level to Dscam::GFP . Both Dscam::GFP ( left ) and DscamΔCyto::GFP ( right ) are trafficked to presynaptic terminals . In addition , presynaptic terminal overgrowth is observed 100% of the time when Dscam::GFP is overexpressed , while presynaptic terminal overgrowth is never observed when DscamΔCyto::GFP is overexpressed . Top image shows merged images mCD8::mRFP ( red ) and either Dscam::GFP or DscamΔCyto::GFP ( green ) . Bottom images show Dscam::GFP or DscamΔCyto::GFP only . Scale bar is 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05196 . 008 We then asked whether Dscam and Abl physically interact through the Dscam cytoplasmic domain . We found that Dscam and Abl proteins co-immunoprecipitated from transfected Drosophila Schneider 2 ( S2 ) cells expressing these two proteins ( Figure 2B , second lane from right ) . In contrast , DscamΔCyto did not co-immunoprecipitate Abl ( Figure 2B , furthest right lane ) . These results suggest that Dscam and Abl proteins form a complex through Dscam's cytoplasmic domain . Next , to test the in vivo interaction of Dscam and Abl in presynaptic terminals specifically , we determined whether Abl localization in presynaptic terminals was altered by the expression of Dscam or DscamΔCyto ( Figure 2C ) . When expressed alone or with DscamΔCyto::GFP , Abl::Myc was diffusely distributed in the presynaptic terminals , with little colocalization with DscamΔCyto::GFP ( Figure 2C , middle and bottom ) . However , when expressed with Dscam::GFP , Abl::Myc became more punctate and clearly colocalized with Dscam::GFP ( Figure 2C , top ) . We used Manders' Correlation Coefficients to quantify the colocalization of Dscam::GFP and Abl::Myc . Colocalization analysis revealed a significant increase in both M1 and M2 ( Figure 2C , bottom right ) when Abl::Myc was coexpressed with Dscam::GFP as compared to when Abl::Myc was coexpressed with DscamΔCyto::GFP , where M1 represents the fraction of Abl that overlaps with Dscam , and M2 represents the fraction of Dscam that overlaps with Abl . These findings support the idea that Abl and Dscam interact in presynaptic terminals in vivo . Do increased Dscam levels activate Abl kinase ? In mammals , autophosphorylation of Abl at tyrosines 245 and 412 ( Y245 and Y412 ) stabilizes the active conformation of the kinase ( Brasher and Van Etten , 2000; Tanis et al . , 2003 ) . As a result , phospho-specific antibodies raised against Y412 have been employed to detect active Abl kinases ( Brasher and Van Etten , 2000 ) . This approach has been used successfully to recognize the phosphorylation of the corresponding tyrosines ( Y539/522 ) in Drosophila as an assay for Abl kinase activation ( Stevens et al . , 2008 ) . Since the ability of Abl to instruct presynaptic terminal growth relies on Abl kinase activity , we tested whether Dscam activates Abl using a phosho-Y412-Abl ( p-Abl ) antibody . We found that Abl kinase activation was significantly increased ( 2 . 6 fold ) when Abl and Dscam were co-expressed in S2 cells ( Figure 3A ) . Furthermore , unlike wild-type Dscam , DscamΔCyto did not increase Abl kinase activation . In fact , it appears to act as a dominant-negative , as Abl activity was significantly decreased from control ( Figure 3A , right ) . As a negative control , no signal was detected when the kinase-dead Abl-K417N was blotted with p-Abl antibody in the same assay , suggesting that our assay specifically reported Abl activation ( Figure 3—figure supplement 1 ) . These results suggest that Dscam enhances Abl kinase activity . To investigate whether the same is true in presynaptic terminals in vivo , we devised a novel method of reporting Abl activation specifically in C4da presynaptic terminals . To achieve this , we used a previously reported probe that reports Abl activity , Pickles2 . 31 ( Mizutani et al . , 2010 ) . Pickles2 . 31 is composed of a fragment of a characteristic Abl substrate , CrkL , sandwiched between the fluorescent proteins Venus and enhanced CFP ( ECFP ) ( Figure 3B ) . It has previously been reported that activated Abl phosphorylates Pickles2 . 31 on the Y207 residue of the CrkL fragment , which can be detected with an antibody against CrkL-phospho-Y207 ( p-CrkL ) ( Mizutani et al . , 2010 ) . After expressing Pickles2 . 31 specifically in C4da neurons with the ppk-Gal4 driver , we dissected the larval CNS and immunoprecipitated Pickles2 . 31 from the lysates . Since the cell bodies of C4da neurons reside in the body wall , using only the larval CNS allowed us to monitor Pickles2 . 31 phosphorylation only in the C4da neuron presynaptic terminals ( Figure 3C ) . We found that overexpression of Dscam in C4da neurons led to an increase in Y207 phosphorylation of Pickles2 . 31 in the presynaptic terminals , while overexpression of DscamΔCyto was indistinguishable from control ( mCD8-mRFP ) ( Figure 3D ) . Consistent with the notion that Pickles2 . 31 is an Abl activity indicator , overexpression of BCR-Abl led to a robust increase in phospho-Y207 levels as compared to the control . These results suggest that Dscam activates Abl both in culture and in C4da presynaptic terminals in vivo , and that this activation requires the cytoplasmic domain of Dscam . 10 . 7554/eLife . 05196 . 009Figure 3 . Dscam activates Abl kinase in culture and in vivo . ( A ) Dscam activates Abl in cultured S2 cells . Abl activation was examined in S2 cell lysates transfected with indicated constructs by using anti-phospho-Y412-Abl antibody . The intensity of phospho-Abl was quantified , normalized to total Abl::Myc , and presented as bar graph ( n = 3 ) ( A , right ) . ( B ) Schematic of Pickles2 . 31 , an Abl activity reporter that uses phosphorylation of CrkL to report Abl kinase activity . Pickles2 . 31 is composed of a fragment of human CrkL that contains an Abl phosphorylation site , Y207 , sandwiched between ECFP and Venus . Phosphorylation of Pickles2 . 31 by Abl can be detected with an anti-phospho-Y207-CrkL ( p-CrkL ) antibody . ( C ) Schematic of in vivo assay for detecting Abl activity in C4da presynaptic terminals . Pickles2 . 31 is specifically expressed in C4da neurons . As can be appreciated from the larval fillet diagram ( left ) , the cell bodies and dendrites of C4da neurons reside in the larval body wall while their presynaptic terminals reside in the CNS . To assay Abl activity only in presynaptic terminals , larval CNS are dissected out and solubilized into lysates . Pickles2 . 31 in the presynaptic terminals is then immunoprecipitated with an anti-Venus antibody ( left ) . After running on an SDS-PAGE gel , Pickles2 . 31 expression level can be assayed using an anti-Venus antibody , while the phosphorylation of Y207 , a proxy for Abl activity level , can be ascertained by western blotting with a p-CrkL antibody . ( D ) Dscam activates Abl in presynaptic terminals in vivo . Overexpression of BCR-Abl leads to a robust increase in p-CrkL staining of Pickles2 . 31 when compared to the mCD8-mRFP control . Similarly , overexpression of Dscam leads to consistent , though less extreme , increase in p-CrkL when compared to control . In contrast , overexpression of DscamΔCyto is indistinguishable from the mCD8-mRFP control . This is a representative blot of three experimental repeats . DOI: http://dx . doi . org/10 . 7554/eLife . 05196 . 00910 . 7554/eLife . 05196 . 010Figure 3—figure supplement 1 . Phospho-Y412-Abl antibody specifically reports Abl activation . S2 cells were transfected with either Abl::Myc or Abl-K417N::Myc . Myc was blotted to report total Abl::Myc or Abl-K417N::Myc level ( middle ) , while phosho-Y412-Abl ( p-Abl ) was blotted to report Abl kinase activation ( top ) . While Abl::Myc displays a characteristic two-band pattern at the correct molecular weight when blotted for p-Abl , no signal is detected for Abl-K417N . This demonstrates that p-Abl specifically reports Abl activation . DOI: http://dx . doi . org/10 . 7554/eLife . 05196 . 010 These results raised the interesting possibility that targeting Abl might be a viable therapy for brain disorders caused by increased Dscam expression . Abl is a well-established target for treating chronic myeloid leukemia , and there are multiple Abl inhibitors that are approved by the US Food and Drug Administration ( FDA ) . As a proof-of-concept experiment , we attempted to rescue the developmental defects caused by Dscam overexpression using Abl inhibitors . We first tested nilotinib , which is a FDA-approved second-generation Abl kinase inhibitor that can cross the blood–brain barrier ( Weisberg et al . , 2005; Hebron et al . , 2013 ) . Using cultured S2 cells overexpressing Abl , we found that nilotinib robustly inhibited Drosophila Abl ( Figure 4A ) . Based on these results , we tested whether administration of nilotinib to developing larvae could rescue the effects of increased Dscam expression in C4da presynaptic terminals in vivo . To do this , we performed MARCM to visualize single C4da neurons in animals fed nilotinib or vehicle and assessed presynaptic terminal length . While overexpression of Dscam caused increased ( 152% ) presynaptic terminal length in animals fed vehicle ( Figure 4B–D ) , the effect was significantly rescued ( to 115% of control ) by feeding the animals with nilotinib ( Figure 4B , E ) . Consistent with the idea that these effects were due to inhibition of Abl activity rather than a reduction in Dscam expression , nilotinib did not change the expression of the Dscam transgene ( Figure 4—figure supplement 1A ) . 10 . 7554/eLife . 05196 . 011Figure 4 . Pharmacological inhibition of Abl mitigates the neuronal defects caused by increased Dscam expression in vivo . ( A ) Nilotinib inhibits Drosophila Abl kinase . S2 cells were transfected with either Myc-vector or Abl::Myc , and then treated with either vehicle ( DMSO ) or 5 μM nilotinib for 6 hr . Total lysates were subjected to western blot analysis with phospho-Y412-Abl ( p-Abl ) ( top ) and Myc antibodies ( bottom ) . ( B ) Quantification of the presynaptic terminal length of the indicated genotypes and drug treatment . Sample number is shown inside each bar . ( C–H ) Nilotinib treatment mitigates presynaptic arbor enlargement caused by Dscam overexpression ( OE Dscam , D and E ) and by dFMRP mutations ( dFMRPΔ50M , G and H ) . Nilotinib treatment alone does not affect presynaptic terminal growth ( F ) . The arrowhead in each panel points to the location where an axon elaborates the presynaptic terminal arbor . The MARCM technique was used to generate and visualize single presynaptic terminals of mutant C4da neurons . Drosophila larvae were raised in the presence of either 380 μM nilotinib or vehicle ( DMSO ) for 4 days before the analysis . Scale bar is 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05196 . 01110 . 7554/eLife . 05196 . 012Figure 4—figure supplement 1 . Nilotinib and bafatinib do not reduce Dscam transgene expression . Example images of C4da presynaptic terminals expressing Dscam::GFP in animals fed either vehicle ( A and B , top ) , 380 µM nilotinib ( A , bottom ) , or 125 µM bafetinib ( B , bottom ) throughout larval development . Images of mCD8::mRFP are shown to indicate the neuropil regions used for the quantifications ( white dotted line ) . Scale bar is 10 μm . Quantification of the fluorescence of the Dscam::GFP transgene in neuropil region is shown on the right . Sample number is shown inside each bar . DOI: http://dx . doi . org/10 . 7554/eLife . 05196 . 01210 . 7554/eLife . 05196 . 013Figure 4—figure supplement 2 . Nilotinib treatment does not cause defects in dendritic development or adult viability . ( A and B ) Nilotinib does not affect dendritic development . After egg collection , the animals were raised on food containing either vehicle ( DMSO ) or 380 µM nilotinib for 4 days . C4da dendrites were visualized by expressing mCD8::GFP with ppk-Gal4 ( A ) . Total dendritic length was measured , quantified , and presented in the bar graph ( B ) . Sample number is shown inside each bar . Scale bar is 50 µm . ( C and D ) Nilotinib does not affect the development of the flies . After egg collection , the animals were raised on food containing either vehicle ( DMSO ) or 380 µM nilotinib . Eclosed adults were counted on a daily basis . Total number and cumulative number of adults are shown in ( C ) and ( D ) respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 05196 . 01310 . 7554/eLife . 05196 . 014Figure 4—figure supplement 3 . Nilotinib and bafetinib act through Abl inhibition to mitigate Dscam-induced presynaptic arbor enlargement in vivo . The MARCM technique was used to generate and visualize single presynaptic terminals of mutant C4da neurons . Drosophila larvae were raised in the presence of 380 µM nilotinib , 125 µM bafetinib , or vehicle ( DMSO ) for 4 days before the analysis . Scale bar is 10 µm . ( A–D ) Nilotinib acts through Abl inhibition to mitigate presynaptic arbor enlargement in Dscam overexpressing neurons . Wt ( wild-type , FRT2A ) , OE Dscam ( overexpression of Dscam ) , OE Dscam , abl1 ( overexpression of Dscam in abl1 homozygous mutations ) . Note that nilotinib does not further decrease the size of presynaptic arbors in abl1 neurons overexpressing Dscam ( C and D ) . ( E and F ) Bafetinib mitigates presynaptic arbor enlargement in Dscam overexpressing neurons . ( G ) Quantification of the presynaptic terminal length of the indicated genotype and drug treatment . Sample number is shown below the x-axis . DOI: http://dx . doi . org/10 . 7554/eLife . 05196 . 014 Administration of nilotinib to developing larvae did not lead to adverse effects on overall development and neuronal growth . At the dose we used , nilotinib did not cause a change in presynaptic terminal growth ( Figure 4F ) or dendritic growth ( Figure 4—figure supplement 2A , B ) in wild-type larvae . Moreover , it did not impact the number of adults that eclosed or the dynamics of eclosion when compared to vehicle-fed flies ( Figure 4—figure supplement 2C , D ) . Although frequently used to inhibit pathological increases in Abl activity in patients , nilotinib is known to have several off-targets , including c-Kit , PDGFR , Arg , NQ02 , and DDR1 ( Hantschel et al . , 2008 ) . Consistent with the idea that nilotinib acts on Abl rather than on an off-target molecule to rescue presynaptic terminal growth , administering nilotinib to larvae overexpressing Dscam in abl1 clones did not lead to a further decrease in presynaptic terminal length when compared to vehicle-fed control ( Figure 4—figure supplement 3C , D , G ) . To further rule out the possibility that the observed rescue of presynaptic terminal length by nilotinib was the result of an off-target effect , we tested bafetinib , another Abl inhibitor with non-overlapping off-targets , Fyn and Lyn ( Kimura et al . , 2005 ) . Bafetinib has also been shown to cross the blood brain barrier ( Santos et al . , 2010 ) . Like nilotinib , administration of bafetinib to Dscam-overexpressing larvae led to a significant decrease in presynaptic terminal length ( Figure 4—figure supplement 3A , B , F , G ) without changing the expression of the Dscam transgene ( Figure 4—figure supplement 1B ) . Bafetinib alone did not change presynaptic terminal length in wild-type larvae when compared to wild-type larvae fed vehicle ( Figure 4—figure supplement 3E , G ) . Taken together , these results suggest that pharmacological inhibition of Abl mitigates the consequences of increased Dscam signaling in vivo . We next sought to test the efficacy of nilotinib treatment in a model of a disease associated with dysregulated Dscam expression , FXS . Thus , we tested whether administration of nilotinib could rescue the presynaptic overgrowth caused by increased Dscam expression in dFMRP mutants . We found that , while dFMRP mutants fed vehicle showed a significant increase ( 130% ) in presynaptic terminal length ( Figure 4B , G ) , administration of nilotinib to dFMRP mutants almost completely rescued ( to 103% of control ) the exuberant presynaptic terminal growth to wild-type levels ( Figure 4B , H ) . These results suggest that pharmacological inhibition of Abl kinase is effective for mitigating the effects of increased Dscam level in an in vivo model of FXS . In this study , we show that Dscam requires Abl to promote presynaptic terminal growth in vivo and that the binding of Abl to the Dscam cytoplasmic domain leads to Abl kinase activation . Furthermore , we show that treating larvae with Abl inhibitors rescues the developmental defects caused by increased Dscam levels in vivo in both Dscam-overexpressing neurons and disease-relevant models . Taken together , these results suggest that Abl is a potential drug target for the treatment of brain disorders associated with dysregulated Dscam expression , including DS and FXS . abl1 ( Gertler et al . , 1989 ) , abl4 ( Bennett and Hoffmann , 1992 ) , ppk-Gal4 ( Kuo et al . , 2005 ) , UAS-Dscam[3 . 36 . 25 . 2]::GFP ( Yu et al . , 2009 ) , UAS-Abl , UAS-BCR-Abl , UAS-Abl-K417N ( Wills et al . , 1999b ) , and dFMRPΔ50M ( Zhang et al . , 2001 ) were used in this study . To generate pUASTattB-Abl::Myc for expression in S2 cells , the coding region of Abl was recovered from UAS-Abl transgenic flies by PCR , subcloned into pUASTattB-Myc by using the InFusion cloning system following manufacturer's protocol ( Clontech , Mountain View , California ) . We generated pUASTattB-Abl-K417N::Myc by PCR mutagenesis as previously described ( O'Donnell and Bashaw , 2013 ) from pUASTattB-Abl::Myc . UAS-Dscam[3 . 36 . 25 . 2]::GFP was previously generated as described ( Kim et al . , 2013 ) . To generate UAS-DscamΔCyto , the Dscam coding region was digested with SstI and ligated with the GFP cDNA . Pickles2 . 31 was generously provided by Dr Yusuke Ohba at RIKEN Brain Science Institute ( Mizutani et al . , 2010 ) . To generate UAS-Pickles2 . 31 , the Pickles2 . 31 coding region was subcloned from pCAGGS-Pickles2 . 31 into pUASTattB using the InFusion cloning system following the manufacturer's protocol ( Clontech ) . Transgenic flies carrying UAS-DscamΔCyto , UAS-Abl::Myc , and UAS-Pickles2 . 31 were generated by germline transformation with support from BestGene , Inc . The MARCM technique was used to visualize single neurons homozygous for abl1 , abl4 , or dFMRPΔ50 , and overexpressing Dscam[3 . 36 . 25 . 2]::GFP as previously described ( Kim et al . , 2013 ) . Immunostaining of third-instar larvae was accomplished as previously described ( Ye et al . , 2011 ) . Antibodies used include chicken anti-GFP ( Aves , Tigard , Oregon ) and rabbit anti-RFP ( Rockland , Limerick , Pennsylvania ) . Samples were dehydrated and mounted with DPX mounting media ( Electron Microscopy Sciences , Hatfield , Pennsylvania ) . Confocal imaging was completed with a Leica SP5 confocal system equipped with a resonant scanner and 63× oil-immersion lens ( NA = 1 . 40 ) . Images were collected and quantified as previously described ( Kim et al . , 2013 ) . Drosophila S2 cells were maintained in Drosophila Schneider's medium supplemented with 10% fetal bovine serum at 25°C in a humidified chamber . Cells were transfected with indicated DNA constructs together with tubulin-Gal4 ( Lee and Luo , 2001 ) by using Lipofectamine 2000 ( Life Technologies , Grand Island , New York ) according to manufacturer's protocol . To perform co-immunoprecipitation , transfected S2 cells were harvested and lysed on ice with lysis buffer ( 50 mM Tris-HCl/pH 7 . 4 , 150 mM NaCl , 2 mM sodium vanadate , 10 mM sodium fluoride , 1% Triton X-100 , 10% glycerol , 10 mM imidazole and 0 . 5 mM phenylmethylsulfonyl fluoride ) . Lysates were centrifuged for 15 min at 20 , 000×g , 4°C and the resulting supernatant was incubated with Protein A/G PLUS-Agarose beads ( Santa Cruz Biotechnology , Paso Robles , California ) conjugated to mouse monoclonal anti-GFP clone 20 ( Sigma-Aldrich , St . Louis , Missouri ) for 4 hr at 4°C . After washing once with lysis buffer , twice with lysis buffer containing 0 . 1% deoxycholate , and 3 times with lysis buffer lacking Triton X-100 , the immunoprecipitates and total lysates were resolved on 7 . 5% SDS-PAGE gels followed by western blot analysis as previously described ( Kim et al . , 2013 ) . Primary antibodies used in western blotting were mouse monoclonal anti-tubulin ( Sigma ) , mouse anti-Myc ( Sigma-Aldrich ) , mouse monoclonal anti-Aequorea Victoria GFP JL-8 ( Clontech ) , and rabbit anti-phospho-Tyr412-c-Abl ( Cell Signaling , Beverly , Massachusetts ) . To assay in vivo Abl activation , UAS-Pickles2 . 31 was expressed specifically in C4da neurons using ppk-Gal4 along with other UAS transgenes . The CNS was dissected from third-instar larvae into ice-cold PBS with 2 mM sodium vanadate ( ∼100 per experimental condition ) . After a brief centrifugation , larval CNSs were transferred into lysis buffer as described above in immunoprecipitation and western blotting . Cells were disrupted using a pestle followed by brief sonication . Immunoprecipitation and western blotting of Pickles2 . 31 was then accomplished as described above . Primary antibodies used were rabbit anti-eGFP ( a gift from Dr Yang Hong ) and rabbit anti-phospho-Tyr 207-CrkL ( Cell Signaling ) . Nilotinib ( Abcam , United Kingdom ) and bafetinib ( ApexBio Technology , Houston , Texas ) were dissolved in dimethyl sulfoxide ( DMSO ) at 94 mM and 50 mM , respectively , as stock solutions before adding to S2 cells or fly food . S2 cells transfected with Abl::Myc were treated with either 5 μM nilotinib or the same volume of DMSO as a vehicle control for 6 hr before harvested and subjected to western blot analysis . Nilotinib and bafetinib were administered to larvae by rearing the larvae on standard corn meal food containing different concentrations of the drugs . The highest concentrations that did not affect overall larval development were used . Fly viability on nilotinib treatment was performed by counting the number of adult flies . Seven virgin female and seven male flies were crossed and transferred to standard corn meal food containing either 380 μM nilotinib or the same volume of DMSO ( 0 . 4% final concentration ) . Embryos were collected for 24 hr and allowed to develop . Eclosed adult flies were counted on a daily basis . The MARCM technique was used to generate and visualize mutant single C4da neurons as described above except that Drosophila embryos were collected and raised for 4 days on standard corn meal food containing either 380 μM nilotinib , 125 μM Bafetinib , or 0 . 4% DMSO . Sample preparation , imaging , and quantification were then completed as described above . Colocalization of Dscam and Abl was quantified with Manders' Correlation Coefficients using the Just Another Colocalization Plugin ( JACoP ) ( Bolte and Cordelieres , 2006 ) in ImageJ . Images were analyzed in three dimensions . Manders' Correlation Coefficients vary between 0 and 1 , with 0 representing no overlap between images and 1 representing complete colocalization . M1 and M2 describe the overlap of each channel with the other ( Bolte and Cordelieres , 2006 ) . M1 presents a measure of the fraction of Abl::Myc that overlaps Dscam ( ΔCyto ) ::GFP , while M2 presents a measure of the fraction of Dscam ( ΔCyto ) ::GFP that overlaps Abl::myc . Two-way student's t- test was used for statistical analysis . *: p < 0 . 05; **: p < 0 . 01; ***: p < 0 . 001; ****: p < 0 . 0001; ns: not significant .

Information is transmitted through the brain by cells called neurons , which are connected into specific circuits and networks . As the brain develops , several different signaling molecules control how the connections between neurons develop . If these signals occur at the wrong time or wrong place , or in the wrong amount , the neurons may not connect in the right way; this is the cause of several brain disorders . One of the signaling molecules that helps neural circuits to form in the developing brain is the Dscam protein . Having too much Dscam has been linked to neurons with enlarged presynaptic terminals . Presynaptic terminals are the parts of each neuron that send information on to the next cell , and when they are enlarged it results in the neuron not being able to communicate with other neurons in a targeted way . People with brain disorders including Down syndrome , epilepsy and possibly fragile X syndrome often have excessive amounts of Dscam . It was not known precisely how Dscam signals within neurons . Sterne , Kim and Ye have now investigated this by exploring the effects of Dscam on a group of well-known neurons in the larvae of the fruit fly species Drosophila . The presynaptic terminals of single neurons in this group were labeled in the larvae using a genetic marker . This revealed that the neurons of larvae that had been engineered to produce too much Dscam had larger presynaptic terminals than normal . Further investigation showed that for Dscam to influence how a presynaptic terminal grows , it must interact with another signaling protein called Abelson tyrosine kinase ( or Abl for short ) . Therefore , the larger presynaptic terminals seen in larvae that produce too much Dscam are a result of the Dscam protein activating too much Abl . There are several drugs that are approved for use in humans that suppress the activity of Abl . Sterne , Kim and Ye used two of these to treat fruit fly larvae , and found that they reversed the detrimental effects of extra Dscam on the larvae's neural circuit . Furthermore , the drugs fixed neural defects in a fruit fly model designed to reproduce the symptoms of fragile X syndrome . Overall , the results presented by Sterne , Kim and Ye suggest that suppressing the abnormally high activity of the Abl protein could be a way of treating the brain disorders caused by having excessive amounts of the Dscam protein . The next step is to test whether Dscam and Abl interact in the same way in mammals and whether the proposed treatment is effective in treating mammalian models of disorders that involve dysregulated Dscam signaling .

2015

Dysregulated Dscam levels act through Abelson tyrosine kinase to enlarge presynaptic arbors

The development and morphology of vascular plants is critically determined by synthesis and proper distribution of the phytohormone auxin . The directed cell-to-cell distribution of auxin is achieved through a system of auxin influx and efflux transporters . PIN-FORMED ( PIN ) proteins are proposed auxin efflux transporters , and auxin fluxes can seemingly be predicted based on the—in many cells—asymmetric plasma membrane distribution of PINs . Here , we show in a heterologous Xenopus oocyte system as well as in Arabidopsis thaliana inflorescence stems that PIN-mediated auxin transport is directly activated by D6 PROTEIN KINASE ( D6PK ) and PINOID ( PID ) /WAG kinases of the Arabidopsis AGCVIII kinase family . At the same time , we reveal that D6PKs and PID have differential phosphosite preferences . Our study suggests that PIN activation by protein kinases is a crucial component of auxin transport control that must be taken into account to understand auxin distribution within the plant . The synthesis and proper distribution of the hormone auxin within the growing plant body is essential for basically all differentiation processes throughout plant development as well as for the plant's tropic responses . As such , proper plant development and morphology strictly require the directed cell-to-cell transport of auxin , which is achieved by a system of auxin influx and efflux transporters ( Teale et al . , 2006 ) . AUXIN RESISTANT1 ( AUX1 ) /LIKE-AUX1 ( LAX ) proteins are auxin influx transporters and PIN-FORMED ( PIN ) proteins , which have been proposed to act in concert with ABC transporters , are the proposed auxin efflux transporters ( Galweiler et al . , 1998; Friml et al . , 2002; Noh et al . , 2003; Geisler et al . , 2005; Bainbridge et al . , 2008; Peret et al . , 2012 ) . The directed transport of auxin throughout the plant is critically determined by the—in many cells–asymmetric plasma membrane distribution of PINs and plant developmental processes have been successfully modeled based on the knowledge of PIN distribution and PIN protein behavior ( Jonsson et al . , 2006; Smith et al . , 2006; Wisniewska et al . , 2006; Blakeslee et al . , 2007; Grieneisen et al . , 2007 ) . We have previously identified and studied Arabidopsis protein kinases of the AGCVIII family designated D6 PROTEIN KINASE ( D6PK ) ( Zourelidou et al . , 2009 ) . The D6PK family is comprised of four functionally redundant members , namely D6PK , D6PK-LIKE1 ( D6PKL1 ) , D6PKL2 and D6PKL3 . Although D6PKs are devoid of any sequence features indicative for an association of these protein kinases with the plasma membrane , D6PKs colocalize with PIN proteins at the basal ( rootward ) plasma membrane in cells of the root cortex and stele , the hypocotyl and main inflorescence stem as well as the shoot apical meristem ( Zourelidou et al . , 2009; Barbosa et al . , 2014 ) . D6PKs phosphorylate PIN proteins in vitro and PIN phosphorylation is reduced in d6pk mutants in vivo without affecting PIN distribution or strongly affecting PIN abundance ( Zourelidou et al . , 2009; Willige et al . , 2013; Barbosa et al . , 2014 ) . Just as the PINs , D6PK constitutively cycles intracellularly between endosomal compartments and the plasma membrane but both , PINs and D6PK , traffic via distinct intracellular routes and seemingly encounter each other only at the basal plasma membrane ( Barbosa et al . , 2014 ) . Since PIN phosphorylation , as assessed by evaluating overall PIN1 and PIN3 phosphorylation levels , rapidly reacts to the presence and absence of D6PK at the plasma membrane , we postulated that D6PKs directly activate auxin transport by PIN phosphorylation ( Willige et al . , 2013; Barbosa et al . , 2014 ) . This hypothesis has , however , never been tested . Another subfamily of AGCVIII kinases comprises the proteins PINOID ( PID ) , WAG1 , and WAG2 ( Christensen et al . , 2000; Benjamins et al . , 2001; Santner and Watson , 2006; Galvan-Ampudia and Offringa , 2007 ) . Phosphorylation of PINs by PID/WAGs has previously been proposed to control PIN polarity ( Friml et al . , 2004; Michniewicz et al . , 2007; Dhonukshe et al . , 2010; Huang et al . , 2010 ) . PID/WAGs phosphorylate PINs at three highly conserved phosphosites , designated S1–S3 ( Dhonukshe et al . , 2010; Huang et al . , 2010 ) . Modulating PIN phosphorylation either by PID or WAG overexpression or by introducing phosphorylation-mimicking mutants in PIN1 seemingly results in a basal-to-apical shift in PIN polar distribution ( Michniewicz et al . , 2007; Dhonukshe et al . , 2010; Huang et al . , 2010 ) . The proposed loss of PIN phosphorylation in the pid mutant has been used to explain the phenotypic similarity between pin1 and pid mutants: pin1 mutants , on the one side , have a pin-formed inflorescence because they are devoid of the central auxin efflux protein required for shoot meristem differentiation ( Galweiler et al . , 1998 ) ; pid mutants , on the other side , are deficient in PIN1 phosphorylation , which seemingly prevents the essential basal-to-apical polarity switch required to redirect auxin fluxes during differentiation at the shoot meristem ( Friml et al . , 2004 ) . The PID/WAG-mediated repolarization of PIN proteins is also important for phototropic responses ( Ding et al . , 2011 ) . During phototropic bending of the hypocotyl , the polarity of the relevant PIN3 protein changes upon light exposure and this polarity switch is required for auxin redistribution in the hypocotyl and for efficient phototropism . This PIN3 polarity change requires the activity of PID/WAG protein kinases and it has been proposed that PID/WAG-dependent PIN3 phosphorylations directly control this process ( Ding et al . , 2011 ) . We showed previously that D6PKs also play a critical role in this process: d6pk mutants are strongly impaired in phototropic hypocotyl bending and the inability of d6pk mutants to efficiently transport auxin from the cotyledons to the hypocotyl may be responsible for this tropism defect ( Willige et al . , 2013 ) . Importantly , the light-induced and PID/WAG-dependent PIN3 polarity changes required for hypocotyl bending can still take place in the absence of D6PKs suggesting that the function of PID/WAGs in auxin transport and phototropism can be uncoupled from that of the D6PKs and that both kinases may control PINs independently and differentially ( Willige et al . , 2013 ) . While the differential biological function of D6PK and PID/WAGs in the context of phototropism may be explained by the two kinases being active in different tissues or during different stages of the phototropism response , there is also evidence that the two kinases have differential biochemical activities . While the overexpression of PID and WAG kinases results in a basal-to-apical PIN shift , the overexpression of D6PKs does not affect PIN distribution ( Zourelidou et al . , 2009; Dhonukshe et al . , 2010 ) . Inversely , the loss of PID function results in strong differentiation defects of the primary inflorescence , which are not apparent in the d6pk mutants . Thus , there is evidence for a differential biochemical activity of D6PKs and PID/WAGs but the molecular basis of this differential activity remains to be determined . The auxin efflux activity of PINs has previously been demonstrated by passive loading of yeast , plant , or mammalian cells with radiolabeled auxin ( Petrasek et al . , 2006; Wisniewska et al . , 2006; Mravec et al . , 2008; Yang and Murphy , 2009 ) . In these experiments , the auxin efflux activity of PINs was deduced from the reduced amount of radiolabeled auxin that accumulated in cells ( over- ) expressing certain PIN proteins in comparison to control samples . Because these experiments used passive loading of auxin , it is unclear if the differences in intracellular auxin accumulation observed in these experiments are truly a result of differences in auxin efflux or a consequence of differences in auxin uptake . In other studies , auxin efflux was shown based on differences in auxin retention after passive loading and subsequent transfer to auxin-free medium , thereby reversing the electrochemical gradient . In these studies , background transport activities could not be ruled out and differences became apparent only at endpoint steady-state levels . To date , there has been no report of a heterologous expression system that allows measuring auxin export directly in the linear phase . Here , we report the results from direct auxin efflux experiments with radiolabeled auxin ( indole-3-acetic acid , IAA ) injected into Xenopus oocytes . We find that PINs are unable to promote auxin efflux in this system unless PINs become activated by specific protein kinases of the Arabidopsis AGCVIII family . We map the phosphosites of these kinases in the PINs and further show that phosphorylation of conserved phosphosites is required for the efficient activation of PIN1 and PIN3 . Our study strongly suggests that the activation of PIN-mediated auxin efflux by protein kinases is a crucial component of auxin transport control that must be taken into account to understand auxin distribution within the plant . In Arabidopsis thaliana , the four AGCVIII kinases of the D6PK subfamily D6PK , D6PK-LIKE1 ( D6PKL1 ) , D6PKL2 and D6PKL3 redundantly control auxin transport-dependent growth ( Zourelidou et al . , 2009; Willige et al . , 2013 ) . Mutants with defects in multiple D6PK genes such as d6pk d6pkl1 ( d6pk01 ) double and d6pk d6pkl1 d6pkl2 ( d6pk012 ) triple mutants are severely impaired in several developmental processes including tropic responses ( d6pk01 and d6pk012 ) and lateral root differentiation ( d6pk012 ) ( Zourelidou et al . , 2009; Willige et al . , 2013 ) . In inflorescence stems , auxin is transported primarily in a basipetal ( rootward ) direction ( Teale et al . , 2006 ) . To understand the contribution of the individual D6PK genes to auxin transport in inflorescence stems , we measured basipetal auxin transport in primary inflorescence stems of a selected set of d6pk single , double and triple mutants that represented a previously established phenotypic series ( Zourelidou et al . , 2009; Willige et al . , 2013 ) . In these experiments , we noted a decrease in auxin transport in mutants of increased mutant complexity ( Figure 1 ) . While auxin transport defects were comparatively subtle in d6pk single mutants , the decrease in basipetal auxin transport was as strongly impaired in the d6pk012 triple mutant as in mutants of PIN1 , a major PIN protein in this tissue ( Figure 1 ) . Furthermore , we found that D6PKs are coexpressed with PINs in stems ( Figure 1—figure supplement 1 ) and that both , D6PK and PIN1 , localize to the basal plasma membrane in cells where auxin levels are high as suggested by the auxin response reporter DR5:GFP ( Figure 1—figure supplement 2 ) . Based on these observations , we concluded that D6PKs have an essential role in auxin transport regulation in inflorescence stems . 10 . 7554/eLife . 02860 . 003Figure 1 . Basipetal auxin transport is impaired in d6pk and pin1 mutants . ( A ) Basipetal auxin transport measured in inflorescence stems of 5-week-old Arabidopsis plants . Segment numbers refer to the 5 mm stem segments dissected from the primary inflorescence stem where segment 1 is the 5 mm segment closest to the radiolabeled auxin . The 5 mm segment directly in contact with the radiolabeled auxin is not included . Mutant nomenclature: d6pk0 , d6pk-1; d6pk1 , d6pkl1-1; d6pk01 , d6pk-1 d6pkl1; d6pk012 , d6pk-1 d6pkl1 d6pkl2-2 . A linear mixed-effects model analysis ( fixed factor ) revealed statistically significant differences ( p<0 . 01 ) in the transport rates between the wild type and all mutant genotypes , between the d6pk single mutants and the higher order d6pk mutants as well as between the d6pk01 double mutant and the d6pk012 triple mutant . d6pk012 and pin1 are not significantly different ( p=0 . 43 ) . ( B ) Amount of radiolabeled auxin found in all segments of the plants shown in ( A ) . An ANOVA revealed highly significant differences between groups ( p<0 . 001 ) . An all-pairwise post hoc analysis ( Holm-Sidak ) allowed the assignment of three significance levels indicated by letters ( p≤0 . 05 between levels ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02860 . 00310 . 7554/eLife . 02860 . 004Figure 1—figure supplement 1 . D6PKs and PINs are coexpressed in vascular bundles of inflorescence stems . Representative GUS-reporter stainings of promoter GUS-fusions of the four D6PK genes and PIN1 , PIN3 , PIN4 and PIN7 from inflorescence stem sections of 5-week-old plants . Scale bar = 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 02860 . 00410 . 7554/eLife . 02860 . 005Figure 1—figure supplement 2 . D6PK and PIN1 localize to the basal plasma membrane in xylem parenchyma cells . Representative confocal images of sectioned inflorescence stems expressing PIN1p:PIN1:YFP ( PIN1:YFP ) , D6PKp:YFP:D6PK ( YFP:D6PK ) and the auxin response reporter DR5:GFP . Shown are fluorescent images , bright field images , and the overlay of the two images . Arrowheads mark the accumulation of the translational fusions at the basal plasma membrane . Please note that the auxin response reporter is not expected to show polar distribution in these tissues . xpc , xylem parenchyma cell; v , xylem vessel . Scale bar = 30 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 02860 . 005 Since auxin transport is impaired in d6pk mutant inflorescence stems and since we had previously accumulated evidence that D6PK directly phosphorylates PINs ( Zourelidou et al . , 2009; Willige et al . , 2013; Barbosa et al . , 2014 ) , we hypothesized that D6PK may directly activate auxin transport by PIN phosphorylation in vivo . To test this hypothesis , we established a heterologous test system for measuring auxin efflux using Xenopus laevis oocytes . In this assay , in vitro transcribed cRNAs for the proteins under investigation were injected into the oocytes 4 days prior to the experiment to allow for protein synthesis . At the beginning of the experiment , radiolabeled IAA was injected and the amount of residual radiolabel was measured in the oocytes after incubation for up to 30 min . PIN as well as D6PK protein accumulated at the plasma membrane also in oocytes as shown by immunoblots for PINs and confocal microscopy for D6PK ( Figure 2A , B ) . An inherent feature of this assay system was the gradual loss of the injected radiolabeled IAA from the oocytes over time–in the absence of exogenous proteins–which we attributed to the leakiness of the plasma membrane for IAA ( Figure 2C–F ) . Interestingly , when we tested PIN1 or PIN3 alone , we did not observe any measurable auxin efflux that differed from the background , suggesting that the PINs are inactive auxin transporters in the oocyte system . However , when we co-expressed D6PK with the PINs we observed a significant and kinase activity-dependent activation of auxin efflux . This activation correlated with the appearance of high molecular weight bands for PIN1 and PIN3 that appeared in anti-PIN immunoblots only in the presence of the active D6PK kinase ( Figure 2B ) . In line with an activation of PINs by D6PK through direct PIN phosphorylation , a kinase-dead variant of D6PK could not activate auxin efflux in this system ( Figure 2D , F ) . In summary , these experiments showed that D6PK is an activator of PIN-mediated auxin efflux in the oocyte expression system . 10 . 7554/eLife . 02860 . 006Figure 2 . D6PK activates PIN-mediated auxin efflux in Xenopus oocytes . ( A ) Representative confocal microscopy images of oocytes expressing YFP:D6PK ( D6PK ) and YFP:D6PKin ( D6PKin ) reveals localization of the proteins at the plasma membrane . ( B ) Anti-PIN immunoblots of protein extracts from microsomal membrane ( MF ) fractions ( and where applicable cytoplasmic fractions [CF] ) from oocytes expressing PIN1 , PIN3 , YFP:D6PK ( D6PK ) and kinase-dead YFP:D6PK ( D6PKin ) . ( C ) – ( F ) . Results of representative auxin efflux assays conducted in Xenopus oocytes expressing PIN1 , PIN3 , YFP:D6PK ( D6PK ) and kinase-dead YFP:D6PKin as specified . Each data point represents the mean and standard error of at least 10 oocytes . DOI: http://dx . doi . org/10 . 7554/eLife . 02860 . 006 Using mass spectrometry , we next identified D6PK-dependent phosphosites in the PINs after in vitro phosphorylation of the cytoplasmic loops ( CL ) of PIN1 , PIN2 , PIN3 and PIN4 . These analyses resulted in the identification of two novel serine residues as conserved PIN phosphosites , S4 and S5 , as well as three further serine phosphosites , S1–S3 , that had previously been identified as phosphosites of the PID/WAG kinases ( Figure 3A; Figure 3—figure supplement 1; Figure 3—source data 1; Dhonukshe et al . , 2010; Huang et al . , 2010 ) . Whereas S1 , S2 and S3 are conserved in all four PINs tested , S4 and S5 are not conserved in PIN2 where the corresponding protein sequence motifs are divergent when compared to PIN1 , PIN3 , PIN4 and PIN7 and when compared to the strong conservation of the S1–S3 phosphosites in all PINs including PIN2 ( Figure 3A ) . Furthermore , S5 was not conserved in PIN1 but aligned with a strongly conserved region of PIN1 . At the position of S5 , PIN1 had an aspartic acid ( D; D215 ) and we speculated that D215 might be a natural phosphomimic variant of the S5 site ( Figure 3A ) . 10 . 7554/eLife . 02860 . 007Figure 3 . PIN S4 and S5 are phosphorylated by D6PK . ( A ) Sequence alignment of PIN cytoplasmic loop fragments indicating the PIN phosphosites identified after in vitro phosphorylation with D6PK . ( B ) Results of in vitro phosphorylation experiments with synthetic wild type and mutant peptides confirm the D6PK-dependent phosphorylation of sites corresponding to S4 ( right panels ) and S5 ( left panels ) in the PINs where the respective sites are conserved . PIN3 and PIN7 are sequence identical at the S4 phosphosite . Each reaction was spotted in duplicate . Amino acid sequences of the respective wild type and mutant peptides are shown on the right of each panel , their peptide identification numbers are shown in the upper left corner ( Supplementary file 1B ) . The amino acid exchange in the respective peptide pair is shown in bold typeface . The N- and C-terminal amino acids Y–A and K–K were added to allow for peptide quantification after synthesis and to facilitate attachment of the peptide to the negatively charged P81 paper . DOI: http://dx . doi . org/10 . 7554/eLife . 02860 . 00710 . 7554/eLife . 02860 . 008Figure 3—source data 1 . Mass spectrometric analyses of PIN cytoplasmic loop phosphorylation by D6PK . List of peptides including the phosphorylation state ( PO3H2 ) of the respectively modified amino acid after phosphorylation of the recombinant PIN1–PIN4 cytoplasmic loops ( CL ) by D6PK . For each peptide identified from the PIN CLs , mass , charge state , and SEQUEST scores ( cross-correlation values , X-corr , and Delta Correlation values , DeltaCN ) are listed . For reasons of clarity , posttranslational modifications other than phosphorylation that were considered for peptide identification and mass calculation are not shown . DOI: http://dx . doi . org/10 . 7554/eLife . 02860 . 00810 . 7554/eLife . 02860 . 009Figure 3—figure supplement 1 . Summary of the mass spectrometric analyses of PIN cytoplasmic loop phosphorylation by D6PK . PIN cytoplasmic loop sequences used for the in vitro phosphorylation analyses . Peptides identified by mass spectrometry as listed in Figure 3—source data 1 are presented in color . Different colors are used to distinguish neighboring peptides . Phosphorylated residues as identified in this analysis are marked in bold . The conserved S1–S5 phosphosites are marked in bold and underlined . Please note that phosphosites other than S1–S5 as detected in PIN1 and PIN3 are not conserved among the different PINs and were therefore not considered in our analyses . DOI: http://dx . doi . org/10 . 7554/eLife . 02860 . 009 We next tested the identity and relevance of S4 and S5 in in vitro phosphorylation experiments using synthetic peptides as well as recombinant PIN CL fragments as substrates ( Figure 3B ) . In the experiments with the synthetic peptides , we could confirm the identity and phosphorylation of the novel S4 and S5 phosphosites using mutant peptides as negative controls where the respective S had been replaced by an alanine ( A ) ( Figure 3B ) . Since S5 from PIN3 , PIN4 and PIN7 corresponded to D215 in PIN1 and since D215 was embedded in an otherwise highly conserved part of the protein , we were also interested in testing whether a serine ( S ) in a PIN1 D215S variant could be phosphorylated by D6PK . Indeed , while a synthetic peptide comprising PIN1 D215 could not be phosphorylated by D6PK in vitro , the D215S peptide variant was efficiently phosphorylated indicating that , although the respective S5 phosphosite was not conserved , the sequence conservation in this region was sufficient for phosphorylation by D6PK . This was suggestive for an overall structural conservation of this PIN1 protein domain ( Figure 3A ) . In contrast , PIN2-specific peptides corresponding to the S4 or S5 phosphosites could not be phosphorylated by D6PK despite the fact that their sequences also contained serine residues . Phosphorylation of the corresponding peptides also failed when an asparagine ( N ) at the respective position was replaced by a serine ( Figure 3B ) . Thus , S4 and S5 are novel PIN protein phosphosites that are differentially conserved in the five plasma membrane-resident PIN proteins with a role in promoting auxin efflux . When we examined the contribution of the individual phosphosites to PIN1 phosphorylation in the context of the PIN1 cytoplasmic loop ( CL ) fragment , we found that PIN1 CL phosphorylation by D6PK was already strongly reduced ( 40% of wild type levels ) in a mutant variant where only PIN1 S4 was replaced by an alanine ( S4A; Figure 4A ) . In turn , mutations of the phosphosites PIN1 S1 , PIN1 S2 or PIN1 S3 alone impaired phosphorylation by D6PK to a lesser extent ( ca . 80% ) and only mutation of all three sites in PIN1 S1A S2A S3A led to a clear reduction of PIN1 CL phosphorylation ( 58%; Figure 4A ) . Finally , the mutation of all four PIN1 phosphosites under investigation in PIN1 S1A S2A S3A S4A abolished phosphorylation by D6PK almost completely ( 2 . 7%; Figure 4A ) . Based on these analyses , we concluded that S4 is a major phosphosite for D6PK in PIN1 . 10 . 7554/eLife . 02860 . 010Figure 4 . In vitro phosphorylation of PIN1 and PIN3 . ( A ) – ( D ) Representative experiments with recombinant purified GST:D6PK ( D6PK ) or GST:PID ( PID ) and wild type or mutant PIN1 ( A and B ) or PIN3 ( C and D ) CL fragments in the presence of radiolabeled [©-32P]ATP . AR , autoradiography; CBB , Coomassie Brilliant Blue-stained gel , loading control . Percentage values represent the amount of radiolabel incorporated into the PIN1 ( A ) and ( B ) and PIN3 ( C ) and ( D ) mutant proteins relative to the respective wild type protein after normalization to the loading control . Asterisks mark non-specific background bands or degradation products . DOI: http://dx . doi . org/10 . 7554/eLife . 02860 . 010 Since PIN1 S1 , S2 , and S3 had previously been identified as phosphorylation targets of PID , we also examined and quantitatively compared the effects of the phosphosite mutations with those of D6PK . In the case of PID , the phosphorylation of PIN1 CL by PID was not altered in the PIN1 S4A mutant when compared to the wild type ( 100%; 40% for D6PK ) but already strongly affected by the PIN1 S1A mutation ( 61%; ca . 80% for D6PK ) and even more by PIN1 S1A S2A S3A ( 18%; 58% for D6PK; Figure 4B ) . Thus , D6PK and PID have an overlapping but also differential preference for specific phosphosites in PIN1 . When we examined the effects of S4 and S5 site mutations in the context of PIN3 , we detected a similar phosphosite preference . Whereas a PIN3 CL S4A S5A variant was still efficiently phosphorylated by PID its phosphorylation by D6PK was severely impaired ( 28%; Figure 4C , D ) . Thus , mutations of the five phosphosites have differential effects in the case of D6PK or PID . We next evaluated the importance of S1–S3 and S4 for PIN1- and D6PK-dependent auxin efflux in oocytes . For this purpose , we calculated the transport rates of PIN1 and the S to A mutants as described in Figure 5—figure supplement 1 . In line with the proposed important role of S4 for PIN1 phosphorylation , we found that a PIN1 S4A mutant was already significantly impaired in auxin efflux activation by D6PK in the auxin efflux experiments ( Figure 5A , B ) . At the same time , the requirement of PIN1 S1 , S2 , and S3 for D6PK activation was not obvious with a PIN1 S1A S2A S3A mutant but became apparent in the presence of the S4A mutation where the PIN1 activation defect of the S4A mutation was further enhanced in the presence of mutations of the other three sites ( Figure 5A ) . We thus concluded that PIN1 S4 is an important site for D6PK-dependent PIN1 activation but that all four phosphosites are required for full PIN1 activation . Also , in line with the results obtained in the in vitro phosphorylation experiments , we found that a PIN3 S4A S5A variant showed reduced responsiveness to D6PK when compared to wild type PIN3 providing further support for the importance of the S4 and S5 phosphosites for PIN activation by D6PK ( Figure 5C ) . 10 . 7554/eLife . 02860 . 011Figure 5 . D6PK activates auxin transport through phosphorylation of specific serine residues . ( A ) Results of quantitative analyses from oocyte auxin efflux assays with D6PK and wild type or mutant PIN1 . The averages of at least three independent measurements are shown after normalization to the mock control . Student's t test: *p=0 . 022; **p=0 . 005; ***p<0 . 001; n . s . , not significant . ( B ) Anti-PIN1 immunoblots of microsomal membrane ( MF ) and cytoplasmic fractions ( CF ) of the corresponding oocytes used in ( A ) . ( C ) PIN3 S4 S5 are required for full activation by D6PK . Results of quantitative analyses from oocyte auxin transport assays with D6PK and wild type PIN3 or the PIN3 S4A S5A mutant . The averages of at least three independent biological replicates are shown after normalization to the mock control . Student's t test * , p=0 . 016; n . s . , not significant . ( D ) PIN1 D215 does not contribute to the auxin transport activity of PIN1 . Results of oocyte auxin efflux assays with wild type and mutant PIN1 together with YFP:D6PK ( D6PK ) as specified . Each data point represents the mean and standard error of measurements from at least 10 oocytes . DOI: http://dx . doi . org/10 . 7554/eLife . 02860 . 01110 . 7554/eLife . 02860 . 012Figure 5—figure supplement 1 . Quantification of auxin efflux in Xenopus oocytes . Graph of a typical auxin transport experiment conducted in oocytes . Values represent the average of the measurements of at least 10 individual oocytes per time point . Linear regression was performed to calculate the relative efflux rate , that is the change in concentration over time . This rate ( n = 1 ) corresponds to one biological replicate . The bar graphs in Figures 5A , C and 8A , C were calculated based on the efflux rates of at least three biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 02860 . 012 Since S5 corresponded to an aspartic acid residue in PIN1 ( D215 ) and because we could demonstrate that a peptide with a D215S replacement was efficiently phosphorylated by D6PK ( Figure 3 ) , we speculated that D215 might be a natural phosphomimic variant of the S5 phosphosite . We reasoned that PIN1 D215S might show a differential behavior in the auxin efflux experiments in the absence and presence of D6PK because the D215S mutant variant could show a stronger dependency on kinase activation . However , we found that the auxin efflux ( activation ) of the wild type PIN1 protein was indistinguishable from the behavior of the PIN1 D215S mutant in these oocyte experiments ( Figure 5D ) . We therefore rejected this hypothesis . To examine the biological significance of S4 and S5 for PIN function , we introduced wild type and mutant transgenes for the expression of PIN1 and PIN3 under the control of their respective promoters into pin1 and pin3 pin4 pin7 ( pin347 ) mutants , respectively ( Figure 6 ) . In support of an important but not exclusive role of S4 phosphorylation for PIN1 function , we detected only a partial rescue of the auxin transport defect in inflorescence stems of pin1 mutants transformed with PIN1 S4A compared to a full rescue with the wild type PIN1 . While the mutant and the wild type transgene were able to complement the PID-dependent inflorescence differentiation defect of the pin1 mutant ( Figure 6C , D ) , D6PK-dependent basipetal auxin transport in the stem was compromised ( Figure 6A , B ) . Since the mutation of the S4 phosphosite may potentially interfere with the polar distribution or the intracellular transport of the constantly trafficking PIN1 protein , we analyzed the polar distribution of PIN1 S4A and its sensitivity to the trafficking inhibitor Brefeldin A ( BFA ) ( Figure 6—figure supplement 1 ) . Since PIN1 S4A showed an identical behavior to wild type PIN1 in these experiments , we concluded that changes in PIN1 polarity , PIN1 trafficking or PIN1 abundance at the plasma membrane may not be causal for the observed differences in basipetal auxin transport . We also evaluated the effects of PIN3 phosphosite mutations using the ability of PIN3 transgenes to complement the strong phototropism defect of the pin3 pin4 pin7 ( pin347 ) triple mutant ( Figure 6E , F; Willige et al . , 2013 ) . When we measured the ability of wild type PIN3 and mutant PIN3 S4A S5A to complement the pin347 mutant when expressed from a PIN3 promoter fragment , we found that the phototropism defect of the pin347 mutant was only partially complemented by the PIN3 S4A S5A transgene while it was fully complemented by wild type PIN3 ( Figure 6E ) . This finding was in line with the hypothesis that D6PK-dependent PIN3 S4 and S5 phosphorylations are required for efficient basipetal auxin transport in the hypocotyls of dark-grown seedlings , which is a prerequisite for efficient hypocotyl bending . Consistent with the predominant role of the PID phosphosite phosphorylation at S1–S3 , we found that the mutation of the PIN3 S1–S3 phosphosites as well as mutation of all five PIN3 phosphosites , S1–S5 , fully impaired the ability of the PIN3 transgene to complement the pin347 mutation ( Figure 6F ) . This finding can be explained by the essential role of PID-dependent PIN3 polarity changes in the hypocotyl that take place after light exposure and that are required for phototropic hypocotyl bending . As we had previously shown , the PID-dependent PIN3 polarity change after phototropic stimulation is a distinct process that is independent from the regulation of basipetal auxin transport in the dark-grown seedling ( Ding et al . , 2011; Willige et al . , 2013 ) . In summary , this experiment supported the conclusion that the novel phosphosites , PIN3 S4 and S5 , are required for full PIN3 activity , most likely by interfering primarily with basipetal auxin transport in the hypocotyls of dark-grown seedlings . 10 . 7554/eLife . 02860 . 013Figure 6 . PIN1 S4 and PIN3 S4 S5 are required for full pin mutant complementation . ( A ) Basipetal auxin transport measured in inflorescence stems of 5-week-old Arabidopsis plants . Segment numbers refer to the 5 mm stem segments dissected from the inflorescence stem where segment 1 is the 5 mm segment closest to the radiolabeled auxin . The 5 mm segment directly in contact with the radiolabeled auxin was discarded . The values represent the mean and standard error of six biological replicates , except pin1 and NPA-treated wild type ( n = 2 ) . A linear mixed-effects model analysis ( fixed factor ) revealed statistically significant differences ( p<0 . 05 ) in the transport rates between the plant lines complemented with the PIN1 S4A construct and the other genotypes as indicated by the significance levels in ( B ) . ( B ) Amount of radiolabeled auxin found in all segments of the plants shown in ( A ) . An ANOVA revealed highly significant differences between groups ( p<0 . 001 ) . An all pairwise post hoc analysis ( Holm-Sidak ) allowed the assignment of three significance levels indicated by letters ( p≤0 . 036 between levels ) . ( C ) Phenotypes of 5-week-old pin1 mutants complemented with a transgenic construct expressing wild type PIN1 and PIN1 S4A under control of a PIN1 promoter fragment . Scale bar = 10 cm . ( D ) PIN1 immunoblot detects comparable PIN1 protein levels between the wild type and PIN1 transgenic lines . ( E ) and ( F ) Analysis for the rescue of phototropic hypocotyl bending defects of a pin3 pin4 pin7 mutant carrying wild type and mutant transgenes for the expression of wild type and mutant PIN3 under control of a PIN3 promoter fragment . Seedlings were exposed for 6 hr ( E ) or 20 hr ( F ) to unilateral white light before quantification . To assess differences between genotypes a Kruskal–Wallis ANOVA on ranks was performed . The differences in the median values among the different genotypic groups was highly significant ( p<0 . 001 ) . Different letters in indicate different significance levels ( p<0 . 01 ) calculated by an all-pairwise multiple comparison ( Dunn's Method ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02860 . 01310 . 7554/eLife . 02860 . 014Figure 6—figure supplement 1 . BFA-sensitivity of PIN1 and PIN1 S4A . Confocal images of root cells from 6 day-old seedlings examining the localization of YFP-tagged PIN1 and PIN1 S4A in a time course experiment ( 15 min and 60 min ) without ( mock ) and with a BFA [50 μM] treatment . FM4-64 [2 μM] was used as an endocytosis marker and included in all treatments . Shown are the fluorescent images from the YFP and the FM4-64 channels as well as the merged images . The open arrowheads mark the polarly localized PIN1 proteins , the filled arrowheads point at selected BFA-compartments that are detected with both fluorescent proteins after BFA-treatment . To ensure that stele cells are stained at the onset of the treatment , seedlings were stained with FM4-64 for 5 min before BFA-treatment . Scale bar = 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 02860 . 014 Next , we were interested in examining the phosphorylation at PIN1 S4 and PIN3 S4 and S5 in vivo and to examine the phosphorylation at these sites in the presence and absence of D6PKs . To this end , we employed selected reaction monitoring ( SRM ) , a mass spectrometry technique that allows detection and quantification of specific peptides and their phosphorylated variants in total protein preparations ( Picotti and Aebersold , 2012 ) . In these experiments , we detected a strong reduction in the in vivo abundance of the PIN1 S4 as well as PIN3 S4 phosphorylations that increased with increasing d6pk mutant complexity ( Figure 7A , B ) . This decrease in S4 phosphorylation could not be explained by changes in the overall abundance of PIN proteins as shown by quantitative SRM analyses of the unphosphorylated PIN1 and PIN3 S4 peptides and analyses of internal control peptides ( Figure 7A , B ) . Furthermore , introducing a D6PK transgene expressing D6PK under control of a D6PK promoter fragment rescued the PIN1 and PIN3 S4 phosphorylation defects ( Figure 7—figure supplements 1 and 2 ) . We also examined phosphorylation at PIN3 S5 using the same methodology and observed that the abundance of phosphorylation at these sites was as strongly reduced in the d6pk012 triple mutant as observed for the S4 site . Again , the phosphorylation defect could not be explained by changes in PIN3 abundance and was rescued by a D6PK transgene as described above ( Figure 7—figure supplement 3 ) . Most importantly , the observed decreases in PIN1 and PIN3 phosphorylation were in good agreement with the reductions in auxin transport that we had detected in the same tissue of d6pk mutants ( Figure 1 ) . We therefore concluded that D6PKs are the major kinases targeting PIN1 S4 , PIN3 S4 , and PIN3 S5 in Arabidopsis inflorescence stems and that the reduced phosphorylation at these sites may be causal for the reduced auxin transport of d6pk mutants in this tissue . 10 . 7554/eLife . 02860 . 015Figure 7 . In vivo phosphorylation of PIN1 and PIN3 . ( A ) and ( B ) Results of SRM analysis for the quantification of PIN1 S4 and PIN3 S4 phosphorylation in inflorescence tissue of 5-week-old Arabidopsis wild type and d6pk mutant plants ( PIN1_pS4 , phosphorylated form of PIN1 S4; PIN1_S4 , unphosphorylated form etc ) . Internal peptides allow for an estimation of the overall PIN protein levels , standard deviations were calculated based on average variation of standard peptide abundances across all samples , columns represent the sum of individual fragment ions that are shown in different gray scales . DOI: http://dx . doi . org/10 . 7554/eLife . 02860 . 01510 . 7554/eLife . 02860 . 016Figure 7—figure supplement 1 . Auxin-dependent phosphorylation at PIN1 S4 . Quantification of PIN1 S4 phosphorylation in IAA-treated [10 µM; 1 hr] inflorescence stem tissue of 5-week-old Arabidopsis plants . PIN1_pS4 , phosphorylated form; PIN1_S4 , unphosphorylated form . Internal peptides allow for quantification of the overall protein levels . DOI: http://dx . doi . org/10 . 7554/eLife . 02860 . 01610 . 7554/eLife . 02860 . 017Figure 7—figure supplement 2 . Auxin-dependent phosphorylation at PIN3 S4 . Quantification of PIN3 S4 phosphorylation in IAA-treated [10 µM; 1 hr] inflorescence stem tissue of 5-week-old Arabidopsis plants . PIN3_pS4 , phosphorylated form; PIN1_S4 , unphosphorylated form . Internal peptides allow for quantification of the overall protein levels . DOI: http://dx . doi . org/10 . 7554/eLife . 02860 . 01710 . 7554/eLife . 02860 . 018Figure 7—figure supplement 3 . Auxin-dependent phosphorylation at PIN3 S5 . Quantification of PIN3 S5 phosphorylation in IAA-treated [10 µM; 1 hr] inflorescence stem tissue of 5-week-old Arabidopsis plants ( PIN3_pS5 , phosphorylated form; PIN3_S5 , unphosphorylated . Internal peptides controls , shown in Figure 7—figure supplement 2 allow for quantification of the overall protein levels . DOI: http://dx . doi . org/10 . 7554/eLife . 02860 . 018 D6PKs belong to a larger family of AGCVIII kinases in Arabidopsis ( Galvan-Ampudia and Offringa , 2007 ) . Besides D6PKs and the already introduced PID/WAGs , other AGCVIII kinases such as the phototropin blue light receptors phot1 and phot2 as well as UNICORN ( UCN ) have known biological functions ( Inoue et al . , 2008; Enugutti et al . , 2012 ) . We were interested in testing the ability of these protein kinases to activate PIN-mediated auxin efflux and examined PID , WAG2 as well as phot1 and UCN together with PIN1 in the oocyte auxin transport assay ( Figure 8 ) . Interestingly , PID and WAG2 but not phot1 or UCN were able to activate PIN1-mediated auxin efflux ( Figure 8A , B , Figure 8—figure supplement 1 ) . We thus concluded that PID and WAG2 have a role in PIN activation besides their previously reported role in the control of PIN polarity ( Friml et al . , 2004; Dhonukshe et al . , 2010 ) . 10 . 7554/eLife . 02860 . 019Figure 8 . Capability of various AGCVIII kinases to active PIN1-mediated auxin efflux . ( A ) and ( C ) Results of quantitative auxin efflux assays performed in the oocyte system with PIN1 and different AGCVIII kinases ( A ) or mutant PIN1 and PID ( B ) . The averages of at least three independent measurements , calculated as described in Figure 5—figure supplement 1 , are shown after normalization to the mock control . In ( A ) , a one-way ANOVA revealed high differences between groups ( p<0 . 001 ) and a post hoc analysis ( Holm-Sidak ) indicated that the D6PK and PID values were significantly different from control oocytes ( ***p<0 . 001 ) . In ( C ) , a Student's t-test was performed: *p<0 . 027; ***p<0 . 001; n . s . , not significant . ( B ) Immunoblots of total protein extracts prepared from oocytes expressing PIN1 and different AGC kinases . The presence and activation ( phot1 only ) of the non-effective kinases in the membrane ( MF ) and cytoplasmic fraction ( CF ) was confirmed with anti-phot1 , anti-phot1-pS851 ( for phot1 activation ) and anti-UCN . DOI: http://dx . doi . org/10 . 7554/eLife . 02860 . 01910 . 7554/eLife . 02860 . 020Figure 8—figure supplement 1 . WAG2 activates PIN1-mediated auxin transport . ( A ) Results of representative auxin efflux assays conducted in Xenopus oocytes expressing wild type PIN1 alone or together with WAG2 as specified . Each data point represents the mean and standard error of at least 10 oocytes . ( B ) Immunoblots of total protein extracts prepared from oocytes expressing PIN1 without and with WAG2 . DOI: http://dx . doi . org/10 . 7554/eLife . 02860 . 02010 . 7554/eLife . 02860 . 021Figure 8—figure supplement 2 . S4 phosphorylation in PIN1 is not strongly reduced in pid and wag1 wag2 mutants . Results of SRM analysis for the quantification of PIN1 S4 phosphorylation in inflorescence tissue of 5-week-old Arabidopsis wild type and mutant plants . PIN1_pS4 , phosphorylated form of PIN1 S4; PIN1_S4 , unphosphorylated form . Internal peptides allow for quantification of the overall PIN1 protein levels , standard deviations were calculated based on the average variation of standard peptide abundances , columns represent the sum of individual fragment ions shown in different gray scales . DOI: http://dx . doi . org/10 . 7554/eLife . 02860 . 02110 . 7554/eLife . 02860 . 022Figure 8—figure supplement 3 . S4 phosphorylation in PIN3 is also not strongly reduced in pid and wag1 wag2 mutants . Results of SRM analysis for the quantification of PIN3 S4 phosphorylation in inflorescence tissue of 5-week-old Arabidopsis wild type and mutant plants . PIN3_pS4 , phosphorylated form of PIN3 S4; PIN3_S4 , unphosphorylated form . Internal peptides allow for quantification of the overall PIN3 protein levels , standard deviations were calculated based on the average variation of standard peptide abundances , columns represent the sum of individual fragment ions shown in different gray scales . DOI: http://dx . doi . org/10 . 7554/eLife . 02860 . 02210 . 7554/eLife . 02860 . 023Figure 8—figure supplement 4 . Auxin-dependent phosphorylation at PIN1 S1 . Quantification of PIN1 S1 phosphorylation in IAA-treated [10 µM; 1 hr] inflorescence stem tissue of 5-week-old Arabidopsis plants . PIN1_pS1 , phosphorylated form; PIN1_S1 , unphosphorylated form . Internal peptides controls , shown in Figure 7—figure supplement 2 allow for quantification of the overall protein levels . DOI: http://dx . doi . org/10 . 7554/eLife . 02860 . 02310 . 7554/eLife . 02860 . 024Figure 8—figure supplement 5 . Auxin-dependent phosphorylation at PIN3 S1 . Quantification of PIN3 S1 phosphorylation in IAA-treated [10 µM; 1 hr] inflorescence stem tissue of 5-week-old Arabidopsis plants . PIN3_pS1 , phosphorylated form; PIN3_S1 , unphosphorylated form . Internal peptides controls , shown in Figure 7—figure supplement 2 allow for quantification of the overall protein levels . DOI: http://dx . doi . org/10 . 7554/eLife . 02860 . 024 We then examined whether the differential phosphosite preferences of D6PK and PID as observed in the in vitro phosphorylation experiments ( Figure 2C ) would also translate into differential defects in the oocyte auxin transport assay . Indeed , we found , in agreement with the in vitro data , that the PIN1 S1A S2A S3A mutant was less efficiently activated by PID than by D6PK ( Figure 8C ) . Inversely , the PIN1 S4A mutation that strongly affected activation by D6PK did not significantly affect activation by PID . Again , mutation of all four PIN1 phosphosites , PIN1 S1A–S4A , resulted in the strongest impairment of PIN1 activation by PID ( Figure 8C ) . We also used SRM analyses to examine the effects of the loss of PID as well as WAG1 and WAG2 function on the phosphorylation of PIN1 S4 ( Figure 8—figure supplement 2 ) and PIN3 S4 ( Figure 8—figure supplement 3 ) . However , in contrast to the strong defects in PIN S4 phosphorylation that we observed in the d6pk mutants , neither pid nor wag1 wag2 mutants showed a clear reduction in PIN phosphorylation at the S4 phosphosite suggesting that PID and WAG1/WAG2 do not contribute to PIN S4 phosphorylation in this tissue . We also aimed to conduct the complementary SRM analysis experiment of the PIN1 and PIN3 S1 , S2 , and S3 phosphosites but , for technical reasons , had to restrict our efforts to SRM measurements of PIN1 S1 ( Figure 8—figure supplement 4 ) and PIN3 S1 ( Figure 8—figure supplement 5 ) : Whereas the peptides comprising the S3 phosphosites of PIN1 and PIN3 were unsuitable for chemical peptide synthesis as predicted based on their primary amino acid sequence , we repeatedly failed to obtain synthetic peptides for the PIN1 and PIN3 S2 phosphosites . Our analysis of PIN1 and PIN3 S1 phosphorylations , however , showed that the phosphorylation at the S1 phosphosites was not affected when comparing the d6pk012 mutant with the d6pk012 mutant expressing a complementing D6PK transgene suggesting that D6PK does not contribute to the phosphorylation of S1 in vivo ( Figure 8—figure supplements 4 and 5 ) . Since our phosphosite analyses indicated that D6PK and PID share their PIN target but display differential preferences for these phosphosites , we analyzed the functional redundancy of these two kinases in promoter swap experiments by expressing them under the control of the genes' promoter fragments in the d6pk012 and the pid mutant background , respectively . These experiments demonstrated that D6PK and PID cannot functionally replace each other when expressed from the promoter of the respective other gene ( Figure 9 ) . Whereas the expression of PID from a PID promoter fragment was sufficient to complement the phenotypes of a pid mutant , the expression of D6PK under control of the PID promoter fragment failed to complement pid ( Figure 9A ) . Inversely , D6PK but not PID expression from a D6PK promoter fragment was sufficient to complement the d6pk012 mutant ( Figure 9B ) . In summary , these genetic experiments supported our conclusion that D6PK and PID/WAGs are functionally divergent and these findings and conclusions are in line with previous observations on the differential effects of these two kinases in PIN polarity control ( Dhonukshe et al . , 2010 ) . These differential phosophosite preferences as detected in in vitro as well as in vivo phosphosite analyses may be the basis of the distinct roles of the two kinases in the control of PIN polarity and plant growth control . 10 . 7554/eLife . 02860 . 025Figure 9 . PID and D6PK are functionally non-redundant in vivo . ( A ) and ( B ) Test for genetic suppression of ( A ) the inflorescence phenotype of the pid mutant ( 5-week-old plants ) and ( B ) the lateral root formation defect ( 8 day-old seedlings ) of d6pk012 triple mutants with PID and D6PK expressed from the PID ( PIDp ) and D6PK ( D6PKp ) promoters . The suppression of d6pk012 by D6PKp:YFP:D6PK demonstrates the functionality of the YFP:D6PK translational fusion employed in other experiments . Scale bars = 1 mm ( A ) and 1 cm ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02860 . 025 Since auxin had previously been shown to regulate auxin transport at the level of PIN transcription and PIN endocytosis control , we were also interested in examining the role of auxin on PIN phosphorylation . In these analyses , we detected concentration- , time- and D6PK-dependent increases in the phosphorylation of PIN1 S4 , PIN3 S4 and PIN3 S5 already 15 min after auxin application ( Figure 10A , B , Figure 10—figure supplements 1–4 ) . While these increases were clearly observed in the wild type , only marginal increases in PIN phosphorylation at these sites were observed in the phosphorylation deficient d6pk012 mutant . At the same time , phosphorylation at the preferential PID target site S1 was neither strongly impaired in d6pk012 mutants when compared to a d6pk012 mutant expressing a complementing D6PK transgene nor clearly induced by auxin ( Figure 8—figure supplements 4 and 5 ) . Furthermore , in agreement with an auxin-dependent control of PIN phosphorylation at S4 and S5 , we detected increased phosphorylation at S4 and S5 in the auxin-overproducing yucca mutant ( Figure 10A , B , Figure 10—figure supplements 1–4; Zhao et al . , 2001 ) . Although the analyses of the control peptides showed that there is also an overall increase in PIN abundance in yucca , the relative increases in phosphosite phosphorylations exceeded the increases in overall PIN abundance suggesting that PIN phosphorylation is activated in this mutant when compared to the wild type . 10 . 7554/eLife . 02860 . 026Figure 10 . Dose- and time-dependent phosphorylation of PIN1 S4 after auxin treatment . ( A ) . Quantification of PIN1 S4 phosphorylation in as a function of IAA concentration in inflorescence tissue of 5-week-old Arabidopsis plants . PIN1_pS4 , phosphorylated form; PIN1_S4 , unphosphorylated form . ( B ) SRM analysis of time-dependent S4 phosphorylation after IAA [10 µM] treatment of inflorescence tissue of 5-week-old Arabidopsis plants . DOI: http://dx . doi . org/10 . 7554/eLife . 02860 . 02610 . 7554/eLife . 02860 . 027Figure 10—figure supplement 1 . Dose-dependent phosphorylation of PIN3 S4 after auxin treatment . Quantification of PIN3 S4 phosphorylation as a function of IAA concentration in inflorescence tissue of 5-week-old Arabidopsis plants . PIN3_pS4 , phosphorylated form; PIN3_S4 , unphosphorylated form . The internal peptide controls allow for quantification of the overall protein levels . DOI: http://dx . doi . org/10 . 7554/eLife . 02860 . 02710 . 7554/eLife . 02860 . 028Figure 10—figure supplement 2 . Time-dependent phosphorylation of PIN3 S4 after auxin treatment . SRM analysis of time-dependent PIN3 S4 phosphorylation after IAA [10 µM] treatment of inflorescence tissue of 5-week-old Arabidopsis plants . PIN3_pS4 , phosphorylated form; PIN3_S4 , unphosphorylated form . The internal peptide controls allow for quantification of the overall protein levels . DOI: http://dx . doi . org/10 . 7554/eLife . 02860 . 02810 . 7554/eLife . 02860 . 029Figure 10—figure supplement 3 . Dose-dependent phosphorylation of PIN3 S5 after auxin treatment . Quantification of PIN3 S5 phosphorylation as a function of IAA concentration in inflorescence tissue of 5-week-old Arabidopsis plants . PIN3_pS5 , phosphorylated form; PIN3_S5 , unphosphorylated form . The internal peptide controls , shown in Figure 10—figure supplement 1 , allow for quantification of the overall protein levels . DOI: http://dx . doi . org/10 . 7554/eLife . 02860 . 02910 . 7554/eLife . 02860 . 030Figure 10—figure supplement 4 . Time-dependent phosphorylation of PIN3 S5 after auxin treatment . SRM analysis of time-dependent PIN3 S5 phosphorylation after IAA [10 µM] treatment of inflorescence tissue of 5-week-old Arabidopsis plants . PIN3_pS5 , phosphorylated form; PIN3_S5 , unphosphorylated form . The internal peptide controls , shown in Figure 10—figure supplement 2 , allow for quantification of the overall protein levels . DOI: http://dx . doi . org/10 . 7554/eLife . 02860 . 030 We had previously reported that D6PK is a plasma membrane-associated protein that cycles between the plasma membrane and the cytoplasm or intracellular compartments ( Zourelidou et al . , 2009; Willige et al . , 2013; Barbosa et al . , 2014 ) . This cycling is highly sensitive to the trafficking inhibitor Brefeldin A ( BFA ) and in selected BFA-treatment conditions D6PK can be depleted from the plasma membrane without significantly affecting the plasma membrane abundance of PIN1 ( Figure 11; Barbosa et al . , 2014 ) . The differential BFA-sensitivity of D6PK and PIN allowed us testing the contribution of plasma membrane-resident D6PK to PIN phosphorylation . For this purpose , we generated a phosphosite-specific antibody for PIN1 S4 that efficiently detected S4 phosphorylated PIN1 at the plasma membrane but failed to detect PIN1 S4A ( Figure 11—figure supplement 1 ) . Importantly , we found that PIN1 S4 phosphorylation was strongly decreased already minutes after BFA treatment when D6PK had become dissociated from the plasma membrane ( Figure 11 ) . Thus , PIN1 S4 phosphorylation depended on the presence of D6PK or other BFA-sensitive protein kinases at the plasma membrane . 10 . 7554/eLife . 02860 . 031Figure 11 . PIN1 pS4 is dependent on D6PK presence at the plasma membrane . Representative confocal images of root stele cells after immunostaining highlighting ( arrowheads ) the presence of YFP:D6PK ( D6PK ) , S4-phosphorylated PIN1 ( PIN1 pS4 ) and PIN1 at the plasma membrane before and the absence of D6PK and PIN1 pS4 after BFA treatment . Note that unphosphorylated PIN1 can still readily be detected in a polarized manner after S4-phosphorylation was efficiently removed . DOI: http://dx . doi . org/10 . 7554/eLife . 02860 . 03110 . 7554/eLife . 02860 . 032Figure 11—figure supplement 1 . α-PIN1 pS4 is a PIN1 S4 phosphosite-specific antibody . Representative confocal images of root stele cells after immunostaining , highlighting ( arrowheads ) the presence and absence of PIN1 and S4-phosphorylated PIN1 ( PIN1 pS4 ) in the wild type and the PIN1 S4A mutant , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 02860 . 032 In this study , we examined the functional roles of the D6PK protein kinases in PIN phosphorylation and auxin transport activation . We showed that d6pk mutants are impaired in basipetal auxin transport in inflorescence stems and postulated that PINs may be directly activated by D6PKs . This hypothesis was supported by the facts that D6PK colocalized with the basally localized PIN1 and PIN3 proteins in various cell types , that D6PKs phosphorylated PINs in vitro and that D6PKs influence PIN1 and PIN3 phosphorylation in vivo ( Zourelidou et al . , 2009; Willige et al . , 2013; Barbosa et al . , 2014 ) . Here , we tested this hypothetical functional relationship by examining PIN1 or PIN3 activity and auxin transport at various levels . We showed that basipetal auxin transport was reduced in inflorescence stems of d6pk mutants and that PIN-mediated auxin efflux was activated by D6PK in Xenopus oocytes . Furthermore , we could rule out that the decreases in auxin transport as measured in inflorescence stems are the consequence of changes in PIN abundance as demonstrated using confocal imaging , immunoblotting , and SRM analyses of PIN proteins . We furthermore demonstrated that D6PK-dependent PIN activation was dependent on specific phosphosites in PIN1 and PIN3 . Taken together , all our findings support the conclusion that D6PKs are major regulators of PIN-mediated auxin transport in inflorescence stems . Since d6pk mutants have a number of phenotypes such as gravitropism defects in the root , negative gravitropism defects in the hypocotyl , phototropism defects in the hypocotyl as well as defects in lateral root initiation ( Zourelidou et al . , 2009; Willige et al . , 2013; Barbosa et al . , 2014 ) , we are tempted to speculate that these other d6pk mutant phenotypes are also the consequence of reduced auxin transport activity of PINs in the absence of the PIN-activating D6PK kinases . The detailed analysis of the S1–S5 phosphosites in in vitro phosphorylation experiments and in oocyte auxin transport experiments revealed that PIN1 S4 as well as PIN3 S4 and S5 are major target sites for D6PK . This conclusion found support also in the analysis of the in vivo phosphorylation levels at these sites since phosphorylation at S4 and S5 was strongly reduced in the d6pk012 mutant . Interestingly , the preferential D6PK phosphosites S4 and S5 are not conserved in PIN2 and it is striking that the respective domains in PIN2 carry small insertions when compared to the other PIN proteins . Thus , the activation of PINs by phosphorylation may be regulated by the presence and abundance of activating kinases such as D6PK but also by the availability and conservation of phosphosites in their PIN targets . Besides S4 and S5 , PINs must have other phosphosites that are targeted by D6PK since auxin transport defects in pin mutants expressing the respective PIN S4A and S5A mutant variants are partially complemented and not as severe as those observed in the d6pk012 loss-of-function mutants . Besides phosphorylations at S1 , which are not affected in d6pk012 mutants , S2 and S3 would be other possible target sites since their mutation further impairs D6PK-dependent PIN phosphorylation in vitro and auxin transport in the oocyte system . In this respect , it is unfortunate that we were unable , for technical reasons , to measure phosphorylation at S2 and S3 in the d6pk and pid mutants . Our study also addressed the functional role of PID and the PID-related WAG1/WAG2 kinases in the control of auxin transport . While we found that PID and WAG2 activate PIN-mediated auxin efflux in Xenopus oocytes , we showed at the same time that PID has different phosphosite preferences when compared to D6PK . We observed these phosphosite preferences when analyzing PIN phosphorylation at S1–S5 in in vitro phosphorylation experiments , auxin transport in oocytes , and PIN phosphorylation by quantitative mass spectrometry in the d6pk012 , pid and wag1 wag2 mutants . Whereas D6PK appeared to have a preference for the S4 and S5 sites in PIN1 and PIN3 , PID preferentially phosphorylated the previously identified S1–S3 phosphosites . S1–S3 are highly related to each other and also highly conserved among all five plasma membrane-resident PIN auxin efflux carriers including PIN2 . Although D6PK had a preference for S4 and S5 phosphorylation , our in vitro phosphorylation experiments as well as the auxin transport experiments in oocytes further suggested that the phosphorylation of S1–S3 also contributes to full PIN phosphorylation and activation by D6PK . The respective inverse observations were made with PID . Whereas PID phosphorylation and activation of PIN1 was strongly impaired when S1–S3 were mutated , full impairment of phosphorylation and activation could only be achieved after S4 mutation . In this respect , we consider the complementation experiments of the pin1 mutant with the wild type PIN1 and the mutant PIN1 S4A transgenes particularly insightful . Here , we found that basipetal auxin transport in the inflorescence stem was partially impaired when PIN1 S1 was mutated whereas the strong inflorescence differentiation phenotype of the pin1 mutant was rescued . The partial complementation of the pin1 auxin transport defect indicates that PIN1 S4 is not the only phosphosite required for D6PK-dependent PIN1 activation and basipetal auxin transport . As such , this result is in agreement with the results of our in vitro phosphorylation and oocyte auxin transport experiments , which showed that D6PK can also activate PIN1 through S1–S3 phosphorylation . On the other side , the rescue of the differentiation defect can be explained because the PIN1 S4A protein still contained the preferential phosphorylation sites for PID . As shown in the in vitro phosphorylation experiment as well as in the oocyte auxin transport experiment , the PIN1 S4A mutant variant is neither strongly impaired in its phosphorylation by PID nor in its activation by PID . Thus , the essential phosphosites required for PID-dependent PIN activation and PIN polarity changes are retained in PIN1 S4A . Therefore , the selective functionality of this mutant PIN1 in the context of inflorescence development indirectly supports the findings of our other analyses . Along the same lines , we also studied the ability of a PIN3 S4A S5A transgene to rescue the strong phototropism defect of the pin347 triple mutant . We had previously shown that d6pk d6pkl1 double mutants as well as pin347 triple mutants are severely compromised in phototropic hypocotyl bending ( Willige et al . , 2013 ) . We had shown that this phenotype could be explained by a strong defect in basipetal auxin transport and the apparent accumulation of auxin in the cotyledons of dark-grown seedlings , which , in turn , correlated with the absence of an auxin maximum in the bending zone of the hypocotyl ( Willige et al . , 2013 ) . At the same time , PID-dependent PIN3 polarity changes could still take place in the d6pk012 mutant indicating that PID can function independently from D6PK on PIN3 . Our observation that the PIN3 S4A S5A transgene could only partially rescue the pin347 triple mutant phenotype supports the notion that phosphorylation at these sites is important for PIN3 activation but suggests further that other phosphosites , such as S1–S3 , may also be targeted by D6PK . This partial inactivation of an S4A S5A mutated PIN3 as observed in planta is in agreement with the partial impairment of PIN3 S4A S5A phosphorylation in in vitro phosphorylation experiments with D6PK as well as the fact that there is still a residual activation when PIN3 S4A S5A is activated with D6PK in the oocyte system . With regard to the relevance of S1–S3 for PIN3 function , we found that mutation of PIN3 S1–S3 or S1–S5 rendered this PIN3 non-functional when introduced as a transgene into pin347 . Since these mutant PIN3 variants would be expected to be impaired in D6PK-dependent basipetal auxin transport as well as in PID-dependent PIN3 polarity changes , it is difficult based on the present depth of analysis to judge whether the non-functionality of PIN3 S1A–S3A or PIN3 S1A–S5A is primarily caused by a defect in basipetal auxin transport , a defect in changing PIN3 polarity or a combination of both . In our experiments , PIN phosphorylation led to a direct activation of auxin efflux in oocytes . The analysis of auxin transport in Arabidopsis inflorescence stems suggested that this might indeed be the primary function of this modification since auxin transport was strongly impaired in d6pk loss-of-function mutants while PIN abundance at the plasma membrane was not altered . This observation does , however , not rule out that PIN phosphorylation has other regulatory roles such control of PIN polarity by PID or WAG-dependent phosphorylation ( Friml et al . , 2004; Dhonukshe et al . , 2010 ) . Changes in PIN polarity as they are observed after PID or WAG2 overexpression are not observed after D6PK overexpression ( Zourelidou et al . , 2009; Dhonukshe et al . , 2010; Barbosa et al . , 2014 ) . The differential effect of D6PK and PID/WAGs on PIN may have its molecular basis in the differential phosphosite preferences of the two kinases . Common to both kinases seems , however , the fact that their phosphorylation activity is antagonistically regulated by phosphatases . While the phenotypic effects of PID can be antagonized by PP2A phosphatases ( Michniewicz et al . , 2007 ) , removal of the D6PK from the plasma membrane through BFA treatment resulted in an almost immediate decrease in PIN1 phosphorylation . Thus , it can be speculated that also D6PK-dependent PIN phosphorylation is antagonized by phosphatases , the identities of which remain to be determined . Our data also suggest that PIN1 and PIN3 phosphorylation is not only controlled by the presence of D6PK at the plasma membrane but also by auxin itself . Using quantitative SRM analyses , we could show that PIN1 S4 as well as PIN3 S4 and S5 phosphorylation increases in response to auxin treatment in the wild type . In the d6pk012 mutant , the loss of phosphorylation at these preferential D6PK phosphosites could not be compensated by auxin application suggesting that these auxin-dependent phosphorylations are D6PK-dependent and may be mediated either directly by D6PK or by D6PK acting as an indirect auxiliary factor . Although we observed a minor increase in PIN phosphorylation at the D6PK phosphosites in the d6pk012 triple mutant , these increases were comparatively minor and may be attributed to phosphorylation through D6PKL3 , which is still expressed in the d6pk012 mutant . Alternatively , they may be attributed to the activity of other PIN-regulatory kinases such as the PID/WAGs or other as yet uncharacterized protein kinases . Theoretically , it could be envisioned that the auxin-dependent increases in PIN phosphorylation are the consequence of the previously reported inhibitory effects of auxin on PIN endocytosis ( Paciorek et al . , 2005 ) . In this case , PIN phosphorylating kinases would encounter their PIN targets simply for a longer period of time thereby increasing the chances for phosphorylation . Unraveling the identity of the underlying auxin-sensory mechanism and its molecular details will be an interesting avenue for future investigations ( Dharmasiri et al . , 2005; Parry et al . , 2009; Robert et al . , 2010 ) . We recently reported that auxin treatment led to a transient dissociation of D6PK from the plasma membrane in root cells There , this auxin response correlates with a slight decrease in PIN1 phosphorylation as judged by immunoblots ( Barbosa et al . , 2014 ) . In contrast , we report here that auxin promotes PIN phosphorylation in inflorescence stems as determined by quantitative mass spectrometry of PIN1 S4 , PIN3 S4 and PIN3 S5 . It is at present difficult for us to reconcile these two apparently contrasting observations . We can therefore only argue that PIN phosphorylation is controlled by different auxin-dependent regulatory mechanisms in different tissues . In summary , our study provides evidence that PIN-mediated auxin efflux requires activation by PIN phosphorylating kinases such as D6PK and PID/WAGs . Several of our findings point at a differential biochemical activity of these two AGCVIII kinase representatives on PINs that may explain their differential effects in controlling PIN polarity , auxin transport , and plant growth . The differential PIN-dependent distribution of auxin within the plant is of pivotal importance for the regulation of a multitude of processes in plant growth and development . It is our view that the activation of PINs by D6PKs and PID/WAGs is a crucial component of the control of auxin transport that must be taken into account to understand auxin transport within the plant and to ultimately understand plant growth . The following mutant alleles were used for this study: Single , double and triple mutants of d6pk-1 ( d6pk0; SALK_061847 ) , d6pkl1-1 ( d6pk1; SALK_056618 ) , d6pkl2-2 ( d6pk2; SALK_086127 ) ( Zourelidou et al . , 2009 ) ; d6pk012 triple mutants with a complementing D6PKp:YFP:D6PK transgene ( Willige et al . , 2013 ) ; DR5:GFP ( Jonsson et al . , 2006 ) ; pid-14 ( SALK_049736 ) ; pin1 ( SALK_047613 ) ; pin3-3 pin4-101 pin7-102 ( Willige et al . , 2013 ) ; PIN1:GFP ( Grieneisen et al . , 2007 ) ; wag1 wag2 ( Santner and Watson , 2006 ) . D6PKp:GUS transgenic lines expressing the ß-glucuronidase ( GUS ) reporter under control of D6PK promoter ( D6PKp ) fragments as previously described ( Zourelidou et al . , 2009 ) . PIN1p:GUS , PIN3p:GUS , PIN4p:GUS and PIN7p:GUS ( Vieten et al . , 2005 ) were obtained from the Nottingham Arabidopsis Stock Center ( NASC ) or are a gift from Christian Luschnig ( Vienna , Austria ) . Primer sequences for cloning , insert amplification and site-directed mutagenesis are listed in Supplementary file 1A . For in vitro transcription prior to protein translation in oocytes , PIN1 and PIN3 were inserted into the expression vector pOO2 ( Broer , 2010 ) . To this end , the genes were amplified from cDNA templates and cloned as blunt-ended Phusion polymerase-amplified ( Biozym , Hessisch Oldendorf , Germany ) PCR fragments into the SmaI or EcoRV site of pOO2 . The S to A mutations were introduced by PCR-based site-directed mutagenesis using primers carrying mutations for the respective S to A replacements . YFP:D6PK and kinase-dead YFP:D6PKin were amplified in a similar manner as described for PIN1 and PIN3 from previously described vector templates ( Zourelidou et al . , 2009 ) and inserted into the EcoRV site of pOO2 . A PCR-fragment of the PID coding sequence was first inserted into pJET1 . 2 ( Fisher Scientific , Schwerte , Germany ) and from there transferred as an Xho1/Xba1 fragment into pOO2 . A PCR-fragment of the WAG2 CDS was cloned directly into pOO2 after Xba1/Nco1 digestion . The phot1 CDS was cloned as a BamH1/Pst1-digested PCR fragment into the corresponding sites of pOO2 . Constructs for the expression and purification of glutathione-S-transferase ( GST ) -tagged PIN and D6PK were previously described ( Zourelidou et al . , 2009; Huang et al . , 2010; Willige et al . , 2013 ) . GST:PID was obtained by inserting a Gateway-compatible PCR-fragment obtained from a PID cDNA with the primers PID-GW-FW and PID-GW-RV into pDONR201 before transferring the PID insert to pDEST15 ( Life Technologies , Carlsbad , CA ) . D6PKp:YFP:D6PK was generated by inserting YFP:D6PK including a terminator sequence as an XhoI/NotI fragment excised from pEXTAG-YFP-GW1 into pGREEN022912 to generate pGREEN-YFP:D6PK . A 1977 bp D6PK promoter-PCR fragment ( D6PKp ) was cloned into pCR2 . 1 ( LifeTechnologies , Carlsbad , CA ) and subsequently inserted as KpnI/XhoI fragment into pGREEN-YFP:D6PK to generate D6PKp:YFP:D6PK . To obtain D6PKp:D6PK , the D6PK CDS plus terminator sequence was excised from p35SGW-MYC as an Xho1/Spe1 fragment ( Zourelidou et al . , 2009 ) and inserted into the respective sites of pGREEN0229 containing D6PKp . In a similar manner , the PID promoter ( 2344 bp ) was amplified by PCR from genomic DNA , inserted into pCR2 . 1 ( Life Technologies , Carlsbad , CA ) and from there transferred as a Kpn1/Xho1 fragment into pGREEN0229 . To obtain D6PKp:PID and PIDp:PID , the PID CDS was inserted as a Gateway-technology compatible insert into p35SGW-MYC and from there cloned as an Xho1/Spe1 insert into pGREEN0229 containing the D6PKp or PIDp . Genomic PIN1 constructs were prepared by insertion of a 3558 bp Sal1/Not1-digested PCR fragment including the PIN1 open reading frame and terminator into pGREEN0229 . Subsequently , a 2081 bp PIN1 promoter fragment was inserted upstream from PIN1 as a Kpn1/Sal1-digested PCR fragment . Mutations for the S4A replacement were introduced by site-directed mutagenesis ( Sawano and Miyawaki , 2000 ) . The constructs were transformed into heterozygous PIN1/pin1 ( SALK_047613 ) plants by Agrobacterium-mediated transformation13 and pin1 homozygous lines carrying the PIN1 transgenes were isolated from the progeny . Plants expressing comparable levels of the PIN1 protein were identified by immunoblotting . Constructs for the expression of PIN3 under the control of its own promoter were obtained by amplifying a fragment spanning the region from 1776 bp upstream of the PIN3 translation start site to 621 bp downstream of the PIN3 stop codon with the primers PIN3g-ApaI-FW and PIN3g-NotI-RV and inserted into the KpnI and NotI sites of pGREEN0229 . The S1A through S5A mutations were introduced into the wild type construct by PCR-based site-directed mutagenesis using the primers listed in Supplementary file 1A . The constructs were transformed by Agrobacterium-mediated transformation into pin3 pin4 pin7 mutants ( Willige et al . , 2013 ) and phototropism experiments were performed on T2 progeny seedlings segregating for the PIN3 transgenes in the pin3 pin4 pin7 mutant background . Assuming that 25% of the segregating population represent non-transgenic pin3 pin4 pin7 segregants , the 25% of the seedlings of the analysed population ( n >50 for pin3 pin4 pin7 PIN3 S4A S5A; n >25 for pin3 pin4 pin7 PIN3 S4A S5A and pin3 pin4 pin7 PIN3 S1A S2A S3A S4A S5A ) with the lowest hypocotyl angle were excluded from the analysis . The T2 progeny of at least three independent transgenic lines was analysed for each transgene , and in each case the three lines with the strongest phenotypic suppression were chosen for the graphic representations and statistical analyses . The variance between the individual transgenic populations was analysed with a Kruskal–Wallis ANOVA on ranks ( Kruskal and Wallis , 1952 ) . To measure auxin transport in Arabidopsis inflorescence stems , 25-mm stem sections were cut above the rosette of 5-week-old plants and placed , in inverted orientation , into 30 μl auxin transport buffer containing 500 pM IAA , 1% ( wt/vol ) sucrose , 5 mM 2- ( N-morpholino ) ethanesulfonic acid ( MES ) , [pH 5 . 5] with or without 100 μM 1-N-naphthylphthalamic acid ( NPA ) . At the beginning of the transport experiment , the stem segments were transferred to 30 µl auxin transport buffer containing 417 nM ( 11 kBq ) [3H]-IAA ( American Radiolabeled Chemicals , St . Louis , MO ) . After 2 hr , 5-mm segments were dissected from the inflorescence stem , the lowermost 5-mm segment was discarded , and the remaining segments were macerated overnight in 3 ml QuickSafe A ( Zinsser Analytic , Frankfurt , Germany ) . [3H]-IAA was quantified using a liquid scintillation analyzer ( Tri Carb 2100TR; Perkin–Elmer ) . The results presented are the average and standard deviation of at least four biological replicate measurements in the case of wild type , d6pk mutants , pin1 PIN1 and pin1 PIN1S4A , and at least two biological replicates in the case of pin1 or the NPA-treated wild type . The experiments were repeated with comparable results and the result of a comparable experiment is shown . Where relevant , auxin transport measurements were compared using the linear mixed-effects model analysis ( fixed factors ) using the R software package . Xenopus laevis oocyte collection was performed as previously described and cRNA injection was carried out the day after surgery ( Kottra et al . , 2009 ) . cRNA was synthesized using the mMessage Machine SP6 Kit ( Life Technologies , Carlsbad , CA ) and cRNA concentration was adjusted to 300 ng/µl PIN and 150 ng/µl protein kinase , respectively . Oocytes were injected with ∼50 nl of a 1:1 mixture of cRNAs for PIN and protein kinase . If only PIN or protein kinase cRNA was injected , the cRNA was mixed 1:1 with water ( mock control ) . Following injection , oocytes were incubated in Barth's solution containing 88 mM NaCl , 1 mM KCl , 0 . 8 mM MgSO4 , 0 . 4 mM CaCl2 , 0 . 3 mM Ca ( NO3 ) 2 , 2 . 4 mM NaHCO3 , 10 mM HEPES ( pH 7 . 4 ) supplemented with 50 µl gentamycine at 16°C for 4 days to allow for protein synthesis . An outside medium buffer at pH 7 . 4 was chosen to prevent passive rediffusion of IAA into the oocytes , which would take place at acidic pH . At the beginning of the oocyte experiment , 10 oocytes were injected for each time point with 50 nl of a 1:5 dilution ( in Barth's solution ) of [3H]-IAA , 25 Ci/mmol; 1 mCi/ml ( ARC , St . Louis , MO ) to reach an intracellular oocyte concentration of ∼1 µM [3H]-IAA based on an estimated oocyte volume of 400 nl ( Broer , 2010 ) . After [3H]-IAA injection , oocytes were placed in ice-cold Barth's solution for 10 min to allow substrate diffusion and closure of the injection spot . Subsequently , oocytes were washed and transferred to Barth's solution at 21°C to allow for auxin efflux . To stop auxin efflux , oocytes were washed twice and lysed individually in 100 µl 10% SDS ( wt/vol ) at selected time points and the residual amount of [3H]-IAA in each oocyte was determined by liquid scintillation counting . At least 10 oocytes were measured per time point and mock as well as other negative controls were performed with the same oocyte batch to account for differences between batches . The relative transport rates of an experiment were determined by linear regression as shown in Figure 5—figure supplement 1 . Transport rates of different biological replicates ( i . e . oocytes collected from different female donors ) were averaged and are presented as mean and standard error of at least three biological replicates . Comparability in protein expression between the respective wild type and mutant protein variants and between the experiments was confirmed using immunoblots or confocal laser scanning microscopy with a Axiovert 200 M microscope equipped with a LSM 510 META laser scanning unit ( Zeiss , Jena , Germany ) . For protein extraction from Xenopus laevis oocytes , up to 25 oocytes were homogenized by trituration on ice in a homogenization buffer containing 50 mM Tris–HCl , 100 mM NaCl , 1 mM EDTA , 1 mM Pefabloc ( 400 µl/oocyte ) . The homogenate was centrifuged at 2000×g for 10 min at 4°C and the supernatant was transferred to a polyallomer microfuge tube ( Beckman Instruments , Fullerton , CA ) . Membrane proteins were pelleted from this supernatant at 150 , 000 g for 30 min at 4°C . The supernatant ( CF , cytoplasmic fraction ) was recovered and the pellet ( MF , microsomal membrane fraction ) was resuspended in homogenization buffer supplemented with 4% ( wt/vol ) SDS ( 8 µl per oocyte ) . The equivalent of 1/16th oocyte was loaded for immunoblots . To detect PIN phosphorylation , the homogenization buffer was supplemented with PhosSTOP phosphatase inhibitor cocktail ( Roche , Penzberg , Germany ) and the samples were immediately subjected to immunoblot analysis with the following antisera: anti-PIN1 ( 1:5000; NASC ) , anti-PIN3 ( 1:3000; NASC ) , anti-GFP ( 1:2000; Life Technologies , Carlsbad , CA ) , anti-UCN ( 1:2000; a gift from Kay Schneitz , Technische Universität München , Germany [Enugutti et al . , 2012] ) , anti-phot1 and anti-S851-phot1 antisera ( 1:1000; a gift from Shin-ichiro Inoue , Nagoya University and Ken-ichiro Shimazaki , Kyushu University , Japan [Inoue et al . , 2008] ) . Secondary detection was performed with donkey anti-sheep IgG-HRP ( 1:5000; Dianova , Hamburg , Germany ) and goat anti-rabbit IgG-HRP ( 1:5000; Santa Cruz Biotechnology , Santa Cruz , CA ) . PIN1 western blots from PIN1 transgenic plants were performed as previously described ( Willige et al . , 2011 ) . For GUS staining , inflorescence stem segments were sectioned with a razor blade , fixed for 15 min in heptane and stained for 4 hr or overnight ( PIN7p:GUS only ) with GUS-staining solution ( 100 mM Na-phosphate buffer pH 7 . 0 , 0 . 1% ( vol/vol ) Triton X-100 , 0 . 2 mg/ml 5-bromo-4-chloro-3-indolyl β-D-glucuronic acid ) and subsequently destained in 70% ( vol/vol ) ethanol . Images were taken with a Leica MZ16 microscope ( Leica Microsystems , Heerbrugg , Switzerland ) . Peptides for phosphorylation experiments as listed in Supplementary file 1B were synthesized by standard automated solid phase chemistry following the Fmoc ( Fluorenylmethyloxycarbonyl ) strategy ( Multipep , Intavis , Cologne ) . Phosphorylation experiments were performed using 0 . 5 µg purified GST:D6PK and 50 µM synthetic peptide in a reaction buffer containing 125 mM Tris pH 7 . 5 , 25 mM MgCl2 , 1 mM EDTA , 1 × Complete protease inhibitor cocktail ( Roche , Penzberg , Germany ) , 0 . 09 mM ATP , 0 . 0125% xylene cyanol and 0 . 1 µl [©−32P]ATP ( 370 MBq , specific activity 185 TBq; Hartmann Analytic , Braunschweig , Germany ) . The reactions were incubated for 1 hr at 30°C and 2 µl of the 20 µl reaction were spotted in duplicates on P81 ion exchange chromatography paper ( GE Healthcare , Freiburg , Germany ) . Air-dried chromatography papers were washed with 0 . 85% phosphoric acid , dried and exposed to X-ray film . Phosphorylation experiments with recombinant PIN cytoplasmic loop substrates were performed using 0 . 2 µg GST:D6PK or GST:PID and 0 . 5 µg GST:PIN substrate in a reaction buffer containing 25 mM Tris pH 7 . 5 , 5 mM MgCl2 , 0 . 2 mM EDTA , 1 × cOmplete protease inhibitor cocktail ( Roche , Penzberg , Germany ) , and 0 . 5 µCi [©−32P]ATP ( 370 MBq , specific activity 185 TBq; Hartmann Analytic , Braunschweig , Germany ) . Reactions were incubated for 1 hr at 30°C and separated on 10% SDS-PAGE . Gels were dried using a vacuum drier and exposed to X-ray film . Band intensities were quantified using MultiGauge v . 3 . 0 and normalized to the band intensities of the wild type . Phosphorylation experiments with recombinant PIN cytoplasmic loop substrates for mass spectrometric analysis were performed at 30°C for 1 hr in a non-radioactive reaction buffer containing 25 mM Tris pH 7 . 5 , 5 mM MgCl2 , 0 . 2 mM EDTA , 1 × cOmplete protease inhibitor cocktail ( Roche , Penzberg , Germany ) , 0 . 15 mM ATP , 1 × PhosSTOP phosphatase inhibitor cocktail ( Roche , Penzberg , Germany ) with 5 µg purified recombinant D6PK and 5 µg purified recombinant PIN cytosolic loop . For subsequent mass spectrometric analyses , the reactions were separated on a 10% SDS-PAGE and stained with Coomassie Brilliant Blue . PIN bands were cut from the gel , destained with two washes of H2O and two washes of 50% acetonitrile/50 mM NH4HCO3 pH 8 at 37°C . The bands were then sliced into small pieces ( 1 mm2 ) and transferred to a low binding microcentrifuge tube . The gel pieces were then covered in a solution with 50 mM dithiothreitol ( DTT ) , 50 mM NH4HCO3 and incubated for 1 hr at 60°C . After cooling to room temperature , the solution was replaced by 100 mM iodoacetamide in 50 mM NH4HCO3 and incubated for at least 1 hr in the dark . Subsequently , the gel pieces were washed three times by vortexing for 10 min in 50 mM NH4HCO3 , pH 8 . Following removal of the wash solution , the gel pieces were dried in a SpeedVac concentrator for 30 min and then incubated overnight in 10 µl Bovine Sequencing Grade Trypsin ( Roche , Penzberg , Germany ) dissolved in 50 mM NH4HCO3 , 1 mM CaCl2 . The trypsin solution was subsequently removed and transferred to a low binding tube . 10 µl of trifluoroacetic acid ( TFA; 5% wt/vol H2O ) were then added to the gel pieces and after sonication for 1 min the supernatant was transferred to the tube containing the previous liquid . The same procedure was repeated by adding 10 µl 15% acetonitrile/1% TFA to the gel pieces and combining the liquid with the previous supernatants . Mass spectrometry was performed using an nLC-LTQ-Orbitrap tandem mass spectrometer at Biqualys ( Wageningen , The Netherlands ) , and the data were analysed using the Bioworks software ( Thermo Fisher Scientific , Ulm , Germany ) . PIN protein alignments was performed using the ClustalW alignment option of the Geneious ( Biomatters , Auckland , New Zealand ) software package . Seedlings were grown in the dark at 22°C on vertically oriented half-strength Murashige and Skoog ( MS ) agar ( 0 . 8% ) plates for 3 to 4 days . Agravitropically growing seedlings were reoriented toward the gravity vector in safe green light 2 to 4 hr before the experiment . The seedlings were then transferred to GroBank growth chambers ( CLF Plant Climatics , Wertingen , Germany ) and illuminated with unilateral white light ( 100 µmol m−2 s−1 ) . Plates were subsequently scanned and hypocotyl bending was measured for each seedling using the NIH ImageJ software . The rabbit anti-PIN1 pS4 antibody was generated with the phosphorylated synthetic peptide SGGGRN-S ( PO3H2 ) -NFGPGE followed by affinity-purifications against the non-phosphorylated and phosphorylated peptide at Eurogentec ( Liege , Belgium ) . Immunostaining was performed on roots of 5-day-old seedlings as previously described ( Sauer et al . , 2006 ) using rabbit anti-S4-P ( dilution 1:300 ) , goat anti-PIN1 ( 1:400; NASC Nottingham , UK ) , and mouse anti-GFP ( 1:300; Roche , Penzberg , Germany ) , and as secondary antibodies anti-rabbit Cy3 ( 1:500; Dianova , Hamburg , Germany ) , anti-goat FITC ( 1:100; Dianova , Hamburg , Germany ) , anti-rabbit FITC ( 1:300; Dianova , Hamburg , Germany ) as well as anti-mouse ALEXA FLUOR 488 ( 1:500; LifeTechnologies , Carlsbad , CA ) . For BFA treatment , seedlings were immersed in BFA [50 µM]-containing liquid MS medium prior to analysis . All images were taken with an Olympus FV1000 confocal microscope ( Olympus , Hamburg , Germany ) . The experiment was repeated several times with reproducible outcome , representative images are shown . Live imaging of DR5:GFP or fluorescent protein-tagged proteins was performed as previously described ( Barbosa et al . , 2014 ) . Protein extracts for SRM analyses were prepared from 5 cm primary inflorescence stem segments , excised at the base of the infloresence stems , from 5 week-old Arabidopsis plants grown in continuous light . Total protein extracts were prepared in an extraction buffer containing 50 mM Tris–HCl pH 7 . 5 , 150 mM NaCl , 0 . 5% Triton X-100 , 0 . 1 mM MG132 ( Z-Leu-Leu-Leu-al ) , 1 mM PMSF ( phenylmethylsulfonyl fluoride ) , protease inhibitor cocktail ( Sigma-Aldrich , St . Louis , MO ) and PhosSTOP ( Roche , Mannheim , Germany ) . From each sample , 150 µg total protein were prepared in 100 µl extraction buffer and precipitated with 10 ng/µl glycogen , 400 µl ethanol ( HPLC Gradient Grade , Roth , Karlsruhe , Germany ) , 25 mM NaOAc 2 . 5 M pH5 . 2 for 4 hr at room temperature . Subsequently , the samples were centrifuged at 10 , 000×g and air-dried for subsequent SRM analysis . Protein pellets were resuspended in 6 M urea , 2 M thiourea , pH 8 . Protein disulfide bridges were reduced by adding DTT and free cysteine residues were subsequently alkylated using iodacetamide . 150 µg protein was then digested using sequencing grade trypsin ( Promega ) and desalted over C18 tips22 . Phosphopeptides were enriched over titanium dioxide23 and eluted phosphopeptides as well as flow-through after peptide binding to titanium dioxide were kept for analysis . Synthetic peptides with fully 13C and 15N-labeled C-terminal K or R were synthesized ( Thermo Fisher Scientific , Ulm , Germany; Supplementary file 1C ) and spiked into the tryptic peptide mixture at concentrations ranging from 40 to 250 fmol depending on peptide ionization properties . Tryptic peptide mixtures including heavy standard peptides were then analysed by SRM using nanoflow HPLC ( Easy nLC , Thermo Scientific , Ulm , Germany ) coupled to a triple quadrupole mass spectrometer as mass analyser ( TSQ Quantum Discovery Max , Thermo Scientific , Ulm , Germany ) . Peptides were eluted from a 75 µm analytical column ( Easy Columns , Thermo Scientific , Ulm , Germany ) on a linear gradient running from 10% to 30% acetonitrile in 60 min and were ionized by electrospray directly into the mass spectrometer . Specifically , phosphorylated and non-phosphorylated peptides were selected as targets of analysis after optimization of ionization conditions using the standard peptides . Visible transitions were selected from acquired mass spectra of the synthetic standard peptides . A list of transitions used for each ( phospho ) peptide sequence is available as Supplementary file 1C . The quadrupole Q1 was set as a mass filter for the respective parent ion , while Q3 was set to monitor specific fragment ions . For each peptide , at least three fragment ions were used . Mass width for Q1 and Q3 was 0 . 7 Da , scan time 5 ms . Data analysis involving merging of fragment ion information to a parent ion sum of intensities and calculation of peak areas was done using the Software Pinpoint v . 1 . 0 ( Thermo Scientific , Ulm , Germany ) . For quantitative analysis of peptide abundance , ion intensity sums of the measured transitions were used and averaged between up to three biological replicates . Ion intensity sums of spiked-in heavy peptide were used to normalize for sample-to-sample variation .

In plants , a hormone called auxin controls the growth of the stems and roots . This chemical is transported from cell to cell , and its flow though the plant is redirected continuously as the plant is developing . Auxin is pumped out of cells by proteins in the cell membrane called ‘auxin efflux carriers’ . These proteins are usually found on one side of each cell and this is what gives the direction to auxin transport . Zourelidou , Absmanner et al . now report that being positioned on the correct side of a plant cell is not enough to enable an efflux carrier to do its job—it must also be turned on by kinases before it can pump auxin out of cells . Kinases are enzymes that add phosphate groups to specific sites on other proteins , and plants without certain kinases are unable to transport auxin . When Zourelidou , Absmanner et al . produced the efflux carrier and a plant kinase—which turns the efflux carrier on—in immature egg cells from frogs , auxin was rapidly pumped out of the cells . However , cells that contained the efflux carrier but not the kinase could not transport the hormone . Importantly egg cells from frogs do not normally transport auxin , but these cells are commonly used in experiments because they are large , which makes them easier to work with in the lab . One of at least two kinases must tag a number of sites on the efflux carrier to ensure that it is switched on . It was already known that some of these sites are involved in making sure that the efflux carrier is located on the correct side of the cell . Zourelidou , Absmanner et al . also found that auxin itself encourages the addition of phosphate groups onto the efflux carrier . Though it was thought that knowing where the auxin transporters are was enough to explain the direction of auxin transport in plants , it is now clear that activation by the kinases needs to be taken into account too . And since these kinases may activate the transporters to different extents , identifying how these proteins are controlled , for example by auxin itself , will be the next challenge in the field .

2014

Auxin efflux by PIN-FORMED proteins is activated by two different protein kinases, D6 PROTEIN KINASE and PINOID

Mitochondrial deficits in energy production cause untreatable and fatal pathologies known as mitochondrial disease ( MD ) . Central nervous system affectation is critical in Leigh Syndrome ( LS ) , a common MD presentation , leading to motor and respiratory deficits , seizures and premature death . However , only specific neuronal populations are affected . Furthermore , their molecular identity and their contribution to the disease remains unknown . Here , using a mouse model of LS lacking the mitochondrial complex I subunit Ndufs4 , we dissect the critical role of genetically-defined neuronal populations in LS progression . Ndufs4 inactivation in Vglut2-expressing glutamatergic neurons leads to decreased neuronal firing , brainstem inflammation , motor and respiratory deficits , and early death . In contrast , Ndufs4 deletion in GABAergic neurons causes basal ganglia inflammation without motor or respiratory involvement , but accompanied by hypothermia and severe epileptic seizures preceding death . These results provide novel insight in the cell type-specific contribution to the pathology , dissecting the underlying cellular mechanisms of MD . Leigh syndrome ( LS ) is the most frequent pediatric mitochondrial disorder , leading to defective mitochondrial energy metabolism . LS affects 1 in 40 , 000 births ( Rahman et al . , 1996 ) , although adult onset has also been described ( McKelvie et al . , 2012 ) . Mutations in more than 75 genes have been described to cause LS ( Lake et al . , 2016 ) . To date , no effective treatment or cure exists . Clinically , albeit highly variable , LS symptoms usually include failure to thrive , hypotonia , rigidity , seizures , ataxia , lactic acidosis , encephalopathy and premature death ( Rahman et al . , 1996; Lake et al . , 2016; Sofou et al . , 2014 ) . LS is characterized by restricted anatomical and cellular specificity ( Arii and Tanabe , 2000 ) , a common feature shared by mitochondrial diseases , affecting high energy-requiring tissues such as muscle and brain ( Molnar and Kovacs , 2017 ) . Pathologically , LS is characterized by the presence of bilateral symmetrical lesions predominantly in the brainstem and basal ganglia ( Arii and Tanabe , 2000 ) . Neuronal damage is responsible for most of the fatal symptoms , including respiratory failure and seizures ( Barends et al . , 2016 ) . However , the identity of the affected cellular populations and the molecular determinants of neuronal vulnerability have not been adequately elucidated , representing a challenge for the development of efficient treatments . Mutations affecting the NDUFS4 subunit of mitochondrial Complex I , a key structural component for the assembly , stability and activity of the complex ( Calvaruso et al . , 2012 ) , are commonly associated with a severe , early-onset LS phenotype ( Ortigoza-Escobar et al . , 2016 ) . Although a late-onset case of LS has been recently reported ( Bris et al . , 2017 ) , prognosis is usually poor and most of the patients die in early childhood ( Sofou et al . , 2014 ) . Animals with a global deletion of Ndufs4 ( Ndufs4KO mice ) develop a fatal encephalomyopathy , which recapitulates the classical signs of LS , including motor alterations , respiratory deficits , epilepsy and premature death ( Quintana et al . , 2010; Kruse et al . , 2008 ) . Behavioral and neuropathological characterization of Ndufs4KO mice revealed the pivotal role of the dorsal brainstem , particularly the vestibular nucleus ( VN ) , in disease manifestation and progression ( Quintana et al . , 2012 ) . However , the genetic identity of the neuronal populations and circuitries involved in the plethora of symptoms observed have not yet been identified . Here , we describe the contribution of genetically defined , discrete neuronal populations to the fatal phenotype of Ndufs4KO mice . To that end , we generated three mouse lines using a conditional genetic approach that selectively inactivates Ndufs4 in glutamatergic ( Vglut2-expressing ) , GABAergic ( Gad2-expressing ) or cholinergic ( ChAT-expressing ) neurons . The results reveal distinct , lethal phenotypes for the glutamatergic and GABAergic neuronal populations . To dissect the neuronal cell types contributing to the neuropathology observed in Ndufs4KO mice ( Quintana et al . , 2010; Kruse et al . , 2008; Quintana et al . , 2012 ) , we generated three mouse lines lacking Ndufs4 selectively in glutamatergic ( expressing Vglut2 ) , GABAergic or cholinergic neurons . We did this by crossing Ndufs4 exon2-floxed mice with Cre-driver lines of mice expressing either Slc17a6-Cre , Gad2-Cre or Chat-Cre as described in Figure 1A and Figure 1—figure supplement 1; the affected mice are referred to here as: Vglut2:Ndufs4cKO , Gad2:Ndufs4cKO or ChAT:Ndufs4cKO mice and their respective controls are Vglut2:Ndufs4cCT , Gad2:Ndufs4cCT or ChAT:Ndufs4cCT . Ndusf4 gene inactivation in glutamatergic or GABAergic neurons of both male and female mice resulted in failure to thrive and premature death ( Figure 1B–F ) ; however , there was no effect on survival , body weight , or motor function when Ndufs4 expression was abolished in cholinergic neurons ( Figure 1—figure supplement 1 ) . Selective deletion of Ndufs4 in glutamatergic or GABAergic neurons was confirmed by western blot analysis of NDUFS4 levels in brain areas where Vglut2 or Gad2 are preferentially expressed ( Lein et al . , 2007 ) ( Figure 1—figure supplement 2 ) . Vglut2:Ndufs4cKO mice had a median lifespan of 67 days with a mortality rate of 90% at postnatal day ( P ) 128 . Similarly , Gad2:Ndufs4cKO mice had a median lifespan of 60 days ( 90% mortality at P70 ) ( Figure 1B ) . This premature death was preceded by a reduction in body weight gain in male and female mice of both genotypes ( Figure 1C–F ) . At 7 weeks of age for females and 9 weeks of age for males , Vglut2:Ndufs4cKO mice stopped gaining weight , which resulted in an overall reduction in body weight when compared to age-matched controls ( Figure 1C , D ) . Similarly , male and female Gad2:Ndufs4cKO mice body weight reached a plateau 2–3 weeks before manifesting a sudden unexpected death ( Figure 1E , F ) . Both Vglut2:Ndufs4cKO and Gad2:Ndufs4cKO mice were also significantly smaller than their littermate controls ( Figure 1G–H ) . This lack of weight gain and reduced size appeared to be due to decreased food intake in both genotypes ( Figure 1I , J ) ; however , this was not significantly different when food consumption was normalized to body weight ( Figure 1—figure supplement 2 ) . The reduced lifespan and decreased body weight observed in both Vglut2:Ndufs4cKO and Gad2:Ndufs4cKO mice were the result of two prominently different clinical presentations . Gad2:Ndufs4cKO mice were , for the most part , phenotypically indistinguishable from controls , without any overt clinical alteration beyond a reduced growth rate for a few weeks prior to a sudden premature death . On the other hand , Vglut2:Ndufs4cKO mice manifested progressive motor and respiratory deterioration with most of the clinical signs visibly apparent ( Table 1 ) . Vglut2:Ndufs4cKO mice presented body tremor and a decline in balance as early as at 5 weeks ( early stage of the disease ) , which dramatically worsened as the disease progressed . In a mid-stage of the disease , mice increased body tremor and showed a prominent decline in balance and motor coordination ( Table 1 ) . Subsequently , animals started exhibiting ataxia and a progressive loss of the righting and hindlimb extension reflexes ( clasping ) ( Figure 1—figure supplement 3 ) . At a late stage of the disease , these mice showed increased tremor , were completely docile and hypotonic ( Table 1 ) . Animals also had difficulty maintaining a regular breathing pattern at an early stage of the disease . These breathing abnormalities worsened as the disease progressed with mice presenting noticeably shorter and deeper respirations at advanced stages of the disease ( Video 1 ) . Furthermore , at a late stage of the disease ( over P60 ) , about 40% of Vglut2:Ndufs4cKO developed hindlimb dragging and eventually hindlimb paralysis . We observed increased glial reactivity but no neuronal loss in the spinal cord of these mice when compared to Vglut2:Ndufs4cCT mice , which was determined by immunoblot analysis of spinal cord lysates from Vglut2:Ndufs4cCT and Vglut2:Ndufs4cKO mice using antibodies against GFAP ( glial fibrillary acidic protein ) , Iba-1 ( ionized calcium-binding adapter molecule 1 ) and NSE ( neuronal specific enolase ) ( Figure 1—figure supplement 3 ) . This glial reactivity was further confirmed by immunofluorescence analysis using anti-GFAP and anti-Iba-1 antibodies in the spinal cords of the Vglut2:Ndufs4cKO mice that exhibited hindlimb dragging ( Figure 1—figure supplement 3 , top panels ) . However , no signs of demyelination or immune cell infiltration were observed ( Figure 1—figure supplement 3 , mid and bottom panels ) . LS is characterized by symmetrical brain lesions and neuroinflammation in select nuclei , predominantly brainstem and/or basal ganglia ( Arii and Tanabe , 2000 ) . Accordingly , late-stage Ndufs4-deficient mice present overt lesions in brainstem ( namely vestibular nucleus ( VN ) , cerebellar fastigial nucleus ( FN ) and inferior olive ( IO ) ) , olfactory bulb and basal ganglia ( Quintana et al . , 2010; Quintana et al . , 2012; Chen et al . , 2017a ) . To define the contribution of Ndufs4 deficiency in either excitatory or inhibitory neurons to the overall neuroinflammatory phenotype and identify the specific brain areas with exacerbated astroglial and microglial reactivity , we performed immunofluorescence analysis using anti-GFAP and anti-Iba1 antibodies on brain sections of Vglut2:Ndufs4cKO and Gad2:Ndufs4cKO mice , and their respective controls ( Figure 2A–D ) . Analysis of these sections showed marked glial reactivity in VN , IO and FN ( Figure 2A , C; Figure 2—figure supplement 1 ) , accompanied by increased caspase eight activation in affected areas ( such as the VN ) in Vglut2:Ndufs4cKO mice ( Figure 2E , F ) , recapitulating most of the neuroinflammatory profile described for the global Ndufs4KO mice ( Quintana et al . , 2010 ) . In contrast , Gad2:Ndufs4cKO mice presented a more restricted glial reactivity pattern , including marked microglial and astroglial reactivity in primarily GABAergic nuclei such as the external globus pallidus ( GPe ) in the basal ganglia and the substantia nigra pars reticulata ( SNr ) ( Figure 2B , D ) , without affecting neighboring non-GABAergic areas like the dopaminergic substantia nigra pars compacta ( SNc ) ( Figure 2—figure supplement 2 ) . Other prominently GABAergic areas such as the olfactory bulb also showed increased immunoreactivity for GFAP and Iba-1 in Gad2:Ndufs4cKO mice when compared to control mice ( Figure 2B , D; Figure 2—figure supplement 1 ) . In a percentage of Gad2:Ndufs4cKO mice , prominent microglial reactivity with Purkinje neuron loss ( as assessed by calbindin staining ) was also observed in the cerebellar vermis and flocculus ( Figure 2—figure supplement 2 ) . Normal Slc17a6 ( Vglut2 ) or Gad2 transcript abundance was observed in the VN and OB of late-stage Vglut2:Ndufs4cKO or Gad2:Ndufs4cKO mice , respectively , suggesting preservation of Ndufs4-deficient neuronal populations ( Figure 2G , H ) . The extensive neuroinflammation present in the brainstem of Vglut2:Ndufs4cKO mice , and its similarity to the phenotype of the global Ndufs4KO , allowed us to further define this inflammatory phenotype using whole-tissue transcriptional profiling . Gene expression analysis in the brainstem of late-stage ( over P68 ) Vglut2:Ndufs4cKO mice using Illumina Beadchips ( MouseRef-8 V2; Illumina ) showed that differentially expressed ( DE ) mRNAs were , for the most part , upregulated in Vglut2:Ndufs4cKO mice ( p<0 . 05 , Fold Change > 2 ) ( Figure 2—figure supplement 2 ) . Selected genes included transcripts directly involved in the regulation of the immune system such as chemokines and their receptors ( Ccl3 , Ccl4 and Cxcr3 ) , toll-like receptors ( Tlr2 , Tlr7 ) , complement proteins ( C1qa , C1qb , C3 and C4b ) , surface antigens ( Cd84 , Cd86 and Ly9 ) , and markers of the myeloid cell lineage ( infiltrating macrophages and microglia ) including Lyz , Lyz2 , Lyzs and Slc11a1 , among others . Functional enrichment analysis of differentially expressed mRNAs ( 1 . 4-fold or higher ) using overrepresentation analysis ( ORA ) ( Wang et al . , 2017 ) showed that the top 10 most overrepresented Gene Ontology ( GO ) categories ( biological process , non-redundant ) were all related to defense and immune responses , and also uncovered components of the adaptive immune response ( GO:0002250; p=2 . 0177e-12 , FDR = 7 . 2033e-10 ) including ‘leucocyte mediated immunity’ ( GO:0002443; p=1 . 7533e-12 , FDR = 7 . 2033e-10 ) and ‘myeloid leukocyte activation’ ( GO:0002274; p=8 . 7458e-10 , FDR = 9 . 0537e-8 ) ( Figure 2—figure supplement 2 ) . Therefore , to further characterize the immune cell composition in these whole tissue gene expression profiles , we applied recently developed deconvolution tools that use leucocyte gene expression signature matrices to computationally infer the relative proportions of each immune cell type in gene expression mixtures ( Newman et al . , 2015; Chen et al . , 2017b ) ( Figure 2—figure supplement 2 ) . This analysis revealed a gene expression profile consistent with an increased proportion of proinflammatory CD4 cells ( Follicular cells , Th1 , Th17 and Treg ) , dendritic cells , mast cells and macrophages , and an underrepresentation of CD4 Th2 cells , CD8 cells and NK cells in the brainstem of Vglut2:Ndufs4cKO mice compared to controls , showing that Ndufs4 deficiency in glutamatergic neurons promotes a neuroinflammatory environment that involves a commitment to distinct proinflammatory TH cell lineages and a defined profile of tissue defense cells . Motor dysfunction is a prominent feature in LS and Ndufs4KO mice pathology ( Sofou et al . , 2014; Quintana et al . , 2010 ) . To genetically define the neuronal cell types mediating this functional disorder , we assessed motor coordination in Vglut2:Ndufs4cKO mice , Gad2:Ndufs4cKO mice , and their respective controls ( Figure 3 ) . Starting at P40 , and concurring with the onset of clinical signs , Vglut2:Ndufs4cKO mice showed impaired rotarod performance when compared to control littermates ( Figure 3A ) . While control mice maintained rotarod performance , Vglut2:Ndufs4cKO mice failed to properly execute the task , presenting a progressive decline in motor coordination , in line with the clinical phenotype . Conversely , and in agreement with the lack of apparent clinical signs , no differences in rotarod performance were observed in Gad2:Ndufs4cKO mice compared to control littermates ( Figure 3B ) . Exposure to a novel environment revealed a hypoactive phenotype in Vglut2:Ndufs4cKO mice as assessed by a reduction in the total distance traveled and the speed of exploratory movement in the open-field test ( Figure 3C–E ) . The severity of this phenotype in Vglut2:Ndufs4cKO mice positively correlated with age and disease stage ( Figure 3—figure supplement 1 ) . In contrast , no significant differences were observed in either distance traveled or speed in the open-field test between Gad2:Ndufs4cKO mice and their respective controls ( Figure 3F–H ) . No differences in the innervation of neuromuscular junctions were observed in the gastrocnemius muscle of either Vglut2:Ndufs4cKO or Gad2:Ndufs4cKO mice ( Figure 3I–K ) , suggesting a central origin of the motor phenotype . Respiratory abnormalities are frequently associated with disease mortality in LS patients and global Ndufs4KO mice ( Arii and Tanabe , 2000; Quintana et al . , 2012; Gerards et al . , 2016 ) . To provide insight into the genetic identity of the neurons responsible for this respiratory phenotype , we assessed respiratory function by unrestrained whole-body plethysmography in awake Vglut2:Ndufs4cKO and Gad2:Ndufs4cKO mice . Vglut2:Ndufs4cKO mice exhibited erratic plethysmographic recordings ( Figure 4A ) . In these mice , the frequency of respiration ( fR ) was markedly reduced at a mid-stage ( P40-P60 ) of the disease and worsened as disease progressed ( Figure 4C ) . In addition , significant differences were also seen in the volume of air inspired by the animal during one breath ( tidal volume , VT ) . VT was increased in Vglut2:Ndufs4cKO mice at a mid-stage of the disease and became significantly larger than in Vglut2:Ndufs4cCT at late disease stages ( Figure 4D ) . In contrast , Gad2:Ndufs4cKO mice did not exhibit irregular plethysmographic traces and had breathing patterns ( fR and VT ) that did not differ from controls ( Figure 4B and E–F ) . These data reveal that the motor impairment and respiratory deficits observed in the Ndufs4KO mouse are mediated by Vglut2-expressing excitatory neuronal populations . We have shown that neuronal inactivation of Ndufs4 in the VN promotes breathing abnormalities ( Quintana et al . , 2012 ) . This observation , along with the extensive neuroinflammation observed in the VN of late-stage Vglut2:Ndufs4cKO mice , prompted us to assess the activity of Vglut2-expressing neurons in the VN using in vivo electrophysiology . Cell-firing activity was recorded during an open-field session before identification by laser stimulation of Channelrhodopsin-2 ( ChR2 ) -expressing vestibular glutamatergic neurons in Vglut2:Ndufs4cKO and Vglut2:Ndufs4cCT mice ( Figure 5A ) . Electrophysiological recordings in vivo showed that the firing rate for vestibular Vglut2-expressing neurons from Vglut2:Ndufs4cKO mice was not significantly different than controls when mice were at rest . However , when the mice were actively moving in the open-field , the Vglut2-expressing VN neurons in the control mice approximately doubled their firing rate to about 40 Hz but the experimental group did not ( Figure 5B ) . The failure of VN glutamatergic neurons in Vglut2:Ndufs4cKO mice to respond to motor activity may account for the breathing abnormalities . Dysregulation of body temperature homeostasis , such as hypothermia , is commonly observed in LS patients ( Finsterer , 2008 ) . Accordingly , global Ndufs4KO mice present reduced body temperature starting at P30 ( Quintana et al . , 2010 ) . However , the genetic identity of the neuronal population involved in this phenotype was unknown . Hence , using telemetric temperature monitoring , we observed that Vglut2:Ndufs4cKO presented normal resting body temperature during early life , decreasing only as the disease progressed to mid-late stage ( P40 onwards , Figure 6A ) . In contrast , we identified a severe reduction in resting body temperature of Gad2:Ndufs4cKO as early as P20-P30 , which was maintained at all time points analyzed ( Figure 6B ) . Therefore , our results suggest a central role of GABAergic neurons in central body temperature regulation in the context of Ndufs4 deficiency . In Gad2:Ndufs4cKO mice , spontaneous seizure-like events were observed during routine husbandry practices such as lifting the animal by the tail or cage cleaning ( Video 2 ) . To define whether fatal seizures were the cause of the unanticipated death observed in Gad2:Ndufs4cKO mice , starting at P40 animals were continuously video-recorded to monitor for potential seizures leading to an abrupt death . Analysis of the recordings revealed that all deaths in Gad2:Ndufs4cKO mice consistently followed a severe generalized tonic-clonic convulsion . In contrast , spontaneous , sporadic or lethal seizures were not observed in either controls ( data not shown ) or Vglut2:Ndufs4cKO mice at any stage of the disease ( Figure 7—figure supplement 1 ) . Detailed visualization and analysis of the images unveiled that seizures in Gad2:Ndufs4cKO mice were mostly generalized and of multiple semiology ( unilateral , primary bilateral , secondary bilateral or alternating ) . Subsequent electroencephalographic ( EEG ) and electromyographic ( EMG ) characterization showed the presence of spontaneous generalized tonic-clonic ( GTC ) seizures ( Figure 7A ) , interictal spikes and myoclonic seizures in Gad2:Ndufs4cKO mice ( Figure 7B–C ) , which are hallmarks of epilepsy in mitochondrial disorders ( Rahman , 2015 ) , with 40–60% of mitochondrial disease patients manifesting seizures ( Koenig , 2008; El Sabbagh et al . , 2010; Canafoglia et al . , 2001 ) . In addition , local field potential ( LFP ) recordings identified the presence of seizure-like events ( Queiroz et al . , 2009 ) in the GPe of Gad2:Ndufs4cKO mice as soon as P35 , while being absent in control mice ( Figure 7D ) , suggesting that subcortical alterations in affected regions may precede the development of generalized seizures . In individuals with mitochondrial disease , seizures are commonly observed after stressors such as febrile events ( Bindoff and Engelsen , 2012 ) . Hence , susceptibility of Gad2:Ndufs4cKO mice to thermally-induced seizures was assessed ( Figure 7E–F ) . Gad2:Ndufs4cKO mice were highly sensitive to temperature compared to Gad2:Ndufs4cCT mice , with some mice displaying seizures already at basal temperature . The number of seizures significantly increased with temperature; 50% of the mice exhibited seizures at 39 . 5°C and all Gad2:Ndufs4cKO manifested seizures by 41 . 5°C . Both spontaneous and temperature-induced seizures were always preceded by a myoclonic ( MC ) seizure ( Figure 7A and F ) with the latter being more severe ( Racine scale stage 5 ) than the former ( Racine scale stage 4 ) ( data not shown ) . Both spontaneous and thermally-evoked seizures were characterized by EEG hyperactivity . Similarly to controls , Vglut2:Ndufs4cKO did not show susceptibility to temperature-induced seizures up to 42°C ( data not shown ) . Peri-onset administration ( from P40 onwards ) of the anti-epileptic drugs levetiracetam ( 60 mg/kg ) , perampanel ( 0 . 75 mg/kg ) or carbamazepine ( 40 mg/kg ) to Gad2:Ndufs4cKO mice failed to prevent fatal epileptic events , resulting in similar lifespan ( Figure 7—figure supplement 2 ) . Drug-treated mice showed a median survival of 63 , 62 and 66 days , respectively , which was not statistically significant when compared to the median lifespan of vehicle-treated mice ( 60 days ) . Neuronal vulnerability is one of the hallmarks of Leigh Syndrome . However , the restricted anatomical distribution of brain lesions indicates a clear gradation in neuronal susceptibility to LS-causing mutations . Specific neuronal populations in affected regions , such as brainstem or basal ganglia , are highly vulnerable to mitochondrial dysfunction and may underlie the plethora of neurological signs observed in LS . The scarcity and high variability of patients has limited our knowledge on the genetic identity and relative contribution of the affected neuronal populations to the phenotype; thus , a model system with consistent neuropathological features resembling mitochondrial disease is a valuable research tool . We have shown ( Quintana et al . , 2010; Quintana et al . , 2012 ) that mice lacking the Ndufs4 gene ( Ndufs4KO mice ) recapitulate the clinical signs of the human disease ( Ortigoza-Escobar et al . , 2016; Quintana et al . , 2012 ) . Ndufs4KO mice present overt lesions predominantly in the brainstem ( Quintana et al . , 2010; Quintana et al . , 2012; Johnson et al . , 2013 ) , but also in the striatum in late stages of the disease ( Quintana et al . , 2012 ) . Hence , we hypothesized that a concerted role of diverse neuronal populations was necessary to drive the plethora of symptoms observed in Ndufs4KO mice . In this study , we use a conditional genetic approach to selectively ablate NDUFS4 in ChAT-expressing cholinergic neurons ( ChAT:Ndufs4cKO ) , Vglut2-expressing glutamatergic neurons ( Vglut2:Ndufs4cKO ) or Gad2-expressing GABAergic neurons ( Gad2:Ndufs4cKO ) , thus restricting Complex I deficiency to some of the most abundant neuronal populations . This approach allowed us to provide a comprehensive dissection of the neuronal involvement in the phenotype of LS . Animals lacking Ndufs4 in cholinergic neurons ( ChAT:Ndufs4cKO ) were phenotypically equivalent to controls , indicating no overt contribution of this cell type to the pathology observed in Ndufs4KO mice . In contrast , Vglut2:Ndufs4cKO and Gad2:Ndufs4cKO mice had reduced lifespan and body weight , which was accompanied by a decrease in food intake , which are common clinical signs that appear in Leigh Syndrome patients ( Rahman et al . , 1996; Smeitink , 2003 ) . Recent reports have shown that neuronal cell-specific NDUFS4 knock down in Drosophila also leads to severe feeding abnormalities and premature death ( Foriel et al . , 2018 ) . Our results indicate a conserved role for neurons in the onset and progression of the pathological condition of global Ndufs4 deficiency and reveal that both glutamatergic and GABAergic systems contribute to the growth and lethality phenotype . Noteworthy , the slight reduction in phenotype severity of conditional neuronal Ndufs4 deficiency , with relative neuronal preservation , compared to the global Ndufs4KO mice suggests a concerted contribution of different cell types . In this regard , our previous work has shown the importance of neuron-astrocyte crosstalk in the development of neurodegeneration in the context of mitochondrial disease ( Liu et al . , 2015 ) . However , astrocytic Ndufs4 deficiency is not sufficient to recapitulate the phenotype of global Ndufs4KO mice ( Ramadasan-Nair et al . , 2019 ) , underscoring the central role of neurons in the disease . Even so , recent studies have shown enhanced neuronal survival in global Ndufs4KO after disruption of hepatic S6K1 ( Ito et al . , 2017 ) . Hence , the role of cell and tissue non-autonomous effects on disease progression have to be taken into account to fully understand the phenotype of neuronal-specific conditional Ndufs4KO mice . Apart from the premature death and feeding deficits , Vglut2:Ndufs4cKO and Gad2:Ndufs4cKO mice present two markedly distinct clinical entities ( summarized in Table 2 ) . With Vglut2:Ndufs4cKO mice , the lethality was associated with severe motor and respiratory alterations , whereas with Gad2:Ndufs4cKO mice , sudden unexpected death was associated with epilepsy ( SUDEP ) ( Abdel-Mannan et al . , 2019; Manolis et al . , 2019; DeGiorgio et al . , 2019 ) with no overt clinical alteration beyond the weight loss . Histologically , Vglut2:Ndufs4cKO mice present with prominent neuroinflammation and lesions in areas of the brainstem such as the VN , IO and the cerebellar FN , reminiscent of the pathology found in Ndufs4KO mice ( Quintana et al . , 2010 ) . We have identified a critical role of brainstem lesions in the development of fatal breathing alterations observed in the Ndufs4KO mice ( Quintana et al . , 2012 ) , in agreement with human LS patients ( Arii and Tanabe , 2000 ) . Glutamatergic signaling in the brainstem has been shown to regulate breathing ( Whitney et al . , 2000 ) . In addition , VN glutamatergic neurons have been suggested to modulate respiratory responses ( Xu et al . , 2002 ) . Furthermore , the Pre-Bötzinger ( PreBotC ) complex , a key respiratory center , is composed of glutamatergic neurons ( Stornetta et al . , 2003 ) and receives extensive glutamatergic inputs ( Bochorishvili et al . , 2012 ) . We have shown that Ndufs4KO mice present intrinsic PreBotC alterations and that vestibular projections to the PreBotC are necessary for maintaining respiration ( Quintana et al . , 2012 ) . Our electrophysiological recordings in Vglut2-expressing VN neurons from Vglut2:Ndufs4cKO mice show relatively normal basal firing rate but fail to increase spiking in response to locomotor activity . Neuronal firing is a highly energy-requiring process , mostly dependent in mitochondrial function ( Harris et al . , 2012 ) , especially in glutamatergic synapses ( Juge et al . , 2010; Zimin et al . , 2016 ) . Hence , our results suggest that glutamatergic VN neurons are unable to achieve the energy requirements of increased firing rates . Thus , it is likely that failure of glutamatergic VN projections to the PreBotC allowing appropriate responses to physiological needs underlies the breathing deficits observed in Vglut2:Ndufs4cKO mice . Development of brainstem lesions correlate with motor deficits in animals constitutively lacking Ndufs4 ( Quintana et al . , 2010 ) . Accordingly , strategies that improve motor symptoms in Ndufs4KO mice , such as AAV-based gene therapy ( Quintana et al . , 2012; Di Meo et al . , 2017 ) , rapamycin administration ( Johnson et al . , 2013 ) , or hypoxia ( Jain et al . , 2016; Ferrari et al . , 2017 ) cause a marked reduction in brainstem lesions . Here , we have identified a critical role for Vglut2-expressing glutamatergic neurons , in the brainstem and cerebellum , in the development of the motor deficits observed after Ndufs4 deficiency . In keeping with this , conditional deletion of Ndufs4 in dopaminergic neurons does not cause cell loss ( Kim et al . , 2015 ) or motor deficits ( Choi et al . , 2017 ) . However , other areas , such as the striatum , may participate in the delayed , mild , progressive motor dysfunction observed in LS patients ( Chen et al . , 2017a; Di Meo et al . , 2017 ) . Our gene expression analysis in brainstem of Vglut2:Ndufs4cKO mice has enabled the generation of an in-depth profile of the transcriptomic landscape in an affected brain area after Ndufs4 deficiency . This analysis revealed a prominent increase in inflammatory mediators in the affected tissue . However , anti-inflammatory or immunotherapeutic approaches have been mostly ineffective as treatments for LS ( Johnson et al . , 2013; Finsterer and Zarrouk-Mahjoub , 2017 ) with only a few successful cases reported ( Chuquilin et al . , 2016 ) . Our deconvolution data show a marked infiltration of distinct leucocyte populations , underscoring the complex cellular milieu elicited by mitochondrial dysfunction that may underlie the failure of global anti-inflammatory approaches . Delineation of the immune cells recruited to the brain lesions may lead to novel therapeutic approaches tailored for LS . The Gad2-expressing GABAergic neurons do not participate in the appearance of respiratory or motor deficits in Ndufs4 deficiency . However , they are critical for body temperature control and the onset and development of the fatal epileptic seizures , features that are observed in both global Ndufs4KO mice ( Quintana et al . , 2010 ) and LS patients ( Finsterer , 2008; Koenig , 2008; Finsterer and Zarrouk Mahjoub , 2012 ) . Lack of Ndufs4 in Gad2-expressing neurons leads to the appearance of neuroinflammation in the basal ganglia nuclei such as the GPe and SNr , in agreement with the increased vulnerability of basal ganglia neurons to mitochondrial dysfunction ( Gubellini et al . , 2010 ) . Furthermore , we show that electrophysiological alterations in the GPe neurons predate cortical epileptic events , suggesting a primary role of the basal ganglia circuitry in the development of epileptic seizures in Gad2:Ndufs4cKO mice . Basal ganglia are involved in epilepsy ( Rektor et al . , 2012; Badawy et al . , 2013; Vuong and Devergnas , 2018 ) , likely by acting as an inhibitory input to cortical seizure spread via feedback mechanisms ( Rektor et al . , 2012 ) . Hence , we hypothesize that basal ganglia inhibitory network is affected in Gad2:Ndufs4cKO mice , being unable to control the activity of cortical excitatory neurons , thus leading to epilepsy . Gad2:Ndufs4cKO mice are resistant to different antiepileptic approaches , such as the widely-used antiepileptic drugs carbamazepine , perampanel , and levetiracetam . Although earlier administration of these drugs may have led to a better antiepileptic outcome , epilepsy-induced death in mitochondrial disease patients is usually linked to refractory epileptic seizures ( Finsterer and Zarrouk Mahjoub , 2012 ) . Hence , Gad2:Ndufs4cKO mice may represent an excellent model to study epileptic mechanisms in LS , a much needed area of research , especially considering that most commonly used antiepileptic drugs may promote mitochondrial toxicity ( Finsterer , 2017 ) . As described , both LS patients ( Finsterer , 2008 ) and NDUFS4-LS patients ( Ortigoza-Escobar et al . , 2016 ) present predominant basal ganglia and brainstem affectation . Accordingly , basal ganglia and brainstem lesions are prominent features in global Ndufs4KO ( References: Quintana et al . , 2010; Quintana et al . , 2012; Table 2 ) , and GABAergic and glutamatergic conditional Ndufs4KO mice , respectively . However , alterations in other areas such as thalamus , cerebellum and spinal cord are also frequently observed , contributing to the clinical complexity of the pathology ( Arii and Tanabe , 2000; Lake et al . , 2015 ) . In line with our previous studies ( Quintana et al . , 2010; Quintana et al . , 2012 ) , here we show the glutamatergic origin of cerebellar and spinal cord alterations , probably contributing motor deficits observed in LS and Ndufs4KO . Clinically , LS patients commonly present hypothermia and failure to thrive ( Finsterer , 2008 ) , which are recapitulated in global Ndufs4KO mice ( Quintana et al . , 2010 ) . Our work shows reduced body weight and hypophagia in both Vglut2:Ndufs4cKO and Gad2:Ndufs4cKO , while hypothermia is mainly restricted to the latter . Central control of food intake and thermoregulation heavily rely on glutamatergic and GABAergic hypothalamic neuronal populations ( Tan and Knight , 2018; Sternson and Eiselt , 2017; Zhao et al . , 2017 ) . Hence , our results indicate , that even in the absence of overt neuroinflammation , neuronal Ndusf4 deficiency may lead to hypothalamic impairment , as observed in LS patients ( Zinka et al . , 2010 ) . One remaining question is the characterization of the underlying molecular mechanisms leading to the specific vulnerability of defined neuronal populations to Ndufs4 deficiency . In this regard , increased oxidative stress is one of the hallmarks of the phenotype ( Quintana et al . , 2010 ) . Accordingly , antioxidant treatments have proven moderate effectivity in the global Ndufs4KO mice ( Liu et al . , 2015; de Haas et al . , 2017 ) . Initiation of extrinsic apoptotic cascades has also been found in Ndufs4KO mice ( Finsterer and Zarrouk-Mahjoub , 2017 , and this work ) , even though EM imaging demonstrated mostly necrotic death in affected brain regions of Ndufs4KO mice ( El Sabbagh et al . , 2010 ) . In this regard , the remarkable immune cell infiltration described in this work may contribute to the initiation of these cascades . Finally , different studies have pointed at metabolic dysregulation as a potential contributor to the pathology . In this regard , mTOR inhibition or hypoxic conditions modify glycolytic levels , leading to clinical sign amelioration and extended lifespan in Ndufs4KO mice ( Johnson et al . , 2013; Jain et al . , 2016 ) . In conclusion , we provide new insights on the genetic identity of affected neuronal populations in LS by dissecting the associated cell type-specific molecular , biochemical , clinical and behavioral features in a model of LS . Our work highlights the importance of addressing mitochondrial disease at the cell type-specific level . The advent of new tools to assess transcriptomic and biochemical changes at this level of resolution ( Sanz et al . , 2009; Bayraktar et al . , 2019 ) bodes well for more progress . Hence , our work broadens current understanding of the etiology of LS and paves the way for future studies at the cell type-specific level to unravel the molecular determinants of neuronal pathology in LS . All experiments were conducted following the recommendations in the Guide for the Care and Use of Laboratory Animals and were approved by the Animal Care and Use Committee of the Seattle Children´s Research Institute and the Universitat Autònoma de Barcelona . Mice were maintained with Teklad Global rodent diet No . 2014S ( HSD Teklad Inc , Madison , Wis . ) and water available ad libitum in a vivarium with a 12 hr light/dark cycle at 22°C . The following mouse lines were used in this study: Slc17a6Cre ( BAC-Vglut2::Cre ) ( Borgius et al . , 2010 ) mice were generated by Ole Kiehn . Gad2Cre/+ ( Gad2-IRES-Cre ) ( Taniguchi et al . , 2011 ) and ChatCre/+ ( Chat-IRES-Cre ) ( Rossi et al . , 2011 ) mice were obtained from The Jackson Laboratory ( Stock No: 028867 and 031661 , respectively ) ( Bar Harbor , ME ) . Ndufs4lox/lox and Ndufs4Δ/+ were previously generated by our group ( Quintana et al . , 2010; Kruse et al . , 2008 ) . Male and female mice of different ages were used in this study . Sex and age of the animals are described in the figure legends . All mice were on a C57BL/6J background after backcrossing for at least 10 generations . Mice with conditional deletion of Ndufs4 in Vglut2-expressing glutamatergic neurons ( Slc17a6Cre , Ndufs4Δ/lox or Vglut2:Ndufs4cKO ) were generated by crossing mice with one Ndufs4 allele deleted and expressing a codon-improved Cre recombinase ( iCre ) under the Slc17a6 promoter ( Slc17a6Cre , Ndufs4Δ/+ ) to mice with two floxed Ndufs4 alleles ( Ndufs4lox/lox ) . Mice with conditional deletion of Ndusf4 in Gad2-expressing GABAergic neurons ( Gad2Cre/+ , Ndufs4lox/lox or Gad2:Ndufs4cKO ) were obtained by crossing mice with one floxed Ndufs4 allele and expressing Cre recombinase under the control of the Gad2 promoter ( Gad2Cre/+ , Ndufs4lox/+ ) to mice carrying two floxed Ndufs4 alleles ( Ndufs4lox/lox ) . Similarly , mice with conditional Ndufs4 deletion in ChAT-expressing cholinergic neurons ( ChatCre/+ , Ndufs4lox/lox or ChAT:Ndufs4cKO ) were obtained by crossing mice carrying one floxed Ndufs4 allele and expressing Cre recombinase driven by the ChAT promoter ( ChatCre/+ , Ndufs4lox/+ mice ) to Ndufs4lox/lox mice . Littermate controls were Slc17a6Cre , Ndufs4lox/+ ( Vglut2:Ndufs4cCT ) ; Gad2Cre/+ , Ndufs4lox/+ ( Gad2:Ndufs4cCT ) and ChatCre/+Ndufs4lox/+ ( ChAT:Ndufs4cCT ) mice . In all cases , genotype of the offspring and absence of ectopic recombination ( i . e . presence of recombination bands in tail DNA samples ) was determined by PCR analysis . Primer sequences have been described ( Kruse et al . , 2008 ) . Vglut2:Ndufs4cKO and Gad2:Ndufs4cKO mice were examined every other day for clinical signs resulting from cell type-specific Ndufs4 inactivation . Physiological ( body weight ) and behavioral ( locomotor activity , motor coordination , gait/postural alterations ) parameters were evaluated in more than 50 animals for each mouse line and were grouped into the following categories based on visual observation: ‘+++” severe manifestation of the clinical sign , ‘++” moderate manifestation of the clinical sign , ‘+” mild clinical sign , “- “absence of clinical sign . Mice were humanely euthanized after losing 20% of their peak body weight . Only Vglut2:Ndufs4cKO mice presented the overt and progressive clinical signs . Albeit the presence of individual variability in the development of the disease , early stage was defined in the range P20-P40 , mid-stage between P40-P60 , and late stage at ages over P60 in Vglut2:Ndufs4cKO . Food consumption was recorded from 7 to 11 weeks of age using a Physiocage system ( Panlab , Spain ) . Data at 8 weeks of age ( right before the median survival value ) are presented , including enough individuals to ensure sufficient statistical power . For immunofluorescence , mouse brains were collected and fixed overnight in 4% paraformaldehyde ( PFA ) in PBS ( pH 7 . 4 ) . Subsequently , brains were cryoprotected in a PBS solution containing 30% sucrose and frozen in dry ice . Frozen brains were embedded in OCT , sectioned at 30 μm in a cryostat and rinsed in PBS prior to staining . For western blot analysis , brain areas ( olfactory bulb , thalamus , spinal cord and globus pallidus ) were dissected according to the Paxinos mouse brain atlas ( Paxinos and Frank , 2013 ) and flash-frozen in liquid nitrogen before homogenization . Tissue sections were rinsed in PBS-0 . 2% Triton X-100 ( PBST ) solution . Non-specific binding was blocked with 10% normal donkey serum ( NDS ) in a PBST solution for 60 min at room temperature , followed by overnight incubation at 4°C with primary antibodies diluted in 1% NDS-PBST ( 1:2000 for mouse anti-GFAP , Sigma; 1:1000 for chicken anti-GFAP , Abcam; 1:1000 for anti-Iba-1 , Wako; 1:1000 for anti-TH , Millipore ) . Sections were then washed in PBST and incubated for 1 hr at room temperature with the corresponding Cy- ( 1:200 , Jackson Immunoresearch ) or Alexa Fluor-conjugated secondary antibodies ( 1:500 , Thermo Fisher Scientific ) in 1% NDS-PBST . Sections were finally washed in PBS and rinsed in water before mounting onto slides with Fluoromount G ( Electron Microscopy Sciences ) for microscopic analysis . Gastrocnemius muscles were sectioned in 60 µm longitudinal sections , collected in 24-well plates in sequential series of 4 slices per well in antifreezing solution . Sections were then blocked with PBS-0 . 3%Triton-5%Normal Donkey serum and incubated 48 hr at 4°C with primary antibodies anti-synaptophysin ( 1:500; AB130436 , Abcam , UK ) and anti-neurofilament 200 ( NF200 , 1:1000; AB5539 , Millipore , USA ) . After washes , sections were incubated overnight with Alexa 594-conjugated secondary antibody ( 1:200; A11042 , Invitrogen , USA ) and Alexa 488 conjugated alfa-bungarotoxin ( 1:200; B13422 , Life Technologies , USA ) . Slides with the sections were then mounted in Fluoromount-G ( Southern Biotech , USA ) . Confocal images were captured with a scanning confocal microscope ( LSM 700 Axio Observer , Carl Zeiss 40x/1 . 3 Oil DIC M27 , Germany ) . Maximum projections images shown in this study were created from 1 . 5 µm z projections . For neuromuscular junctions analysis , the proportion of fully occupied endplates was determined by classifying each endplate as fully innervated ( when presynaptic terminals overly the endplate ) , partially innervated ( when presynaptic terminals were not clearly within the endplate ) or vacant ( no presynaptic label in contact with the endplate ) . At least 3–4 fields with more than 80 endplates were analyzed per each muscle . Brain tissue samples were homogenized in iced-cold RIPA buffer ( Santa Cruz Biotechnology ) and protein concentration determined by the BCA assay ( Thermo Fisher Scientific ) . Thereafter , 20 μg of protein lysates were heat-denatured in Laemmli sample buffer ( Bio-Rad Laboratories , Inc ) , subjected to 4–20% gradient SDS-PAGE and transferred to nitrocellulose membranes ( Bio-Rad Laboratories , Inc ) . Membranes were then blocked for 1 hr with 5% ( w/v ) dried skimmed milk in Tris-buffered saline containing 0 . 1% Tween-20 ( TBS-T ) and incubated overnight at 4°C with primary antibodies against NDUFS4 ( Abcam , mouse , 1:500 ) , NSE ( Dako , mouse , 1:1 , 000 ) , GFAP ( Sigma , mouse , 1:50 , 000 ) , Iba1 ( Wako , rabbit , 1:10 , 000 ) , Active ( cleaved ) caspase 8 ( Cell Signaling Technologies , 1:1000 ) , β-actin ( Sigma , mouse , 1:20 , 000 ) or GAPDH ( GeneTex , mouse , 1:40 , 000 ) . After incubation with the corresponding HRP-conjugated secondary antibodies ( 1∶10 , 000; Jackson ImmunoResearch ) , membranes were washed in TBS-T and developed using an enhanced chemiluminescence ( ECL ) detection system ( Pierce ) . Bands were quantified using Image J software ( National Institutes of Health , USA ) . For WGGEX analysis , 150 ng of total RNA extracted from the brainstem of late-stage ( over P68 ) Vglut2:Ndufs4cKO ( n = 4 ) and Vglut2:Ndufs4cCT mice ( n = 4 ) was amplified and biotin-labeled using the Illumina TotalPrep RNA Amplification kit ( Ambion ) . 750 ng of the labeled cRNA was hybridized to MouseRef-8 v2 expression beadchips ( Illumina ) for 16 hr before washing and analyzing according to the manufacturer's directions . Signal was detected using a BeadArray Reader ( Illumina ) , and data were analyzed for differential expression using the GenomeStudio data analysis software ( Illumina ) . Average normalization , the Illumina custom error model , and multiple testing corrections using the Benjamini and Hochberg false discovery rate were applied to the analysis . Only transcripts with a differential score of >13 ( p<0 . 05 ) were considered . Normalized and raw data have been deposited in the National Center for Biotechnology Information Gene Expression Omnibus ( accession number GSE125470 ) . Functional enrichment analysis of differentially expressed mRNAs ( 1 . 4 fold or higher ) using overrepresentation analysis ( ORA ) was accomplished using WebGestalt ( http://www . webgestalt . org ) ( Wang et al . , 2017 ) . Characterization of the immune cell composition in these gene expression profiles was accomplished using the computational algorithm ImmuCC ( Chen et al . , 2017b ) . Mice underwent survival surgery to implant EEG and EMG electrodes under isoflurane anesthesia with subcutaneous bupivacaine ( 1 mg/kg ) for analgesia as described ( Kalume et al . , 2013 ) . Using aseptic technique , a midline incision was made anterior to posterior to expose the cranium . Each EEG electrode consisted of a micro-screw attached to a fine diameter ( 130 µm bare; 180 µm coated ) silver wire . The screw electrodes were placed through the small cranial burr holes at visually identified locations: left and right frontal cortices approximately 1 mm anterior to the bregma and 3 mm lateral of the sagittal suture . EMG electrodes were placed in back muscles . A reference electrode was placed at the midline cerebellum and a ground electrode was placed subcutaneously over the back and the skin was closed with sutures . All electrodes connected to a micro-connector system and their impedances were typically <10 kΩ . After electrode placement , the skin was closed with sutures and the mice were allowed to recover from surgery for 2–3 days . Recording approach was performed as described ( Kalume et al . , 2013 ) . Simultaneous video-EEG-EMG records were collected in conscious mice on a PowerLab 8/35 data acquisition unit using Labchart 7 . 3 . 3 software ( AD Instruments , Colorado Spring , Co ) . All bio-electrical signals were acquired at 1 KHz sampling rate . The EEG signals were processed off-line with a 1–80 Hz bandpass filter and the EMG signals with a 3 Hz highpass filter . Video-EEG-EMG data collected were analyzed using Labchart software . Mouse body temperature was controlled using a rectal temperature probe and a heat lamp attached to a temperature controller in a feedback loop ( Physitemp Instruments Inc , NJ ) . Briefly , as described ( Oakley et al . , 2009 ) , body temperature was increased by 0 . 5°C every 2 min until seizure occurred or a 42°C temperature was reached . Mice were immediately cooled using a small fan . Levetiracetam ( Keppra parental formulation ) , perampanel ( Clinisciences ) and carbamazepine ( Sigma-Aldrich ) antiepileptic drugs were injected intraperitoneally in Gad2:Ndufs4cKO mice daily , starting at P40 . Levetiracetam and perampanel were dissolved in saline solution and injected at a dose of 60 mg/kg and 0 . 75 mg/kg , respectively . Carbamazepine was slowly diluted in PEG300 , dissolved in saline , and injected at 40 mg/kg . A control group for each drug was obtained by injecting each respective vehicle to Gad2:Ndufs4cKO mice . qRT-PCR assays were performed as described ( Sanz et al . , 2015 ) . Briefly , equal amounts of RNA were assayed using the Power SYBR green RNA-to-Ct 1-Step Master Mix ( Applied Biosystems ) or the TaqMan RNA-to-Ct 1-Step Master Mix ( Applied Biosystems ) , depending on the system used ( SYBR or Taqman ) . Relative expression values were obtained using the standard curve method and normalized to Actb levels . Amplification efficiencies were calculated using the AriaMx software ( Agilent ) and were within accepted parameters ( 80–120% ) . Gad2 mRNA was determined using a specific Taqman assay ( Mm00484623_m1; Applied Biosystems ) , and sequences for the different primer sets used in SYBR assays ( Slc17a6 and Actb ) were obtained from Primerbank ( Spandidos et al . , 2010 ) . Data are shown as the mean ± SEM . GraphPad Prism v5 . 0 software was used for statistical analyses . Appropriate tests were selected depending on the experimental design as stated in the figure legends . Statistical significance , when reached ( p<0 . 05 was considered significant ) , is stated in the figure legends .

Mitochondria are often described as the power plants of cells because they generate most of the energy that a cell needs to survive . But one in every 5 , 000 children is born with a mutation that leads to faulty mitochondria , which generate less energy than their healthy counterparts . This is particularly problematic for tissues with high energy demands , such as the brain and muscles . Children with such mutations are said to have mitochondrial disease , and one of the most common and severe forms is Leigh syndrome . Children with Leigh syndrome suffer from epilepsy , and have difficulties with movement and breathing . There is no treatment for Leigh syndrome , and most of those affected will die in childhood . The brains of children with Leigh syndrome show a characteristic pattern of damage and inflammation , symmetrical across both hemispheres , with two areas of the brain affected the most . First , the brainstem , which connects the brain with the spinal cord and is responsible for many vital functions such as breathing , maintaining the heart rate or swallowing . Secondly , a group of neurons deep within the brain called the basal ganglia , which has a role in voluntary movement . But although all of a patient’s neurons carry the mutation responsible for their Leigh syndrome , not every neuron is harmed by it . Knowing which neurons are affected , and why , could help develop treatments . Bolea , Gella , Sanz et al . therefore introduced the same Leigh syndrome mutation into different groups of neurons in three groups of mice . The first group had the mutation in the neurons that activate other cells , called glutamatergic or 'go' neurons . The second group had the mutation in the neurons that inhibit other cells , known as GABAergic , or 'stop' , neurons . The third had the mutation in cholinergic neurons , which carry information from the brain to the organs . Examining the mice revealed that having faulty mitochondria in GABAergic neurons from the basal ganglia and in glutamatergic neurons of the brainstem , but not in cholinergic neurons , leads to the symptoms of Leigh syndrome . The fault in the GABAergic neurons causes the epilepsy associated with the syndrome , while faulty mitochondria in the glutamatergic neurons give rise to the observed impairments in movement and breathing . This work could help researchers identify the cellular mechanisms that make neurons more or less resistant to the effects of faulty mitochondria . This in turn will provide a stepping stone to developing new treatments , which can then be tested on the mice developed for these experiments .

2019

Defined neuronal populations drive fatal phenotype in a mouse model of Leigh syndrome

In the human fungal pathogen Cryptococcus neoformans , sex can benefit its pathogenicity through production of meiospores , which are believed to offer both physical and meiosis-created lineage advantages for its infections . Cryptococcus sporulation occurs following two parallel events , meiosis and differentiation of the basidium , the characteristic sexual structure of the basidiomycetes . However , the circuit integrating these events to ensure subsequent sporulation is unclear . Here , we show the spatiotemporal coordination of meiosis and basidial maturation by visualizing event-specific molecules in developing basidia defined by a quantitative approach . Monitoring of gene induction timing together with genetic analysis reveals co-regulation of the coordinated events by a shared regulatory program . Two RRM family regulators , Csa1 and Csa2 , are crucial components that bridge meiosis and basidial maturation , further determining sporulation . We propose that the regulatory coordination of meiosis and basidial development serves as a determinant underlying the production of infectious meiospores in C . neoformans . Sex is pervasive throughout eukaryotes , including fungi . In the ubiquitous human fungal pathogen Cryptococcus neoformans , a model organism for fungal sex studies , sexual reproduction is considered to play an important role in promoting its infections ( Idnurm et al . , 2005; Heitman et al . , 2014 ) . For instance , sexual spores in C . neoformans are presumed infectious particles due to their special physical features , including oxidative stress resistance and small size , which enables compatible deposition in the pulmonary alveoli after inhalation ( Giles et al . , 2009; Velagapudi et al . , 2009; Botts and Hull , 2010; Kozubowski and Heitman , 2012; Ballou and Johnston , 2017 ) . Notably , sporulation in C . neoformans has not been observed during the mitotic life cycle under laboratory condition or in nature but is exclusively associated with sexual ( meiotic ) reproduction ( Kozubowski and Heitman , 2012; Huang and Hull , 2017 ) . This feature is mechanistically different from that of many human fungal pathogens in which asexual reproduction serves as the major route to produce genetically identical spore progenies ( Huang and Hull , 2017 ) . By comparison , due to meiotic recombination , meiospore progenies appear to have more diverse genomes , thereby potentially providing a lineage benefit associated with Cryptococcus infections and drug resistance ( Ni et al . , 2013 ) . C . neoformans has two defined sexual programs underlying sporulation , bisexual and unisexual reproduction ( also named haploid fruiting ) ( Kwon-Chung , 1976; Lin et al . , 2005; Wang and Lin , 2011; Fu et al . , 2015 ) . Bisexual reproduction occurs between cells from two opposite mating types ( α and a ) ( Kwon-Chung , 1976 ) , while unisex only involves cells from a single mating type ( mostly α ) ( Lin et al . , 2005 ) . Both reproduction modes involve similar sequential morphological differentiation and molecular events ( Fu et al . , 2015 ) ( Figure 1A ) . These sexual cycles are initiated by the mating MAPK pathway in response to mating-inducing signals . The activated mating cascade subsequently induces a transition of a subpopulation of yeast cells in the mating community to hyphae , including invasive and aerial hyphae ( Wang et al . , 2014 ) . Aerial hyphae stochastically differentiate into basidia ( also named fruiting bodies ) on their apexes , which represent a hallmark sexua l structure of the phylum Basidiomycota . In many basidiomycetes including C . neoformans , meiosis usually occurs and progresses within basidia , leading to production of four meiotic products ( Kües , 2000; Wang et al . , 2014 ) . The meiotic products undergo repeated rounds of mitosis and yield multiple nuclei that are packaged into meiospores ( basidiospores ) , ultimately generating four chains of spores from the tip of the basidium ( Fu et al . , 2015 ) . As two pre-sporulation events , the formation of basidia and the meiotic cycle individually offer a developmental basis and genetically distinct genomes for the production of meiospores . Thus , cryptococcal sporulation likely requires successful integration of these two events during sexual development ( Figure 1A ) . This hypothesis remains to be proven due to the fact that the genetic basis for orchestrating meiosis and basidial differentiation remains poorly understood in C . neoformans and other basidiomycetes . In this work , we developed a quantitative phenotypic method and identified a novel basidium indicator molecule to define the various stages during basidial differentiation . Using these approaches , we confirmed that basidial maturation and the meiotic cycle are spatiotemporally coordinated in C . neoformans . By profiling gene induction during mating differentiation , we further revealed that the coordination of these events is likely attributed to integrated control mediated by a shared regulatory program . Two RNA recognition motif ( RRM ) family RNA-binding proteins , Csa1 and Csa2 , are the central members that specifically orchestrate meiosis and basidial differentiation , further directing sporulation . Together , our findings provide important insight into how the regulatory coordination of meiosis and basidial development is genetically determined to ensure the formation of resistant meiospores , which are predicted to have both physical and lineage advantages benefiting Cryptococcus infections . The cellular features of the different phases during basidial maturation in basidiomycetes are poorly understood due to the lack of methods to define them . In this regard , we developed a criterion ( BMS: basidial maturation score ) to quantitatively evaluate basidial maturation using the ratio of the diameters between the basidium and its connected hyphae ( Figure 1B ) . In both reproduction modes , the average BMS level gradually increased over time ( Figure 1C ) . This result reflects a dynamic enlargement of basidia during sexual reproduction . In C . neoformans , spores are produced at the tip of basidia after the completion of fruiting body differentiation . Thus , sporulated basidia represent the mature state of basidial development . The BMS of sporulated basidia was significantly higher than that of un-sporulated basidia ( p<0 . 0001 , Kolmogorov-Smirnov test , two sided ) ( Figure 1D ) . Moreover , there was no evident change in the average BMS in the sporulated basidia population throughout both unisexual and bisexual development ( data not shown ) . This finding suggests that , as the final state of basidial maturation , sporulated basidia cannot undergo further enlargement . These data indicate that BMS analysis represents a reliable approach for evaluating basidial maturation in C . neoformans . In this study , a BMS of 1 . 6 was set as the threshold to define the mature state or late stage of fruiting body development , as all the sporulated basidia tested showed a BMS over 1 . 6 ( Figure 1D ) . Next , we sought to evaluate the meiosis activity at different stages during basidial maturation , which were assessed by the BMS method . The meiosis-specific recombinase , Dmc1 , with basidium-specific expression , has been employed as a molecular indicator of meiosis , due to its conserved function in meiotic recombination among divergent eukaryotes and high abundance during the meiotic cycle ( Lin et al . , 2005; Wang et al . , 2014 ) . The fluorescent signal of mCherry-fused Dmc1 was measured throughout basidial maturation . Interestingly , in both sexual cycles , Dmc1 displayed similar expression patterns during basidial maturation ( r = 0 . 84 , p<0 . 05 , Pearson’s test ) . Using a quantitative fluorescence assay , Dmc1 expression was first detected after basidial initiation and dramatically increased during the enlargement of basidia from BMS 1 . 2 to 1 . 6 , while it began to decline as the BMS exceeded 1 . 6 ( Figure 1E ) . These data indicate an explicit correlation between basidial maturation and meiosis-specific gene expression , supporting the hypothesis that meiotic progression and basidial differentiation are spatiotemporally coordinated in C . neoformans . Next , we questioned whether meiotic progression and basidial maturation are genetically associated in C . neoformans and whether this association ensures their coordination in space and time . To tackle this question , we first investigated the effect of meiosis-specific components ( Dmc1 and Spo11 ) on basidial maturation ( Lin et al . , 2005; Feretzaki and Heitman , 2013 ) . Compared with the wild-type strain , the dmc1Δ and spo11Δ mutant showed only a slightly increased population of large basidia during unisexual reproduction ( Figure 1―figure supplement 1 ) . This finding indicates that the absence of these meiosis essential genes cannot impair basidial maturation . Thus , basidial differentiation can be achieved independently of meiotic progression per se . Instead , the coordination of the meiotic cycle and basidial differentiation may potentially be attributed to integrated control by a shared regulatory program that orchestrates these events . Our previous study has shown that the Pumilio family RNA binding protein Pum1 is important for the expression of the meiosis protein Dmc1 and post-meiotic sporulation ( Wang et al . , 2014; Kaur and Panepinto , 2016 ) . This suggested us to test whether Pum1 is also involved in basidial maturation . A violin-plot analysis revealed that disruption of PUM1 led to a significant decrease in mature basidial populations during both unisexual and bisexual development ( 95% CI: 0 . 02 , 0 . 16 , p=0 . 01 for unisex; 95% CI: 0 . 10 , 0 . 29 , p=9 . 1 × 10−5 for bisex; two tailed Student’s-t test ) ( Figure 2A ) . Considering Pum1’s contribution to basidial maturation , we speculated that Pum1 may play an important regulatory role in modulating the expression of the proteins residing in basidia . During sexual development , a highly dynamic expression has been observed in many genes encoding secretory or cell surface proteins , which contain signal peptide destined towards the secretory pathway . Some of these proteins exhibit specificity for enrichment in disparate morphotypes ( yeast , hypha and basidium ) from the Cryptococcus mating community ( Wang et al . , 2014 ) . In a previous microarray study , Pum1 strongly upregulated the expression of multiple genes , whose products contain signal peptides predicted using SignalP and WoLF PSORT programs . These products include Fas1 , Dha1 and Fad1 ( Figure 2B ) ( Wang et al . , 2014 ) . The ample expression of mCherry-labeled Dha1 and Fas1 has been visualized in the fruiting body , while they can also be clearly detected in other morphotypes ( Wang et al . , 2014; Gyawali et al . , 2017 ) . These findings were recapitulated by our fluorescence microscopy analysis ( Figure 2B and Figure 2—figure supplement 1 ) . This result led us to examine whether Fad1 , another important target of Pum1 , displays a similar basidium-enriched expression feature . Indeed , we observed abundant Fad1 in almost all basidia examined . However , in contrast to Dha1 and Fas1 , which can be strongly expressed in other morphotypes in addition to basidium , Fad1-mCherry could not be observed in yeast cells and most hyphae ( Figure 2B and Figure 2—figure supplement 1 ) . Only a small hyphal population appeared to very weakly express this protein ( Figure 2—figure supplement 1 ) . This finding indicates that Fad1 functions as a basidium-enriched protein . In basidia , this protein displayed four major subcellular localization patterns ( Figure 2B and C ) . To our surprise , these patterns were highly dynamic during basidial maturation and were strongly correlated with specific basidial stages . For instance , as basidial development began , Fad1 was observed nearly exclusively in a diffuse form throughout the basidial cytoplasm ( pattern I ) or a condensed form as foci-like structures ( pattern II ) . Subsequently , Fad1 accumulated on the ‘neck’ region connecting the basidia and hyphae ( pattern III ) during basidial enlargement , and was eventually localized to the surface of the upper region of the fruiting body ( pattern IV ) , which was predominant in sporulated basidia or large fruiting body populations with a BMS above 1 . 6 ( Figure 2D and E ) . Considering that the localization patterns of Fad1 are greatly related to specific basidial phases , Fad1 could be used as an indicator to define the various phases during basidial maturation in C . neoformans . Given that Fad1 is enriched in the basidium , Fad1 may play a role in fruiting body differentiation or subsequent sporulation . The detailed phenotypic assays indicated that disrupting FAD1 cannot compromise basidial maturation or morphogenesis but indeed perturbed post-meiotic sporulation . In both reproduction modes , shorter spore chains were produced in the fad1Δ mutant , in which spores from different chains adhered to each other , leading to intertwined spore chains coiling at the top of the basidia ( Figure 2F ) . Thus , Fad1 is required for proper sporulation and spore dispersal . In addition to meiosis and basidial differentiation , Pum1 is also involved in other mating processes , such as filamentation and α-a cell-cell fusion ( Wang et al . , 2014 ) . Thus , Pum1 appears to act as a pleiotropic regulator that coordinates the sequential stages of sexual development rather than specifically connecting meiosis and fruiting body differentiation . To reveal the regulatory program that specifically and critically integrates the meiotic cycle and basidial maturation , we employed a high-coverage strand-specific RNA-sequencing analysis to compare whole-genome expression at five time points ( 6 hr , 12 hr , 24 hr , 48 hr and 72 hr ) throughout unisexual development in the XL280 background ( Figure 3A ) . No later time point was tested because a certain number of basidia have been formed at 72 hr post-mating stimulation ( Figure 1C ) , and the genes activating this process should be transcriptionally induced earlier . We obtained 8177 XL280 unigenes from five time points from either sense or antisense transcripts , which include 6964 genes predicted to encode proteins . These putative protein-coding genes were further aligned against well-annotated genome sequence of another C . neoformans isolate JEC21 to reduce false positives due to variation in gene prediction process ( see Materials and methods for details ) . We identified 2228 protein-coding genes displaying remarkably upregulated expression ( log2|fold-change| > 1 . 0 , q value < 0 . 01 , TPM ( transcripts per million mapped ) >5 ) at least at one time point tested ( Supplementary file 1 ) . These genes were further divided into four groups based on the time point when their transcription was first induced ( Figure 3B ) . We reasoned that each group may consist of sets of genes that specifically activate given molecular or differentiation events . Supporting this hypothesis , we found that the four groups individually contained genes responsible for different processes throughout sexual development ( filamentation , meiosis and sporulation ) ( Figure 3B ) ( Bahn et al . , 2005; Lin et al . , 2005; Wang et al . , 2012; Huang et al . , 2015 ) . A Gene Ontology ( GO ) analysis using BiNGO program was further performed to systemically explore the biological functions of the genes belonging to the different groups . Genes involved in the response to pheromones , which is determined by the pheromone MAPK pathway , are significantly enriched in group I ( Figure 3B ) . In addition , this group also comprised many targets activated by Mat2 and Znf2 , which dominate the Cryptococcus mating response ( early mating process ) and filamentous growth ( middle mating process ) , respectively ( p<0 . 001 , Fisher’s exact test ) ( Figures 3B and 4A ) . These findings support the hypothesis that the genes in group I are likely responsible for or related to early or middle mating events . Among the four groups , group II represents the largest gene set and contains 840 genes , which first displayed expression induction at 24 hr after the mating stimulation . Compared with the transcriptomic data from any other time points , the gene expression profile at 24 hr was remarkably different ( Figure 3A ) . This finding raises the possibility that this time course likely reflects an important developmental switch , which involves highly organized processes . Indeed , this gene set can be further divided into four sub-groups based on detailed expression signature analysis using tree clustering ( Figure 3B and Figure 3—figure supplement 1 ) . The GO analysis identified various GO terms in these sub-groups and , as expected , revealed enrichment of meiosis genes . In addition , these terms also involve cell wall part ( GO: 0044426 ) , fatty acid metabolism ( GO: 0006631 ) and vacuolar protein catabolic process ( GO: 007039 ) , which are potentially related to remodeling cell wall or offering energy and metabolic support during basidial maturation . Compared to the group II genes , genes displaying an initial induction at 48 hr ( group III ) and 72 hr ( group IV ) are only associated with a few biological processes . These processes mainly include rRNA/protein metabolic processes , which may reflect cellular protein turnover in response to stress after extended inoculation on V8 juice agar , a relatively nutrition-restrictive medium . Considering the temporal overlap of meiosis and basidial differentiation , we hypothesized that the genes dedicated to the coordination of these two events are likely included in meiosis gene-enriched group II . In this gene group , we specially focused on ‘regulatory genes’ , whose products contain domains associated with DNA or mRNA binding activities , since DNA or mRNA-binding proteins generally function as core regulatory determinants of meiotic progression and fungal cellular differentiation ( van Werven and Amon , 2011; Wang and Lin , 2012 ) . Based on the InterProScan program predication , 19 predicted regulatory genes were identified in gene group II ( Figure 4—source data 1 ) . We speculated that among these regulatory genes , the ones bridging meiosis and basidial maturation are likely controlled by Pum1 due to its important role in orchestrating these processes . To identify the Pum1-controlled group II regulatory genes , we performed whole-genome RNA-seq profiling at 24 hr after mating induction when group II genes first showed expression induction ( Figure 3B ) . It has been previously shown that Pum1 is almost exclusively expressed in hyphae , which constitute only a minor cell population in a mating colony ( Wang et al . , 2014 ) . To avoid noise caused by massive mating cells that do not or poorly express PUM1 , the PRPL2B-PUM1 strain , which constitutively expresses the PUM1 gene , was applied to the RNA-seq profiling . The profiling revealed 907 differentially expressed genes ( DEGs ) in response to the overexpression of Pum1 ( log2|fold-change| > 1 . 0 , q value < 0 . 01 ) . The number of Pum1 targets explored by the RNA-seq assay was much larger than that identified by the previous transcriptomic assay ( 85 DEGs ) of the PUM1 overexpression strain based on the microarray analysis , which normally has a lower sensitivity and specificity than RNA-seq technology ( Wang et al . , 2014 ) . Besides , it is notable that the previous microarray experiment was performed at 72 hr post mating stimulation when the expression of many group II genes considerably decreased compared to that at 24 hr ( Figure 3B ) . The inappropriate time point used in the previous microarray assay could have led to a failure in comprehensively exploring the group II genes activated by Pum1 . Indeed , only 15 genes from group II were found to be upregulated by Pum1 based on the previous transcriptomic assay . By comparison , the current gene expression profiling approach revealed 94 group II genes induced by Pum1 , indicating a better sensitivity . Moreover , the group II genes are significantly enriched in Pum1-induced regulon ( p=3 . 95 × 10−7 , Fisher’s exact test ) ( Figure 4A ) , and many of these genes are related to the meiotic cycle or sexual sporulation ( Figure 4B ) , supporting the key role of Pum1 in governing these late sexual events . RNA-seq analysis identified eight group II regulatory genes ( Figure 4C ) that displayed remarkably induced expression in response to Pum1 overexpression . These genes included four potential RNA-binding protein coding genes and four genes predicted to produce DNA-binding protein ( Figure 4C ) . We deleted these genes individually , and the impact of the resulting deletion mutants on sequential unisexual differentiation processes ( hyphal initiation , hyphal extension and sporulation ) were assessed by quantitative or semi-quantitative phenotypic approaches ( Figure 4D , Figure 4―figure supplement 1 ) . We speculated that the regulator involved in the coordination of meiosis and basidial maturation must be important for downstream sporulation but does not affect the earlier differentiation processes , such as filamentation . Our phenotypic assays showed that four of these eight regulators ( 50% ) are involved in post-meiotic sporulation ( Figure 4D , Figure 4—figure supplement 1 ) . Among these four regulators , CNA00260 , CNJ00760 , and CNB02060 are strictly required for the formation of spores , and we did not observe spore formation in these mutants even after extending the incubation time up to one month . CNA00260 , which encodes a ZnF_GATA DNA-binding family protein , exerted a dramatic effect not only on sporulation but also on filamentation , suggesting that it is not specific to Cryptococcus fruiting . In contrast , deletion of CNJ00760 or CNB02060 prevented formation of meiospores and led to self-filamentation with an abundance similar to the wild-type level during unisexual reproduction ( Figure 4D , Figure 4—figure supplement 1 , Figure 4—figure supplement 2 ) . Similarly , these two genes are also critical for bisexual sporulation in bilateral mating assays ( Figure 5A ) and dispensable for bisexual filamentation , which was remarkably defective in bilateral crosses of pum1Δ ( α pum1Δ × a pum1Δ ) ( Figure 5—figure supplement 1 ) . Domain searches using the Motif scan and InterProScan programs revealed that the products of both CNJ00760 and CNB02060 belong to the RRM RNA-binding protein family ( Glisovic et al . , 2008 ) . We named these genes CSA1 ( CNJ00760 ) and CSA2 ( CNB02060 ) ( Cryptococcus sporulation activator ) . To test whether CSA1-activated or CSA2-activated sporulation is unique to the XL280 ( serotype D ) background , CSA1 and CSA2 were individually mutated in the JEC21 ( serotype D ) and H99 ( serotype A ) backgrounds . Both the serotype D and serotype A strains belong to the Cryptococcus neoformans species complex and are considered to have diverged from each other for at least 18 . 5 million years ( Xu et al . , 2000 ) . Expectedly , the absence of either of Csa1 or Csa2 in these Cryptococcus strains abolished sporulation ( Figure 5A , Figure 5—figure supplement 2 ) . These data demonstrated that the requirement of Csa1 and Csa2 for the formation of meiospores is not limited to XL280 but is conserved among strains in the C . neoformans species complex . The RT-PCR analysis of the mRNA levels of CSA1 and CSA2 at seven different time points post unisexual induction indicated that their gene expression patterns during unisexual reproduction significantly overlapped ( r = 0 . 96 , p=6 . 4 × 10−5 , Pearson’s test ) ( Figure 5B ) , which is suggestive of co-regulation . This concept was further supported by the transcriptional evidence showing that the mRNA levels of CSA1 and CSA2 were co-upregulated by Pum1 ( Figure 5C ) . Next , we asked if defective sporulation in the csa1Δ and csa2Δ mutants is due to failure in the orchestration of meiosis and basidial maturation , particularly given the important effect of their upstream regulator Pum1 on these two events . To address this question , we first detected the expression of Dmc1-mCherry in the csa1Δ and csa2Δ mutants during unisexual reproduction . The fluorescence signal was undetected when either of these two genes was absent , suggesting that they are both required for meiotic progression ( Figure 5D ) . Early studies have shown that disruption of meiosis-specific genes causes a greatly reduced number of spores or spore chains but cannot completely prevent sporulation , which can be otherwise achieved by the deletion of CSA1 or CSA2 ( Figures 4D and 5A ) . This finding highlights the possibility that the blocked sporulation observed in the csa1Δ and csa2Δ mutants is not only due to defective meiosis but also involves additional mechanism . This idea suggested us to examine whether CSA1 and CSA2 affect the formation of the basidium , which offers the physical base underlying the formation of spore chains . The BMS assay indicated that deleting either CSA1 or CSA2 dramatically dampened basidial formation and enlargement ( maturation ) during both unisexual and bisexual reproduction ( Figure 2A ) . Basidia , especially mature basidia ( BMS >1 . 6 ) , were reduced to a much lower level in the csa1Δ and csa2Δ mutants compared with those in the wild-type strain and even the pum1Δ mutant ( Figures 2A and 6A ) . To gain a further insight into the effect of Csa1 and Csa2 on basidial maturation , we investigated the csa1Δ and the csa2Δ mutants for the patterns of the subcellular localizations of Fad1 , which can be used to define the different phases during fruiting body maturation , particularly the early ( pattern I and II ) and late/mature phases ( pattern IV ) ( Figure 2C–E ) . In the csa1Δ mutant , ~53 . 3% and~34 . 1% of basidia exhibited patterns identical to patterns I and II , respectively , after 7 days incubation on mating-inducing V8 medium . These frequencies were much higher than those detected in the wild-type strain ( I =~4 . 7% and II =~9 . 7% ) . Furthermore , up to ~61 . 9% of basidia in the wild-type XL280α strain achieved the late stage represented by localization pattern IV , while this pattern was detected in only 2 . 7% of basidia in the absence of Csa1 ( Figure 6B ) , suggesting that Csa1 is important for Cryptococcus basidial maturation . Despite the dramatic change in the localization features of Fad1 caused by the deletion of CSA1 , Csa1 cannot control Fad1 expression as revealed by our quantitative fluorescence imaging analysis ( Figure 6—figure supplement 1 ) . Unlike Csa1 , Csa2 contributed to the full expression of Fad1-mCherry , whose abundance was greatly decreased in the mutant lacking CSA2 ( Figure 6—figure supplement 2A ) . Consistently , Csa2 is important for the transcription of FAD1 during both bisexual and unisexual mating , demonstrating its role as an upstream regulator of FAD1 ( Figure 6—figure supplement 2B ) . Moreover , the csa2Δ mutant was nearly devoid of the maturation-represented pattern of Fad1-mCherry ( pattern IV ) and almost exclusively exhibited early stage patterns ( I =~84 . 3% and II =~13 . 6% ) ( Figure 6B ) . We next performed a detailed phenotypic analysis for evaluating the impact of Csa1 and Csa2 on basidial morphogenesis . Surprisingly , a vast majority of basidia exhibited aberrant morphologies in the absence of Csa1 or Csa2 during unisexual mating ( csa1Δ:~72 . 2% and csa2Δ:~82 . 4% ) ( Figure 6C ) . Most of the irregular basidia in the mutants displayed a ‘snake-head’-like phenotype . ( Figure 5A ) . In contrast , ‘cup-shaped’ or ‘spindle-shaped’ basidia were usually observed in the wild-type XL280 strain in which only ~2 . 8% of basidia showed morphological abnormalities . Collectively , our findings suggest that Csa1 and Csa2 are critical for basidial formation , maturation and morphogenesis . The significance of both Csa1 and Csa2 in the coordination of meiosis and basidial differentiation led us to investigate their genetic interactions . First , we assessed the reciprocal effect of the disruption of one gene on the transcript level of the other . The RT-PCR analysis indicated that the two regulators did not appear to affect the expression of each other , indicating that they may function in parallel ( Figure 6—figure supplement 3 ) . To further address this hypothesis , we simultaneously deleted CSA1 and CSA2 in the XL280 background . The resulting csa1Δ/csa2Δ mutants were applied to compare to the effect of each of the single-deletion mutants on the spatiotemporal expression of meiosis and basidium indicators , as well as basidial maturation and morphogenesis . The expression of Dmc1-mCherry remained undetectable in the csa1Δ/csa2Δ mutant ( Figure 5D ) . This result was expected because deletion of either genes was sufficient to block the expression of Dmc1 ( Figure 5D ) . Of note , compared with either of the single deletion , the double-deletion resulted in a significantly smaller population of large basidia ( BMS >1 . 6 ) , particularly during bisexual mating ( Figures 2A and 6A ) . Consistent with this finding , a modestly higher frequency of irregular basidia was detected in the double-deletion mutant than in either of the single deletion mutants ( Figure 6C ) . Furthermore , in the csa1Δ/csa2Δ mutant , we failed to observe the mature basidium-specific pattern IV and intermediate pattern ( III ) , which could be detected in the single deletion mutants , although at a much lower level than that in the wild-type strain ( Figure 6B ) . In the double-deletion mutant , Fad1-mCherry exclusively displayed the localization features indistinguishable from the patterns reflecting the early stage of basidial differentiation ( I =~93 . 8% and II =~6 . 2% ) . These data suggest that Csa1 can function in concert with Csa2 during fruiting body maturation . In many microbes , genetically identical cells from a sibling community can be remarkably distinct in cellular shape or physiology ( Mitri and Foster , 2013; Wang and Lin , 2015 ) . Such heterogeneity underlies a source of diversity maximizing microbial survival under numerous environmental and host stresses . For instance , in C . neoformans , cells with different morphologies co-exist during the differentiation of the mating community ( McClelland et al . , 2004; Wang and Lin , 2015 ) . These morphotypes differ in their tradeoffs related to virulence potential and resistance to given natural or host stress ( Wang and Lin , 2015 ) . Among these morphotypes , spores , as the final products of mating community differentiation , are considered important infectious propagules due to their resistant nature and size , which is compatible with alveolar deposition ( Sukroongreung et al . , 1998; Giles et al . , 2009; Velagapudi et al . , 2009 ) . Considering its importance in terms of Cryptococcus biology and infections , important efforts have been undertaken to identify new genes engaged in sporulation ( Bahn et al . , 2005; Liu et al . , 2011; Feretzaki and Heitman , 2013; Wang et al . , 2014; Huang et al . , 2015; Huang and Hull , 2017 ) . However , much less is known about the regulatory determinant and commitment factor underlying the formation of spores in this important fungal pathogen . During sexual development , spores are stochastically formed after two parallel events , basidial maturation and meiotic progression . Early studies have reported that sporulation can be perturbed in mutants lacking genes dedicated to meiosis , such as DMC1 and SPO11 ( Lin et al . , 2005; Feretzaki and Heitman , 2013 ) . This observation suggests that the successful completion of the meiotic cycle is important for sporulation . Notably , mutation of these meiosis-specific genes greatly impairs sporulation but cannot completely abolish it . This finding strongly suggests that C . neoformans also involves other commitment mechanism to form spores . The basidium , a hallmark structure of the phylum Basidiomycota that comprises more than 30 , 000 species , physically supports the formation of spore chains during sexual development ( Kües , 2000; Kües and Liu , 2000; Wang and Lin , 2011; Fu et al . , 2015 ) . Accordingly , basidial maturation may serve as a commitment process for spore production . We showed that meiosis and basidial maturation are coordinated spatiotemporally during both unisex and bisex in C . neoformans ( Figure 1E ) . Profiling gene induction during mating differentiation further unveiled a special gene group ( group II ) potentially responsible for the coordination of these two processes ( Figure 3B ) . Gene Ontology analysis identified a strong enrichment of cell wall-related genes in this group ( Figure 3B ) . This finding probably mirrors the dynamic remodeling of cell wall components during basidial differentiation in C . neoformans , particularly given that the re-organization of the cell wall is normally associated with fungal cellular differentiation ( Wang and Lin , 2012 ) . In addition , multiple genes from group II encode enzymes involved in fatty acid metabolism , including three paralogous genes ( CNA05200 , CNL05760 and CNF04660 ) that are predicted to produce peroxisomal/mitochondrial carnitine acetyltransferase ( CAT ) . During fatty acid β-oxidation , CAT is a key enzyme that mediates acetyl-carnitine shuttle to enable the production of energy via the TCA cycle ( Strijbis and Distel , 2010 ) . An early study has shown that mutants lacking CAT displayed an attenuated production of fruiting body in the cereal fungal pathogen Gibberella zeae , suggesting an important role of fatty acid catabolism during sexual structure formation ( Son et al . , 2012 ) . Considering its significance for the production of energy and metabolic intermediates , fatty acid metabolism likely contributes to sustaining fruiting body by providing energy and metabolic supply . Furthermore , the genes associated with lipid/fatty acid metabolism have also been found to be induced during fruiting body differentiation in other basidiomycetes , such as in Schizophyllum commune and Lentinula edodes ( Ohm et al . , 2010; Wang et al . , 2018 ) . This may be indicative of the conserved importance of fatty acid metabolism during fruiting body development in divergent fungi . Despite temporal overlap between meiosis and fruiting body differentiation , the absence of meiosis-specific Dmc1 or Spo11 does not affect basidial initiation and maturation ( Figure 1―figure supplement 1 ) . This finding demonstrates that basidial differentiation can be achieved independently of meiotic progression and that a shared regulatory program may be responsible for the coordination of these events ( Figure 7 ) . This hypothesis was confirmed by the identification of the regulatory circuitry formed by Pum1 , Csa1 and Csa2 , which are the targets of Pum1 ( Figure 7 ) . Compared with their upstream regulator , Csa1 and Csa2 are more specific and essential for directing the co-regulation of meiosis and basidial differentiation ( Figure 4E , Figure 4—figure supplement 1 , Figure 4—figure supplement 2 , Figure 5A and Figure 5—figure supplement 1 ) . The domain prediction analysis indicated that both Csa1 and Csa2 belong to the RRM protein family . In addition to Csa1 and Csa2 , many RRM family members in different eukaryotic kingdoms have been reported to control sexual development or meiosis , but most of them are not similar in their protein sequences , except for Mei2 and its homologs ( Jeffares et al . , 2004 ) . Mei2 has been demonstrated to be the master regulator of meiosis in Schizosaccharomyces pombe , and the genes encoding its orthologs were found in several groups of eukaryotes ( Jeffares et al . , 2004 ) . Thus , MEI2-like genes are considered to have arisen early in the eukaryotic evolution . Based on blast analysis , Csa2 , but not Csa1 , displays significant similarity with Mei2 in the C-terminal RRM motif . The functions conducted by Csa2 and its ortholog in S . pombe during sexual differentiation are not identical , although both are required for meiosis . Csa2 governs basidium formation in C . neoformans ( Figures 2A and 5D ) . In contrast , Mei2 appears not to be required for the formation of ascus ( Nakayama et al . , 1985 ) , which is the sexual structure of S . pombe analogous to basidium in C . neoformans , suggesting a divergent evolution . Consistently , accumulating studies on Mei2-like proteins in plants have demonstrated that their functions are not limited to meiosis but also associated with other biological processes , such as leaf development and vegetative growth ( Kaur et al . , 2006; Kawakatsu et al . , 2006 ) . The functional divergence is very common among orthologs of mRNA binding proteins , and is likely achieved through altered downstream targets or interaction partners ( Hogan et al . , 2015 ) . During meiosis in fission yeast , Mei2 binds the noncoding RNA meiRNA and sequesters Mmi1 , an inhibitor of meiosis , through Mei2-Mmi1 interactions ( van Werven and Amon , 2011 ) . However , there is no gene from C . neoformans genome showing significant homology to the ones in S . pombe that produce meiRNA or Mmi1 . The divergence of targets controlled by Mei2 and Csa2 is probably attributed to considerable differences between their sequences beyond the C-terminal RRM motif . Consistent with this notion , the region 429–532 upstream of the C-terminal RRM contains two residues ( Ser 438 and Thr 527 ) for phosphorylation by the kinase Pat1 and is essential for the function of Mei2 ( Watanabe et al . , 1997 ) , but this region is missing from Csa2 protein . Blast analysis indicated that Csa2 orthologs from different basidiomycetes share a high similarity in the full protein sequence ( greater than 50% coverage ) . Intriguingly , most fungi harboring Csa2-like protein coding genes also have genes producing Csa1 homologs ( greater than 30% identity and greater than 50% coverage ) . Phylogenetic analysis demonstrated that the fungal species containing both CSA1-like and CSA2-like genes belong to the Tremellales clade ( Figure 7—figure supplement 1 ) . This may suggest the conserved and concerted function of these homologs in coordinating basidial maturation and meiotic progression in Tremellales . The strains used in this study are listed in the Key Resources Table . Cryptococcus yeast cells were cultured on YPD solid medium ( 1% yeast extract , 2% Bacto peptone , 2% dextrose , and 2% Bacto agar ) at 30°C for routine growth . Unisexual and bisexual mating assays were carried out on V8 solid medium ( 0 . 5 g/liter KH2PO4 , 4% Bacto agar , and 5% V8 juice ) in the dark at 25°C ( V8 pH seven agar for serotype D strains and V8 pH five agar for serotype A H99 strain ) . YPD media containing nourseothricin ( NAT ) , G418 ( NEO ) or hygromycin ( HYG ) were used for selecting the Cryptococcus transformants generated by electroporation and biolistic transformation . For bisexual filamentation and sporulation assays , congenic α and a cells ( XL280α/a , JEC21α/a and H99α/a ) were cultured on YPD medium separately overnight at 30°C . Cells were then collected by centrifugation . Equal numbers ( OD600 = 1 . 0 ) of collected cells from opposite mating types were co-incubated on V8 medium in the dark at 25°C for mating stimulation . For self-filamentation and unisexual sporulation assays ( serotype D XL280α strain background ) , the cells were spotted on V8 medium alone . Both bisexual and unisexual mating phenotypes were examined microscopically for production of mating hyphae and chains of basidiospores . For the BMS analysis , α cells alone ( unisexual mating ) or α-a cell mixtures ( α-a bisexual mating ) were cultured on V8 medium in the dark at 25°C to stimulate mating . In most BMS assays performed in this study , the cells were harvested at 7 days post mating stimulation , unless otherwise indicated . In both reproduction modes , the cells displayed evident heterogeneity in morphotypes ( mostly yeast and hyphae ) , and ample hyphae are normally formed on the edge of the mating colony , especially during unisexual mating . Regardless of their morphotypes , the cells were entirely scraped off the edge of mating patches to avoid bias . All cells were harvested , vortexed and suspended in 20 μl fixative ( 3 . 7% formaldehyde and 1% Triton X-100 in PBS buffer ) . Mating cells ( 2 μl ) were subsequently dropped onto a glass slide for microscopic examination . Among the cells , most hyphae tended to form ‘hyphal clusters’ due to cell aggregation . Over 100 hyphae with or without basidia from different ‘clusters’ in each sample were randomly chosen for the BMS calculation . A BMS of 1 . 0 was arbitrarily set as the threshold to define the basidial state . In the BMS assays of sporulated basidia that constitute a minority of the basidial population , at least 50 basidia were examined for each sample . The diameters of basidia and their connected hyphae were measured using a Zeiss Imager A2-M2 imaging system with AxioCam MRm camera software Zen 2011 ( Carl Zeiss Microscopy ) . For the gene disruption , overlapping PCR products were generated with a NEO or NAT resistance cassette and 5′ and 3′ flanking sequences ( 1 . 0 ~ 1 . 5 kb ) of the coding regions of selected genes from Cryptococcus strains as we previously described ( Wang et al . , 2012; Wang et al . , 2013 ) . The PCR products were introduced into the Cryptococcus strains via biolistic transformation . The resulting mutants , in which the genes were replaced by homologous recombination , were confirmed by PCR . For Pum1 overexpression , PUM1 gene open reading frame were amplified by PCR , and the amplified fragments were digested with FseI and PacI . The digested fragment was then introduced into the copper-inducible plasmid pXC ( Wang et al . , 2013 ) . The plasmid was digested with Not1 and FseI to remove the copper-inducible promoter ( PCTR4 ) , which was replaced by the promoter region of RPL2B by ligation to generate the PRPL2B-PUM1 overexpression system . The primers used for the gene disruption and overexpression are listed in the Key Resources Table . The mCherry protein was fused to the C terminus of Dmc1 , Fad1 , Fas1 , and Dha1 . The coding regions of the mCherry-fused products were placed under the control of their native promoters . The constructs were introduced into Cryptococcus cells using electroporation ( Wang et al . , 2012 ) . The strains LL174α ( PDMC1-DMC1-mCherry ) and LL168α ( PFAD1-FAD1-mCherry ) were subsequently used as the parental strains to generate the isogenic mutant strains ( HG516α , LL178α , HG519α and LL194α ) , in which selected genes were knocked out . The strains used in this study are listed in the Key Resources Table . To examine the protein subcellular localization , the cells were placed onto glass slide and visualized by a Zeiss Axioplan two imaging system with AxioCam MRm camera software Zen 2011 ( Carl Zeiss Microscopy ) . SEM was performed with the assistance of a Beijing Regional Center of Life Science Instrument , Chinese Academy of Sciences . The samples were prepared for SEM as previously described ( Fu and Heitman , 2017 ) . For all SEM assays performed in this study , the cell samples were obtained from bilateral mating . The α-a mixtures were cultivated on V8 solid medium at 25°C for 7 days in the dark . The colonies were excised and fixed in phosphate-buffered glutaraldehyde ( pH 7 . 2 ) at 4°C for more than 2 hr . Samples were then rinsed with ddH2O three times for 6 min , 7 min and 8 min , respectively . The rinsed samples were dehydrated through a graded ethanol series ( 50% , 70% , 85% and 95% ) for 14 min for each concentration and then 100% ethanol three times for 15 min . After dehydration , the cells were critical-point dried with liquid CO2 ( Leica EM CPD300 , Germany ) and sputter coated with gold-palladium ( E-1045 ion sputter , Hitachi , Japan ) . The samples were viewed under a Quanta200 scanning electron microscope ( FEI , America ) . For the RNA-seq analysis , the wild-type XL280 strain and isogenic Pum1 overexpression mutants were cultured in YPD liquid medium ( extremely mating-suppressing condition ) at 30°C overnight . The overnight culture was then washed with cold water and dropped on V8 agar ( pH = 7 ) for mating induction . The cells were collected at different time points post mating stimulation for the isolation of total RNA . The total RNA was extracted using TRIzol Reagent ( CW0580M , CWBIO ) and an Ultrapure RNA Kit ( CW0581M , CWBIO ) according to the manufacturer’s instructions . Total RNA ( 2 μg ) was subjected to gDNase treatment , and single-stranded cDNA was synthesized by a Fastquant RT Kit ( with gDNase , KR106-02 , Tiangen ) according to the manufacturer's instructions . The relative mRNA level of selected genes was measured by real time RT-PCR using RealMaster Mix ( SYBR Green , FP202-02 , TIANGEN ) in a CFX96 TouchTM Real-time PCR detection system ( Bio-Rad ) . The primers used for qPCR in this study are listed in the Key Resources Table . The relative transcript levels were normalized to those of the reference housekeeping gene TEF1 and determined using the 2-ΔΔCT approach . The total RNA from each sample was purified as previously described ( Wang et al . , 2012 ) . RNA purity was assessed using a Nano Photometer spectrophotometer ( IMPLEN , CA , USA ) , and the RNA concentration was measured using Qubit RNA Assay Kit in a Qubit 2 . 0 Fluorometer ( Life Technologies , CA , USA ) . The RNA integrity was assessed using the RNA Nano 6000 Assay Kit of a Bioanalyzer 2100 system ( Agilent Technologies , CA , USA ) . The transcriptome library for sequencing was generated using a VAHTSTM Stranded mRNA-seq v2 Library Prep Kit for Illumina ( Vazyme Biotech Co . , Ltd , Nanjing , China ) following the manufacturer's recommendations . The clustering of the index-coded samples was performed using VAHTS RNA Adapters set1/set2 for Illumina ( Vazyme Biotech Co . , Ltd , Nanjing , China ) according to the manufacturer's instructions . After clustering , the libraries were sequenced on an Illumina platform . The raw images were transformed into raw reads by base calling using CASAVA ( http://www . illumina . com/support/documentation . ilmn ) . Then , the raw reads in a fastq format were first processed using in-house Perl scripts . Clean reads were obtained by removing the reads with adapters , such as the reads in which unknown bases exceeded 5% . The low-quality reads were defined by a low-quality base , and the sequencing quality score was no more than 10 . Additionally , the Q20 , Q30 , and GC contents of the clean data were calculated . The quality of sequenced clean data was assessed using FastQC v0 . 11 . 5 software . Then , ~2 GB clean data for each sample ( representing over 100 x coverage ) were mapped to the genome sequence of C . neoformans XL280 ( XL280α ) using Hisat2 v2 . 1 . 0 . The gene expression level was measured in TPM by Stingtie v1 . 3 . 3 to determine unigenes . All unigenes were subsequently aligned against the well-annotated genome of JEC21α ( which is congenic to JEC20a that served as the parent strain to generate XL280α through a cross with B3501α ) . The protein coding genes found in both genomes of JEC21α and XL280α were kept for the following bioinformatics analysis . The DEGs were assessed using DEseq2 v1 . 16 . 1 of the R package and defined based on the fold change criterion ( log2|fold-change| > 1 . 0 , q value < 0 . 01 ) . The gene ontologies and P-values of the GO terms were calculated by BiNGO v3 . 0 . 3 using a hypergeometric distribution with Benjamini-Hochberg false discovery rate ( FDR ) correction . In all RNA-seq assays performed in this study , two biological replicates were included . Statistical analyses were performed using R , version 3 . 4 . 2 , and the statistical tests are indicated in the corresponding figure legends or Results section . We used two-tailed Student’s t-test to compare the mean florescence intensity or transcript levels between two groups . Fisher’s exact test was utilized to evaluate the significance of the overlap between two sets of genes . A two-sided Kolmogorov-Smirnov test was used to verify the normality of the distribution of the continuous variables . *p-values<0 . 05 were considered significant , and ***p-values<0 . 001 were considered very significant . In all figures , the error bars represented the mean ±standard deviation ( SD ) from at least three independent experiments .

Many microbes that cause disease form spores to survive during and between infections . These include the fungus Cryptococcus neoformans , which is the leading cause of fungal meningitis worldwide . This fungus produces spores via sexual reproduction , meaning the genes from two living strains of the fungi combine to create new lives with unique genetics . By diversifying the fungus’s genetics , sexual reproduction in Cryptococcus is considered to accelerate drug resistance . Several processes must be coordinated for Cryptococcus to reproduce sexually . Genetic information recombines through a process called meiosis , the spore-making cell ( known as the sexual structure ) matures and later spores are produced . Scientists have identified many genes involved in each of these processes . Yet it is not known how these processes are coordinated to ensure the proper sequence of events . Liu , He , Chen et al . studied the physical changes in Cryptococcus cells when they lost certain genes . Two genes , which the researchers named CSA1 and CSA2 , were found to regulate the parallel progression of meiosis and maturation of the sexual structure . Both processes need to be complete before spore production begins . Further investigation showed that these genes are important across various strains of infectious Cryptococcus . This research highlights sexual reproduction as a target to stop Cryptococcus forming spores and starting infections . The results also show that these processes change little through evolution within a large group of fungi . The next step will be to see how these systems operate across species and the effect this has on spore production .

2018

Genetic basis for coordination of meiosis and sexual structure maturation in Cryptococcus neoformans

Visual predators rely on fast-acting optokinetic responses to track and capture agile prey . Most toothed whales , however , rely on echolocation for hunting and have converged on biosonar clicking rates reaching 500/s during prey pursuits . If echoes are processed on a click-by-click basis , as assumed , neural responses 100× faster than those in vision are required to keep pace with this information flow . Using high-resolution biologging of wild predator-prey interactions , we show that toothed whales adjust clicking rates to track prey movement within 50–200 ms of prey escape responses . Hypothesising that these stereotyped biosonar adjustments are elicited by sudden prey accelerations , we measured echo-kinetic responses from trained harbour porpoises to a moving target and found similar latencies . High biosonar sampling rates are , therefore , not supported by extreme speeds of neural processing and muscular responses . Instead , the neurokinetic response times in echolocation are similar to those of tracking responses in vision , suggesting a common neural underpinning . Response speed critically determines the outcome of interactions between mobile prey and pursuit predators . Prey must react rapidly to survive while predators must counter evasive prey movements quickly to gain sustenance . The fitness implications of these interactions have led to an evolutionary escalation of response times with the fastest-responding individuals being the most likely to survive and reproduce ( Dawkins and Krebs , 1979 ) . However , sensory and motor requirements are asymmetric for predators and prey . Responding to close-approaching predators , prey may trade accuracy for speed , relying on imprecise ballistic motor actions triggered by strong sensory cues that require little neural processing ( Domenici and Batty , 1997; Turesson et al . , 2009 ) . This has led to extremely fast responses employing short efferent pathways linking sensors directly to muscles , such as the Mauthner-cell-mediated C-start response of teleost fish to fluid motion from oncoming predators ( Eaton and Hackett , 1984 ) . C-start responses are characterised by sudden accelerations ( Domenici and Blake , 1997 ) and unpredictable trajectories ( Moore and Biewener , 2015 ) , with response latencies as low as 5–10 ms ( Eaton et al . , 1977 ) . In contrast , predators must sacrifice speed for accuracy , typically requiring greater sensory resolution and motor-planning capabilities to track and successfully pursue evasive prey . The increased processing needed to locate prey in complex natural scenes , along with the typically larger body size of predators , inevitably results in slower movement responses compared to prey , and these are often partly offset by ingenious capture tactics ( Catania , 2009; Wiley et al . , 2011 ) , cooperative hunting ( Domenici et al . , 2000; Benoit-Bird and Au , 2009 ) , or sensory/cognitive superiority ( Bailey et al . , 2012 ) . For most macro-predators , binocular vision is the predominant sensory modality for hunting , and prey movements are tracked within complex visual scenes by a combination of smooth and stepped ( saccadic ) movements of the eye muscles that manipulate gaze direction and depth of view ( Land , 1999 ) . This dynamic tracking is achieved by a set of optokinetic responses in humans and other primates with latency of 50–250 ms ( Miles et al . , 1986; Erkelens , 2006; Kirchner and Thorpe , 2006 ) that have been described as ultra-fast ( Girard et al . , 2008 ) . The evolution of echolocation in bats and toothed whales has allowed erstwhile visual predators to occupy foraging niches with low light levels such as deep or murky waters or at night . In contrast with vision , which relies on exogenous continuous light energy , echolocation is a discrete-time active sense in which ultrasonic pulses are used to sample the environment . These intense sounds potentially offer an auditory cue to prey , and many insects targeted by bats have developed ultrasonic hearing provoking an acoustic arms race ( Goerlitz et al . , 2010 ) . In contrast , very few marine organisms have ultrasonic hearing ( Wilson et al . , 2007 ) . This allows echolocating toothed whales in dark waters to approach most prey without being detected until close enough that prey can sense them via hydromechanical cues ( Wilson et al . , 2018 ) . Such short-range detection necessitates rapid responses from prey which must be countered by fast biosonar-informed locomotory adjustments if toothed whales are to capture agile prey . Echolocators control information flow , in the form of returning echoes , by the rate at which sound transients are produced to probe the environment . Both bats and toothed whales appear to avoid ambiguous echo information by adjusting their sampling interval , effectively expanding and contracting the acoustic depth of field , to accommodate changes in the two-way acoustic travel time as they approach targets ( Madsen and Surlykke , 2013; Stidsholt et al . , 2021 ) . Clicks are accordingly produced slowly for long-range echolocation ( long sensing range ) , but more rapidly and with lower intensity ( short sensing range ) for tracking nearby targets ( Griffin , 1958; Wisniewska et al . , 2014; Salles et al . , 2020 ) . In a remarkable convergence , toothed whales and bats both use rapidly accelerating series of clicks , known as buzzes , when approaching prey ( Madsen and Surlykke , 2013 ) , thereby obtaining enhanced temporal resolution to track evasive prey at the expense of a short sensing range ( Wisniewska et al . , 2016 ) . In analogy with engineered sonars , the working model for animal biosonar assumes that neural echo processing occurs on a click-by-click basis in which the next click is produced after echoes from the current click are detected and processed ( Au , 1993 ) . During buzzes , bats can click at up to about 160/s ( Elemans et al . , 2011 ) requiring very short neural latencies to maintain such click-synchronised processing . However , in toothed whales , the click rates attained during buzzes ( >500/s in porpoises , Wisniewska et al . , 2012 ) would require neural and muscular responses two orders of magnitude faster than primate visual reflexes ( Kawano , 1999 ) to keep up with the information flow . This raises the question of whether toothed whales have an extreme processing capability for acoustic information or if they instead integrate echo information over successive clicks ( Kothari et al . , 2018; Ladegaard et al . , 2019 ) giving a processing time more similar to vision . In the latter case , why produce so many clicks during close approaches ? The modulation of clicking rate in echolocation to control information flow serves a similar function to short-latency eye movements in vision , leading us to hypothesise that echo information must guide two inter-related control loops during hunting ( Figure 1 ) : a kinematic response loop controls the heading , posture , and locomotor rate so as to intercept prey , while a sensor-motor response loop maintains attention fixed on the moving target by continual adjustment of the clicking rate . On that basis , we predict that sudden prey movements during close approaches should provoke tightly coupled mechanical and sensor-motor responses in echolocating animals similar to those shown by visual predators . However , despite echolocation being the main sensory mode for hunting in one of four species of mammals ( Madsen and Surlykke , 2013 ) , the sensory feedback that echolocating predators receive from movements of their prey has received little attention . Wisniewska et al . , 2016 reported clicking rate dynamics of wild harbour porpoises during buzzes that appeared to be associated with prey movements but offered no analysis . The only reported study of predator responses during echolocation buzzes is for wild bats approaching suspended prey that were moved by hand . Rather than adjust clicking rates , these bats aborted buzzes some 100 ms after strong target movements ( Geberl et al . , 2015 ) , perhaps indicative of a neural processing delay , but the decision to end a buzz may well result from different neural processing than that involved in prey tracking . The scant literature on sensory feedback in echolocation owes much to the inherent difficulty of measuring simultaneously the motor responses of predators and movements of prey at high time resolution . This has seldom been achieved outside of instrumented enclosures where movements can be tracked with high-speed cameras , but where ecologically relevant stimuli are hard to emulate . Echolocating toothed whales provide an excellent model system in which to measure the coupled kinetics of free-ranging predators and prey . Sound and movement logging tags ( DTAGs; Johnson and Tyack , 2003 ) record the outgoing echolocation clicks of toothed whale species and , for some species , also detect returning echoes from prey ( Johnson et al . , 2004; Wisniewska et al . , 2016 ) . They simultaneously record the fine-scale movements of the tagged animal , allowing quantification of both the sensory and locomotor responses of predators to movements of their prey . Crucially , these tags sample the sensory scene at exactly the same rate as it is acquired by the animal , that is , the rate at which clicks are produced . Here we used high-resolution DTAGs on two species of echolocating toothed whales living in very different habitats to study biosonar responses to prey movements in the wild . Specifically , we tested the hypotheses that prey movements trigger neuromotor feedback during buzzes , and that this feedback operates at the extreme speeds needed to keep pace with the high clicking rate in buzzes . We show that both species make stereotyped biosonar adjustments when prey attempt to escape during close approaches , but the apparent latency of these echo-kinetic responses is much longer than the inter-click intervals ( ICIs ) in buzzes . Given the evolutionary importance of such a feedback system , we further hypothesised that it would be stimulated by any rapidly moving target during a close approach , facilitating controlled studies of biosonar responses . To test this , we trained harbour porpoises to approach a moveable target while wearing a biologging tag , enabling direct measurement of biosonar-mediated sensory and kinematic response latencies as a function of target movement . Echograms visualising the acoustic scene during prey capture buzzes in wild harbour porpoises ( Phocoena phocoena , abbreviated to Pp ) and Blainville’s beaked whales ( Mesoplodon densirostris , abbreviated to Md ) frequently show evidence of evasive prey that launch sudden escape attempts as the predator approaches ( Figure 2A and B , Figure 2—figure supplements 1 and 2 ) . Prey that accelerate away from the predator can quickly move beyond the acoustic depth of field ( i . e . , the buzz ICI times one half of the sound speed ) requiring a rapid increase in ICI by the predator to avoid ambiguous target ranging . The ICIs used in buzzes when targeting evasive prey show dynamics that seem to correspond to changes in prey range ( Figure 2A and B , Figure 2—figure supplements 1 and 2 ) , suggesting a tight sensor-motor feedback loop . To verify that these ICI changes are linked to prey escapes , we plotted the proportion of outward depth-of-field adjustments ( i . e . , increasing ICIs ) in time bins synchronised with the first detectable prey movement in buzzes . Echograms with clear echo traces showing sudden prey movements suitable for this timing analysis are sparse . Pooling data from six harbour porpoises , we found 76 buzzes in which the prey escape speed exceeded the predator closing speed by more than 0 . 25 m/s , leading to clear V-shaped echo traces . These buzzes ( Figure 2C ) showed strong positive ICI adjustments that differ significantly from control buzzes ( i . e . , with randomly reassigned target movement times ) beginning in the 50–100 ms time bin after the start of prey escapes . Suitable echograms for Blainville’s beaked whales were less common . After relaxing the selection criteria to accept buzzes with any sudden detectable increase in prey speed away from the whale , we pooled 36 examples from seven individuals . The resulting ICI distributions ( Figure 2D ) showed strong positive adjustments that differed significantly from control intervals beginning 100–200 ms after initial prey movement . Thus , in both toothed whale species foraging in the wild , evasive movements by prey appear to be consistently matched by biosonar adjustments that maintain the prey within the unambiguous depth of field with a feedback loop latency of ~50–200 ms . ICIs during buzzes decrease to <2 . 5 ms for harbour porpoises and <3 . 5 ms for Blainville’s beaked whales , meaning that at least 20 clicks are produced during this latency time . Energetic prey targeted by harbour porpoises can make multiple escape attempts within a buzz providing an opportunity to examine ICI responses to repeated cues ( Figure 3A ) . Plotting prey range against the unambiguous depth of field ( equivalent to an input-output plot of a control system ) revealed distinctive counterclockwise loops due to the ICI response latency ( Figure 3B ) . To estimate this latency , we advanced the ICI time series in 5 ms steps until the areas within the loops were minimised . An advance of 90 ms collapsed most of the loops in the delay compensation plot ( Figure 3C ) , suggesting that successive prey movements elicit responses with self-similar latency . In comparison , the magnitude of the ICI responses varied widely . Although ICI changes proportionate with prey escape movements occurred in some cases ( Figure 3A ) , over-compensation was more typical ( Figure 2—figure supplements 1 and 2 ) , leading to depths of field that extended well beyond prey range when prey attempted to escape . However , echoes from schools of prey , or from reflectors such as the seafloor or sea surface , can have dynamics which greatly exceed the speed of single prey , requiring rapid outward ICI adjustments to avoid range ambiguity ( Figure 4 ) . In wild predator-prey interactions , predators frequently strike at prey at about the same time that the prey responds , raising the possibility that ICI adjustments are timed based on predator strikes ( i . e . , implying an anticipatory or feed-forward control scheme ) rather than based on prey movement . To exclude this potential confound , we designed an experiment in which captive harbour porpoise approached a target that could be moved suddenly . Echograms during target approaches ( Figure 5 ) show that this experimental design successfully replicated the sharp speed changes of prey in wild porpoise chases and thus provide reliable cues for timing biosonar and kinematic responses . Responses to fast target movement were evaluated in 43 and 31 trials for two captive porpoises , Freja and Sif , respectively . ICI responses , parameterised by the proportion of positive ICI changes , showed similar latency as for wild porpoises with response times of 50–100 ms for Freja and 100–150 ms for Sif ( Figure 6A and B ) . Whole-body kinematic responses , inferred from the differential of on-animal acceleration measurements ( jerk ) , were detectable with latencies of 0–50 ms for the two porpoises , although strong jerk responses were more clearly evident at 50–100 ms ( Figure 6C and D ) . As the target was pulled at varying speeds by hand , the trials could be ranked according to the magnitude of the initial target movement cue based on the signal recorded from the built-in accelerometer . Using root-mean-square ( RMS ) target acceleration as a proxy for this cue , we plotted ICI and RMS jerk as a function of cue magnitude ( Figure 7 ) . Whereas the analysis of Figure 6 included all responses , here we set detection thresholds so as to focus on large biosonar and movement adjustments that might be broadly comparable to saccades in vision . The latency of these strong responses was consistently longer than the responses shown in Figure 6 and tended to increase with decreasing target acceleration . Using a 5 ms threshold ( corresponding to a 3 . 75 m depth of field ) to detect strong outward ICI adjustments , the latency of these large-scale responses was inversely correlated with RMS target acceleration ( r2 = 0 . 3 , p<0 . 0001 ) . Kinematic response latencies , using a 300 m/s3 threshold on RMS jerk to detect the onset of strong responses , were similarly inversely correlated with RMS target acceleration ( r2 = 0 . 2 , p<0 . 001 ) , albeit with more variability . Toothed whales use echolocation in a deliberative mode to stalk unwary prey from long ranges , but must transition to a reactive mode when close enough that their bow-wave can be detected by prey ( Wisniewska et al . , 2012 ) . In this discrete-time sensory system , there is an unavoidable trade-off between sensing range and temporal resolution ( Madsen and Surlykke , 2013 ) : to detect rapid prey movements , echolocators must sample at a high rate . However , under the prevailing click-by-click model for echolocation processing ( Au , 1993 ) , such fast sampling would require infeasible neural and muscular speeds , leaving uncertain how toothed whales respond to evasive prey . Our results from both wild animals and controlled trials demonstrate that toothed whales make sensor-motor responses to sudden target movements during echolocation buzzes but that the response latencies , although fast compared to other mammalian sensory systems , span multiple clicks . Our sample size ( eight Md , and six wild and two captive Pp ) is constrained by the difficulty of tagging wild toothed whales and by the availability of suitably trained captive animals . However , the consistent speed and stereotypy of responses in both natural foraging interactions and controlled trials strongly support an acute echo-mediated sensory feedback loop that is responsive to the evasive manoeuvres of prey . We propose the term echo-kinetic for these responses in analogy with the optokinetic responses that control ocular tracking in visual predators . Our results show that high-sampling-rate buzzes , which are a defining characteristic of toothed whale echolocation ( Wisniewska et al . , 2014 ) , delimit a reactive sensory mode in which the greatly increased temporal resolution allows detection of fast prey motion while rapid feedback control of click rate maintains unambiguous tracking of escaping prey . The universality of this sensory strategy in toothed whales is supported by the similarity of our results from two toothed whale families that diverged some 21 mya ( McGowen et al . , 2009 ) and occupy very different niches: Blainville’s beaked whales dive to great depths to forage on fish and squid in and below the deep scattering layer ( Arranz et al . , 2011 ) , whereas harbour porpoise forage epipelagically and benthically on small shallow-dwelling fish ( Wisniewska et al . , 2016 ) . Moreover , similar biosonar responses were consistently elicited from captive porpoises by sudden movements of an artificial target hinting that this reflex-like behaviour is deeply embedded in the neural substrate of toothed whales . This leads us to propose that acute sensor-motor feedback during buzzes is a fundamental feature of toothed whale echolocation that has enabled hunting of nutritionally-valuable muscular but reactive prey . The measured response latencies ( Figure 2 , Figure 6 ) show that tight sensor-motor feedback in echolocation buzzes can be achieved without extreme neural processing speeds . Echo-kinetic response latencies of 50–200 ms in our study are comparable to short-latency eye movements in primate vision ( Land , 1999; Kirchner and Thorpe , 2006; Erkelens , 2006 ) and to vocal response latencies to passive acoustic cues in dolphins ( Ridgway , 2011 ) and porpoises ( Wensveen et al . , 2014 ) . However , these response latencies are more than 20× longer than the 2 . 5–3 . 5 ms ICI during buzzes , demonstrating that echolocating whales process and respond to echo information during prey approaches much more slowly than they acquire it . Put another way , the maximum information bandwidth ( i . e . , 1/ ( 2 × ICI ) Hz , by Nyquist theorem ) is some 40× greater than the maximum control bandwidth that can be achieved given the response delay , that is , approximately 1/ ( 4× response latency ) Hz ( Åström , 1997 ) . This implies strongly that echo processing and control decisions during buzzes are decoupled from click rate rather than occurring on a click-synchronous basis as widely assumed ( Au , 1993 ) . This conjecture is consistent with the proposed processing mode of click packets produced during long-range echolocation of dolphins ( Ladegaard et al . , 2019; Finneran , 2013 ) . The dramatic bandwidth mismatch between information gathering and feedback control in echolocation buzzes begs the questions of how these processes are decoupled and what purpose this serves . Unlike bats , which generate calls individually by contractions of super-fast vocal muscles ( Elemans et al . , 2011 ) , we propose that the extreme click rates in toothed whale buzzes are achieved by a free-running pneumatic oscillator comprising the pressurised pre-narial air space and phonic lips ( Au and Suthers , 2014 ) . This oscillator produces buzz clicks at a rate determined by the air pressure and the tension of the phonic lips , both of which can likely be controlled asynchronously with respect to click production . In this model , control decisions , formed after observing echoes from tens of clicks during buzzing , modulate the rate of clicking with the objective of maximising temporal resolution subject to the constraint that the ICI is consistently greater than the two-way travel time to the target . The high clicking rate in buzzes enables rapid detection of prey responses but may provide other benefits , when combined with appropriate feedback mechanisms , such as ( i ) signal-to-noise ratio improvement of weak echoes by integration over multiple clicks ( i . e . , using integral control ) ; ( ii ) speed-based processing of echoic scenes to predict target motion ( via differential control ) ; and ( iii ) detection of modulations in echo level ( e . g . , due to prey tail-beats , Wisniewska et al . , 2016 ) that may be the earliest cues of prey responses , while avoiding aliasing in this discrete-time sensor . The high click rate in buzzes effectively forms a temporal fovea , akin to the spatial fovea in many visual predators , matched to the burst movement rates of the relatively small prey targeted by most toothed whales . This ensures the observability of prey behaviour and enables control tactics that counter unpredictable prey movement . Echograms recorded during wild encounters with evasive prey ( Figures 2—4 ) hint at two control tactics that may be employed . When the change in target range is dominated by predator movements , a gradual upward adjustment of clicking rate is sufficient to track the changing spatial relationship of predator and prey . In comparison , unexpected rapid prey movements often provoke large adjustments in the biosonar rate in which the acoustic depth of field is rapidly expanded at the expense of temporal resolution . This suggests a layered control with smooth tracking during stalking and when prey move predictably , but occasional saccade-like ballistic increases in clicking rate during chases . Such layered control actions may also accommodate the dynamics of schooling prey , which can quickly switch between cohesion when being pursued and dispersion when escaping ( Couzin and Krause , 2003 ) . Similar to vision ( Erkelens , 2006 ) , the saccadic biosonar adjustments occurred with longer latencies in captive trials , compared to average responses , suggesting that large adjustments may be employed when targets move away sufficiently rapidly for there to be a risk of ambiguous echo ranging , that is , the echo delay exceeding the time between outgoing clicks . While information bandwidths in echolocation are likely linked to prey dynamics , the control bandwidths ( i . e . , the speed with which the system can respond to changes ) may be more matched to the size and manoeuvrability of the predator given that size influences both the rotational inertia of the body and the length , and therefore contraction rate , of muscles ( Domenici , 2001 ) . We hypothesised the existence of two control loops in echolocation-guided hunting , controlling , respectively , the acoustic depth of field and the swimming kinematics ( Figure 1 ) . We have been able to demonstrate the biosonar feedback loop in both wild and controlled settings , but full-body kinematic responses to prey movement are confounded in wild predator-prey interactions by the predator’s own striking actions . However , our controlled studies demonstrate that porpoises make an accelerative response to target movement with latency roughly comparable to the biosonar response and it seems very likely that wild animals would have similar kinematic responses . Therefore , our finding of longer biosonar response latencies in Blainville’s beaked whales , which are three times the size of harbour porpoises , suggests that control bandwidths may scale inversely with predator size . In effect , selection pressure on higher control bandwidths may be opposed by the increasing energetic cost of fast movements in large animals . An additional constraint on biosonar control bandwidth arises in the largest toothed whale species , sperm whales , due to the separation of the brain and sound source ( located at the anterior tip of the nose ) which may be more than 3 m apart . Even with highly myelinated nerves , the conduction delay ( perhaps 30 ms ) of neural signals to the sound source will be comparable to , or exceed , the time between successive clicks in buzzes ( e . g . , 10–20 ms , Fais et al . , 2016 ) . The decoupling between information flow and feedback control proposed here , in concert with a self-running pneumatic oscillator at the sound source , may have been instrumental in permitting such extreme cranial telescoping . Thus , despite the overt differences between echolocation and vision , the response bandwidths and layered control inferred here for toothed whale echolocation are remarkably similar to those in primate vision ( Kirchner and Thorpe , 2006 ) , with response times of the order of 0 . 1 s , likely limited in both auditory and visual senses by higher-order processing and muscle contraction speeds . The apparent universal use of buzzes during capture of moving prey by echolocators ( Madsen and Surlykke , 2013 ) suggests that extreme sensory sampling rates , guiding fast echo-kinetic responses , may have been a critical development , parallel to optokinetic responses in visual predators , enabling echolocation to be used to hunt agile prey , as opposed to just navigation and prey search . Our results , therefore , reveal strongly convergent neural sensor-motor feedback loops between vision and echolocation that are key for sensing dynamic spatial relationships with small prey . The non-invasive experimental approach developed here enables measurement of neuro-sensory dynamics while animals solve vital real-world problems , opening the way for a deeper understanding of ecological drivers on sensor performance in the wild . Sound and movement recording DTAGs were attached with suction cups to the anterio-dorsal surface of wild harbour porpoises ( Pp , n = 6 ) in inner Danish waters between 2012 and 2018 , and Blainville’s beaked whales ( Md , n = 8 ) off El Hierro in the Canary Islands between 2004 and 2013 . DTAGs ( v3 and v4 ) were attached to harbour porpoises bycaught in pound nets as they were removed from nets ( for details , see Wisniewska et al . , 2016 ) . For beaked whales , DTAGs ( v2 and v3 ) were delivered to free-swimming animals using a hand pole from a small inflatable boat ( for details , see Aguilar de Soto et al . , 2012 ) . The tags sampled sound from mono or stereo hydrophones with 16-bit resolution and a sampling rate of 500 or 576 kHz ( Pp ) , and 192 or 240 kHz ( Md ) ( clipping levels of 170–175 dB re 1 μPa ) . Tags also sampled depth sensors , and tri-axial accelerometers and magnetometers , at sampling rates of 200–625 Hz/channel ( Pp ) and 50 Hz/channel ( Md ) . Tags automatically released from animals after a pre-programmed interval and were recovered by VHF radio tracking . Data processing was performed in Matlab 2016a ( MathWorks Inc ) . Spectrograms of the on-animal sound recordings were examined to identify rapid click sequences ( buzzes ) during foraging . For each buzz , the production times of clicks were determined using a supervised click detector with approximately 50 µs accuracy . Clicks from the tagged animal were distinguished from those of other nearby animals by the consistent angle-of-arrival ( on stereo tags ) and broader frequency range of the former . Buzzes were defined as intervals in which the ICI was below 0 . 013 s ( Pp ) ( Wisniewska et al . , 2016 ) or 0 . 1 s ( Md ) ( Johnson et al . , 2006 ) for at least 0 . 5 s . Echograms were formed for each buzz by first bandpass filtering the sound ( Pp: 100–250 kHz; Md: 25–60 kHz ) and then computing the amplitude envelope using the Hilbert transform . Segments of envelope synchronised to each click were extracted and displayed as coloured bars with width equal to the click’s ICI , resulting in an echogram display with axes of time and distance ( Johnson , 2014 ) . Body movement during buzzes was quantified from the norm of the jerk , that is , the vector magnitude of the change rate of the tri-axial acceleration signals ( Johnson et al . , 2004 ) . Buzz echograms were selected for timing analysis based on visual inspection . Echograms with unclear prey echo traces or with substantial interference ( e . g . , echoes from the sea surface , seafloor , or other organisms ) were rejected . The remaining echograms were examined for indications of prey escape attempts . These appear as sudden changes in the slope of prey echo traces ( Figure 2—figure supplement 1 and Figure 2—figure supplement 2 ) reflecting a step change in the closing speed between predator and prey as the prey accelerates away ( Wisniewska et al . , 2016 ) . As prey reactions are typically fast , the onset time of the slope change in the echo trace is usually well-defined . The first such reaction time in each buzz echogram was selected manually with ~10 ms accuracy , and traces with unclear or gradual slope changes were rejected . Potential biosonar adjustments to these prey movements were quantified by the proportion of positive changes in ICI ( suggesting an outward adjustment of the depth of field ) in 50 ms ( Pp ) or 100 ms ( Md ) bins , spanning from 500 ms before to 500 ms after each prey response time . As ICI varies continuously throughout buzzes , these bin sizes were chosen as a compromise between temporal resolution and rejection of noise from routine ICI variations . The wider bin size for Md reflects the longer ICIs produced during buzzes by these larger animals . To determine the probability of chance associations between target movement and ICI changes , a bootstrap method was applied for each species . The same biosonar response metric was computed 1000 times for randomly selected pairs of buzzes in which the prey movement time from one buzz was applied to another buzz . Specifically , the time elapsed between the start of the buzz and the onset of the prey response in the first buzz of each pair was added to the start of buzz time in the second buzz to give a mock prey move time from which to reference the analysis time bins . A significant deviation from chance was concluded for each time bin in which >95% of the observed proportions exceeded the randomised proportions . Experiments were carried out in an 8 × 12 m semi-natural facility at Fjord & Bælt , Kerteminde , Denmark , in May 2017 with two harbour porpoises ( Sif and Freja , both female ) . At the time of the experiments , Sif was 1 . 6 m in length , 14 years old , and had been housed at Fjord & Bælt since 2004 . Freja was 1 . 58 m in length , 20 years old , and had been held at the facility since 1997 . Both porpoises were trained to locate and intercept a 50 . 8-mm-diameter aluminium sphere while wearing opaque soft silicone eye cups . The target sphere contained an embedded hydrophone ( flat [±2 dB] frequency response from 1 to 160 kHz ) and two-axis accelerometer ( flat [+0/–3 dB] frequency response from 0 to 2 kHz , axes oriented horizontally ) , and was suspended in the water via a 0 . 8 mm nylon string to a depth of approximately 1 . 5 m . A 1 . 2-mm-diameter screened cable carrying the accelerometer and hydrophone signals from the target was loosely attached to the nylon string and connected out of the water to a three-channel synchronous 16-bit analog-to-digital convertor sampling at 500 kHz ( National Instruments , Austin , TX ) . A second nylon line running horizontally from the sphere to the side of the pool was used to move the target during trials . Animals were equipped with a DTAG v4 sound and movement recording tag attached 5 cm behind the blow-hole with silicone suction cups . This tag contains a single hydrophone sampled at 576 kHz ( flat [±2 dB] frequency response from 1 to 150 kHz , clipping level of 175 dB re 1 μPa ) together with a tri-axial accelerometer sampled synchronously at 200 Hz and a tri-axial magnetometer and pressure sensor sampled at 50 Hz . For each session , one of the two porpoises was introduced into the pool and stationed approximately 8 m from the target until given a cue to perform the target interception task . If the animal intercepted the target by touching it , it was bridged with a whistle and received a fish reward upon returning to station . In randomly selected trials , the target was moved manually approximately 30 cm by pulling vigorously on the horizontal line when the porpoise approached within one body length . The line was held slack prior to this to limit any early anticipatory target movement . Target movement was selected pseudo-randomly for each trial between fast , slow , and no movement , with a maximum run of two equal conditions . Up to 20 trials were performed with each animal per day for a total of 150 trials over 4 days for the two animals . Echolocation clicks were detected in the animal-attached DTAG recordings using a supervised detector . The time offset between the tag and the National Instruments recordings was then determined for each trial by matching click sequences between the tag and the target hydrophones ( max . timing error due to acoustic propagation of ~1 ms ) . Echograms were then assembled from the tag data as described above . Sudden changes in the closing speed between the target and the porpoise due to rapid movement of the target generated the same distinctive slope changes in echo traces as observed in wild predator-prey interactions . To maximise timing accuracy , trials were only selected for analysis if the target attained a speed greater than the porpoise’s approach speed ( approximately 1 m/s ) , resulting in a V-shaped echo trace . The apex of the V was then taken as the reference time for calculating response latencies . The target acceleration ( as measured by the embedded accelerometer ) began ~100 ms before this due to tightening of the line and rotation of the spherical target to align the tie point with the pull direction . Given the thin line and spherical target , neither of these movements generated significant echo signatures and the porpoise was therefore unlikely to detect the target motion until it is underway . We accordingly view the apex of the V in the echogram as a close indicator of the time at which the porpoise was first able to detect the target movement . As the target movement varied in each trial , the root-mean-squared target acceleration was calculated as a proxy for target motion . This was computed from the accelerometer embedded in the target by removing the fixed gravity component from each axis and then summing the squared signals from both axes over the 500 ms following the first acceleration transient . Biosonar responses to target movement were quantified as for the wild toothed whales using the proportion of positive ICI changes in 50 ms intervals . Locomotor responses to target movement were assessed from the norm-jerk ( Ydesen et al . , 2014 ) calculated from the DTAG accelerometer data sampled at 200 Hz . To assess the probability of chance associations between target movement and ICI changes , the same bootstrap method was applied as for wild harbour porpoise and beaked whales ( see above ) , that is , randomly selecting 1000 pairs of trials and applying the elapsed time between buzz start and target movement from one trial to the buzz of another trial .

In the animal world , split-second decisions determine whether a predator eats , or its prey survives . There is a strong evolutionary advantage to fast reacting brains and bodies . For example , the eye muscles of hunting cheetahs must lock on to a gazelle and keep track of it , no matter how quickly or unpredictably it moves . In fact , in monkeys and primates , these muscles can react to sudden movements in as little as 50 milliseconds – faster than the blink of an eye . But what about animals that do not rely on vision to hunt ? To find food at night or in the deep ocean , whales and porpoises make short ultrasonic sounds , or ‘clicks’ , and then listen for returning echoes . As they close in on a prey , they need to click faster to get quicker updates on its location . What is unclear is how fast they react to the echoes . Just before a kill , a harbour porpoise can click over 500 times a second: if they wait for the echo from one click before making the next one , they would need responses 100 times faster than human eyes . Exploring this topic is difficult , as it requires tracking predator and prey at the same time . Vance et al . took up the challenge by building sound and movement recorders that attach to whales with suction cups . These were used on two different hunters: deep-diving beaked whales and shallow-hunting harbour porpoises . Both species adapted their click rate depending on how far they were from their prey , but their response times were similar to visual responses in monkeys and humans . This means that whales and porpoises do not act on each echo before clicking again: instead , they respond to groups of tens of clicks at a time . This suggests that their brains may be wired in much the same way as the ones of visual animals . In the ocean , increased human activity creates a dangerous noise pollution that disrupts the delicate hunting mechanism of whales and porpoises . Better understanding how these animals find their food may therefore help conservation efforts .

2021

Echolocating toothed whales use ultra-fast echo-kinetic responses to track evasive prey

The function of microtubules relies on their ability to switch between phases of growth and shrinkage . A nucleotide-dependent stabilising cap at microtubule ends is thought to be lost before this switch can occur; however , the nature and size of this protective cap are unknown . Using a microfluidics-assisted multi-colour TIRF microscopy assay with close-to-nm and sub-second precision , we measured the sizes of the stabilizing cap of individual microtubules . We find that the protective caps are formed by the extended binding regions of EB proteins . Cap lengths vary considerably and longer caps are more stable . Nevertheless , the trigger of instability lies in a short region at the end of the cap , as a quantitative model of cap stability demonstrates . Our study establishes the spatial and kinetic characteristics of the protective cap and provides an insight into the molecular mechanism by which its loss leads to the switch from microtubule growth to shrinkage . Microtubules are protein filaments found in all eukaryotic cells . They consist of typically 13 protofilaments forming a tube . Their function depends on their ability to switch between growing and shrinking states , called dynamic instability , which is essential for intracellular space exploration , mitosis and migration ( Howard and Hyman , 2009 ) . The dynamic properties of microtubules are tightly controlled and a target of numerous anti-cancer drugs ( Peterson and Mitchison , 2002; Seligmann and Twelves , 2013 ) . However , the molecular mechanism of microtubule state switching is not understood . The switch-like nature of microtubule stability is thought to depend on the existence of a stabilising GTP ( or GDP-Pi ) cap at the end of growing microtubules as a result of addition of GTP-loaded tubulin heterodimers and subsequent GTP hydrolysis and phosphate release ( Carlier , 1982; Mitchison and Kirschner , 1984 ) . However GTP cannot be visualised in individual dynamic microtubules; therefore the cap size and their relation to individual microtubule stability remains unknown . A variety of models exist , assuming either long caps ( several tens of tubulin layers , i . e . hundreds of nanometers long , [Carlier et al . , 1984; Chen and Hill , 1983; Mitchison and Kirschner , 1984] ) or short caps ( 1-2 tubulin layers , i . e . 8–16 nm only , [Bayley et al . , 1989; Bolterauer et al . , 1999; Caplow and Shanks , 1996; Drechsel and Kirschner , 1994; Walker et al . , 1991] ) . Long caps would be a result of random GTP hydrolysis , causing the cap size to increase with growth speed . Short caps would be a consequence of coupled GTP hydrolysis , which would require GTP hydrolysis to accelerate when microtubules grow faster , resulting in a short cap of constant size ( Bayley et al . , 1989 ) . Alternative models exist that assume either currently not observable structural events for loss of stability , such as formation of cracks ( Flyvbjerg et al . , 1996; Li et al . , 2014; Margolin et al . , 2012 ) or defects , or local biochemical criteria at the very end of the microtubule , introducing a distinction between the total GTP cap and its part that is critical for stabilisation ( Bowne-Anderson et al . , 2013; Gardner et al . , 2011b; Brun et al . , 2009 ) . Proteins of the EB1 family ( EBs ) transiently bind to growing microtubule ends as their conformation matures in a nucleotide state-dependent manner ( Bieling et al . , 2007; Kumar and Wittmann , 2012; Maurer et al . , 2011 ) , potentially binding to the GTP ( or GDP-Pi ) cap ( Maurer et al . , 2012; Seetapun et al . , 2012; Zhang et al . , 2015 ) . Before microtubules switch from growth to depolymerisation ( catastrophe ) , the size of the EB binding region tends to decrease ( Maurer et al . , 2012 ) , potentially suggesting a link between microtubule stability and the size of the EB binding region . As this region extends over several hundred nanometers , an EB binding site cap would be a 'long cap' . The recent observation of large microtubule growth fluctuations ( Gardner et al . , 2011a; Kerssemakers et al . , 2006; Schek et al . , 2007 ) challenged 'short cap' models because transient shrinking episodes during growth phases of up to ~ five tubulin layers ( which would remove such short caps ) did not cause catastrophe ( Schek et al . , 2007 ) . On the other hand , long cap models have been challenged by tubulin dilution experiments ( Voter et al . , 1991; Walker et al . , 1991 ) . Tubulin dilution causes microtubules to stop growing and hence prevents the addition of new capping tubulins , resulting in microtubules undergoing a catastrophe after a delay of several seconds ( Walker et al . , 1991 ) . This delay time is thought to be the time it takes for the stabilising structure at the microtubule end to disappear and therefore is a measure of momentary stability at the instant of dilution . In these experiments , delay times were reported to be insensitive to the microtubule growth speed , apparently contradicting 'long cap' models that predict higher stability with increasing growth speed due to longer caps ( Padinhateeri et al . , 2012 ) . However , the cap size itself could not be visualised in these previous experiments and no quantitative explanation could be given for the observed values of the delay times . Given the reported contradicting results , the size of the stabilising cap and especially its relationship with microtubule stability is still unclear . Ultimately , this is due to a lack of experiments measuring the protective cap size and relating it to the stability of individual microtubules . To simultaneously measure momentary microtubule stability and EB cap size with high spatio-temporal resolution , we developed a new microfluidics-assisted two-colour TIRF microscopy assay for fast and complete tubulin removal , combining automated microtubule end tracking and EB protein monitoring . For the first time , this allowed us to measure with close-to-nm and sub-second precision the instantaneous stability of individual microtubules , and the instantaneous size of their EB binding site caps , and to directly investigate their relationship . We find that the EB binding region is indeed the stabilising cap . We demonstrate that microtubules with longer caps are more stable and we present a quantitative kinetic model of momentary microtubule stability . Catastrophe is induced when the number of EB binding sites at the very end of the cap has fallen to 15–30% of its steady state value . This establishes the basic mechanism of catastrophe induction based on measured protective cap properties . In a microfluidic device , we immobilised Alexa568-labelled stabilised microtubule seeds on a functionalised glass surface . We then observed their growth in the presence of 20 µM Alexa568-labelled tubulin and GTP by time-lapse total internal reflection fluorescence ( TIRF ) microscopy ( Figure 1A , Figure 1—figure supplement 1A ) . Solutions were exchanged within ~200 ms ( Figure 1—figure supplement 1B ) ; the exchange itself did not affect microtubule growth ( Figure 1—figure supplement 1C & D ) . Sudden and complete removal of tubulin stopped growth , and induced a catastrophe after a delay of typically several seconds ( Figure 1B , Video 1 ) , similar to earlier observations after tubulin dilution ( Walker et al . , 1991 ) . This delay is thought to be caused by the temporary survival of the stabilising cap . 10 . 7554/eLife . 13470 . 003Figure 1 . Momentary microtubule stability assayed by fast tubulin washout and nm precision plus end tracking . ( A ) Schematic of the microfluidic TIRF microscopy setup . ( B ) TIRF microscopy image sequence of an Alexa568-microtubule before and after washout , with growth , delay and fast depolymerisation periods indicated . Tubulin concentration was changed from 20 to 0 μM at washout . Time in seconds , scale bar is 3 µm . ( C ) Illustration of sub-pixel precision microtubule end tracking using a 2D fitting procedure ( Materials and methods , [Bohner et al . , 2015] ) . ( D ) Plots of the background fluorescence intensity ( top ) and end position data ( middle ) of a washout experiment together with the fits ( solid black lines ) used to extract the derived parameter values ( see Materials and methods ) . ( E ) Histograms of the growth speeds at tubulin washout ( left ) and of the subsequent delay times before catastrophe ( right ) ( n = 101 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13470 . 00310 . 7554/eLife . 13470 . 004Figure 1—figure supplement 1 . Fast and complete microfluidics-controlled solution exchange . ( A ) Scaled layout of microfluidic channels used in this study . The channel width was 300 µm and the height was 95 ± 5 µm . ( B ) Representative kymograph ( left ) and a plot of the time course ( right ) of 20 µM Alexa568-tubulin ( 12 . 5% labelled; 2 . 5 µM in total ) fluorescence intensity as measured by TIRF microscopy , showing fast and complete solution exchange ( imaged at 7 . 7 Hz ) : from 9 experiments , the average time for 90% buffer exchange ( 5% - 95% ) was 199 ( ± 32 s . e . m ) ms . ( C ) Left: Time series of TIRF microscopy images showing microtubules ( red ) growing from a surface-immobilised seed during repeated sudden microfluidics-controlled exchanges between solutions containing 75 nM Mal3-GFP ( green ) and not containing any Mal3-GFP , in the constant presence of 15 µM Alexa568-tubulin . Right: a corresponding kymograph ( top: GFP channel only , bottom: merge ) . Horizontal and vertical scale bars are 3 µm and 1 min , respectively . Persistent microtubule growth is not affected by solution exchanges . Time in seconds , recording frequency was 0 . 5 Hz . ( D ) Kymographs showing a Alexa568-microtubule growing from an immobilised seed; the growth speed changes abruptly in response to a sudden microfluidics-controlled change of the Alexa568-tubulin concentration from 12 μM to 25 μM ( left ) or from 25 μM to 12 μM ( right ) , again demonstrating that solution exchange does not cause catastrophesDOI: http://dx . doi . org/10 . 7554/eLife . 13470 . 00410 . 7554/eLife . 13470 . 005Figure 1—figure supplement 2 . Delay times and microtubule orientations . ( A ) Top: distribution of individual microtubule orientations , each averaged over 10 s growth before tubulin washout , for the conditions without Mal3 ( Figures 1 , 2 ) and with 200 nM Mal3-GFP ( Figures 3 , 4 ) . The microchannel axis is 0° and corresponds to the flow direction . Bottom: Delay times after washout versus orientation before washout , for the individual microtubules shown in the top panel . ( B ) As ( A ) for the tubulin concentration variation datasets ( Figure 6A ) . For each dataset , the Pearson’s r correlation coefficients were calculated using the magnitude of the orientations relative to the distribution mean . DOI: http://dx . doi . org/10 . 7554/eLife . 13470 . 00510 . 7554/eLife . 13470 . 006Video 1 . Tubulin washout experiment , corresponding to Figure 1 . A drop in background fluorescence marks the time point , when tubulin is removed . Microtubules undergo catastrophe with a delay . Time is in seconds , where 0 is set to the washout time point . DOI: http://dx . doi . org/10 . 7554/eLife . 13470 . 006 We imaged 101 microtubules in the presence of 20 μM tubulin at 4 Hz and tracked their plus ends with a precision of ~30 nm ( Figure 1C ) . Fitting traces of the end position and fluorescence background ( Materials and methods ) allowed us to determine the washout and catastrophe times with sub-sampling time precision , as well as the instantaneous growth speeds measured over a 10 s time period just before washout ( Figure 1D , Figure 1—figure supplement 1B ) . Growth speeds varied considerably around a mean of 28 nm/s ( Figure 1E , left ) . The delay times between washout and catastrophe also showed a broad and non-exponential distribution , with a mean of 7 . 3 s ( Figure 1E , right ) , similar to previous dilution experiments ( Walker et al . , 1991 ) . As bending can affect the material properties of microtubules ( Schaedel et al . , 2015 ) , we tested whether the measured delay times were influenced by mechanical stress , potentially induced by microtubule bending in our assay . We determined the orientation of the growing microtubule end regions relative to the flow direction before tubulin washout , and found that the variation of orientations was relatively small ( mean orientation 3 . 1° with a standard deviation of 7 . 3° ) indicative of good microtubule alignment . No correlation between the delay time and the magnitude of the orientation was observed ( Figure 1—figure supplement 2A , blue data ) , indicating that mechanical stress is not responsible for the observed variations of the momentary microtubule stabilities in our assay . However , in contrast to a previous report ( Walker et al . , 1991 ) , the delay times clearly increased with increasing growth speed , provided that speeds were measured directly before washout ( Figure 2A , Figure 2B left , Spearman's rank correlation coefficient ρ = 0 . 69 , p<10–15 ) . This demonstrates that microtubules are more stable when they grow faster . Interestingly , the strong correlation decreased when the growth speed was measured at earlier times before tubulin washout ( Figure 2A , B right ) , as quantitatively demonstrated by a gradual decrease of the correlation coefficient and a gradual increase of its p-value ( Figure 2C ) . Therefore , the measured delay times reflect microtubule stability at the moment of tubulin washout and this stability varies over time as the growth speed fluctuates . This might explain why previous tubulin dilution experiments failed to detect the correlation between delay times and growth speeds; there average growth speeds were measured over long time intervals ( Walker et al . , 1991 ) . 10 . 7554/eLife . 13470 . 007Figure 2 . Momentary stability increases with growth speed at washout time . ( A ) Example end position-time plot . Growth speeds were measured by linear fits over a 10 s time window directly before washout ( Δt = 0 ) or for comparison also at earlier times up to 14 s before washout . ( B ) Scatter plots showing a positive correlation between delay times and growth speeds when measured directly before washout ( Δt = 0 ) ; correlation strength ( Spearman's rank correlation coefficient ρ ) is significantly reduced for Δt = 14 s . Histograms of growth speeds ( top ) and delay times ( right ) are also shown . ( C ) Spearman's rank correlation coefficient ρ ( top ) and corresponding p-values ( below ) between delay time and growth speed indicate that the correlation strength decreases progressively when the time window over which speed is determined is shifted away from the washout time point . ( D ) Averaged microtubule end position trace after washout ( alignment with respect to catastrophe times ) . Errors are s . e . m . ( E ) Top: Example trace of a shrinkage episode for an individual microtubule after washout , with fit ( Materials and methods ) illustrating the definition of shrinkage length . Blue and magenta lines indicate washout and catastrophe times , respectively . Bottom: scatter plot of the shrinkage lengths versus growth speeds . DOI: http://dx . doi . org/10 . 7554/eLife . 13470 . 00710 . 7554/eLife . 13470 . 008Figure 2—figure supplement 1 . Determination of catastrophe and shrinkage parameters after tubulin washout . ( A ) Top: End position - time plots showing four example catastrophes ( offset in position ) from the data set in Figure 2 , with fits as described in the Materials and methods ( dashed lines ) . Bottom: corresponding speeds and transition widths , σ , determined from the fits above . Good fits are produced for the end position traces for a large range of transitions between slow shrinkage to fast depolymerisation at catastrophe . ( B ) Enlarged view of a catastrophe transition ( dashed box in panel A , illustrating the definitions of the shrinkage length between washout and catastrophe used in the text: the time point of 25% change in speed is defined as the catastrophe time ( magenta dashed lines ) . The difference in microtubule end position between the washout time and the catastrophe time is defined as the measured shrinkage length ( here 174 nm ) . In contrast , for simplicity the predicted shrinkage length ( here 140 nm ) is obtained from an extrapolated linear fit to the slow shrinkage episode between washout and catastrophe . For gradual transitions between slow shrinkage to fast depolymerisation , the predicted shrinkage length is systematically a few tens percent shorter than the measured shrinkage length , but nevertheless provides a good estimate of the range of lengths . ( C ) Distribution of individual shrinkage speeds after washout . Histogram of slow shrinkage speeds after washout as determined from a linear fit to the end position trace between tubulin washout and catastrophe , corresponding to the data set in Figure 2 . The mean shrinkage speed obtained from a fit to the average when all tracks are aligned at the catastrophe time point ( Figure 2D ) is very similar to the mean speed from fits to all individual shrinking phases ( bar graph , inset ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13470 . 008 A mean microtubule end trajectory was generated by averaging all tracks after washout , aligned with respect to the catastrophe time ( Figure 2D ) . This average track demonstrated that microtubules did not simply pause between washout and catastrophe as previous lower resolution measurements indicated ( Walker et al . , 1991 ) ; instead microtubules shrank slowly on average at 26 nm/s before catastrophe . This compared to a much faster depolymerisation speed of 480 nm/s after catastrophe . Our resolution allowed us to also measure slow shrinkage at the level of individual microtubules ( Figure 2E top ) , yielding a similar mean shrinkage speed ( Figure 2—figure supplement 1C ) . The observed slow shrinkage agrees with the established view that in the presence of tubulin , growth results from the difference of fast tubulin association and slower dissociation events ( Gardner et al . , 2011a; Walker et al . , 1988 ) ; we observe the latter here directly for the first time . We also measured the nanometer-range distances that microtubules shrank between washout and catastrophe . These shrinkage lengths also tended to increase with growth speed ( Figure 2E bottom , ρ = 0 . 51 , p<10–7 ) , supporting the idea that faster microtubules are more stable , apparently due to longer stabilising caps . The mean shrinkage length was 220 nm , i . e . more than 25 tubulins per protofilament . This directly excludes models based on short stabilising caps . Interestingly the observed shrinkage length before catastrophe was in the same range as previously reported 'comet lengths' of the EB binding regions in cells ( Seetapun et al . , 2012 ) and in vitro ( Bechstedt et al . , 2014; Bieling et al . , 2007; Dixit et al . , 2009 ) , further supporting the possibility of a link between the EB binding region and microtubule stability , as suggested recently ( Maurer et al . , 2012 ) . To visualise the EB binding region before and after tubulin washout , we added 200 nM GFP-labelled Mal3 , the fission yeast EB ( Beinhauer et al . , 1997; Busch and Brunner , 2004 ) , keeping its concentration constant throughout the tubulin washout experiment ( Figure 3A , Video 2 ) . EBs do not only report on the size of their binding region , but also affect the kinetic processes determining its size: they mildly increase microtubule growth speeds and accelerate the turnover of their own binding sites , effectively shortening the region they bind to ( Maurer et al . , 2011; Maurer et al . , 2014 ) . Analysis of 139 microtubules showed that after tubulin washout , the mean delay time was now reduced approximately twofold to 3 . 5 s ( Figure 3B ) , despite a mild increase of the growth speed to 33 nm/s ( Figure 3B , inset ) . Hence , binding of Mal3-GFP reduces the momentary microtubule stability , reminiscent of the EB-induced stimulation of catastrophes at steady state ( Komarova et al . , 2009; Maurer et al . , 2014; Mohan et al . , 2013 ) . This suggests that the acceleration of microtubule end maturation that shortens the EB binding region is , at least in part , responsible for decreased momentary microtubule stability . 10 . 7554/eLife . 13470 . 009Video 2 . Tubulin washout experiment in the presence of 200 nM Mal3-GFP ( yellow ) , corresponding to Figure 3 . Microtubules are depicted in blue . The microtubule with the higher Mal3-GFP signal at the time point of washout exhibits a longer delay until catastrophe occurs . Time is in seconds , where 0 is set to the washout time point . To enhance clarity of presentation , background subtraction ( imageJ , 30 pixel rolling ball plug-in ) was applied to suppress the large difference between total intensities in the microtubule channel before and after washout . DOI: http://dx . doi . org/10 . 7554/eLife . 13470 . 00910 . 7554/eLife . 13470 . 010Figure 3 . Mal3 shortens the delay between tubulin washout and catastrophe . ( A ) TIRF microscopy image sequence of a Alexa568-microtubule ( red ) in a tubulin washout experiment in the constant presence of 200 nM Mal3-GFP ( green ) . Tubulin was present at 20 µM before washout . Background has been subtracted; the raw background intensity of the tubulin/microtubule channel is depicted on the right hand side ( [Tub] ) , indicating when tubulin is removed; other conditions are as in Figures 1 and 2 . Time in seconds , scale bar is 3 µm . ( B ) Cumulative delay time distributions in the presence of 200 nM Mal3-GFP ( green ) and in its absence ( black ) . Inset: bar graph of the corresponding mean growth velocities before washout; error bars are s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 13470 . 010 At the moment of tubulin washout the Mal3-GFP intensities at microtubule end regions varied considerably ( Figure 4A , blue histogram ) . At a constant Mal3-GFP concentration the fraction of occupied binding sites is constant; additionally , the Mal3-GFP binding/unbinding turnover is fast compared to other processes ( Bieling et al . , 2007 ) : hence , the observed variation of the Mal3-GFP intensities results mainly from variations of the EB binding site numbers at different microtubule ends ( Figure 4—figure supplement 1 ) . Higher Mal3-GFP intensities were observed for faster growth speeds ( ρ = 0 . 74 , p<10–15 ) ( Figure 4B ) . This indicates that individual microtubules that happened to be growing faster at the moment of tubulin washout had longer EB site caps , under otherwise identical conditions . This extends previous observations made at the ensemble level showing that the mean EB comet length increases with mean growth speed when varying the tubulin concentration ( Bieling et al . , 2007 ) . 10 . 7554/eLife . 13470 . 011Figure 4 . Microtubules with longer EB site caps are more stable . ( A ) Histograms of the Mal3-GFP intensity at microtubule ends at the moment of tubulin washout ( blue ) and catastrophe ( magenta ) ( n = 139 ) . The corresponding lattice intensity at tubulin washout ( grey ) is also shown . Mal3-GFP was present at 200 nM throughout the experiment and tubulin at 20 µM before washout . ( B ) . Scatter plot showing the positive correlation of Mal3-GFP intensities and growth speeds determined directly before washout . ( C ) Two representative end position time plots ( red dots ) with the corresponding Mal3-GFP signals ( green line ) . The microtubule with an initially stronger Mal3-GFP signal at washout ( top ) has a longer delay time until catastrophe compared to the microtubule with weaker Mal3-GFP signal at washout ( bottom ) . ( D ) Scatter plots of Mal3-GFP intensities at microtubule ends at the time of tubulin washout ( top ) , at microtubule ends at the time of catastrophe ( middle ) , and on the microtubule lattice ( 1 . 5 µm from the microtubule end ) ( bottom ) versus the delay times . At the moment of catastrophe , on average 29% of the EB binding sites present at tubulin washout were left , calculated by comparing the Mal3-GFP intensity at washout and catastrophe , from each of which the lattice intensity was subtracted . ( E ) Scatter plot showing the positive correlation of delay times and growth speeds determined directly before washout . Delay times were independent of small variations of the microtubule orientation ( Figure 1—figure supplement 2A , green data ) . ( F ) Decrease of correlation strength ρ between Mal3-GFP intensities and delay times ( open green symbols ) or growth speeds and delay times ( solid red symbols ) when the intensities/speeds were measured at earlier time points before washout , similar to Figure 2 . Spearman's rank correlation coefficients ρ ( top ) and corresponding p-values ( bottom , log scaled ) are depicted . DOI: http://dx . doi . org/10 . 7554/eLife . 13470 . 01110 . 7554/eLife . 13470 . 012Figure 4—figure supplement 1 . The main source of Mal3-GFP intensity fluctuations at microtubule ends growing in the presence of GTP are fluctuations of the size of the EB binding site cap . ( A ) Time series of TIRF microscopy images ( GFP channel only ) of 75 nM Mal3-GFP binding to microtubules growing in the presence of either GTPγS ( left ) or GTP ( right ) . GTPγS is a slowly hydrolysable GTP or GDP-Pi analogue , transforming the entire microtubule into a mimic of a microtubule end with a constant density of EB binding sites ( Maurer et al . , 2011 ) . Tubulin concentration was 22 µM . Time is in seconds , the frame rate was 9 . 9 Hz . ( B ) Bar graph showing the relative variance , D , ( see Materials and methods ) of Mal3-GFP intensities at growing microtubule end for the conditions in A . Error bars are standard deviations . For each condition , 15 , 000 image frames were analyzed . Because the EB binding site density on GTPγS microtubules is maximal and constant , the larger relative variance measured at microtubule ends growing in the presence of GTP demonstrates that the broad distribution of Mal3-GFP intensities under this condition ( data set in Figure 4 ) is largely due to the variability of the number of high affinity EB binding sites at growing microtubule ends and not to stochastic binding/unbinding events or measurement noise . Therefore , the measured Mal3-GFP intensities at growing microtubule ends are a good measure for the size of the EB binding region . DOI: http://dx . doi . org/10 . 7554/eLife . 13470 . 01210 . 7554/eLife . 13470 . 013Figure 4—figure supplement 2 . The instant EB cap length defines momentary stability . ( A ) Example plot of a tubulin washout track with the microtubule end position in red and the corresponding Mal3-GFP intensity in green , from the data set in Figure 4 . Mal3-GFP intensities were measured over time windows of 1 s , which were successively shifted away from the washout time point: ( B ) the intensity at Δt = 0 ( directly before washout ) shows a strong positive correlation with the delay time ( top ) , which decreases at larger shifts ( bottom Δt = 14 s ) . ( C ) The correlation between growth speed and delay time shows a similar trend with increasing shift , also consistent with the data presented in in Figure 2 in the absence of Mal3-GFP . The data in B top and C top are the same as in Figure 4D top and E . These data demonstrate that both the instantaneous EB cap length and growth speed are indicative of the 'momentary' stability of the microtubule . DOI: http://dx . doi . org/10 . 7554/eLife . 13470 . 013 Immediately after tubulin washout the Mal3-GFP intensity in the microtubule end region started decreasing ( Figure 4C , green traces ) . Strikingly , microtubules with initially smaller EB site caps displayed shorter delay times than those with larger caps ( Figure 4C , compare top vs bottom ) , as demonstrated quantitatively by a strong correlation between delay times and initial Mal3-GFP end intensities ( ρ = 0 . 78 , p<10–30 ) ( Figure 4D top , Figure 4—figure supplement 2 ) . Consistently , the more stable microtubules were again the faster growing ones ( ρ = 0 . 66 , p<10–15 ) ( Figure 4E , Figure 4—figure supplement 2C top ) . Both correlations were strongest when the EB cap size and the growth speed were measured directly at tubulin washout; the correlation coefficients decreased and p-values increased for growth velocities and Mal3-GFP intensities measured at earlier times before washout ( Figure 4F , Figure 4—figure supplement 2 ) . This demonstrates again that microtubule stability fluctuates over seconds as the growth state and the size of the EB binding region fluctuate . At the time point of catastrophe , the total Mal3-GFP intensity , and hence the size of the EB site caps , was on average reduced to 29% of the initial size ( Figure 4D , compare top vs middle , considering background shown in bottom ) . The similarity of the observed reduction in size of the EB binding regions at the moment of catastrophe suggests that catastrophe occurs when a critical threshold of EB sites is reached . We averaged both microtubule end trajectories and Mal3-GFP intensity traces after tubulin removal , all aligned with respect to the catastrophe time . The mean Mal3-GFP intensity decay was roughly mono-exponential until catastrophe , with a measured rate of 0 . 33 s-1 ( Figure 5A ) . This rate could be quantitatively explained as a consequence of two kinetic processes removing the cap ( Figure 5B ) : ( i ) slow shrinkage of the microtubule after washout ( mean vs = 40 nm/s in the presence of Mal3-GFP ) which removes tubulins from the end; ( ii ) continued microtubule maturation after washout which transforms EB binding sites everywhere in the cap into mature lattice ( km = 0 . 16 s-1 ) . The latter can be determined from the analysis of averaged Mal3-GFP intensity profiles ( comet analysis ) ( Figure 5C , Supplementary file 1A , Materials and methods ) ( Maurer et al . , 2014 ) . Together with the measured mean growth speed vg = 33 nm/s before washout , these known kinetic parameter values predict a theoretical decay rate for the size of the EB cap of kMal3 = km ( vs/vg+1 ) = 0 . 35 s-1 ( Materials and methods ) , in good agreement with the measured value of 0 . 33 s-1 . 10 . 7554/eLife . 13470 . 014Figure 5 . When an EB cap size threshold is reached catastrophe is induced . ( A ) Averaged microtubule end position traces ( red ) and Mal3-GFP intensity traces ( green ) after tubulin washout ( alignment with respect to the catastrophe time ) , and fits ( black lines ) ( Materials and methods ) . Same data set as in Figure 4 . ( B ) Top: Scheme illustrating the two processes leading to EB cap loss after tubulin washout: disassociation of tubulin from the end ( koff ) resulting in slow shrinkage ( vs ) , and maturation of EB binding sites into mature lattice ( km ) . Together these processes define the kinetics of the decay of EB binding sites ( Materials and methods ) . Bottom: simple model of average distributions of the EB binding sites at washout ( two ) and catastrophe ( tcat ) times . ( C ) Comet analysis: Average ( green ) of 5500 Mal3-GFP intensity profiles aligned with respect to the microtubule end position as described in ( Maurer et al . , 2014 ) and a fit to the data ( black ) yielding the comet length , which together with the growth velocity provides the maturation rate ( Materials and methods ) . ( D ) Histograms of the distribution of measured delay times ( left ) and shrinkage lengths ( right ) overlaid with theoretical estimates ( red solid line ) from an empirical kinetic model ( Materials and methods ) based on the measured GFP intensity distributions ( Figure 4A ) and the measured kinetic parameters from panel A and C . DOI: http://dx . doi . org/10 . 7554/eLife . 13470 . 014 Knowing the rate of the Mal3-GFP end intensity decay and the relative size of the EB binding region at catastrophe compared to the moment of tubulin washout , the theoretically expected delay time can be calculated ( Figure 5D left , Materials and methods ) , assuming that catastrophe is reached when 29% of the EB cap is left under the conditions studied here ( same data as in Figure 4 ) . This demonstrates that the mean delay time is determined by the growth characteristics of the microtubule ( growth speed vg and shrinkage speed vs ) and by the kinetics of microtubule end maturation ( maturation rate km ) . Similarly the mean shrinkage length ( 165 nm in the presence of Mal3-GFP , or ~20 tubulin lengths ) could also be calculated ( Figure 5D right , Figure 2—figure supplement 1B , Materials and methods ) . Furthermore a rough estimate of the standard deviations of the delay times and the shrinkage lengths can be obtained assuming that they are mostly determined by the observed variations of the size of the EB binding region at tubulin washout and catastrophe ( Figure 5D , Materials and methods ) . The good agreement between the measured and predicted delay times and shrinkage lengths shows that the temporal and spatial scale of the response of the microtubules to a sudden stop of growth can be quantitatively explained . We observed that a large part of the EB binding region is lost at catastrophe , however part of it is still present when catastrophe occurs . This suggests that a threshold may have to be reached to induce catastrophe , raising the question of what exactly constitutes this threshold . Two extreme possibilities can be envisaged: A minimal total number of cap sites anywhere in the entire cap might be required for stability . Alternatively a minimal density of cap sites only at its very end where the cap site density is highest might be needed for stability of the cap . In other words , either the entire cap or only its highest density region could be critical for stability . These two scenarios predict different dependencies of the delay times on the growth speed and hence cap length ( Materials and methods ) . To explore the momentary microtubule stabilities over a larger range of cap sizes , we performed tubulin washout experiments at a range of different tubulin concentrations from 10 μM to 35 μM , and at an increased magnesium ion concentration to further increase the velocity range ( O'Brien et al . , 1990 ) . In total 210 microtubules were analysed . As the growth speed distributions at different tubulin concentrations overlapped strongly ( Figure 6A , top ) , we speed-sorted the data when calculating averages as a function of speed ( Figure 6—figure supplement 1A ) ( Maurer et al . , 2014 ) . Average delay times extracted for seven speed groups from over 200 individual microtubule tracks displayed the expected positive correlation between delay times and growth speeds , however showing a weaker dependence especially for the higher speed range ( Figure 6A bottom ) . This correlation was masked when displaying the delay times simply as a function of tubulin concentration ( Figure 6—figure supplement 1B ) , again emphasizing the importance of correlating delay times with momentary speeds in the presence of growth fluctuations . 10 . 7554/eLife . 13470 . 015Figure 6 . An end density threshold explains the dependence of the mean delay times on growth speed . ( A ) Top: Histograms of overlapping growth speed distributions at tubulin washout for 10 , 14 , 20 and 35 μM tubulin . Bottom: Scatter plots of the corresponding individual delay times versus growth speeds ( open symbols ) and of the averages for 7 different speed groups ( filled symbols ) ( see Figure 6—figure supplement 1A ) . Error bars are s . e . m . n = 210 . Fits are shown for kinetic models assuming that catastrophe is induced when either a critical EB binding site density in the end region ( solid line , Equation 11 in Materials and methods ) or a critical total number of EB sites ( dashed line , Equation 13 in Supplemental Materials and methods ) is reached ( Materials and methods ) . As for the other datasets , delay times were independent of small variations of the microtubule orientation ( Figure 1—figure supplement 2B ) . ( B ) Schemes illustrating the two threshold scenarios for a fast ( top ) and slowly ( bottom ) growing microtubule . The total number of EB binding sites ( shaded areas under the curves ) is different for the two speeds when the same critical end density of sites ( bold vertical lines indicating the maximum amplitude of the curves ) has been reached . This illustrates that the threshold values of a kinetic model assuming a critical 'end density' threshold and of a model assuming a critical 'total number' threshold have different dependencies on the initial length of the EB binding region before tubulin washout . This leads to different predictions of the dependence of the delay times on the growth speed as shown in ( A ) . ( C ) Scatter plots of the data previously shown in Figure 2B left and Figure 4—figure supplement 2C top , without and with Mal3-GFP ( green and red symbols , respectively ) : individual delay times versus growth speeds ( open symbols ) and of the averages for 4 different speed groups ( filled symbols ) . Error bars are s . e . m . Solid lines are fits to the data using the end density threshold model ( Equation 11 in Materials and methods ) . The delay times of the data without Mal3 differ slightly from those shown in ( A ) as a consequence of different Mg concentrations . ( D ) Global fits to the speed-sorted data in A , C and the speed-sorted Mal3-GFP intensities at tubulin washout and catastrophe ( from Figure 4D top and middle ) using the L- model that considers as a threshold the number of EB cap sites within a region of length L behind the microtubule end . Reduced χ2 values are shown for a range of L values . Example fits at L = 8 nm , 80 nm and 320 nm are shown in Figure 6—figure supplement 2 . Inset: Scheme showing the region considered in the L model and the number of EB sites within that region ( Materials and methods ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13470 . 01510 . 7554/eLife . 13470 . 016Figure 6—figure supplement 1 . Growth and shrinkage speeds and delay times of the data sets with varied tubulin concentrations . ( A ) Illustration of speed-sorting: Histogram showing all growth speeds from the 4 data sets with different tubulin concentrations presented in Figure 6A . The data was then sorted into 7 groups each containing 31 microtubules , according to their measured speed before washout , irrespective of the original tubulin concentration . ( B ) Delay times as a function of tubulin concentration: box plots depicting the delay times of microtubules grown at different tubulin concentrations . Same data set as in Figure 6A . ( C ) Scatter plot of the slow shrinkage speed vs after washout against the growth speed vg before washout . A mild correlation was observed; the explicit relationship between vs and vg was necessary for the model fits to the experimental data in Figure 6A and Figure 6—figure supplement 2 . This observation potentially supports earlier proposals that predict higher koff rates for faster and more tapered microtubule ends ( Coombes et al . , 2013; Gardner et al . , 2011a ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13470 . 01610 . 7554/eLife . 13470 . 017Figure 6—figure supplement 2 . Threshold model fits considering the number of binding sites within a specific length , L . Simultaneous fits to the delay times of the tubulin concentration variation data ( left , top ) in Figure 6A , the delay times of the data with and without Mal3-GFP ( right , top ) in Figure 6C , and speed-sorted and background-subtracted Mal3-GFP intensity data ( bottom ) from Figure 4D , using the L model shown in Figure 6D ( Equation 17–19 ) . Solid , dashed and dotted lines are for L = 8 , 80 , and 320 nm , respectively , supporting the idea that the front part of the cap – and not its entirety – is important for stability . DOI: http://dx . doi . org/10 . 7554/eLife . 13470 . 01710 . 7554/eLife . 13470 . 018Figure 6—figure supplement 3 . Threshold model fits considering 2-step maturation of the microtubule end . ( A ) Averaged delay times as a function of microtubule growth speed as shown in Figure 6A ( tubulin concentration variation dataset ) are fitted using either an 'end density' or 'total number' threshold model , either assuming simple 1-step maturation of microtubule end regions , as throughout this study , or assuming 2-step maturation kinetics ( Maurer et al . , 2014 ) , as indicated . Both end density models yield good fits , predicting thresholds of 0 . 20 and 0 . 25 for the 1-step and 2-step maturation model , respectively . Both 'total number' models yielded unsatisfactory fits . ( B ) Averaged delay times as shown in Figure 6C ( Mal3 dataset and its control ) are fitted using an 'end density' threshold model either assuming 1-step or 2-step microtubule end maturation , as indicated . Predicted thresholds: Mal3 data – 0 . 28 and 0 . 29 for the 1-step and 2-step model , respectively; control without Mal3 – 0 . 14 and 0 . 17 for the 1-step and 2-step model , respectively . For the fits of the data without Mal3 , k1 = 5km; for the fits of the data with Mal3 , k1 = 20km ( Maurer et al . , 2014 ) . The other fit parameters were as in Figure 6A and 6C . DOI: http://dx . doi . org/10 . 7554/eLife . 13470 . 018 Together with the measured end maturation rate ( Supplementary file 1A ) , growth and shrinkage speeds ( Figure 6—figure supplement 1C ) for this condition , the dependence of the mean delay times on growth speed could be quantitatively explained by assuming that catastrophe is induced when a critical 'end density' of cap sites ( Figure 6B , ncrit ) is reached in the decaying cap . A fit ( Figure 6A , bottom , solid red line ) ( Materials and methods ) predicts that independent of the initial length of the stabilising cap , catastrophe occurred when around a fifth of the initial density of cap sites ( fcrit = 0 . 20 ) was left at the end of the decaying cap , i . e . ~3 per tubulin layer . The longer the caps at washout , i . e . the faster microtubules grow , the less dependent the momentary microtubule stability becomes on cap length . This is because EB site maturation becomes more important relatively to end shrinkage for reaching the 'end density' threshold , with increasing cap length at washout . In contrast , an alternative model assuming that catastrophe occurs when a critical 'total number' of sites anywhere in the cap ( Figure 6B , Ncrit ) is reached did not agree with the data ( Figure 6A , dashed line ) : this model predicts a stronger increase of delay times with growth speed for longer caps as a consequence of increasing total EB binding site numbers with cap length at washout . Fits to the other speed-sorted mean delay times as a function of mean growth speed ( from data shown in Figure 2B left and Figure 4E ) produced also good agreement with the data for a simple critical 'end density' threshold ( Figure 6C ) . This analysis revealed furthermore that the addition of Mal3-GFP destabilised the cap by increasing the threshold from 14% to 28% of the initial end density , i . e . from ~2 to ~4 cap sites per tubulin layer ( Supplementary file 1B ) . To gain insight into the length along the microtubule end over which a critical minimal density of cap sites is required for stability , we examined the data using a third version of the simple maturation model: this explicitly considered the number of cap sites within an extended but finite distance of the shrinking end ( Figure 6D , inset ) . Good global fits to our complete dataset , consisting of all delay times and also the Mal3-GFP intensities at washout and catastrophe , were obtained for fixed end region lengths ranging between 8 and ~80 nm ( Figure 6D , Figure 6—figure supplement 2 ) . This puts an upper limit on the size of the critical part of the cap over which a minimal density needs to be maintained for microtubule stability . For simplicity we have assumed here throughout that the EB binding site region starts directly at the microtubule end ( one-step end maturation ) . Previously , a detailed analysis of fluorescent EB end profiles has revealed an additional small non-binding region at the very end of the microtubule before the actual EB binding region ( Maurer et al . , 2014 ) . This region could be accounted for by a two-step end maturation process consisting first of fast generation of EB binding sites , and a subsequent slower maturation into lattice sites . Applying this more complex model here did not improve the quality of the fits ( Figure 6—figure supplement 3 ) and confirmed that the threshold of stability is defined by a critical end density of cap sites and not by a critical total number of cap sites ( Figure 6—figure supplement 3A ) . In the absence of Mal3 , the 2-step maturation model predicted threshold values for the end density that were ~25% larger compared to the simpler 1-step model . In the presence of Mal3 the two models predicted the same threshold value ( within error ) due to the first maturation step being very fast in the presence of EB1 family proteins ( Maurer et al . , 2014 ) . Therefore , the conceptually simpler end maturation model is sufficient to capture the basic principles determining momentary microtubule stability , especially for the more physiological condition in the presence of an EB1 family protein . We have performed tubulin washout experiments with high spatial and temporal resolution . By simultaneously monitoring the size of the EB binding region we provide compelling evidence that the protective cap is the EB binding region . Previously , it was noted that the size of the EB binding region decreased before catastrophe at steady state ( Maurer et al . , 2012 ) . Here , using sudden tubulin removal , we directly investigated for the first time the relationship between the size of the protective cap and the momentary stability of individual microtubules . We found that faster growing microtubules have longer EB caps and are more stable after tubulin washout . Therefore , our results provide the first direct experimental demonstration for the original proposal that microtubule stability would increase with growth speed as a consequence of longer protective caps ( Carlier et al . , 1984; Mitchison and Kirschner , 1984 ) . Our observations resolve the previous contradiction between tubulin dilution experiments , where delay times had been reported to be independent of growth speeds ( Voter et al . , 1991; Walker et al . , 1991 ) , and steady state lifetime measurements , where longer life times were observed for faster growth ( Gardner et al . , 2011b; Janson et al . , 2003; Walker et al . , 1988 ) . This apparent discrepancy led to the development of various 'short cap' models of microtubule stability or to more elaborate models postulating currently non-observable structural defects or cracks as being critical for catastrophe induction ( Bolterauer et al . , 1999; Bowne-Anderson et al . , 2013; Brun et al . , 2009; Flyvbjerg et al . , 1996; Li et al . , 2014; Margolin et al . , 2012; Piette et al . , 2009 ) . However , these models are inconsistent with the dependence of microtubule stability on instantaneous growth speed ( Figure 2B , 4E and 6A ) and are incompatible with the measured slow shrinking phase after tubulin washout removing ~25 tubulin layers on average from microtubule ends before catastrophe onset ( Figure 2 and 5 ) . These observations and the increase of steady state microtubule lifetimes ( Gardner et al . , 2011b; Janson et al . , 2003; Walker et al . , 1988 ) with increasing growth speed as well as the recently observed shrinking episodes during steady state growth ( Schek et al . , 2007 ) are all consistent with the view that the protective cap is hundreds of nanometers long and that a random process limits its length as proposed early on for the 'random GTP hydrolysis' model ( Carlier et al . , 1984; Mitchison and Kirschner , 1984; Padinhateeri et al . , 2012 ) . Whether this random process is GTP hydrolysis or rather phosphate release remains still to be determined . Given the high affinity of EBs to berylliumfluoride-microtubules ( often used as a GDP-Pi mimic [Carlier et al . , 1989] ) and GTPγS microtubules , but not to GMPCPP microtubules ( thought to be the canonical GTP analogue for microtubules [Alushin et al . , 2014; Hyman et al . , 1992] ) it is plausible that both the GTP and GDP-Pi nucleotide states are stabilising , with GDP-Pi tubulins forming the majority of the cap to which EBs bind with high affinity , consistent with the EB binding pattern along microtubule ends ( Maurer et al . , 2011; 2014; Zhang et al . , 2015 ) . We discovered that the correlations between microtubule growth speed , EB cap length and momentary microtubule stability ( delay time after tubulin removal ) were lost when growth speed or EB cap size were measured several seconds before tubulin washout ( Figure 2 , 4 , Figure 4—figure supplement 2 ) . This indicates that the momentary microtubule stability fluctuates as growth speed varies and that EB caps store a 'memory of stability' over the time scale of several seconds , probably as a consequence of the lifetimes of the EB binding sites being in this range . The observation of this memory excludes memoryless models of microtubule dynamics ( Bayley et al . , 1989; Bolterauer et al . , 1999; ) . It also explains why lower correlations between mean growth speeds measured over longer time intervals and momentary microtubule stability were observed in previously experiments with lower resolution ( Walker et al . , 1991 ) , emphasizing the dynamic nature of microtubule stability , being a prerequisite for dynamic instability . Our study shows that the entire EB binding site region serves as the stabilising cap . In space , its stability decreases gradually with binding site density . Large parts of the cap can be lost before it loses stability , strictly excluding cap models where the entire cap size is small ( Bayley et al . , 1989; Bolterauer et al . , 1999; Bowne-Anderson et al . , 2013 ) . The total cap is however much larger than what is required for its stability . This property allows the cap to withstand strong growth fluctuations that include even short depolymerisation episodes ( Schek et al . , 2007 ) . The observed delay times after tubulin washout were more sensitive to growth speed variations in the lower growth speed regime ( Figure 6A ) . Kinetic threshold models revealed that this behaviour can be accounted for if only the highest density region of the cap , i . e . its part closest to the microtubule end , determines its stability . This agrees with earlier conclusions that the minimal critical cap required for stability is short ( Caplow and Shanks , 1996; Drechsel and Kirschner , 1994 ) . We found that for the microtubule to be stable on average 15–30% cap sites needed to be left in the end region of the cap which had a length of up to ~10 tubulin layers at most . This conclusion drawn from our tubulin washout experiments is in good agreement with previous observations during steady state growth for the reduction of EB binding sites before catastrophes ( Maurer et al . , 2014 ) , demonstrating that it is independent of the specifics of the tubulin washout experiment . We observed considerable variation of individual delay times around their average values indicating the influence of additional stochastic effects that our simple kinetic threshold model does not capture . Such variability can be expected to result for example from differences in the exact spatial distributions of stabilising tubulins in the cap region and from the structural consequences these different distributions have for stability . Nevertheless our model captures the fundamental characteristics of momentary microtubule stability and provides a quantitative link with the kinetic properties of microtubule growth and GTPase-linked microtubule end maturation . At steady state , the catastrophe frequency has been reported to depend on the time of microtubule growth , i . e . on the microtubule age ( Gardner et al . , 2011b; Odde et al . , 1995 ) . Here we analyzed only microtubules which grew for similar time periods before tubulin washout . This allowed us to neglect ageing effects in our analysis . Currently there is no agreement on the mechanistic origin of microtubule ageing under steady state conditions ( Bowne-Anderson et al . , 2013; Coombes et al . , 2013; Zakharov et al . , 2015 ) . In the future , tubulin washout experiments as presented here can be expected to provide valuable novel insights into the effects of ageing on microtubule stability which will then likely require an extension of our current model of the momentary microtubule stability . Here , we gained also more mechanistic insight into how binding of EB proteins to the protective cap region influences microtubule dynamics . EBs destabilize microtubules in three ways: ( i ) acceleration of conformational end maturation ( Maurer et al . , 2014 ) , ( ii ) acceleration of tubulin dissociation from the microtubule end ( compare Figure 5A and Figure 2D ) , possibly due to effects of EBs on the microtubule end structure ( 'taper' or 'sheet' ) during growth ( Vitre et al . , 2008 ) , and ( iii ) direct structural destabilisation of the cap ( i . e . increase of the end density threshold ) ( Figure 6C ) , which agrees with recent structural studies indicating that EBs induce strain in the microtubule lattice ( Zhang et al . , 2015 ) . These findings further support the view that end binding proteins of the EB1 family may have evolved to modulate the stability of the functionally essential protective end structure of the microtubule and may then have gained the additional function to recruit a variety of other unrelated plus end binding proteins ( +TIPs ) , whereas in parallel more effective , for example ATP-dependent modulators of cap stability such as depolymerases of the kinesin-13 family evolved ( Akhmanova and Steinmetz , 2015; Brouhard and Rice , 2014; Duellberg et al . , 2013; Gardner et al . , 2011b; Howard and Hyman , 2007; Maurer et al . , 2012 ) . While the debate about the size of the stabilising cap was ongoing , microtubule end binding proteins were discovered ( Beinhauer et al . , 1997; Su et al . , 1995 ) and later used to facilitate microtubule end tracking in living cells ( Busch and Brunner , 2004; Komarova et al . , 2009; Mimori-Kiyosue et al . , 2000; Seetapun et al . , 2012 ) . Our quantitative demonstration that the EB binding region is the protective microtubule cap means that EBs can be used to monitor the momentary stability of individual microtubules . This will be useful for the further investigation of the momentary microtubule stability in reconstituted systems , for example for the study of its dependence on the microtubule growth time , but also for studies of how microtubule stability is modulated in living cells in real time . For the future , such experiments combined with quantitative analysis promise to yield a deeper mechanistic understanding of how microtubule ageing , and the presence of regulatory proteins and drugs control microtubule dynamics . Mal3-GFP was expressed and purified as described ( Maurer et al . , 2011 ) . Porcine tubulin was purified ( Castoldi and Popov , 2003 ) and labelled with biotin ( Pierce ) or Alexa568 ( Life technologies ) using NHS esters following standard protocols ( Hyman et al . , 1991 ) . Microfluidic channels with a functionalised glass surface were fabricated combining glass surface chemistry ( Bieling et al . , 2010 ) and soft lithography techniques ( Whitesides et al . , 2001 ) . The overall design was inspired by a recent study on actin filaments ( Jegou et al . , 2011 ) . 24 mm x 60 mm glass coverslips ( Menzel ) were covalently passivated ( against non-specific protein adsorption ) with polyethylene glycol ( PEG ) and functionalised with biotin as described ( Bieling et al . , 2010 ) . Silicon moulds with negative channel patterns were produced by deep reactive ion etching ( Holmes et al . , 2014 ) . Channel dimensions are detailed in Figure 1—figure supplement 1A . A mixture of polydimethylsiloxane ( PDMS ) and curing agent ( both from Dow Corning; 10:1 ratio w/w ) was poured over a mould , degassed for 2 hr at 4°C ( Modulyo 4K , Edwards ) and polymerised for ~12 hr at 68°C . The structured side of the peeled-off PDMS elastomer and of the functionalised surface of the glass coverslip were treated with air plasma ( Diener , Femto ) for 42 s . A small area of 4 mm x 8 mm , where imaging was to occur , was protected from plasma radiation by a small PDMS block to ensure the integrity of the surface functionalization . Immediately after plasma treatment the exposed sides were bonded to form the channels , and holes for the inlet and outlet channels were created using biopsy punchers ( Harris Uni-Core I . D . 0 . 75 mm ) . Tubing ( Tygon; diameter: 0 . 5 mm ) was connected to the PDMS inlets/outlet via short metal extensions ( 21 gauge hollow cylinders , cut from 'microlances' Becton Dickinson ) and Hamilton gas tight syringes ( 500 or 1000 μl total volume ) . The microfluidic devices were used for TIRF microscopy experiments immediately after assembly . All experiments were performed using a previously described TIRF microscopy set-up at a constant temperature at 30°C ( Duellberg et al . , 2014; Maurer et al . , 2014 ) . Fast solution exchange in the micro-channel was achieved by switching the flow from three different inlets that were controlled by independent syringe pumps ( Aladdin , World Precision Instruments ) and manual valves ( Cole-Parmer ) between pumps and inlets . The microfluidic set-up and all solutions were pre-warmed to 30˚C just before the experiment . To assemble a sample , short GMPCPP-stabilised , Alexa568 and biotin-labelled microtubule 'seeds' were introduced through one inlet and allowed to attach to the functionalised glass surface via neutravidin ( Bieling et al . , 2010 ) . In brief , channels were purged with 100 µl assay buffer , 30 µl assay buffer supplemented with neutravidin ( Life Technologies , 0 . 05 mg/ml ) , 100 µl assay buffer , 30 µl GMPCPP seeds in assay buffer and again 100 µl assay buffer using always the same inlet . To initiate microtubule growth , 10–35 µM Alexa568-labelled tubulin ( labelling ratio was always 12 . 5% ) in final imaging buffer with or without 200 nM Mal3-GFP was introduced through the second inlet at a constantly maintained flow rate of 15 µl/min . Unless stated otherwise , the final imaging buffer was 80 mM K-PIPES ( pH 6 . 85 ) , 1 mM EGTA , 1 mM dithiothreitol , 90 mM KCl , 0 . 5 mM MgCl2 , 5 mM 2-mercaptoethanol , 10 mM ascorbic acid , 0 . 1% ( w/v ) methylcellulose , 2 mM GTP , 50 µg/ml β-casein , 20 mM glucose , 0 . 25 mg/ml catalase , 0 . 5 mg/ml glucose oxidase . For data presented in Figure 6A . the MgCl2 concentration was increased to 4 mM . Microtubules were allowed to grow for ~100 s until tubulin was quickly washed out by switching the flow to the third inlet leaving all other buffer constituents unchanged . Alexa568-microtubules and Mal3-GFP images were recorded simultaneously in separate channels at a frame rate of 4 Hz with 100 ms exposure time per image , unless stated otherwise . For some controls , the tubulin concentration was changed from 12 to 25 μM and vice versa ( Figure 1—figure supplement 1D ) or the tubulin concentration was kept constant at 15 µM while 75 nM Mal3-GFP was washed in and out to demonstrate that solution exchange does not interfere with microtubule growth or cause catastrophes by itself ( Figure 1—figure supplement 1C ) Growing microtubule ends were automatically tracked using a previously described script ( Maurer et al . , 2014; Ruhnow et al . , 2011 ) . Briefly , microtubules were coarsely identified by the user at the start of each movie , including microtubule seed position and polarity . A two-dimensional model was then automatically fit to the intensity data for the microtubule plus-end in all subsequent frames . For each tracked microtubule , this gave the position of the microtubule end at every time point . From simulations , the tracking precision is estimated to be 20–30 nm ( Bohner et al . , 2015 ) at the typical signal-to-noise levels of >3 determined for these movies . When present , the simultaneously imaged Mal3-GFP intensity was quantified at the microtubule end position , and the lattice intensity was quantified 1 . 5 μm from the end along the microtubule . To generate soluble tubulin background time series , the spatially averaged background intensity of soluble Alexa568-tubulin was determined during the fitting procedure of the microtubule end , at every time point . The microfluidic channel was aligned with the camera axis before imaging . Only plus ends of microtubules that were aligned with the flow direction along the x-axis and did not show appreciable bending were considered for analysis to avoid potential mechanical effects on microtubule dynamic behaviour due to bending ( Schaedel et al . , 2015 ) . For each microtubule , the delay time between tubulin washout and catastrophe , the corresponding depolymerisation length and the kinetic parameters characteristic of its dynamics were quantified using a fully automated custom script ( Matlab , Mathworks ) , as follows . Firstly , the washout time point was identified from the inflection point of an error-function fit to the background intensity of the tubulin fluorescence channel ( top panel in Figure 1D and Figure 1—figure supplement 1B ) . The average time for 90% buffer exchange ( 5% - 95% ) was 199 ( ± 32 s . e . m ) ms , as determined from 9 washout profiles measured at 7 . 7 Hz ( Figure 1—figure supplement 1B ) . The pre-washout instantaneous microtubule growth speed was found from a linear fit to the end position data over a 10 s time window exactly before washout . After washout , the catastrophe time was found by fitting the position data with the integral of the following function for the speed of the shrinking microtubule end ( Figure 1C , middle panel ) 0 . 5 ( v1+v2 ) +0 . 5 ( v1−v2 ) erf[ ( t0−t ) /2σ] This describes a transition between a period of slow linear shrinkage v1 to a period of fast depolymerisation v2 , through an error function centred at t0 , with transition width σ . The catastrophe point was defined as the time at which there was a 25% change between the slow shrinkage and the fast depolymerisation phase ( Figure 1D , bottom panel , Figure 2—figure supplement 1A ) . This point was found to be a sensible empirical measure for the fitted microtubule end trajectory visibly departing from the noise of the linear slow shrinkage phase between tubulin washout and catastrophe ( Figure 2—figure supplement 1B ) . Thence the measured delay time was defined as the difference between the washout and catastrophe times , and the measured shrinkage length defined as the corresponding difference in fitted microtubule end positions ( Figure 2E , Figure 2—figure supplement 1B ) . A linear approximation for the slow shrinkage speed between washout and catastrophe vs was obtained from a linear fit to this part of the track . v2 was directly taken as the fast depolymerisation speed after catastrophe vf . Spearman's rank coefficient was used to examine correlations between the above parameters as well as the Mal3-GFP end and lattice intensities at tubulin washout and catastrophe . To investigate the importance of measuring growth speeds and Mal3-GFP end intensities 'instantaneously' at washout for the magnitude of their correlation with the delay times , we also determined growth velocities and Mal3-GFP intensities over 10 s and 1 s time windows , respectively , at various times before washout . All reported correlation coefficients are summarised in the Supplementary file 1C . For each assay condition , plots showing the average microtubule end position around catastrophe were made by aligning all individual tracks at their determined catastrophe time and catastrophe end position , followed by resampling and averaging the data . Corresponding average Mal3-GFP intensity plots were made by aligning individual Mal3-GFP intensity profiles at the catastrophe time , followed by resampling and averaging the data . For each tracked microtubule , the 2D fit in the tracking procedure gave the orientation of the microtubule end region at every time point , with the positive x-axis of the image defined as zero degrees . The pre-washout average orientation and corresponding standard deviation were calculated over a 10 s time window just before washout , along with the subsequent delay time until catastrophe . For each assay condition , scatter plots of delay times versus the orientations of the microtubule end regions were produced . For analysis of statistical correlations , the magnitude of the orientation relative to the mean was used . To compare the Mal3-GFP intensity variability at growing microtubule ends ( Figure 4—figure supplement 1 ) in the presence of GTP and on GTPγS microtubules , microtubules growing in the presence of 22 µM Alexa568-labelled tubulin and 75 nM Mal3-GFP were imaged in standard flow chambers consisting of one functionalised and one passivated glass separated by double-sided sticky tape , as described previously ( Bieling et al . , 2010 ) . The chambers were filled manually; other conditions were as above , except that the final assay buffer was supplemented with 3 . 5 mM MgCl2 , did not contain ascorbic acid and contained either 1 mM GTPγS ( Roche ) or 1 mM GTP ( Fermentas ) . Microtubule end positions and Mal3-GFP intensities were recorded and analysed as detailed in the previous section . To compare these data sets , we determined the relative variance , D , of GFP intensities , where D is the variance divided by the mean . To confirm that the broad distribution of Mal3-GFP signals at microtubule ends reflects differences in the number of high affinity binding sites , we compared microtubules grown in the presence of GTP and GTPγS , the latter being a GTP or GDP/Pi analogue with very slow hydrolysis kinetics which mimics a microtubule end with a roughly constant number of EB binding sites ( Maurer et al . , 2011 ) . For each microtubule ( n = 9 , from 3 independent experiments , at least 1500 frames per microtubule ) , GFP intensities were measured for each frame and the mean and variance were determined . To determine the conformational maturation rates at microtubule ends , Mal3-GFP intensity profiles ( 'comets' ) were analysed ( Maurer et al . , 2014 ) . For tubulin washout experiments in the presence of 200 nM Mal3-GFP , the growth episodes before washout were directly used for analysis . To obtain estimates for maturation rates in washout experiments without Mal3-GFP , we performed independent experiments under essentially identical conditions as in tubulin washout experiments without Mal3-GFP , however with added low 'spike' concentrations of 0 . 75 nM or 1 nM Mal3-GFP , as indicated in the Supplementary file 1A , i . e . concentrations well below the Kd of Mal3-GFP binding to microtubule ends ( Maurer et al . , 2011 ) . To generate averaged Mal3-GFP comets , speed-sorted subsets of microtubule tracks were generated . Images of the microtubule end and corresponding Mal3-GFP signal were cropped from each movie frame and aligned at the detected end position: average images were produced from a total of ~5500 individual image frames , from 38 separate microtubules ( 200 nM Mal3-GFP condition ) . Average one-dimensional Mal3-GFP comet profiles were generated and analysed as described ( Maurer et al . , 2014 ) . For the growth speed range of the microtubules used for comet analysis here , a simple kinetic 'one-step’ model of microtubule maturation ( Maurer et al . , 2014 ) describes the distribution of EB binding sites sufficiently well: a microtubule growing at speed vg is assumed to form EB binding sites approximately instantly , which subsequently mature into weak ( lattice ) binding sites at a rate km . Hence the microtubule will have an approximately mono-exponential decay of binding sites in space x , giving a spatial profile of the linear density of all types of binding sites ( 1 ) n ( x ) =nx0e−xkm/vg+nlat where nx0 is the maximal binding site density , vg/km = Lcomet is the decay length of the binding region or the 'comet length' , and nlat is a step function representing a constant background signal due to lattice binding ( the affinity of lattice binding sites is roughly ten-fold reduced compared to end binding sites [Maurer et al . , 2011] ) . The measured average fluorescent intensity profile , I ( x ) , is a convolution of the binding site distribution , n ( x ) , with a Gaussian function , g ( x ) , that accounts for the optical PSF of the microscope and effects of averagingg ( x ) =e−12 ( x−xcσ ) 2 ( 1a ) I ( x ) =g ( x ) * n ( x ) where σ is the width of the Gaussian , and xc accounts for any spatial offset of the start of the EB binding sites from the fitted microtubule end position . Supplementary file 1A summarises the kinetic rates found from the average comets for each examined condition . We consider here the evolution of the number of the high affinity EB binding sites or their density at the end of the microtubule . We assume here that a microtubule with npf protofilaments , each composed of tubulin dimers of length ldim behaves as a single one-dimensional filament with an effective linear density of potential binding sites of nx0=npf/ldim . We further assume that at the moment of tubulin washout ( t = 0 ) , a microtubule with growth velocity vg and maturation rate has an approximately mono-exponential profile of EB binding sites in space with ( 1b ) n ( x ) =nx0e−xkm/vg Hence , integrating [1b] along the whole microtubule , the total number of EB binding sites at washout is ( 2 ) N ( t=0 ) =∫0∞nx0e−xkm/vgdx=nx0vg/km=nx0Lcomet As Figure 5B illustrates , after removal of free tubulin , all binding sites decay in time independently at a rate km as a consequence of continued maturation giving ( in the absence of shrinkage ) a spatio-temporal profile ( 3 ) n ( x , t ) =nx0e−xkm/vge−kmt Additionally binding sites ( and non-binding site tubulins ) are lost from the microtubule end due to shrinking at a speed vs ( see Figure 5B ) . Defining the shrinking end position as x0'=vst , the total number of sites at a given time after washout is ( 4 ) N ( t ) =∫vst∞nx0e−xkm/vge−kmtdx=N ( 0 ) e−km ( vs/vg+1 ) t The measured Mal3-GFP intensity ( corrected for background intensity ) is proportional to this total number of binding sites . Thus , from [4] the predicted temporal decay of the Mal3-GFP intensity is ( 5 ) IMal3 ( t ) =IMal3 ( 0 ) e−kMal3t with ( 6 ) kMal3=km ( vs/vg+1 ) as used for the analysis in Figure 5A . Hence , the average time to catastrophe can be estimated ( Figure 5D left ) using ( 7 ) TMal3=−ln[IMal3 ( cat ) IMal3 ( 0 ) ]kMal3 Assuming that the dispersion of delay times depends mostly on the variation of the size of EB binding site caps at washout and catastrophe , a rough estimate for the dispersion of delay times ( Figure 5D left ) can be produced in two steps: ( i ) The measured dispersion of the Mal3-GFP end signals is a result of measurement noise and the true variability of cap sizes . To extract an estimate for the intrinsic standard deviation of the number of EB binding sites σEB , we assume that the standard deviation of the Mal3-GFP signal on the microtubule lattice σlattice can be considered a rough estimate of the measurement noise . Hence the intrinsic standard deviation of the EB sites at washout and at catastrophe is approximatelyσEB=σMal32−σlattice2 ( ii ) kMal3 is calculated from km , vs , vg assuming that their errors are small compared to the dispersion of cap sizes . The standard deviations of the Mal3-GFP intensities at washout and catastrophe are then propagated according to standard rules of error propagation . The expected mean shrinkage length is ( 8 ) Lshrink=vsTMal3 and a rough estimate for its standard deviation is obtained by error propagation of the standard deviation of TMal3 ( Figure 5D right ) . For simplicity , we have so far approximated the density distribution of the EB binding sites by a mono-exponential decay , as described above . We previously reported evidence for a more complex distribution suggesting a delay between tubulin addition and EB binding site generation ( Maurer et al . , 2014 ) . This delay could be described by an additional kinetic step with a rate k1 that is fast compared to the decay rate , km , of the EB binding sites . This 2-step kinetic model leads to modified expressions for the threshold models: Assuming a kinetic process A→k1B→kmC , where A is a non-binding but stabilizing end state , B is the EB binding state , and C is the mature lattice state , one finds the time course after washout of the end density of the sum of A and B sites: ( 20 ) nA+B ( x0' , t ) = ( 13/8nm ) / ( k1−km ) [k1e−km ( vs/vg+1 ) t−kme−k1 ( vs/vg+1 ) t] Setting t=Tend2 ( the delay time according to the 2-step end density threshold model ) and nA+B ( x0' , Tend2 ) =ncrit2 ( the critical threshold density required for microtubule stability according to the 2-step threshold model ) , we performed a fit to the data solving Tend2=f ( k1 , km , vs , vg , ncrit2 ) numerically using Matlab . For the time course of the total number of all A and B sites one finds ( 21 ) NA+B ( x0' , t ) = ( 13/8nm ) vg/ ( k1−km ) [ ( k1/km ) e−km ( vs/vg+1 ) t− ( km/k1 ) e−k1 ( vs/vg+1 ) t] As above , setting t=Ttot2 and NA+B ( x0' , Ttot2 ) =Ncrit2 , we performed a fit to the data solving Ttot2=f ( k1 , km , vs , vg , Ncrit2 ) numerically using Matlab . Comparing Equation 20 and 9 , and Equation 21 and 4: for k1≫km the expressions for the end density and total number of A and B sites for a 2-step maturation process transform into the expressions of the simpler models used throughout this study that assume only one maturation step with rate constant km .

Much like the skeleton supports the human body , a structure called the cytoskeleton provides support and structure to cells . Part of this cytoskeleton is made up of small tubes called microtubules that – unlike bones – can shrink and grow very quickly . This allows the cell to change shape , move and split into two new cells . Exactly how the microtubules switch between growing and shrinking was not clear . One suggestion is that a protective cap at the end of microtubule allows it to keep growing and prevents it from shrinking . However , the nature and size of this cap have been debated . Now , Duellberg et al . have measured the caps of microtubules with high precision by combining the techniques of microfluidics , TIRF microscopy and recently developed image analysis tools . This revealed that the cap sizes change , with longer caps being more stable . In addition , proteins called end-binding proteins can destabilize the cap by binding to it . This allows microtubules to switch from a growing to a shrinking state more often . Future work could now investigate how changes in cap length cause the microtubules to switch from growing to shrinking . It also remains to be seen whether other proteins also influence the cap to control this switching behaviour .

2016

The size of the EB cap determines instantaneous microtubule stability

We generated a library of ~1000 Drosophila stocks in which we inserted a construct in the intron of genes allowing expression of GAL4 under control of endogenous promoters while arresting transcription with a polyadenylation signal 3’ of the GAL4 . This allows numerous applications . First , ~90% of insertions in essential genes cause a severe loss-of-function phenotype , an effective way to mutagenize genes . Interestingly , 12/14 chromosomes engineered through CRISPR do not carry second-site lethal mutations . Second , 26/36 ( 70% ) of lethal insertions tested are rescued with a single UAS-cDNA construct . Third , loss-of-function phenotypes associated with many GAL4 insertions can be reverted by excision with UAS-flippase . Fourth , GAL4 driven UAS-GFP/RFP reports tissue and cell-type specificity of gene expression with high sensitivity . We report the expression of hundreds of genes not previously reported . Finally , inserted cassettes can be replaced with GFP or any DNA . These stocks comprise a powerful resource for assessing gene function . Knowing where a gene is expressed and where the encoded protein is localized within the cell provides critical insight into the function of almost any gene ( Kanca et al . , 2017 ) . The use of antibodies and molecular manipulation of genes have provided key tools to assess gene expression and protein localization in Drosophila . For example , thousands of P-element mediated enhancer detectors have been used to assess expression patterns ( Bellen et al . , 2011; Bellen et al . , 1989; Bier et al . , 1989; O'Kane and Gehring , 1987; Wilson et al . , 1989 ) . The original enhancer trap vectors were based on the presence of a relatively weak , neutral promoter driving lacZ that can be acted upon by adjacent enhancers as P elements often insert in 5’ regulatory elements ( Bellen et al . , 2011; Spradling et al . , 2011 ) . In adapting a powerful binary expression system first developed in yeast ( Fischer et al . , 1988 ) for use in Drosophila , Brand and Perrimon ( 1993 ) replaced lacZ with GAL4 to induce expression of UAS-effectors ( e . g . GFP , cDNAs , shRNAs ) . They showed that this technology allowed labeling of cells to assess gene expression patterns and drive expression of cDNAs ( Brand and Perrimon , 1993 ) . This binary system has been used to perform tissue-specific knockdown using UAS-RNAi constructs ( Dietzl et al . , 2007; Ni et al . , 2009 ) , carry out intersectional approaches to refine expression patterns in select neuronal populations via Split-GAL4 technology ( Luan et al . , 2006 ) , perform stochastic neuronal labeling approaches via MARCM ( Mosaic Analysis with a Repressible Cell Marker ) ( Lee and Luo , 2001 ) , block synaptic transmission or induce neuronal excitation to assess neuronal activity ( Rosenzweig et al . , 2005; Sweeney et al . , 1995 ) , as well as numerous other manipulations ( Venken et al . , 2011b ) . We previously developed the MiMIC ( Minos-Mediated Insertion Cassette ) technology to permit integration of any DNA cassette at a site where the MiMIC transposable element is inserted ( Venken et al . , 2011a ) . We created fly stocks with nearly 17 , 500 MiMIC insertions and characterized their properties ( Nagarkar-Jaiswal et al . , 2015b; Venken et al . , 2011a ) . MiMICs contain two ϕC31 attP sites that can be used to exchange the integrated cassette with diverse cassettes containing two attB sites through Recombinase Mediated Cassette Exchange ( RMCE ) ( Bateman et al . , 2006; Groth et al . , 2004; Kanca et al . , 2017; Nagarkar-Jaiswal et al . , 2015b; Venken et al . , 2011a ) . We used RMCE to generate a library of protein trap lines where we inserted a cassette consisting of SA ( Splice Acceptor ) -GFP-SD ( Splice Donor ) ( short for SA-GSS-EGFP-FIAsH-StrepII-TEV-3XFlag-GSS-SD , also abbreviated GFSTF , GFP-tag ) into 400 MiMICs inserted in coding introns ( introns flanked by two coding exons ) ( Nagarkar-Jaiswal et al . , 2015a; Nagarkar-Jaiswal et al . , 2015b ) . The synthetic GFP exon is spliced into the mRNA of the gene , leading to the translation of a protein with an internal GFP tag . This intronic GFP tagging approach allows us to determine which cells express the corresponding gene/protein and assess subcellular protein distribution . Importantly , ~75% of intronically tagged genes appear functional ( Nagarkar-Jaiswal et al . , 2015b ) . These endogenous GFP-tagged lines provide an excellent tool to survey subcellular distribution of the encoded proteins . In addition , the GFP tagged proteins can be knocked down in a spatially and temporally restricted fashion , and loss of the GFP-tagged protein is reversible using the deGradFP technique as long as the gene is actively transcribed ( Caussinus et al . , 2011 ) , allowing elegant in vivo manipulation ( Nagarkar-Jaiswal et al . , 2015b ) . More recently , Diao et al . ( 2015 ) developed a T2A-GAL4 technology , named Trojan GAL4 , that integrates a cassette consisting of a SA-T2A-GAL4-polyA ( polyadenylation signal ) in coding introns of genes that carry MiMICs to assess the expression pattern of genes and measure or block neuronal activity ( Diao et al . , 2015; Gnerer et al . , 2015 ) . The polyA should arrest transcription of the gene in which the MiMIC is inserted , generating a truncated transcript . T2A is a viral ribosomal skipping site that arrests translation , which becomes reinitiated after the site , producing untagged GAL4 protein ( Diao and White , 2012 ) . The ability to replace intronic MiMICs with T2A-GAL4 opens many avenues that are complementary to tagging genes that carry intronic MiMICs with SA-GFP-SD ( the GFSTF tag ) . Indeed , T2A-GAL4 could allow determination of expression patterns , notably including in tissues or cells where genes are expressed at such low levels that they cannot easily be detected using the GFSTF tag approach . Although , driving UAS-GFP with GAL4 amplifies expression levels and greatly increases sensitivity , subcellular localization information is lost . In addition , SA-T2A-GAL4-polyA should cause a severe loss-of-function mutation ( i . e . a truncated transcript due to the polyA signal ) unless the SA allows exon skipping ( Rueter et al . , 1999 ) or the truncated protein is functional . Moreover , integration of a transgene carrying a UAS-cDNA for the gene that is mutated ( GOI , gene of interest ) should rescue phenotypes induced by insertion of a SA-T2A-GAL4-polyA cassette , allowing quick and efficient structure-function analyses ( Bellen and Yamamoto , 2015 ) . Finally , numerous other manipulations based on GAL4/UAS technology can be explored to assess function including those of species homologues , to query neuronal connectivity , impair activity , ablate cells , or assess gene or cellular functions , as well as various other applications ( Kanca et al . , 2017; Venken et al . , 2011b ) . So far , about 50 genes have been reported to be tagged with a Trojan-GAL4 cassette ( Chao et al . , 2017; Conway et al . , 2018; Diao et al . , 2015; Diao et al . , 2016; Hattori et al . , 2017; Krüger et al . , 2015; Lee et al . , 2018; Li et al . , 2017; Liu et al . , 2017; Poe et al . , 2017; Skeath et al . , 2017; Toret et al . , 2018; Wu et al . , 2017; Yoon et al . , 2017 ) . Hence , the power and generality of this technology remains to be explored . The potential usefulness of a large collection of T2A-GAL4 insertion fly stocks led us to create a large library; assess the features , properties , and robustness of the T2A-GAL4 method; and explore some of the potential applications of the technology . Here , we report the conversion of 619 intronic MiMICs with T2A-GAL4 . Given that there are only ~1860 genes containing MiMICs inserted between coding exons that can be used for tagging with T2A-GAL4 ( Nagarkar-Jaiswal et al . , 2015b ) , we tested a number of vectors for CRISPR-mediated integration and eventually developed a vector and an efficient , gene-specific protocol for T2A-GAL4 insertion that we named CRIMIC ( CRISPR-Mediated Integration Cassette ) . Using this approach , we tagged 388 genes using CRIMIC . We characterized genetic features associated with these T2A-GAL4 insertions , document numerous novel expression patterns , and provide compelling evidence that this library of ~1000 strains will permit a wide variety of elegant and highly valuable genetic , cell biological , and neurobiological applications . As a part of the Gene Disruption Project , we created and sequenced the flanks of ~15 , 660 MiMIC insertions ( Nagarkar-Jaiswal et al . , 2015b; Venken et al . , 2011a ) . Of these 2854 are intronic insertions that permit tagging of 1862 different genes . We classified 1399 insertions as ‘Gold’ as they are predicted to tag all transcripts annotated in FlyBase , 550 are ‘Silver’ and tag more than 50% of all gene transcripts , whereas 193 are ‘Bronze’ and tag less than 50% of the transcripts . As some genes are tagged with multiple MiMICs , the total is greater than 1862 . We prioritized the tagging of 881 genes that have one or more human homolog ( DIOPT Score ≥4 ( Hu et al . , 2011 ) ) and are part of the ‘Gold’ collection ( Nagarkar-Jaiswal et al . , 2015b; Yamamoto et al . , 2014 ) . In addition , 139 ‘Gold’ MiMICs in genes with low-confidence orthologs ( DIOPT Score ≤3 ) or not conserved in humans were also selected , along with a number of ‘Silver’ and ‘Bronze’ insertions ( see Flypush: http://flypush . imgen . bcm . tmc . edu/MIMIC/lines . php ) . We successfully tagged 611 genes with GFSTF ( Nagarkar-Jaiswal et al . , 2015a; Nagarkar-Jaiswal et al . , 2015b; Venken et al . , 2011a ) , and 211 in this work . We previously showed that conversion of MiMICs with GFSTF allows for efficient tagging of genes that carry intronic MiMICs and that 90% of intronically GFP-tagged proteins show robust GFP signals in third instar larval brains ( Nagarkar-Jaiswal et al . , 2015b ) . However , staining of adult brains revealed robust expression in only ~19% of the GFP-tagged genes tested ( 114/611 , Figure 1—figure supplement 1 ) . To achieve higher adult brain expression we prioritized genes based on the presence of human homologs and converted 619 MiMIC insertions to SA-T2A-GAL4-polyA ( see Flypush: http://flypush . imgen . bcm . tmc . edu/pscreen/rmce ) . We generated both GFP ( GFSTF ) and T2A-GAL4 tagged lines by converting the same original MiMIC line through RMCE and compared the expression patterns for 104 genes , to assess if expression was consistently increased . Figure 1A shows expression in third instar larvae and adult brains of four proteins tagged with GFP . The expression and localization of the proteins encoded by nAChRalpha1 , dpr15 , Pxn and Gprk2 are easily detectable in third instar larval brains and ventral nerve cords , yet exhibit weak or no detectable signals in adult brains . In contrast , the gene expression pattern visualized using T2A-GAL4 converted MiMICs and assayed with UAS-mCD8::GFP ( Figure 1B and C ) exhibits robust GFP signals in third instar and adult brains . This method of integrating the T2A-GAL4 is very efficient and is less time consuming than integrating GFSTF ( Nagarkar-Jaiswal et al . , 2015b ) , as RMCE-mediated conversion events can be easily detected by scoring insertion events crossed to UAS-2xEGFP and screening for expression in any tissue in embryos , larvae , or adults ( Diao et al . , 2015 ) . We previously showed that genes tagged with GFSTF faithfully reproduce the expression and subcellular distribution pattern of all tagged proteins tested ( Nagarkar-Jaiswal et al . , 2015b ) . We confirmed this observation as the similarities between GFSTF localization ( Figure 1—figure supplement 1 ) and published antibody staining for Cactus ( Zhou et al . , 2015 ) , Rgk1 ( Murakami et al . , 2017 ) , Discs large 1 , and Bruchpilot ( Nagarkar-Jaiswal et al . , 2015b ) in the brain are obvious . However , GAL4 strongly amplifies the expression of UAS-mCD8::GFP when compared to the endogenous GFP tagged proteins but the subcellular protein distribution is lost . As shown in Figure 1—figure supplement 2 , in non-neuronal tissue the expression patterns as gauged with mCD8::GFP driven by T2A-GAL4 or antibody staining overlap significantly for arm in larval wing disc , Mhc in larval muscle , and osa in larval eye-antenna imaginal discs ( Figure 1—figure supplement 2A ) when assessed at low resolution . However , for trio , which encodes a Rho guanyl-nucleotide exchange factor that regulates filamentous actin , the expression patterns do not overlap extensively , even at low resolution . Trio is known to play a role in the mushroom body ( MB ) neurons ( Awasaki et al . , 2000 ) as well as in motor neurons at neuromuscular junctions ( NMJs ) ( Ball et al . , 2010 ) . However , the localization of the Trio protein ( Figure 1—figure supplement 2A red , bottom row ) in the larval central brain and ventral nerve cord ( VNC ) appears different from the GAL4 >UAS-mCD8::GFP pattern since GFP is strongly expressed throughout the MB and VNC , whereas the expression of Trio protein is low in VNC and the protein is localized to NMJs ( red staining , insert ) . Similarly , we observe that mCD8::GFP driven by T2A-GAL4 is also present at the NMJs ( green staining , inset ) . In summary , the data are consistent and suggest that Trio is expressed in many neurons , including the motor neurons . A comparison of the expression patterns of four genes tagged with both GFSTF and T2A-GAL4>mCD8::GFP exemplifies differences in the expression patterns . As shown in Figure 1—figure supplement 2B , the patterns of SIFaR , zip , VGlut and mbl are difficult to reconcile without further characterization . In summary , both the T2A-GAL4 and the GFSTF conversions provide valuable information and should permit different applications . In order to vastly expand the collection of MiMIC-tagged genes , we initially tried to use CRISPR technology to insert MiMIC-like constructs and developed two vectors , pM14 and pM36 . pM14 contains a MiMIC-like cassette ( attP-FRT-SA-3XSTOP-polyA-3xP3-EGFP-FRT-attP ) whereas pM36 lacks the FRT sites present in pM14 ( Figure 2A ) . Homology arms approximately 500–1000 bp in length were added to each side of these cassettes by Golden Gate Assembly ( GGA ) ( Engler et al . , 2008 ) to generate donor plasmids for homology directed repair ( Figure 2B ) . To ensure similar and clean genetic backgrounds for all transformation experiments , we isogenized the second and third chromosomes of the nos-Cas9 flies into which we injected our constructs . We used the FindCRISPR tool which is based on a pre-computed database of CRISPR sgRNA designs requiring the presence of a PAM sequence at the end and a unique seed region ( Housden et al . , 2015 ) . All sgRNA designs used the reference genome from FlyBase . Homology arms were amplified from genomic DNA from the isogenized nos-Cas9 injection lines . The mix of sgRNAs and donor vectors was injected into embryos expressing Cas9 , under the nanos promoter ( nos-Cas9 ) , to ensure germline expression ( Kondo and Ueda , 2013; Ren et al . , 2013 ) for integration into introns of the GOI in a directional manner ( Casini et al . , 2015 ) . We injected constructs for 89 genes with pM14 with a success rate of 57% , and 114 genes with pM36 with a success rate of only 26% ( Figure 2—figure supplement 1A ) . The insertion efficiencies of these constructs were deemed too low , and thus they are no longer used in our production pipeline . The utility of the T2A-GAL4 lines generated by RMCE of MiMICs encouraged us to use CRISPR/Cas9 ( Zhang et al . , 2014 ) to insert SA-T2A-GAL4-polyA in introns of GOI using the CRISPR/Cas9 system , similar to the T-GEM vector developed by Diao et al . ( 2015 ) . However , we added flanking FRT sites to allow excisions of the cassette with Flippase . We therefore designed a set of vectors with a swappable MiMIC-like cassette that contains attP-FRT-SA-T2A-GAL4 ( with phases 0 , 1 , and 2 ) -polyA- 3xP3-EGFP-FRT-attP named pM37 ( Figure 3A ) . Upon many trials we settled on injecting 25 ng/µl of a single sgRNA and 150 ng/µl of the -SA-T2A-GAL4-polyA- donor construct ( pM37 ) with ~1 kb homology arms on either side in isogenized nos-Cas9 flies ( Housden et al . , 2016; Housden and Perrimon , 2016 ) . As summarized in Figure 4A , we injected approximately 500 embryos for each of 557 different genes . The fly crosses for each target chromosome are documented in Supplementary file 1 . The percentage of injected embryos surviving to first instar was 23% and on average 4 . 6 flies expressing GFP in the eye ( 3xP3-GFP ) were recovered per injection . Molecular analysis of lines started from each individual GFP+ fly revealed that at least one insertion in the GOI was obtained for nearly 70% of the genes ( Figure 4A ) . All insertions were confirmed by PCR ( see Materials and Methods or Flypush for protocols and corresponding primers; Figure 2—figure supplement 1B ) . Note that the efficiency is higher if we omit the data for genes that map to the third chromosome as the nos-Cas9 transgene insertion on the second chromosome carries a recessive lethal mutation , reducing the efficiency significantly . Alternative nos-Cas9 insertions on the second and X chromosomes are being tested to improve the efficiency . To assess expression patterns of the GOIs , we crossed the transgenic flies to UAS-mCD8::RFP , which labels cell membranes ( Belenkaya et al . , 2008 ) and thus can be easily distinguished from the 3XP3-GFP tag , which is used as a selectable marker for transgenesis and is sparsely expressed in the nervous system ( Figure 3B ) . As shown in Figure 3C , the insertions in different genes produce a variety of expression patterns . For ten genes picked at random , several different independently isolated and sequenced insertions for a given gene exhibited very similar expression patterns , suggesting that the method is robust . The design of pM37 and the ability to use CRISPR should provide the following advantages: first , the ability to insert the CRIMIC cassettes in sites that affect all transcripts encoded by a gene and create severe loss-of-function or null alleles ( Figure 4—figure supplement 1A ) ; second , the ability to excise the mutagenic cassette in vivo ( revert ) using UAS-FLP under the control of GAL4 inserted in the GOI to assess if the CRIMIC cassette is indeed responsible for the observed phenotypes ( Figure 4—figure supplement 1B ) ; third , the ability to revert loss-of-function phenotypes in any tissues at any time to assess when a protein is required and if loss of the gene causes a permanent or reversible phenotype at the time of excision; fourth , the ability to choose an integration site that does not disrupt protein domains upon retagging with GFSTF ( Figure 4—figure supplement 1C ) ; fifth , the ability to insert any DNA flanked by attB sites and replace the SA-T2A-GAL4-polyA cassette . These include the following available cassettes: GFSTF , mCherry , GAL80 , LexA , QF , and split-GAL4 ( Diao et al . , 2015; Venken et al . , 2011a ) . Finally , the ability to test for rescue of the mutant phenotypes by driving the corresponding UAS-cDNA , a feature that also allows for structure-function analysis ( Figure 4—figure supplement 1D ) . Insertion of a SA-T2A-GAL4-polyA in a coding intron should arrest transcription at the polyA signal ( PAS or AATAAA ) unless the site is masked ( Berg et al . , 2012 ) . Hence , MiMIC and CRIMIC T2A-GAL4 insertions should cause a severe loss-of-function mutation in most but not all cases , depending on where the SA-T2A-GAL4-polyA is inserted and whether or not all transcripts are effectively disrupted by the cassette ( Figure 4—figure supplement 1A ) . To test the mutagenic capacity of the T2A-GAL4 cassette , we selected insertions in 100 genes ( 82 MiMIC-derived insertions and 18 CRIMICs , Supplementary file 2 ) that are annotated in FlyBase ( http://flybase . org/ ) as essential genes , based on previous publications . Of these , 80 were categorized as ‘Gold’ , 14 as ‘Silver’ and six as ‘Bronze’ ( Supplementary file 2 ) . We performed complementation tests using 99 molecularly defined deficiencies ( Dfs ) that remove the affected gene ( Parks et al . , 2004; Ryder et al . , 2004 ) and one P-element insertion for Cka ( Supplementary file 2 ) . As shown in Figures 4B , 90 insertions fail to complement the lethality , five are semi-lethal ( less than 5% escapers ) , and five are viable ( see Discussion ) . Because the SA-T2A-GAL4-polyA cassette should prematurely terminate transcription , and as the cassette in CRIMICs is flanked by FRT sequences , we next tested if the lethality associated with eleven insertions can be reverted by using the GAL4 to drive UAS-FLP ( Figure 4—figure supplement 1B ) . We tested excision of 11 CRIMIC T2A-GAL4 insertions in essential genes on the X chromosome by simply crossing them with UAS-FLP . As shown in Figure 4C , eight out of eleven hemizygous lethal insertions on the X chromosome produced numerous viable flies when crossed to UAS-FLP . To assess the efficiency of FLP/FRT mediated CRIMIC cassette excision for the three genes for which we did not observe viable flies ( Dsor1 , Raf and Marf ) , we tested if the T2A-GAL4/+;+/+;UAS-FLP/+ females lacked the 3xP3-GFP marker associated with the T2A-GAL4 insertions . As shown in Figure 4—figure supplement 2 , these flies did not express or barely expressed GFP in the eye , indicating that the efficiency of FLP-mediated excision is high . Given the rescue failure , we also tested whether these lines carry second-site recessive lethal mutations . However , all three are rescued by a genomic P[acman] clone ( Table 1 ) indicating that these chromosomes do not carry second-site lethal mutations . All together , we conclude that cassette excision can revert the phenotype in most cases , providing a simple and powerful tool to assess the requirement for a gene product in a variety of cells and assess if the phenotype of interest is caused by the loss-of-function of the GOI ( see Discussion ) . Expression of GAL4 may allow rescue of the lethality associated with an insertion by driving expression of a UAS-cDNA in a pattern that corresponds to the gene ( Figure 4—figure supplement 1D ) . However , this may not be effective in many cases as the vast majority of genes have more than one splice isoform , and rescue with any one isoform encoded by a UAS-cDNA construct might not work ( Table 1 ) . In addition , many cDNAs are tagged at the C-terminal end and it has been estimated that about 22% of the genes tagged with 3XHA ( Bischof et al . , 2013 ) and 33% tagged with GFP disrupt gene function ( Sarov et al . , 2016 ) . Moreover , since the GAL4/UAS system is an over-expression system , cDNA rescue may not be possible for genes that are sensitive to dosage . Nevertheless , we assessed the ability of a single UAS-cDNA per gene to rescue mutant phenotypes associated with disruption of 36 genes for which we were able to find a UAS-cDNA ( Bischof et al . , 2013; Gramates et al . , 2017 ) . For 11 genes on the X-chromosome , we assessed rescue of male lethality , whereas for genes on the second and third chromosomes , we assessed rescue of SA-T2A-GAL4-polyA-induced lethality over the corresponding Dfs that fail to complement the lethality . To ensure that the lethality of the genes on the X-chromosome is indeed associated with the insertions , we first performed genomic rescue using the 80 kb P[acman] BAC transgenic lines ( Venken et al . , 2010 ) . The lethality of all genes on the X-chromosome was rescued with the corresponding P[acman] clones ( Table 1 ) , indicating that these chromosomes are very unlikely to carry second-site mutations . Of the 36 essential genes that carry SA-T2A-GAL4-polyA , 25 ( ~70% ) could be rescued by a single UAS-cDNA driven by the endogenous GAL4 ( Figure 4D; Table 1 ) . The sensitivity of T2A-GAL4 tagging allows us to determine where genes are expressed , especially when expression levels in specific cell populations are low , as shown for the adult brain in Figure 1 . We therefore determined the expression patterns of 550 genes in adult brains and documented expression patterns of many genes that have not been previously reported ( Gramates et al . , 2017 ) ( Figure 5; Figure 5—figure supplements 1–3 ) . Nearly 80% of all tested tagged genes are expressed in adult brains . The smallest category of genes ( 9/550 or 2% ) corresponds to genes expressed in trachea , a tubular system that provides oxygen to all tissues ( Varner and Nelson , 2014 ) . For example , breathless ( btl ) encodes a protein kinase expressed specifically in the trachea and is involved in tracheal branching ( Lee et al . , 1996 ) . A comparison of the GAL4>UAS-mCD8::GFP expression pattern of a GAL4 based P-element enhancer detector in btl ( P[GawB]btlNP6593 ) ( Hayashi et al . , 2002 ) and the T2A-GAL4 insertion ( MI03286-TG4 . 0 ) in the brain and thoracicoabdominal ganglion ( TAG ) show very similar mesh-like tracheal patterns . Another gene previously documented to be expressed in trachea , empty spiracles ( emp ) , also shows that the T2A-GAL4 insertion drives expression in trachea ( Hart and Wilcox , 1993 ) . In Figure 5 and Figure 5—figure supplement 1 , we report the expression of seven other genes that have not been reported to be expressed in trachea ( FlyBase 2 . 0/FB2017_06 ) . Hence , nine genes out of 550 tested are expressed in trachea and for seven of these , detection of expression in the trachea is novel ( Frl , CG8213 , sprt , geko , ex , Samuel , Cad96Ca ) . The next most frequent category consists of genes whose expression are mostly confined to a subtype of cells corresponding to glia . Glia account for about 10% of the cells in the fly brain ( Kremer et al . , 2017 ) and about 50% of cells in the mammalian brain ( von Bartheld et al . , 2016 ) . To assess various glial patterns in the brain upon UAS-mCD8::GFP expression , we selected five known glial cell GAL4 drivers as controls: repo-GAL4 ( all glia except midline glia ) , gcm-GAL4 ( embryonic glia ) , NP2222-GAL4 ( cortex glia ) , NP6520-GAL4 ( ensheathing glia ) and NP1243-GAL4 ( astrocyte-like glia ) ( Awasaki and Lee , 2011 ) . We identified 19/550 genes that are mostly or specifically expressed in one or several types of glia cells . Seven were previously shown to be expressed in glia: CIC-a , loco , CG10702 , CG6126 , Gs2 , Egfr and Tret1-1 ( Figure 5; Figure 5—figure supplement 2 ) , whereas 12 have not previously been associated with glial expression based on available data ( bdl , Zasp52 , rols , ine , CG5404 , CG14688 , CG31663 , ry , CG4752 , βTub97EF , CG32473 , LManII; Figure 5 and Figure 5—figure supplement 2 ) . Note that ry ( rosy ) is known to be expressed in pigment cells of the eye ( Keller and Glassman , 1965 ) , and that these cells function as glial cells in this organ ( Liu et al . , 2017 ) . Finally , about 80% of lines showed expression patterns in adult brain neurons . Given the complexity of the brain and the sheer number of different expression patterns in neurons , we decided to focus on a single neuronal population that is easily identifiable and on expression patterns that were not previously documented . We selected the neurons of the pars intercerebralis ( PI ) , which are located on the dorsal medial side of the brain and project to the tritocerebrum and the corpora cardiaca in the middle central area ( Nässel et al . , 2013 ) . They secrete a variety of neuropeptides as well as Drosophila Insulin Like Peptides or DILPs ( Rulifson et al . , 2002 ) . This cluster of neurons is a neuroendocrine command center that not only controls cell growth by releasing DILPs but also controls fly behaviors , including aggression , via secreted neuropeptides ( Davis et al . , 2014; de Velasco et al . , 2007 ) . The gene IIp2 encodes Insulin-like peptide 2 . An Ilp2 promotor GAL4 fusion ( P{Ilp2-Gal4} ) ( Broughton et al . , 2005 ) was used to express mCD8::GFP in a subset of PI neurons as a positive control . The expression of GFP in PI neurons driven by T2A-GAL4 insertions in AstA-R2 ( Allatostatin A receptor 2 ) and Lkr ( Leucokinin receptor ) , agrees well with previous observations of their expression in these neurons ( Cannell et al . , 2016; Hentze et al . , 2015 ) , In addition , we found 18 genes ( CG31547 , if , NimB2 , Lerp , CG7744 , cnc , CG2656 , spin , gem , Fs , Aldh-III , CG33056 , grsm , CG31075 , Pi3K68D , Dh44-R2 , Lgr4 , Atg16 ) that are expressed in PI neurons and yet have not been previously described as such ( FlyBase 2 . 0/FB2017_06 ) ( Figure 5; Figure 5—figure supplement 3 ) . Here , we report the creation of ~1000 T2A-GAL4 lines by two different methods: 619 generated by RMCE of MiMIC insertions and 388 by CRIMIC , a novel CRISPR-mediated strategy . Our success rate of MiMIC T2A-GAL4 conversion was 68 . 1% ( 543/797 ) upon a single attempt and 41 . 1% ( 76/185 ) upon a second attempt . Hence , we failed twice for 109 out of 797 genes . The T2A-GAL4 insertions not only provide a GAL4 driver that reveals the cells in which the targeted genes are expressed with great sensitivity but also allow many useful applications for testing gene function . We show that the CRIMIC technology is as powerful and reproducible as converting MiMICs with T2A-GAL4 , and we should therefore be able to tag at least half of the genes in the Drosophila genome with the T2A-GAL4 CRIMIC approach as they carry suitable introns that are large enough . While the conversion of MiMICs depends on the presence of intronic MiMIC insertions , the CRIMIC approach allows us to select many genes that do not carry a MiMIC but contain an intron that is large enough and has proper sgRNA target sites to introduce a cassette that carries SA-T2A-GAL4-polyA flanked by FRT sites . The cloning success rate for the donor vector was about 80% on a first attempt , but significantly higher when repeated for another intron . This should allow us to tag about ~45–50% of all fly genes as those with short coding introns or without introns cannot be targeted using this strategy . By injecting ~535 embryos/construct we average a 70% successful integration rate . If we exclude the data for the third chromosome , where the nos-Cas9 isogenized strain used was sub-optimal , our success rate is ~80% . We do not anticipate that we will be able to improve this much in the future except for the third chromosome . However , we are currently developing strategies with much shorter homology arms to avoid cloning and reduce the number of injected embryos , as our approach is labor-and cost-intensive . Indeed , we estimate that each line requires approximately ~50 hr of work for technicians , postdoctoral fellows , and bioinformaticians to obtain a single characterized stock deposited in the BDSC . This technology is based on the properties of the SA-T2A-GAL4-polyA cassette . Issues with efficiency of those properties may limit the use of this cassette . First , skipping of the SA would reduce or abolish the gene-trap function of this cassette , leading to hypomorphic or neutral alleles of the GOI . The SA used here corresponds to intron 18 of Mhc ( Hodges and Bernstein , 1992 ) , a SA that has been used before ( Diao et al . , 2015; Morin et al . , 2001; Nagarkar-Jaiswal et al . , 2015a; Nagarkar-Jaiswal et al . , 2015b; Venken et al . , 2011a; Zhang et al . , 2014 ) . We show that this SA is quite effective , as lethal insertions in essential genes fail to complement the lethality of known alleles and deficiencies in 90% of the cases tested . These data also indicate that a second feature of the cassette , the polyA signal , is efficient at arresting transcription . As previously shown for a few genes ( Diao et al . , 2015; Gnerer et al . , 2015 ) , GAL4 drives UAS-GFP or RFP expression efficiently in all cases tested and permits detection of expression in cells that express low mRNA and protein levels ( Figure 1 and Figure 3 ) . Although the GAL4/UAS binary system strongly enhances the detection sensitivity when compared to the expression of the endogenous gene in the adult head tagged with GFSTF , this is much less the case in the third instar larval CNS ( Figure 1 and Figure 1—figure supplement 2B ) ( Nagarkar-Jaiswal et al . , 2015b ) . We have no obvious explanation for this discrepancy . In summary , although it is impossible to prove that the GAL4 is faithfully mimicking the endogenous expression given its enhanced sensitivity , the data we have compiled so far indicate that these insertions accurately represent the expression of the vast majority of genes . The latter feature is important , as current GAL4-driver resources developed at the Janelia Research Campus and Vienna Drosophila Resource Center ( Jenett et al . , 2012; Pfeiffer et al . , 2008 ) are based on very different premises . The driver transgenes were engineered to label few neurons . Indeed sparse labeling is a prerequisite to study neural networks . Given that the regulatory elements of genes used to create these collections are removed from their endogenous context it is difficult to determine which enhancers mimic a portion of the expression pattern of the gene they have been derived from , as repressors may not be present and enhancers may be truncated or not tested ( FlyLight ) ( Jory et al . , 2012 ) . Hence , it should now be possible to compare the patterns of the genes presented here with those based on GAL4 patterns driven by the ~2–3 kb fragments used in these studies ( Jory et al . , 2012; Pfeiffer et al . , 2008 ) . Given the caveats associated with CRISPR technology ( Doench et al . , 2016 ) , it is important to demonstrate that an observed phenotype is indeed associated with the insertion . In addition , we have previously shown that the genetic manipulations based on MiMIC can induce a significant number of second-site mutations ( Nagarkar-Jaiswal et al . , 2015b; Venken et al . , 2011a ) . We therefore attempted to rescue the lethal phenotypes associated with CRIMIC T2A-GAL4 insertions with UAS-FLP , as this should excise the cassette . We found that 8 of 11 CRIMIC insertions that cause lethality were reverted with UAS-FLP ( Figure 4C ) , providing a quick tool to assess genetic background load . The results of this experiment also indicate that the cassette can be excised with other FLP drivers like LexA or hsp70 promoter driven FLP . Hence , most chromosomes engineered through CRISPR in this study do not carry second-site lethal mutations and this was confirmed with genomic P[acman] rescue constructs: all mutations tested were rescued with the corresponding P[acman] clones ( Venken et al . , 2010 ) ( Table 1 ) . The data also indicate that the delay between FLP production by GAL4 and excision is not critical for most essential genes . Finally , we note that the failure to rescue lethality was not due to a failure of excision for Dsor1 , Raf and Marf . Indeed , flies that express GAL4 and FLP lack GFP expression in the eyes ( Dsor1 ) or produce very little GFP derived from the 3xP3-EGFP marker ( Raf and Marf ) ( Figure 4—figure supplement 2 ) , suggesting that excision of the T2A-GAL4 cassette was successful in all cases tested . Hence , tissue-specific excision should easily be induced using hs-FLP or another binary system ( Venken et al . , 2011b ) , allowing one to perform conditional rescue experiments and assess in some cases when and where proteins are required . In summary , combining the features of T2A-GAL4 with the FLP-mediated excision system provides numerous possibilities . One of the most useful applications of T2A-GAL4 may be the ability to use SA-T2A-GAL4-polyA with UAS-cDNAs to perform structure-function analyses , that is , test the consequences of removal of protein domains and/or of introducing point mutations into the UAS-cDNA construct , or even to test the rescue ability of human cDNAs and variants ( Bellen and Yamamoto , 2015 ) . The odds that this strategy will be effective for the majority of genes seem limited at first glance given the following issues: the test is done with a single cDNA yet two or more protein isoforms are encoded by the vast majority of genes ( Table 1 ) ; there may be issues with expression levels as shown for UAS-GFP versus GFSTF; timing of protein production may be delayed; and finally , tagging of cDNAs ( HA or GFP ) has been documented to impair function for ~20–30% of the tagged cDNAs ( Bischof et al . , 2013; Sarov et al . , 2016 ) . Nevertheless , as shown in Table 1 , about 70% of the UAS-cDNA constructs were able to rescue lethality , despite the fact that nearly all genes tested encode more than one protein isoform . In addition , no obvious pattern emerged from these data with respect to the presence or absence of a tag ( Chi sq . = 0 . 0004 , p=0 . 98 ) , and no pattern emerged with respect to rescue of lethal mutations , as these genes encode anywhere from 1 to 20 protein isoforms but often could be rescued with a single cDNA ( Table 1 ) . Establishing that there is complete rescue of all phenotypes , not just lethality , would be time consuming and require detailed studies including longevity , fertility , and numerous behavioral assays beyond the scope of this work . We note that we also previously showed that intronic tagging with GFSTF disrupted about 25% of the genes ( Nagarkar-Jaiswal et al . , 2015b ) . Hence , we recommend that both tagged and untagged cDNAs be tested whenever possible . In summary , this library provides a set of ~1000 gene-specific GAL4 drivers for the fly community . We are in the process of creating numerous other T2A-GAL4 insertions as part of the Gene Disruption Project and we prioritize genes based on the nomination from scientific community through a web site ( http://www . flyrnai . org/tools/crimic/web/ ) . The GAL4/UAS system is a very well established binary approach and this T2A-GAL4 library will provide numerous additional tools to survey gene and circuit function in combination with many other existing genetic tools such as UAS-RNAi , UAS-fly cDNA , UAS-GCaMP ( Nakai et al . , 2001 ) , UAS-ChR ( Schroll et al . , 2006 ) , UAS-shits ( Kitamoto , 2001 ) and so on . For an estimated 90% of the genes tested , the insertion of SA-T2A-GAL4-polyA causes a severe loss-of-function mutation and only three insertions displayed dominant phenotypes out of ~1000 genes tested . Finally , the T2A-GAL4 flies provide a very useful platform for functional testing of fly as well as human genes and their possible disease variant ( s ) ( Chao et al . , 2017; Chen et al . , 2016; Luo et al . , 2017; Sandoval et al . , 2014; Wangler et al . , 2017; Yoon et al . , 2017 ) . Fly stocks were maintained on standard cornmeal-yeast-agar medium at 25°C , and on a 12/12 hr light/dark cycle . The MiMIC and CRIMIC flies were created in the Bellen lab ( see Flypush or Supplementary file 2 ) . UAS-2xEGFP , hs-Cre , vas-dϕC31 , Trojan T2A-GAL4 triplet flies were from Dr . Ben White ( Diao et al . , 2015 ) . The RMCE conversion of MiMICs with GFSTF and T2A-GAL4 cassettes was described in previous studies ( Diao et al . , 2015; Nagarkar-Jaiswal et al . , 2015a; Nagarkar-Jaiswal et al . , 2015b ) . The crossing schemes for CRIMICs are shown in Supplemental Information 1 . btl-GAL4 , Ilp2-GAL4 , repo-GAL4 , gcm-GAL4 , UAS-mCD8::GFP , UAS-mCD8::RFP , P[acman] flies , and UAS-FLP flies were obtained from the Bloomington Drosophila Stock Center ( BDSC , USA ) . UAS-if was from Dr . Celeste Berg ( Beumer et al . , 1999 ) . NP1243-GAL4 , NP2222-GAL4 , and NP6250-GAL4 are from Kyoto Stock Center ( Kyoto DGGR , Japan ) . Dfs flies were from BDSC or Kyoto DGGR . UAS-cDNA flies were from BDSC or FlyORF ( Switzerland ) . y , w;attP40 ( y+ ) {nos-Cas9 ( v+ ) }/CyO ( Kondo and Ueda , 2013 ) and y , w;+/+; attP2 ( y+ ) {nos-Cas9 ( v+ ) } ( Ren et al . , 2013 ) were isogenized in this work . See Supplementary file 2 for the genotypes and stock numbers of fly stocks . All references to FlyBase are based on FlyBase 2 . 0/FB2017_06 ( Gramates et al . , 2017 ) . TypeIISRE-attP-FRT-SA-3xStop-SV40-3xP3-GFP-SV40-FRT-attP-TypeIISRE fragment was synthesized in two parts by GENEWIZ ( www . genewiz . com ) in the pUC57 vector ( pM5 and pM7 were synthesized by GENEWIZ ) . Next , the ~1 . 2 kb fragment of attP-FRT-SA-3xStop-SV40-3xP3 in pM5 was digested with BstXI and EagI . The ~1 . 3 kb fragment of GFP-SV40-FRT-attP in pM7 was digested with EagI and EcoRV . To generate pM14 , these two DNA fragments were separated and purified from agarose gel and ligated with pBS-deltaBsaI vector which was digested with BstXI and EcoRV ( molar ratio of insert:vector = 5:1 ) . The ligation mix ( 1 μg/8 μl total DNA + 1 μl 10xT4 DNA Ligase Buffer + 1 μl T4 DNA ligase ) was incubated at 16°C overnight then transformed into NEB® Stable E . coli competent cells . Cells were raised on ampicillin ( 50 μg/ml ) /LB agar plate at 37°C overnight . pM14 plasmids were checked by double digestion of BstXI and EcoRV . pM36 was modified from pM14 by removing two FRT sites in pM14 by mutagenesis . pM36 was modified from pM14 by sequentially adding 25 nucleotides flanking each of the attP sites for sequencing the inserted homology arms and mutating the two FRT sites to render them nonfunctional . In brief , a NsiI-EcoRI fragment containing the necessary modifications was cloned by PCR from oligos ( DLK256 = taaatATGCATcgatcgtctggtactacattcacgcGTACTGACGGA CACACCGAAGCccc ( fwd ) and DLK331 = AGAGAGAATTCCTACATGGTAATGT TACTAGAGAATAGGAACTTCTCGCGCTC ( rev ) ) using pM14 as a template and inserted between the NsiI and EcoRI sites to replace the original pM14 sequence , followed by cloning a XbaI-SphI fragment from pM14 with the necessary modifications for the downstream site using the oligos ( DLK332 = TATTCTCTAGAAACATTACCATGTAGTCGCGCTCGCGCGACTGACG ( fwd ) and DLK255 = GGTAGGAAGACAACGCGCAGTGAAGGACGAGAGGTAGTACC GCATGCGTACTGACGGACACACCG ( rev ) ) and replacing the pM14 sequences between the XbaI and SphI sites . pM37 vectors were modified from pM14 by replacing 3xStop with T2A-GAL4 of different phases from pT-GEM vectors of the corresponding phase ( Diao et al . , 2015 ) . Briefly the EcoRI-PstI fragment of pM14 was subcloned in pBluescript SK and mutagenized by PCR mutagenesis to replace 3XStop sequences with AscI restriction enzyme site and subcloned back in pM14 vector . T2A-GAL4 sequences were PCR amplified from pT-GEM vector and cloned in EcoRI/MfeI and AscI sites in mutated pM14 , generating pM14 T2A-GAL4 vector . AscI-SbfI fragment of T2A-GAL4 was resynthesized to remove Type IIS RE sites by substituting base pairs corresponding to Type IIS REs with synonymous mutations eliminating the sites . The resulting fragment was subcloned in pM14 T2A-GAL4 vector . pM14 and pM36 vectors were found to be unstable in bacteria , frequently recombining out the 3XP3-GFP cassette . Further analysis showed that 3XP3 promoter fragment of pM14 and pM36 was 290 bps longer than other vectors that use the same marker . Shortening this fragment by PCR and replacing the AscI-FseI fragment with the shortened fragment improved stability of the vector in bacteria , creating the pM37 vector . Sequences of pM14 , pM36 and pM37 can be found in Supplementary file 3 . We analyzed the introns of all protein-coding genes of Drosophila melanogaster annotated at FlyBase and selected the genes that have at least one CDS intron that is >100 bp and is shared by all isoforms . Based on FlyBase release 6 . 16 , there are 5822 protein-coding genes that meet these criteria . Then , we removed the genes that are covered by the MIMIC Gold collection and prioritized the genes if their human ortholog ( s ) are disease-related ( Hu et al . , 2011 ) . We also prioritized genes based on the nomination from scientific community through a web site ( http://www . flyrnai . org/tools/crimic/web/ ) . sgRNA targeting the qualified CDS introns were selected based on their efficiency score and specificity annotated at Find CRISPR Tool ( Ren et al . , 2013 ) . The homology arms upstream or downstream of the cutting site were designed using Primer3 ( Untergasser et al . , 2012 ) . We required that the homology arms are between 500 and 1200 bp in length , less than 40 bp apart from each other , and free of one or more of the three restriction enzymes ( BsaI , BbsI , BsmBI ) used for cloning . Donor constructs were generated as previously described ( Housden and Perrimon , 2016 ) . Briefly , homology arms were PCR amplified from genomic DNA using Q5 or Phusion polymerase ( NEB ) , run on an agarose gel and purified with the QIAquick Gel Extraction Kit ( Qiagen ) . The homology arms , pBH donor vector and pM14/pM36/pM37 cassette were combined by Golden Gate assembly ( Engler et al . , 2008 ) using the appropriate type IIS restriction enzyme ( BbsI , BsaI , or BsmBI ) . The resulting reaction products were transformed into Stbl3 or TOP10 Chemically Competent Cells ( ThermoFisher ) , and plated overnight under kanamycin selection . Colonies were cultured for 24 hr at 30°C and DNA prepared by miniprep . The entire homology arm sequence and 300–500 bps of the adjacent cassette sequence were verified prior to injection . sgRNA constucts were generated as previously described ( Housden et al . , 2016 ) . Briefly , sense and antisense oligos containing the 20 bp guide target sequence were annealed and phosphorylated with T4 Polynucleotide Kinase ( NEB ) , then inserted between BbsI sites in the pl100 sgRNA expression vector ( Ren et al . , 2013 ) . Ligation products were transformed into TOP10 Competent Cells ( ThermoFisher ) , and plated overnight . Colonies were cultured , DNA prepared by miniprep , and sequences verified prior to injection . We injected a mix of 25 ng/μl sgRNA and 150 ng/μl donor DNA in isogenized fly embryos of the following genotypes y , w; attP40 ( y+ ) {nos-Cas9 ( v+ ) }/CyO ( Kondo and Ueda , 2013 ) and y , w; +/+; attP2 ( y+ ) {nos-Cas9 ( v+ ) } ( Ren et al . , 2013 ) to generate CRIMIC insertions ( Housden et al . , 2016; Housden and Perrimon , 2016 ) . For validation of MiMIC conversion and CRIMIC cassette insertion events , the genomic DNA was extracted from ~20 adult flies using the PureLink Genomic DNA Mini Kit ( Invitrogen ) . For MiMIC conversions , four reactions of PCR were performed with tag-specific primers and MiMIC specific primers as described previously ( Diao et al . , 2015; Venken et al . , 2011a ) . The PCR reaction mix was: 1 μl genomic DNA ( ~10 ng ) , 1 μl primer 1 ( 10 μM ) , 1 μl primer 2 ( 10 μM ) , 4 . 5 μl H2O , and 7 . 5 μl GoTaq Green Master Mix ( Promega ) . Hot start PCR conditions in C100 Touch Thermal Cycler ( Bio-Rad ) were: denaturation at 95° for 1 min , 34 cycles at 95° for 30 s , 56° for 30 s and 72° for 60 s , and post-amplification extension at 72° for 10 min . For CRIMIC cassette insertion , two reactions of PCR were performed with target-specific primers ( see our website at Flypush ) and attP-R primer ( 5’-CCCCAGTTGGGGC-3’ ) ( Figure 2—figure supplement 1 ) . PCR reaction mix was: 1 μl genomic DNA ( ~10 ng ) , 1 μl primer 1 ( 10 μM ) , 1 μl primer 2 ( 10 μM ) , 4 . 5 μl H2O , and 7 . 5 μl GoTaq Green Master Mix ( Promega ) . Hot start PCR conditions in C100 Touch Thermal Cycler ( Bio-Rad ) were: denaturation at 95° for 1 min , 40 cycles at 95° for 30 s , 56° for 30 s and 72° for 2 min 30 s , and post-amplification extension at 72° for 10 min . Virgin female pM37 flies were collected and crossed with male flies carrying a UAS-FLP on the third chromosome . The adult eyes of pM37/+;+/+;UAS-FLP/+ for insertions in Dsor1 , Raf and Marf were imaged with a fluorescent microscope ( Zeiss SteREO Discovery . V20 ) . Confocal imaging was performed as described previously ( Lee et al . , 2011 ) . In brief , dissected adult brains or VNCs were fixed in 4% paraformaldehyde/1xPBS at 4°C overnight , transferred to 2% Triton X-100/1xPBS at room temperature , vacuumed for 1 hr and left overnight in the same solution at 4°C . The larvae brains or other tissues were fixed in 4% paraformaldehyde/1 xPBS at 4°C for at least 2 hr , transferred to 0 . 5% Triton X-100/1xPBS at 4°C for overnight . For immunostaining , the samples were blocked in 10% NGS/0 . 5% Triton X-100/1xPBS and incubated with primary antibodies ( 1:50 ~ 200 dilution ) at 4°C for overnight with shaking , then washed with 0 . 5% Triton X-100/1xPBS for 5 min three times . The secondary antibodies conjugated with Alexa-488 or Alexa-647 ( Jackson ImmunoResearch ) were diluted 1:100 ~ 500 in 0 . 5% Triton X-100/1xPBS and incubated with samples at 4°C for overnight with shaking . For immunostaining of GFP , the samples were incubated with anti-GFP antibody conjugated with FITC ( 1:500 ) ( Abcam ) in 1xPBS with 0 . 5% Triton X-100 for overnight . Samples were cleared and mounted in RapiClear ( SunJin Lab Co . ) and imaged with a Zeiss LSM 880 Confocal Microscope under a 20x or 40x C-Apochromat water immersion objective lens .

Determining what role newly discovered genes play in the body is an important part of genetics . This task requires a lot of extra information about each gene , such as the specific cells where the gene is active , or what happens when the gene is deleted . To answer these questions , researchers need tools and methods to manipulate genes within a living organism . The fruit fly Drosophila is useful for such experiments because a toolbox of genetic techniques is already available . Gene editing in fruit flies allows small pieces of genetic information to be removed from or added to anywhere in the animal’s DNA . Another tool , known as GAL4-UAS , is a two-part system used to study gene activity . The GAL4 component is a protein that switches on genes . GAL4 alone does very little in Drosophila cells because it only recognizes a DNA sequence called UAS . However , if a GAL4-producing cell is also engineered to contain a UAS-controlled gene , GAL4 will switch the gene on . Lee et al . used gene editing to insert a small piece of DNA , containing the GAL4 sequence followed by a ‘stop’ signal , into many different fly genes . The insertion made the cells where each gene was normally active produce GAL4 , but – thanks to the stop signal – rendered the rest of the original gene non-functional . This effectively deleted the proteins encoded by each gene , giving information about the biological processes they normally control . Lee et al . went on to use their insertion approach to make a Drosophila genetic library . This is a collection of around 1 , 000 different strains of fly , each carrying the GAL4/stop combination in a single gene . The library allows any gene in the collection to be studied in detail simply by combining the GAL4 with different UAS-controlled genetic tools . For example , introducing a UAS-controlled marker would pinpoint where in the body the original gene was active . Alternatively , adding UAS-controlled human versions of the gene would create humanized flies , which are a valuable tool to study potential disease-causing genes in humans . This Drosophila library is a resource that contributes new experimental tools to fly genetics . Insights gained from flies can also be applied to more complex animals like humans , especially since around 65% of genes are similar across humans and Drosophila . As such , Lee et al . hope that this resource will help other researchers shed new light on the role of many different genes in health and disease .

2018

A gene-specific T2A-GAL4 library for Drosophila

Flatworms number among the most diverse invertebrate phyla and represent the most biomedically significant branch of the major bilaterian clade Spiralia , but to date , deep evolutionary relationships within this group have been studied using only a single locus ( the rRNA operon ) , leaving the origins of many key clades unclear . In this study , using a survey of genomes and transcriptomes representing all free-living flatworm orders , we provide resolution of platyhelminth interrelationships based on hundreds of nuclear protein-coding genes , exploring phylogenetic signal through concatenation as well as recently developed consensus approaches . These analyses robustly support a modern hypothesis of flatworm phylogeny , one which emphasizes the primacy of the often-overlooked ‘microturbellarian’ groups in understanding the major evolutionary transitions within Platyhelminthes: perhaps most notably , we propose a novel scenario for the interrelationships between free-living and vertebrate-parasitic flatworms , providing new opportunities to shed light on the origins and biological consequences of parasitism in these iconic invertebrates . The true flatworms ( Platyhelminthes ) are one of the major phyla of invertebrate animals , the significance of which may be practically measured in terms of their species diversity and body-plan disparity , as well as from a more theoretical perspective by their role in broader-scale discussions of metazoan phylogeny and as models of fundamental concepts in developmental and stem cell biology , parasitology , and invertebrate zoology . As small acoelomate animals , the free-living members of this phylum ( ‘turbellaria’ ) almost without exception rely on their fully ciliated , non-cuticularized epidermis for all locomotory , respiratory , and circulatory functions , fundamentally constraining them to protected aquatic or humid habitats ( Hyman , 1951 ) . Despite this restriction , they have successfully radiated in almost all marine and continental aquatic habitats and many humid terrestrial settings , today numbering perhaps tens of thousands of free-living species ( Appeltans et al . , 2012; Tyler et al . , 2012 ) , of which about 6500 are currently described . The acoelomate condition of Platyhelminthes , among other traits ( e . g . , their blind gut ) , has also historically positioned them prominently as figures of supposedly ‘primitive’ Bilateria . While molecular phylogenetics has for over a decade nested this taxon well within the protostome clade Spiralia ( Carranza et al . , 1997; Baguñà and Riutort , 2004 ) , displacing them from their classical position as early-branching bilaterians , modern manifestations of the debate over the relevance of such characters continue , with the role of acoelomate early-branching bilaterians ( but see Philippe et al . , 2011 ) being taken over by Xenacoelomorpha ( Hejnol et al . , 2009; Srivastava et al . , 2014 ) , themselves formerly Platyhelminthes . This fragmentation of the phylum is not , however , fully incompatible with the classical interpretation of the ‘primitive’ nature of some aspects of platyhelminth organization , and indeed interest in this debate is resurging with , for example , recent molecular phylogenetic evidence for the paraphyly of ‘Platyzoa’ ( an assemblage of small acoelomate and pseudocoelomate spiralians including Platyhelminthes , Gastrotricha , and Gnathifera [Struck et al . , 2014; Laumer et al . , 2015] ) . Irrespective of the broader evolutionary implications of pan-platyhelminth characteristics , the clade is also widely known for those of its members which have been adopted as models of fundamental zoological concepts . Freshwater planarians such as Schmidtea mediterranea ( Tricladida ) have a long history of utility in classical zoology , and modern molecular genetic appropriations of this system , as well as the more recently developed model Macrostomum lignano ( Macrostomorpha ) ( Ladurner et al . , 2005 ) , have provided insights into especially non-embryonic developmental processes inaccessible in other familiar invertebrate models , such as whole body regeneration ( Sánchez Alvarado , 2012 ) , stem-cell maintenance ( Sánchez Alvarado and Kang , 2005 ) , tissue homeostasis ( Pellettieri and Alvarado , 2007; Reddien , 2011 ) , and aging ( Mouton et al . , 2011 ) . The marine polyclad flatworms ( Polycladida ) have also been a subject of perennial study , not least due to their compelling reproductive biology: although they engage in ( an often elaborately achieved [Michiels and Newman , 1998] ) internal fertilization unlike most other marine macroinvertebrates , their embryos show a clear quartet spiral cleavage and cell fate ( Boyer et al . , 1998 ) , and many species present a long-lived planktotrophic larva ( Rawlinson , 2014 ) with well-developed ciliary bands and cerebral ganglia , which have been homologized to the trochophora larvae of other Spiralia ( Nielsen , 2005 ) . Furthermore , polyclads , due to their large clutch sizes , endolecithal yolk ( Laumer and Giribet , 2014 ) , and thin eggshells , represent the only platyhelminth lineage in which experimental manipulation of embryonic development is possible . Lastly , but far from least , platyhelminths have been long considered masters of parasitism ( Kearn , 1997 ) . Although nearly all ‘turbellarian’ lineages evince some symbiotic representatives ( Jennings , 2013 ) , the flatworm knack for parasitism reaches is zenith in a single clade , Neodermata ( Ehlers , 1985 ) . Indeed , the obligate vertebrate parasitism manifested by this group of ecto- and endoparasitic flukes ( Polyopisthocotylea , Monopisthocotylea , Digenea , and Aspidogastrea ) and tapeworms ( Cestoda ) is perhaps the single most evolutionarily successful adoption of a parasitic habit in the animal kingdom ( in contrast to the case of the nematodes , in which vertebrate parasitism has multiple evolutionary origins [Dieterich and Sommer , 2009] ) . Central among the adaptations responsible for the success of Neodermata—reflected in its some 40 , 000–100 , 000 estimated species ( Rohde , 1996; Littlewood , 2006 ) —was the invention ( among other synapomorphies [Littlewood , 2006; Jennings , 2013] ) of the eponymous ‘neodermis’ , a syncytial tegument which plays specialized roles in host attachment , nutrient appropriation , and immune system evasion ( Tyler and Tyler , 1997; Mulvenna et al . , 2010 ) . The neodermis has intimately ( and ostensibly , irreversibly [Littlewood , 2006] ) tied the evolutionary success of this lineage to that of its hosts , and as a result , neodermatans appear to have outstripped the diversification of their free-living ancestors by nearly an order of magnitude , with evidence that most vertebrate species ( not to mention many species of intermediate hosts from diverse animal phyla ) are infected by at least one neodermatan flatworm ( Poulin and Morand , 2000; Littlewood , 2006 ) , sometimes with startling host specificity ( particularly in monogenean trematodes ) . Human beings and their domesticated animals have also not escaped the depredations of neodermatans , which include the etiological agents of several diseases of profound incidence , morbidity , and socioeconomic impact ( Berriman et al . , 2009; Torgerson and Macpherson , 2011; Tsai et al . , 2013 ) , such as schistosomiasis ( Gryseels et al . , 2006 ) , the second-most globally important neglected tropical disease ( after malaria ) , affecting almost 240 million people worldwide . Despite their scientific preeminence , however , planarians , polyclads , and neodermatans remain merely the best-known branches of a much larger and deeper phylogenetic diversity of platyhelminths ( Hyman , 1951; Karling , 1974; Rieger et al . , 1991 ) . Indeed , these three lineages are among the only flatworms to exhibit large ( >1–2 mm ) body size; accordingly , the 9–10 other flatworm orders are usually collectively referred to as ‘microturbellarians’ , a practical term acknowledging their shared , albeit plesiomorphic , adaptations to interstitial habitats ( Giere , 2015 ) . No one microturbellarian taxon shows the remarkable regenerative capacity of some triclad species ( Egger et al . , 2007 ) , nor the clear , experimentally accessible spiral cleavage of polyclads ( Martín-Durán and Egger , 2012 ) , nor the profound commitment of neodermatans to parasitic habits ( Jennings , 2013 ) , but several taxa do exhibit lessened or modified versions of some or all of these traits . Understanding the broader evolutionary significance and initial emergence of these emblematic flatworm traits , therefore , requires phylogenetically constrained comparisons between these familiar taxa and their relatively obscure ‘microturbellarian’ relatives . To this end , the internal phylogeny of Platyhelminthes has gained much clarity in recent years through the analysis of rRNA sequence data ( Littlewood et al . , 1999; Lockyer et al . , 2003; Baguñà and Riutort , 2004; Littlewood , 2006; Laumer and Giribet , 2014 ) , for instance via the demonstration of the polyphyly of taxa such as Seriata ( Tricladida , Proseriata , and Bothrioplanida; [Sopott-Ehlers , 1985] ) and Revertospermata ( Fecampiida and Neodermata; [Kornakova and Joffe , 1999] ) , as well as through support for some classically defined scenarios such as the sister-group relationship between Catenulida and Rhabditophora ( Larsson and Jondelius , 2008 ) , and more recently , evidence for Macrostomorpha ( Haplopharyngida+Macrostomida ) and the early-branching position of ‘lecithoepitheliates’ within a clade of ectolecithal flatworms ( Euneoophora; [Laumer and Giribet , 2014] ) . Nonetheless , even with these taxon-rich data sets , several key deep phylogenetic splits remain lacking in resolution . While it is clear that polyclads occupy a relatively basally branching position within Rhabditophora , their relationship to Macrostomorpha and possibly Prorhynchida remains unclear . Similarly , though rRNA support for a clade called Adiaphanida ( comprising Tricladida , Fecampiida , and Prolecithophora; [Norén and Jondelius , 2002] ) is nearly unequivocal , the internal relationships within this clade remain poorly supported . Finally , and most importantly , while rRNA-based phylogenies have proven effective in falsifying previous hypotheses on the origins of Neodermata—perhaps the most intensely researched goal of platyhelminth systematics ( Baguñà and Riutort , 2004; Littlewood , 2006 ) —to date they have not been successful in producing practically useful , well-supported alternative hypotheses . Instead , available phylogenies have largely indicated that Neodermata has no close relationship with any ‘turbellarian’ order: its sister group is currently understood to be a diversified clade consisting of Tricladida , Prolecithophora , Fecampiida , Rhabdocoela , and perhaps Proseriata ( Littlewood et al . , 1999; Lockyer et al . , 2003; Baguñà and Riutort , 2004; Littlewood , 2006; Laumer and Giribet , 2014 ) , thereby complicating comparisons between free-living flatworms and Neodermata , and implying an ancient origin of obligate vertebrate parasitism in the phylum . To test the so far almost entirely rRNA-based molecular phylogeny of Platyhelminthes , we conducted an RNA-seq survey encompassing representatives of all orders of ‘turbellarian’ flatworms and related spiralian outgroups ( Supplementary file 1 ) . Such data represent a potent source of protein-coding sequence data suitable for phylogenetic analysis ( Dunn et al . , 2013; Lemmon and Lemmon , 2013; Yang and Smith , 2014 ) , and given the much smaller target size and the comparative dearth of complex repeat landscapes in transcriptome data , may be assembled de novo with much less required sequencing depth and computational difficulty than genome data ( Hittinger et al . , 2010; Haas et al . , 2013 ) , despite the complicating phenomena of alternative splicing and wide variance in gene expression levels . From such de novo assemblies , we have selected a single representative open reading frame per putative unigene per species , and employed a sensitive , graph-based orthology assignment algorithm ( Roth et al . , 2008 ) to organize these peptides , together with gene models derived from annotated genomes , into putatively orthologous sets . These ortholog groups were then subjected to multiple sequence alignment , with masking of poorly aligned regions , and a subset of 516 such groups were selected for phylogenetic analysis using an automatic matrix reduction procedure ( Misof et al . , 2013 ) . We constructed a supermatrix representing a concatenation of this subset , composed of 132 , 299 amino acids ( with 48 . 58% matrix completeness ) which was analyzed under site-heterogeneous models ( Lartillot and Philippe , 2004; Le et al . , 2012 ) using both maximum likelihood ( ML ) ( Stamatakis and Aberer , 2013 ) and Bayesian inference ( BI ) approaches ( Lartillot et al . , 2013 ) ( Figure 1 ) . For these concatenated analyses , we also employed several approaches to control for systematic errors , for example , by trimming sites that fail tests of compositional heterogeneity ( Foster , 2004; Criscuolo and Gribaldo , 2010 ) or by leveraging models built to control the effects of heterotachous substitution ( Philippe et al . , 2005; Pagel and Meade , 2008 ) . We also considered phylogenetic signal from a gene-tree centric perspective , inferring individual ML trees for each gene , and summarizing the predominant ( and sometimes , conflicting; [Fernández et al . , 2014] ) splits in this set of unrooted , incomplete gene trees using both quartet supernetworks ( Grunewald et al . , 2013 ) ( Figure 2 ) and an efficient species-tree algorithm ( Mirarab et al . , 2014 ) ( Figure 3 ) . Such approaches may mitigate the inter-gene heterogeneity in branch length and amino acid frequency introduced by concatenation ( Liu et al . , 2015 ) , albeit at the cost of introducing a greater sampling error into gene-tree estimation ( a cause of apparent gene-tree incongruence perhaps more prevalent at this scale of divergence than the genuine incongruence modeled by most species-tree approaches , namely incomplete lineage sorting ) . We also performed taxon deletion experiments to test for the effects of long-branch attraction in influencing the placement of the fast-evolving Neodermata within the phylogeny ( Figures 4 , 5 ) . Considered together , our analyses provide a consistent signal of deep platyhelminth interrelationships , demonstrating a combination of groupings familiar from the eras of classical morphological systematics and rRNA phylogenetics , as well as several novel but nonetheless well-supported clades , whose provenance and broader evolutionary significance we now consider ( Figure 6 ) . 10 . 7554/eLife . 05503 . 003Figure 1 . Phylogenetic relationships of Platyhelminthes , encompassing 25 ‘turbellarian’ species , 8 representatives of Neodermata , and 7 spiralian outgroups . Phylogram represents results from a maximum likelihood ( ML ) analysis of 516 predicted orthogroups ( 120 , 527 aligned amino acid sites trimmed of nonstationary and poorly aligned residues ) , analysed in ExaML v1 . 0 . 0 under LG4M+F ( from which the phylogram shown was taken ) , and also in phyML ( v20130927 ) under LG+FR4+F+IL , with support values to the right of each node representing clade frequency in 100 bootstraps or aBayes supports , respectively ( complete support in both measures indicated by a white dot ) . Values below internodal edges represent the number of genes available to test each clade ( decisive genes ) ; values above such edges represent the percentage of these genes whose ML trees are consistent with this clade ( congruence frequency ) . Colored circles at tips of tree are given with diameter in proportion to the number of orthogroups represented for that taxon in the supermatrix , and colored as per the legend in the lower left . DOI: http://dx . doi . org/10 . 7554/eLife . 05503 . 00310 . 7554/eLife . 05503 . 004Figure 1—figure supplement 1 . Maximum likelihood phylogram resulting from analysis of untrimmed 516-gene matrix . 132 , 299 aligned amino acid sites , analyzed in ExaML under LG4M+F and also in phyML ( v20130927 ) under LG+FR4+F+IL , with support values to the right of each node representing clade frequency in 100 bootstraps or aBayes supports , respectively ( complete support in both measures indicated by a white dot ) . The clade indicated with an asterisk was not recovered in the phyML analysis; rather , this analysis recovered Austrognathia sp . as sister to Rotifera+Gastrotricha , with an aBayes support value of 0 . 8117 . DOI: http://dx . doi . org/10 . 7554/eLife . 05503 . 00410 . 7554/eLife . 05503 . 005Figure 1—figure supplement 2 . Majority rule consensus ( MRC ) phylogram of a Bayesian Markov Chain Monte Carlo sampling with the CAT+GTR+Γ4 model in PhyloBayes-MPI v1 . 4e . Phylogram shown represents a summary of 517 and 219 trees sampled from two chains which had run for 4236 and 3638 generations , respectively; the first 3200 trees sampled were discarded as ‘burn-in’ , and every other remaining tree was included in the posterior summary . All clades showed maximum posterior probability ( denoted with a white circle ) . Convergence diagnostics assessed via bpcomp showed a maximum and mean bipartition discrepancy of 0 between these chains . An alternative pair of chains run for 3931 and 3554 showed identical evidence for convergence , but a separate topology . However , the MRC phylogram from this pair of chains differed from that of the pair shown only in the positions of Proseriata and Rhabdocoela , which were inverted relative to the topology sampled in the chain shown; all clades were maximally supported in this second pair of chains as well . The negative log likelihood from the first pair of chains ( which generated the MRC phylogram shown ) has a smaller average than that sampled by the second pair ( −1 . 878634E + 06 vs −1 . 880144E + 06 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05503 . 00510 . 7554/eLife . 05503 . 006Figure 2 . Quartet supernetworks built from 516 individual ML gene trees , showing predominant inter-gene conflicts in genes selected for concatenation . A qualitatively nearly identical topology ( not shown ) was recovered using bootstrap majority rule consensus trees as input . Edge weights were calculated in SuperQ v1 . 1 , with the ‘balanced’ linear objective function , and no filter applied . Supernetwork was visualized in SplitsTree v4 using default settings . DOI: http://dx . doi . org/10 . 7554/eLife . 05503 . 00610 . 7554/eLife . 05503 . 007Figure 3 . ASTRAL species tree . Constructed under default settings from 516 input unrooted partial gene trees inferred in RAxML v8 . 0 . 20 . Nodal support values reflect the frequency of splits in trees constructed by ASTRAL from 100 bootstrap replicate gene trees using the -b flag; gene- and site-level bootstrapping ( -g ) was not performed . DOI: http://dx . doi . org/10 . 7554/eLife . 05503 . 00710 . 7554/eLife . 05503 . 008Figure 4 . ML phylogram inferred from a version of the BMGE-trimmed matrix in which all taxa of Neodermata have been deleted . Tree inferred in ExaML v1 . 0 . 0 under the LG4M+F model; nodal support values represent the frequency of splits in 100 bootstrap replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 05503 . 00810 . 7554/eLife . 05503 . 009Figure 5 . ML phylogram inferred from a version of the BMGE-trimmed matrix from which Bothrioplana semperi has been deleted . Tree inferred in ExaML v1 . 0 . 0 under the LG4M+F model; nodal support values represent the frequency of splits in 100 bootstrap replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 05503 . 00910 . 7554/eLife . 05503 . 010Figure 6 . Summary depiction of the phylogenetic results presented in the present study . All terminal taxa ( turbellarian orders , neodermatan classes , and outgroup clades ) are shown with equilateral triangles log-scaled in proportion to described species number ( largely following [Martín-Durán and Egger , 2012] ) . Clades deemed to have equivocal or only preliminary support in these analyses are shown with a dashed subtending branch . Named clades discussed in the text are labeled . Clades with known homoplasy-free synapomorphies are labeled with an open square , whereas those with proposed synapomorphies showing inferred losses are labeled with a single diagonal line , and clades without any known synapomorphies are shown with a cross . Traits commonly referred to in discussions of platyhelminth phylogenetics are given to the right of the phylogeny; terminal taxa are shown filled in if any representatives have been reported in published studies to manifest the trait in question ( although numerous reversals apparently occurring subsequent to the origin of a clade are not depicted ) , and clades for which no published observations are available are labeled with a question mark . The orders Prorhynchida and Gnosonesimida are given with a ‘*’ in the ‘Bulbous pharynx’ column to indicate that although the pharyngeal types manifested in these orders have been described under separate names ( variable and coniform pharynx , respectively ) , both represent , by definition , a bulbous pharynx . Line drawings in Figure 6 modified with permission from the following sources:Gastrotricha ( unspecified chatonotoid ) – BIODIDAC database ( http://biodidac . bio . uottawa . ca/ ) Gnathostomulida ( Gnathostomula peregrina ) – Kirsteuer , 1969Annelida ( Hirudinea sp . ) – BIODIDAC database ( http://biodidac . bio . uottawa . ca/ ) Prorhynchida ( after Geocentrophora marcusi ) – originalAll others – Images modified from original illustrations provided courtesy of Janine Caira ( Caira and Littlewood , 2013 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05503 . 010 Platyhelminthes , in its modern conception , is comprised of two major clades , Catenulida and Rhabditophora , each themselves morphologically well-defined , which however do not share any known morphological apomorphies ( Ehlers , 1985; Smith et al . , 1986 ) . Nonetheless , in rRNA phylogenies to date ( Larsson and Jondelius , 2008 ) , as well as in the present analyses ( Figures 1 ) , the monophyly of Platyhelminthes finds nearly unequivocal support . The precise position of the phylum within Spiralia remains controversial , though recent studies have argued for a sister-group relationship with Gastrotricha within a paraphyletic ‘Platyzoa’ ( Struck et al . , 2014; Laumer et al . , 2015 ) . As we intended only to resolve relationships within Platyhelminthes , our outgroup sampling is insufficient to test the status of Platyzoa , as we lack more distant outgroups to Spiralia ( members of Ecdysozoa ) . Nonetheless , in all our analyses , our sampled platyzoan taxa fall between Platyhelminthes and our representatives of Trochozoa ( Annelida and Mollusca ) , indicating either mono- or paraphyly of this taxon ( Struck et al . , 2014; Laumer et al . , 2015 ) . It is , however , interesting to note the comparatively long branch distance separating Catenulida and Rhabditophora , which may imply that future efforts to test the placement of Platyhelminthes within Spiralia would do well to sample Catenulida , if long-branch attraction artifacts are to be avoided . Pre-cladistic classifications emphasized the separation of the parasitic flatworms from their free-living ancestors ( the paraphyletic [Ehlers , 1985] ‘Class Turbellaria’ ) , in recognition of the vast phenetic differences between these lineages . Our identification of B . semperi as the closest free-living relative of Neodermata , and the nuclear genomic evidence we present for Cercomeromorpha , will help to narrow this artificial gap , by clarifying the relevant comparisons that should be made , and by setting taxonomic priorities for future research . Bothrioplanida and Neodermata may , for instance , bear evidence of common ancestry in aspects of their morphology: at the ultrastructural level , B . semperi resembles Neodermata both in the structure of its excretory system ( namely , its protonephridial flame bulbs , which are composed of two cells with extensions that interdigitate to form an ultrafiltration weir in much the same way in both taxa [Kornakova , 2010] ) , as well as in the structure of its monociliary epidermal sensory receptors ( which bear an electron-dense collar in both taxa [Kornakova and Joffe , 1996] ) . Further morphological investigation of the relatively obscure B . semperi may reveal other shared derived characters of these taxa , although certain character systems may prove elusive ( e . g . , spermatogenesis in a purportedly parthenogenetic species ) . Fortunately , knowledge of Bothrioplanida need not necessarily be restricted to B . semperi , given the evidence for at least one undescribed putative Bothrioplana species ( Kawakatsu and Mack-Firă , 1975 ) ; further representatives may also be recovered by studies of cryptic molecular diversity and continued exploratory taxonomic surveys of freshwater microturbellarians , which remain poorly known outside of the Palearctic ( Artois et al . , 2011 ) . Indeed , a new species of Bothrioplanida apparently capable of normal spermatogenesis has been recently reported from mainland China ( Ning et al . , In press ) . Comparison between Bothrioplanida and the extant Neodermata can also extend beyond the search for synapomorphies: they may inform hypotheses on the route by which the earliest vertebrate-parasitic associations of Platyhelminthes arose . As a cosmopolitan species able to colonize temporary , chemically diverse , and spatially isolated freshwaters , B . semperi appears to be remarkably well adapted to frequent long-distance passive dispersal ( perhaps via vertebrate , especially waterfowl , vectors [Sluys and Ball , 1985; Artois et al . , 2011] ) , an ecological challenge at least analogous to the sweepstakes game that each succeeding generation of parasite plays during the colonization of a new host . It is therefore tempting to speculate that at least some adaptations to these similar ecological challenges could have been present in the most recent common ancestor of Bothrioplanida and Neodermata , and may have ‘pre-adapted’ early neodermatans to a parasitic lifestyle . For example , if stem Neodermata possessed a resistant , presumably quinone-tanned egg capsule similar to that used by B . semperi in passive dispersal , this could have facilitated enteric infection early in the history of this lineage ( Llewellyn , 1965 ) ; indeed , extant freshwater microturbellarians produce resting eggs that are known to retain high viability subsequent to passage through vertebrate digestive tracts ( Artois et al . , 2011 ) . Llewelyn's original formulation of this hypothesis posited that precocious emergence of the larvae of such swallowed eggs would provide a simple route to endoparasitism . Indeed , compared to their marine relatives , freshwater flatworms are regularly exposed to a much wider range of variation in temperature , salinity , pH , and dissolved gas content ( Hutchinson , 1957 ) , facts which would seem to facilitate the colonization of an internal environment . This speculation hence implies that endoparasitism would be plesiomorphic to Neodermata , a possibility at least consistent with the evidence presented here for a relatively more derived position of Monogenea within Neodermata ( Cercomeromorpha ) . One important counter-indication to this scenario , however , lies in the fact that most neodermatans , prior to the development of the neodermis , hatch as fully or partially ciliated larvae ( miracidia or oncomiracida ) which spend their earliest minutes-to-hours searching an aquatic medium for a new primary host in a functionally free-living dispersive mode . Definitive conclusions on the precise mode of parasitism employed by the common ancestor of extant Neodermata remain , in any case , premature , pending resolution of the mono- or non-monophyly of Monogenea . Whatever the nature of the most recent common ancestor of Neodermata , it must be emphasized that the symbiosis presented by the neodermatan crown-group may be only dimly reflective of the form of symbiosis employed in its stem lineage . Traces of this earliest transition , may , moreover , be sparse , given the timescale of the divergence . While Bothrioplanida entirely lacks a fossil record , there are at least a few indications of the geological antiquity of crown Neodermata . The earliest direct fossil evidence of the clade is an assemblage of sclerotic hooks resembling those of Cercomeromorpha , recovered from lower Frasnian ( ∼380 Mya ) freshwater acanthodians and placoderms ( Upeniece , 2001 ) . However , if the suggestion of codivergence ( notwithstanding subsequent host-switching events ) with their gnathostome hosts in the deepest splits of several neodermatan clades ( Cestoda [Hoberg , 1999] and both groups of Monogenea [Jovelin and Justine , 2001; Bentz et al . , 2003] ) were correct , then the diversification of Neodermata must precede that of crown-group gnathostomes in the midst of the Middle Cambrian ( ∼525 Mya [Blair and Hedges , 2005] ) , implying that its common ancestor with Bothrioplanida has a still earlier origin . Thus , it must be noted that , despite the low amino acid substitution rate of Bothrioplanida , evolution and extinction have had considerable opportunity to erase the primitive organismic characteristics of this common ancestor in both descendant lineages , inherently constraining efforts to reconstruct the nature of this ancestor . Nonetheless , it may be significant that the sister-taxon of Neodermata is a freshwater species . Although many aspects of the ecology of B . semperi have not been thoroughly studied , the species is not known to engage in any symbioses; its mode of dispersal is thought to be essentially passive . However , dispersal among disconnected habitats remains a fundamental challenge of all freshwater invertebrates , and one specific form of association , phoresis , has become a common adaptation to the necessity of dispersal in diverse groups ( Bilton et al . , 2001 ) . Phoretic associations in freshwater invertebrates range from purely commensal to explicitly parasitic , with the life cycles of several higher taxa ( Unionida , Hydrachnidia , Nematomorpha ) including both a free-living adult phase and an ecto- or endoparasitic larval phase . We propose that a similar ecological mode may also have characterized stem Neodermata prior to their transition to dedicated parasitism . This hypothesis presumes that Neodermata originally colonized freshwater or diadromous hosts . Given a well-sampled and well-resolved internal phylogeny of all Neodermata , and an explicit attempt at ancestral state reconstruction in host habitat , this suggestion could be straightforwardly tested . In this light , it is remarkable that many of the early-branching taxa within each major clade of Neodermata ( e . g . , Iagotrematidae , Sundanonchidae , and Pseudomurraytrematidae in Monopisthocotylea [Olson and Littlewood , 2002; Bentz et al . , 2003] , Polystomatidae in Polyopisthocotylea [Jovelin and Justine , 2001] , Amphilinidea , Caryophyllidea , and Diphyllobothriidea+Haplobothriidea in Cestoda [Waeschenbach et al . , 2012] , Aspidogastridae in Aspidogastrea [Littlewood , 2006] , several higher taxa within the digenean clade Diplostomida [Olson et al . , 2003] ) are primarily or exclusively found in freshwater hosts ( principally teleosts and amphibians ) . To date , discussions of the emergence of platyhelminth parasitism have focused on organismic and morphological traits—in other words , those character systems for which data have been historically available . However , principally because of their importance as human pathogens , genomic data are now available from all major lineages of Neodermata , including well-curated assemblies , annotation efforts , and experimental protocols for species such as Echinococcus multilocularis ( Brehm , 2010; Olson et al . , 2012; Tsai et al . , 2013 ) and Schistosoma mansoni ( Collins et al . , 2013; Wang et al . , 2013 ) . With such data available , there has been much discussion of the genome-level adaptations to parasitism , with suggestions of many apparent losses , including several homeobox genes , vasa , tudor , and piwi orthologs , fatty and amino acid biosynthesis pathways , and peroxisome components; proposed gains include the evolution of a neodermatan-specific Argonaute subfamily and micro-exon gene organization ( Tsai et al . , 2013; Hahn et al . , 2014 ) . It is , however , essential to recognize that , in the absence of a well-founded platyhelminth phylogeny , associating any of these common genomic features to parasitism per se is not possible , as any of them may have a deeper , ‘turbellarian’ history . Discerning the molecular-level changes specifically associated with the origin of parasitism , therefore , requires comparison of neodermatan genome biology—initially , from the perspective of simple gene presence/absence , but eventually incorporating information on gene expression , function , regulation , and selection history—with the genome biology of their nearest free-living relative , as well as with more distant free-living flatworm orders . Fortunately , under the topology we have recovered , the distribution of model systems is nearly ideally suited to disentangle the molecular foundations of platyhelminth parasitism: only one taxon ( indeed , a monospecific taxon ) stands between Neodermata and Adiaphanida , within which triclads such as the powerful experimental system S . mediterranea ( Collins and Newmark , 2013 ) are recovered as the earliest-branching lineage . Thus , if further evidence bears out the topology we have recovered here , we suggest that establishment of genomic resources and experimental protocols for B . semperi would thus provide the best available point of comparison to clearly understand the ecological , physiological , morphological , developmental , and genomic changes that took place during the single evolutionary transition that led to the most spectacular form of vertebrate parasitism in the modern world . Using a class of molecular sequence data fundamentally different in quality and much larger in quantity than the rRNA data sets that have been heretofore available , we have provided a modern hypothesis of ‘deep’ platyhelminth phylogeny ( Figure 6 ) , one which agrees in many respects with concepts from the eras of classical morphological and rRNA-based phylogenetics , but which also presents a number of unexpected relationships , not least of which is a novel scenario for origin of the vertebrate-parasitic Platyhelminthes . However , while several clades remain deserving of further investigation—in particular , ( a ) the interrelationships between Proseriata , Rhabdocoela , and the remaining Euneoophora , given the analytical instability surrounding these splits , and ( b ) the interrelationships within Adiaphanida , considering our sparse sampling of Fecampiida and Prolecithophora—most interrelationships in our phylogeny demonstrate remarkably high support and robustness across analytical modes . Thus , we would argue that , though this phylogeny certainly bears further testing using expanded taxon sampling , the most significant advancements in informing platyhelminth evolution henceforth will come not only from systematics , but particularly from those disciplines which take a known species tree as an input—for example , morphology , evolutionary developmental biology , and evolutionary genomics . It is our hope that such research will give due attention to the many poorly known lineages of ‘microturbellaria’ with which the most prominent members of this diverse phylum share their heritage . RNA was extracted from live single or pooled specimens , or specimens flash-frozen in TRIzol reagent or RNAlater ( Ambion , Inc . , Carlsbad , CA ) . Polyadenylated mRNAs were either directly extracted using the Dynabeads DIRECT kit ( using ‘standard’ or ‘mini’ scaling as recommended by the manufacturer , Life Technologies , Inc . , Carlsbad , CA ) or were purified from total RNA generated using a standard TRIzol extraction using the Dynabeads mRNA purification kit ( Life Technologies , Inc ) . mRNA concentration was quantified and quality was assessed on an Agilent Bioanalyzer 2100 mRNA Pico kit; however , some successfully sequenced libraries prepared using the IntegenX mRNA kit were derived from mRNAs that were undetectably dilute in this protocol . cDNAs from the taxa sequenced on the MiSeq platform ( see below ) were produced by the phi29-mRNA amplification ( PMA ) method , amplifying autoligated cDNAs generated from poly-A selected mRNA as described by ( Pan et al . , 2013 ) ; Qiagen's Repli-G single cell kit was used to amplify these cDNAs . Nonstranded libraries were generated using the TruSeq RNA Sample Prep v2 kit ( Illumina , Inc . , San Diego , CA ) , using 5 µl mRNA per sample , and following the manufacturer's guidelines to fragment full-length cDNAs to a mean insert size of 500 bp in a Covaris S220 . PMA-amplified cDNAs were fragmented to a mean insert size of 800 bp and libraries were prepared following a modification of Neiman et al . ( 2012 ) in which dual indexing adapters ( derived from the Illumina TruSeq DNA HT adapter sequences ) were used in place of the original single-indexing adapters . Stranded libraries were generated using the IntegenX directional mRNA kit on the Apollo324 instrument , using up to 250 ng of mRNA per library ( or the maximum value of 18 µl per library when mRNA quantity was undetectable ) . Details on specimen source and the library construction method used for each species are available in MCZBase ( mczbase . mcz . harvard . edu ) using the accession numbers provided in Supplementary file 1 . Most cDNA libraries were completed with a single multiplexing index and sequenced with up to six libraries per lane; these libraries were read as normal 101 bp paired end runs on the Illumina HiSeq 2000 or as rapid runs on an Illumina HiSeq 2500 in the Harvard FAS Center for Systems Biology . cDNA libraries amplified using the PMA method ( Supplementary file 1 ) were sequenced on an Illumina MiSeq as paired end 250 bp reads ( v2 chemistry ) ; read quality in these MiSeq runs was poor ( only 48% reads with Phred ≥ 30 ) , perhaps reflecting a combination of unusually large insert sizes and expired reagents . Raw reads were stringently demultiplexed by staff at the FAS Center for Systems Biology . Quality control was performed using the NGSQC Toolkit v2 . 3 ( Waeschenbach et al . , 2012 ) to maintain parity between reads during QC , retaining no ambiguous bases ( AmbiguityFilter . pl -c set to 0 ) , and trimming reads with a PHRED quality score threshold of 20 ( TrimmingReads . pl -q set to 20 ) , and using the IlluQC_PLL . pl script to remove primer/adapter sequences from a custom file constructed of all adapters/primers used in library construction . Libraries sequenced on the MiSeq platform were quality controlled using Trimmomatic v0 . 32 called with the following run parameters: ILLUMINACLIP:TruSeq3-PE . fa:2:30:10 LEADING:3 TRAILING:3 SLIDINGWINDOW:4:20 MINLEN:36 . Following cleanup , paired reads were assembled in Trinity ( Grabherr et al . , 2011; Haas et al . , 2013 ) ( released 05 October 2012 or in the case of the PMA libraries , 14 April 2014 ) , using the –RF flag when appropriate for stranded libraries ( example command: Trinity . pl --seqType fq --left Mfus_R1_QC . fastq --right Mfus_R2_QC . fastq --SS_lib_type RF --CPU 6 --JM 150 G --bflyCPU 6 --output /n/Giribet_Lab/claumer/Mfus_trinity ) . The finished de novo assemblies were subjected to an initial round of redundancy reduction using CD-HIT-EST v3 . 1 , clustering with a 5% global sequence similarity threshold , as a heuristic means of clustering transcripts expressed from the same putative locus but not clustered into the same subcomponent by Trinity . From this redundancy-reduced assembly , we extracted putative open reading frames ( ORFs ) using the transcripts_to_best_scoring_ORFs . pl script provided in the TransDecoder plugin to trinity . From the ‘best_candidates . eclipsed_orfs_removed . pep’ file , we used a custom Python script ( choose_longest . py ) to parse the fasta headers , and remove all putative ORFs but the longest per subcomponent . Statistics displayed in Supplementary file 1 were computed using the TrinityStats . pl script distributed with Trinity v2014-04-13 , and with the fastq-stats program from the ea-utils package v2014-04-22 . For the species sampled using public Sanger EST data , we downloaded all available reads from the NCBI dbEST ( 52 , 772 for Brachionus plicatilis , 6729 for Neobenedenia melleni ) . Reads for each species were first sanitized using the SeqClean pipeline , removing all sequences with a hit against the NCBI UniVec database . These were then assembled using the TGI Clustering Tools ( TGICL ) package , with redundancy reduction and mitochondrial removal performed as described above . ORFs were predicted using the TransDecoder pipeline separately on ‘contigs’ and ‘singlets’ from the TGICL assemblies , and on the contigs , the longest ORF per cluster was retained using a modification of the aforementioned Python script . These ORFs from filtered contigs were then concatenated with singlet ORFs to generate a final peptide fasta file for use in OMA standalone . Predicted proteins from the S . mansoni , E . multilocularis , Taenia solium , and Hymenolepis microstoma genomes ( Berriman et al . , 2009; Tsai et al . , 2013 ) were downloaded from GeneDB ( Logan-Klumpler et al . , 2011 ) . Predicted peptides from the draft G . salaris ( v1 . 0 ) genome were downloaded from http://invitro . titan . uio . no/gyrodactylus/downloads . html . Protein predictions from the draft Clonorchis sinensis genome ( Wang et al . , 2011 ) ( BioProject PRJDA72781 ) , as well as from the genomes of Lottia gigantea ( PRJNA175706 ) , Capitella teleta ( PRJNA175705 ) , and Helobdella robusta ( Simakov et al . , 2012 ) ( PRJNA175704 ) were downloaded from NCBI . RNA-Seq reads from Fascioloides magna ( Cantacessi et al . , 2012 ) were downloaded from SRA ( accession SRX147910 ) and were subjected to the same pipeline as our newly generated RNA-Seq data sets . 454 cDNA reads from Stylochoplana maculata ( Struck et al . , 2014 ) were quality-trimmed using the fastx-toolkit ( v0 . 0 . 13 . 2 ) , trimming reads to a Phred score of 30 and discarding reads trimmed to shorter than 30 bp; these were assembled using Trinity ( release 14 April 2014; see Ren et al . , 2012 ) . Prior to orthology prediction , all peptides containing nonsense characters ( e . g . , # ) and intercalary stop codons were removed , and closely related peptides ( e . g . , isoforms ) whose headers grouped them into a cluster were filtered using the retain-one-per-cluster Python script described above . Predicted peptides derived from the workflows described above were assigned into orthologous groups using the graph-based OMA algorithm ( Roth et al . , 2008 ) , implemented in OMA standalone v0 . 99x ( http://omabrowser . org/standalone/ ) , with the LengthTol parameter set to 0 . 61 and the MinSeqLen parameter set to 50 . This yielded 86 , 808 OMA groups ( putative orthologous groups by definition containing one sequence per species ) , of which 11 , 960 contained more than four species each ( our criterion for potential inclusion in phylogenetic analysis ) . These were aligned using the L-INS-i algorithm implemented in MAFFT ( Katoh and Toh , 2008 ) v6 . 853 . Alignment certainty was quantified using the pair Hidden Markov Model framework implemented in ZORRO ( Wu et al . , 2012 ) , using FastTree to derive guide trees for each alignment . A Python script was used to trim residues with an alignment uncertainty score below 0 . 5 , and all alignments with a sequence-masked length greater than 50 AA ( 11 , 913 ) were considered for further analysis . We concatenated this set of genes using Phyutility v2 . 2 . 6 ( Smith and Dunn , 2008 ) and utilized MARE v0 . 1 . 2-a ( Misof et al . , 2013 ) to select a subset of ‘tree-like’ genes for concatenation , using default settings , except for a taxon weighting ( ‘-t’ ) of 100 , which we selected to ensure even representation among taxa . This procedure selected a subset of 516 orthogroups , which when concatenated yielded a supermatrix of 132 , 299 amino acid residues , of which 48 . 58% were occupied . We also constructed a trimmed version of this matrix using Block Mapping and Gathering with Entropy ( BMGE ) ( Criscuolo and Gribaldo 2010 ) v1 . 1 . 1 , as a means of removing aligned sites showing evidence of substitutional saturation and compositional heterogeneity . BMGE was called with the ‘fast’ test of compositional heterogeneity ( -s FAST ) , calculating entropy scores against the BLOSUM30 matrix , and retaining all gaps ( -g 1 ) . This trimming yielded a matrix with 120 , 527 residues , of which 49 . 24% were occupied . Maximum likelihood inference for both the trimmed and untrimmed 516-ortholog supermatrix was conducted as a unpartitioned analysis under the best-fitting LG4M+F substitution model implemented in ExaML v1 . 0 . 0 , with 100 bootstrap replicates , with the tree search set to begin from an initial tree ( ‘-t’ ) constructed using FastTree v2 . 1 . 7 ( with the ‘--wag’ and ‘--gamma’ options selected ) . Bootstrap support for each clade in the ML tree was summarized using the sumtrees . py script in Dendropy v3 . 12 ( Sukumaran and Holder , 2010 ) . To account for the effects of heterotachous substitution on phylogenetic inference , we also undertook maximum likelihood analyses using the ‘integrated length’ ( --il ) option implemented in phyML v20130927 to estimate branch lengths by integrating over geometric Brownian trajectories as described by Guindon ( 2013 ) . We performed this inference on both untrimmed and BMGE-trimmed versions of the concatenated matrix , using a combination of NNI and SPR heuristic searches , and inferring under the LG+F model with four ‘free’ ( --free-rates ) categories of rate variation . Branch support values were calculated as aBayes ( Bayesian-like transformation of aLRT ) . Bayesian phylogenetic analysis was carried out on the untrimmed 516-gene matrix in PhyloBayes-MPI v1 . 4e ( Lartillot et al . , 2013 ) . We ran four independent chains under the CAT+GTR+Γ4 model , removing constant sites using the -dc option , as a means of improving convergence as recommended in the PhyloBayes-MPI manual . Chains were run for 3554 to 4236 generations , until convergence on the posterior distribution between at least two chains appeared to have been achieved . We defined convergence as the point at which the maxdiff statistic from the bpcomp program ( with a burnin of at least 3000 trees ) fell below 0 . 1 for two or more chains . By this criterion , two pairs of the four chains appeared to mutually converge ( with bpcomp showing a maxdiff of 0 with a burn-in of greater than 3300 ) , but both chains in each pair remained divergent ( maxdiff = 1 ) with the other two chains in the remaining pair . The negative log likelihood from the first pair of chains , however , showed a smaller average than that sampled by the second pair ( 1 . 878634E + 06 vs 1 . 880144E + 06 ) . Individual gene trees were inferred in RAxML 8 . 0 . 2 , using the AUTO function to select the optimal substitution matrix , with empirical amino acid frequencies ( ‘-m PROTGAMMAAUTOF’ ) . For each of the 516 genes selected by MARE for supermatrix construction , trees were created for 100 rapid bootstrap replicates , with branch length optimization specified for these bootstraps ( ‘-k’ ) , and the final ML tree search was conducted using these bootstrap trees as starting trees ( ‘-f a’ ) . We constructed quartet supernetworks to summarize phylogenetic conflicts within our set of 516 best-found ML trees visually using SuperQ ( Grunewald et al . , 2013 ) v1 . 1 , which permits partial gene trees as input . This is noteworthy as many other approaches of visualizing or quantifying among gene tree conflict ( including the available RAxML implementation of internode certainty [Salichos et al . , 2014] ) require input gene trees to be complete in all taxa . We constructed supernetworks using the balanced linear secondary objective , imposing no filter . We also constructed supernetworks using identical settings , taking as input bootstrap majority rule consensus trees constructed using the sumtrees . py program of DendroPy python library ( Sukumaran and Holder , 2010 ) , v3 . 12 . This supernetwork appeared qualitatively very similar to the ML-tree quartet supernetwork and is hence not displayed . Supernetworks were visualized in SplitsTree v4 using default settings . We used the ASTRAL species tree algorithm ( Mirarab et al . , 2014 ) , which , like SuperQ , decomposes input gene trees into quartets , but which finds for these input quartets an optimal , fully bifurcating species tree . For this reason ASTRAL also operates efficiently on large sets of incomplete , unrooted gene trees , which remains a limitation of other currently available species tree methods . ASTRAL was run using default settings , given as input all 516 best-found ML trees; we also calculated bootstrap species trees using the bootstrapped gene trees , bootstrapping only at the site level ( i . e . , not at the site-and-gene level , [Seo , 2008] ) . To quantify the number of genes potentially decisive for a given split , and also the number within this set that actually support the split in question ( Figure 1 ) , we employed a custom Python script ( parse_gene_trees . py ) , built using the ETE2 library ( Huerta-Cepas et al . , 2010 ) , to parse the set of individual ML gene trees ( see also Fernández et al . , 2014 ) . If a tree contained at least one species in both descendant clusters for a given clade , plus at least two distinct species basal to the node in question , it was considered potentially decisive ( forming minimally a quartet; [Dell’Ampio et al . , 2013] ) ; if the descendants were monophyletic with respect to their relative outgroups , that gene tree was considered congruent with the node in question ( although this count is agnostic to the topology within the node in question ) . We displayed these counts for all nodes reflecting interordinal relationships in the 516-gene , BMGE-trimmed supermatrix ML topology ( Figure 1 ) . Original sequence reads and have been deposited to SRA accessible at NCBI ( http://www . ncbi . nlm . nih . gov/bioproject/ ) via the BioProject IDs provided in Supplementary file 1 . Scripts used in this study are available at https://github . com/claumer . Concatenated and individual orthogroup amino acid alignments , all inferred trees , and original Trinity transcriptome assemblies used in this study are available at Data Dryad ( doi:10 . 5061/dryad . 622q4; Laumer et al . , 2015b ) .

Flatworms are relatively simple invertebrates with soft bodies . They can be found living in nearly every aquatic environment on the planet , are well-known for their ability to regenerate , and some species live as parasites in humans and other animals . Studies of the physical characteristics of flatworms have provided us with clues about how some groups , for example , the parasitic flatworms , have evolved , but the evolutionary origins of other groups of flatworms are less clear . The genetic studies of flatworm evolution have focused on a single gene that makes a molecule called ribosomal ribonucleic acid , which is required to make all the proteins in flatworms and other animals . By comparing the sequences of this gene in different species of flatworm , it is possible to infer how they are related in evolutionary terms—that is , species with shared gene sequence features are likely to be more closely related than two species with less similar gene sequences . Although this approach has proved to be useful , it has also produced some results that conflict with the conclusions of previous studies . Here , Laumer et al . studied the evolution of flatworms using an approach called RNA sequencing . This approach made it possible to sequence many hundreds of genes in all major groups of flatworms , and compare these genes in different species . Laumer et al . used the data to build a ‘phylogenetic tree’ that infers the evolutionary relationships between the different groups of flatworms . This tree provides evidence that supports some of the ideas about flatworm evolution produced by the previous studies based on both physical features and ribosomal ribonucleic acid . It also presents several unexpected evolutionary relationships; for example , it suggests that the parasitic flatworms are most closely related to a group of small flatworms called Bothrioplanida , which are predators of other invertebrates . Bothrioplanida can live in many freshwater environments , and the physical characteristics that allow them to survive might resemble those found in the earliest parasitic flatworms . The phylogenetic tree made by Laumer et al . represents a guide for researchers seeking clues to the origins of the genetic and developmental innovations that underlie the various physical features found in different flatworms .

2015

Nuclear genomic signals of the ‘microturbellarian’ roots of platyhelminth evolutionary innovation

Lysergic acid diethylamide ( LSD ) has agonist activity at various serotonin ( 5-HT ) and dopamine receptors . Despite the therapeutic and scientific interest in LSD , specific receptor contributions to its neurobiological effects remain unknown . We therefore conducted a double-blind , randomized , counterbalanced , cross-over studyduring which 24 healthy human participants received either ( i ) placebo+placebo , ( ii ) placebo+LSD ( 100 µg po ) , or ( iii ) Ketanserin , a selective 5-HT2A receptor antagonist , +LSD . We quantified resting-state functional connectivity via a data-driven global brain connectivity method and compared it to cortical gene expression maps . LSD reduced associative , but concurrently increased sensory-somatomotor brain-wide and thalamic connectivity . Ketanserin fully blocked the subjective and neural LSD effects . Whole-brain spatial patterns of LSD effects matched 5-HT2A receptor cortical gene expression in humans . Together , these results strongly implicate the 5-HT2A receptor in LSD’s neuropharmacology . This study therefore pinpoints the critical role of 5-HT2A in LSD’s mechanism , which informs its neurobiology and guides rational development of psychedelic-based therapeutics . Funded by the Swiss National Science Foundation , the Swiss Neuromatrix Foundation , the Usona Institute , the NIH , the NIAA , the NARSAD Independent Investigator Grant , the Yale CTSA grant , and the Slovenian Research Agency . NCT02451072 . Disorders of perception and the form and content of thought are important contributors to the global burden of disease ( Murray et al . , 2012 ) . Mechanistic studies of consciousness may be undertaken using psychedelic drugs as pharmacologic probes of molecular signaling within cortical networks underlying perception and thought . In particular , lysergic acid diethylamide ( LSD ) is a psychedelic drug with predominantly agonist activity at serotonin ( 5-HT ) 2A/C , −1A/B , −6 , and −7 and dopamine D2 and D1 receptors ( R ) . Its administration produces characteristic alterations in perception , mood , thought , and the sense of self ( Marona-Lewicka et al . , 2002; Nichols , 2004 ) . Despite its powerful effects on consciousness , human research on LSD neurobiology stalled in the late 1960s because of a narrow focus on the experiential effects of hallucinogenic drugs , combined with a lack of understanding of its effects on molecular signaling mechanisms in the brain . However , renewed interest in the potentially beneficial clinical effects of psychedelics ( Carhart-Harris et al . , 2016a; Gasser et al . , 2014; Griffiths et al . , 2016 ) warrants a better understanding of their underlying neuropharmacology . Nevertheless , major knowledge gaps remain regarding LSD’s neurobiology in humans as well as its time-dependent receptor neuropharmacology . To address this critical gap , the current study aims to comprehensively map time-dependent pharmacological effects of LSD on neural functional connectivity in healthy human adults and compare them to the spatial expression profile of genes coding for receptors interacting with LSD . The goal is to leverage the statistical properties of the slow ( <1 Hz ) intrinsic fluctuations of the blood-oxygen-level-dependent ( BOLD ) signal hemodynamics at rest ( i . e . resting-state functional connectivity ( rs-fcMRI ) ) . Critically , rs-fcMRI analyses are able to reveal the functional architecture of the brain , which is organized into large-scale systems exhibiting functional relationships across space and time ( Biswal et al . , 2010; Buckner et al . , 2013; Yang et al . , 2014 ) . Rs-fcMRI measures have furthermore revealed potential biomarkers of various neural disorders ( Murrough et al . , 2016; Yang et al . , 2016a ) , as well as proven sensitive to the effects of neuropharmacological agents ( Driesen et al . , 2013a; Anticevic et al . , 2015 ) . Focused analyses on specific regions revealed effects of intravenously administered LSD on functional connectivity between V1 and distributed cortical and subcortical regions ( Carhart-Harris et al . , 2016b ) . However , such ‘seed-based’ approaches rely on explicitly selecting specific regions of interest based on a priori hypotheses . Therefore , such an approach has limited ability to detect pharmacologically-induced dysconnectivity not predicted a priori . To characterize LSD effects on functional connectivity in the absence of strong a priori hypotheses , the current study employed a fully data-driven approach derived from graph theory called Global Brain Connectivity ( GBC ) ( Anticevic et al . , 2014b ) . In essence , GBC computes the connectivity of every voxel in the brain with all other voxels and summarizes that in a single value . Therefore , areas of high GBC are highly functionally connected with other areas and might play a role in coordinating large-scale patterns of brain activity ( Cole et al . , 2010 ) . Reductions in GBC may indicate decreased participation of a brain area in larger networks , whereas increased GBC may indicate a broadening or synchronization of functional networks ( Anticevic et al . , 2014b ) . One focused study examined GBC after intravenously administered LSD in a sample of 15 participants , revealing connectivity elevations across higher-order association cortices ( Tagliazucchi et al . , 2016 ) . While compelling , this preliminary study did not take into account the influence of global signal ( GS ) artefacts ( e . g . via global signal regression , GSR ) , which are known to exhibit massive differences in clinical populations and following pharmacological manipulations ( Power et al . , 2017; Yang et al . , 2016b; Lewis et al . , 2017; Driesen et al . , 2013b ) . Specifically , GS is hypothesized to contain a complex mixture of non-neuronal artefacts ( e . g . , physiological , movement , scanner-related ) , which can induce elevated relationships across the brain ( Yang et al . , 2014 ) . No study has examined LSD-induced changes as a function of GS removal . To inform this knowledge gap a major objective here was to study data-driven LSD-induced dysconnectivity in the context of GS removal . Another aim of the current study was to determine the extent to which the neural and behavioral effects of LSD are mediated by 5-HT2A receptors . Preclinical studies suggest that LSD binds potently to many neuroreceptors including 5-HT2A , 5-HT2C , 5-HT1A , D2 , and other receptors ( Marona-Lewicka et al . , 2002; Passie et al . , 2008 ) . Yet , a recent paper from our group ( Preller et al . , 2017 ) reported that the psychedelic effects of LSD were entirely blocked in humans by ketanserin ( Ket ) , a selective antagonist at 5-HT2A and α-adreno receptors ( Leysen et al . , 1982 ) . This would suggest that the neural effects of LSD should be blocked by Ket . It also suggests that networks modulated by LSD should highly be associated with the distribution of 5-HT2A receptors in the brain and not closely associated with the distribution of receptors unrelated to the mechanism of action of LSD . Here we leverage recent advances ( Burt et al . , 2018 ) in human cortical gene expression mapping to inform the spatial topography of neuropharmacologically-induced changes in data-driven connectivity . We hypothesized that the LSD-induced GBC change will quantitatively match the spatial expression profile of genes coding for the 5-HT2A receptor . In turn , we hypothesized that this effect will be preferential for the 5-HT2A but not other receptors and that the spatial match will be vastly improved after artefact removal . In doing so , this convergence of neuropharmacology and gene expression mapping validates the contribution of the 5-HT2A receptor to LSD neuropharmacology . In turn , it also highlights a general method for relating spatial gene expression profiles to neuropharmacological manipulations , which has direct and important implications for the rational refinement of any receptor neuropharmacology . Collectively , this pharmacological neuroimaging study addresses the following major knowledge gaps in our understanding of LSD neurobiology , by demonstrating: ( i ) data-driven LSD effects across brain-wide networks , which are exquisitely sensitive to GS removal , ( ii ) the subjective and neural effects of LSD neuropharmacology are attributable to the 5-HT2A receptor , and ( iii ) the cortex-wide LSD effects can be mapped onto the spatial expression profile of the gene coding for the 5-HT2A receptor . The main effect of drug on GBC computed with global signal regression ( GSR ) revealed significant ( TFCE type I error protected , 10000 permutations ) widespread differences in GBC between drug conditions in cortical and subcortical areas ( Figure 1 ) . Comparing LSD to Ketanserin+LSD ( Ket+LSD ) +Placebo ( Pla ) conditions across sessions shows that LSD induces hyper-connectivity predominately in sensory and somatomotor areas , that is the occipital cortex , the superior temporal gyrus , and the postcentral gyrus , as well as the precuneus . Hypo-connectivity was induced in subcortical areas as well as cortical areas associated with associative networks , such the medial and lateral prefrontal cortex , the cingulum , the insula , and the temporoparietal junction . All changes in connectivity were expressed bilaterally ( Figure 1A ) . Figure 1B shows mean connectivity strength ( Fz ) for each drug condition and the distribution of Fz values for grayordinates ( i . e . either a surface vertex ( node ) or a volume voxel in gray-matter ) showing significant hyper- and hypo-connectivity for LSD compared to ( Ket+LSD ) +Pla conditions . Mean Fz values do not differ between Pla and Ket+LSD conditions either in hyper-connected or in hypo-connected areas . Figure 1C depicts the comparison between LSD and Pla conditions and Figure 1D the comparison between LSD and Ket+LSD conditions . Similarly to the comparison LSD vs . ( Ket+LSD ) +Pla shown in Figure 1A , LSD compared to both Pla and Ket+LSD separately induced a connectivity pattern characterized by significant ( TFCE type I error protected , 10000 permutations ) hyper-connectivity in predominantly sensory areas and significant hypo-connectivity in associative networks . The similarity between the LSD>Pla and LSD>Ket+LSD contrasts is corroborated by a significant positive correlation ( r = 0 . 91 , p<0 . 001 ) between the respective Z-maps ( Figure 1E ) . Furthermore , only negligible differences were observed when comparing Ket+LSD and Placonditions directly ( Figure 1—figure supplement 1 ) . Furthermore , we tested if the directionality of LSD-induced effects on GBC ( hyper-connectivity across sensory and somatomotor networks , hypo-connectivity across associative networks ) are separable effects or result from functionally related systems-level perturbations . To this end , we correlated the mean connectivity strength difference between Pla and LSD in hyper-connected regions with mean connectivity strength difference in hypo-connected regions across subjects ( based on the LSD vs . ( Ket+LSD ) +Pla contrast ) . There was a significant correlation between hypo- and hyper-connectivity ( r = −0 . 90 , p<0 . 001 , Figure 1F ) indicating that participants with the highest LSD-induced coupling within sensory and somatomotor networks also showed the strongest LSD-induced de-coupling in associative networks . This suggests that LSD-induced alterations in information flow across these networks may result from systems-level perturbations . Together , these results indicate that LSD-induced GBC alterations are predominantly attributable to its agonistic activity onto the 5-HT2A receptor . In line with this , a repeated-measures ANOVA ( drug condition×scale ) was conducted for the retrospectively administered Altered States of Consciousness ( 5D-ASC ) questionnaire , and revealed significant main effects for drug condition ( F ( 2 , 46 ) =88 . 49 , p<0 . 001 ) and scale ( F ( 10 , 230 ) =14 . 47 , p<0 . 001 ) , and a significant interaction of drug condition×scale ( F ( 20 , 460 ) =13 . 02 , p<0 . 001 ) . Bonferroni corrected simple main effect analyses showed increased ratings on all 5D-ASC scales in the LSD condition compared to Pla and Ket+LSD conditions ( all p<0 . 05 ) except for the scales spiritual experience and anxiety ( all p>0 . 20 ) . Pla and LSD+Ket scores did not differ on any scale ( all p>0 . 90 ) ( Figure 1G ) . To investigate the influence of GSR on LSD results , we repeated the analyses presented above without GSR ( i . e . the effect of drug condition on GBC shown in in Figure 1 ) . The main effect of drug on GBC computed without GSR revealed significant predominantly left-hemispheric widespread differences in GBC between drug conditions ( Figure 2A , TFCE type I error protected , 10000 permutations ) . Figure 2B shows mean Fz for each drug condition and the distribution of Fz values within voxelgrayordinates showing significant hyper- and hypo-connectivity for LSD compared to ( Ket+LSD ) +Pla conditions . Mean Fz values for hypo-connected grayordinates differed significantly between Pla and Ket+LSD conditions . Mean Fz values for hyper-connected grayordinates differed significantly between Pla and Ket+LSD , and LSD and Ket+LSD conditions . Figure 2C depicts the comparison between LSD and Pla conditions . Without GSR LSD induced hypo-connectivity mainly in the right insula and hyper-connectivity predominantly in the cerebellum . Figure 2D shows the comparison between LSD and Ket+LSD conditions with LSD-induced hypo-connectivity in the left insula and widespread predominantly left-hemispheric hyper-connectivity in the frontal and temporal cortex , the tempoparietal junction , and the cerebellum . Comparing Ket+LSD and Pla conditions revealed Ket+LSD induced hyper-connectivity predominantly in the right hemisphere ( Figure 2E ) . LSD>Pla and LSD>Ket+LSD Z-maps were significantly correlated ( r = 0 . 81 , p<0 . 001 , Figure 2F ) . To test the relationship between hyper- and hypo-connectivity when GSR was not performed , we correlated the mean Fz difference between Pla and LSD without GSR in hyper-connected regions with the mean Fz difference in hypo-connected areas ( based on the LSD vs . ( Ket+LSD ) +Pla contrast ) across subjects . There was a significant positive correlation between hyper- and hypo-connectivity ( r = 0 . 92 , p<0 . 001 , Figure 2G ) . In contrast to the analysis performed with GSR showing a negative relationship between hyper- and hypo-connectivity change scores , the analysis without GSR indicates that participants with the highest LSD-induced hyper-connectivity showed the weakest LSD-induced de-coupling . Correlating the combined hyper- and hypo-connectivity values with GSR with those without GSR showed that these are not significantly related within subjects ( r = 0 . 003 , p=0 . 99 , Figure 2H ) . Furthermore , we tested the consistency of the hyper/hypo relationships with and without GSR by examining the areas that survived the type I error correction following TFCE with data that has undergone GSR ( Figure 1 ) . Here we focused on the areas showing hyper vs . hypo effects , which we used as masks to extract values for each person prior to GSR . If GSR altered or induced the hyper/hypo effect then we would hypothesize the correlation would weaken prior to GSR . The effect was not consistent with this null hypothesis – namely that the hyper/hypo individual difference remained highly stable even without GSR ( r = 0 . 92 , p<0 . 001 , Figure 2I ) . Put differently , this is not consistent with the hypothesis that the hyper/hypo changes are an artefact of the GSR process . We reported robust and widespread differences between GBC analyses results with and without GSR following LSD administration . This discrepancy calls into question interpretations regarding the directionality of LSD on sensory-somatomotor vs . association cortices , at least when assayed via the GBC metric . To inform this question , we conduced three additional analyses designed to investigate the influence of LSD and GSR on BOLD signal properties , which help inform and constrain GBC interpretations: First , we investigated if the amplitude of the BOLD signal is influenced by GSR across different experimental conditions ( i . e . LSD vs . Pla ) . Specifically , we quantified ‘amplitudes’ using a measure of local grayordinate-wise variance – an approach validated in our prior work in the context of clinical neuroimaging and effects of GSR in such datasets ( Yang et al . , 2014 ) . Figure 3—figure supplement 1 shows the change in local grayordinate-wise variance under LSD vs . Pla . The effect illustrates a very weak alteration in local variance ( min/max Z = −1 . 54/+2 . 28 ) . No areas survived whole-brain correction . This result in not consistent with the hypothesis that LSD markedly alters grayordinate-wise amplitudes/variance relative to Pla . Second , we calculated the mean variance of the GS across all grey matter grayordinates ( as opposed to local grayordinate-wise variance ) . We achieved this by defining the mean of all gray matter signal for a given subject based on their FreeSurfer segmentation and then computed the variance of the BOLD signal time course , averaged over all grayordinates in this global greymatter mask . Results indicate that variance of the GS does not differ significantly between conditions on average when computing the mean across all grey matter grayordinates [F ( 2 , 46 ) =0 . 71 , p>0 . 49 ) ] . Third , we investigated the possibility that the GS itself may exhibit a shifted topography on LSD , as shown in prior work ( Yang et al . , 2016b ) . Specifically , the two analyses above reveal that grayordinate-wise and average GS variance do not markedly differ for LSD vs . Pla . However , the mean GS analysis above cannot address the possibility that the GS signal itself has a distinct spatial configuration following LSD administration . In other words , which areas are maximally contributing to the mean GS may not be the same after LSD administration . To investigate the possibility of a shifted topography of the GS , as shown in prior work ( Yang et al . , 2016b ) , we computed the beta map of the GS for each subject ( see Materials and methods ) . This GS beta map allowed us to compare the spatial topography of GS under LSD vs . Pla conditions ( Figure 3A ) . As evident from the figure , the GS beta contrast was quite robust , especially when compared to the local grayordinate-wise variance results ( min/max Z = −5 . 73/+7 . 74 ) . Critically , the map revealed a bi-directional spatial shift of the GS under LSD where associative cortices and large areas of sub-cortex showed an elevated GS contribution . In contrast , the blue areas showed a reduced GS contribution under LSD . This map correlated highly with the spatial organization of the LSD-induced changes on GBC . This is unsurprising as GBC is highly sensitive to shared brain-wide signal shifts . Put differently , a GBC measure will be sensitive to the change in the mean shared signal across the brain . If LSD is altering this mean shared signal topography , then the GBC effect should be similarly affected . To quantify this we calculated the relationship between the LSD-Pla contrast GBC map before ( Figure 3C ) and after GSR ( Figure 3B ) and the LSD-Pla contrast GS beta map . The LSD-Pla contrast GS beta map exhibited a highly significant negative correlation with the LSD-Pla contrast GBC map after GSR ( r = 0 . 65 , p<0 . 001 ) , but a highly significant positive correlation before GSR ( r = 0 . 66 , p<0 . 001 ) . This result provides evidence consistent with the hypothesis that LSD induces a transformation in the GS beta map itself , which is contributing the GBC effect pre/post GSR . However , this analysis still does not inform the ‘ground truth’ effect of LSD on baseline connectivity – namely if LSD reduces or elevates connectivity across sensory and somatomotor versus associative cortices . There is a core limitation to the GBC metric in relation to the GS topography inherent to the way it is computed: Specifically , GBC yields the mean shared signal from a given grayordinate to all other grayordinates . This calculation is therefore affected by the shared variance across all grayordinates ( i . e . the map of the GS ) . If this shared GS variance structure is shifted in one condition versus the next , then the GBC calculation will shift accordingly in a spatially ordered way corresponding to the GS spatial shift . Therefore , it is not known from the GBC effects alone if LSD elevates or reduces mean connectivity in associative vs . sensory and somatomotor cortices . This interpretational challenge stems from the presented GS beta map analyses because it is not clear if GS beta map transformations on the GBC effect under LSD are primarily neural or artefactual . To address this , we designed a complementary analysis , which yields a map that is interpretationally consistent irrespective of GS-related shifts . Here we focused on the thalamo-cortical system leveraging a well-established effect that is not affected by GSR transformations ( Yang et al . , 2014; Anticevic et al . , 2014a; Woodward et al . , 2012 ) . To examine brain-wide thalamic coupling in session one we computed a seed-based map by extracting average time-series across all grayordinates in each subject's anatomically defined bilateral thalamus ( via FreeSurfer segmentations ) . To examine between-drug differences , thalamic maps were entered into a second level analysis as done for the GBC measures ( mean thalamic connectivity maps for each condition are displayed in Figure 4—figure supplement 1 and all contrasts in Figure 4—figure supplement 2 ) . Here we used both the correlation and covariance as methods of statistical association , which we conjuncted ( Cole et al . , 2016 ) . We did this because covariance reflects a non-normalized measure of shared BOLD signal across time ( which is scale free and unaffected by variance structure ) whereas correlation is inherently normalized by pooled variance . Furthermore , given that GS induces mean signal shifts ( i . e . it may induce anti-correlation ) , we also obtained the top and bottom 10% of all thalamic connections from both correlation and covariance maps before and after GSR . This final 4-way conjunction map ensured that the resulting regions in the top/bottom ranges exhibit thalamic coupling irrespective of processing ( i . e . GSR/noGSR ) or statistical method ( i . e . r or cov ) . This map was then used to calculate LSD induced effects on top 10% and bottom 10% of seed thalamic brain-wide connections . The prediction was that LSD would decrease thalamic connections that were in the top 10% ( i . e . highly positive thalamic connections at baseline , which represent thalamo-associative coupling , Figure 4A ) . In turn , we predicted that LSD would elevate connections that were in the bottom 10% ( i . e . very weak thalamic connections at baseline , which represent thalamo-sensory coupling , Figure 4A ) . Figure 4B and C shows the difference in the average signal between drug conditions for the correlation method after GSR . Here , LSD consistently decreases coupling in associative areas and increases FC in sensory-somatomotor regions ( Figure 4A ) . Critically , this effect was preserved for the LSD- ( Ket+LSD ) analysis ( Figure 4B ) . Without GSR however , inconsistent results emerged ( Figure 4—figure supplement 2 ) . To reconcile the interpretation of LSD-induced directionality we leveraged the conjunction map that was robust to processing method and statistical approach . This conjunction map was used as a mask to extract the average signal across these regions in the LSD-Pla and LSD- ( Ket+LSD ) contrast before and after GSR in the seed-based correlation/covariance analyses as well as GBC correlation/covariance analyses . Figure 4F shows the difference in the average signal between these drug conditions across analyses methods ( thalamic seed FC/GBC , correlation/covariance ) after GSR . Here , LSD consistently decreased thalamic coupling in associative areas and increased thalamic coupling in sensory-somatomotor regions irrespective of analysis method . Importantly , the thalamic seed analyses matched GBC effects . Without GSR however ( Figure 4G ) , seed thalamic coupling and GBC results were inconsistent . Furthermore , in contrast to results after GSR , LSD did not consistently decrease connections that were in the top 10% or elevate connections that were in the bottom 10% without GSR . To investigate individual differences , we computed the correlation between the top and bottom connections before and after GSR across participants ( Figure 4H , full connectivity matrix is presented in Figure 4—figure supplement 3 ) . The prediction was that individuals with biggest elevation for bottom 10% would should the biggest drop in the top 10% under LSD . Predicted negative individual differences emerged after GSR but without GSR results were not compatible with individual difference predictions based either for thalamic seed analysis or GBC . Collectively , these results are consistent with the hypothesis that , following GS cleanup , LSD reduces shared signals for association cortices but elevates shared signals for sensory and somatomotor areas across both seed-based thalamic and GBC analyses . To investigate the time course of subjective effects , a short version of the 5D-ASC was administered 180 min , 250 min , and 360 min after the second drug administration . A repeated-measures ( drug condition×time×scale ) ANOVA for the short-version 5D-ASC questionnaire revealed significant main effects for drug condtition ( F ( 2 , 44 ) =58 . 32 , p<0 . 001 ) , time ( F ( 2 , 44 ) =26 . 61 , p<0 . 001 ) , and scale ( F ( 4 , 88 ) =14 . 83 , p<0 . 001 ) and significant interactions for drug condtion×time ( F ( 4 , 88 ) =16 . 89 , p<0 . 001 ) , treatmentdrug condition×scale ( F ( 8 , 176 ) =12 . 82 , p<0 . 001 ) , time×scale ( F ( 8 , 176 ) =4 . 05 , p<0 . 001 ) , and drug condition×time×scale ( F ( 16 , 352 ) =2 . 22 , p<0 . 01 ) . Bonferroni-corrected simple main effect analyses revealed that score in the LSD treatment condition differed significantly from score in the Pla and Ket+LSD treatment conditions for the blissful state scale , disembodiment scale , elementary imagery scale , and changed meaning of percepts scale at 180 and 250 min after treatment intake ( all p<0 . 05 ) . 360 min after intake , score on the disembodiment scale and elementary imagery scale was significantly greater in the LSD treatment condition than in the Pla and Ket+LSD treatment conditions . Scores did not differ between the Pla and Ket+LSD treatment conditions for any scale at any time point ( all p>0 . 90; Figure 5 , Figure 5—source data 1 ) . Test-retest reliability of these measures is high . Within each drug condition , mean scores over time correlated highly and significantly ( all Pearson’s r > 0 . 40 , max r = 0 . 99 ) . Figure 5—figure supplement 1 shows the correlation coefficients between scores on the five subscales in the LSD condition at the three time points . To investigate the potentially distinct temporal phases of LSD pharmacology ( Marona-Lewicka et al . , 2005; Marona-Lewicka and Nichols , 2007 ) , two resting-state scans were conducted on each test day: 75 min ( session 1 ) and 300 min ( session 2 ) after the second drug administration . No significant differences in GBC were observed when comparing session 1 and 2 within the Pla and the LSD condition ( Figure 6—figure supplement 1 ) . Within the Ket+LSD condition , participants showed significant decreases in GBC in session two compared to session one predominantly in occipital areas . Increases in GBC in session two were found in cortical regions such as the anterior and posterior cingulate cortex , and the temporoparietal junction , as well as subcortical structures including the thalamus and the basal ganglia ( Figure 6A ) . Figure 6B shows the significant ( p<0 . 05 ) difference in mean Fz between session 1 and session two in hyper- and hypo-connected areas and the distribution of Fz values for grayordinates within hyper- and hypo-connected areas for both sessions ( hyper- and hypo-connected areas are derived from the LSD vs . ( Ket+LSD ) +Pla contrast , see Figure 1A ) . We next specifically investigated mean Fz ( with and without GSR ) for all drug conditions and sessions for grayordinates within seven functionally-defined networks using parcellations derived by Yeo et al . ( 2011 ) , Buckner et al . ( 2011 ) and Choi et al . ( 2012 ) ( Figure 7 ) . This parcellation contains both sensory ( visual and somatomotor ) and associative ( dorsal attention , ventral attention , limbic , frontoparietal control , and default mode ) networks . Repeated-measures ANOVAs revealed significant main effects for drug condition for all networks ( all p<0 . 05 ) except for the dorsal attention network when including GSR , with the LSD condition differing significantly from both , Pla and Ket+LSD conditions ( all p<0 . 05 , Bonferroni corrected ) , except for the somatomotor network , where LSD differed significantly only from Ket+LSD . Pla and Ket+LSD conditions did not differ significantly in any network . Without GSR , main effects for drug were found in the frontoparietal control network ( F ( 2 , 46 ) =4 . 09 , p<0 . 03 ) with significantly lower values in the Ket+LSD condition than in both , the LSD and Pla condition , and dorsal attention network ( F ( 2 , 46 ) =3 . 86 , p<0 . 04 ) with significantly lower values in the Ket+LSD condition than in the Pla condition . To evaluate the relationship between LSD-induced changes in GBC in functional networks and subjective LSD-induced effects , Fz mean connectivity change ( LSD–Pla condition , session 2 , with GSR ) in the seven functional networks ( see Figure 7 ) was correlated with the mean 5D-ASC short version score at 250 mins ( assessment closest in time to resting-state data collection , see Figure 1—figure supplement 2 and Figure 5 ) . Correlating measures at session two allows high stability in LSD-induced effects . Bonferroni corrected correlations showed a significant relationship between the change in Fz connectivity in the somatomotor network and subjective LSD-induced effects ( r = 0 . 81 , p<0 . 001 , Bonferroni corrected , Figure 8A and Figure 8B ) . Correlations between mean 5D-ASC score and Fz connectivity in the other six networks and did not reveal significant relationships ( all p>0 . 16 , Bonferroni corrected ) . To further investigate the contribution of specific LSD-induced symptoms to the relationship with somatomotor network Fz connectivity , we calculated the correlation between Fz mean connectivity change in the somatomotor network with each 5D-ASC short version scale separately . All five scale scores ( blissful state , disembodiment , changed meaning of percepts , elementary imagery , spiritual experience ) were significantly correlated with Fz mean connectivity change in the somatomotor network ( all p<0 . 05 , Bonferroni corrected , Figure 7C–G ) , indicating that the relationship between somatomotor network Fz connectivity and subjective effects was not driven by a specific LSD-induced symptom alone . The five scale scores were moderately to strongly correlated with each other ( r = 0 . 39 – 0 . 82 ) . Correlating mean Fz connectivity changes without GSR in the seven functional networks with subjective effects did not reveal any significant result ( all p>0 . 3 , unc ) . LSD stimulates not only 5-HT2A receptors but also 5-HT2C , −1A/B , −6 , and −7 and dopamine D2 and D1 receptors . These receptors are differentially expressed across the cortex . To further investigate LSD’s receptor pharmacology , we tested the correlation between unthresholded Z-score map for LSD condition vs . ( Ket+LSD ) +Pla condition with and without GSR and six available receptor gene expression maps of interest ( DRD1 , DRD2 , HTR1A , HTR2A , HTR2C , and HTR7 ) derived from the Allen Human Brain Atlas ( Burt et al . , 2018; Hawrylycz et al . , 2012 ) . Figure 9A shows the average GBC Z-score with and without GSR and the mean gene expression of the genes of interest within the seven functionally-defined networks using parcellations derived by Yeo et al . ( 2011 ) , Buckner et al . ( 2011 ) and Choi et al . ( 2012 ) ( see also Figure 7 ) , indicating that gene expression is distinct by network . Next , we investigated whether there is a common pattern of distribution between the six gene expression maps . Correlation analyses showed that the expression of the main gene of interest ( HTR2A ) is highly negatively correlated with the expression of HTR7 ( r = −0 . 68 , p<0 . 001 , Bonferroni corrected , Figure 9B ) . Figure 9C illustrates the cortical distribution of HTR2A gene expression . This HTR2A cortical gene expression map is highly correlated with the unthresholded GBC Z-score map for the LSD condition vs . ( Ket+LSD ) +Pla condition with GSR ( r = 0 . 50 , p<0 . 001 ) , and higher than all other candidate serotonin receptor genes . While the correlation between the unthresholded GBC Z-score map without GSR and the HTR2A cortical gene expression map also reached significance ( r = 0 . 18 , p<0 . 001 ) , this correlation was significantly weaker than between the Z-score map with GSR and the HTR2A gene expression map ( p<0 . 05 , Bonferroni corrected , Figure 9F ) . Taking into account all available gene expression maps the correlation between the Z-score map with GSR and the HTR2A gene expression map was higher that 95 . 9% of all possible correlations . The GBC Z-score map with GSR and the HTR7 gene expression map was lower than 99 . 8% of all possible correlations , indicating a strong negative relationship ( r = −0 . 63 , p<0 . 001 , Figure 9D ) . The correlation between both HTR2A and HTR7 with the GBC Z-Score map is not surprising considering the strong negative correlation between HTR2A and HTR7 gene expression maps ( Figure 9B ) . Lastly , Figure 9F illustrates that correlation coefficients between gene expression maps and GBC Z-score maps were significantly stronger with GSR ( all p<0 . 05 , Bonferoni corrected ) , except for DRD1 expression map , where the absolute value of the correlation coefficient increased when correlated with the Z-score map without GSR . We show across conditions that LSD induces hyper-connectivity predominantly in sensory and somatomotor areas ( i . e . the occipital cortex , the superior temporal gyrus , and the postcentral gyrus ) . LSD-induced hypo-connectivity was observed across subcortical areas ( with the exception of the amygdala and sensory thalamus ) as well as cortical areas associated with associative networks ( i . e . the medial and lateral prefrontal cortex , the cingulum , the insula , and the temporoparietal junction ) ( Figure 1 ) . These results are consistent with the hypothesis that LSD induces a de-synchronization of associative networks whereas sensory and somatomotor areas exhibit elevated brain-wide shared signal ( Anticevic et al . , 2014b ) . This is in line with previous seed-based studies reporting increased V1 resting-state connectivity with the rest of the brain after LSD administration ( Carhart-Harris et al . , 2016b ) . Additionally , decreases of connectivity within the DMN were reported after ayahuasca , a tea containing the hallucinogenic 5-HT2A receptor agonist N , N-Dimethyltryptamine , intake ( Palhano-Fontes et al . , 2015 ) . Furthermore , desynchronization within the DMN was reported after psilocybin infusion measured with magnetoencephalography ( Muthukumaraswamy et al . , 2013 ) . As noted , while subcortical areas predominantly show hypo-connectivity under LSD there were key exceptions: the amygdala exhibited brain-wide hyper-connectivity under LSD . Amygdala neurons abundantly express 5-HT2A receptors , and alterations in amygdala activity and connectivity have been hypothesized to be important for potential beneficial clinical effects of psychedelics ( Rainnie , 1999; Kraehenmann et al . , 2016 ) . Furthermore , results showed that participants with the highest LSD-induced coupling within sensory and somatomotor networks also showed the strongest LSD-induced de-coupling in associative networks . This suggests that LSD-induced alterations in information flow across these networks probably results from linked systems-level perturbations , as opposed to being due to dissociable mechanisms across subjects . This pattern of hyper-and hypo-connectivity may underlie the psychedelic state , suggesting increased processing of sensory information which is not counterbalanced by associative network integrity . Consequently , this may result in an altered state of consciousness whereby internal and external sensory computations are not integrated , leading to psychedelic symptoms . A virtually identical pattern of hyper-connectivity in sensory networks and hypo-connectivity in associative networks is revealed when contrasting LSD effects with the condition where LSD is blocked by pre-administration of Ket . This brain-wide net effect of LSD was virtually indistinguishable from LSD vs . Pla condition . Put differently , pre-treatment of LSD with Ketanserin induced only negligible changes compared to LSD vs . Pla , indicating that Ket blocked virtually all LSD-induced alterations in GBC . Ketanserin has high antagonistic properties particularly on the 5-HT2A receptor . Thus , these results indicate that LSD-induced alterations in neural and behavioral effects are highly dependent on stimulation of the 5-HT2A receptor ( Leysen et al . , 1982 ) . This is in line with data on subjective LSD-induced effects which were normalized by Ket pretreatment ( Figure 1G ) . Collectively , these data extend previous studies by revealing that the described pattern of brain-wide dysconnectivity may be directly attributable to stimulation of the 5-HT2A receptor . Specifically , the effect can be characterized by brain-wide integration of sensory networks and dis-integration of associative networks , which presumably underlie LSD-induced altered state of consciousness . Furthermore , these data highlight the importance of the 5-HT2A receptor system in LSD-induced neural and behavioral alterations . One small study by Tagliazucchi et al . has previously investigated the effects of intravenously administered LSD and reports that association cortices ( partially overlapping with the default-mode , salience , and frontoparietal attention networks ) and the thalamus showed increased GBC under LSD ( Tagliazucchi et al . , 2016 ) . These results are partially contradictory to the data presented in Figure 1 . Importantly , this previous study did not quantify the influence of the GS , which likely contains a complex mixture of non-neuronal artefacts ( e . g . respiration , which may be increased under LSD [Power et al . , 2017; Glasser et al . , 2017] ) . Such artefacts can induce spuriously high statistical association across the brain ( Yang et al . , 2014; Coyle , 2006 ) . Due to these discrepancies , we studied the data as a function of GS regression to inform how this methodological step affects results ( Tagliazucchi et al . , 2016 ) . These analysis support the observations by Tagliazucchi et al . ( 2016 ) when GS was not considered: results showed increased GBC in fronto-parietal , temporal , and subcortical areas ( Figure 2 ) . The analysis without GSR showed a positive correlation between hyper- and hypo-connectivity change scores , indicating that participants with the highest LSD-induced hyper-connectivity showed the weakest LSD-induced de-coupling . This was in contrast to results after GS removal . Furthermore , connectivity values with GSR and those without GSR were not significantly correlated within subjects ( Figure 2 ) . Notably , without GS regression , LSD-Pla differences were not observed when examining seven functionally pre-defined networks ( Figure 7 ) . These data suggest that GS related effects cannot be explained by a mean-shift in connectivity values on average , but instead may reflect a process within each subject . One hypothesis is that GSR statistically attenuates non-neural arts and therefore provides a method to better isolate functional networks in pharmacological resting-state connectivity studies ( Yang et al . , 2014; Coyle , 2006 ) . This interpretation is consistent with the absence of a neural-symptom correlation without GSR . Finally , spatial correlations with gene expression maps ( discussed below ) were notably attenuated for without GSR . That said , this dataset is not well-suited for drawing conclusions about GSR suitability for pharmacological neuroimaging . In fact , it can be argued that results are more replicable across session 1 and 2 without GSR . However , the statistical phenomenon of ‘artefactual’ replication is not surprising , if one considers that GSR is designed to attenuate sources of spatially pervasive structured artefacts which may persist across sessions; f . e . elevated respiratory artefacts ) . Put differently , there is a key nuance between ‘noise’ and ‘artefact’ . Pure unstructured noise can be signal-averaged out and would not yield a consistent ‘artefactual’ effect . In contrast , ‘structured’ artefact represents a signal that can induce the same spurious effect multiple times ( Power et al . , 2018; Glasser et al . , 2018 ) . Therefore , if a structured artefact is large across both measurements ( session 1 and 2 ) , then this artefact will spuriously drive the effect and will be replicable . Experiments that manipulate variables such as breathing rate and vigilance will be key to fully characterize the effects of GSR on pharmacological neuroimaging data and help separate neuropharmacological effects associated with ‘global artefacts’ versus those affecting ‘global neural signal’ ( Glasser et al . , 2018 ) . Furthermore , there are open knowledge gaps regarding LSD’s effects on neurovascular coupling and the hemodynamic response function properties . This pitfall needs to be addressed in experiments incorporating measures specifically designed to investigate changes in hemodynamic coupling . Here animal studies , which offer the possibility to combine neuronal recordings with simultaneous measurement of hemodynamics , will be critical to help interpret LSD effects inhumans . As expected , the GS analyses indicate that a GBC metric is highly sensitive to global shared signal , which is altered under LSD . This raises the question regarding the direction of the LSD-induced effects on association vs . sensory-somatomotor areas . To inform the directionality of LSD modulations of GBC we completed an additional analyses based on seed-based thalamic functional connectivity , which yielded a map that was robust to GS transformations . The reason for this phenomenon is that the thalamus exhibits strong bi-directional brain-wide shared signal . Furthermore , we constructed a conjunction measure that identified baseline ( i . e . Pla ) thalamic FC that was interpretationally consistent irrespective of GS-related shifts ( Figure 4 ) . This seed-based conjunction analysis revealed that LSD-induced changes were consistent after GSR and comparable to GBC effects . Without GS removal , neither the thalamic nor GBC effects converged across metrics . This observation , however , does not rule out the possibility that GS removal in fact attenuates signal components that are neuronal in origin and may be relevant to important LSD-induced properties ( Tagliazucchi et al . , 2016 ) . Careful manipulation and measurement of respiratory-related artefacts during pharmacological fMRI is needed to disambiguate the amount of GS variance that relates to neuronal vs . artefactual LSD effects . Animal studies suggest distinct temporal phases of LSD pharmacology ( Marona-Lewicka et al . , 2005; Marona-Lewicka and Nichols , 2007 ) . Therefore , we investigated the time-dependent effects of LSD on subjective effects as well as on GBC . As shown in Figure 5 , subjective effects were highest 180 min after LSD administration and decreased in intensity 250 and 360 min after administration as expected ( Passie et al . , 2008; Schmid et al . , 2015 ) . No differences were found between Pla and Ket+LSD conditions at any time point , with subjective effects in both conditions being very low in intensity ( <4 . 7% ) . This shows that Ket blocked subjective LSD-induced effects over the whole time course , indicating that subjective effects are most likely attributable to 5-HT2A receptor stimulation . To investigate the time course of LSD-induced effects on GBC , two resting-state scans were analyzed , conducted 75 min ( session 1 ) and 300 min ( session 2 ) after the second drug administration . While no significant differences were observed when comparing session 1 and 2 within the Pla and the LSD condition ( Figure 6—figure supplement 1 ) , participants showed significant changes in GBC in session two compared to session one in the Ket+LSD condition . Taken together with time-dependent results observed in the functionally defined networks ( Figure 6 ) , the blocking effect of ketanersin is particularly evident in the session one across all networks . Specifically , Ket not only blocks LSD effects in session one but also augments the effects seen in Pla , indicating the opposite mechanism of action from that seen by LSD – namely 5-HT2A receptor antagonism ( Leysen et al . , 1982; Kometer et al . , 2013 ) . On the other hand , it seems possible that there exist two distinct pharmacological time phases as described in animal studies . The first phase may be modulated by 5-HT2A receptor activation and the second phase possibly by D2 receptor activation , as suggested by preclinical work ( Marona-Lewicka et al . , 2005; Marona-Lewicka and Nichols , 2007 ) . This hypothesis of time-dependent complex receptor pharmacology awaits further testing . First , studies are needed to investigate the effect of Ket alone on GBC to verify the preferential effects of 5-HT2A antagonism . Second , studies using pre-treatment of LSD by antagonists on receptors other than 5-HT2A are needed to determine if the second phase is indeed modulated by another receptor system . Lastly , indications of different pharmacological phases are not evident from subjective drug effects which remain completely blocked by Ket . Studies using higher doses of LSD are therefore needed to investigate if the potential effect of LSD’s action on other receptors becomes more pronounced and therefore subjectively accessible . The LSD-induced change in connectivity in the somatomotor network correlated significantly with general and specific subjective LSD-induced effects ( mean across all scales , blissful state , disembodiment , changed meaning of percepts , elementary imagery , spiritual experience ) . Participants with increased connectivity in the somatomotor network also showed higher subjective effects . On average , the change in connectivity between the LSD and Pla condition in the somatomotor network was not significant . However , this is likely explained by the heterogeneous connectivity changes within this network: while the pre- and postcentral gyrus predominantly showed increases in GBC , medial areas were hypo-connected . Connectivity changes in other functional networks were not significantly correlated with subjective effects . This points to the importance somatomotor network brain regions and their connectivity with the rest of the brain for psychedelic experiences . This is in line with previous results obtained from task-related data showing that the supplementary motor area is associated with LSD-induced alterations in meaning and personal relevance processing ( Preller et al . , 2017 ) . These results also support broader theories of consciousness emphasizing the importance of the sensorimotor system for the perception of presence and agency , and therefore a sense of self ( e . g . , interoceptive predictive coding model of conscious presence ( Seth et al . , 2011 ) , comparator model ( Frith et al . , 2000; Allen et al . , 2016; Blanke and Metzinger , 2009 ) . Furthermore , alterations in sensorimotor gating have been suggested to underlie psychedelic experiences ( Quednow et al . , 2012; Ludewig et al . , 2003 ) . Somatomotor system activity and connectivity has also been implicated in the pathophysiology of schizophrenia ( Anticevic et al . , 2014a ) , an illness characterized by delusions and alterations in the sense of self , potentially arising from alterations in sensorimotor gating deficits in an inferential mechanism that allows distinguishing whether or not a sensory event has been self-produced ( Synofzik et al . , 2010 ) . The current results corroborate and extend these previous findings by showing that somatomotor network connectivity is also closely associated with an LSD-induced psychedelic state . To further investigate LSD’s receptor pharmacology we specifically used the threshold-free Z-score map of LSD effects relative to Ket blockade and Pla . The logic here is that such a map may reflect Ket-specific contributions to LSD blockade , which is hypothesized to involve the 5-HT2A receptor . This map was then correlated with gene expression maps of receptors that may be stimulated by LSD ( Nichols , 2004 ) . LSD-induced changes in functional connectivity after GSR exhibited strong positive relationships with HTR2A expression ( higher than 95 . 9% of all possible gene expression correlations , Figure 9 ) . These results show that LSD-induced changes in GBC quantitatively match the spatial expression profile of genes coding for the 5-HT2A receptor , supporting the central role of this receptor system in LSD’s neuronal and subjective effects . LSD-induced changes in functional connectivity were also highly negatively correlated with HTR7 gene expression ( lower than 99 . 8% of all possible gene expression correlations , Figure 9 ) . This can be explained by the highly anti-correlated expression of these two genes ( Figure 9 ) . However , it is also possible that the 5-HT7 receptor functionally contributes to LSD-induced effects . In contrast to its agonistic properties on the 5-HT2A receptor , LSD has been reported to be an antagonist in the 5-HT7 receptor ( Wacker et al . , 2013 ) . Since previous studies have shown that 5-HT7 receptor antagonists have anti-psychotic potential ( Waters et al . , 2012; Abbas et al . , 2009 ) , it seems very unlikely that LSD’s effects have a strong and appreciable contribution on the 5-HT7 receptor . However , future studies should examine 5-HT7 receptor pharmacology more carefully as they may reveal a role of this receptor system in pro-cognitive effects that contrast those of LSD . While the current results strongly implicate the involvement of the 5-HT2A receptor in LSD-induced effects , it must be noted that no further conclusions can be drawn regarding the functional contribution of other receptors agonized or antagonized by LSD . This limitation needs further investigation in future studies by blocking serotonin and dopamine receptors involved in the pharmacology of LSD beyond the 5-HT2A receptor . Furthermore , the contribution of these receptors to the effects of different doses of LSD still need to be studied . Finally , we show that the spatial match between gene expression maps and GBC maps is significantly improved after GSR , even though correlation coefficients in particular for DRD1 , DRD2 , HTR1A , and HTR2C remain moderate . These results also highlight the validity of this approach as a general method for relating spatial gene expression profiles to neuropharmacological manipulations in humans . An important next step allowing further methodological validation is comparing LSD-induced alterations in GBC with receptor maps provided by Positron Emission Tomography ( PET ) ( Saulin et al . , 2012; Ettrup et al . , 2016; Ettrup et al . , 2014 ) , preferably using MR scanners that are both MR and PET compliant allowing for cross-validation across BOLD and PET modalities within the same person . In summary , the current results close major knowledge-gaps regarding the neurobiology and neuropharmacology of LSD . First , we show that LSD induces widespread alterations of GBC in cortical and subcortical brain areas , characterized by a synchronization of sensory and somatomotor functional networks and dis-integration of associative networks . We show that this effect is sensitive to GSR , which has important implications for future pharmacological resting-state studies . Second , we investigated the receptor-pharmacology of LSD , showing that the 5-HT2A receptor plays a critical role in subjective and neuronal LSD-induced effects . However , analyzing the time course of LSD-induced alterations in functional connectivity , it seems likely that at a later phase , modulation by receptors other than the 5-HT2A receptor is involved . The comparison of LSD-induced effects on functional connectivity and receptor-gene expression maps underscores the interpretations of 5-HT2A pharmacology and points to potentially impactful studies on 5-HT7 receptor pharmacology . Lastly , in line with various theories of consciousness we showed that the somatomotor system in particular is related to LSD-induced psychedelic effects . Collectively , these results deepen our understanding of psychedelic compounds and offer important directions for development of novel therapeutics . Participants were recruited through advertisements placed in local universities from March to July 2015 and underwent a screening visit before inclusion in the larger study protocol ( Preller et al . , 2017 ) . All included subjects were healthy according to medical history , physical examination , blood analysis , and electrocardiography . The Mini-International Neuropsychiatric Interview ( MINI-SCID ) ( Sheehan et al . , 1998 ) , the DSM-IV self-rating questionnaire for Axis-II personality disorders ( SCID-II ) ( Fydrich et al . , 1997 ) , and the Hopkins Symptom Checklist ( SCL-90-R ) ( Franke , 1995 ) were used to exclude subjects with present or previous psychiatric disorders or a history of major psychiatric disorders in first-degree relatives . Participants were asked to abstain from the use of any prescription or illicit drugs for a minimum of two weeks prior to the first test day and for the duration of the entire study , and to abstain from drinking alcohol for at least 24 hr prior to test days . Urine tests and self-report questionnaires were used to verify the absence of drug and alcohol use . Urine tests were also used to exclude pregnancy . Participants were furthermore required to abstain from smoking for at least 60 min before MRI assessment and from drinking caffeine during the test day . Further exclusion criteria included left-handedness , poor knowledge of the German language , cardiovascular disease , history of head injury or neurological disorder , history of alcohol or illicit drug dependence , magnetic resonance imaging ( MRI ) exclusion criteria including claustrophobia , and previous significant adverse reactions to a hallucinogenic drug . Twenty-five participants took part in the study . One subject was excluded due to failure in registration caused by an improper head position . Therefore a sample of 24 participants was included in the final analysis ( n = 19 males and n = 5 females; mean age = 25 . 00 years; standard deviation ( SD ) = 3 . 60 years; range 20 – 34 years ) . All participants provided written informed consent statements in accordance with the declaration of Helsinki before participation in the study . Subjects received written and oral descriptions of the study procedures , as well as details regarding the effects and possible risks of LSD and Ket treatment . The Swiss Federal Office of Public Health , Bern , Switzerland , authorized the use of LSD in humans , and the study was approved by the Cantonal Ethics Committee of Zurich ( KEK-ZH_No: 2014_0496 ) . The study was registered at ClinicalTrials . gov ( NCT02451072 ) . No substantial side effects were recorded during the study . Four participants reported transient headaches after drug effects had worn off . One participant reported transient sleep disturbances for the first two nights after drug administration . Participants were contacted again three months after the last drug administration . No further side effects were recorded . The study employed a fully double-blind , randomized , cross-over design ( see Figure 1—figure supplement 2 ) . Randomization was completed by a study nurse who had no other role in the trial . Sample size ( n = 24 ) was determined based on a previous study reporting LSD-induced effects on functional brain connectivity ( Tagliazucchi et al . , 2016 ) . Recruitment was stopped after the determined sample size was reached . Specifically , participants received either: ( i ) placebo +placebo ( Pla ) condition: placebo ( 179 mg Mannitol and Aerosil 1 mg po ) after pretreatment with placebo ( 179 mg Mannitol and Aerosil 1 mg po ) ; ( ii ) placebo +LSD ( LSD ) condition: LSD ( 100 µg po ) after pretreatment with placebo ( 179 mg Mannitol and Aerosil 1 mg po ) , or ( iii ) Ketanserin +LSD ( Ket+LSD ) condition: LSD ( 100 µg po ) after pretreatment with the 5-HT2A antagonist Ket ( 40 mg po ) at three different occasions two weeks apart . Pretreatment with placebo or Ket occurred 60 min before treatment with placebo or LSD . The resting-state scan was conducted 75 and 300 min after treatment administration . Participants were asked to not engage in repetitive thoughts such as counting and close their eyes during the resting state scan . Compliance to this instruction was monitored online using eye tracking ( NordicNeuroLab VisualSystem , http://www . nordicneurolab . com/ ) . The 5D-ASC ( a retrospective self-report questionnaire ) ( Dittrich , 1998 ) was administered to participants 720 min after drug treatment to assess subjective experience after drug intake . In addition , a short version of the 5D-ASC was administered 180 , 250 , and 360 min after drug treatment to assess the time course of subjective experience . Magnetic resonance imaging ( MRI ) data were acquired on a Philips Achieva 3 . 0T whole-body scanner ( Best , The Netherlands ) . A 32-channel receive head coil and MultiTransmit parallel radio frequency transmission was used . Images were acquired using a whole-brain gradient-echo planar imaging ( EPI ) sequence ( repetition time = 2 , 500 ms; echo time = 27 ms; slice thickness = 3 mm; 45 axial slices; no slice gap; field of view = 240 × 240 mm2; in-plane resolution = 3 × 3 mm; sensitivity-encoding reduction factor = 2 . 0 ) . 240 volumes were acquired per resting state scan resulting in a scan duration of 10 mins . Additionally , two high-resolution anatomical images were acquired using T1-weighted and T2-weighted sequences . T1-weigthed images were collected via a 3D magnetization-prepared rapid gradient-echo sequence ( MP-RAGE ) with the following parameters: voxel size = 0 . 7×0 . 7×0 . 7 mm3 , time between two inversion pulses = 3123 ms , inversion time = 1055 ms , inter-echo delay = 12 ms , flip angle = 8° , matrix = 320×335 , field of view = 224×235 mm2 , 236 sagittal slices . Furthermore T2-weighted images were collected using via a turbo spin-echo sequence with the following parameters: voxel size = 0 . 7×0 . 7×0 . 7 mm3 , repetition time = 2500 ms , echo time = 415 ms , flip angle = 90° , matrix = 320×335 , field of view = 224×235 mm2 , 236 sagittal slices . Structural and functional MRI data were first preprocessed according the methods provided by the Human Connectome Project ( HCP , RRID:SCR_006942 ) , outlined below , and described in detail by the WU-Minn HCP consortium ( WU-Minn HCP Consortium et al . , 2013 ) . These open-source HCP algorithms , optimized for our specific acquisition parameters and Yale’s High Performance Computing resources , represent the current state-of-the-art in distortion correction , registration , and maximization of high-resolution signal-to-noise ( SNR ) . Here we briefly describe the processing steps . Complete details are outlined by Glasser and colleagues ( WU-Minn HCP Consortium et al . , 2013 ) . First , the T1w/T2w images were corrected for bias-field distortions and warped to the standard Montreal Neurological Institute-152 ( MNI-152 ) brain template through a combination of linear and non-linear transformations using the FMRIB Software Library ( FSL , RRID:SCR_002823 ) linear image registration tool ( FLIRT ) and non-linear image registration tool ( FNIRT ) ( Jenkinson et al . , 2002 ) . Then , FreeSurfer’s recon-all pipeline was employed to compute brain-extraction , within-subjects registration , and individual cortical and subcortical anatomical segmentation ( Reuter et al . , 2012 ) . T1w/T2w images were then converted to the Connectivity Informatics Technology Initiative ( CIFTI ) volume/surface ‘grayordinate’ space . Raw BOLD images were first corrected for field inhomogeneity distortion , phase encoding direction distortions and susceptibility arts using the pair of reverse phase-encoded spin-echo field-map images implemented via FSL’s TOPUP algorithm ( Andersson et al . , 2003 ) . Motion-correction was then performed by registering each volume in a run to the corresponding single-band reference image collected at the start of each run . BOLD images were then registered to the structural images via FLIRT/FNIRT , and a brain-mask was applied to exclude signal from non-brain tissue . After processing in NIFTI volume space , BOLD data were converted to the CIFTI gray matter matrix by sampling from the anatomically-defined gray matter ribbon . Following these minimal HCP preprocessing steps , a high-pass filter ( >0 . 008 Hz ) was applied to the BOLD time series in order to remove low frequencies and scanner drift . In-house MATLAB ( RRID:SCR_001622 ) tools were used to compute the average variation in BOLD signal in the ventricles and deep white matter . This signal was regressed out of the gray matter time series as a nuisance variable because any BOLD signal change in those structures was likely due to pervasive rather than cortical activity . Finally , mean gray matter time series ( i . e . the global signal ) was also regressed to address spatially pervasive artefacts , such as respiration . There still remains considerable controversy regarding the utility of mean signal de-noising strategies ( Power et al . , 2017; Yang et al . , 2016b ) , with clear pros/cons . While there are several emerging approaches in the literature that attenuate and/or remove sources of global artefacts in BOLD data ( Glasser et al . , 2018 ) , the field-wide gold-standard approach still uses a univariate framework for removing variance from each grayordinate’s time series by computing the mean across grayordinates and regressing it from each grayordinate’s time course ( Power et al . , 2018 ) . GSR was performed using these standard procedures , explicitly excluding ventricles and white matter ( which are defined as separate nuisance regressors ) . The GS and its first derivative ( with respect to time ) were used as nuisance predictor terms within a multiple linear regression model along with other nuisance predictor terms ( ventricular signal , white matter signal , movement parameters , and the first derivatives of each of these , as noted above ) . Finally , all data were motion-scrubbed as recommended by Power et al . ( 2013 ) . As accomplished previously ( Anticevic et al . , 2012 ) , all image frames with possible movement-induced artual fluctuations in intensity were identified via two criteria: first , frames in which the sum of the displacement across all six rigid body movement correction parameters exceeded 0 . 5 mm ( assuming 50 mm cortical sphere radius ) were identified . Second , root mean square ( RMS ) of differences in intensity between the current and preceding frame was computed across all voxels and divided by mean intensity . Frames in which normalized RMS exceeded 1 . 6 times the median across scans were identified . The frames flagged by either criterion , as well as the one frame preceding and two frames following each flagged frame , were marked for exclusion ( logical or ) . Subjects with more than 50% frames flagged were completely excluded from all analyses . All the included subjects in the final samples passed these criteria . Most connectivity studies focus on pre-defined areas ( i . e . seed-based approaches ) . Such approaches assume ‘dysconnectivity’ across similar regions or networks . However , functional dysconnectivity induced by LSD , especially across heterogeneous associative cortical circuits , may exhibit variability across people . To address this , here we applied recently optimized neuroimaging analytic techniques to identify dysconnectivity in a data-driven fashion , termed global brain connectivity ( GBC ) ( Anticevic et al . , 2013; Anticevic et al . , 2014b; Cole et al . , 2011 ) . GBC is a measure that examines connectivity from a given grayordinate ( or area ) to all other voxelgrayordinates ( or areas ) simultaneously by computing average connectivity strength – thereby producing an unbiased approach as to the location of dysconnectivity . Also , unlike typical seed approaches , GBC involves one statistical test per grayordinates ( or area ) rather than one test per grayordinate-to-grayordinate pairing , substantially reducing multiple comparisons . These improvements dramatically increase the chances of identifying pharmacologically-induced dysconnectivity , or individual differences correlated with symptoms , as we demonstrated by our prior studies conducted in clinical populations ( Cole et al . , 2010; Anticevic et al . , 2013; Cole et al . , 2011 ) . By extension , this approach can be readily applied to pharmacological neuroimaging studies . Specifically , the GBC approach ( Cole et al . , 2010; Cole et al . , 2011 ) was applied using in-house Matlab tools studies ( Anticevic et al . , 2013; Anticevic et al . , 2014b; Cole et al . , 2011 ) , extended across all grayordinates in the brain , as defined via the CIFTI image space , which was obtained via an adapted version of FreeSurfer software fine-tuned by the HCP pipelines ( Fischl et al . , 2002 ) . Finally , for each grayordinate in the CIFTI image space , we computed a correlation with every other whole-brain grayordinate , transformed the correlations to Fisher z-values , and finally computed their mean . This calculation yielded a GBC map for each subject where each grayordinates value represents the mean connectivity of that grayordinate with all other grayordinates in the brain . We also verified that differences in variance of BOLD signals did not drive our GBC results , as predicted by our prior computational modeling work ( Yang et al . , 2014 ) . To this end , we computed GBC using a non-normalized covariance measure , which did not alter effects . Appropriate whole-brain type I error correction was computed via FSL’s PALM tool ( see second - Level Group Comparisons below ) . To examine the thalamus coupling with all grayordinates in the brain in session one we computed a seed-based thalamus correlation and covariation map by extracting average time-series across all grayordinates in each subject's bilateral thalamus and then correlating/covarying these with each grayordinate . For details on this approach see ( Anticevic et al . , 2014a ) . Because of emerging findings suggesting that clinical populations exhibit elevated GS variability ( Yang et al . , 2014 ) , we separately examined results without GSR implemented . This demonstration is particularly important given recent reports suggesting that the GS may be abnormally altered in specific clinical populations ( Yang et al . , 2014; Gotts et al . , 2013 ) , but also that it may contain major elements of respiratory artefacts ( Power et al . , 2017 ) , which could influence GBC analyses . To obtain global signal ( GS ) beta values , we first performed GS regression ( GSR ) using standard widely adopted procedures ( Anticevic et al . , 2013; Cole et al . , 2011 ) . The GS timeseries for each subject was obtained by calculating mean raw BOLD signal averaged over all grayordinates for each time point , explicitly excluding ventricles and white matter signal . This GS timeseries was used as nuisance predictor term within a multiple linear regression model . More formally , we used the following multiple regression analysis:BOLDkraw ( t ) =b0+∑i=1nbiXi+ BOLDkpreprocessed ( t ) , where BOLDkraw ( t ) represents the raw BOLD signal in grayordinate k as a function of time , t . b0 is the intercept , Xi represents the ith nuisance ( e . g . GS at that time point ) , bi is the corresponding beta weight computed for regressor Xi . The last term is the residual signal that is not accounted for by the regressors . In other words , the residual represents the preprocessed BOLD signal at grayordinate k . In our model the regressor of interest is GS ( t ) . BOLDkrawt=b0+ bGSGSt+ BOLDkpreprocessedt , The GS beta weights reported are represented by the bGS values obtained from this multiple regression . GS ( t ) is the spatial average of time-varying BOLD signal across all gray matter grayordinates:GSt=∑kmBOLDk ( t ) m The ‘mean GS beta weight’ computation in Figure 3 is done by fitting a generalized linear model ( GLM ) to each grayordinate’s BOLD time series to obtain the GS beta weight ( bGS ) as shown above . In that sense , the grayordinate-wise whole-brain map of GS beta weights is more interpretable as a task-evoked GLM analysis than to a functional connectivity measure . In other words , GS beta weights are not functional connectivity values and should not be interpreted as such – instead they represent the amount of GS variance accounted for by that grayordinate for a given subject . This ‘GS beta map’ was then entered into a second level analysis as done for the functional connectivity dependent measures . This comparison tests the hypothesis that the spatial contribution to the GS is altered under LSD vs . Pla , as done in our prior work ( Yang et al . , 2016b ) . For quality assurance purposes we computed the following measures: ( i ) signal-to-noise ratio ( defined as mean signal over the entire BOLD time series for a given grayordinate divided by its standard deviation ) , and ( ii ) the percentage of ‘scrubbed’ images . In turn , we correlated these measures with mean Fz-connectivity with and without GSR for the first and second session in the LSD condition ( Figure 1—figure supplement 3 ) . All correlations were non-significant indicating that changes in GBC induced by LSD are not attributable to motion and image arts . GBC maps for each subject , condition , and session were entered into a 2 × 3 repeated-measures ANOVA and tested using FSL’s Permutation Analysis of Linear Models ( PALM , [Winkler et al . , 2014] ) . Threshold-free cluster enhancement ( TFCE ) was used to avoid the need to define clusters using arbitrary thresholds for cluster size . We report the default TFCE parameters that were used in the permutation , which are fully described in the PALM code ( https://github . com/andersonwinkler/PALM/blob/master/palm_defaults . m ) . The statistical images were thresholded at p<0 . 05 ( family-wise error corrected ) , with 10000 permutations . For further analysis connectivity strength ( Fz ) values for hyper- and hypo-connected areas ( based on the LSD vs . ( Ket+LSD ) +Pla contrast ) were averaged across grayordinates for each participant and condition . Furthermore , Fz values for grayordinates within seven functionally-defined networks using parcellations derived by Yeo et al . ( 2011 ) , Buckner et al . ( 2011 ) and Choi et al . ( 2012 ) were averaged for each participant and condition . All analyses were performed with and without GSR . Results were visualized using the Connectome Workbench software ( https://www . humanconnectome . org/software/connectome-workbench . html ) . The 5D-ASC comprises 94 items that are answered on visual analogue scales ( Dittrich et al . , 2006 ) . Scores were calculated for 11 recently validated scales ( Studerus et al . , 2010 ) : experience of unity , spiritual experience , blissful state , insightfulness , disembodiment , impaired control and cognition , anxiety , complex imagery , elementary imagery , audio-visual synesthesia , and changed meaning of percepts . The short version of the 5D-ASC includes the 45 items that comprise the spiritual experience , blissful state , disembodiment , elementary imagery , and changed meaning of percepts scales . 5D-ASC score was analyzed using a repeated-measures ANOVA with treatment condition ( Pla , LSD , and Ket+LSD ) and scale as within-subject factors . 5D-ASC short-version score was analyzed using a repeated-measures ANOVA with treatment condition ( Pla , LSD , and Ket+LSD ) , scale , and time ( 180 , 250 , and 360 min ) as within-subject factors . The 5D-ASC short-version scores of one participant could not be analyzed due to missing data at 360 min after administration . Bonferroni-corrected Pearson correlations were conducted to investigate the relationship between Fz values within the seven functionally defined networks at session two and subjective drug effects ( 5D-ASC short version at 250 min ) . To relate LSD-related neuroimaging effects to the cortical topography of gene expression for candidate receptors , we used the Allen Human Brain Atlas ( AHBA , RRID:SCR_007416 ) . The AHBA is a publicly available transcriptional atlas containing gene expression data , measured with DNA microarrays , that were sampled from hundreds of histologically validated neuroanatomical structures across six normal post-mortem human brains ( Hawrylycz et al . , 2012 ) . All reported analyses were performed on group-averaged gene expression maps in the left cortical hemisphere , which were generated following a previously reported procedure ( Burt et al . , 2018 ) . In brief , a group-averaged , dense cortical expression map was constructed through a neurobiologically informed approach using a surface-based Voronoi tessellation combined with a 180-area unilateral parcellation with the Human Connectome Project’s Multi-Modal Parcellation ( MMP1 . 0 ) ( Glasser et al . , 2016 ) ( Figure 9—figure supplement 1 ) .

The psychedelic drug LSD alters thinking and perception . Users can experience hallucinations , in which they , for example , see things that are not there . Colors , sounds and objects can appear distorted , and time can seem to speed up or slow down . These changes bear some resemblance to the changes in thinking and perception that occur in certain psychiatric disorders , such as schizophrenia . Studying how LSD affects the brain could thus offer insights into the mechanisms underlying these conditions . There is also evidence that LSD itself could help to reduce the symptoms of depression and anxiety disorders . Preller et al . have now used brain imaging to explore the effects of LSD on the brains of healthy volunteers . This revealed that LSD reduced communication among brain areas involved in planning and decision-making , but it increased communication between areas involved in sensation and movement . Volunteers whose brains showed the most communication between sensory and movement areas also reported the strongest effects of LSD on their thinking and perception . Preller et al . also found that another drug called Ketanserin prevented LSD from altering how different brain regions communicate . It also prevented LSD from inducing changes in thinking and perception . Ketanserin blocks a protein called the serotonin 2A receptor , which is activated by a brain chemical called serotonin that , amongst other roles , helps to regulate mood . By mapping the location of the gene that produces the serotonin 2A receptor , Preller et al . showed that the receptor is present in brain regions that show altered communication after LSD intake , therefore pinpointing the importance of this receptor in the effects of LSD . Psychiatric disorders that produce psychotic symptoms affect vast numbers of people worldwide . Further research into how LSD affects the brain could help us to better understand how such symptoms arise , and may also lead to the development of more effective treatments for a range of mental health conditions .

2018

Changes in global and thalamic brain connectivity in LSD-induced altered states of consciousness are attributable to the 5-HT2A receptor

Understanding the fate of exogenous cells after implantation is important for clinical applications . Preclinical studies allow imaging of cell location and survival . Labelling with nanoparticles enables high sensitivity detection , but cell division and cell death cause signal dilution and false positives . By contrast , genetic reporter signals are amplified by cell division . Here , we characterise lentivirus-based bi-cistronic reporter gene vectors and silica-coated gold nanorods ( GNRs ) as synergistic tools for cell labelling and tracking . Co-expression of the bioluminescence reporter luciferase and the optoacoustic reporter near-infrared fluorescent protein iRFP720 enabled cell tracking over time in mice . Multispectral optoacoustic tomography ( MSOT ) showed immediate biodistribution of GNR-labelled cells after intracardiac injection and successive clearance of GNRs ( day 1–15 ) with high resolution , while optoacoustic iRFP720 detection indicated tumour growth ( day 10–40 ) . This multimodal cell tracking approach could be applied widely for cancer and regenerative medicine research to monitor short- and long-term biodistribution , tumour formation and metastasis . Non-invasive optical imaging methods for preclinical in vivo research include bioluminescence ( BLI ) and fluorescence as well as photoacoustic/optoacoustic tomography , a technology that has only been developed recently ( Deliolanis et al . , 2014; Wang and Yao , 2016; Weber et al . , 2016 ) . These imaging modalities have enabled great progress in the tracking of labelled cells longitudinally over time in animal models of disease , which has become especially relevant for cancer research and cell-based regenerative medicine therapies ( de Almeida et al . , 2011; James and Gambhir , 2012; Sharkey et al . , 2016 ) . The resolution and sensitivity of optical imaging in animals is limited by autofluorescence , absorption and scattering of excitation and/or emission light , especially in deep tissues . The optimal window for in vivo optical imaging lies in the near infrared ( NIR ) spectrum ( ~650–900 nm ) , since absorption through the main endogenous chromophores ( oxy-haemoglobin , deoxy-haemoglobin , melanin , water and lipids ) are minimal in this spectral range ( Weber et al . , 2016 ) . For permanent cell labelling and tracking , genetic modification with reporter genes is the method of choice , although fluorescent tags and nanoparticles have been developed recently for sensitive short-term cell tracking over a period of a few cell divisions ( Comenge et al . , 2016; Dixon et al . , 2016 ) . Using luciferase reporter genes , bioluminescence constitutes the most sensitive optical modality due to its excellent signal-to-noise ratio , as light emission only occurs in the presence of a functional enzyme and its required co-factors . Firefly , luciferase has become the most widely used reporter as its substrates , D-luciferin or CycLuc1 ( Evans et al . , 2014 ) , are very well tolerated by animals and , compared to other luciferases , its peak light emission at around 562 nm is closest to the infrared window for in vivo imaging ( de Almeida et al . , 2011 ) . Although highly sensitive in vivo cell tracking via bioluminescence imaging of firefly luciferase is well established ( de Almeida et al . , 2011; Mezzanotte et al . , 2013 ) , this modality provides poor information about the spatial localisation of cells . Fluorescence has recently gained importance for animal imaging , since novel near-infrared fluorescent proteins ( iRFPs ) were developed from bacterial phytochrome photoreceptors ( Shcherbakova et al . , 2015; Shcherbakova and Verkhusha , 2013 ) . Similar to bioluminescence imaging , fluorescence only allows limited spatial resolution due to the high scattering coefficient of photons in tissues . On the other hand , photoacoustic imaging is based on the generation of ultrasound waves after absorption of light emitted by a pulsed laser . The sound waves are well transmitted in fluid media and less prone to scattering through tissues than emitted light . In fact , acoustic scattering is three orders of magnitude less than photon scattering ( Wang and Hu , 2012 ) , which overcomes deep tissue spatial resolution drawbacks of other optical-based imaging technologies . Interestingly , some iRFPs , such as iRFP720 , have an absorption profile in the NIR window , thus enabling their use as reporter genes for photoacoustic imaging , and allowing deep tissue imaging and tumour monitoring in mice ( Deliolanis et al . , 2014; Jiguet-Jiglaire et al . , 2014 ) . For example , new iRFPs have been proven to be useful genetic photoacoustic reporters in mammary gland and brain tumour monitoring , which establishes them as dual-modality imaging probes ( Deliolanis et al . , 2014; Filonov et al . , 2012; Krumholz et al . , 2014 ) . In addition , in multispectral optoacoustic tomography ( MSOT ) , a rapid multiwavelength excitation allows the distinction between different absorbers simultaneously after applying multispectral unmixing algorithms ( Tzoumas et al . , 2014 ) . Hence , a number of endogenous ( e . g . deoxy- and oxyhaemoglobin ) or exogenously introduced targets can be imaged in vivo via MSOT , including fluorescent proteins and nanoparticles . Since light absorption triggers the photoacoustic response , contrast agents with a high molar extinction coefficient facilitate their distinction from endogenous chromophores ( Lemaster and Jokerst , 2017 ) . In this regard , GNRs are an attractive choice since the extinction coefficients are in the range of 4–5 . 5 × 109 M−1cm−1 for longitudinal surface plasmon between 728 and 845 nm ( Orendorff and Murphy , 2006 ) , that is orders of magnitude higher than endogenous absorbers or other types of contrast agents ( e . g . AlexaFluor750 and iRFP720 coefficients are 2 . 5 × 105 and 9 . 6 × 104 M−1cm−1 , respectively [Shcherbakova and Verkhusha , 2013; Ntziachristos and Razansky , 2010] ) . We recently demonstrated how , after coating GNRs with a 35 nm silica shell , the optical properties of GNRs are maintained even in harsh environments such as the endosomal vesicles ( Comenge et al . , 2016 ) . This enabled very sensitive photoacoustic imaging of labelled cells , pushing the limits of detection into a range useful for biologically-relevant applications including cell tracking ( Wang and Jokerst , 2016 ) . However , cell division and cell death result in a decrease of GNR concentration and therefore a decrease of photoacoustic intensity . Thus , for long-term photoacoustic tracking of cells , adequate reporter genes are required . Here , we characterise newly generated lentivirus-based bi-cistronic reporter gene vectors and silica-coated GNRs as synergistic tools for co-labelling of cells and tracking in multiple imaging modalities at different stages after cell administration . The GNR-860 offered an excellent photoacoustic sensitivity that enabled detection of a very low number of cells . Reporter gene vectors mediated co-expression of the most sensitive bioluminescent reporter , firefly luciferase , and the most red-shifted near-infrared fluorescent protein , iRFP720 , which also acted as a photoacoustic reporter in MSOT . Whilst GNR-860 enabled early detection of cells by photoacoustic tomography after systemic injection , reporter genes were needed to monitor any potential tumour growth with the same technique . In addition , co-expression of luciferase allowed bioluminescence imaging . Although this modality does not offer the spatial resolution of photoacoustic tomography , it was used here to continuously monitor cell viability as well as for validating the photoacoustic results . The combination of the two cell labelling strategies ( nanoparticles and genetic labelling ) allowed short-term and long-term multimodal imaging of deep tissues with high spatial resolution without compromising sensitivity . It enabled the non-invasive monitoring of the initial cell biodistribution on a sub-organ level as well as the detection of tumour formation from very early stages , thus opening opportunities for gaining a better understanding of the safety of stem cell therapies , which is required for moving towards clinical translation ( Ankrum and Karp , 2010 ) . Two types of GNRs with distinct optical spectra were prepared following previously published protocols ( Nikoobakht and El-Sayed , 2003; Ye et al . , 2012 ) . The rationale for evaluating two types of gold nanorods in this study is as follows . First , detection of GNRs at different wavelengths would support the versatility of the approach , for example when combining them with other reporter molecules , that is it is not dependent upon preparing the exact same nanorods as reported here . Second , having two different probes provided us with a powerful control for false positive signals as explained in more detail below in the context of Figure 4 . The first batch consisted of GNRs with a longitudinal surface plasmon resonance ( LSPR ) peak at 710 nm ( GNR-710 ) . Their core size was 60 . 0 ± 7 . 5 nm by 25 . 7 ± 3 . 7 nm with an average aspect ratio of 2 . 3 ± 0 . 3 . The second batch had a LSPR peak at 860 nm ( GNR-860 ) . In this case , their size was 93 . 8 ± 9 . 3 nm by 25 . 2 ± 4 . 4 nm with an average aspect ratio of 3 . 8 ± 0 . 5 ) ( Figure 1a ) . Both GNRs were prepared to have LSPR bands in the NIR to favour their detection deep inside tissues . We have recently demonstrated that a 35 nm silica shell is needed to minimise plasmon coupling between GNRs after cell uptake ( Comenge et al . , 2016 ) . This enables the preservation of intrinsic optical properties resulting in an increased sensitivity for photoacoustic detection . Hence , both GNR batches were coated with a silica shell as previously described ( Comenge et al . , 2016 ) . Specifically , GNR-710 and GNRs-860 nm have a silica shell thickness of 38 . 7 ± 1 . 9 nm and 35 . 7 ± 1 . 6 nm , respectively ( Figure 1a ) . A murine MSC line was selected for labelling and tracking , since these cells have the capability to differentiate into mesenchymal derivatives , but can also form tumours occasionally ( Comenge et al . , 2016; Kuzma-Kuzniarska et al . , 2012 ) . The cells , which had first been transfected with a reporter gene vector ( see below ) , were incubated with different concentrations of GNR-710 and GNR-860 for 24 hr . To provide a better comparison between GNRs of different sizes , we standardize and report the GNR concentrations by the optical density ( O . D . ) of the peak , because absorption is what triggers the photoacoustic response . For reference , O . D . = 4 corresponds to ~100 pM GNR-710 and ~72 . 5 pM GNR-860 ( calculated by relating the absorbance at 400 nm of GNRs before silica coating to concentrations of molecular gold ) ( Pastoriza-Santos and Liz-Marzán , 2013 ) . To characterize their uptake , cells were incubated with GNRs at 0 . D . = 3 for 24 hr . TEM images of cells show GNRs localised inside vesicles , with silica shells separating the gold cores ( Figure 1b , and Figure 1—figure supplement 1 ) . In line with our previous report , the optical properties of GNRs were largely preserved after cell uptake ( Figure 1c , d ) . Preservation of the absorbance intensity and the shape of the plasmon band are prerequisite for an optimal photoacoustic detection of GNR-labelled cells since it relies on a spectral deconvolution to differentiate intrinsic absorbers from probes . To confirm that the cell labelling does not cause any toxicity , cell viability after exposure to a range of GNR concentrations up to O . D . = 4 was assessed ( Figure 1e ) . Only at the highest GNR concentration was a slight decrease in cell viability observed: 92 . 0 ± 2 . 6% and 91 . 7 ± 1 . 6% for GNR-710 and GNR-860 , respectively . These results are in agreement with our previous work , in which we showed that silica-coated GNRs did not affect cell viability , proliferation or differentiation potential of the mMSCs ( Comenge et al . , 2016 ) . To combine the sensitivity of bioluminescence imaging with the spatial resolution of MSOT , we generated bi-cistronic lentivirus vectors for expression of firefly luciferase and iRFP720 from the general eukaryotic EF1α promoter ( Figure 2a ) . We compared vectors containing either an internal ribosome entry site ( IRES ) ( Hellen and Sarnow , 2001 ) or a self-cleaving 2A element ( Szymczak and Vignali , 2005; Szymczak et al . , 2004 ) , to determine the most efficient method of translating the second open reading frame ( ORF; encoding luciferase ) from a single mRNA . Following transfection with lentivirus vectors , the mMSCs were selected by FACS for iRFP720 fluorescence , to obtain a pure IRES-vector and a pure E2A-vector transfected cell population , respectively , with similar levels of expression of the first ORF protein ( iRFP720 ) ( Figure 2b ) . Co-expression of iRFP720 and luciferase in individual cells was confirmed by immunocytochemistry ( Figure 2c ) , which also indicated that , in contrast to luciferase , iRFP720 was preferentially localised in the nucleus . To compare the levels of expression of luciferase from the IRES-vector and E2A-vector transfected cells , fluorescent and bioluminescent signals were quantified in cell suspensions using an IVIS Spectrum imager . For both types of reporter cells ( IRES-vector and E2A-vector transfected ) , the fluorescence signal intensity for iRFP720 was the same , but the E2A-vector-transfected cells showed a five times higher luciferase bioluminescence ( Figure 2d , e ) . These data indicate that in the context of our bi-cistronic lentivirus vectors , a superior translation efficiency of the second ORF from a single mRNA was achieved via an E2A element as compared to an IRES element . We therefore used the pHIV-iRFP720-E2A-Luc vector in all further experiments described herein . Proof of principle of the application of this vector for multi-modal imaging was tested in preliminary experiments , in which cells were injected subcutaneously and monitored with bioluminescence , fluorescence and photoacoustic imaging ( Figure 2—figure supplement 1 ) . Detection of labelled cells after subcutaneous injection has provided indications that gold nanorods ( our previous work [Comenge et al . , 2016] ) and genetic reporters ( Figure 2—figure supplement 1 ) could enable cell tracking . Similar proof of principle of the utility of nanoparticles as contrast agents for photoacoustic cell detection has often been obtained by imaging labelled cells implanted locally ( Jokerst et al . , 2012; Nam et al . , 2012 ) . Here , we address the much more challenging case of a systemic injection where cells spread through the body and the signal , therefore , becomes diluted . Specifically , 1 . 5 × 106 mMSCs transduced with the iRFP720-E2A-Luc lentivirus and labelled with GNRs were administered via ultrasound-guided intracardiac injection into the left ventricle . Left ventricle administration was chosen for cells to enter the arterial circulation rather than the venous . This allows cells to pass through all organs before returning to the right ventricle and subsequently lungs , where they become sequestered in the pulmonary microvasculature ( Kean et al . , 2013; Fischer et al . , 2009 ) . To provide a detailed assessment of cell localisation , we combined photoacoustic tomography imaging at 1 mm steps with bioluminescence detection . While bioluminescence imaging provides excellent sensitivity , its anatomical resolution is poor , mainly due to the high scattering of photons by tissue ( Patterson et al . , 1989; Ntziachristos et al . , 2005 ) . By contrast , photoacoustic imaging provides superior spatial resolution ( 150 µm in our system ) ( Ntziachristos and Razansky , 2010 ) . To demonstrate the potential of this multimodal approach to precisely determine the localisation of distributed nanorod-labelled cells , we show the correlation between bioluminescence imaging and MSOT imaging ( Figure 3 ) . Cell localisation in the head , liver and kidney regions was clearly observed with both imaging modalities ( Figure 3a , b ) , and the cell presence within organs was confirmed in the photoacoustic scans through the respective regions ( Figure 3c ) . This distribution is in agreement with other , different types of assessments in previous reports ( Basse et al . , 1988 ) . The presence of cells in the brain region is probably due to cell trapping in small capillaries , as reported also in other studies after intracardiac injection of cells ( Heyn et al . , 2006 ) . As shown in our parallel work ( Scarfe et al . , 2017 ) , the mMSCs remain in the lumen of the brain capillaries and do not cross the blood-brain barrier . It should be noted here that there was no difference in the bioluminescence signal biodistribution when cells with or without gold nanorods were injected , indicating that GNR uptake does not affect the overall behaviour of the cells ( Figure 3—figure supplement 1 ) . To further test the robustness and specificity of the signal , additional controls were performed . First , following the same imaging protocols , we observed a similar biodistribution pattern for mice injected with either GNR-710 or GNR-860-labelled cells ( Figure 3—figure supplement 1 , and Figure 3—figure supplement 2 ) . Second , when we applied the unmixing algorithm with the ‘wrong’ spectrum ( i . e . the GNR-710 spectrum to an animal injected with GNR-860-labelled cells , and vice versa ) , no GNR-specific signals were detected in photoacoustic images pre- and post-injection ( Figure 4 ) . Third , when the photoacoustic spectra of regions of interest before and after injection of cells were extracted , an increase in the photoacoustic intensity was only observed in the range of wavelengths in which corresponding GNRs have the higher absorbance , while intensity in the other wavelengths remained similar ( Figure 4—figure supplement 1 ) . It has to be noted that in some cases there were regions with an endogenous absorbance similar to GNRs ( e . g . food in the intestines has similar spectra as GNR-710 ) , which can lead to misinterpretation ( Figure 3—figure supplement 2 ) . For this reason , a scan was performed before cells were injected , which allowed detection of any potential endogenous interference with the GNR signal . Using a multimodal imaging approach ( in this case luminescence and photoacoustic ) also assisted in discounting any false positive signals . As expected , the photoacoustic signal provided by iRFP720 expression was not strong enough to detect the cells immediately after injection by means of photoacoustic imaging ( Figure 4 ) . One day after GNR-labelled cell infusion , only between 15–22% of the initial bioluminescence signal in whole mice persisted ( Figure 5b ) , decreasing to less than 4% by day 5 . Such substantial amounts of cell death following cell infusion were expected from previous reports ( Tögel et al . , 2008; Bentzon et al . , 2005 ) . A similar degree of cell death was observed with cells not labelled with GNRs , indicating that cell loss in vivo was not caused by GNR labelling ( Figure 5—source data 1 ) . Interestingly , after 24 hr , the bioluminescence signal and the photoacoustic signal for GNRs were no longer localised in the same regions/organs ( Figure 5a–c ) , indicating that the vast majority of GNRs followed a different fate than the remaining luciferase-expressing cells . Specifically , although the luminescence intensity had decreased substantially by 24 hr , the distribution of the signal remained the same ( Figure 5b ) . By contrast , the MSOT signal for GNR-860 increased dramatically in the liver over the same timescale , but was substantially reduced in regions such as the kidneys and head where it had been predominant immediately after cell injection ( Figure 5a , c ) . Whilst luminescent signal is based on living cells , photoacoustic signal can be produced from GNRs independently , irrespective of whether they are within cells , or if the cells are viable . Thus , these data suggest that the liver plays an important role in the clearance of GNRs and potentially associated cell debris following the massive loss of cells during the first 24 hr after injection , as previously reported ( Eggenhofer et al . , 2012; Khabbal et al . , 2015 ) . In addition , these findings are in line with reports describing nanoparticle clearance through the hepato-biliary tract ( Longmire et al . , 2008; Hirn et al . , 2011; Lankveld et al . , 2011 ) . We observed that the clearance could be continuously monitored in a longitudinal fashion since the photoacoustic signal in the liver dropped by 47% after 5 days and by 73% after 25 days ( Figure 5d ) . Following the substantial loss in cell viability over the first 24 hr , signal intensity declined further and by day 5 cells were no longer detectable by BLI in the regions where they had first accumulated ( i . e . brain , kidneys and spinal cord , Figure 3 ) . However , we now observed BLI signals from cells that had settled in different locations ( Figure 6—figure supplement 1 ) . The signal intensity increased over the following weeks as cells grew into tumours ( Figure 6—figure supplement 1 ) . Since the mMSC cell line had been reported to form osteoid structures in vivo ( Kuzma-Kuzniarska et al . , 2012 ) , and recent results from our group have revealed the capacity of the cells to form osteosarcomas ( Scarfe et al . , 2017 ) , we anticipated the possibility of tumour development . In other studies , this process has been typically monitored by bioluminescence imaging ( Jenkins et al . , 2005 ) , which allows the intensity of the signal to be correlated with the number of living cells . However , bioluminescence is limited by the poor spatial resolution at depths beyond 1 mm due to photon scattering . We therefore evaluated the potential of MSOT to determine the precise localisation of internal tumours as well as their size and shape . We used the constitutive expression of iRFP720 for the long-term monitoring of tumour growth by means of photoacoustic imaging . Presence of iRFP720 resulted in a change of the photoacoustic spectra in the regions where cells were forming tumours ( Figure 6 , and Figure 6—figure supplement 2 ) . The fact that the iRFP720 absorption spectrum is very different from the main endogenous absorbers ( a sharp band peaking at 690–700 nm , Figure 6—figure supplement 2 ) facilitated its detection even with minimal changes of photoacoustic intensity compared to the background . Mice developed tumours in different locations whether the cells were labelled with GNRs or not ( Figure 6 without GNRs and Figure 8 with GNRs ) . This process could be monitored longitudinally by bioluminescence and MSOT as shown in Figure 6 , where a number of tumours formed in a mouse 30 days after administration of cells labelled with the iRFP720-Luciferase vector . After scanning the animal in 1 mm steps , multispectral processing of the corresponding cross sections ( Figure 6b ) revealed the presence of tumours on the right shoulder ( 1 ) , spinal cord ( 2 ) , in the region dorsal of the kidney ( 2a and 2b ) , back ( 3 ) , hip ( 4 and 5 ) , and right leg ( 6 ) . Tumours 2a and 2b are very illustrative of the outcomes that can be achieved with this approach . From the in vivo bioluminescence images ( Figure 6a ) , it is difficult to determine the localisation of the tumours as either subcutaneous or in deep tissues , and with which organs they are associated . Analysis of photoacoustic images revealed that those tumours were growing in an unusual position close to and dorsal of the kidneys ( Figure 6b ) . Ex vivo luminescence imaging confirmed the observations made with in vivo photoacoustic tomography ( e . g . tumour masses 2a and 2b were found attached to the anterior-dorsal part of the kidneys ) ( Figure 6c ) . Expression of iRFP720 was also confirmed by ex vivo fluorescence imaging of these tumours ( Figure 6—figure supplement 3 ) . In addition to improving the anatomical localisation of internal tumours , photoacoustic tomography using iRFP720 expression also allowed precise longitudinal monitoring of tumour development from very early stages . The example in Figure 7 shows the growth of tumour 4 ( Figure 6 ) from day 13 to day 30 after cell injection . The excellent spatial resolution enabled the localisation of this tumour in the vicinity of the pelvic ilium from day 13 . The tumour mean diameter at this stage was 1 . 2 mm . By day 19 , 23 and 30 it had expanded to 2 . 0 , 2 . 2 and 3 . 3 mm , respectively . Furthermore , changes in the shape of the tumour were also monitored from day 23 onwards when it branched off into two ramifications , which grew attached to the main part of the tumour ( Figure 7 and Figure 7—video 1 ) . In a second example , the presence of several tumours was determined 40 days after injection of GNR-860 and reporter gene-labelled cells ( Figure 8 ) . Although GNRs were not used as contrast agents for tumour imaging , since the vast majority of them were cleared out following cell death , tumour growth was monitored with high spatial resolution via iRFP720 expression . The mouse shown in Figure 8 developed tumours in both shoulders , in the right dorsal kidney/adrenal gland region , towards the liver , and in several positions of the hip region . A late developing ( days 33–40 ) , but fast-growing tumour in the left shoulder was used to demonstrate the capabilities of iRFP720 MSOT imaging to obtain detailed spatial information in 3D reconstructions as shown in Figure 8c . Overall , our analysis reveals that MSOT provides an excellent method to monitor tumour growth longitudinally from very early stages onwards ( from sub mm size ) . Only some tumours growing in very peripheral positions of the animal could not be monitored due to technical or procedural limitations of the MSOT imaging system . For example , the nose could not be imaged as it was placed in the nose cone for air and anaesthesia supply during the imaging process . Also , since the animal was placed on its back , it was very difficult to avoid an air layer being trapped between the legs and the main body , which impaired sound propagation . Abundant use of ultrasound gel in this region minimised this effect and allowed data to be obtained as shown in Figure 6 ( tumour 6 ) , although the quality of the image was not as good as in other regions of the body . Any tumours in the lungs could not have been monitored with photoacoustic imaging due to a different sound propagation in air . However , the presence of tumours in the lungs could be ruled out from the bioluminescence images . We demonstrate here a multimodal imaging approach , which utilises a combination of GNRs and reporter genes to track cells immediately after systemic injection and longitudinally over time during the development of tumours . We show that multispectral optoacoustic tomography achieves high spatial resolution of the initial cell distribution through analysis of the GNR signal . Whilst expression of iRFP720 is initially undetectable , due to the scattered distribution of cells , it provides long-term tumour monitoring capability in MSOT as clustered iRFP720-positive cells lead to a stronger photoacoustic signal . Additionally , bioluminescence imaging of luciferase allowed for the most sensitive detection of live cells for comparison and validation of the MSOT results . The silica-coated GNRs provided the excellent photoacoustic sensitivity required to monitor the broad distribution of labelled cells shortly after systemic intra-cardiac injection . Because the GNRs maintained their optical signature when taken up by the cells , spectral unmixing could be applied in MSOT , which allowed GNR signals to be distinguished from other reporters ( e . g . iRFP720 ) and endogenous absorbers , resulting in highly specific cell detection . Over the medium term , that is several days post-injection , photoacoustic GNR signals and luciferase bioluminescence signals became separated , indicating nanoparticle clearance and surviving cells , respectively . The GNR signal became localised to the liver , which is in line with the known clearance pathway of nanoparticles and associated cell debris through the hepato-biliary system . Using the genetic reporters , longitudinal monitoring of surviving cells and tumour formation was achieved via luciferase bioluminescence and iRFP720 photoacoustic tomography . Compared to other optical-based imaging technologies , the superior deep-tissue spatial resolution of MSOT provided precise information about anatomical localisation and morphology of tumours from very early stages ( <1 mm diameter ) . Beyond the specific study of the biodistribution and tumour development of an exemplary mesenchymal cell line performed here , we emphasise the technological potential of these labelling tools for non-invasive in vivo imaging and cell tracking . In particular , research fields such as cell-based regenerative medicine therapies and cancer biology might find new applications using these labelling reagents and imaging approaches for cell monitoring and safety studies . The following chemicals were purchased from Sigma-Aldrich ( Gillingham , UK ) . HAuCl4 ·3H2O ( >99 . 0% ) , NaBH4 ( >99 . 99% ) , AgNO3 ( 99 . 0% ) , hexadecyltrimethylammonium bromide ( CTAB , >99% ) , L-ascorbic acid ( reagent grade ) , tetraethyl orthosilicate ( TEOS , 99 . 999% ) , O-[2- ( 3-Mercaptopropionylamino ) ethyl]-O′-methylpolyethylene glycol ( mPEG-SH , MW: 5000 Da ) , 5-bromosalicylic acid . Monocarboxy ( 1-mercaptoundec-11-yl ) hexaethylene glycol ( PEG-COOH , MW 526 . 73 Da ) was obtained from Prochimia ( Sopot , Poland ) . Dulbecco's modified eagle's medium ( DMEM ) , phosphate-buffered saline ( PBS ) , penicillin-streptomycin , biliverdin and polybrene were also obtained from Sigma Aldrich . Foetal bovine serum ( FBS ) was purchased from Life-Technologies . D-Luciferin was purchased from Promega ( Southampton , UK ) and CycLuc1 from Glixx Laboratories ( Southborough , USA ) . GNR synthesis were based on previously published protocols ( Nikoobakht and El-Sayed , 2003; Ye et al . , 2012 ) . First , seeds were prepared by adding 0 . 6 mL of ice-cold NaBH4 ( 0 . 01 M ) to a mixture of 5 mL CTAB ( 0 . 2 M ) and 5 mL of HAuCl4 ( 0 . 5 mM ) under vigorous stirring . The growth solution for GNR-710 was prepared by mixing 50 mL of CTAB ( 0 . 2M ) , 1 . 8 mL of AgNO3 ( 4 mM ) , 50 mL of HAuCl4 ( 1 mM ) and 0 . 8 mL of ascorbic acid ( 0 . 1 M ) . Finally , 0 . 2 mL of freshly synthesised seeds was added to the growth solution . The reaction was kept in a water bath at 28°C for 3 hr . The growth solution for GNR-860 was prepared by adding 440 mg of 5-bromosalicylic acid to 100 mL of CTAB ( 0 . 1 M ) and heat up the mixture to 60°C . After cooling down the solution to 30°C , 4 . 8 mL of AgNO3 ( 4 mM ) was added and left undisturbed for 15 min . Then , 50 mL of HAuCl4 ( 1 mM ) was added and left under low stirring for 15 min at 30°C . Afterwards , 256 µL of ascorbic acid was added under vigorous agitation for 30 s . Finally , 160 µL of seeds were added and the solution was kept at 28°C for 12 hr . Silica coating was performed as described in our previous work ( Comenge et al . , 2016 ) , which was based on slight modifications of a protocol by Fernández-López et al ( Fernández-López et al . , 2009 ) . GNRs were visualized using a Tecnai G3 Spirit transmission electron microscope ( TEM ) at 120 keV . Formvar/carbon-coated 200 mesh copper grid ( TAAB ) were dipped in a solution of the GNRs of interest and left to dry in air . More than 100 GNRs were considered for image analysis . Lentivirus vectors were constructed on the pHIV-Luciferase backbone ( a gift from Bryan Welm , Addgene plasmid # 21375 ) . The iRFP720 ORF was obtained from plasmid piRFP720-N1 ( a gift from Vladislav Verkhusha , Addgene plasmid # 45461 ) ( Shcherbakova and Verkhusha , 2013 ) and cloned as an EcoRI/XbaI fragment upstream of the IRES element of pHIV-Luciferase , resulting in the pHIV-iRFP720-IRES-Luc vector . To achieve a more efficient translation of the second ORF , the IRES element was replaced with an E2A peptide motif ( Szymczak and Vignali , 2005; Szymczak et al . , 2004 ) using a two-step PCR protocol to create the pHIV-iRFP720-E2A-Luc plasmid . The complete sequence of the E2A plasmid version has been submitted to the NCBI GenBank database ( accession number: MF693179 ) . Lentivirus particles were produced in HEK293T cells , which were purchased for this work from ATCC ( ATCC CRL-3216 ) and free of mycoplasma when tested with the e-Myco Mycoplasma PCR Detection Kit ( iNtRON Biotechnology , cat . no . 25235 ) . Cells were transfected via calcium phosphate co-transfection of the reporter gene plasmid , packaging plasmid psPAX2 and envelope plasmid pMD2 . G ( both a gift of Didier Trono , Addgene plasmids #12260 and #12259 , respectively ) as described ( Kutner et al . , 2009 ) . Virus-containing cell culture supernatant was collected three days post-transfection , centrifuged at 500 g and filtered through 0 . 45 µm pores . We chose the mouse MSC line D1 ORL UVA [D1] ( ATCC CRL-12424 ) for labelling and tracking , since these cells have the potential to differentiate into mesenchymal derivatives , but can occasionally also form tumours as previously shown ( Comenge et al . , 2016; Kuzma-Kuzniarska et al . , 2012 ) . The mMSC cell line was purchased for this work from ATCC and free of mycoplasma when tested with the e-Myco Mycoplasma PCR Detection Kit ( iNtRON Biotechnology , cat . no . 25235 ) . Furthermore , the mMSC cells were re-authenticated by Science Exchange – IDEXX BioResearch using short tandem repeat ( STR ) profiling . They were found to be of mouse origin and no other mammalian interspecies contamination was detected . They were grown in DMEM supplemented with 10% FBS and 2 mM L-Glutamine . Cells were transfected with purified reporter gene virus particles for 24 hr . iRFP720-expressing cells were sorted using a BD FACSAria III with red laser and APC-CyTM7 filter . Cells expressing high levels of iRFP720 were selected and maintained for all further experiments . Before IVIS imaging of cells , biliverdin was added to the culture medium for 24 hr to increase its concentration beyond levels present in FBS and to improve its uptake by cells for incorporation into iRFP720 as a chromophore . For analysis of reporter gene expression by immunofluorescence , mMSCs were grown on glass cover slips , fixed with 4% paraformaldehyde/PBS , washed with PBS , blocked and stained with an anti-firefly luciferase antibody ( Abcam , Cambridge , UK , ab21176 , diluted in 1:500 in PBS/10% donkey serum/0 . 25% Triton-X100 ) and a donkey-anti-rabbit-AlexaFluor488 secondary antibody ( Molecular Probes , A21206 ) . Fluorescence of iRFP720 was assessed directly using a Zeiss LSM510 Multiphoton microscope . GNR-labelling was performed as previously published ( Comenge et al . , 2016 ) . Briefly , cells were treated with cell medium containing GNRs at the concentration/optical density indicated in the Results ( always 79% cell medium , 20% GNRs in water , 1% penicillin-streptomycin ) . Unless otherwise stated , cells were labelled to a GNR O . D . = 3 . 0 . Then , cells were dissociated with trypsin , resuspended in fresh medium , washed twice with PBS , and counted using an automated cell counter ( TC10 , BioRad , Watford , UK ) . Cell viability was assessed with Cell Titer Glo ATP Assay ( Promega ) . Cells were labelled as described above by seeding 104 mMSCs cells in 96-well plates . Cells were incubated with GNRs-710 and GNR-860 at a final O . D . in media = 1 , 2 , 3 . 2 , and 4 ( 80% media and 20% GNRs solution in each case ) for 24 hr . After labelling , cells were washed three times with PBS . 50 μL of medium were added to each well and then 25 μL of the ATP reagent was added . The plate was mixed in an orbital shaker and , after 10 min , the contents of the plate were transferred to white , opaque , 96-well plates and the luminescence measured with a plate reader ( Fluostar Omega , BMG Labtech , Aylesbury , UK ) . Each condition was assessed in triplicate and results are given as %±SD relative to cells that were incubated without GNRs as described above . Cells were fixed with a solution containing 1% paraformaldehyde and 3% glutaraldehyde in 0 . 1 M cacodylate buffer ( pH 7 . 4 ) . Then , cells were incubated with a reduced osmium staining solution , containing 2% OsO4 and 1 . 5% K4[Fe ( CN ) 6] , for 1 hr . This was followed by a second 1 hr osmium staining ( 2% OsO4 ) step and overnight staining with 1% uranyl acetate . Cells were washed with water for 3 min , three times after every staining step . Samples were then dehydrated in graded ethanol ( 30% , 50% , 70% , 90% and 2 × 100% ) for 5 min each . Finally , samples were infiltrated with medium TAAB resin 812 and embedded within the same resin . The resin was cured for 48 hr at 60°C . Finally , ultrathin sections of 350 μm x 350 μm x 74 nm were cut and placed in 200-mesh Formvar/Carbon filmed grids . They were post-stained with uranyl acetate ( 4% UA in a 50:50 ethanol/water solution ) and Reynolds lead citrate before TEM imaging . 8–10 week-old female SCID hairless outbred ( SHO ) mice ( Charles River , Margate , UK ) were housed in individually ventilated cages at a 12 hr light/dark cycle , with ad libitum access to food and water . Experimental animal protocols were performed in accordance with the guidelines under the Animals ( Scientific Procedures ) Act 1986 ( licence PPL70/8741 ) and approved by the University of Liverpool Animal Welfare and Ethical Review Body . The tumour burden was monitored and kept within recommended limits in accordance with guidelines for the welfare and use of animals in cancer research ( Workman et al . , 2010 ) . Experiments are reported in line with the ARRIVE guidelines . These experiments aimed at evaluating the potential of MSOT to track cells over the short and long-term . The number of animals was chosen so that a range of tumour positions and sizes could be observed ( Table 1 ) . Image quantification is presented to demonstrate the information that can be extracted from this approach . In each case , the numbers correspond to the particular animal presented . The data for these exemplar animals as well as for the other animals are available in the data repository Zenodo ( Comenge et al . , 2017 ) . Transfected cells were labelled with GNR-710 and GNR-860 at O . D . = 2 . 4 as detailed above . 1 . 5 × 106 cells in 100 µL PBS were prepared for intracardiac injection as described below . All cell injection and imaging procedures of mice were carried out under general isoflurane/oxygen anaesthesia . The imaging routine at day 0 was composed of the following sequence of steps: 1 ) a baseline MSOT scan; 2 ) the ultrasound-guided ( Prospect imaging system , S-Sharp , New Taipei City , Taiwan ) intracardiac injection of cells into the left ventricle; 3 ) a second MSOT scan after 10 min for acclimatisation; 4 ) BLI , for which the mouse received an intraperitoneal injection of D-luciferin ( approximately 150 µg/g body weight ) 15 min before whole body imaging using an IVIS Spectrum system ( Perkin Elmer , Seer Green , UK ) . All data were analysed with Living Image ( Perkin Elmer ) and data are displayed in radiance units . On subsequent days , the MSOT scan and BLI were repeated in the same manner . For intracardiac injection of cells , mice were positioned supine on a heated platform . Fur around the chest area was removed using depilatory cream and limbs were taped down to keep the mouse position fixed . Ultrasound gel was applied liberally to the chest area and the ultrasound transducer ( Prospect imaging system , S-Sharp ) was positioned above the chest so the long axis view of the left ventricle was visible . 100 µl of cell suspension was drawn up into an insulin syringe ( 29 G ) and , using the ultrasound image as guidance , was inserted into the left ventricle of the heart . Cell suspension was then administered slowly over a period of approximately 30 s . For optoacoustic imaging , an MSOT inVision 256-TF small animal imaging system ( iThera Medical GmbH , Munich , Germany ) was used ( Morscher et al . , 2014 ) . After acclimatisation for 10 min inside the water bath , a whole-body scan was performed on the mouse with 1 mm steps and the following wavelengths for acquisition: 660 , 670 , 680 , 690 , 700 , 705 , 710 , 715 , 720 , 725 , 735 , 750 , 765 , 780 , 795 , 810 , 820 , 830 , 840 , 850 , 860 , 870 , 880 , 890 , 900 , 910 , 920 , 930 , 940 , 1025 , 1050 , 1075 , and 1100 nm . Heavy water was used in the water bath due to its low absorbance at wavelengths > 910 nm ( contrary to regular water ) . Linear-mode-based reconstruction and guided ICA multispectral processing were applied using viewMSOT software v3 . 6 ( iThera Medical GmbH ) .

Many scientists are studying the possibility of using human cells to treat diseases . For example , using stem cells to regenerate damaged body parts or genetically engineered immune cells to destroy cancer . Scientists need new tools to track what happens to these cells once they have been injected into a laboratory animal . This will help them understand how they work and make sure these potential treatments are safe . One concern with using cells as a treatment is that they might form cancerous tumors . To track these cells in a laboratory animal , scientists need two things: a way to distinguish the treatment cells from the animal’s normal cells and an imaging tool that allows them to see where the cells are in a living animal . One way to differentiate treatment cells from normal cells is to genetically engineer them to make a fluorescent protein called iRFP720 . Another way is to fill the cells with gold nanorods . Both , the fluorescent protein and the gold nanorods , absorb light in the infrared range . Scientists can use a technique called multispectral optoacoustic tomography , which transforms infrared light into ultrasound signals to create an image , to see where these markers are in the body . Now , Comenge et al . showed that the gold nanorods and multispectral optoacoustic tomography track the cells immediately after injection into the blood stream of a mouse . Most of the injected cells die within a few days , and the nanorods are progressively eliminated from the body through the liver . But some of the injected cells live on , multiply , and form tumors within a month . This was expected because the cells they used were chosen for their ability to sometimes form tumors . Using multispectral optoacoustic tomography to track the cells making iRFP720 , Comenge et al . were able to see exactly where the tumors are deep inside the body . Together , gold nanorods and iRFP720 could allow scientists to track where the cell-based therapies for cancer or other diseases go in the short and long term . This may help them prove whether these treatments work , and whether they have harmful effects . Comenge et al . are helping other scientists to use these techniques by distributing their tool for making iRFP720-producing cells .

2018

Multimodal cell tracking from systemic administration to tumour growth by combining gold nanorods and reporter genes

Acquisition of pluripotency is driven largely at the transcriptional level by activators OCT4 , SOX2 , and NANOG that must in turn cooperate with diverse coactivators to execute stem cell-specific gene expression programs . Using a biochemically defined in vitro transcription system that mediates OCT4/SOX2 and coactivator-dependent transcription of the Nanog gene , we report the purification and identification of the dyskerin ( DKC1 ) ribonucleoprotein complex as an OCT4/SOX2 coactivator whose activity appears to be modulated by a subset of associated small nucleolar RNAs ( snoRNAs ) . The DKC1 complex occupies enhancers and regulates the expression of key pluripotency genes critical for self-renewal in embryonic stem ( ES ) cells . Depletion of DKC1 in fibroblasts significantly decreased the efficiency of induced pluripotent stem ( iPS ) cell generation . This study thus reveals an unanticipated transcriptional role of the DKC1 complex in stem cell maintenance and somatic cell reprogramming . The acquisition of pluripotency in the epiblast , a transient population of cells with unrestricted developmental potential during early embryogenesis , is controlled by a core set of transcription factors that include OCT4 , SOX2 and NANOG ( Nichols et al . , 1998; Avilion et al . , 2003; Chambers et al . , 2003; Mitsui et al . , 2003; Silva et al . , 2009 ) . This undifferentiated , pristine stem state can be captured as embryonic stem ( ES ) cells ( Evans and Kaufman , 1981; Martin , 1981; Brook and Gardner , 1997 ) , regenerated from somatic cells by cell fusion and nuclear transfer ( Yamanaka and Blau , 2010 ) , or by the ectopic expression of defined transcription factors ( Takahashi and Yamanaka , 2006; Yu et al . , 2007 ) . These reprogrammed pluripotent cells display a transcriptome that is highly similar to ES cells . Not surprisingly , OCT4 , SOX2 and NANOG also play key roles in the maintenance of pluripotency in ES cells and its reacquisition in induced pluripotent stem ( iPS ) cells by targeting a common set of genes that underpin the pluripotent state ( Boyer et al . , 2005; Loh et al . , 2006; Chen et al . , 2008 ) . However , execution of these complex stem cell-specific gene expression programs also require a growing list of co-regulators including enhancer binding transcription factors ( KLF4 ( Jiang et al . , 2008 ) , SALL4 ( Wu et al . , 2006; Zhang et al . , 2006 ) , ESRRB ( Zhang et al . , 2008; Festuccia et al . , 2012 ) ) , coactivators ( Mediator ( Chia et al . , 2010; Kagey et al . , 2010 ) , YAP ( Lian et al . , 2010 ) , TAFs/TFIID ( Fong et al . , 2011; Liu et al . , 2011; Pijnappel et al . , 2013 ) ) , chromatin remodelers ( esBAF ( Ho et al . , 2009 ) ) , and histone modifiers ( p300/CBP ( Chen et al . , 2008 ) , the trithorax histone methyltransferase ( Ang et al . , 2011 ) ) . Perhaps the involvement of this rather elaborate collection of cofactors arose from the need for ES cells to significantly expand their transcriptional repertoire in order to accommodate the wide range of gene expression responses governing self-renewal and the transition into diverse differentiated cell-types ( Fong et al . , 2012 ) . Intriguingly , recent studies have implicated additional cofactors that have not been traditionally associated with transcriptional regulation , such as the XPC DNA repair complex ( Fong et al . , 2011 ) , as well as microRNAs and long non-coding RNAs as part of the pluripotency regulatory network ( Wilusz et al . , 2009; Orkin and Hochedlinger , 2011; Jia et al . , 2013 ) . Unconventional transcriptional coactivators like the XPC complex and YAP are often found to be multifunctional . For example , the XPC complex safeguards genome integrity of self-renewing stem cells as well as their differentiated progeny by scanning the genome for DNA damage and initiating excision repair ( de Laat et al . , 1999; Riedl et al . , 2003; Sugasawa , 2011 ) , while YAP controls the expansion of stem cells by sensing diffusible signals and external cues in the niche ( Lian et al . , 2010; Dupont et al . , 2011; Schlegelmilch et al . , 2011; Mori et al . , 2014 ) . It therefore seems reasonable to speculate that co-opting these protein complexes into performing gene regulatory functions may represent a prevalent evolutionary strategy that allows rapidly dividing stem cells to expand and enhance the pluripotency network while coping with the enormous pressure to maintain genome stability and cellular homeostasis . Indeed , coactivators like the XPC complex and YAP are highly enriched in ES and iPS cells perhaps because they are performing double duty ( Ramalho-Santos et al . , 2002; Lian et al . , 2010; Fong et al . , 2011 ) . Not surprisingly , depletion of these multifaceted complexes compromises pluripotency gene expression , stem cell maintenance , and somatic cell reprogramming ( Lian et al . , 2010; Fong et al . , 2011 ) . Therefore , it appears that a critical threshold level of these coactivators may be required for a stem cell to maintain its pluripotency . Likewise , high levels of these cofactors may be necessary to establish an appropriate gene regulatory environment for a somatic cell to re-enter the cell cycle and become responsive to transcription factor-mediated reprogramming . Somatic cell reprogramming by a small cadre of specific transcription factors is thought to be a stochastic and inefficient process where only a small fraction ( 0 . 1–1% ) of somatic cells become iPS cells ( Buganim et al . , 2013 ) . However , recent data suggests that the induction of these rare , reprogramming-permissive somatic cells is not entirely a random event but may depend in part on cell-intrinsic determinants that are somehow restricted to a privileged few ( Guo et al . , 2014 ) . These privileged somatic cells exhibit ultrafast cell duplication and express higher levels of proteins involved in DNA repair , RNA processing and cell cycle control ( Guo et al . , 2014 ) . It is thought that these enrichments are required to fuel the rapid cellular proliferation necessary to overcome some major bottlenecks in reprogramming ( Banito et al . , 2009; Hanna et al . , 2009; Hong et al . , 2009; Utikal et al . , 2009; Ruiz et al . , 2011 ) . Another roadblock to cellular reprogramming is the requisite early reactivation of a robust transcriptional circuitry governed by OCT4 and SOX2 ( Buganim et al . , 2012 ) . Although this process can be enhanced by a number of transcription factors , reprogramming efficiency remains stubbornly low . It seems likely , therefore , that some key components of reprogramming remain undiscovered and there is a need to better define the molecular mechanisms by which OCT4 and SOX2 activate a stem cell-specific transcriptional program in ES and iPS cells . To directly screen in an unbiased manner for cofactor requirements that support OCT4 and SOX2 mediated activation , we developed an in vitro transcription assay that faithfully recapitulates OCT4/SOX2 and coactivator-dependent gene activation observed in ES cells using purified components to reconstitute the human transcriptional apparatus ( Fong et al . , 2011 ) . Deploying this sensitive biochemical complementation assay , we recently detected two additional stem cell coactivators ( SCC-A and -B ) that , in concert with the XPC coactivator complex , co-dependently stimulate the transcriptional activation of the Nanog gene by OCT4 and SOX2 . Here we report that SCC-A activity is delivered by a subset of the dyskerin ribonucleoprotein complexes ( DKC1 RNPs ) . We examined the specific activity of the various endogenous DKC1 RNPs assembled with distinct small nucleolar RNAs ( snoRNAs ) by in vitro transcription . Furthermore , we combined promoter occupancy data with pluripotency gene expression profiles from loss-of-function studies to directly link the DKC1 complex to transcriptional coactivator function in ES cells . In addition to its well-documented role in regulating the proliferative capacity of stem cells , our studies unveil a previously unrecognized direct role of non-coding snoRNAs and the DKC1 complex in regulating transcription initiation with important implications for understanding the cell-intrinsic determinants conducive to cellular reprogramming . We previously have shown an activity present in a partially purified protein fraction , Q0 . 3 , that is required for the XPC coactivator complex to stimulate a full , synergistic activation of the human Nanog proximal promoter by OCT4 and SOX2 but is dispensable for basal or Sp1-activated transcription ( Rodda et al . , 2005; Fong et al . , 2011 ) . Q0 . 3 separated from the XPC complex at the Poros-HQ anion exchange chromatographic step ( Figure 1A , B ) . Although Q0 . 3 appeared to migrate as a single activity on a size exclusion column with an apparent molecular mass ( Mr ) of ∼500 kDa ( Figure 1C ) , this coactivator activity splits again into two distinct chromatographic fractions on a Poros-Heparin ( Poros-HE ) cation exchanger . One cofactor ( SCC-B ) eluted at ∼0 . 4 M KCl whereas the second activity ( SCC-A ) eluted at ∼0 . 6 M KCl ( Figure 1D ) . Taken together , it appears that at least three distinct stem cell coactivators ( one being the XPC complex ) are required to generate a full , OCT4/SOX2-dependent transcriptional response . Starting with nuclear extracts prepared from 400 L of a pluripotent embryonal carcinoma ( EC ) cell line NTERA-2 ( NT2 ) , we used the reconstituted transcription system supplemented with recombinant XPC complex , purified OCT4 , SOX2 , and a modified human Nanog template , to purify SCC-A over six successive chromatographic columns resulting in >30 , 000-fold increase in specific activity ( Figure 1A ) . Silver staining of the peak Poros-HE purified fractions revealed a distinct pattern of four major polypeptides that consistently co-purified with SCC-A activity ( Figure 1E ) . For the remainder of this report , we focused on the identification and functional characterization of SCC-A in vitro and in vivo . 10 . 7554/eLife . 03573 . 003Figure 1 . Purification of Stem Cell Coactivator-A ( SCC-A ) required for OCT4/SOX2-dependent activation of the Nanog gene . ( A ) Chromatography scheme for purification of Q0 . 3 from NT2 nuclear extracts ( NT2 NE ) . NT2 NE is first subjected to ammonium sulfate precipitation ( 55% saturation ) followed by a series of chromatographic columns including cation exchangers phosphocellulose ( P11 ) , heparin ( Poros-HE ) , the anion exchanger Poros-HQ , hydroxyapatite ( HAP ) , and gel filtration medium Superose 6 . ( B ) Input ( IN , Ni-NTA flowthrough ) , buffer control ( − ) and fractions containing Q0 . 3 eluted from a Poros-HQ anion exchanger ( fraction number indicated ) are assayed in the presence of OCT4 , SOX2 , and recombinant XPC complex in in vitro transcription assays . ( C ) Q0 . 3 appears to migrate as a single activity . Superose 6 fractions are assayed as in ( B ) . Mobilities of peak activity ( 400–600 K ) and gel filtration protein standards are shown at bottom . ( D ) Q0 . 3 is composed of two distinct coactivator activities , SCC-A and SCC-B . Transcription reactions contain buffer control ( − ) , Poros-HE fractions and are assayed as in ( B ) . SCC-A activity elutes in fractions 39–43 . ( E ) Silver-stained 10% Bis-Tris polyacrylamide gel of the active SCC-A fractions . Filled arrowheads indicate polypeptides that co-migrate with SCC-A activity . The bottom panel shows the same fractions separated on a 12% SDS-PAGE gel to show the smallest subunit of SCC-A . Insulin added to Poros-HE fractions as a protein stabilizer is indicated by asterisk . DOI: http://dx . doi . org/10 . 7554/eLife . 03573 . 00310 . 7554/eLife . 03573 . 004Figure 1—figure supplement 1 . The DKC1 and the XPC coactivator complexes are highly enriched in the transcriptionally active phosphocellulose 1 M KCl ( P1M ) and Ni-NTA flowthrough ( Ni-FT ) fractions . Comparative western blot analysis of NT2 nuclear extract ( NE ) , phosphocellulose 0 . 3 M , 0 . 5 M , 1 M KCl fractions ( P0 . 3 , P0 . 5 , P1M , respectively ) , and Ni-NTA flowthrough ( Ni-FT ) using antibodies against DKC1 , GAR1 , NHP2 , NOP10 , XPC , and RAD23B . Each lane contains 5 μg of protein as determined by Bradford protein assay ( Bio-Rad ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03573 . 004 To identify the polypeptides comprising the SCC-A complex , peak Poros-HE fractions were pooled , concentrated and separated by SDS-PAGE . Tryptic digestion of the four excised gel bands followed by mass spectrometry analysis revealed SCC-A to be the dyskerin ( DKC1 ) complex comprised of DKC1 , GAR1 , NHP2 , and NOP10 subunits ( Figure 2A ) ( Meier , 2005 ) . Identification of the DKC1 complex as the active constituent of SCC-A activity was unexpected because it has not been previously linked to transcription . To corroborate the mass spectrometry data , we carried out western blot analysis to track the early chromatographic behavior of the three stem cell coactivators on a phosphocellulose column ( Figure 1A ) . Consistent with our previous observation that the bulk of OCT4/SOX2 coactivator activity resides in the 1 M KCl fraction ( P1M ) ( Fong et al . , 2011 ) , core subunits of the DKC1 and XPC complexes ( and SCC-B , data not shown ) were highly enriched in P1M compared to total nuclear extract or the transcriptionally inactive 0 . 3 and 0 . 5 M fractions ( Figure 1—figure supplement 1 ) . 10 . 7554/eLife . 03573 . 005Figure 2 . SCC-A is the dyskerin ( DKC1 ) complex . ( A ) Silver stained SDS-PAGE gel of Poros-HE peak activity fraction in Figure 1E with protein identities determined by mass spectrometry analysis . ( B ) Silver stained SDS-PAGE gel of recombinant DKC1 complex reconstituted in insect Sf9 cells infected with baculoviruses expressing His6-tagged DKC1 , untagged GAR1 , FLAG-tagged NHP2 , and untagged NOP10 . ( C ) Silver stained SDS-PAGE gel of recombinant partial and holo DKC1 as well as NAF1-containing intermediate complexes reconstituted in E . coli cells expressing various epitope-tagged subunits and untagged DKC1 as denoted . A prominent partial fragment of DKC1 co-purifies extensively with the full length DKC1 complexes . ( D ) Recombinant DKC1 complexes enhance OCT4/SOX2-activated transcription of Nanog . Buffer control ( − ) , bacterial DKC1-NOP10 heterodimer ( lanes 2 and 3 ) , DKC1-NHP2-NOP10 trimer ( lanes 4 and 5 ) , NAF1-DKC1-NHP2-NOP10 intermediate ( lanes 6 and 7 ) , and holo DKC1-GAR1-NHP2-NOP10 ( lanes 8 and 9 ) , recombinant holo DKC1 complex purified from Sf9 cells ( lanes 10 and 11 ) , and endogenous holo-complex from NT2 ( Poros-HE peak activity fraction 40 , lanes 12 and 13 ) are assayed over a twofold concentration range . Transcription reactions contain OCT4 , SOX2 , recombinant XPC complex , and a Poros-HE fraction containing SCC-B . DOI: http://dx . doi . org/10 . 7554/eLife . 03573 . 00510 . 7554/eLife . 03573 . 006Figure 2—figure supplement 1 . 5’ end radiolabeling of RNAs co-purified from recombinant DKC1 complexes . ( A ) Schematics of RNA labeling procedure . RNAs are treated with tobacco acid pyrophosphatase ( TAP ) to remove 5′ cap and dephosphorylated with APex heat-labile alkaline phosphatase before being radiolabeled by T4 polynucleotide kinase ( T4 PNK ) . ( B ) Recombinant DKC1 complexes reconstituted in E . coli lack detectable co-purifying RNAs . 5′ end radiolabeled RNAs from DKC1 complexes purified from Sf9 and E . coli cells are separated on a 6% urea-polyacrylamide gel . Size markers ( M ) are in nucleotides . DOI: http://dx . doi . org/10 . 7554/eLife . 03573 . 006 The DKC1 complex is an evolutionarily conserved , four-subunit protein complex that interacts with a large heterogeneous class of small non-coding RNAs called H/ACA small nucleolar RNAs ( snoRNAs ) ( Meier , 2005; Terns and Terns , 2006 ) . The assembly of a DKC1 RNP in vivo follows an elaborate , multi-step process mediated by the protein chaperones SHQ1 and NAF1 ( Darzacq et al . , 2006; Grozdanov et al . , 2009 ) . The GAR1 subunit subsequently replaces NAF1 in the intermediate complex containing NAF1 , DKC1 , NHP2 , and NOP10 to form the mature RNP only after snoRNAs are incorporated and properly processed ( Kiss et al . , 2010 ) . These H/ACA snoRNAs guide sequence-specific pseudouridylation of ribosomal RNAs ( rRNAs ) and spliceosomal small nuclear RNAs ( snRNAs ) by the catalytic subunit DKC1 ( Liang and Li , 2011 ) . The DKC1 complex also plays a key role in the biogenesis of telomerase by binding and promoting the processing and intranuclear trafficking of telomerase RNA ( TERC ) ( Egan and Collins , 2012 ) . Given the intimate association of the DKC1 complex with numerous RNAs and the multiple factors required to assemble the RNP in vivo , it is remarkable that an RNA-free , ternary ‘apo-complex’ can be generated in vitro . Indeed , several crystal structures of the archeal and yeast partial and holo-complexes of DKC1 revealed direct protein–protein contacts among the four subunits independent of RNA ( Li and Ye , 2006; Li et al . , 2011 ) . To firmly establish that the DKC1 complex rather than some trace contaminants present in the purified SCC-A fractions was responsible for the coactivator activity detected in our in vitro transcription assays , we reconstituted the human DKC1 complex from recombinant gene products expressed in insect ( Sf9 ) and bacterial cells . Using a combination of conventional chromatography and affinity purification procedures , we were able to efficiently purify the recombinant DKC1 complex from Sf9 cells to near homogeneity ( Figure 2B ) . It is important to point out that a significant amount of Sf9 RNAs co-purified with the human DKC1 complex as determined by 5′ end radiolabeling of the purified RNA species ( Figure 2—figure supplement 1 ) . This suggests that the biogenesis pathway and machinery for DKC1 RNP assembly are at least partly conserved between human and Sf9 cells . To determine whether specific snoRNAs are required for DKC1 coactivator function , we next attempted to reconstitute the DKC1 complex in Escherichia coli . However , the lack of dedicated chaperones ( i . e . SHQ1 and NAF1 ) and accessory factors for the assembly of a DKC1 complex in E . coli made this task rather challenging , resulting in low yields after purification . Nonetheless , the holo-DKC1 complex isolated from E . coli showed similar subunit stoichiometry compared to DKC1 complexes purified from NT2 cells ( Figure 2A , C ) . More importantly , it did not appear to contain detectable amounts of any associated RNAs ( Figure 2—figure supplement 1B ) . To examine the contribution of individual subunits of the DKC1 complex in supporting OCT4/SOX2-activated transcription , we attempted to express and purify each of them individually in E . coli . However , all four gene products were either insoluble or remained tightly associated with bacterial heat shock proteins , suggesting that the individual protein subunits were not properly folded ( data not shown ) . This is consistent with recent structural and functional analyses of free GAR1 and NOP10 showing that they are largely unfolded proteins ( Hamma et al . , 2005; Li et al . , 2011 ) . To circumvent this problem , we reconstituted partial complexes representing the different assembly intermediates during the biogenesis of DKC1 complexes in vivo in order to address the minimal protein subunit requirement for coactivator function ( Figure 2C ) . These hetero-dimeric ( DKC1-NOP10 ) , -trimeric ( DKC1-NHP2-NOP10 ) , ternary ( NAF1-DKC1-NHP2-NOP10 ) , and holo-DKC1 complexes were tested for their ability to potentiate OCT4/SOX2-dependent transcriptional activation of Nanog in vitro . Remarkably , all partial and complete recombinant complexes whether produced in E . coli or Sf9 cells exhibited similar specific activities for coactivation , but were reproducibly less active than the purified native endogenous DKC1 complex from NT2 cells ( Figure 2D ) . It was not clear whether the reduced specific activities of the recombinant purified complexes resulted from poorly folded or assembled subunits , presence of inhibitory RNAs , or both . Nevertheless , these results using purified recombinant subunits confirm that at least the protein components of the DKC1 complex represent a major contributor of the SCC-A coactivator function . Indeed , it appears that the largest subunit DKC1 and the smallest protein NOP10 are sufficient to provide the bulk of the transcriptional coactivator function and that an RNA component may not be strictly required for this moonlighting activity of the DKC1 complex . Although snoRNAs may not be essential for conferring coactivator competence to the recombinant DKC1 complexes , we note that the endogenous DKC1 complexes are twofold to threefold more active than their recombinant counterparts , suggesting that some mammalian snoRNAs may play a role in enhancing the transcriptional activity of the DKC1 complex . The DKC1 RNPs in mammalian cells are highly heterogeneous—with more than 100 known H/ACA snoRNAs that form an equally large number of distinct RNPs by associating with the same four core protein subunits , some of which with unknown functions ( i . e . orphan snoRNAs that lack base complementarity to rRNAs or snRNAs ) ( Kiss et al . , 2010 ) . New classes of snoRNAs have also been identified and shown to directly participate in disparate cellular processes from pre-mRNA splicing to chromatin decondensation ( Jady et al . , 2012; Schubert et al . , 2012; Yin et al . , 2012 ) . Furthermore , as much as 60% of snoRNAs can be processed into microRNAs ( miRNAs ) , most of which have unknown targets ( Ender et al . , 2008; Taft et al . , 2009 ) . Thus , our understanding of the full repertoire of H/ACA snoRNAs and their ‘non-canonical’ functions remains limited . It is also unclear if the binding of different snoRNAs to the human DKC1 complex induces structural changes or masks protein surfaces that may positively or negatively impact coactivator function . Given that most , if not all , GAR1-containing DKC1 complexes are mature RNPs in vivo ( Kiss et al . , 2006 ) , it seemed prudent for us to examine the range and specific activity of these native but heterogeneous mixtures of human DKC1 RNPs . Even though these 100 or more DKC1 RNPs have highly similar if not identical core protein composition and architecture , we reasoned that these RNPs are likely to display distinct chromatographic properties due to their unique snoRNAs and/or associated factors . In an attempt to biochemically fractionate this heterogeneous population of DKC1 RNPs , a partially purified fraction prepared from 200 L of NT2 cells that contains >95% of the total population of human DKC1 RNPs ( Ni-FT; Figure 1—figure supplement 1 ) was applied to a Poros-HQ anion exchange column and fractionated using a salt gradient ( Figure 3A ) . As expected , DKC1 RNPs were found to elute in a broad profile from 0 . 3 to 0 . 9 M KCl with the majority of the complexes eluting at ∼ 0 . 5 M ( Figure 3B ) , consistent with extensive heterogeneity of the DKC1 RNPs in NT2 cells . We next immuno-affinity purified the various DKC1 RNPs from different salt eluted Poros-HQ fractions using a monoclonal antibody against human DKC1 followed by peptide elution . The various affinity-purified DKC1 RNP pools all contain stoichiometric amounts of the four core protein subunits indicating that they are likely mature RNPs ( Figure 3C ) . However , we failed to detect any other major associated polypeptides in these purified samples . Therefore , differences in protein composition alone are unlikely to fully account for the observed chromatographic heterogeneity of the DKC1 RNPs separated by the salt gradient on a Poros-HQ column . Instead , we strongly suspect the differential chromatographic behavior of the endogenous human DKC1 RNP complexes to derive from association with distinct RNA species . Indeed , 5′ end radiolabeling of the purified RNA species from the various affinity-purified DKC1 RNP preparations revealed distinct patterns of associated RNAs ( Figure 3D ) . The DKC1 RNPs purified from high salt eluted fractions ( # 22 , 26 , and 30 ) were enriched for longer RNAs ( >180 nucleotides ) and some select shorter RNAs ( <100 nucleotides ) . The 130–140 nucleotide-long snoRNA clusters were recovered from DKC1 immunoprecipitates from multiple fractions spanning a wide spectrum of the salt gradient . Thus , it appeared that parameters in addition to RNA length may contribute to the observed differential chromatographic properties of different DKC1 RNPs . Of note , the DKC1 RNPs purified from fraction 9 did not appear to contain significant amounts of RNA ( Figure 3D ) . This is unexpected because the presence of GAR1 usually signifies that some RNA species should have been loaded into the complex in the normal course of DKC1 RNP assembly . However , we cannot exclude the possibilities that , although unlikely , RNAs were present but somehow refractory to labeling at both 5′ ( Figure 3D ) and 3′ ends ( data not shown ) . It remains possible that some snoRNAs were degraded or had dissociated from a small fraction of the DKC1 RNPs during purification . 10 . 7554/eLife . 03573 . 007Figure 3 . DKC1-associated small RNAs modulate transcriptional coactivator activity . ( A ) Purification scheme of endogenous DKC1 ribonucleoprotein complexes ( RNPs ) . A partially purified fraction ( Ni-FT ) containing the bulk of the DKC1 RNPs in NT2 cells ( See Figure 1—figure supplement 1 ) is fractionated over a Poros-HQ anion exchange column followed by affinity purification using a monoclonal antibody against DKC1 and peptide elution . ( B ) Extensive heterogeneity of the endogenous DKC1 RNPs from NT2 cells . Western blotting of input ( IN ) , flowthrough ( FT ) , and various salt-eluted Poros-HQ fractions using antibodies against DKC1 , NHP2 , and NOP10 . Filled inverted triangles indicate fraction numbers used for affinity purification . Salt concentrations ( [K+] in M ) of selected fractions are indicated . ( C ) Silver stained SDS-PAGE gel of the DKC1 RNPs affinity-purified from indicated Poros-HQ fractions . A proteolytic fragment of DKC1 is denoted by asterisk . ( D ) 5′ end labeling of RNAs isolated from affinity-purified DKC1 RNPs from indicated Poros-HQ fractions . Radiolabeled RNAs were separated on a 6% denaturing urea-polyacrylamide gel . Size markers are in nucleotides . ( E ) Buffer control ( − ) or affinity-purified DKC1 RNPs from salt-eluted Poros-HQ fractions over a threefold concentration range are assayed using in vitro transcription . Reactions contain OCT4 , SOX2 , recombinant XPC complex , a Poros-HE fraction containing SCC-B . DOI: http://dx . doi . org/10 . 7554/eLife . 03573 . 007 These highly purified pools of DKC1 RNPs were assayed over a threefold dose–response range in our fully reconstituted in vitro transcription reactions containing OCT4 , SOX2 , recombinant XPC complex and SCC-B . Remarkably , DKC1 RNPs purified from higher salt eluted Poros-HQ fractions ( fractions 26 and 30 ) displayed significantly higher specific activity than those from lower salt fractions ( fractions 9 and 14 ) ( Figure 3E ) . We estimated a ∼sixfold enhancement in the specific activities of DKC1 RNPs purified from fraction 30 compared to fraction 9 , which , as we had shown in Figure 3D , contained no detectable RNAs ( Figure 3E , compare lanes 3 and 13 ) . It is unclear if this endogenous apo-complex lacking any detectable RNA component is physiologically relevant or an experimental artifact . However , the fact that this apo-complex activated transcription with reduced specific activity ( Figure 3E , compare lanes 1 and 3 ) is consistent with our previous observation that the bacterial apo-complex is less active than DKC1 RNPs purified from NT2 cells in supporting transcription ( Figure 2D ) . Paradoxically , recombinant DKC1 RNPs purified from Sf9 cells contained insect snoRNAs ( Figure 2—figure supplement 1 ) but exhibited low specific activities similar to the bacterial and apo-complexes suggesting that some RNAs may be inhibitory . Taken together , these results uncover a previously unrecognized gene regulatory role of the DKC1 RNP complex wherein a subset of mammalian non-coding snoRNAs may enhance the DKC1 coactivator function while other RNAs may inhibit its transcription activity . Identification of the DKC1 RNP and the XPC DNA repair complexes as co-dependent coactivators for OCT4/SOX2 was unexpected on two fronts . These two multi-subunit protein assemblies had not been previously implicated in directing stem cell-specific transcription nor had they been functionally linked to each other in any cellular processes . We therefore set out to determine the functional relationship between these newly identified stem cell coactivators and their mechanisms of coactivation in vitro and in vivo . Our ability to recombinantly express and purify these coactivators ( including purified SCC-B which will be the subject of a future study ) allowed us to systematically test the contribution of each coactivator alone in supporting OCT4/SOX2-activated transcription in vitro . Addition of individual coactivator complexes only marginally activated Nanog transcription ( Figure 4A ) . However , when the DKC1 complex was supplemented with the XPC complex , we observed a noticeable increase in transcriptional output that was substantially further enhanced by adding the third coactivator , SCC-B ( Figure 4A ) . These results confirmed the co-dependent nature of these three coactivators in supporting an optimal , synergistic activation of the Nanog gene by OCT4 and SOX2 . To further explore the mechanism by which the DKC1 complex cooperates with the XPC complex in OCT4/SOX2 activated transcription , we co-expressed both complexes along with ( or without ) OCT4 and SOX2 in 293T cells and performed co-immunoprecipitation assays to probe for a potential interaction between these two coactivators . Immunoprecipitation of the XPC complex pulled down the DKC1 complex both in the presence and absence of the activators . This finding suggests that the DKC1 complex may function as an OCT4/SOX2 coactivator in part through a direct physical interaction with the XPC complex , which in turn binds OCT4 and SOX2 ( Figure 4B ) . In support of this observation , a recent global proteomic study using large scale biochemical fractionation of human cell extracts to isolate stable protein complexes identified WDR79 , a known accessory protein of the mature DKC1 RNP ( Tycowski et al . , 2009; Jady et al . , 2012 ) , as a candidate XPC-interacting protein ( Havugimana et al . , 2012 ) . Whether the DKC1 complex also forms direct contacts with OCT4 and SOX2 in the absence of XPC is unclear . Our attempt to address this was hampered by the fact that we could not express any of the four subunits of the DKC1 complex to a significant level in 293T or several other cell lines ( data not shown ) . However , the fact that co-expression of OCT4/SOX2 did not increase the amount of DKC1 pulled down by the XPC complex argues against a stable tripartite complex wherein the coactivators interact with each other and form independent contacts with the activators . 10 . 7554/eLife . 03573 . 008Figure 4 . Mechanism of Coactivation by the DKC1 Complex . ( A ) Co-dependent activation of Nanog transcription by the XPC complex , the DKC1 complex , and SCC-B . Recombinant XPC and DKC1 complexes purified from Sf9 cells , and SCC-B purified from bacteria were added individually ( lanes 2–4 ) , or in various combinations ( lanes 5–7 ) to in vitro transcription reactions containing OCT4 and SOX2 . Strongest synergistic activation is observed when all three coactivators are present in the transcription reaction ( lane 7 ) . ( B ) DKC1 interacts with the XPC complex independent of OCT4 and SOX2 in vivo . Control ( − ) , plasmids expressing mouse XPC complex ( XPC ) , mouse DKC1 complex ( DKC1 ) , and STEMCCA were co-transfected into 293T cells . Cell lysates are immunoprecipitated with anti-RAD23B antibody . Input extracts ( 2% ) and RAD23B-bound proteins were analyzed by western blotting . ( C ) Schematic diagrams showing the two structural domains in DKC1 ( TruB and PUA ) and mutations in DKC1 and NOP10 selected for functional analyses in ( D ) . All mutations except D125A are identified in patients with dyskeratosis congenita ( DC ) . ( D ) Wild type , pseudouridine synthase inactive ( D125A ) , and DC mutant DKC1 complexes ( Dkc1 A353V , Dkc1 L37del , Dkc1 Δ22C , and Nop10 R34W ) are reconstituted in Sf9 cells and assayed over a threefold concentration range in in vitro transcription reactions containing OCT4 , SOX2 , recombinant XPC complex , and SCC-B . DOI: http://dx . doi . org/10 . 7554/eLife . 03573 . 00810 . 7554/eLife . 03573 . 009Figure 4—figure supplement 1 . Micrococcal nuclease ( MNase ) -treated recombinant DKC1 complexes remain structurally intact . Recombinant wild-type ( WT ) and various mutant DKC1 complexes are mock treated ( − ) or digested extensively with MNase ( + ) , and washed extensively to remove any dissociated RNAs prior to FLAG peptide elution . Eluted protein complexes are analyzed by Coomassie Blue staining . DOI: http://dx . doi . org/10 . 7554/eLife . 03573 . 00910 . 7554/eLife . 03573 . 010Figure 4—figure supplement 2 . DKC1-associated RNAs in recombinant DKC1 complexes are resistant to extensive MNase digestion . RNAs co-purified from mock and MNase-treated recombinant DKC1 complexes in Figure 4—figure supplement 1 are 5′ end radiolabeled and separated on a 6% urea-polyacrylamide gel as described in Figure 2—figure supplement 1 . Note that the prominently labeled 130–140 nucleotide ( nt ) -long RNA clusters are resistant to complete nuclease digestion . It appears that these RNAs are cut on average once by MNase to generate two new smaller clusters ( 80–90 nt and 30–55 nt ) that remain stably associated with the DKC1 complexes . Increasing the amount of MNase and/or nuclease digestion time did not change the patterns or disrupt the integrity of the protein complexes ( data not shown ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03573 . 01010 . 7554/eLife . 03573 . 011Figure 4—figure supplement 3 . MNase digestion moderately increases DKC1 coactivator activity . Mock ( − ) or MNase-treated ( + ) WT and Nop10 R34W DKC1 complexes are assayed in in vitro transcription reactions ( over a fourfold concentration range ) supplemented with OCT4 , SOX2 , recombinant XPC complex and SCC-B . DOI: http://dx . doi . org/10 . 7554/eLife . 03573 . 011 Mutations in the Dkc1 , Nhp2 , and Nop10 genes have been linked to dyskeratosis congenita ( DC ) , a rare but fatal human genetic disorder that impairs stem cell function and proliferation generally attributed to defects in telomerase or ribosome biogenesis ( Mitchell et al . , 1999; Mason and Bessler , 2011 ) . Our discovery of a stem cell-specific transcriptional role of the DKC1 complex adds a potentially important alternative mechanism for interpreting the molecular basis of DC disease phenotypes . However , it was unclear if amino acid residues critical for telomerase and ribosome biogenesis impinge on distinct or overlapping domains with respect to our newly uncovered DKC1 transcription coactivator function . To begin to address this potentially important link to disease , we focused on DC mutations in the large DKC1 subunit and the small NOP10 protein because a partial complex of these two subunits was sufficient to activate Nanog transcription in vitro ( Figure 2D ) . We recombinantly expressed and purified a panel of mutant DKC1 complexes in Sf9 cells that are representative of both position ( L37del ( Heiss et al . , 1998 ) , A353V ( Knight et al . , 1999 ) , Δ22C ( Vulliamy et al . , 1999 ) ) and frequency ( A353V ) at which DC mutations occur in Dkc1 ( Marrone et al . , 2005 ) . We also generated an artificial , pseudouridine synthase inactive mutant DKC1 ( D125A ( Gu et al . , 2013 ) ) as well as a mutant NOP10 containing ( R34W ( Walne et al . , 2007 ) ) complex ( Figure 4C; Figure 4—figure supplement 1 ) . Remarkably , all mutant DKC1 RNPs , whether they were mock or nuclease-treated to partially remove the associated Sf9 small RNAs ( Figure 4—figure supplement 2 ) , were consistently more active than the WT holo-complex in potentiating OCT4/SOX2-mediated transcription ( Figure 4D; Figure 4—figure supplement 3 ) . Therefore , it appeared that neither the enzymatic activity nor amino acids mutated in DC are essential for coactivator activity although the enhanced coactivator phenotype could lead to changes in gene expression and altered stem cell function . The transcriptional phenotypes of these DKC1 mutations are highly reminiscent of our findings with the XPC complex in that disease-relevant amino acids and domains critical for DNA repair functions were also dispensable for OCT4/SOX2-activated transcription ( Fong et al . , 2011 ) . To further probe the molecular mechanisms by which the DKC1 complex might function as a transcriptional coactivator for OCT4 and SOX2 in ES cells , we performed chromatin immunoprecipitation ( ChIP ) assays to investigate whether the DKC1 complex is directly recruited to regulatory regions of key OCT4/SOX2-target genes . We found that efficient crosslinking of DKC1 to the Oct4 enhancer by formaldehyde requires the pre-treatment of ES cells with a protein–protein cross-linker ( ethylene glycol bis[succinimidylsuccinate] or EGS ) ( Figure 5—figure supplement 1 ) . ChIP-qPCR analysis revealed that sites of DKC1 occupancy at the Oct4 , Nanog , Sox2 genes indeed coincide with those of OCT4 ( Boyer et al . , 2005; Loh et al . , 2006 ) and SOX2 binding to enhancer and promoter DNA sequences ( Figure 5—figure supplement 2 ) in the mouse ES cell line D3 ( Figure 5A ) . Importantly , DKC1 binding is also enriched at the enhancers of Oct4 and Nanog in human ES cell line H9 ( Figure 5B ) and EC cell line NT2 ( Figure 5C ) , thus confirming the generality of a co-recruitment mechanism to transcriptional regulatory elements in pluripotent stem cells . Curiously , we failed to detect a significant enrichment of DKC1 at some OCT4/SOX2-target genes such as Fgf4 in D3 cells ( Figure 5A ) . This suggests that the DKC1 complex may be differentially employed by OCT4 and SOX2 to regulate a subset of their target genes . Additional experiments such as genome-wide analyses of DKC1 occupancy will be required to ascertain the extent to which DKC1 associates with OCT4/SOX2 target genes in mouse ES cells . Since over 90% of snoRNAs are embedded in the introns of coding and non-coding genes ( Filipowicz and Pogacic , 2002 ) , the DKC1 complex has also been found to localize at gene bodies where it is thought to co-transcriptionally process nascent snoRNAs ( Darzacq et al . , 2002; Ballarino et al . , 2005; Yang et al . , 2005 ) . Now our finding of the DKC1 complex co-occupying pluripotent gene promoters and enhancer elements with sequence-specific activators OCT4 and SOX2 in ES cells strongly suggests a classical coactivator function of the DKC1 complex rather than acting purely as a snoRNP maturation factor . 10 . 7554/eLife . 03573 . 012Figure 5 . The DKC1 complex is recruited to regulatory regions of key pluripotency genes in mouse and human ES cells . ( A ) Co-occupancy of DKC1 , OCT4 , and SOX2 on enhancers of Oct4 , Nanog , Sox2 , but not Fgf4 , in mouse ES cell line D3 . Chromatin immunoprecipitation ( ChIP ) analysis of DKC1 occupancy on control and enhancer regions of the Oct4 , Nanog , Sox2 , and Fgf4 gene loci . Representative data ( n = 3 ) showing the enrichment of DKC1 ( black bars ) compared to control IgGs ( white bars ) are analyzed by qPCR and expressed as percentage of input chromatin . Schematic diagrams of OCT4/SOX2 binding sites of each gene and the relative positions of the amplicons used to detect enriched ChIP fragments are shown at the bottom . Error bars represent standard deviation , n = 3 . Primer sequences can be found in Supplementary file 1 . ( B ) DKC1 is recruited to regulatory regions of Oct4 and Nanog in human ES cell line H9 . Representative ChIP data ( n = 3 ) are analyzed as described in ( A ) . Error bars represent standard deviation , n = 3 . ( C ) DKC1 is enriched on Oct4 promoter in human embryonal carcinoma cell line NT2 . Representative ChIP data ( n = 3 ) are analyzed as described in ( A ) . Error bars represent standard deviation , n = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 03573 . 01210 . 7554/eLife . 03573 . 013Figure 5—figure supplement 1 . Crosslinking DKC1 to chromatin requires protein–protein crosslinker ethylene glycol bis[succinimidylsuccinate] ( EGS ) in addition to formaldehyde ( FA ) . Mouse ES cell line D3 is crosslinked with FA for 5 min ( left ) , 10 min ( middle ) , or with EGS for 30 min followed by FA for 5 min ( right ) . ChIP analysis of DKC1 occupancy on control ( −251 ) and enhancer ( −2000 ) regions of the Oct4 gene locus . Enrichment of DKC1 ( black bars ) compared to control IgGs ( white bars ) are analyzed by qPCR and expressed as percentage of input chromatin . Schematic diagrams of OCT4/SOX2 binding sites of each gene and the relative positions of the amplicons used to detect enriched ChIP fragments are shown at the bottom . Error bars represent standard deviation , n = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 03573 . 01310 . 7554/eLife . 03573 . 014Figure 5—figure supplement 2 . SOX2 is enriched on the regulatory regions of Oct4 , Nanog , and Fgf4 in mouse ES cells . Mouse ES cell line D3 is crosslinked with EGS and FA . Enrichment of SOX2 compared to control IgGs is analyzed as described in Figure 5—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 03573 . 014 Many transcriptional activators ( OCT4 , SOX2 , NANOG ) and coactivators ( Mediator , TAFs/TFIID , the XPC complex ) critical for stem cell pluripotency are often highly enriched in ES cells but become rapidly down-regulated upon differentiation . Dynamic regulation of these transcription factors in ES cells is thought to confer not only stability to the transcriptional circuitry governing self-renewal but also the flexibility to exit the pluripotent state and switch between competing developmental programs during differentiation ( Jaenisch and Young , 2008; Liu et al . , 2011; Fong et al . , 2012 ) . Consistent with the notion that the DKC1 complex is performing as a stem cell-specific coactivator in ES cells , the DKC1 , GAR1 , and NOP10 subunits are highly enriched in pluripotent D3 cells ( Figure 6A ) . Their levels in ES cells decreased rapidly upon retinoic acid ( RA ) -induced differentiation while general transcription factor TFIIB and loading control β-Actin remained unchanged . Importantly , the selective decrease of DKC1 levels was not simply a reflection of a reduced proliferative state or protein translational activity in differentiating ES cells because components of the C/D snoRNP ( FBL and NOP58 ) , another major machinery involved in the ribosome biogenesis pathway , stayed largely constant ( Su et al . , 2013 ) . Indeed , it has been shown that transcription of the Dkc1 gene is regulated by OCT4 and NANOG in ES and iPS cells ( Agarwal et al . , 2010 ) , thus providing a transcriptional mechanism whereby Dkc1 expression levels are tightly coupled to the pluripotent state . 10 . 7554/eLife . 03573 . 015Figure 6 . The DKC1 complex is required for stem cell maintenance . ( A ) Downregulation of the DKC1 complex upon retinoic acid ( RA ) -induced differentiation of mouse ES cell line D3 . Western blot analyses of whole cell extracts prepared from D3 cells ( mESC D3 WCE ) collected at indicated days post LIF withdrawal and RA treatment using antibodies against the DKC1 complex ( DKC1 , GAR1 , and NOP10 ) , XPC , OCT4 , the NOP58/fibrillarin ( FBL ) complex , TFIIB , and β-actin as loading control . ( B ) shRNA-mediated knockdown of the DKC1 complex in mouse ES cells . Whole cell extracts of mouse D3 cells infected with control non-target ( NT ) lentiviruses or with lentiviruses targeting XPC ( shXPC ) or DKC1 ( shDKC1-1 and shDKC1-2 ) are analyzed by western blotting . MOI = 25 . Asterisk denotes non-specific signals . ( C ) ES cell colony morphology and alkaline phosphatase ( AP ) activity are maintained in control non-target shRNA infected D3 cells ( NT ) , but are compromised in XPC ( shXPC ) and DKC1 depleted cells using two independent shRNAs ( shDKC1-1 and shDKC1-2 ) . ( D ) DKC1 and/or XPC depletion in ES cells compromised pluripotency gene expression . Quantification of Nanog , Oct4 , Sox2 , Klf4 , and Fgf4 mRNA levels in single and double knockdown of XPC and DKC1 in D3 cells are analyzed by qPCR and normalized to β-actin ( Actb ) . For double knockdown experiments , a cumulative MOI = 50 is used . Data from representative experiments are shown . Error bars represent standard deviation ( n = 3 ) . ( E ) DKC1 and/or XPC depletion in ES cells induces spontaneous differentiation towards primitive ectoderm and trophectoderm . Quantification of mRNA levels of primitive ectoderm marker Fgf5 , mesoderm marker T , primitive endoderm marker Gata6 , and extraembryonic trophectoderm marker Cdx2 in single and double knockdown of XPC and DKC1 in D3 cells are analyzed as in ( D ) . Primer sequences can be found in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 03573 . 01510 . 7554/eLife . 03573 . 016Figure 6—figure supplement 1 . shRNA-mediated knockdown of DKC1 in mouse ES cells does not compromise housekeeping gene expression . Quantification of housekeeping gene Gapdh mRNA levels in control ( NT ) , XPC , and DKC1 knockdown D3 cells are analyzed as in ( D ) . Error bars represent standard deviation ( n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03573 . 016 To gain additional in vivo evidence that the DKC1 complex is required for the proper expression of genes critical for stem cell self-renewal , we performed loss-of-function studies using lentiviruses expressing two independent short hairpin RNAs ( shRNAs ) specifically targeting DKC1 in mouse D3 ES cells ( Figure 6B ) . We also depleted XPC in D3 cells using a previously characterized shRNA ( Fong et al . , 2011 ) to investigate potential functional interactions between these two coactivator complexes . Interestingly , knockdown of DKC1 but not XPC resulted in co-depletion of the small NOP10 subunit indicating that the stability of individual subunits likely depends on the integrity of the DKC1 complex ( Figure 6B ) . This may also explain why a decrease in protein levels of GAR1 and NOP10 during RA-induced ES cell differentiation follows the same kinetics as DKC1 even though Gar1 and Nop10 do not appear to be direct targets of OCT4 and SOX2 ( Figure 6A ) . Compared to control knockdown D3 cells , shRNA-mediated silencing of XPC ( shXPC ) or DKC1 ( shDKC1-1 and shDKC1-2 ) resulted in pronounced morphological abnormalities including rapid collapse of the tightly packed ES cell colonies and appearance of large , flattened cells with concurrent dramatic reductions in alkaline phosphatase activity , all indicative of enhanced spontaneous differentiation of ES cells ( Figure 6C ) . At this point , we cannot rule out the possibility that the severe phenotype observed in DKC1 knockdown ES cells is at least partially contributed by disruption of other well documented DKC1-dependent cellular processes ( telomerase function and ribosome biogenesis ) in addition to its transcription coactivator function . However , mouse ES cells lacking telomerase activity ( Terc −/− ( Niida et al . , 1998 ) ) or carrying a pathogenic mutation in Dkc1 ( A353V ( Mochizuki et al . , 2004 ) ) can be maintained in culture for over 300 population doublings with no observable impact on growth rate and only a very mild effect on ribosome biogenesis . Since the self-renewal defects observed in DKC1 knockdown ES cells became apparent by 3 days post lentiviral infection ( <9 population doublings ) , cellular senescence or a gross defect in rRNA processing are unlikely to be major contributors to the DKC1 knockdown phenotypes we observed in ES cells . Consistent with the evident morphological changes associated with compromised stem cell identity , single knockdown of XPC or DKC1 in D3 cells resulted in a significant reduction in mRNA levels of core pluripotency genes including Nanog , Oct4 , Sox2 , Klf4 , as well as stem cell marker Fgf4 ( Figure 6D ) , while housekeeping gene Gapdh remain stable ( Figure 6—figure supplement 1 ) . Interestingly , simultaneous knockdown of XPC and DKC1 did not further reduce their expression . This is consistent with the co-dependent nature of the DKC1 and XPC complexes in gene activation wherein the absence of one coactivator severely limited the ability of the other two stem cell coactivators to stimulate Nanog transcription in vitro ( Figure 4A ) . To further explore the spontaneous differentiation phenotype in DKC1 and XPC-deficient ES cells , we performed qPCR analyses to monitor the expression level of lineage-specific markers representing the three germ layers and the trophectoderm . Depletion of DKC1 or XPC upregulated the expression of neuroectodermal maker Fgf5 and trophoblast-specifier Cdx2 at the expense of Gata6 , a primitive endoderm marker , while mesodermal marker T remained unchanged ( Figure 6E ) . Double knockdown of DKC1 and XPC appeared to further augment the expression of Cdx2 but not Fgf5 . The observed differentiation bias in DKC1 and XPC knockdown ES cells may be in part due to the reduced levels of OCT4 and NANOG , both of which have well-documented functions in antagonizing differentiation of extraembryonic lineages including the trophectoderm ( Niwa et al . , 2000; Hay et al . , 2004; Hyslop et al . , 2005; Silva et al . , 2009 ) . The essential role of the DKC1 complex in establishing an OCT4/SOX2-dependent gene expression program in ES cells led us to hypothesize that DKC1 may be required for the reacquisition of pluripotency during cellular reprogramming by ectopic expression of OCT4 , SOX2 , KLF4 , and c-MYC ( Takahashi and Yamanaka , 2006 ) . Of note , recent studies showed that primary human adult fibroblasts ( HFs ) carrying pathogenic mutations in Dkc1 are refractory to cellular reprogramming ( Agarwal et al . , 2010; Batista et al . , 2011 ) . However , it is important to point out several key differences between using MEFs and HFs derived from DC patients to study DKC1 function in somatic cell reprogramming . Unlike MEFs which display high levels of telomerase activity and long telomeres ( >50 kb ( Blasco et al . , 1997 ) ) , HFs lack measurable TERT activity and have relatively short telomeres ( 10–15 kb ( Harley et al . , 1990 ) ) . In fact , telomerase null MEFs can be propagated in culture for more than 200 cell divisions without loss of viability ( Blasco et al . , 1997 ) , which make MEFs a potentially better cell culture system for studying telomerase-independent functions of DKC1 in reprogramming . By contrast , DC patient-specific fibroblasts have shorter telomeres and could also accumulate secondary mutations due to genome instability , which are both detrimental to the reprogramming process ( Fong et al . , 2013 ) . Consistent with this notion , it was shown that ectopic expression of wild type DKC1 ( or TERT ) in a DC mutant fibroblast line ( Dkc1 L37del ) failed to rescue the reprogramming defect phenotype ( Agarwal et al . , 2010 ) . Therefore , it remains unclear what impact , if any , acute DKC1 depletion in MEFs will have on iPS cell generation . To address this question , we infected MEFs with lentiviruses expressing non-targeting control shRNA or two independent shRNAs specific for DKC1 and initiated reprogramming by doxycycline ( dox ) -induced expression of OCT4 , KLF4 , SOX2 and c-MYC ( OKSM ) ( Sommer et al . , 2009 ) . We observed a marked decrease in the number of AP-positive iPS cell colonies ( ∼20–50-fold reduction ) whether or not we plated the induced DKC1 knockdown MEFs directly onto gelatin coated plates ( where the surrounding DKC1 knockdown MEFs refractory to reprogramming acted as feeder cells ) or onto mitomycin-treated feeder cells ( Figure 7A; Figure 7—figure supplement 1 ) . This suggests that failure of DKC1-deficient MEFs to acquire pluripotency is likely a cell autonomous phenomenon . Flow cytometry analysis showed that the majority of both control and DKC1 knockdown cells down-regulated fibroblast-associated cell surface marker THY1 indicating a loss of MEF identity ( Figure 7B ) . However , unlike control cells where many of them became SSEA1+ and ultimately gave rise to AP and NANOG-positive iPS cell colonies , most DKC1 knockdown cells do not ( Figure 7B , C ) . Because of the observed early arrest in reprogramming associated with DKC1-depleted MEFs , we next asked whether these cells were able to undergo the mesenchymal-to-epithelial transition ( MET ) , a requisite initiating event prior to expression of SSEA1 antigen ( Li et al . , 2010; Samavarchi-Tehrani et al . , 2010; Golipour et al . , 2012; Polo et al . , 2012 ) . By day 14 post dox-induction , control knockdown MEFs showed reduced expression of fibroblast-enriched , pro-mesenchymal genes Slug and Snail , but their levels remained noticeably higher than those in ES cells ( Figure 7D ) . This is likely due to contaminating partially reprogrammed iPS cells and residual fibroblasts present in the induced cell culture ( Figure 7B ) . These non-target knockdown cells also acquired epithelial characteristics indicated by elevated levels of Ecad and Epcam ( Figure 7D ) , as expected , given that THY1-/SSEA1+ partially and fully reprogrammed iPS cells represent the bulk of these control cells ( Figure 7B ) . By contrast , depletion of DKC1 in MEFs blocked the reactivation of epithelial genes ( Ecad and Epcam ) without significantly perturbing the silencing of mesenchymal genes ( Figure 7D ) , thus effectively uncoupling the otherwise tightly coordinated MET induced by OKSM ( Liu et al . , 2013 ) . These data taken together suggest that DKC1 could be required for reprogrammed MEFs to acquire an epithelial identity during the critically important mesenchymal-to-epithelial transition . 10 . 7554/eLife . 03573 . 017Figure 7 . The DKC1 complex is required for mesenchymal-to-epithelial transition ( MET ) during somatic cell reprogramming . ( A ) Depletion of DKC1 blocks somatic cell reprogramming . CF-1 mouse embryonic fibroblasts ( MEFs ) are infected with lentiviruses expressing OCT4 , KLF4 , SOX2 , and c-MYC ( STEMCCA ) and reverse tetracycline-controlled transactivator ( rtTA ) together with control non-target shRNA ( NT ) or two independent shRNAs targeting DKC1 ( shDKC1-1 and shDKC1-2 ) . Infected MEFs are plated onto gelatin coated 24-well plates ( Experiment 1 ) or 24-well plates containing mitomycin-treated feeder MEFs ( Experiment 2 ) ; cellular reprogramming is initiated by the addition of doxycycline ( dox ) . Cells are stained for AP activity and counted after 14 days ( 11 days with dox followed by 3 days without dox ) post induction ( dpi ) . ( B ) Single cell suspensions of 14 dpi CF-1 MEFs as described in ( A ) are stained with anti-mouse SSEA-1 and anti-THY-1 antibodies and analyzed by flow cytometry . ( C ) Representative confocal images of NANOG stained colonies 17 dpi ( 14 days with dox followed by 3 days without dox ) as described in ( A ) . The same acquisition settings—excitation laser intensity , gain , and exposure time—were used for all NANOG images . Scale bar , 100 μm . ( D ) DKC1-depleted MEFs are arrested at the MET during iPS cell generation . Somatic cell reprogramming of CF-1 MEFs is performed as in ( A ) . Cells were collected at 14 dpi ( 11 days with dox followed by 3 days without dox ) . mRNA levels of Dkc1 , epithelial markers Ecad ( also known as Cdh1 ) and Epcam , and mesenchymal markers Slug and Snail are compared with that of uninduced MEFs and mouse ES cell line D3 by qPCR . Values are normalized to expression levels in control non-target knockdown samples . Error bars represent standard deviation ( n = 3 ) . Primer sequences can be found in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 03573 . 01710 . 7554/eLife . 03573 . 018Figure 7—figure supplement 1 . Somatic cell reprogramming is blocked by DKC1 depletion . Induced MEFs depleted of DKC1 by two independent shRNAs ( shDKC1-1 and shDKC1-2 ) as described in Figure 7A are plated along with MEFs infected with control non-target lentiviruses onto 24-well plates at indicated cell numbers on feeders . AP-positive colonies are stained 14 days post induction ( dpi ) ( 11 days with dox followed by 3 days without dox ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03573 . 018 To address whether the early reprogramming arrest observed in DKC1-depleted MEFs can be attributed to a gross defect in cellular proliferation , we labeled control and DKC1 knockdown MEFs with a stable dye ( CFSE ) . The doubling time of these cells was measured by monitoring the decrease in dye intensity resulting from cell division over a 4 day period ( Figure 8A ) . Although a lengthening of doubling time of DKC1-depleted MEFs by both targeting shRNA hairpin was observed , compared to shDKC1-1 , knockdown by shDKC1-2 has a significantly more pronounced effect on iPS cell generation while having a minimal impact on cellular proliferation ( Figure 7—figure supplement 1; Figure 8A ) . This suggests that reprogramming efficiency does not strictly correlate with changes in proliferation rates caused by DKC1 depletion . However , we cannot exclude the possibility that differences in doubling rates could be due to differential off-target effects of the two hairpins . We also note that DKC1 depletion does not cause abrupt growth arrest of all MEFs but appears to selectively impair the proliferation of the faster cycling subpopulation without affecting the rest of the slower-dividing MEFs ( Figure 8A ) . Therefore , factors in addition to growth impairment are at least contributing to the observed defect in somatic cell reprogramming . These observations are also consistent with recent findings suggesting that essentially all of the reprogramming potential in OKSM-induced MEF cultures is confined to a small fraction ( 1–8% ) of cells characterized by an ultrafast cell cycle ( Smith et al . , 2010; Guo et al . , 2014 ) . Given the multiple functions of the DKC1 complex in regulating cellular proliferation ( Alawi et al . , 2011 ) , MET ( Figure 7D ) and pluripotency gene expression ( Figure 6D ) , we asked whether DKC1 might also be involved in overcoming barriers to deterministic cellular reprogramming . After 4 days of dox-induced expression of OKSM in MEFs , we labeled cells with CFSE and continued dox treatment for another 2 days before subjecting them to FACS ( Guo et al . , 2014 ) . We identified and characterized four distinct cell populations bearing variegated dye concentrations ( Figure 8B ) . The fastest dividing population ( CFSE-Lo ) had undergone at least 4 more cell divisions than the bulk MEFs and gave rise to substantially more AP-positive iPS cell colonies than the slower-dividing populations ( Figure 8B; Figure 8—figure supplement 1 ) . Strikingly , CFSE-Lo cells also expressed the highest levels of Dkc1 reaching that of ES cells ( Figure 8C ) . These cells have lost their mesenchymal identity and initiated the transition into cells of epithelial origin ( Figure 8D ) . By contrast , the slower dividing populations expressed significantly lower levels of Dkc1 and failed to fully silence mesenchymal genes or robustly reactivate epithelial markers indicating a delayed or abortive MET . Importantly , using MEFs carrying an integrated dox-inducible OKSM expression cassette ( Carey et al . , 2010 ) , we observed a similar preferential enrichment of Dkc1 in the fastest-dividing population ( CFSE-Lo ) despite uniform Oct4 expression levels among CFSE-Lo , Med , and Hi cells ( Figure 8—figure supplement 2 , 3 ) . Therefore , an early onset of MET appears to be a defining property of these ultrafast cycling cells wherein appropriately high levels of DKC1 are necessary and likely serve as an important gene-specific transcriptional coactivator . 10 . 7554/eLife . 03573 . 019Figure 8 . Fast cycling somatic cell state conducive to iPS cell generation requires DKC1 . ( A ) MEFs depleted of DKC1 by two independent shRNAs ( shDKC1-1 and shDKC1-2 ) , along with MEFs infected with control non-target lentiviruses , were analyzed using the CellTrace CFSE Proliferation Assay ( Life Technologies ) . The doubling time for each population was calculated using the mean fluorescence intensity of each timepoint over 96 hr . ( B ) Induced MEFs ( light purple ) are treated with dox for 4 days , labeled with CFSE , and continuously cultured in the presence of dox for an additional 48 hr prior to FACS . Populations for ultrafast ( Lo ) , fast ( Med-Lo ) , medium slow ( Med-Hi ) , and slow ( Hi ) cycling dox-induced MEFs are sorted based on CFSE intensity and denoted by dashed boxes . CFSE-intensity of MEFs immediately after labeling ( black ) , unlabeled MEFs ( brown ) , and uninduced MEFs 48 hr post-labeling ( dark purple ) are shown and used as controls . ( C ) mRNA levels of Dkc1 and Oct4 in sorted MEF populations ( Lo-Hi ) are compared to D3 ES cells and uninduced MEFs by qPCR . Results are normalized to Actb . Error bars represent standard deviation ( n = 3 ) . ( D ) Ultrafast ( CFSE-Lo ) cycling MEFs undergo early MET . mRNA levels of mesenchymal genes ( Slug , Snail , and Zeb1; left ) , and epithelial genes ( Ecad and Epcam; right ) in sorted CFSE-labeled cell populations are compared to D3 ES cells and uninduced MEFs by qPCR . Data are analyzed as in ( C ) . Primer sequences can be found in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 03573 . 01910 . 7554/eLife . 03573 . 020Figure 8—figure supplement 1 . Ultrafast cycling MEF population contains the bulk of reprogramming activity . Sorted MEF populations ( CFSE-Lo , Med-Lo , Med-Hi , and Hi ) in Figure 7B are plated on feeders in 24-well plates at indicated cell numbers . AP-positive colonies are stained as described in Figure 7A . DOI: http://dx . doi . org/10 . 7554/eLife . 03573 . 02010 . 7554/eLife . 03573 . 021Figure 8—figure supplement 2 . Ultrafast cycling somatic cell state can be induced in secondary OKSM MEFs . ( A ) Labeling and sorting of CFSE-labeled MEFs carrying an integrated dox-inducible transgene expressing OCT4 , KLF4 , SOX2 , and c-MYC ( OKSM MEFs ) are performed and analyzed as described in Figure 8B . ( B ) Sorted OKSM MEF popuations ( CFSE-Lo , Med , and Hi ) in ( A ) are plated onto 24-well plate . Cells are stained for AP activity after 14 days dpi ( 11 days with dox followed by 3 days without dox ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03573 . 02110 . 7554/eLife . 03573 . 022Figure 8—figure supplement 3 . Ultrafast cycling OKSM MEFs have elevated Dkc1 levels . Dkc1 expression is upregulated in the ultrafast cycling MEFs upon OKSM expression . mRNA levels of Dkc1 and Oct4 in sorted MEF populations described in ( C ) are compared to D3 ES cells and uninduced MEFs by qPCR . Results are normalized to Actb . Error bars represent standard deviation ( n = 3 ) . Primer sequences can be found in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 03573 . 022 Our de novo identification of the DKC1 complex as a transcriptional coactivator for OCT4/SOX2 underscores the expanding repertoire of this multifunctional ribonucleoprotein complex ( RNP ) in stem cells . Beyond its well-documented role in ribosome and telomerase biogenesis , the DKC1 complex has been shown to effect diverse cellular processes including internal ribosome entry site ( IRES ) -dependent translation ( Yoon et al . , 2006 ) and base excision of 5-hydroxymethyluridines in rRNA by uracil-DNA glycosylase 1 ( SMUG1 ) ( Jobert et al . , 2013 ) . Interestingly , the telomerase complex itself has been implicated in the regulation of MYC and WNT/β-catenin associated gene expression programs critical for stem cell function ( Choi et al . , 2008; Park et al . , 2009 ) . However , the reverse transcriptase TERT , but curiously not its catalytic activity , was reported to be required for gene activation ( Choi et al . , 2008 ) , a finding that remains somewhat controversial in the telomerase field ( Listerman et al . , 2014 ) . In light of our findings that the core DKC1 complex possesses transcriptional coactivator activity , it is tempting to speculate that in the context of WNT-responsive genes , TERT may function to tether the DKC1 complex ( as part of the telomerase RNP ) to gene promoters and activate transcription by binding to a β-catenin-TCF3 activating complex ( Park et al . , 2009 ) , an integral component of the core stem cell-specific regulatory circuitry ( Cole et al . , 2008 ) . It is , however , unlikely that the coactivator activity detected in our assay is dependent on TERT because we did not detect TERT or any other known components of the telomerase complex ( Fu and Collins , 2007 ) in our purified fractions by mass spectrometry ( data not shown ) . This is further supported by our observation that recombinant DKC1-NOP10 heterodimers purified from bacteria was active in transcription . Instead , we favor the model whereby the DKC1 complex ( free of any accessory factors ) can be recruited to key pluripotency genes via a direct interaction with the XPC complex as we observed in co-immunoprecipitation experiments . This DKC1-XPC assembly is , in turn , recruited to target gene promoters via activator–coactivator interactions with OCT4 and SOX2 ( Fong et al . , 2011; Gao et al . , 2012 ) . The mechanism by which different activators recruit the same coactivator by targeting distinct subunits or protein surfaces within the DKC1 complex may represent a common strategy that is frequently observed with other transcriptional coactivators such as Mediator ( Taatjes et al . , 2002 ) and TAFs/TFIID ( Liu et al . , 2009 ) . Therefore , the DKC1 complex may coordinate diverse transcriptional outputs contributing to stemness by cooperating with both stem cell-specific and cell-ubiquitous activators and coactivators ( Mediator , the XPC complex , SCC-B ) . Interestingly , the Dkc1 gene itself is also a target of OCT4 and NANOG ( Agarwal et al . , 2010 ) . Integrating Dkc1 into the core regulatory circuitry could further stabilize the autoregulatory loops established by OCT4 , SOX2 and NANOG that are postulated to confer stability to self-renewing ES cells without sacrificing their responsiveness to developmental cues during differentiation ( Boyer et al . , 2005; Loh et al . , 2006 ) . An increasing number of non-coding RNAs ( ncRNAs ) have emerged as critical players in gene regulation in both mammals ( 7SK , Alu , B2 ) and bacteria ( 6S ) ( Storz et al . , 2011; Kugel and Goodrich , 2012 ) . With few exceptions , these ncRNAs function to inhibit the transcriptional activity of their target proteins or protein complexes by forming a stable RNP . The precise mechanism by which snoRNAs may positively or negatively regulate the transcriptional activity of the DKC1 complex is unclear . Additional experiments will be required to determine how the lengths , sequences , and/or secondary/tertiary structures of a subset of snoRNAs may confer coactivator competence to the DKC1 complex . Because as much as 90% of snoRNAs are embedded within introns of protein-coding genes in humans ( Dieci et al . , 2009 ) , their repertoire and relative abundance are directly coupled to the expression of their host genes . Indeed , many snoRNAs are differentially expressed during neural differentiation of mouse ES cells in vitro ( Skreka et al . , 2012 ) . Therefore , it is conceivable that the coactivator activity ( and potentially specificity ) of the DKC1 snoRNP can be coordinately regulated by cell type-dependent modulation of snoRNA composition . Considering that many of the disease-causing mutations found in Dkc1 have been shown to disrupt the binding and/or stability of a select subset of mammalian snoRNAs which we have shown could play a critical role in conferring coactivator competence to the DKC1 complex , it was somewhat surprising that neither the enzymatic activity nor amino acids mutated in dyskeratosis congenita ( DC ) patients negatively impacted coactivator activity . Although targeted disruption of Dkc1 in mice is lethal ( He et al . , 2002 ) , many DC patients carrying mutations in the DKC1 complex live well into their teens and beyond , indicating that these mutations are hypomorphic and can be tolerated during embryogenesis . These findings suggest that amino acid residues in DKC1 critical for transcription are likely largely distinct from those mutated in DC patients . In cases where such mutations do overlap , their effect , if any , on transcription are expected to be subtle because mutations that significantly compromise coactivator function would likely be severely detrimental to the tightly regulated process of mammalian development . However , we note that the various mutant complexes tested in our in vitro transcription assays were reconstituted in insect cells and co-purified with significant amounts of insect snoRNAs ( Figure 4—figure supplement 2 ) . Because very little is known about the snoRNA repertoire in Sf9 cells and the extent of functional conservation between insect and human snoRNAs , it is unclear how the incorporation of these insect RNAs into human DKC1 complexes might impact structure and function of such hybrid DKC1 RNPs . Therefore , it will be prudent in the future to re-examine these disease relevant mutations in the context of human-protein , human-snoRNA DKC1 RNPs . Acute depletion of DKC1 in mouse ES cells rapidly down-regulated key pluripotency genes well in advance of telomere attrition and the ensuing cellular senescence that one would expect to occur ( after >300 population doublings ) due to compromised telomerase function ( Niida et al . , 1998; Mochizuki et al . , 2004 ) . Taken together with our ChIP results showing specific recruitment of the DKC1 complex to OCT4/SOX2 enhancers of core pluripotency genes in both mouse and human ES cells , as well as the strong dependence of OCT4/SOX2-activated transcription on the DKC1 complex in vitro , we suggest that defects in pluripotency gene expression and stem cell self-renewal upon DKC1 knockdown are at least in part due to compromised transcriptional activation rather than a sole consequence of telomerase deficiency . Given the importance of establishing a robust OCT4/SOX2-depedent transcriptional circuitry during iPS cell induction , it is perhaps not surprising that DKC1 knockdown also severely limits reprogramming capacity of MEFs . Our data , however , indicate that reprogramming of DKC1-deficient MEFs by OKSM aborted at a rather early stage—during the mesenchymal-to-epithelial transition ( MET ) . Specifically , DKC1 appears to be required for the proper induction of epithelial markers like Ecad ( also known as Cdh1 ) critical for iPS cell generation ( Chen et al . , 2010 ) even though key negative regulators of epithelial gene expression , Snail and Slug , were already repressed in DKC1 knockdown MEFs ( Thiery et al . , 2009 ) . Because SOX2 ( or KLF4 ) alone is sufficient to induce Ecad expression in MEFs ( Liu et al . , 2013 ) , and OKSM cotarget many MET genes early in the reprogramming process ( Soufi et al . , 2012 ) , the DKC1 complex could cooperate with SOX2 and other reprogramming factors to activate an epithelial gene expression program . It has also been reported that restoring telomerase activity in DC patient-specific fibroblasts carrying a loss-of-function mutation in Dkc1 ( L37del ) by overexpressing TERT ( Wong and Collins , 2006 ) is insufficient to overcome the reprogramming defect associated with these cells ( Agarwal et al . , 2010 ) . Given that L37del fibroblasts show normal rRNA pseudouridine content as well as rRNA processing kinetics ( Wong and Collins , 2006 ) , these data taken together strongly suggests a telomerase and ribosome independent mechanism by which the DKC1 complex participates in somatic cell reprogramming . We propose that the DKC1 complex may function to promote the requisite MET during iPS cell generation by activating pro-epithelial genes consistent with its transcriptional coactivator function . Direct reprogramming of fibroblasts into iPS cells is a slow and presumably stochastic process ( Yamanaka , 2009 ) . However , accumulating evidence suggests that it is nonetheless amenable to acceleration ( and thereby enhanced efficiency ) by manipulating pathways that promote cell division ( Banito et al . , 2009; Hanna et al . , 2009; Hong et al . , 2009; Utikal et al . , 2009 ) . These highly proliferative cells competent for reprogramming are also found to exist naturally or become primed by OKSM in a small subset of so called ‘privileged’ somatic cells within a largely homogeneous population that can proceed through reprogramming in a non-stochastic manner with shorter latency ( Guo et al . , 2014 ) . However , the cell-intrinsic determinants conducive to this privileged state remain unclear , but are likely distinct from factors previously implicated in deterministic cellular reprogramming including SOX2 ( Buganim et al . , 2012 ) and MBD3 ( Liu et al . , 2013; Rais et al . , 2013 ) . Here we show that a subpopulation of MEFs proliferating at a significantly faster rate is especially sensitive to DKC1 depletion ( Figure 8A ) . Similarly , a small fraction of ultrafast cycling MEFs enriched for Dkc1 and depleted of mesenchymal signatures also emerged upon OKSM expression . Although the relationship between these two highly proliferative MEF populations is unclear , we surmise that they could represent a similar privileged somatic cell state . Therefore , the intrinsic variable levels of DKC1 in regular MEFs , or the ability of some MEFs to upregulate Dkc1 to a critically high threshold level in response to ectopic expression of OKSM , could be a limiting factor in the acquisition of this rare somatic cell state possibly by facilitating an early MET . The precise role of DKC1 in establishing this privileged state in MEFs is unclear but may involve regulating both cellular proliferation and gene expression critical for the early phase of iPS cell formation . In summary , using an unbiased biochemical approach to probe the transcriptional regulation of the Nanog gene by OCT4 and SOX2 , we uncovered an unanticipated transcriptional coactivator role of the DKC1 complex and a subset of its associated snoRNAs in ES and iPS cells . We surmise that the DKC1 complex could be one of the cell-intrinsic determinants that impinges on somatic cells during reprogramming by coupling cellular proliferation to stem cell-specific transcription . cDNAs for human and mouse DKC1 , GAR1 , NHP2 , and NOP10 were obtained from cDNA libraries generated from total RNAs isolated from human NTERA-2 ( NT2 ) and mouse ES D3 cells . Mammalian expression plasmids encoding all four subunits of the DKC1 complex were derived from the pHAGE-EF1α-STEMCCA construct ( Sommer et al . , 2009 ) , wherein OCT4 , KLF4 , SOX2 , and c-MYC were replaced with N-terminal FLAG-tagged DKC1 , NHP2 , GAR1 , and NOP10 , respectively ( pHAGE-EF1α-DKC1 ) . Expression plasmid for overexpressing the XPC complex ( pHAGE-EF1α-XPC ) was described ( Fong et al . , 2011 ) . For expressing the DKC1 complex in insect Sf9 cells , N-terminal His6- tagged human DKC1 ( wild-type and various disease-associated mutants ) , untagged GAR1 , N-terminal FLAG-tagged NHP2 , and untagged NOP10 were inserted into a modified pFastBAC Dual vector ( Invitrogen , Carlsbad , CA ) . For expressing partial and holo DKC1 complexes in E . coli , untagged human DKC1 , N-terminal HA-tagged GAR1 , N-terminal FLAG-tagged NHP2 , and C-terminal His6-tagged NOP10 were cloned into a pST44 polycistronic expression plasmid ( Tan et al . , 2005 ) . Of note , GAR1 cDNA was reengineered using Quikchange II Site Directed Mutagenesis Kit ( Agilent , Santa Clara , CA ) to replace codon-pairs for diglycine residues with sequences that are more favorable for translation ( Li et al . , 2012 ) . For the NAF1-containing intermediate complex , N-terminal HA-tagged NAF1 was inserted in place of HA-GAR1 . Polyclonal antibodies against GAR1 ( 11 , 711 ) , NHP2 ( 15 , 128 ) , FBL ( 16 , 021 ) , NOP58 ( 14 , 409 ) , and CETN2 ( 15 , 877 ) were purchased from ProteinTech Group; XPC ( 122A ) , RAD23B ( 306A ) from Bethyl Laboratories , Montgomery , TX; DKC1 ( H-300 ) , TFIIB ( C-18 ) , and OCT4 ( N-19 ) from Santa Cruz Biotechnology ( Dallas , TX ) ; SOX2 ( AB5603 ) from Millipore ( Billerica , MA ) . Purified rabbit IgGs were purchased from Jackson ImmunoResearch Laboratories ( West Grove , PA ) . Monoclonal antibodies against β-actin ( AC-74 ) were purchased from Sigma Aldrich ( St . Louis , MO ) , DKC1 ( H-3 ) from Santa Cruz Biotechnology , and NOP10 ( 6547-1 ) from Epitomics ( Burlingame , CA ) . Anti-FLAG ( M2 ) monoclonal antibodies were purchased from Sigma Aldrich and anti-HA antibodies ( MMS-101P ) from Covance ( Dedham , CA ) . Antibody against mouse RAD23 was generated in guinea pigs ( Fong et al . , 2011 ) . The human embryonal carcinoma NTERA-2 ( NT2 ) cell line was obtained from ATCC . NT2 , 293T , and HeLa cells were cultured in DMEM high glucose with GlutaMAX ( Invitrogen ) supplemented with 10% fetal bovine serum ( FBS; HyClone , Piscataway , NJ ) . Large scale culture of NT2 cells were described ( Fong et al . , 2011 ) . Mouse ES cell line D3 was purchased from ATCC ( Manassas , VA ) and adapted to feeder-free condition as described ( Fong et al . , 2011 ) . Differentiation of D3 cells was induced by maintaining cells in LIF-free ES cell medium containing 2–5 mM all-trans retinoic acid ( Sigma Aldrich ) for up to 7 days . Human ES cell line H9 ( WiCell , Madison , WI ) was maintained in feeder-independent conditions , using Synthemax SC-II Substrate ( Corning ) and grown in TeSR-E8 ( Stemcell Technologies , Canada ) . Media was changed daily and cell cultures were passaged using Dispase ( Stemcell Technologies ) , according to the manufacturer's protocol . All steps were performed at 4°C . Nuclear extracts were prepared from 400 l of NT2 cells . Partially purified P11-phosphocellulose 1 M KCl and Ni-NTA flowthrough ( Ni-FT ) fractions were prepared as described ( Fong et al . , 2011 ) . The Ni-FT fraction was dialyzed against buffer D at 0 . 2 M KCl with 0 . 0025% NP-40 and 10% glycerol ( all buffers from then on contained 0 . 0025% NP-40 and 10% glycerol unless otherwise stated ) . This Ni-FT fraction was applied to a Poros 20 HQ column ( Applied Biosystems , Carlsbad , CA ) , subjected to a 4 column volume ( CV ) linear gradient from 0 . 2 M to 0 . 4 M KCl ( Q0 . 3 ) , washed at 0 . 52 M KCl , and developed with a 13 CV linear gradient from 0 . 52 M to 1 . 0 M KCl . Transcriptionally active Q0 . 3 fraction ( 0 . 32–0 . 4 M ) were pooled and applied directly to hydroxyapatite ( HAP ) type II ceramic resin ( Bio-Rad , Hercules , CA ) , washed first at 0 . 38 M , then lowered to 0 . 1 M KCl in 3 CV . HAP column buffer was then exchanged and washed extensively with buffer D at 0 . 03 M KPi , pH 6 . 8 without KCl and NP-40 . The HAP column was subjected to a 20 CV linear gradient from 0 . 03 M to 0 . 6 M KPi . Active HAP fractions eluting from 0 . 2–0 . 3 M KPi were pooled and separated on a Superose 6 XK 16/70 gel filtration column ( 130 ml , GE Healthcare , Piscataway , NJ ) equilibrated with buffer D + 0 . 1 mM EDTA at 0 . 15 M KCl . Active Superose 6 fractions with an apparent molecular mass of 400–600 kDa were pooled and supplemented with 0 . 25 mg/ml insulin ( Roche , Indianapolis , IN ) . Pooled fractions were applied to a Poros 20 HE column ( Applied Biosystems ) equilibrated in buffer D + 0 . 1 mM EDTA at 0 . 15 M KCl , subjected to a 34 CV linear gradient from 0 . 15 M to 1 M KCl . SCC-A containing HE fractions eluted from 0 . 56–0 . 62 M KCl . For affinity purification of endogenous DKC1 complexes , Ni-FT derived from 200 l of NT2 cells was applied to a Poros 20 HQ column , subjected to a 22 CV linear gradient from 0 . 2 M to 1 M KCl . Fractions with low levels of DKC1 were first concentrated using a Spin-X UF concentrator ( Corning , Tewksbury , MA ) before they were used for immune-affinity purification . Various Poros 20 HQ fractions ( adjusted to 0 . 05% NP-40 ) were incubated with 10 μg of anti-DKC1 monoclonal antibody immobilized on Protein G Sepharose ( GE Healthcare ) for 16 hr in the presence of RNase inhibitors ( RNasin Plus , Promega , Madison , WI ) , washed extensively with 0 . 6 M KCl HEMG buffer ( 25 mM HEPES , pH 7 . 9 , 0 . 1 mM EDTA , 12 . 5 mM MgCl2 , 10% glycerol ) with 0 . 2% NP-40 , then equilibrated with 0 . 3 M KCl HEMG with 0 . 1% NP-40 before elution with peptides . Peak Poros 20 Heparin fractions were pooled , concentrated using a Spin-X centrifugal concentrator , separated by SDS-PAGE , stained , protein bands excised , digested with trypsin , and extracted . Peptide pools from each gel slice were analyzed by matrix-assisted laser desorption time-of-flight mass spectrometry ( MALDI-TOF MS; Bruker Reflex III ) . Selected mass values were used to search protein databases linked to PROWL ( Rockefeller University ) using ProFound and protein databases linked to ExPASy ( Swiss Institute of Bioinformatics , Geneva ) using PeptIdent . In vitro transcription reactions , DNA template , purification of activators OCT4 and SOX2 , general transcription factors , RNA polymerase , and recombinant XPC complex were described ( Fong et al . , 2011 ) . DKC1-associated small RNAs were isolated using TRIzol reagent ( Life Technologies , Carlsbad , CA ) . RNAs were treated with tobacco acid pyrophosphatase ( TAP ) ( Epicentre , Madison , WI ) to remove the 5′ m7G cap followed by dephosphorylation with APex Alkaline Phosphatase ( Epicentre ) . Purified RNAs were labeled with T4 polynucleotide kinase ( PNK ) ( New England Biolabs , Ipswich , MA ) and γ-32P-ATP in the presence of RNase inhibitors ( RNAsin Plus , Promega ) at 37°C for 1 . 5 hr . RNAs were precipitated and washed with 75% ethanol to remove free γ-32P-ATP . Labeled RNAs were separated on a 6% denaturing Urea-polyacrylamide gel , and visualized by radiography . Recombinant Bacmid DNAs for expressing wild-type and mutant DKC1 complexes were generated from pFastBAC constructs ( described above ) according to manufacturer's instructions ( Invitrogen ) . Recombinant baculovirus for the infection of Sf9 cells was generated using the Bac-to-Bac Baculovirus Expression System ( Invitrogen ) . Baculoviruses were amplified three times in Sf9 cells . 2 l of Sf9 cells ( ∼2 × 106/ml ) were infected with baculoviruses , collected at 48 hr post infection , washed once with ice-cold PBS , lysed in six packed cell volume of 0 . 3 M NaCl buffer HGN ( 50 mM HEPES , pH 7 . 9 , 10% glycerol , 0 . 5% NP-40 ) , and sonicated briefly . Cleared lysate was supplemented with 10 mM imidazole and incubated with Ni-NTA resin pre-equilibrated with 0 . 5 NaCl HGN and 10 mM imidazole for 16 hr . Resin slurries were poured into gravity columns , washed with 0 . 5 NaCl HGN ( 0 . 1% NP-40 ) with 20 mM imidazole , and bound DKC1 complexes were eluted with buffer 0 . 3 M NaCl HGN ( 0 . 1% NP-40 ) containing 0 . 25 M Imidazole . Peak fractions were loaded immediately to a gravity column containing Heparin Sepharose 6 Fast Flow ( GE Healthcare ) pre-equilibrated with 0 . 3 M NaCl HEGN ( 25 mM HEPES , pH 7 . 9 , 0 . 1 mM EDTA , 10% glycerol , 0 . 02% NP-40 ) . Column was washed extensively at 0 . 3 M NaCl HEGN , then with 0 . 5 M NaCl HEGN . The DKC1 complexes were eluted with 1 M NaCl HEGN . Peak fractions containing all four subunits of the DKC1 complex , as determined by western blotting , were pooled and incubated with anti-FLAG ( M2 ) agarose ( Sigma Aldrich ) for 3–4 hr , washed at 0 . 5 M NaCl HEGN and re-equilibrated with micrococcal nuclease ( MNase ) digestion buffer ( 25 mM Tris–HCl , pH 7 . 9 , 20 mM NaCl , 60 mM KCl , 2 mM CaCl2 , 0 . 01% NP-40 , 10% glycerol ) . Bound DKC1 complexes were treated with 300 U of MNase ( Thermo Scientific , Waltham , MA ) or buffer at room temperature and nutated for 1 hr . MNase digestion was terminated with 20 mM EGTA . Mock and MNase-treated DKC1 complexes were washed extensively with 0 . 6 M NaCl HEMG with 0 . 2% NP-40 and 20 mM EGTA and equilibrated with 0 . 3 M NaCl HEMG with 0 . 1% NP-40 followed by FLAG peptide elution . For purification of bacterial DKC1 complexes , pST44 expression plasmids were transformed into BL21-Codon Plus RIPL competent cells ( Agilent ) . Expression of hetero-dimeric ( FLAG-DKC1/NOP10-His6 ) , -trimeric ( untagged DKC1/FLAG-NHP2/ NOP10-His6 ) , holo ( untagged DKC1/HA-GAR1/FLAG-NHP2/ NOP10-His6 ) DKC1 complexes as well as NAF1-containing intermediate DKC1 complex ( untagged DKC1/HA-NAF1/FLAG-NHP2/ NOP10-His6 ) were induced at 30°C for 4 hr with 0 . 5 mM IPTG . Cell pellets were lysed in high salt lysis buffer HSLB ( 50 mM Tris–HCl pH 7 . 9 , 0 . 5 M NaCl , 0 . 6% TritonX-100 , 0 . 05% NP-40 , 10% glycerol ) with imidazole ( 10 mM ) and lysozyme ( 0 . 5 mg/ml ) . Sonicated lysates were cleared by ultracentrifugation and incubated with Ni-NTA resin for 16 hr . Bound proteins were washed extensively with HSLB with 20 mM imidazole , equilibrated with 0 . 25 M NaCl HGN ( 25 mM HEPES , pH 7 . 9 , 10% glycerol , 0 . 01% NP-40 ) with 20 mM imidazole , and eluted with 0 . 25 M imidazole in 0 . 25 M NaCl HGN . Peak fractions were pooled and applied to a Poros 50 Heparin ( HE ) column , washed extensively with 0 . 25 M and 0 . 5 M NaCl HGN , and subjected to a 4 CV linear gradient from 0 . 5 M to 1 M NaCl . Fractions containing the desired subunits of the DKC1 complexes were detected by western blotting , pooled , and incubated with anti-FLAG agarose for 3–4 hr at 4°C . Bound proteins were washed extensively at 0 . 7 M NaCl HGN with 0 . 1% NP-40 and re-equilibrated with 0 . 3 M NaCl HGN with 0 . 1% NP-40 before elution with FLAG peptides . For holo and NAF1-containing DKC1 complexes , HE fractions were first incubated with anti-HA resin , washed and eluted with HA peptides before proceeding to the anti-FLAG affinity immunoprecipitation step as described . 32 , 25 , 5 , and 2 l of E . coli cultures were required to generate ∼0 . 5 μg of purified holo DKC1 , NAF1 intermediate , hetero-trimeric , and–dimeric complexes , respectively . pHAGE-EF1α-STEMCCA , pHAGE-EF1α-mXPC , and pHAGE-EF1α-mDKC1 expression plasmids were co-transfected into 293T cells using Lipofectamine 2000 ( Invitrogen ) . Transfected cells on 10 cm dishes were lysed directly on plates with 1 ml of lysis buffer ( 200 mM NaCl , 50 mM HEPES-KOH , pH 7 . 9 , 0 . 1 mM EDTA , 0 . 5% NP-40 and 10% glycerol ) 40 hr post-transfection . Cell lysates were collected and homogenized by passing through a 25-gauge needle five times . Lysates were cleared by centrifugation at 15k rpm for 25 min at 4°C . 3 μg of anti-RAD23B antibodies were coupled to Protein A sepharose ( GE Healthcare ) in PBS containing 0 . 05% NP-40 for 1 hr at room temperature . Antibody-coupled beads were washed and equilibrated with lysis buffer before incubating with 0 . 5 ml of cleared cell lysates for 16 hr at 4°C . Sepharose beads were then washed extensively with lysis buffer and bound proteins were eluted with SDS/sample buffer and analyzed by western blotting . For lentivirus production , non-target control and pLKO plasmids targeting mouse DKC1 ( and XPC ) ( Sigma Aldrich ) were co-transfected with packaging vectors into 293T cells using lipofectamine 2000 ( Invitrogen ) . Supernatants were collected at 48 hr , and again at 72 hr . Virus preparation , titer determination , and infection of D3 mouse ES cells were performed as described ( Fong et al . , 2011 ) , except at a multiplicity of infection ( MOI ) of 25 . For DKC1 knockdown reprogramming experiments , MEFs were transduced at a MOI of 5 prior to iPS cell induction . Detection of alkaline phosphatase activity of knockdown ES cells was carried out using a commercial kit ( Millipore ) . Mouse ES cell line D3 and human ES cell line H9 were first crosslinked with ethylene glycol bis[succinimidylsuccinate] ( EGS , 3 mM , Pierce ) for 30 min and then with formaldehyde ( 1% ) for 5 min in fixing buffer ( 50 mM HEPES , pH 7 . 5 , 0 . 1 M NaCl , 1 mM EDTA , 0 . 5 mM EGTA ) to capture protein–protein and protein-DNA interactions ( Zeng et al . , 2006 ) . Crosslinking was then terminated by glycine ( 0 . 125 M ) . Cells were washed twice with PBS , scraped , and centrifuged at 150×g for 5 min at 4°C , resuspended in lysis buffer ( 50 mM HEPES , pH 7 . 9 , 0 . 14 M NaCl , 1 mM EDTA , 10% glycerol , 0 . 5% NP-40 , 0 . 25% Triton X-100 ) with Halt Protease Inhibitor Cocktail ( Pierce , Waltham , MA ) , and nutated at 4°C for 10 min . Nuclei were pelleted at 1700×g for 5 min , washed twice with wash buffer ( 10 mM Tris–HCl , pH 8 . 1 , 0 . 2 M NaCl , 1 mM EDTA , 0 . 5 mM EGTA ) and twice with shearing buffer ( 0 . 1% SDS , 1 mM EDTA , 10 mM Tris–HCl , pH 8 . 1 ) . Nuclei were resuspended in shearing buffer , transferred to Covaris TC 12 × 12 mm tubes with AFA Fiber , and sonicated with a Covaris S2 Focused Ultrasonicator to obtain DNA fragments averaging 300–500 bp in length . Cleared chromatin extracts were adjusted to 0 . 15 M NaCl and 1% Triton X-100 and immunoprecipitated overnight at 4°C with 3 μg of purified rabbit IgGs or anti-DKC1 antibody . Immunoprecipitated DNA was captured with pre-equilibrated Protein A sepharose ( GE Healthcare ) , washed extensively with high salt wash buffer ( 0 . 1% SDS , 1% Triton X-100 , 2 mM EDTA , 20 mM HEPES , pH 7 . 9 , 0 . 5 M NaCl ) , LiCl wash buffer ( 100 mM Tris–HCl , pH 7 . 5 , 0 . 5 M LiCl , 1% NP-40 , 1% sodium deoxycholate ) , and TE buffer ( 10 mM Tris–HCl , pH 8 . 0 , 0 . 1 mM EDTA ) . Supernatant from control IgG immunoprecipitates was saved as input . Input chromatin and immunoprecipitated DNA were reversed crosslinked overnight at 50°C with Proteinase K ( Invitrogen ) , RNase A ( Thermo Scientific ) , and 0 . 3 M NaCl . DNA was purified using a Qiaquick PCR Purification Kit ( Qiagen , Netherlands ) . Purified DNA was quantified by real time PCR with SYBR Select Master Mix for CFX ( Life Technologies ) and gene specific primers ( Supplementary file 1 ) using a CFX Touch Real-Time PCR Detection System ( Bio-Rad ) . The position of each amplicon relative to transcription start site of mouse Nanog , Oct4 , Sox2 , Fgf4 , and human Oct4 and Nanog is indicated in Figure 5 . Cells were rinsed once with PBS . Total RNA was extracted and purified using TRIzol reagent ( Life Technologies ) followed by DNase I treatment ( Invitrogen ) . cDNA synthesis was performed with 1 μg of total RNA using iScript cDNA Synthesis Kit ( Bio-Rad ) and diluted 10-fold . Real time PCR analysis was carried out with SYBR Select Master Mix for CFX ( Life Technologies ) and gene specific primers ( Supplementary file 1 ) using the CFX96 Touch Real-Time PCR Detection System ( Bio-Rad ) . Results were normalized to β-actin . CF-1 MEFs ( Charles River , Wilmington , MA ) were transduced with inducible STEMCCA and rtTA lentivirus-containing supernatants overnight in 8 μg/ml polybrene ( Sigma Aldrich ) . Alternatively , MEFs isolated from mice carrying an integrated dox-inducible transgene expressing OCT4 , KLF4 , SOX2 , and c-MYC ( Jackson Laboratories , Bar Habor , ME ) were also used . Doxycycline ( Sigma Aldrich; 2 μg/ml ) was supplemented to complete mouse ES cell media to induce expression of OKSM . Reprogramming was assayed by alkaline phosphatase staining ( Millipore ) , NANOG staining ( Abcam , United Kingdom , ab80892 ) , or by flow cytometry analysis using anti-CD90 . 2/Thy1 . 2 ( Biolegends , San Diego , CA ) and anti-SSEA1 ( Biolegends , San Diego , CA ) on a BD LSRFortessa , performed according to the manufacturers' protocols . To determine the doubling time , MEFs were labeled with CFSE-Violet ( Life Technologies ) at a working concentration of 5 . 0 μM , according to the manufacturers' protocol . Cells were analyzed for remaining fluorescence on a BD LSRFortessa every day for 4 days . Induced MEFs were labeled with CFSE-Violet ( Life Technologies ) at a working concentration of 7 . 5 μM , as described in Guo et al . , 2014 . CFSE-labeled MEFs were sorted into distinct fast to slow dividing populations at the UC Berkeley Li Ka Shing Flow Cytometry Facility . MEFs cultured in the absence of doxycycline were used as controls .

The stem cells found in an embryo are able to develop into any of the cell types found in the body of the animal: an ability called pluripotency . When a cell becomes a specialized cell type , such as a nerve cell or a muscle cell , it loses this ability . However , mature cells can be reprogrammed back to a pluripotent state by artificially introducing certain proteins ( known as ‘reprogramming factors’ ) into the mature cells . A core group of reprogramming factors are known to activate networks of genes that are normally only expressed in stem cells , and by doing so trigger and maintain a pluripotent state . Other proteins help these core factors to regulate these networks of genes . In 2011 , researchers discovered that a protein complex called XPC—which is normally involved in DNA repair—also helps two core reprogramming factors to activate an important gene related to pluripotency . Now , Fong et al . , including several of the researchers involved in the 2011 work , have identified another unexpected partner for the same two core reprogramming factors . The protein complex , called DKC1 , has a number of known functions related to the processing of RNA molecules . This complex has also been linked to a fatal , rare human disorder called dyskeratosis congenita—a condition that affects many parts of the body , including the skin and bone marrow . Fong et al . found that when embryonic stems cells from mice are depleted of the DKC1 complex , the activation of important pluripotency-related genes by two of the core reprogramming factors is markedly reduced . The XPC and DKC1 protein complexes were found to interact in pluripotent cells , and together they can activate a pluripotency-related gene to a greater extent than either can individually . Fong et al . propose that DKC1 binds to XPC , which in turn binds to two of the core reprogramming factors . The DKC1 complex also binds to RNA molecules , and Fong et al . found that when the DKC1 complex binds to certain RNAs it is more able to help reprogramming factors activate pluripotency-related genes . On the other hand , other RNA molecules seem to inhibit the complex's ability to activate these genes . Mutations identified in people with dyskeratosis congenita can prevent the DKC1 complex from binding to a subset of human RNA molecules . Moreover , the activity of stem cells is impaired in people with this developmental condition . As such , one of the next challenges will be to investigate if these mutations and RNA binding could be linked to problems with the activation of genes related to pluripotency in stem cells .

2014

The dyskerin ribonucleoprotein complex as an OCT4/SOX2 coactivator in embryonic stem cells

Gene expression is precisely regulated during the inflammatory response to control infection and limit the detrimental effects of inflammation . Here , we profiled global mRNA translation dynamics in the mouse primary macrophage-mediated inflammatory response and identified hundreds of differentially translated mRNAs . These mRNAs’ 3’UTRs have enriched binding motifs for several RNA-binding proteins , which implies extensive translational regulatory networks . We characterized one such protein , Zfp36 , as a translation repressor . Using primary macrophages from a Zfp36-V5 epitope tagged knock-in mouse generated by CRISPR/Cas9-mediated genome editing , we found that the endogenous Zfp36 directly interacts with the cytoplasmic poly ( A ) -binding protein . Importantly , this interaction is required for the translational repression of Zfp36’s target mRNAs in resolving inflammation . Altogether , these results uncovered critical roles of translational regulations in controlling appropriate gene expression during the inflammatory response and revealed a new biologically relevant molecular mechanism of translational repression via modulating the cytoplasmic poly ( A ) -binding protein . Precise generation of gene products is essential for cells to mount proper responses to physiological and pathological stimuli . This accuracy is achieved by multiple layers of regulations that occur synchronously at different steps of gene expression , including transcription , splicing , translation , mRNA stability , and protein modification and degradation . Importantly , malfunction of these regulatory circuits contributes to many pathological conditions . Thus , characterizing the regulatory networks controlling gene expression is of both biological and clinical significance . Macrophages are an important type of innate immune cell involved in both detecting exogenous and endogenous danger signals and initiating proper immune responses ( reviewed in Medzhitov and Janeway , 2000 ) . These danger signals are recognized as pathogen-associated molecular patterns ( PAMPs ) by pattern-recognition receptors ( PRRs ) . PAMPs are molecules that are absent in host cells but broadly shared by pathogens . For example , macrophages can detect bacterial infection by recognizing bacterial lipopolysaccharide ( LPS ) , a cell membrane component present in gram-negative bacteria , by a PRR , toll-like receptor 4 . This recognition activates signaling transduction cascades that result in rapid expression of inflammatory response genes to control infection . Many gene products from the inflammatory response , such as the proinflammatory cytokine TNF ( tumor necrosis factor alpha ) , however , are also toxic to healthy tissues ( reviewed in Kalliolias and Ivashkiv , 2016 ) . Thus , the expression of these inflammatory genes is tightly controlled by multiple regulatory circuits to maintain a delicate balance between managing infection/injury and damaging normal tissues ( reviewed in Hanada and Yoshimura , 2002 ) . Critically , malfunction of these regulations can result in pathologic inflammation that is associated with a wide variety of human diseases , including obesity , cardiovascular and neurodegenerative diseases , and cancer ( Reviewed inKotas and Medzhitov , 2015 ) . Thus , characterizing regulatory programs in the inflammatory response will both reveal fundamental mechanisms of gene expression and provide insights into inflammation-associated diseases . Previous studies have identified multiple programs controlling mRNA production and degradation in macrophage-mediated inflammatory responses ( reviewed in Carpenter et al . , 2014; Medzhitov and Horng , 2009 ) . For example , stimulation of PRRs by PAMPs can activate several transcription factors , such as nuclear factor-κB , interferon-regulatory factors , and CCAAT/enhancer-binding protein-δ , which results in the rapid induction of inflammatory response genes . These genes include both cytokines ( e . g . TNF ) to control infection and negative regulators of inflammatory signaling pathways ( e . g . suppressors of cytokine signaling proteins ) to limit the detrimental effects of inflammation . Besides transcriptional regulations , inflammatory response genes are also subject to controls at the mRNA stability level . The 3’-untranslated regions ( UTRs ) of many inflammatory cytokine mRNAs have sequence motifs , such as AU-rich elements ( AREs ) and constitutive decay elements ( CDE ) that can promote rapid mRNA degradation ( Garneau et al . , 2007; Stoecklin et al . , 2003 ) . These mRNA decay mechanisms help to prevent sustained expression of potentially harmful cytokines , thereby contributing to resolving inflammation . Together , these regulations at the mRNA production and degradation levels ensure appropriate transcriptomic responses during inflammation . In eukaryotic cells , mRNA degradation is intimately linked with mRNA translation ( reviewed in Roy and Jacobson , 2013 ) . Interestingly , temporal regulation of translation and mRNA decay is an important mechanism of inhibiting gene expression . For example , certain microRNAs down-regulate the expression of their target mRNAs by first inhibiting translation and then promoting mRNA degradation ( Bazzini et al . , 2012; Djuranovic et al . , 2012 ) . Compared to our knowledge of mRNA stability regulations , our understanding of mRNA translational control during the inflammatory responses , however , is still very limited . Recent observations from activated macrophage-like cell lines indicate that several mRNAs are differentially translated ( Schott et al . , 2014 ) , which implies important roles of controlling mRNA translation during the inflammatory response . The trans-acting factors regulating the translation of these mRNAs , however , are still largely unknown . In this study , we explored translational regulation of the inflammatory response mediated by mouse primary bone-marrow-derived-macrophages ( BMDMs ) . Using ribosome profiling , we identified hundreds of differentially translated mRNAs . Interestingly , the 3’UTRs of these mRNAs have significantly enriched binding motifs for several RNA-binding proteins ( RBPs ) expressed in the activated BMDMs , which suggests widespread mRNA translational regulatory networks . We functionally characterized one such RBP , Zfp36 , which is required for resolving inflammation , as a translational repressor . Using primary BMDMs from a V5-epitope tag knock-in mouse generated by CRISPR/Cas9-mediated genome editing , we identified the target mRNAs of the endogenous Zfp36 by CLIP-seq ( cross-linking immunoprecipitation followed by high-throughput sequencing ) . Moreover , using quantitative proteomics , we found that the endogenous Zfp36 predominantly interacts with the cytoplasmic poly ( A ) -binding protein ( Pabpc1 ) in an RNA-independent manner in primary activated BMDMs . Critically , this interaction is required for the translational repression of the Zfp36 target mRNAs , including the mRNAs encoding several important proinflammatory cytokines , in the activated BMDMs . Collectively , these results highlight critical roles of translational regulations in controlling appropriate gene expression during the inflammatory response and reveal a new biologically relevant molecular mechanism of translational repression via modulating Pabpc1 . We used the response of mouse primary BMDMs to LPS as a model to study translational regulations in inflammation , because this is one of the best-characterized inflammatory responses ( reviewed in Medzhitov and Janeway , 2000 ) . Specifically , mouse primary BMDMs were treated with 100 ng/ml LPS and collected at 0 , 1 , 2 , 4 , and 6 hr after LPS stimulation . Cells harvested at each time point were split into two fractions , which were subjected to either RNA-seq or ribosome profiling ( Ribo-seq ) ( Figure 1A ) . In RNA-seq , ribosomal-RNA-depleted total RNAs were randomly fragmented , followed by directional cloning for sequencing . In Ribo-seq , the ribosome-protected mRNA fragments ( RPFs ) were isolated by sucrose-density gradients and then subjected to directional cloning and sequencing . The resulting reads were then aligned to a well-annotated mouse transcriptome , with more than 300 million reads mapped in total . Thereby , we profiled mRNA expression and the ribosome-mRNA association in parallel during the BMDM-mediated inflammatory response . 10 . 7554/eLife . 27786 . 003Figure 1 . Global translation profiling of mouse primary BMDM-mediated inflammatory response . ( A ) Workflow of parallel ribosome and RNA profiling during the BMDM-mediated inflammatory response . ( B ) Metagene plots show the rise and fall in 28-nt ribosome footprints ( RFPs ) density ( reads per million uniquely mapped reads , RPM ) near starts and stops of annotated CDS , respectively . The 12-nt and 15-nt offsets ( indicated by the red arrows ) from starts and stops reflect distances from RFP 5’ termini to the ribosome P- and A-site codons at translation initiation and termination , respectively . ( C ) Subcodon resolution of ribosome footprints . Note that 3-nt codon periodicity relative to the known CDSs is seen for 28-nt RFPs but not RNA-seq reads . ( D ) Ribosome footprints are highly specific to coding regions . Boxplots show density of Ribo-seq and RNA-seq reads at 5’UTRs , 3’UTRs , and introns relative to that of the associated CDSs . ( E ) Clustering analysis of the translation efficiency ( TE ) of the 724 genes differentially translated ( empirical p<0 . 05 ) in the inflammatory response . Heatmap displays mean row-centered log2 TE values at 0 , 1 , 2 , 4 , 6 hr post LPS treatment . ( F ) RNA-binding protein motifs enriched within the 3’UTRs of the 724 translationally regulated genes . The 10 most enriched motifs of macrophage-expressed RBPs are shown along with enrichment statistics . DOI: http://dx . doi . org/10 . 7554/eLife . 27786 . 00310 . 7554/eLife . 27786 . 004Figure 1—figure supplement 1 . Reproducibility of Ribo-seq and GO analysis of differentially translated mRNAs in the inflammatory response . ( A ) The ribosome profiling datasets in BMDM-mediated inflammatory response are highly reproducible . Scatterplots ( lower left ) show pairwise comparisons of the ribosome footprint ( RFP ) density estimates ( log2 reads per million mapped reads , RPM ) for annotated transcripts ( >10 reads in every Ribo-seq sample replicate ) among samples after 0 , 1 , 2 , 4 , and 6 hr post LPS treatment; histograms ( middle ) show range of values for each sample replicate , and a graphical display of the Pearson’s correlation matrix for pairwise comparisons is shown to the upper right . ( B ) Gene ontology analysis of the differentially translated mRNAs during the inflammatory response . The 724 differentially translated mRNAs were analyzed by the Enrichr ( ) using the WikiPathways 2016 . The resulting top five pathways are listed . DOI: http://dx . doi . org/10 . 7554/eLife . 27786 . 00410 . 7554/eLife . 27786 . 005Figure 1—figure supplement 2 . TNF mRNA is translationally repressed during late stages of inflammatory response . Reads from RNA-seq and Ribo-seq at each time point during the inflammatory response in the TNF locus were displayed . DOI: http://dx . doi . org/10 . 7554/eLife . 27786 . 00510 . 7554/eLife . 27786 . 006Figure 1—figure supplement 3 . Verification of translational repression of TNF mRNA at late time points in LPS-stimulated BMDMs . ( A ) The apparent TEs of TNF mRNA in the BMDMs at 0 , 1 , 2 , 4 , 6 hr post LPS stimulation . The variation of the two biological replicates at 0 hr is likely due to low TNF mRNA expression at this time point . ( B ) Polysome profiles of BMDMs with 1 hr LPS stimulation and BMDMs with 6 hr LPS stimulation . ( C ) TNF mRNA and Gapdh mRNA distribution across the sucrose gradient . The results represent the means ( ± SD ) of three independent measurements . *p<0 . 05 , n . s . not significant ( p>0 . 05 ) by the Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 27786 . 00610 . 7554/eLife . 27786 . 007Figure 1—figure supplement 4 . Zfp36 is a translational repressor . The luciferase reporter was co-transfected into 293 T cells with either the lambdaN-Zfp36 expressing plasmid or the lambda-GFP expressing plasmid . The luciferase activity , the FLuc mRNA level , and the translatability ( luciferase activity/FLuc mRNA level ) were determined at 40 hr post transfection . The results represent the means ( ± SD ) of three independent measurements . DOI: http://dx . doi . org/10 . 7554/eLife . 27786 . 007 Our Ribo-seq datasets were highly reproducible , as indicated by comparing the results from two biological replica sets of BMDMs ( Figure 1—figure supplement 1A ) . Moreover , the following fundamental features of translation can be captured at single-nucleotide resolution from these data . First , the RPFs precisely delineate known coding sequences ( CDSs ) and their exons , with 12-nt and 15-nt offsets upstream of the translation start codons and termination codons , reflecting known distances from RFP 5’-termini to the P- and A- site codons , respectively ( Figure 1B ) . Second , the RPFs have 3-nt codon periodicity , a key feature of translocating ribosomes , but the RNA-seq reads do not ( Figure 1C ) . Third , the RPFs are highly specific to CDS regions compared with the RNA-seq reads ( Figure 1D ) . Collectively , these observations indicate good data quality for global analysis of translational regulation in the inflammatory response . To identify differentially translated mRNAs in the inflammatory response , we calculated the apparent translation efficiency ( TE ) for each mRNA expressed above a threshold ( FPKM ≥10 ) at each time point . Here , we defined the apparent TE as the enrichment of RPFs over RNA-seq reads in the CDS regions . We found 724 mRNAs with significant ( empirical p<0 . 05 ) TE changes across the five time-points during the LPS treatment ( Figure 1E ) ( Supplementary file 1 ) . Gene ontology analysis revealed that these mRNAs are involved in diverse pathways ( Figure 1—figure supplement 1B ) , including the TNF and NF-kB signaling pathways that are critical in inflammatory responses . Indeed , TNF mRNA , encoding the essential inflammatory cytokine TNF , is translationally regulated . This mRNA is transcriptionally induced from 0 to 1 hr upon LPS stimulation and then decreases from 1 to 2 hr , and the mRNA level stays relatively stable untill 6 hr post-stimulation . The RPFs of TNF mRNA , however , decreases significantly from 1 to 6 hr , with few TNF RPFs at 6 hr ( Figure 1—figure supplement 2 ) , which indicates TNF mRNA is translationally repressed in the late stages of inflammation . Indeed , the apparent TE of TNF mRNA in the BMDMs had a significant decrease from 1 hr to 6 hr post LPS stimulation ( Figure 1—figure supplement 3A ) . To further confirm this result , we performed polysome analysis followed by qRT-PCR to monitor the distribution TNF mRNA across the sucrose gradient ( Figure 1—figure supplement 3B ) . Compared with TNF mRNA at the 1 hr time point , the TNF mRNA had a significant shift from the polyribosome region to the mRNP region on the sucrose gradient at 6 hr post the LPS treatment ( Figure 1—figure supplement 3B and C ) , indicating that this transcript was translationally repressed . Importantly , the distribution of a control mRNA , Gapdh mRNA , did not change between the 1 hr and the 6 hr time points ( Figure 1—figure supplement 3C ) , which indicates the specificity of the translational repression on the TNF mRNA . Collectively , these observations indicated that a large number of mRNAs are regulated at the translational level during the inflammatory response . Translation of mRNA can be modulated by cis-elements in UTRs and trans-acting factors , such as RBPs and microRNAs . To identify the potential translational regulators in the inflammatory response , we performed the following computational analysis . First , we searched the 3’UTRs of the 724 differentially translated mRNAs for significantly enriched motifs ( adjusted p<10−100 , Fisher’s exact test ) that match with the known binding sites of ~200 RBPs identified through in vitro binding studies ( Ray et al . , 2013 ) , which generated a list of RBPs that can bind these mRNAs . Next , from this list of RBPs , we chose those that are expressed ( FPKM ≥10 ) in BMDMs , which resulted in a group of RBPs ( Figure 1F ) that may be potential translational regulators in the inflammatory response . Among these RBPs are several known translational regulators , such as Pcbp1 , Pcbp2 , and Cpeb4 ( Hu et al . , 2014; Makeyev and Liebhaber , 2002 ) , which supports the validity of this approach . Here , we focused on Zfp36 ( TTP ) , an ARE ( AU-rich element ) -binding protein , for several reasons . First , genetic studies in mouse indicated that Zfp36 is required to resolve inflammation ( Taylor et al . , 1996 ) . Second , Zfp36 is abundantly expressed in activated BMDMs ( FPKM >200 ) . Third , although Zfp36 was characterized to promote target mRNA degradation ( reviewed in Brooks and Blackshear , 2013 ) , its role in mRNA translational control is not well understood in primary BMDMs . To test whether Zfp36 can regulate mRNA translation , we performed the tethering experiments , which allow us to determine the function of a target RBP without knowing its substrate mRNAs ( Coller and Wickens , 2007 ) . Specifically , using the robust and specific interaction between the bacteriophage λN polypeptide and the BoxB RNA motif , we tethered a λN-Zfp36 fusion protein to the 3’UTR of a firefly luciferase ( FLuc ) mRNA ( Figure 1—figure supplement 4 ) . Luciferase activity and the FLuc mRNA level were then assayed in 293 T cells . Tethering Zfp36 reduced both the luciferase activity and the Fluc mRNA level ( Figure 1—figure supplement 4 ) . Importantly , the translatability of the Fluc mRNA , defined as the luciferase activity normalized to the Fluc mRNA level , was significantly decreased in a Zfp36-dependent manner ( Figure 1—figure supplement 4 ) , which indicated that Zfp36 can repress mRNA translation . This result is consistent with previous observations in the Zfp36 knockout BMDMs that the TNF mRNA , a Zfp36 target , increases ~two-fold compared with that in wild-type BMDMs; but the TNF protein increased ~five-fold ( reviewed in Taylor et al . , 1996 ) , which indicates that Zfp36 also represses target mRNA translation in primary BMDMs . Next , we aimed to determine how Zpf36 represses mRNA translation in primary BMDMs and to identify the mRNA targets of endogenous Zfp36 . Previous studies using immunoprecipitation ( IP ) of overexpressed Zfp36 or the yeast two-hybrid system have identified many proteins with which Zpf36 can interact ( reviewed in Brooks and Blackshear , 2013 ) . Although useful mechanistic insights can be obtained , there are three caveats in interpreting these previous results: ( 1 ) cell lines may not represent the corresponding primary cells; ( 2 ) proteins binding to the overexpressed Zfp36 may not interact with the endogenous Zfp36; and ( 3 ) for many identified interactions , their effect on gene expression are still unknown . Similar limitations also apply to the mRNA targets identified in Zpf36-overexpressing macrophage-like and nonmacrophage cell lines ( Mukherjee et al . , 2014; Tiedje et al . , 2016 ) . One significant technical challenge of studying the endogenous Zpf36 , however , is the unavailability of IP-grade antibodies for specific and efficient Zfp36 isolation . To characterize the endogenous Zfp36-mediated regulation of gene expression in a biologically relevant setting and to overcome the technical obstacles , we created a V5-epiptope tag knock-in mouse using CRISPR/Cas9-mediated genome editing ( Figure 2A ) . Specifically , a sgRNA targeting a genomic site near the stop codon region of the Zfp36 locus was coinjected into mouse zygotes with the Cas9 enzyme and an oligonucleotide containing a 51-nt V5-epitotpe tag with a ~ 60 nt homologous arm on each side of the V5 sequence . The double-stranded DNA break generated by the Cas9 facilitated in-frame knock-in of the V5-epitope tag via homologous recombination . The genetically modified zygotes were transferred into foster mouse mothers , and the Zfp36-V5 knock-in allele in the resulting mice was monitored by PCR ( Figure 2B ) . We genetically purified this Zfp36-V5 knock-in allele via backcrosses . The homozygous Zfp36-V5 mice were born in the predicted Mendelian ratio from heterozygous matings ( Figure 2—figure supplement 1A ) . Moreover , unlike Zfp36-deficient BMDMs ( Taylor et al . , 1996 ) , BMDMs from the Zfp36-V5 mice have the same response to LPS as those from wild-type mice in terms of inflammatory cytokine induction and Zfp36 expression ( Figure 2—figure supplement 1B ) , which indicates that the Zfp36-V5 in the knock-in mice is expressed at the same endogenous level as Zfp36 is in wild-type mice . Furthermore , the Zfp36-V5 mice are phenotypically indistinguishable from the wild-type mice . These observations are consistent with the notion that the 51-nt V5-tag knock-in sequence neither alters the Zfp36 promoter nor changes the regulatory elements in the 3’UTR . 10 . 7554/eLife . 27786 . 008Figure 2 . Generation of a Zfp36 V5-epitope tag knock-in mouse for mechanistic studies on the endogenous Zfp36 . ( A ) Workflow for generating a Zfp36-V5 knock-in mouse via CRISPR/Cas9-mediated genome editing . ( B ) Genotyping of the Zfp36-V5 knock-in mice using the two primers shown in ( A ) . ( C ) Specific and unambiguous detection of Zfp36 using the V5-epitope in BMDMs from the Zfp36-V5 mice . BMDMs from wild-type and Zfp36-V5 mice were treated with 100 ng/ml LPS for 4 hr , followed by Western blot using an anti-V5 antibody . The doublet bands of Zfp36 are due to its post-translational modifications . ( D ) Efficient isolation of endogenous Zfp36 using the V5-epitope in BMDMs from the Zfp36-V5 mice . BMDMs from the Zfp36-V5 mice were stimulated with 100 ng/ml LPS for 4 hr , followed by UV254 crosslinking . An anti-V5 antibody and two different IgGs were used to IP the endogenous Zfp36 from the cell lysates , respectively . The input , IP samples , and supernatants were subjected to SDS-PAGE and Western blot analysis using an anti-V5 antibody . The following figure supplement is available for Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 27786 . 00810 . 7554/eLife . 27786 . 009Figure 2—figure supplement 1 . The Zfp36-V5 mouse is normal . ( A ) The Zfp36-V5 mice were born in the predicted Mendelian ratio from heterozygous matings . The Zfp36-V5 mice from heterozygous matings were genotyped at 3–4 weeks old . ( B ) BMDMs from the Zfp36-V5 mice have the normal inflammatory response . BMDMs from the Zfp36-V5 mice and the wild-type mice were treated with 100 ng/ml LPS , and the cells were harvested at 0 , 1 , 2 , 4 , 6 , 8 hr post LPS treatment . The levels of the indicated mRNAs were quantified by qRT-PCR , and 18S rRNA was used for normalization . The results represent the means ( ±SD ) of three independent experiments . ( C ) Zfp36 can be unambiguously detected in the Zfp36-V5 mouse . BMDMs from the Zfp36-V5 mice were treated with 100 ng/ml LPS , and cells were harvested at 0 , 2 , 4 , 6 hr post LPS treatment for SDS-PAGE and Western blot using the indicated antibodies . DOI: http://dx . doi . org/10 . 7554/eLife . 27786 . 009 The V5-tag allows precise detection and efficient isolation of endogenous Zfp36 . Using a V5 antibody , we could unambiguously detect the endogenous Zfp36 in the BMDMs from the Zfp36-V5 mice ( Figure 2C , Figure 2—figure supplement 1C ) . Importantly , unlike a Zfp36 antibody , the V5 antibody resulted in no nonspecific bands ( Figure 2—figure supplement 1C ) , which indicates the high specificity of Zfp36 detection . Furthermore , using the V5 antibody , we could also efficiently isolate endogenous Zfp36 from UV-crosslinked ( see below ) LPS-treated BMDMs , as indicated by the distribution of Zfp36 in the input , IP , and supernatant samples ( Figure 2D ) . Finally , the Zfp36-V5 mice provide an almost unlimited source of primary cells ( that is , BMDMs ) . Collectively , these advantages make the Zfp36-V5 mouse a valuable tool for mechanistic characterization of endogenous Zfp36 in biologically relevant settings . To determine how Zfp36 regulates gene expression in primary BMDMs , we first identified the proteins that the endogenous Zfp36 interacts with by using IP followed by quantitative proteomics . Specifically , BMDMs from the Zfp36-V5 mice were harvested after 4 hr of LPS stimulation , when Zfp36-V5 is abundantly expressed ( Figure 2—figure supplement 1C ) . The Zfp36 and its associated proteins were IPed using the V5 antibody or IgG as control . Due to Zfp36’s RNA-binding ability , the IP was performed with RNaseA , which disrupts RNA-mediated interactions , so that only protein-protein interactions can be isolated . The IPed proteins were subject to tandem mass-tag ( TMT ) -based quantitative proteomic analysis ( Thompson et al . , 2003 ) to identify the proteins specifically associated with Zfp36 ( Figure 3A ) . In this method , the proteins isolated by the V5 antibody or IgG were digested by trypsin . The resulting peptides from each sample were labeled with a different isobaric mass tag , followed by pooling together for mass spectrometry ( isobaric tags for relative and absolute quantification , iTRAQ ) identification and quantification . The unique mass tag of each sample enables comparison of the abundance of the identified proteins in the V5 IP versus the IgG control samples . Using this approach , we identified several proteins that are significantly enriched ( >2 fold in all the three biological replicas ) in the V5 IP samples ( Supplementary file 2 ) . As the positive control , Zfp36 itself had greater than 20-fold enrichment , which indicates the validity of this method . The next-most enriched ( >4 fold ) and abundantly detected protein was Pabpc1 , followed by 14-3-3θ ( Figure 3B ) . Hereafter , we focused on Pabpc1 , the cytoplasmic poly ( A ) binding protein 1 , because of its roles in controlling mRNA translation and stability ( Mangus et al . , 2003 ) . 10 . 7554/eLife . 27786 . 010Figure 3 . Endogenous Zfp36 interacts with Pabpc1 in LPS-stimulated BMDM . ( A ) Workflow for the tandem mass tag ( TMT ) quantitative proteomics to identify the proteins associated with endogenous Zfp36 in LPS-stimulated BMDMs . ( B ) Zfp36-associated proteins identified by the TMT quantitative proteomics . ( C ) Zfp36 interacts with endogenous Pabpc1 in LPS-stimulated BMDMs . BMDMs from the Zfp36-V5 mice or wild-type ( WT ) mice were treated with 100 ng/ml LPS for 4 hr , followed by IP with ( + ) or without ( - ) RNaseA ( 200 ng/ml ) using an anti-V5 antibody . The input ( 5% ) and IP products were subject to SDS-PAGE and Western blot analysis using the indicated antibodies . ( D ) The Zfp36-Pabpc1 interaction is independent of RNA . The plasmid expressing the FLAG-tagged wild-type Zfp36 or a mutant , Zfp36 ( F118N ) , was co-transfected with a plasmid expressing the HA-tagged Pabpc1 into 293 T cells , respectively . IP was performed with ( + ) or without ( - ) RNaseA ( 200 ng/ul ) using either an anti-FLAG antibody or an IgG . SDS-PAGE and Western blot were used to analyze the indicated proteins in the input and IP samples . ( E ) Schematic presentation of Zfp36 truncations for mapping the region interacting with Pabpc1 . ( F ) Identification of the region on Zfp36 that binds Pabpc1 . The plasmids expressing FLAG-tagged Zfp36 truncations shown in ( E ) were co-transfected with a plasmid expressing the HA-tagged Pabpc1 into 293 T cells , respectively . IP was performed with RNaseA ( 200 ng/ml ) using an anti-FLAG antibody or an IgG . SDS-PAGE and Western blot were used to analyze the indicated proteins in the input and IP samples . The following figure supplement is available for Figure 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 27786 . 01010 . 7554/eLife . 27786 . 011Figure 3—figure supplement 1 . The Zfp36-Pabpc1 interaction is independent of RNA . A HA-Pabpc1 expressing plasmid was co-transfected into 293 T cells with a Zfp36-FLAG expressing plasmid . IP was performed using either an IgG control or an anti-FLAG antibody in the presence ( + ) or absence ( - ) of RNase1 ( 0 . 5 U/ul ) . Western blot was used to examine the indicated proteins in the input and IP samples . DOI: http://dx . doi . org/10 . 7554/eLife . 27786 . 011 To verify the RNA-independent interaction between Zfp36 and Pabpc1 , we performed two additional experiments . First , we performed IPs in stimulated BMDMs from the Zfp36-V5 mice and wild-type mice with or without RNaseA using the V5 antibody . Endogenous Pabpc1 was detected by Western blot in the IP samples from the Zfp36-V5 BMDMs in an RNaseA-insensitive manner but not in that from the wild-type BMDMs ( Figure 3C ) . Importantly , Gapdh , an abundant cytoplasmic protein , was not detected in the IP samples , which indicates the specificity of the IPs ( Figure 3C ) . This result showed that the endogenous Zfp36-Pabpc1 interaction is not mediated by RNA . To further confirm that the interaction between these two RBPs is RNA-independent and to rule out the possibility that this interaction is mediated by the V5 tag , we performed IPs in 293 T cells expressing an HA-tagged Pabpc1 with either a FLAG-tagged Zfp36 or a FLAG-tagged RNA-binding-deficient Zfp36 mutant ( Zfp36[F118N] ) ( Lai et al . , 2002 ) . We found that the HA-tagged Pabpc1 was specifically co-IPed with both the wild-type and the mutant Zfp36 in the presence of RNaseA ( Figure 3D ) . To further rule out the possibility that the Zfp36-Pacbpc1 interaction is mediated by poly ( A ) sequences that RNaseA cannot degrade , we performed the IP experiment with RNAse1 , which degrades RNA in a non-sequence-specific manner . We found that Zfp36 still interacted with Pabpc1 under this condition ( Figure 3—figure supplement 1 ) . Collectively , these experiments revealed that the Zfp36-Pabpc1 interaction is truly RNA-independent . Since Pabp1c is the most abundant Zfp36-associated protein identified by the quantitative proteomics , it argues that the endogenous Zfp36-Pabpc1 interaction is a direct interaction in the activated BMDMs . To define a minimal region on Zfp36 that can specifically interact with Pabpc1 , we performed structure-function mappings . In detail , a series of FLAG-tagged Zfp36 truncations ( Figure 3E ) were coexpressed in 293 T cells with an HA-tagged Pabpc1 for co-IP , respectively . We took two steps to ensure that the identified interaction is RNA-independent . First , all the Zfp36 truncations had the F118N mutation ( Zfp36-m ) , which abolishes RNA-binding ability ( Lai et al . , 2002 ) . Second , the IPs were performed with RNaseA . This analysis resulted in a minimal region from amino acids 39 to 196 on Zfp36 ( Zfp36-mF ) that can be stably expressed and specifically interacts with Pabpc1 ( Figure 3F ) . To test whether the Zfp36-Pabpc1 interaction is required for the Zfp36-mediated translational repression , we performed two experiments . First , using the λN polypeptide and the BoxB RNA motif tethering system , we tethered Zfp36 and 2 Zfp36 truncations that cannot interact with Pabpc1 ( Figure 4—figure supplement 1 ) , separately , to the 3’UTR of an FLuc mRNA , and then measured the luciferase activity , the Fluc mRNA level , and the translatability ( Figure 4A ) . We found that unlike Zfp36 , the two truncations did not specifically repress the translation of the Fluc mRNA ( Figure 4B ) . This result is consistent with the notion that the Zfp36-Pabpc1 interaction is functional . 10 . 7554/eLife . 27786 . 012Figure 4 . The Zfp36-Pabpc1 interaction is required for Zfp36-mediated translational repression . ( A ) Luciferase reporters and Zfp36 and its truncations used in the tethering experiment . The blue box represents the λN polypeptide . ( B ) Luciferase activity , the FLuc mRNA , and translatability determined in the tethering experiment . The FLuc-5BoxB reporter plasmid or the control plasmid was co-transfected with either the λN-GFP plasmid or plasmids expressing λN-Zfp36 and its truncations shown in ( A ) into 293 T cells , respectively . The luciferase assay and the mRNA measurement were performed at 24–30 hr post-transfection . The translatability was calculated as the luciferase activity normalized by the FLuc mRNA level . The luciferase activity , FLuc mRNA , and the translatability from the λN-GFP expressing cells were set as 1 for relative quantification , respectively . ( C ) The FLuc-5BoxB-MALAT1 reporter mRNA is a poly ( A ) - transcript . Total RNA from the 293 T cells transfected with the FLuc-5BoxB-MALAT1 reporter plasmid were fractionated by oligod ( T ) 25 magnetic beads . RNAs were quantified by qRT-PCR in the poly ( A ) + and poly ( A ) -fractions . The RNA level in the poly ( A ) - fraction was set as 1 for relative RNA level calculation . ( D ) IP of HA-Pabpc1 in 293 T cells . A HA-Pabpc1 expressing plasmid was co-transfected into 293 T cells with either the FLuc-5BoxB-SV40pA reporter or the FLuc-5BoxB-MALAT1 reporter . IP was performed using an anti-HA antibody , and Western blot was used to examine the indicated proteins in the input and IP samples . ( E ) Pabpc1 does not bind the FLuc-5BoxB-MALAT1 reporter mRNA . qRT-PCR was performed on indicated mRNAs from the total RNA isolated from the input and IP samples of ( D ) . ( F ) Zfp36 cannot repress the translation of the FLuc-5BoxB-MALAT1 mRNA . The luciferase , mRNA level , and translatability of the FLuc-5BoxB-MALAT1 mRNA were determined in the tethering experiment as described in ( B ) . All results represent the means ( ± SD ) of three independent experiments . *p<0 . 05 , n . s . not significant ( p>0 . 05 ) by the Student’s t-test . The following figure supplement is available for Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 27786 . 01210 . 7554/eLife . 27786 . 013Figure 4—figure supplement 1 . The Zfp36 N-terminus and C-terminus fragments do not interact with Pabpc1 . ( A ) Schematic presentation of Zfp36 full length , the Zfp36 N-terminus fragment , and the Zfp36 C-terminus fragment for the CoIP experiment . ( B ) The Zfp36 N-terminus and C-terminus fragments do not interact with Pabpc1 . The Zfp36 full length and fragments shown in ( A ) were transfected into 293 T cells with a plasmid expressing Pabpc1-HA , respectively . Then IP was performed using an anti-FLAG antibody with RNaseA , followed by SDS-PAGE and Western blot to detect the indicated proteins . The bands labeled with * are likely to be protein degradation products . DOI: http://dx . doi . org/10 . 7554/eLife . 27786 . 013 Next , we speculated that if the Zfp36-Pabpc1 interaction is important , then Zpf36 could not repress the translation of the transcripts that are devoid of Pabpc1 , such as mRNAs without poly ( A ) tails . To test this hypothesis , we created a poly ( A ) minus ( poly ( A ) - ) mRNA by replacing the SV40 polyadenylation signal on the FLuc reporter ( FLuc-5BoxB-SV40pA ) with a sequence from the 3’ end of MALAT1 , which can result in cytoplasmic transcripts without poly ( A ) tails ( Figure 4C ) ( Wilusz et al . , 2012 ) . Indeed , when the resulting FLuc-5Box-MALAT1 mRNA was fractionated by oligod ( T ) magnetic beads , most ( >95% ) remained in the poly ( A ) - fraction , which is similar to poly ( A ) - transcripts , such as histone mRNA ( H2ab ) and 18S rRNA but different from the Gapdh transcript , a poly ( A ) + mRNA ( Figure 4C ) . To further determine that the poly ( A ) - FLuc-5Box-MALAT1 mRNA is not associated with Pabpc1 , we performed RNA IPs on the 293 T cells expressing an HA-Pabpc1 with either the FLuc-5BoxB-SV40pA mRNA or the FLuc-5BoxB-MALAT1 mRNA . We found that the HA-Pabpc1 can pull down significant amounts of the poly ( A ) + transcripts , such as the FLuc-5BoxB-SV40pA mRNA and Gapdh mRNA , but not the poly ( A ) - transcripts , such as the FLuc-5BoxB-MALAT1 mRNA and histone mRNA ( H2ab ) ( Figure 4D and E ) . These results indicated that Pabpc1 does not bind the poly ( A ) - FLuc-5BoxB-MALAT1 mRNA . When Zfp36 was tethered to the FLuc-5BoxB-MALAT1 mRNA , we found that no change in luciferase activity , FLuc mRNA level , or translatability compared with those from tethering a control protein ( GFP ) ( Figure 4F ) , which indicates that Zfp36 cannot repress the translation of the poly ( A ) - mRNA that is not associated with Pabpc1 . Together , these results argue for the functional importance of the Zfp36-Pabpc1 interaction in the Zfp36-mediated translational repression . To determine the functional relevance of the Zfp36-Pabpc1 interaction on the expression of Zfp36 target mRNAs , we first identified the mRNAs that endogenous Zfp36 binds in activated BMDMs . Specifically , we performed CLIP-seq ( Darnell , 2010 ) in Zfp36-V5 BMDMs . In this method , the activated BMDMs were first crosslinked by UV ( 254 nm ) , which only introduces covalent bonds between RNA and proteins that are in direct contact with each other ( Darnell , 2010 ) . The protein-free regions of RNAs were digested by RNase1 in the cell lysate . Then , the Zfp36 and its associated RNA fragments were IPed by the V5 antibody . The covalent link between Zfp36 and its associated RNA fragments allows stringent washes during the IP step , which substantially decreases nonspecific interactions . The RNA fragments were then isolated and cloned for high-throughput sequencing . A total of ~5 million uniquely mapped reads were obtained from two independent CLIP-seq experiments ( Figure 5A ) . Importantly , the CLIP results from these two biological replicas were similar to each other but different from RNA-seq on total RNA , which indicates high reproducibility of the data ( Figure 5B ) . 10 . 7554/eLife . 27786 . 014Figure 5 . Transcriptome-wide identification of Zfp36 target mRNAs using CLIP-seq . ( A ) A Table summarizing the number of uniquely mapped reads to the mouse genome for the CLIP-seq duplicates and RNA-seq . ( B ) Log plots of the number of uniquely mapped reads per gene from the CLIP-seq duplicates and the mRNA-seq . Each dot represents a gene . Pearson’s correlation coefficients ( Cor ) are indicated . ( C ) CLIP-seq reads distribution within protein coding genes . For the CLIP-seq reads uniquely mapped to mRNAs , their percentages in the 3’UTR , 5’UTR , CDS , and intron regions are shown . ( D ) Over-represented Zfp36-binding motifs identified by CLIP-seq . The motif was identified by MEME analysis on the Zfp36 binding clusters within mRNAs . The E-value is 1 . 6e-22 . ( E ) Gene ontology analysis of the Zfp36 target mRNAs in activated BMDMs . DOI: http://dx . doi . org/10 . 7554/eLife . 27786 . 014 We identified 280 mRNAs with clear Zfp36-binding peaks present in both of the two independent CLIP-seq experiments ( Supplementary file 3 ) . These mRNAs include well-characterized Zfp36 targets , such as TNF , IL-6 , and Zfp36 itself . Most of the binding peaks are located within the 3’UTRs of the target mRNAs ( Figure 5C ) . Critically , motif analysis revealed that the most significantly enriched sequence motif within the Zfp36-binding peaks is UAUUUAUU ( Figure 5D ) . Although UV-C ( such as UV254nm ) induced crosslinking preferentially occurs at uridines ( Sugimoto et al . , 2012 ) , the U-rich motif we identified ( Figure 5D ) is consistent with the Zfp36-binding motifs determined from in vitro RNA-binding studies on Zfp36 ( reviewed in Brooks and Blackshear , 2013 ) , strongly arguing that the UAUUUAUU sequence is also the in vivo binding motif of Zfp36 in the activated BMDMs . Gene ontology analysis revealed that the Zfp36 target mRNAs are involved in several important pathways in inflammatory responses , such as cytokine and chemokine signaling pathways ( Figure 5E ) . Collectively , these results indicated the validity of the CLIP-seq data and identified mRNA targets that endogenous Zfp36 binds in the activated BMDMs . To determine whether the Zfp36-Pabpc1 interaction is functionally relevant in controlling Zfp36 target mRNA expression in activated BMDMs , we first attenuated this endogenous interaction . Specifically , we expressed the minimal Zfp36 region ( Zfp36-mF ) that can interact with Pabpc1 ( Figure 3E , F ) in Zfp36-V5 BMDMs using a retroviral vector ( Figure 6A ) . Importantly , the Zfp36-mF contains a point mutation ( F118N ) that abolishes the RNA-binding activity ( Lai et al . , 2002 ) , which eliminates the complications of competitive binding for Zfp36 target mRNAs . IP with RNaseA indicated that the endogenous Zfp36-Pabpc1 interaction was decreased by half in the Zfp36-mF-expressing BMDMs compared with that in control BMDMs ( Figure 6B , C ) . 10 . 7554/eLife . 27786 . 015Figure 6 . The Zfp36-Pabpc1 interaction is important for Zfp36-mediated translational repression in LPS-stimulated BMDM . ( A ) Expression of the Zfp36-mF fragment in LPS-stimulated BMDMs . BMDMs from the Zfp36-V5 mice were transduced with either an empty retroviral vector or a retroviral vector expressing the Zfp36-mF fragment ( Figure 3E ) . The transduced BMDMs were treated with LPS for 4 hr , and the expression level of the Zfp36-mF was monitored by Western blot using the indicated antibodies . ( B ) Zfp36-mF attenuates the endogenous Zfp36-Pabpc1 interaction in activated BMDMs . The Zfp36 was IPed with RNaseA ( 200 ng/ml ) from the cell lysates of the LPS-stimulated transduced BMDMs in ( A ) using an anti-V5 antibody . SDS-PAGE and Western blot were used to analyze the indicated proteins in the input and IP samples . ( C ) Quantification of the Zfp36-Pabpc1 interaction . The IPed endogenous Pabpc1 were quantified by the ImageJ , and the Pabpc1 intensity from the BMDMs expressing the empty vector was set as 1 for relative quantification . ( D ) Polysome analysis of Zfp36-mF expressing primary BMDMs . ( E ) Zfp36 target mRNAs are translationally up-regulated in Zfp36-mF expressing BMDMs . The mRNA distribution in the mRNP , the 80S , and the polyribosome fractions ( shown in D ) were quantified by qRT-PCR in the activated BMDMs expressing either Zfp36-mF ( + ) or an empty retroviral vector ( - ) , respectively . ( F ) Zfp36 target mRNA levels were not dramatically changed in Zfp36-mF expressing BMDMs . TNF and IL-6 mRNAs were quantified by qRT-PCR in the activated BMDMs expressing either Zfp36-mF or an empty vector , respectively . 18S rRNA was used for normalization . ( G ) Proteins from Zfp36 target mRNAs are increased upon attenuating the Zfp36-Pabpc1 interaction . TNF and IL-6 proteins were quantified by ELISA in the cell lysate and supernatant of the Zfp36-mF expressing BMDMs and the control BMDMs , respectively . The results represent the means ( ± SD ) of three independent experiments . *p<0 . 05 , and n . s . not significant by the Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 27786 . 015 Next , we monitored the ribosome association of Zfp36 target mRNAs by sucrose-density-gradient-mediated polysome profiling . Expression of the Zfp36-mF fragment did not change the global polysome profile in the activated BMDMs ( Figure 6D ) , consistent with the notion that this fragment does not bind RNA . We then examined the distribution of seven Zfp36 target mRNAs across the gradient . These mRNAs are abundantly expressed and have strong Zfp36-binding peaks within their 3’UTRs . Upon weakening the Zfp36-Pabpc1 interaction , all seven of the Zfp36 target mRNAs had significant shifts from the mRNP region , which is devoid of translocating ribosome , to the polyribosome region , where active translation occurs , on the sucrose gradient ( Figure 6E ) . Critically , the distribution of Gapdh mRNA , which is not a Zfp36 target , did not change upon Zfp36-mF expression ( Figure 6E ) . These observations indicated that attenuation of the Zfp36-Pabpc1 interaction resulted in a specific increase of ribosome association on Zpf36 target mRNAs . To further verify the functional importance of the Zfp36-Pabpc1 interaction , we measured the mRNA and protein levels of two Zfp36 targets: TNF and IL-6 . These two cytokines are critical players in the inflammatory response . Quantitative RT-PCR revealed that IL-6 mRNA was not significantly altered and TNF mRNA was slightly decreased ( to ~70% of control ) when the Zfp36-mF was expressed ( Figure 6F ) . Interestingly , however , at the protein level , both TNF and IL-6 were significantly increased in the cell lysates from the Zfp36-mF expressing BMDMs ( Figure 6G ) . Similar increases were observed in the supernatant from the Zfp36-mF-expressing cells ( Figure 6G ) , although the increase in IL-6 was not statistically significant . Thus , the combined results from the mRNA level and the protein level indicated that the translation efficiency of these two Zfp36 target mRNAs increased when the endogenous Zfp36-Pabpc1 interaction was attenuated . Collectively , these results revealed that the Zfp36-Pabpc1 interaction is required for the translational repression of Zfp36 target mRNAs in activated BMDMs . To determine which step ( s ) of translation Zfp36 interferes with , we used bicistronic reporters containing IRES ( internal ribosome entry site ) elements . IRES elements are structured RNAs present in many viruses that can position and activate eukaryotic translation initiation ( Fraser and Doudna , 2007 ) . Importantly , compared with canonical translation , different viral IRES elements require different subsets of translation initiation factors . Thus , by examining the sensitivity of various IRES-mediated translations to Zfp36 , we aimed to dissect which step ( s ) of translation are repressed by Zfp36 . Here , we used two IRES elements: the hepatitis C virus ( HCV ) IRES , which requires all the initiation factors except eIF4G and eIF4E for translation initiation , and the cricket paralysis virus ( CrPV ) IRES , which only needs the 40S ribosome subunit for translation initiation . These two IRES elements were inserted between an FLuc CDS and a renilla luciferase ( RLuc ) CDS on the bicistronic reporter , respectively ( Figure 7A ) . Thus , FLuc translation is controlled by the canonical cap-dependent translation , whereas RLuc is translated by IRES-mediated translation . In addition , the BoxB RNA motifs in the 3’UTR of the reporter enable specific tethering of either Zfp36 or a control protein ( GFP ) on the bicistronic reporter mRNAs using the λN polypeptide . When Zfp36 was tethered , the reporter mRNA levels did not change compared with those of GFP tethering ( Figure 7B ) . Interestingly , however , both the FLuc and the RLuc luciferase activities were significantly decreased in either the HCV-IRES reporter or the CrPV-IRES reporter ( Figure 7C ) . Since the 40S ribosome subunit is the only factor needed by the CrPV IRES element to initiate translation ( Fraser and Doudna , 2007 ) , these results strongly argue that Zfp36 either interferes with 40S ribosome subunit recruitment during translation initiation or inhibits the events at or after the 60S subunit joining step during mRNA translation . Interestingly , Pabpc1 also regulates these steps in facilitating mRNA translation ( Mangus et al . , 2003 ) . Thus , these observations indicate that Zfp36 represses translation at similar step ( s ) as Pabpc1 regulates translation . 10 . 7554/eLife . 27786 . 016Figure 7 . Zfp36 represses translation at similar steps as Pabpc1 regulates translation . ( A ) Schematic presentation of the bicistronic luciferase reporter system for dissecting how Zfp36 represses translation . ( B ) Tethering Zfp36 does not change the mRNA levels of the bicistronic reporters . The bicistronic reporter plasmid ( either the HCV reporter or the CrPV reporter ) was co-transfected with the λN-GFP plasmid or the λN-Zfp36 plasmid into 293 T cells , respectively . The reporter mRNA levels from the transfected 293 T cells were quantified by qRT-PCR , and 18S rRNA was used for normalization . ( C ) Zfp36 inhibits both canonical translation and IRES-mediated translation . The bicistronic reporter plasmid was co-transfected with either the λN-GFP plasmid or the λN-Zfp36 plasmid into 293 T cells , respectively . The firefly ( FLuc ) and renilla ( RLuc ) luciferase activities were measured at 24–30 hr post-transfection . The translatability is calculated as the luciferase activity normalized by the reporter mRNA level . ( D ) A model for Zfp36-mediated regulation of gene expression in activated BMDMs . All the quantification results represent the means ( ± SD ) of three independent experiments . *p<0 . 05 , and n . s . not significantly different by the Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 27786 . 016 Gene expression is precisely regulated during the inflammatory responses for controlling infections and limiting detrimental effects of inflammation . Previous studies identified two two layers of regulation at the transcriptomic level . First , in the early stage of inflammation , several critical transcriptional factors are activated to rapidly produce mRNAs of inflammatory response genes ( reviewed in Medzhitov and Horng , 2009 ) . Second , in the late stage of inflammation , mRNA stability regulatory circuits promote the degradation of mRNAs of many inflammatory cytokines thus preventing overproduction of cytokines that are potentially toxic to normal tissues and thereby contributing to resolving inflammation ( reviewed in Carpenter et al . , 2014 ) . Using parallel ribosome profiling and RNA-seq , here we identified hundreds ( 724 ) of differentially translated mRNAs in the primary BMDM-mediated inflammatory response . We think that this number may be an underestimate of mRNAs regulated at the translation level due to rapid degradation of translationally repressed mRNAs . In contrast to a few unique biological systems , such as oocytes , in which mRNA decay activity is minimal , in most somatic cells it is challenging to fully identify all the translationally regulated mRNAs at a steady state due to efficient and dynamic mRNA degradation . Nonetheless , among the 3’UTRs of the differentially translated mRNAs we identified , there are significantly enriched binding motifs of several RNA-binding proteins expressed in the activated BMDMs . These matched cis-elements and trans-acting factors strongly argue for the presence of translational regulatory networks during the inflammatory response . Zfp36 is an important factor we characterized in these translational regulatory networks . Previous studies from multiple labs indicate that Zpf36 not only is an inducer of mRNA degradation by recruiting mRNA decay factors , such as mRNA deadenylases and decapping factors ( reviewed in Brooks and Blackshear , 2013 ) , but also can function as a translational silencer ( Brooks and Blackshear , 2013; Franks and Lykke-Andersen , 2007; Fu et al . , 2016; Qi et al , 2012; Tao and Gao , 2015; Tiedje et al . , 2016 ) . Based on the results predominantly from cell lines and overexpression systems , multiple mechanisms of Zfp36-mediated translational repression have been proposed . For example , Zfp36 can repress target mRNA translation via interacting with the translation initiation factor 4E2 ( Tao and Gao , 2015 ) , binding the mRNA decapping activator RCK/p54 ( Qi et al , 2012 ) , recruiting the 4EHP-GYF2 cap-binding complex ( Fu et al . , 2016 ) , and delivering target mRNAs to processing bodies ( Franks and Lykke-Andersen , 2007 ) . Here , we focus on how endogenous Zfp36 regulates gene expression in mouse primary BMDMs . Using a knock-in mouse with a V5 epitope tag sequence inserted in-frame near the stop codon of Zfp36 , we found that Pabcp1 , the cytoplasmic poly ( A ) tail binding protein , is the most abundant RNA-independent binding partner with the Zfp36 in primary activated BMDMs . Critically , the Zfp36 expressed in the knock-in mouse is at its endogenous level , because the 51-nt knock-in V5 tag sequence neither changes the Zfp36 promoter nor eliminates the regulatory elements in the 3’UTR . Thus , this Zfp36-Pabpc1 interaction is an endogenous protein interaction in activated BMDMs . Surprisingly , however , we did not detect any mRNA decay factors or the previously identified proteins involved in translational repression in our quantitative proteomic analysis on endogenous Zfp36 . How can this result be reconciled with previous findings on the proteins associated with Zfp36 ? We think that this difference can be explained by the several possibilities . First , previous observations of the interactions between Zfp36 and mRNA decay factors and translational regulators were predominantly from the experiments of Zfp36 overexpression in cell lines , followed by either Co-IP or IP and mass spectrometry; or yeast two-hybrid screens ( reviewed in Brooks and Blackshear , 2013 ) . These results indicate that Zfp36 can interact with those factors . Our study , however , we focused primarily on the protein ( s ) that interact with endogenous Zfp36 in primary activated BMDMs . The differences in both the biological setting and the Zfp36 expression levels may contribute to the different results . Second , it is also possible that the interaction between Zfp36 and mRNA decay factors is transient and beyond detection by the quantitative proteomics we used . Consistent with this notion , mRNA decay is a highly dynamic process . Once the mRNA is destroyed , the decay factors will be released and assemble on the next transcript . Thus , it is likely that Zpf36 binds to the mRNA decay factors within a very specific time window during the life of the target mRNAs , whereas the proteins we identified are steady-state partners that interact with endogenous Zfp36 in primary activated BMDMs . The Pabpc1-interacting region on Zfp36 overlaps with its RNA-binding domain . Critically , we confirmed that the Zfp36-Pabpc1 interaction is not mediated by RNA . Interestingly , the Pabpc1-interacting domain ( in the middle ) and the domains that can interact with mRNA decay factors ( at the N-terminus and the C-terminus ) are located in different regions on Zfp36 . This observation suggests that Zfp36’s binding to Pabcp1 to repress mRNA translation ( which is characterized in this study ) and recruiting decay factors to promote mRNA degradation ( which is described in the literature ( reviewed in Brooks and Blackshear , 2013 ) are two different steps . Considering that mRNA translation repression can accelerate mRNA degradation ( reviewed in Coller and Parker , 2004; Richter and Coller , 2015 ) , we speculate that these two steps are intimately related and may occur in a sequential order in the Zfp36-mediatd regulation of gene expression . Thus , we propose the following model ( Figure 7D ) : once bound to target mRNAs , Zfp36 interacts with Pabpc1 to attenuate translation; then mRNA decay factors are recruited to destroy the transcripts . Thereby , Zfp36 facilitates the transition of its target mRNAs from a translation-competent state in the initial stage of inflammation to a degradation-prone state in the late stages of the inflammatory response . Since many Zfp36 target mRNAs are proinflammatory cytokines , such as TNF , we speculate that this Zfp36-mediated transition of its target mRNAs from an active state to an inactive state of protein production facilitates resolving inflammation . Targeting poly ( A ) -binding proteins by sequence-specific RBPs to repress mRNA translation is a new mechanism of translational control . Unlike our knowledge of regulating mRNA translation through the classical translational initiation factors , such as eIF4E and eIF4G , our understanding of modulating translation via Pabpc1 is very limited . Here , we determined that the Zfp36-Pabpc1 interaction is required for the Zfp36-mediated translational repression in primary activated BMDMs . Using reporter assays , we found that Zfp36 represses translation at either late stage ( s ) of initiation or postinitiation step ( s ) . This result is consistent with Pabpc1’s roles in translation by enhancing ribosome recruitment to the mRNA at both the 40S ribosome subunit recruitment step and the 60S subunit joining step ( Kahvejian et al . , 2005; Mangus et al . , 2003 ) . Thus , we speculate that Zfp36 represses mRNA translation by compromising Pabpc1’s ability to maintain mRNA in a translation-competent state . Interestingly , Pabpc1 functions at the interface of controlling mRNA translation and stability by protecting mRNA from deadenylases and stimulating mRNA translation ( Mangus et al . , 2003 ) . These unique functions make Pabpc1 an ideal target for shutting down gene expression posttranscriptionally . We think that in the activated BMDMs , through modulating Pabpc1 , Zfp36 can rapidly inhibit the translation of its target mRNAs and promote their degradation to resolve inflammation . Among the 280 Zfp36 target mRNAs identified by CLIP-seq in the activated BMDMs , we noticed that not every one is translationally repressed during the inflammatory response ( the differentially translated genes in the Supplemental file 1 versus the mRNAs listed in the Supplemental file 3 ) . This can be explained by several possibilities . First , it is possible that multiple RBPs can bind on the same mRNA , and the ultimate fate of the transcript is determined by the combinatory functions of all the associated proteins . Indeed , besides Zfp36 , we also predicted several other RBPs as potential translation regulators . Thus , it would be interesting to explore the functional interactions between Zfp36 and other RBPs that can regulate the same transcripts . These results will provide insight into how Zfp36’s activity is antagonized or stimulated during the inflammatory response . Second , not every Zfp36’s binding event may result in expression changes of the target mRNAs . Third , when we determined the differentially translated mRNAs during the inflammatory response , we focused on the mRNAs with the expression levels ≥ 10 FPKM at all the time points . We set this threshold because the apparent TEs ( Ribo-seq divided by RNA-seq ) of lowly expressed mRNAs have large variations between the two biological replicate samples . Although have high expression levels at late time-points in the inflammatory response , some of the Zfp36 target mRNAs , such as the IL-6 mRNA and the IL-1b mRNAs , are expressed at low level ( <10 FPKM ) at 0 hr . Thus , these transcripts are excluded from the list of the identified differentially translated mRNAs . It is important to mention that the mechanistic insights we obtained regarding Zfp36-mediated regulation of gene expression are different from previous studies in Zfp36 overexpressed cell lines ( Franks and Lykke-Andersen , 2007; Fu et al . , 2016; Qi et al , 2012; Tao and Gao , 2015 ) . The Zfp36-V5 knock-in mouse allowed us to focus on the endogenous Zfp36 in primary innate immune cells , making the molecular mechanisms from this study biologically relevant . We believe similar approaches can be applied to characterize molecular mechanisms of other RBPs that lack high-quality antibodies under biologically relevant settings . In summary , our study revealed widespread translational regulations of the primary macrophage-mediated inflammatory response . We found that Zfp36 , a critical RBP for resolving inflammation , mechanistically functions as a translation repressor by targeting the poly ( A ) -tail binding protein , Pabpc1 , to inhibit proinflammation gene production . Besides Zfp36 , we also identified additional RBPs that may be involved in translational regulatory networks during the inflammatory response . Future functional and mechanistic characterization of these RBPs will provide novel insights into gene expression in the innate immune response . The Zfp36-V5 knock-in mouse was generated via CRISP/Cas9-mediated genomic editing using the protocols described previously ( Wang et al . , 2013 ) . Specifically , Cas9 mRNA ( TriLink BioTechnologies , San Diego , CA ) ( 100 ng/μl ) , sgRNA derived from oWH2056 ( 50 ng/μl ) , and a donor oligo ( owH2057 ) containing the V5 tag sequence surrounded by a 60nt homologous arm on each side ( 50 ng/μl ) , were co-injected into mouse zygotes . Then the injected zygotes were transferred to mouse foster mothers . The resulting mice were genotyped at 3–4 weeks old to check the V5 tag integration at the Zfp36 locus . In this study , we predominantly used primary cells ( that is , mouse BMDMs ) isolated from 8 to 12 weeks B6 mice except for the 293 T cells used in the transfection-based assays . The 293 T cells used in this study were obtained from ATCC ( ATCC CRL-11268 ) ( RRID:CVCL_0063 ) . The cell identity was authenticated via short tandem repeat ( STR ) profiling . Mycoplasma contamination was tested as negative by a PCR method . BMDMs isolation , culture , and retroviral transduction were performed in accordance with the published protocols ( Weischenfeldt and Porse , 2008 ) . All the procedures on the isolation of BMDMs from mice were performed under a protocol ( A35015 ) approved by the Mayo Clinic animal welfare committee . The RiboZol RNA Extraction Reagent ( Amresco , Solon , OH ) was used for total RNA isolation . The residual genomic DNA was removed by DNase1 ( NEB , Ipswich , MA ) digestion . Reverse transcription was performed using the SuperscriptII ( Invitrogen , Waltham , MA ) and random hexamers . Quantitative PCR was performed using the 2 X SYBR Green qPCR Master Mix ( BioRad , Hercules , CA ) on a Bio-Rad CFX Real-Time PCR Detection System . The Protein G Dynabeads ( LifeTechnologies , Carlsbad , CA ) were used for all the immunoprecipitation assays in this study . The TMT-based quantitative mass spectrometry was performed at the proteomic core facility of the Whitehead Institute for Biomedical Research ( Cambridge , MA , USA ) . CLIP-seq was performed using the previously described protocols ( Moore et al . , 2014 ) with the following modifications: First , the BMDMs were crosslinked with 400 mJ/cm2 UV254 . Second , the IPed protein complexes were washed using the high stringent conditions . Ribosome profiling was performed as previously described ( Ingolia et al . , 2012 ) except that we used a sucrose density gradient ( 5–50% w/v ) to isolate the ribosome and the ribosome protected fragments . The computational analysis was performed in accordance with the pipelines described previously ( Alvarez-Dominguez et al . , 2017 ) . The sequencing data from these two sets of experiments are deposited at GEO ( Gene Expression Omnibus ) with the accession number GSE99787 . Polysome analysis was performed as described previously ( Hu et al . , 2014 ) . Briefly , BMDM cells were lysed in polysome lysis buffer ( 10 mM Tris-HCl , pH 7 . 4 , 12 mM MgCl2 , 100 mM KCl , 1% Tween-20 , and 100 mg/ml cycloheximide ) , and then the cell lysate was clarified by centrifugation at 21 , 130 g for 10 min at 4°C . The resulting supernatant was loaded onto a 5–50% ( w/v ) linear sucrose-density gradient and centrifuged at 39 , 000 rpm ( 260 , 000 g ) for 2 hr at 4°C in a Beckman SW-41Ti rotor . The gradient was fractionated using a Gradient Station ( BioComp ) with an ultraviolet detector ( Bio-Rad EM-1 ) . RNA from each collected fraction was isolated using the method described above .

DNA sequences called genes produce RNA molecules , some of which ( the “messenger RNAs” ) go on to be ‘translated’ to make proteins . This gene activity enables cells to react to their surroundings . For example , immune cells called macrophages produce hundreds of RNA molecules and proteins as part of an inflammatory response that defends the body against an infection . However , many of these molecules can also damage healthy tissue , so many layers of regulation control when , and how much of , these molecules are made . There are several ways to control how many proteins a cell produces . For example , cells might regulate how many messenger RNA molecules ( also called mRNAs for short ) are produced from a gene , or control how many proteins are translated from those mRNA molecules . Previous studies of how inflammatory responses are regulated have largely focussed on how mRNA production is controlled . Much less is known about the role that regulating mRNA translation has on the inflammatory response . By studying mouse macrophages , Zhang , Chen et al . have now identified hundreds of proteins whose production is regulated during an inflammatory response by controlling their translation from mRNA molecules . A group of RNA-binding proteins produced by the macrophages perform this regulation . Further observation revealed that a particular RNA-binding protein called Zfp36 prevents the translation of several important mRNAs , thereby helping to end an inflammatory response . Zhang , Chen et al . then genetically engineered mice to produce a version of Zfp36 that has a ‘tag’ attached to it that makes the protein easier to detect . Studying the activity of Zfp36 in these mice revealed that this RNA-binding protein works by interacting with another protein that normally binds to structures known as poly ( A ) tails at the end of the mRNA molecules . Zhang , Chen et al . believe that similar genetic engineering approaches could help researchers to study how other RNA-binding proteins work in living animals . In addition , by better understanding how inflammatory responses are regulated it may be possible to investigate new ways of treating conditions where this response is prolonged , such as in autoimmune disorders like rheumatoid arthritis .

2017

Translation repression via modulation of the cytoplasmic poly(A)-binding protein in the inflammatory response

Mutational activation of the BRAF proto-oncogene in melanocytes reliably produces benign nevi ( pigmented ‘moles’ ) , yet the same change is the most common driver mutation in melanoma . The reason nevi stop growing , and do not progress to melanoma , is widely attributed to a cell-autonomous process of ‘oncogene-induced senescence’ . Using a mouse model of Braf-driven nevus formation , analyzing both proliferative dynamics and single-cell gene expression , we found no evidence that nevus cells are senescent , either compared with other skin cells , or other melanocytes . We also found that nevus size distributions could not be fit by any simple cell-autonomous model of growth arrest , yet were easily fit by models based on collective cell behavior , for example in which arresting cells release an arrest-promoting factor . We suggest that nevus growth arrest is more likely related to the cell interactions that mediate size control in normal tissues , than to any cell-autonomous , ‘oncogene-induced’ program of senescence . Activating BRAF mutations ( e . g . BRAFV600E ) are the most common oncogenic mutations in melanoma , seen in about 66% of cases ( Davies et al . , 2002 ) . Curiously , the same mutation is found in 89% of melanocytic nevi ( Pollock et al . , 2003 ) —the benign , pigmented ‘moles’ found on the skin of most individuals . In animal studies , melanocyte-specific expression of BRAFV600E efficiently produces nevi , but only very rarely melanoma ( Dankort et al . , 2009; Dhomen et al . , 2009; Patton et al . , 2005 ) . The widely-accepted explanation is that transformed melanocytes undergo oncogene-induced senescence ( OIS ) , arresting proliferation before additional oncogenic events can occur ( e . g . Bennett , 2003; Huang et al . , 2017; Kaplon et al . , 2014; Michaloglou et al . , 2005 ) . Nevus melanocytes are indeed growth-arrested , but the assumption that OIS is the cause remains untested , in part because of a lack of criteria to rigorously define OIS in vivo ( Damsky and Bosenberg , 2017 ) . Initially studied as a consequence of forced expression of oncogenes in cell cultures ( Serrano et al . , 1997 ) , OIS has come to be seen as a distinctive cellular stress response characterized by a phenotype of growth arrest , morphological and metabolic changes , chromatin alterations , and secretion of growth factors , chemokines , cytokines and proteases ( Campisi and d'Adda di Fagagna , 2007; Gorgoulis et al . , 2019; Ito et al . , 2017; Kuilman et al . , 2010 ) . Given an abundance of ‘hallmarks’ of senescence , one might think that recognizing this cell state in vivo should be straightforward . Yet no single hallmark distinguishes senescence from other growth-arrested cell states . Phenotypes once thought to be ‘gold standards’ , such as expression of lysosomal beta-galactosidase , cyclin-dependent kinase inhibitors , or p53 , commonly mark only subsets of senescent cells ( Wiley et al . , 2017 ) , as well as non-senescent cells ( Tran et al . , 2012 ) . Moreover , observations of supposedly senescent cells resuming proliferation ( e . g . Beauséjour et al . , 2003 ) , imply that permanent cell cycle exit cannot be used as a distinguishing feature . In vivo senescence , as a result , is currently somewhat of a Gestalt diagnosis , that is assessed by a collection of traits , no subset of which is necessary or sufficient . Yet there is no clear consensus on which traits are best to assess , and recent meta-analyses of gene expression suggest that some of the most commonly assessed features are not ‘core’ to senescence at all ( Hernandez-Segura et al . , 2017 ) . The reason it is important to clarify how BRAF-transformed nevus melanocytes stop growing is that it shapes how we think about the origins of melanoma . OIS is usually portrayed in cell-intrinsic terms: oncogene expression within a transformed cell produces a stress within that cell , which triggers it to senesce . Even those who acknowledge a possible role for paracrine signals ( Acosta et al . , 2013; Elzi et al . , 2012; Ito et al . , 2017; Wajapeyee et al . , 2008 ) still portray the process as something initiated and orchestrated by cell-autonomous responses to oncogenes . This naturally leads to an approach to melanoma prevention and treatment that focuses on understanding how oncogenes derange intracellular processes; how those derangements elicit stress responses; and what might enable cancer cells to circumvent those responses ( e . g . Bennett , 2003; Damsky and Bosenberg , 2017; Vredeveld et al . , 2012; Yu et al . , 2018 ) . In contrast , as we argue below , it is possible that the growth arrest displayed by nevus melanocytes has little to do with oncogene-induced stress , and may have more to do with networks of cell–cell communication that are characteristic of melanocytes , independent of whether they are transformed . In this case , the most effective path to understanding how to prevent or treat melanoma could be to better elucidate the normal physiology of melanocytes in their environment . Here , we investigate the details of nevus growth arrest in a model in which melanocyte-specific Braf activation generates hundreds of nevi on the skin of mice ( Dankort et al . , 2009 ) . By examining both single-cell transcriptomes and the dynamics of growth arrest in nevus-associated melanocytes , we make two key observations: First , patterns of gene expression in arrested nevus melanocytes fail to identify them as any more senescent than other skin cells or normal melanocytes , arguing against a primary role for any form of senescence in their arrest . Second , the timing and statistics of nevus formation effectively argue against any relatively simple cell-autonomous process as being the cause of growth arrest . Ultimately , we propose a model in which arrest is driven not by oncogene stress , but by feedback mechanisms similar to those commonly involved in normal tissue homeostasis . Characterizing the dynamics of nevus growth and arrest requires observing nevi that started growing at known times . We took advantage of a mouse model in which Cre-mediated recombination introduces the activating V600E mutation into the endogenous Braf locus . When crossed onto a background carrying a Tyr-CreER transgene , the mice acquire the BrafV600E mutation only in cells of the melanocytic lineage , and only after Cre activation by 4-hydroxytamoxifen ( 4-OHT ) , applied either systemically or through painting on the skin . As shown previously ( Dankort et al . , 2009 ) , 4-OHT treatment of these mice leads to development of numerous pigmented nevi . Visualization of nevi is hindered , however , by the strong pigmentation in hair follicles which , except at microscopic resolution , can be difficult to distinguish from nevi . One way to circumvent this difficulty is to observe nevi only during the telogen phase of the hair cycle , when follicle-associated pigment is not present ( conveniently , synchronization of hair cycles may be maintained on a large patch of skin through depilation ) . As shown in Figure 1 , in mice whose back skin was treated with topical 4-OHT at postnatal day 2 ( P2 ) , P3 and P4 , nevi were apparent macroscopically at telogen ( P50; Figure 1A ) . Live imaging , using multi-photon microscopy ( MPM; Saager et al . , 2015 ) , revealed that , like human nevi , mouse nevi consist of scattered nests of pigment-containing cells ( Figure 1B ) . Nevi could also be visualized post-mortem , using a dissecting microscope , on the undersurface of pieces of telogen-stage back skin ( Figure 1C ) . An alternate approach to visualization that did not require hair synchronization was to generate nevi by painting 4-OHT on glabrous ( hairless ) skin , such as the ventral surface of the paw , permitting tracking of individual nevi on a daily basis . As shown in Figure 1D , when forepaws were treated with 4-OHT from P2 through P4 , tiny nevi could be detected as early as P6 . Serial observation indicated that most nevi reach a maximum size somewhere between P16 and P21 ( Figure 1D , Figure 1—figure supplement 1A ) . This suggests that , in the mouse , BrafV600E-transformed melanocytes arrest within 2–3 weeks . To confirm this , we used BrdU labeling to monitor DNA synthesis . Because melanin readily obscures immunohistochemical signals , these experiments were done in an albino ( unpigmented ) genetic background , using premelanosome protein ( Pmel ) staining to identify melanocytes . As shown in Figure 1F , albino mice generate nests similar to those seen in pigmented mice . In such animals , BrdU readily incorporated into hair follicle melanocytes ( Figure 1E , Figure 1—figure supplement 1B ) , whereas by p21 nests within nevi were uniformly negative for BrdU , implying growth arrest ( Figure 1F , Figure 1—figure supplement 1C ) . The conclusion that Braf-induced nevi are already growth-arrested by P21 agrees with the reports of others ( Damsky and Bosenberg , 2017 ) , and is lent further support by time course measurements of nest size by MPM ( Figure 1C ) , which show that nest size distributions change insignificantly between P21 and P50 ( Figure 4—figure supplement 1A-B ) . As discussed above , senescence is usually accompanied by distinctive gene expression . Various gene expression ‘signatures’ have been developed to help investigators identify senescent cells and distinguish them from cells that have become growth-arrested by other processes . We considered several of these ( Source data 1 ) : To determine whether any of these proposed signatures fits nevus melanocytes , we performed single-cell RNA sequencing on dissociated cells from the back skin of nevus-bearing mice at both P30 and P50 ( i . e . after nevi have stopped growing ) , using wildtype skin as a control . Using known cell-type marker genes ( Figure 2—figure supplement 1A-B ) , we identified 14 different cell types in the skin , including melanocytes ( Figure 2A ) . Unsupervised clustering further sub-divided the melanocytes into four groups ( Figure 2B ) : Two of them , Mel 0 and Mel 1 , were composed of cells found only in nevus-bearing , and not wildtype , skin ( Figure 2C ) ; they are highly similar in gene expression , primarily differing in having a slightly lower level of pigment gene expression in Mel 1 versus Mel 0 ( Figure 2D ) . We identify them as the ‘nevus melanocytes’ , because they are seen only when nevi are present , and are by far the predominant melanocyte population in such animals . Mel 2 cells express the lowest levels of pigmentation genes ( Figure 2D ) , and are seen in both genotypes ( Figure 2C ) at all stages ( although expanded in number in nevus-bearing animals ( Figure 2F ) ) . Their pattern of gene expression bears a strong resemblance to one recently published for melanocyte stem cells isolated from telogen-stage hair follicles ( Zhang et al . , 2020 ) . In particular , they express Cd34 , which has been proposed to be a marker for bulge-associated melanocyte stem cells ( Joshi et al . , 2019 ) . Finally , cells of cluster Mel 3 , which express the highest levels of pigment genes ( Figure 2D ) , are found in both mutant and wildtype mice , but only at the P30 time point ( Figure 2E–F ) . We thus identify them as mature hair follicle melanocytes , as such cells are present exclusively during anagen phase of the hair cycle ( P30 ) , and disappear during telogen ( P50 ) . Because gene signatures are based on the idea of up- and downregulation of expression relative to some baseline state , to test whether nevus melanocytes fit a known signature it is necessary to have comparison transcriptomes . We made two types of comparisons: nevus melanocytes versus every other cell type in the skin ( which , with the possible exception of mature keratinocytes , we would not expect to be senescent ) ; and the four melanocyte subclusters ( two of which are nevus-associated and two of which are not ) versus each other . In each case we computed average expression for each gene in every cell type or cluster , together with a standard error of the mean as a measure of dispersion . Expression values were then normalized to average expression across all of the cells being compared ( i . e . all skin cells , or all melanocytes , depending on which comparison was being done ) and log2-transformed , so that positive values signify upregulation ( relative to the average for that gene ) , and negative downregulation . Gene expression was then visualized using heat maps ( Figure 3A and Figure 3—figure supplements 1–2 , with positive values in blue and negative in red ) . Because gene expression levels inferred from single-cell RNA sequencing tend to be noisy , particularly for genes with low expression , we ranked all genes by their minimum level of noise ( i . e . normalized standard error of the mean in the least noisy cell type ) , and used this value ( ‘n-SEM’ , which is also presented graphically as a bar to the right of each heatmap ) to sort gene lists , so that maps vary from most to least reliable as one goes from top to bottom . Figure 3A shows the results for the ‘Classical’ and ‘Universal Up’ signatures ( heat maps for the other signatures are shown in Figure 3—figure supplement 1 ) . Here we see no strong enrichment of blue over red signals in nevus melanocytes , nor in most other cells . When compared with whole skin , using the ‘Classical’ signature , only Cdkn2a stands out as strongly upregulated in nevus melanocytes , but it is similarly upregulated in skin fibroblasts ( it also has the noisiest data among genes in the signature ) . With the ‘Universal Up’ signature , more genes are downregulated than upregulated in nevus melanocytes . To quantify such impressions , we summed the log2-transformed data in each column in every heat map , producing the bar graphs in Figure 3B . We reasoned that summation of log-transformed data would emphasize consistent trends in the data while suppressing effects of noise ( random positive and negative variation would tend to cancel out ) . The results suggest that skin fibroblasts better fit the ‘Classical’ senescent signature than any skin cell type , including nevus melanocytes , or indeed melanocytes of any cluster . As a control—to demonstrate the ability of this approach to correctly associate cell types with gene signatures— we analyzed the same data using a signature of cell proliferation , ‘meta-PCNA’ , that represents 129 human genes most positively correlated with proliferation marker PCNA in a compendium of normal tissues ( Venet et al . , 2011 ) . As shown in Figure 3B ( also see Figure 3—figure supplement 1 ) , this signature ( 122 genes of which had unambiguous mouse orthologs; Source data 1 ) identified two keratinocyte populations ( ‘IFE-cycling’ and ‘Outer Bulge 1’ ) as highly proliferative ( in agreement with Joost et al . , 2020 ) , and mature ( postmitotic ) keratinocytes as non-proliferative . Importantly , it also correctly identified nevus melanocytes as non-proliferative—and other melanocytes as proliferative—in agreement with Figure 1E–F . Interestingly , the relatively high expression of proliferation-associated genes in non-nevus , hair follicle melanocytes ( Mel 3 ) when compared with nevus melanocytes , was consistent between BrafV600E-expressing and control mice ( Figure 3—figure supplement 2 ) , suggesting that tissue context plays a role in whether BrafV600E-expressing cells even arrest growth . Figure 3C and Figure 3—figure supplement 1 extend this analysis to the remaining eight potential signatures of senescence ( five consisting of genes that are upregulated; three of genes that are downregulated ) . In five of the eight cases , nevus melanocytes rank as less senescent than the average skin cell; in two of the cases nevus melanocytes are about average . In only one case ( ‘Chatsirisupachai Down’ ) does nevus melanocyte gene expression go in the predicted direction for senescence . However , the Chatsirisupachai signatures had not been curated to remove genes associated simply with proliferation/quiescence ( Chatsirisupachai et al . , 2019 ) , and inspection of the ‘Chatsirisupachai Down’ gene list shows that 61 of its 250 genes are shared with the 129-gene meta-PCNA signature; that is it is more likely a signature of proliferation than ‘non-senescence’ ( note the strong similarity between the ‘Chatsirisupachai Down’ bar graph in Figure 3C and the meta-PCNA graph in Figure 3B ) . To confirm that the senescence-associated gene expression signatures used here truly could identify melanocytes that had become senescent , we also analyzed published data comparing gene expression in BRAFV600E-transduced and normal human melanocytes in culture , under conditions in which the former developed definitive morphological characteristics of senescence ( Pawlikowski et al . , 2013; see Figure 3—figure supplement 3 ) . Of the 23 ‘Universal Down’ signature genes , 15 were significantly differentially expressed , and 100% of these were decreased in expression . Of the 31 ‘Universal Up’ genes , 19 were significantly differentially expressed , and 84% of these were elevated in expression . Together these data do not support the view that any sort of senescence—oncogene-induced or otherwise—is characteristic of nevus melanocytes and therefore a possible cause of their growth arrest . As discussed above , OIS is usually presented as a cell-autonomous process ( e . g . Dankort et al . , 2009; Dhomen et al . , 2009; Michaloglou et al . , 2005; Serrano et al . , 1997 ) . The simplest cell-autonomous process that one might imagine is a probabilistic switch: Once oncogene activation commences , cells arrest with a fixed probability ( per time or per cell cycle ) . Regardless of the molecular details , such a model makes distinctive predictions about clonal dynamics . Consider the clonal descendants of a single oncogene-transformed founder cell . For any value of the per-cell-cycle senescence probability ( which we denote here as ‘s’ ) , how many cells should we expect that clone to contain at any given time ? How many cell cycles should it take before all of the cells in most clones should have arrested ? Such questions are well studied in mathematics ( Athreya and Ney , 1972 ) , and easily solved by computer simulation . For this particular problem , there are two key results ( Figure 4 ) . First , the time after which one can expect clones to have stopped growing ( e . g . when all cells will have arrested in , say , 95% or 99% of clones ) is a steep function of s . If s < 0 . 5 , ( i . e . less than a 50% chance of arrest per cell cycle ) , then some clones will never stop growing . If s is , say , 0 . 53 , all clones will eventually stop growing , but one must wait 51 cell cycles before 99% of them do so ( Figure 4A ) . Given typical lengths of postnatal mammalian cell cycles , and the fact that we observe cessation of mouse nevus growth in about 2–3 weeks , we may consider 30 to be a generous estimate for the maximum number of cell cycles by which nevi stop growing . To achieve 99% clonal arrest by 30 cell cycles , s must be around 0 . 56 or higher; to achieve arrest in 95% of clones , s must be greater than 0 . 52 ( Figure 4A ) . From the same simulations one may also calculate predicted distributions of clone sizes . There is a clear reciprocal relationship between mean clone size and the fraction of clones that arrest by 30 cell cycles of time ( Figure 4B ) . For example , a value of s that enables 95% of clones to arrest produces a mean clone size of only 18 . 5 cells . For comparison , we estimate cell numbers per nevus to be in the range of 100–1000 cells ( see Materials and methods ) . The explanation for the small mean clone sizes produced by simulations can be appreciated by examining the full size distributions . As shown in Figure 4C–D , such distributions are extremely heavy-tailed , with a very large number of very small clones and a small number of very large clones ( the histograms in Figure 4C–D are plotted with logarithmic abscissa to facilitate display of all clone sizes ) . Qualitatively , this is very different from what we observed for nevi on the backs of p21 mice . Nevi displayed a mean radius of 76 . 8 µm ( corresponding to an area of 0 . 019 mm2 , in excellent agreement with the results of Damsky et al . , 2015 ) and , when plotted on a logarithmic scale , individual radii displayed a Gaussian-like distribution ( Figure 4E; a Gaussian shape on a logarithmic axis is usually referred to as ‘log-normal’ ) . Interestingly , nest sizes ( quantified by MPM ) also seem to be log-normally distributed ( Figure 4—figure supplement 1 ) whether at P21 ( panel A ) or P50 ( panel B ) . So are the nests within human melanocytic nevi , despite the latter being an order of magnitude larger than those in mice ( Figure 4—figure supplement 1C ) . It should be noted that using different units to represent simulation results ( cell numbers ) and empirical data ( linear dimension ) in Figure 4 and its supplement does not confound comparing the shapes of distributions , thanks to the logarithmic abscissa: as long as cell number scales as some power of linear dimension , values associated with log-transformed bins are simply scaled by a constant factor . The above comparison of observed distributions with the results of computer simulation is not entirely fair , however: Simulations track all clones , no matter how small , whereas empirical distributions undoubtedly omit nevi with sizes below some threshold of observability . To correct for this , one can truncate simulated distributions to remove clones smaller than some threshold size . With no a priori way to know what threshold to use , we examined the entire range of plausible truncations ( up to the largest clone sizes ) . As it turns out , the relative shapes of simulated distributions were about the same regardless of where they were truncated . The reason for this behavior can be understood by displaying simulated distributions ( with bin sizes of one cell ) on a log-log plot , and observing that they fall , over most clone size ranges , on a straight line ( see Supplemental Material ) . This implies an approximately ‘power law’ relationship which , by definition , is scale-free , that is has the same relative shape over any range of observations . In fact , for s reasonably close to 0 . 5 , the approximate probability of observing any clone size can be shown analytically to vary inversely with the 3/2-power of size ( for derivation , see Appendix 1 ) . These data imply that observed nevus size distributions cannot be generated by any cell-autonomous , random , time independent , one-step process . But they do not speak to whether a more complicated random process , for example , one with several steps , might suffice . To address this , one can simulate clonal evolution under multi-step models . Again , dynamic predictions can be made . First , to achieve clonal stopping times within 30 cell cycles , the minimum per-step transition probability increases with the number of steps ( Figure 4F ) . For example , if it takes three random events to arrest growth in 99% of clones , the average probability of each event needs to be at least 0 . 66 per cell cycle; with five random events that number is 0 . 74 . Second , although distributions generated by such models still tend to be heavy-tailed ( Figure 4G ) , they become less so as the number of steps increases ( Figure 4H ) , gradually approaching something that looks log-normal . This makes mathematical sense: as per-cell-cycle probabilities approach unity , the system approaches a clock that simply ticks off a fixed number of cell cycles before stopping . A scenario in which all clones stop at roughly the same time , plus or minus some variation , necessarily produces a log-normal distribution , since the logarithm of cell size will be proportional to the number of cell divisions . To determine how many independent steps would be required for a random cell-autonomous process to produce distributions that fit those we observed for nevi , we simulated up to eight random stages , over a variety of transition probabilities ( Figure 4I ) . We subjected the results to a range of possible truncations , from 0 to 1600 cells , to mimic any observability cutoffs in the empirical data , and recorded the median clone sizes produced under each of these scenarios . As described below ( see Materials and methods ) , we estimate that the average nevus has about 500–1000 cells , but given possible errors in the estimate , we consider here a range of values between 100 and 3000 ( gray-shaded area in Figure 4I ) . Subject to the constraint that enough clones must arrest within 30 cell cycles , and that observation thresholds cannot be so high that the observed median is less than twice the threshold , we find that , to produce nevi of even 200 cells requires 4–5 independent events ( stages ) , depending on whether one requires 95% or 99% clonal arrest; to reach 500 cells requires 6–7 events . To reach even larger numbers—as would be found in human nevi , or in other mouse models ( Chai et al . , 2014 ) —would require even more stages . The above results indicate that , to generate in vivo-like distributions of nevi , a process something like a clock is needed , with cells either counting elapsed divisions ( or time ) since oncogene activation , or progressing through a sufficiently large sequence of random processes , with tightly controlled probabilities , so that the net outcome is clock-like . Cell-autonomous counting of cell cycles ( up to about 12 ) can occur in early , cleavage-stage embryos ( Tadros and Lipshitz , 2009 ) , but no mechanism has been described to enable growing ( as opposed to merely cleaving ) cells to track more than a small handful of divisions ( or the equivalent amount of time ) . Erosion of telomeres can mark the passage of large amounts of time in some cells , but this does not seem to occur to any significant degree in nevus melanocytes ( Michaloglou et al . , 2005 ) . In contrast , if growth arrest is not cell-autonomous , but driven by cell–cell communication , then clock-like behavior is easily achieved , without any sort of intrinsic cell memory: Consider a simple communication circuit in which every cell’s arrest probability is simply a monotonically increasing function of the number of cells around it that have already arrested ( Figure 5A ) . This mechanism describes a dynamically well-understood feedback process that normal tissues use to control size ( Lander , 2011; Lander et al . , 2009 ) . Termed ‘renewal control’ ( Buzi et al . , 2015 ) , because differentiated cells control the probability that progenitor cells self-renew , the process is often mediated by secreted TGF-β superfamily members such as myostatin , activin and GDF11 ( Gokoffski et al . , 2011; Lee et al . , 2005 ) . Because it implements the engineering principle of ‘integral negative feedback’ , renewal control produces highly robust final population sizes that are independent of parameters such as cell cycle speed or the starting numbers of cells ( Buzi et al . , 2015; Lander , 2011; Lander et al . , 2009 ) . When growth arrest due to renewal control is simulated as a probabilistic process ( Figure 5B ) , the observed size distributions of clones are very close to log-normal . This is because renewal control effectively enforces cell cooperation , so that once a small fraction of a clone has arrested , the entire clone stops soon thereafter . The resulting narrow distribution of stopping times produces size distributions that are approximately log-normal , that is that emulate a clock . This behavior is a generic outcome of feedback control , and does not depend strongly on the details of how feedback is implemented . Similar distributions are obtained whether we model nevi as progressing reversibly or irreversibly through more than one proliferative stage , or use agent-based simulations in which renewal is controlled by the concentration of a secreted molecule that accumulates according to the laws of diffusion and local uptake ( Figure 5C–D ) . The point of these simulations is not to argue in favor of a specific feedback mechanism , but rather to show that , where cell-autonomous mechanisms of arrest struggle to fit nevus dynamics , almost any sort of ( collective ) feedback does so easily . Although nevus size distributions alone cannot shed light on the molecular details of how feedback might be implemented in nevi , it is interesting that those cells that we identify as nevus melanocytes ( Figure 2 ) express multiple genes encoding ligands with known or suspected growth inhibitory activities , together with the receptors for those ligands . These include TGFβ superfamily members Gdf11 , Gdf15 , Tgfb1 , Tgfb2 , and Tgfb3 , as well as other genes associated with growth inhibition , such as Angptl2 , Angptl4 , Il6 , Sema3a , Sema3b , and Sema3f ( Attisano and Wrana , 1996; Neufeld et al . , 2016; Santulli , 2014 ) . The evaluation of such candidates ( as well as other genes expressed at too low a level to be reliably detected by single-cell RNA sequencing ) will no doubt require further study . In the meantime , we reasoned that any feedback mechanism based on secreted , diffusible factors should induce spatial correlations among clones . In particular , when clones ( or subclones ) get close enough to each other , they should inhibit each other’s growth , leading to a smaller final size . The distance over which such effects could occur should reflect the spatial decay length of diffusible molecules in the skin , which is thought to be on the order of no more than a few hundred microns ( Chen et al . , 2015 ) . Although our data on macroscopic nevi ( Figure 1D ) , which had been collected in a manner that included spatial coordinates , did not contain enough examples of nevus spacings in this range to test this hypothesis , our data on the nests within individual nevi did , as the median spacing between nests at postnatal day 21 was approximately 79 microns . To assess whether nests were significantly smaller when located near other nests ( especially large ones ) , we extracted the coordinates and nest areas from seven separate fields at P21 ( representing 122 individual nests ) and , modeling each nest as a disk of equivalent area , calculated the mean sizes of neighboring nests falling within successively larger annuli around each target nest . We compared the distributions of mean neighbor sizes near the 52 smallest nests ( radius <20 µm ) with the equivalent distributions around the 70 largest nests . As shown in Figure 5E–F , within annuli extending 45 µm away from the perimeters of target nests , we saw fewer examples of large neighbors ( radius >30 µm ) around large nests ( panel E ) than around small ones ( panel F ) . To determine whether the difference was statistically significant , we used a permutation test in which we randomly swapped nest areas ( but not locations ) within each field 5000 times , and recalculated the distributions . This allowed us to plot the envelope enclosing the 5th and 95th percentiles for permuted data , onto which we overlaid the observed data . Unlike the observed data , the envelopes of the permuted data ( blue zones in Figure 5E–F ) look similar whether target nests are large or small . Moreover , the observed data extended outside of the envelopes only for median neighbor sizes > 30 µm , with the data for small target nevi extending well above the relevant envelope and the data for large target nevi lying at to the bottom of the envelope ( Figure 5E–F , arrows ) . These results argue that proximity is associated with a small , but significant decrease in nest size , supporting the view that nests inhibit each other’s growth . Interestingly , when we repeated the same analysis using annulus sizes of 150 µm , differences in the sizes of neighbors of small and large nests were not seen , consistent with the view that whatever is promoting coordination among nests has a spatial range of <150 µm ( Figure 5—figure supplement 1 ) . Studies in man , mouse and fish establish that most melanocytic nevi form by mutational activation of BRAF , which triggers proliferation followed by growth arrest ( Damsky and Bosenberg , 2017; Dankort et al . , 2009; Dhomen et al . , 2009; Kaufman et al . , 2016; Michaloglou et al . , 2005; Patton et al . , 2005; Shain and Bastian , 2016 ) . Nevus growth is often considered a paradigmatic example of OIS , but here we question two of the major tenets of the OIS hypothesis: that nevus melanocytes are actually senescent; and that growth arrest is a direct effect of oncogene action on the individual cell . To assess whether nevi are senescent , we used single-cell RNA sequencing in a mouse model of Braf-driven nevus formation , comparing gene expression of nevus melanocytes with that of other cell types . Across a wide variety of gene expression signatures , especially those developed to distinguish senescence from other growth-arrested cell states , we failed to find any evidence in support of the OIS hypothesis . By gene expression criteria , nevus melanocytes were less senescent than many other normal skin cells , including non-nevus melanocytes ( Figure 3 , and its supplements ) . These results support earlier work that also questioned , based on immunohistochemical staining of human nevi for markers including lysosomal β-galactosidase , Ki67 , p16INK4a ( CDKN2A ) , γ-H2AX and p53 , whether nevus melanocytes should be considered senescent ( Tran et al . , 2012 ) . In agreement with other studies , we do find that Cdkn2A is highly expressed in nevus cells; it is in fact the only ‘classical’ senescence marker that clearly distinguishes nevus melanocytes from other melanocytes ( Figure 3A ) . Yet , as others have shown , Cdkn2A is neither necessary nor sufficient for oncogene-mediated melanocyte growth arrest ( Haferkamp et al . , 2009; Zeng et al . , 2018 ) . Thus , to the extent that nevus melanocytes do execute even part of a common senescence program , there is little to support the view that this why they stop proliferating . As for the question of exactly how Braf-induced nevus growth arrest occurs , Figure 6 presents a continuum of models: In model A , oncogene action elicits a cell-autonomous stress response which , after some time lag , triggers senescence that shuts proliferation down . This is the form in which the OIS hypothesis is most frequently presented . In model C , growth arrest is not a direct effect of oncogene action , but rather a consequence of growth itself . This type of feedback is commonly used by adult tissues to maintain constant size , and also enables developing tissues to produce precise numbers of differentiated cells ( Kunche et al . , 2016; Lander , 2011; Lander et al . , 2009 ) . Because of the collective nature of this mechanism—cells that have stopped dividing tell other cells in their vicinity to do likewise—it naturally produces semi-synchronous arrest of spatially-coherent cell clones , and the distinctive log-normal clone size distribution that comes along with that . In contrast , a purely cell-autonomous mechanism ( panel A ) has great difficulty producing such distributions ( Figure 4 ) , either necessitating the operation of some kind of multi-cell-cycle clock , or requiring cells to complete a long sequence of independent probabilistic events prior to arresting ( Figure 4 ) . One can , of course , build a model in between these two extremes ( model B ) , in which oncogenes induce growth arrest directly , but paracrine signals ( i . e . SASP factors ) help maintain it . If the paracrine role is important enough , this mechanism might also produce clone size distributions that are approximately log-normal , so we cannot categorically rule this model out . However , our gene expression data do not support any versions of it that have been explicitly proposed for nevi . So far , several groups ( working predominantly from in vitro observations ) have claimed a critical role for SASP factors in melanocyte OIS: Wajapeyee et al . , 2008 argued that IGFBP7 plays a necessary role in the establishment of BRAF-V600E-induced melanocyte senescence ( a conclusion disputed by some [Scurr et al . , 2010] ) ; Feuerer et al . , 2019 proposed that MIA ( melanoma inhibitory activity ) secreted by senescent melanocytes is required to maintain senescence; and Damsky and Bosenberg have proposed that IL1 , IL6 , IL8 ( encoded in mouse by Cxcl15 ) , and type 1 interferons produced by nevus cells play a role in their arrest ( Damsky and Bosenberg , 2017 ) . Our in vivo data do not support any of these hypotheses . For example , we observed that the vast majority of Igfbp7 transcripts are produced by fibroblasts and endothelial cells and that , among melanocytes , nevus melanocytes express lower levels of Igfbp7 than non-nevus melanocytes . We did not detect any Mia transcripts in nevus melanocytes , although it was expressed at detectable levels in non-nevus melanocytes and various other skin cells . Likewise , of Il1 family members , only Il1a transcripts were detected in nevus melanocytes and they were at levels lower than in many other skin cell types . Il6 was also only weakly expressed in nevus melanocytes , especially when compared with other cells . Transcripts for type one interferons were not detected in any melanocytic cells , and Cxcl15 transcripts were not detected in any skin cells at all . Of course , the accuracy of single-cell RNA sequencing can be limited for weakly expressed genes , so we cannot completely eliminate the possibility that these factors play some role in nevus growth arrest . But given these results , and the evidence that nevus melanocytes are not senescent , we strongly favor the renewal-feedback model ( model C ) . Adopting this model also makes it easier to accommodate long-standing evidence that nevus growth arrest is not permanent ( Shain and Bastian , 2016 ) . For example , it is known that nevi may exhibit a low level of mitoses ( Glatz et al . , 2010 ) ; that they can grow in response to stimuli such as UV light ( Rudolph et al . , 1998 ) or immunosuppression ( Shain and Bastian , 2016 ) and , perhaps most tellingly , they can re-grow after incomplete surgical resection—stopping again when they reach a typical nevus size ( Vilain et al . , 2016 ) . The latter result is inherently problematic for any non-feedback model , but is precisely what renewal feedback predicts ( Lander , 2011; Lander et al . , 2009 ) . Because feedback control of renewal implements a generic strategy ( integral negative feedback [Lander , 2011] ) , it places no constraints on the molecular details of feedback , short of the fact that whatever is mediating it needs to rise with the number of cells already arrested . One possibility is that nevi are responding to some of the same signals that are used in melanocyte homeostasis . For instance , during anagen phase of the hair cycle , melanocyte stem cells produce progeny that migrate out of the hair follicle bulge as they differentiate , leaving functional stem cells behind for future cycles . A variety of experimental and pathological circumstances that allow small numbers of melanocytes to differentiate within the bulge cause differentiation and loss of the entire stem cell pool ( with concomitant hair graying [Nishimura et al . , 2005] ) . This sort of behavior—where differentiated cells drive the differentiation of their progenitors—is exactly the sort of behavior that drives feedback models of renewal ( Buzi et al . , 2015; Lander , 2011; Lander et al . , 2009 ) . Nevi are but one of many types of benign , clonal , proliferative lesions that arise due to the activation of oncogenes , but rarely if ever progress to malignancy ( Adashek et al . , 2020 ) . Notwithstanding the disruptive influence that oncogenes can have on cell physiology , the existence of such lesions suggest that homeostatic mechanisms persist and function at many stages along the road to cancer . New avenues for cancer prevention and treatment are likely to follow from the detailed elucidation of such mechanisms . BrafV600E , Tyr-CreER ( C56BL/6 ) mice ( RRID:MGI:5902125 ) were genotyped by PCR as previously described ( Bosenberg et al . , 2006; Dankort et al . , 2007 ) . The primers used in this study are: Braf forward 5’-TGAGTATTTTTGTGGCAACTGC −3’ , Braf reverse 5’-CTCTGCTGGGAAAGCGCC −3’ , Cre forward 5’- GGTGTCCAATTTACTGACCGTACA-3’ and Cre reverse 5’- CGGATCCGCCGCATAACCAGTG −3’ . Topical administration of 4-hydroxytamoxifen ( 4-OHT; 25 mg/mL or 75 mg/mL in DMSO; 98% Z-isomer , Sigma-Aldrich ) was administered to pups on their back and/or paws at ages P2 , P3 , and P4 . Images of nevi on back and paw skin were taken with a digital camera at the indicated ages . Nevi from the underside of the skin were imaged using a dissection microscope . All mouse procedures were approved by UCI’s IACUC . Mice were sedated , shaved , and depilated with wax strips at the indicated ages ( during a telogen phase ) and the dorsal skin was imaged to capture the intrinsic fluorescent signal from keratin , melanin , as well as the second-harmonic-generation signal from collagen , using the LSM 510 NLO Zeiss system . Excitation was achieved with a femtosecond Titanium: Sapphire ( Chameleon-Ultra , Coherent ) laser at 900 nm . Emission was detected at 390–465 nm for second harmonic generation ( blue ) and 500–550 nm ( green ) and 565–650 ( red ) for fluorescence . BrdU was prepared in sterile PBS at 10 mg/mL and injected intraperitoneally into mice that were 20 days old at 100 mg/kg of body weight . 24 hr later the mouse was shaved , depilated with wax strips and the skin was removed and fixed in 10% formalin for 16 hr . Formalin fixed paraffin embedded skins were sectioned 8 µm thick , deparaffinized with Xylene , and dehydrated in a series of increasing concentration of ethanol washes . Antigen retrieval was performed with 10 mM citric acid buffer at pH 6 . 0 for 10 min in a steamer . Samples were washed with PBS , incubated with TrueBlack for 30 s to reduce autofluorescence , and washed again with PBS . All antibodies were diluted at a 1:500 and incubated overnight at 4°C . Samples were washed and incubated with the appropriate secondary antibody . Melanocytes were identified with a Pmel antibody ( EP4863 ( 2 ) ; ab137078 , Abcam; RRID:AB_2732921 ) . Cells that incorporated BrdU were visualized with a BrdU antibody ( ab6326 , Abcam; RRID:AB_305426 ) . BrafWT , Tyr-CreER or BrafV600E , Tyr-CreER mice were euthanized at either P30 ( n = 2 of each genotype ) or P50 ( n = 3 of each genotype ) , shaved , and depilated . A 2 × 3 cm section of the dorsal skin was removed , and the fat scraped off from the underside . The piece was then diced into smaller pieces and suspended in dissociation buffer ( RPMI , liberase 0 . 25 mg/mL , Hepes 23 . 2 mM , Sodium Pyruvate 2 . 32 mM , Collagenase:Dispase 1 mg/mL ) for 50 min at 37°C with gentle agitation . After incubation , DNaseI ( 232U ) was added for 10 min and then inactivated with fetal bovine serum and EDTA ( 1 mM ) . The tissue suspension was further dissociated mechanically with the GentleMACS using the setting m_imptumor_04 . 01 , which runs for 37 s at various speeds . Single-cell suspensions were filtered twice through a 70 µm strainer and dead cells removed by centrifugation at 300 x g for 15 min . The live cells were washed with 0 . 04% UltraPure BSA:PBS buffer , gently re-suspended in the same buffer , and counted using trypan blue . Libraries were prepared using the Chromium Single Cell 3’ v2 protocol ( 10X Genomics ) . Briefly , individual cells and gel beads were encapsulated in oil droplets where cells were lysed and mRNA was reverse transcribed to 10X barcoded cDNA . Adapters were ligated to the cDNA followed by the addition of the sample index . Prepared libraries were sequenced using paired end 100 cycles chemistry for the Illumina HiSeq 4000 . FASTQ files were generated from Illumina’s binary base call raw output with Cell Ranger’s ( v2 . 1 . 0; RRID:SCR_017344 ) `cellranger mkfastq` command and the count matrix for each sample was produced with `cellranger count` . All ten samples ( four samples from P30 [two control ( wild type ) and two mutant] and six samples from P50 [three control and three mutant] ) were aggregated together with the `cellranger aggr` command to produce one count matrix that includes all samples . Data analysis was performed with Scanpy [v1 . 3 . 6; RRID:SCR_018139] ( Wolf et al . , 2018 ) . Cells with fewer than 200 detected genes , and genes detected in less than three cells , were discarded . We calculated the percent mitochondrial gene expression and kept cells with less than 13% mitochondrial gene expression , and cells with fewer than 4000 genes/cell ( 35 , 141 cells ) . Each cell was normalized to total counts over all genes . In the final preprocessing step , we regressed out cell-cell variation driven by mitochondrial gene expression and the number of detected UMI . To identify clusters , we first performed principal component analysis on log-transformed data , using highly variable genes , Louvain clustering ( Levine et al . , 2015 ) , and visualization with t-distributed stochastic neighbor embedding ( tSNE ) . To quantify the sizes of nevi in mice , dorsal skin was excised and the underside visualized using a dissecting microscope . Nevi were traced , and area calculated using ImageJ . Nest sizes were quantified in live mice by MPM . Sizes of human nests were measured from histological samples ( n = 5 ) obtained from the UCI Department of Dermatology . Samples were stained with hematoxylin and eosin and imaged with a microscope . A dermatologist manually identified the nests on each slide , and nest area was quantified using ImageJ . Human studies were performed under IRB protocol HS# 2019–5054 . Estimates of cell numbers for mouse nevi were obtained in two different ways: First , we used estimates from Chai et al . , 2014 for melanocytic nuclei per square area of mouse nevus , together with our observed median nevus radius of 76 . 8 µm; this approach led to an estimate of 897 cells/nevus . As the data of Chai et al . , 2014 come from a different genetic model , we also estimated cell number as follows: Using 8 µm sections of back skin from Albino BrafV600E mice , we used fluorescence microscopy to measure the sizes of 194 Pmel-stained melanocytes within the nests of nevi , obtaining an average cell diameter of 5 . 68 µm , and counted approximately 14 . 4 cells per 104 µm3 of nest . In pigmented animals , we measured by MPM an average nest cross sectional area of 1385 µm2 , an average nest volume of 38792 µm3 , and an average number of nests per nevi of approximately 12 , yielding an estimate of 672 cells/nevus . Given uncertainties in these measures , analyses in the manuscript take into account the possibility of an average that falls anywhere between 100 and 1000 cells . Stochastic , non-spatial simulations of renewal control were obtained by Monte Carlo simulation , in which cells duplicated every cell cycle , and then chose randomly whether to differentiate or continue dividing according to a probability modified by feedback from non-dividing cells . A Hill function , with Hill coefficient = 1 , was used to represent the feedback . To model feedback in a spatial context , we used CompuCell3D ( RRID:SCR_003052 ) , an open-source platform for Cellular Potts modeling ( Swat et al . , 2012 ) . In CPM , every generalized object or ‘cell’ is associated with a list of attributes such as cell type , surface area , volume , etc . These enter into the calculation of an effective energy , which can be summarized as the sum of the contact energy between neighboring cells and the effective energy due to volume constraints . Simulations were initialized by seeding a single cell , with a size of 25 pixel2 , in the center of a 300 × 300 pixel lattice , which grew and divided according to rules and parameters summarized in Source data 2 . To add variability to cell growth , cells randomly chose one of three different growth rates after every cell division . To add variability to cell division times , cells randomly chose a target area , between 72 and 80 pixel2 , at which to divide . Growth rates were chosen to be sufficiently slow that the mean time between cell divisions came out to approximately 382 time steps . At division , each cell was divided in half by a randomly-oriented division plane . Upon division , a cell either remained dividing or became permanently arrested . All cells had a minimum 1% probability of arrest per cell division . Once an arrested cell was generated , it began continuous secretion of a signaling molecule that diffuses and promotes the transition from dividing to non-dividing ( Kunche et al . , 2016 ) . Diffusion and decay of the feedback factor was modeled deterministically , with parameters chosen to produce a steady state decay length of 15 pixels . The concentration of this factor at the center of mass of each cell then augmented the arrest probability of that cell by an amount determined by a Hill function ( see Source data 2 ) . Statistical analyses for single-cell RNA sequencing were performed using Scanpy ( RRID:SCR_018139 ) . Other statistical testing was done using Mathematica ( RRID:SCR_014448 ) . For the spatial analysis in Figure 5E–F , nest areas in each field were randomly swapped , with positions held constant , 5000 times , and the distributions of neighboring nest locations and sizes recalculated each time . This allowed us to generate an envelope enclosing the 5th and 95th percentiles for the permuted data , at each target nest size , and compare the observed data with the bounds of that envelope .

Melanocytes are pigment-producing cells found throughout the skin . Mutations that activate a gene called BRAF cause these cells to divide and produce melanocytic nevi , also known as “moles” . These mutations are oncogenic , meaning they can cause cancer . Indeed , BRAF is the most commonly mutated gene in melanoma , a deadly skin cancer that arises from melanocytes . Yet , moles hardly ever progress to melanoma . A proposed explanation for this behavior is that , once activated , BRAF initiates a process called “oncogene-induced senescence” in each melanocyte . This process , likened to premature aging , is thought to be what causes cells in a mole to quit dividing . Although this hypothesis is widely accepted , it has proved difficult to test directly . To investigate this notion , Ruiz-Vega et al . studied mice with hundreds of moles created by the same BRAF mutation found in human moles . Analyzing the activity of genes in individual cells revealed that nevus melanocytes that have stopped growing are no more senescent than other skin cells , including non-mole melanocytes . Ruiz-Vega et al . then analyzed the sizes at which moles stopped growing , estimating the number of cells in each mole . The data were then compared with the results of a simulation and mathematical modeling . This revealed that any model based on the idea of cells independently shutting down after a number of random events could not reproduce the distribution of mole sizes that had been experimentally observed . On the other hand , models based on melanocytes acting collectively to shut down each other’s growth fit the observed data much better . These findings suggest that moles do not stop growing as a direct result of the activation of BRAF , but because they sense and respond to their own overgrowth . The same kind of collective sensing is observed in normal tissues that maintain a constant size . Discovering that melanocytes do this not only sheds light on why moles stop growing , it could also help researchers devise new ways to prevent melanomas from forming .

2020

Dynamics of nevus development implicate cell cooperation in the growth arrest of transformed melanocytes

Cancer poses danger because of its unregulated growth , development of resistance , and metastatic spread to vital organs . We currently lack quantitative theory for how preventive measures and post-diagnostic interventions are predicted to affect risks of a life threatening cancer . Here we evaluate how continuous measures , such as life style changes and traditional treatments , affect both neoplastic growth and the frequency of resistant clones . We then compare and contrast preventive and post-diagnostic interventions assuming that only a single lesion progresses to invasive carcinoma during the life of an individual , and resection either leaves residual cells or metastases are undetected . Whereas prevention generally results in more positive therapeutic outcomes than post-diagnostic interventions , this advantage is substantially lowered should prevention initially fail to arrest tumour growth . We discuss these results and other important mitigating factors that should be taken into consideration in a comparative understanding of preventive and post-diagnostic interventions . Mathematical models play an important role in describing and analysing the complex process of carcinogenesis . Natural selection for increases in tumour cell population growth can be represented as the net effect of increased cell division rates and/or decreased apoptosis ( e . g . , Wodarz and Komarova , 2007 ) . Relatively rare driver mutations confer such a net growth advantage , whereas numerically dominant passenger mutations with initially neutral or mildly deleterious effects ( Marusyk et al . , 2012; Bozic et al . , 2013; McFarland et al . , 2013 ) can increase in frequency due to genetic hitchhiking or subsequent positive selection . Amongst the many passengers in a growing tumour , some can contribute to chemoresistance , and sufficiently large tumours could contain different clones that , taken as a group , can resist some , if not most , chemotherapies ( see Michor et al . , 2005 for resistance to imatinib ) . Chemotherapeutic remission followed by relapse suggests that these resistant cells are often present at low frequencies prior to therapy , either due to genetic drift or costs associated with resistance . Resistant phenotypes subsequently increase in frequency during radiotherapy or chemotherapy , and through competitive release they may incorporate one or more additional drivers , resulting in accelerated growth compared to the original tumour ( for related discussion on pathogens , see Huijben et al . , 2013 ) . Previous mathematical studies have considered alternatives to attempting to minimize or eradicate clinically diagnosed cancers with maximum tolerated doses ( MTDs ) of chemotherapeutic drugs . This body of work indicates that MTD is particularly prone to select for chemoresistance ( e . g . , Foo and Michor , 2009; Foo and Michor , 2010; Lorz et al . , 2013 ) , and what little empirical work exists supports this basic prediction ( Turke et al . , 2010 ) , but see ( Kouyos et al . , 2014 ) for other disease systems . Numerous alternatives to the goal of cancer minimization/eradication have been proposed and investigated ( e . g . , Maley et al . , 2004; Komarova and Wodarz , 2005; Foo and Michor , 2009; Gatenby et al . , 2009a , 2009b; Bozic et al . , 2013; Jansen et al . , 2015 ) . For example , Komarova and Wodarz ( 2005 ) considered how the use of one or multiple drugs could prevent the emergence or curb the growth of chemoresistance . They showed that the evolutionary rate and associated emergence of a diversity of chemoresistant lineages is a major determinant in the success or failure of multiple drugs vs a single one . Lorz and co-workers ( Lorz et al . , 2013 ) recently modelled the employment of cytotoxic and cytostatic therapies alone or in combination and showed how combination strategies could be designed to be superior in terms of tumour eradication or managing resistance than either agent used alone . Foo and Michor ( 2009 ) evaluated how different dosing schedules of a single drug could be used to slow the emergence of resistance given toxicity constraints . One of their main conclusions is that drugs slowing the generation of chemoresistant mutants and subsequent evolution are more likely to be successful than those only increasing cell death rates . These and other computational approaches have yet to consider the use of preventive measures to reduce cancer-associated morbidity and mortality whilst limiting resistance . Prevention includes life-style changes and interventions or therapies in the absence of detectable invasive carcinoma ( e . g . , Etzioni et al . , 2003; Lippman and Lee , 2006; William et al . , 2009; Hochberg et al . , 2013 ) , for example , reduced cigarette consumption ( Doll and Peto , 1976 ) or chemoprevention ( Steward and Brown , 2013 ) . In depth consideration of preventive measures and their likely impact on individual risk and epidemiological trends is important given the likelihood that all individuals harbour pre-cancerous lesions , some of which may transform into invasive carcinoma ( Bissell and Hines , 2011; Greaves , 2014 ) , and concerns as to whether technological advances will continue to make significant headway in treating clinically detected cancers ( Gillies et al . , 2012; Vogelstein et al . , 2013 ) . Here , we model how continuous , constant measures affect tumour progression and the emergence of resistant lineages . We assume that an individual can contract at most a single cancer , originating from a single lesion . Importantly , we consider cases where the measure may select for the evolution of resistant phenotypes and cases where no resistance is possible . Our approach is to quantify the daily extent to which a growing neoplasm must be arrested in order to either eradicate it or to delay a potentially lethal cancer . Several authors have previously argued how constant or intermittent low toxicity therapies either before or after tumour discovery could be an alternative to MTD chemotherapies ( Wu and Lippman , 2011; Hochberg et al . , 2013 ) , but to our knowledge , no study has actually quantified based on empirical parameter estimates , the extent to which cancer cell population growth needs to be arrested for such approaches to succeed ( see related discussion in Bozic et al . , 2010; Gerstung et al . , 2011; Bozic et al . , 2013 ) . Below we employ the terms ‘treatment’ , ‘measure’ , and ‘therapy’ interchangeably , all indicating intentional measures to arrest cancer cell population growth . We first derive analytical expressions for the expected total number of cells within a tumour at any given time . We explore dynamics of tumour sizes at given times , and times to detection for given tumour sizes . Specifically , we show that the expected mean tumour size in a population of subjects can be substantially different from the median , since the former is highly influenced by outliers due to tumours of very large size . We then consider constant preventive measures and show that treatment outcome is sensitive to initial conditions , particularly for intermediate-sized tumours . Importantly , we provide approximate conditions for tumour control both analytically and numerically using empirical parameter estimates . We next consider post-diagnostic interventions in which tumour resection either is not complete and leaves residual cells or undetected metastases are present . We contrast these with prevention scenarios where ( 1 ) there is no difference in the age at which either prevention or post-diagnostic intervention commences , and ( 2 ) prevention and post-diagnostic interventions are alternatives , that is , the former always occurs before the latter . We show as expected that therapeutic outcomes are generally superior under prevention vs post-diagnostic intervention , and that higher impacts on the cancer cell population are usually required for post-diagnostic interventions to achieve a level of control comparable to prevention . Moreover , we find that should resection leave sufficiently large numbers of residual cells ( or metastases are not discovered ) , then a range of the most successful outcomes under prevention is not attainable under post-diagnostic intervention , regardless of potential cell arrest . Finally and importantly , whereas there is little gained in terms of outcomes in post-diagnostic intervention beyond approximately 0 . 3% cell arrest per day for both small ( 10 , 000 ) and large ( 1 million ) cancer cell populations , prevention outcomes may achieve continual gains for the latter cell number , up to about 0 . 6% cell arrest per day . Previous study has evaluated the effects of deterministic and stochastic processes on tumour growth and the acquisition of chemoresistance ( Komarova and Wodarz , 2005; Bozic et al . , 2010; Reiter et al . , 2013 , see review Beerenwinkel et al . , 2015 ) . We first consider both processes through exact solutions and numerical simulations of master equations , using the mean field approach ( see Appendix 1 for details ) . A mean field approach assumes a large initial number of cells ( Krapivsky et al . , 2010 ) and averages any effects of stochasticity , so that an intermediate state of the system is described by a set of ordinary differential equations ( i . e . , master equations; Gardiner , 2004 ) . Solutions to these are complex even in the absence of the explicit consideration of both drivers and passengers ( Antal and Krapivsky , 2011; Kessler and Levine , 2013 ) . We do not explicitly model the different pre-cancerous or invasive carcinoma states . Rather , our approach follows the dynamics of the relative frequencies of subclones , each composed of identical cells ( Baake and Wagner , 2001; Saakian and Hu , 2006 ) . We simulate tumour growth using a discrete time branching process for cell division ( Athreya and Ney , 1972; Bozic et al . , 2010 ) . For each numerical experiment , we initiate a tumour of a given size and proportion of resistant cells . Briefly , the model framework is as follows . Each cell in a population is described by two characteristics . The first is its resistance status to the measure , which is either ‘not resistant’ ( j = 0 ) or ‘resistant’ ( j = 1 ) . The second property is the number of accumulated driver mutations ( maximum N ) in a given cell line . At each time step of 4 days , cells either divide or die , and when a cell divides , its daughter cell has a probability u of producing a driver mutation and v of producing a resistant mutation . We assume no back mutation , and that cells do not compete for space or limiting resources . The fitness function fij , the difference between the birth and death rates of a cell , is defined by the number of accumulated drivers ( i = 0 , 1 , … , N ) and resistance status ( j = 0 , 1 ) : a sensitive cancerous cell with a single driver has selective advantage s , and any accumulated driver adds s to fitness , while resistance is associated with a constant cost c . Exposure to a single treatment affects only non-resistant cells ( j = 0 ) , incurring a loss σ to their fitness . Thus , the fitness function is:fij=s ( i+1 ) −σ ( 1−j ) −cj . The assumption of driver additivity is a special case of multiplicative fitness , and both are approximately equivalent for very small s . We conducted numerical experiments , each with the same initial states but each using a unique set of randomly generated numbers of a branching process . For each simulation and each time step , the number of cells at time ( t + 1 ) was sampled from a multinomial distribution of cells at time t ( see Bozic et al . , 2010 for details ) . Table 1 presents baseline parameter values employed in this study . Hereafter , we refer to σ as the treatment intensity ( applied once every cell cycle of 4 days ) , while the corresponding daily arrest level to non-resistant cells is approximated by σ/4 . 10 . 7554/eLife . 06266 . 024Table 1 . Baseline parameter values used in this studyDOI: http://dx . doi . org/10 . 7554/eLife . 06266 . 024ParameterVariableValueRangeRef . Time step ( cell cycle length ) T4 days3–4 days ( Bozic et al . , 2010 ) Selective advantages0 . 4%0 . 1–1 . 0% ( Bozic et al . , 2010 ) Cost of resistancec0 . 1%Mutation rate to acquire an additional driveru3 . 4 × 10−510−7–10−2 ( Bozic et al . , 2010 ) Mutation rate to acquire resistancev10−610−7–10−2 ( Komarova and Wodarz , 2005 ) Maximal number of additional driversN5 ( Figures 1 , 2 ) 9 ( other figures ) 0–9Initial cell populationM0106 cells–Pre-resistance levelκ0 . 01%– ( Iwasa et al . , 2006 ) Number of replicate numerical simulations ( excluding extinctions ) –106–Detection thresholdM109 cells107–1011 ( Beckman et al . , 2012 ) ‘Range’ is values from previous study and employed in the present study . We first study preventive interventions where a patient has a high risk of developing a cancer and/or a biomarker that indicates the probable presence of a cancer . In either case , so that we can compare and contrast different intervention levels , we assume that the ( undetected ) tumour contains M0 cells when prevention commences . We examine effects on the mean by considering the distribution of tumour sizes at different times using mean-field dynamics ( see Appendix 1 ) . Numerical experiments were then conducted by assuming that tumours initially contained M0 = 106 identical cells ( i = 0 ) , of which 0 . 01% were resistant . These assumptions are obviously oversimplifications , and we relax some of them below and in the next sections . There is an excellent correspondence between analytical and numerical results for σ varied in range of s ( Appendix 1—figure 1A ) . A more detailed study of the distribution of tumour sizes reveals that the mean diverges considerably from median behaviour in the majority of cases , since the former is strongly influenced by outliers with high-tumour cell numbers ( see Appendix 1—figure 1B ) . Figure 1 shows four examples of numerical experiments . An untreated tumour reaches the assumed detection threshold of 109 cells by about 18 years on average and because it is not subject to strong negative selection ( we assume low c ) , any emerging resistant cell-lines are likely to remain at low frequency ( 0 . 03% at the detection time in the example of Figure 1A ) . In Figure 1B , low-treatment intensity delays tumour growth and thus time of detection by approximately 16 years , while an increase in dose tends to result in tumours dominated by resistant cells ( Figure 1C ) . Despite being unaffected by treatment , resistant cell populations are sometimes observed to go extinct stemming from stochasticity ( Figure 1D ) , and this tends to occur more at high-treatment levels , because there are fewer sensitive tumour cells to seed new ( mutant ) resistant cell populations . 10 . 7554/eLife . 06266 . 003Figure 1 . Treatments curb or eliminate tumours . Examples of single patient tumour growth for ( A ) no treatment . ( B ) σ = 0 . 6% . ( C ) σ = 1 . 0% . ( D ) σ = 2 . 0% . The shaded area shows the change in total tumour size and the hatched area , the resistant part of a tumour . The treatment intensity σ in this and all other figures are represented as cell arrest per day ( σ/4 ) . Parameter values as in Table 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06266 . 003 We next considered how therapies affected the distribution of tumour detection times in cases where the cancer cell population attained a threshold of 109 cells . The magnitude of the selective advantage s shows that tumour growth is largely driven by its non-resistant part for relatively low-impact treatments σ < 2s ( Figure 2A ) . Importantly , the tumour shifts from being mainly non-resistant to resistant at σ ≈ 2s , which is reflected by the inflection point in the trajectory of the median ( indicated by point B in Figure 2A , B ) . Notice that detection times are also most variable at σ ≈ 2s . The median changes smoothly at high-treatment levels ( σ > 2s ) , tending to a horizontal asymptote . This is explained by the fact that the sensitive part is heavily suppressed at high-treatment levels , meaning that the dynamics are strongly influenced by the actual time point at which the first resistance mutation occurs . 10 . 7554/eLife . 06266 . 004Figure 2 . Treatment level affects both detection time and frequency of resistance . The median ( lines ) and 90% confidence intervals ( shaded areas ) of detection times , measured as years beyond the initiation of the preventive measure . Effects of: ( A ) the selective advantage of each additional driver and ( B ) the cost of resistance . ( C ) Samples of the distribution of detection times ( in relative frequencies , adjusted for 3-month bins ) for corresponding points , indicated in A and B . Dashed black line is the mean and the dashed-and-dotted line is the median . The colour-code indicates the average level of resistance in detected tumours over 3 month intervals ( see inset in B ) . All cells j = 0 at t = 0 . Other parameters as in Table 1 . Detection time is log-transformed in A and B . DOI: http://dx . doi . org/10 . 7554/eLife . 06266 . 00410 . 7554/eLife . 06266 . 005Figure 2—figure supplement 1 . Sensitivity analysis for several key parameters . ( A ) Maximal number of additionally accumulated drivers . ( B ) Initial cell number . ( C ) Level of initial partial resistance of a tumour . ( D ) Presence or absence of resistant cell-lines . Point colour-codes indicate the average level of resistance in detected tumours over 3 month intervals ( see inset in B ) . For simplicity , only the median is indicated in B and C for the baseline case ( blue line ) . Lines and shading otherwise as in Figure 2 . Unless otherwise stated , parameter values as in Table 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06266 . 00510 . 7554/eLife . 06266 . 006Figure 2—figure supplement 2 . Effects of initial neoplasm size ( A , B ) and resistance level ( C ) on preventive measure success . Success is defined as tumour non-detection by 50 years . Daily effect of treatment on cellular arrest is assumed to be 0 . 25% . Unless otherwise stated , parameter values as in Table 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06266 . 006 We find , counterintuitively , that early-detected tumours are more likely to be resistant under constant treatments than those detected at later times ( A , B , and C in Figure 2C ) . This is because tumours under treatment that by chance obtain resistance early grow faster than those that do not . By the time of detection , non-resistant tumours usually accumulate up to 4 additional drivers on average , while resistant tumours have fewer . For larger values of cost c , an additional non-regularity emerges at σ ≈ 3s ( segment DEF in Figure 2B ) , and is associated with tumours having a majority of cells with maximum numbers of drivers . This region is also characterized by a different transition to complete resistance ( cf . Videos 1 , 2 for relatively low and high costs of resistance , respectively ) . For example , at point D , tumours with a majority of non-resistance have less variable detection times than tumours with a majority of resistant cells ( points E and F in Figure 2B and corresponding panels in Figure 2C ) . Treatment levels along the segment DEF result in tumours that are more likely to be resistant as one goes from the centre to the tails of the distribution of detection times . This differs qualitatively from the previous case of a lower cost of resistance , where the tumours are less resistant in the tail of the distribution of detection times ( cf segments ABC and DEF in Figure 2B and corresponding panels in Figure 2C ) . 10 . 7554/eLife . 06266 . 007Video 1 . Treatment level affects both detection time and frequency of resistance . ( A ) The median ( thick line ) and 90% confidence intervals ( shaded areas with dashed boundaries ) for the distribution of detection times . ( B ) Arbitrary samples of the distribution of detection times and distribution of the mean number of accumulated drivers . The colour-code indicates the average level of resistance in detected tumours over 3 month intervals . Parameters as in Table 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06266 . 00710 . 7554/eLife . 06266 . 008Video 2 . Treatment level affects both detection time and frequency of resistance . ( A ) The median ( thick line ) and 90% confidence intervals ( shaded areas with dashed boundaries ) for the distribution of detection times . ( B ) Arbitrary samples of the distribution of detection times and distribution of the mean number of accumulated drivers . The colour-code indicates the average level of resistance in detected tumours over 3 month intervals . Parameters as in Table 1 except for the cost of resistance c = 0 . 4% . DOI: http://dx . doi . org/10 . 7554/eLife . 06266 . 008 The inflection point at σ ≈ 2s in Figure 2A is due to the accumulation of additional drivers within tumours and associated increases in the likelihood that the tumour eventually resists treatment . Since the initial population consists of 106 cells , in the absence of treatment , a mutant cell with one additional driver and associated fitness 2s will appear very rapidly . Such a tumour can be suppressed only if σ > 2s . This is supported by additional numerical experiments where we vary the maximal number of additional driver mutations N: the inflection point σ ≈ 2s disappears when N = 0 ( Figure 2—figure supplement 1A ) . The inflection points at σ = 3s , 4s emerge at treatment levels that effectively suppress sensitive subclones with the most drivers before resistance mutations are obtained ( cf Figure 2—figure supplement 1A–C with Figure 2—figure supplement 1D and Video 3 ) . Specifically , the peaked distributions , corresponding to better therapeutic outcomes , tend to disappear when resistant subclones emerge . 10 . 7554/eLife . 06266 . 009Video 3 . Treatment level effects on detection times assuming no resistance is possible . ( A ) The median ( thick line ) and 90% confidence intervals ( shaded areas with dashed boundaries ) for the distribution of detection times . ( B ) Arbitrary samples of the distribution of detection times and the distribution of the mean number of accumulated drivers . The colour-code indicates the average level of resistance in detected tumours over 3 month intervals . The resistance mutation is knocked out ( v = 0 ) . Otherwise parameters as in Table 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06266 . 009 The initial cancer cell number M0 affects both the median and distribution of detection times ( Figure 2—figure supplement 1B ) . For large initial tumours , growth is deterministic and exponential . As the initial size is decreased from 106 to 105 , stochastic effects are increasingly manifested by greater variability in tumour inhibition and an inflection point observed at the 95th percentile . Moreover , we find that a tumour is likely to be eradicated under a range of constant treatments when M0 = 105 or fewer initial cells; in contrast , a tumour is virtually certain to persist regardless of treatment level for M0 = 107 cells or greater ( Figure 2—figure supplement 2A , B ) . In other words , our model indicates that tumours that are c . 1% the size of most clinically detectable , internal cancers will typically be impossible to eradicate by single molecule chemoprevention when resistance is possible . Given the mutation rates assumed here , many tumours with 1 million cells will either already contain or rapidly subsequently acquire resistant cells ( Iwasa et al . , 2006 ) . It is therefore not surprising that the initial fraction of resistant cells in a tumour has little impact on dynamics ( Figure 2—figure supplement 1C ) . In contrast , another measure of success in control ( the fraction of persons with tumours that remain undetected after 50 years of growth ) improves substantially with lower numbers of initial resistance mutations , particularly at higher treatment levels ( Figure 2—figure supplement 2C ) . This is because the initial phases of treatment have a major impact on the potential for new resistant mutants: should few be initially present or emerge , they will either go stochastically extinct or will not grow to detection levels ( 1 billion cells ) in the 50 year time frame of these numerical experiments . We conducted further sensitivity analyses by varying accumulation rates u of additional driver mutations . We find that tumours exhibit more or less deterministic growth depending on the initial number of cells M0 and driver mutation rate u whereby the larger the population ( Figure 2—figure supplement 1B ) or the higher the mutation rate ( Appendix 1—figure 4A ) , the less apparent are stochastic effects . The corresponding analysis is presented in ’Varying mutation rate and initial tumour size‘ in Appendix 1 and Appendix 1—figure 4 . Finally , we considered scenarios where the cost of resistance is dose-dependent and specifically situations of drug addiction ( Das Thakur et al . , 2013 ) . Numerical studies presented in more detail in ’A simple form of drug addiction for resistant cell-lines‘ in Appendix 1 show that under dose-dependent costs , a drug treatment only applied when the number of non-resistant cells exceeds the number of resistant cells ( e . g . , a metronomic therapy [Fischer et al . , 2015] ) leads to slower long-term tumour growth than does a constant therapy . We next investigated how a post-diagnostic measure ( usually some form of chemotherapy or radiation therapy , but could also involve adjuvants after an initial therapy ) affects the probability of treatment success , the distribution of times for tumour relapse , and resistance levels . We assume that a tumour grows from one cell ( i = 0 , j = 0 ) and is discovered either at 109 ( early ) or 1011 ( very late ) cells , whereupon the primary tumour is removed , leaving a small number ( 104 or 106 ) of residual , and/or undetected or inoperable neighbouring micro-metastatic cells , and/or distant metastatic cells . Below , we contrast this with prevention without discriminating the age at which either intervention type commences , whereas in the following section , we consider these as competing alternatives . Figure 3A and Figure 3—figure supplement 1A present the distributions of driver mutations for each scenario . ( Recall that in the previous section , we assumed that when a measure commenced , tumours had no additional drivers ( i = 0 ) ) . 10 . 7554/eLife . 06266 . 010Figure 3 . Effects of preventive and post-diagnostic interventions against tumours consisting of 1 million cells . ( A ) The distribution of mean sizes of subclones ( hatched bars = before removal and solid bars = post removal ) . ( B ) The time distribution of cases in which either intervention type fails to control the tumour below the detection threshold after 50 years ( thick line = median , filled area with dashed boundaries = 90% CIs ) for different constant treatment intensities . ( C ) The percentage of cases where the tumour consists of less than 100 resistant cells at 4 years after treatment commences ( solid lines ) , and the percentage of cases where tumour size is below the detection threshold 20 years after the measure begins ( dashed-and-dotted lines ) . ( D ) The mean number of accumulated drivers within a tumour at the time of detection . Parameter values as in Table 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06266 . 01010 . 7554/eLife . 06266 . 011Figure 3—figure supplement 1 . Effects of preventive and post-diagnostic interventions against tumours consisting of 10 , 000 cells . Same as Figure 3 , except interventions against 104 cancer cells . ( A ) The distribution of mean sizes of subclones for different constant treatment intensities ( hatched bars = before removal and solid bars = post removal ) . ( B ) The time distribution of cases in which either intervention type fails to control the tumour below the detection threshold after 50 years ( thick lines = medians , shaded areas with dashed boundaries = 90% CIs ) . ( C ) The percentage of cases when a tumour consists of less than 100 resistant cells at 4 years post-resection ( solid lines ) and the percentage of cases when tumour sizes are below the detection threshold 20 years after the measure commences ( dashed-and-dotted lines ) . ( D ) The mean number of accumulated drivers within a tumour at the time of detection . Parameter values as in Table 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06266 . 01110 . 7554/eLife . 06266 . 012Figure 3—figure supplement 2 . Time to first discovery as a predictor of post-diagnostic treatment success . Time to tumour relapse following resection as function of the time it takes for the initial cancer cell to attain 109 cells ( i . e . , the point at which the tumour is discovered , resected , and treatment begins ) . Each dot represents a numerical simulation from the yellow distribution in Figure 3B ( only 1 , 000 simulation results out of 106are shown ) . Four different treatment levels are considered . Black solid line is a simple linear regression , and grey area with dashed boundaries indicates extrapolation of high and low bounds accounting for 95% of observations ( prediction interval ) . The fitted linear regression model gives an intercept of 7 . 5 years , a slope of 1 . 6° and R2 of 0 . 024 in ( A ) , 10 . 4 years , 2 . 2° and R2 of 0 . 017 in ( B ) , 12 . 9 years , 3 . 0° and R2 of 0 . 009 in ( C ) , and 13 . 1 years , 3 . 3° and R2 of 0 . 008 in ( D ) . Parameters as in Table 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06266 . 01210 . 7554/eLife . 06266 . 013Figure 3—figure supplement 3 . The R2 of regressions from numerical experiments for different treatment levels of time to tumour relapse following resection as function of the mean number of drivers in a resected tumour . Time to tumour discovery is generally more predictive of post-diagnostic therapeutic outcome for lower treatment levels . See Figure 3—figure supplement 2 for details . DOI: http://dx . doi . org/10 . 7554/eLife . 06266 . 01310 . 7554/eLife . 06266 . 014Figure 3—figure supplement 4 . Mean number of additionally accumulated drivers in resected tumour as a predictor of post-diagnostic treatment success . The fitted negative exponential regression model y = ae-bx gives a = 13 . 5 years , b = 0 . 3 and R2 = 0 . 696 in ( A ) , 18 . 95 years , 0 . 3 and R2 = 0 . 537 in ( B ) , 23 . 0 years , 0 . 29 and R2 = 0 . 262 in ( C ) , and 23 . 9 years , 0 . 3 and R2 = 0 . 224 in ( D ) . See Figure 3—figure supplement 2 for details . DOI: http://dx . doi . org/10 . 7554/eLife . 06266 . 01410 . 7554/eLife . 06266 . 015Figure 3—figure supplement 5 . The R2 of regressions from numerical experiments for different treatment levels of time to tumour relapse following resection as function of the mean number of drivers in a resected tumour . Time to tumour discovery is more predictive of post-diagnostic therapeutic outcome for lower treatment levels . See Figure 3—figure supplement 4 for details . DOI: http://dx . doi . org/10 . 7554/eLife . 06266 . 015 First , we examine the case where post-diagnostic resection leaves 106 cells . As suggested by our studies above on prevention , 1 million cells have a high probability of already containing resistant subclones , and deterministic effects dominate subsequent tumour growth dynamics . Comparing the median expectations of years from tumour excision to relapse , early discovery ( at 109 cells ) yields an additional 3 . 4 years compared to late discovery ( at 1011 cells ) at σ = 1 . 5% ( medians for low vs high detection thresholds are 14 . 8 and 11 . 4 years , respectively; Figure 3B ) . Consider the following example: 20 years after resection and commencing treatment , the probability of tumour non-detection ( i . e . , the tumour is either eradicated or does not reach the detection threshold ) is close to zero , regardless of treatment intensity ( Figure 3C ) . Contrast this with cases of prevention starting at the same cancer cell population size ( 106 cells ) but which fail to control the incipient tumour for the 50 years of the simulation: the detection time of these potentially life-threatening tumours is substantially longer than either of the excision cases ( median 25 . 5 years for σ = 1 . 5% , i . e . , 0 . 3–0 . 4% potential cell arrest per day ) , and tumours are managed below the detection threshold after 20 years in more than 80% of cases for any σ > 1 . 0% ( Figure 3C ) . Now consider a residual population of 1/100th the previous case , that is , 104 cells . Here , stochastic effects play a more important role in dynamics ( Figure 3—figure supplement 1A , B ) . Due to initial heterogeneity ( i . e . , the co-occurrence of many subclones ) , when there are 4 and 5 ( 5 and 6 ) additional drivers in the dominant subclones of a residual cancer from an excised tumour of 109 ( 1011 ) cells , we observe a double peak at 4s and 5s ( 5s and 6s ) ( cf Figure 3—figure supplement 1B ) . These peaks in variability of outcomes are a result of the stochastic nature of the appearance of the first resistance mutations and of additional driver mutations . Interestingly , the secondary detection times ( i . e . , when residual or metastatic cells grow to form a new tumour ) are more variable for small initial tumours compared to larger ones ( cf the median 35 . 8 years , 90% CIs [17 . 0 , 70 . 5] years vs 22 . 4 , [13 . 7 , 37 . 0] years for 109 vs 1011 , respectively , with σ = 1 . 5% ) . This effect is due to resistance emergence in more aggressive subclones for larger tumours , such that the tumour relapses more deterministically ( i . e . , with less variability and faster on average ) . The probability of tumour non-detection after 20 years and the distribution of the mean number of accumulated drivers within tumours are shown in Figure 3—figure supplement 1C , D , respectively ( cf with the previous case , shown in Figure 3C , D ) . Importantly , for both thresholds of tumour excision , subsequent cancer cell arrest levels beyond approximately σ = 1 . 5% make little difference in terms of tumour growth ( Figure 3B-D , Figure 3—figure supplement 1B-D ) , since virtually all of the sensitive cells post-excision will be arrested or killed by the measure beyond this level , leaving uncontrollable resistant cells to grow and repopulate the primary tumour site and/or metastases . ( Note that this level is above that found in the previous section . This is because drivers accumulate throughout tumour growth in the results given in Figure 3 , whereas tumours were assumed to only start accumulating the first drivers after growth from M0 cells in Figure 2 and Figure 2—figure supplements 1 , 2 ) . Moreover , we find that for post-diagnostic interventions knowledge about the number of drivers at the time of tumour discovery is a far better predictor of outcome than information about the time from tumour initiation to discovery , and that increases in treatment intensity tend to decrease predictive accuracy ( Figure 3—figure supplements 2–5 ) . The above results consider preventive measures and post-diagnostic interventions as independent rather than alternative approaches . Thus , although prevention delays tumour growth for longer times on average than does post-diagnostic intervention , because prevention is always initiated before diagnosis , when considering the relative benefits and risks of each , the actual time gained by the former relative to the latter in terms of cancer-free life will be less than the differences reported in Figure 3B and Figure 3—figure supplement 1B . Figure 4 presents a hypothetical comparative scenario of prevention vs post-diagnostic intervention . Prevention may either succeed without recurrence , or should the measure initially fail and a tumour be clinically detected , the patient has a ‘second chance’ whereby the tumour is resected and treatment continued ( assumed at the same treatment intensity σ ) , either to a further relapse ( failure ) or non-detection ( success ) ( Figure 4A ) . Compare this scenario with the more standard post-diagnostic resection followed by treatment , which either results in relapse or detection-free life ( Figure 4B ) . These numerical experiments assume the same starting point ( time at which the cancer cell population equals M0 , and drivers and resistant subclones are present ) for each tumour , and because of a ‘second chance’ following initial failure in prevention , are run for a maximum of 50 years after the starting point ( same as the numerical studies in the previous section ) . We also assume , as before , that potential therapeutic resistance mechanisms to all intervention types are identical . 10 . 7554/eLife . 06266 . 016Figure 4 . Hypothetical process of preventive ( with a ‘second chance’ ) and post-diagnostic measures . A tumour is initiated by one cell and grows to size M0 ( either 104 or 106 cells in our numerical studies ) . Prevention ( A ) arrests tumour growth at intensity σ ( daily level = σ/4 ) . Should the tumour grow to 109 cells , it is diagnosed and resected to M = M0 cells and then treated again at intensity σ . Post-diagnostic intervention ( B ) does not discover the growing tumour until 109 cells ( i . e . , σ=σ^=0 ) , whereupon it is resected to M = M0 cells and then treated at intensity σ > 0 . Either intervention finally ‘fails’ should the tumour attain 109 cells a second time , no later than 50 years after the initial lesion of size M0 . Should the tumour be eradicated or not exceed 109 cells by 50 years after the initial lesion , then the intervention is deemed a ‘success’ . DOI: http://dx . doi . org/10 . 7554/eLife . 06266 . 016 Figure 5 presents the comparative outcomes ( see also Videos 4 , 5 ) . When prevention starts at ( or tumour resection misses ) relatively large cancer cell populations ( 1 million cells ) , only small comparative gains occur from higher cell arrest in terms of outright treatment success ( Figure 5A ) , whereas interventions starting at much smaller cancer cell numbers ( 10 , 000 ) result in considerably greater outright success ( Figure 5B ) . Looking at situations of relapse only for prevention vs post-diagnostic intervention , the former generally results in superior outcomes in terms of delaying tumour growth , particularly for large residual cell populations ( cf Figure 5C , D ) . In contrast , for lower numbers of residual cells , some post-diagnostic resected tumours in the sample will be initially resistance free ( cf Figure 5—figure supplement 1A , B ) . This , together with fewer accumulated drivers in the highest driver subclones , contributes to improved outcomes should relapse occur ( Figure 5D ) and overall treatment success at sufficiently high treatment intensities ( Figure 5A , B , E , F ) . Importantly , resected tumours in both the prevention ( when it initially fails ) and post-diagnostic scenarios may contain numerous resistant cells ( example of 0 . 25% daily cellular arrest: Figure 5—figure supplements 2 , 3 ) . Prior selection for resistance in initially failed prevention generally results in larger residual resistant cell populations than pre-therapeutic residual populations in post-diagnostic situations ( filled bars , cf captions A and B in Figure 5—figure supplements 2 , 3 ) , but smaller residual resistant cell populations than treatment failures following post-diagnostic resection ( hatched bars , cf captions A and D in Figure 5—figure supplements 2 , 3 ) . Note that , as expected , secondary failures are associated with larger percentages of resistant subclones and a shift in the distributions towards more drivers ( cf captions C and D in Figure 5—figure supplements 2 , 3 ) . 10 . 7554/eLife . 06266 . 017Figure 5 . Comparison of preventive ( blue lines and shading ) and post-diagnostic ( red lines , yellow shading ) interventions . Tumours are either treated at M0 = 106 cells ( left panels ) or M0 = 104 cells ( right panels ) . ( A , B ) Probability of treatment success , defined as the proportion of cases where the tumour remains undetected ( either extinct or below 109 cells ) by 50 years after the initial lesion of M0 cells . ( C , D ) Distribution of times to relapse for treatment failures . ( E , F ) Distribution of detection times for all cases including relapsed tumours and tumours remaining undetected prior to and after 50 years ( detection times are assigned to 50 years in the latter case ) . Parameters as in Table 1 . See Figure 3 for details . DOI: http://dx . doi . org/10 . 7554/eLife . 06266 . 01710 . 7554/eLife . 06266 . 018Figure 5—figure supplement 1 . Resistant cell populations after initial failure . Tumours are either treated at M0 = 106cells ( A ) or M0 = 104cells ( B ) . Red lines and yellow shading = population following resection . Parameter values as in Table 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06266 . 01810 . 7554/eLife . 06266 . 019Figure 5—figure supplement 2 . Distribution of mean sizes of subclones . ( A ) Cases where a tumour is detected and resected following prevention . ( B ) Cases where a tumour is detected and resected with no prevention . ( C ) Cases of relapse following resection and secondary treatment to initially failed prevention . ( D ) Cases of relapse following resection and primary treatment in cases where there was no prevention . Hatched bars indicate cell numbers in the tumour and solid bar numbers after resection and 106 residual or metastatic cells . Daily arresting level assumed to be 0 . 25% . Parameter values as in Table 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06266 . 01910 . 7554/eLife . 06266 . 020Figure 5—figure supplement 3 . Distribution of mean sizes of subclones . Same as Figure 5—figure supplement 2 , except M0 is 104 cancer cells . DOI: http://dx . doi . org/10 . 7554/eLife . 06266 . 02010 . 7554/eLife . 06266 . 021Video 4 . Comparison of preventive ( blue lines and shading ) and post-diagnostic ( red lines , hatched ) interventions . Tumours are treated at M0 = 106 cells . ( A ) The median ( thick line ) and 90% confidence intervals ( shaded areas with dashed boundaries ) for the distribution of times to relapse for treatment failures . ( B ) and ( C ) Arbitrary samples of the distribution of detection times for preventive and post-diagnostic interventions , respectively . The colour-code indicates the mean number of accumulated drivers over a period of 1 year . The rectangles on the top of B and on the bottom of C show the fifth and 95th percentiles , the blue circle indicates the median , and the red line is the mean . Parameters as in Table 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06266 . 02110 . 7554/eLife . 06266 . 022Video 5 . Comparison of preventive ( blue lines and shading ) and post-diagnostic ( red lines , hatched ) interventions . Tumours are treated at M0 = 104 cells . ( A ) The median ( thick line ) and 90% confidence intervals ( shaded areas with dashed boundaries ) for the distribution of times to relapse for treatment failures . ( B ) and ( C ) Arbitrary samples of the distribution of detection times for preventive and post-diagnostic interventions , respectively . The colour-code indicates the mean number of accumulated drivers over a period of 1 year . The rectangles at the top of B and the bottom of C shows the fifth and 95th percentiles , the blue circle indicates the median , and the red line is the mean . Parameters as in Table 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06266 . 022 Figure 5E , F shows the distributions of detection times for all numerical experiments . We see that when both non-relapse ( Figure 5A ) and relapse ( Figure 5C ) are taken into account for large cancer cell populations ( 1 million cells ) , treating preventively at levels beyond about 0 . 3% arrest per day increases median delays in detection times due to outright success ( i . e . , survival beyond 50 years ) but has no effect on the lower 95th percentile ( Figure 5E ) . ( Although not shown , arrest beyond approximately 0 . 6% per day does not yield further gains ) . In contrast , post-diagnostic intervention improves only marginally beyond daily arrest levels of about 0 . 3% ( Figure 5E ) . Figure 5F shows the corresponding results for smaller cancer cell populations ( based on integrating the results in Figure 5B , D ) , whereby a high median probability of full success is obtained >0 . 1% and >0 . 3% daily arrest for prevention and post-diagnostic intervention , respectively ( Figure 5F ) . Thus for both cell population levels , prevention generally results in better outcomes compared to post-diagnostic intervention . Whereas primary prevention is becoming an increasingly significant approach in reducing risk of certain cancers ( e . g . , Colditz and Bohlke , 2014 ) , chemopreventive therapies are uncommon , despite empirical support for their effects ( William et al . , 2009 ) . Several theoretical and in vitro experimental studies indicate that chemoprevention can reduce risks of life threatening cancers . For example , Silva and colleagues ( Silva et al . , 2012 ) parameterized computational models to show how low doses of verapamil and 2-deoxyglucose could be administered adaptively to promote longer tumour progression times . These drugs are thought to increase the costs of resistance and the competitive impacts of sensitive cells on resistant cancer cell subpopulations . However , some of the most promising results have come from studies employing non-steroidal anti-inflammatory drugs ( NSAIDs ) , including experiments ( Ibrahim-Hashim et al . , 2012 ) , investigations of their molecular effects ( Galipeau et al . , 2007; Kostadinov et al . , 2013 ) , and their use ( Cuzick et al . , 2015 ) . For example , Ibrahim and co-workers ( Ibrahim-Hashim et al . , 2012 ) studied the action of NSAIDs and specifically sodium bicarbonate in reducing prostate tumours in male TRAMP mice ( i . e . , an animal model of transgenic adenocarcinoma of the mouse prostate ) . They showed that mice commencing the treatment at 4 weeks of age had significantly smaller tumour masses , and that more survived to the end of the experiment than either the controls or those mice commencing the treatment at an older age . Kostadinov et al . ( 2013 ) showed how NSAID use in a sample of people with Barrett's oesophagus is associated with reductions in somatic genomic abnormalities and their growth to detectable levels . It is noteworthy that it is not known to what extent reductions in cancer progression under NSAIDs are due to either cytotoxic or cytostatic effects or both . Although we do not explicitly model cytotoxic or cytostatic impacts , therapies curbing net growth rates but maintaining them at or above zero could be interpreted as resulting from the action of either cytotoxic and/or cytostatic processes . In contrast , therapies reducing net growth rates substantially below zero necessarily have a cytotoxic component . Our model , or modifications of it to explicitly include cytotoxic and cytostatic effects , could be used in future research to make predictions about optimal dose and start times to achieve acceptable levels of tumour control ( or , e . g . , the probability of a given tumour size and heterogeneity level by a given age ) . Decisions whether or not to employ specific chemopreventive therapies carry with them the risk of a poorer outcome than would have been the case had another available strategy ( or no treatment at all ) been adopted ( Esserman et al . , 2004 ) . This issue is relevant to situations where alterations in life-style , removal or treatment of pre-cancerous lesions , or medications potentially result in unwanted side effects or induce new invasive neoplasms ( e . g . , Berrington de Gonzalez et al . , 2011 ) . Chemopreventive management prior to clinical detection would be most appropriate for individuals with genetic predispositions , familial histories , elevated levels of specific biomarkers , or risk-associated behaviours or life-styles ( Hemminki and Li , 2004; Lippman and Lee , 2006; Sutcliffe et al . , 2009; William et al . , 2009; Hochberg et al . , 2013 ) . Importantly , our approach presupposes that the danger a nascent , growing tumour presents is proportional to its size and ( implicitly , all else being equal ) a person's age . Due caution is necessary in interpreting our results , since studies have argued that metastatic potential rather than tumour size may be a better predictor of future survival ( Hynes , 2003; Foulkes et al . , 2010; Sethi and Kang , 2011 ) . However , given the expectation that prevention typically confronts smaller , less heterogeneous neoplasms , which are less likely to have resistant clones and to have metastasised ( Hochberg et al . , 2013; Gerlinger et al . , 2014 ) , support our basic conclusion that prevention is generally a superior strategy in terms of cancer-free survival compared to post-diagnostic intervention . Over the past decade , several alternative approaches to MTD have been proposed , where the objective is to manage rather than eradicate tumours ( e . g . , Maley et al . , 2004; Komarova and Wodarz , 2005; Gatenby , 2009; Gatenby et al . , 2009a , 2009b; Foo and Michor , 2010; Jansen et al . , 2015 ) . Tumour management attempts to limit cancer growth , metastasis , and reduce the probability of obtaining resistance mutations through , for example , micro-environmental modification , or competition with non-resistant cancer cell populations or with healthy cells . These approaches usually involve clinically diagnosed cancers: either inoperable tumours or residual or metastatic cancers after tumour excision . In the former situation , tumours are typically large enough in size to contain numerous resistance mutations . In many , if not most , cases , these neoplasms will have metastasized , meaning greater variability both in terms of phenotypes and potential resistance to chemotherapies , and in penetrance of therapeutic molecules to targeted tumour cells ( Klein et al . , 2002; Byrne et al . , 2005 ) . In contrast , the latter situation involves smaller , residual , or metastatic cancer cell populations , composed of high frequencies of resistant variants or dormant cells ( Klein et al . , 2002 ) . According to our results , both scenarios are likely to involve populations with large numbers of accumulated driver mutations ( or , although not considered in our study , fewer driver mutations but each with larger selective effect ) , which ostensibly contribute to the speed of relapse . Thus , management of clinically detected tumours need not only limit the proliferation and spread of refractory subpopulations but should also aim to control the growth of multi-driver subclones ( Figure 5—figure supplements 2 , 3 ) . In other words , in addition to actual resistance mutations ( j = 1 ) , subclones with q drivers will be effectively resistant to therapeutic interventions if q s ≫ σ ( Figure 6 ) . We therefore suggest that the frequency distribution of driver mutations and the distribution of resistant subclones within a heterogeneous cancer cell population could be used to instruct decisions of the time course of treatment levels , with the aims of curbing tumour growth , metastasis , and resistance . We found that tumours typically achieve several additional driver mutations by the time they reach detection ( Figure 3A; Figure 3—figure supplement 1A; Figure 5—figure supplements 2 , 3 ) , which approximates certain estimates ( Stratton et al . , 2009 ) but falls short of others ( Sjoblom et al . , 2006 ) . Our results indicate that the two most important variables in determining therapeutic outcome are ( 1 ) the size of the initial cancer cell population ( i . e . , when prevention commences and/or post-diagnosis , following resection ) , ( 2 ) associated tumour heterogeneity in terms of accumulated drivers , and the presence of resistance phenotypes . This highlights the importance of biomarkers as accurate indicators of otherwise undetectable malignancies ( Roukos et al . , 2007 ) , and the accurate assessment of local or distant metastases ( Pantel et al . , 1999 ) . We suggest that if order-of-magnitude estimates of cell populations and intra-tumour heterogeneity are possible , then low dose , continuous , constant approaches could be established that lower and possibly minimize risks of the emergence of future , life-threatening cancers . According to our model , such options will generally be superior to more aggressive chemotherapies if therapeutic resistance is a risk factor . The framework proposed here is sufficiently general to portray major events in different types of cancer with emphasis on solid tumours . However , some aspects of cancerous tumour growth are considered only implicitly , and further research is required to formulate more realistic models to include , for example , spatial aspects of tumour growth ( Orlando et al . , 2013 ) , competition/cooperation between different subclones ( Korolev et al . , 2014 ) , combinational ( multidrug ) resistance ( Gillet and Gottesman , 2010; Bozic et al . , 2013 ) , drug-addiction , observed for example in certain melanomas ( Das Thakur et al . , 2013 ) , or advantageous resistant mutations , observed in some leukemias ( Michor et al . , 2005 ) . Moreover , future studies should investigate alternatives to the traditional post-diagnostic therapeutic scenarios considered here ( e . g . , molecularly targeted therapies [Yap , 2015] ) . Our study nevertheless predicts that the main hurdle to post-diagnostic MTD interventions remains resistant subclones , since beyond minimal impacts on the order of 0 . 3% per day for the larger of the two residual or metastatic cell populations simulated here ( which are still very small by clinical diagnostic standards—c 1 mm3 ) , increased therapeutic intensity selects disproportionally for resistance and has negligible benefits in terms of delaying life-threatening cancers .

About one person in every two will get cancer during their lives . Surgery and chemotherapy have long been mainstays of cancer treatment . Both , however , have substantial downsides . Surgery may leave behind undetected cancer cells that can grow into new tumours . Furthermore , in response to chemotherapy drugs , some cancer cells may emerge that resist further treatment . There is therefore interest in whether preventive strategies—including lifestyle changes and medications—could reduce the likelihood of confronting a life-threatening cancer . Now , Akhmetzhanov and Hochberg have developed a mathematical model to help compare the effectiveness of preventive strategies and traditional cancer treatments . The model—which assumes that a person can only develop a single cancer from a single region of pre-cancerous cells—suggests that long-term cancer prevention strategies reduce the risk of a life-threatening cancer by more than traditional treatment that begins after a tumour is discovered . The preventive measures may be less effective in some cases compared to traditional treatments if they initially fail to stop a tumour growing , although on average they still work better than treating the cancer after detection . According to Akhmetzhanov and Hochberg's model , surgical removal followed by chemotherapy is less likely to be successful than prevention , and when successful , requires larger impacts on the cancer ( and therefore creates more side-effects for the patient ) to achieve the same level of control as prevention . The model also suggests that even at very low levels of impact on residual cancer cells , chemotherapies are likely to be counterproductive by boosting the subsequent emergence of treatment-resistant tumours . Akhmetzhanov and Hochberg's model predicts how effective preventive measures need to be in terms of slowing the growth of cancer cells to result in given reductions in the future risk of a life-threatening cancer . Future work should test this model by measuring the effects on tumour growth of prevention and of traditional therapies .

2015

Dynamics of preventive vs post-diagnostic cancer control using low-impact measures

Human sound localization is an important computation performed by the brain . Models of sound localization commonly assume that sound lateralization from interaural time differences is level invariant . Here we observe that two prevalent theories of sound localization make opposing predictions . The labelled-line model encodes location through tuned representations of spatial location and predicts that perceived direction is level invariant . In contrast , the hemispheric-difference model encodes location through spike-rate and predicts that perceived direction becomes medially biased at low sound levels . Here , behavioral experiments find that softer sounds are perceived closer to midline than louder sounds , favoring rate-coding models of human sound localization . Analogously , visual depth perception , which is based on interocular disparity , depends on the contrast of the target . The similar results in hearing and vision suggest that the brain may use a canonical computation of location: encoding perceived location through population spike rate relative to baseline . A fundamental question of human perception is how we perceive target locations in space . Through our eyes and skin , the activation patterns of sensory organs provide rich spatial cues . However , for other sensory dimensions , including sound localization and visual depth perception , spatial locations must be computed by the brain . For instance , interaural time differences ( ITDs ) of the sounds reaching the ears allow listeners to localize sound in the horizontal plane . In the ascending mammalian auditory pathway , the first neural processing stage where ITDs are encoded , on the timescale of microseconds , is the medial superior olive ( MSO ) . Here , temporally precise binaural inputs converge , and their ITDs are converted to neural firing rate ( Goldberg and Brown , 1968; Yin and Chan , 1990; Spitzer and Semple , 1995; Pecka et al . , 2010; Day and Semple , 2011 ) . The shape of the MSO output firing rate curves as a function of ITD resembles that of a cross-correlation operation on the inputs to each ear ( Batra and Yin , 2004 ) . How this information is interpreted downstream of the MSO has led to the development of conflicting theories on the neural mechanisms of sound localization in humans . One prominent neural model for sound localization , originally proposed by Jeffress ( 1948 ) , consists of a labelled line of coincidence detector neurons that are sensitive to the binaural synchronicity of neural inputs from each ear , with each neuron maximally sensitive to a specific magnitude of ITD ( Figure 1A ) . This labelled-line model is computationally equivalent to a neural place-code based on bandlimited cross-correlations of the sounds reaching both ears ( Domnitz and Colburn , 1977 ) . Several studies support the existence of labelled-line neural place-code mechanisms in the avian brain ( Carr and Konishi , 1988; Overholt et al . , 1992 ) , and versions of it have successfully been applied in many engineering applications predicting human localization performance ( e . g . Durlach , 1963; Hafter , 1971; Stern and Trahiotis , 1995; Breebaart et al . , 2001; Hartmann et al . , 2005 ) . A growing literature proposes an alternative to the labelled-line model to explain mammalian sensitivity to ITD ( Lee and Groh , 2014 ) . One reason for an alternative is that two excitatory inputs should suffice to implement the labelled-line model , but evidence from experiments on Mongolian gerbils shows that in addition to bilateral excitatory inputs , sharply tuned bilateral inhibitory inputs to the MSO play a crucial role in processing ITDs ( Brand et al . , 2002 ) . Moreover , to date no labelled-line type neurons encoding auditory space have been discovered in a mammalian species . Indeed , using a population rate-code , several studies proposed that mammalian sound localization can be modeled based on differences in firing rates across the two populations of neurons that are tuned to opposing hemispheres ( Figure 1B; van Bergeijk , 1962; McAlpine and Grothe , 2003; Devore et al . , 2009 ) . Rate-based models generally predict that neuronal responses carry most information at the steepest slopes of neural-discharge-rate versus ITD curves , where neural discharge changes most strongly ( Stecker et al . , 2005 ) , consistent with the observation that the peak ITDs of rate-ITD curves often fall outside the physiologically plausible range ( McAlpine and Grothe , 2003; Grothe et al . , 2010; but see also Joris et al . , 2006 ) . In addition , some authors have suggested that how mammalian sound localization adapts to stimulus history further supports a rate-based neural population code , as assessed behaviorally or via magnetoencephalography ( Phillips and Hall , 2005; Stange et al . , 2013; Salminen et al . , 2010 ) . It is unknown which of the two competing models , broadly characterized as labelled-line versus rate-code model , describes human sound localization better . Here , we observe that the two different models predict different dependencies of sound localization on sound intensity . By combining behavioral data on sound intensity dependence in normal-hearing listeners with numerical predictions of human sound lateralization from both models , we attempt to disentangle whether human auditory perception is based on a place-code , akin to the labelled-line model , or whether it is instead more closely described by a population rate-code . An extensive physiology literature characterizes labelled-line versus population-rate type neurons and suggests that , at least from the perspective of evolution , birds and mammals use different neural mechanisms to calculate sound direction ( review: Grothe , 2003 ) . Thus , we searched the avian and mammalian physiology literature and identified two studies that characterized labelled-line versus population rate-code neurons at low sound levels and as a function of both sound level and ITD ( Peña et al . , 1996; Zwiers et al . , 2004 ) . Both Peña et al . ( 1996 ) and Zwiers et al . ( 2004 ) report neural firing rate in response to acoustic noise stimuli and are thus suitable for predicting each model’s sensitivity to the acoustic noises we tested in the current study . Here , we ran a meta-analysis , reconstructing simulated neurons with response characteristics from each of the two studies and using maximum likelihood estimation to predict source laterality from these previous findings . To predict how lateralization depends on sound intensity from the responses of labelled-line neurons , we estimated neural firing rates from previous recordings in the nucleus laminaris in barn owl ( Peña et al . , 1996 ) . To estimate lateralization’s dependence on level based on a population rate-code , we used previous recordings from the inferior colliculus of rhesus macaque monkey and calculated hemispheric differences in firing rate ( Zwiers et al . , 2004 ) . The labelled-line neurons predicted that , as sound intensity decreases , perceived source laterality would converge towards similar means for low versus high sound intensities , with increased response variability at decreasing sound intensities ( Figure 1C ) . In contrast , the hemispheric-difference model predicted that as sound intensity decreases to near threshold levels , perceived laterality would become increasingly biased toward the midline reference ( Figure 1D , for example note the shallower slope and thus compressed laterality percepts for red versus blue curves ) . At higher overall sound intensities , both models predicted that lateralization would be intensity invariant ( see insets in Figure 1C versus D ) . Therefore , analyzing how sound intensity affects perceived sound direction near sensation threshold offers an opportunity to disentangle whether our human auditory system relies on a place-based or rate-based population code for localizing sound based on ITD . A listener’s ability to discriminate ITD can vary with sound intensity ( Dietz et al . , 2013 ) . However , it is difficult to interpret previous findings linking sensitivity to ITD and a listener’s judgement of sound source direction as a function of sound intensity . Some reported decreased perceived source laterality near sensation threshold ( Teas , 1962; Sabin et al . , 2005 ) , but others reported weak or no level effects on perceived lateralization ( Von Békésy and Wever , 1960; Mickunas , 1963; Hartmann and Rakerd , 1993; Macpherson and Middlebrooks , 2000; Inoue , 2001; Vliegen and Van Opstal , 2004; Brungart and Simpson , 2008; Gai et al . , 2013 ) . Several factors complicate the interpretation of these previous findings in the context of the current hypothesis . For instance , assuming an approximately 30 dB dynamic range of rate-level function either at the MSO or downstream in the binaural pathway ( e . g . medial superior olive: Goldberg and Brown , 1968; inferior colliculus: Zwiers et al . , 2004 ) , for stimuli at higher sensation levels ( SL ) where the rate-level functions saturate , both the labelled-line and the hemispheric difference model predict level invariance . This could explain how studies that tested for the role of sound level over a range of high intensities did not see an effect . Moreover , when presented in the free field , sounds also contain interaural level differences and spectral cues , in addition to ITD . For low-frequency sound , listeners rely dominantly on ITD when judging lateral source angle . However , for broadband sound , listeners integrate across all three types of spatial cue ( Wightman and Kistler , 1992; Ihlefeld and Shinn-Cunningham , 2011 ) . Unlike ITDs , interaural level differences and overall sound intensity both decrease with increasing source distance , raising the possibility that for stimuli with high-frequency content , listeners judged softer sounds to be more medial because they interpreted them to be farther away than louder sounds . Further , at low sound intensities , the sound-direction-related notches of the spectral cues at high-frequencies should have been less audible than at higher sound intensities , increasing stimulus ambiguity . A resulting increase in response variability may have obscured the effect of sound intensity on ITD coding . Finally , some historic studies used only two or three listeners , suggesting that they may have been statistically underpowered . Thus , the literature provides insufficient evidence on how ITD-based lateralization varies with sound level near sensation threshold . Here , we contrasted two competing hypotheses toward the goal of disentangling whether ITD-based human sound localization relies on a labelled-line versus a population rate-place neural code . The labelled-line code hypothesis predicted that the mean perceived direction based on ITD would be intensity invariant , even at intensities close to SL . Using a psychophysical paradigm , we studied lateralization based on ITD as a function of sound intensity in a group of ten normally hearing listeners ( experiment 1 ) . Stimuli consisted of low-frequency noise tokens that were bandlimited to cover most of the frequency range where humans can discriminate ITD ( Brughera et al . , 2013; here , corner frequencies from 300 to 1200 Hz , shown in Figure 2A ) . In each one-interval trial , listeners had to indicate perceived laterality across a range of ITDs from −375 to 375 μs . Lateralization was measured as function of SL . To examine how sound intensity affects perceived ITD coding of source direction , we modelled perceived laterality with a nonlinear mixed effect model ( NLME ) that included fixed effects of ITD and sound intensity as well as a random effect of listener . Figure 2B depicts lateralization performance with spectrally flat noise at two sound intensities for a representative listener ( TCW ) . Figure 2C shows raw data ( circles ) and NMLE fits ( lines ) across all listeners . Error bars show one standard error of the mean across listeners , and shaded ribbons indicate one standard error of the mean fit across listeners . This model predicts 80 . 6% of the variance in the measured responses and is deemed an appropriate fit of the data . Table 1 lists all NLME parameters . Perceived laterality scores increased with increasing ITD , as expected . With decreasing sound intensity , percepts were increasingly biased towards midline ( compare order of colored lines , magnified in the inset of Figure 2C ) . These trends were supported by the NLME model , which revealed significant effects of ITD ( p<0 . 001; αx1 ) and sound intensity ( p<0 . 001; αy1 ) on the maximal extent of laterality , confirming the predicted trend from the hemispheric difference model and rejecting our null hypothesis . Average pure tone audiometric thresholds affect perceived laterality , albeit mildly ( p<0 . 001; αy2 = 0 . 01 ) . Sound intensity did not significantly affect the slope of the psychometric functions ( p=0 . 14; αx2 ) . In a second experiment , we examined whether these results were robust to the spectral details of the stimuli . A caveat of testing spectrally flat noise at low sound intensities is that parts of the spectrum may be inaudible , and this may contribute to the slight but significant effect of audibility on laterality ( αy2 ) . Therefore , the results of experiment 1 could potentially be confounded by the fact that the bandwidth of the audible portion of the noise tokens decreased with decreasing sound intensity . Alternatively , the effect of absolute pure tone detection thresholds that we observe in our normal-hearing listeners may reflect differences in neural function beyond audibility . As a control for perceived stimulus bandwidth , the same listeners were tested again , using inverse A-weighted noises ( experiment 2 ) . Inverse A-weighting boosts sound energy at each frequency in rough proportion to the human threshold . Resulting inverse A-weighted sensitivity thus achieves nearly constant sensation level across frequency . All of the original ten listeners from experiment 1 completed experiment 2 . Methods were similar as in the first experiment , except that the stimuli consisted of inversely A-weighted noise ( compare magnitude spectra in Figure 2A ) . The data and NLME model fits for the second experiment are shown in Figure 2D ( color key identical to Figure 2C ) , and coefficients are listed in Table 2 . This second model accounts for 80 . 4% of the variance in the data , closely fitting the measured responses . All NLME coefficients are significant ( p<0 . 001 for αx1; αx2; αy1 and αy2 ) . The intercept coefficient ( αx0 estimate = −0 . 60 , SE = 0 . 03 , p<0 . 001 ) revealed a slight leftward response bias , consistent with a slight narrowband interaural level difference in our stimuli due to precision limits of our test system . The fact that αx2 is significant shows that when all noise portions are approximately equally audible , as here , with inverse A-weighted noise , both perceived laterality and the slope linking the change in laterality to ITD decrease with decreasing sound intensity . This is consistent with the interpretation that by controlling for audibility across-frequency , the sensitivity of the task to sound level increases , revealing a medial bias effect not only for the most lateral but also for more medial source angles . Thus , the results confirm the effect of biasing perceived laterality toward midline with decreasing sound intensity . Therefore , for both spectrally flat noise and A-weighted noise , statistical analyses , which partialed out overall differences between listeners , are inconsistent with a labelled-line model of human sound localization . Population rate-coding to compute sensory dimension may not be unique to the auditory system . In analogy to sound localization based on the comparison of signals from the two ears ( Figure 3A ) , visual depth is computed in the cerebral cortex based on signals from the two eyes ( Figure 3B; Poggio , 1995; Parker and Cumming , 2001; Parker , 2016 ) . Specifically , in both primary V1 and extrastriate V3a cortex of rhesus macaque monkeys , three types of neurons are thought to encode binocular disparity . ‘Tuned-excitatory’ neurons respond best to zero spatial disparity between the two eyes , whereas ‘near cells’ respond more vigorously when an object approaches , increasing crossed disparity between the eyes ( Parker and Cumming , 2001 ) . Finally , ‘far cells’ fire more vigorously as uncrossed disparity increases . In V1 , the most frequently encountered type of binocular neurons are of the tuned-excitatory type . However , in V3a the large majority of neurons is stereo-specific ( Poggio et al . , 1988 ) and most neurons are either near or far cells . Functional magnetic resonance imaging experiments on human stereoscopic vision found that unlike V1 activity , the activity in cortical area V3a predicts behavioral performance on tasks involving stereoscopic depth ( Backus et al . , 2001 ) . These observations lead us to propose that in order to compute perceptual space from sensory input , the central nervous system has evolved a canonical computation that is common to different sensory modalities . Specifically , we propose that near and far cells encode visual distance from the fixation plane in a way similar to how inferior colliculus neurons encode auditory azimuthal angle away from midline reference: firing rate increases monotonically with distance from perceptual reference anchor or fixation . We observe that in both the auditory and the visual system , the same cells that are tuned to binaural ITD or binocular disparity also have intensity-response functions . A rate-code based on a population of these cells should cause ambiguities when stimulated below the saturation firing rate , either at low sound intensity or at low contrast ( Figure 3C ) . Thus , based on the analogies between the stereo-depth computation and the azimuth-ITD computation , we hypothesized that low visual contrast might affect the computation of depth in a manner analogous to the effect of low sound levels in sound localization—there might be a bias to lower perceived depth at lower contrast ( Figure 3D ) . Indeed , one study found such an effect , but only in some observers ( Cisarik and Harwerth , 2008 ) . A confounding factor in that earlier study is that perceived depth is a complicated neural computation , not only dependent on stereoscopic disparity but also on monocular cues including contrast ( Parker , 2016 ) . Several studies on depth perception indicate that low contrast is interpreted by the brain as a cue for distance; lower contrast targets are perceived farther away ( e . g . Schor and Howarth , 1986; Rohaly and Wilson , 1999 ) . However , experiments that controlled for low contrast bias demonstrated that low contrast causes perceived depth to shrink , both for near and far deviations from baseline ( Chen et al . , 2016 ) . Thus , there is a link between population rate-coding and stimulus intensity in perceived visual depth as in perceived auditory azimuth , two perceptual spatial dimensions computed by the brain . To illustrate how rate-based decoding of target location varies with sound intensity , we here chose a rate-coding model that compares firing rates across two populations of neurons , tuned to opposite hemifields . This read-out is a direct realization of the original canonical rate-based model for ITD decoding ( van Bergeijk , 1962 ) . Alternative rate-code readouts exist ( for a recent summary of binaural models , see Dietz et al . , 2018 ) . Most of these rate-code models rely on subtractive comparisons between populations of neurons that are tuned to opposite hemifields , inherently sharing ambiguous readouts at low suprathreshold sound intensities . In contrast , divisive comparisons between ipsi- and contralaterally tuned neural populations are less likely to predict the observed behavioral bias due to stimulus intensity ( Groh , 2001 ) . Future work will need to delineate how specific implementations of rate-based readouts shape the intensity-induced bias of sound localization . Moreover , it has been suggested that depending on perceptual task , the mammalian brain could combine place- and rate-codes ( Porter and Groh , 2006; Goodman et al . , 2013 ) . For instance , the mammalian auditory pathway may convert place- into rate-codes and vice versa ( Groh et al . , 2003; Porter and Groh , 2006 ) . However , downstream from the inferior colliculus , rate-coding seems to be maintained , at least in the superior colliculus of rhesus macaque ( Werner-Reiss and Groh , 2008; Lee and Groh , 2014 ) . Moreover , our psychophysical and computational results suggest that for sound localization based on ITD at low sound levels , cortical maps do not play a role . However , there is good evidence for spatial map-like signals in higher order auditory cortical fields when interaural level differences are present , at medium to high sound levels ( Higgins et al . , 2010 ) . How these interaural-level-difference-based cortical maps influence sound localization behavior is yet to be determined . An additional factor restricting rate-based readouts is that auditory cortex units display nonlinear rate-intensity functions . For instance , excitatory-excitatory ( EE ) cells in auditory cortex that are tuned to sound locations near midline are also often tuned for sound intensity ( Semple and Kitzes , 1993; Pollak et al . , 2002; Zhang et al . , 2004; Razak and Fuzessery , 2010; Higgins et al . , 2010 ) . This intensity tuning may complicate rate-based decoding at higher sound intensities . However , it is not apparent at the very low sound intensities needed to explain the perceptual bias observed here . There are additional fascinating findings in the neurophysiological literature regarding frequency and intensity tuning , and interesting correlations between non-monotonicity in the azimuthal and intensity dimensions ( Woods et al . , 2006 ) , but a detailed discussion of these points is beyond the scope of the present behavioral-computational study . In summary , unlike predictions from a rate-code neuronal readout , labelled-line coding predicts that sound localization is intensity invariant . Our experimental results show that for low frequency noise , where ITDs are the dominant localization cue , and at low sound intensities , sound lateralization based on ITD is not intensity invariant; it becomes increasingly medially biased with decreasing SL . The observed localization bias is overall small in magnitude , showing that the brain can robustly localize based on ITD across a large range of sound intensities . However , this bias is of theoretical importance as it confirms the prediction of a subtractive rate-based neuronal readout . Moreover , our auditory finding parallels a phenomenon of visual fixation bias when calculating visual distance from binocular disparity at low contrast . This casts doubt on the idea that the neural mechanism of ITD-based sound localization and binocular disparity-based visual distance estimation are based on place-based coding . Instead , our perceptual data on auditory localization together with previously published data on visual distance perception are parsimonious with the idea that a population rate-code underlies the brain’s computation of location . Twelve naïve normal-hearing listeners ( ages 18–27 , five females ) were enrolled in this study and paid for their time . Their audiometric thresholds , as assessed via a calibrated GSI 39 Auto Tymp device ( Grason-Stadler ) , were 25 dB hearing level or better at octave frequencies from 250 to 8000 Hz , and did not differ by more than 10 dB across ears at each octave frequency . This study was approved by and all testing was administered according to the guidelines of the Institutional Review Board of the New Jersey Institute of Technology , protocol F217-14 . All listeners gave written informed consent both to participate in the study and to publish the results with confidential listener identity . Listeners were seated in a double-walled sound-attenuating booth ( Industrial Acoustics Company ) with a noise floor of 20 . 0 dB SPL ( wideband LAFeq ) . Stimuli were digitally generated in Matlab R2016b ( The MathWorks , Inc ) , D/A converted through an external sound card ( Emotiva Stealth DC-1 ) at a sampling frequency of 192 kHz , with a resolution of 24 bits per sample , and presented to the listener through ER-2 insert earphones ( Etymotic Research Inc ) . The equipment was calibrated using an acoustic mannequin ( KEMAR model , G . R . A . S . Sound and Vibration ) with a precision of less than ±5 μs ITD and less than ±1 dB interaural level difference . Foam eartips were inserted following guidelines provided by Etymotic Research to encourage equal representation of sounds to both ears and minimize interaural leakage . Each session lasted approximately 60 min . Listeners kept the insert earphones placed inside their ears throughout testing . Insert earphones were replaced by the experimenter after each break . Throughout this study , to generate stimuli , tokens of uniformly distributed white noise were generated and bandpassed using a zero-phase Butterworth filter with 36 dB/octave frequency roll-off , and 3 dB down points at 300 and 1200 Hz . Each noise token was 1 s in duration , including 10 ms long squared cosine ramps at the onset and offset . At the beginning of each session , and , as a re-test control , mid-way through each session , each individual listener’s SL was measured for the type of sound that was later on used for training and testing , via one run of adaptive tracking . On each one-interval trial of each track , a new noise token was generated and presented diotically . Trials were spaced randomly in time ( uniform distribution , inter-token intervals from 3 to 5 . 5 s ) . Listeners pressed a button when they heard a sound . No response feedback was given . On each trial , a response was scored a ‘hit’ if a listener responded with a button push before the onset of the subsequent trial , and a ‘miss’ if the listener did not respond during the interval . If a listener’s response changed from hit to miss or from miss to hit across sequential trials , this was interpreted as a response reversal . Using one-up-one-down adaptive tracking , the noise intensity was increased or decreased after each reversal , with a step size of 5 dB ( decreasing ) or 2 . 5 dB ( increasing ) . Each listener completed ten adaptive-track reversals , with SL threshold equaling the median of the final six reversals . Each SL was used as reference intensity for the subsequent 30 min of testing . If detection thresholds changed between initial test and re-test control by more than 5 dB , this indicated that an insert earphone moved , and the experimenter replaced the earphones . Thresholds generally did not change by more than 5 dB . To train listeners on consistently reporting their perception of ITD , using adaptive tracking , listeners matched the perceived laterality of a variable-ITD pointer to that of a fixed-ITD target . Target token intensity was set relative to the listener’s own diotic sensation threshold , at 10 or 25 dB SL , and presented with 0 dB interaural level difference . The pointer intensity was fixed at 25 dB SL . Target ITDs spanned the range from −375 to 375 μs , in 75 μs steps . Target ITDs and SLs were randomly interleaved across runs , but held fixed throughout each adaptive run . In each two-interval trial of a run , the pointer token was presented in the first , and the target token in the second interval . The start ITD of the pointer token at the beginning of each run equaled 0 μs . Using a hand-held controller ( Xbox 360 wireless controller for Windows , Microsoft Corp . ) , listeners adjusted the ITD of the pointer token . Specifically , listeners pushed the directional keys ( D-pad ) either to the left or right in order to move and match the pointer direction with that of the target sound . When a listener indicated a left- or right-ward response , the pointer ITD was decreased or increased . Initial ITD step size equaled 100 μs , then 50 ±5 μs ( uniformly distributed ) after the first reversal . By the end of the second reversal , ITD step size was reduced to 25 ±5 μs ( uniformly distributed ) and remained the same for all of the following reversals . Listeners were instructed to ‘home in’ on the target by moving the pointer initially to a position more lateral than the target , then more medial than the target with the goal of centering on the target . No response feedback was provided . A run was completed after a listener had completed a total of five adaptive-tracking reversals . For each target ITD , the matched pointer ITD was estimated by averaging the pointer ITDs of the final two reversals . Each listener performed three sessions of training: In the first session only a subset of target ITDs were presented ( −375 , –150 , 0 , 150 and 375 μs ) , whereas the two following sessions included all of the eleven ITDs . Per training session , each ITD was presented once at 10 and 25 dB SL , for a total of 54 adaptive tracking runs across all training sessions . To familiarize listeners with the experimental task ( described below ) , at the end of second and third sessions of training listeners performed an additional 5 blocks of the experimental testing task , without response feedback . These task training data were not used for statistical analysis . To assess whether listeners could reliably report their lateralization percepts , training performance was evaluated for each listener by calculating the Pearson correlation coefficient between target ITD and matched pointer ITD in the final training session . Criterion correlation equaled 0 . 9 ( N = 11 ITDs , significance level = 0 . 01 , power = 0 . 95 ) . Ten listeners reached criterion , suggesting that they were able to consistently report where they perceived the sounds based on ITD . Two of the originally recruited twelve listeners failed to reach training criterion ( R2 = <0 . 84 , 0 . 87> ) and were excluded from testing . Using the method of fixed stimuli , we tested lateralization in two experiments . Except for the stimuli , which consisted of spectrally flat noise tokens in experiment 1 and A-weighted noise tokens in experiment 2 , the methods were similar across the two experiments . Noise tokens were generated from a statistically similar noise distribution as those presented during both SL measurements and training ( see Overall Design ) . A touchscreen monitor ( Dell P2314T ) displayed the response interface at about 40 cm distance from the listener . Using a precise touch stylus ( MEKO Active Fine Point Stylus 1 . 5 mm Tip ) , listeners indicated perceived laterality of noise in a one-interval task . Noise tokens were presented at 5 , 10 , 15 , 20 , and 25 dB SL . ITDs varied randomly from trial to trial , in 75 μs steps spanning the range from −375 μs to 375 μs . On each trial , a new token of noise was generated . Each listener performed 20 blocks of 55 trials each ( 11 ITDs at each of the five sound intensities ) , with SL measured both before the first and the eleventh block . ITDs and sound intensity were randomly interleaved from trial to trial such that each combination of ITD and sound intensity was presented once before all of them were repeated in a different random order . We estimate the combined effects of ITD and sound intensity on predicted source laterality both in avian labelled-line type units and in binaurally sensitive units of a mammalian auditory system . The sound intensities where we expect to see an effect of overall sound level fall below 30 dB SPL , because only in this range would most auditory neurons fire below saturation , allowing us to disambiguate labelled-line versus hemispheric rate-difference coding . However , scant data exist for either type of unit at sound pressure levels below 30 dB SPL . We identified two prior studies that have measured neural discharge rate as a function of ITD at these very low sound intensities . Both studies used noise as acoustic stimuli , and the neural response statistics they report are thus suitable for estimating what type of information would be available to either type of coding mechanism with the type of noise stimuli that human listeners lateralized in the behavioral experiments here . One study in barn owl shows that the output functions of nucleus laminaris neurons can be modeled through interaural cross-correlation functions , even at very low sound intensities ( Peña et al . , 1996 ) . That study reports Pearson correlation coefficients between the neural response function of nucleus laminaris units at 50 dB SPL versus all other tested sound levels . To reconstruct the spatial information realistically available from the output of labelled-line neurons , across both a range of −375 to 375 μs ITD in 20 μs steps , we first constructed biologically plausible interaural cross-correlation functions at 50 dB SPL and then added internal noise to the resulting curves to mimic the Pearson correlation coefficients reported by Peña et al . ( 1996 ) . Our model predictions pertain to sound intensities spanning the range from 10 to 70 dB SPL , similar to previous work ( Peña et al . , 1996 ) . Due to overall scarcity of available data at low dB SPL , here we use firing rate characteristics for unit # 0123795–530 . 02 ( Peña et al . , 1996 ) with a nominal best frequency of 1 kHz . To generate the acoustic inputs to the labelled-line model , we initially generated a Gaussian noise token , duplicated it and introduced a variable ITD , spanning a range from −375 to 375 μs , with 20 μs step size and 0 dB interaural level difference . To simulate ITD information available after cochlear processing , we then processed both noises with a 1/3-octave wide bandpass filter with 24 dB/octave frequency roll-off , followed by half-wave rectification and low-pass filtering at 1500 Hz . We then simulated internal noise by adding uniformly distributed dichotic noise tokens with mean spontaneous firing rates of 5% of the root mean square value of the signal , resulting in left ( L ) and right ( R ) inputs to the binaural cross-correlation neurons , called xL ( t ) and xR ( t ) . To establish 50 dB SPL reference functions , at each simulated ITD , we then calculated the binaural cross-correlation function cc ( τ ) of xL ( t ) and xR ( t ) , as follows: cc ( τ ) =300+ ( 450−300 ) ∫−∞+∞ xL ( t ) xR ( t+τ ) dt ) max|∫−∞+∞ xL ( t ) xR ( t+τ ) dt| , with τ signifying the best ITD of each neuron , and extrema scaled such that cc ( τ ) spans a range from 300 to 450 spikes/sec , approximating nucleus laminaris firing rates at 50 dB SPL ( Peña et al . , 1996 ) . To simulate non-sound driven neural discharge , we then added uniformly distributed random noise cc^ref ( τ ) =cc ( τ ) +U ( 0 , μ ) , with a mean discharge of μ = 5 spikes/sec , ( Peña et al . , 1996 ) . The resulting signal is our reference cross-correlation function at 50 dB SPL , called ccref ( τ ) , shown in Figure 1A as yellow bold line for a representative simulated neuron . For each sound level and ITD , we then statistically reconstructed a family of interaural cross-correlation functions that match the originally reported functions ( Peña et al . , 1996 ) . Specifically , we added scaled dichotic uniformly distributed noise tokens nL ( t ) ←U ( 0 , μ ) and nR ( t ) ←U ( 0 , μ ) to the xL ( t ) and xR ( t ) , such that the monaural inputs to the binaural cross-correlation functions equal x^L , R ( t ) =αxL , R ( t ) +1−α2nL , R ( t ) . The resulting cross-correlation function for each sound level and ITD is then cc^ ( τ ) = ( xL⋆xR ) ( τ ) , shown for a representative neuron in Figure 1A as blue , brown and red lines corresponding to 70 , 30 and 10 dB SPL . We then searched through the space of scaling coefficients α until the Pearson correlation coefficient between ccref ( τ ) and cc^ ( τ ) matched the coefficients originally reported by Peña et al . ( 1996 ) with a precision error of less than 10% . To estimate predicted sound laterality as a function of sound intensity for these simulated labelled-line neurons , at each intensity , we then identified the τ where cc^ ( τ ) =max ( cc^ ( τ ) ) . For each sound level and ITD , we calculated predicted sound laterality in 100 repetitions of these simulations . Figure 1C shows mean estimated laterality across these 100 simulations , with ribbons showing one standard error of the mean across simulations . To estimate source laterality based on rate-coding , we assayed the mammalian auditory system , where one previous study reports firing statistics for 81 inferior colliculus units in rhesus macaque as a function of ITD and over a wide range of sound intensities , including very low sound intensities ( Zwiers et al . , 2004 ) . From the previously published linear regression parameters , we initially reconstruct linear regression functions linking ITD , sound intensity and firing rate ( Zwiers et al . , 2004 ) . However , while linear regression fits afford statisticial convenience , they cannot fully capture the sigmoidally shaped firing rate functions in mammalian inferior colliculus units . Therefore , we multiplied the original linear reconstructions with sigmoid functions . Specifically , consistent with prior literature , each simulated sigmoidal output function saturates over a 30 dB dynamic range , has linear growth over the physiologically plausible range of contralateral ITDs , has a threshold between uniformly distributed between 0 and 10 dB SPL , and a spontaneous non-sound-evoked discharge of between 2 and 10 spikes/second ( e . g . Ramachandran et al . , 1999 ) . The inset of Figure 1B shows a representative simulated inferior colliculus unit ( color denotes sound intensity , dark shading shows contralateral responses ) , whereas Figure 1B shows the differences in firing rates for contra minus ipsi-lateral simulated firing rates , averaged across all 81 simulated inferior colliculus units . From these resulting differences in contra versus ipsi firing rates we calculated , collapsed across sound intensities from 0 to 80 dB SPL , the probability density of the firing rate for each inferior colliculus unit as a function of source ITD . Assuming an ideal observer , we then classified the sound azimuth as a function of sound intensity via maximum likelihood estimation . To calculate the mean and variance of predicted ITD as a function of sound intensity , we then ran a bootstrapping analysis , sampling with replacement 100 times . Figure 1D shows the across-simulation average predicted source laterality , with ribbons showing one standard error of the mean across simulations . Growth curve analysis was used to analyze perceived laterality scores as a function of ITD and sound intensity . For each of the two noise conditions , the perceived laterality scores were fitted with an NLME model . The model included fixed effects α and random effects β . Equation 1 describes a sigmoidal function linking ITD to perceived laterality , with a score from left ( −1 ) to right ( 1 ) . The effect of sound intensity on the maximal extent of lateralization is αy1 . To factor out across-listener differences in absolute hearing thresholds , for each listener , we calculated the pure tone average ( PTA ) detection threshold in quiet , averaged across ears , and across 500 and 1000 Hz . Weight αy2 models the contribution of PTA . The slope terms are αx1 for perceived laterality changes attributed to ITD , and αx2 for laterality-ITD slopes attributed to sound intensity . Our stimuli were initially calibrated to have a broadband interaural level difference of 0 dB . However , because the transfer function of our sound card was not perfectly flat across frequency , fluctuations of ±1 dB interaural level difference occurred across frequency , on the same order of magnitude as the minimal threshold for human interaural level difference discrimination ( Francart and Wouters , 2007 ) . Thus , parameter αx0 factors out central response bias from the lateralization scores . Random effects of individual differences across listeners were used to model both the maximal extent of lateralization , βy0 , listener , and the perceived midline , βx0 , listener , centering the sigmoid ( Equation 1 ) : ( 1 ) response∼αy2×PTA+αy1×intensity+βy0 , listener1+e−[αx2×intensity+αx1× ( ITD−αx0−βx0 , listener ) ]−0 . 5 To better conform with the assumptions of the NLME model , prior to fitting , ITD and sound intensity parameters were scaled by subtracting the mean stimulus value , and dividing by the standard deviation of stimulus parameters , resulting in distributions of stimulus parameters with zero-mean and a variance of one . Laterality scores were then fitted using these normalized parameters , with the nlme package , programmed in RStudio 1 . 1 for Windows ( RStudio Inc , Boston , MA , USA ) . All data and analysis code are available at Dryad ( http://doi . org/10 . 5061/dryad . t8c381f ) .

Being able to localize sounds helps us make sense of the world around us . The brain works out sound direction by comparing the times of when sound reaches the left versus the right ear . This cue is known as interaural time difference , or ITD for short . But how exactly the brain decodes this information is still unknown . The brain contains nerve cells that each show maximum activity in response to one particular ITD . One idea is that these nerve cells are arranged in the brain like a map from left to right , and that the brain then uses this map to estimate sound direction . This is known as the Jeffress model , after the scientist who first proposed it . There is some evidence that birds and alligators actually use a system like this to localize sounds , but no such map of nerve cells has yet been identified in mammals . An alternative possibility is that the brain compares activity across groups of ITD-sensitive nerve cells . One of the oldest and simplest ways to measure this is to compare nerve activity in the left and right hemispheres of the brain . This readout is known as the hemispheric difference model . By analyzing data from published studies , Ihlefeld , Alamatsaz , and Shapley discovered that these two models make opposing predictions about the effects of volume . The Jeffress model predicts that the volume of a sound will not affect a person’s ability to localize it . By contrast , the hemispheric difference model predicts that very soft sounds will lead to systematic errors , so that for the same ITD , softer sounds are perceived closer towards the front than louder sounds . To investigate this further , Ihlefeld , Alamatsaz , and Shapley asked healthy volunteers to localize sounds of different volumes . The volunteers tended to mis-localize quieter sounds , believing them to be closer to the body’s midline than they actually were , which is inconsistent with the predictions of the Jeffress model . These new findings also reveal key parallels to processing in the visual system . Visual areas of the brain estimate how far away an object is by comparing the input that reaches the two eyes . But these estimates are also systematically less accurate for low-contrast stimuli than for high-contrast ones , just as sound localization is less accurate for softer sounds than for louder ones . The idea that the brain uses the same basic strategy to localize both sights and sounds generates a number of predictions for future studies to test .

2019

Population rate-coding predicts correctly that human sound localization depends on sound intensity

We dissected the importance of human telomerase biogenesis and trafficking pathways for telomere maintenance . Biological stability of human telomerase RNA ( hTR ) relies on H/ACA proteins , but other eukaryotes use other RNP assembly pathways . To investigate additional rationale for human telomerase assembly as H/ACA RNP , we developed a minimized cellular hTR . Remarkably , with only binding sites for telomerase reverse transcriptase ( TERT ) , minimized hTR assembled biologically active enzyme . TERT overexpression was required for cellular interaction with minimized hTR , indicating that H/ACA RNP assembly enhances endogenous hTR-TERT interaction . Telomere maintenance by minimized telomerase was unaffected by the elimination of the telomerase holoenzyme Cajal body chaperone TCAB1 or the Cajal body scaffold protein Coilin . Surprisingly , wild-type hTR also maintained and elongated telomeres in TCAB1 or Coilin knockout cells , with distinct changes in telomerase action . Overall , we elucidate trafficking requirements for telomerase biogenesis and function and expand mechanisms by which altered telomere maintenance engenders human disease . Eukaryotic chromosome stability relies on end-protective telomeres ( Arnoult and Karlseder , 2015 ) . In most organisms , end-protective telomeric chromatin assembles on a tract of simple-sequence DNA repeats ( de Lange , 2010 ) , for example the human repeat of TTAGGG on the strand with 3’OH terminus . Maintenance of this telomeric repeat array requires de novo repeat synthesis by the ribonucleoprotein ( RNP ) reverse transcriptase telomerase to balance the repeat erosion inherent in DNA-dependent DNA-polymerase replication of the genome ( Blackburn et al . , 2006; Hug and Lingner , 2006 ) . Telomerase extends chromosome 3' ends by copying a template within the telomerase RNA subunit ( hTR in human cells ) , using an active site in the telomerase reverse transcriptase protein ( TERT ) . The intricate co-folding and co-function of telomerase RNA and TERT obliges a step-wise RNP assembly process and generates a network of protein- and RNA-domain interactions ( Blackburn and Collins , 2011 , Schmidt and Cech , 2015 ) . Cellular RNP biogenesis involves transit through and concentration in specific nuclear bodies ( Mao et al . , 2011; Machyna et al . , 2013 ) . Trafficking pathways differ depending on the diverse steps of RNA processing , modification , and RNP assembly that give a transcript its fate and function . Among the best-studied RNP transit points are Cajal bodies , defined as foci of the protein Coilin ( Nizami et al . , 2010; Machyna et al . , 2015 ) . Enzymes resident in Cajal bodies catalyze numerous RNA processing and modification reactions as well as RNP assembly and remodeling ( Machyna et al . , 2013 ) . Beyond RNA processing and RNP biogenesis factors , Cajal bodies also recruit regulatory complexes such as CDK2-cyclinE ( Liu et al . , 2000 ) and have widespread influence on gene expression ( Wang et al . , 2016 ) . Despite the multiplicity of functions ascribed to Cajal bodies , including critical roles in vertebrate telomerase function described below , it remains unclear whether their formation is a cause or consequence of associated RNP biogenesis pathways . Curiously , ciliate , fungal , and vertebrate telomerases follow entirely different RNP biogenesis pathways , which are directed by telomerase RNA interaction with a La-motif protein , Sm proteins , or H/ACA proteins , respectively ( Egan and Collins , 2012a ) . In human cells , telomerase shares the same mature H/ACA proteins ( dyskerin , NHP2 , NOP10 , GAR1 ) and H/ACA RNP biogenesis chaperones as the intron-encoded small nucleolar ( sno ) or small Cajal body ( sca ) RNPs that catalyze cleavage and pseudouridylation of ribosomal and small nuclear RNAs ( Kiss et al . , 2010 ) . Because precursor hTR is released from its site of synthesis as an autonomous transcript rather than the spliced intron lariat of other human H/ACA RNAs , it is sensitized to degradation in dyskeratosis congenita ( DC ) patient cells with a mutation of an H/ACA protein ( Egan and Collins , 2012b; Armanios and Blackburn , 2012; Sarek et al . , 2015 ) . Also , unlike other H/ACA RNAs , hTR requires a 5' trimethylguanosine cap to prevent 5'-3' exonuclease processing ( Mitchell et al . , 1999 ) . Models for vertebrate telomerase RNA trafficking suggest an initial transit of Cajal bodies , where 5' trimethylguanosine cap modification is thought to occur , followed by localization to nucleoli ( Egan and Collins , 2012a ) . Subsequent RNP trafficking from nucleoli to steady-state concentration in Cajal bodies depends on the binding of the Cajal body chaperone and telomerase holoenzyme protein TCAB1/WDR79/WRAP53β to an hTR 3' stem-loop CAB-box motif ( Venteicher et al . , 2009; Tycowski et al . , 2009; Zhong et al . , 2011 ) , which is present in both stem-loops of an H/ACA scaRNA ( Kiss et al . , 2010 ) . Overall , this trafficking complexity could represent only a subset of the necessary cellular directions for human telomerase biogenesis and function . The human telomerase holoenzyme subunits that localize active RNP to Cajal bodies are considered crucial for telomerase action at telomeres ( Schmidt and Cech , 2015 ) . Transient telomere colocalization with a Cajal body can be detected in S-phase , when telomerase acts at chromosome ends ( Jády et al . , 2006; Tomlinson et al . , 2006 ) . Evidence for Cajal body delivery of telomerase to telomeres builds from studies depleting TCAB1 or Coilin using RNA interference , which reduced or eliminated hTR colocalization with telomeres and induced telomere shortening ( Venteicher et al . , 2009; Zhong et al . , 2011; Stern et al . , 2012; Zhong et al . , 2012 ) . DC patient cells with biallelic TCAB1 mutations have short telomeres and fail to maintain telomere length even with the up-regulated telomerase expression in induced pluripotent cells ( Batista et al . , 2011 ) . With all of this experimental support for Cajal body delivery of human telomerase to telomeres , it is puzzling that mouse telomerase RNA does not localize to Cajal bodies ( Tomlinson et al . , 2010 ) . Also , recently , Coilin gene knock-out ( KO ) in HeLa cells generated two clonal cell lines that maintained telomeres ( Chen et al . , 2015 ) . Here , we address the significance of human telomerase biogenesis and trafficking pathways for active RNP assembly and function at telomeres . As a new approach , we first bypassed the endogenous hTR stability requirement for H/ACA proteins . Remarkably , we found that a minimal hTR ( hTRmin ) containing only binding sites for TERT can assemble active RNP in cells , and this active RNP can maintain stable telomere length homeostasis . TERT overexpression was required for cellular assembly with hTRmin , suggesting that hTR H/ACA RNP assembly enhances TERT interaction at scarce endogenous telomerase subunit levels . KO of TCAB1 or Coilin did not alter hTRmin RNP function at telomeres . Surprisingly , TCAB1 KO or Coilin KO was also permissive for hTR telomerase to maintain telomeres . In cancer and pluripotent stem cell lines with endogenous hTR and TERT expression , TCAB1 KO resulted in a slow decline of telomere length followed by stable telomere length homeostasis . We conclude that aside from conferring hTR stability and facilitating active telomerase assembly in cells with scarce TERT , hTR assembly as H/ACA RNP is not essential . Also , Cajal body localization is not essential for hTRmin or hTR telomerase to maintain stable telomere length homeostasis . Our findings illuminate the influences of nuclear trafficking on human telomerase biogenesis and action at telomeres , account for why telomerase biogenesis pathways can be so divergent in eukaryotic evolution , and give new interpretation to the mechanisms by which different telomerase subunit mutations impose human disease . To test the significance of H/ACA RNP biogenesis for human telomerase function at telomeres , we sought to bypass the essential role of this pathway in the protection of hTR from degradation while preserving hTR-TERT interaction and RNP catalytic activity . The H/ACA-motif 5' stem of hTR ( Figure 1A ) separates two hTR regions that are critical for binding of TERT and catalytic activity: the template/pseudoknot ( t/PK ) and conserved regions ( CR ) 4/5 ( Zhang et al . , 2011 ) . Previously we joined these two regions with a spacer to create hTRmin ( Figure 1B ) , which in combination with TERT reconstitutes a minimized active telomerase in rabbit reticulocyte lysate ( Wu and Collins , 2014 ) . To allow cellular accumulation of hTRmin , the most successful strategy was to append 3' processing and protection motifs from the human long non-coding RNA MALAT1 ( Brown et al . , 2012; 2014 ) . 10 . 7554/eLife . 18221 . 003Figure 1 . Human telomerase RNA can accumulate without H/ACA RNP biogenesis . ( A , B ) Diagrams of hTR and hTRmin secondary structure and bound proteins . ( C ) Parts list for a cellular minimal telomerase RNA . Components are presented in 5’ to 3’ order . ( D , E ) Northern blot assay for RNA accumulation in transfected 293T cells . Endogenous hTR was also detected . Loading control ( LC ) is a cellular RNA non-specifically detected by the Northern blot probe used for normalization . In ( D ) , all hTRmin variants are without a 5’ leader and with a 6 nt spacer . In ( E ) , all constructs had the RNA triplex and RNase P site . hTRmin accumulation was normalized to LC to quantify relative accumulation ( Rel ) . ( F ) Copurification of hTR or LhTRmin with tagged telomerase holoenzyme subunits co-overexpressed in transfected VA-13 cells . RNPs were purified from cell lysate using FLAG antibody resin and analyzed by immunoblot and Northern blot . F indicates 3xFLAG peptide , ZZ indicates tandem Protein A domains . ( G ) Northern blot assay for RNA accumulation in transfected VA-13 cells . RNA folding during extensive gel electrophoresis gives mature hTR two mobilities ( a doublet of bands ) . ( H ) FISH detection of hTR or LhTRmin in transfected HCT116 cells . Untagged TERT was coexpressed . ( I ) Northern blot assay for RNA expressed from transgenes integrated at; in HCT116 cells . Endogenous hTR was also detected . DOI: http://dx . doi . org/10 . 7554/eLife . 18221 . 00310 . 7554/eLife . 18221 . 004Figure 1—figure supplement 1 . Cellular assembly of hTRmin telomerase . ( A ) U2OS cells were transiently transfected with constructs encoding F-TERT and either hTRmin , hTR , or empty vector ( mock ) . TERT was immunopurified on anti-FLAG agarose from cell extract 48 hr post-transfection . An aliquot of the bound samples were treated with TRIzol to purify RNA , which was analyzed by Northern blot . In parallel , bound samples were tested for telomerase activity by primer extension using 32P dGTP for radiolabeling . Products were precipitated and resolved by denaturing gel electrophoresis . A radioactive oligonucleotide recovery control ( RC ) was added prior to precipitation . For the hTR cell extract , different amounts of extract were assayed relative to the hTRmin sample set at 100% . All lanes are from the same gel . ( B ) HCT116 hTR KO#2 cells were transiently transfected to express hTR , CAB-box-mutant hTR ( G414C ) , LhTRmin , or hTRmin . FISH was performed to detect the RNAs ( green ) . Nuclei were counterstained with DAPI ( blue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18221 . 004 The overall design strategy to optimize hTRmin accumulation involved testing combinations and variations of RNA modules ( Figure 1C ) . For example , we interchanged the hTRmin 5’ end as no leader , endogenous hTR quadruplex-forming leader ( LeaderG ) , or a mutant leader ( LeaderC ) that eliminates one of the four adjacent guanosine tracts ( Sexton and Collins , 2011 ) . We also tested 6 or 14 nucleotide ( nt ) lengths of single-stranded RNA spacer separating the t/PK and CR4/5 regions . For 3'-end formation , we tested the MALAT1 triplex motif with or without an adjacent RNase P cleavage site mimicking a pre-tRNA ( Brown et al . , 2012 ) . We also omitted or included a hepatitis delta virus ribozyme ( HDV RZ ) , which can increase hTR accumulation by stabilizing the precursor 3' end with an exonuclease-resistant 2' , 3' cyclic phosphate ( Egan and Collins , 2012b ) . We first assessed hTRmin accumulation in transiently transfected 293T cells , using an expression context previously optimized for hTR that exploits the U3 snoRNA promoter to direct transcription by RNA Polymerase II ( Fu and Collins , 2003 ) . Accumulation of hTRmin to a level detectable by Northern blot required both the RNA triplex and the RNase P site ( Figure 1D ) . We next tested if this cellular hTRmin could form active RNP . We transiently co-expressed hTRmin or hTR with 3xFLAG-tagged ( F ) TERT in the telomerase-negative U2OS cell line and assayed for telomeric primer extension by immunopurified TERT . Cellular reconstitution with hTRmin yielded lower RNA accumulation and proportionally lower overall activity than reconstitution with hTR , but hTRmin and hTR telomerase RNPs had similar profiles of repeat synthesis ( Figure 1—figure supplement 1A ) . For continued optimization , we next compared hTRmin accumulation with no 5’ leader , LeaderG , or LeaderC and with 6 nt versus 14 nt spacer ( Figure 1C ) , all with the RNA triplex and RNaseP cleavage site . The presence of a 5’ leader increased accumulation , and the quadruplex-mutant LeaderC was as good or better than the native hTR LeaderG ( Figure 1E ) . Also , the 6 nt spacer between t/PK and CR4/5 was as good or better than the 14 nt spacer ( Figure 1E ) . HDV RZ addition downstream of the MALAT1 3' processing elements improved hTRmin accumulation from some transfected constructs but did not notably increase accumulation of the optimal LeaderC-containing hTRmin with 6 nt spacer ( Figure 1E ) , which in subsequent experiments we designate LhTRmin for distinction from the leader-less hTRmin . To confirm that removal of the H/ACA motif eliminated all interactions with H/ACA proteins , we expressed epitope-tagged TERT , dyskerin , or TCAB1 with hTR or LhTRmin in telomerase-negative VA-13 cells , immunopurified the tagged protein using FLAG antibody resin , and detected bound RNA by Northern blot ( Figure 1F ) . As expected , TERT , dyskerin , and TCAB1 each bound hTR whereas only TERT bound LhTRmin . As in 293T cells , in VA-13 cells both leader-free hTRmin and LhTRmin accumulated , with LhTRmin being optimal ( Figure 1G ) . The robust 3' end protection activity of the MALAT1 triplex was evident when it was appended to full-length hTR ( Figure 1G; note that mature hTR sometimes migrates as a doublet due to in-gel folding ) . To compare the cellular localizations of LhTRmin and hTR , we overexpressed LhTRmin or hTR with TERT by transient transfection . While hTR concentrated in nuclear puncta as expected , LhTRmin distributed throughout the nucleoplasm ( Figure 1H ) . Transfection of cells lacking endogenous hTR ( see below ) confirmed that hTRmin and LhTRmin have a diffuse nucleoplasmic distribution distinct from that of hTR or hTR with a CAB-box mutation ( Figure 1—figure supplement 1B ) . Together the studies above establish that cellular hTRmin and LhTRmin escape the confines of H/ACA RNP assembly . We next tested whether hTRmin or LhTRmin could accumulate when stably expressed from an integrated transgene . In the near-diploid HCT116 human colon carcinoma cell line , we used a zinc finger nuclease that introduced a double-stranded DNA break at the AAVS1 safe-harbor locus ( Hockemeyer et al . , 2009 ) to integrate an RNA expression cassette and a neomycin resistance cassette . After selection , the polyclonal population of targeted cells was assayed for RNA accumulation by Northern blot . Consistent with results from transient transfection , LhTRmin accumulation exceeded that of hTRmin , with or without a downstream HDV RZ ( Figure 1I ) . Stably expressed hTRmin was barely detectable , and even LhTRmin accumulated to a level lower than endogenous hTR ( Figure 1I ) . Therefore , the MALAT1 3' processing elements do not confer as much accumulation to minimized hTR as does an embedded H/ACA RNP . Successful cellular biogenesis of hTRmin telomerase RNP allowed us to test whether H/ACA proteins have post-biogenesis influences on telomerase function at telomeres . We first programmed Cas9 for cleavage of the endogenous hTR locus ( TERC ) in HCT116 cells in the presence of donor plasmid that would integrate a puromycin resistance cassette ( Figure 2A ) . After selection , clonal cell lines were established and screened by PCR for biallelic gene disruption ( Figure 2—figure supplement 1A ) . Two independent cell lines with homozygous TERC KO were confirmed to lack telomerase catalytic activity in cell extract , as assayed by the PCR-based telomeric repeat amplification protocol performed with radiolabeled dGTP ( hotTRAP ) ( Figure 2B and Figure 2—figure supplement 1B , left ) . Initially , these two hTR KO clonal cell lines showed no growth defect . 10 . 7554/eLife . 18221 . 005Figure 2 . An hTRmin RNP can functionally substitute for hTR . ( A ) Schematic of Cas9-mediated disruption of the hTR locus with a PURO selection cassette . ( B ) Northern blot and hotTRAP assays of hTR KO#1 and KO#2 clonal HCT116 cell lines . An internal control ( IC ) to normalize PCR amplification was always included in hotTRAP assays . ( C ) AAVS1 donor constructs used for transgene rescue of hTR KO HCT116 cells . RNA expression used the U3 promoter and terminated within 500 bp of transplanted genomic region from immediately downstream of endogenous hTR ( g500 ) . mRNA expression used the CAGGS promoter terminated with a polyadenylation element ( pA ) . ( D ) Immunoblot , northern blot , and hotTRAP characterization of HCT116 cell lines 51 days after hTR KO targeting . ( E ) Brightfield microscopy images of HCT116 cell lines during the die-off interval of telomerase-negative cell cultures . Yellow arrowheads indicate membrane blebbing . ( F ) Chart of survival fate of HCT116 hTR KO cell lines with the indicated transgene at AAVS1 . DOI: http://dx . doi . org/10 . 7554/eLife . 18221 . 00510 . 7554/eLife . 18221 . 006Figure 2—figure supplement 1 . Generation of cell lines expressing hTRmin telomerase . ( A ) Schematic and assay results for PCR detection of hTR locus KO . ( B ) Telomerase activity comparison by hotTRAP for the hTR KO cell lines with TERT OE alone or in combination with hTRmin or LhTRmin . Input total protein is whole cell lysate . Assays were from polyclonal populations following selection for transgene integration and extended post-targeting culture ( 114 days post-targeting for hTR KO ) . ( C ) QTRAP comparison of telomerase activity levels in the hTR KO cell lines rescued by AAVS1 hTRmin TERT OE or LhTRmin TERT OE . Values are set relative to parental HCT116 ( n = 3 ) . ( D ) Immunoblot , northern blot , and hotTRAP characterization of HCT116 hTR KO cells with or without TERT OE from an AAVS1 transgene and with lentiviral hTR , LhTRmin , or GFP . ( E ) Time course of TRF in the HCT116 hTR KO with TERT OE and lentiviral LhTRmin or negative control GFP . Days post-targeting is relative to hTR KO; 65 days post-targeting was the last time point of collection prior to die-off of the telomerase-negative cells . DOI: http://dx . doi . org/10 . 7554/eLife . 18221 . 006 In each hTR KO cell line , we targeted transgene integration at AAVS1 with expression cassette ( s ) encoding GFP , N-terminally F-tagged TERT , hTR , LhTRmin , hTRmin , or LhTRmin or hTRmin coexpressed with TERT ( Figure 2C ) . RNA expression was directed by the U3 snoRNA promoter , and protein expression was driven by the strong and constitutive CAGGS promoter . Polyclonal populations of transgene-containing cells were selected and immediately assayed for telomerase subunit expression and telomerase activity ( Figure 2D ) . TERT overexpression ( OE ) was detected by immunoblot , and transgene-encoded RNA expression was detected by Northern blot . Telomerase catalytic activity was assayed by hotTRAP . Transgene expression of wild-type hTR restored telomerase catalytic activity to the hTR KO cells , rescuing endogenous TERC locus disruption ( Figure 2D , lane 3 ) . Transgene expression of GFP , LhTRmin , hTRmin , or TERT alone did not ( Figure 2D , lanes 2 and 4–6 ) . However , LhTRmin or hTRmin with TERT OE did generate active telomerase ( Figure 2D , lanes 7–8 ) . TERT co-expression did not affect hTRmin biological accumulation ( Figure 2D , compare lanes 4–5 to lanes 7–8 ) . Parallel results were confirmed using the independent hTR KO clonal cell line ( Figure 2—figure supplement 1B , C ) . In addition , we tested telomerase activity rescue of hTR KO cells or hTR KO cells with AAVS1 TERT OE by lentiviral introduction of LhTRmin . Only with TERT OE did lentiviral expression of LhTRmin produce active telomerase , whereas lentiviral expression of hTR rescued hTR KO without TERT OE ( Figure 2—figure supplement 1D , E ) . Telomerase activity levels in the hTRmin telomerase cell lines were within an order of magnitude of the telomerase level in parental HCT116 cells ( Figure 2—figure supplement 1B , C ) . From these experiments we conclude that hTR assembly as H/ACA RNP strongly stimulates hTR interaction with TERT when both subunits are at very low expression levels , but this role of H/ACA RNP assembly can be bypassed by increasing the cellular availability of TERT . The hTR KO cell lines lacking active telomerase ultimately entered an interval of pervasive cell death with dramatic membrane blebbing ( Figure 2E , yellow arrowheads ) at ~70 days post-targeting , corresponding to ~70 population doublings . In stark contrast , all of the telomerase-positive polyclonal cell cultures and clonal cell lines proliferated over many months of continuous passage with normal morphology and doubling time ( Figure 2F ) . Therefore , the catalytically active hTRmin telomerase RNPs conferred indefinite cellular proliferative capacity . We next tested the hTR KO cell lines with AAVS1 transgenes for rescue of telomere shortening . As expected , hTR KO cells re-expressing hTR rapidly gained telomere length , whereas hTR KO cells expressing the negative control GFP did not ( Figure 3A , lanes 1–3 and Figure 3—figure supplement 1A ) . All cell cultures lacking active telomerase had short telomeres that continued to shorten until eventually all cells in the culture died ( Figure 3A , B ) . Telomere length was heterogeneous in the polyclonal population of hTR KO cells rescued by LhTRmin with TERT OE ( Figure 3A , lane 7 and Figure 3—figure supplement 1B ) . We generated clonal cell lines from the hTR KO cell lines complemented by LhTRmin or hTRmin with TERT OE and characterized their maintenance of telomere length . These clonal cell lines had distinct but stable telomere lengths ( Figure 3C , D and Figure 3—figure supplement 1C ) . 10 . 7554/eLife . 18221 . 007Figure 3 . Telomerase with hTRmin supports stable telomere length maintenance . ( A ) Southern blot detection of telomere restriction fragment lengths ( TRF ) for HCT116 cell lines after release from selection . The telomerase-negative cell line TRFs were analyzed before cultures commenced cell death . ( B ) Time course of TRF shortening in the telomerase-negative HCT116 cell lines . The 65 days post-targeting time point was the final cell collection before culture death . All lanes are from the same gel . Red dashed lines separate different genotypes . ( C ) TRF analysis for multiple clonal cell lines expressing LhTRmin with TERT OE or hTRmin with TERT OE , all cultured in parallel and assayed at the same time point 111 days post-targeting . ( D ) Time course of TRF in the clonal cell lines expressing LhTRmin with TERT OE or hTRmin with TERT OE . Days post-targeting refers to the hTR KO . ( E ) Immunoblot and Northern blot analysis of telomerase subunit expression levels across the clonal cell lines expressing LhTRmin with TERT OE or hTRmin with TERT OE , performed using cells at 111 days post-targeting . ( F ) Comparison of telomerase activity measured by QTRAP ( purple bars ) with telomerase subunit expression levels quantified from blots in ( E ) . QTRAP ( averaged , n = 5 ) and RNA signals were normalized to parental HCT116 cell line activity and endogenous hTR . TERT OE signals were set relative to LhTRmin TERT OE clone #1 . DOI: http://dx . doi . org/10 . 7554/eLife . 18221 . 00710 . 7554/eLife . 18221 . 008Figure 3—figure supplement 1 . Characterization of telomere length maintenance and telomerase activity levels in hTRmin telomerase cell lines . ( A ) Northern blot and TRF timecourse of HCT116 hTR KO#2 cell line rescue by transgene hTR . ( B ) Time course of HCT116 TRF after introduction of a transgene for TERT OE alone or with LhTRmin or hTRmin; cell cultures were polyclonal after transgene introduction . Cells with TERT OE alone died shortly after 65 days post-targeting , counted from the original hTR KO . ( C ) Time course of TRF in clonal cell lines of HCT116 hTR KO with AAVS1 TERT OE and hTRmin . Days post-targeting refers to the original hTR KO . ( D ) QTRAP data from Figure 3F with error bars from technical replicates ( n = 5 ) . ( E ) Telomerase activity measured by hotTRAP for the HCT116 clonal cell lines with TERT OE and LhTRmin or hTRmin . In each panel , all lanes are from the same gel . DOI: http://dx . doi . org/10 . 7554/eLife . 18221 . 008 To investigate the clonal cell line variance in telomere length at length homeostasis , we profiled levels of telomerase subunits and activity across the clonal cell lines . All of the LhTRmin cell lines had higher steady-state RNA accumulation than the hTRmin lines ( Figure 3E ) . LhTRmin levels varied more than hTRmin , and TERT levels varied more in LhTRmin lines than hTRmin lines ( Figure 3E and Figure 3F , symbols ) . Telomerase activity measured by fluorescence quantification of telomeric repeat amplification ( QTRAP ) ( Figure 3F , bars and Figure 3—figure supplement 1D for replicates ) or hotTRAP ( Figure 3—figure supplement 1E ) was generally greatest in cell lines with high telomerase subunit expression ( Figure 3F , compare symbols to bars ) . Higher telomerase activity in cell extract correlated generally but not absolutely to longer telomere length at homeostasis ( compare Figure 3F , bars to Figure 3C ) . We suggest that differences in telomerase subunit expression levels could in part reflect epigenetic differences introduced upon transgene integration , which was an independent event in each clonal cell line . In addition , differences in subunit expression levels could result from the stochastic fluctuation established to occur in many model systems , including HeLa cells ( Bryan et al . , 1998 ) . We confirmed that the hTRmin telomerase RNP assembled in the stable cell lines did not have the endogenous hTR RNP interaction with TCAB1 , tested by immunopurification of TCAB1 from cell lysate and subsequent telomerase activity assay ( Figure 4—figure supplement 1A ) . However , because telomere-associated telomerase would be a small fraction of the total telomerase RNP pool , we sought another approach to demonstrate that hTRmin telomerase maintained telomeres without a requirement for TCAB1-mediated recruitment to Cajal bodies . To this end , we disrupted the genes encoding TCAB1 and Coilin in HCT116 cells expressing hTRmin telomerase . We programmed Cas9 for cleavage of the endogenous TCAB1 or Coilin locus in the presence of donor plasmid that would integrate a hygromycin resistance cassette ( Figure 4A ) . Targeting and selection were highly efficient , resulting in polyclonal populations of LhTRmin TERT OE cells and hTRmin TERT OE cells with little TCAB1 or Coilin , as detected by immunoblot ( Figure 4B ) or immunofluorescence ( Figure 4C ) . TCAB1 KO cells retained a normal level of Coilin , and Coilin KO cells retained a normal level of TCAB1 ( Figure 4B ) . Cells lacking TCAB1 retained Cajal bodies ( Figure 4C ) , in agreement with some previous findings ( Venteicher et al . , 2009; Zhong et al . , 2011 ) but contrary to others ( Mahmoudi et al . , 2010; Wang et al . , 2016 ) . Also , nuclear foci of SMN remained detectable in the TCAB1 KO cells but not Coilin KO cells ( Figure 4—figure supplement 1B ) . We conclude that in HCT116 cells , TCAB1 KO did not disrupt Cajal bodies , and Coilin KO did not induce TCAB1 degradation . 10 . 7554/eLife . 18221 . 009Figure 4 . TCAB1 and Cajal bodies are not required for telomere maintenance by hTRmin telomerase . ( A ) Schematic of Cas9-mediated disruption of TCAB1 and Coilin ( COIL ) loci with a HYGRO selection cassette . ( B ) Immunoblot and QTRAP analysis of the polyclonal populations of LhTRmin or hTRmin TERT OE cells selected for disruption of TCAB1 or COIL loci . QTRAP values were normalized to the cell line before TCAB1 or COIL disruption ( n = 3 ) . ( C ) Immunofluorescence localization of TCAB1 and Coilin in HCT116 cells . ( D ) Immunoblot analysis for TCAB1 and Coilin in clonal KO cell lines with LhTRmin + TERT OE or hTRmin + TERT OE . ( E ) QTRAP assay of the clonal cell lines in ( D ) . QTRAP values were normalized to the cell line before TCAB1 or COIL disruption ( n = 3 ) . ( F ) Stable TRF lengths in clonal cell lines lacking TCAB1 or Coilin . Days post-targeting refers to the TCAB1 or COIL KO . DOI: http://dx . doi . org/10 . 7554/eLife . 18221 . 00910 . 7554/eLife . 18221 . 010Figure 4—figure supplement 1 . Characterization of HCT116 hTRmin with TERT OE cell lines with TCAB1 KO or Coilin KO . ( A ) Immunopurification from HCT116 hTR KO cell lines expressing LhTRmin with TERT OE or hTRmin with TERT OE . Protein A/G beads were coupled with anti-TCAB1 antibody and used to immunopurify TCAB1 complexes . Rabbit IgG and anti-FLAG antibody were used as negative and positive controls , respectively . Immunopurified samples were assayed by hotTRAP . ( B ) Immunofluorescence localization of SMN in HCT116 TCAB1 KO and COIL KO cell lines . ( C , D ) Immunoblot and PCR genotyping of HCT116 TCAB1 KO and COIL KO cell lines . The immunoblots are also shown in main figures . The TCAB1 KO clonal cell line #2 in the LhTRmin TERT OE#3 background had one allele with target-site mutagenesis that produced an early translation stop codon . ( E ) Extended culture TRF analysis of HCT116 TCAB1 KO and COIL KO clonal cell lines from the LTRmin and hTRmin telomerase backgrounds . DOI: http://dx . doi . org/10 . 7554/eLife . 18221 . 010 We isolated clonal cell lines from the polyclonal KO cell populations and validated homozygous TCAB1 or Coilin KO by genomic locus PCR and protein immunoblots ( Figure 4D and Figure 4—figure supplement 1C , D ) . These clonal cell lines retained telomerase catalytic activity in cell extract ( Figure 4E ) and stably maintained telomeres ( Figure 4F and Figure 4—figure supplement 1E ) . Some heterogeneity was evident comparing the telomere lengths maintained in different clonal cell lines , which could result from stochastic variation ( Bryan et al . , 1998 ) . Importantly , neither TCAB1 KO nor Coilin KO affected cell viability , morphology , or proliferation in a readily detectable manner . These findings support the conclusion that an RNP of minimized hTRmin and TERT functions at telomeres without dependence on H/ACA RNP biogenesis or localization pathways . In parallel , we generated TCAB1 , Coilin , and TERT KO HCT116 cells with endogenous hTR expression using Cas9 for TCAB1 and Coilin or a zinc finger nuclease developed previously for TERT ( Sexton et al . , 2014 ) . Clonal cell lines with homozygous KO were identified using genomic locus PCR ( Figure 4—figure supplement 1C , D and Figure 5—figure supplement 1A ) and validated for loss of TCAB1 or Coilin but not hTR ( Figure 5A , B ) . TERT KO was validated by loss of telomerase activity from cell extract ( Figure 5C ) . As expected , TCAB1 KO cells retained Coilin , and Coilin KO cells retained TCAB1 ( Figure 5A ) . We also confirmed that immunopurification of TCAB1 from Coilin KO cell extract , but not from TCAB1 KO cell extract , enriched active telomerase ( Figure 5—figure supplement 1B ) . The loss of TCAB1 or Coilin did not alter telomerase activity in cell extract by more than a few-fold extent that could be within the range of clonal variation ( Figure 5D and Figure 5—figure supplement 1C ) . HCT116 cells with TCAB1 KO or Coilin KO showed no change in growth rate , cell viability , or morphology over more than half a year of continuous culture . In contrast , the TERT KO cells underwent cell death at ~70 days after targeting , paralleling the fate of the hTR KO HCT116 cells . 10 . 7554/eLife . 18221 . 011Figure 5 . TCAB1 and Cajal bodies are not essential for telomere maintenance by endogenous telomerase . ( A ) Immunoblot analysis of HCT116 TCAB1 KO and COIL KO clonal cell lines . ( B ) Northern blot for hTR in HCT116 TCAB1 KO and COIL KO clonal cell lines . ( C ) Lack of hotTRAP telomerase activity detection in the HCT116 TERT KO clonal cell lines . ( D ) QTRAP analysis of telomerase activity in the HCT116 TCAB1 KO and COIL KO clonal cell lines . Values were normalized to the parental HCT116 cell line ( n = 3 ) . ( E ) Time course of TRF in the HCT116 TERT KO clonal cell lines . ( F , G ) Time course of TRF in the HCT116 TCAB1 and COIL KO clonal cell lines . ( H ) Immunoblot analysis of hESC TCAB1 KO and COIL KO clonal cell lines . Wild-type refers to an hESC clonal cell line subjected to Cas9 electroporation but retaining a wild-type genotype . ( I ) QTRAP analysis of telomerase activity in the hESC TCAB1 KO and COIL KO clonal cell lines . Values were normalized to the parental hESC line ( n = 3 ) . ( J , K ) Time course of TRF in hESC TCAB1 KO and COIL KO clonal cell lines . Note that the long telomeres in mouse cells from the hESC feeder layer contribute some blot signal ( indicated MEFs ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18221 . 01110 . 7554/eLife . 18221 . 012Figure 5—figure supplement 1 . Characterization of HCT116 TERT KO , TCAB1 KO , and COIL KO cell lines with endogenous hTR . ( A ) PCR genotyping of HCT116 TERT KO clonal cell lines . ( B ) Immunopurification of TCAB1 complexes from parental , TCAB1 KO , and COIL KO HCT116 cell lines followed by hotTRAP . ( C ) HotTRAP of HCT116 TCAB1 KO and COIL KO cell lines . DOI: http://dx . doi . org/10 . 7554/eLife . 18221 . 01210 . 7554/eLife . 18221 . 013Figure 5—figure supplement 2 . Characterization of hESC TCAB1 KO and COIL KO lines with endogenous hTR . ( A ) PCR genotyping of hESC TCAB1 KO and COIL KO clonal cell lines . Wild-type refers to an hESC clonal cell line subjected to Cas9 electroporation but retaining a wild-type genotype . ( B ) HotTRAP of the TCAB1 KO hESC lines . ( C ) Immunofluorescence localization of Coilin and TCAB1 in hESC lines . DOI: http://dx . doi . org/10 . 7554/eLife . 18221 . 01310 . 7554/eLife . 18221 . 014Figure 5—figure supplement 3 . Additional controls for TCAB1 loss-of-function in TCAB1 KO cell lines . ( A ) Immunoblot and QTRAP analysis of HCT116 parental and TCAB1 KO cell lines expressing F-tagged TCAB1 or GFP from transgenes at AAVS1 . QTRAP values were normalized to the GFP cell line signal ( n = 3 ) . Days post-targeting refers to transgene integration . ( B ) Time course of TRF in cell lines from ( A ) . ( C ) Immunoblot and QTRAP analysis of hESC parental and TCAB1 KO cell lines expressing F-tagged TCAB1 transgene at AAVS1 . QTRAP values were normalized to the unrescued TCAB1 KO cell line signal ( n = 3 ) . Days post-targeting refers to transgene integration . ( D ) Time course of TRF in cell lines from ( C ) . ( E ) TCAB1 double-exon-disruption schematic with TCAB1 exon 2 and exon 8 KO genotyping by PCR . TCAB1 exon 8 targeting was performed in the HCT116 TCAB1 KO#2 background , which has a hygromycin resistance cassette inserted in TCAB1 exon 2 . ( F ) TRF of HCT116 parental , TCAB1 KO#2 background , and exon 8 disrupted TCAB1 KO#2 ( double-exon-disruption ) cell lines . DNA was purified from double-exon-disruption cells at 43 days post-targeting the second KO . ( G ) Immunofluorescence staining for Coilin ( green ) and TCAB1 ( red ) in double-exon-disruption cells . DOI: http://dx . doi . org/10 . 7554/eLife . 18221 . 014 As expected the TERT KO cells experienced rapid , progressive telomere attrition ( Figure 5E ) . Surprisingly , telomeres shortened more gradually in the TCAB1 KO clonal cell lines , followed by stable telomere length homeostasis ( Figure 5F ) . The starting point of telomere length in TCAB1 KO clonal cell lines mirrored the amount of telomerase activity in cell extract ( compare Figure 5D and Figure 5F , left panel ) , but after more than half a year of continuous growth , telomere lengths in all of the TCAB1 KO clonal cell lines stabilized at a few kbp shorter than telomeres in the parental cell line ( Figure 5F , right panel ) . In comparison , telomeres in Coilin KO clonal cell lines cultured in parallel remained near the length of telomeres in the parental HCT116 cells ( Figure 5G ) . We conclude that in HCT116 cells , neither TCAB1 nor Coilin is required for a stable telomere length homeostasis . Stable telomere length maintenance in the TCAB1 KO cells with endogenous telomerase was unexpected . To determine whether this finding is general , we investigated the consequence of TCAB1 or Coilin KO in the human embryonic stem cell ( hESC ) line WIBR#3 ( Chiba and Hockemeyer , 2015 ) . Clonal hESC lines with TCAB1 or Coilin KO were generated using the same approach that was successful in HCT116 cells ( Figure 5—figure supplement 2A ) . As observed for HCT116 cells , hESC lines lacking TCAB1 or Coilin were viable with no evident change in cell morphology or proliferation . TCAB1 accumulated in Coilin KO cells , and Coilin accumulated in TCAB1 KO cells ( Figure 5H ) . Neither TCAB1 KO nor Coilin KO affected telomerase activity assayed in cell extract beyond the range of clonal variation ( Figure 5I and Figure 5—figure supplement 2B ) . Coilin remained localized to Cajal bodies in TCAB1 KO cells ( Figure 5—figure supplement 2C ) . Over many months of continuous culture , hESCs lacking TCAB1 experienced very gradual telomere shortening followed by telomere length maintenance ( Figure 5J ) . The rate of telomere shortening was slow compared to telomere shortening in TERT KO hESC ( Sexton et al . , 2014 ) . The hESCs lacking Coilin retained telomere lengths comparable to the parental hESC culture ( Figure 5K ) . Clonal hESC lines that had undergone targeting but retained the wild-type genotype also did not demonstrate telomere shortening ( Figure 5J , lanes labeled wild-type ) . To confirm that telomere shortening in TCAB1 KO cells was directly linked to the loss of TCAB1 , we introduced a TCAB1 transgene to TCAB1 KO cells by integration at AAVS1 . TCAB1 KO HCT116 cells ectopically expressing F-tagged TCAB1 regained telomere length ( Figure 5—figure supplement 3A , B ) . Likewise , TCAB1 KO hESCs ectopically expressing F-tagged TCAB1 regained telomere length ( Figure 5—figure supplement 3C , D ) . As a control for complete TCAB1 loss-of-function , we targeted the TCAB1 KO cells above , which have an exon 2 disruption shortly after the translation start codon ( Figure 4A ) , for successful homozygous disruption of downstream exon 8 ( Figure 5—figure supplement 3E ) . Cells with homozygous disruptions of exon 2 and exon 8 showed no proliferation defect , no change in telomere length from the exon 2 KO , and no loss of Cajal bodies ( Figure 5—figure supplement 3F , G ) . Overall , based on the assays described above , we conclude that TCAB1 and Cajal bodies are not essential for hTR telomerase to maintain telomeres at a stable length homeostasis . In human somatic cells that remain telomerase-positive , stimulation to proliferate can be accompanied by dramatic telomerase activation and telomere length gain ( Weng et al . , 1997 ) . This rapid increase in telomere length could depend on telomerase assistance by TCAB1 and/or Cajal bodies in a manner not required for maintaining length homeostasis . We therefore tested TCAB1 and Coilin KO cells for their ability to support rapid telomere elongation upon an increase in telomerase expression , achieved by integrating transgenes for hTR and TERT overexpression at AAVS1 . The parental , TCAB1 KO , and Coilin KO HCT116 cells with integrated hTR and TERT transgenes acquired ~2-fold elevated hTR and ~5-fold elevated telomerase catalytic activity ( Figure 6A ) . These High-Telomerase ( HiT ) polyclonal cell cultures had increased telomere length by the first time point after selection ( Figure 6B ) . Telomere elongation was rapid in TCAB1 KO HiT cell cultures , reaching the limit of length discrimination almost immediately . Rapid telomere elongation in TCAB1 KO cells is consistent with telomere elongation in wild-type HeLa cells by overexpression of CAB-box-mutant hTR ( Fu and Collins , 2007; Cristofari et al . , 2007 ) . In comparison , Coilin KO HiT cells had less telomere elongation ( Figure 6B ) . These results were replicated in an independently performed HiT conversion of the same cell lines ( not shown ) . Telomeres in both TCAB1 KO and Coilin KO cells elongated upon expression of a truncated POT1 compromised for binding to single-stranded telomeric-repeat DNA ( Loayza and De Lange , 2003 ) , expressed by transgene integration at AAVS1 ( data not shown ) . No obvious hTR foci were detected in HCT116 HiT cells lacking TCAB1 or Coilin ( Figure 6C ) , indicating that the hTR foci detected in HeLa cells lacking Coilin ( Chen et al . , 2015 ) are challenging to detect . We conclude that if telomerase is abundant , TCAB1 is not required to support telomerase-mediated elongation of even relatively long telomeres . Integrating all of the findings of this study , we conclude that human telomerase H/ACA RNP assembly is essential not only for the biological stability of hTR but also for active RNP biogenesis at scarce subunit expression levels . In addition , H/ACA RNP assembly gives endogenous-level hESC telomerase the ability to maintain long telomeres . Remarkably , all of these H/ACA RNP assembly roles can be bypassed using hTRmin ( Figure 7 ) . This plasticity of telomerase RNP biogenesis and holoenzyme composition informs mechanisms of telomerase diversification across eukaryotes . 10 . 7554/eLife . 18221 . 015Figure 6 . Cajal bodies promote telomere elongation upon increased telomerase expression level . ( A ) Immunoblot , northern blot , and QTRAP characterization of TCAB1 KO and COIL KO HCT116 cells overexpressing hTR and TERT at AAVS1 ( HiT ) . QTRAP values were normalized to parental HCT116 ( n = 3 ) . ( B ) Time course of TRF in the HiT HCT116 TCAB1 KO and COIL KO cell cultures polyclonal following HiT transgene introduction . Days post-targeting refers to HiT transgene introduction . The lanes labeled 'con' are the indicated cell line without HiT transgene introduction . ( C ) FISH for hTR localization in HiT HCT116 cell lines . DOI: http://dx . doi . org/10 . 7554/eLife . 18221 . 015 Telomerase holoenzymes from different eukaryotes share the cellular requisites of stable RNP biogenesis and active RNP recruitment to telomeres in S-phase , but the mechanisms that underlie these requirements for telomerase function vary greatly ( Egan and Collins , 2012a ) . We set out to uncover the rationale for a vertebrate telomerase evolutionary gain of H/ACA RNP biogenesis . H/ACA RNP biogenesis confers hTR biological stability , but across eukaryotes , telomerase RNA stability can be conferred by diverse other RNP assembly pathways . To bypass the essential role of H/ACA proteins in hTR cellular stability , we designed hTRmin RNAs containing the activity-essential hTR motifs and a 3’ triplex structure but not an H/ACA motif . Although hTRmin did not bind H/ACA proteins or TCAB1 , it did assemble catalytically active telomerase . The opposite TER redesign strategy was used to test the significance of the relative positioning of a large number of holoenzyme protein binding motifs in S . cerevisiae TLC1 ( Zappulla et al . , 2005 ) . S . cerevisiae Mini-T retains all of the holoenzyme protein interactions but condenses the length of spacing between them ( Zappulla et al . , 2005 ) . Initially we expected the biological function of hTRmin-TERT RNPs to require tethering to Cajal bodies or telomeres . Instead , hTRmin-TERT RNPs supported stable telomere maintenance without any supplementary trafficking instructions ( Figure 7 , left ) . Therefore , most telomerase holoenzyme proteins may serve indirect roles in telomerase function that are readily adapted to evolutionary divergence of nuclear architecture and its cell cycle regulation . Other than hTR biological stability , the contributions of H/ACA RNP assembly , TCAB1 , and Cajal bodies to endogenous human telomerase function could all be accounted for by changes in subunit and RNP distribution within the nucleus ( Figure 7 , right ) . 10 . 7554/eLife . 18221 . 016Figure 7 . Multiple traffic pathways for human telomerase biogenesis and action at telomeres . At left , hTRmin telomerase assembles and acts at telomeres without supplemental trafficking instructions . At right , endogenous hTR and TERT are trafficked for their assembly and for telomerase action at telomeres . DOI: http://dx . doi . org/10 . 7554/eLife . 18221 . 016 Studies here point to stabilization and efficient interaction of low-abundance telomerase subunits as a main rationale for the vertebrate telomerase H/ACA RNP biogenesis pathway . Cellular assembly of minimized hTRmin into catalytically active telomerase RNP required a higher than endogenous expression level of TERT . One simple model for this influence would be that H/ACA RNP biogenesis directs trafficking of assembly-competent hTR to meet assembly-competent TERT . Based on the cell cycle regulation of human telomerase subunit synthesis and subunit exchange in assembled RNPs , we have suggested that active RNP assembly may occur with only newly synthesized TERT and/or hTR subunits prior to their steady-state distribution ( Vogan and Collins , 2015 ) . Newly synthesized subunits could meet in Cajal bodies , as pre-mature hTR transits to acquire 5' cap trimethylation , or in nucleoli . Additional non-exclusive possibilities for cellular role ( s ) of hTR H/ACA RNP assembly in active RNP biogenesis include hTR folding , which would parallel the precedent from ciliate telomerase biogenesis ( Stone et al . , 2007 ) , and/or exclusion of non-productive hTR-protein interactions . In both HCT116 cells and hESCs , TCAB1 KO led to telomere shortening followed by telomere length homeostasis . As one model to account for these results , TCAB1 could influence the subnuclear distribution of active RNP between nucleoli and nucleoplasm ( Figure 7 , right ) . This change in distribution would have different influence on the amount of new telomeric repeat synthesis depending on telomerase expression level and other variables across human cell types . Of note , we have not ruled out an influence of TCAB1 related to its proposed function in DNA damage repair ( Henriksson and Farnebo , 2015 ) . However , in studies here it is striking that TCAB1 KO affected telomere length only in cells with hTR telomerase , not in cells with hTRmin telomerase . Therefore , TCAB1 functions directly related to active telomerase assembly and trafficking are more logical hypotheses . Consistent with a contemporary study of Coilin KO in HeLa cells ( Chen et al . , 2015 ) , we found that Coilin KO in HCT116 cells or in hESCs was permissive for stable telomere length maintenance . Coilin KO clonal cell lines generated in this work had telomere lengths similar to the parental cell lines . Because Coilin KO cells with endogenous telomerase levels have apparently unperturbed telomere length homeostasis , we suggest that they will be useful for visually tracking telomerase interactions with telomeres without the complication of Cajal body clustering of active with inactive hTR . Curiously , Coilin KO HCT116 cells had dampened telomere elongation upon an increase in telomerase expression . This could reflect less synergistic telomerase RNP loading on telomeres . Alternately , Cajal bodies could coordinate telomerase with other factors important for telomere synthesis , or they could safeguard telomere integrity . Our results support a paradigm of nuclear bodies as zones that draw factors together to fine-tune the likelihood of their physical association or functional cooperation rather than their interaction or reaction specificity . Previous studies did not anticipate the genetic consequences of TCAB1 and Coilin KOs . For example , opposite the impact of TCAB1 and Coilin KOs on telomere lengths described above , previous studies found that endogenous hTR colocalization with telomeres was reduced more by Coilin depletion than TCAB1 depletion and overexpressed hTR colocalization with telomeres was reduced more by TCAB1 depletion than by Coilin depletion ( Zhong et al . , 2012; Stern et al . , 2012 ) . Also , TCAB1 mutations result in severe DC ( Zhong et al . , 2011; Armanios and Blackburn , 2012 ) . We suggest that TCAB1 mutations reduce telomere length in early human development and in somatic cells with long telomeres . Overall , insights from this work inform strategies of therapy for human disease . HCT116 , 293T , VA-13 , and U2OS cells were cultured in DMEM with GlutaMAX ( Thermo , Waltham , MA ) supplemented with 10% FBS and 100 µg/ml Primocin ( Invivogen , San Diego , CA ) . HCT116 cells were obtained from the Molecular and Cell Biology Department tissue culture facility ( UC Berkeley ) . The 293T , VA-13 , and U2OS cell lines are long-term Collins lab stocks . The hESC line WIBR#3 ( NIH stem cell registry #0079 , originating at the Whitehead Institute for Biomedical Research ) was maintained on inactivated mouse embryonic fibroblasts in DMEM/F12 under conditions previously described ( Chiba and Hockemeyer , 2015 ) . None of the cell lines have been authenticated recently at the genome sequence level , but all have the expected cell morphology and doubling time . HCT116 and hESC cell lines were tested for mycoplasma . Transient transfection , was performed using calcium phosphate or Lipofectamine 2000 ( Thermo ) . RNA was purified using TRIzol ( Thermo ) and resuspended in distilled water . Concentration was determined by absorbance at 260 nm . Formamide loading buffer was added to RNA samples , followed by heat denaturation at 95°C for 1 min and ice for 5 min before loading on a denaturing gel ( 5% of 19:1 acrylamide:bis-acrylamide , 0 . 6X TBE , 32% formamide , and 5 . 6 M urea ) . Following electrophoresis , RNA was transferred to nylon membrane by electroblotting . RNA was then UV crosslinked and the membrane was blocked in Church's buffer ( 1% BSA , 1 mM EDTA , 0 . 5 M NaP04 , 7% SDS ) with 15% formamide at 50°C . 32P-labeled , 2'O-methyl RNA probe complementary to the hTR template region was then added ( Fu and Collins , 2003 ) . In some experiments , 32P-labeled DNA oligonucleotides complementary to regions in the hTR pseudoknot ( 5'-TAGAATGAACGGTGGAAGGCGGCAGGCCGAGGCT-3' ) and hTR CR4/5 ( 5'-TCGCGGTGGCAGTGGGTGCCTCCGGAGAAGCCC-3' ) were also added to enhance signal detection . Membranes were hybridized overnight at 50°C , followed by extensive washing with 1X SSC + 0 . 1% SDS at 50°C . Blots were then exposed on phosphorimager screens and subsequently scanned on a Typhoon ( GE Healthcare , Chicago , IL ) . Scans were processed using ImageJ software ( NIH ) . The Northern blot loading control ( LC ) is a non-specifically detected cellular RNA . Protein was resolved in 10% Bis-Tris SDS-PAGE gels in MOPS buffer ( 250 mM MOPS pH 7 . 0 , 250 mM Tris , 5 mM EDTA , 0 . 5% SDS ) before being transferred to nitrocellulose membrane . The membranes were then blocked in immunoblot buffer consisting of 4% nonfat milk ( Carnation ) in TBS buffer ( 150 mM NaCl , 50 mM Tris pH 7 . 5 ) . Membranes were then incubated with primary antibodies diluted in immunoblot buffer at 4°C overnight . After primary antibody incubation , membranes were washed extensively in TBS , followed by secondary antibody incubation in immunoblot buffer for 1 hr at room temperature . After washing in TBS , blots were scanned using a LI-COR Odyssey imager . Immunoblot primary antibodies included mouse anti-tubulin ( 1:500 , DM1A , Calbiochem , Billerica , MA ) , mouse anti-FLAG ( 1:4 , 000 , F1804 , Sigma ) , rabbit anti-TCAB1 ( 1:2000 , NB100-68252 , Novus Biologicals , Littleton , CO ) , mouse anti-Coilin ( 1:250 , IH10 , Abcam , Cambridge , United Kingdom ) , and mouse anti-TERT ( 1:2000 ( Wu et al . , 2015 ) ) . Secondary antibodies included goat anti-mouse Alexa Fluor 680 ( 1:2000 , Life Technologies , Carlsbad , CA ) and goat anti-rabbit IR Dye 800 ( 1:10 , 000 , Rockland Immunochemicals , Pottstown , PA ) . HLB150 buffer ( 20 mM HEPES pH 8 . 0 , 2 mM MgCl2 , 0 . 2 mM EGTA , 10% glycerol , 1 mM DTT , 0 . 1% Igepal , Sigma protease inhibitor cocktail , 150 mM NaCl ) was used throughout . Samples were normalized to 2 mg/ml and 200 µl was incubated at 4°C for 1–2 hr with 4 µl magnetic anti-FLAG M2 beads ( Sigma ) or 4 µl magnetic Protein A/G beads ( BioTool , Houston , TX ) pre-bound with antibody against TCAB1 ( Novus Biologicals ) , antibody against FLAG ( Sigma ) , or rabbit IgG ( Sigma ) . After binding and washing , the beads were resuspended in 20 µl HLB150 buffer . For hotTRAP , the resuspended beads were diluted 1:10 in HLB150 , and 2 µl of the diluted sample was used per hotTRAP reaction . For Northern blot analysis , the washed beads were processed with TRIzol ( Thermo ) . Live cell imaging was performed on a Zeiss Axio Observer A1 with a 40X phase contrast objective . For RNA and protein localization , cells were fixed in 4% paraformaldehyde for 10 min at room temperature . Cells were then washed with PBS twice , before additional fixation and permeabilization with 100% methanol for 10 min at room temperature . For RNA FISH , the methanol was aspirated and RNA probes in RNA FISH buffer were directly added to cells . Detection of hTR and hTRmin used Cy3-labeled RNA probes complementary to the template , pseudoknot , and CR4/5 spanning hTR nt 36–70 , 129–162 , and 249–281 , respectively . Ten ng of each probe was diluted in RNA FISH buffer consisting of 2X SSC , 10% formamide , and 10% dextran sulfate . Probes were incubated with samples overnight at 37°C in a humidified chamber . Samples were then washed 6 times with 2X SSC before being mounted on coverslips using ProLong Gold with DAPI ( Thermo ) . For immunofluorescence , cells were washed three times in PBS following methanol incubation and then rehydrated and blocked in 4% BSA in PBS for 1 hr at room temperature or overnight at 4°C . Primary antibodies diluted in 4% BSA in PBS were then added and incubated with samples for 1 hr at room temperature . Primary antibodies used included rabbit anti-TCAB1 ( 1:300 , NB100-68252 , Novus Biologicals ) , mouse anti-Coilin ( 1:250 , IH10 , Abcam ) , and mouse anti-SMN ( 1:300 , sc-15320 , Santa Cruz Biotechnology , Santa Cruz , CA ) . After washing in PBS , samples were incubated with Alexa Fluor 488 or Alexa Fluor 568 ( Thermo ) secondary antibodies diluted in 4% BSA in PBS for 1 hr at room temperature . Samples were then washed in PBS and mounted on coverslips using ProLong Gold with DAPI . Images were acquired using a Zeiss LSM510 Meta confocal microscope with a ×100/1 . 49 Apo objective , with 364- , 488- , and/or 543-nm laser excitation . Cell extract was prepared by hypotonic freeze thaw lysis and salt extraction . Primer extension and hotTRAP were performed as previously described ( Sexton et al . , 2014 ) . Unless otherwise noted , 200 ng total protein was used per hotTRAP or QTRAP reaction , quantified using the Bio-Rad protein assay . An internal control ( IC ) to normalize PCR amplification was always included in hotTRAP assays . QTRAP used the iTaq universal SYBR green Supermix ( Bio-Rad ) and a CFX96 Touch Real-Time PCR Detection System ( Bio-Rad , Hercules , CA ) as previously described ( Vogan and Collins , 2015 ) . Relative telomerase activity measured by QTRAP was calculated by delta Ct to a reference sample . All error bars shown are standard error of the mean and statistical significance was calculated using ANOVA with Tukey’s multiple comparison test in GraphPad Prism 6 . Single-guide RNA sequences for Cas9 targeting of protein coding genes were designed for selection cassette insertion downstream of the start codon . Optimal guide sequences were generated using the CRISPR Design tool ( http://crispr . mit . edu/ ) . Guide sequences were then inserted into the PX330 plasmid ( Cong et al . , 2013 ) . TERC guide sequence , followed by the PAM: 5’-TTCAGCGGGCGGAAAAGCCT CGG-3' . TCAB1 exon 2 guide sequence , followed by the PAM: 5'-TTTATTCATCGGGGAAGCGT GGG-3' . TCAB1 exon 8 guide sequence , followed by the PAM: 5’-TGAGAAGAAGCGGTTGCCAT CGG-3’ . COIL exon 1 guide sequence , followed by the PAM: 5'-AAGCCGTAGCCTAACCGTCT CGG-3' . Donor plasmid constructs were designed by flanking a puromycin resistance cassette or a hygromycin resistance cassette with sequence 500–600 bp upstream and downstream of the genomic DNA cut site . The TCAB1 targeting sites do not overlap with the upstream p53 gene . Zinc-finger nuclease ( ZFN ) mediated disruption of the TERT gene and transgene integration at AAVS1 have been previously described ( Sexton et al . , 2014 ) . For the disruption of TERT gene exon 1 , a donor plasmid carrying a hygromycin resistance cassette flanked by homology arms of approximately 500 bp upstream and downstream of the cut site was transfected with plasmids expressing the TERT-targeting ZFN . For AAVS1 transgene integration , donor plasmids carrying the transgene ( s ) with an upstream neomycin or puromycin resistance cassette , together flanked AAVS1 homology arms , were transfected with plasmids expressing the AAVS1-targeting ZFN . Genome engineering in HCT116 cells was performed by Lipofectamine 3000 transfection according to manufacturer guidelines ( Thermo ) using a 2:1 ratio of donor plasmid to nuclease plasmid . HCT116 cell lines were selected with 300 µg/ml hygromycin , 1 µg/ml puromycin , or 300 µg/ml G418 . hESC gene editing was performed by electroporation as previously described ( Sexton et al . , 2014; Chiba and Hockemeyer , 2015 ) . Lentivirus was produced in 293T cells by calcium phosphate transfection with the packaging plasmid , psPAX2 , the envelope plasmid , pMD2 . G , and with transgene constructs in the DUET011 backbone ( Zhou et al . , 2007 ) . Cell transfection media was replaced at 24 hr post-transfection and virus was harvested 48 hr post-transfection . Virus-containing media was applied to HCT116 cells in the presence of 5 µg/ml polybrene ( Sigma ) for 24 hr before a media change . At 48 hr post-infection , transduced cells were selected with 300 µg/ml hygromycin . Genomic DNA was prepared as described above for telomere restriction fragment analysis . Between 50–100 ng of genomic DNA was used as the template for PCR using the Q5 polymerase ( NEB , Ipswich , MA ) . For PCR confirmation of gene editing , PCR primers were designed to generate amplicon size differences between loci with or without a drug resistance cassette . Paired PCR primers complementary to genomic loci had one primer complementary to a region also in the donor plasmid and the other primer complementary to a region beyond the donor plasmid homology . A third primer against either the hygromycin or puromycin resistance cassette was included that would generate an amplicon size for cassette-containing alleles that was either smaller or larger than amplicons from alleles lacking the cassette . For TCAB1 exon 2 and Coilin exon 1 PCRs , a hygromycin cassette primer ( PGK hygro: 5'-AGGCTGATCAGCGGTTTAAACTTAGCCTCCCCTACCCGGTAGAATTC-3'; or hygro pA: 5’-CTAGTGGATCCGAGCTCGGTACCAGATGCGGTGGGCTCTATGGC-3’ ) was paired with the following locus-specific primers: Coil_FWD: 5'-TAGTGGATCCGAGCTCGGTACCACCACTGCTCCTGGCCTCTAGTTAC-3' , Coil_REV: 5'-AGGCTGATCAGCGGTTTAAACTTAAGAACTGAAGCCGAAGCGCTGG-3' , TCAB_FWD1: 5'-CTAGTGGATCCGAGCTCGGTACCAGGAAGGCTTTCCGTAATATCACACCCTAACG-3' , and TCAB_REV1: 5'-AGGCTGATCAGCGGTTTAAACTTACAGAAAGTTCTTGCTCCTCGATTCGAGGACTC-3' . An alternative set of TCAB1 locus primers was also used for additional validation of the lines: TCAB1_FWD2: 5’-CTAGTGGATCCGAGCTCGGTACCAGCGGTGCTAAGGAACACAGTGCTTTCAAAAG-3’ , and TCAB1_REV2: 5’-AGGCTGATCAGCGGTTTAAACTTAGGCATCCCTCTCCTAGAAAACTGG-3’ . For TCAB1 exon 8 PCR , a primer against the puromycin resistance cassette ( PGK_PURO: 5’-GGCGCACCGTGGGCTTGTACTCGGTCATGGTGGCGGGATGCAGGT CGAAAGGCCCG-3’ ) was combined with two loci primers ( TCAB1_ex8_FWD: 5’-CCAAGGCCAACCAGCTGGTCAAAGGACTGCTTC-3’ , and TCAB1_ex8_REV:5’-CTCAGCATCCTGGAGACAAGGAACAGGACCTGGAGT-3’ ) . For TERT locus PCR , another primer against the hygromycin cassette was used ( TERT_hygro: 5'-CTCACCGCGACGTCTGTCGAGAAG-3' ) with the following locus-specific primers: TERT_FWD 5'-CTAGTGGATCCGAGCTCGGTACCAGCGGCGCGAGTTTCAGGCAG-3' , and TERT_REV 5'-AGGCTGATCAGCGGTTTAAACTTAAACGGCAGACTTCGGCTGGCAC-3' . For TERC locus PCR , a puromycin resistance cassette primer ( 5’-TGAAGCCGAGCCGCTCGTAGAA-3’ ) was combined with locus-specific primers: hTR_FWD 5’-GTGGATCCGAGCTCGGTACCACCCACTGAGCCGAGACAAGATTC-3’ and hTR_REV 5’-GAAAGCGAACTGCATGTGTGAGCCG-3’ . Some primers had 5’ regions complementary to the pcDNA3 . 1 vector to facilitate cloning , for which DNA was gel-excised , purified , and ligated into pcDNA3 . 1 ( Thermo ) . DH5a cells were then transformed and several colonies were sequenced . Cell pellets were lysed in RIPA buffer ( 150 mM NaCl , 50 mM Tris at pH 7 . 5 , 1 mM EDTA , 1% Triton X-100 , 0 . 5% sodium deoxycholate , 0 . 1% SDS , 1 mM DTT ) , treated with RNase A for 30 min at 37°C , followed by proteinase K treatment for 4 hr to overnight at 50°C . DNA was then purified using phenol-isoamyl alcohol-chloroform , followed by precipitation with isopropanol and NaCl . After pelleting , the DNA was washed twice with ethanol and resuspended in TE buffer ( 10 mM Tris , 1 mM EDTA ) . DNA concentration was determined by absorbance at 260 nm . Two to 8 µg of DNA was then digested with MboI and AluI for 6 hr to overnight before electrophoresis in a 0 . 7% agarose gel in 1X TAE . The agarose gel was then vacuum dried at 50°C for 1 hr . The dried gel was then denatured with 0 . 5 N NaOH , 1 . 5 M NaCl for 30 min at 50°C . The gel was washed twice with 4X SSC + 0 . 1% SDS and subsequently blocked with Church's buffer for 30 min at 50°C . 32P-end-labeled telomeric repeat probe ( T2AG3 ) 3 and 32P-labeled probe made by random-priming of HinDIII lambda phage digest ladder ( NEB ) and/or the O'generuler 1 kbp plus DNA ladder ( Thermo ) were added . Probes were hybridized overnight at 50°C . The membrane was then extensively washed in 4X SSC + 0 . 1% SDS at 40°C before screen exposure and imaging on a Typhoon scanner ( GE Healthcare ) .

Most cells in the human body can only divide a certain number of times before they die . This is because regions called telomeres at the ends of the cell’s DNA get shorter every time the cell divides , to the point that they disappear and halt cell growth . Particular types of cells – including some stem cells and cancer cells – can avoid death and continue to divide indefinitely because they produce an enzyme called telomerase that extends the telomere regions . The process by which the telomerase enzyme binds to and lengthens the DNA has several stages and involves many different proteins . One of the stages involves moving telomerase from the sites where it is assembled within the cell to a place where it can find telomeres in need of elongation ( different areas within the cell compartment called the nucleus ) . Structures inside the nucleus called Cajal bodies were thought to help the enzyme bind to the telomeres . It is not clear why the process of extending telomeres is so complex . Vogan et al . engineered altered versions of telomerase that use simpler pathways to bind to and act on telomeres and inserted them into ‘pluripotent’ stem cells and cancer cells from humans . The experiments show that a pathway that helps to move the enzyme from its normal storage place in the nucleus is less important for extending telomeres in cancer cells than in pluripotent stem cells . Unexpectedly , Cajal bodies are not critical for bringing telomerase into contact with the telomeres in either cell type . The findings show that many of the proteins involved in extending telomeres in cells are not strictly essential . The simplified pathway developed by Vogan et al . opens up new opportunities to study the details of how telomerase extends telomeres .

2016

Minimized human telomerase maintains telomeres and resolves endogenous roles of H/ACA proteins, TCAB1, and Cajal bodies

Zika virus was discovered in Uganda in 1947 and is transmitted by Aedes mosquitoes , which also act as vectors for dengue and chikungunya viruses throughout much of the tropical world . In 2007 , an outbreak in the Federated States of Micronesia sparked public health concern . In 2013 , the virus began to spread across other parts of Oceania and in 2015 , a large outbreak in Latin America began in Brazil . Possible associations with microcephaly and Guillain-Barré syndrome observed in this outbreak have raised concerns about continued global spread of Zika virus , prompting its declaration as a Public Health Emergency of International Concern by the World Health Organization . We conducted species distribution modelling to map environmental suitability for Zika . We show a large portion of tropical and sub-tropical regions globally have suitable environmental conditions with over 2 . 17 billion people inhabiting these areas . Zika virus ( ZIKV ) is an emerging arbovirus carried by mosquitoes of the genus Aedes ( Musso et al . , 2014 ) . Although discovered in Uganda in 1947 ( Dick et al . , 1952; Dick , 1953 ) ZIKV was only known to cause sporadic infections in humans in Africa and Asia until 2007 ( Lanciotti et al . , 2008 ) , when it caused a large outbreak of symptomatic cases on Yap island in the Federated States of Micronesia ( FSM ) , followed by another in French Polynesia in 2013–14 and subsequent spread across Oceania ( Musso et al . , 2015a ) . In the 2007 Yap island outbreak , it was estimated that approximately 20% of ZIKV cases were symptomatic . While indigenous transmission of ZIKV to humans was reported for the first time in Latin America in 2015 ( Zanluca et al . , 2015; World Health Organisation , 2015 ) , recent phylogeographic research estimates that the virus was introduced into the region between May and December 2013 ( Faria et al . , 2016 ) . This recent rapid spread has led to concern that the virus is following a similar pattern of global expansion to that of dengue and chikungunya ( Musso et al . , 2015a ) . ZIKV has been isolated from 19 different Aedes species ( Haddow et al . , 2012; Grard et al . , 2014 ) , but virus has been most frequently found in Ae . aegypti ( Monlun et al . , 1992; Marchette et al . , 1969; Smithburn , 1954; Pond , 1963; Faye et al . , 2008; Foy et al . , 2011b; WHO Collaborating Center for Reference and Research on Arboviruses and Hemorrhagic Fever Viruses: Annual Report , 1999 ) . These studies were based upon ancestral African strains of ZIKV , but the current rapid spread of ZIKV in Latin America is indicative of this highly efficient arbovirus vector ( Marcondes and Ximenes , 2015 ) . The relatively recent global spread of Ae . albopictus ( Benedict et al . , 2007; Kraemer et al . , 2015c ) and the rarity of ZIKV isolations from wild mosquitoes may also partially explain the lower frequency of isolations from Ae . albopictus populations . Whilst virus transmission by Ae . albopictus and other minor vector species has normally resulted in only a small number of cases ( Kutsuna et al . , 2015; Roiz et al . , 2015 ) , these vectors do pose the threat of limited transmission ( Grard et al . , 2014 ) . The wide geographic distribution of Ae . albopictus combined with the frequent virus introduction via viraemic travellers ( McCarthy 2016; Bogoch et al . , 2016; Morrison et al . , 2008; Scott and Takken , 2012 ) , means the risk for ZIKV infection via this vector must therefore also be considered in ZIKV mapping . The fact that ZIKV reporting was limited to a few small areas in Africa and Asia until 2007 means that global risk mapping has not , until recently , been a priority ( Pigott et al . , 2015b ) . Recent associations with Guillain-Barré syndrome in adults and microcephaly in infants born to ZIKV-infected mothers ( World Health Organisation , 2015; Martines et al . , 2016 ) have revealed that ZIKV could lead to more severe complications than the mild rash and flu-like symptoms that characterize the majority of symptomatic cases ( Gatherer and Kohl , 2016 ) . Considering these potentially severe complications and the rapid expansion of ZIKV into previously unaffected areas , the global public health community needs information about those areas that are environmentally suitable for transmission of ZIKV to humans . Being a closely related flavivirus to DENV , there is furthermore the potential for antigen-based diagnostic tests to exhibit cross-reactivity when IgM ELISA is used for rapid diagnosis . Although ZIKV-specific serologic assays are being developed by the U . S . Centers for Disease Control , currently the only method of confirming ZIKV infection is by using PCR on acute specimens ( Lanciotti et al . , 2008 , Faye et al . , 2008 ) . Awareness of suitability for transmission is essential if proper detection methods are to be employed . In this paper , we use species distribution modelling techniques that have been useful for mapping other vector-borne diseases such as dengue ( Bhatt et al . , 2013 ) , Leishmaniasis ( Pigott et al . , 2014b ) , and Crimean-Congo Haemorrhagic Fever ( Messina et al . , 2015b ) to map environmental suitability for ZIKV . The environmental niche of a disease can be identified according to a combination of environmental conditions supporting its presence in a particular location , with statistical modelling then allowing this niche to be described quantitatively ( Kraemer et al . , 2016 ) . Niche modelling uses records of known disease occurrence alongside hypothesized environmental covariates to predict suitability for disease transmission in regions where it has yet to be reported ( Elith and Leathwick , 2009 ) . Contemporary high spatial-resolution global data representing a variety of environmental conditions allows for these predictions to be made at a global scale ( Hay et al . , 2006 ) . Figure 1A shows the locations of the 323 standardized occurrence records in the final dataset , classified by the following date ranges: ( i ) up until 2006 ( before the outbreak in FSM ) ; ( ii ) between 2007 ( the year of the FSM outbreak ) and 2014; and ( iii ) since 2015 , the first reporting of ZIKV in the Americas . This map is accompanied by the graph in Figure 1B , showing the number of reported occurrence locations globally by year . These figures highlight the more sporadic nature of reporting until recent years , with the majority of occurrences in the dataset ( 63% ) coming from the recent 2015–2016 outbreak in Latin America . 10 . 7554/eLife . 15272 . 003Figure 1 . ( A ) Map showing the distribution of the final set of 323 ZIKV occurrence locations entered into the ensemble Boosted Regression Tree modelling procedure . Locations are classified by year of occurrence to show those which took place ( i ) prior to the 2007 outbreak in Federated States of Micronesia; ( ii ) between 2007–2014; and ( iii ) during the 2015–2016 outbreak; ( B ) the total number of locations reporting symptomatic ZIKV occurrence in humans globally over time . DOI: http://dx . doi . org/10 . 7554/eLife . 15272 . 00310 . 7554/eLife . 15272 . 004Figure 1—figure supplement 1 . Maps of all covariates entered into the 300 BRT models . ( A ) probability of being urban , 2015; ( B ) enhanced vegetation index; ( C ) minimum relative humidity; ( D ) cumulative annual precipitation ( mm ) ; ( E ) temperature suitability for dengue via Ae . aegypti; ( F ) temperature suitability for dengue via Ae . albopictusDOI: http://dx . doi . org/10 . 7554/eLife . 15272 . 004 The final map that resulted from the mean of 300 ensemble Boosted Regression Tree ( BRT ) models is shown in Figure 2A ( with greater detail shown for each region in Figures 2B–D ) . Figure 2—figure supplement 1 shows the distribution of uncertainty based upon the upper and lower prediction quantiles from the 300 models . We restricted our models to make predictions only within areas where i ) mosquito vectors ( in this case Ae . aegypti ) were able to persist and ii ) where temperature was sufficient for arboviral replication within the mosquito . The former of these was calculated by taking the Ae . aegypti probability of occurrence ( Kraemer et al . , 2015c ) value that incorporated 90% of all known occurrences ( Kraemer et al . , 2015b ) ( giving a threshold value of 0 . 8 and greater ) while the latter was evaluated using a mechanistic mosquito model ( Brady et al . , 2013; 2014 ) , which identified regions where arboviral transmission could be sustained for at least 355 days ( one year minus the human incubation period ) in an average year . Figure 3 is a country-level map distinguishing between those countries that are currently reporting ZIKV , those which have reported ZIKV in the past , those which have highly suitable areas for transmission , and those which are unsuitable . Our models predicted high levels of risk for ZIKV in many areas within the tropical and sub-tropical zones . Large portions of the Americas are suitable for transmission , with the largest areas of risk occurring in Brazil , followed by Colombia and Venezuela , all of which have reported high numbers of cases in the 2015–2016 outbreak . In Brazil , where the highest numbers of ZIKV are reported in the ongoing epidemic , the coastal cities in the south as well as large areas of the north are identified to have the highest environmental suitability of ZIKV . The central region of Brazil , on the other hand , has low population densities and smaller mosquito populations , which is reflected in the relatively low suitability for ZIKV transmission seen in the map . Although ZIKV has yet to be reported in the USA , a large portion of the southeast region of the country , including much of Texas through to Florida , is also highly suitable for transmission . Potential risk for ZIKV transmission is high in much of sub-Saharan Africa , with continuous suitability in the Democratic Republic of Congo and surrounding areas and several sporadic case reports in western sub-Saharan countries since the 1950s . Although no symptomatic cases have yet been reported in India , a large portion of this country is at potential risk for ZIKV transmission ( over 2 million square kilometres ) , with environmental suitability extending from its northwest regions through to Bangladesh and Myanmar . The Indochina region , southeast China , and Indonesia all have large areas of environmental suitability as well , extending into Oceania . While only representing less than ten percent of Australia’s total land area , the area shown to be suitable for ZIKV transmission in its northernmost regions is considerable ( comprising nearly 250 , 000 square kilometres ) . 10 . 7554/eLife . 15272 . 005Figure 2 . Maps of ( A ) global environmental suitability for ZIKV , ranging from 0 ( grey ) to 1 ( red ) , showing greater detail for ( B ) the Americas , ( C ) Africa , and ( D ) Asia and Oceania . DOI: http://dx . doi . org/10 . 7554/eLife . 15272 . 00510 . 7554/eLife . 15272 . 006Figure 2—figure supplement 1 . Uncertainty around Zika suitability predictions displayed in main manuscript – Figure 2 , ranging from less than 0 . 01 ( very little uncertainty ) to 0 . 94 ( greatest uncertainty ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15272 . 00610 . 7554/eLife . 15272 . 007Figure 2—figure supplement 2 . Effect plots for each covariate entered into the ensemble of 300 BRT models . ( A ) minimum relative humidity; ( B ) cumulative annual precipitation ( mm ) ; ( C ) enhanced vegetation index; ( B ) probability of being urban ( % ) ; ( E ) temperature suitability for dengue via Ae . aegypti; ( F ) temperature suitability for dengue via Ae . albopictus . DOI: http://dx . doi . org/10 . 7554/eLife . 15272 . 00710 . 7554/eLife . 15272 . 008Figure 2—figure supplement 3 . Environmental suitability for Zika virus transmission to humans , not taking into account temperature suitability for dengue via Aedes albopictus . Covariate effects are as follows: cumulative annual precipitation ( 67 . 4% ) ; temperature suitability for dengue via Ae . aegypti ( 16 . 9% ) ; probability of being urban , 2015 ( 8 . 2% ) ; enhanced vegetation index ( 5 . 1% ) ; minimum relative humidity ( 2 . 4% ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15272 . 00810 . 7554/eLife . 15272 . 009Figure 2—figure supplement 4 . Map showing areas predicted to have greater dengue suitability ( from Bhatt et al . , 2013 , Nature ) vs those which are predicted to have greater Zika suitability in the current study . These values are restricted to areas where both diseases had non-zero predictions . DOI: http://dx . doi . org/10 . 7554/eLife . 15272 . 00910 . 7554/eLife . 15272 . 010Figure 3 . Status of ZIKV reporting as of 2016 by country , showing countries that are highly environmentally suitable ( having a suitable area of more than 10 , 000 square kilometres ) but which have not yet reported symptomatic cases of ZIKV in humans . 'Currently reporting' countries are those having reported cases since 2015 . DOI: http://dx . doi . org/10 . 7554/eLife . 15272 . 010 Our models showed ZIKV risk to be particularly influenced by annual cumulative precipitation , contributing 65 . 0% to the variation in the ensemble of models . The next most important predictor in the model was temperature suitability for DENV transmission via Ae . albopictus , contributing 14 . 6% . These are followed by urban extents ( 8 . 3% ) , temperature suitability for DENV via Ae . aegypti ( 5 . 7% ) , the Enhanced Vegetation Index ( EVI; 3 . 8% ) , and minimum relative humidity ( 2 . 5% ) . Effect plots for each covariate are provided in Figure 2—figure supplement 2 . Validation statistics indicated high predictive performance of the BRT ensemble mean map evaluated in a 10-fold cross-validation procedure , with area under the receiver operating characteristic ( AUC ) of 0 . 829 ( ± 0 . 121 SD ) . Due to the uncertainty about Ae . albopictus as a competent vector for ZIKV , we also provide results for an ensemble of models which did not include temperature suitability for dengue via this mosquito species in Figure 2—figure supplement 3 . A threshold environmental suitability value of 0 . 397 in our final map was determined to incorporate 90% of all ZIKV occurrence locations . This was used to classify each 5 km x 5 km pixel on our final map as suitable or unsuitable for ZIKV transmission to humans . Using high-resolution global population estimates ( WorldPop , 2015; SEDAC , 2015 ) , we summed the populations living in Zika-suitable areas and have identified 2 . 17 billion people globally living within areas that are environmentally suitable for ZIKV transmission . Table 1 shows a breakdown of this figure by major world region , also showing the top four contributing countries to the potential population at risk . Asia has the most people living in areas that are suitable for ZIKV transmission at 1 . 42 billion , accounted for in large part by those living in India . In Africa , roughly 453 million people are living in areas suitable for ZIKV transmission , the largest proportion of which live in Nigeria . In the Americas , more than 298 million people live in ZIKV-suitable transmission zones , with approximately 40 percent of these people living in Brazil . Within the majority of environmentally suitable areas for ZIKV in the Americas , prolonged year-round transmission is possible . Southern Brazil and Argentina , however , are more likely to see transmission interrupted throughout the year , as is the case with the USA should autochthonous ZIKV transmission occur there . Using high-resolution data on births for the year 2015 ( WorldPop , 2015 ) , we also estimate that 5 . 42 million births will occur in the Americas over the next year within areas and times of environmental suitability for ZIKV transmission . 10 . 7554/eLife . 15272 . 011Table 1 . Population living in areas suitable for ZIKV transmission within each major world region and top four countries contributing to these populations at risk . DOI: http://dx . doi . org/10 . 7554/eLife . 15272 . 011Region/CountryPopulation living in areas suitable for ZIKV transmission ( millions ) Africa452 . 58Nigeria111 . 97Democratic Republic of the Congo68 . 95Uganda33 . 43United Republic of Tanzania22 . 70Americas298 . 36Brazil120 . 65Mexico32 . 22Colombia29 . 54Venezuela22 . 22Asia1 , 422 . 13India413 . 19Indonesia226 . 04China213 . 84Bangladesh133 . 29World2 , 173 . 27 In this study , we produced the first global high spatial-resolution map of environmental suitability for ZIKV transmission to humans using an assembly of known records of ZIKV occurrence and environmental covariates in a species distribution modelling framework . While it is clear that much remains to be understood about ZIKV , this first map serves as a baseline for understanding the change in the geographical distribution of this globally emerging arboviral disease . Knowledge of the potential distribution can encourage more vigilant surveillance in both humans and Aedes mosquito populations , as well as help in the allocation of limited resources for disease prevention . Public health awareness campaigns and advice for mitigation of individual risk can also be focused in the areas we have predicted to be highly suitable for ZIKV transmission , particularly during the first wave of infection in a population . The maps we have presented may also inform existing travel advisories for pregnant women and other travellers . The maps and underlying data are freely available online via figshare ( http://www . figshare . com ) . Information about the locations of ZIKV occurrence in humans was extracted from peer-reviewed literature , case reports , and informal online sources following previously established protocols ( Kraemer et al . , 2015b; Messina et al . , 2014; 2015a ) . To collate the peer-reviewed dataset , literature searches were undertaken using PubMed ( http://www . ncbi . nlm . nih . gov/pubmed ) and ISI Web of Science ( http://www . webofknowledge . com ) search engines using the search term 'Zika' . No language restrictions were placed on these searches; however , only those citations with a full title and abstract were retrieved , resulting in the review of 148 references ranging in publication dates between 1951 and 2015 . In-house language skills allowed review of all English , French , Portuguese and Spanish articles for useable location information for human ZIKV occurrence . ProMED-mail ( http://www . promedmail . org ) was also searched using the term 'Zika' , resulting in the review of 139 reports between 27 June 2007 and 18 January 2016 . Additionally , the most current database of ZIKV case locations in Brazil was obtained directly from the Brazilian Ministry of Health . From all sources , only laboratory confirmation of symptomatic ZIKV infection in humans was entered into the dataset ( mention of suspected cases was not entered ) . Serological evidence from healthy individuals could represent a past infection , with transmission potentially occurring in a different location to that where the individual currently resides ( Darwish et al . , 1983 ) , or could be an artefact from possible cross-reactivity with a variety of different viruses ( Smithburn et al . , 1954 ) . As a result , these less reliable diagnoses of ZIKV were excluded . All available location information was extracted from each peer-reviewed article and ProMED case report . The site name was used together with all contextual information provided about the site to determine its latitudinal and longitudinal coordinates using Google Maps ( https://www . maps . google . com ) . If the study site could be geo-positioned to a specific place , it was recorded as a point location . If the study site could only be identified at an administrative area level ( e . g . province or district ) , it was recorded as a polygon along with an identifier of its administrative unit . If imported cases were reported with information on the site of infection , they were geo-positioned to this site; if imported cases were reported with no information about the site of infection , they were not entered into the dataset . Informal online data sources were collated automatically by the web-based system HealthMap ( http://www . healthmap . org ) as described elsewhere ( Freifeld et al . , 2008 ) . Alerts for ZIKV were obtained from HealthMap for the years 2014–2016 , and then manually checked for validity . In total , usable location information was extracted from 110 sources . Information was also collected about the status of symptoms in each reported occurrence , distinguishing between those where symptomatic cases were being reported , versus those where only seroprevalence was detected in healthy individuals . Due to the potential for multiple independent reports referring to the same cases temporal and spatial standardization was required , as we have described previously in detail for dengue mapping efforts ( Messina et al . , 2014 ) . In brief , an occurrence was defined as a unique location with one or more confirmed cases of ZIKV occurring within one calendar year ( the finest temporal resolution available across all records ) . Point locations were considered to be overlapping if they lay on the same 5 km x 5 km pixel , and polygon locations were identified by a unique administrative unit code . Furthermore , all polygons whose geographic area was greater than one square decimal degree ( approximately 111 square kilometers at the equator ) were removed from the dataset to avoid averaging covariate values over very large areas , and only those occurrences comprising symptomatic individuals were retained for modelling purposes to ensure an accurate location of infection . In total , the final occurrence dataset contained 323 unique occurrences to be entered into our BRT modelling procedure . A map of the final set of occurrence locations is provided as Figure 1A . Separate maps of the relative probability of occurrence of Ae . aegypti and Ae . albopictus ( Kraemer et al . , 2015c ) were used to compute a combined metric of the relative probability of vector occurrence , by taking the maximum value from the two layers for all 5 km x 5 km gridded cells globally . The inverse of this combined-Aedes occurrence probability layer ( higher values indicating greater certainty of absence ) was then used to draw a biased sample of 10 , 000 background locations . As such , a greater number of background points were sampled in areas where we are more certain that Ae . aegypti or Ae . albopictus do not occur , and therefore where ZIKV is less likely to be transmitted to humans . While it has been demonstrated that predictive accuracy from presence-background species distribution models can be improved by biasing background record locations toward areas with greatest reporting probabilities ( Phillips et al . , 2009 ) , information on possible reporting biases , or proxies of such spatial bias , are currently unavailable for ZIKV . These 10 , 000 background locations were combined with the standardized occurrence dataset to serve as comparison data locations in the BRT species distribution modelling procedure . The background locations were weighted such that their total sum was equal to the total number of occurrence locations ( n=237; pseudo-absence weighting=0 . 0237 ) , in order to aid in the discrimination capacity of the model ( Barbet-Massin et al . , 2012 ) . A set of six covariates hypothesized to influence the global distribution of ZIKV transmission to humans were used in our models to establish an empirical relationship between ZIKV presence or absence and underlying environmental conditions . These six covariates included: ( i ) an index of temperature suitability for dengue transmission to humans via Ae . aegypti; ( ii ) temperature suitability for dengue transmission to humans via Ae . albopictus; ( iii ) minimum relative humidity; ( iv ) annual cumulative precipitation; ( v ) an enhanced vegetation index ( EVI ) ; and ( vi ) urban versus rural habitat type . The underlying hypothesis behind each of the covariates is discussed in more detail below , along with a description of data sources and any processing that was undertaken before entering these covariates into our models . Maps of each covariate layer are provided in the supplementary information in Figure 1—figure supplement 1 . The boosted regression tree ( BRT ) modelling procedure combines regression trees with gradient boosting ( Friedman , 2001 ) . In this procedure , an initial regression tree is fitted and iteratively improved upon in a forward stagewise manner ( boosting ) by minimising the variation in the response not explained by the model at each iteration . It has been shown to fit complicated response functions efficiently , while guarding against over-fitting by use of extensive internal cross-validation . As such , this approach has been successfully employed in the past to map dengue and its Aedes mosquito vectors , as well as other vector-borne diseases ( Bhatt et al . , 2013; Pigott et al . , 2014b; Messina et al . , 2015b; Kraemer et al . , 2015c ) . To increase the robustness of model predictions and quantify model uncertainty , we fitted an ensemble ( Araújo and New , 2007 ) of 300 BRT models to separate bootstraps of the data . We then evaluated the central tendency as the mean across all 300 BRT models ( Bhatt et al . , 2013 ) . Each of the 300 individual models was fitted using the gbm . step subroutine in the dismo package in the R statistical programming environment ( Elith et al . , 2008 ) . All other tuning parameters of the algorithm were held at their default values ( tree complexity= 4 , learning rate= 0 . 005 , bag fraction= 0 . 75 , step size= 10 , cross-validation folds=10 ) . Each of the 300 models predicts environmental suitability on a continuous scale from 0 to 1 , with a final prediction map then being generated by calculating the mean prediction across all models for each 5 km x 5 km pixel . Cross-validation was applied to each model , whereby ten subsets of the data comprising 10% of the presence and background observations were assessed based on their ability to predict the distribution of the other 90% of records using the mean area under the curve ( AUC ) statistic . This AUC value was then averaged across the ten sub-models and finally across all 300 models in the ensemble in order to derive an overall estimate of goodness-of-fit . Additionally , to avoid AUC inflation due to spatial sorting bias , a pairwise distance sampling procedure was used , resulting in a final AUC which is lower than would be returned by standard procedures but which gives a more realistic quantification of the model’s ability to extrapolate predictions to new regions ( Wenger and Olden , 2012 ) . We restricted our models to make predictions only within areas where either Ae . aegypti probability of occurrence ( Kraemer et al . , 2015c ) is more than 0 . 8 or temperature is conducive to transmission for at least 355 days in an average year . A second ensemble of 300 models was executed which did not take into account temperature suitability for dengue transmission via Ae . albopictus , due to the uncertainty of this species as a competent ZIKV vector . The results of this ensemble of models are provided in Figure 2—figure supplement 3 . To calculate the number of people located in an area that is at any level of risk for ZIKV transmission , the global ZIKV environmental suitability map was combined with fine-scale global population surfaces ( SEDAC , 2015; WorldPop , 2015 ) . Firstly , the continuous ZIKV environmental suitability map ( ranging from 0 to 1 ) was converted into a binary surface indicating whether there is any risk of transmission . To do this , we carried out a protocol previously used in ( Pigott et al . , 2015a ) , choosing a threshold environmental suitability value that encompasses 90% of the ZIKV occurrence point locations . This threshold cut-off of 90% was chosen ( rather than 100% ) to reflect potential errors or inaccurate locations in the occurrence point dataset . Every 5 km x 5 km pixel in the suitability map with a value above this threshold value was considered at risk for ZIKV transmission . Finally , to estimate the population at risk , we multiplied this binary ZIKV risk map by the global population counts ( aligned and aggregated to the same 5 x 5 km grid ) for the year 2015 and summed across all cells . We next estimated the maximum number of births potentially affected by ZIKV in Latin America , as this region is the focus of the recent outbreak and the first to point to a possible association with microcephaly in newborn infants to mothers infected with ZIKV . In order to do this , we first identified the proportion of the year that is suitable for ZIKV transmission within areas that are predicted to be suitable in the binary ZIKV risk map . This proportion was derived from existing temperature suitability models ( Brady et al . , 2013; 2014 ) , which predict the total number of days within an average year that arbovirus transmission can be sustained in Ae . aegypti , assuming there is a local human reservoir of infection . While the intra-mosquito viral dynamics in this model were parameterised for dengue virus , the limited information currently available on other arboviruses suggests that their dynamics are similar ( Lambrechts et al . , 2011 ) . Using the resulting 5 km x 5 km map showing the proportion of the year suitable for ZIKV transmission to humans , we then multiplied this by a map ( also at a 5 km x 5 km resolution ) of the number of births in the Americas for the year 2015 , updated from ( Tatem et al . , 2014; UNFPA , 2014 ) . The resulting map indicates the number of births in the Americas potentially at risk for ZIKV ( for 2015 ) , assuming ZIKV currently fully occupies its environmental niche and that births are evenly distributed throughout the year .

Zika virus is transmitted between humans by mosquitoes . The majority of infections cause mild flu-like symptoms , but neurological complications in adults and infants have been found in recent outbreaks . Although it was discovered in Uganda in 1947 , Zika only caused sporadic infections in humans until 2007 , when it caused a large outbreak in the Federated States of Micronesia . The virus later spread across Oceania , was first reported in Brazil in 2015 and has since rapidly spread across Latin America . This has led many people to question how far it will continue to spread . There was therefore a need to define the areas where the virus could be transmitted , including the human populations that might be risk in these areas . Messina et al . have now mapped the areas that provide conditions that are highly suitable for the spread of the Zika virus . These areas occur in many tropical and sub-tropical regions around the globe . The largest areas of risk in the Americas lie in Brazil , Colombia and Venezuela . Although Zika has yet to be reported in the USA , a large portion of the southeast region from Texas through to Florida is highly suitable for transmission . Much of sub-Saharan Africa ( where several sporadic cases have been reported since the 1950s ) also presents an environment that is highly suitable for the Zika virus . While no cases have yet been reported in India , a large portion of the subcontinent is also suitable for Zika transmission . Over 2 billion people live in Zika-suitable areas globally , and in the Americas alone , over 5 . 4 million births occurred in 2015 within such areas . It is important , however , to recognize that not all individuals living in suitable areas will necessarily be exposed to Zika . We still lack a great deal of basic epidemiological information about Zika . More needs to be known about the species of mosquito that spreads the disease and how the Zika virus interacts with related viruses such as dengue . As such information becomes available and clinical cases become routinely diagnosed , the global evidence base will be strengthened , which will improve the accuracy of future maps .

2016

Mapping global environmental suitability for Zika virus

Event-related potentials ( ERPs ) are widely used in both healthy and neuropsychiatric conditions as physiological indices of cognitive functions . Contrary to the common belief that cognitive ERPs are generated by local activity within the cerebral cortex , here we show that an attention-related ERP in the frontal cortex is correlated with , and likely generated by , subcortical inputs from the basal forebrain ( BF ) . In rats performing an auditory oddball task , both the amplitude and timing of the frontal ERP were coupled with BF neuronal activity in single trials . The local field potentials ( LFPs ) associated with the frontal ERP , concentrated in deep cortical layers corresponding to the zone of BF input , were similarly coupled with BF activity and consistently triggered by BF electrical stimulation within 5–10 msec . These results highlight the important and previously unrecognized role of long-range subcortical inputs from the BF in the generation of cognitive ERPs . Event-related potential ( ERP ) represents the stereotypical response of the electroencephalogram ( EEG ) activity to an internal or external stimulus , and reflects reproducible and highly coherent large-scale activity patterns in the underlying cerebral cortical network ( Davis , 1939; Rohrbaugh et al . , 1990; Luck , 2005a; Luck and Kappenman , 2012 ) . ERPs have been widely used in both healthy and neuropsychiatric conditions as robust physiological indices of cognitive functions because the amplitudes of late ERP components are modulated by various cognitive functions including attention ( Hillyard and Kutas , 1983; Näätänen , 1988; Schreiber et al . , 1992; Iragui et al . , 1993; Polich , 1997; Herrmann and Knight , 2001; Barry et al . , 2003; Caravaglios et al . , 2008; Luck and Kappenman , 2012 ) . Despite the broad applications of ERP in both basic and clinical research , there remains considerable debate regarding the source mechanisms that generate cognitive ERPs ( Snyder , 1991; Miltner et al . , 1994; Pascual-Marqui et al . , 2002; Luck , 2005b; Nunez and Srinivasan , 2006; Cohen et al . , 2009; Riera et al . , 2012 ) . It is commonly assumed that cognitive ERPs are generated by , and therefore reflect the functions of , local activity within the cerebral cortex . However , inferring the underlying sources based on the skull surface EEG pattern , generally referred to as the ‘inverse problem’ , is difficult and lacks a unique mathematical solution because the same EEG pattern can be generated by many different configurations of underlying sources ( Helmholtz , 1853; Luck , 2005b; Nunez and Srinivasan , 2006 ) . For this reason , it has remained difficult to experimentally demonstrate how cognitive ERPs are generated . While most studies have focused on identifying cortical activities responsible for generating cognitive ERPs , one neglected possibility is that subcortical inputs play a significant role . In this study , we explored an alternative hypothesis that ERPs are driven by subcortical inputs from the basal forebrain ( BF ) , one of the major cortically-projecting neuromodulatory systems ( Semba , 2000; Zaborszky , 2002; Jones , 2003 ) . This hypothesis is motivated by recent findings that a subset of BF neurons is robustly activated by motivationally salient stimuli that attract animals’ attention ( Richardson and Delong , 1991; Lin and Nicolelis , 2008; Avila and Lin , 2014 ) , and that the phasic bursting response of these BF neurons is tightly coupled with LFP activity in the frontal cortex ( Lin et al . , 2006 ) . These discoveries led us to the hypothesis that ERPs elicited by motivationally salient stimuli are generated by the bursting response of BF neurons . To test this hypothesis , we trained rats to perform an auditory oddball task that is commonly used in human ERP studies and observed a robust ERP response in the frontal cortex that was coupled with behavioral performance . We simultaneously recorded skull surface EEG and BF single neuron activity , and investigated whether frontal ERP and BF bursting activity were correlated , in single trials , in terms of amplitude and timing . To better understand how the frontal ERP was generated , we recorded LFPs throughout all layers of frontal cortical regions , and identified the layer-specific LFP response that was coupled with the frontal ERP and BF bursting activity . Finally , we tested whether activating BF via electrical stimulation was sufficient to trigger the layer-specific LFP response in the frontal cortex . The results support that the frontal ERP is correlated with , and likely generated by , subcortical inputs from the basal forebrain . We first developed a rat version of the auditory oddball task commonly used in human ERP studies , with the goal of reproducing characteristic ERP responses . In the auditory oddball task ( Figure 1A ) , a standard tone ( 10 kHz ) was presented once every 2 s , and occasionally once every 6–14 s a deviant oddball tone ( 6 kHz ) was presented instead . Adult Long Evans rats ( n = 8 ) were trained to discriminate the standard tone from the oddball tone that signaled the availability of liquid reward . After training , rats showed high hit rates toward the oddball tone and a low false alarm rate toward the standard tone ( Figure 1B ) . 10 . 7554/eLife . 02148 . 003Figure 1 . The frontal ERP in an auditory oddball task is behaviorally relevant and functionally coupled with BF bursting . ( A ) In the auditory oddball task , a standard tone ( 10 kHz ) was presented once every 2 s , and occasionally once every 6–14 s a deviant oddball tone ( 6 kHz ) was presented that signaled reward if responded to within a 3-s window ( yellow ) . ( B ) Behavioral performance in the oddball task from eight rats ( mean ± SEM ) . ( C ) Grand-average ERPs in the frontal cortex relative to tone onsets in three trial types , plotted separately while rats actively performed the oddball task or when the tones were presented passively without any reward . ( D ) The frontal ERP in hit trials from eight individual sessions ( six rats ) . Sessions were sorted by peak ERP latency , the same as in ( E ) and ( G ) . ( E ) Frontal ERP amplitude reliably discriminated hit from miss responses to the oddball tone in individual sessions , calculated based on signal detection theory . Only bins reaching statistical significance ( p<0 . 01 ) were plotted . ( F ) Grand-average of BF bursting response relative to tone onsets in three trial types . BF bursting occurred at the same time window as the frontal ERP , and showed similar amplitude modulation between trial types . ( G ) Population BF bursting response in hit trials from the same eight sessions . DOI: http://dx . doi . org/10 . 7554/eLife . 02148 . 00310 . 7554/eLife . 02148 . 004Figure 1—figure supplement 1 . ERP in the frontal cortex , but not the primary visual cortex , provides information about behavioral performance . ( A ) Grand-average ERPs relative to tone onset in three trial types , plotted separately for the frontal cortex ( same as in Figure 1C ) and the primary visual cortex ( V1 ) . V1 ERP was recorded as a control site to illustrate the specificity of the frontal ERP , as well as for studying processing of visual stimuli in related experiments . ( B ) Frontal ERP amplitude reliably discriminated oddball from standard tone trials in individual sessions , calculated based on signal detection theory . Only bins reaching statistical significance ( p<0 . 01 ) were plotted . Conventions and the order of sessions ( n = 8 ) are the same as in Figure 1E . ( C ) Single session V1 ERP in hit trials ( 7 sessions from 6 rats ) , sorted in the same session order as in ( B ) . V1 EEG signal from one session , corresponding to the third from the top in ( B ) , was contaminated with excessive movement artifact and removed from this analysis . ( D–E ) V1 ERP amplitude provided little information about hit vs miss responses in oddball trials ( D ) , or information about oddball vs standard trials ( E ) . Conventions as in ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02148 . 00410 . 7554/eLife . 02148 . 005Figure 1—figure supplement 2 . BF electrode locations and the classification of BF bursting neurons . ( A ) Each set of color boxes represent the locations of bilateral BF electrode bundle in one rat ( n = 8 ) . BF electrodes were located between −0 . 12 mm–−0 . 84 mm relative to Bregma throughout multiple subregions , including the ventral part of globus pallidus ( GP ) , ventral pallidum ( VP ) , substantia innominata ( SI ) , nucleus basalis of Meynert ( NBM , or B ) , magnocellular preoptic nucleus ( MCPO ) and horizontal limb of the diagonal band ( HDB ) . This widespread spatial distribution of BF bursting neurons is consistent with the location of cortically-projecting BF neurons as revealed by placing retrograde tracers in the prefrontal cortex ( Gritti et al . , 1997 ) . ( B ) PSTHs of all 168 BF neurons in response to oddball tone onset in hit trials . BF neurons were sorted by their bursting index , defined as the ratio of the bursting amplitude ( average firing rate in the [50 , 200] msec window ) over the average firing rate in the entire session . Bursting index of 2 . 5 ( indicated by the white dashed line ) was used as the cutoff for classifying BF bursting neurons . ( C ) Scatter plot of the baseline firing rate vs bursting amplitude for all BF neurons . Each dot represents one BF neuron . The red dashed line indicates the 2 . 5 bursting index cutoff used for classifying BF bursting neurons . DOI: http://dx . doi . org/10 . 7554/eLife . 02148 . 005 We found a prominent negative ERP response in the frontal cortex when rats correctly responded to the oddball tone ( hit ) ( Figure 1C–D ) . This ERP response started at 50 msec , peaked around 100 msec , and was absent when the same stimuli were not behaviorally relevant ( Figure 1C ) . Furthermore , the amplitude of this frontal ERP reliably discriminated hit from miss responses to oddball tones ( Figure 1E ) , and also discriminated oddball from standard tones ( Figure 1—figure supplement 1 ) . These results demonstrate that this frontal ERP is highly relevant for behavioral performance and does not simply reflect sensory processing . In 120 BF neurons recorded simultaneously with frontal EEG activity , 55% ( 66/120 ) showed robust bursting response to the oddball tone in hit trials and were classified as BF bursting neurons ( Figure 1—figure supplement 2; Table 1 ) . These BF neurons showed stronger bursting responses in hit trials than in miss trials , and in oddball trials than in standard trials ( Figure 1F ) , consistent with the encoding of motivational salience as previously reported ( Lin and Nicolelis , 2008; Avila and Lin , 2014 ) . Interestingly , the timing of the BF bursting response as well as the modulation of BF bursting amplitude between different trial types ( Figure 1F , G ) were highly similar to those of the frontal ERP , suggesting that BF bursting may be functionally coupled with the frontal ERP . 10 . 7554/eLife . 02148 . 006Table 1 . List of animals and sessions used for each analysisDOI: http://dx . doi . org/10 . 7554/eLife . 02148 . 006Animal IDBF neurons ( Bursting/All ) EEG ( Frontal/V1 ) LFPBF stim ( awake ) BF stim ( isofurane ) Rat #17/8Frontal/V1YESYESYESRat #210/18Frontal/V1YESYESYESRat #28/14Frontal/V1YESRat #39/14–YESYESYESRat #35/7YESRat #44/11Frontal/V1YESYESYESRat #516/27–YESYES–Rat #614/31Frontal/V1YES––Rat #74/6Frontal/V1YES––Rat #86/15Frontal/-–––Rat #813/17Frontal/V18 rats 11 sessionsn = 96/168 neuronsFX = 8/V1 = 7 sessions; n = 66/1209 sessions n = 77/1365 sessions4 sessionsFigure 1BFigure 1—figure supplement 2Figures 1–3 , Figure 1—figure supplement 1 , Figure 3—figure supplement 1Figure 4 , Figure 4—figure supplement 1 , Figure 4—figure supplement 2Figure 5 , Figure 4—figure supplement 1 , Figure 4—figure supplement 2Figure 5 , Figure 4—figure supplement 1 , Figure 4—figure supplement 2A detailed list of all 11 sessions from 8 rats used for the current study . Animal IDs correspond to those used in Figures 4 , 5 , Figure 4—figure supplement 1 , Figure 4—figure supplement 2 . Two sessions were recorded in three animals ( Rat #2 , 3 , 8 ) with BF electrodes positioned at two different depths to record from independent BF neuronal ensembles . The number of neurons indicated by n . Data used for each analysis are indicated in the bottom row . Five rats ( Rats #1 , 2 , 4 , 7 , 8 ) were also recorded in one session of passive oddball task in which the stimuli were not behaviorally relevant ( Figure 1C ) . To determine whether the amplitudes of BF bursting and the frontal ERP were coupled in single trials , we first visualized their relationship by sorting oddball and standard trials based on the amplitude of BF population bursting response in the 50–200 msec window . The example depicted in Figure 2A illustrates the frequent observation that trials with strong BF bursting responses were accompanied by strong frontal ERPs , while trials with weak BF bursting were accompanied by weak frontal ERPs . Indeed , significant single trial amplitude correlation ( p<0 . 001 ) was observed in 95% ( 63/66 ) of BF bursting neurons ( Figure 2B ) . Furthermore , the amplitude coupling relationships between BF bursting and the frontal ERP were well described by a highly homogeneous linear scaling function ( Figure 2B ) despite the variable bursting amplitude among BF bursting neurons ( Figure 1—figure supplement 2 ) . This linear scaling relationship indicates that the amplitude coupling between BF bursting and the frontal ERP remains invariant in the face of different types of stimulus ( oddball vs standard ) or behavioral response ( hit vs miss ) , and therefore likely reflects a true functional coupling between BF bursting and the frontal ERP . 10 . 7554/eLife . 02148 . 007Figure 2 . Single trial amplitude coupling between BF bursting and frontal ERP . ( A ) An example session showing single trial activities relative to tone onsets in the BF bursting neuron population ( left ) , frontal EEG ( middle ) and one LFP channel in the deep layer of the frontal cortex ( right ) . Trials were sorted based on the amplitude of BF population bursting response ( left ) . ( B and C ) For each BF bursting neuron ( B ) and non-bursting neuron ( C ) , the mean amplitudes of BF activity and ERP in each quintile of trials were normalized by their respective average amplitudes in hit trials . Each neuron is represented by a gray line , with the population average ( ±SEM ) for BF bursting neurons and non-bursting neurons shown in red ( B ) or blue ( C ) , respectively . 63/66 BF bursting neurons and 11/54 non-bursting BF neurons showed significant single trial amplitude correlation ( p<0 . 001 ) . BF bursting neurons showed a highly homogeneous linear scaling relationship with ERP amplitude . ( D ) The scatter plot and histograms of R- and p-values for BF-ERP amplitude correlations for all BF neurons . Significantly correlated neurons were shown in filled symbols . DOI: http://dx . doi . org/10 . 7554/eLife . 02148 . 007 In contrast to the homogeneous amplitude coupling relationship in BF bursting neurons , only 20% ( 11/54 ) of non-bursting BF neurons showed significant single trial amplitude correlation with the frontal ERP , which consisted of heterogeneous coupling relationships ( Figure 2C ) that were less correlated with the frontal ERP compared to BF bursting neurons ( Figure 2D ) . This finding is evidence that the frontal ERP is coupled specifically with the subset of BF neurons that encode motivational salience using the phasic bursting response ( Lin and Nicolelis , 2008; Avila and Lin , 2014 ) . In addition to amplitude coupling , BF bursting and the frontal ERP were also coupled temporally . At the gross temporal scale , BF bursting and the frontal ERP always occurred in the same time window despite the variable latency of the frontal ERP responses between sessions and across trial types ( Figure 1D–G ) . These coupling relationships are compatible with two concurrently plausible explanations: First , it may be that both the BF bursting and frontal ERP are controlled by common inputs from un-observed third brain region ( s ) . A second possibility is that the frontal ERP is driven by BF activity with a very short delay . To further examine these two possibilities , we determined the time lag that gave the maximal cross correlation between each BF bursting neuron and the frontal ERP ( Figure 3 , Figure 3—figure supplement 1 ) . This analysis showed that , while the activity of many BF bursting neurons either preceded or followed the frontal ERP , the BF neurons that temporally led the frontal ERP were better correlated with the frontal ERP and showed stronger bursting responses . In other words , the contributions from BF bursting neurons that led or trailed the frontal ERP were not equal . These results support the hypothesis that the frontal ERP may be driven by the strongly bursting BF neurons that led the ERP by 5–10 msec through a fast circuit mechanism . 10 . 7554/eLife . 02148 . 008Figure 3 . Fine temporal relationship between BF bursting and the frontal ERP . ( A ) Cross correlation between the activity of individual BF bursting neurons and the frontal EEG activity . Each BF bursting neuron is represented by one gray line , with dark gray indicating correlation exceeding statistical significance ( p<0 . 01 in 65/66 BF bursting neurons , permutation test ) and red dots indicating the maximum correlation . ( B ) Correlation between the BF-ERP delay that produced the maximum correlation against the maximum correlation coefficient in each BF bursting neuron . The significant positive correlation shows that the BF bursting neurons whose activity consistently led the frontal ERP were better correlated with the frontal ERP , compared to those that trailed the frontal ERP . ( C and D ) Correlation between the BF-ERP delay against the BF bursting index ( C ) and baseline firing rate ( D ) . BF bursting neurons whose activity consistently led the frontal ERP showed stronger bursting responses ( C ) compared to those that trailed the frontal ERP , while having similar baseline firing rates ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02148 . 00810 . 7554/eLife . 02148 . 009Figure 3—figure supplement 1 . Additional analysis on the fine temporal relationship between BF bursting and the frontal ERP . ( A ) The average of statistically-significant correlation functions across all BF bursting neurons from Figure 3A . The peak of the average correlation function occurred when BF activity led the frontal ERP by 5 msec . ( B ) Histogram of the BF-ERP delay that gave the maximum cross correlation in each BF bursting neurons from Figure 3A . Considerable variability in the BF-ERP delay was present among BF bursting neurons , with some significant proportions leading or following the frontal ERP . The mean of the BF-ERP delay occurred when the frontal ERP led BF activity by 4 msec . ( C ) Correlation between the BF-ERP delay against the magnitude of the maximum cross correlation coefficient in each BF bursting neuron ( same as in Figure 3B ) . Blue filled circles indicate the correlation coefficient of the least correlated BF bursting neuron in each session . ( D ) To further ensure that the correlation was not driven by a few outliers , we normalized the magnitude of correlation coefficients by the correlation coefficient of the least correlated BF bursting neuron in that session . This normalization equalized the contribution of different sessions that had different ranges of correlation magnitude , and showed stronger correlation with the BF-ERP delay . These results support that the contributions from the BF bursting neurons that lead or trail the frontal ERP are unequal , and suggest that a simple histogram of the BF-ERP delay that treats all BF bursting neurons equally , as in ( B ) , does not provide a fair estimate of the BF-ERP temporal relationship . The coupling between BF-ERP delay and the strength of BF bursting ( Figure 3C ) further suggests that the result in ( B ) will be significantly affected by the choice of cutoff criteria for classifying BF bursting neurons ( Figure 1—figure supplement 2B , C ) , such that a more liberal criteria will include more BF neurons whose activity trails the frontal ERP , while a more conservative criteria will include only the strongly bursting BF neurons whose activity leads the frontal ERP . In contrast , results in ( A ) , ( C ) and ( D ) are less susceptible to changes in the cutoff criteria . DOI: http://dx . doi . org/10 . 7554/eLife . 02148 . 00910 . 7554/eLife . 02148 . 010Figure 3—figure supplement 2 . A model of how BF bursting activity generates the frontal ERP . Schematic of one possible model of the BF-ERP interaction that is consistent with our findings . This model consists of 10 BF bursting neurons ( green to red traces show the PSTHs ) , where the ones with stronger bursting also have earlier onset latencies . Specifically , we set the peak bursting amplitudes to range from 100 to 20 spikes per second , and the onset latency to stagger by 5 msec . The population average is shown in black . In this model , we assume that all BF bursting neurons contribute to the generation of the frontal ERP with a fixed delay of 5 msec . The contribution of individual BF bursting neurons to the frontal ERP , indicated by dashed traces , cannot be directly observed and only the summed ERP response ( black ) can be experimentally observed . Two example trials are shown to illustrate how the BF bursting amplitude linearly scales with the frontal ERP ( Figure 2 ) : one with bursting amplitude set at 100% ( left ) to resemble a hit trial , and the other with bursting amplitude set at 50% ( right ) to resemble a miss or standard trial . The relative amplitude and relative timing between BF bursting neurons , and also between BF and the frontal ERP , are unaffected by the amplitude scaling . Raster plots for neuron#1 and #10 are shown below the two example trials ( 30 repeats each ) . Baseline tonic activity of BF bursting neurons is omitted for clarity . In this model , the bursting amplitudes of all BF neurons are linearly scaled with the frontal ERP amplitude . The subset of BF neurons with stronger bursting responses temporally leads the frontal ERP , while the activity of the majority of BF neurons trails the frontal ERP , even though all BF neurons causally contribute to ERP generation . This model further predicts higher magnitudes of BF-ERP cross correlation coefficient in BF neurons with stronger bursting responses , even though each of the smoothed PSTHs should be perfectly and equally correlated with the frontal ERP . This prediction arises because the higher firing rate in strongly bursting BF neurons allows the underlying spike trains to better approximate the smoothed PSTH function in single trials ( e . g . , neuron #1 ) , while the stochastic spike trains provide a poor approximation of the smoothed PSTH especially when the BF bursting rate is low ( e . g . , neuron #10 ) , partly because no spike was generated in a significant proportion of trials . Clearly , this model is only one of many possible models compatible with our findings . The purpose of this model is to show that our results are fully compatible with the scenario that all BF neurons causally contributes to the frontal ERP , even for the BF bursting neurons whose activity trails the frontal ERP in the cross correlation analysis . Alternatively , our results are compatible with the model that the subset of BF neurons trailing the frontal ERP may instead be driven by inputs from the frontal cortex . DOI: http://dx . doi . org/10 . 7554/eLife . 02148 . 010 To further understand how the frontal ERP was generated , we recorded LFP activity across all cortical layers in frontal cortical regions below the EEG electrode . Concomitant with the frontal ERP and the BF bursting response , we observed clear LFP responses in a subset of cortical layers ( Figure 4A , B , Figure 4—figure supplement 1 , Figure 4—figure supplement 2 ) . The non-uniform distribution of LFP responses across cortical layers indicates that these LFP responses were not volume conducted from distant sources . 10 . 7554/eLife . 02148 . 011Figure 4 . BF bursting is associated with LFP responses in deep layers of the frontal cortex . ( A and B ) LFP responses to tone onset across cortical layers in the frontal cortex in two representative rats . LFP responses occurred in the same window as BF bursting ( red shaded area ) despite variability in the latency of BF bursting in different animals and trial types . LFP responses were shown in different amplitude scales in oddball ( left ) and standard trials ( right ) to highlight the similar layer profile . ( C ) The layer profiles of LFP responses were similar ( 9/9 sessions ) between oddball and standard trials . Similarity is defined as the cosine of the angle between the two 32-dimension LFP layer profile vectors . ( D ) 66/77 BF bursting neurons showed significant ( p<0 . 001 ) single trial amplitude correlation and linear amplitude scaling with LFP responses . Conventions as in Figure 2B . ( E ) The LFP layer profiles overlaid on histological reconstructions show that the prominent positive LFP responses were located in the deep layers of the frontal cortex . DOI: http://dx . doi . org/10 . 7554/eLife . 02148 . 01110 . 7554/eLife . 02148 . 012Figure 4—figure supplement 1 . LFP layer profiles in the oddball task are similar to those elicited by BF electrical stimulation . LFP layer profiles across cortical layers in the frontal cortex are shown for 9 sessions ( 7 rats ) , including the examples shown in Figures 4 and 5 . The conventions are the same as in Figures 4 and 5 . The two left columns show LFP layer profiles in oddball and standard trials . LFP responses always occurred in the same window as BF bursting ( red shaded areas ) despite variability in the latency of BF bursting response in different animals and trial types . LFP responses were shown in different amplitude scales in oddball and standard trials to highlight the similar layer profile . The similarity of the LFP layer profiles between oddball and standard trials were shown in Figure 4C . Note that two sessions were recorded in Rat#2 and Rat#3 with the BF electrodes at two different depths to sample from distinct BF neuronal populations . These data showed that the LFP layer profiles were highly stable across sessions in the same animal , and that the same LFP layer profile was coupled with the bursting activity of different populations of BF bursting neurons . Also note that the amplitude of BF bursting was the smallest in standard trials of Rat#7 , which was accompanied by an ill-defined LFP layer profile that resembled the oddball LFP layer profile the least . The two middle columns show the overall LFP layer profile overlaid on the reconstructed locations of the linear probe electrodes . The LFP layer profile was defined as the mean LFP activity in the 60 msec window around the peak of BF bursting , averaged across both oddball and standard trials . The amplitudes of LFP layer profiles were normalized to its peak positive value . The prominent positive LFP responses were restricted to the deep layers of the frontal cortex . Arrows indicate the tip of the linear probes . Note that the linear probe in Rat #4 was positioned more ventrally than the tip of the probe in the corresponding photo . This was because some of the frontal cortex sections were not recovered during histology , and therefore the linear probe was positioned at 6 mm below cortical surface according to our surgical record . The two right columns show the LFP responses elicited by a single pulse of BF electrical stimulation , delivered during wakefulness or under isoflurane anesthesia . BF electrical stimulation was conducted in 5/7 rats . BF electrical stimulation generated responses across cortical layers that resembled the LFP layer profile associated with BF bursting in the oddball task . Significantly similar layer profile ( thick red trace ) started , on average , 8 . 4 ± 2 . 7 msec after BF electrical stimulation ( indicated by blue circles ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02148 . 01210 . 7554/eLife . 02148 . 013Figure 4—figure supplement 2 . Additional examples . DOI: http://dx . doi . org/10 . 7554/eLife . 02148 . 013 Furthermore , we found that the layer profiles of LFP responses associated with BF bursting were highly similar between oddball and standard trials ( Figure 4A–C , Figure 4—figure supplement 1 , Figure 4—figure supplement 2 ) . The amplitudes of LFP responses and BF bursting were significantly correlated in single trials ( p<0 . 001 ) in 86% ( 66/77 ) of BF bursting neurons recorded simultaneously with frontal LFPs ( Figure 4D ) . These results indicate that the LFP response amplitudes were linearly scaled between oddball and standard trials , and also linearly scaled with BF bursting amplitudes , similar to what we observed between BF bursting and the frontal ERP ( Figure 2B ) . Therefore , the characteristic LFP layer profile associated with BF bursting response likely reflects the organization of local activity dipoles that generate the frontal ERP at the skull surface . The most prominent feature of the LFP layer profile was a strong positive LFP response located within the deep cortical layers of the frontal cortex ( Figure 4E , Figure 4—figure supplement 1 , Figure 4—figure supplement 2 ) , which are known to be the predominant target layers of cortical projections from the BF ( Henny and Jones , 2008 ) . Finally , to further determine whether BF activity was sufficient to trigger the frontal ERP with a short delay , we investigated whether artificially inducing BF bursting activity can produce the characteristic LFP response in the oddball task . To reliably activate BF bursting neurons , electrical stimulation was chosen over more selective methods such as optogenetics because the neurochemical identity of BF bursting neurons needed for selective activation techniques remains to be established ( Lin and Nicolelis , 2008 ) . We found that a single pulse of BF electrical stimulation in the absence of any auditory stimulus was able to elicit highly reliable LFP responses in frontal cortical regions , both under isoflurane anesthesia and during wakefulness ( Figure 5 , Figure 4—figure supplement 1 , Figure 4—figure supplement 2 ) . BF electrical stimulation generated positive LFP responses in the same deep cortical layers that showed strong positive LFP responses in the oddball task within 5–10 msec ( Figure 5 , Figure 4—figure supplement 1 , Figure 4—figure supplement 2 ) , thus recapitulating the signature features of oddball LFP responses—in terms of timing , layer profile and polarity of LFP responses . BF electrical stimulation also generated an earlier LFP response , possibly due to activation of other neuronal populations in the BF ( Gritti et al . , 1997 , 2003 ) or from antidromic activation of frontal cortex projections to the BF ( Zaborszky et al . , 1997 ) . Together , these results support the idea that the BF bursting in the oddball task is sufficient to generate and account for key features of the frontal ERP response . 10 . 7554/eLife . 02148 . 014Figure 5 . BF electrical stimulation mimics the layer profile of LFP responses associated with BF bursting . ( A and B ) LFP responses to a single pulse of BF electrical stimulation ( 70 µA per electrode ) , delivered either during wakefulness ( left ) or under isoflurane anesthesia ( right ) , in the same two rats shown in Figure 4A , B . The top panels show the similarity between the LFP layer profile evoked by BF electrical stimulation and the LFP layer profile in the oddball task ( red traces ) , defined as the scalar projection of the two layer profile vectors normalized by its peak amplitude . Gray shaded areas indicate the 95% confidence interval ( permutation test ) . ( C ) Significantly similar layer profiles ( thick red traces ) started , on average , 8 . 4 ± 2 . 7 msec after BF electrical stimulation ( blue circles ) . Each trace represents the similarity of layer profiles in one session . BF stimulation sessions during wakefulness and isoflurane are indicated by solid and dotted traces , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 02148 . 014 ERP is one of the oldest and also most widely used methods in neuroscience research ( Davis , 1939; Rohrbaugh et al . , 1990; Luck , 2005a; Luck and Kappenman , 2012 ) . Contrary to the common belief that cognitive ERPs are generated by local activity within the cerebral cortex , we found that a frontal ERP response closely linked to behavioral performance in the auditory oddball task ( Figure 1 ) is tightly coupled with BF bursting activity in single trials in terms of both amplitude ( Figure 2 ) and timing ( Figure 3 ) . The coupling between BF bursting and the frontal ERP remains invariant in the face of different types of stimulus ( oddball vs standard ) and behavioral response ( hit vs miss ) . The frontal ERP is likely generated by the underlying layer-specific LFP response , which is similarly coupled with BF bursting activity ( Figure 4 ) . BF electrical stimulation reproduces the key features of LFP responses in the oddball task ( Figure 5 ) , suggesting that the BF bursting activity is sufficient to generate and account for most of the frontal ERP response in the oddball task . The BF bursting activity therefore provides a novel neural correlate of the frontal ERP response . This is the first time , to the best of our knowledge , that a key component of cognitive ERPs has been linked to the activity of a well-defined neuronal population located outside of the cerebral cortex . These findings highlight the important and previously unrecognized role of long-range subcortical inputs from the BF in generating cognitive ERPs . While the strong correlation between the frontal ERP and BF bursting activity does not rule out the possibility that the observed correlation might be generated by common inputs from other un-sampled structures , our finding that the activity of strongly bursting BF neurons temporally precedes the frontal ERP by 5–10 msec ( Figure 3 ) is more compatible with BF driving the frontal ERP , and less compatible with the common input hypothesis . We suggest that even the weakly bursting BF neurons whose activity trails the frontal ERP may causally contribute to the frontal ERP with a short delay ( Figure 3—figure supplement 2 ) . Another possibility is that the subset of BF bursting neurons trailing the frontal ERP may be driven by inputs from the frontal cortex ( Zaborszky et al . , 1997 ) . The hypothesis that BF bursting activity generates the frontal ERP is further supported by our finding that BF electrical stimulation triggers the same layer profile of oddball LFP responses within 5–10 msec ( Figure 5 ) , the same temporal delay found in the correlation analysis ( Figure 3 ) . Given the complex anatomical connection network associated with the BF region ( Semba , 2000; Zaborszky , 2002; Jones , 2003 ) , the goal of this study was not to fully dissect the causal contributions of various connection pathways . Rather , our goal was to test the specific hypothesis that the BF bursting activity alone , through the prominent projection to the frontal cortex , is sufficient to trigger and account for the oddball ERP response . While our results do not rule out the possible contribution of common inputs , our findings support the conclusion that the BF bursting activity alone is sufficient to trigger and account for most of the frontal ERP in the auditory oddball task . Two key features of the coupling between BF bursting and the frontal ERP—the short temporal delay and the LFP layer profile concentrating in deep cortical layers—are consistent with the specific properties of BF projections to the frontal cortex . The average conduction latency of BF neurons to the frontal cortex , measured by antidromic stimulation , is less than 5 msec ( Aston-Jones et al . , 1985; Reiner et al . , 1987 ) . Moreover , the fast-conducting population of cortically-projecting BF neurons ( latencies 1–4 msec ) are most easily activated ( with lowest activation threshold ) when the stimulation electrode is placed in deep cortical layers of the frontal cortex ( Aston-Jones et al . , 1985; Reiner et al . , 1987 ) . Deep cortical layers in the frontal cortex ( layer V–VI ) are also the predominant target layers of cortical projections from the BF , which contain more than 70% of axonal varicosities from the BF ( Henny and Jones , 2008 ) . Among the BF axonal varicosities , the most abundant ( ∼50% ) is the large GABAergic terminal , while cholinergic terminals from the BF are smaller and fewer ( 15–20% ) ( Freund and Meskenaite , 1992; Henny and Jones , 2008 ) . These specific properties of BF projection to the frontal cortex support our hypothesis that the frontal ERP is generated by BF inputs to deep cortical layers through a fast circuit mechanism . A likely candidate to mediate the fast circuit mechanism is the non-cholinergic BF inputs because of their anatomical abundance ( Freund and Meskenaite , 1992; Gritti et al . , 1997; Henny and Jones , 2008 ) and their fast ionotropic actions . While studies of the BF have traditionally focused on its cholinergic neurons ( Everitt and Robbins , 1997 ) , the majority of BF corticopetal projections are non-cholinergic neurons , consisting mostly of GABAergic neurons and a smaller subset of glutamatergic neurons ( Gritti et al . , 1997; Henny and Jones , 2008 ) . The GABAergic BF neurons in particular preferentially innervate intracortical GABAergic interneurons and therefore may rapidly enhance cortical activity through disinhibition ( Freund and Meskenaite , 1992; Henny and Jones , 2008 ) . Consistent with such a suggestion , salience-encoding BF bursting neurons represent a physiologically homogeneous group of non-cholinergic BF neurons , which , unlike cholinergic BF neurons , do not change their mean firing rates between awake and sleep states ( Lin et al . , 2006; Lin and Nicolelis , 2008 ) . The widespread spatial distribution of bursting neurons throughout multiple brain regions of the BF ( Figure 1—figure supplement 2 ) is also consistent with the location of cortically-projecting BF neurons ( Aston-Jones et al . , 1985; Reiner et al . , 1987; Gritti et al . , 1997 ) . Therefore , our results highlight the potential functional significance of the poorly understood non-cholinergic BF neurons ( Sarter and Bruno , 2002; Lau and Salzman , 2008 ) , and the importance in determining the neurochemical identity of salience-encoding BF bursting neurons . Future studies are needed to elucidate the precise nature of this ERP response , and test whether the frontal ERP reflects , as a result of BF modulation , the synchronous activation of pyramidal neuronal ensembles in the frontal cortex as commonly assumed ( Snyder , 1991; Miltner et al . , 1994; Pascual-Marqui et al . , 2002; Luck , 2005b; Nunez and Srinivasan , 2006; Riera et al . , 2012 ) , or , alternatively , reflects the direct postsynaptic response of cortical neurons to non-cholinergic BF inputs . The single trial coupling between BF bursting activity and the frontal ERP in the current study suggests that both neural signals are functionally homologous and reflect the motivational salience of attended stimuli ( Lin and Nicolelis , 2008; Avila and Lin , 2014 ) . In support of this view , behaviorally irrelevant stimuli presented passively to the animal elicit little response in both the BF neurons ( Lin and Nicolelis , 2008 ) and the frontal ERP ( Figure 1C ) . This observation indicates that both neural signals do not reflect passive sensory processing . Instead , the BF bursting activity , and likely the frontal ERP response , reflects motivational salience irrespective of sensory modality or stimulus identity ( Lin and Nicolelis , 2008; Avila and Lin , 2014 ) . This motivational salience signal is highly relevant for behavioral performance , which predicts successful detection of a near-threshold sound ( Lin and Nicolelis , 2008 ) as well as performance in the oddball task ( Figure 1E ) . It is important to point out that the relationship between BF activity and behavioral performance is more complex and involves more than the phasic bursting response that encodes motivational salience . In behavioral contexts that require inhibition , such as Nogo trials in the Go/Nogo task , BF neurons display an initial bursting response followed by a subsequent sustained inhibition , with the latter response better correlated with behavior ( Lin and Nicolelis , 2008 ) . Future studies will need to determine the contribution of BF inhibitory responses on behavioral performance when animals need to stop or cancel a behavioral response instead of making a response to motivationally salient stimuli . It will also be important to determine whether the BF inhibitory response leads to a corresponding frontal ERP signature . Based on these observations , we propose that the main function of non-cholinergic BF neurons is to serve as a signal amplifying mechanism for motivationally salient stimuli . This amplification mechanism is needed because most sensory stimuli animals and humans encounter are behaviorally irrelevant and not processed further by the brain . For the subset of stimuli that are motivationally salient and predict important outcomes , the brain therefore requires an additional system to amplify its processing based on the motivational , but not perceptual , salience of the stimulus . The non-cholinergic BF neurons represent an ideal candidate for this amplification mechanism , which not only needs to encode the motivational salience of the attended stimulus but also needs to powerfully modulate and amplify cortical activity as fast as possible . The current study supports this hypothesis by demonstrating that the motivational salience information encoded by BF bursting activity is transformed into the frontal ERP response with a minimum delay , which likely provides powerful modulation and amplification on the early stages of information processing . Future studies need to determine precisely how task-related single neuron activity in the frontal cortex is modulated by BF inputs and the resulting frontal ERP . Ultimately , this proposed amplification mechanism should lead to faster and better decision making . In support of this prediction , we recently showed that stronger BF motivational salience signal is quantitatively coupled with faster and more precise decision speed ( Avila and Lin , 2014 ) . The findings in this study likely can be extended to primates because BF neurons with bursting responses to motivationally salient stimuli are similarly found in monkeys ( DeLong , 1971; Wilson and Rolls , 1990; Richardson and Delong , 1991 ) , and BF electrical stimulation in monkeys similarly triggers LFP responses in the frontal cortex ( Richardson and Fetz , 2012 ) . Despite the many differences in rodent and human ERP studies ( such as species , the use of primary reward and punishment in rodents but not in humans , the choice of electrical reference , etc ) , the major ERP components are broadly similar in rodents and humans ( Yamaguchi et al . , 1993; Ehlers et al . , 1994; Shinba , 1997; Sambeth et al . , 2003 ) . The rodent frontal ERP described in this study is most similar to the processing negativity or Nd ERP component in humans ( Hansen and Hillyard , 1980; Näätänen , 1982; Näätänen et al . , 1987 ) because both the rodent frontal ERP and the human processing negativity responses share several features—negative polarity of the ERP , early timing in the N1 ERP window , spatial distribution centered on the frontal cortex , as well as reflecting motivational salience or selective attention toward the stimulus . Further studies are needed to establish the functional homology between the rodent and human frontal ERP components , and determine the potential role of non-cholinergic BF neurons in selective attention in humans . An important implication of this study is that rodent ERPs may serve as a unique translational platform to bridge the human ERP literature and the neurophysiology literature in animal models . Studies in rodent models may provide unique mechanistic insights that can inform human ERP studies . For example , localizing the sources of human ERP responses , that is solving the ‘inverse problem’ , requires assumptions about the number and spatial distribution of underlying sources ( Pascual-Marqui et al . , 2002; Luck , 2005b ) . Our results suggest that , at least in the case of the frontal ERP component , the underlying source may reflect the anatomical projection pattern of the salience-encoding BF neurons , which project extensively to broad regions of the cerebral cortex ( Gritti et al . , 1997; Zaborszky , 2002; Henny and Jones , 2008; Chandler et al . , 2013 ) . Our results also raise the novel hypothesis that the decline of frontal ERP amplitude and associated cognitive functions in conditions such as schizophrenia , ADHD , Alzheimer’s disease and cognitive aging ( Schreiber et al . , 1992; Iragui et al . , 1993; Polich , 1997; Barry et al . , 2003; Caravaglios et al . , 2008 ) may result from the functional impairment of the poorly understood non-cholinergic basal forebrain system or a dysfunctional BF-cortical interaction . These results also suggest that the increasing use of deep brain stimulation of the BF to ameliorate dementia ( Hescham et al . , 2013; Salma et al . , 2014 ) may be mediated in part by activating non-cholinergic BF neurons and perhaps compensating for their functional impairment in these conditions . Eight male Long Evans rats ( Charles River , NC ) , aged 3–6 months and weighing 300–400 grams at the start of the experiment , were used for this experiment . Rats were housed under 12/12 day/night cycle with ad libitum access to rodent chow and water in environmentally controlled conditions . During training and recording procedures , rats were mildly water restricted to their 90% weight and were trained in a daily session of 60–90 min in length , five days a week . Rats received 15 min water access at the end of each training day with free access on weekends . All experimental procedures were conducted in accordance with the National Institutes of Health ( NIH ) Guide for Care and Use of Laboratory Animals and approved by the National Institute on Aging Animal Care and Use Committee . 12 plexiglass operant chambers ( 11′L × 8 ¼ ′W × 13′H ) , custom-built by Med Associates Inc ( St . Albans , VT ) were contained in sound-attenuating cubicles ( ENV-018MD ) each with an exhaust fan that helped mask external noise . Each chamber was equipped with one photo-beam lickometer reward port ( CT-ENV-251L-P ) located in the center of the front panel , with its sipper tube 7 . 5 cm above the grid floor . Two infrared ( IR ) sensors were positioned to detect reward port entry and sipper tube licking , respectively . Water reward was delivered through a custom-built multi-barrel sipper tube . The delivery system was controlled by pressurized air ( 2 . 6 psi ) and each solenoid opening ( 10 msec ) was calibrated to deliver a 10 μl drop of water . The reward port was flanked by two nosepoke ports ( ENV-114M ) , located 6 . 6 cm to each side and 5 . 9 cm above the grid floor . The nosepoke ports were inactive in the behavioral tasks . Each chamber was equipped with two ceiling-mounted speakers ( ENV-224BM ) to deliver auditory stimuli controlled by an audio generator ( ANL-926 ) , and two stimulus lights ( ENV-221 ) positioned above the reward port in the front panel . Behavior training protocols were controlled by Med-PC software ( Version IV ) , which stored all event timestamps at 2 msec resolution and also sent out TTL signals to neurophysiology recording systems to register event timestamps . Rats were initially trained in a simple tone-reward association task , in which a 6 kHz tone ( 0 . 5 s long , 70 dB ) signaled the availability of reward if responded within 3 s . Rats were rewarded with 3–5 drops of water reward starting at the 3rd lick of the sipper tube . The inter-trial intervals ( ITI ) were randomly drawn from 5 , 8 , 11 , or 14 s . If the rat licked outside the 3 s reward window ( false alarm ) , the ITI timer was reset . After completing more than 180 trials in a 60 min session , rats were transitioned to the auditory oddball task . The auditory oddball task contains infrequent-rewarded ( oddball , 6 kHz ) as well as frequent-unrewarded ( standard , 10 kHz ) tones , delivered through the same speaker . The time between stimulus presentations was 2 s , and the number of standards that were presented in between oddball tones was uniformly drawn from 2 , 3 , 4 , 5 , and 6 , corresponding to 6–14 s ITI between oddball tones . When the oddball tone was presented , the rules for receiving reward were the same as that for the tone-reward association task . False alarms during the oddball task reset the ITI timer . Correct behavioral response to the oddball tone ( hit ) led to reward delivery , as well as a temporary cessation of any tone presentation and the ITI timer until the end of reward consumption . In the passive oddball task ( Figure 1C ) , the same stimuli were presented to rats that were not water deprived , and the access to reward port was physically blocked to prevent responding . After reaching asymptotic behavioral performance , rats were taken off water restriction for at least 3 days before undergoing stereotaxic surgery for chronic electrode implant . Rats were anesthetized with isoflurane ( 4% isoflurane induction followed with 1–2% maintenance ) and received atropine ( 0 . 02–0 . 05 mg/kg , i . m . ) to reduce respiratory secretion . Ophthalmic ointment was applied to prevent corneal dehydration . A heating pad was used to maintain body temperature at 37°C . Rats were placed in a stereotaxic frame ( David Kopf Instrument , CA ) fitted with atraumatic earbars . After the skull was exposed , up to three types of probes were chronically implanted in one animal: ( 1 ) EEG skull screws were implanted in contact with the dura over the frontal cortex ( AP 3 . 0 mm , ML 3 . 0 mm relative to Bregma [Paxinos and Watson , 2007] ) and the primary visual cortex ( AP–7 . 0 mm , ML 4 . 5 mm ) ; ( 2 ) a NeuroNexus linear probe ( A1-style , 100 μm spacing , 32-channel ) was slowly lowered into the frontal cortex ( AP 3 . 0–4 . 0 mm , ML–3 . 0 mm ) with the target depth at 4–6 . 5 mm below cortical surface; ( 3 ) A custom-built 32-wire multi-electrode moveable bundle was implanted into bilateral BF ( AP–0 . 5 mm , ML +/− 2 . 25 mm , DV 7 mm below cortical surface ) . The electrode consisted of two moveable bundles , each containing 16 polyimide-insulated tungsten wires ( California Fine Wire ) ensheathed in a 28-gauge stainless steel cannula and controlled by a precision microdrive . Eight of the wires in a bundle were 38 µm in diameter and the other eight were 16 µm diameter , with 0 . 1–0 . 3 MΩ impedance measured at 1 kHz ( niPOD , NeuroNexusTech , MI ) . During surgery , the cannulae were lowered to DV 6–6 . 3 mm below cortical surface using a micropositioner ( Model 2662 , David Kopf Instrument ) , and the electrodes were advanced to 7 mm below cortical surface . Two of the 32 channels were replaced by silver wire pigtails that were connected with skull screws during surgery to record EEG signals . A common ground screw and a separate reference screw were placed over the right cerebellum ( AP–10 mm , 3 . 0 mm ) and left cerebellum ( AP–10 mm , −3 . 0 mm ) , respectively . The electrode and screws were covered with dental cement ( Hygenic Denture Resin ) . Rats received acetaminophen ( 300 mg/kg , oral ) and topical antibiotics after surgery for pain relief and prevention of infection . Water restriction and behavioral training resumed 7–10 days after surgery . Cannulae and electrode tip locations were verified with cresyl violet staining of histological sections at the end of the experiment . The tips of the linear probes in the frontal cortex are indicated by arrows in Figure 4—figure supplement 1 and Figure 4—figure supplement 2 . BF electrodes were located between −0 . 12 mm–−0 . 84 mm relative to Bregma , as shown in Figure 1—figure supplement 2A . Electrical signals were referenced to a common skull screw placed over the cerebellum . Electrical signals were filtered ( 0 . 03 Hz—7 . 5 kHz ) and amplified using Brighton Omnetics or Cereplex M digital headstages and recorded using a Neural Signal Processor ( Blackrock Microsystems , UT ) . EEG and LFP signals were continuously digitized and saved to disk at a rate of 1 kHz or 2 kHz ( band-pass filtering between 0 . 7–500 Hz ) . In general , 32 channels were devoted to laminar ( Neuronexus probe ) LFP recordings , 2 channels for skull EEG recording , and 30 channels for LFP & single unit recording in the BF region . Spike waveforms were further filtered ( 250 Hz–5 kHz ) and digitized at 30 kHz and stored to disk only when the waveform exceeded a user-defined amplitude threshold . Spike waveforms were sorted offline using OfflineSorter ( Plexon Inc , TX ) . Only single units with clear separation from the noise cluster and with minimal ( <0 . 1% ) spike collisions were used for further analyses . Additional cross correlation analysis was used to remove duplicate units recorded simultaneously over multiple electrodes . When multiple sessions from the same animal were used , BF electrodes were advanced at least 125 µm in between sessions to sample from distinct BF single neuron ensembles . The linear probe in the frontal cortex remained fixed across sessions . A total of 168 BF neurons was recorded simultaneously with EEG or LFP activity in the frontal cortex . Table 1 provides detailed information about the types of neural signals recorded in each animal , and how the data were used for different analyses in each figure . Signal detection theory ( Green and Swets , 1966 ) was used to quantify neuronal discrimination of successful ( hit ) vs failed ( miss ) tone detections . Receiver Operating Characteristic ( ROC ) curves , in particular the area under curve ( AUC ) measure , was used to determine how neuronal responses in hit trials and miss trials differed . The non-parametric AUC measure , referred to as choice probability in Figure 1E and Figure 1—figure supplement 1 , quantified the difference of the average EEG activity within a 100 msec window between hit trials vs miss trials ( or between oddball vs standard trials ) . A choice probability of 0 . 5 represents a complete overlap between the two distributions , while a choice probability of 1 ( or 0 ) indicates complete non-overlap between the two distributions , with one distribution having larger ( or smaller ) values . The statistical significance of the choice probability was established by comparing against a null distribution of choice probability values generated by randomly permuting the identity of hit and miss trials 1000 times ( p<0 . 01 , two-sided ) . The choice probability ( and its statistical significance ) was generated for each EEG signal , at time lags between [−0 . 2 , 0 . 4] sec of tone onsets with 10 msec steps . Therefore , a significant choice probability >0 . 5 indicated that the EEG signal at that time lag ( within a 100 msec window ) was significantly more positive for hit trials compared to miss trials . Conversely , a significant choice probability <0 . 5 indicated that the EEG signal at that time lag was significantly more negative for hit trials compared to miss trials . To identify BF bursting neurons , we calculated a bursting index for each BF neuron , defined as the ratio of the bursting amplitude ( average firing rate in the [50 , 200] msec window in hit trials ) over the average firing rate of the neuron in the entire session . Based on the peri-stimulus time histograms ( PSTHs , Figure 1—figure supplement 2B ) , bursting index of 2 . 5 was chosen as the cutoff for classifying BF bursting neurons , which accounted for 57% ( 96/168 ) BF neurons . BF bursting neurons have highly homogeneous baseline firing rates ( 3 . 77 ± 1 . 48 spikes/sec , mean ± std , maximum 6 . 36 spikes/sec ) , while other BF neurons have heterogeneous baseline firing rates ( 5 . 34 ± 6 . 25 spikes/sec , mean ± std , maximum 31 . 4 spikes/sec ) . To determine the single trial amplitude correlation between the activity of individual BF neurons and the amplitude of EEG/LFP signals , we calculated the number of spikes for each BF neuron and the average amplitude of EEG/LFP activity in the [50 , 200] msec window following the onset of all oddball and standard tones . The statistical significance of BF-ERP amplitude correlation was determined using Pearson correlation ( p<0 . 001 ) . R- and p-values for all BF neurons are shown in Figure 2D . For the purpose of visualizing the linear amplitude scaling relationship between BF bursting and EEG/LFP ( Figure 2B , C and 4D ) , all trials were first binned into five quintiles based on the population bursting amplitude of all BF bursting neurons recorded simultaneously in a session . The mean amplitude of BF bursting and EEG/LFP in the five quintiles were then normalized by their respective average amplitudes in hit trials , such that the average hit response amplitude corresponded to 100% . To determine the fine temporal relationship between BF bursting response and the frontal ERP at msec temporal resolution ( Figure 3 , Figure 3—figure supplement 1 ) , we calculated cross correlations between the activity of individual BF bursting neurons and the frontal EEG signal at the [0 , 170] msec window relative to tone onsets . To focus the analysis on BF bursting , the [0 , 170] msec window was chosen to center on the BF bursting response . Only trials ( both oddball and standard ) with BF population bursting amplitude greater than 30% of the average hit trial bursting amplitude were included in this analysis . Normalized cross correlation ( correlation coefficient ) was calculated between two concatenated vectors corresponding to BF and EEG activity ( described next ) , between lags [−50 , 50] msec at 1 msec resolution . Specifically , for each BF bursting neuron , the spike train was binned at 1 msec , and the binned activity in the [0 , 170] msec window from selected trials were concatenated into a long vector , each trial flanked by a 100 msec segment set to a fixed value corresponding to the mean firing rate of this neuron . For the EEG signal , EEG was first down sampled to 1 kHz sampling rate ( 1 msec bin ) , and the activity in the [0 , 170] msec window from selected trials were similarly concatenated in a long vector , each trial flanked by a 100 msec segment set to a fixed value corresponding to the mean of the EEG signal . The 100 msec fixed value flanking each trial ensured that the activity in different trials remains segregated in the cross-correlation analysis . Statistical significance of the correlation coefficient was established by comparing against a null distribution of correlation coefficients generated by randomly permuting the trial order of EEG activity 1000 times ( p<0 . 01 , two-sided ) . To determine the layer profile of LFP responses in association with BF bursting , we identified a 60 msec window around the peak of BF bursting response in oddball and standard trials , indicated by the red shaded areas in Figure 4A , B and Figure 4—figure supplement 1 . The LFP layer profile was defined as the mean LFP response in this 60 msec window , represented by a 32-dimension vector . The similarity index ( Figure 4C ) between LFP layer profiles in oddball and standard trials was defined as the cosine of the angle between the two 32-dimension vectors . This similarity index measures how well the two vectors are aligned with each other irrespective of amplitude . The maximum value of this similarity index is 1 when the angle is 0° , and the minimum value is −1 when the angle is 180° . Statistical significance of the similarity index was established by comparing against a null distribution of similarity index values generated by randomly permuting the 32 dimensions of the LFP layer profile vector 1000 times ( p<0 . 01 , two-sided ) . Since the LFP layer profiles in oddball and standard trials were statistically similar in all sessions ( 9/9 ) ( Figure 4C ) , the overall LFP layer profile for each session was generated by averaging all trials , including both oddball and standard trials . The amplitudes of the overall LFP layer profile were normalized to its peak positive value among all cortical layers . As a result , the prominent positive LFP responses restricted in deep cortical layers of the frontal cortex have values of up to 1 ( Figure 4E , Figure 4—figure supplement 1 , Figure 4—figure supplement 2 ) . To assess whether BF activity was sufficient to generate the frontal ERP response , we used BF electrical stimulation to mimic BF ensemble bursting activity with the goal of recreating the major features of the layer-specific LFP responses in the oddball task . Five rats were used for BF electrical stimulation over nine sessions . BF electrical stimulation was delivered while rats were awake but not performing any behavioral task ( n = 5 ) or under isoflurane anesthesia ( n = 4 ) ( Table 1 ) . BF electrical stimulation was conducted in separate sessions after BF single unit activity in the oddball task has been recorded . BF stimulation was delivered through all BF electrodes used in the recording experiment . This was intended to mimic the widespread presence of BF bursting neurons throughout the recording region , representing an ensemble-bursting event of the entire population ( Lin et al . , 2006; Lin and Nicolelis , 2008; Avila and Lin , 2014 ) . Furthermore , individual BF neurons tend to project to multiple subregions in the frontal cortex , unlike single neurons in other neuromodulatory systems which tend to project to one single subregion in the frontal cortex ( Chandler et al . , 2013 ) . This result suggests that the activation of any subset of cortically-projecting BF neurons by electrical stimulation should provide similar modulation of the entire frontal cortex , irrespective of the exact location of BF electrical stimulation electrodes within this region . One pulse of stimulation was delivered every 2 s , without any accompanying auditory stimuli . Individual stimulation pulse was a biphasic charge-balanced pulse ( 0 . 1 ms each phase ) delivered through a constant current stimulator ( stimulus isolator A365R , World Precision Instruments , FL ) , driven by a Master-8-VP stimulator ( AMPI , Israel ) . Currents were flowing through all BF electrodes in one hemisphere against all BF electrodes in the other hemisphere , and never through EEG skull screws , or through ground or reference skull screws . Stimulation current level was set at 62–71 µA per electrode , resulting in a total of 1 mA over all electrodes . LFP signals from the frontal cortex were recorded using a unit gain analog headstage ( Blackrock Microsystem ) , sampled at 2 kHz with 0 . 7–500 Hz filter . When BF electrical stimulation was delivered under isoflurane anesthesia , rats were first anesthetized with 3% isoflurane for at least 5 min before the start of the experiment , and maintained at 2% isoflurane under a nose cone . The body of the rat was wrapped in an aluminum foil shield lined with paper towels such that the body did not make contact with any metal . To assess the similarity of the LFP responses elicited by BF stimulation with the LFP layer profile in the oddball task , we projected the 32-dimension LFP layer vector elicited by BF stimulation onto the 32-dimension LFP layer profile in the oddball task at each time point , and calculated the length of the vector projection ( scalar projection ) . The amplitude of the scalar projection was then normalized by its peak amplitude ( excluding the 5 msec around electrical stimulation to avoid stimulation artifact ) , which we referred to as normalized similarity index ( Figure 5 , Figure 4—figure supplement 1 , Figure 4—figure supplement 2 ) . The normalized similarity index takes into account both the angle between the two LFP layer vectors and also the amplitude of the LFP responses elicited by BF electrical stimulation . Statistical significance of the normalized similarity index was established by comparing against a null distribution of scalar projections generated by randomly permuting the 32 dimensions of the LFP layer profile vector 1000 times ( p<0 . 01 , two-sided ) .

The vertebrate nervous system coordinates an animal’s involuntary and voluntary actions , and is responsible for transmitting signals between different parts of the body . Two different cell types , glial cells and neurons , make up the nervous system: glial cells play a metabolic or structural role , whereas neurons are responsible for physically transmitting the signals . Signals are sent quickly and precisely between neurons in the form of electrochemical waves , and it is this electrical activity that allows researchers to measure and study the communication of signals throughout the body . At the center of the nervous system is the brain . The brain receives and integrates information from all parts of the body , and coordinates appropriate responses to various stimuli . The outermost layer of the brain , which is known as the cerebral cortex , enables an organism to perceive and interact with the world in meaningful ways , and is also responsible for the body’s movement , as well as for thought and cognition . Composed of an estimated 30 billion neurons , the cerebral cortex generates a tremendous amount of electrical activity during sensory , motor or cognitive events . Researchers interested in evaluating brain function or assessing the level of activity in the cerebral cortex often use a non-invasive technique called electroencephalography . This technique has provided insight into the regions of the brain that process sensory and motor events , but it has been difficult to work out where cognitive events are processed . To date , most studies have tried to identify the region of the cerebral cortex that is responsible for generating the electrical activity associated with cognitive events , ignoring the possibility that other regions of the brain could play a significant role in producing this activity that is observed in the cerebral cortex . Now Nguyen and Lin have shown that the basal forebrain , a region of the brain located beneath the cerebral cortex , is responsible for generating some of the electrical activity that happens during cognitive events . The experiments involved making measurements on rats as they performed a cognitive task . Nguyen and Lin found that a form of electrical activity called an event-related potential occurred in the frontal lobe of the cerebral cortex at the same time as the activity of single neurons in the basal forebrain . Stimulating neurons in the basal forebrain also resulted in an event-related potential being measured within the frontal cortex . These findings raise the possibility that the impairment of these basal forebrain neurons is involved in conditions such as Schizophrenia , ADHD and Alzheimer’s disease , and also in normal cognitive aging .

2014

A frontal cortex event-related potential driven by the basal forebrain

Transcriptome and genome data from twenty stony coral species and a selection of reference bilaterians were studied to elucidate coral evolutionary history . We identified genes that encode the proteins responsible for the precipitation and aggregation of the aragonite skeleton on which the organisms live , and revealed a network of environmental sensors that coordinate responses of the host animals to temperature , light , and pH . Furthermore , we describe a variety of stress-related pathways , including apoptotic pathways that allow the host animals to detoxify reactive oxygen and nitrogen species that are generated by their intracellular photosynthetic symbionts , and determine the fate of corals under environmental stress . Some of these genes arose through horizontal gene transfer and comprise at least 0 . 2% of the animal gene inventory . Our analysis elucidates the evolutionary strategies that have allowed symbiotic corals to adapt and thrive for hundreds of millions of years . Reef-building stony corals ( Scleractinia ) and their cnidarian ancestors have created many thousands of square kilometers of biomineralized marine habitat in shallow tropical seas since their extensive radiation in the Middle Triassic period ~240 million years ago ( Ma ) ( Veron , 1995 ) . Coral reefs provide a significant source of ecosystem-based services ( Moberg and Folke , 1999 ) that stabilize coastlines and provide habitat for an astounding variety of flora and fauna ( Connell , 1978 ) . To better understand the evolutionary strategies underpinning the evolutionary success of reef-building corals , we analyzed genomic and transcriptomic information from twenty stony corals that contain intracellular photosynthetic dinoflagellate symbionts of the genus Symbiodinium ( https://comparative . reefgenomics . org/ ) ( Figure 1 , and Figure 1—source data 1 ) . In addition , bilaterian reference gene sets and genomes from other cnidarians , ctenophores , sponges , a choanozoan , and a placozoan were integrated into our analysis . The comprehensive reference database used for our study included 501 , 991 translated protein sequences from 20 coral species , 98 , 458 proteins from five other cnidarians such as sea anemone and sea fan , and 91 , 744 proteins from seven basal marine metazoan lineages such as sponges and ctenophores . These publicly available genomic and transcriptomic data , which showed large disparities in terms of numbers of predicted protein sequences per species were ‘cleaned’ of contaminants and poor quality data with the use of stringent filters and selection criteria ( see Materials and methods ) . This procedure resulted in a reasonably comprehensive coverage of corals ( i . e . , 20 species in total , 11 robust clade species including 2 genomes , 9 complex clade species including 1 genome ) with and average of 21 , 657 protein sequences per species . Given the challenges associated with inferring conclusions based on the absence of genes ( in particular when analyzing transcriptomic data ) , our approach focused on identifying ortholog groups present in different taxonomic categories to reach conclusions about genes associated with coral specific traits . This analysis yielded a set of 2485 'root' orthologs , 613 'Non-Cnidaria' orthologs , 462 'Cnidaria' orthologs , 1436 'Anthozoa' orthologs , 1810 'Hexacorallia' orthologs , 172 'A' orthologs , 4751 'Scleractinia' orthologs , 1588 'complex coral' orthologs , and 6 , 970 'robust coral' orthologs ( available at http://comparative . reefgenomics . org/ ) . These orthologs were analyzed to address four major issues in coral evolution: 1 ) the basis of aragonite exoskeletal accretion that results in reef formation; 2 ) environmental sensing mechanisms of the cnidarian host; 3 ) evolution of the machinery necessary to accommodate the physiological risks as well as the benefits associated with the photosynthetic algal symbionts that create a hyperoxic environment when exposed to light; and 4 ) given the rich microbial flora associated with the coral holobiont ( Fernando et al . , 2015 ) , the contribution of horizontal gene transfer ( HGT ) to coral evolution . Here we examine novel insights gained in each of these key areas . 10 . 7554/eLife . 13288 . 003Figure 1 . Multigene maximum likelihood ( RAxML ) tree inferred from an alignment of 391 orthologs ( 63 , 901 aligned amino acid positions ) distributed among complete genome ( boldface taxon names ) and genomic data from 20 coral species and 12 outgroups . The PROTGAMMALGF evolutionary model was used to infer the tree with branch support estimated with 100 bootstrap replicates . Robust and complex corals are shown in brown and green text , respectively , and non-coral metazoan species are shown in blue text . DOI: http://dx . doi . org/10 . 7554/eLife . 13288 . 00310 . 7554/eLife . 13288 . 004Figure 1—source data 1 . Coral genomic data compiled in this study and their attributes . DOI: http://dx . doi . org/10 . 7554/eLife . 13288 . 004 Relying on conserved proteins as queries in BLAST searches against our genomic database , we identified major components of the coral biomineralization toolkit and reconstructed their evolutionary origins using standard phylogenetic methods ( see Material and methods ) . These results are presented in the Discussion section below and summarized in Figures 2A and 3 . We also identified major components of the ion trafficking systems in human genomes , and searched for their orthologs in corals ( Figure 2B and Figure 2—source data 1 ) . Finally , using the approach described above , we identified stress response genes in corals and other cnidarians ( listed in Supplementary file 1 ) . 10 . 7554/eLife . 13288 . 005Figure 2 . The mechanism of ( A ) coral biomineralization based on data from physiological and molecular approaches and ( B ) the major components of the human ion trafficking system that were identified in the coral genomic data ( Figure 2—source data 1 for details ) . Here , in ( A ) Biomineralization , 1 = carbonic anhydrases ( orange ) ; 2 = bicarbonate transporter ( green ) ; 3 = calcium-ATPase ( purple ) ; 4 = organic matrix proteins ( shown as protein structures ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13288 . 00510 . 7554/eLife . 13288 . 006Figure 2—source data 1 . Major components of the human ion trafficking system identified in the coral genomic data . DOI: http://dx . doi . org/10 . 7554/eLife . 13288 . 00610 . 7554/eLife . 13288 . 007Figure 2—figure supplement 1 . Bayesian consensus trees of SLC26 . Bayesian posterior probabilities are indicated when greater than 50% . For this analysis and for the trees shown in Figure 2—figure supplements 2–4 , MrBayes v3 . 1 . 2 was used with a random starting tree and the LG model of amino acid substitution . Trees were generated for 6 , 000 , 000 generations and sampled every 1000 generations with four chains to obtain the consensus tree and to determine the posterior probabilities at the internal nodes . DOI: http://dx . doi . org/10 . 7554/eLife . 13288 . 00710 . 7554/eLife . 13288 . 008Figure 2—figure supplement 2 . Bayesian consensus trees of SLC4 . Bayesian posterior probabilities ( ×100 ) are indicated when greater than 50% . DOI: http://dx . doi . org/10 . 7554/eLife . 13288 . 00810 . 7554/eLife . 13288 . 009Figure 2—figure supplement 3 . Bayesian consensus trees of Cav . Bayesian posterior probabilities ( ×100 ) are indicated when greater than 50% . DOI: http://dx . doi . org/10 . 7554/eLife . 13288 . 00910 . 7554/eLife . 13288 . 010Figure 2—figure supplement 4 . Bayesian consensus trees of coral and outgroup Ca-ATPase proteins . Bayesian posterior probabilities ( ×100 ) are indicated when greater than 50% . DOI: http://dx . doi . org/10 . 7554/eLife . 13288 . 01010 . 7554/eLife . 13288 . 011Figure 2—figure supplement 5 . Evolution of CARPs and other coral acid-rich proteins . ( A ) Maximum likelihood ( RAxML ) tree showing extensive history of duplication of genes encoding CARP 5 that predates the split of robust ( brown text ) and complex ( green text ) corals . ( B ) RAxML tree showing the origin of CARP 1 in robust ( brown text ) and complex ( green text ) corals from a reticulocalbin-like ancestor by the evolution of a novel acid-rich N-terminaldomain . The non-coral species in both trees are shown in blue text . DOI: http://dx . doi . org/10 . 7554/eLife . 13288 . 01110 . 7554/eLife . 13288 . 012Figure 2—figure supplement 6 . Scatter plot of isoelectric points of collagens from Seriatopora , Stylophora , Nematostella , and Crassostrea gigas . DOI: http://dx . doi . org/10 . 7554/eLife . 13288 . 01210 . 7554/eLife . 13288 . 013Figure 2—figure supplement 7 . Maximum likelihood ( ML ) trees of galaxin and amgalaxin . ( A ) ML tree of best galaxin hits from 19 coral species ( brown for robust corals and green for complex corals ) and 11 non-coral species ( blue text ) . ( B ) ML tree of best amgalaxin hits from 13 coral species . No outgroup blast hits were found against the acidic region of Acropora millepora amgalaxin 1 or 2 ( Genbank accession numbers ADI50284 . 1 and ADI50285 . 1 , respectively ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13288 . 01310 . 7554/eLife . 13288 . 014Figure 3 . Comparison of robust coral ( brown text ) and complex coral ( green text ) and non-coral ( blue text ) genomes with respect to percent of encoded proteins that contain either >30% or >40% negatively charged amino acid residues ( i . e . , aspartic acid [D] and glutamic acid [E] ) . The average composition and standard deviation of D + E is shown for the two cut-offs of these estimates . On average , corals contain >2-fold more acidic residues than non-corals . This acidification of the coral proteome is postulated to result from the origin of biomineralization in this lineage . DOI: http://dx . doi . org/10 . 7554/eLife . 13288 . 014 To elucidate the impact of foreign gene acquisition in coral evolution , we estimated the extent of HGT in the genomic data using a conservative phylogenomic approach ( see Materials and methods ) . This procedure was followed by localization of key HGT candidates to genomic contigs to validate their provenance ( Figure 4 ) . Using the A . digitifera and Seriatopora sp . proteomes independently as queries resulted in 13 , 256 and 19 , 700 alignments of which 21 and 41 , respectively ( i . e . , in A . digitifera , Seriatopora sp . ) , supported HGT ( 62/32 , 956 trees = 0 . 2% ) . After accounting for gene duplicates and redundancy between the trees , we discovered 41 unique instances of foreign gene acquisition from bacteria and algae ( Table 1 ) . Of these candidates , 28 genes were present in the anthozoan common ancestor ( i . e . , were shared with anemone and/or sea fan ) and 13 were specific to corals . In all cases , the HGT-derived genes were shared by multiple anthozoan species and the phylogenies of these genes were largely consistent with the reference tree shown in Figure 1 . 10 . 7554/eLife . 13288 . 015Figure 4 . Analysis of a genomic region in Acropora digitifera that encodes a putative HGT candidate . ( A ) The genome region showing the position of the HGT candidate ( PNK3P ) and its flanking genes . ( B ) Maximum likelihood trees of PNK3P ( polynucleotide kinase 3 phosphatase , pfam08645 ) domain-containing protein and the proteins ( RNA-binding and GTP-binding proteins ) encoded by the flanking genes . Robust and complex corals are shown in brown and green text , respectively , and non-coral metazoan and choanoflagellate species are shown in blue text . Photosynthetic lineages , regardless of phylogenetic origin , are shown in magenta text and all other taxa are in black text . GenBank accession ( GI ) or other identifying numbers are shown for each sequence . The PNK3P domain plays a role in the repair of DNA single-strand breaks by removing single-strand 3'-end-blocking phosphates ( Petrucco et al . , 2002 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13288 . 01510 . 7554/eLife . 13288 . 016Figure 4—figure supplement 1 . Maximum likelihood trees of a DEAD-like helicase and the protein encoded by the flanking gene . The bacterium-derived DEAD-like helicase genes in coral are nested within bacterial sequences , whereas the upstream host-derived gene ( encoding mannosyl-oligosaccharide 1 , 2-alpha-mannosidase IB ) is monophyletic with homologous genes from other metazoan species . The downstream Acropora digitifera-specific gene has no detectable homolog in other species . Robust and complex corals are shown in brown and green text , respectively , and non-coral metazoan and choanoflagellate species are shown in blue text . Photosynthetic lineages , regardless of phylogenetic origin , are shown in magenta text and all other taxa are in black text . GenBank accession ( GI ) or other identifying numbers are shown for each sequence . DOI: http://dx . doi . org/10 . 7554/eLife . 13288 . 01610 . 7554/eLife . 13288 . 017Figure 4—figure supplement 2 . Maximum likelihood tree of an exonuclease-endonucease-phosphatase ( EEP ) domain-containing protein ( A ) , an ATP-dependent endonuclease ( B ) , a tyrosyl-DNA phosphodiesterase 2-like protein ( C ) , and DNA mismatch repair ( MutS-like ) protein ( D ) . Robust and complex corals are shown in brown and green text , respectively , and non-coral metazoan and choanoflagellate species are shown in blue text . Photosynthetic lineages , regardless of phylogenetic origin , are shown in magenta text and all other taxa are in black text . GenBank accession ( GI ) or other identifying numbers are shown for each sequence . DOI: http://dx . doi . org/10 . 7554/eLife . 13288 . 01710 . 7554/eLife . 13288 . 018Figure 4—figure supplement 3 . Maximum likelihood trees of glyoxalase I ( or lactoylglutathione lyase ) and the proteins encoded by the flanking genes ( top image ) in Acropora digitifera . Robust and complex corals are shown in brown and green text , respectively , and non-coral metazoan and choanoflagellate species are shown in blue text . Photosynthetic lineages , regardless of phylogenetic origin , are shown in magenta text and all other taxa are in black text . GenBank accession ( GI ) or other identifying numbers are shown for each sequence . DOI: http://dx . doi . org/10 . 7554/eLife . 13288 . 01810 . 7554/eLife . 13288 . 019Figure 4—figure supplement 4 . Maximum likelihood tree of a second glyoxalase I ( or lactoylglutathione lyase ) and the proteins encoded by the flanking genes ( top image ) in Acropora digitifera . The coral glyoxalase gene gene was derived from a bacteria-specific gene type . Robust and complex corals are shown in brown and green text , respectively , and non-coral metazoan and choanoflagellate species are shown in blue text . Photosynthetic lineages , regardless of phylogenetic origin , are shown in magenta text and all other taxa are in black text . GenBank accession ( GI ) or other identifying numbers are shown for each sequence . DOI: http://dx . doi . org/10 . 7554/eLife . 13288 . 01910 . 7554/eLife . 13288 . 020Figure 4—figure supplement 5 . Maximum likelihood tree of an algal-derived short-chain dehydrogenase/reductase ( A ) , and a dinoflagellate-derived phosphonoacetaldehyde hydrolase ( B ) . Robust and complex corals are shown in brown and green text , respectively , and non-coral metazoan and choanoflagellate species are shown in blue text . Photosynthetic lineages , regardless of phylogenetic origin , are shown in magenta text and all other taxa are in black text . GenBank accession ( GI ) or other identifying numbers are shown for each sequence . DOI: http://dx . doi . org/10 . 7554/eLife . 13288 . 02010 . 7554/eLife . 13288 . 021Table 1 . The list of non-redundant anthozoan genes derived via HGT . DOI: http://dx . doi . org/10 . 7554/eLife . 13288 . 021No . AncestorGenesProtein productsSupportSource ( s ) 1CoralA . digitifera_2036PNK3P100CA2CoralA . digitifera_8849SDR100CA3CoralSeriatopora_31861DEAD-like helicase100Bact4CoralSeriatopora_16594glyoxalase100CA5CoralSeriatopora_17147acyl- dehydrogenase100Bact6CoralSeriatopora_17703carbonic anhydrase96Dino7CoralSeriatopora_19477fatty acid or sphingolipid desaturase100CA8CoralSeriatopora_3957atpase domain-containing protein100Bact9CoralSeriatopora_7060sam domain-containing protein100Bact10CoralSeriatopora_7928atp phosphoribosyltransferase100CA/Fungi11CoralSeriatopora_8296glyoxalase98Bact12CoralSeriatopora_225962-alkenal reductase92Bact13CoralSeriatopora_28321histidinol-phosphate aminotransferase96Unclear14AnthozoaA . digitifera_418duf718 domain protein100CA15AnthozoaA . digitifera_15871peptidase s4996Algae/Bact16AnthozoaA . digitifera_14520predicted protein100CA/Bact17AnthozoaA . digitifera_7178rok family protein/fructokinase93Red algae18AnthozoaA . digitifera_10592Phospholipid methyltransferase100CA/Viri19AnthozoaA . digitifera_13390predicted protein100Bact20AnthozoaA . digitifera_313malate synthase98CA/Bact21AnthozoaA . digitifera_1537hypothetical protein100Bact22AnthozoaA . digitifera_13577gamma-glutamyltranspeptidase 1-like100Unclear23AnthozoaA . digitifera_5099Isocitrate lyase ( ICL ) 100Bact24AnthozoaA . digitifera_13467uncharacterized iron-regulated protein100CA25AnthozoaA . digitifera_68663-dehydroquinate synthase98CA26AnthozoaA . digitifera_11675intein c-terminal splicing region protein100Bact27AnthozoaSeriatopora_10994penicillin amidase100Bact28AnthozoaSeriatopora_14009nucleoside phosphorylase-like protein100Bact29AnthozoaSeriatopora_14494phosphonoacetaldehyde hydrolase100Dino30AnthozoaSeriatopora_15303exonuclease-endonuclease-phosphatase99CA/Viri31AnthozoaSeriatopora_15772fmn-dependent nadh-azoreductase99Dino32AnthozoaSeriatopora_19888had family hydrolase97Algae/Bact33AnthozoaSeriatopora_20039chitodextrinase domain protein92Dino34AnthozoaSeriatopora_20146glutamate dehydrogenase100CA/Bact35AnthozoaSeriatopora_20479thif family protein100Bact36AnthozoaSeriatopora_21195ATP-dependent endonuclease100Dino37AnthozoaSeriatopora_8585chitodextrinase domain protein92Bact38AnthozoaSeriatopora_24047aminotransferase100Bact39AnthozoaSeriatopora_25961d-alanine ligase99Bact40AnthozoaSeriatopora_26478quercetin 3-o-methyltransferase100Viri41AnthozoaSeriatopora_29443diaminopimelate decarboxylase100CABact: Bacteria; CA: chlorophyll c-containing algae; Dino: dinoflagellates; Viri: Viridiplantae . The most obvious feature of corals over geological time is their fossilized calcium carbonate skeletons , of which the original mineral component is aragonite . It has been hypothesized for many years that the precipitation of aragonite is catalyzed by and organized on an extracellular organic matrix containing a suite of proteins , lipids , and polysaccharides ( Mann , 2001; Watanabe et al . , 2003 ) . This process is precisely controlled and occurs in the calcifying fluid lined by the ectodermal calicoblastic cells that initiate and control the precipitation reaction . Four major components are involved in the process and will be described below: a source of inorganic carbon , a source of calcium ions , proteins that catalyze the nucleation reaction , and proteins and other organic molecules that organize the crystals to form macroscopic structures ( Figure 2A ) . In this figure , only the transcellular pathway at the level of the calicoblastic cells is shown . Calcium presumably enters the cells via a calcium channel ( Zoccola et al . , 1999 ) and exits through a calcium ATPase which is proposed to remove protons from the site of calcification ( Zoccola et al . , 2004 ) . Whereas part of the dissolved inorganic carbon ( DIC ) can enter the cells via a bicarbonate transporter ( Furla et al . , 2000 ) , the major source of DIC comes from metabolic CO2 , which either diffuses out of the cells through the membranes or is intracellularly converted into HCO3-due to a favorable pH ( Venn et al . , 2009 ) , a reaction which is accelerated by carbonic anhydrases ( Bertucci et al . , 2013 ) . This bicarbonate can then exit the cells via a bicarbonate transporter ( Zoccola et al . , 2015 ) . At the site of calcification carbonic anhydrases can also play a role in the kinetics of the interconversion between carbon dioxide and bicarbonate ( Bertucci et al . , 2013 ) according to the extracellular pH ( Venn et al . , 2011 ) . The organic matrix which plays different roles in the biological precipitation of carbonates , comprises a set of proteins including CARPs ( Mass et al . , 2013; ) , collagens ( Drake et al . , 2013 ) , galaxins ( Fukuda et al . , 2003 ) , and carbonic anhydrase related proteins ( Drake et al . , 2013 ) . More broadly , inorganic carbon in seawater in the upper ocean is approximately 2 mM with 95% in the form of bicarbonate ions and is delivered to the site of calcification from an internal pool within the host animal ( Erez , 1978; Furla et al . , 2000 ) . This happens either by diffusion of CO2 or by active transport of HCO3- following CO2 hydration ( Tambutté et al . , 1996 ) . The hydration reaction is catalyzed by an intracellular carbonic anhydrase ( CA ) ( Bertucci et al . , 2013 ) . To help facilitate calcification , calicoblastic cells concentrate dissolved inorganic carbon ( DIC ) in the calcifying fluid ( Allison et al . , 2014 ) . Analysis of our genome data shows two distinct families of bicarbonate anion transporters ( BATs ) in the coral Stylophora pistillata ( Zoccola et al . , 2015 ) . Three isoforms belong to the SLC26 family ( Figure 2—figure supplement 1 ) and 5 isoforms belong to the SLC4 family ( Figure 2—figure supplement 2 ) . One isoform , SLC4γ , is restricted to scleractinians and is only expressed in the calicoblastic cells ( Zoccola et al . , 2015 ) , strongly suggesting that this protein plays a key role in calcification . This bicarbonate transporter could either supply DIC at the site of calcification , or aid in pH regulation in addition to a calcium ATPase ( see below ) . Furthermore , the two BAT gene families are split along phylogenetic lines between the robust and complex clades of scleractinians . The concentration of calcium ions in seawater is 10 mM , with these ions being actively transported by the calicoblastic cells to the calcifying fluid ( Tambutté et al . , 1996 ) . Radiocalcium ( 45Ca ) and inhibitor studies demonstrate that calcium entry in calicoblastic cells by facilitated diffusion is dependent on voltage-gated calcium channels ( Tambutté et al . , 1996 ) . Based on their alpha 1 subunit ( Cavα1 ) these channels can be phylogenetically divided into three groups . Specific inhibitors ( dihydropyridines ) strongly suggest that these channels belong to the voltage-dependent L-type family Cav 1 and have been characterized at the molecular level and localized by immunohistochemistry in the calicoblastic cells ( Zoccola et al . , 1999 ) . We constructed a phylogeny of the alpha 1 subunit of all types of Cav ( Figure 2—figure supplement 3 ) and found orthologs in most of the datasets used here , as previously shown for the actinarian Nematostella vectensis and the scleractinian Acropora millepora ( Moran and Zakon , 2014 ) . Calcium efflux from the calicoblastic cells to calcifying fluid likely occurs through a plasma membrane calcium ATPase ( Ca-ATPase ) ( Zoccola et al . , 2004 ) . This enzyme is also responsible for removing protons and increasing pH in the calcifying fluid in order to increase the aragonite saturation state to promote calcification ( Zoccola et al . , 2004; Venn et al . , 2011; Davy et al . , 2012 ) . For this enzyme ( Figure 2—figure supplement 4 ) as well as for calcium channels and bicarbonate transporters , there is a division between the robust and complex clades of scleractinians . As described in the two previous paragraphs , for the analysis of the source of inorganic carbon and calcium transport for biomineralization , we focused on the molecules which were previously characterized both by pharmacological and physiological studies in order to link molecular characterization to function . Our data clearly show that transporters such as calcium channels and calcium ATPases and some bicarbonate transporter isoforms are ubiquitously present in the calcifying and non-calcifying cnidarians ( scleractinian corals and sea anemones ) . Based on the genomic analysis of bicarbonate transporters families in two scleractinian corals and one sea anemone , Zoccola et al . ( 2015 ) observed that one isoform of the bicarbonate transporter family SLC4γ , was restricted to scleractinians . The current transcriptomic analysis of calcifying and non-calcifying cnidarian species confirms this result , which underlines the role of this transporter in biomineralization . Additional studies are however needed to localize this transporter in different coral species and to determine whether , as for S . pistillata , it is also specifically expressed in the calicoblastic cells . Another important piece of information is that for all the different enzymes and transporters studied , there is generally a division in the phylogenetic tree between the robust and the complex clades of scleractinian corals . This suggests that the different calcification traits observed for the two clades ( for example , complex corals have less heavily calcified skeletons than robust corals ) , are due to differences in the biochemical characteristics of these enzymes and transporters . The skeletal structure of corals contains an embedded organic matrix with a set of proteins that have a high proportion of aspartic and glutamic acids ( Mitterer , 1978; Weiner , 1979; Mann , 2001; Weiner and Dove , 2003; Gotliv et al . , 2005 ) . These coral acid-rich proteins ( CARPs ) ( Mass et al . , 2013 ) show sequence similarity across Scleractinia ( Drake et al . , 2014 ) and have functional analogs across the biomineralizing tree of life ( Gorski , 1992; Sarashina and Endo , 2001; Kawasaki et al . , 2009 ) . CARPs contain >28% aspartic and glutamic acids and have isoelectric points less than pH 5 ( Table 1 in Mass et al . , 2013 ) . Each of these proteins can individually catalyze the precipitation of calcium carbonate in vitro in natural seawater ( Mass et al . , 2013 ) , hence , they appear to be responsible for initiating biomineralization . Our results show that the average composition of aspartic and glutamic acids in scleractinian corals is >2-fold higher than in 12 non-calcifying invertebrates , with no obvious difference between the robust and complex clades of scleractinians ( Figure 3 ) . Moreover , phylogenetic analysis reveals that four CARP genes ( CARPs 2–5 ) are widely distributed among scleractinians , suggesting they are derived from homologs present in non-calcifying anthozoans . Extensive duplication of genes encoding CARPs predated the split of robust and complex corals can be seen for CARPs 3–5 ( Figure 2—figure supplement 5 ) , whereas CARP 2 appears to be unique to robust corals . A previous hypothesis that CARP 1 resulted from a gene ( domain ) fusion ( Mass et al . , 2013 ) is supported by these extensive genome data . CARP 1 is derived from a reticulocalbin-like gene present in all metazoans that underwent the fusion of an acidic N-terminal domain , resulting in a modular gene that is found only in corals ( Figure 2—figure supplement 5 ) . Our data suggest that the enrichment of highly negatively charged proteins is a major distinguishing feature of stony corals . At the nanoscale , the biological precipitation of aragonite crystals is insufficient to form the highly organized , stable macrostructures that characterize corals . The crystals are organized by a series of proteins that act as ‘glues’ . One of these protein families , found in the skeletons of corals is collagen ( Jackson et al . , 2010; Drake et al . , 2013 ) . In basal invertebrates , there are three families of collagen ( fibrillar , multiplexins , and type IV ) that are also present in vertebrates . Other than their structural function , collagens play an important role in the regulation of cell-cell adhesion , differentiation , and wound healing ( Heino et al . , 2009 ) . Collagens in the alpha IV subfamily have been identified in the organic matrix of coral skeletons ( Ramos-Silva et al . , 2013; Drake et al . , 2013 ) . Alpha IV collagens form networks of fibers that are an important component of the extracellular matrix . Using the complete genome data from S . pistillata and Seriatopora sp . , we identified 230 and 208 predicted open reading frames ( ORFs ) , respectively , that contained a collagen Pfam domain . Of these , 52 S . pistillata proteins contain an extracellular secretion signal , in comparison to 17 from Seriatopora sp . By plotting the isoelectric point ( IP ) of the secreted collagens from both corals we identified four acid-rich collagens in Seriatopora sp . and five in S . pistillata that have an IP < 7 ( Figure 2—figure supplement 6 ) . This analysis strongly suggests that these collagens play a critical role in tethering aragonite crystals in coral skeletons similar to their role in bone formation ( Nudelman et al . , 2010 ) . In addition to collagens , stony corals secrete a variety of other adhesion proteins into the calcifying milieu ( Ramos-Silva et al . , 2013; Drake et al . , 2013 ) . These include cadherins , which facilitate cell-cell or cell-substrate adhesion , vitellogenin , and zonadhesin proteins . As part of the biomineralization toolkit , these proteins bind the calicoblastic cells to the newly formed skeleton and may assist in the binding of CARPs to other functional proteins . Interestingly , the first protein sequenced from coral skeleton , galaxin , is neither acidic nor calcium binding , and its function remains unknown ( Fukuda et al . , 2003 ) . Originally sequenced from Galaxea fascicularis , but more recently identified in the A . millepora skeleton , galaxin is a 30–40 kDa glycosylated protein with a signal peptide , suggesting it is secreted ( Fukuda et al . , 2003; Ramos-Silva et al . , 2013 ) . The primary sequence contains ~20 paired cysteine ( CC ) residues . Usherin , found in vertebrates has a similar high number of paired cysteine motifs ( Baux et al . , 2007 ) and binds type IV collagens ( Bhattacharya et al . , 2004 ) , suggesting a potential role for this galaxin . Galaxin was originally suggested to be coral-specific ( Fukuda et al . , 2003 ) , however , galaxin-like proteins are found in non-calcifying taxa outside Cnidaria ( e . g . , Sanchez et al . , 2007; Heath-Heckman et al . , 2014 ) . Therefore , it has been proposed that the precursor to modern coral galaxin homologs was recruited as a biomineralization protein when Scleractinia diverged from non-biomineralizing taxa during the Triassic ( Foret et al . , 2010 ) . Our sequence analysis supports this hypothesis , suggesting that not only is coral galaxin derived from a common ancestor with non-calcifying metazoans , but that it is polyphyletic within corals ( Figure 2—figure supplement 7 ) , and independently recruited for a role in biomineralization multiple times in coral evolution . The first evidence for stony corals occurs in the Triassic and fossil evidence shows a rapid proliferation of taxa ( reviewed by Stanley , 2003 ) ; this was also a time of ‘aragonite seas’ when geochemical conditions were favorable to the formation and evolution of aragonitic coral skeletons ( Stanley and Hardie , 1998 ) . A second type of galaxin , amgalaxin , has an N-terminal acidic domain that precedes the galaxin domain ( Reyes-Bermudez et al . , 2009 ) . However , unlike galaxin , amgalaxin appears to function only in the early larval stages of biomineralization and has not been observed in the coral skeleton ( Reyes-Bermudez et al . , 2009; Ramos-Silva et al . , 2013 ) . This pattern is similar to the mollusk and coral proteins nacrein and CARP1 ( see above ) , in which an acidic domain is fused to an existing gene ( Miyamoto et al . , 1996; Mass et al . , 2013 ) . Unlike galaxin , the acidic portion of amgalaxin appears to be limited to corals ( Figure 2—figure supplement 7 ) . This result suggests that the attachment of an acidic region to galaxin is unique to stony corals and that amgalaxin , like CARP1 , emerged from a gene fusion event . Corals typically produce planktonic or ‘crawl-away larvae’ that calcify when they settle on an appropriate benthic substrate , and have thereby effectively determined their future physical environment for the life of the organism . Hence , habitat selection is one of the most critical elements in the survival and success of individual corals . To help accommodate variations in habitat on time scales varying from hours to years , corals have evolved a suite of environmental sensing and response systems . One of the most critical environmental cues for coral success is light ( Dubinsky and Falkowski , 2011 ) . Stony corals use diel periodicity and light sensing capabilities as cues for spawning , feeding , and orienting the polyps . Perhaps not surprisingly , the host animal has genes encoding a circadian clock . However , the light sensing signal cascades in zooxanthellate corals are particularly complex because of their symbiotic relationship with dinoflagellates , which also have a circadian clock . Coral environmental response genes are coupled to the dinoflagellate circadian clock , anticipating changes in the intracellular milieu such as the coral tissue becoming hyperoxic due to zooxanthellate photosynthesis and near-hypoxic at night due to host and symbiont respiration . Numerous chaperones such as heat shock protein ( hsp ) 40 , hsp70 , hsp90 , grp94 , hsp90b1 , calreticulin , and protein disulfide isomerase are ‘hard-wired’ to this photosynthesis/respiration clock and the high level of synchrony of circadian transcription of chaperones and antioxidant genes reflects the diurnal preparedness of the coral to the consequences of oxidative protein damage imposed by photosynthesis of the algal symbionts ( Levy et al . , 2011 ) . Symbiosis also indirectly imposes diurnal gene expression fluctuations , most likely via the hypoxia inducible factor ( HIF ) system . In a wide array of animals , glycolytic enzymes are regulated by HIF1-alpha transcription factor , a clear ortholog of which is present in the 20 coral genomic datasets . The HIF system is unique to animals , and HIF itself is a target of calpain-mediated degradation in vertebrates . Calpains are Ca2+-dependent regulatory proteases and corals linkage of calpain expression to the HIF system potentially enables them to utilize cellular calcium levels to modulate expression of other HIF targets when hypoxia dominates ( Levy et al . , 2011 ) . The casein kinase I ( CK1 ) family consists of serine/threonine protein kinases that are key regulators of circadian timing in bilaterian animals , fungi , and green algae ( van Ooijen et al . , 2013 ) . CK1-like encoding genes are found in most corals and were suggested to be components of the coral circadian gene network along with CLOCK , GSK3B/Sgg , and CSNK1D ( Vize , 2009 ) . The proteins ADCI , GNAQ , GNAS , GNB1 , CREB1 , and NOS1 are related to G-protein coupled receptor signaling and can act on neuropeptide/GPCR-coupled signaling mechanisms . This is consistent with neurohormones playing a role in synchronized spawning events in tropical abalone ( York et al . , 2012 ) and in coral larvae settlement ( Grasso et al . , 2011 ) . Other proteins such as PPEF1 and GRIN1 respond to light stimulus , whereas MTNR1A and MTNR1B are melatonin receptors , whereas PRKAA2 is a protein kinase that responds to peptide hormone stimulus and is responsive to circadian rhythms . The circadian processes are impacted by catabolic process; i . e . , S . pistillata glycolysis is controlled by ARNT and HIF1-alpha that provide feedback that affects the circadian loop . Surprisingly , BLASTp analysis of the 20 coral genomic datasets did not turn up the Period gene as reported in other cnidarians . Therefore , the core circadian clock architecture of the negative feedback loop in basal metazoans such as corals may differ significantly from animal lineages that diverge after corals . Although fluxes of calcium and bicarbonate ions into the calicoblastic space are part of the biomineralization system , these and other ion pumps also generate electrochemical gradients that allow stony corals to sense the environment and initiate complex and specific signaling cascades ( Hille , 1986 ) . This ion trafficking landscape and downstream signaling components are comprised of channels , transporters , exchangers , pumps , second messenger generators , and transcriptional response elements . Many of these ion transporters act as direct physicochemical sensors providing intra-cellular and intra-organismal regulation and the critical linkage between external environmental changes and cytoplasmic and organellar events , cascades and transcriptional regulation . We identified major components of the ion trafficking systems in human genomes , and searched for their orthologs in corals ( Figure 2B ) . Ion channel sensors such as the transient receptor potential ( TRP ) channels ( TRPA , TRPV , TRPM , TRPC ) ( Ramsey et al . , 2006; Nilius and Szallasi , 2014 ) and acid sensing channels ( ASICs ) are present in corals ( Krishtal , 2015 ) . Most of these are either direct , or indirect , physicochemical sensors of environmental parameters such as temperature , pH and oxygen tension . Organelle ion regulators such as two-pore channels ( TPCN ) ( Wang et al . , 2012; Horton et al . , 2015 ) , mucolipin ( MCOLN ) are also present and are thought to maintain intraorganellar pH and ion environments . In summary , most , if not all of these components sense environmental changes and implement signaling cascades that lead to the activation of specific transcriptional programs that allow the organism to physiologically respond to environmental signals . Symbiotic corals thrive in oligotrophic tropical and subtropical seas in large part because their intracellular , symbiotic dinoflagellates provide a significant portion of their photosynthesis-derived fixed carbon to the host animal . However , this benefit comes with significant costs . The ecological stability of the symbiotic association is dependent on it being stable in the face of environmental extremes . This symbiosis has been widely described as living close to the upper extremes of thermal tolerance that , when exceeded , leads to a cascade of cellular events resulting in ‘coral bleaching’ , whereby corals lose their symbiotic algae and consequently one of their main sources of carbon ( Lesser , 2006; 2011 ) . Other environmental extremes can lead to coral bleaching including exposure to ultraviolet radiation ( UVR ) and ocean acidification ( Lesser , 2004; Hoegh-Guldberg et al . , 2007 ) . Proximately , in this cascade of events , many physiological studies on bleaching in corals and other symbiotic cnidarians have shown that photosynthetically produced hyperoxic conditions act synergistically with solar radiation , especially UVR , and thermal stress to produce reactive oxygen species ( ROS ) and reactive nitrogen species ( RNS ) in both host tissues and Symbiodinium sp . beyond their capacity to quench these toxic products ( Lesser , 2006; 2011 ) . Ultimately a series of fairly well described stress response events involving cell cycle arrest and apoptosis , in both the algal symbionts and host , appear to be responsible for the massive expulsion of dinoflagellates from the host , and ultimately , host mortality if the environmental insult is severe enough or of prolonged duration ( Lesser , 2006; 2011 ) Therefore , the ecological stability of the symbiotic association in zooxanthellate corals requires increased stability in the face of environmental extremes . Previous coral genomic studies have identified genes involved in the stress response of cnidarians ( Shinzato et al . , 2011 ) , but here we show that corals contain highly conserved genes involved in oxidative stress , DNA repair , the cell cycle and apoptosis ( Supplementary file 1 ) . For instance we identify both the extrinsic and intrinsic apoptotic pathways , characteristic of many vertebrates including humans . These genes are not derived by HGT in the Cnidaria , because of their presence in poriferans and other sister taxa ( see HGT discussion below ) . Corals exposed to oxidative stress , or UVR , accumulate DNA damage , whereby cell cycle arrest occurs and cell repair is initiated ( Lesser and Farrell , 2004 ) . If DNA damage is too severe , then a cellular cascade leading to genetically programmed cell death by apoptosis occurs via an intrinsic , or mitochondrial , pathway . Whereas the intrinsic pathway is considered a response to stress ( e . g . , thermal stress ) , the extrinsic , or death-receptor pathway is a cellular process by which cell to cell communication activates apoptosis via ligand binding to cell surface receptors , as in the well described immunological response to cancer cells or pathogens . Genes present in cnidarians and active in the vertebrate intrinsic DNA damage induced apoptotic pathway include: ATM , p53 ( and many of its important regulator proteins and transcriptional products ) , Hausp , Bax , Bcl-2 , AIF , cytochrome C , APAF1 , procaspase 9 , procaspase 3 , ICAD and CAD ( Supplementary file 1 ) . The activity of these genes in cnidarians comprises the cellular machinery necessary to accomplish the following: mitochondrial catastrophe , apoptosome formation , breakdown of the nuclear pores , intra-nuclear DNA disassembly and flipping of phosphotidylserine from the inner to the outer leaflet of the plasma membrane that in humans permits macrophage recognition of apoptotic cells . In addition , we identified a complete nitric oxide synthase ( NOS; EC 1 . 14 . 13 . 39 ) in corals . This gene is derived from a metazoan ancestor and is thought to play a key role in the stress response that leads to breakdown of the symbiosis and coral bleaching ( Trapido-Rosenthal et al . , 2005; Hawkins et al . , 2013 ) . Another significant finding of our analysis of multiple taxa is that Bid ( BH3; Bcl-2 domain of homology 3 ) , the only protein that allows the extrinsic and intrinsic pathways in vertebrates to directly communicate with each other , is not present in the coral data . Previous research on apoptosis in invertebrates , particularly on the intrinsic pathway , demonstrated the conserved nature of the molecular machinery in ancestral metazoans ( Bender et al . , 2012 ) . Cnidarians encode all the genes for both pathways known to be expressed and active in vertebrates , but appear to lack the ability to communicate between them . This function is mediated by p53 , the gatekeeper for cell growth and division , through Bid in vertebrates ( Sax et al . , 2002 ) that is present in 20 of 25 cnidarian datasets examined here . The antiquity of the intrinsic pathway is striking and along with the recent demonstration of a functional extrinsic pathway in cnidarians ( Quistad et al . , 2014 ) reveals the importance of these apoptotic pathways in metazoan evolution . Interestingly , tumor necrosis factor ( TNF ) , an essential mediator of the extrinsic death-receptor pathway , was present in only 7 of the 32 datasets examined in this study ( Supplementary file 1 ) . Lastly , the presence of the major genes in the human extrinsic and intrinsic pathways suggests that cnidarians may be a potential model system for studying transcriptionally induced apoptosis , when compared to Caenorhabditis elegans and Drosophila melanogaster . In these latter animal models , the available functional data indicate that genes in the cellular senescence , DNA editing , and repair pathways that are governed by the transcriptional activation domain ( TAD ) of p53 are only 2% ( D . melanogaster ) and 33% ( C . elegans ) conserved when compared to human p53 ( Walker et al . , 2011 ) . This result suggests limited control of somatic cell apoptosis in these organisms perhaps because their adult somatic cells do not divide by mitosis . The primary function of the HGT candidates we identified in stony corals is to extend the existing stress related pathways in these animals . These foreign genes encode proteins that provide protection from UVR and stress from reactive species ( Banaszak and Lesser , 2009; Nesa et al . , 2012 ) . It has already been reported that corals and sea anemones acquired a pathway that produces photo-protective mycosporine amino acids that absorbs UVR ( Shinzato et al . , 2011 ) . Our results show additions to the DNA repair pathway , including a polynucleotide kinase 3-phosphatase ( PNK3P ) of algal origin ( Figure 4 ) and a DEAD-like helicase of bacterial origin ( Figure 4—figure supplement 1 ) . These two genes are flanked by eukaryotic or coral-specific genes in their respective contigs in the draft genome of A . digitifera ( Figure 4 and Figure 4—figure supplement 1 ) . Two DNA repair genes that were transferred from algal sources were found in the anthozoan ancestor . These encode an exonuclease-endonuclease-phosphatase ( EEP ) domain-containing protein and an ATP-dependent endonuclease ( Figure 4—figure supplement 2 ) . Furthermore , two DNA repair genes are shared between Anthozoa and sponges or choanoflagellates , but are missing from a large diversity of Bilateria; these encode a tyrosyl-DNA phosphodiesterase 2-like protein and a DNA mismatch repair ( MutS-like ) protein ( Figure 4—figure supplement 2 ) . Our results fit in well with the so-called Public Goods Hypothesis that posits important genetic resources , such as mechanisms of DNA repair , are distributed widely among taxa via both vertical and horizontal evolution ( McInerney et al . , 2011 ) . Protection against reactive species in corals , in addition to the multiple homologs we found with antioxidant functions such as superoxide dismutase ( Supplementary file 1 ) , is provided by two genes derived via HGT that encode glyoxalase I . One of these has an algal ( Figure 4—figure supplement 3 ) and the other a bacterial provenance ( Figure 4—figure supplement 4 ) . Interestingly , the latter gene is physically located between a DNA repair gene ( encoding RAD51 ) and a tRNA modification gene on scaffold 2777 in the A . digitifera draft assembly ( Figure 4—figure supplement 4 ) . Glyoxalase I belongs to a system that carries out the detoxification of reactive carbonyls ( RC ) , such as highly cytotoxic methylglyoxal , produced by sugar metabolism and the Calvin cycle ( Shimakawa et al . , 2014 ) . Methylglyoxal production in plastids increases with light intensity ( Takagi et al . , 2014 ) . Another gene encoding a putative RC scavenger ( Shimakawa et al . , 2014 ) is short-chain dehydrogenase/reductase ( SDR ) that was derived in corals from an algal source ( Figure 4—figure supplement 5 ) . Other alga-derived HGTs were from species containing plastids of red algal secondary endosymbiotic origin ( i . e . , chlorophyll c-containing lineages such as stramenopiles ) ( Table 1 ) . Given the coral-Symbiodinium symbiosis , it is also notable that several of the HGT candidates appear to be derived from dinoflagellates ( e . g . , Figure 4—figure supplement 5 ) . The gene contribution from chlorophyll c-containing lineages suggests a long history of interaction between these algae and the anthozoan lineage . Cnidarians enter the fossil record about 545 Ma in the latest Ediacaran Period and have been an important component of marine ecosystems throughout the Phanaerozoic , surviving five major mass extinctions and many smaller biotic crises . Although reefs have often disappeared during each of these events , various coral clades have persisted . Our analysis of a subset of coralliform cnidarians , the symbiotic Scleractinia , reveals how their genomic information has provided the basis for adapting to changes in ocean temperature and pH , while maintaining the ability to calcify . This is significant because scleractinians survived throughout the Cenozoic despite atmospheric CO2 levels reaching 800 ppm 50–34 Ma , and tropical sea temperatures of 30º–34ºC from 45 to 55 Ma ( Norris et al . , 2013 ) . This interval coincides with a reef gap , but reefs were quickly re-established thereafter . The resilience of corals in the face of extraordinary changes in ocean conditions clearly bespeaks a gene inventory that is highly adaptive as exemplified by the diversification of CARPs and genes recruited through HGT . Human activity has the potential to further reduce the abundance of these organisms in coming decades; indeed , there is compelling evidence of human destruction of corals worldwide . However , the diverse genetic repertoire of these organisms will potentially allow them to survive the expected changes in thermal structure and pH in the coming centuries ( Stolarski et al . , 2011 ) , assuming that their populations and habitats are not physically destroyed by humans . Coral genomic and transcriptome data compiled in this study are summarized in Figure 1—source data 1 . All data were filtered to remove assembled contigs <300 bp . ORFs were predicted with TransDecoder ( Haas et al . , 2013 ) yielding amino acid sequences . Protein duplicates were subsequently removed with CD-HIT ( Fu et al . , 2012 ) . With regard to coral sequence datasets , potential contaminant sequences from the algal symbiont , Symbiodinium were removed with script psytrans . py ( https://github . com/sylvainforet/psytrans ) using training sets retrieved from Symbiodinium microadriaticum ( Baumgarten et al . , 2013 ) and Acropora digitifera ( Shinzato et al . , 2011 ) . Successful separation of coral and algal sequences was validated by GC-content plots that showed a clear bimodal data distribution ( results not shown ) . Filtered sequence data were searched against SwissProt ( Boutet et al . , 2007 ) , TrEMBL ( Bairoch and Apweiler , 2000 ) , NCBI nr databases using BLASTp ( Basic Local Alignment Search Tool , e-value cut-off = 1e-03 ) ( Altschul et al . , 1990 ) and retaining annotations from databases in this order . BLAST2GO ( Conesa et al . , 2005 ) was queried to provide GO annotations , and KEGG ( Kanehisa and Goto , 2000 ) , Pfam ( Bateman et al . , 2002 ) , InterProScan ( Zdobnov and Apweiler , 2001 ) were searched to further annotated gene sets . Filtered and annotated genomic and transcriptomic data are available at comparative . reefgenomics . org . Orthologs were identified using InParanoid ( Ostlund et al . , 2010 ) on pairwise BLASTp ( e-value cutoff = 1e-05 ) yielding a list of pairwise orthologs that was subsequently queried with QuickParanoid ( http://pl . postech . ac . kr/QuickParanoid/ ) for automatic ortholog clustering among multiple species . QuickParanoid input files were filtered according to the following rules: A ) Only orthologs sets were retained with a confidence score of 1 , and B ) Pairwise comparisons were retained if only one sequence is present in each of the two involved species . To make more robust inferences based on transcriptomic data , we filtered our ortholog dataset such that any ortholog from a given phylogenetic grouping ( i . e . , robust corals , complex corals , Scleractinia , Actiniaria , Hexacorallia , Anthozoa , Cnidaria , non-cnidarian , root ) was considered to be an ortholog in this group if it was present in this group and absent in all other groups . The QuickParanoid-derived ortholog clusters were sorted into the following categories based on the constituent taxa and known species tree ( Figure 1 ) : 1 . ) 2 , 485 ‘root’ orthologs , 2 . ) 613 ‘Non-Cnidaria’ orthologs , 3 . ) 462 ‘Cnidaria’ orthologs , 4 . ) 1436 ‘Anthozoa’ orthologs , 5 . ) 1 , 810 ‘Hexacorallia’ orthologs , 6 . ) 172 ‘Actiniaria’ orthologs , 7 . ) 4 , 751 ‘Scleractinia’ orthologs , 8 . ) 1 , 588 ‘complex coral’ orthologs , and 9 . ) 6 , 970 ‘robust coral’ orthologs ( available at http://comparative . reefgenomics . org ) . For phylogenetic tree building , we selected ‘root’ orthologs that were present in at least 50% of the species of any lineage ( i . e . Root , Non-Cnidarian , Cnidarian , Anthozoa , Hexacorallia , Actiniaria , Scleractinia , Complex corals , Robust corals ) yielding 391 distinct orthologs over 7970 sequences . Orthologs were aligned individually on the protein level via MAFFT ( Katoh and Standley , 2013 ) in ‘LINSI’ mode . The resulting alignments were concatenated and then trimmed with TrimAl in the automated mode ( -automated ) ( Capella-Gutierrez et al . , 2009 ) . The resulting alignment ( 63 , 901 amino acids ) was used for phylogenetic tree building with RAxML ( Stamatakis , 2014 ) under PROTGAMMALGF model with 100 bootstrap replicates for the estimation of branch supports ( -T 32 -f a -x 1234 -p 1234 -N 100 -m PROTGAMMALGF ) . Human ionome protein reference sequences were identified and downloaded from Genbank at NCBI . Using BLASTStation-Local64 ( v1 . 4 , TM software , Inc , Arcadia , CA 91007 , USA ) , a coral protein database was generated . This contained all protein sequences available from the reefgenomics website ( http://comparative . reefgenomics . org/ ) . The human ionome protein sequences were then used as queries to search ( Basic Local Alignment Search Tool , BLAST ) against this local database using BLASTp ( no filter , Expect: 10; Word Size 3; Matrix: BLOSUM63; Gap Costs: Existence 11 extension 1 ) using BLASTStation-Local64 . The resulting matching coral proteins were saved in multi-FASTA format files , and then re-BLASTed against the NCBI Refseq protein database ( Pruitt et al . , 2012 ) limited to human-only proteins ( taxid:9606 ) on the NCBI BLAST webportal ( algorithm BLASTp , default parameters; Expect: 10; Word Size 3; Matrix: BLOSUM62; Gap Costs: Existence 11 extension 1 ) ( Camacho et al . , 2009 ) . The results were viewed for each coral protein from the input file , and a summary was generated , indicating which human protein was identified as a top hit , and in which coral species it was found . The coral multi-FASTA file was copied and annotated manually with the gene symbols of the human protein identified . If a protein coral sequence was not identified as the original human protein sequence , it was deleted , if other gene family members were identified this information was also annotated , and entered into the summary table . These multi-FASTA files were then stored for future analysis ( e . g . , generating phylogenetic trees ) . The results from the coral to human BLASTp alignments were also stored . Protein sequences in RefSeq ( version 58 ) were downloaded from NCBI FTP site ( ftp://ftp . ncbi . nlm . nih . gov/refseq/ ) . When sequences were available from more than one ( sub ) species in a genus ( e . g . , Arabidopsis thaliana and A . lyrata in the genus Arabidopsis ) , the species ( e . g . , A . thaliana ) with largest number of sequence were retained , whereas others ( e . g . , A . lyrata ) were all removed . This dataset was combined with algal sequences collected from Cryptophyta [Guillardia theta ( Curtis et al . , 2012 ) ] , Haptophyceae [Emiliania huxleyi ( Read et al . , 2013 ) ] , Rhizaria [Bigelowiella natans ( Curtis et al . , 2012 ) and Reticulomyxa filose ( Glockner et al . , 2014 ) ] , Stramenopiles [Nannochloropsis gaditana ( Radakovits et al . , 2012 ) and Aureococcus anophagefferens ( Gobler et al . , 2011 ) ] and dinoflagellates [Alexandrium tamarense ( Keeling et al . , 2014 ) , Karenia brevis ( Keeling et al . , 2014 ) , Karlodinium micrum ( Keeling et al . , 2014 ) , Symbiodinium minutum ( Shoguchi et al . , 2013 ) ] , Glaucophyte [Cyanophora paradoxa ( Price et al . , 2012 ) ] , Viridiplantae [Bathycoccus prasinos ( Moreau et al . , 2012 ) , Chlorella variabilis ( Blanc et al . , 2010 ) , Coccomyxa subellipsoidea ( Blanc et al . , 2012 ) , Micromonas pusilla ( Worden et al . , 2009 ) , Glycine max ( Schmutz et al . , 2010 ) ] and all red algal sequences collected in the previous study ( Qiu et al . , 2015 ) . We further clustered similar sequences ( sequence identity ≥85% ) among taxa from each order ( e . g . , Brassicales or Primates ) , retained the longest sequence and removed all other related sequences in the same cluster using CD-HIT version 4 . 5 . 4 ( Li and Godzik , 2006 ) . This non-redundant database , combined with protein sequences derived from three coral genomes ( Acropora digitifera and Seriatopora sp . and Stylophora pistillata ) was designated as ‘Ref58+Coral’ database . The protein sequences from A . digitifera and Seriatopora sp . genomes were used as query to search against the ‘Ref58+Coral’ database using BLASTp ( e-value cut-off = 1e-05 ) . Up to 1000 top hits ( query-hit identity ≥27 . 5% ) were recorded . These hits were sorted according to query-hit identity in a descending order among those with query-hit alignment length ( ≥120 amino acids ) . Hit sequences were then retrieved from the queried database with no more than three sequences for each order and no more than 12 sequences for each phylum . The resulting sequences were aligned using MUSCLE version 3 . 8 . 31 ( Edgar , 2004 ) under default settings and trimmed using TrimAl version 1 . 2 ( Capella-Gutierrez et al . , 2009 ) in an automated mode ( -automated1 ) . Alignment positions with ≥50% gap were discarded . We removed sequence alignments with <80 amino acid sites and those with <10 sequences . The remaining alignments were used for phylogenetic tree building using FastTree version 2 . 1 . 7 ( Price et al . , 2010 ) under the defaulting settings ( except that WAG model was used instead of JTT model ) . The resulting trees were parsed to search for coral sequences that were nested within metazoan sequences with ≥0 . 9 local support values estimated using the Shimodaira-Hasegawa test ( Shimodaira and Hasegawa , 1999 ) using in-house tools . All such coral sequences were considered to represent metazoan host genes and were discarded from downstream analyses . We conducted a second run of phylogenomic analysis using an expanded database comprising ‘Ref58+Coral’ database and all metazoan sequences collected in this study ( http://comparative . reefgenomics . org/datasets . html ) . The analyses were performed following the aforementioned procedure except that phylogenetic trees were constructed using RAxML ( Stamatakis , 2014 ) under PROTGAMMALGF model with branch supports estimated using 100 bootstrap replicates . With these RAxML trees , we searched for coral sequences that were nested within non-metazoan sequences ( with ≥60% bootstrap support ) . The resulting phylogenetic trees were manual inspected to identify HGT candidates . HGT cases that were unique to the query species ( not shared with any other coral taxa ) were discarded . The tree topologies for the resulting candidates were confirmed by re-building the trees using IQtree version 0 . 96 ( Nguyen et al . , 2015 ) under the best amino acid substitution model selected by the build-in model-selection function . Branch supports were estimated using ultrafast bootstrap ( UFboot ) approximation approach ( Minh et al . , 2013 ) using 1500 bootstrap replicates ( -bb 1500 ) . Coral sequences were considered to have a HGT origin if they were nested within non-metazoan sequences with ≥90% UFboot support . When phylogenetic trees derived from the A . digitifera data and those derived from Seriatopora sp . showed the same HGT event ( i . e . , an ancient transfer that occurred before the split of these two species ) , they were manually grouped into a shared non-redundant group . The same was the cases for phylogenetic trees that resulted from recent gene duplications . This process gave rise to 21 A . digitifera sequences and 41 Seriatopora sp . sequences that represent 41 independent HGTs from non-metazoan sources ( Table 1 ) . The key HGT genes involved in stress response were mapped to A . digitifera genome browser using the BLAST function therein ( http://marinegenomics . oist . jp/acropora_digitifera ) . The corresponding phylogenetic trees were rebuilt with inclusion of representative sequences ( if available ) from more algal taxa ( Pyrodinium bahamense pbaha01 , Gambierdiscus australes CAWD149 , Goniomonas Pacifica CCMP1869 , Togula jolla CCCM725 , Pleurochrysis carterae CCMP645 , Ceratium fusus PA161109 ) that were generated from the Marine Microbial Eukaryote Transcriptome Sequencing Project ( Keeling et al . , 2014 ) . The alignments were carried out using MUSCLE version 3 . 8 . 31 ( Edgar , 2004 ) followed by manual trimming and curation ( e . g . , with the removal of highly divergent sequences and redundant sequences from highly sampled groups ) . The corresponding ML trees were built using IQtree ( Nguyen et al . , 2015 ) as aforementioned . The phylogenetic trees for the flanking genes ( if any ) were generated likewise .

For millions of years , reef-building stony corals have created extensive habitats for numerous marine plants and animals in shallow tropical seas . Stony corals consist of many small , tentacled animals called polyps . These polyps secrete a mineral called aragonite to create the reef – an external ‘skeleton’ that supports and protects the corals . Photosynthesizing algae live inside the cells of stony corals , and each species depends on the other to survive . The algae produce the coral’s main source of food , although they also produce some waste products that can harm the coral if they build up inside cells . If the oceans become warmer and more acidic , the coral are more likely to become stressed and expel the algae from their cells in a process known as coral bleaching . This makes the coral more likely to die or become diseased . Corals have survived previous periods of ocean warming , although it is not known how they evolved to do so . The evolutionary history of an organism can be traced by studying its genome – its complete set of DNA – and the RNA molecules encoded by these genes . Bhattacharya et al . performed this analysis for twenty stony coral species , and compared the resulting genome and RNA sequences with the genomes of other related marine organisms , such as sea anemones and sponges . In particular , Bhattacharya et al . examined “ortholog” groups of genes , which are present in different species and evolved from a common ancestral gene . This analysis identified the genes in the corals that encode the proteins responsible for constructing the aragonite skeleton . The coral genome also encodes a network of environmental sensors that coordinate how the polyps respond to temperature , light and acidity . Bhattacharya et al . also uncovered a variety of stress-related pathways , including those that detoxify the polyps of the damaging molecules generated by algae , and the pathways that enable the polyps to adapt to environmental stress . Many of these genes were recruited from other species in a process known as horizontal gene transfer . The oceans are expected to become warmer and more acidic in the coming centuries . Provided that humans do not physically destroy the corals’ habitats , the evidence found by Bhattacharya et al . suggests that the genome of the corals contains the diversity that will allow them to adapt to these new conditions .

2016

Comparative genomics explains the evolutionary success of reef-forming corals

Mitochondrial Ca2+ uptake , a process crucial for bioenergetics and Ca2+ signaling , is catalyzed by the mitochondrial calcium uniporter . The uniporter is a multi-subunit Ca2+-activated Ca2+ channel , with the Ca2+ pore formed by the MCU protein and Ca2+-dependent activation mediated by MICU subunits . Recently , a mitochondrial inner membrane protein EMRE was identified as a uniporter subunit absolutely required for Ca2+ permeation . However , the molecular mechanism and regulatory purpose of EMRE remain largely unexplored . Here , we determine the transmembrane orientation of EMRE , and show that its known MCU-activating function is mediated by the interaction of transmembrane helices from both proteins . We also reveal a second function of EMRE: to maintain tight MICU regulation of the MCU pore , a role that requires EMRE to bind MICU1 using its conserved C-terminal polyaspartate tail . This dual functionality of EMRE ensures that all transport-competent uniporters are tightly regulated , responding appropriately to a dynamic intracellular Ca2+ landscape . Ca2+ regulation of key mitochondrial processes such as ATP production and initiation of apoptosis is controlled by precise balance of Ca2+ influx and efflux across the mitochondrial inner membrane ( Gunter et al . , 2000; Rizzuto et al . , 2012 ) . Studies in the 1960s and '70s established that mitochondria from most eukaryotes , except for certain yeast species , can take up large quantities of Ca2+ from the cytosol into the matrix through a mechanism that is membrane potential dependent and strongly inhibited by ruthenium compounds such as Ru360 ( Carafoli and Lehninger , 1971; Deluca and Engstrom , 1961; Ying et al . , 1991 ) . A few years ago , the field witnessed a groundbreaking achievement — identification of the MCU gene ( Baughman et al . , 2011; De Stefani et al . , 2011 ) . The 35-kDa MCU protein oliogomerizes with unknown stoichiometry to form a Ca2+-selective pore ( Baughman et al . , 2011 ) . MCU possesses two transmembrane helices ( TMHs ) connected by a short loop that hosts a signature sequence ( DIME ) thought to contribute to a Ca2+-selective permeation site . The N- and C-terminal regions of MCU are exposed to the mitochondrial matrix , each with a coiled-coil sequence of unknown function . It was subsequently found that MCU forms a complex with the mitochondrial Ca2+ uptake protein 1 ( MICU1 ) , which has co-evolved with MCU since early eukaryotic evolution ( Baughman et al . , 2011; Bick et al . , 2012 ) . In humans , MICU1 has two additional homologues , MICU2 and the neuron-specific MICU3 ( Plovanich et al . , 2013 ) . The MICUs serve as the Ca2+-sensing gate that confers Ca2+-dependence to opening of the Ca2+-selective pore ( Csordas et al . , 2013; Mallilankaraman et al . , 2012 ) . In resting cellular conditions , where cytoplasmic Ca2+ is low , MICUs shut the pore to prevent excessive Ca2+ influx into the matrix , a dangerous process that could diminish inner membrane potential and trigger apoptotic cell death . Transient elevation of Ca2+ to the low μM range , detected by EF-hands in MICUs , releases this inhibition to open the channel ( Csordas et al . , 2013; Kamer and Mootha , 2014 ) . To avoid confusion on nomenclature , we henceforth refer to the Ru-360 sensitive mitochondrial Ca2+ channel complex as the 'uniporter complex , ' a molecular assembly of the pore-forming MCU protein along with associated regulatory subunits . Recently , using quantitative mass spectroscopy , Mootha and colleagues discovered yet another component of the uniporter complex: the essential MCU regulator ( EMRE ) , a small ( ~10 kDa ) inner membrane protein found only in metazoa ( Sancak et al . , 2013 ) . EMRE possesses a single TMH and a highly conserved C-terminal polyaspartate tail , typically composed of one glutamate followed by 5–7 aspartates . In humans , MCU-EMRE interaction is absolutely required for Ca2+ permeation via MCU ( Kovacs-Bogdan et al . , 2014; Sancak et al . , 2013 ) . However , an MCU homologue in D . discoideum , a species belonging to the EMRE-lacking Amoebazoa group in protists , is fully capable of conducting Ca2+ ( Kovacs-Bogdan et al . , 2014 ) . The question naturally arises: what might be the physiological importance for MCU to become strictly dependent on EMRE for function in humans ? What would be the consequence if human MCU could transport Ca2+ without EMRE ? We address these questions by mounting an extensive investigation of EMRE . We first sought to determine the protein's transmembrane topology , a problem that cannot be definitively resolved by standard protease digestion assays ( Baughman et al . , 2011; Vais et al . , 2016 ) due to the small size of the protein’s extra-membrane regions . Two alternative strategies – directed mass-tagging and MCU-EMRE fusion construction - establish that EMRE exposes its N-terminal region to the matrix and C-terminus to the intermembrane space ( IMS ) . Mutagenesis screening and domain-interaction analysis further demonstrate that EMRE supports Ca2+ transport by using its TMH to bind to MCU through its first TMH ( TMH1 ) . Moreover , EMRE also interacts with MICU1 via its C-terminal polyaspartate tail , a molecular contact that turns out to be crucial to retain MICUs in the uniporter complex to gate the MCU pore . These results lead to a molecular model wherein the dual 'MCU-activating' and 'MICU-retaining' functionalities of EMRE together play a crucial role in orchestrating uniporter responses to intracellular Ca2+ signaling . To study uniporter subunits without interference from native mitochondrial proteins , we employed CRISPR/Cas9 to produce MCU-knockout ( KO ) , EMRE-KO , or MCU/EMRE double KO ( ME-KO ) HEK 293 cell lines . A standard Ca2+ flux assay was used to evaluate uniporter activity . In a typical experiment , HEK cells were permeabilized with digitonin in the presence of a Ca2+-sensing fluorophore ( CG-5N ) and then treated with 10 μM extracellular Ca2+ ( Figure 1A ) . In WT cells , Ca2+ is rapidly sequestered by mitochondria , and Ru360 , a potent MCU inhibitor , arrests the uptake immediately ( Figure 1A ) . ( Henceforth for clearer data presentation , only the response of permeabilized cells to Ca2+ will be presented , as in the red box in Figure 1A , with arrowheads indicating Ru360 addition . ) Consistent with previous reports ( Baughman et al . , 2011; De Stefani et al . , 2011; Sancak et al . , 2013 ) , EMRE- , MCU- , or ME-KO mitochondria are completely devoid of uniporter activity , a deficiency rescued by supplying the deleted genes ( Figure 1B ) . 10 . 7554/eLife . 15545 . 003Figure 1 . Functional analysis of uniporter in various species . ( A ) A representative fluorescence-based Ca2+ flux experiment . ( B ) Characterization of ME-KO HEK 293 cells . Left: western analysis comparing EMRE , MCU , and actin expression in WT or ME-KO cells . Right: Loss of MCU-mediated Ca2+ uptake in ME-KO cells , and rescue by delivering both MCU and EMRE genes . ( C ) Activity of uniporters in species indicated . Ca2+ flux experiments were performed using ME-KO cells expressing MCU alone or MCU and EMRE from the same species . DOI: http://dx . doi . org/10 . 7554/eLife . 15545 . 003 Genome sequences imply that MCU and MICU proteins are ancient in eukaryotic evolution , while EMRE sequences appear only among metazoa . Indeed , MCU homologues from an amoeba , D . discoideum , and a green plant , A . thaliana , both of which lack EMRE , can alone mediate rapid , Ru360-sensitive mitochondrial Ca2+ uptake in ME-KO cells ( Figure 1C ) . In contrast , metazoan MCU homologues from C . elegans and D . melanogaster require EMRE to transport Ca2+ ( Figure 1C ) , as in humans . This striking difference raises questions regarding the biological purpose of EMRE in animals and offers opportunities for attacking questions of molecular mechanism . To approach the physiological importance of EMRE’s regulatory function , we first ask how it assembles with other uniporter subunits into a channel complex . The orientation of EMRE in the inner membrane was determined by a cysteine-modification , mass-tagging method . In a typical experiment , mitoplasts ( outside-out submitochondrial vesicles lacking the outer membrane ) prepared from HEK 293 cells are incubated with a 5-kDa , membrane impermeant thiol-reactive reagent , polyethylene glycol maleimide ( PEGM ) . If the protein of interest has a cysteine exposed to the external solution , PEGM would react with this cysteine and thus increase the protein’s molecular weight . We first validated the assay on the known orientation of MCU ( Kamer and Mootha , 2015; Murgia and Rizzuto , 2015 ) . Human MCU possesses five cysteines , all in the N-terminal domain . If this region faces outward towards the IMS , PEGM would increase MCU’s mass by 5 kDa per residue modified . Experiments ( Figure 2A ) , however , show that MCU mobility on SDS-PAGE is not altered by PEGM treatment unless the mitoplast membrane is first disrupted by the mild detergent dodecyl maltoside ( DDM ) . The results thus confirm MCU’s Nin-Cin orientation , with both the N- and C-termini residing in the matrix . 10 . 7554/eLife . 15545 . 004Figure 2 . Transmembrane orientation of MCU and EMRE . ( A ) Western blot analysis of WT-MCU response to PEGM , in the absence or presence of DDM detergent , with molecular weight marker positions indicated on left . ( B ) PEGM treatment of WT , I49C , or I97C EMRE . EMRE’s molecular weight is ~10 kDa . ( C ) Cartoon illustrating the proposed membrane orientation of MCU and EMRE . The N-terminus of EMRE is fused to the C-terminus of MCU ( dashed line ) . Blue circles: native cysteines . Green boxes: soluble region . ( D ) WT-MCU or MCU-EMRE fusion protein ( Fus ) probed with anti-MCU ( left ) or anti-EMRE ( right ) antibodies . ( E ) MCU-EMRE fusion treated with PEGM . In the presence of DDM , PEGM treatment produces 4 bands , representing fusion proteins with various numbers of cysteines modified . ( F ) Mitochondrial Ca2+ uptake in ME-KO cells with ( red ) or without ( black ) expression of the MCU-EMRE fusion protein . See also Figure 2—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 15545 . 00410 . 7554/eLife . 15545 . 005Figure 2—figure supplement 1 . Uniporter function supported by indicated EMRE mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 15545 . 005 To determine EMRE topology , we constructed mutants with unique cysteines ( I49C or I97C ) engineered on either side of the TMH of the naturally cysteineless EMRE . Like WT , these mutants support MCU-dependent Ca2+ flux in EMRE-KO cells ( Figure 2—figure supplement 1 ) . In intact mitoplasts free of native EMRE protein , PEGM readily labels I97C-EMRE but fails to react with I49C or the cysteineless WT protein , while after detergent pretreatment both mutants are labeled ( Figure 2B ) . EMRE thus adopts a Nin-Cout orientation , with its C-terminal tail facing the IMS . Figure 2C summarizes the inner-membrane topology of MCU and EMRE inferred here . This was further verified by fusing EMRE onto the C-terminus of MCU , thus forcing the orientation of the two proteins in tandem to conform to the above molecular picture ( Figure 2C ) . The fusion construct was tested in ME-KO cells , where it was expressed as a full-length protein detectable by both MCU and EMRE antibodies ( Figure 2D ) . As with WT MCU , PEGM fails to modify any of the five native cysteines in the N-terminus without detergent pretreatment ( Figure 2E ) , thus implying that the fusion protein is inserted homogeneously into the inner membrane in a proper Nin-Cout orientation . Moreover , the MCU-EMRE fusion protein mediates robust , Ru360-sensitive mitochondrial Ca2+ uptake ( Figure 2F ) , a powerful result corroborating the transmembrane orientation cartooned in Figure 2C . We next investigate how EMRE interacts with MCU to support Ca2+ permeation . This issue is addressed using three EMRE variants ( Figure 3A ) , with N- or C-termini largely deleted by replacing it with a foreign 'C8' epitope ( PRGPDRPEGIEE ) ( Abacioglu et al . , 1994 ) into either region , or with the TMH substituted by an artificial transmembrane 'WALP' helix ( GWWLALALALALALALWWA ) ( Killian et al . , 1996 ) . These mutants , named ΔN- , ΔC- , or WALP-EMRE , are all properly targeted to EMRE-KO mitochondria ( Figure 3B ) . Both ΔN- and ΔC-EMRE fully support uniporter function , but cells expressing WALP-EMRE exhibit no uniporter activity ( Figure 3C ) . The results are surprising , as the strict conservation of EMRE’s C-terminal polyaspartate tail implies that it should carry out some sort of crucial function . 10 . 7554/eLife . 15545 . 006Figure 3 . MCU-interacting residues in EMRE . ( A ) EMRE constructs with indicated regions substituted by either the C8 epitope or the WALP helix . MTS: mitochondrial targeting sequence . Green boxes: soluble regions . Pink ovals: polyaspartate tail . ( B ) The presence of these mutants in whole cell lysate ( W ) or isolated mitochondria ( M ) . ΔC-EMRE is not detectable by the EMRE antibody because the C-terminal truncation removes the epitope . ( C ) Mitochondrial Ca2+ uptake in EMRE-KO cells expressing WALP- , ΔN- , or ΔC-EMRE . ( D ) Diagram summarizing Trp scan of the EMRE TM helix . Red shows positions where Trp substitution reduces the rate of Ca2+ uptake by >50% . ( E ) Sequence alignment of EMRE TM helix . Yellow indicates residues intolerant to Trp substitutions in human EMRE . ( F ) Co-IP experiments using 1D4-tagged MCU immobilized in 1D4 affinity columns to pull down indicated EMRE mutants . IP: elution , analyzed using indicated antibody . CL: whole cell lysate input . Upper panel: proteins being expressed in ME-KO cells . Leftmost lane: MCU-free control to rule out non-specific binding of EMRE in the 1D4 column . See also Figure 3—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 15545 . 00610 . 7554/eLife . 15545 . 007Figure 3—figure supplement 1 . Functional impact of mutations in EMRE’s TM helix . ( A ) Expression of indicated EMRE mutants . ( B ) Ca2+ flux traces in EMRE-KO cells expressing EMRE mutants . ( C ) Bar chart summarizing Ca2+ uptake activity in cells expressing various EMRE mutants . Dashed red line: 50% WT activity . DOI: http://dx . doi . org/10 . 7554/eLife . 15545 . 007 To locate EMRE residues critical for MCU-mediated Ca2+ transport , we performed tryptophan scanning mutagenesis to cover the entire transmembrane region ( S64 to A92 ) . This classical strategy posits that the large tryptophan side chain introduced at a protein interface is more likely to disrupt helical packing and hence function than when projecting towards lipid ( Hong and Miller , 2000; Sharp et al . , 1995 ) . The results ( Figure 3D , Figure 3—figure supplement 1 ) highlight a rather clean segregation of Trp-sensitive vs insensitive positions on a helical-wheel diagram . In particular , Trp substitution at G81 or S85 completely eliminates Ca2+ uptake via the uniporter complex . These two residues belong to a conserved Gxxx[G/A/S] motif ( Figure 3E ) , frequently found to mediate packing of TMHs in membrane protein structures ( Russ and Engelman , 2000 ) . Co-immunoprecipitation ( co-IP ) experiments further confirm the GxxxS sequence as crucial to MCU-EMRE complex formation . EMRE variants were co-expressed with MCU carrying a C-terminal '1D4' epitope ( TETSQVAPA ) ( MacKenzie et al . , 1984 ) in ME-KO cells , and the MCU-EMRE complex was immobilized on a 1D4 affinity column for downstream analysis . WT EMRE is captured by MCU , but disruption of the GxxxS region by the G81W or S85W mutation prevents this association , while Trp substitution of residues elsewhere on the TMH does not ( Figure 3F ) . These results taken together show that EMRE binds to MCU using a Gxxx[G/A/S] motif in the C-terminal half of its TMH , and that this binding is necessary to activate the Ca2+-conducting pore in the uniporter complex . To examine the MCU side of the interaction with EMRE , we launched Trp-perturbation mutagenesis of both TMHs ( residues L234 – T254 in TMH1; Y268 – V283 in TMH2 ) . All mutants were expressed to near WT levels in MCU-KO cells ( Figure 4—figure supplement 1 ) , and were classified as either low- or high-impact on Ca2+ transport function ( Figure 4A , Figure 4—figure supplement 1 ) . Of the 16 Trp mutations in TMH2 , only two ( F269W , T271W , located near the N-terminal end of the helix ) induce severe functional defects , as if most TMH2 residues project either to the lipid bilayer or an aqueous cavity . In contrast , the six Trp-sensitive positions in TMH1 tend toward one side of a helical wheel diagram , suggesting that this high-impact face might pack against other TMHs in the MCU-EMRE complex . 10 . 7554/eLife . 15545 . 008Figure 4 . Interactions between EMRE and MCU’s TMH1 . ( A ) Helical projection diagram summarizing functional impact of Trp substitutions in TMHs of MCU . Trp mutation that reduces Ca2+ uptake by >70% is defined as high impact ( red ) , and <30% as low impact ( black ) . Red arc highlights proposed helical surface sensitive to Trp substitutions . ( B ) Residue swap showing an impaired MCU mutant ( A241F ) forming a highly functional uniporter complex with an EMRE mutant ( F77A ) . Left: Ca2+ uptake in ME-KO cells expressing indicated MCU and EMRE mutants . Right: Ca2+ uptake ( upper ) and expression of key uniporter proteins ( lower ) in cells transfected with A241F-MCU and WT- or F77A-EMRE . ( C ) Co-IP experiments comparing complex formation of A241F-MCU with WT- or F77A-EMRE . ( D ) Ca2+ flux in a hMCU-ceMCU chimera ( human portion: yellow , C . elegans portion: blue ) , coexpressed with either hEMRE or ceEMRE in ME-KO cells . See also Figure 4—figure supplements 1–3 . DOI: http://dx . doi . org/10 . 7554/eLife . 15545 . 00810 . 7554/eLife . 15545 . 009Figure 4—figure supplement 1 . Functional impact of mutations in MCU’s TM helices . ( A ) Expression of selected MCU mutants . ( B–C ) The effect of mutations in MCU’s TMH1 ( B ) or TMH2 ( C ) on Ca2+ flux via the uniporter . Dashed red line: 30% WT activity . DOI: http://dx . doi . org/10 . 7554/eLife . 15545 . 00910 . 7554/eLife . 15545 . 010Figure 4—figure supplement 2 . Trp mutations in D . discoideum MCU . ( A ) Sequence alignment of TMH1 and TMH2 of human and D . discoideum MCU homologues . Red indicates residues selected for Trp substitution . ( B ) Characterization of D . discoideum MCU mutants . Left: Ca2+ flux in ME-KO cells . Right: expression of the mutants . ( C ) The activity of Trp mutants compared with WT D . discoideum MCU . DOI: http://dx . doi . org/10 . 7554/eLife . 15545 . 01010 . 7554/eLife . 15545 . 011Figure 4—figure supplement 3 . Uniporter formation by F77A EMRE and MCU mutants . ( A ) Uniporter activity in EMRE-KO cells expressing F77A EMRE and MCU mutants as indicated . Activity is normalized to that produced by WT MCU and WT EMRE . ( B ) Co-IP test of complex formation by A241W MCU and WT- or F77A-EMRE . DOI: http://dx . doi . org/10 . 7554/eLife . 15545 . 011 If a Trp mutation in MCU perturbs the function of human uniporter solely by interrupting EMRE binding , it should cause negligible functional impact when introduced at corresponding positions in D . discoideum MCU , because this homologue transports Ca2+ without EMRE ( Figure 1C ) . We therefore introduced each of the eight high-impact Trp substitutions ( Figure 4A ) into D . discoideum MCU to test the mutational effect in ME-KO cells . The results ( Figure 4—figure supplement 2 ) show that six of these mutations are functionally disruptive while two are fully active; these two active mutants correspond to L240W and A241W in TMH1 of human MCU , suggesting that these residues in human MCU contact EMRE . A steric clash between a substituted Trp in MCU and a native residue in EMRE could in principle be alleviated by reducing the side-chain volume of that particular EMRE residue . We chose F77 in EMRE to test this idea , as it is on the same helical face as the GxxxS sequence identified above , and since the large Phe residue enables substantial shortening of the side chain . Accordingly , F77A EMRE was coexpressed with each of the 6 functionally defective MCU Trp mutants of TMH1 . This EMRE mutant , which forms a functional channel with WT MCU , rescues Ca2+ transport with A241W but not with any of the other mutants ( Figure 4—figure supplement 3 ) . Similarly , the impaired uniporter function induced by A241F in MCU is rescued by F77A in EMRE . Co-IP experiments further show that A241F ( or A241W ) MCU pulls down F77A but not WT EMRE ( Figure 4C , Figure 4—figure supplement 3 ) . These results demonstrate that the combination of a large and a small side chain ( F or W , A ) on EMRE position 77 and MCU position 241 leads to proper transport function regardless of which protein the residues occupy . This 'side chain swap' experiment argues strongly that A241 in MCU’s TMH1 is in close proximity to F77 in the TMH of EMRE in the uniporter complex . While analyzing MCU/EMRE from various species ( Figure 1C ) , we noticed that human MCU ( hMCU ) forms functional Ca2+ channels with human EMRE ( hEMRE ) or with C . elegans EMRE ( ceEMRE ) , but C . elegans MCU ( ceMCU ) supports mitochondrial Ca2+ uptake only with ceEMRE ( Figure 4D ) . These results present an opportunity to test which of ceMCU’s two TMHs is responsible for discriminating against hEMRE . Accordingly , we produced two MCU chimeras , with TMH1 or TMH2 of ceMCU substituted by the corresponding region in hMCU . The chimera containing hMCU TMH1 , though well expressed , is transport-inactive in the presence of either hEMRE or ceEMRE . The chimera containing hMCU TMH2 , however , is activated by ceEMRE , while remaining unresponsive to hEMRE ( Figure 4D ) . This result independently supports the proposal that TMH1 of MCU contains the interaction region for EMRE . With the membrane disposition of the MCU-EMRE complex in hand , we now ask how EMRE interacts with MICU proteins ( Sancak et al . , 2013 ) , which contain no apparent transmembrane sequences . To tackle this problem , we began by testing whether MICUs bind to EMRE from the matrix- or IMS-side of the inner membrane . Currently , the submitochondrial localization of MICU1 is unsettled ( Csordas et al . , 2013; Foskett and Madesh , 2014; Hoffman et al . , 2013; Hung et al . , 2014; Waldeck-Weiermair et al . , 2015 ) , and that of MICU2 has not been approached with rigorous methods ( Vais et al . , 2016 ) . Moreover , although MICU1 is known to be a peripheral membrane protein ( Csordas et al . , 2013 ) , it remains unclear if MICU2 is similarly attached to the inner membrane . Inner membrane association of MICUs was probed by stripping mitoplasts of peripheral membrane proteins using alkaline Na2CO3 treatments . Figure 5A shows that MICU1 and cytochrome C ( Cyt-C ) , but not the integral membrane protein Letm1 , are extracted into Na2CO3 solution , a result consistent with a previous report ( Csordas et al . , 2013 ) that MICU1 is a peripheral membrane protein . Furthermore , nearly all MICU1 is membrane-bound , virtually none appearing in the IMS without Na2CO3 treatment , in contrast to Cyt-C , which is found in both the IMS and the membrane ( Figure 5A ) . Similar experiments demonstrate that MICU2 remains membrane-associated even after harsher Na2CO3 extraction conditions ( Figure 5B ) . The results thus establish that MICU1 and MICU2 are both confined to the inner-membrane surface under physiological conditions . 10 . 7554/eLife . 15545 . 012Figure 5 . Localization of MICUs and interaction with the pore region . ( A–B ) Carbonate extraction ( pH 10 . 5 or 11 . 5 ) of MICU1 at 4°C ( A ) or MICU2 at room temperature ( B ) for 1 h , showing membrane pellet ( M ) , proteins extracted into supernatant ( S ) , and control ( con ) with mitoplasts treated at pH 7 . 0 . ( C– D ) PEGM modification of MICU1 or MICU2 . Both MICUs are detected at monomer ( ~64 kDa ) or dimer ( ~115 kDa ) positions . ( E ) Co-IP experiments using immobilized Flag-tagged MICU1 to pull down MCU or EMRE . For all experiments shown in this figure , MICU1 and MICU2 were Flag- and V5-tagged , respectively , and were detected using corresponding Flag and V5 antibodies . See also Figure 5—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 15545 . 01210 . 7554/eLife . 15545 . 013Figure 5—figure supplement 1 . MICU2 interaction with other subunits in the uniporter complex . ( A ) Co-IP experiment showing complex formation of MICU1 and MICU2 . MICU2 was used as bait . ( B ) Co-IP experiments using immobilized MICU2 to pull down MCU or EMRE . ( C ) Immobilized MICU2 was used to pull down EMRE with MICU1 also expressed . For all Co-IP experiments , ME-KO cells were used , and proteins being expressed are indicated at the upper panel above the Western results . MICU1 and MICU2 were Flag and V5 tagged , respectively , and were detected with corresponding antibodies . IP: elution . CL: cell lysate . DOI: http://dx . doi . org/10 . 7554/eLife . 15545 . 013 To test if MICU1 is exposed to the mitochondrial matrix or the IMS , we incubated mitoplasts with PEGM , relying on MICU1's native cysteines to report accessibility to the reagent from the mitoplast exterior . PEGM readily reacts with MICU1 ( Figure 5C ) , showing that this subunit resides on the outer , IMS side of the inner membrane . Two issues regarding this experiment require comment . First , the PEGM-treated sample appears as a smear instead of a defined band , consistent with heterogeneous modification of the protein's 7 native cysteines . Second , a significant western-blot signal is observed roughly corresponding to MICU1 dimers , stable in SDS-PAGE conditions , as reported previously ( Patron et al . , 2014 ) . This signal also shifts upward after PEGM treatment , suggesting that MICU1 dimers or higher oligomers are also localized to the IMS . Similar results in parallel experiments with MICU2 ( Figure 5D ) argue that both MICU1 and MICU2 are associated with the outer leaflet of the mitochondrial inner membrane , a location in harmony with the cytoplasmic Ca2+-sensing function of these proteins . It has been established , as we also confirm here ( Figure 5—figure supplement 1 ) , that MICU1 forms a stable complex with MICU2 ( Kamer and Mootha , 2014; Patron et al . , 2014 ) , and that MICU1 , but not MICU2 , is tightly associated with the MCU-EMRE complex ( Kamer and Mootha , 2014; Sancak et al . , 2013 ) . It however remains uncertain if MICU1 interacts with MCU or EMRE ( Hoffman et al . , 2013; Sancak et al . , 2013 ) , a problem addressed using co-IP to examine association of FLAG-tagged MICU1 with MCU or EMRE expressed individually in ME-KO cells . As illustrated in Figure 5E , MICU1 can precipitate EMRE without MCU present and MCU without EMRE present . This observation rules out a required MICU1-interacting surface contributed by both MCU and EMRE . It also invites us to search for the molecular determinants mediating MICU1-EMRE interaction in the absence of MCU . The IMS localization of MICU1 ( Figure 5C ) implies that EMRE binds to MICU1 via its C-terminal , IMS-exposed region containing the polyaspartate tail ( EDDDDDD ) . This highly charged tail alerts us to a complementary polybasic sequence ( KKKKR ) , which though conserved in MICU1 , is absent in MICU2 ( Hoffman et al . , 2013 ) . Indeed , co-IP experiments demonstrate that MICU1 pulls down ΔN- but not ΔC-EMRE ( Figure 6A ) . Moreover , a MICU1 mutant carrying an electrostatically neutered sequence ( KKKKR =>EQEQR ) readily complexes with MICU2 , but not with EMRE ( Figure 6B ) . These results strongly argue that the EMRE-MICU1 interaction is mediated by this electrostatic pair . The strict conservation of these charged sequences further suggests that the EMRE-MICU1 interaction plays an important , previously unappreciated physiological role . 10 . 7554/eLife . 15545 . 014Figure 6 . Functional importance of the MICU1-EMRE interaction . ( A–B ) Co-IP experiments with WT- or EQEQ-MICU1 ( Flag-tagged ) used to pull down WT or mutant EMRE proteins in ME-KO cells . MICU1 was detected using anti-Flag , MICU2 by anti-V5 , and ΔN- or ΔC-EMRE by anti-C8 . ( C–D ) The effect of MICU1 knockdown on mitochondrial Ca2+ uptake in WT HEK 293 cells at high ( C ) or low ( D ) Ca2+ conditions . Con: cells with no MICU KD . sh1-3 indicates three stable cell lines expressing distinct shRNAs against MICU1 mRNA . ( E ) Mitochondrial Ca2+ uptake ( 30 µM Ca2+ ) using untransfected EMRE-KO cells , or cells expressing WT- , ΔN- , or ΔC-EMRE as indicated . ( F–G ) Ca2+ flux ( 0 . 5 µM Ca2+ ) via MCU complexed with WT- or ΔN-EMRE ( F ) , or ΔC-EMRE ( G ) in the presence or absence of stable MICU1 KD by shRNA2 . Data shown in C–G represent mean ± s . e . m . of 3–4 independent measurements . See also Figure 6—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 15545 . 01410 . 7554/eLife . 15545 . 015Figure 6—figure supplement 1 . Biochemical characterization of MICU1 knockdown cells . ( A ) the mRNA level of MICU1 in MICU1 knockdown cells compared with WT control , as determined by qPCR . ( B ) MCU and EMRE expression in these cells . DOI: http://dx . doi . org/10 . 7554/eLife . 15545 . 015 The MICU proteins act as the Ca2+-sensing gate in the uniporter complex , shutting the pore at resting cellular Ca2+ concentrations and opening it when cytoplasmic Ca2+ exceeds ~1 µM . By binding to both MCU and MICU1 , EMRE might serve as an anchor to retain the MICU1-MICU2 pair near the Ca2+-conducting pore . If so , disrupting the MICU1-EMRE interaction would yield a population of channels free of MICUs , allowing unregulated , constitutive Ca2+ permeation from the cytosol into the matrix . To examine this idea , we quantify MCU-dependent Ca2+ uptake by following accumulation of the 45Ca2+ radioisotope into mitochondria in digitonin-permeabilized cells , an approach that allows free extramitochondrial Ca2+ to be buffered at well-defined submicromolar concentrations without sacrificing sensitivity . We first examine the effect of MICU1 knockdown ( KD ) on mitochondrial Ca2+ uptake . Three HEK293 cell lines stably expressing distinct short hairpin RNAs ( sh1 – 3 ) were generated , all exhibiting at least 70% decrease of MICU1 mRNA and normal levels of MCU or EMRE protein ( Figure 6—figure supplement 1 ) . Although insufficiently sensitive antibodies frustrate quantification of MICU1 , the MICU1-KD cell lines show a profound functional alteration in 45Ca2+ uptake . At high Ca2+ ( 30 μM ) , rates of Ca2+ transport into WT and MICU1-KD mitochondria are similar , while as expected , uptake is virtually undetectable in ME-KO cells ( Figure 6C ) . At low Ca2+ ( 0 . 5 μM ) , however , MICU1-KD mitochondria show massive Ca2+ accumulation in the matrix ( Figure 6D ) , in dramatic contrast to WT , where very little Ca2+ uptake occurs . These observations confirm previous studies that identified MICU1 as a molecular element controlling Ca2+-dependent activation of the uniporter complex ( Csordas et al . , 2013; Mallilankaraman et al . , 2012 ) . Functional manifestations of the EMRE-MICU1 interaction were further examined by comparing 45Ca2+ uptake supported by EMRE variants expressed in EMRE-KO cells . As above , WT- , ΔN- , or ΔC-EMRE all activate MCU-dependent Ca2+ uptake to a similar degree at high Ca2+ ( Figure 6E ) . At low Ca2+ , uptake is suppressed in mitochondria hosting WT or ΔN-EMRE , but is enhanced over 50-fold by MICU1 knockdown ( Figure 6F ) . In contrast , Ca2+ rapidly enters mitochondria containing ΔC-EMRE , which cannot bind MICU1 , and the rate is only trivially increased after MICU1 KD ( Figure 6G ) . We thus conclude that MICU1 must bind EMRE to maintain uninterrupted engagement with the MCU pore , thus conferring Ca2+-dependent gating upon what would otherwise be constitutive Ca2+ leakage into the mitochondrial matrix . In the past few years , the mitochondrial Ca2+ transport field has witnessed a molecular dawn following a half-century of functional phenomenology . It is now firmly established that the human mitochondrial Ca2+ uniporter is a Ca2+-activated Ca2+ channel composed of at least four proteins: MCU , EMRE , MICU1 , and MICU2 . The present work focuses mainly on EMRE , the least understood component of the channel complex . Impressed by the functional dependence on EMRE arising as metazoan uniporters evolved , we have endeavored to enrich our current view of the uniporter subunits and the physiological purposes of the domain interactions mediating their assembly . Results here establish ( 1 ) EMRE orientation in the inner membrane , ( 2 ) the molecular contacts governing EMRE interactions with MCU and MICU1 , ( 3 ) the disposition of the MICU1-MICU2 complex on the outer surface of the inner-membrane , and ( 4 ) the functional purpose of the EMRE-MICU1 interaction . These findings lead to a molecular model ( Figure 7 ) featuring a central role of EMRE in orchestrating uniporter responses to intracellular Ca2+ signaling . 10 . 7554/eLife . 15545 . 016Figure 7 . The physiological role of EMRE . Cartoon summarizing main findings here , illustrating how the two functions of EMRE , MCU activation and MICU1 retention , together prevent Ca2+ leakage through the uniporter complex in resting cellular conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 15545 . 016 When a Ca2+ channel in the plasma membrane opens , or when an intracellular store releases Ca2+ , a cytoplasmic Ca2+ wave is generated . Once the wave hits mitochondria , Ca2+ can rise above ~1 μM , activating the uniporter to catalyze rapid Ca2+ entry into the mitochondrial matrix . Mitochondria can therefore serve as a buffer to shape intracellular Ca2+ signals ( Demaurex et al . , 2009 ) . Moreover , Ca2+ entry boosts ATP output by accelerating the citric acid cycle , but excessive , sustained Ca2+ accumulation in the matrix triggers caspase-dependent apoptosis ( Rizzuto et al . , 2012 ) . Thus , mitochondria can also decode Ca2+ stimulation as either metabolic or death signals . Failure of the uniporter to appropriately respond to Ca2+ signals would perturb these crucial physiological processes and could also produce other serious problems . For instance , if the uniporter fails to stay inactive under resting conditions , the large negative inner membrane potential of energized mitochondria would drive continual Ca2+ influx . Removing these Ca2+ ions requires the action of Na+/Ca2+ and Na+/H+ exchangers at a cost of 3 H+ entering into the matrix for each Ca2+ expelled . Unregulated , ‘leaky’ uniporters would therefore divert protons away from the FoF1-ATPase , partially uncoupling electron transport from ATP synthesis . The recent discovery of MICU proteins ( Baughman et al . , 2011; Mallilankaraman et al . , 2012 ) has enhanced our mechanistic understanding of how cellular Ca2+ signals control uniporter activity . It is now clear that the mitochondrial response to physiological Ca2+ is mediated by the Ca2+-sensing gate formed by the MICU subunits that engage the pore-lining MCU proteins at the external face of the inner membrane . This modular design comes with a potential danger: that the gating and ion-conducing subunits might dissociate sporadically , producing unregulated Ca2+ channels . How could this problem be prevented ? It is likely that MICU1 has a high intrinsic affinity to MCU , since at least part of the MICU1-MCU complex survives lengthy co-IP experiments . Moreover , diffusion of MICUs confined to two dimensions within the inner membrane raises the local density of MICUs near the MCU pore . These factors combined could in principle reduce the population of MICU-free , unregulated uniporters — a condition that might be particularly helpful to protist and plant mitochondria , where MCU and MICUs are the only components of the uniporter complex . The small , single-pass membrane protein EMRE emerges in animals as a new subunit of the uniporter complex . A crucial finding here is that the presence of EMRE in the uniporter complex ensures that the channel conducts Ca2+ only when cytoplasmic Ca2+ rises above resting levels . This requires EMRE to use its polyaspartate tail to bind MICU1 , as if it functions as 'molecular glue' to prevent dissociation of MICU1 from the MCU pore , a circumstance that would produce catastrophic Ca2+ leakage ( Figure 7 ) . Alternatively , EMRE might allosterically transmit the Ca2+ signal from MICUs to the pore; in this case , disrupting EMRE-MICU1 interaction would also prevent MICUs from properly gating the Ca2+ pathway . We consider this allosteric scenario unlikely , as the mechanism by which MICU1 gates MCU probably evolved in early eukaryotic evolution when EMRE was absent . The understanding that EMRE safeguards mitochondria against inappropriate Ca2+ uptake helps us appreciate the physiological importance of the strict EMRE-dependence of uniporter function appearing in animals . As in any multisubunit protein , it is inevitable that EMRE might occasionally dissociate from MCU , and some tissues under natural or pathological conditions might express MCU in excess of EMRE . Under these situations , a population of EMRE-free uniporters could arise . These channels would also lack MICUs , which would no longer be EMRE-linked to the pore . But the MCU-activation function of EMRE would ensure that these channels would become inactive , preventing them from wreaking havoc on normal cell physiology ( Figure 7 ) . We should point out that our results clash with several published assertions regarding the uniporter complex . First , single-channel recordings in planar lipid bilayers have been used to argue that the human MCU protein alone is sufficient to reconstitute a Ca2+ channel without EMRE ( De Stefani et al . , 2011; Patron et al . , 2014 ) . These recordings , however , obtained with in vitro-expressed protein of uncharacterized purity , show channel properties vastly different from uniporter currents directly patch-recorded from intact mitoplasts ( Kirichok et al . , 2004 ) . Second , a recent study ( Vais et al . , 2016 ) using protease digestion argues that EMRE adopts a Nout-Cin orientation , opposite to that deduced here . Interpretation of the assay , however , is based on an unjustified assumption that the N- and C- termini of EMRE are digested at similar rates . In contrast , our results are supported by two lines of direct and independent evidence – mass tagging of substituted cysteines and the functional competence of the MCU-EMRE fusion protein ( Figure 1 ) . This same study also claims that EMRE uses its C-terminal tail to sense matrix Ca2+ , producing a biphasic response of uniporter activity to matrix Ca2+ . However , this phenomenon , to our best knowledge , has not been observed in any mitochondrial Ca2+ uptake experiments in the literature or in previous patch recordings ( Kirichok et al . , 2004 ) . We also note a recent study appearing when our work was under review ( Yamamoto et al . , 2016 ) that deduced , using an epitope-tagging method , a Nin-Cout EMRE orientation fully consistent with our results . That study also shows that a Pro-to-Ala substitution 3 residues N-terminal to the predicted TMH - a region left unperturbed in our 22-residue deletion ΔN-EMRE construct - abolishes EMRE-MCU interaction . Thus , a small portion in the N-terminus of EMRE near the TMH might also be involved in binding MCU . In summary , our results argue that EMRE mediates two distinct functions – MCU activation and MICU retention - through two distinct types of subunit-subunit interactions . These functions conspire to achieve a single physiological outcome: obligatory linkage of the Ca2+-conducting and Ca2+-sensing machineries , a necessary condition for the uniporter complex to respond rapidly and accurately to the elaborate Ca2+-signaling network that has evolved in animal cells . Site-directed mutagenesis was performed using the QuickChange mutagenesis kit ( Agilent ) . HEK 293 cells were grown in Dulbecco’s modified Eagle’s medium supplemented with 10% FBS , and were incubated at 37°C , 5% CO2 . The HEK-293 cell line was supplied by Dr . D . E . Clapham and authenticated by short tandem repeat profiling conducted by ATCC , and was free of mycoplasma as determined by PCR based detection using a kit supplied by ATCC ( 30-1012K ) . Transient transfection was performed using Lipofectamine 3000 ( Life Technologies ) , following manufacturer’s instructions . Cells were used for downstream analysis 1-2 days after transfection . Stable knockdown was achieved by lentivirus , using the transfer vector pLKO . 1 puro ( Sigma ) for U6-driven shRNA expression . The viral titer was determined with a p24 ELISA kit ( Clontech , Mountain View , CA ) . WT HEK293 cells were exposed to the virus for 12 hr , using a multiplicity of infection of 5–10 . Afterwards , the culture was incubated with 2 μg/mL puromycin for 2 days to eliminate untransduced cells . The efficiency of knockdown was evaluated by quantitative PCR ( qPCR ) . Detailed qPCR procedure and the shRNA sequences are reported in Extended Experimental Procedures . Gene knockout by CRISPR/Cas9 was performed using the published protocol ( Ran et al . , 2013 ) . In brief , the pSpCas9 ( BB ) vector containing the 20-nucleotide guide sequence was transfected into HEK 293 cells . After two days of incubation , single cells were isolated by serial dilutions , and expanded for 2–4 weeks . Gene KO was assessed by sequencing and Western blot . Two sets of guide sequences ( see Supplementary file 1 ) were used to rule out off-target effects . For Western blot , proteins on an SDS gel were transferred onto nitrocellulose membranes , which were blocked by 5% milk in TBS , and then incubated with the primary antibody diluted in TBST ( TBS + 0 . 1% Tween-20 ) . Signal development was done using alkaline phosphatase conjugated secondary antibodies ( Pierce ) and the NBT/BCIP substrate ( Life Technologies ) . The primary antibody and dilution used: α-MCU ( Sigma , HPA016480 , 1:2000 ) , α-EMRE ( Santa Cruz , 86337 , 1:400 ) , α-FLAG ( Sigma , F1804 , 1:4000 ) , α-V5 ( Life Technologies , 46–0705 , 1:5000 ) , α-Cyt-C ( Santa Cruz , 13156 , 1:1000 ) , α-β-actin ( Santa Cruz , 69879 , 1:500 ) , α-Letm1 ( Abcam , 55434 , 1:2000 ) . Monoclonal anti-1D4 and -C8 antibodies were produced in house . All co-IP experiments were performed at 4 °C . Transfected HEK 293 cells were grown in a 10-cm dish to confluency , were harvested , and then lysed in 1-mL solubilization buffer ( SB , 100 mM NaCl , 20 mM Tris , 1 mM EGTA , 25 mM DDM , pH 7 . 5-HCl ) , supplemented with an EDTA-free protease inhibitor cocktail ( cOmplete Ultra , Roche ) . The cell lysate was clarified by centrifugation , and a small portion of the sample was taken for whole cell lysate analysis . Antibody-conjugated Sepharose beads ( 25 μL ) were added , and after 1 h , the beads were collected on a mini column , washed with 2-mL SB , and eluted with 200-μL SDS-gel loading buffer for Western blot . Antibody affinity gel used: FLAG ( Sigma , A2220 ) , V5 ( Sigma , A7345 ) . 1D4 and C8 affinity gels were produced using 20-mg 1D4 or C8 antibody per 1-g Sepharose 4B ( GE Healthcare ) . Mitoplasts were formed at 4°C by standard procedures that yield outside-out , stable transport vesicles . Protease inhibitor ( cOmplete Ultra , Roche ) was present in all steps . HEK 293 cells from a 15-cm dish were pelleted , resuspended in 2-mL mitochondria resuspension buffer ( MRB , 250 mM sucrose , 5 mM HEPES , 1 mM EGTA , pH 7 . 2-KOH ) , and lysed by passing through a 27 . 5 g needle 15 – 20 times . Nuclei and cell debris were removed by spinning the cell lysate at 1000 g for 10 min . The supernatant was spun down at 10 , 000 g for 10 min , resuspended in 2-mL MRB , and then spun down again to pellet crude mitochondria . To obtain mitoplasts , mitochondria were resuspended in 800-μL hypotonic shock buffer ( 5 mM sucrose , 5 mM HEPES , 1mM EGTA , pH 7 . 2-KOH ) , and subjected to osmotic shock for 10 min . Then 200 μL of high-salt storage buffer ( 750 mM KCl , 100 mM HEPES , 2 . 5 mM EGTA , pH 7 . 2-KOH ) was added , and mitoplasts were subsequently sedimented by centrifugation at 20 , 000 g for 10 min . The supernatant , which contains proteins in the outer membrane and the intermembrane space , was collected if further analysis is required . Mitoplasts were resuspended in the thiol-modification buffer ( 100 mM NaCl , 50 mM MOPS , pH 7 . 0-NaOH ) , to which 1 mM PEGM ( Sigma ) in the presence or absence of 1 mM DDM ( Anatrace ) was added . The samples were incubated for 1–4 hr at RT before the reaction was quenched with 5 mM cysteine . All reagents were prepared fresh before experiments . Mitoplasts were resuspended either in carbonate extraction buffer ( 120 mM NaCO3 , pH 10 . 5- or 11 . 5-NaOH ) or in a control solution ( 250 mM sucrose , 25 mM Tris , pH 7 . 0-HCl ) . The samples were incubated at 4°C or RT for 1 hr , and then spun down with ultracentrifugation at 200 , 000 g for 1 hr . The supernatant contains proteins extracted by carbonate , while the membrane pellet containing integral membrane proteins . All Ca2+ uptake assays were repeated at least 3 times on multiple preparations , and traces in figures show typical responses . For the fluorescence-based assay , 107 HEK 293 cells were suspended in 10-mL Ca2+ flux wash buffer ( CWB , 120 mM KCl , 25 mM HEPES , 2 mM KH2PO4 , 1 mM MgCl2 , 50 μM EGTA , pH 7 . 2-KOH ) , pelleted at 1000 g for 5 min , and resuspended in 2 . 5-mL recording buffer ( RB , 120 mM KCl , 25 mM HEPES , 2 mM KH2PO4 , 1 mM MgCl2 , 5 mM succinate , pH 7 . 2-KOH ) . 2 mL of the cell suspension was loaded into a stirred quartz cuvette in a Hitachi F-2500 spectrophotometer ( ex: 506 nm , ex-slit: 2 . 5 nm , em: 532 nm , em-slit: 2 . 5 nm , sampling frequency: 2 Hz ) , with the temperature maintained at 37°C by a circulating bath . In a typical experiment , reagents were added into the cuvette in the following order: 0 . 5-μM calcium green 5N ( Life Technologies ) , 30-μM digitonin ( Sigma ) , 10-μM CaCl2 , and 2-μM Ru360 ( Santa Cruz ) . Under these conditions , peak free Ca2+ concentrations were close to 10 μM ( 11 + 6 SD , N=40 , as determined by calibration ) . Because of uncertainties in protein concentration , Ca2+ uptake activity is reported as a linear fit to the fluorescent signal obtained in the first 10 s after addition of 10-μM CaCl2 . Activity was not altered by 5 µM thapsigargin . For the 45Ca2+ based uptake assay , 2 x 106 cells were suspended in 1 . 5-mL CWB , spun down at 2 , 000 g for 1 min , and resuspended again in 200-μL CWB , supplemented with 5 μM thapsigargin and 30 μM digitonin . To initiate Ca2+ flux , 100 μL of the cell suspension was transferred to either 400-μL high-Ca2+ flux buffer ( RB + 10 μM EGTA and 40 μM 45CaCl2 ) or 400-μL low-Ca2+ flux buffer ( RB + 0 . 69 mM EGTA , 0 . 5 mM CaCl2 , and 20 uM 45CaCl2 , pH 7 . 2-KOH ) , with both solution containing 5-μM thapsigargin and 30-μM digitonin . At desired time points , 100 μL of the reaction mixture was added into 5-mL ice-cold CWB , and then filtered through 0 . 45-μm nitrocellulose membranes ( EMD-Millipore ) on a vacuum filtration manifold ( Millipore model 1225 ) . The membrane was washed immediately with 5-mL ice cold CWB , and later transferred into scintillation vials for counting . 45Ca2+ radioisotope was purchased from Perkin Elmer , with a specific activity of 12 . 5 mCi/mg . Sequences of MCU or EMRE homologues were collected using PSI-BLAST search of ~100 species . EMRE was identified by the presence of the polyaspartic tail , and MCU by the conserved DIME loop . Multiple sequence alignment was performed using the ClustalW2 online server ( Larkin et al . , 2007 ) . The helical wheels were plotted using Antheprot v 6 . 4 ( Deleage et al . , 2001 ) . Mitochondrial targeting sequence prediction was carried out using the TargetP 1 . 1 online server ( Emanuelsson et al . , 2007 ) . Whole cell RNA was extracted from HEK 293 cells grown in 6-well plates using TRIzol . Residual DNA was removed using the TURBO DNA-free kit ( Life Technologies ) . First strand cDNA synthesis was performed with 1 μg RNA using M-MuLV reverse transcriptase ( NEB ) , following manufacturer’s instructions . The sample was subsequently digested with RNaseH ( NEB ) . qPCR was performed with SsoFast EvaGreen Supermixes ( BIO-RAD ) , using 0 . 5 μM β-actin or MICU1 primers , and 0 . 5 , 2 . 5 , 5 , or 10 ng RNA for producing a standard curve . Detection of the PCR product was done with a CFX96 real-time PCR detection system ( BIO-RAD ) , using the following protocol: 95°C for 30 s , 50 cycles of 95°C for 5 s and 57°C for 5 s . The sequence of the primers is provided in Supplementary file 1 . ΔCt was calculated by subtracting the Ct for β-actin from the Ct for MICU1 , with 3 independent RNA extractions and qPCR measurements using 2 . 5 ng whole RNA . ΔΔCt was calculated by subtracting the mean ΔCt for control WT cells from the mean ΔCt for each stable MICU1 knockdown cells . The results were presented as the percentage of MICU1 mRNA in MICU1 knockdown cells relative to MICU1 mRNA in WT control , using the equation% mRNA = 1/2|ΔΔCt| .

Like all power plants , mitochondria – the compartments inside our cells that supply energy – must adjust their energy output to match fluctuations in demand . Inside cells , the levels of calcium ions in the cytoplasm often signal such demands . Mitochondria therefore control their calcium ion levels with tightly regulated , membrane-embedded proteins that move calcium ions into and out of the mitochondria . One of these membrane machines , the mitochondrial calcium uniporter ( MCU ) complex , is a "smart channel" that admits calcium ions into the mitochondria only when their cytoplasmic levels exceed a threshold . The MCU complex contains four essential proteins: MCU , which forms the pore through which the calcium ions enter the mitochondrion; MICU1 and MICU2 , which act as “gatekeepers” , opening the pore only when the cell contains high levels of calcium ions; and EMRE , a small , mysterious protein . Why is EMRE required for the channel's operation , and how does it fit into the four-protein complex ? By comparing EMRE proteins from different species , constructing mutant forms of EMRE , and recording calcium ion transport in mitochondria from cultured human cells , Tsai , Phillips et al . show that EMRE has two key roles . First , it snuggles up against the MCU protein and forms an essential part of the calcium ion-selective pore . Second , it acts as molecular glue to fix the calcium ion-sensing MICU gatekeepers to the pore . These two linked functions ensure that the MCU complex switches on only when the cell contains high levels of calcium ions , preventing the cell becoming catastrophically overloaded with calcium ions and cell death . Challenges for the future are to purify the MCU complex and reconstitute its ability to transport calcium ions from its component parts . This will help to determine the structure of the channel .

2016

Dual functions of a small regulatory subunit in the mitochondrial calcium uniporter complex

Missense mutations in ATM kinase , a master regulator of DNA damage responses , are found in many cancers , but their impact on ATM function and implications for cancer therapy are largely unknown . Here we report that 72% of cancer-associated ATM mutations are missense mutations that are enriched around the kinase domain . Expression of kinase-dead ATM ( AtmKD/- ) is more oncogenic than loss of ATM ( Atm-/- ) in mouse models , leading to earlier and more frequent lymphomas with Pten deletions . Kinase-dead ATM protein ( Atm-KD ) , but not loss of ATM ( Atm-null ) , prevents replication-dependent removal of Topo-isomerase I-DNA adducts at the step of strand cleavage , leading to severe genomic instability and hypersensitivity to Topo-isomerase I inhibitors . Correspondingly , Topo-isomerase I inhibitors effectively and preferentially eliminate AtmKD/- , but not Atm-proficientor Atm-/- leukemia in animal models . These findings identify ATM kinase-domain missense mutations as a potent oncogenic event and a biomarker for Topo-isomerase I inhibitor based therapy . ATM kinase is a tumor suppressor that has a central role in the DNA damage responses . Germline inactivation of ATM causes ataxia-telangiectasia ( A-T ) , which is associated with greatly increased risk of lymphoma and leukemia ( Lavin , 2008 ) . Somatic mutations of ATM occur frequently in mantle cell lymphomas ( MCL ) , chronic lymphoblastic leukemia ( CLL ) and T-cell prolymphocytic leukemia ( TPLL ) ( Wang et al . , 2011; Beà et al . , 2013; Stilgenbauer et al . , 1997 ) . In lymphomas , ATM mutations often occur with concurrent heterozygous deletion of 11q23 including ATM ( Wang et al . , 2011; Beà et al . , 2013; Stilgenbauer et al . , 1997 ) , potentially leading to the expression of mutated ATM in the absence of the wild-type ( WT ) protein . Recent sequencing studies also identified recurrent ATM mutations in 2–8% of breast , pancreas or gastric cancers ( Roberts et al . , 2012; Cremona and Behrens , 2014 ) . While the majority of A-T patients ( ~90% ) have truncating ATM mutations that result in little or no ATM protein expression ( Concannon and Gatti , 1997 ) , missense ATM mutations are more common in cancers and with the exception of the few that cause A-T , their biological functions are unknown . As a serine/threonine protein kinase , ATM is recruited and activated by DNA double strand breaks ( DSBs ) through direct interactions with the MRE11 , RAD50 and NBS1 ( MRN ) complex ( Lee and Paull , 2004; Paull , 2015; Stewart et al . , 1999; Carney et al . , 1998 ) . Activated ATM phosphorylates >800 substrates implicated in cell cycle checkpoints , DNA repair , and apoptosis to suppress genomic instability and tumorigenesis . ATM activation is also associated with inter-molecular autophosphorylation ( Bakkenist and Kastan , 2003; Kozlov et al . , 2011 ) . Studies in human cells suggest that auto-phosphorylation is required for ATM activation ( Bakkenist and Kastan , 2003; Kozlov et al . , 2011 ) . However , alanine substitutions at one or several auto-phosphorylation sites do not measurably affect ATM kinase activity in transgenic mouse models ( Daniel et al . , 2008; Pellegrini et al . , 2006 ) , leaving the biological function of ATM auto-phosphorylation unclear . In this context , we and others generated mouse models expressing kinase dead ( KD ) ATM protein ( Atm-KD ) ( Yamamoto et al . , 2012; Daniel et al . , 2012 ) . In contrast to the normal development of Atm-/- mice , AtmKD/- and AtmKD/KD mice die during early embryonic development with severe genomic instability ( Yamamoto et al . , 2012; Daniel et al . , 2012; Barlow et al . , 1996 ) , implying that the expression of Atm-KD in the absence of WT ATM further inhibits DNA repair beyond the loss of ATM . Among the major DNA DSB repair pathways , non-homologous end-joining ( NHEJ ) is uniquely required for the repair phases of lymphocyte specific V ( D ) J recombination and class switch recombination ( CSR ) . ATM promotes NHEJ ( Zha et al . , 2011b; Bredemeyer et al . , 2006 ) , and Atm-/- mice and A-T patients suffer from primary immunodeficiency ( Jiang et al . , 2015b; Zha et al . , 2011a ) . The end-joining defects in V ( D ) J recombination and CSR are similar in AtmKD/- and Atm-/- lymphocytes , suggesting that Atm-KD protein is not dominant-inhibitory for NHEJ . Meanwhile , AtmKD/- cells show moderate yet significant hypersensitivity to PARP inhibitors in comparison to both Atm-/- and Atm+/+ cells ( Daniel et al . , 2012 ) , and ATM kinase inhibitor , but not loss-of-ATM , reduced homologous recombination ( HR ) as measured by sister chromatid exchange ( SCE ) ( White et al . , 2010 ) and the DR-GFP HR reporter ( Rass et al . , 2013; Kass et al . , 2013 ) , suggesting a role of ATM auto-phosphorylation in replication associated homology dependent repair . Yet , it is unknown how ATM auto-phosphorylation contributes to DNA replication and HR beyond its previously identified signaling roles . Finally , consistent with the 'inter'-molecular autophosphorylation model , Atm+/KD mice are largely normal ( Yamamoto et al . , 2012 ) , suggesting that ATMKD mutation carriers could be asymptomatic and somatic loss of the WT allele in the carriers might create the Mut/Del status reported in human cancers and lymphomas . Here we report that 72% of human cancer-associated ATM mutations are missense mutations that are highly enriched in the kinase domain . We further show that conditional expression of Atm-KD protein alone ( somatic inactivation of the conditional allele ( AtmC ) in AtmKD/C mice to generate the AtmKD/- cells ) in murine hematopoietic stem cells ( HSCs ) is more oncogenic than the complete loss of ATM ( Atm-/- ) . AtmKD/- cells are selectively hypersensitive to Topo Isomerase I ( Topo1 ) inhibitors , in part because the Atm-KD protein physically blocks replication-dependent strand cleavage upon Topo1 inhibition . Correspondingly , Topo1 inhibitor selectively eradicates Notch1-driven AtmKD/- leukemia , but not the isogenic parental Atm proficient leukemia , identifying Topo1 inhibitors as a targeted therapy for human cancers carrying missense ATM kinase domain mutations . Among the 5402 cases in The Cancer Genome Atlas ( TCGA ) , we identified 286 unique non-synonymous mutations of ATM . While truncating ( nonsense/frameshift ) mutations compose 83% ( 373/447 ) of A-T associated point mutations ( >1000 patients ) , 72% ( 206/286 ) of non-synonymous point mutations of ATM in TCGA are missense mutations ( Figure 1A , Supplementary file 1A , B ) . Permutation analyses show that ATM gene is not hyper-mutated , but the kinase-domain is mutated 2 . 5 fold more frequently than otherwise expected in TCGA ( Figure 1—figure supplement 1A , p<0 . 01 ) . The mutation density calculated using the Gaussian Kernel model revealed that cancer associated missense ATM mutations in TCGA cluster around the C-terminal kinase domain , while truncating mutations ( in A-T or TCGA ) span the entire ATM protein ( Figure 1B and Figure 1—figure supplement 1B ) . Given the severe phenotype of AtmKD/- , but not AtmKD/+ cells , we further analyzed the subset ( ~105/286 ) of ATM missense mutations in TCGA that are concurrent with heterozygous loss of ATM ( shallow deletion ) or truncating mutations in the same case , and found that , again , ATM missense mutations cluster around the C-terminal kinase domain even in this smaller subset ( Figure 1B ) . The kinase and FATC domains of ATM share 31% sequence identity with mTOR , a related phosphatidylinositol 3-kinase-related protein kinase ( PIKK ) for which the high resolution crystal structure is available ( Yang et al . , 2013 ) . Homology modeling using mTOR ( PDB 4JSP ) ( Yang et al . , 2013 ) revealed that 64% ( 27/42 ) ( at 18 unique amino acids ) of ATM kinase domain missense mutations from TCGA , affect highly conserved residues and 50% ( 21/42 ) of the mutations ( red on the ribbon structure ) likely abolish kinase activity based on structural analyses ( Figure 1C , Figure 1—figure supplement 1C ) . Specifically , residues K2717 , D2720 , H2872 , D2870 , N2875 and D2889 of human ATM are predicted to bind ATP or the essential Mg+ ion ( Figure 1—figure supplement 1D ) . Notably , N2875 is mutated in two TCGA cases at the time of initial analyses . One of the two cases have concurrent shallow deletion in this region ( Supplementary file 1B ) . Since then , one additional N2875 mutation was reported in a prostate cancer case ( TCGA-YL-A8S9 ) with an allele frequency of 0 . 92 , consistent with homozygosity . Mutations corresponding to N2875K of human ATM were previously engineered into the AtmKD/+ mice together with the corresponding D2870A mutation of human ATM . This combination was found to abolish ATM kinase activity without significantly affecting ATM protein levels ( Yamamoto et al . , 2012; Canman , 1998 ) . Finally , immunoblotting confirmed the expression of catalytically inactive ATM protein in several human cancer cell lines with missense mutations around the kinase domain ( CCLE , Broad Institute ) ( Figure 1—figure supplement 1E–F ) . 10 . 7554/eLife . 14709 . 003Figure 1 . Cancer-associated ATM mutations are enriched for kinase domain missense mutations . ( A ) Classification of unique and nonsynonymous mutations reported for A-T patients ( n > 1000 ) in the Leiden open variable database ( http://chromium . lovd . nl/LOVD2/home . php ? select_db=ATM ) ( Supplementary file 1A ) or unique mutations from 5402 cancer cases in TCGA ( Supplementary file 1B ) . 'Others' includes splice-site and stop codon mutations . ( B ) Heatmaps for the frequency of cancer-associated truncating mutations ( upper panel ) or nonsynonymous substitutions ( missense ) mutations ( lower panel ) along the human ATM polypeptide obtained using Gaussian kernel based analyses ( see Materials and methods ) . Warmer colors indicate denser mutation distribution . The shadowed curves in the middle and the bottom panel reflect the ratio of actual over expected mutation frequencies , calculated by Gaussian kernel method in all TCGA ATM missense mutations ( middle ) or only missense mutations with concurrent shallow deletion/truncating mutations ( bottom ) . The confidence intervals ( shadows ) are estimated by fitting a binomial distribution ( see Materials and methods ) . The dashed line at 1 represents the expected rate of mutations . The x-axis is the amino acid number ( from 1–3057 , including stop code ) along the human ATM protein . ( C ) Homology model of the human ATM kinase domain is shown in two views rotated by 90° . The N-lobe ( aa 2640–2774 ) is shown in green; the C-lobe ( aa 2775–3057 , ) is shown in yellow; the activation loop ( aa 2883–2907 ) is shown in cyan; the FATC domain ( aa 3029–3056 ) is shown in purple and the ka9b ( aa 2942–2956 ) and ka10 ( aa 3002–3026 ) is shown in pink . The αC-helix ( aa 2722–2741 ) , which is a part of the N-lobe and contains residues critical for catalysis is shown in blue . ATP and residues critical for catalysis are shown in the stick model . Mg2+ ions are shown as spheres . The black dotted loops indicate the disordered region between ka9b and ka10 . The amino acids that are conserved between human ATM and mTOR and mutated in TCGA cases are shown in red . DOI: http://dx . doi . org/10 . 7554/eLife . 14709 . 00310 . 7554/eLife . 14709 . 004Figure 1—figure supplement 1 . Cancer associated ATM mutations are enriched for kinase domain missense mutations that disrupt kinase activity . ( A ) Permutation analyses of actual truncation or missense mutation frequencies within the ATM gene or within the ATM kinase domain over the expected mutation frequency based on the size of the gene and the frequency of non-synonymous mutation . The dash line marks the expected mutation rate . ( B ) The density distribution of truncation/frame shift mutations in A-T patients ( see Figure 1B legend and method for more details ) . ( C ) Superposition of the human ATM model ( yellow ) and the crystal structure of human mTOR ( blue ) suggest close structural homology , especially around the catalytic center . Mutations conserved between ATM and mTOR are shown in red . ( D ) The active site architecture of human ATM predicted based on homology modeling is colored as follows: ATP is shown in red , Mg2+ ion is shown in purple , the catalytic residues from ATM are shown in blue and the substrate is colored black . ( E ) Table with the annotated ATM mutations in selected human cancer cells . ( F ) Western blots of total cell lysates harvested from the indicated human tumor lines for ATM ( upper ) , and pKap1 ( lower ) with ( + ) or without ( - ) 5 Gy Irradiation ( IR ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14709 . 004 To determine the oncogenic properties of ATM-KD protein and circumvent the embryonic lethality of the AtmKD/- mice ( Yamamoto et al . , 2012 ) , we generated littermate matched VavCre+AtmC/KD ( VKD ) and VavCre+AtmC/C ( VN , N for null ) mice . In these mice , the HSC-specific Cre recombinase ( VavCre ) ( de Boer et al . , 2003 ) efficiently inactivates the Atm conditional ( AtmC ) allele ( Zha et al . , 2008; Callén et al . , 2009 ) in all blood cells including lymphocytes ( Figure 2—figure supplement 1A ) . VKD and VN mice display defects in B cell CSR and T cell development similar to Atm-/- mice , namely 50% decrease of CSR ( to IgG1 ) , reduced expression of surface-TCRβ/CD3 in CD4+CD8+ double-positive ( DP ) cells , and partial blockage at DP to CD4+CD8- or CD4-CD8+ single-positive ( SP ) T cell transitions ( Borghesani et al . , 2000; Lumsden et al . , 2004; Reina-San-Martin et al . , 2004 ) ( Figure 2—figure supplement 1B–C ) , supporting efficient inactivation of AtmC allele in VKD and VN lymphocytes . Immature T cells with productive TCRβ rearrangements undergo rapid proliferation and expansion in the CD4-CD8- double-negative ( DN ) 3 stage ( CD44-CD25+ ) . VKD thymocytes , but not VN or Atm-/- thymocytes , are partially blocked at the DN3 stage , consistent with proliferation defects in AtmKD/- cell ( Yamamoto et al . , 2012 ) ( Figure 2A–B ) . As a result , thymocyte number in VKD mice was reduced ~60% compared to their VN littermates ( Figure 2—figure supplement 1C ) . 10 . 7554/eLife . 14709 . 005Figure 2 . Expression of ATM-KD protein ( AtmKD/- ) is more oncogenic than loss of ATM ( Atm-/- ) . ( A ) Representative FACS analyses of DN thymocytes from 4-week old VavCre-AtmC/KD ( Ctrl , littermate ) , germline Atm-knockout ( Atm-/- ) , VavCre+AtmC/C ( VN ) and VavCre+AtmC/KD ( VKD ) mice . DN cells are defined as thymocytes negative for surface staining of CD8 , CD4 and TCRγδ . ( B ) The average relative DN3% ( CD44-CD25+ ) ( among all DN thymocytes ) in VN ( n = 3 ) and VKD ( n = 4 ) mice . The relative DN3% was calculated by normalizing to the absolute DN3% to the DN3% of control littermate VavCre- ( cre negative ) mice stained at the same time . The error bars represent SEMs . ( C ) The averages and standard SEMs of total thymocyte number from 4–8 weeks old VN ( n = 3 ) and VKD ( n = 3 ) mice . ( D ) The life time risk of B or T cell lymphomas in VN , VKD mice in 400 days . ( E ) Kaplan-Meier ( K-M ) survival curve of littermate matched VN ( n = 18 ) and VKD ( n = 24 ) mice . The red and blue dots denote thymic lymphomas from the VN and VKD cohorts , respectively . The yellow dots denote the B cell lymphomas in the VKD cohort . The T1/2 for thymic lymphoma is 139 . 5 days ( VN ) and 104 . 0 days ( VKD ) . The asterisk marks the difference in T1/2 of thymic lymphoma development between the two cohorts . *p=0 . 03 per Mantel-Cox/log-rank test . ( F ) Copy number analyses of the region around Pten ( chromosome 19 , mm8: 32 , 779 , 983–32 , 908 , 796 ) . The y-axis is the natural log ratio of tumor/kidney genomic DNA from the same mouse . Log ratio of −0 . 67 ( dotted line ) indicates heterozygous deletion . ( G ) Immunoblot of Pten , total Atm and β-actin for Atm+/+ ( WT Thy ) , VKD , and VN thymic lymphomas . ( H ) Telomere FISH analyses of metaphases spreads from ConA activated T-cells ( 72 hr ) . Bar graphs represent the average and SEMs of the number of cytogenetic aberrations per metaphase ( MP ) . All p-values in this figure were calculated using a two-tailed student’s t-test assuming unequal variances , unless otherwise noted . DOI: http://dx . doi . org/10 . 7554/eLife . 14709 . 00510 . 7554/eLife . 14709 . 006Figure 2—figure supplement 1 . Lymphocyte development in VKD and VN mice . ( A ) Southern Blot for the Atm conditional vs deleted allele in kidney , bone marrow , splenocytes , thymocytes and activated splenic B cells ( CD43- , LPS stimulated ) of a representative VN mouse , indicating complete deletion of Atm in lymphoid lineage . KpnI-digested genomic DNA was probed with ATMCKO3’ probe . ( B ) Representative FACS analyses of bone marrow , thymus and spleen show normal B cell and myeloid development , thymocyte development defects and CSR defects in VN and VKD mice . ( C ) Quantification of IgG1+% of purified B cells stimulated with LPS and IL-4 for 4 . 5 days . ( D ) FACS analyses of thymocytes harvested from VN and VKD mice with thymic lymphomas . The dash line marks the level of surface TCRβ expression in control thymocytes . ( E ) Southern blot analyses of thymocyte DNA digested with EcoRI and probed with TCRJβ1 . 6 and TCRJβ2 . 7 probes . Marker: Molecular Weight Marker . Ctrl: Kidney DNA harvested from a VavCre- AtmC/KD mouse , GL-Germline . DOI: http://dx . doi . org/10 . 7554/eLife . 14709 . 00610 . 7554/eLife . 14709 . 007Figure 2—figure supplement 2 . Analyses of the T cell lymphomas from VKD and VN mice . ( A ) CGH analyses of 4 VKD and 1 VN thymic lymphomas . The Y axis is the Log ratio ( base 2 ) of tumor/kidney DNA from the same mouse . The red arrow heads indicate trisomy 15 and regions amplified on chromosome 14 , and the green arrow heads point to the deletion in chromosomes 12 and the deletion around the Pten gene . Each chromosome is demarcated by gray lines , and the chromosome numbers are marked at the top of the CGH panels . ( B ) Representative FACS analyses of the splenocytes from control and two VKD mice with B-cell lymphomas . ( C ) Southern blot analyses of splenocyte DNA harvested from VKD mice with B-cell lymphomas digested with EcoRI and probed using JH4 , Myc-A , and TCRJβ1 . 6 probes . Ctrl: Kidney DNA harvested from a VavCre- AtmC/KD mouse , GL-Germline . DOI: http://dx . doi . org/10 . 7554/eLife . 14709 . 007 Despite low thymocyte counts , 75% of VKD mice succumbed to lymphomas , representing a 34% increase over the 56% life-time risk among VN mice ( Figure 2D ) . Furthermore , the median survival of thymic lymphoma bearing VKD mice is ~35 days shorter than that of VN mice ( 104 and 139 days respectively , p=0 . 03 ) ( Figure 2E ) . The thymic lymphomas from the VN mice and VKD mice are both clonal , immature ( TCRβ/CD3low ) ( Figure 2—figure supplement 1D–E ) with frequent trisomy 15 , have focal amplifications around the TCRα/δ loci , and have hemizygous deletion of the telomeric portion of chromosome 12 , as is the case for previously characterized Atm-/- thymic lymphomas ( Jiang et al . , 2015b; Zha et al . , 2010 ) ( Figure 2—figure supplement 2A ) . About 8% of the VKD , but none of the VN mice , also developed clonal pro-B cell lymphomas ( B220+CD43+IgM- ) ( Figure 2—figure supplement 2B–C ) . Comparative genome hybridization ( CGH ) showed that all VKD thymic lymphomas , but none of the VN controls , carried deep deletions at the Pten locus ( Figure 2F and Figure 2—figure supplement 2A ) . Immunoblotting of additional lymphomas revealed severe reductions of Pten protein levels in all but one of the VKD tumors , while most VN tumors retained Pten expression . The only Pten-expressing VKD tumor ( 6684 ) appeared to have lost Atm-KD expression ( Figure 2G ) . Pten deletion was previously found in ~25% of germline Atm-/- or Tp53-/- thymic lymphomas and was thought to occur in the early progenitors before T cell commitment ( Zha et al . , 2010; Dudgeon et al . , 2014 ) . Based on our findings , we propose that expression of Atm-KD protein further increases genomic instability in lymphoid progenitors/HSCs , eventually led to frequent Pten deletion and early lymphomas in the VKD mice . Consistent with this model , frequency of chromatid breaks are significantly higher in VKD T cells , than in VN or Atm+/+ T cells ( Figure 2H ) . We hypothesized that the additional DNA repair defects in AtmKD/- cells would confer hypersensitivity to certain genotoxic chemotherapy that could be used to target ATM-mutated cancers . To test this , we derived Rosa+/CreERT2Atm+/C , Rosa+/CreERT2AtmC/- and Rosa+/CreERT2AtmC/KD murine embryonic fibroblasts ( MEFs ) , in which 4-Hydroxytamoxifen ( 4-OHT , 200 nM ) induces nuclear translocation of ER-fused Cre-recombinase and effectively inactivates the AtmC allele to generate Atm+/- , Atm-/- and AtmKD/- MEFs ( Figure 3—figure supplement 1A ) ( Yamamoto et al . , 2012 ) . All the experiments were repeated ≥3 times in several independently derived and freshly deleted cells to avoid secondary alterations in genomically unstable Atm-deficient cells . While both AtmKD/- and Atm-/- MEFs were similarly sensitive to ionizing radiation ( IR ) , AtmKD/- cells were significantly more sensitive to the Topo1 inhibitor camptothecin ( CPT ) than Atm-/- and Atm+/- cells ( Figure 3A–B ) . The CPT hypersensitivity of AtmKD/- cells does not appear to reflect an inability to process blocked DNA ends or overcome replication blockage in general , as AtmKD/- cells were not hypersensitive to etoposide , a Topoisomerase II inhibitor , nor to two replication blocking agents , hydroxyurea ( HU ) or aphidicolin ( APH ) , ( Figure 3C and Figure 3—figure supplement 1B ) . Moreover , AtmKD/- primary T cells and the human cancer cell lines that express ATM-KD protein identified in Figure 1—figure supplement 1F are also hypersensitive to CPT compared to corresponding ATM+/+ or Atm-/- controls ( Figure 3D and Figure 3—figure supplement 1E ) . Finally , shRNA knockdown of mutated ATM in Granta519 human lymphomas cell lines , moderately , yet significantly , rescued the CPT hypersensitivity and CPT induced apoptosis associated with the expression of catalytically inactive ATM ( Figure 3E and Figure 3—figure supplement 1C and D ) . 10 . 7554/eLife . 14709 . 008Figure 3 . AtmKD/- murine embryonic fibroblasts ( MEFs ) and human cancer cell lines are hypersensitive to Topo isomerase I inhibitors in vitro . Representative sensitivity plots of Atm+/- , Atm-/- and AtmKD/- MEFs to ( A ) IR , ( B ) Camptothecin ( CPT ) and ( C ) Etoposide . *p<0 . 001 . At least three independent experiments on two independently derived MEF lines from each genotype were performed . ( D ) Sensitivity of 6 human tumor cell lines with wildtype ( ATM+/+ ) or mutant ATM ( ATMMut/- or Mut/Mut ) to 2 . 5 nM CPT observed after 48 hr of treatment . ( E ) shRNA mediated knockdown of ATM in Granta591 human lymphoma cell lines rescued the CPT sensitivity in vitro ( after 72 hr of 5 nM CPT treatment ) . Three independent experiments were performed . The p value were obtained based on student t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 14709 . 00810 . 7554/eLife . 14709 . 009Figure 3—figure supplement 1 . Analyses of immortalized AtmKD/- and control MEFs . ( A ) Treatment scheme of immortalized Rosa-CreERT2 MEFs . PCR and Western blot analyses of the MEFs before and after three rounds of 4OHT treatment ( 200 nM , 48 hr each round ) confirmed the efficient ATM deletion and the expression of kinase-dead ATM protein . The Western blot also shows the normal expression of DNA-PKcs , Mre11 and Kap1 in the AtmKD/- cells . ( B ) Representative result of the sensitivity assay for Hydroxyurea and Aphidocolin . See method section for details . ( C ) Representative FACS analyses for apoptotic cells ( PI staining ) in HBL2 and Granta591 cells with or without shRNA knockdown of ATM . ( D ) CPT ( 5 nM , 48 hr ) induced apoptosis in HBL2 and Granta591 cells measured by PI staining . ( E ) Relative survival of ConcanavalinA ( ConA ) -activated purified ( CD43+ ) splenic T-cells to Olaparib ( PARPi , 1 µM ) or CPT ( 0 . 1 µM ) , or the combination ( PARPi+CPT ) . Cells were stimulated for 24 hr and exposed to drugs for another 48 hr and relative survival was determined by MTT assay . DOI: http://dx . doi . org/10 . 7554/eLife . 14709 . 009 In animal models , irinotecan , a clinical Topo1 inhibitor , eradicated activated-Notch1 ( Notch1-ΔE-IRES-GFP ) driven AtmKD/- T cell leukemia , but not the isogenic parental AtmKD/C leukemia in vivo , determined by the absence of GFP+ leukemia cells in the spleen , reduced spleen size and weight , lack of infiltrated leukemia blasts and increased TUNEL+ apoptotic cells ( Figure 4A–F and Figure 4—figure supplement 1A–B ) . When similar experiments were performed on isogenic AtmC/C and corresponding Atm-/- leukemia , irinotecan has no significant benefit in either groups ( Figure 4—figure supplement 1C–F , p≥0 . 05 in all pairs ) , suggesting loss of ATM-mediated DNA damage responses alone is not sufficient to explain the hypersensitivity of AtmKD/- leukemia to Topo1 inhibitors . Together these findings identified the expression of ATM-KD as a biomarker for Topo1 inhibitors sensitivity . 10 . 7554/eLife . 14709 . 010Figure 4 . AtmKD/- leukemia are hypersensitive to Topo isomerase I in vivo . ( A ) Experimental scheme outlining the treatment of isogenic AtmKD/- and AtmKD/C leukemia in vivo . Please see on-line methods for details . Clonal Rosa+/CreERT2 AtmC/KD leukemia was established by transducing the bone marrow with activated Notch and GFP . Isogenic AtmKD/- or AtmKD/C leukemia were then established and confirmed in secondary recipients upon exposure to oral tamoxifen , which induces Cre nuclear translocation . 1 × 10^6 AtmKD/- and AtmKD/C leukemic blasts were injected into tertiary recipients , which were treated with daily doses of Irinotecan ( 10 mg/kg ) or vehicle for 5 consecutive days . Representative FACS analyses for GFP+ leukemia blast ( B ) and pictures ( D ) of spleens from mice transduced with isogenic AtmKD/C or AtmKD/- leukemia and treated or not treated with irinotecan ( Irino , 10 mg/Kg for 5 days ) . The untreated mice received vehicle ( Veh ) . Quantification ( Average ± SEMs ) of GFP+% splenocytes ( C ) and spleen weights ( E ) of mice transduced with AtmKD/C or AtmKD/- leukemia and treated with either vehicle ( black ) or Irinotecan ( grey ) for 5 days . ( F ) Representative field of H&E ( left 2 panels ) and TUNEL-stained spleen sections Scale Bars: 200 µm ( in H&E section ) , 20 µm ( TUNEL ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14709 . 01010 . 7554/eLife . 14709 . 011Figure 4—figure supplement 1 . Analyses of isogenic paired AtmKD/- vs AtmKD/C or Atm-/- vs AtmC/C leukemia . ( A ) PCR confirming the AtmKD/- and AtmKD/C genotypes of the leukemia harvested from secondary recipients after the tamoxifen treatment . ( B ) Southern blot using the Jβ2 . 7 probe on AtmKD/- and AtmKD/C tumors harvested after vehicle ( V ) or Irinotecan ( Iri ) treatment . *denoted degraded DNA . Spleens from Irinotecan treated mice with AtmKD/- leukemia had very few leukemia cells and only the germline TCRβ bands . ( C ) Representative spleens harvested from tertiary recipients transduced with isogeneic AtmC/C or Atm-/- leukemia treated with either Vehicle ( Veh ) or Irinotecan . Quantification ( Average ± SEMs ) of GFP+% splenocytes ( D ) and spleen weights ( E ) of mice transduced with AtmC/C or Atm-/- leukemia and treated with either vehicle ( black ) or Irinotecan ( grey ) for 5 days . ( F ) Representative fields of H&E stained spleen sections . Scale Bars: 200 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 14709 . 011 Topo1 inhibitors prevent DNA-religation by Topo1 , and trap Topo1 in a covalent DNA complex ( Top1cc ) at the 3’-end of single-strand DNA ( ssDNA ) nicks generated by Topo1 cleavage ( Pommier , 2006 ) . If not removed , Top1cc interferes with transcription and replication to elicit cytotoxic effects . We found that AtmKD/- , but not Atm-/- or Atm+/+ MEFs accumulated Top1cc even at low CPT concentrations ( 0 . 1 µM ) ( Figure 5A and Figure 5—figure supplement 1A and B ) . ATM kinase inhibitor ( ATMi , KU55933 , 15 µM ) also leads to CPT-dependent Top1cc-accumulation in Atm+/- MEFs ( Figure 5A ) . At high CPT concentration ( 15 µM ) , the Top1cc level is greatest in AtmKD/- cells , followed by Atm-/- , then Atm+/+ ( Figure 5A and Figure 5—figure supplement 1A and B ) . Meanwhile , CPT-induced degradation of full-length Topo1 , necessary for its subsequent removal ( Pommier , 2006 ) , is not significantly affected in AtmKD/- or Atm-/- MEFs ( Figure 5B ) . Tyrosyl-DNA phosphodiesterase 1 ( TDP1 ) or 3’ flapase ( e . g . XPF ) can remove Top1cc independent of DNA replication ( Pommier , 2006 ) ( Figure 5C ) . In quiescent neurons , ATM promotes transcription-dependent removal of Top1cc potentially by facilitating Topo1 degradation independent of its kinase activity or MRN ( Alagoz et al . , 2013; Katyal et al . , 2014 ) . While this function of ATM might contribute to the moderate accumulation of Top1cc in CPT treated Atm-/- MEFs at high CPT concentrations ( Figure 5A ) , the kinase-dependent and protein-dependent role of ATM in Top1cc removal revealed in proliferating AtmKD/- MEFs appears to be independent of Topo1 degradation ( Figure 5B ) . In this context , a recent study using Xenopus extract suggests that collision of the replication forks with protein conjugated to DNA would lead to rapid and robust proteolytic degradation of the conjugated proteins , which might mask the moderate effects of ATM in Top1cc-proteolytic degradation described in none proliferative neuronal cells ( Duxin et al . , 2014 ) . During replication , it is proposed that the collision of the DNA replication forks with Top1cc blocks the replication on the involved strand , leads to uncoupling of the leading strand and lagging strand syntheses and fork reversal . Eventually a subset of the stalled replication forks are cleaved to remove Top1cc ( as diagrammed in Figure 5C ) . The resulting single-ended DSBs generated from fork cleavage can only be repaired by HR , not NHEJ . Resolution of the holiday junction intermediates during HR generates crossover ( CO ) , thus CPT is known to induce SCEs ( Figure 5C ) . CPT ( 1 µM ) induced SCEs were reduced in AtmKD/- MEFs , but not Atm-/- MEFs , and ATM inhibitor decreased CPT-induced SCEs in Atm+/- , but not in Atm-/- MEFs ( Figure 5D ) , suggesting a defect at replication-dependent removal of Top1cc in the AtmKD/- cells . ATM kinase inhibitor did not suppress SCEs in Blmhelicase-deficient ( Blm-/- ) MEFs , suggesting that the lack of SCEs could not be simply explained by the bias in CO-generating holiday junction resolvases ( e . g . MUS81 , SLX1/4 , GEN1 ) ( Figure 5—figure supplement 1C ) ( Chan and West , 2014 ) . 10 . 7554/eLife . 14709 . 012Figure 5 . Accumulation of Top1cc and reduced homologous recombination in AtmKD/- cells . ( A ) Representative In-vitro Complex of Enzymes ( ICE ) assay ( see Materials and method ) . The top panel was blotted by anti-Topo1 antibody and the bottom panel was probed by p32 labeled total mouse genomic DNA . The amounts of genomic DNA ( in μg ) loaded on each row are marked on the left . ( B ) Western blot for full length Topo1 on MEFs treated with indicated concentrations of CPT for 2 hr . ( C ) Diagram of replication-dependent ( blue shade ) and replication-independent ( pink shade ) Top1cc removal pathways . Top1cc is removed by TDP1 or 3’ flapases to generate a single stand gap , which can be repaired by PARP1 and Ligase3-mediated pathways independent of replication . In replicating cells , replication forks collide with Top1cc , lead to uncoupling of the leading and the lagging strand , accumulation of single strand DNA ( blue dashed box ) , which promotes fork reversal and eventually a subset of the forks were cleaved . The stalled fork could also be directly converted to breaks ( top row , dash arrow ) . The resulting single-ended DSBs ( grey shade circle ) can only be repaired by HR ( through end resection , homology search and resolution ) to generate cross over ( CO ) ( scored as SCEs ) or non–crossover . ATM-KD protein selectively inhibits fork breakage in a MRN dependent manner , and could also potentially suppress the same nucleases ( Mus81/SLX4 ) implicated in holiday junction resolution in a later step of HR . ( D ) Representative images and the average ( and SEMs ) of SCEs per chromosome for vehicle ( DMSO ) or CPT ( 1 µM ) treated MEFs . Red diamonds point to exchange events . ( E ) Diagram of the DR-GFP reporter , representative FACS plots of the GFP+ cells 48 hr after I-SceI transfection . The bar graphs represent the adjusted GFP% , calculated based on raw GFP% and the I-SceI transfection efficiency . At least two independent ES cell lines per genotype were assayed in >3 independent experiments . Bar graphs represent the average ± SEMs . In the right panel , the cells were either treated with vehicle ( DMSO ) or ATMi ( 15 µM Ku55933 ) for 36 hr ( 12 hr after I-SceI transfection ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14709 . 01210 . 7554/eLife . 14709 . 013Figure 5—figure supplement 1 . Analyses of immortalized AtmKD/- MEFs and Mre11-deficient cells . ( A and B ) Two independent In-vitro Complex of Enzymes ( ICE ) assay results ( see Materials and method ) . The top panel was blot by anti-Topo1 antibody and the bottom panel was probed by P32 labeled total mouse genomic DNA . The amounts of genomic DNA ( in μg ) loaded on each row are marked on the left . The ** lanes represents Atm+/- cells treated with 15 μM CPT that were lost during centrifugation due to an equipment error . ( C ) The number of sister chromatid exchange ( SCE ) per chromosome in Blm+/+ ( WT ) or Blm-/- cells treated with vehicle ( DMSO ) or ATMi ( 15 µM KU55933 for second doubling period ) for 24 hr . DOI: http://dx . doi . org/10 . 7554/eLife . 14709 . 013 In this context , an integrated ( at the Pim1 locus ) DR-GFP reporter ( Moynahan et al . , 1999 ) revealed a 50% reduction of HR in AtmKD/- embryonic stem ( ES ) cells , but not in Atm-/- cells . ATM inhibitor reduced HR in Atm+/+ , but not Atm-/- ES cells ( Figure 4D ) ( Rass et al . , 2013; Kass et al . , 2013 ) , consistent with potential HR defects . End resection that reveals single strand DNA ( ssDNA ) and loading of Rad51 protein that is necessary for homology search are two early events required for HR . Yet IR induced phosphorylated-RPA ( T21 ) and Rad51 foci , were not significantly compromised in AtmKD/- cells ( Figure 6A and Figure 6—figure supplement 1A ) ( Shakya et al . , 2011 ) . These findings imply that Atm-KD does NOT block RAD51 loading to resected DSBs and suggest that either a later step of HR is affected in AtmKD/- cells or other mechanisms exist to explain the lack of CPT-induced SCEs in AtmKD/- cells . In this regard , we found that CPT-induced Rad51 foci were markedly decreased in AtmKD/- MEFs ( Figure 6B ) . Moreover CPT-induced DSBs measured by phosphorylated H2AX ( γ-H2AX ) and the neutral comet assay were also attenuated in AtmKD/- cells ( but not Atm-/- cells ) and by ATM kinase inhibitor ( Figure 6C–E ) . In the same cells , CPT effectively increased alkaline comet tails , an indicator of both single and double stand breaks , in Atm+/- , Atm-/- and AtmKD/- cells ( Figure 6F ) . Together these findings indicate that ATM-KD protein suppressed CPT-induced DSBs formation during replication and thereby reduced SCEs . 10 . 7554/eLife . 14709 . 014Figure 6 . ATM-KD blocks CPT-induced double stand breaks formation in replicating cells . ( A ) Representative images and the frequency of cells with >5 Rad51 foci per cell ( Average ± SEMs ) at 10 hr after 5 Gy of IR . Brca1SF/SF cells homozygous for the S1598F mutation ( Shakya et al . , 2011 ) were used as a control for the lack of Rad51 foci . Scale bar ~10 µm . ( B ) The proportion of cells with >5 Rad51 foci after CPT ( 0 . 1 µM , 2 hr ) treatment . The bar graph represents the average and SEMs from 3 independent experiments . All p-values in this figure were calculated based on two-tailed student’s t-test assuming unequal variances . ( C ) Representative images and quantification of CPT-induced γH2AX foci ( 0 . 1 µM , 2 hr ) . The dot plot represents average and SEMs in three biological replicates with >200 nuclei per genotype per condition . Scale bar ~10 µm . ( D ) Western blots for pRPA ( T21 ) , γH2AX , and total H2AX of cells treated with either Vehicle ( − , DMSO ) or CPT ( + , 0 . 1 µM ) for 2 hr . ATM inhibitor ( 15 µM , Ku55933 ) was added 1 hr prior to CPT/vehicle treatment as indicated . ( E ) Quantification of Neutral COMET tail lengths in cells treated with Vehicle ( − , DMSO ) or CPT ( + , 0 . 1 µM ) for 1 hr . The graph represents the average ± SEMs from two independent experiments with over 50 comets quantified for each . ( F ) Quantification of Alkaline COMET tail lengths of cells treated with Vehicle ( − , DMSO ) or CPT ( + , 0 . 1 µM ) for 1 hr . Two replicates of the experiments were performed with over 50 comets quantified for each . DOI: http://dx . doi . org/10 . 7554/eLife . 14709 . 01410 . 7554/eLife . 14709 . 015Figure 6—figure supplement 1 . Analyses of cell cycle dependent reduction of CPT induced DSBs . ( A ) Proportion of cells with greater than 5 pRPA ( pT21 ) foci 2 hr post 5 Gy IR treatment . ( B ) Representative FACS plots for Bromodeoxyuridine ( BrdU ) and Propidium Iodide ( PI ) of ATM+/- MEFs growing at log phase ( ctrl ) , undergoing double thymidine ( DT ) Block ( G1 ) , and cells released into fresh medium for 6 hr after wash-off ( S/Release ) . For DT block , cells were treated for two intervals of 24 hr with 2 mM Thymidine , with a 9 hr wash-off in between with fresh medium . Western blot of whole-cell lysates collected from MEFs treated with indicated concentrations of CPT for 2 hr . DOI: http://dx . doi . org/10 . 7554/eLife . 14709 . 015 Upon DNA DSBs , Mre11 , along with the MRN complex recruits ATM to the DSBs and activates ATM kinase activity . Notably , Mre11 also binds to ssDNA and the affinity of Mre11 to ssDNA is higher than to dsDNA and is independent of DNA ends ( de Jager et al . , 2001; Usui et al . , 1998; Paull and Gellert , 1999 ) . Extended ssDNA ( potentially without ends ) is likely the structure that accumulated at stalled and uncoupled replication forks ( Figure 5C ) . Correspondingly , recent iPOND experiments have identified MRE11 at the stalled replication fork in vivo ( Sirbu et al . , 2011 ) . In this context , we found that ATM kinase inhibitor prevents CPT induced DSBs formation in Mre11+/- but not in Mre11-/- MEFs . The presence of a nuclease deficient Mre11 ( Mre11nuc/- ) that binds and recruits ATM ( Buis et al . , 2008 ) is sufficient to abolished CPT induced DSBs in the presence of ATM kinase inhibitor ( Figure 7A , Figure 7—figure supplement 1A–C ) . Together these findings support a model , in which ATM-KD protein physically blocks CPT induced DSBs formation at replication forks upon recruitment by MRN . 10 . 7554/eLife . 14709 . 016Figure 7 . Atm-KD physically blocks the cleavage of CPT-stalled replication fork independent of fork reversal . ( A ) Western blots of Mre11+/- , Mre11-/- or Mre11H129N/- ( refereed as Nuc/- in the text ) MEFs treated with CPT ( C , 0 . 1 µM , 2 hr ) or pretreated with ATM kinase inhibitor ( CA , ATMi 15 µM , 1 hr , then together with CPT 0 . 1 µM , 2 hr ) . ( B ) Visualization and quantification of ssDNA at replication fork junctions in not treated ( NT ) cells and upon treatment with CPT with and without the ATM inhibitor ( 15 µM ) . Statistical analysis t-test according to Mann–Whitney results are *p≤0 . 1; **p≤0 . 01; ***p≤0 . 001 . ( C ) Representative images and quantification of fork reversal by electron microscopy upon CPT ( 25 or 100 nM ) and CPT+ATMi ( Ku55933 , 15 µM ) treatment . P: parental DNA; D: daughter DNA strands , and R: reversed fork DNA . 70 replication forks were quantified for each genotype at each condition . Statistical analysis t test according to Mann–Whitney results are *p≤0 . 1; **p≤0 . 01; ***p≤0 . 001 . ( D ) Representative images and quantification of the DNA fiber assay . Cells were incubated 30 min with 25 µM IdU , followed by 250 µM of CldU along with either vehicle ( DMSO ) or CPT ( 1 µM ) . The dot plot represents the summary of two independent experiments . *p<0 . 0001 . DOI: http://dx . doi . org/10 . 7554/eLife . 14709 . 01610 . 7554/eLife . 14709 . 017Figure 7—figure supplement 1 . Replication fork analyses in cells treated with CPT . ( A ) Mre11+/C , Mre11C/- and Mre11C/H129N immortalized MEFs were stably infected with retrovirus expressing ER-CRE- IRES-hCD2 , MACS purified for hCD2+ cells and then induced with 4OHT ( 200 nM ) for 48 hr . PCR confirmed the efficient conversion of the Mre11 conditional allele to the null alleles . PCRs were performed as previously described ( Buis et al . , 2008 ) . ( B and C ) Independent Western blot analyses of Mre11+/- , Mre11-/- or Mre11H129N/- MEFs treated with CPT ( C , 0 . 1 µM , 2 hr ) or pretreated with ATM kinase inhibitor ( CA , ATMi 15 µM , 1 hr , then together with CPT 0 . 1 µM , 2 hr ) . ( D ) Western blot analyses of Atm+/- , Atm-/- or AtmKD/- MEFs treated with CPT ( C , 0 . 1 µM , 2 hr ) or pretreated with Mre11 nuclease inhibitors ( CM for Mirin at 500 µM , or CP for PFM01 ( Shibata et al . , 2014 ) ( 200 µM , for 1 hr ) then with CPT ( 0 . 1 µM ) together for 2 more hours . ( E ) Western blot analyses of cells treated with Aphidicolin ( A ) or CPT ( C ) or first Aphidicolin for 1 hr , then together with CPT for 2 hrs ( AC ) . ( F ) Representative sensitivity assay to mitomycin C ( MMC ) . ( G ) Western blot analyses of immortalized MEFs treated with CPT ( C ) or CPT and ATM kinase inhibitor ( CA ) together for 2 hr . DOI: http://dx . doi . org/10 . 7554/eLife . 14709 . 017 Collision with Top1cc uncouples the progression of the leading and lagging strands , and causes accumulation of ssDNA ( Figure 5C ) , which were thought to be the precursor of 'fork reversal' that converts a replication fork into a four-way junction ( Figure 4—figure supplement 1A ) ( Zellweger et al . , 2015; Berti et al . , 2013 ) . Fork reversal slows replication . In one hand , fork reversal could allow the replication machinery to bypass the lesion using the newly synthesized stand as the template . On the other hand , the four-way junctions , if they persist , could be recognized by structural specific nucleases ( e . g . MUS81 , SLX1/4 ) implicated in CPT induced DSBs ( Regairaz et al . , 2011 ) . In this regard , we reported that CPT induced RPA phosphorylation at Thr 21 , a marker for ssDNA and ssDNA induced CHK1 phosphorylation by ATR are normal in AtmKD/- cells and ATM kinase inhibitor-treated Atm+/- cells ( Figure 6D , Figure 6—figure supplement 1B , Figure 7—figure supplement 1E ) . Moreover , we found that ATM kinase inhibitor did not alter the frequency or size of CPT induced ssDNA , or the frequency of actual fork reversal in Atm+/- cells measured by electron-microscopy ( Zellweger et al . , 2015 ) ( Figure 7B and C ) . Correspondingly , CPT significantly reduced fork velocity in AtmKD/- as well as control Atm+/+ or Atm-/- cells ( Figure 7D ) , consistent with fork reversal . We noted that baseline fork progression is slowest in AtmKD/- cells ( Figure 7D ) , potentially reflecting defects in resolving spontaneous Top1cc or related lesions , such as Topo1-processed ribonucleotide mis-incorporation ( Kim et al . , 2011 ) . Together , our findings support a model , in which ATM is recruited to stalled replication fork by MRE11 ( or MRN complex ) , where ATM kinase activity is necessary to release ATM and allow fork cleavage without affecting fork uncoupling and fork reversal . This role of ATM at the replication fork explains the increased genomic instability , especially chromatid breaks , and hypersensitivity to CPT in murine cells as well as human cancer cells that express catalytically inactive ATM protein . ATM is a tumor suppressor gene that is frequently inactivated in human cancers . While previous studies have mostly focused on the complete loss of ATM ( Atm-/- , truncating mutations ) , recent sequencing analyses identified a large number of missense ATM mutations in human cancers with limited information on their biological function . Here we report that cancer-associated ATM missense mutations are highly enriched in the kinase domain . Using murine models , we showed that expression of ATM-KD causes cancer more frequently and rapidly than loss of ATM , and the AtmKD/- cancers are selectively hypersensitive to Topo1 inhibitors . As such , Topo1 inhibitors could potentially be a targeted therapy for tumors with ATM kinase domain mutations . Mechanistically , we found that ATM-KD physically blocks CPT-induced DSBs formation during replication in a MRN-dependent manner . Given that both AtmKD/- cells and ATM kinase inhibitor treated Atm+/- cells display defects in replication-dependent Top1cc removal , our data revealed not only a neomorphic function associated with kinase domain point mutations of ATM , but also a previously unappreciated auto-phosphorylation dependent structural function of normal ATM that is not apparent in Atm-/- cells ( due to the lack of both ATM protein and ATM kinase activity ) . This new function of ATM expands the repertoire through which ATM functions as a tumor suppressor and has implications on DNA repair and cancer therapy . Despite the ~two fold increased breast cancer risk in heterozygous ATM mutation carriers ( Renwick et al . , 2006; Ahmed and Rahman , 2006 ) and 1–2% carrier rate in Caucasians , truncating ATM mutations are exceedingly rare ( <1/400 , in contrast to the predicted 8% ) in early-onset breast cancers patients ( FitzGerald et al . , 1997 ) . This and other findings ( Spring et al . , 2002 ) led to the speculation that ATM mutations are different in cancer compared to A-T ( Gatti et al . , 1999 ) . Here we reported that , while Atm+ ( C ) /KD mice are normal ( Yamamoto et al . , 2012 ) , somatic inactivation of the AtmC allele in VKD mice led to aggressive lymphomas . As such , while missense ATM mutations in the kinase domain are not compatible with embryonic development when homozygous and are not found in A-T , they could predispose carriers to cancers , and maybe have more potent oncogenic potential than A-T causing truncating mutations . Notably , loss of ATM preferentially affects NHEJ ( Franco et al . , 2006 ) and predisposes patients primarily to lymphomas . Here we show that AtmKD also compromises HR , a pathway implicated in breast , ovarian , pancreatic and prostate cancers , in which recurrent ATM missense mutations were recently identified ( Roberts et al . , 2012; Cremona and Behrens , 2014 ) . MRE11 binds to ssDNA independent of DNA ends in vitro ( de Jager et al . , 2001; Usui et al . , 1998; Paull and Gellert , 1999 ) and MRE11 has been identified at stalled replication forks in vivo ( Sirbu et al . , 2011 ) . Our results also revealed that the ATM-KD protein physically blocks strand cleavage in a MRN-dependent manner , further suggesting the level of mutant ATM protein and MRN status can modify both the cancer risk and therapeutic responses . ATM kinase inhibitor seems to reduce CPT-induced DSBs more efficiently in Mre11Nuc/- cells , than in Mre+/- cells ( Figure 7A , Figure 7—figure supplement 1A–C ) . This result suggests that Mre11 nuclease activity might contribute to the removal of ATM or the termination of ATM activation . Further investigation is likely needed to clarify this issue . We noted that loss of sae2 , a modulator of Mre11 nuclease activity in yeast , causes persistent DNA damage responses mediated by tel1/mec1 , yeast orthologs of ATM and ATR . Two recent studies have identified mre11 mutations that rescued the hyper activation of mec1 in sae2 deficient yeast . ( Chen et al . , 2015; Puddu et al . , 2015 ) . Mre11 nuclease might also contribute to fork cleavage directly . In this context , we noted that Mre11 nuclease inhibitors – either PFM01 or Mirin ( Shibata et al . , 2014 ) , reduced CPT induced γ-H2AX in both Atm+/- and Atm-/- cells , supporting an ATM independent function of Mre11 in fork cleavage ( Figure 7—figure supplement 1D ) . However , given the toxicity associated with Mre11 nuclease inhibitors , further experiments are warranted to understand the molecular details of MRE11 function in fork cleavage . Together , these findings support our model in which MRE11 , together with the MRN complex recruits ATM to the stalled replication forks , while the kinase activity of ATM is important for fork cleavage in the presence of ATM protein . The mechanism identified here also suggests that AtmKD/- tumors should be hyper-sensitive to agents targeting genes involved in Top1cc removal/repair ( e . g . PARPs , TDP1 ) ( Pommier , 2009; Murai et al . , 2014 ) andgenotoxic agents that require similar 'strand-cleavage' mechanisms for repair ( e . g . crosslink agents –like Mitomycin C ) ( Zhang and Walter , 2014 ) . Consistent with these predictions , AtmKD/- T cells are hypersensitive to PARP inhibitor alone or in combination with CPT ( Figure 3—figure supplement 1E ) and AtmKD/- , but not Atm-/- MEFs , are hypersensitive to Mitomycin C ( Figure 7—figure supplement 1F ) . Together , these findings further expand the therapeutic options for ATM mutated cancers . The similarity between AtmKD/- cells and ATM kinase inhibitor treated Atm+/- ( + ) cells further suggest that ATM inhibitors , which have recently entered clinical trials , could potentially increase the efficacy of Topo1 inhibitors in ATM+/+ cancers . Finally , the lack of CPT-induced γ-H2AX described for AtmKD/- cells could be used as a biomarker to assess the function of ATM mutations , including the potential confounding MRN status . While the lack of CPT-induced DSBs might explain the lack of CPT-induced Rad51 foci and SCEs in Atm-/KD cells , the reason for which I-SceI nuclease induced 'clean' DSBs could not be efficiently repaired by HR in the AtmKD/- cells is not apparent . The efficient accumulation of IR induced Rad51 foci in Atm-/KD cells , suggests that Atm-KD does not block the loading of Rad51 to resected DSBs , an early step required for HR ( Figure 6 ) . However , Rad51 loading or filament stability at a stalled replication fork might also be regulated differently from those at resected DSBs ( from IR ) and might occur independent of DSBs ( Schlacher , 2011 ) . In any case , our findings suggest that ATM-KD probably suppresses a later step of HR , especially holiday junction resolution , because the same structure specific nucleases ( MUS81 and SLX1/4 ) , are implicated in both holiday junction resolution and strand cleavage during Interstrand Crosslink ( ICL ) or Top1cc repair ( Zhang and Walter , 2014; Rouse , 2009 ) . The lack of MUS81 or SLX1/4 function could explain both the HR defects and the Top1cc removal defects in Atm-/KD cells . Accordingly , we found that Slx4-/- and , to a lesser extent Mus81-/- , MEFs also display reduced numbers of CPT-induced DSBs ( Figure 7—figure supplement 1G ) . In addition , the slow replication fork progression in Atm-/KD cells ( Figure 7D ) might also indirectly reduce HR efficiency measured by DR-GFP . Intermolecular autophosphorylation is a common feature of ATM , DNA-PKcs and ATR ( Liu et al . , 2011; Dobbs and Tainer , 2010; Jiang et al . , 2015a; Bakkenist and Kastan , 2003 ) . We previously reported that DNA-PKcs recruited by Ku to DSBs block NHEJ in the absence of autophosphorylation ( Jiang et al . , 2015a ) . Here , ATM-KD recruited by MRE11 blocks replication fork cleavage upon collision with Top1cc in a remarkably similar manner , suggesting that autophosphorylation of PIKKs might have a common function in promoting their release from their respective activation partners ( Ku for DNA-PKcs , MRN for ATM ) and license distinct repair events ( NHEJ for DNA-PKcs , strand cleavage for ATM ) . In the AtmKD/- cells , the defects in Top1cc removal combined with the lack of ATM kinase activity-mediated DNA damage response leads to severe genomic instability , embryonic lethality , and enhanced oncogenic potential . ATM somatic mutations were collected from 5 , 402 cancer patients from 24 tumor types in TCGA , including ACC , BRCA , CHOL , COADREAD , ESCA , HNSC , KIRC , LAML , LIHC , MESO , PAAD , PRAD , SARC , STAD , THYM , UCS , BLCA , CESC , COAD , DLBC , GBM , KICH , KIRP , LGG , LUAD , OV , PCPG , READ , SKCM , TGCT , UCEC , and UVM . All maf files of those tumor types were downloaded using firehose ( http://gdac . broadinstitute . org ) . To unify the annotation format , we first used the liftover tool ( https://genome . ucsc . edu/cgi-bin/hgLiftOver ) to map all coordinates to hg19 , and then re-annotated variant effect with VEP ( http://www . ensembl . org/info/docs/tools/vep/ ) under transcript NM_000051 . 3 . Mutation types Missense_Mutation , In_Frame_Del , and In_Frame_Ins were classified as missense , while Nonsense_Mutation , Frame_Shift_Del , Frame_Shift_Ins , and Splice_Site were classified as truncating mutations . Germline ATM mutations are from A-T patients in the Leiden open variable database LOVD2 ( http://chromium . lovd . nl/LOVD2/home . php ? select_db=ATM ) . Protein mutation position was curated and annotated based on the mutation effect provide by LOVD2 . Start_Codon , Frame_Shift_Del , Large_DEL , Nonsense_Mutation , Frame_Shift_Ins , Splice_Site , and Stop_Codon are labeled as Truncating in our analysis . To measure the mutation density of different ATM positions , we applied a Gaussian kernel smoother to smooth the number of mutations of each amino acid site . If we use y ( xi ) ( i=1 , 2 , ⋯ , n ) to represents the number of mutations in each site xi , then the Gaussian kernel smoother of y ( x ) is defined by:y^ ( xi ) =∑j=1nK ( xi , xj ) y ( xj ) ∑j=1nK ( xi , xj ) where K ( xi , xj ) =exp ( − ( xi−xj ) 22b2 ) is the Gaussian kernel , and b indicates the window size ( we use 80 in the current study ) . Missense mutations and truncating mutations were separately considered . To estimate the expected number of mutations in ATM , we assume the ratio between silent and non-silent mutation ( e . g . missense mutation ) is constant , and then the frequency , as well as its confidence interval , to observe a missense mutation can be estimated by fitting a binomial distribution with n trials . The shadowed curve is the 95% confidence interval of fold change between observation and expectation after a Bonferroni Correction . The supplementary table 1 shows the result of the permutation analyses for the expected and observed mutations in ATM or ATM kinase domain ( AA2711-2962 , Supplementary file 1B ) . Homology model of human ATM was generated based on the crystal structure of human mTOR ( PDB 4JSP ) ( Yang et al . , 2013 ) using the ROBETTA server , setting default parameters ( Kim et al . , 2004; Katyal et al . , 2014 ) . Briefly , the AAs 2520–3056 of human ATM was submitted to the ROBETTA server and the mTOR crystal structure was identified as a template . After aligning the query sequence with the parent structure , a template was generated and the variable regions were modelled in the context of the fixed template using Rosetta fragment assembly . Finally , the model was subjected to several rounds of optimization by Rosetta’s relax protocol . All structure analyses and visualizations were performed with pymol ( http://pymol . org/ ) . Human cancer cell lines – BT-474 ( Breast cancer ) , Granta519 ( MCL ) , HEC251 ( Endometrium tumor ) and SNU283 ( Colon Cancer ) were selected , since the CCLE ( Broad Institute ) have identified mis-sense mutations in or around the ATM kinase domains in these cell lines . We have verified the identity of these cell lines to the best of our knowledge – based on ATM mutation status , ATM expression and activities ( Figure 1—figure supplement 1E ) . HBL2 cell lines were selected as the MCL control , since similar to Granta519 , HBL2 has deregulated expression of CyclinD1 and is WT for ATM . Other cell lines ( Hela and U2OS ) were also validated for normal expression of ATM and ATM kinase activity . None of the used cell lines were in the list of commonly misidentified cell lines maintained by the International Cell Line Authentication Committee ( 2016 May version ) . All the mouse alleles used have been previously described ( Yamamoto et al . , 2012; de Boer et al . , 2003; Zha et al . , 2008; de Luca et al . , 2005 ) , AtmC , AtmKD , VavCre+ , RosaERCre . All the animal work was approved by and performed according to the regulations of the Institutional Animal Care and Use Committee ( IACUC ) of Columbia University . For development analyses , VN , VKD and littermate matched control mice were euthanized at 4–6 weeks of age . Total lymphocyte counts , flow cytometry analyses and in vitro class switch recombination were performed as described previously ( Li et al . , 2008 ) . Briefly , splenic B and T-cells were isolated using Magnetic-Activated Cell Sorting ( MACS ) for CD43- and CD43+ fractions , respectively , following the manufacturer’s protocol ( Miltenyi Biotec ) . B-cells ( CD43- ) were stimulated with LPS ( 20 µg/ml ) and IL-4 ( 20 ng/ml ) for 4 . 5 days at a cellular concentration no more than 1 × 106 cells/ml , and the T-cells ( CD43+ ) were activated by Con A ( 2 µg/ml ) for 3 . 5 days ( Li et al . , 2008 ) . Metaphases were collected at the end of the cytokine stimulation period after 4 hr of colcemid ( 100 ng/ml KaryoMax , Gibco ) treatment , and stained with Telomere FISH probes as previously described ( Yamamoto et al . , 2012; Franco et al . , 2006 ) . Lymphoma- bearing animals were identified based on enlarged lymph nodes ( B cell lymphomas ) or severely hunched postures ( thymic lymphomas ) and analyzed . For the tumor studies , the investigators were not blinded . MEFs were harvested at embryonic day 14 . 5 ( E14 . 5 ) based on timed breeding and immortalized using retrovirus expression of the large and small SV40 T antigen . Immortalized culture was established after three passages ( 1:3 dilutions ) after the primary infection . To induce 4-hydroxytamoxifen ( 4OHT ) dependent nuclear translocation of ER-Cre recombinase ( Figure 3—figure supplement 1A ) , the cells were plated at low density ( ~30% confluence ) , and passed three times ( 48 hr interval ) in the presence of 4OHT ( 200 nM ) . Complete deletion of the conditional allele was confirmed using PCR as described previously ( Yamamoto et al . , 2012 ) . For drug sensitivity assays , the immortalized MEF treated for 4OHT ( 3 rounds ) were seeded in gelatinized 96-well plates ( 6 × 103/well ) . Twenty-four hours after the initial seeding , the cells were treated with the compounds at indicated concentrations for 48 hr . The cell number was quantified using CyQuant ( Molecular Probes ) , a nucleotide stain , per the manufacturer’s instructions . The relative survival was calculated relative to the cell number in the untreated wells . Immunofluorescence was performed on 4% formaldehyde/PBS fixed , 1% Triton-X/PBS treated cells using the following antibodies: Rad51 ( Clone PC130 , 1:200 , Calbiochem ) phospho-RPA pT21 ( Cat . # ab109394 , 1:500 , Cell Signaling ) γH2AX ( Cat . #07–164 , 1:500 , EMD Millipore ) , and CyclinA2 ( Clone CY-A1 , 1:500 , Sigma-Aldrich ) . Slides were scanned on the Carl Zeiss Axio Imager Z2 equipped with a CoolCube1 camera ( Carl Zeiss , Thornwood , NY ) . Metafer 4 software ( MetaSystems , Newton , MA ) was used for automated quantification of Rad51 , pRPA , γH2AX foci . More than 500 cells were quantified for every experimental replicate . Images were exported and processed on ISIS software ( MetaSystems ) . The following antibodies were used for Western blots on whole cell lysate: ATM ( clone MAT3 , 1:5000 ) , β-Actin ( clone A5316 , 1:20 , 000 ) and Vinculin ( clone V284 , 1:10 , 000 ) from Sigma-Aldrich; phospho-Kap1 ( Cat . #A300-767A , 1:1000 ) and Top-I ( Cat . # A302-590 , 1:1000 ) from Bethyl laboratory; phospho-RPA pT21 ( Cat . # ab109394 , 1:5000 ) from Abcam; RPA ( clone RPA30 , 1:5000 ) , γH2AX ( Cat . #07–164 , 1:1000 ) and H2AX ( Cat . #07–627 , 1:1000 ) from EMD Millipore; phospho-Chk1 ( clone 133D3 , 1:500 ) , Mre11 ( Cat . #4895S , 1:1000 ) , Kap1/TIF1β ( clone C42G12 , 1:1000 ) , and Pten ( clone D4 . 3 XP , 1:1000 ) from Cell Signaling and PCNA ( clone PC10 , 1:1000 ) from Santa Cruz . To confirm deletion of the ATM conditional allele in VKD and VN mouse models , Southern blot was performed on KpnI digested genomic DNA and probed with the ATMCKO 3’ probe ( Zha et al . , 2008 ) . For clonal analyses of the thymic lymphomas or the B cell lymphomas , Southern blot was performed on EcoRI-digested genomic DNA , and blotted with Jβ1 . 6 , Jβ2 . 7 probes ( Zha et al . , 2008; Khor and Sleckman , 2005 ) or Jh4 and myc-A probe ( Gostissa et al . , 2009 ) . Cells were incubated for two doubling times ( 48 hrs ) with Bromodeoxyuridine ( BrdU , Sigma-Aldrich , 5 µg/ml ) . During the second doubling period ( hours 25–48 ) , the cells were also incubated with either vehicle ( DMSO ) or CPT ( 1 µM ) . BrdU-incorporated sister chromatids were quenched using Hoechst33258 ( 50 µg/ml ) treatment followed by UV exposure , then stained with DAPI ( Vectashield ) . Slides were scanned for metaphase spreads using the Carl Zeiss Axio Imager Z2 equipped with a CoolCube1 camera ( Carl Zeiss , Thornwood , NY ) and Metafer 4 software ( MetaSystems , Newton , MA ) . Spreads were quantified and images exported via ISIS software ( MetaSystems ) . Over 40 metaphases were quantified for each genotype . A total of three independent replicates were performed . Neutral and alkaline COMET assays were performed per manufacturer’s protocol ( Trevigen , Gaithersburg , MD ) . Slides were scanned for COMETs using Metafer 4 and tail lengths quantified using ISIS . Covalent complexes of TopI-DNA were detected as previously described ( Nitiss et al . , 2012 ) . Briefly , cells were gently lysed in 1% Sarkosyl/TE solution , ultracentrifuged through a CsCl2 gradient ( 6 M ) and the resulted genomic DNA was dissolved in TE . Equivalent amounts of DNA were loaded on 0 . 45 µm Nitrocellulose membranes , and probed for Top-1 ( Cat . # A302-590 , 1:1000 , Bethyl ) . Loading controls were probed with radiolabeled total mouse genomic DNA . Cells were first incubated with 25 µM IdU ( 30 min ) then with additional 250 µM CldU ( Sigma ) , trypsinized and harvested in ice-cold 1 x PBS , lysed using 1% SDS/Tris-EDTA and stretched along glass coverslips . Slides were stained using primary Anti-IdU ( B44 , BD Biosciences ) , CldU ( BU1/75 ( ICR1 ) , Abcam ) antibodies and corresponding secondary anti-Rat Alexa594 , and anti-mouse Alexa488 ( Molecular Probes ) antibodies . DNA fibers were analyzed on the Nikon Eclipse 80i microscope equipped with remote focus accessory and CoolSNAP HQ camera unit using the 60 x /1 . 30 NA oil Plan Fluor lens . All images were processed and quantified with NIS-Elements AR . Lengths of fibers were calculated with a base pair length of 3 . 4 Å ( 340 pm ) . 1 µg of differentially labeled genomic DNA from tumor and non-tumor control ( kidney ) derived from the same animal were hybridized on the 244 k Mouse Genome CGH microarray platform ( Agilent ) and analyzed on Agilent Genomic Workbench and Microsoft Excel as previously described ( Zha et al . , 2010; Yamamoto et al . , 2015 ) . In vivo psoralen cross-linking , isolation of genomic DNA from mammalian cells , enrichment of replication intermediates , and analysis of data was performed as previously described ( Neelsen et al . , 2014; Thangavel et al . , 2015 ) . Briefly , 5–10 × 106 Mouse Embryonic Fibroblasts ( Rosa+/ERCRE Atm+/C ) cells were harvested and genomic DNA was cross-linked by three rounds of incubation in 10 µg/ml 4 , 5′ , 8-trimethylpsoralen ( Sigma-Aldrich ) and 3 min of irradiation with 366 nm UV light on a precooled metal block . Cells were lysed and cellular proteins were digested with 2 mg proteinase K ( Life technologies ) in 5 ml digestion buffer . DNA was purified by isopropanol precipitation , restriction digested with PvuII HF for 4 hr at 37°C , and replication intermediates were enriched using benzolylated naphthoylated DEAE-cellulose ( Sigma–Aldrich ) in 3 ml BioRad poly-prep chromatography columns . Samples were prepared by spreading DNA on carbon-coated grids in the presence of benzyl-dimethyl-alkylammonium chloride and formamide . Rotary platinum shadowing was performed using the Balzers BAF400 with Quartz crystal thin film monitor . Images were acquired on a JOEL 1200 EX transmission electron microscope with side-mounted camera ( AMTXR41 supported by AMT software v601 ) and analyzed with ImageJ ( National Institutes of Health ) . To determine the therapeutic effect of clinically used Topo1 inhibitor – irinotecan in vivo , we generated isogenic leukemia by transducing bone marrow from Rosa+/CreERT2 AtmC/KD mice treated with 5-Fluorouracil ( 5-FU ) were transduced with retrovirus encoding activated Notch1 ( ΔE-Notch1 ) and GFP ( MSCV-ΔENOTCH-IRES-GFP ) ( Tzoneva et al . , 2013 ) and transplanted into semi-lethal irradiated ( 8 . 5 Gy ) recipient mice ( n = 3 ) . Once the leukemia developed ( monitored by peripheral blood GFP+ cells ) , ~1 × 10^6 GFP+ leukemia cells ( from one mouse to ensure clonality ) were transduced to 2nd recipients ( n = 4–6 ) . 4 days later , the 2nd recipients were fed with oral tamoxifen ( n = 3 ) or carrier alone ( sun flower oil , n = 3 ) for two days . When the 2nd recipients developed leukemia , the leukemia cells were genotyped for ATM status ( KD/- or KD/C ) and analyzed for clonality ( TCRβ rearrangements by Southern ) . Once the isogenic ( defined by identical TCRβ rearrangements ) leukemia was obtained , they were transduced to tertiary recipients ( n = 5–10 per genotype ) . 3 days after the transplantation , the tertiary recipients were treated with IP injection of irinotecan ( 10 mg/kg ) for 5 days or vehicle alone ( for 5 days ) and observed for leukemia development . In ~10 days after the last drug injection , the vehicle groups displayed terminal diseases and both the vehicle control and the irinotecan treated mice were euthanized and analyzed for GFP+ cells in the spleen , histology and spleen weight and size . Similar experiments were performed on Rosa+/CreERT2 AtmC/C bone marrow derived , isogenic AtmC/C vs Atm-/- Notch driven leukemia and representative results are shown in Figure 4—figure supplement 1C–F .

Cancer is a genetic disease . To remain healthy , therefore , it is essential that cells do not accrue too many dangerous mutations in their DNA that allow cancers to grow and develop . An enzyme called ATM helps to do just that . DNA damage activates ATM , which , in turn , adds phosphate groups to other proteins . These newly tagged proteins then stop cells dividing until the DNA has been repaired . Human cancers often switch off ATM , either by completely deleting the enzyme or mutating it . This renders ATM unable to add phosphate groups to proteins , and so allows the cancer cells to continue proliferating even in the face of DNA damage . Yamamoto et al . wanted to know whether cancers that completely lack ATM behave differently from cancers that contain an inactive version of the enzyme . Studying mice that were engineered to have an inactive version of ATM in their blood cells showed that such mice developed blood cancers faster than mice have no ATM in their blood cells . In particular , cancer cells with the inactive form of the ATM enzyme accumulated more DNA damage than cells that lacked the enzyme completely . Using biochemical techniques , Yamamoto et al . then showed that the inactive form of ATM can prevent other enzymes from repairing DNA . Drugs that inhibit one of these repair enzymes – called topo-isomerase I – are already used in cancer treatments . These drugs were particularly effective on tumors with the inactive version of the ATM enzyme . As ATM is commonly mutated in human cancers , the next steps that follow on from this research are to develop methods to test which cancers contain the inactive form of the ATM enzyme . Clinical trials could then investigate how effectively topo-isomerase I inhibitors treat these specific types of cancer .

2016

Kinase-dead ATM protein is highly oncogenic and can be preferentially targeted by Topo-isomerase I inhibitors

The epithelial sodium channel ( ENaC ) , a member of the ENaC/DEG superfamily , regulates Na+ and water homeostasis . ENaCs assemble as heterotrimeric channels that harbor protease-sensitive domains critical for gating the channel . Here , we present the structure of human ENaC in the uncleaved state determined by single-particle cryo-electron microscopy . The ion channel is composed of a large extracellular domain and a narrow transmembrane domain . The structure reveals that ENaC assembles with a 1:1:1 stoichiometry of α:β:γ subunits arranged in a counter-clockwise manner . The shape of each subunit is reminiscent of a hand with key gating domains of a ‘finger’ and a ‘thumb . ’ Wedged between these domains is the elusive protease-sensitive inhibitory domain poised to regulate conformational changes of the ‘finger’ and ‘thumb’; thus , the structure provides the first view of the architecture of inhibition of ENaC . The fine-tuning of Na+ homeostasis is largely mediated by epithelial sodium channels ( ENaC ) that are related in amino acid sequence to acid-sensing ion channels ( ASIC ) found in eukaryotes , degenerin channels ( DEG ) of Caenorhabditis elegans , and the FMRF-amide peptide-gated channels ( FaNaCh ) of mollusk ( Driscoll and Chalfie , 1991; Chalfie and Wolinsky , 1990; Kellenberger and Schild , 2002; Waldmann et al . , 1997a; Waldmann et al . , 1997b; Krishtal and Pidoplichko , 1981; Chelur et al . , 2002; Garty and Palmer , 1997; Cottrell et al . , 1990; Lingueglia et al . , 1995 ) . These ion channels belong to the voltage-independent , Na+-selective , and amiloride-sensitive ENaC/DEG superfamily which together perform diverse cellular functions in different organisms . In humans , ENaCs are expressed at the apical surface of epithelial tissues throughout the body , and play critical roles that range from regulation of total-body salt , water , and blood volume , to modulating airway surface liquid clearance in epithelial cells in the lungs ( Büsst , 2013; Ismailov et al . , 1996; Rossier et al . , 2015; McDonald et al . , 1994 ) . The importance of ENaC in Na+ homeostasis is highlighted by gain-of-function mutations causing severe hypertension , as in Liddle syndrome , or loss-of-function mutations causing the neonatal salt-wasting disorder pseudohypoaldosteronism type 1 ( PHA-1 ) ( Gründer et al . , 1997; Hansson et al . , 1995; Shimkets et al . , 1994; Chang et al . , 1996; Edelheit et al . , 2005; Kerem et al . , 1999 ) . More subtle ENaC dysfunction contributes to diseases as diverse as essential hypertension , heart failure , and nephrotic syndrome ( Soundararajan et al . , 2010; Hamm et al . , 2010; Zheng et al . , 2016 ) . ENaCs require three different subunits to form a functional channel , α , β , and γ ( Canessa et al . , 1994 ) . Despite decades of study , the number of subunits in an active channel remains unclear ( Shobair et al . , 2016 ) . Unique among the ENaC/DEG channels , ENaCs are activated by proteolysis of peptidyl tracts embedded in the extracellular domain ( ECD ) , which releases inhibitory peptides . The cleavage event increases channel opening probability , Po ( Orce et al . , 1980; Vallet et al . , 1997; Vallet et al . , 2002; Vuagniaux et al . , 2002; Hughey et al . , 2004; Hughey et al . , 2003; Caldwell et al . , 2004; Passero et al . , 2010; Kleyman et al . , 2009 ) . Amino acid sequence alignments and biochemical analyses in the ECD have so far revealed that only the β subunit lacks the characteristic motifs for protease recognition . ENaCs are widely known as substrates of serine proteases like furin , and a growing list of proteases that recognize sites in ENaC suggests a multifaceted regulation of channel function ( Rossier and Stutts , 2009 ) . Indeed , the complexities of ENaC function involving the requisite heteromeric subunit assembly and asymmetric subunit modification via differential proteolytic processing are critical to ion channel gating . Thus , to define subunit arrangement and stoichiometry , and elucidate the molecular architecture of ENaC inhibition , we determined the structure of ENaC in the uncleaved state by single-particle cryo-electron microscopy ( cryo-EM ) . We first assessed the expression of full-length ( FL ) ENaC by small-scale expression in adherent HEK293S GnTI- cells and fluorescence-detection size-exclusion chromatography ( FSEC ) ( Kawate and Gouaux , 2006 ) . We found a low , wide peak , indicating a poorly expressing polydisperse population unsuitable for cryo-EM ( Figure 1a ) . We thus screened a number of deletions and mutations in each ENaC subunit , harnessing information derived from previous biochemical and functional experiments gauging the propensity for heterotrimeric formation of ENaC and its susceptibility to proteolytic processing ( Canessa et al . , 1994; Orce et al . , 1980; Vallet et al . , 1997; Vallet et al . , 2002; Vuagniaux et al . , 2002; Hughey et al . , 2004; Hughey et al . , 2003; Caldwell et al . , 2004; Passero et al . , 2010 ) , before arriving at the construct referred to here as ΔENaC ( Figure 1a–c , Figure 1—figure supplement 1 , Figure 1—figure supplement 2 ) . ΔENaC is composed of αβγ subunits truncated at the N- and C-termini ( Figure 1b , c ) . Additionally , the Δα and Δγ subunits possess mutations in the identified furin and prostasin sites which prevent subunit cleavage and channel activation ( Hughey et al . , 2003; Bruns et al . , 2007 ) . For protein purification , neither Δα nor Δβ were modified with affinity tags because there is strong evidence that the α subunit can readily form functional homomeric channels , and the termini of Δβ are sensitive to perturbations ( Canessa et al . , 1993 ) . As a result , Δγ contains both GFP and a Strep-II tag at the N-terminus ( Figure 1c ) , minimizing contamination by homomeric Δα channels during purification . This construct provided a homogeneous and highly-expressing population . However , the inherent pseudosymmetry from common secondary and tertiary structures between the α , β , and γ subunits of human ENaC hindered particle alignment ( Figure 1d ) . To evaluate biochemical integrity and to facilitate cryo-EM three-dimensional reconstruction of ΔENaC , we generated subunit-specific monoclonal antibodies ( mAbs ) that bind to three-dimensional epitopes in ΔENaC and FL-ENaC . For immunization , we exploited the high-expressing chicken ASIC ( cASIC ) by adding the first 22 N-terminal amino acids of cASIC to Δβ , which tolerated the fusion . This construct is referred to hereafter as ΔβASIC . Together , Δα , ΔβASIC , and Δγ comprise ΔENaCASIC ( Figure 1—figure supplement 3a , b ) ( Jasti et al . , 2007 ) . Two fragment-antigen binding domains ( Fabs ) were isolated that recognize different epitopes ( 7B1 and 10D4 ) . While these antibodies were raised against ΔENaCASIC ( Figure 1—figure supplement 3a , b ) , both Fabs bind to both ΔENaC expressed in HEK 293S GnTI- and FL-ENaC expressed in HEK293T/17 , which indicates that ΔENaC is properly folded and that the Fabs do not bind to the ASIC segment ( Figure 1—figure supplement 3c , d; Figure 1—figure supplement 4a , b ) . Inclusion of 7B1 and 10D4 allowed for proper alignment of the particles ( Figure 1e ) . Moreover , maps of the particles with only 10D4 ( monoFab ) compared to those with both 10D4 and 7B1 ( diFab ) show that each Fab recognizes only one subunit ( Figure 1—figure supplement 4c–f ) . We monitored and compared grid conditions and the resulting data quality ( including ice thickness , sample quality , particle distribution , and orientation ) between the monoFab and the diFab complexes of ENaC and discovered that the diFab complex was a more promising complex for cryo-EM analysis . We investigated ΔENaC function by two-electrode voltage-clamp electrophysiology ( TEVC ) and whole-cell patch clamp electrophysiology in oocytes and GnTI- HEK cells , respectively ( Figure 2 , Figure 2—figure supplements 1 and 2a and b ) . Unlike FL-ENaC ( Figure 2a , Figure 2—figure supplement 2a ) , ΔENaC does not exhibit amiloride-sensitive currents in oocytes and HEK cells , and the only Na+-specific currents resemble those from uninjected oocytes ( Figure 2—figure supplement 1a , b ) . Similarly , oocytes expressing ENaC channels with restored protease sites in the Δ subunits ( Δα* and Δγ* ) to form Δ*ENaC did not present amiloride-sensitive currents ( Figure 2—figure supplement 1c ) . Because HEK cells are better suited to defining whether ΔENaC traffics to the plasma membrane , we examined surface expression of ΔENaC and FL-ENaC expressed in GnTI- HEK cells using confocal microscopy . To ensure robust expression , we transduced the HEK cells with baculovirus encoding the ΔENaC and FL-ENaC proteins , taking advantage of the N-terminal eGFP in the Δγ subunit and the N-terminal eGFP in all three FL subunits to visualize expression , respectively . Based on eGFP fluorescence , we observed robust expression of both ΔENaC and FL-ENaC ( Figure 2—figure supplement 2c , d ) . We employed tetramethylrhodamine ( TRITC ) -labeled 10D4 mAb , an antibody that binds to the extracellular domain of ENaC , to probe the plasma membrane localization of ENaC channels . Indeed , we observed overlapping signals from both eGFP and TRITC-10D4 mAb in cells expressing FL-ENaC but not in cells expressing ΔENaC . Based on the confocal imaging results , ΔENaC is not trafficked to the plasma membrane , in agreement with the electrophysiology results in HEK 293S GnTI- cells and oocytes ( Figure 2—figure supplement 2 ) . We further examined whether disruption of the channel by mutagenesis also caused the absence of ΔENaC current . We tested channels comprising a single Δ or Δ* subunit in complex with the two complementary FL-ENaC subunits . Channels comprising Δα-FLβ-FLγ conduct amiloride-sensitive Na+ currents which increase approximately 5-fold upon trypsin treatment ( compared with 2 . 2-fold for FL-ENaC , Figure 2b , Figure 2—figure supplements 2 and 3 , and Figure 2—source data 1 ) . Since this trypsin response could be a result of cleavage of FLγ , we also tested channels of Δα*-FLβ-FLγ ( Figure 2c ) . These channels show an increase in total current compared to Δα-FLβ-FLγ , and demonstrate a more archetypal ENaC current trace ( Figure 2a , Figure 2—figure supplement 3 , Figure 2—source data 1 ) . These results , in addition to the cleavage pattern of an anti-α immunoblot ( Figure 2—figure supplement 4 ) indicate that Δα adopts a biologically relevant fold , capable of forming active channels with other full-length subunits , and that it is likely cleaved once at its N-terminal furin site ( RSRA in Δα ) but not the C-terminal furin site ( AAAA in Δα , Figure 1—figure supplement 1 ) . By restoring the protease sites , as in Δα* , the inhibitory peptide was effectively removed . The FLα-Δβ-FLγ channels conducted amiloride-sensitive Na+ currents with a post-trypsin/pre-trypsin ratio of 1 . 5 ( Figure 2d , Figure 2—figure supplement 3 , Figure 2—source data 1 ) , similar to that of FL-ENaC . Moreover , an anti-β immunoblot shows no cleavage of Δβ , as expected ( Figure 2—figure supplement 5 ) . The FLα-FLβ-Δγ channel also conduct an amiloride-sensitive Na+ current with approximately 9 . 5-fold increase upon trypsin treatment ( Figure 2e , Figure 2—figure supplements 2 and 3 , and Figure 2—source data 1 ) . Although the Δγ subunit has the canonical furin and prostasin sites mutated ( AAAA and QQQQ respectively , Figure 1—figure supplement 1 ) , there are other basic residues near the furin and prostasin sites that could be cleaved by trypsin . This hypothesis is further supported by the immunoblot showing significant trypsin digestion in Δγ ( Figure 2—figure supplement 6 ) as well as the even higher trypsin activation of FLα-FLβ-Δγ* ( approximately 13 . 3-fold , Figure 2f and Figure 2—figure supplement 3 and Figure 2—source data 1 ) . Nevertheless , the results are a promising direction for future studies . Importantly , the combination of TEVC traces of each Δ subunit and the α and γ Δ* counterparts supports ΔENaC representing a biologically relevant channel . We solved the structure of ΔENaC diFab complex in n-Dodecyl β-D-maltoside ( DDM ) by cryoEM ( Figure 3 , Figure 3—source data 1 ) . We first carried out cycles of 2D and 3D classifications to remove ice contamination , micelles , and denatured complexes . The remaining particles were subjected to unsupervised ab initio 3D classification and refinement in cryoSPARC ( Punjani et al . , 2017 ) as well as 3D classification and refinement in cisTEM ( Grant et al . , 2018 ) to arrive at the cryo-EM potential map with a nominal resolution of 4 . 2 Å from both programs , based on the gold standard FSC = 0 . 143 and solvent adjusted FSC = 0 . 143 criteria , respectively ( Figure 3—figure supplement 1 ) . Additionally , we conducted a masked refinement excluding the flexible Fc domains of the Fabs and micelle in cisTEM ( Figure 3a ) , and obtained a map at 3 . 9 Å , as determined by the solvent adjusted FSC = 0 . 143 criterion ( Figure 3b , c ) , with local resolution estimates generated by BSoft ( Heymann and Belnap , 2007 ) indicating regions of the map with a resolution of 3 . 7 Å ( Figure 3d ) . The cryo-EM potential map has three major regions into which the two Fabs and homology models of ΔENaC were manually fitted ( Figure 4 ) . Alignment of predicted glycosylation sites and aromatic residues to distinct features in the map allowed for the correct assignment of the homology models of the ENaC ECD , generated from the desensitized state of ASIC ( PDB: 2QTS , Figure 4—figure supplements 1–5 , Figure 4—video 1 ) . The β subunit is predicted to have 11 glycosylation sites by primary sequence , considerably more than α or γ . Six prominent glycosylation sites were used to assign β ( as opposed to the three each in α and γ ) , whereas a glycosylation on the β9-α4 loop distinguished α from γ ( Figure 1—figure supplement 2 , Figure 4—figure supplement 1 ) . Guided by these features and the 10D4 monoFab ΔENaC map , we assigned the identity of 7B1 and 10D4 as binding α and β subunits , respectively ( Figure 4 , Figure 1—figure supplement 4 ) . Forming a trimeric ensemble , the α-β-γ subunits arrange in a counterclockwise manner , as reported by previous studies ( Collier and Snyder , 2011; Collier et al . , 2014; Chen et al . , 2011 ) ( Figure 4b , d ) . The overall domain organization within each subunit of ΔENaC concurs with that of ASIC , which was first illustrated in the crystal structure of chicken ASIC ( cASIC ) ( Jasti et al . , 2007 ) ( chicken ASIC shares 18 – 20% sequence identity with human ENaC; Figure 5 , Figure 5—figure supplement 1 ) . Each subunit of ΔENaC harbors a cysteine-rich ECD resembling a hand with the palm , knuckle , finger , and thumb domains clenching a ‘ball’ of β strands . This compact organization is stabilized by eight disulfide bridges in the ECDs of α and γ and nine in β . Seven of the disulfide bonds are conserved throughout the ENaC/DEG family ( Figure 1—figure supplement 1 , Figure 1—figure supplement 2 , Figure 5a–c ) . The eighth is unique to the three ENaC subunits . For the purpose of consistency in the following discussion , domain and secondary structure assignment in ENaC follow those of ASIC ( Figure 5d ) . At the center of the trimeric architecture of the ECD are β-sheets formed by β1 , β3 , β6 , and β9-β12 that constitute the palm domain , which are divided into two sections , the upper and lower palm domains . The upper palm domain cradles the β-ball , which is composed of β2 , β4 , β5 , β7 , and β8 in all three subunits , contrary to previous findings which suggested that the α subunit lacked the β4 and β5 strands ( Stockand et al . , 2008 ) . Completing the ‘clench’ around the β-ball are the α1 – 3 of the finger , α4 – 5 of the thumb , and α6 of the knuckle domains . The lower palm is directly linked to the transmembrane domain ( TMD ) via β1 and β12 and to the α4 and α5 of the thumb through β9 and β10 . The thumb and the lower palm converge to forge interactions with the TMD at a juncture called the ‘wrist’ ( Figure 5—figure supplement 1 ) . Underscoring the importance of the wrist region and the critical roles that disulfide bridges play in maintaining the structural and functional integrity of ENaC , alterations of a conserved cysteine , α-Cys479 to an Arg , causes Liddle syndrome due to a missense mutation that not only eliminates a disulfide bridge located at the juncture of the thumb and palm domains but also introduces a bulky , positively charged residue ( Salih et al . , 2017 ) ( Figure 5a ) . ENaC differs significantly from ASIC in both structure and primary sequence at the knuckle and finger domains ( Figure 1—figure supplement 1 , Figure 1—figure supplement 2 , Figure 5—figure supplement 1 ) . Each knuckle domain in ENaC makes extensive interactions with the α1 and α2 helices of the finger domain in the adjacent subunit ( Figure 6 ) . Together , the knuckle and finger domains of all three subunits form a ‘collar’ at the top of the ECD . Sequence alignment of the three subunits demonstrate divergence in amino acid sequence at the C-terminal end of both α1 and α6 in all three subunits , which results in distinct types of molecular interactions at each subunit interface that are , perhaps , associated with assembly and stability of the ENaC . The contact between the finger and thumb domains is mediated by long antiparallel helices α1 and α2 , which form a barrier between the thumb domain and the β6-β7 loop with α2 making the primary contacts with the thumb domain ( Figure 5—figure supplement 1 , Figure 6 ) . The domain arrangement observed in the ΔENaC structure agrees with the functional study probing Na+ binding sites in the α subunit of ENaC ( Kashlan et al . , 2015 ) . The α2 helix makes an almost 90° turn towards the palm domain breaking the helix . This architecture marks another departure from ASIC , in which contacts between the finger and thumb domains are largely mediated by α1 , α3 , α5 , and the α4-α5 loop ( Figure 5—figure supplement 1 ) . The TMD is not well ordered , hampering our ability to model the entire TMD region and assign a functional state of the channel . Nevertheless , the EM map offers a glimpse of the positions of TM1 and TM2 on the extracellular side from each subunit ( Figure 6—figure supplement 1 ) . The overall configuration of the TMD shows that TM2 of all three subunits are positioned near the central axis , poised to mediate ion conduction in agreement with the crystal structures of ASIC and previous elegant functional studies probing ion selectivity and channel block ( Kellenberger and Schild , 2002 ) . Strikingly , the potential map for γ-TM1 on the extracellular side illustrates clear map for the main chain preceding the β1 strand validating a sequence disparity between the γ subunit and the other ENaC and ASIC subunits ( Figure 6—figure supplement 1 ) . The γ subunit lacks two residues preceding the palm domain ( Figure 6—figure supplement 1f ) . Consequently , interactions within the wrist region in the γ subunit may differ from that of α and β subunits . Previous studies of ENaC have probed stretches of amino acids and their roles in ENaC function by perturbing known protease sites , observing changes in molecular weight , recording channel activity , and conducting cross-linking studies ( Bruns et al . , 2007; Carattino et al . , 2006 ) . The structure of ΔENaC indicates that these stretches of 20 – 40 amino acids are pieces of larger domains located in the periphery of the ECD near subunit interfaces ( Figure 7 ) . These stretches of amino acids , located between α1 and α2 are unique to ENaC and are responsible for channel Gating Relief of Inhibition by Proteolysis and will hereafter be referred to as the GRIP domain . Each GRIP domain is composed of a core of β strands that forge interactions with the finger and thumb domains forming a β-sheet ‘blanket’ that conceals the α2 helix of the finger ( Figure 7—figure supplement 1 ) . Surprisingly , although the β subunit is not known to gate the channel via proteolysis , it also possesses a GRIP domain with similar organization to those of the α and γ subunits . In all three subunits , the GRIP domains comprise two antiparallel β strands stapled together by a disulfide bond located in the loop that rests against the thumb domain ( Figure 7b–d , Figure 7—figure supplement 1b–d ) . Furthermore , an additional disulfide bond in the loop near the β-γ interface stabilizes the GRIP domain of the β subunit . We suspect that this additional disulfide bond contributes to the well-ordered behavior of the β GRIP domain , allowing the resolution of nearly the whole segment between α1 and α2 in the β subunit . Moreover , the 10D4 Fab binds the β GRIP domain , allowing us to resolve two additional antiparallel β strands . In the α and γ subunits , we can only identify one stretch of residues that adopt an extended conformation . Based on the shared features observed in all three subunits , it is plausible that the α and γ subunits also contain a fourth β strand . With four possible β strands in the GRIP domains , each strand or stretch of peptides is referred to here as P1-4 ( Figure 7 ) . Structural insight gleaned from the β GRIP domain reveals the possible positions of the functionally well-characterized but structurally elusive inhibitory tracts and furin and other protease sites in the α and γ GRIP domains . Studies by the Kleyman group have identified 8- and 11-mer peptide tracts within the α and γ GRIP domains , respectively , which are implicated in channel gating ( Passero et al . , 2010; Carattino et al . , 2008b; Kashlan et al . , 2010 ) . Sequence comparison between the three subunits suggests that the inhibitory tracts contain the P1 strand ( Figure 7b – d , Figure 7—figure supplement 1b–d , Figure 1—figure supplement 1 ) . Based on this configuration , the first furin site lies at the N-terminal side of P1 whereas the second protease site ( furin for α and other protease sites for γ ) is likely located at the C-terminal side of P2 . The anti-parallel organization of P1 and P2 strands places the two protease sites in close proximity to each other . We speculate that this arrangement allows for efficient proteolysis , especially for the cleavage of the α subunit by furin . The first crystal structure of ASIC identified the finger and the thumb domains as major players in ion channel gating . Rearrangements of these domains are coupled to the TMD via the wrist . Additionally , the crystal structure provided insight into the domain essential for fine-tuning ASIC pH-response , deemed the acidic pocket , formed by the β-ball , finger , and thumb domains of one subunit , and the palm domain of the adjacent subunit ( Vullo et al . , 2017 ) . While the acidic pockets in ASIC are lined with negatively charged residues , the equivalent crevices in ENaC are replete with aromatic residues . In fact , aside from Ser428 of Δβ ( Asp346 in ASIC ) , the equivalent sites in the thumb domain that are acidic in ASIC are occupied by tyrosines in all three subunits ( Figure 8 , Figure 1—figure supplement 1 , Figure 1—figure supplement 2 ) ( Jasti et al . , 2007 ) . Accordingly , the pocket that is largely occupied by α2 in ΔENaC is referred to here as the aromatic pocket . Tucked in the aromatic pocket , α2 makes contacts with all critical elements of the gating machinery in ENaC . This observation is consistent with studies finding that site-directed mutagenesis perturbing residues in the α2 results in changes in Na+ self-inhibition and binding of the P1 segment of the GRIP domain ( Kashlan et al . , 2010 ) . In all three ENaC subunits , the α2 forms contacts with the thumb , α1 helix , the β6-β7 loop and the GRIP domain , and with the knuckle and the upper palm domains in the adjacent subunit ( Figure 8 ) . Moreover , studies using synthesized 8-mer ( LPHPLQRL ) and 11-mer peptides ( RFLNLIPLLVF ) ( Passero et al . , 2010 ) , the inhibitory peptides of α and γ subunits , respectively , have identified residues in α2 to be critical to the binding of the inhibitory peptides . These peptides pack against α2 and form a wedge between the thumb domain and α1 in the ΔENaC structure ( Figure 7b–d and Figure 7—figure supplement 1b–d ) . These inhibitory peptides contain prolines that introduce a kink within the tract that may serve as a point that divides P1 into two segments: the N-terminal side , which interacts with the finger and thumb; and the C-terminal side , which interacts primarily with α2 and P3 . The observed orientation of the P1 segment is consistent with the cross-linking experiments by Kashlan et al . , which provided two major findings: ( 1 ) the inhibitory tracts adopt an extended conformation and  ( 2 ) the N-terminal side of the peptide binds near the thumb/finger interface ( Kashlan et al . , 2010; Kashlan et al . , 2012 ) . The potential map for α-P1 suggests that the N-terminal side mirrors that of β-P1 forging contacts with the α1 helix ( Figure 7—figure supplement 1b , c ) . In contrast , the potential map for the γ-P1 suggests that the peptide interacts with the thumb and α1/α2 more extensively and extends toward α3 ( Figure 7—figure supplement 1d ) . These distinct points of contact with the finger and thumb domains between the α- and γ-P1 segments may influence the extent to which the subunits influence channel Po . While removal of the inhibitory tract in α transitions the channel to an intermediate Po state , excision of the γ-P1 segment places the channel in the high Po state; this high Po state can be accomplished without the removal of the α-P1 ( Carattino et al . , 2008a ) . The visual evidence of direct interactions between P1 and the finger and thumb domains demonstrated in the ΔENaC structure sheds light into how these inhibitory tracts can modulate channel function in ENaC . ΔENaC follows a common organization that was first observed in ASICs: a scaffolding structure in the upper palm , the flexible lower palm which is tethered to the TM and thumb , and the β-ball ( Figure 5—figure supplement 1 ) ( Jasti et al . , 2007 ) . However , the specialized finger domains deviate from what is observed in ASIC and such deviations accommodate the distinct functions between the proton sensors of ASIC and the protease-sensitive regulators of ENaC . Key gating structures are preserved , albeit with specific structural configurations in both ASIC and ENaC supporting the idea that the superfamily of ENaC/DEG channels conform to a gating scheme that involves conformational changes of the finger and thumb domains , rearrangements that are propagated to the ion channel via the wrist ( Jasti et al . , 2007 ) . In the case for ENaC , a speculative model for gating involves proteolysis and the subsequent removal of the P1 segment , which serves as a wedge , inducing rearrangements of the finger and thumb domains ( Figure 9 ) . The structural work presented here provides new insight into ENaC assembly and gating . The structure unveils the positions of the GRIP domains , specifically the key peptidyl tracts that inhibit ENaC activity , and the distinct interactions that they mediate with the finger and thumb domains . Furthermore , it reveals that there are different interactions between the finger and knuckle domains at each subunit interface , and between the base of the thumb and the TMD in the wrist region suggesting that each subunit differentially contributes toward gating the channel , supporting electrophysiological findings . Importantly , the structure provides the first molecular model for protease-dependent regulation of ENaC opening and Na+ and water homeostasis . The cDNA encoding the full length α , β and γ subunits of human ENaC were cloned into pEG BacMam expression vector harboring an N-terminal eGFP ( Goehring et al . , 2014 ) . The Δα was generated by removing both N- and C-terminal segments and modifying the furin sites obtaining a mutant variant of the α subunit lacking 42 and 89 residues at N- and C- termini , respectively . While the Δβ was designed to possess truncations of 30 and 79 residues at the N- and C-terminal regions , ΔβASIC has the same truncated regions as Δβ but contains an additional N-terminal 2 – 22 residues of cASIC upstream of Δβ . Lastly , Δγ lacks 20 and 79 residues at the N- and C-terminal domains , has modified furin and prostasin sites , and includes a Strep-tag II , an octa-histidine tag , eGFP , and Thrombin cleavage site at the N-terminus . Mouse monoclonal antibodies 7B1 and 10D4 were generated using standard procedure by Dan Cawley at the Vaccine and Gene Therapy Institute ( OHSU ) . Liposomes containing asolectin:cholesterol:lipidA:brain polar lipid extract ( BPLE ) ( 16:4 . 6:1:5 . 3 ) were prepared in 20 mM Tris , 150 mM NaCl at pH 8 . 0 at a concentration of 40 mg/ml . The mixture was subjected to repeated freeze-thaw cycles followed by extrusion through a 200-nm filter . Purified ΔENaCASIC ( Δα , ΔβASIC , Δγ ) protein was added to the liposome mixture in the presence of 400 mM NaCl and 0 . 8% Na-cholate and passed through a PD-10 desalting column to remove excess salt and detergent . Mice were immunized with approximately 30 μg of the reconstituted ΔENaCASIC for generation of hybdridoma cell lines ( Figure 1—figure supplement 3 ) . Monoclonal antibodies were screened by FSEC and BioDot blot to identify clones that recognize tertiary or primary epitopes . The 7B1 and 10D4 mAbs were selected because they recognize tertiary epitopes of ENaC . The mAbs were purified , and their Fabs were generated by papain cleavage . Fab 7B1 was isolated by anion exchange using HiTrap Q HP column while Fab 10D4 was eluted using Protein A column to remove Fc . After isolation , both Fabs were dialyzed in 200 mM NaCl and 20 mM Tris at pH 8 . 0 . Human embryonic kidney cells lacking N-acetylglucosaminyltransferase I ( HEK293S GnTI- cells ) were grown in suspension at a density of 2−4 × 106 cells / ml in Freestyle medium with 2% FBS and transduced with the virus ( Δα , Δβ and Δγ ) at a multiplicity of infection ( MOI ) of 1 and incubated at 37°C . Eight hours post-transduction , sodium butyrate and phenamil mesylate were added to 10 mM and 500 nM , respectively , and cells were incubated at 30°C . After 36 hr , the cells were collected by centrifugation at 4790 xg for 15 min . The pellet was washed with 20 mM Tris , 200 mM NaCl and followed by a second round of centrifugation at 4790 xg for 15 min . Cells were homogenized with a dounce homogenizer and sonicated in 20 mM Tris , 200 mM NaCl , 5 mM MgCl2 , 25 μg/ml DNase I and protease inhibitors . Lysed cells were centrifuged at 9715 xg for 20 min; the resulting supernatant containing the membrane fraction was further centrifuged at 100 , 000 xg for 1 hr . Membrane pellets were resuspended and solubilized in 20 mM TRIS pH 8 , 200 mM NaCl , 20 mM n-dodecyl-β-D-maltopyranoside ( DDM , Anatrace ) , 3 mM cholesteryl hemisuccinate ( CHS ) , 2 mM ATP , 2 mM MgSO4 , protease inhibitor and 25 U/mL nuclease for 1 hr at 4°C . The solubilized fraction was isolated by ultracentrifugation 100 , 000 xg for 1 hr , and ΔENaC was bound to streptactin resin packed into an XK-16 column . The column was washed with 20 mM TRIS , 200 mM NaCl , 0 . 5 mM DDM , 75 μM CHS and 25 U/mL nuclease , followed by an additional wash of the same buffer containing 2 mM ATP , and eluted with 2 . 5 mM desthiobiotin . The eluted fractions were concentrated and then incubated with either one Fab 10D4 ( monoFab complex ) or two Fabs 7B1 and 10D4 ( diFab complex ) in a 1:3 molar ratio of ENaC:Fab for 10 min , and clarified by ultracentrifugation 100 , 000 xg for 1 hr . The supernatant was injected onto a Superose 6 Increase 10/300 GL column equilibrated in 20 mM TRIS pH 8 . 0 , 200 mM NaCl , 0 . 5 mM DDM , 75 µM CHS and 1 mM TCEP to isolate the protein complex by size-exclusion chromatography . Monodispersed fractions were pooled and concentrated to 2 . 2 mg/mL . For FSEC experiments analyzing peak shifts of the ΔENaC and FL-ENaC with 7B1 and 10D4 , ΔENaC was expressed in HEK 293S GnTI- , as described above , while FL-ENaC was expressed in HEK 293T/17 . The HEK 293T/17 cells were grown in suspension at a density of 2−4 × 106 cells / ml in Freestyle medium with 2% FBS and transduced with the virus ( FL-α , FL-β and FL-γ ) at a multiplicity of infection ( MOI ) of 1 and incubated at 37°C . Eight hours post-transduction , 500 nM phenamil mesylate was added , and cells were incubated at 30°C . After 36 hr post-transduction , the cells were collected by centrifugation at 4790 xg for 15 min . The pellet was washed with 20 mM Tris , 200 mM NaCl and followed by a second round of centrifugation at 4790 xg for 15 min . Cell pellets were resuspended and solubilized in 20 mM TRIS pH 8 , 200 mM NaCl , 20 mM n-dodecyl-β-D-maltopyranoside ( DDM , Anatrace ) , 3 mM cholesteryl hemisuccinate ( CHS ) , protease inhibitor and 25 U/mL nuclease for 1 hr at 4°C . The solubilized fraction was isolated by ultracentrifugation 100 , 000 xg for 1 hr , then incubated with either one Fab ( 7B1 or 10D4 ) or two Fabs ( 7B1/10D4 ) in a 1:3 molar ratio of ENaC:Fab for 10 min , and clarified by ultracentrifugation 100 , 000 xg for 1 hr . The supernatant was injected onto a Superose 6 Increase 10/300 GL column for FSEC analysis . Aliquots of 7 µg purified ENaC were incubated with 2 . 5 µg/mL trypsin for 10 min at room temperature . These samples were then run through 4 – 20% Criterion SDS-PAGE gels and blotted onto nitrocellulose membranes according to manufacturer’s instructions ( Bio-Rad ) . After blocking overnight in 5% non-fat milk in TBS , membranes were incubated in primary antibody ( ENaC α subunit , 6 µg/blot SC-21012; ENaC β subunit , 6 µg/blot SC-21013; ENaC γ subunit , 10 µg/blot abcam ab133430 ) for 2 hr . The membranes were then incubated in 1 µg/blot IRDye 800CW goat anti-rabbit IgG ( Licor ) for 1 hr . Purified ΔENaC-Fab complexes were applied to glow-discharged Quantifoil holey carbon grids ( Au 1 . 2 µm/1 . 3 µm hole space/hole separation , 300mesh ) , blotted using a Vitrobot Mark III ( FEI ) with the following conditions , 7 s wait time , and 5 s blot time at 100% humidity , and then plunge-frozen in liquid ethane cooled by liquid nitrogen . All images were collected on a Titan Krios electron microscope operating at 300 kV at the Multiscale Microscopy Core ( OHSU ) . Images were recorded by a Gatan K2 Summit direct electron detector operating in super-resolution mode , and the images were collected using the automated acquisition program SerialEM ( Mastronade , 2003 ) . Magnification of the recorded images corresponded to a pixel size of 1 . 33 Å in counting mode ( 0 . 665 Å in super-resolution mode ) . For the ∆ENaC-10D4 complex , two data sets were acquired and were initially processed separately , and subsequently combined for 3D reconstruction . Each image in the first dataset was dose-fractionated to 30 frames with 0 . 5 s per frame and a total exposure time and dose of 15 s and 54 e-/Å2 , respectively . The second dataset was collected in counting mode , and was therefore not binned when combined with the first where each image was dose-fractionated to 60 frames with 0 . 25 s per frame and a total exposure time and dose of 15 s and 50 e-/Å2 , respectively . Similarly , two separate datasets were obtained for ∆ENaC-7B1/10D4 complex in super-resolution mode . Like in the monoFab complex , each data set was processed separately and later combined for further analysis and 3D reconstruction . The images of the first dataset of the diFab complex were dose-fractionated to 40 frames with 0 . 25 s per frame and a total exposure time and dose of 10 s and 62 e-/Å2 , respectively , while the images of the second dataset were dose-fractionated to 48 frames with 0 . 25 s per frame and a total exposure time and dose of 12 s and 71 e-/Å2 , respectively . The ∆ENaC-10D4 data set collected in super-resolution mode was binned 2 × 2 while the ∆ENaC-10D4 data set collected in counting mode was left unbinned . Both data sets were motion corrected using MotionCor2 ( Zheng et al . , 2017 ) , and automated particle selection was performed using DoGPicker ( Voss et al . , 2009 ) . Defocus values for individual particles were estimated using Gctf ( Zhang , 2016 ) , and particles belonging to low-abundance classes were removed via 2D classification and 3D classification in RELION ( Scheres , 2012 ) . The final set of particles was further analyzed in cryoSPARC and refined to a nominal resolution of 5 . 4 Å ( Punjani et al . , 2017 ) . For the ∆ENaC-7B1/10D4 data sets , super-resolution counting images were 2 × 2 binned , and motion corrected using MotionCor2 . Manual and automated particle selections were performed where DoGPicker was utilized for the latter resulting in a total of 667 , 984 particles . Defocus values for individual particles were estimated using Gctf , and particles of low-abundance classes via 2D classification in RELION were removed . For 3D classification in RELION , a reference model of a low-resolution map of ENaC-7B1/10D4 obtained from a data set ( 14 . 4 Å ) was low-pass filtered to 50 Å , and particles were classified into two classes where the major class contained 385 , 997 particles . Duplicates ( as a result of RELION2 . 0 re-centering particles after 2D classification ) and particles close to micrograph edges were removed , resulting in 329 , 180 particles that were subjected to ab initio 3D classification in cryoSPARC ( Punjani et al . , 2017 ) , and 3D classification and refinement in cisTEM ( Grant et al . , 2018 ) . Particles belonging to the low abundance class in cryoSPARC and cisTEM were discarded yielding 244 , 223 and 290 , 007 particles , respectively . Using default settings in cryoSPARC , particles with class probability of > 0 . 9 were used for refinement; thus , final reconstruction and refinement used 244 , 223 particles . For cisTEM , initial 3D classification and refinement was done using a refinement threshold of 8 Å and applying a mask during the last few iterations that excluded the constant domain ( Fc ) of the Fabs . During this process , we noticed that extraneous features , such as the micelle , were having a strong influence on alignment and classification , so the cisTEM particles were then re-processed using a mask that excluded both the micelle and the Fc of the Fabs , and aligned with a 5 . 4 Å limit . This dataset consisted of 302 , 263 particles and improved the resolution , as determined by the FSC = 0 . 143 criterion ( ~3 . 9 Å ) . More importantly , the electrostatic potential map was notably improved in the regions of interest . The resolutions reported in Figure 3—source data 1 are based on the FSC = 0 . 143 criterion ( gold-standard in the case of RELION and cryoSPARC ) . Final resolution reported in Figure 3—source data 1 are solvent adjusted FSC = 0 . 143 criterion . No symmetry was applied during data processing . Homology models of the human α , β , and γ subunits were generated with the crystal structure of cASIC ( Jasti et al . , 2007 ) ( PDB code: 2QTS , chain A ) as a template using SWISS-MODEL server and homology models for the Fabs were also generated by SWISS-MODEL ( Arnold et al . , 2006 ) . All models were docked into the EM potential in UCSF Chimera then rigid-body fitted into the EM potential using Coot ( Pettersen et al . , 2004; Emsley and Cowtan , 2004 ) . We incorporated models generated from Rosetta ( DiMaio et al . , 2011 ) into manual fitting and adjustments during model-building in Coot to build the palm , knuckle , TM , thumb , β-ball , and finger domains . To build the GRIP domains , we integrated analysis from Jpred4 , PSIPRED v3 . 3 , and QUARK online ab initio protein structure prediction to support our analysis of the cryo-EM map ( Buchan et al . , 2013; Xu and Zhang , 2012; Drozdetskiy et al . , 2015 ) . The final model was subjected to refinement using the module phenix . real_space_refine in PHENIX ( Adams et al . , 2011 ) . The cryo-EM map in all three subunits preceding the N-terminal side of the α2 helices was unambiguous and showed features consisting of two β-strands connected by a loop , the P3 and P4 segments of the GRIP domains ( Figure 4—figure supplement 5 ) . Secondary structure prediction analysis by online servers Jpred4 , PSIPRED v3 . 3 , and QUARK supported this observation ( Buchan et al . , 2013; Xu and Zhang , 2012; Drozdetskiy et al . , 2015 ) . We found that the potential map in the β subunit had the best-defined feature demonstrating four β strands ( Figure 4—figure supplement 5 ) . Based on the cryo-EM map , the P1 segments in α and γ adopt β strand-like conformations , like in the β subunit , which is also supported by the secondary structure prediction servers ( Figure 4 , Figure 4—figure supplement 5 ) . The regions between P1 and P3 in the α and γ subunits , however , are disordered in the cryo-EM map . We built stretches of residues into the P1 potential maps in the α- and γ-GRIP domains using sequence alignment with the β-GRIP domain and cryo-EM potential map features as guides . For validation , FSC curves were calculated between the final model and the EM map as well as the two half maps generated by cisTEM . We implemented MolProbity to analyze geometries of the atomic model ( Chen et al . , 2010 ) . All figures of map and atomic model were prepared using UCSF Chimera and Pymol ( Pettersen et al . , 2004 ) . All constructs used for two-electrode voltage clamp electrophysiology ( TEVC ) experiments were cloned into pGEM vector , linearized and transcribed to mRNA using mMESSAGE mMACHINE T7 Ultra Kit ( Ambion ) procedure . Xenopus laevis oocytes purchased from Ecocyte were injected with a volume of 50 nL containing either 0 . 5 – 1 . 0 ng of each FL-ENaC subunit mRNA or 5 ng of each ∆ENaC and Δ*ENaC subunit mRNA . For experiments containing combinations of FL-ENaC and ∆ENaC , 5 ng of each subunit mRNA was injected . Oocytes were incubated at 16°C for 12 – 48 hr in the presence of 100 µM amiloride and 250 µg/mL amikacin . The recordings were performed using two different ionic solutions with or without 100 µM amiloride ( 110 mM KCl and 110 mM NaCl ) where all buffers additionally contained 1 . 8 mM CaCl2 and 10 mM HEPES ( pH 7 . 4 ) . Macroscopic ENaC currents are defined as the difference between inwards currents obtained in the absence and in the presence of 100 μM amiloride . To test full activation of ∆ENaC constructs , 2 . 5 μg/mL Trypsin was perfused for 5 min in the presence of 100 µM amiloride . Amiloride-sensitive currents were recorded prior to Trypsin treatment as well as after in order to determine the increase in current amplitude . All recording experiments were carried out at a holding potential of −60 mV and repeated independently at least three times . HEK293S GnTI- cells were grown in suspension at a density of 2−4 × 106/ml in Freestyle medium with 2% FBS and transduced with the virus ( Δα , Δβ , and Δγ; or FL-α , FL-β , and FL-γ ) at a multiplicity of infection ( MOI ) of 1 and incubated in the presence of 500 nM phenamil mesylate at 30°C for 12 hr . Five hours before recording , cells were transferred to wells containing glass coverslips at a density of 0 . 3 – 0 . 5 × 106 cells/ml and in Dulbecco’s Modified Eagle Medium supplemented with 5% FBS and 500 nM phenamil mesylate . Whole-cell recordings were carried out 17 – 24 hr after transduction . Pipettes were pulled and polished to 2 – 2 . 5 MΩ resistance and filled with internal solution containing ( in mM ) : 150 KCl , 2 MgCl2 , 5 EGTA and 10 HEPES ( pH 7 . 35 ) . External solution contained ( in mM ) : 150 NaCl , 2 MgCl2 , 2 CaCl2 , 10 HEPES ( pH 7 . 4 ) , and 0 . 1 amiloride . Test external solution did not contain 0 . 1 mM amiloride . As in TEVC experiments , macroscopic ENaC currents are defined as the difference between inwards currents obtained in the absence and in the presence of 100 μM amiloride . Holding potential was at −60 mV . Six mg of mAb 10D4 was dialyzed into 0 . 2 M carbonate-bicarbonate ( Na2CO3/NaHCO3 ) solution buffered at pH 9 . 0 . The dialyzed mAb was concentrated to 6 mg/mL . Tetramethylrhodamine ( TRITC , ThermoFisher 46112 ) was dissolved in DMSO at a final concentration of 1 mg/mL . To the 10D4 solution , 35 μg of TRITC was slowly added and mixed thoroughly . The 10D4-TRITC mix was incubated at room temperature in the dark for 2 hr followed by gel filtration to remove excess TRITC . The carbonate-bicarbonate buffer was exchanged with Tris-buffered saline buffer ( 200 mM NaCl , 20 mM TRIS , pH 8 . 0 ) using PD-10 desalting column . The dye:protein molar ratio of the final TRITC-labeled mAb 10D4 in TBS buffer was approximately 2 . 8 . HEK293S GNTI- cells were resuspended from DMEM into 2 mL HBSS media , stained with 10 µg ( 5 µg/μL stock ) of WGA Alexa Fluor 647 conjugate ( ThermoFisher W32466 ) and 170 µg ( 4 . 9 µg/μL stock ) of 10D4-TRITC and subsequently incubated at 37°C for 10 min . The cells were then washed with PBS two times before resuspended in 1 mL HBSS . Live cell imaging was performed on a Yokogawa CSU-W1 spinning disk confocal microscope using a 60 × 1 . 4 Plan Apo VC objective . Images were acquired at a pixel size of 0 . 108 μm for three different wavelengths , starting at 640 nm , 561 nm and then 488 nm . Exposure time varied depending on sample intensity , but remained the same for each wavelength between the two samples of infected cells ( FL-ENaC and ∆ENaC ) , 400 ms for 640 nm , 2 s for 561 nm and 600 ms for 488 nm . Images were imported into Fiji for image analysis .

The bodies of humans and other animals contain many different fluids that play vital roles in the body , such as blood , saliva and the fluids that surround cells in organs . These fluids all contain particles called ions , which can affect the flow of water into and out of cells and alter the activity of proteins . Therefore , in order to survive , an animal must tightly regulate the levels of ions in its body . Epithelial cells line the surface of organs , and the inside of the digestive system and other cavities in the human body . A channel known as ENaC is found on the surface of epithelial cells and controls the volume of the fluid surrounding cells , blood pressure and the volume of liquid in the airways . This channel spans the membrane surrounding each epithelial cell and allows sodium ions to pass into the cell . To promote the opening of the channel , enzymes remove portions of the ENaC called extracellular domain , which sits on the outside surface of an epithelial cell . Three components ( or ‘subunits’ ) called alpha , beta and gamma are needed to form an ENaC , but it is not clear how they fit together to form a single working unit . Noreng et al . used a technique called cryo-electron microscopy to study the three-dimensional structure of the human ENaC . This revealed that a single channel contains one alpha , one beta and one gamma subunit , which sit next to each other to form a narrow tube through the membrane and a large extracellular domain . When viewed from the outside of the cell the subunits form a narrow ring in a counter-clockwise manner . Further analysis of the structure suggested that when enzymes remove pieces of the extracellular domain of ENaC , it becomes easier for the rest of the channel to adopt a shape that allows sodium ions to move through the pore . A next step will be to study the three-dimensional structure of ENaC when it takes on different shapes to better understand how it works .

2018

Structure of the human epithelial sodium channel by cryo-electron microscopy

The bacterial flagellar motor , a cell-envelope-embedded macromolecular machine that functions as a cellular propeller , exhibits significant structural variability between species . Different torque-generating stator modules allow motors to operate in different pH , salt or viscosity levels . How such diversity evolved is unknown . Here , we use electron cryo-tomography to determine the in situ macromolecular structures of three Gammaproteobacteria motors: Legionella pneumophila , Pseudomonas aeruginosa , and Shewanella oneidensis , providing the first views of intact motors with dual stator systems . Complementing our imaging with bioinformatics analysis , we find a correlation between the motor’s stator system and its structural elaboration . Motors with a single H+-driven stator have only the core periplasmic P- and L-rings; those with dual H+-driven stators have an elaborated P-ring; and motors with Na+ or Na+/H+-driven stators have both their P- and L-rings embellished . Our results suggest an evolution of structural elaboration that may have enabled pathogenic bacteria to colonize higher-viscosity environments in animal hosts . The bacterial flagellum is a macromolecular machine that transforms the movement of ions ( H+ , Na+ or other cations ) across the cell membrane into a mechanical torque to move the bacterial cell through its environment ( Ito and Takahashi , 2017; Sowa and Berry , 2008 ) . In general , the flagellum consists of a cell-envelope-embedded motor , a hook which acts as a universal joint and a long propeller-like filament ( Berg , 2003; Erhardt et al . , 2010 ) . The motor is composed of a rotor and a stator: while the stator is the part of the motor that remains static , the rotor is the part that rotates and can rotate the filament in either a counterclockwise or clockwise direction . For cells with a single flagellum this drives the cell forward or backward; for peritrichous cells this results in ‘run’ or ‘tumble’ movements . Flagella can also exhibit a more complex behavior; it was recently reported that the Shewanella putrefaciens flagellum can wrap around the cell to mediate a screw-like motion that allows the cell to escape narrow traps ( Kühn et al . , 2017 ) . Besides their role in motility , bacterial flagella participate in other vital activities of the cell such as biofilm formation ( Belas , 2014 ) . Moreover , the virulence of many human pathogens depends directly on their flagella , with flagellated strains of Pseudomonas aeruginosa and Legionella pneumophila causing more serious infections with higher mortality rates ( Appelt and Heuner , 2017; Feldman et al . , 1998 ) . P . aeruginosa lacking fully-assembled flagella cause no mortality and are 75% less likely to cause pneumonia in mice ( Feldman et al . , 1998 ) . The best-studied flagellar motor , in Salmonella enterica , consists of several sub-complexes , which we will describe in order from the inside out . On the cytoplasmic side are the inner-membrane-embedded MS ring ( formed by the protein FliF ) and the C-ring ( aka the switch complex , formed by FliN , FliM and FliG ) . The C-ring encircles a type III secretion system ( T3SS ) export apparatus ( FliH , FliI , FliJ , FlhA , FlhB , FliP , FliQ and FliR ) . Spanning the space from the inner membrane to the peptidoglycan cell wall is the ion channel ( called the stator ) , a complex of two proteins ( MotA and MotB ) with 4:2 stoichiometry ( Koebnik , 1995; Kojima , 2015; Morimoto and Minamino , 2014 ) . This complex is anchored to the peptidoglycan and converts the flux of ions across the bacterial membrane into a torque through the interaction of the so-called ‘torque-helix’ in MotA with FliG in the C-ring . Previous studies have shown that cycles of protonation/deprotonation of a certain aspartate residue in the cytoplasmic end of MotB induce conformational changes in MotA , which in turn interacts with the C-terminus of FliG . How the torque that is generated through this interaction is transferred to the other parts of the motor remains unclear with different models suggested ( see Berg , 2003; Stock et al . , 2012 and references therein for details ) . The MS ring is coupled to the extracellular hook ( FlgE ) through the rod ( FlgB , FlgC , FlgF and FlgG ) . The rod is further surrounded by two other rings: the P- ( peptidoglycan , FlgI ) and the L- ( lipopolysaccharide , FlgH ) rings which act as bushings during rod rotation . Extending from the hook is the filament ( FliC ) which is many micrometers in length . In addition to these components , the assembly of the whole flagellar motor is a highly synchronized process that requires a plethora of additional chaperones and capping proteins ( Altegoer and Bange , 2015; Evans et al . , 2014; Jones and Macnab , 1990; Kaplan et al . , 2018; Kubori et al . , 1992; Macnab , 1999 ) . Recently , the development of electron cryo-tomography ( ECT ) ( Gan and Jensen , 2012; Oikonomou and Jensen , 2017; Pfeffer and Förster , 2018 ) has allowed the determination of the complete structures of flagellar motors in their cellular milieu at macromolecular ( ~5 nm ) resolution . ECT studies of many different bacterial species have revealed that while the core structure described above is conserved , the flagellar motor has evolved many species-specific adaptations to different environmental conditions ( Beeby et al . , 2016; Chaban et al . , 2018; Chen et al . , 2011; Minamino and Imada , 2015; Terashima et al . , 2017; Zhao et al . , 2014; Zhu et al . , 2017 ) . For example , extra periplasmic rings were found to elaborate the canonical P- and L-rings in the motor of the Gammaproteobacteria Vibrio species . These rings are called the T-ring ( MotX and Y ) and H-ring ( FlgO , P and T ) ( Terashima et al . , 2010; Terashima et al . , 2006 ) . Unlike the S . enterica motor described above , which is driven by H+ ions , the motors of Vibrio and other marine bacteria employ different stators ( PomA and PomB ) which utilize Na+ . These Na+-dependent stators generate higher torque ( ~2200 pN ) than H+-dependent stators ( ~1200 pN ) , driving the motor at higher speeds ( up to 1 , 700 Hz compared to ~300 Hz in H+-driven motors ) ( Magariyama et al . , 1994 ) . Many flagellated bacteria use a single stator system – either H+-driven or Na+-driven , depending on their environment . Interestingly , it has also been shown that the single stator system of Bacillus clausii KSM-K16 is able to use both Na+ and H+ at different pH levels ( Terahara et al . , 2008 ) . Additionally , some species ( like alkaliphilic Bacillus alcalophilus AV1934 and Paenibacillus sp . TCA20 ) , can use other cations to generate the energy required for torque generation depending on their environment ( Ito and Takahashi , 2017; Terahara et al . , 2012 ) . Some species , however , such as Vibrio alginolyticus , use two distinct types of motors to move in different environments: a polar Na+-driven flagellum and lateral H+-driven flagella . Still other species employ dual stator systems with a single flagellar motor , conferring an advantage for bacteria that experience a range of environments ( see Thormann and Paulick , 2010 and references therein ) . For example , P . aeruginosa employs a dual H+-driven stator system ( MotAB and MotCD ) . While the MotAB system is sufficient to move the cell in a liquid environment ( Doyle et al . , 2004 ) , MotCD is necessary to allow the cell to move in more viscous conditions ( Toutain et al . , 2007 ) . Shewanella oneidensis MR-1 combines both Na+- and H+-dependent stators in a single motor , enabling the bacterium to move efficiently under conditions of different pH and Na+ concentration ( Paulick et al . , 2015 ) . How these more elaborate motors may have evolved remains an open question . Here , we used ECT to determine the first in situ structures of three Gammaproteobacteria flagellar motors with dual stator systems: in L . pneumophila , P . aeruginosa and S . oneidensis MR-1 . L . pneumophila and P . aeruginosa have dual H+-dependent stator systems and S . oneidensis has a dual Na+-H+-dependent stator . This imaging , along with bioinformatics analysis , shows a correlation between the structural elaboration of the motor and its stator system , suggesting a possible evolutionary pathway . To determine the structures of the flagellar motors of L . pneumophila , P . aeruginosa , and S . oneidensis we imaged intact cells of each species in a hydrated frozen state using ECT . We identified clearly visible flagellar motors in the tomographic reconstructions and performed sub-tomogram averaging to enhance the signal-to-noise ratio , generating a 3D average of the motor of each species at macromolecular resolution ( Figure 1 and Figure 1—figure supplement 1 ) . Although the three motors shared the conserved core structure of the flagellar motor , they exhibited different periplasmic decorations surrounding this conserved core . While the S . oneidensis and P . aeruginosa averages showed clear densities corresponding to the stators ( Figure 1E , F , K and L , dark orange density ) , none were visible in the L . pneumophila average , suggesting that they were more variable , or dynamic and therefore are not visible in the average ( see e . g . , ( Chen et al . , 2011; Zhu et al . , 2017 ) ) . Interestingly , we observed a novel feature in the S . oneidensis motor: an extra ring outside the outer membrane ( Figure 1A–F , purple density ) . Although in some tomograms two extracellular rings appeared to be present ( see Figure 1A and C ) , only one ring was visible in the sub-tomogram average which could be either because one of the rings is more dynamic or substoichiometric ( Figure 1B , D and Figure 1—figure supplement 2 for more examples of single motors ) . This structure is reminiscent of the O-ring ( outer membrane ring ) described recently in the sheathed flagellum of Vibrio alginolyticus ( Zhu et al . , 2017 ) . However , while the V . alginolyticus O-ring was associated with a 90° bend in the outer membrane , no such outer membrane bend was seen in the unsheathed S . oneidensis flagellum , so the function of this structure remains mysterious . The most striking difference between the three motor structures was the L- and P-rings , which were highly elaborated in S . oneidensis . The P . aeruginosa and L . pneumophila motors lacked additional rings associated with the L-ring , but showed smaller elaborations of their P-rings . To determine whether flagellar motor structure correlates with stator type , we compared our three new ECT structures with those of the five previously-published Gammaproteobacteria motors ( Figure 2 ) . Two motors ( Escherichia coli and S . enterica ) have a single H+-driven stator system , two motors have dual H+-dependent stator systems ( P . aeruginosa and L . pneumophila ) , three motors have Na+-driven systems ( the three Vibrio species ) and one motor has a dual Na+-H+-driven system ( S . oneidensis ) . Interestingly , we found that motors with similar stator type also shared similar structural characteristics . While the two motors with a single H+-dependent stator system did not show any periplasmic elaborations beyond the conserved flagellar core , the dual H+-dependent stator systems had an extra ring surrounding their P-ring , with no embellishment of the L-ring . The Na+-dependent motors of the Vibrio spp . , together with the Na+-H+-dependent motor of S . oneidensis , have extra components surrounding both their P- and L- rings . In Vibrio , these extra periplasmic rings are known as the T-ring ( surrounding the P- ring and formed by the MotX and MotY proteins ) and the H-ring ( surrounding the L-ring and consisting of the FlgO , FlgP and FlgT proteins ) . The presence of the T- and H-rings was suggested to be specific to the Na+-driven Vibrio motors ( Minamino and Imada , 2015 ) with the FlgT protein required for the formation of both rings ( Terashima et al . , 2013 ) . Previous studies showed that MotX and MotY are important for flagellar rotation in S . oneidensis but it was not known whether they form part of the motor or not ( Koerdt et al . , 2009 ) . Similarly , bioinformatics analysis and biochemical studies showed that MotY is involved in the function of the P . aeruginosa motor , but the structural basis of this role was not known ( Doyle et al . , 2004 ) . We therefore performed a bioinformatics search for candidate homologs of MotX , MotY , FlgO , FlgP and FlgT in the genomes of P . aeruginosa , L . pneumophila and S . oneidensis to examine whether there is a correlation between the presence of homologous genes and the extra periplasmic rings observed in the ECT structures . While we found candidates for all five proteins constituting the T- and H-rings in S . oneidensis as previously suggested ( Wu et al . , 2011 ) , only MotY candidates were found in L . pneumophila and P . aeruginosa ( Table 1 ) . This is in accordance with our ECT structures , which showed that L . pneumophila and P . aeruginosa motors have a ring surrounding only their P-rings while the S . oneidensis motor has rings surrounding both the P- and L-rings . These rings are likely T- and H-rings , respectively , as in Vibrio . The lack of candidate MotX homologs in the genomes of L . pneumophila and P . aeruginosa ( Table 1 ) is consistent with their lack of PomB , the component of the Na+-dependent stator with which MotX interacts . Interestingly , the absence of candidates for FlgT in the L . pneumophila and P . aeruginosa genomes suggests that it may not be required for the recruitment of MotY as in Vibrio species . To see whether these correlations hold more broadly , we expanded our bioinformatics analysis to additional species of Gammaproteobacteria ( Williams et al . , 2010 ) . We examined the genomes of species with single H+-driven stator systems ( Table 2 ) , dual H+-driven stator systems ( Table 3 ) and Na+-driven stator systems ( Table 4 ) . These species were identified either by Blasting the sequence of the stator proteins ( MotA , B , C and D and Pom A and B , see Materials and methods ) against the genome of the species or based on previous studies ( Thormann and Paulick , 2010 ) . In all species we examined , we observed the same pattern: ( i ) genomes of species with single H+-driven stator systems lacked homologs of H- or T-ring components; ( ii ) genomes of species with Na+ ( or Na+-H+ ) stator systems contained homologs of all H- and T-ring components , and ( iii ) genomes of species with dual H+-driven stator systems contained candidate homologs only for the T-ring component MotY . The sole exception to this rule was Chromohalobacter salexigens DSM 3043 , which contained a homolog of FlgO in addition to MotY . Also , while Serratia proteomaculans and Psychromonas ingrahamii have candidates for single MotAB stator system they also have candidates for MotY ( see Supplementary file 1 and 2 ) . None of the thirteen species with dual H+-driven stator systems we examined contained a homolog of FlgT , further suggesting that it is not essential for MotY stabilization in this group . Together , our results from ECT imaging of flagellar motors in situ and bioinformatics analysis reveal a correlation between the structural elaboration of the flagellar motor of Gammaproteobacteria and the type of its torque-generating unit , the stator ( summarized in Figure 3 ) . Low-speed motors with single H+-stator systems have only the P- and L-rings , while high-speed motors using Na+ have two extra periplasmic rings , the T- and H-rings . Unexpectedly , we find that motors with dual H+-driven stator systems represent a hybrid structure between the two , elaborating their P-rings with one of the five components of the T- and H-rings , MotY . It is important to note that the presence of these extra elaborations in the motor is encoded in the genome and is not related to whether or not a stator subunit is recruited on the motor . This extra MotY ring might help to stabilize the motor under conditions of increased load , as in the viscous environment of the pulmonary system encountered by L . pneumophila and P . aeruginosa . These results therefore suggest an evolutionary pathway in which these pathogenic Gammaproteobacteria species could have borrowed a motor stabilization strategy from related Na+-driven motors to allow them to colonize animal hosts . Finally , It would be interesting to investigate whether our observation here holds for other bacterial species that use different cations as their energy source ( Ito and Takahashi , 2017 ) and whether it extends to other bacterial species with more than two stator systems or other classes . Legionella pneumophila ( strain Lp02 ) cells were grown on plates of ACES [N- ( 2-acetamido ) −2-aminoethanesulfonic acid]-buffered charcoal yeast extract agar ( CYE ) or in ACES-buffered yeast extract broth ( AYE ) with 100 μg/ml thymidine . Ferric nitrate and cysteine hydrochloride were added to the media . For ECT experiments , cells were harvested in early stationary phase . Shewanella oneidensis MR-1 cells belonging to the strains listed in Supplementary file 3 were used in this study . They were grown using one of the following methods: Luria–Bertani ( LB ) broth culture , chemostat , the batch culture method or in a perfusion flow imaging platform . Detailed descriptions of these methods can be found in Subramanian et al . ( 2018 ) . Briefly , in the chemostat method , 5 mL of a stationary-phase overnight LB culture was injected into a continuous flow bioreactor containing an operating liquid volume of 1 L of a defined medium ( Pirbadian et al . , 2014 ) , while dissolved oxygen tension ( DOT ) was maintained at 20% . After 20 hr , and as the culture reached stationary phase , continuous flow of the defined medium ( Pirbadian et al . , 2014 ) was started with a dilution rate of 0 . 05 hr−1 while DOT was still maintained at 20% . After 48 hr of aerobic growth under continuous flow conditions , the DOT was manually reduced to 0% . O2 served as the sole terminal electron acceptor throughout the experiment . pH was maintained at 7 . 0 , temperature at 30°C , and agitation at 200 rpm . Either 24 or 40 hr after DOT reached 0% , samples were taken from the chemostat for ECT imaging . In the batch culture method , 200 μL of an overnight LB culture of S . oneidensis cells was added to each of two sealed and autoclaved serum bottles containing 60 mL of a defined medium ( Pirbadian et al . , 2014 ) . One of the two bottles acted as a control and was not used for imaging . To this control bottle , 5 μM resazurin was added to indicate the O2 levels in the medium . The bottles were then placed in an incubator at 30°C , with shaking at 150 rpm until the color due to resazurin in the control bottle completely faded , indicating anaerobic conditions . At this point , samples were taken for ECT imaging from the bottle that did not contain resazurin . For the perfusion flow imaging experiments , S . oneidensis cells were grown overnight in LB broth at 30°C to an OD600 of 2 . 4–2 . 8 and washed twice in a defined medium ( Pirbadian et al . , 2014 ) . A glow-discharged , carbon-coated , R2/2 , Au NH2 London finder Quantifoil EM grid was glued to a 43 mm × 50 mm no . 1 glass coverslip using waterproof silicone glue ( General Electric Company ) and let dry for ~ 30 min . Using a vacuum line , the perfusion chamber ( model VC-LFR-25; C and L Instruments ) was sealed against the grid-attached glass coverslip . A total of ~10 mL of the washed culture was injected into the chamber slowly to allow cells to settle on the grid surface , followed by a flow of sterile defined medium from an inverted serum bottle through a bubble trap ( model 006BT-HF; Omnifit ) into the perfusion chamber inlet . Subsequently , the flow of medium was stopped and the perfusion chamber was opened under sterile medium . The grid was then detached from the coverslip by scraping off the silicone glue at the grid edges using a 22-gauge needle and rinsed by transferring three times in deionized water , before imaging by ECT . Samples were also prepared from an aerobic S . oneidensis LB culture grown at 30°C to an OD600 of 2 . 4–2 . 8 . Pseudomonas aeruginosa PAO1 cells were first grown on LB plates at 37°C overnight . Subsequently , cells were inoculated into 5 ml MOPS [ ( 3- ( N-morpholino ) propanesulfonic acid ) ] Minimal Media Limited Nitrogen and grown for ~24 hr at 30°C . Many of the flagellar motors analyzed here were taken from tomograms recorded for more than one purpose . The Shewanella oneidensis MR-1 mutants , for instance , were grown under different growth conditions for the purpose of studying the nanowires formed by these cells ( see Subramanian et al . , 2018 ) , but all their motors were presumably the same , so we included them here to increase the clarity and resolution of our average . Cells ( L . pneumophila , P . aeruginosa and S . oneidensis ) from batch cultures and chemostats were mixed with BSA ( Bovine Serum Albumin ) -treated 10 nm colloidal gold solution ( Sigma-Aldrich , St . Louis , MO , USA ) and 4 μL of this mixture was applied to a glow-discharged , carbon-coated , R2/2 , 200 mesh copper Quantifoil grid ( Quantifoil Micro Tools ) in a Vitrobot Mark IV chamber ( FEI ) . Excess liquid was blotted off and the grid was plunge frozen in a liquid ethane/propane mixture for ECT imaging . Imaging of ECT samples ( S . oneidensis and P . aeruginosa ) was performed on an FEI Polara 300-keV field emission gun electron microscope ( FEI company , Hillsboro , OR , USA ) equipped with a Gatan image filter and K2 Summit counting electron-detector camera ( Gatan , Pleasanton , CA , USA ) . Data were collected using the UCSF Tomography software ( Zheng et al . , 2007 ) , with each tilt series ranging from −60° to 60° in 1° increments , an underfocus of ~ 5–10 μm , and a cumulative electron dose of ~ 130–160 e-/A2 for each individual tilt series . For L . pneumophila samples , imaging was done using an FEI Titan Krios 300 kV field emission gun transmission electron microscope equipped with a Gatan imaging filter and a K2 Summit direct electron detector in counting mode ( Gatan ) . L . pneumophila data was also collected using UCSF Tomography software and a total dose of ~ 100 e-/A2 per tilt series with ~ 6 um underfocus . The IMOD software package was used to calculate three-dimensional reconstructions of tilt series ( Kremer et al . , 1996 ) . Alternatively , the images were aligned and contrast transfer function corrected using the IMOD software package before producing SIRT reconstructions using the TOMO3D program ( Agulleiro and Fernandez , 2011 ) . Sub-tomogram averages with 2-fold symmetrization along the particle Y-axis were produced using the PEET program ( Nicastro et al . , 2006 ) . To obtain the sub-tomogram averages of the flagellar motors we reconstructed 156 tomograms of Pseudomonas aeruginosa , 50 of Legionella pneumophila and ~ 300 of Shewanella oneidensis MR-1 . The averages were obtained by averaging 144 sub-volumes P . aeruginosa , 100 sub-volumes S . oneidensis MR-1 and 45 sub-volumes L . pneumophila . Candidate H- and T-ring component genes were identified by sequence alignment of the following Vibrio cholerae proteins against the fully sequenced genomes of each bacterial species using BLASTP ( https://www . genome . jp/tools/blast/ ) . The Vibrio cholerae proteins used were: MotX ( Q9KNX9 ) , MotY ( Q9KT95 ) , FlgO ( Q9KQ00 ) , FlgP ( Q9KQ01 ) and FlgT ( Q9KPZ9 ) . To check for the stator system candidates in different species , the following proteins were blasted against the genome of the bacterial species: PomAB proteins of V . cholerae ( Q9KTL0 and Q9KTK9 respectivley ) , MotAB proteins of E . coli ( P09348 and P0AF06 respectively ) and MotCD of P . aeruginosa ( G3XD73 and G3XD90 respectively ) using BLASTP . Candidate MotX and MotY homologs identified were adjacent to the flagellar cluster in the genome , and for each stator system candidate homologs were characteristically located in tandem in the genome . The codes in parentheses represent Uniprot IDs . An E-value cutoff of <1×10−10 was used . The raw BLAST results for all species are shown in Supplementary file 1 and 2 . Note that for the stator system , a candidate stator locus was considered only when two neighboring candidates for Mot/B , MotC/D or PomA/B were found .

Bacteria are so small that for them , making their way through water is like swimming in roofing tar for us . In response , these organisms have evolved a molecular machine that helps them move in their environment . Named the bacterial flagellum , this complex assemblage of molecules is formed of three main parts: a motor that spans the inner and outer membranes of the cell , and then a ‘hook’ that connects to a long filament which extends outside the bacterium . More precisely , the motor is formed of the stator , an ion pump that stays still , and of a rotor that can spin . Different rings can also be present in the space between the inner and outer membranes ( the periplasm ) and surround these components . The stator uses ions to generate the energy that makes the rotor whirl . In turn , this movement sets the filament in motion , propelling the bacterium . Depending on where the bacteria live , the stator can use different types of ions . In addition , while many species have a single stator system per motor , some may have several stator systems for one motor: this may help the microorganisms move in different conditions . As microbes colonize environments with a different pH or viscosity , they constantly evolve new versions of the motor which are more suitable to their new surroundings . However , a part of the motor remains the same across species . Overall , it is still unclear how bacterial flagella evolve , but examining the structure of new motors can shed light upon this process . Here , Kaplan et al . combine a bioinformatics approach with an imaging technique known as electron cryo-tomography to dissect the structure of the flagellar motor of three species of bacteria with different stator systems , and compare these to known motors of the same class . The results reveal a correlation between the nature of the stator system and the presence of certain elements . Stators that use sodium ions , or both sodium and hydrogen ions , are associated with two periplasmic rings surrounding the conserved motor structure . These rings do not exist in motors with single hydrogen-driven stators . Motors with dual hydrogen-driven stators are , to some extent , an ‘intermediate state’ , with only one of those rings present . As all the studied species currently exist , it is difficult to know which version of the motor is the most ancient , and which one has evolved more recently . Capturing the diversity of bacterial motors gives us insight into the evolutionary forces that shape complex molecular structures , which is essential to understand how life evolved on Earth . More practically , this knowledge may also help us design better nanomachines to power microscopic robots .

2019

The presence and absence of periplasmic rings in bacterial flagellar motors correlates with stator type

Memory dysfunction is a key symptom of age-related dementia . Although recent studies have suggested positive effects of electrical stimulation for memory enhancement , its potential targets remain largely unknown . In this study , we hypothesized that spatially targeted deep brain stimulation of ventromedial prefrontal cortex enhanced memory functions in a middle-aged rat model . Our results show that acute stimulation enhanced the short- , but not the long-term memory in the novel-object recognition task . Interestingly , after chronic high-frequency stimulation , both the short- and long-term memories were robustly improved in the novel-object recognition test and Morris water-maze spatial task compared to sham . Our results also demonstrated that chronic ventromedial prefrontal cortex high-frequency stimulation upregulated neurogenesis-associated genes along with enhanced hippocampal cell proliferation . Importantly , these memory behaviors were strongly correlated with the hippocampal neurogenesis . Overall , these findings suggest that chronic ventromedial prefrontal cortex high-frequency stimulation may serve as a novel effective therapeutic target for dementia-related disorders . Memory loss is the key symptom of dementia-related disorders along with impaired cognitive functioning such as language or reasoning . It is usually caused by Alzheimer's disease and other age-related dementia . Its prevalence doubles from a low rate in 60–64 age group to 40–50% of those older than 85 ( Lobo et al . , 2000 ) . Dementia is a progressive disease , which has a detrimental impact on the quality of life for patients . To date , pharmacological treatments for dementia have limited effects and there are no known treatments that cure or delay the progression of this memory impairment ( Doody et al . , 2014; Salloway et al . , 2014 ) . Therefore , a novel non-pharmacological approach such as deep brain stimulation ( DBS ) is currently considered as an alternative treatment to reduce the symptomatic and progression of this memory deterioration ( Hescham et al . , 2013a ) . DBS , a technique of minimally invasive surgical implantation of electrodes with delivering of electrical impulses into the brain , has been demonstrated to control a wide range of neurological disorders and neuropsychiatric diseases ( Sesia et al . , 2009; Temel et al . , 2012b; Temel and Lim , 2013 ) . In line with these developments , evidence from recent studies suggests that DBS might enhance memory functions when particular brain areas are stimulated ( Hamani et al . , 2008; Laxton et al . , 2010; Suthana et al . , 2012; Hescham et al . , 2013b ) . Of particular interest , DBS of the subgenual anterior cingulate cortex or the ventromedial prefrontal cortex ( vmPFC ) induced striking antidepressant activity in both patients and animal studies ( Mayberg et al . , 2005; Lozano et al . , 2008; Hamani et al . , 2010b; Kennedy et al . , 2011; Temel and Lim , 2013; Lim et al . , 2015b ) . Despite encouraging results , no studies have shown the putative role of vmPFC DBS in learning and memory performance . In the realm of cognitive function , there is empirical evidence indicating that vmPFC plays an important role in the formation , consolidation , and retrieval of memory , as well as reward and decision making ( Maviel et al . , 2004; Euston et al . , 2012 ) . Based on the human imaging and rodent studies , the vmPFC was significantly activated during the recall of remote memory ( Bontempi et al . , 1999; Takashima et al . , 2006a , 2006b; Gais et al . , 2007 ) , while its inactivation caused memory impairment when tested in the radial arm-maze ( Maviel et al . , 2004 ) , the Morris water-maze ( MWM ) ( Teixeira et al . , 2006 ) , and the contextual fear conditioning ( Frankland et al . , 2004 ) . In line with these studies , malfunctioning has also been reported in the hippocampus and the vmPFC ( which received robust projections from the hippocampal formation ) in early stages of Alzheimer's disease , frontotemporal dementia , and healthy aging-related memory impairments ( Salat et al . , 2001; Lindberg et al . , 2012 ) . Given the potential mechanisms involved by DBS including the increase of hippocampal brain-derived neurotrophic factor ( BDNF ) levels ( Hamani et al . , 2012; Ying et al . , 2012 ) and neurogenesis-related functions ( Toda et al . , 2008; Kadar et al . , 2011; Stone et al . , 2011 ) , we tested the hypothesis that vmPFC DBS-enhanced memory function by modulating the hippocampal neurogenic activity in the middle-aged rat model with aging-related memory impairment . The use of this animal model was supported by previous data that showed aged-related deficits in both the memory and the hippocampal functioning ( Rex et al . , 2005; Kaczorowski and Disterhoft , 2009 ) . In acute DBS , animals were tested with either high- or low-frequency stimulation ( HFS or LFS ) at various amplitudes using the conventional novel-object recognition ( NOR ) test . Subsequently , another set of animals was used to assess the chronic stimulation effects on memory enhancement using the NOR and the MWM tests . For investigation of the underlying mechanism , we analyzed the effects of chronic stimulation on the molecular and cellular levels of hippocampal neurogenesis-related functions . Progressive age-related memory decline has been previously described for human ( Davis et al . , 2003 ) and animal studies ( Sloane et al . , 1997; Ward et al . , 1999; Kaczorowski and Disterhoft , 2009 ) . We compared the short- and long-term memory functions in young ( n = 20 ) and middle-aged ( n = 15 ) rats using the NOR test ( Figure 1A ) . Three-way ANOVA ( group age × retention interval × object ) with repeated-measures showed significant effects for object ( F ( 1 , 82 ) = 18 . 043 , p < 0 . 001 ) , retention interval ( F ( 2 , 82 ) = 13 . 956 , p < 0 . 001 ) , and the interaction group age × retention interval × object ( F ( 1 , 82 ) = 4 . 160 , p = 0 . 019 ) ( Figure 1C , D ) . No differences were observed for group ( F ( 1 , 82 ) = 0 . 009 , p = n . s . ) . With regard to the duration of object exploration , there was no significant difference between the young and middle-aged rats in the acquisition phase ( t ( 28 ) = −0 . 742 , p = n . s . ) , see Supplementary file 1A . However , a decrease in the duration of novel object exploration was observed for the middle-aged group when compared to the young in the long-term ( t ( 26 ) = 4 . 129 , p < 0 . 001 ) , but not the short-term ( t ( 29 ) = 0 . 014 , p = n . s . ) memory . Interestingly , the young animals spent relatively more time with the novel object as compared to the familiar object in both the short- ( t ( 18 ) = −5 . 23 , p < 0 . 001 ) and long-term ( t ( 14 ) = −8 . 722 , p < 0 . 001 ) phase ( Figure 1C , D ) . In the middle-aged rats , no significant effect was found for discrimination between the novel and familiar objects in the short- and long-term memory retention interval ( all t ( 11 ) > −2 . 058 , p = n . s . ) , indicating a possible manifestation of memory deficit in the middle-aged animals . 10 . 7554/eLife . 04803 . 003Figure 1 . Experimental protocol of the novel-object recognition test ( A ) , and representative illustration of the stimulating electrode localization in the vmPFC ( B ) . The box plots show the comparisons between young ( 4 month old ) and middle-aged ( 12 month old ) animals on the short- and long-term memory retention interval in the novel-object recognition task ( C , D ) . Note: there was a decrease of time spent in the novel object exploration in the middle-aged animals as compared to the young rats , suggesting a possible manifestation of memory deficit in this animal model . Indication: * , significant difference from the middle-aged rats; # , significant difference from the familiar object of respective age animals , ( p < 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04803 . 003 To test the hypothesis that electrical stimulation enhances memory functions , the middle-aged animals were stereotactically implanted with electrodes in the vmPFC region . All localization of electrode tips were verified within the vmPFC target as illustrated in a representative Figure 1B . Animals with electrode misplacement or detachment in the acute stimulation ( HFS: 50 μA , n = 2; 100 μA , n = 2; 400 μA , n = 3; and LFS: 50 μA , n = 1; 400 μA , n = 2; and Sham , n = 4 ) and chronic stimulation ( vmPFC HFS , n = 3; Sham , n = 3 ) experiments were excluded from data analysis . Overall , the final number of rats per group was as follows for the acute stimulation with either HFS ( 50 μA , n = 8; 100 μA , n = 8; 200 μA , n = 10; and 400 μA , n = 7 ) or LFS ( 50 μA , n = 11; 100 μA , n = 12; 200 μA , n = 12; and 400 μA , n = 8 ) in comparison with the sham animals ( n = 12 ) . In chronic stimulation experiment , the final number of rats per group was as follows for the vmPFC HFS ( n = 12 ) and the sham ( n = 9 ) animals . For determination of acute stimulation efficacy in memory functions , animals were tested with either HFS or LFS at amplitudes varying across 50 , 100 , 200 , and 400 μA . Both the short- and long-term memory functions were assessed using the NOR test . In the HFS animals , repeated-measures three-way ANOVA ( group × retention interval × object ) showed main effects for group ( F ( 4 , 116 ) = 5 . 873 , p < 0 . 001 ) , retention interval ( F ( 2 , 116 ) = 16 . 670 , p < 0 . 001 ) , object ( F ( 1 , 116 ) = 9 . 552 , p = 0 . 003 ) , and the interaction group × retention interval × object ( F ( 8 , 116 ) = 2 . 342 , p = 0 . 023 ) . Similarly , in the LFS animals , there were significant effects for object ( F ( 1 , 145 ) = 63 . 815 , p < 0 . 001 ) , retention interval ( F ( 2 , 145 ) = 16 . 418 , p < 0 . 001 ) , group ( F ( 4 , 145 ) = 2 . 544 , p = 0 . 042 ) , and interaction group × retention interval × object ( F ( 8 , 145 ) = 3 . 112 , p = 0 . 003 ) . In the acquisition phase , no differences were found in the duration of object exploration for both the HFS ( F ( 4 , 40 ) = 1 . 509 , p = n . s . ) and the LFS ( F ( 4 , 50 ) = 0 . 478 , p = n . s . ) groups , see Supplementary file 1B , C . In acute HFS , stimulation at 200 μA significantly increased the duration of novel object exploration in the short- ( F ( 4 , 39 ) = 7 . 995 , p < 0 . 001 ) , but not the long-term ( F ( 4 , 39 ) = 1 . 553 , p = n . s . ) memory when compared to the sham ( Figure 2A , B ) . The novel object exploration was also higher compared to the familiar object ( t ( 9 ) = −14 . 636 , p < 0 . 001 ) . In acute LFS , stimulation at 50 , 200 , and 400 μA induced a longer duration of novel object exploration in the short-term memory ( F ( 4 , 49 ) = 4 , 432 , p = 0 . 004 ) when compared to the sham ( Figure 2C , D ) . As for the long-term memory , Bonferroni post-hoc test revealed no significant difference for the novel object exploration between groups . Interestingly , for comparisons of discrimination between novel and familiar objects , LFS at 50 , 100 , 200 , and 400 μA ( all t ( 6–11 ) > −6 . 250 , p < 0 . 001 ) increased the duration of novel object exploration in the short-term memory . In terms of the long-term memory , an increase for novel object exploration was found with LFS at 50 , 100 , and 200 μA ( all t ( 9–11 ) > −3 . 440 , p < 0 . 029 ) , but not at 400 μA ( all t ( 6 ) = 0 . 969 , p = n . s . ) when compared to the familiar object . 10 . 7554/eLife . 04803 . 004Figure 2 . The box plots show the effects of either high- ( A , B ) or low-frequency ( C , D ) stimulation at amplitudes varying across 50 , 100 , 200 , and 400 μA in the middle-aged animals . Both the short- and long-term memory functions were tested using the novel-object recognition test . Note: HFS ( 100 Hz ) at 200 μA and LFS ( 10 Hz ) at 50 , 200 , 400 μA significantly increased the novel object exploration as compared to the sham animals , respectively . Indication: * , significant difference from the sham rats; # , significant difference from the familiar object of respective stimulation amplitude , ( p < 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04803 . 004 In this study , expression of neurogenesis-related genes was quantified using qPCR assay . The selection of candidate genes for qPCR was based on our previous microarray data ( Kadar et al . , 2011 ) . We found significant effects for group [F ( 1 , 10 ) = 52 . 948 , p < 0 . 001] , genes [F ( 8 , 80 ) = 386 . 955 , p < 0 . 001] , and the interaction group × genes [F ( 8 , 80 ) = 2 . 443 , p = 0 . 02] . Remarkably , vmPFC HFS upregulated genes related with neurogenesis and neuroplasticity ( NeuN/Rbfox3 , t ( 10 ) = −7 . 018 , p < 0 . 001; Syn , t ( 10 ) = −4 . 660 , p = 0 . 001; Dcx , t ( 10 ) = −2 . 860 , p = 0 . 012; Nes , t ( 10 ) = −3 . 214 , p = 0 . 009 ) , genes related with neuronal differentiation ( Angpt2 , t ( 10 ) = −3 . 520 , p = 0 . 006; and S100a4 , t ( 10 ) = −3 . 372 , p = 0 . 007 ) , as well as genes related with migration and neuroprotective functions ( Angpt2: t ( 10 ) = −3 . 520 , p = 0 . 006 ) in the hippocampus ( Figure 4A ) . No changes were found for Timp1 , Ccl2 , and BDNF ( all t ( 10 ) > −1 . 781 , p = n . s . ) . Calculation for the fold-change values using 2-Delta ( Delta Ct ) method indicated that vmPFC HFS induced approximately 4 . 8-fold ( Angpt2 ) , 2-fold ( NeuN/Rbfox3 ) , and >1-fold ( Syn , Dcx , Nes , S100a4 ) increase of gene expression relative to the sham ( Figure 4B ) . Although no differences were observed for Timp1 , Ccl2 , and BDNF genes , their fold-changes were increased by approximately 1 . 8-fold for Timp1 , 2-fold for Ccl2 , and 1 . 2-fold for BDNF , respectively . Interestingly , the gene expression for Syn was significantly correlated with the Nes and Dcx genes ( all r2 > −0 . 831 , p < 0 . 046; Figure 4C , D ) , indicating a close association between these genes for induction of neuroplasticity in the hippocampus after chronic vmPFC HFS . 10 . 7554/eLife . 04803 . 006Figure 4 . Effects of chronic vmPFC HFS on the mRNA gene expression related to neuroplasticity in the hippocampus ( A ) . Note: vmPFC HFS upregulated genes involved in proliferation and neurogenesis-related functions including the NeuN , Syn , Dcx , Nes , Angpt2 , and S100a4 relative to sham . No changes were found for Timp1 , Ccl2 , and BDNF . Calculation for the fold-change values indicating that vmPFC HFS induced approximately 4 . 8-fold ( Angpt2 ) , 2-fold ( NeuN ) , and >1-fold ( Syn , Dcx , Nes , S100a4 ) increase of gene expression relative to the sham ( B ) . Interestingly , scatter plots show significant correlation between the Syn and the Nes/Dcx ( C , D ) , indicating that these genes are strongly related to each other for neuroplasticity in the hippocampus after chronic vmPFC HFS . Gene expression was expressed as the change in Ct of the gene of interest compared to the sham ( Delta Ct ) ; and relative expression was calculated using the comparative CT method with fold change 2-Delta ( Delta Ct ) . Indication: * , significant difference from the sham rats , ( p < 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04803 . 006 The present findings confirm the results of previous studies of progressive age-related memory impairment in the middle-aged rats ( Rex et al . , 2005; Kaczorowski and Disterhoft , 2009 ) . We next conducted electrical stimulation in this animal model of memory deficit , using HFS and LFS with various stimulation current intensities in the NOR test . Our results showed that HFS at 200 μA and LFS at 50 , 200 , and 400 μA significantly enhanced the memory functions in the short- , but not the long-term memory retention interval when compared to the sham . We next carried out chronic stimulation in this middle-aged rat model . We hypothesized that chronic stimulation would increase both the short- and long-term memory functions via a mechanism of enhanced hippocampal neuroplasticity . Previous studies have shown that memory deficits were partly due to the disruption of the hippocampal neuroplasticity ( Deupree et al . , 1993; Rex et al . , 2005 ) . Therefore , our hypothesis was driven by findings that DBS in various brain targets increased BDNF level ( Hamani et al . , 2012; Ying et al . , 2012 ) and enhanced neurogenesis-related functions ( Toda et al . , 2008; Kadar et al . , 2011; Stone et al . , 2011 ) in the hippocampus . Further , this hypothesis was also supported by the fact that the increased BDNF level and neurogenesis function in the hippocampus were strongly correlated with the hippocampal-dependent memory tests ( Drapeau et al . , 2003; Erickson et al . , 2011 ) . Thus , after chronic stimulation , we found significant improvement in both the short- and long-term memories in the NOR test , as well as the spatial memory performances during the MWM task as compared to the sham . In the acute stimulation experiment , we found that the behavioral effects of vmPFC stimulation were dependent on the stimulation frequency and current intensity . Although these findings are consistent with previous reports ( Hamani et al . , 2010a; Hescham et al . , 2013b ) , the mechanisms of the stimulation parameter dependency in regulation of memory functions remain largely obscure . Nonetheless , it is postulated that the neurons in the vmPFC are highly sensitive to specific stimulation settings for induction of either long-term potentiation ( LTP ) or depression ( LTD ) . This notion was highly supported by the findings in which the vmPFC synapses are potentially vulnerable to LTP or LTD strengthening after specific type of stimulation ( Herry and Garcia , 2002 ) . It has been found that the prefrontal cortex neurons could undergo LTP with rapid and stable potentiation in the prelimbic synapses after high-frequency tetanic stimulation of the hippocampal CA1/subicular region ( Laroche et al . , 1990 ) . Such a LTP-like plasticity in the vmPFC after hippocampal HFS has been shown to be dependent on α-amino-3-hydroxy-5-methyl-4-isoxazole propionic acid ( AMPA ) and N-methyl-D-aspartate ( NMDA ) -receptor , which indicates a crucial role for the synaptic potentiation in the hippocampal-vmPFC pathway in rapid memory consolidation ( Laroche et al . , 2000 ) . Although HFS and LFS at specific amplitudes were effective for memory enhancement , it is noteworthy that LFS induced negative effects on the anxiety-related behaviors ( Lim et al . , 2015b ) . Furthermore , HFS of the vmPFC has been shown to reduce conditioned fear and enhance the extinction of aversive memory ( Milad and Quirk , 2002; Milad et al . , 2004 ) , in contrast to LFS , which induced impairment in extinction of conditioned fear ( Shehadi and Maroun , 2012 ) . Since previous studies of vmPFC HFS produced robust antidepressant-like behaviors ( Lim et al . , 2015a , 2015b ) , we therefore applied this stimulation parameter in the chronic stimulation experiment to test the hypothesis that it would restore the memory deficits for both the short- and long-term memory functions of the middle-aged rats . As might be expected for the effects after chronic stimulation , there was an improvement on memory function in the short- but not the long-term memory retention interval when the animals were tested with no-HFS prior to the NOR task . Strikingly , we found a remarkable reversal of memory deficits in both the short- and long-term memory of the animals that received HFS prior to the NOR test as compared to the non-stimulated sham . Congruently , chronic vmPFC HFS significantly improved the spatial navigation performances when conducted in a hippocampal-dependent memory MWM test . In the present data , we clearly distinguished a role for the vmPFC in regulation of both the short- and long-term memory functions . In most studies , short-term memory is generally regarded as memory spanning from seconds to several minutes or hours , while long-term memory is usually last from hours to several days or longer ( Nagai et al . , 2007; Euston et al . , 2012 ) . Albeit many studies have indicated that vmPFC is involved in the expression of long-term memory ( Maviel et al . , 2004; Teixeira et al . , 2006 ) , there are also findings supporting that vmPFC is involved in consolidation and retrieval of recently acquired memories ( Blum et al . , 2006; Leon et al . , 2010 ) . Thus , it is likely that the vmPFC HFS potentiated the initial hippocampal encoding during the acquisition phase , which was then followed by enhanced retrieval during the short- and long-term memory recalls . Taken together , these findings further implicate a specific role for the vmPFC HFS in facilitation of rapid consolidation and retrieval of the short- and long-term memory processes in the hippocampus . HFS of the vmPFC has not only been demonstrated for memory enhancement in the middle-aged memory deficit rat model , but also induced profound antidepressant-like behaviors in the experimental animal studies ( Lim et al . , 2015a; 2015b ) . Interestingly , the effects of memory improvement by vmPFC HFS , as characterized by the increased novelty seeking in the NOR and MWM tasks , were highly associated with the stimulation effects on reduction in anxiety behavior . These results confirm previous findings that animals with lower anxiety or fear level displayed higher novelty seeking behavior , particularly in the elevated plus-maze and light–dark box tests ( Kabbaj et al . , 2000; Stead et al . , 2006 ) . In consistent with the data found in the animal model of depression , rats undergoing chronic stress exhibited anxious and low exploratory behaviors ( Lim et al . , 2015b ) , as well as impairment in spatial memory task performances ( Conrad et al . , 1996; Beck and Luine , 1999 ) . Apparently , chronic stress produces pathological alterations on the molecular and morphological levels in the hippocampus that is involved in the regulation of both the spatial memory formation and emotional behaviors . Thus , our experiments found that the memory enhancement effect is likely to be accompanied by the vmPFC DBS-induced anxiolytic effects via the mechanisms of neurogenesis and dendritic remodeling in the hippocampal neurons . Given the prominent anatomical connectivity between the hippocampus and vmPFC ( Jay and Witter , 1991; Cenquizca and Swanson , 2007 ) , it was observed that vmPFC HFS drove the local neural activity as characterized by c-Fos-ir activation in the subiculum , DG , as well as a marginal increase in the CA1 field of the hippocampus . Although there is no direct connection from vmPFC to the hippocampus , it is possible that the effects of vmPFC HFS on the hippocampal neural activity are mediated by a reciprocal bisynaptic pathway through the nucleus reunions or the lateral entorhinal cortex ( Burwell and Amaral , 1998; Vertes et al . , 2007 ) . Besides , the neural activity in the hippocampus could probably be activated by either antidromic or orthodromic stimulation that possibly achieved by a current spread from the vmPFC structure to the neighboring axon bundles—the minor forceps of the corpus callosum ( Lim et al . , 2015b ) . As a result , the hippocampal regions would eventually be activated for an induction of LTP to strengthen its synaptic plasticity for memory processes . Of particular interest , a tractography analysis by diffusion tensor imaging provides evidence for this structural connectivity that HFS of the subgenual cingulate gyrus ( generally considered to be homologue of the rat vmPFC ) , showed connections to the medial frontal cortex , anterior and posterior cingulate , and the anterior medial temporal lobe ( i . e . , amygdala-hippocampus ) ( Gutman et al . , 2009 ) . The connections of the hippocampus and amygdala to the vmPFC have been previously investigated by anatomical and electrophysiological studies ( Laroche et al . , 1990; Jay and Witter , 1991 ) . It has been shown that the excitatory and inhibitory inputs from the amygdala and hippocampus were converged and interact in the vmPFC ( Ishikawa and Nakamura , 2003 ) , implying that activation of the amygdalar-hippocampal neurons might be crucial for vmPFC neurons in memory regulation . More importantly , the electrophysiological studies have provided concrete evidence of functional interaction between the vmPFC and hippocampus in which their theta oscillations were highly synchronized as measured by both the spike-theta phase locking and local field potential coherence , during memory acquisition and retrieval in spatial tasks ( Jones and Wilson , 2005; Siapas et al . , 2005; Benchenane et al . , 2010 ) . Based on this evidence , we suggested that the rapid encoding and retrieval of memory depend largely on the bidirectional regulation of synaptic connectivity between the vmPFC and hippocampus; while the disruption of its connection affect the learning and memory functions , which are commonly identified in patients suffering from dementia ( Salat et al . , 2001 ) . In line with our findings , HFS of the anterior nucleus of the thalamus and the entorhinal cortex has been demonstrated to increase neurogenesis ( Toda et al . , 2008 ) and memory functions , particularly spatial memory measured in the MWM ( Stone et al . , 2011 ) and enhanced performance on a delayed non-matching to sample task ( Hamani et al . , 2011 ) . In this study , we found that vmPFC HFS induced upregulation of neurogenesis-associated genes with increased neural progenitors cells and dendritic spines in the DG of the hippocampus . In agreement with our previous microarray data ( Kadar et al . , 2011 ) , chronic HFS has been shown to modulate the hippocampal genes that involved in the proliferation and neurogenesis-related functions , as well as genes supporting for neural differentiation , migration and maturation . Although we observed no differences for the Timp1 ( neuroprotection ) , Ccl2 ( neural differentiation ) , and BDNF ( synaptic plasticity ) gene expression , there was an overall increase in their fold-change with approximately 1 . 2–2 fold after chronic vmPFC HFS . Notably , we found a strong induction of upregulation in the Angpt2 gene ( 4 . 8-fold ) that promotes neuronal differentiation , migration and neuroprotection ( Liu et al . , 2009 ) , as well as NeuN/Rbfox3 gene ( 2-fold ) , which plays an essential role for neural progenitor cells differentiation and maturation ( Kim et al . , 2009 , 2013 ) . Further , our present observations were also supported by earlier works demonstrating that vmPFC HFS induced significant increase of BDNF and serotonin ( 5-HT ) levels in the hippocampus ( Hamani et al . , 2012 ) . Recent studies have shown that the increase of 5-HT and BDNF expression in the hippocampus regulate synaptic plasticity , as well as cognitive and mood-related behaviors . It is well-known that both the 5-HT and BDNF promote neurogenesis by enhancing the synaptogenesis , neuronal differentiation , and survival particularly for memory acquisition and consolidation ( Gaspar et al . , 2003; Lu and Chang , 2004; Pang et al . , 2004 ) . Although no microdialysis data were provided with regard to the extracellular levels of 5-HT and BDNF after chronic vmPFC HFS , the increased neurogenesis effects as obtained from this study clearly indicate that its plausible mechanism is facilitated by the release of 5-HT and enhanced BDNF levels ( Hamani et al . , 2012 ) . Importantly , the vmPFC HFS effects on the behavioral observation of improved memory functions and the neurogenesis have indicated that a strong synaptic network circuitry has been established within the DG for integration of new information and memory storage . In support of this notion , previous studies have demonstrated that the increased neurogenesis was highly associated with enhanced learning and memory , while its decrease caused memory impairment ( van Praag et al . , 1999; Shors et al . , 2001 ) . Although the present study has identified the DBS-induced memory enhancement by neurogenesis , there is a possibility that these effects are mediated by other non-neurogenic mechanisms such as modulation of the neurotransmission ( via acetylcholine , dopamine , 5-HT , etc ) , synaptic potentiation by the AMPA/NMDA receptor trafficking , and enhancement of the BDNF or CREB ( cAMP response element-binding protein ) function ( Hamani et al . , 2012; Stern and Alberni , 2012 ) . The above findings prompted us to further examine whether the increase of neurogenesis-related functions is associated with the effects of memory enhancement by vmPFC HFS . Our correlation analysis of gene expression revealed that the Syn was strongly associated with the Nes and Dcx genes , suggesting that vmPFC HFS might possibly cause an alteration of increased dendritic synaptogenesis , particularly in the hippocampal DG where transcriptional process for Nes and Dcx genes occurs . Consistent with this interpretation , we have demonstrated the increase of dendritic spines in the Golgi-impregnated cells of the hippocampal DG . This observation was supported by a previous study which showed that electrical stimulation affected the axonal path in cultured Xenopus neurons that was mediated by elevation of both cytoplasmic Ca2+ and cyclic adenosine monophosphate levels ( Ming et al . , 2001 ) . Importantly , our behavioral correlates show that the spatial memory performances were associated with the cell proliferation in the DG , indicating that neurogenesis in the DG is vital for the hippocampal-dependent learning and memory processes . In conclusion , our findings suggest that chronic vmPFC HFS induces long-lasting effects on memory performances and its underlying mechanism is possibly mediated by an enhanced neurogenesis in the hippocampus . Despite the fact that this structure has been previously shown for antidepressant-like activities , it might as well serve as a new effective DBS target for aged-related memory deficits . Thus , this translational research provides an additional window for a possibility of future clinical trials on this potential brain target for memory enhancement . Male Sprague–Dawley rats ( 12 month old , n = 144; and 4 month-old , n = 20; National University of Singapore , Singapore ) were individually housed with ad libitum access to food and water . The animal colony was maintained under controlled temperature ( about 24–26°C ) , humidity ( 60–70% ) , and 12 hr dark/light cycle ( lights-off at 0700 ) . All procedures were approved by the Institutional of Animals Care and Use Committee of Nanyang Technological University . The NOR test was performed to compare the short- and long-term memory functions between the young ( n = 20 ) and middle-aged ( n = 15 ) rats ( Figure 1 ) . In the acute DBS experiments , animals ( n = 102 ) received either HFS ( n = 40 ) or LFS ( n = 46 ) , at various stimulation amplitudes ( n = 10–12 per group ) ( Figure 2 ) . The sham group ( n = 16 ) was similarly operated with electrode implantation in the vmPFC . In the chronic DBS experiment , animals ( vmPFC HFS , n = 15; sham , n = 12 ) received stimulation 1 hr daily for a period of 4 weeks . In week 1 , animals were intraperitoneally injected with 5-bromo-2′-deoxyuridine ( BrdU , Sigma , Missouri , USA; 150 mg/kg per injection; at 2-hr intervals for three times per day ) on day 1 , 3 , 5 , and 7 ( Figure 8 ) . The first injection dose was performed immediately before the 1-hr stimulation , followed by the second and third dose at 2-hr intervals . The memory functions were tested using the hippocampal-dependent memory tasks—NOR and the MWM tests on days 18–27 . 10 . 7554/eLife . 04803 . 010Figure 8 . Schematic representation of the experimental design for chronic stimulation and behavioral testing of memory functions in the middle-aged animals . DOI: http://dx . doi . org/10 . 7554/eLife . 04803 . 010 Surgery was performed as previously described ( Lim et al . , 2010 , 2015a ) . In brief , rats were anesthetized ( 2 . 5% isoflurane inhalation ) and placed in a stereotactic frame ( Vernier Stereotaxic Instrument , Leica Biosystems , Nussloch , Germany ) . A bilateral stimulating electrode was implanted in the vmPFC ( AP: +2 . 70 mm; L: ±0 . 60 mm; V: 4 . 60 mm ) based on the rat brain atlas of Paxinos and Watson ( 1998 ) . All animals were given a 2-week recovery period . For stimulation , a bipolar stimulating electrode ( Synergy , Singapore ) was constructed using an inner platinum-iridium core wire with a gold-plated cannula ( Technomed , Beek , Netherlands ) ( Lim et al . , 2009; Tan et al . , 2010 ) . A digital stimulator DS8000 and stimulus isolators DLS100 ( World Precision Instruments , Sarasota , USA ) were used to deliver the electrical stimuli . In the acute DBS experiment , either HFS ( 100 Hz ) or LFS ( 10 Hz ) with stimulation amplitudes of 50 , 100 , 200 , and 400 μA was used . The pulse width was set at 100 μs . In the chronic DBS experiment , the stimulation parameter ( HFS , 200 μA amplitude , and 100 μs pulse width ) , derived from the present acute DBS study ( Figure 2 ) and previous findings ( Hamani et al . , 2010a; Lim et al . , 2015b ) , was used to test the hypothesis that chronic stimulation enhances both the short- and long-term memory functions . In behavioral experiments , all animals in the acute and chronic stimulation studies received electrical stimulation for approximately 30 min immediately prior to each of the NOR test phases , and MWM training and probe tests . The behavioral testing was conducted in a dimly lighted room from 8:00 till 14:00 hr . In the chronic stimulation experiments , all animals were stimulated daily for 1 hr during 14:00–19:00 hr . However , animals that had already received the 30-min stimulation prior to the behavioral testing were again stimulated for another 30 min during 14:00–19:00 hr , so that each animal received a total of 1-hr stimulation per day . Sham animals were similarly connected to cables , without electrical stimulation . Animals were handled daily to avoid unnecessary stress during behavioral experiments . 2 days after the last behavioral test , the animals were stimulated for 2 hr and immediately decapitated with isoflurane anesthesia . The brains were subsequently removed , frozen in liquid nitrogen , and stored at −80°C for gene expression and morphological studies . For Golgi study , the brains ( vmPFC HFS , n = 4; Sham , n = 3 ) from the chronic stimulation experiment were removed after decapitation , processed for rapid Golgi staining , and coronal sections of 100 μm were obtained as previously described ( Vyas et al . , 2003 ) . For immunohistochemistry , the brains ( vmPFC HFS , n = 8; Sham , n = 6 ) from the chronic stimulation experiment were serially cut on a cryostat CM3050 ( Leica Microsystems , Wetzlar , Germany ) into 20-μm coronal sections and collected on gelatin-coated slides . For neurogenesis-related gene expression study , the hippocampal region from the chronic stimulation experiment was separately collected in Eppendorf tubes and micro-dissected ( 400 μm thickness ) for real-time quantitative polymerase chain reaction ( qPCR ) assay . Before staining , all sections were incubated with 4% paraformaldehyde for 1 hr , followed by 0 . 3% H2O2 treatment for 10 min . For BrdU staining , sections were first incubated in 2N HCl for 30 min at 37°C . The primary antibody incubation was performed using a rabbit anti-c-Fos antibody ( diluted 1:400 ) , mouse-anti-BrdU antibody ( diluted 1:50 ) , and goat anti-Dcx antibody ( diluted 1:50 ) ( all primary antibodies , Santa Cruz Biotechnology , Inc , Dallas , USA ) , for 3 days at 4°C . After rinsing , all sections were incubated with a corresponding secondary biotinylated goat anti-rabbit antibody , biotinylated horse anti-mouse antibody , or biotinylated rabbit anti-goat antibody ( all dilutions 1:200; Vector Laboratories , Burlingame , CA , USA ) for 1 day . Next , the sections were incubated with an avidin-biotin-peroxidase complex ( diluted 1:200 in standard Vectastain Elite ABC kit; Vector Laboratories ) for 4 hr , followed by incubation in solution 3 , 3′-diaminobenzidine tetrahydrochloride ( DAB Substrate Kit; Vector Laboratories ) with nickel chloride enhancement for visualization of the immune complex of the horseradish peroxide reaction product . Finally , all sections were dehydrated , and cover-slipped with Permount ( Thermo Fisher Scientific , Waltham , USA ) . For histological quantification , the counting for c-Fos immunoreactive ( c-Fos-ir ) , BrdU and Dcx-positive cells ( Lim et al . , 2008; Hestermann et al . , 2014 ) , and measurement of dendritic spine density ( Vyas et al . , 2003 ) were performed using previously established methods with minor modifications . In brief , the c-Fos-ir cells per mm2 ( six sections per animals ) was counted in the subiculum , dentate gyrus , CA1 , and CA3 field of the hippocampus using image analysis program ‘Image J’ ( version 1 . 47 , NIH , USA ) . For BrdU- and DCX-positive cells , quantification ( six sections per animal ) was performed using a bright-field microscope ( Olympus , Japan ) . For Golgi measurement of dendritic spine density , the quantifications were restricted to the primary and the secondary dendrite branches ( each , six sections per animal ) . The spines were counted along a 50-μm stretch of the dendrite starting from the origin of the soma or secondary branch at 100× magnification ( Olympus BX43 microscope , 100× Objective , Tokyo , Japan ) . The immunofluorescence staining was performed based on previously established protocols ( Temel et al . , 2012a; Lim et al . , 2015b ) . After pre-blocking for 30 min in PBS-Triton ( PBS-T ) with 10% normal donkey serum , two double labeling stainings were carried out on the vmPFC HFS hippocampal sections using goat anti-Dcx ( 1:50 ) and rabbit anti-c-Fos antibody ( 1:500 ) ; as well as mouse anti-BrdU ( 1:50 ) and goat anti-c-Fos ( 1:100 ) as primary antibodies ( all antibodies , Santa Cruz Biotechnology , Inc . ) in 5% normal donkey serum for 3-day incubation . After rinsing , the sections were incubated with corresponding Alexa Fluor secondary antibodies ( Alexa Fluor 594 donkey anti-rabbit , Alexa Fluor 488 donkey anti-goat , and Alexa Fluor 594 horse anti-mouse; each 1:200; Vector Laboratories ) for 2 hr at room temperature . Finally , the sections were mounted on the Superfrost micro-slides ( VWM , Illinois , USA ) and cover-slipped with Vectashield ( Vector Laboratories ) . To analyze sections for co-localization of cells , photographs were taken using a digital camera that was connected to a laser-scanning confocal microscope ( Carl Zeiss , Oberkochen , Germany ) . Total RNA was isolated from the hippocampal area ( 400 μm ) of frozen tissue using TRIzol reagent ( Life Technologies , Carlsbad , USA ) as recommended and converted into cDNA . Real-time qPCR for neuroplasticity-related gene expression ( Neuronal nuclei , NeuN/Rbfox3; Synaptophysin , Syn; Doublecourtin , Dcx; Nestin , Nes; Angiopoietin-2 , Angpt2; S100-calcium-binding protein a4 , S100a4; TIMP metallopeptidase inhibitor-1 , Timp1; Chemokine [C–C motif] ligand-2 , Ccl2; and BDNF ) was performed using thermal cycler ( Applied Biosystems 7500 , Foster City , USA ) and SYBR Green quantitative PCR mix ( Applied Biosystems , Life Technologies , Warrington , UK ) . For details of primer sequences used , see Table 1 . Data analysis of relative gene expression with reference to internal control by real-time PCR ( Delta Ct ) was quantified . Fold change was calculated using the comparative CT method as the ratio of the 2− and the 2 ( -Delta Delta C ( T ) ) method . 10 . 7554/eLife . 04803 . 011Table 1 . The primers sequences used for real-time quantitative-PCR analysisDOI: http://dx . doi . org/10 . 7554/eLife . 04803 . 011Gene symbol5′–3′ primer sequenceNeuN ( Rbfox3 ) Fwd 5′–GGCTGGAAGCTAAACCCTGT–3′; Rev 5′–TCCGATGCTGTAGGTTGCTG–3′SynFwd 5′–GTGCCAACAAGACGGAGAGT–3′; Rev 5′–TTGGTAGTGCCCCCTTTGAC–3′DcxFwd 5′–ACGACCAAGACGCAAATGGA–3′; Rev 5′–AGGCCAAGGATCTGACTTG –3′NesFwd 5′–TAAGTTCCAGCTGGCTGTGG–3′; Rev 5′–ATAGGTGGGATGGGAGTGCT–3′Angpt2Fwd 5′–GGACCCTGCAGCTACACATT–3′; Rev 5′–TGTCACAGTAGGCCTTGACC–3′S100a4Fwd 5′–CTTGGTCTGGTCTCAACGGT–3′; Rev 5′–GCAGCTTCGTCTGTCCTTCT–3′Timp1Fwd 5′–ACGCTAGAGCAGATACCACG–3′; Rev 5′– GATCGCTCTGGTAGCCCTTC–3′Ccl2Fwd 5′–AGCCAACTCTCACTGAAGCC–3′; Rev 5′–TGGGGCATTAACTGCATCTGG–3′BDNFFwd 5′–AGGACAGCAAAGCCACAATGTTC–3′; Rev 5′–TTGCCTTGTCCGTGGACGTTTG–3′HPRTFwd 5′–AGGCCAGACTTTGTTGGATT–3′; Rev 5′–GCTTTTCCACTTTCGCTGAT–3′ Data analysis was performed using the IBM SPSS Statistics 20 . The results were presented in box plots ( with interquartile ranges and S . E . M ) or mean ± S . E . M , unless otherwise indicated . Kolmogorov–Smirnov test was used to examine the data normality distribution . The behavior data were analyzed by either three-way or four-way ANOVA with repeated-measures , and Bonferroni post-hoc tests or independent sample t-test was used for detailed comparisons , as appropriate . Paired sample t-test was used to compare differences between the novel and familiar objects . The data for gene expression and morphological study were analyzed by either independent sample t-test or non-parametric Mann–Whitney U test , as appropriate . Pearson correlation coefficients with Bonferroni correction were calculated to examine the relationship between different variables related with the hippocampal neuroplasticity and the behavioral measures . All p-values <0 . 05 were considered significant .

Memory loss in older people is a serious and widespread problem that affects up to 50% of those over the age of 85 . It is a key symptom of dementia , but despite the growing impact of this disease on society , there are no treatments currently available that can effectively stop or delay the progression of the symptoms . One therapy that may reduce memory loss is called deep brain stimulation . Electrodes are implanted into the brain and used to stimulate brain cells in particular areas of the brain to alter mental and emotional processes . Deep brain stimulation is already used to treat Parkinson's disease , depression and other conditions that affect how the brain works . Liu et al . studied the effect of deep brain stimulation on memory in rats . The experiments show that middle-aged rats performed less well in short- and long-term memory tests than young rats . Next , Liu et al . investigated whether deep brain stimulation could improve memory in the middle-aged rats . The electrodes were positioned to stimulate a region near the front of the brain called the ‘ventromedial prefrontal cortex’; this region is important for the formation and recall of memories . Liu et al . then gave the rats a series of memory tasks that tested their recall after 90 minutes ( to test their short-term memory ) , and after 24 hours ( to test long-term memory ) . The experiments reveal that a brief stimulation of brain cells in this region of the brain improved the rats' short-term memory , but not their long-term memory . However , more sustained stimulation of this region of the brain improved both the short-term and long-term memory of the rats . Furthermore , deep brain stimulation led to the formation of new brain cells in another region of the brain called the hippocampus , which is also involved in memory . The hippocampus had not been in direct contact with the electrodes so the increase in brain cells was due to its connections with the stimulated ventromedial prefrontal cortex . Liu et al . 's findings suggest that deep brain stimulation of the ventromedial prefrontal cortex has the potential to be developed into a therapy to treat dementia and other conditions that lead to memory loss in humans .

2015

Ventromedial prefrontal cortex stimulation enhances memory and hippocampal neurogenesis in the middle-aged rats

Alteration of antibiotic binding sites through modification of ribosomal RNA ( rRNA ) is a common form of resistance to ribosome-targeting antibiotics . The rRNA-modifying enzyme Cfr methylates an adenosine nucleotide within the peptidyl transferase center , resulting in the C-8 methylation of A2503 ( m8A2503 ) . Acquisition of cfr results in resistance to eight classes of ribosome-targeting antibiotics . Despite the prevalence of this resistance mechanism , it is poorly understood whether and how bacteria modulate Cfr methylation to adapt to antibiotic pressure . Moreover , direct evidence for how m8A2503 alters antibiotic binding sites within the ribosome is lacking . In this study , we performed directed evolution of Cfr under antibiotic selection to generate Cfr variants that confer increased resistance by enhancing methylation of A2503 in cells . Increased rRNA methylation is achieved by improved expression and stability of Cfr through transcriptional and post-transcriptional mechanisms , which may be exploited by pathogens under antibiotic stress as suggested by natural isolates . Using a variant that achieves near-stoichiometric methylation of rRNA , we determined a 2 . 2 Å cryo-electron microscopy structure of the Cfr-modified ribosome . Our structure reveals the molecular basis for broad resistance to antibiotics and will inform the design of new antibiotics that overcome resistance mediated by Cfr . A large portion of clinically relevant antibiotics halt bacterial growth by binding to the ribosome and inhibiting protein synthesis ( Tenson and Mankin , 2006; Wilson , 2009; Arenz and Wilson , 2016 ) . Since antibiotic binding sites are primarily composed of ribosomal RNA ( rRNA ) , rRNA-modifying enzymes that alter antibiotic binding pockets are central to evolved resistance ( Vester and Long , 2013; Wilson , 2014 ) . The rRNA-methylating enzyme Cfr modifies an adenosine nucleotide located within the peptidyl transferase center ( PTC ) , a region of the ribosome essential for catalyzing peptide bond formation and consequently , a common target for antibiotics ( Schwarz et al . , 2000; Kehrenberg et al . , 2005 ) . Cfr is a radical SAM enzyme that methylates the C8 carbon of adenosine at position 2 , 503 ( m8A2503 , Escherichia coli numbering ) ( Jensen et al . , 2009; Kaminska et al . , 2010; Yan et al . , 2010; Yan and Fujimori , 2011; Grove et al . , 2011 ) . Due to the proximal location of A2503 to many antibiotic binding sites , introduction of a single methyl group is sufficient to cause resistance to eight classes of antibiotics simultaneously: phenicols , lincosamides , oxazolidinones , pleuromutilins , streptogramin A ( PhLOPSA ) , in addition to nucleoside analog A201A , hygromycin A , and 16-membered macrolides ( Long et al . , 2006; Smith and Mankin , 2008; Polikanov et al . , 2015 ) . Among rRNA modifying enzymes , this extensive cross-resistance phenotype is unique to Cfr and presents a major clinical problem . Cfr emergence in human pathogens appears to be a recent event , with the first case reported in 2007 from a patient-derived Staphylococcus aureus isolate ( Toh et al . , 2007; Arias et al . , 2008 ) . Since then , the cfr gene has been identified across the globe in both gram-positive and gram-negative bacteria , including E . coli ( Shen et al . , 2013; Vester , 2018 ) and has been associated with several clinical resistance outbreaks to the oxazolidinone antibiotic , linezolid ( Morales et al . , 2010; Locke et al . , 2010; Bonilla et al . , 2010; Cai et al . , 2015; Layer et al . , 2018; Lazaris et al . , 2017; Dortet et al . , 2018; Weßels et al . , 2018 ) . The vast spread of Cfr is attributed to its association with mobile genetic elements and relatively low impact on bacterial fitness , suggesting that cfr can be rapidly disseminated within bacterial populations ( LaMarre et al . , 2011; Schwarz et al . , 2016 ) . Due to the ability of Cfr to confer resistance to several antibiotics simultaneously , it is critical to understand how bacteria may adapt under antibiotic pressure to enhance Cfr activity and bolster protection against ribosome-targeting molecules . Identification of Cfr mutations that improve resistance will also be critical for informing clinical surveillance and designing strategies to counteract resistance . A major limitation in our current understanding of Cfr-mediated resistance is the lack of structural insight into changes in the ribosome as a result of Cfr modification . Steric occlusion of antibiotic binding has been proposed as a model to rationalize altered antibiotic susceptibility ( Polikanov et al . , 2015 ) . Additionally , the observation that A2503 can adopt both syn and anti-conformations in previously reported ribosome structures suggests that methylation may regulate conformation of the base , as previously proposed ( Toh et al . , 2008; Schlünzen et al . , 2001; Tu et al . , 2005; Stojković et al . , 2020 ) . However , direct evidence for how m8A2053 alters antibiotic binding sites to inform the design of next-generation molecules that can overcome Cfr resistance is lacking . In this study , we identified mechanisms that enhance antibiotic resistance by performing directed evolution of a cfr found in a clinical MRSA isolate under antibiotic selection ( Barlow and Hall , 2003 ) . The obtained highly resistant Cfr variants show increased rRNA methylation , driven primarily by robust improvements in Cfr cellular levels , achieved either by higher transcription or increased translation and improved cellular stability . In particular , mutation of the second Cfr amino acid to lysine strongly enhances translation and resistance . Finally , we used an evolved variant which achieves near-stoichiometric rRNA methylation to generate a high-resolution cryo-electron microscopy ( EM ) structure of the Cfr-modified E . coli ribosome . The obtained structural insights provide a rationale for how m8A2503 causes resistance to ribosome antibiotics . To perform directed evolution of Cfr , we used error-prone PCR ( EP-PCR ) to randomly introduce 1–3 mutations into the cfr gene obtained from a clinical MRSA isolate ( Toh et al . , 2007 ) , herein referred to as CfrWT ( Figure 1a ) . Mutagenized cfr sequences were then cloned into a pZA vector where Cfr was expressed under tetracycline-inducible promoter Ptet introduced to enable precise control of Cfr expression ( Wellner et al . , 2013 ) . The resulting library of ~107 E . coli transformants was selected for growth in the presence of increasing amounts of tiamulin , a pleuromutilin antibiotic to which Cfr confers resistance . During each round , a subset of the surviving colonies was sequenced to identify new mutations . After two rounds of evolution , wild-type Cfr was no longer detected , indicating that the introduced mutations provide enhanced survivability in the presence of tiamulin . After five rounds of mutation and selection , we performed two rounds of selection without mutagenesis , and with high tiamulin concentrations , thus leading to fixation of mutations that provide robust resistance . Analysis of surviving cfr sequences from the final rounds of selection revealed notable trends ( Supplementary file 1 ) . Three positions were primarily mutated: N2 , I26 , and S39 . By homology modeling , these mutational hotspots appear distal from the enzyme active site ( >12 Å; Figure 1b ) . In fact , these mutations reside in what has been predicted to be an N-terminal accessory domain separate from the radical-SAM catalytic domain ( Kaminska et al . , 2010 ) . Second , ~28% of sequences contained alterations to the promoter . These alterations consist of either Ptet duplication , or insertion of a partial Ptet sequence ( Supplementary file 1 ) . We selected seven evolved Cfr variants , referred herein as CfrV1–V7 , as representative mutational combinations for further characterization ( Figure 1c ) . All selected Cfr variants contain mutations in the cfr open reading frame ( ORF ) while CfrV6 and CfrV7 also harbor Ptet alterations ( Figure 1d ) . Compared to CfrWT , these variants confer ~2-fold to ~16-fold enhanced resistance to PhLOPSA antibiotics and hygromycin A , with no changes in susceptibility to trimethoprim , an antibiotic that does not inhibit the ribosome ( Figure 1e , Figure 1—figure supplement 1 ) . Interestingly , the promoter alterations enable CfrV7 to be expressed and confer resistance to tiamulin in the absence of inducer ( Figure 1—figure supplement 2 ) . The robustness of resistance , and the absence of active-site mutations , suggests Cfr variants do not act as dominant-negative enzymes that inhibit C-2 methylation of A2503 , as observed in a previous directed evolution experiment ( Stojković et al . , 2016 ) . Furthermore , the specificity of resistance to PhLOPSA antibiotics suggests that these Cfr variants elicit their effects through PTC modification rather than triggering a stress response that confers global resistance . To test the hypothesis that Cfr variants mediate higher resistance by increasing the fraction of ribosomes with m8A2503 , we evaluated the methylation status of A2503 by mass spectrometry . Specifically , we expressed Cfr in E . coli and used oligonucleotide protection to isolate a 40-nt fragment of 23S rRNA containing A2503 ( Andersen et al . , 2004; Stojković and Fujimori , 2015 ) . The isolated fragment was then enzymatically digested and analyzed by MALDI-TOF mass spectrometry ( Figure 2a , Figure 2—figure supplement 1 ) . As expected , an empty vector produces a 1013 m/z fragment corresponding to the mono-methylated m2A2503 , modification installed by the endogenous enzyme RlmN . Upon expression of Cfr , we observe a reduction in the 1013 m/z peak and the emergence of a new peak at 1027 m/z , corresponding to m2A2503 conversion into hypermethylated m2m8A2503 . CfrWT is able to convert less than ~40% of m2A2503 into the hypermethylated m2m8A2503 product . In contrast , the evolved variants achieve ~50%–90% methylation of A2503 , indicating that variants are more active than CfrWT in vivo . The ability of evolved Cfr variants to achieve enhanced ribosome methylation in vivo could be attributed to enhanced enzymatic activity and/or higher levels of functional enzyme . To test the hypothesis that Cfr variants achieve higher turnover number , we anaerobically purified and reconstituted CfrWT and a representative evolved variant , CfrV4 . We then evaluated the ability of CfrWT and CfrV4 to methylate a 23S rRNA fragment ( 2447–2625 ) in vitro by monitoring the incorporation of radioactivity from [3H-methyl] S-adenosylmethionine ( SAM ) into RNA substrate under saturating conditions ( Bauerle et al . , 2018 ) . However , no significant difference in kcat between CfrWT ( 3 . 45×10–2±3 . 2×10–3 min–1 ) and CfrV4 ( 2 . 25×10–2±1 . 3×10–3 min–1 ) was observed ( Figure 2—figure supplement 2 ) . Given these findings , we hypothesized that the variants might alter protein levels . To monitor Cfr protein levels , we inserted a flexible linker followed by a C-terminal FLAG tag , which does not alter resistance ( Supplementary file 1 ) . Interestingly , immunoblotting against FLAG revealed that in addition to full-length Cfr , N-terminally truncated Cfr proteins are also produced ( Figure 2b ) . The truncations result from translation initiation at internal methionines but do not contribute to resistance ( Figure 2—figure supplement 3 ) , indicating that they are non-functional enzymes unable to methylate A2503 . The smaller molecular weight truncation is present in higher levels for all Cfr variants compared to CfrfWT ( Figure 2—figure supplement 4 ) . Interestingly the larger molecular weight truncation is present only in CfrV1/V4/V6 and is generated by the I26M mutation introduced during directed evolution . Quantification of resistance-causative , full-length Cfr proteins alone revealed that the evolved variants achieve ~20–100-fold higher steady-state protein levels than CfrWT ( Figure 2b ) . We measured transcript levels for all variants to assess the contribution of altered transcription to increased protein levels . For Cfr variants with promoter alterations , enhanced production of the Cfr transcript is a large contributor to Cfr protein expression , as CfrV6 and CfrV7 exhibit ~6-fold and ~10-fold enhancement in Cfr mRNA levels compared to CfrWT , respectively ( Figure 2c ) . We also observe a ~2–3-fold increase in mRNA levels for CfrV1-5 . Of note , increased Cfr transcript levels likely explain the higher expression of the larger molecular weight truncation observed for all variants ( Figure 2c , Figure 2—figure supplement 4 ) . Despite the observed increase in mRNA levels for CfrV1-5 , this alone cannot explain the multi-fold improvement in protein expression and indicates that these variants also boost protein levels through a post-transcriptional process . This is further supported by the expression profiles for CfrV1-5 , which are dominated by the full-length protein ( Figure 2d ) . Interestingly , enhanced production of Cfr protein correlates with larger fitness defects in E . coli , with an increase in doubling time of ~4 min for CfrV7 compared to empty vector in the absence of antibiotics ( Figure 2e ) . Given that the evolved variants achieve robust enhancement in Cfr expression we sought to elucidate the mechanism ( s ) by which this occurs . To evaluate the importance of promoter alterations , we generated a construct where the Ptet* promoter sequence from CfrV6 was inserted upstream of CfrWT ORF , herein referred to as Ptet*V6 . The insertion of Ptet* alone was sufficient to elicit improvement in Cfr expression ( Figure 3a ) . Furthermore , E . coli expressing Ptet*V6 resembled CfrV6 in its ability to survive in the presence of chloramphenicol ( Figure 3b ) . Taken together , these results suggest the altered promoter drives expression and resistance for CfrV6 . To investigate the contributions of mutations within the Cfr protein , we generated constructs containing Cfr mutations N2K/I , I26M , and S39G in isolation . Interestingly , we observe that mutations at the second position , N2K and N2I , display the largest enhancements in expression , ~27-fold and ~12-fold , respectively ( Figure 3a ) . Similarly to the evolved variants , in addition to the full-length Cfr protein , we also observe expression of the truncation that results from initiation at M95 ( Figure 3—figure supplement 1 ) . The dominance of the second position mutants in driving antibiotic resistance is further manifested by E . coli expressing CfrN2K , but not I26M or S39G , exhibiting survival similar to that of the triple mutant , CfrV3 , in the presence of chloramphenicol ( Figure 3c ) . Similarly , E . coli expressing CfrN2I also exhibits increased resistance to chloramphenicol when compared to the corresponding directed evolution variant , CfrV5 , albeit weaker than CfrN2K ( Figure 3—figure supplement 2a ) . Taken together , these results suggest that the second position mutations drive the robust expression and resistance observed for CfrV1-5 . Of note , ribosome methylation by the produced Cfr does not impact the translation of CfrN2K , as this mutant and its corresponding catalytically inactive double mutant protein CfrN2K_C338A are similarly highly expressed ( Figure 3—figure supplement 2b-c ) . The Cfr coding mutations drive enhanced steady-state protein levels of Cfr protein through a post-transcriptional process . However , because levels at steady-state reflect the net effect of protein synthesis and degradation , we sought to evaluate how Cfr mutations impact both processes , especially since the nature of N-terminal amino acids and codons can greatly influence both translation and degradation in bacteria ( Gottesman , 2003; Bhattacharyya et al . , 2018; Bentele et al . , 2013; Tuller et al . , 2010; Verma et al . , 2019; Goodman et al . , 2013; Boël et al . , 2016; Looman et al . , 1987; Stenström et al . , 2001b; Stenström et al . , 2001b; Stenström and Isaksson , 2002; Sato et al . , 2001 ) . To test the hypothesis that second position mutations enhance translation of mutants , we used polysome profiling to evaluate the relative abundance of Cfr mRNA in polysome fractions . Polysome profiles derived from 10% to 55% sucrose gradients appear similar across biological conditions , suggesting expression of CfrWT and its evolved mutants do not affect global translation ( Figure 4a–b ) . CfrWT transcripts migrate with low polysomes ( fractions 10 and 11 ) ( Figure 4c ) . In contrast , CfrV4 transcripts are strongly shifted toward high polysomes ( fractions 16 and 17 ) , which indicate that CfrV4 mRNA is associated with a large quantity of ribosomes and is better translated than CfrWT ( Figure 4d ) . Further support that CfrV4 is well-translated is the observation that CfrV4 mRNA co-migrates with mRNA of the well-translated housekeeping gene , recA ( Li et al . , 2014; Figure 4—figure supplement 1 ) . At least in part , this is due to the N2K mutation which shifts transcripts to higher polysomes fractions ( fractions 12 and 13 ) ( Figure 4c ) . The recA control mRNA shows excellent reproducibility across biological samples , indicating that the observed shift of mutant Cfr transcripts toward higher polysomes is due to introduced mutations ( Figure 4b ) . Taken together , these results suggest that enhanced translation is a cumulative effect of N2K and other ORF mutations obtained by directed evolution . To further interrogate the role of second position mutations in Cfr translation , we determined the second codon identity for all sequenced variants from the final rounds of evolution ( Supplementary file 1 ) . Interestingly , all N2K mutations were encoded by an AAA codon , while AUU encoded all N2I mutations . In E . coli , the tRNA molecules that decode K ( AAA ) and I ( AUU ) are slightly more abundant than the wild-type N ( AAU ) , accounting for 3 . 0% and 5 . 4% of the tRNA pool compared to 1 . 9% , respectively ( Dong et al . , 1996 ) . To test if tRNA abundance and codon sequence contribute to enhanced translation , we evaluated the impact of synonymous codons on protein expression . Lysine codons AAA and AAG are decoded by the same tRNALys in E . coli . Interestingly , mutating CfrN2K from AAA to AAG , which increases G/C content , did not significantly impact expression ( Figure 4—figure supplement 2 ) . The isoleucine AUA codon is decoded by the low-abundant tRNAIle2 ( Del Tito et al . , 1995; Nakamura et al . , 2000 ) . Mutation of N2I from AUU to the AUA rare codon resulted in a ~2-fold decrease in Cfr expression , supporting tRNA abundance as a contributing factor ( Figure 4—figure supplement 2 ) . To evaluate the impact of mutations introduced during directed evolution on protein half-life , we monitored changes in protein abundance over time after halting expression with rifampicin ( Figure 4e , Figure 4—figure supplement 3 ) . While CfrWT is rapidly degraded with a half-life of ~20 min , CfrN2K/I exhibit increased half-lives of ~60 min . These results suggest that mutation of the second amino acid to lysine or isoleucine contributes to improved steady-state expression both by enhancing translation and stability of Cfr in the cell . CfrS39G also exhibits an increased half-life of ~60 min . The half-life increase is the most pronounced for the I26M single point mutant and similar to that of the triple-mutant , CfrV3 ( >100 min for both proteins ) . Together , these results suggest that evolved variants achieve higher expression through mutations that enhance translation and decrease the degradation of mutant Cfr proteins . Molecular understanding of Cfr-mediated resistance to antibiotics necessitates structural insights into methylated ribosomes . However , obtaining the structure of a Cfr-modified ribosome has been so far hampered by moderate methylation efficiency of S . aureus Cfr , a challenge that can be addressed by the improved methylation ability of directed evolution variants . Of all characterized evolved variants , CfrV7 achieves the highest levels of antibiotic resistance and methylation of rRNA , providing a unique tool for structural determination . Relative peak quantification of the MALDI spectra revealed that CfrV7 achieved near-stoichiometric ( ~90% ) m8A2503 methylation ( Figure 2—figure supplement 1 ) . Ribosomes were purified from E . coli expressing CfrV7 to obtain a 2 . 2 Å cryo-EM structure of the Cfr-modified 50S ribosomal subunit ( Figure 5a , Table 1 , Figure 5—figure supplement 1 ) . The high-resolution cryo-EM density map enabled modeling all known modified nucleotides including the novel C8 methylation of A2503 ( Figure 5b ) . Furthermore , comparison of the Cfr-modified ribosome with the high-resolution cryo-EM structure of unmodified , wild-type E . coli ribosome we published previously ( Stojković et al . , 2020 ) allowed us to identify with high confidence any structural changes due to the presence of m8A2503 . Importantly , modification of A2503 by Cfr does not affect the conformation or position of the A2503 nucleotide . The adenine ring remains in the syn-conformation and places the newly installed C8-methyl group directly into the PTC to sterically obstruct antibiotic binding ( Figure 5c–d ) . Strikingly , beyond the addition of a single methyl group to the substrate nucleotide , the presence of m8A2503 does not result in any additional structural changes to the PTC region of the ribosome ( Figure 5c ) . Furthermore , the increased resistance provided by CfrV7 appears to be mediated specifically by improved methylation of A2503 . No off-target activity of the evolved variant was observed as manual inspection did not reveal density that could correspond to additional C8-methyl adenosines within the high-resolution regions of the 50S ribosomal subunit . This result was cross-validated using our qPTxM tool ( Stojković et al . , 2020 ) , which identified only A2503 and A556 as possible C8-methyl adenosines . Closer examination of A556 reveals it registered as a false positive ( Figure 5—figure supplement 2a-d ) . Contrary to previous reports , we do not observe changes to methylation of C2498 , a distal PTC nucleotide whose endogenous 2′-O-ribose modification has previously been reported to be suppressed by Cfr methylation of A2503 and hypothesized to alter the PTC through long-range effects ( Kehrenberg et al . , 2005; Jensen et al . , 2009; Purta et al . , 2009 ) . Although it is unclear what percentage of C2498 retains the native modification in our structure , we observe clear density for the methyl group and the nucleotide conformation is unaltered . The density for the methyl group is slightly off of the rotameric position , but the dropoff in density along the methyl bond matches the expected shape ( Figure 5—figure supplement 2e-g ) . Taken together , the results do not indicate that conformational changes to C2498 are involved in Cfr-mediated resistance . Although Cfr has been identified in animal-derived E . coli isolates ( Wang et al . , 2012; Deng et al . , 2014; Liu et al . , 2017; Ma et al . , 2021 ) , the resistance gene has primarily been identified clinically in staphylococcal organisms such as S . aureus ( Vester , 2018 ) . However , given the high sequence and structural conservation within the PTC region ( Figure 5—figure supplement 3 ) , structural impacts of the Cfr m8A2503 modification within E . coli and S . aureus ribosomes are likely conserved . Structural superposition of the Cfr-modified ribosome with ribosomes in complex with PhLOPSA antibiotics , hygromycin A , nucleoside analog A201A , and 16-membered macrolides enables direct identification of chemical moieties responsible for steric collision with m8A2503 for these eight antibiotic drug classes ( Figure 5—figure supplement 4 , Figure 5—figure supplement 5 ) . For example , overlay of a bacterial ribosome in complex with the pleuromutilin derivative tiamulin , the selection antibiotic used during directed evolution , reveals steric clashes between the C10 and C11 substituents of the antibiotic with the Cfr-introduced methyl group ( Figure 5d ) . The pleuromutilin class of antibiotics has recently regained interest for their applications as antimicrobial agents in humans but existing molecules remain ineffective against pathogens with Cfr ( Goethe et al . , 2019 ) . Given recent synthetic advances that enable more extensive modification of the pleuromutilin scaffold ( Murphy et al . , 2017; Farney et al . , 2018 ) , the structural insights we obtained will inform the design of next-generation antibiotics that can overcome Cfr-mediated resistance . By relying on directed evolution under antibiotic selection , we identified strategies that increase the ability of a multi-antibiotic resistance determinant Cfr to cause resistance . Enhanced resistance is associated with improved in vivo methylation of rRNA at the C8 position of A2503 . The positive correlation between extent of rRNA modification and resistance aligns with previous studies that investigated linezolid resistance caused by mutation of rRNA , where the severity of linezolid resistance was proportional to the number of 23S rRNA alleles harboring the resistance mutation ( Besier et al . , 2008; Ebihara et al . , 2014; Lobritz et al . , 2003 ) . While alteration of the antibiotic binding site through mutations and enzymatic modification of 23S rRNA are functionally distinct , dependence on the extent of rRNA modification provides parallels between the two mechanisms . Although Cfr-mediated methylation is an enzymatic process , the ability of Cfr to confer resistance is restricted by ribosome assembly . Since the A2503 is only accessible to Cfr prior to incorporation of 23S rRNA into the large ribosomal subunit ( Yan et al . , 2010 ) , the extent of resistance correlates with the ability of the enzyme to methylate 23S rRNA prior to its incorporation into the 50S subunit . The results of our evolution experiment indicate that increasing the intracellular concentrations of Cfr , rather than improving catalysis of an enzyme with a complex radical mechanism ( Yan and Fujimori , 2011; McCusker et al . , 2012; Grove et al . , 2011; Bauerle et al . , 2018 ) is the preferred strategy to increase the proportion of ribosomes with the protective m8A2503 modification . Investigations into expression levels of CfrWT and its respective mutants revealed that , in addition to full-length protein , a smaller Cfr isoform of ~30 kDa is also produced ( Figures 2b and 3a ) . The truncated product is observed when the expression is driven by both the non-native Ptet and native Pcfr promoter ( Figure 6 , Figure 6—figure supplement 1 ) . The smaller product likely results from translation initiation at an internal start codon , as mutation of Met at position 95 abolishes its production . The sequence upstream of M95 is A-rich , which has been demonstrated to promote translation initiation ( Saito et al . , 2020 ) . However , why an internal region of the Cfr ORF would be recognized as an initiation signal is unclear . Truncation of the first 38 residues of Cfr would eliminate a significant portion of the protein , including the N-terminal accessory domain which is likely involved in substrate recognition ( Boal et al . , 2011; Schwalm et al . , 2016 ) . Elimination of this domain provides rationale for why the smaller Cfr isoform does not contribute to resistance , as the protein would likely exhibit perturbed binding to rRNA . Thus , while the truncated product does not contribute to resistance , the potential function of the smaller protein remains elusive and requires further study . The evolved variants improve the expression of resistance-causative , full-length Cfr using two mechanisms . Improved Cfr expression for CfrV6/7 is driven by increased transcription ( Figure 2c ) due to alterations to the Ptet promoter likely introduced by primer slippage during the EP-PCR step of directed evolution . CfrV6 contains a full duplication of Ptet , providing two sites for transcription initiation , likely responsible for enhanced cfr transcript levels . Interestingly , this result parallels a clinical instance of high Cfr resistance discovered in an S . epidermidis isolate where transcription of cfr was driven by two promoters ( LaMarre et al . , 2013 ) and highlights transcriptional regulation as an important mechanism for modulating the in vivo activity of Cfr . Improved expression for evolved variants CfrV1-5 is mediated by mutations that improve both translational efficiency and protein stability in vivo . Of the tested mutations , I26M provides the largest improvement in stability . Of note , the N-terminally truncated Cfr derived from translation initiation at I26M is rapidly degraded , as no detectable protein is observed after 60 min ( Figure 4e ) . However , these results indicate that the costly production and clearance of this nonfunctional protein is offset by the improved cellular stability of the full-length Cfr carrying the I26M mutation . We also observe modest improvements in protein stability with N2K/I mutants ( Figure 4e ) . In bacteria , the identity of N-terminal residues are important determinants of degradation through N-degron pathways ( Dougan et al . , 2012; Tobias et al . , 1991 ) . During protein synthesis , the N-terminus is co-translationally processed by two enzymes , peptide deformylase to remove the formyl group from Met ( fM ) and methionine aminopeptidase ( Koubek , 2021 ) . Based on previous biochemical work , it is unlikely that CfrWT and CfrN2K/I would have different N-terminal processing , since fMN… and fMK/I… are likely to be efficiently de-formylated ( Ragusa et al . , 1999 ) and resistant to methionine excision ( Hirel et al . , 1989; Frottin et al . , 2006; Xiao et al . , 2010 ) . Although the precise mechanism by which N2K/I improves Cfr stability remains elusive , these mutations may alter recognition by other enzymes important for degradation , such as endopeptidases or L/F-tRNA-protein transferase ( Izert et al . , 2021; Ottofuelling et al . , 2021 ) . Of the mutations investigated , N2K is the largest contributor to enhanced Cfr expression and resistance . Although N2K contributes to cellular stability , our results suggest that improved Cfr translation is the dominant role of this mutation ( Figure 4c ) . The effect of N-terminal residues ( and thus codons near the start site ) on early stages of translation has been well documented . Previous work has demonstrated that minimal secondary structure near the start codon ( Kudla et al . , 2009; Goodman et al . , 2013; Bentele et al . , 2013; Boël et al . , 2016; Bhattacharyya et al . , 2018 ) and presence of A/U-rich elements downstream of the translation start site ( Cifuentes-Goches et al . , 2019; Saito et al . , 2020 ) are important factors for efficient translation initiation . RNA secondary structure predictions of the region proximal to the start codon suggest that the N2K mutation ( AAU to AAA ) could disrupt base pairing between the N2 ( AAU ) codon and the downstream T7 ( ACA ) codon ( Figure 4—figure supplement 4 ) . However , the base pair between the second and seventh codon is predicted to be retained when N2K is encoded by the AAG isocodon ( Figure 4—figure supplement 4 ) , which was also able to increase Cfr protein levels ( Figure 4—figure supplement 2 ) and suggests that alternative mechanisms may be responsible for improved translation . While initiation is a major rate-limiting step in protein synthesis , rates of elongation have also been demonstrated to impact translation efficiency , with several proposed models on how the interconnected factors of codon bias , mRNA structure/sequence , and interactions between the ribosome and nascent chain are involved in modulating protein synthesis ( Rodnina , 2016; Choi et al . , 2018; Samatova et al . , 2020 ) . For example , recent work investigating the role of codons 3–5 identified that both mRNA sequence and amino acid composition are key determinants of proper elongation at the N-terminus ( Verma et al . , 2019 ) . Although the mechanism is poorly understood , previous studies have discovered that presence of an AAA lysine codon after the start site is associated with improved translation efficiency in certain contexts ( Looman et al . , 1987; Stenström et al . , 2001a; Stenström et al . , 2001b; Stenström and Isaksson , 2002; Sato et al . , 2001 ) . Our results indicate that the effect of N2K on early steps of translation elongation may be mediated , at least in part , by tRNA abundance , but the exact impact of Lys2 on translation requires further study . Interestingly , the observed internal translation start sites ( I26M and M95 ) that are responsible for producing Cfr truncations ( Figure 2b , Figure 2—figure supplement 3 ) contain a lysine immediately after methionine , further highlighting the putative role for lysine codons in early steps of translation . To date , only a few S . aureus Cfr variants have been reported and no mutations matching those obtained from directed evolution have been found in clinical isolates . However , enhanced expression through positioning of Lys as the second amino acid of Cfr can be recapitulated by accessing an upstream translational start site found in a native sequence context of cfr ( Figure 6 ) . In the specific case of the pSCFS1 resistance plasmid , the sequence upstream of the annotated start codon , which we validated as the start site under the experimental conditions tested ( Figure 6—figure supplement 1 ) , contains regulatory elements that have been proposed to modulate Cfr expression ( Schwarz et al . , 2000; Kehrenberg et al . , 2007 ) . It is plausible that in response to antibiotics , the upstream start codon is used to add three amino acids ( MKE ) to the N-terminus of Cfr and thus placement of a lysine ( K ) at position 2 of the newly expressed protein , analogous to the N2K mutation . Although start codon selection requires further investigation , N-terminal addition of MKE to Cfr expressed under non-native Ptet promoter phenocopies the N2K directed evolution mutation , resulting in increased expression and resistance compared to CfrWT ( Figure 6 ) . Since our assessment of the evolved variants indicates that an increase in Cfr expression is accompanied by a decrease in fitness ( Figure 2e ) , start site selection in response to antibiotic pressure would mitigate detrimental impact on fitness while enabling higher resistance when acutely needed . Interestingly , mutations obtained through directed evolution have been observed in Cfr homologs that share less than 80% sequence identity with Cfr . Methionine ( M ) at position 26 is observed for the functionally characterized Cfr homologs Cfr ( B ) ( Deshpande et al . , 2015; Marín et al . , 2015; Hansen and Vester , 2015 ) and Cfr ( D ) ( Pang et al . , 2020 ) , which have been recovered from human-derived isolates and share 74% and 64% amino acid identity with Cfr , respectively ( Schwarz et al . , 2021; Figure 4—figure supplement 5 ) . We also observe lysine ( K ) at position 2 , methionine ( M ) at position 26 , and glycine ( G ) at position 39 , akin to N2K , I26M , and S39G mutations , for a number of Cfr homologs that clade with functional Cfr or Cfr-like genes ( Stojković et al . , 2019 ) . While the precise roles of these residues within less well-characterized and more distantly related Cfr proteins requires further study , these observations indicate that directed evolution accessed sequence space that is already being exploited by proteins that are , or are hypothesized to be , functional Cfr resistance enzymes . In addition to identifying mechanisms that increase Cfr-mediated resistance , directed evolution of Cfr also provided an indispensable reagent that enabled structural determination of the Cfr-modified ribosome . The high-resolution cryo-EM structure revealed that broad resistance is due to steric effects of the judiciously positioned methyl group within the shared binding site of PTC-targeting antibiotics . Lack of notable changes in position or orientation of A2503 or surrounding PTC nucleotides upon Cfr methylation suggests that the resulting modification does not obstruct the translation capabilities of the ribosome . This absence of PTC disruption is consistent with the observation that the fitness cost of Cfr acquisition is not due to ribosome modification , but rather results from expression of the exogenous protein ( LaMarre et al . , 2011 ) . Importantly , overlay with existing structures containing PTC-targeting antibiotics provides direct visualization of chemical moieties that are sterically impacted by m8A2503 and will inform the design of antibiotic derivatives that can overcome resistance mediated by Cfr . E . coli ER2267 expressing Cfr from a pZA vector ( Wellner et al . , 2013; Stojković et al . , 2016 ) was used in directed evolution experiments . Antibiotic resistance , fitness , in vivo RNA methylation , and protein/transcript expression , polysome analysis , and protein degradation experiments were conducted with E . coli BW25113 expressing Cfr protein from a pZA vector under the Ptet promoter ( or Pcfr promoter where noted ) . E . coli BW25113 acrB::kan , where the efflux pump acrB was replaced with a kanamycin cassette , was used for antibiotic susceptibility testing of the oxazolidinone antibiotic linezolid and hygromycin A . For experiments for which tagless versions of evolved Cfr variants were used , comparisons were made to the wild-type Cfr protein to which the original C-terminal His tag had been removed . E . coli Rosetta2 BL21 ( DE3 ) pLysS was used for overexpression of N-His6-SUMO-tagged Cfrs from a pET28a vector . E . coli MRE600 was used for the preparation of Cfr-modified ribosomes for structural studies . The wild-type cfr gene ( accession: EF450709 . 1 ) with a C-terminal His6-tag , or pooled cfr genes from the previous round of evolution , were randomly mutagenized by error-prone polymerase chain reaction as described previously ( Stojković et al . , 2016 ) . The mutagenized cfr gene pool was then recloned into a pZA vector and transformed into E . coli ER2267 . The frequency of mutations was determined by sequencing randomly selected library variants and was on average 1–3 mutations per gene . E . coli transformants were then subjected to selection by plating cells on LB agar containing tiamulin ( Wako Chemicals USA ) , in addition to 100 µg/ml ampicillin for plasmid maintenance and 20 ng/ml anhydrotetracycline ( AHT , Sigma-Aldrich ) for induction of Cfr expression . For each round of evolution , the E . coli transformants were divided equally and plated on 4–5 plates of LB agar containing different concentrations of tiamulin and grown at 37 oC for up to 48 hr . The tiamulin concentration was increased in 50–100 μg/ml increments . For example , in the first round of evolution , the transformation was plated on the 150 , 200 , 250 , and 300 μg/ml tiamulin plates , in the last round , we selected on 250 , 350 , 450 , and 550 μg/ml tiamulin plates . About 2 ml was plated on tiamulin deficient plates in order to determine transformation efficiency . In general , colonies isolated from tiamulin plates in which the ≤10% of the transformants grew were taken for the next round . After five rounds of mutagenesis and selection , two rounds of enrichment ( selection without mutagenesis ) using high tiamulin concentrations ( 400–1500 µg/ml ) was conducted . After each round of selection or enrichment , 5–10 randomly selected colonies were sequenced from each plate . Antibiotic resistance experiments by broth microdilution followed established protocols ( Wiegand et al . , 2008 ) . In brief , 2 ml of LB media with selection antibiotic was inoculated with a freshly transformed colony containing either empty plasmid , CfrWT , or Cfr mutants . Cultures were grown at 37°C with shaking for approximately 2 . 5 hr . After measuring the OD600 value , cultures were diluted to 106 cells and 50 µl of this dilution was dispensed into 96-well plates containing 50 µl of LB media with antibiotic of interest , ampicillin ( 100 µg/ml ) , and AHT ( 30 ng/ml ) . Antibiotic resistance of evolved Cfr variants was evaluated using a twofold serial dilution of antibiotic with the following concentration ranges: tiamulin ( 50–6400 µg/ml , Wako Chemicals ) ; clindamycin ( 50–6400 µg/ml , TCI America ) , chloramphenicol ( 0 . 5–64 µg/ml , AllStar Scientific ) , linezolid ( 1–256 µg/ml , Acros ) , hygromycin A ( 2–1024 µg/ml , gifted from Dr . Kim Lewis ) , and trimethoprim ( 0 . 125–0 . 2 µg/ml , Sigma-Aldrich ) . Chloramphenicol resistance of single Cfr mutations was evaluated using concentrations of 1 , 2–12 µg/ml ( in 2 µg/ml step increments ) , followed by 16–64 µg/ml ( twofold dilution ) . The minimum inhibitory concentration ( MIC ) required to inhibit visible bacterial growth was determined after incubating plates at 37°C with shaking for 18 hr . Plate OD600 values were also recorded with a microtiter plate reader ( SpectraMax M5 , Molecular Devices ) . Antibiotic resistance determination on LB agar plates was conducted as described previously ( Stojković et al . , 2016; Wiegand et al . , 2008 ) . In brief , 3 µl of 108 , 106 , and 104 dilutions E . coli harboring Cfr were spotted on LB agar plates containing various concentrations of tiamulin . LB agar plates also contained ampicillin ( 100 µg/ml ) and AHT ( 30 ng/ml ) . LB agar plates were incubated at 37°C for 24–48 hr . E . coli expressing empty plasmid or Cfr were grown at 37°C to an OD600 of 0 . 4–0 . 6 with shaking by diluting an overnight culture 1:100 into LB media containing ampicillin ( 100 µg/ml ) and AHT inducer ( 30 ng/ml ) . Total RNA purification , oligo-protection of the 23S rRNA fragment C2480-C2520 , and RNaseT1 digestion were performed as described previously ( Stojković and Fujimori , 2015; Andersen et al . , 2004 ) . Mass spectra were acquired in positive ion , reflectron mode on an AXIMA Performance MALDI TOF/TOF Mass Spectrometer ( Shimadzu ) . Relative peak intensity values were calculated using the Shimadzu Biotech MALDI-MS software . CfrWT and CfrV4 were expressed , purified , and reconstituted using modified published protocols ( Yan et al . , 2010; Stojković and Fujimori , 2015 ) . In brief , N-His6-SUMO-tagged CfrWT/V4 was overexpressed in minimal media conditions with 800 µM IPTG and 1 , 10-phenanthroline to obtain enzyme lacking a [4Fe-4S] iron-sulfur cluster . Minimal media also contained selection antibiotics kanamycin ( 50 µg/ml ) and chloramphenicol ( 34 µg/ml ) . All purification steps were performed in a glovebox ( MBraun , oxygen content below 1 . 8 ppm ) that was cooled to 10°C . Cfr was purified by Talon chromatography ( Clontech ) from clarified lysates . Following chemical reconstitution of the [4Fe-4S] , the N-His6-SUMO-tag was cleaved by incubating the fusion protein with SenP1 protease ( prepared in-house , 1 mg SenP1:100 mg Cfr ) for 1 h at 10 °C in buffer containing 50 mM EPPS ( pH 8 . 5 ) , 300 mM KCl 15 % glycerol , and 5 mM DTT . The cleaved protein was purified away from SenP1 and the N-His6-SUMO-tag by FPLC on a Mono Q 10/100 GL anion exchange column ( GE Healthcare Life Sciences ) using buffers containing 50 mM EPPS ( pH 8 . 5 ) , 50 mM or 1 M KCl ( low-salt or high-salt ) , 15% glycerol , and 5 mM DTT . Protein was eluted using a linear gradient of 100% low-salt to 100% high-salt buffer over eight column volumes . Fractions containing apo-reconstituted , tag-less Cfr were combined , concentrated using a concentrator cell ( Amicon Ultra- 0 . 5 ml , 30 MWCO ) , and stored at –80°C . Protein concentration was determined by Bradford assay ( Bio-Rad ) . The E . coli 23S rRNA fragment 2447–2624 used for in vitro experiments was prepared using modified published protocols ( Stojković and Fujimori , 2015 ) . The desired DNA fragment was amplified from plasmid pKK3535 using previously established primers ( Yan et al . , 2010 ) and used as the template in the in vitro transcription reaction . Following DNase treatment and purification , RNA was precipitated overnight at −20°C by addition of 1/10th volume of 3 M NaOAc , pH 5 . 5 , and 3 volumes of ethanol ( EtOH ) . The RNA was then pelleted and washed with 70% aqueous EtOH , dried , and resuspended in nuclease-free water to obtain a final concentration of ~6 mg/ml . The rRNA fragment was refolded and purified by size exclusion chromatography . To refold the RNA , the sample was heated at 95°C for 2 min and then cooled to 65°C over 5 min . MgCl2 was subsequently added to a final concentration of 10 mM prior to a final cooling step at room temperature for at least 30 min . After removing insoluble debris , RNA was purified by FPLC on a HiLoad 26/60 Superdex 200 column ( GE Healthcare Life Sciences ) using buffer containing 50 mM HEPES ( pH 7 . 5 ) , 10 mM MgCl2 , and 50 mM KCl . Fractions containing the desired rRNA product were combined and precipitated overnight at −20°C by addition of 1/10th volume of 3 M NaOAc , pH 5 . 5 , and 3 volumes of EtOH . The RNA was then pelleted and washed with ice-cold 80% aqueous EtOH , dried , and resuspended in nuclease-free water . After confirming RNA purity on a denaturing 5% TBE , 7 M Urea-PAGE gel , the RNA sample was concentrated to ~450 mM using a SpeedVac Vacuum Concentrator prior to storage at –80°C . Methylation activity of CfrWT and CfrV4 were assessed by monitoring radioactivity incorporation into RNA . Flavodoxin and flavodoxin reductase enzymes were prepared as described previously ( McCusker et al . , 2012 ) . Prior to assembling reaction components , the RNA substrate was refolded as described above . Reactions were conducted in 52 μl volumes in an anaerobic chamber ( MBraun , oxygen levels less than 1 . 8 ppm ) under the following conditions: 100 mM HEPES ( pH 8 . 0 ) , 100 mM KCl , 10 mM MgCl2 , 2 mM DTT , 50 µM Flavodoxin , 25 µM Flavodoxin reductase , 100 µM rRNA substrate , 2 mM [3H-methyl] S-adenosylmethionine ( 175 . 8 dpm/pmol ) , and 5 µM apo-reconstituted Cfr . Reactions were equilibrated at 37°C for 5 min and subsequently initiated by addition of NADPH ( Sigma-Aldrich , final concentration 2 mM ) . The reaction was allowed to proceed at 37°C and timepoints at 0 , 2 , 4 , 6 , and 8 min of 10 µl volume were quenched by the addition of H2SO4 ( 50 mM final concentration ) . For each time point , the RNA volume was brought up to 100 µl with nuclease-free water and was purified away from other reaction components by an RNA Clean & Concentrator kit ( Zymo Research ) by following the manufacturer’s instructions . Purified RNA eluate was added to Ultima Gold scintillation fluid , and the total amount of radioactivity incorporated in the product was detected using a Beckman–Coulter LS6500 scintillation counter . Amount of product generated at each time point was calculated by subtracting background radioactivity ( t=0 min ) and taking into account that two of the three tritium atoms from [3H-methyl] S-adenosylmethionine would be incorporated into the final methylated RNA product ( Yan and Fujimori , 2011; Bauerle et al . , 2018 ) . E . coli expressing empty plasmid , CfrWT , or Cfr mutants were grown at 37°C to an OD600 of ~0 . 4 with shaking by diluting an overnight culture 1:100 into 10 ml LB media containing ampicillin ( 100 µg/ml ) and AHT inducer ( 30 ng/ml ) . Cells were harvested by centrifugation . Cell pellets were lysed for 15 min using B-PER Bacterial Protein Extraction Reagent ( Thermo Fisher Scientific ) containing DNase I ( New England Biolabs ) and 1× cOmplete , EDTA-free protease inhibitor cocktail ( Roche ) . Whole-cell lysate samples containing 4 µg of protein were fractionated using a 4–20% SDS-PAGE gel ( Bio-Rad ) . Proteins were transferred to a 0 . 2-µm nitrocellulose membrane using a Trans-Blot Turbo transfer system ( Bio-Rad ) with a 7 min , mixed MW protocol . Membranes were incubated with TBST-Blotto buffer ( 50 mM Tris-pH 7 . 5 , 150 mM NaCl , 0 . 1% Tween-20 , 5% w/v Bio-Rad Blotting Grade Blocker ) for 1 hr at room temperature , followed by TBST-Blotto containing two primary antibodies: monoclonal mouse anti-FLAG M2 ( 1:2000 dilution , Sigma-Aldrich ) and monoclonal rabbit anti-RNA polymerase beta ( 1:2000 dilution , Abcam ) for 1 hr at room temperature . After washing 3× for 5 min with TBST , membranes were then incubated overnight at 4°C with TBST-Blotto containing two secondary antibodies: goat anti-rabbit IgG cross-absorbed DyLight 680 ( 1:10 , 000 dilution , Thermo Fisher Scientific ) and goat anti-mouse IgG cross-absorbed IRDye 800CW ( 1:10 , 000 dilution , Abcam ) . Membranes were rinsed 3× for 5 min with TBST and allowed to dry completely prior imaging using a Bio-Rad ChemiDoc Molecular Imager . Quantification was performed using Image Lab Software ( Bio-Rad ) within the linear range of detection . The housekeeping protein RNA polymerase beta , which was stably expressed in all experimental conditions , was used as an internal loading control . E . coli expressing empty plasmid , CfrWT , or Cfr variants were grown at 37°C with shaking by diluting a 50 µl of an overnight culture into 10 ml of LB media containing ampicillin ( 100 µg/ml ) and AHT inducer ( 30 ng/ml ) . OD600 values were recorded every 20 min with a microtiter plate reader ( SpectraMax M5 , Molecular Devices ) . qPCR primer sequences for cfr , recA , and luc were designed using NCBI Primer Blast . Template accession numbers , amplicon length , and primer sequences are described in Supplementary file 1 . Primer sequences for rrsA were used as published previously ( Zhou et al . , 2011 ) . For each primer pair primer , qPCR was performed on a tenfold dilution series of desired samples . Amplification efficiency was calculated from the slope of the graph of Cq values plotted against log10 of the at least four template concentrations . Primers for recA: Y=–3 . 238*X+38 . 46 , R2=0 . 9992 , PCR efficiency=103 . 6% . Primers for luc: Y=–3 . 316*X+34 . 52 , R2=0 . 9967 , PCR efficiency=100 . 2% . Primers for cfr: Y=–3 . 254*X+37 . 52 , R2=0 . 9960 , PCR efficiency=102 . 9% . Primers for rrsA: Y=–3 . 629*X+32 . 24 , R2=0 . 9965 , PCR efficiency=90 . 0% . Lysate preparation and sucrose gradient fractionation were adapted from previously published protocols with modification ( Mohammad and Buskirk , 2019; Li et al . , 2014 ) . Cfr-modified , 70S ribosomal subunit was purified from E . coli MRE600 expressing CfrV7 variant using previously published protocol with modification ( Mehta et al . , 2012; Stojković et al . , 2020 ) . In short , E . coli transformed with pZA-encoded CfrV7 were grown to an OD600 of 0 . 5 in LB media containing ampicillin ( 100 µg/ml ) and AHT inducer ( 30 ng/ml ) at 37°C with shaking . Cells were harvested by centrifugation , washed , and lysed by using a microfluidizer . The lysate was clarified by ultracentrifugation at 30 , 000×g 30 min at 4°C using a Ti45 rotor ( Beckman Coulter ) two times . The recovered supernatant was applied to a 32% w/v sucrose cushion in buffer containing 20 mM Hepes-KOH ( pH 7 . 5 ) , 500 mM NH4Cl , 20 mM Mg ( OAc ) 2 , 0 . 5 mM EDTA , 6 mM β-mercaptoethanol , 10 U/ml SuperASE-In and was ultracentrifuged at 100 , 000×g for for 16 hr at 4°C in a SW Ti41 rotor ( Beckman Coulter ) . After removing the supernatant , the pellet was resuspended slowly at 4°C over 1 hr in Buffer A containing 20 mM Hepes-KOH ( pH 7 . 5 ) , 200 mM NH4Cl , 20 mM Mg ( OAc ) 2 , 0 . 1 mM EDTA , 6 mM β-mercaptoethanol , and 10 U/ml SuperASE-In . Particulates that were not resuspended were removed by centrifugation at 10 , 000 rpm for 10 min at 4°C . Sample concentration was determined by NanoDrop UV spectrophotometer ( Thermo Fisher Scientific ) , where A260=1 corresponds to 24 pmol of 70S ribosome . Tight-coupled 70S ribosomes were purified as described previously ( Khusainov et al . , 2017 ) . In brief , 70S ribosomes were purified on a 15–30% w/v sucrose gradient in Buffer A . Sucrose gradients were generated using a Bio-Comp Gradient Master . 300–400 pmol of 70S ribosomes were loaded on each sucrose gradient . Ultracentrifugation was performed using a SW Ti41 rotor ( Beckman Coulter ) for 75 , 416×g for 16 hr at 4°C . Gradients were fractionated using a Bio-Comp Fractionator in 20 fractions at a speed of 0 . 25 mm/s where absorbance at 260 nm was continuously monitored . Fractions corresponding to 70S ribosomes were combined and precipitated by slow addition at 4°C of PEG 20 , 000 in Buffer A to a final concentration of 9% w/v . Ribosomes were isolated by centrifugation for 10 min at 17 , 500×g . After removing the supernatant , ribosomes were slowly resuspended overnight at 4°C in buffer containing 50 mM Hepes-KOH ( pH 7 . 5 ) , 150 mM KOAc , 20 mM Mg ( OAc ) 2 , 7 mM β-mercaptoethanol , 20 U/ml SuperASE-In . Purified 70S ribosomal subunits were diluted from 2 to 0 . 5 mg/ml in Buffer A , applied to 300-mesh carbon coated ( 2 nm thickness ) holey carbon Quantifoil 2/2 grids ( Quantifoil Micro Tools ) and flash-frozen as described in Khatter et al . , 2015 . Data were collected using serialEM on the in-house Titan Krios X-FEG instrument ( Thermo Fisher Scientific ) operating at an acceleration voltage of 300 kV and a nominal underfocus of Δz=0 . 2–1 . 5 μm at a nominal magnification of 29 , 000 ( calibrated physical pixel size of 0 . 822 Å ) . We recorded 2055 movies using a K2 direct electron detector camera in super-resolution mode with dose fractionation ( 80 individual frames were collected , starting from the first one ) . Total exposure time was 8 s , with a total dose of 80 e- ( or 1 e-/Å2/frame ) . Images in the stack were aligned using the whole-image motion correction and patch motion correction ( 5×5 patches ) methods in MotionCor2 ( Zheng et al . , 2017 ) . Before image processing , all micrographs were checked for quality and 1531 best were selected for the next step of image processing . The contrast transfer function of each image was determined using GCTF ( Zhang , 2016 ) as a standalone program . For particle selection , we have used Relion 3 . 0 autopicking procedure ( Scheres , 2012 ) . For the first steps of image processing , we used data binned by a factor of 8 ( C8 images ) . During the first round of 2D classification , we removed only images with ice or other contaminants . Subsequently , the initial structure was generated using the ab initio procedure in CryoSPARC v2 . 0 . Following this step , we performed Relion 3D classification with bin by four data ( C4 ) in order to exclude bad particles . The resulting 141 , 549 particle images of ribosomes were used for subsequent classification and refinement procedures . For the initial refinement , we used a spherical mask , which was followed by further refinement using a mask around the stable part of 50S ( excluding L1 stalk , L7/L12 region ) . A further improved cryo-EM map was obtained by using CTF-refinement procedure from Relion 3 . 0 . The post-processing procedure implemented in Relion 3 . 0 ( Scheres , 2012 ) was applied to the final maps with appropriate masking , B-factor sharpening ( automatic B-factor estimation was –55 . 86 ) and resolution estimation to avoid over-fitting ( final resolution after post-processing with 50S mask applied was 2 . 7 Å ) . Subsequently , the stack of CTF-refined particles was processed in a new version of CryoSPARC v2 . 0 ( Punjani et al . , 2017 ) . After homogeneous refinement , the same stack of particles was additionally refined in cisTEM ( Grant et al . , 2018 ) . After Auto-Refine ( with automasking within cisTEM ) , we performed local refinement using 50S mask ( the same one used for refinement in Relion ) and also applied per particle CTF refinement as implemented in cisTEM software . After such refinement , the resolution was improved to 2 . 2 Å ( Figure 5—figure supplement 1 ) . This map after Sharpen3D ( Grant et al . , 2018 ) was used for model building and map interpretation . The final model of the 50S subunit was generated by multiple rounds of model building in Coot ( Emsley et al . , 2010 ) and subsequent refinement in PHENIX ( Adams et al . , 2010 ) . The restraints for the novel m2m8A nucleotide for the atomic model fitting and refinements were generated using eLBOW ( Moriarty et al . , 2009 ) . The atomic model of the 50S subunit from the E . coli ribosome structure ( PDB 6PJ6 ) ( Stojković et al . , 2020 ) was used as a starting point and refined against the experimental cryo-EM map by iterative manual model building and restrained parameter-refinement protocol ( real-space refinement , positional refinement , and simulated annealing ) . Final atomic model comprised of ∼92 , 736 atoms ( excluding hydrogens ) across the 3015 nucleotides and 3222 amino acids of 28 ribosomal proteins . Proteins L7 , L10 , L11 , and L31 were not modeled in . In addition , 179 Mg2+ , 2716 water molecules , 1 Zn2+ , and 1 Na+ were included in the final model . Prior to running MolProbity ( Chen et al . , 2010 ) analysis , nucleotides 878–898 , 1052–1110 , 2101–2189 of 23S rRNA , and ribosomal protein L9 were removed , due to their high degree of disorder . Overall , protein residues and nucleotides show well-refined geometrical parameters ( Table 1 ) . Figures were prepared using Pymol Molecular Graphics System , Version 2 . 4 . 1 unless otherwise noted . The final model and map were run through qPTxM ( Stojković et al . , 2020 ) with default parameters except for d_min=2 and cc_threshold=0 . 5 to search for evidence of posttranscriptional modifications . Of a total of 39 sites with density suggesting possible modifications , 2 were C8-methyl adenosines , A556 and A2503 . None of the identified sites were 2′O-methyl cytosines . To calculate expected density dropoff curves for methylated and unmethylated nucleotides , the phenix . fmodel ( Adams et al . , 2010 ) tool was used to generate noise-free maps from models of a single nucleotide in each state , and scripts modified from qPTxM were used to collect measurements of the density at 0 . 1 Å intervals along the vector of the proposed methylation . Means and standard deviations were calculated for densities at the four positions tested by qPTxM on each nucleotide , from which Z-scores were then calculated for selected nucleotides . To measure densities for both the best tested rotamer of m ( 2′O ) C 2498 and the modeled rotamer , densities along the 2′O-methyl bond were compared between the files generated by qPTxM run two times as described above , once with prune=True ( removing the modeled methyl group and placing the rotameric methyl with the strongest density ) and once with prune=False ( leaving the modeled methyl group intact ) .

Antibiotics treat or prevent infections by killing bacteria or slowing down their growth . A large proportion of these drugs do this by disrupting an essential piece of cellular machinery called the ribosome which the bacteria need to make proteins . However , over the course of the treatment , some bacteria may gain genetic alterations that allow them to resist the effects of the antibiotic . Antibiotic resistance is a major threat to global health , and understanding how it emerges and spreads is an important area of research . Recent studies have discovered populations of resistant bacteria carrying a gene for a protein named chloramphenicol-florfenicol resistance , or Cfr for short . Cfr inserts a small modification in to the ribosome that prevents antibiotics from inhibiting the production of proteins , making them ineffective against the infection . To date , Cfr has been found to cause resistance to eight different classes of antibiotics . Identifying which mutations enhance its activity and protect bacteria is vital for designing strategies that fight antibiotic resistance . To investigate how the gene for Cfr could mutate and make bacteria more resistant , Tsai et al . performed a laboratory technique called directed evolution , a cyclic process which mimics natural selection . Genetic changes were randomly introduced in the gene for the Cfr protein and bacteria carrying these mutations were treated with tiamulin , an antibiotic rendered ineffective by the modification Cfr introduces into the ribosome . Bacteria that survived were then selected and had more mutations inserted . By repeating this process several times , Tsai et al . identified ‘super’ variants of the Cfr protein that lead to greater resistance . The experiments showed that these variants boosted resistance by increasing the proportion of ribosomes that contained the protective modification . This process was facilitated by mutations that enabled higher levels of Cfr protein to accumulate in the cell . In addition , the current study allowed , for the first time , direct visualization of how the Cfr modification disrupts the effect antibiotics have on the ribosome . These findings will make it easier for clinics to look out for bacteria that carry these ‘super’ resistant mutations . They could also help researchers design a new generation of antibiotics that can overcome resistance caused by the Cfr protein .

2022

Directed evolution of the rRNA methylating enzyme Cfr reveals molecular basis of antibiotic resistance

To persist in microbial communities , the bacterial pathogen Legionella pneumophila must withstand competition from neighboring bacteria . Here , we find that L . pneumophila can antagonize the growth of other Legionella species using a secreted inhibitor: HGA ( homogentisic acid ) . Unexpectedly , L . pneumophila can itself be inhibited by HGA secreted from neighboring , isogenic strains . Our genetic approaches further identify lpg1681 as a gene that modulates L . pneumophila susceptibility to HGA . We find that L . pneumophila sensitivity to HGA is density-dependent and cell intrinsic . Resistance is not mediated by the stringent response nor the previously described Legionella quorum-sensing pathway . Instead , L . pneumophila cells secrete HGA only when they are conditionally HGA-resistant , which allows these bacteria to produce a potentially self-toxic molecule while restricting the opportunity for self-harm . We propose that established Legionella communities may deploy molecules such as HGA as an unusual public good that can protect against invasion by low-density competitors . Inter-bacterial conflict is ubiquitous in nature , particularly in the dense and highly competitive microenvironments of biofilms ( Davey and O'toole , 2000; Foster and Bell , 2012; Rendueles and Ghigo , 2015 ) . In these settings , bacteria must battle for space and nutrients while evading antagonism by neighboring cells . One strategy for managing these environments is for bacteria to cooperate with their kin cells , sharing secreted molecules as public goods ( Nadell et al . , 2016; Abisado et al . , 2018 ) . However , these public goods are vulnerable to exploitation by other species or by ‘cheater’ bacterial strains that benefit from public goods but do not contribute to their production . For this reason , many bacteria participate in both cooperative and antagonistic behaviors to survive in multispecies biofilms . Bacterial antagonistic factors can range from small molecules to large proteins , delivered directly or by diffusion , and can either act on a broad spectrum of bacterial taxa or narrowly target only a few species . Although narrowly targeted mechanisms may seem to be of less utility than those that enable antagonism against diverse bacterial competitors , targeted strategies can be critical for bacterial success because they tend to mediate competition between closely-related organisms that are most likely to overlap in their requirements for restricted nutrients and niches ( Hibbing et al . , 2010 ) . The bacterium Legionella pneumophila ( Lp ) naturally inhabits nutrient-poor aquatic environments where it undergoes a bi-phasic lifestyle , alternating between replication in host eukaryotes and residence in multi-species biofilms ( Lau and Ashbolt , 2009; Declerck et al . , 2007; Declerck , 2010; Taylor et al . , 2013 ) . If Lp undergoes this lifecycle within man-made structures such as cooling towers , the bacteria can become aerosolized and cause outbreaks of a severe , pneumonia-like disease in humans , called Legionnaires’ disease ( Fraser et al . , 1977; McDade et al . , 1977; Fields et al . , 2002 ) . Because of the serious consequences of Lp colonization , the persistence and growth of Legionella in aquatic environments has been the subject of numerous studies . These studies have examined replication within protozoan hosts ( Rowbotham , 1980; Lau and Ashbolt , 2009; Hoffmann et al . , 2014 ) , survival in water under nutrient stress ( Li et al . , 2015; Mendis et al . , 2015 ) , and sensitivity to biocides ( Kim et al . , 2002; Lin et al . , 2011 ) . Here , we focus on interbacterial competition as an underappreciated survival challenge for Lp . Legionella spp . are not known to produce any antibiotics , bacteriocins , or other antibacterial toxins . Bioinformatic surveys of Legionella genomes have revealed a number of polyketide synthases and other loci that likely produce bioactive metabolites ( Johnston et al . , 2016; Tobias et al . , 2016 ) , but these have not been shown to exhibit any antimicrobial functions . Nevertheless , there are some hints that interbacterial competition is relevant for Lp success within biofilms . For example , one study of artificial two-species biofilms found that viable Lp were able to persist for over two weeks in the presence of some bacterial species ( e . g . Pseudomonas fluorescens , Klebsiella pneumoniae ) but not others ( e . g . Pseudomonas aeruginosa ) ( Stewart et al . , 2012 ) . Additionally , Lp bacteria are often co-resident with other Legionella spp . in man-made structures , with some studies showing that Lp proliferation is correlated with a decrease in other Legionella spp . populations ( Wéry et al . , 2008; Pereira et al . , 2017; Declerck et al . , 2007 ) . These studies suggest that Lp bacteria may compete with other Legionella spp . for similar biological niches . The most direct evidence for interbacterial competition comes from Stewart et al . ( 2011 ) , who found that Lp could antagonize the growth of neighboring Legionella spp . on the same plate . The molecules mediating this competition have not been identified , although previous work suggested a role for Lp’s secreted surfactant , a thin liquid film that facilitates the spread of Lp across agar plates ( Stewart et al . , 2009; Stewart et al . , 2011 ) . Still , it remained unknown if surfactant played a direct or indirect role in inter-Legionella inhibition . Here , we use unbiased genetic approaches to find that homogentisic acid ( HGA ) produced by Lp generates oxidative intermediates that inhibit the growth of neighboring Legionella spp . We find that HGA production co-occurs with surfactant production , but that these are independent , separable phenomena . The redox state of HGA appears to be critical for its activity , as HGA is only toxic in aerobic conditions and fully oxidized HGA-melanin pigment is inactive . Unexpectedly , we find that although Lp secretes abundant HGA , it is also susceptible to HGA-mediated inhibition . We identify one gene– lpg1681– that enhances Lp susceptibility to HGA . Moreover , we find that Lp cells are resistant to HGA at high-density , which is also when they secrete large amounts of HGA . This high-density resistance is cell intrinsic and is independent of growth phase , the stringent response , or the previously described quorum-sensing pathway in Legionella . Based on these findings , we propose that HGA has the potential to play an important role in structuring Legionella communities . Inspired by previous reports ( Stewart et al . , 2011 ) , we investigated how Legionella pneumophila ( Lp ) engages in inter-Legionella competition . We found that Lp inhibited the growth of neighboring Legionella micdadei ( Lm ) plated 1 cm away on solid media , suggesting that it produces a secreted inhibitor ( Figure 1A ) . This inhibition was most robust when we plated the Lp strain on low-cysteine media 3–4 days prior to plating Lm , allowing time for the inhibitory molecule to be produced and spread across the plate . To quantify this inhibition , we recovered Lm grown at different distances from Lp . After 48 hr incubation , we found a 10 , 000-fold difference in growth between Lm antagonized by Lp versus Lm plated outside of the zone of inhibition ( Figure 1B ) . Previous studies ( Stewart et al . , 2011 ) had proposed that inter-Legionella inhibition could be caused by Lp’s secreted surfactant , which is produced by Lp but not Lm ( Stewart et al . , 2009 ) . We tested this hypothesis by deleting a surfactant biosynthesis gene , bbcB , from the Lp genome ( Stewart et al . , 2011 ) . The resulting ∆bbcB strain did not produce surfactant ( Figure 1—figure supplement 1B ) , yet it still inhibited adjacent Lm ( Figure 1A , Figure 1—figure supplement 1C ) . When quantified , we observed nearly identical inhibition from both wild type Lp and ∆bbcB Lp ( Figure 1B and C ) indicating that the surfactant did not enhance inhibition . Furthermore , we observed that the zone of inhibition surrounding wildtype Lp did not always co-occur with the spread of the surfactant front ( Figure 1—figure supplement 1A ) . From these results , we conclude that L . pneumophila can cause strong growth inhibition of neighboring Legionella using an unknown molecule that is distinct from surfactant . To determine which molecule ( s ) might be responsible for inter-Legionella inhibition , we performed an unbiased genetic screen in Lp . We generated Lp mutants using a drug-marked Mariner transposon that randomly and efficiently integrates into the Legionella genome ( O'Connor et al . , 2011 ) . To identify mutants that were defective in producing the inhibitor , we transferred each mutant onto a lawn of L . micdadei on low-cysteine plates and examined the resulting zone of inhibition surrounding each Lp mutant ( Figure 2A ) . After screening 2870 clones , we isolated 19 mutants that produced a smaller zone of inhibition than wild type Lp , as well as five mutants that showed a complete loss of inhibition ( Figure 2B , Supplementary file 1 ) . We refer to these as ‘small zone’ and ‘no zone’ mutants , respectively . Among the ‘small zone’ mutants , some had defects in surfactant spreading on plates , while others showed enhanced surfactant spread ( Figure 2—figure supplement 1A ) , further distinguishing inter-bacterial inhibition from surfactant secretion . We focused on the ‘no zone’ mutants , as these had the strongest defects in inhibition . These five mutants carried transposon insertions in two separate operons ( Figure 2C ) . The first operon had two insertions in the hisC2 gene ( lpg1998 ) , which breaks down tyrosine as part of the HGA-melanin metabolic pathway ( Figure 2D ) . Its downstream gene , pphA , converts phenylalanine to tyrosine in the same pathway . To validate the role of hisC2 in inhibition , we overexpressed this gene in the hisC2 transposon mutant background and found that hisC2 alone was sufficient to complement the mutant phenotype ( Figure 2—figure supplement 1B ) . Having confirmed the role of hisC2 , we turned to the second operon , where we had recovered transposon insertions in two uncharacterized genes , lpg2276 and lpg2277 ( Figure 2C ) . These two genes lie immediately upstream of hpd ( lpg2278 ) , which is known to act with hisC2 in the HGA-melanin pathway ( Steinert et al . , 2001; Gu et al . , 1998 ) ( Figure 2D ) . Because transposon insertions at the beginning of an operon can disrupt the expression of downstream genes via polar effects , we hypothesized that the insertions we recovered in lpg2276 and lpg2277 altered inter-Legionella inhibition via disruption of hpd expression . Indeed , we were able to complement insertions in both genes , which had yielded ‘no zone’ mutants , by overexpressing hpd , despite the fact that hpd overexpression caused a growth defect ( Figure 2—figure supplement 1B ) . In conclusion , all five ‘no zone’ isolates had mutations that disrupted the same metabolic pathway involved in the production of HGA-melanin . Consistent with these findings , we observed defects in HGA-melanin pigmentation in all of the ‘no zone’ mutants as well as some of the ‘small zone’ mutants ( Figure 2—figure supplement 1E ) . The HGA-melanin pathway is found in diverse eukaryotes and bacteria ( Nosanchuk and Casadevall , 2003; Liu and Nizet , 2009 ) including Legionella spp . ( Fang et al . , 1989 ) ( Figure 2—figure supplement 1D ) . This pathway produces homogentisic acid ( HGA ) from the catabolism of phenylalanine or tyrosine ( Steinert et al . , 2001 ) ( Figure 2D ) . HGA can either be further metabolized and recycled within the cell via HmgA-C , or it can be secreted outside of the cell , where it auto-oxidizes and polymerizes to form a black-brown pigment called HGA-melanin , or pyomelanin ( Kotob et al . , 1995 ) ( Figure 2D ) . To test whether intracellular metabolites downstream of HGA are necessary for inhibition , we deleted hmgA , the first gene in the pathway that recycles HGA back into central metabolism . We found that the ∆hmgA strain produced a zone of inhibition that was similar or slightly larger than wild type ( Figure 2—figure supplement 1C ) . We therefore inferred that synthesis of secreted HGA and/or HGA-melanin , but not its recycling and intracellular processing , is required for Lp inhibition of Lm . To our knowledge , the HGA-melanin pathway has not previously been implicated in inter-bacterial competition . To the contrary , prior work has emphasized the beneficial ( rather than detrimental ) effects of HGA-melanin on Legionella growth , by providing improved iron scavenging ( Chatfield and Cianciotto , 2007 ) and protection from light ( Steinert et al . , 1995 ) . We therefore asked whether the active inhibitor produced by the pathway was HGA-melanin , or alternatively a precursor molecule ( Figure 3A ) . We tested the potential inhibitory activity of HGA-melanin pigment from Lp conditioned media in multiple experiments; however , we never observed any inhibition of Lm . We wished to rule out the possibility that the pigment secreted into rich media was too dilute to be active , or that other nutrients in the media might interfere with inhibition . We therefore isolated a crude extract of HGA-melanin from Lp conditioned media via acid precipitation ( as in Chatfield and Cianciotto , 2007 ) , washed and concentrated the pigment approximately 10-fold and repeated the assay; the concentrated pigment also showed no inhibitory activity ( Figure 3B and E ) . The first metabolite secreted by the HGA-melanin pathway is HGA . We tested whether HGA could behave as an inhibitor even though HGA-melanin could not . Indeed , we found that synthetic HGA robustly inhibited Lm growth , both when spotted onto a lawn of Lm and when titrated into AYE rich media ( Figure 3B and C ) , We found that inhibition of Lm by HGA is relatively specific at the molecular level; neither 2-hydroxyphenylacetic acid nor 3-hydroxyphenylacetic acid , two HGA-related molecules that differ from HGA by only a single -OH group , were able to inhibit Lm growth at any concentration tested ( Figure 3—figure supplement 1A ) . Because HGA , but not HGA-melanin , can inhibit Lm growth ( Figure 3B and E ) , we inferred that the oxidative state of HGA might be important to its inhibitory activity . HGA is a reactive molecule , which auto-oxidizes ( Eslami et al . , 2013 ) and polymerizes to form HGA-melanin through a series of non-enzymatic steps that are not genetically encoded ( Steinert et al . , 2001 ) and are therefore undetectable by our genetic screen . Given its auto-oxidative potential , we next tested whether HGA might cause growth inhibition by oxidizing other nearby molecules , either in the media or on bacterial cells . We allowed synthetic HGA to oxidize completely for 24 hr in AYE media before adding Lm ( Figure 4—figure supplement 1D ) . We found that pre-oxidation completely abolished synthetic HGA activity , even at very high HGA concentrations ( Figure 3D , compare to 3C ) . This experiment also ruled out the possibility that HGA acts by causing nutrient depletion or other modifications of the media , since media pre-incubated with HGA for 24 hr is still able to support normal Lm growth . Instead , we infer that Lm inhibition results from direct interactions between bacterial cells and either HGA itself or unstable , reactive intermediates produced during HGA oxidation . Small , reactive , quinone-like molecules similar to HGA are known to react with oxygen to produce H2O2 , which is broadly toxic to bacteria ( Hassan and Fridovich , 1980 ) . In such cases , extracellular catalase has been shown to protect bacteria against the toxic effects of H2O2 ( Hassan and Fridovich , 1980; Imlay , 2013 ) . To test if HGA toxicity occurs via a similar mechanism as H2O2 , we asked if extracellular catalase was sufficient to protect Lm from HGA-mediated toxicity . Even at very high catalase concentrations , we found that catalase provided no protection from HGA ( Figure 3—figure supplement 1B ) , ruling out the production of extracellular H2O2 as a potential mechanism of action for HGA-mediated inhibition . We also considered the possibility that HGA as a weak acid could inhibit Lm indirectly by altering the local pH , but we observed that adding HGA at 1 mM into AYE media or PBS caused little to no change in pH . Given that the redox state of HGA is critical for inhibition , we reasoned that it should be possible to modulate HGA activity by altering the redox state of the media using reducing agents . We accomplished this by titrating L-cysteine from 25% to 200% of the levels in standard AYE media . In the absence of HGA , these altered cysteine concentrations had little impact on Lm growth ( Figure 3F , gray symbols ) . However , lower cysteine concentrations greatly sensitized Lm to HGA , while excess cysteine was completely protective ( Figure 3F ) . These findings may help partially explain why HGA’s inhibitory activity on Legionella has not been previously detected , as Legionella species are typically studied in cysteine-rich media . We found that synthetic HGA is readily able to react with cysteine in vitro ( Figure 3—figure supplement 1D ) , presumably impacting the oxidation state of HGA . Moreover , incubation of HGA with two other reducing agents , DTT ( dithiothreitol ) or reduced glutathione , similarly quenched HGA’s inhibitory activity ( Figure 3—figure supplement 1E ) . From these experiments , we conclude that HGA is less potent in rich media because it reacts with excess cysteine ( or other bystander molecules ) before it can interact with Lm . In sum , these results implicate the reactive activity of HGA and/or its transient , oxidative intermediates in inter-Legionella inhibition . In these experiments , synthetic HGA was a robust inhibitor of Lm . However , this inhibition required relatively high concentrations of HGA ( >50 µM ) . We next quantified the amount of HGA secreted by Lp to determine if these levels were biologically relevant . Compared to other Legionella spp . , Lp produces much more pigment ( Figure 2—figure supplement 1D ) , suggesting that it secretes considerably more HGA . To estimate the quantity of secreted HGA , we created a standard curve of synthetic HGA in AYE rich media . We allowed the HGA added to completely oxidize to HGA-melanin , which can be measured by OD 400 readings . In this way , we can use the pigment levels after oxidation as a reliable measure of total HGA that was produced by a given time-point ( Figure 4—figure supplement 1D ) . Using this calibration , we estimated that wild type Lp had secreted the equivalent of 1 . 7 mM HGA after 48 hr of culture , whereas the hyperpigmented ∆hmgA strain secreted about 2 . 6 mM HGA . Thus , the levels of HGA produced by Lp are considerably higher than the inhibitory concentrations of synthetic HGA used in our assays ( 50–500 µM ) , at least under lab conditions . In contrast , we did not detect any pigment from the non-inhibitory hisC2::Tn strain ( Figure 4—figure supplement 1E ) . From these experiments , we conclude that HGA is an abundant , secreted , redox-active metabolite of Lp , which can accumulate in concentrations that are relevant for inter-Legionella inhibition . Our results so far indicated that HGA can be a potent , redox-active inhibitor of Lm , which is volatile and capable of reacting with many types of thiol-containing molecules . If Lp uses HGA to compete with neighboring Legionella spp . , we anticipated that Lp would have evolved some form of resistance to its own secreted inhibitor . Therefore , we next tested Lp susceptibility to inhibition in low-cysteine conditions , as we had previously done for Lm . Surprisingly , we found that Lp was quite sensitive to inhibition by neighboring Lp that was already growing on the plate ( Figure 4A ) . Indeed , Lp susceptibility closely mirrored the susceptibility of Lm to inhibition ( compare to Figure 1A ) , even though the bacterial cells secreting the inhibitor were genetically identical to the inhibited Lp . In both cases , we observed a sharp boundary at the edge of the zone of inhibition . In contrast , the ‘no zone’ Lp strain hisC2::Tn did not generate a sharp zone of inhibition against neighboring Lp ( Figure 4A ) , suggesting that the HGA-melanin pathway was responsible for both Lm and Lp inhibition . Furthermore , we found that synthetic HGA was able to inhibit Lp in liquid cultures at the same concentrations that were inhibitory to Lm ( compare Figures 4B and 3C ) . However , our comparisons between Lm and Lp revealed one important difference in their response to HGA inhibition . Unlike Lm , the Lp cultures exposed to HGA exhibited a population rebound after a dose-dependent growth delay , measured by both OD600 and plating for viable CFUs ( Figure 4B , Figure 4—figure supplement 1C ) . This rebound response was shared between the KS79 lab strain and the Philadelphia-1 original Lp strain ( Figure 4—figure supplement 1A ) . We hypothesized that the Lp rebound response following HGA inhibition occurred because of selection and outgrowth of HGA-resistant mutants following exposure . To test this possibility , we collected ‘post-rebound’ stationary phase Lp previously exposed to 250 µM HGA and compared their subsequent HGA sensitivity to unexposed Lp ( Figure 4C–D ) . We found that both cultures showed nearly identical susceptibility to HGA inhibition , suggesting that Lp population rebounds were not driven by genetic adaptation . We therefore considered an alternate possibility that Lp populations exposed to HGA remain static until HGA levels fall below inhibitory concentrations , reflecting the auto-oxidation and loss of HGA activity over time . This possibility was supported by CFU measurements , which showed that HGA is bacteriostatic but not bacteriocidal against Lp during the growth delay ( Figure 4—figure supplement 1C ) . Furthermore , we observed a strong , linear correlation between the time required to fully oxidize a given concentration of HGA and the length of the growth delay induced by Lp ( Figure 4E ) . Based on our multiple observations , we favor the parsimonious conclusion that synthetic HGA causes initial , bacteriostatic inhibition of Lp until it has been sufficiently oxidized and thereby inactivated , enabling Lp growth . We note that the liquid culture assays ( Figure 4B ) differ from the co-plating assays , in which we did not observe Lp rebound ( Figure 4A ) . In the latter case , we presume that bacteria continually secrete fresh HGA to replace the oxidized HGA over time . Thus , our results confirm a surprising role for HGA in both interspecies and intraspecies Legionella inhibition . We next investigated the molecular basis of L . pneumophila susceptibility and resistance to HGA using bacterial genetics . First , we tested the role of the HmgA-C proteins , which break down intracellular HGA and recycle it back into central metabolism . We hypothesized that HmgA-C proteins might also be able to deactivate extracellular HGA ( Rodríguez-Rojas et al . , 2009 ) ( Figure 2D ) . Contrary to this hypothesis , we found that the growth response of the ∆hmgA mutant to increasing concentrations of synthetic HGA was nearly identical to that of wild type Lp ( Figure 4—figure supplement 1B ) . These results suggest that the intracellular recycling pathway does not play an appreciable role in Lp resistance to extracellular HGA . Having excluded the obvious candidate pathway for Lp resistance to HGA , we pursued an unbiased forward genetics approach . Because HGA is strongly inhibitory to low density bacteria , we performed a selection for spontaneous , HGA-resistant mutants of Lp and Lm using a high HGA concentration that normally prevents almost all growth for both species . To prevent HGA from reacting with media components and becoming inactive ( as in Figure 3D and F ) , we mixed the bacteria with 1 mM HGA in agar overlays poured onto low-cysteine BCYE plates ( Figure 5A ) . After six days , an average of 53 colonies had grown up on each Lp plate under HGA selection , whereas only 3–5 colonies grew on Lm plates exposed to HGA . Based on these results , we focused on HGA-selected mutants of Lp . We retested the phenotypes of spontaneous mutants on HGA +low cysteine plates and recovered 29 Lp strains that consistently grew better than wild type Lp ( Figure 5B ) . Notably , all recovered mutants had a decrease in HGA sensitivity relative to wild type , but none were completely resistant . The spontaneous mutants also showed improved growth on low-cysteine plates , relative to wild type . To determine the underlying genetic basis of these phenotypes , we sequenced the genomes of all 29 strains plus the starting , wildtype strain of Lp to a median depth of 118x and identified mutations genome-wide . Each mutant strain carried 1 to 3 unique point mutations relative to the starting strain , and most of these mutations were found only in a few shared loci ( Table 1 ) . The most abundant category of mutants was genes related to translation; 19 of 29 resistant Lp strains carried mutations in translation-related machinery , of which 17 carried mutations in either elongation factor P ( lpg0287 ) or enzymes responsible for adding post-translational modifications to elongation factor P ( lpg0607 and lpg0288 ) . Elongation factor P acts to re-start stalled ribosomes at polyproline tracts , and its post-translational modifications are essential for these functions ( Yanagisawa et al . , 2010; Navarre et al . , 2010; Doerfel et al . , 2013; Marman et al . , 2014 ) . Based on the frequency of polyprolines in the Lp proteome , disruptions to elongation factor P function have the potential to impact the expression of about 33% of Lp proteins . The HGA resistance phenotypes we observed in these 17 Lp mutants could therefore result from either a large-scale shift in gene expression , or from the altered expression of specific susceptibility genes . Elongation factor P disruption has also been previously linked to the activation of the stringent response pathway ( Nam et al . , 2016 ) , which is important for coordinating a variety of stress responses in Legionella , including oxidative stress ( Molofsky and Swanson , 2004; Oliva et al . , 2018 ) . We therefore tested if the stringent response pathway is involved in regulating HGA susceptibility or resistance . We assayed HGA susceptibility in mutant Lp strains with disruptions to rpoS , letA , relA , and spoT , which act early in the stringent response pathway to regulate and/or respond to the levels of the alarmone ppGpp ( Figure 7—figure supplement 1C ) ( Bachman and Swanson , 2001; Hammer et al . , 2002; Dalebroux et al . , 2009 ) . We found that the HGA susceptibility of all these mutants was similar to that of wild type; furthermore , complementation with these genes did not alter HGA susceptibility ( Figure 7—figure supplement 1D ) . Thus , we find no evidence of a role for the stringent response in HGA susceptibility or resistance . Instead , we propose that disruptions to elongation factor P result in pleiotropic translation defects that together lead to HGA resistance via still-unknown mechanisms . In addition to the translation-related mutants , we found five missense mutations in secY or secD ( lpg0349 and lpg2001 ) , members of the Sec secretion apparatus that moves polypeptides across the cytosolic membrane; one mutation in aceE pyruvate dehydrogenase ( lpg1504 ) ; and four mutations in a hypothetical gene , lpg1681 ( Table 1 , Figure 5C ) . The Sec apparatus is involved in the secretion of many substrates and mutations to this machinery could also lead to extensive pleiotropic defects . Instead , we focused on the relatively uncharacterized hypothetical gene lpg1681 , which encodes a small , 105 amino acid protein with no predicted domains apart from two transmembrane helices . This gene is adjacent in the genome to lpg1682 , which encodes for a predicted oxidoreductase/dehydrogenase , and lpg1680 , which encodes for the thiol:disulfide exchange protein DsbD2 ( Figure 5C , Figure 5—figure supplement 1 ) . Functional studies of DsbD2 ( aka DiSulfide Bond reductase D2 ) have demonstrated that it interacts with thioredoxin to regulate disulfide bond remodeling in the periplasm ( Inaba , 2009; Kpadeh et al . , 2015 ) . If lpg1681 has a redox function related to its neighboring genes , we expected its syntenic locus to be conserved across bacterial strains and species . Consistent with this prediction , we found that the lpg1680-1682 locus is present and conserved among over 500 sequenced Lp strains currently in NCBI databases . Outside L . pneumophila , lpg1681 is mostly restricted to the Legionella genus , present in about half of the currently sequenced species ( Burstein et al . , 2016 ) ( Figure 5—figure supplement 1 ) . A homolog of lpg1681 is also found in the draft genome of Piscirickettsia litoralis , a gamma proteobacterium and fish pathogen ( Wan et al . , 2016 ) . In all cases , lpg1681 resides upstream of dsbD2 , suggesting a functional link between these proteins ( Figure 5—figure supplement 1 ) and implicating lpg1681 in a role in redox homeostasis . We , therefore , viewed lpg1681 as a promising candidate for a gene involved in HGA susceptibility . We constructed lpg1681 overexpression and deletion strains in Lp and tested the susceptibility of these strains to HGA . Similar to the spontaneous mutants we recovered , we found that the ∆lpg1681 strain was more resistant to synthetic HGA in rich media ( Figure 5D , Figure 5—figure supplement 2 ) . Conversely , overexpression of lpg1681 increased Lp sensitivity , resulting in longer growth delays than wild type at high concentrations of HGA . We therefore conclude that wild type lpg1681 expression sensitizes Lp to inhibition by extracellular HGA . Given its genetic linkage with DsbD , our findings further suggest that alteration of disulfide bond regulation in the periplasm might constitute one means to mitigate HGA susceptibility . Our unbiased genetic screen for Lp resistance to HGA only revealed mutants that were partially resistant to HGA . These mutants had a smaller growth delay than wild type at a given HGA concentration , but all remained qualitatively susceptible to inhibition . These results suggest that the genetic routes for Lp to completely escape from HGA-mediated inhibition are limited . Yet , Lp secretes abundant HGA into its local environment , despite the fact that HGA secretion is not required for Lp growth or metabolism ( as seen by the robust growth of the hisC2::Tn mutants , Figures 2 and 4 ) . Thus , our findings do not provide an adequate explanation for the paradox of how Lp cells can secrete a toxic compound to which they apparently carry no heritable resistance . We therefore considered a distinct mechanism by which Lp might avoid self-inhibition: Lp might produce and secrete HGA only during conditions when it is not susceptible to HGA . To address this possibility , we investigated when and where Lp secretes HGA . We tracked HGA secretion across a growth curve of Lp in rich media with our previously described conditioned media assay . It has long been known that Lp produces abundant HGA-melanin pigment in stationary phase , when the bacteria are undergoing very slow or no growth ( Pine et al . , 1979; Berg et al . , 1985; Wiater et al . , 1994 ) . By comparing to a synthetic HGA standard curve ( Figure 4—figure supplement 1F ) , we estimate that Lp secretes a burst of HGA in stationary phase , producing between 183–266 µM of active HGA within 5 hrs ( Figure 6A ) . HGA secretion then continues after the population has ceased growing . These quantities of HGA are more than enough to be inhibitory to Lp ( Figure 4B ) . For Lp to avoid self-inhibition from HGA , we hypothesized that Lp might be resistant to this inhibitor at high density and/or when the cells are in a stationary phase of growth . We therefore investigated if cell density or growth phase impacted HGA susceptibility ( Figure 6B ) . For Lp exposed to HGA in rich media , we found that the growth phase of bacteria used to inoculate the experiment had little impact on HGA susceptibility ( Figure 6C ) . In contrast , when cells were inoculated at high density ( 10^9/mL instead of 10^8/mL ) , they were resistant even to high concentrations of HGA ( Figure 6C ) , suggesting that cell density might be linked to HGA resistance . However , because both cell density and growth phase are changing over time during our assays in rich media , we could not fully separate their contributions to HGA resistance in Lp . We therefore created a new assay to assess HGA susceptibility . We exposed Lp bacteria to HGA at different dilutions in nutrient-free PBS , which ensured that the bacteria did not grow or change cell density during the course of the experiment . After 24 hr exposure to 125 µM HGA , we assessed bacterial viability by plating for viable CFUs ( Figure 7A ) . No measurable darkening of the HGA was detected in this assay , suggesting that the oxidation and de-activation of HGA was considerably slowed in low-nutrient conditions . We found that Lp bacteria incubated at high density with HGA ( above 10^8 CFU/mL ) were largely protected from inhibition ( Figure 7A ) . However , at lower density ( 10^7 CFU/mL ) , Lp bacteria were extremely sensitive to HGA , with at least a 10^6-fold reduction in CFUs , to below our limit of detection . This result suggests that , although HGA is bacteriostatic in rich media , it appears to be strongly bacteriocidal in PBS . Nevertheless , as in the rich media assay , the resistance to HGA was dependent on cell density and was not altered by inoculum growth phase ( Figure 7A ) . The dramatic loss in viable CFUs in the PBS assay allowed us to further investigate the mechanism behind low-density susceptibility . Specifically , we asked if this loss in viability was due to HGA or one of its transient oxidative intermediates by exposing cells to HGA in both aerobic and anaerobic conditions . We observed that HGA was not inhibitory to Lp in anaerobic conditions ( Figure 7B ) , suggesting that HGA-mediated toxicity comes from the action of a reactive intermediate molecule generated during HGA oxidation . We then considered the density-dependent difference in HGA susceptibility , reasoning that quorum sensing would be most likely to control this phenomenon . When we asked if HGA resistance depended on the previously described Lp quorum sensing response regulator , lqsR ( Tiaden et al . , 2007 ) , we found that deleting lqsR had no detectable impact on HGA susceptibility or resistance ( Figure 7—figure supplement 1B ) . Therefore , the density-dependent susceptibility of Lp to HGA must be independent of the lqsR pathway . We next investigated the basis of high-density resistance to HGA by Lp bacteria , hypothesizing that high-density cells could alter the activity of extracellular HGA , either through the secretion of inactivating compounds or through bulk , non-specific binding of HGA to bacterial biomass , leading to a reduction in its effective concentration . To test both hypotheses , we recovered the supernatants from high-density and low-density bacteria that had been incubated with or without HGA , and applied these supernatants to fresh , low-density Lp to assess HGA activity ( Figure 7C ) . We found that the supernatants from HGA-exposed Lp remained fully inhibitory , even after 24 hr incubation with high-density bacteria . Furthermore , we found that adding heat-killed Lp bacterial cells to low-density viable Lp bacteria did not enhance the latter’s resistance to HGA inhibition ( Figure 7—figure supplement 1A ) . Therefore , we conclude that HGA susceptibility appears to be density-dependent and yet cell-intrinsic . Because Lp bacteria at high density both secrete and are protected from HGA , this strategy of secreting HGA only when Lp cells are conditionally HGA-resistant may allow Lp to produce a broadly active inhibitor while restricting the potential for self-harm . The HGA-melanin pathway is well-studied and widespread among bacteria and eukaryotes . In this study , we identify HGA’s oxidative intermediates as a mediator of inter-Legionella inhibition , both between Legionella species and even between genetically identical populations of L . pneumophila . To our knowledge , this is the first time that HGA has been described to have antimicrobial activity . One reason HGA-mediated inhibition may not have been previously documented is that the active compound ( s ) are redox-active , unstable molecules with transient activity . Our study finds that synthetic HGA can auto-oxidize over the course of an experiment to form inactive HGA-melanin ( Figure 3 , Figure 4—figure supplement 1D ) , allowing exposed Lp populations to rebound following initial inhibition ( Figure 4B ) . Intriguingly , although Lp populations recover upon HGA oxidation , Lm populations do not , suggesting that HGA may cause more harm to Lm cells . The quenching of HGA’s inhibitory activity occurs especially rapidly in the cysteine-rich microbial media typically used to grow Legionella in the lab ( Figure 4—figure supplement 1D ) . Conversely , we find that HGA becomes more potent in low-cysteine media or in PBS ( Figure 3F and 7 ) , where oxidation of HGA into HGA-melanin occurs more slowly; such nutrient-poor conditions may better replicate the nutrient-poor ( oligotrophic ) aquatic environments of Legionella’s natural habitat ( Atlas , 1999; Boamah et al . , 2017 ) . Although dense , stationary-phase cultures of L . pneumophila secrete abundant HGA ( Figure 6A , Figure 2—figure supplement 1D ) , we also find that these bacteria do not possess heritable resistance to HGA ( Figure 4B–D ) and are highly susceptible to HGA-mediated inhibition at low cell density . However , they exhibit conditional , cell intrinsic resistance to HGA at high cell density ( Figure 6 and 7 ) . This lack of heritable resistance makes HGA-mediated inhibition different from classical antibiotics or toxins , which typically are produced by bacteria that also express resistance genes or antitoxins . HGA inhibition is also distinct from that caused by toxic metabolic by-products in two important ways . First , HGA production is not required for efficient growth or metabolism in Legionella species ( Figure 2—figure supplement 1D , and see growth of hisC2::Tn mutant in Figure 4A ) . Second , high-density populations of L . pneumophila that produce HGA are themselves protected from HGA inhibition ( Figure 6 and 7 ) . Only neighboring , low-density Legionella are strongly inhibited ( Figure 4A and 6D ) . The strong density-dependence of Legionella’s susceptibility to HGA may be another reason that it has been previously undiscovered despite intense study of these bacteria . HGA-mediated inhibition of Legionella is reminiscent of the antimicrobial activities of phenazines , another class of small aromatic molecules including pyocyanin from Pseudomonas aeruginosa . Both types of molecules are redox-active ( Hassan and Fridovich , 1980 ) , are produced at high cell density ( Hassan and Fridovich , 1980; Baron et al . , 1989 ) , are able to chemically react with thiol groups ( Cheluvappa et al . , 2008; Heine et al . , 2016 ) , and result in the production of a colored pigment ( Price-Whelan et al . , 2006 ) . Phenazine inhibitory activity is typically thought to come from redox cycling and the production of reactive oxygen species , including H2O2 ( Hassan and Fridovich , 1980; Cheluvappa et al . , 2008 ) . Oddly , HGA-melanin production has previously been implicated both in the production of ( Noorian et al . , 2017 ) and protection from ( Keith et al . , 2007; Orlandi et al . , 2015 ) reactive oxygen species . The catalase experiments presented here have ruled out the production of extracellular H2O2 as a possible mechanism behind HGA inhibition ( Figure 3—figure supplement 1B–C ) . Instead , based on association of lpg1681 and DsbD2 with reduced HGA sensitivity ( Figure 5 ) , the ability of diverse thiols to quench HGA’s activity ( Figure 3—figure supplement 1E ) , and precedents from phenazines ( Heine et al . , 2016 ) , we speculate that HGA’s transient , oxidative intermediates may be toxic by forming adducts on cysteine residues or otherwise disrupting disulfide bonding . Alternatively , HGA-mediated inhibition could occur via the production of other reactive oxygen species , including potentially the generation of intracellular superoxide and/or H2O2 ( Hassan and Fridovich , 1979 ) , which would not be affected by catalase treatment . As both of these mechanisms should be broadly inhibitory to a number of bacterial taxa , it will be interesting to survey HGA susceptibility outside of Legionella . The density-dependence of Lp’s resistance to HGA is unusual and worthy of future study . Because high-density cells do not inactivate or bind up extracellular HGA ( Figure 7C ) and because heat-killed cells cannot protect live , low-density Lp from inhibition ( Figure 7—figure supplement 1A ) , we infer that resistance is cell-intrinsic , resulting from differing physiology and/or gene expression between high- and low-density cells . Two pathways that commonly regulate such defenses include the stringent response pathway , which becomes active under nutrient limitation and stress , and quorum sensing pathways , which become active at high cell density ( Bachman and Swanson , 2001; Hammer et al . , 2002; Tiaden et al . , 2007; Dalebroux et al . , 2009; Hochstrasser and Hilbi , 2017 ) . Although our experiments disrupting these pathways suggest that neither pathway contributes to density-dependent susceptibility or resistance ( Figure 7—figure supplement 1B–D ) , we note that quorum sensing in Legionella remains understudied , and such pathways vary considerably across bacterial taxa ( Hochstrasser and Hilbi , 2017; Miller and Bassler , 2001 ) . Future work using unbiased approaches to investigate the regulation of HGA susceptibility may be able to uncover additional density-sensing pathways , possibly including an undescribed mode of quorum sensing in Legionella . Finally , we note that the density-dependent resistance was observed in well-mixed , liquid cultures in both rich media ( Figure 6C ) and in PBS ( Figure 7A ) . In natural conditions , high-density bacteria may be further protected within anaerobic regions of a biofilm , as HGA was not toxic to L . pneumophila in these conditions ( Figure 7B ) . L . pneumophila is often co-isolated with other Legionella species , which likely compete for similar ecological niches ( Wéry et al . , 2008; Pereira et al . , 2017 ) . HGA-mediated inter-Legionella inhibition therefore has a strong potential to impact the success of Lp in both natural and man-made environments . Because high-density , established Lp bacterial communities are largely resistant to HGA inhibition , these communities might use HGA to protect against low-density , invading Legionella competitors with little harm to themselves ( Figure 7D ) . In this model , motile Lp can disperse , colonize a new surface , and grow into a microcolony using the locally available nutrients . In these early stages , no HGA is produced . After the Lp population grows up and crosses a certain cell density threshold , the cells become HGA-resistant through a cell-intrinsic mechanism . When this dense population enters stationary phase , it also begins to secrete abundant HGA into the local environment . This secreted HGA has minimal impact on the resistant , producer cells . However , it can inhibit the growth of nearby , low-density Legionella , whether the neighboring cells are other Legionella species or even genetically-identical Lp . Given these dynamics observed in the lab , we speculate that HGA and other such inhibitors may be deployed as a bacterial niche-protective strategy . The bacterial strains and plasmids used for this study are listed in Supplementary file 1 . As our wild type Legionella pneumophila ( Lp ) strain , we used KS79 , which was derived from JR32 and ultimately from isolate Philadelphia-1 ( de Felipe et al . , 2008; Sadosky et al . , 1993; Rao et al . , 2013 ) . Compared to JR32 , the KS79 strain has a comR deletion to enable genetic manipulation ( de Felipe et al . , 2008 ) . We used Legionella micdadei ( Lm ) tatlock as our susceptible strain ( Garrity et al . , 1980; Hébert et al . , 1980 ) . Liquid cultures of Legionella were grown shaking in AYE rich liquid media at 37 ˚C ( De Jesús et al . , 2013 ) . Unless otherwise indicated , experiments were inoculated with stationary phase Legionella , grown from a single colony in AYE for 16–18 hr as described , to a density of 3–4 × 10^9 CFU/mL . For experiments with log phase Lp , an overnight culture was diluted into fresh AYE at 1:8 ratio ( to a density of 4–5 × 10^8 CFU/mL ) and allowed to grow to a density of 10^9 CFU/mL before setting up the experiment . To manipulate the redox state of AYE , we altered the amount of cysteine added to the media from 0 . 4 g/L in standard AYE to 0 . 1 , 0 . 2 , and 0 . 8 g/L . On solid media , Legionella were grown either on BCYE agar plates either containing the standard cysteine concentration ( 0 . 4 g/L ) or ‘low cysteine’ ( 0 . 05 g/L ) ( Feeley et al . , 1979; Solomon and Isberg , 2000 ) . E . coli strains used for cloning were grown in LB media . Where indicated , antibiotics were used at the following concentrations in solid and liquid media , respectively; chloramphenicol ( 5 µg/mL and 2 . 5 µg/mL ) , kanamycin ( 40 µg/mL ) and ampicillin ( 50 µg/mL and 25 µg/mL ) . For counter-selection steps while generating deletion strains , 5% sucrose was added to BCYE plates . For agar overlay experiments , we used 0 . 7% agar dissolved in water , which was kept liquid at 50 ˚C before pouring over low cysteine BCYE plates . Genomic knockouts in L . pneumophila were generated as previously described ( Wiater et al . , 1994 ) . Briefly , we used an allelic exchange plasmid ( pLAW344 ) harboring chloramphenicol and ampicillin selection cassettes and the counter-selection marker SacB , which confers sensitivity to sucrose . Into this plasmid , we cloned ~ 1 kb regions upstream and downstream of the gene of interest to enable homologous recombination . Following electroporation and selection on chloramphenicol , we used colony PCR to verify insertion of the plasmid into the chromosome , before counter-selection on sucrose media . From the resulting colonies , we performed PCR and Sanger sequencing to verify clean gene deletion . For complementation , the coding region of a candidate gene was cloned into a plasmid ( pMMB207c ) following a ptac promoter ( Chen et al . , 2004 ) . To induce gene expression , strains carrying pMMB207c-derived plasmids were exposed to 1 mM IPTG . All constructs were assembled using Gibson cloning ( NEB Catalog #E2621 ) . To visualize inhibition between neighboring Legionella on solid media , a streak of approximately 5 × 10^6 CFU of the inhibitory strain of Lp was plated across the center of a low cysteine BCYE plate . After 3 days growth at 37 ˚C , dilutions of susceptible Lp or Lm were plated as 10 µL spots approximately 1 cm and >2 cm from the central line . Once spots were dry , plates were then incubated for an additional 3 days before scoring for inhibition . This assay was also used to quantify the bactericidal inhibition of Lm , with slight modifications . Here , all Lm was plated in 20 µL spots at 10^6 CFU/mL . The time of plating susceptible Lm was treated as t = 0 . Once spots were dry , plugs were extracted from within the Lm spots using the narrow end of a Pasteur pipette . These plugs were transferred into media , vortexed , and plated to quantify CFU . This procedure was repeated after 48 hr at 37 ˚C to compare Lm viability and growth within ( ‘near’ , Figure 1B–C ) or outside ( ‘far’ ) of the zone of inhibition . For inhibition assays on bacterial lawns , we plated 10 µL drops of either live Lp or chemical compounds on top of a lawn of 5 × 10^7 CFU Lm on low cysteine BCYE , and assessed growth of the lawn after 3 days at 37 ˚C . Synthetic HGA ( Sigma: #H0751 ) was dissolved in water at a concentration of 100 mM and filter sterilized before use . To limit the potential for HGA oxidation prior to use , 100 mM aliquots prepared in water were stored frozen at −20C and discarded after 1–2 weeks . HGA-related compounds , 2-hydroxyphenylacetic acid ( Sigma: #H49804 ) and 3-hydroxyphenylacetic acid ( Sigma: #H49901 ) , were prepared in the same way . To test the impact of DTT ( Sigma: #43819 ) and glutathione ( oxidized: Sigma #G4376 , reduced: Sigma #G6529 ) on HGA-mediated inhibition , filter-sterilized solutions dissolved in water were mixed in equimolar ratios with HGA , and incubated shaking at room temperature for 1 hr before spotting onto bacterial lawns . HGA-melanin pigment was prepared from Lp conditioned media as previously described ( Zheng et al . , 2013 ) from KS79 , the unpigmented hisC2::Tn mutant , and the hyperpigmented ∆hmgA mutant . Briefly , conditioned media was collected and sterile filtered from 100 mL cultures of Lp in AYE media grown shaking at 37 ˚C for 3 days . The conditioned media was acidified to a pH of 1 . 5 and transferred to 4 ˚C for 2 hr to precipitate . Precipitated pigment was collected by centrifugation at 4000 x g for 15 min and then washed with 10 mM HCl . Pelleted pigment was then returned to neutral pH and resuspended in PBS at 10X before testing . For random transposon insertion mutagenesis , we used a Mariner transposon from the pTO100 plasmid ( O'Connor et al . , 2011 ) . We electroporated this plasmid into the KS79 strain and allowed cells to recover at 37 ˚C for 5 hr . To select for cells with integrated transposons , cultures were plated on BCYE/Kan/sucrose plates and incubated at 37 ˚C for 3 days before screening individual mutant colonies . To identify transposon mutants with defects in Lm inhibition , we transferred each Lp mutant onto a low cysteine plate with a lawn of 5 × 10^7 CFU Lm and visually screened for those with either small zones of inhibition or no zone of inhibition . This transfer of Lp mutants was achieved either by replica plating using sterile Whatman paper ( Whatman: #1001150 ) or by manual transfer with a sterile toothpick . Plates were then incubated at 37 ˚C for 3 days and scored . All putative mutants underwent clonal re-isolation , were diluted to OD 600 of 0 . 1 , and spotted on fresh Lm lawns to retest their phenotypes . To map the sites of transposon integration , we used arbitrary PCR as described in Chen et al . ( 1999 ) , with primers redesigned to work with the pTO100 transposon ( Supplementary file 1 ) . Briefly , this protocol involved two PCR steps to amplify the DNA flanking the transposon . The first step used low annealing temperatures to allow the arb1 randomized primer to bind many sites in the flanking DNA while the pTO100_F or pTO100_R primer annealed within the transposon , generating multiple products that overlapped the flanking DNA . These products were amplified in the second step PCR using the arb2 and pTO100_Rd2 primers , and we used the pTO100_Rd2 primer for Sanger sequencing . PCR programs and conditions were as in Chen et al . ( 1999 ) . For rich media assays ( e . g . Figures 3C–F , 4B and D ) , overnight cultures of Legionella were diluted to 10^8 CFU/mL in AYE , mixed with synthetic HGA ( at 0 , 62 . 5 , 125 , 250 , or 500 µM final ) or with isolated pigment in 96 well plates , and grown shaking at 425 cpm at 37 ˚C . The cytation three imaging reader ( BioTek CYT3MV ) was used to monitor growth by OD 600 measurements . Because oxidized pigment from synthetic HGA is detected at OD 600 as well , each experiment included bacteria-free control wells containing media and each concentration of HGA . To correct OD 600 readings for pigment development , at each time point we subtracted the control well reading from bacterial wells that received the same concentration of synthetic HGA . For experiments with HGA ‘pre-oxidation’ ( Figure 3D ) , we diluted HGA in AYE media and incubated this solution shaking at 37 ˚C for 24 hr in the plate reader before adding Lm bacteria . Complete oxidation of HGA during the 24 hr was monitored using OD 400 to track the accumulation of HGA-melanin pigment ( Figure 4—figure supplement 1 ) . To test if extracellular catalase could protect from HGA inhibition ( Figure 3—figure supplement 1B ) , we incubated Lm at 10^7 CFU/mL with or without 125 µM HGA and either 0 , 1 , 10 , or 100 U/mL of bovine catalase ( Sigma #C30 ) . As a control to ensure that the catalase was active , we incubated Lp as above with catalase and 2 mM H2O2 ( Sigma #88597 ) . In Lp , HGA inhibition in AYE rich media resulted in a growth delay , similar to an extended lag phase ( Figure 4B ) . To determine if this delay was due to genetic adaptation , we sampled Lp after 70 hr growth with 250 µM HGA or without HGA ( Figure 4C–D ) . These bacteria were washed once and resuspended in fresh AYE , before being diluted back to 10^8 CFU/mL and then exposed to fresh , synthetic HGA as above . To assess the correlation between HGA oxidation and the length of the Lp growth delay ( Figure 4E ) , we pooled data from eight experiments on different days that measured wild type Lp ( KS79 ) exposed to a range of HGA concentrations in AYE . We considered the ‘Time to full HGA oxidation’ as the length of time required for a given concentration of HGA to stop forming additional HGA-melanin , measured as the time until OD 400 readings increased by less than or equal to 0 . 001 units per hour . The ‘Time to mid-log’ was measured as the time when Lp exposed to that concentration of HGA had grown to an OD 600 of 0 . 1 . To compare sensitivity to HGA among Lp strains , we calculated the lag phase from the growth curve of each well using the GrowthRates program ( Hall et al . , 2014 ) . We excluded a small number of samples where the growth curve was not well fit ( R < 0 . 99 ) , and then for each strain used the difference in lag time between the samples with and without HGA to calculate the growth delay due to HGA ( Figure 5D ) . While we were able to manipulate inoculum growth phase and cell density in the AYE assays , during these experiments the bacteria altered their density and growth phase as they grew in rich media . To separate the impacts of cell density and growth phase on HGA susceptibility , we used a complementary assay in which we evaluated Legionella viability when exposed to HGA in nutrient-free PBS at different cell densities . This design ensured that the bacteria maintained a constant cell density throughout the course of the experiment . Stationary phase cultures were washed 1–2 times and re-suspended in PBS . We diluted these bacteria to estimated starting concentrations of 10^9 , 10^8 , and 10^7 CFU/mL and plated for CFU at t = 0 . We distributed the remaining bacteria into 96 well plates with or without 125 µM HGA . Plates were incubated shaking in a plate reader at 425 cpm at 37 ˚C for 24 hr before plating to quantify CFUs on BCYE plates . CFUs were counted after 3–4 days growth at 37 ˚C . To assess HGA toxicity in aerobic vs . anaerobic conditions , Lp stationary phase bacteria were prepared as above and diluted to 10^8 CFU/mL in PBS ± 125 µM HGA in 500 uL volumes in Celltreat bio-reaction tubes ( # 229472 ) . Aerobic samples were incubated shaking in air at 37 ˚C for 24 hr , while anaerobic samples were incubated static at 37 ˚C in an anaerobic chamber ( Coy lab products , #032714 ) filled with 5% hydrogen , 10% CO2 , and 85% N2 . Samples were plated to quantify CFUs as above . To determine if high density bacteria were protected via mass action effects that diluted out the amount of HGA per cell through binding of bulk material , we asked if the addition of dense , heat-killed bacteria could protect low-density Lp from HGA ( Figure 7—figure supplement 1A ) . To prepare high-density heat-killed bacteria , an overnight culture of Lp was washed once in PBS , resuspended to 2 × 10^9 CFU/mL , and incubated at 100–110˚C for 60 min . After heating , this sample was diluted 1:2 and mixed with 10^7 CFU/mL live Lp in PBS ± 125 µM HGA to assess protection . As a control , 10^9 and 10^7 CFU/mL live Lp with or without HGA were tested simultaneously . Cells were incubated and plated as above to assess viability . To determine if high density bacteria were protected via HGA degradation , we tested if the supernatants from HGA-exposed , high-density bacteria retained the potency to inhibit low-density bacteria ( Figure 7C ) . To generate supernatants , we set up 2 mL samples containing 10^8 or 10^7 CFU/mL of Lp in PBS ± 125 µM HGA and incubated them shaking at 37 ˚C for 20 hr . After plating aliquots for viable CFU , we pelleted the remaining bacteria and sterile filtered 1 mL of each supernatant through a 0 . 2 µm filter . Each supernatant was tested in triplicate , incubated with fresh Lp at 10^7 CFU/mL in a 96 well plate as above for 24 hr before plating for CFU . As a control , 10^7 CFU/mL live Lp were incubated in PBS alone . HGA is known to be a redox-active molecule , with a redox potential of +0 . 636V ( Eslami et al . , 2013 ) . As this measurement can be altered by pH and temperature , we assessed the ability for HGA to oxidize cysteine in our experimental conditions using Ellman’s reagent ( also known as 5 , 5’-dithiobis- ( 2-nitrobenzoic acid ) , DTNB , Invitrogen #D8451 ) . Ellman’s reagent reacts in the presence of reduced thiol groups on L-cysteine to form a yellow color , which can be read as 412 nm absorbance . When thiol groups are oxidized , the Ellman’s reagent is colorless . We used the ability for HGA to decrease the amount of reduced cysteine as a proxy for its oxidizing ability . Stock solutions of both 100 mM HGA and 1 . 5 mM L-cysteine ( Sigma #C6852 ) were prepared fresh in PBS at the start of the experiment . Different concentrations of HGA ( from 125 µM to 8 mM ) were incubated in triplicate , shaking at 25°C with 1 . 5 mM cysteine in PBS for 16 hr . These conditions were compared to a standard curve of cysteine from 0 to 2 mM , which were incubated in parallel with the experimental samples to account for cysteine oxidation over time . To quantify the remaining free thiol groups , 180 uL of 0 . 08 mg/mL Ellman’s reagent was mixed with 17 . 65 uL of each experimental or standard sample in a 96 well plate , incubated for 3 hr at 25°C , and read for 412 nm absorbance . The ‘decrease in reduced cysteine’ was calculated as the difference between the initial and final measured cysteine concentrations , based on the standard curve conversion . HGA-melanin is a black-brown pigment that is easily detected at OD 400 . We took advantage of this coloration to estimate the amount of HGA that had been secreted by Lp by comparing the color of conditioned media to a standard curve of oxidized synthetic HGA . To isolate conditioned media from pigment mutant strains , cultures of KS79 , ∆hmgA , and hisC2::Tn were inoculated from fresh colonies from a BCYE plate into 5 mL AYE and were grown shaking at 37 ˚C for 48 hr . We then collected conditioned media by pelleting the bacteria and passing the supernatant through a 0 . 2 µm filter . To harvest conditioned media for a time course , cultures of Lp were inoculated into 5 ml AYE and grown shaking at 37 ˚C . After 15 , 20 , 24 , 39 , 44 , and 48 hr , we measured the OD 600 of the culture and collected conditioned media . To create a standard curve , we diluted synthetic HGA into AYE at the following concentrations: 62 . 5 µM , 125 µM , 250 µM , 500 µM , and 1 mM . The conditioned media and standard curve samples were incubated in a 96 well plate in a plate reader shaking at 37 ˚C for 24 hr to allow the HGA to oxidize . We used OD 400 data from the 24 hr time point to generate a standard curve for each HGA concentration and calculated a line of best fit using linear regression . This equation was used to estimate the amount of secreted HGA that corresponded to the OD 400 of each conditioned media sample . ( Figure 4—figure supplement 1 ) . In these results , we saw that the pool of HGA + HGA melanin increased by 266 µM within 5 hr . Based on the HGA-melanin production kinetics we measured in AYE ( Figure 4—figure supplement 1D ) where 250 µM HGA oxidized ~ 31% in the first 5 hr , we calculated the likely range of active HGA concentrations . If all 266 µM HGA were secreted instantaneously at the beginning of the 5 hr window , by the end of that window 266 x ( 1–0 . 31 ) =183 µM of active HGA would remain . Therefore , we estimate that 183–266 µM of active HGA was produced during the 5 hr window . Because the inhibitory activity of HGA is quenched through interactions with cysteine in rich media ( e . g . Figure 3F ) , it was not possible to select for HGA-resistant mutants by mixing HGA into BCYE agar . Instead , to reduce the potential for HGA to react with media components while allowing sufficient access to nutrients for mutant cells to grow , we selected for HGA-resistant mutants by mixing 4 × 10^7 CFU Legionella with 1 mM HGA in 4 mL of 0 . 7% molten agar and pouring this solution as an overlay on a low cysteine BCYE plate . Plates were incubated at 37 ˚C for 6 days , before candidate resistant colonies were picked and clonally isolated . The HGA resistance and growth of each isolate was re-tested on overlays with or without 1 mM HGA on both regular and low cysteine BCYE . Twenty-nine isolates were more resistant to HGA than wild type Lp upon retesting . We sequenced and analyzed genomic DNA from these isolates and a matched wild type strain as follows . DNA was prepared from each strain using a Purelink genomic DNA mini kit ( Invitrogen , #K1820 ) . DNA concentrations were quantified using Qubit and normalized to 0 . 5 ng/uL . Barcoded libraries were prepared using tagmentation according to Baym et al . , 2015Baym et al . , 2015 , analyzed with QuantIT DNA quantification , pooled , and sequenced with 50 bp paired-end reads on an Illumina HiSeq 2500 machine . Reads were trimmed for quality and to remove Nextera indices with Trimmomatic ( Bolger et al . , 2014 ) and mapped to the Philadelphia-1 genome ( Chien et al . , 2004 ) using Bowtie2 with default parameters ( Langmead et al . , 2009 ) . Coverage plots were generated for each strain using bamcoverage ( Ramírez et al . , 2016 ) and manually examined for evidence of large genomic deletions and amplifications . None were observed , apart from a prophage that was present in the reference genome but missing from all sequenced strains , including our wild type KS79 strain . Variants were detected for each mutant using Naive Variant Caller ( Blankenberg , 2019 ) . Those variants that were detected in mutant strains but not the wild type strain were considered as putative causative mutations . For each of these mutations , we inspected the mapped reads and excluded faulty variant calls that either were adjacent to coverage gaps or that did not appear to be fixed in the clonal mutant and/or wild type sequences , likely due to errors in read mapping . After this manual filtering , 1–3 well-supported mutations remained for each mutant genome . Nine of the mutants were isolated on a different day from the other mutants; in addition to various unshared mutations , these nine strains each carried exactly the same missense mutation in rplX , which we disregarded as a background mutation that likely arose before selection . Following this exclusion , each mutant carried only a single well-supported mutation in a coding region . Most often this coding mutation was the only mutation we detected , although one mutant carried two additional intergenic point mutations . The coding mutations were point mutations or small deletions that resulted in non-synonymous changes , frame shifts , or gene extensions . Across different mutants , the mutations we uncovered were repeatedly found in the same , few loci ( Table 1 ) . The genes in the HGA-melanin synthesis pathway are highly conserved in diverse bacteria and across the Legionella genus , with all genes present in all 41 currently sequenced Legionella spp . genomes ( Burstein et al . , 2016 ) . In contrast , we were able to identify lpg1681 in only 30 Legionella spp . genomes , as well as a single draft genome outside the Legionella genus– Piscirickettsia litoralis , an intracellular fish pathogen ( Wan et al . , 2016 ) . Across the lpg1681-containing genomes , there is evidence for extensive recombination of the flanking loci , yet lpg1681 is always found upstream of dsbD2 . We identified most of these homologs of lpg1681 using a jackhmmr search ( Finn et al . , 2015 ) , followed by cross-referencing the homologs with the Legionella orthology groups defined by Burstein et al . ( 2016 ) . From this starting set , additional lpg1681 orthologs were identified in unannotated , intergenic regions by searching for > 200 bp open reading frames upstream of dsbD orthologs , and confirming the homology of these regions using MAFFT alignments ( Katoh et al . , 2002 ) . Through this method , we located lpg1681 in all currently sequenced Legionella genomes that contain an annotated dsbD2 gene , with the exception of L . shakespearei . We categorized the lpg1681-containing loci into those with similar synteny , based on the orthology group annotations in Burstein et al . ( 2016 ) . We colored and provided names for the neighboring genes in Figure 5—figure supplement 1 if they had a homolog in the L . pneumophila Philadelphia-1 genome that was not annotated as a hypothetical gene . To assess the conservation of the lpg1680-lpg1682 among L . pneumophila strains , we used blastn in the NCBI nr and wgs databases with the full lpg1680-lpg1682 genomic DNA sequence as a query . We found that the full region was conserved with few mutations across 501 currently sequenced L . pneumophila strains .

In the environment , bacteria frequently compete with each other for resources and space . These battles often involve the bacteria releasing toxins , antibiotics or other molecules that make it more difficult for their neighbors to grow . The bacteria also carry specific resistance genes that protect them from the effects of the molecules that they produce . Legionella pneumophila is a species of bacteria that infects people and causes a severe form of pneumonia known as Legionnaires’ disease . The bacteria spread in droplets of water from contaminated water systems such as sink faucets , cooling towers , water tanks , and other plumbing systems . In these water systems , L . pneumophila cells live within communities known as biofilms , which contain many different species of bacteria . These communities often include other species of Legionella that compete with L . pneumophila for similar nutrients . However , L . pneumophila was not known to produce any toxins or antibiotics , so it was not clear how it is able to survive in biofilms . Levin et al . used genetic approaches to investigate how L . pneumophila competes with other species of Legionella . The experiments found that this bacterium released a molecule called homogentisic acid ( HGA ) that reduced the growth of neighboring Legionella bacteria . Unexpectedly , L . pneumophila was not always resistant to HGA , despite secreting large quantities of this molecule . Instead , L . pneumophila cells were only resistant to HGA when the bacteria were living in crowded conditions . Previous studies have shown that HGA is widely produced by bacteria and other organisms – including humans – but this is the first time it has been shown that this molecule limits the ability of bacteria to grow . The work of Levin et al . suggests that HGA may help L . pneumophila bacteria to persist in biofilms , but more work needs to be done to test this idea . A possible next step is to test whether drugs that inhibit the production of HGA can eliminate Legionella bacteria from water systems . If so , similar treatments could potentially be used to stop and prevent outbreaks of Legionnaires’ disease in the future .

2019

Density-dependent resistance protects Legionella pneumophila from its own antimicrobial metabolite, HGA

The cullin-RING ligases ( CRLs ) form the major family of E3 ubiquitin ligases . The prototypic CRLs in yeast , called SCF enzymes , employ a single E2 enzyme , Cdc34 , to build poly-ubiquitin chains required for degradation . In contrast , six different human E2 and E3 enzyme activities , including Cdc34 orthologs UBE2R1 and UBE2R2 , appear to mediate SCF-catalyzed substrate polyubiquitylation in vitro . The combinatorial interplay of these enzymes raises questions about genetic buffering of SCFs in human cells and challenges the dogma that E3s alone determine substrate specificity . To enable the quantitative comparisons of SCF-dependent ubiquitylation reactions with physiological enzyme concentrations , mass spectrometry was employed to estimate E2 and E3 levels in cells . In combination with UBE2R1/2 , the E2 UBE2D3 and the E3 ARIH1 both promoted SCF-mediated polyubiquitylation in a substrate-specific fashion . Unexpectedly , UBE2R2 alone had negligible ubiquitylation activity at physiological concentrations and the ablation of UBE2R1/2 had no effect on the stability of SCF substrates in cells . A genome-wide CRISPR screen revealed that an additional E2 enzyme , UBE2G1 , buffers against the loss of UBE2R1/2 . UBE2G1 had robust in vitro chain extension activity with SCF , and UBE2G1 knockdown in cells lacking UBE2R1/2 resulted in stabilization of the SCF substrates p27 and CYCLIN E as well as the CUL2-RING ligase substrate HIF1α . The results demonstrate the human SCF enzyme system is diversified by association with multiple catalytic enzyme partners . In the ubiquitin-proteasome system ( UPS ) , ubiquitin ligases ( E3s ) bind to protein substrates to direct poly-ubiquitin chain formation and degradation by the 26S proteasome . There are at least 600 E3s in the mammalian proteome ( Deshaies and Joazeiro , 2009 ) , and alterations of the expression levels in several E3s or their substrates have been clinically and biochemically linked to cancer and other diseases in humans ( Skaar et al . , 2014; Wang et al . , 2014b ) . E3 enzymes recruit both the protein substrate as well as a ubiquitin-conjugating enzyme ( E2 ) thioesterified to the protein modifier ubiquitin ( Kleiger and Mayor , 2014 ) . E3s promote the covalent modification of a lysine residue on the E3-bound substrate , in most cases by direct transfer of the ubiquitin moiety from E2 to substrate ( Berndsen and Wolberger , 2014; Metzger et al . , 2014; Vittal et al . , 2015 ) . A chain of at least four ubiquitins is typically required for recognition by the 26S proteasome ( Chau et al . , 1989; Thrower , 2000 ) . Despite much progress , the different roles of E2s and E3s in cells remain enigmatic due to a lack of kinetics , a lack of knowledge of potentially relevant concentrations , and an understanding of how different E2-E3 combinations may regulate substrate polyubiquitylation . Some E2s can link ubiquitin directly to substrate in a priming step , whereas other E2s are dedicated to subsequent cycles of poly-ubiquitin chain extension . Yet other E2s are competent at both priming and chain extension . However , these properties may change dramatically as a function of E2 levels , in particular at the high E2 concentrations typically used in in vitro reconstituted ubiquitylation reactions . Although E2s are clearly harnessed by E3s to transfer ubiquitin ( Pickart and Rose , 1985 ) , the potential combinatorial complexity is overwhelming because many E3s appear to collaborate with multiple E2s to promote substrate ubiquitylation ( Christensen et al . , 2007; Rodrigo-Brenni and Morgan , 2007; Wickliffe et al . , 2011 ) . Defining the roles of particular enzymes and enzyme combinations remains an as yet unmet challenge in understanding how the UPS controls cellular function . The Skp1-cullin-F-box ( SCF ) ubiquitin ligases are a case study for the combinatorial interplay of E2s during substrate ubiquitylation . SCFs are modular multi-subunit complexes that all contain a common cullin 1 ( CUL1 ) core subunit and a ‘really interesting new gene’ ( RING ) domain RBX1 subunit ( Harper and Tan , 2012; Lydeard et al . , 2013; Zimmerman et al . , 2010 ) . CUL1 acts as a scaffold that tethers substrate binding subunits to the E2-binding subunit: the C-terminal segment of CUL1 binds to the RING subunit , which in turn recruits the ubiquitin-charged E2 , while the N-terminal segment of CUL1 binds to adaptors called F-box proteins that recruit substrates . SCFs are the archetypal member of the Cullin-RING ligases ( CRLs ) , which collectively account for at least 200 of the known E3s in humans ( Lydeard et al . , 2013 ) . The original elucidation of Skp1-cullin-F-box ( SCF ) ubiquitin ligase function by genetic and biochemical studies in the budding yeast Saccharomyces cerevisiae , suggested that Cdc34 is the only E2 enzyme needed for SCF activity ( Feldman et al . , 1997; Schwob et al . , 1994; Skowyra et al . , 1997; Verma et al . , 1997 ) . In human cells , an understanding of the relationship between E2s and SCF function has been confounded by the large repertoire of ubiquitin-modifying enzymes associated with SCF complexes . While the yeast E2s Ubc4/5 do not support SCF activity in vitro , several human SCF ubiquitylation reactions have been reconstituted with the corresponding human E2 orthologs UBE2D1/2/3 ( Sakata et al . , 2007; Wu et al . , 2003; Wu et al . , 2000 ) . A ‘hand-off’ model has been proposed where UBE2D3 transfers the first ubiquitin to an SCF-bound substrate , followed by poly-ubiquitin chain elongation catalyzed by the human CDC34 ortholog UBE2R1 or its highly related isoform UBE2R2 ( Wu et al . , 2010 ) . More recently , the E3 enzyme ARIH1 was shown to function with multiple CRLs in a manner similar to UBE2D3 in both humans ( Scott et al . , 2016 ) as well as in Caenorhabditis elegans ( Dove et al . , 2017 ) . Despite the apparent requirement for UBE2D1/2/3 or ARIH1 to initiate a poly-ubiquitin chain on human substrates , UBE2R1/2 are capable of both priming and poly-ubiquitin chain extension on milli-second time scales in in vitro ubiquitylation assays ( Kleiger et al . , 2009b; Pierce et al . , 2009; Saha and Deshaies , 2008 ) . The relative importance of these various mechanisms in human SCF-mediated poly-ubiquitin chain formation on substrates has yet to be addressed . In this study , we investigate the role of the different chain initiating and elongating enzymes implicated in human SCF function . We use single reaction monitoring mass spectrometry to estimate physiological E2 and E3 concentrations in cells , quench flow-based rapid kinetics to analyze the contributions of different enzymes to in vitro reconstituted reactions , and genetic analysis to uncover functional redundancies between different SCF-associated factors . These results uncover three main principles . First , some substrates and/or substrate receptors appear selective for a particular ubiquitin-modifying enzyme even when assayed at physiological concentrations , whereas others are more promiscuously ubiquitylated by different ubiquitin-modifying enzymes . Second , the rates of ubiquitylation can depend dramatically on enzyme concentrations . In particular , UBE2R2 activity is robust when levels are sufficient to saturate the SCF-substrate complex , but virtually undetectable when assayed at physiological levels . Finally , unlike in yeast , human SCF activity appears to be highly buffered such that the loss of any single SCF-associated E2 or E3 is compensated for by redundant factors . We uncover a redundant role for the E2 enzyme UBE2G1 that explains why loss of both UBE2R1 and UBE2R2 can be tolerated by cells . Collectively these results illuminate how the SCF regulatory system has been diversified through evolution and how this diversification has been exploited to allow differential substrate ubiquitylation . We sought to resolve how multiple ubiquitin priming and poly-ubiquitin elongating enzymes collaborate with human SCF ubiquitin ligases by measuring the individual rates of poly-ubiquitin chain priming and elongation using pre-steady state kinetic measurements with different combinations of SCF-associated enzymes . To accomplish this goal , we employed two well-characterized and biologically important SCF complexes based on the βTRCP substrate receptor ( SCFβTRCP ) and the FBW7 substrate receptor ( SCFFBW7; for more details , please see Appendix 1 ) . A critical parameter for measurement of the kinetics of in vitro reconstituted reactions is the concentrations for both ubiquitin chain initiator and elongator E2s and E3s needed to saturate the E3–substrate complex and thereby promote maximal rates of ubiquitin transfer to the substrate . Multi-turnover Michaelis-Menten kinetic assays were used to determine the levels of ARIH1 , UBE2D3 and UBE2R2 necessary to saturate either SCFβTRCP or SCFFBW7 complexes ( Figure 1 , Figure 1—figure supplement 1 , and Table 1 ) . Care was also taken to assess the relative activities of the recombinant enzyme preparations ( Ronchi and Haas , 2012 ) ( please see Appendix 2 for a description and possible impact on interpretation of results ) . A potential caveat with saturating kinetics is that E2 and E3 concentrations in cells may be insufficient to saturate the E3–substrate complex , resulting in slower rates of ubiquitin transfer to the substrate in vivo . To address this issue , single reaction monitoring ( SRM ) mass spectrometry was used to determine the copy numbers of ARIH1 , UBE2D1/2/3 , UBE2R1/2 , as well as the SCF subunits CUL1 and SKP1 in several common tissue culture cell lines , enabling calculation of the protein concentration within the cell ( Table 2 ) . To account for modest nuclear enrichment previously shown for UBE2R1 ( Kleiger et al . , 2009a ) , the activities of ARIH1 , UBE2D3 and UBE2R2 were assayed at twice the cellular concentration estimated by SRM . The rates of ARIH1-catalyzed ubiquitin chain initiation and elongation were obtained by measuring the pre-steady state kinetics of a single encounter ubiquitylation reaction containing SCFFBW7 , single lysine Cyclin E peptide , and 2 . 5 μM ARIH1 to saturate the SCF–substrate complex ( Figure 2a–c ) . The rate of chain initiation was 0 . 5 sec−1 , the second ubiquitin transfer to substrate was only modestly slower ( 0 . 2 sec−1 ) , and the rate of the third ubiquitin transfer was 0 . 08 sec−1 ( Table 3 ) . The kinetics were sufficient to generate products containing up to four ubiquitins on the substrate modified poly-ubiquitin chain prior to substrate dissociation from SCF ( Figures 2b and 3a ) . Reducing the concentration of ARIH1 some 7-fold lower to more physiological levels resulted in only modest changes to the rates of ubiquitin transfer . The rates of the first and the second ubiquitin transfers ( 0 . 2 sec−1 and 0 . 1 sec−1 , respectively ) were only 2-fold slower than in the case where ARIH1 was saturating for SCF , and the rates of the third ubiquitin transfers were comparable ( Table 3 ) . As such , these reactions at lower ARIH1 levels still generated products with up to four ubiquitins on SCF-bound substrates ( Figure 2d , Figure 2—figure supplement 1a , and Figure 3a ) . Assaying ARIH1 with β-Catenin peptide substrate and SCFβTRCP resulted in both similar kinetics as well as ubiquitylated products compared with the reactions with Cyclin E peptide and SCFFBW7 ( Figures 2e , f and 3b , Figure 2—figure supplement 1b , c and Table 3 ) . The rates of chain initiation were comparable when the ARIH1 concentration was 2 . 5 μM or 0 . 36 μM ( 0 . 2 sec−1 and 0 . 1 sec−1 , respectively ) , and the rates of the second ubiquitin transfers were also similar to those for Cyclin E . Taken together , these results demonstrated that ARIH1 rapidly initiates a poly-ubiquitin chain on SCF-bound substrates and is capable of modest chain extension prior to product dissociation , even at ARIH1 levels that reflected intracellular concentrations . The rates of chain initiation and elongation were measured next for UBE2D3 . At a saturating concentration of 10 μM , the rate of chain initiation was 0 . 1 sec−1 for SCFFBW7 and Cyclin E peptide ( Figure 2—figure supplement 2a , b and Table 3 ) , 5-fold slower than the same rate catalyzed by ARIH1 . Similar to ARIH1 , this rate was only reduced 2-fold when UBE2D3 levels were lowered to 3 . 7 μM to more accurately reflect cellular levels ( Figure 2—figure supplement 2c , d and Table 2 ) . However , when 10 μM UBE2D3 was assayed with β-Catenin peptide and SCFβTRCP , the rate of chain initiation ( 5 sec−1 ) was far more rapid than with Cyclin E peptide and SCFFBW7 ( Figure 2—figure supplement 3 and Table 3 ) . Reduction of the concentration of UBE2D3 by approximately 3-fold to physiological levels had a negligible effect on the rate of chain initiation . With chain elongation rates between 0 . 2 sec−1–0 . 5 sec−1 , UBE2D3 was also capable of substantial poly-ubiquitin chain assembly onto β-Catenin substrate , even at reduced levels ( Figure 3b ) . Thus , in contrast with ARIH1 , UBE2D3 activity is sensitive to either the identity of the substrate and/or the SCF substrate receptor . When saturating levels ( 10 μM ) of the ubiquitin-conjugating enzyme UBE2R2 was assayed with both SCFFBW7 and Cyclin E peptide or with SCFβTRCP with β-Catenin peptide , the rates of chain initiation were highly similar ( 0 . 1 sec−1–0 . 2 sec−1 , respectively ) and comparable to the rates generated by ARIH1 ( Figure 2—figure supplement 4 and Table 3 ) . Consistent with prior results , the rate of the second ubiquitin transfer was far more rapid for both substrates and SCF complexes ( 40 sec−1 and 30 sec−1 to mono-ubiquitylated Cyclin E and β-Catenin peptides , respectively ) . The rates of the third ubiquitin transfers to SCF-bound substrates were also rapid , resulting in very long poly-ubiquitin chains on products ( Figure 3 ) , a hallmark of UBE2R1/2-catalyzed ubiquitylation reactions . While reactions containing saturating UBE2R2 for SCF added impressively long poly-ubiquitin chains onto substrate , reduction of the UBE2R2 levels by 20-fold to mimic cellular conditions resulted in the near elimination of any product for both Cyclin E and β-Catenin peptide ( Figure 2—figure supplement 5 and Table 3 ) . Indeed , product levels were so low that we were unable to estimate the rate of poly-ubiquitin chain initiation for either substrate . This result strongly suggested that UBE2R1/2 are incapable of mediating substrate ubiquitylation in vivo without the assistance of other SCF-associated enzymes . We then compared the rates of chain initiation and elongation for reactions containing initiator E2 or E3 in combination with UBE2R2 . Ubiquitylation reactions containing both saturating ARIH1 ( 2 . 5 μM ) and UBE2R2 ( 10 μM ) for SCFFBW7 were performed in the presence of Cyclin E peptide ( Figure 2—figure supplement 6a , b ) . The rate of chain initiation was only 2-fold slower than in comparison with the same reaction containing only ARIH1 ( Table 3 ) . Reduction of both ARIH1 ( 0 . 36 μM ) and UBE2R2 ( 0 . 5 μM ) to more physiological levels did not affect the rate of chain initiation , but the rate of chain elongation was greatly reduced when compared to the reaction containing saturating UBE2R2 alone ( Figure 2—figure supplement 6c , d and Table 3 ) . Similar trends in the rates as well as the effects on poly-ubiquitin chain lengths were observed for β-Catenin peptide and SCFβTRCP ( Figure 2—figure supplement 7 ) . Nevertheless , the rates of chain elongation for ARIH1 in the presence of UBE2R2 were significantly faster than for reactions containing ARIH1 alone , resulting in substrates modified with poly-ubiquitin chains that were substantially longer than those formed by ARIH1 in the absence of UBE2R2 ( Figure 3b ) . These results suggested that while ARIH1 can provide the initiator ubiquitin for subsequent elongation by UBE2R1/2 , it also acts as a competitor for the elongation cycles in in vitro ubiquitylated reactions . The rates of chain initiation and elongation were then determined for analogous reactions containing both 10 μM UBE2D3 and UBE2R2 ( Figure 2—figure supplements 8 and 9; Table 3 ) . While the rates of ubiquitin chain initiation were comparable to those from reactions containing UBE2D3 alone , the rates of the second ubiquitin transfer to substrate were suppressed compared to reactions containing UBE2R2 alone , in particular for reactions with β-Catenin peptide and SCFβTRCP ( 0 . 9 sec−1 versus 30 sec−1 , respectively ) . Reduction of the levels of both UBE2D3 ( 3 . 7 μM ) and UBE2R2 ( 0 . 5 μM ) further slowed the rates of chain elongation; however , long poly-ubiquitin chains were still observed on substrates , especially β-Catenin peptide ( Figure 3 ) . These results suggested that , like ARIH1 , UBE2D3 also acts as a competitor for chain elongation by UBE2R1/2 . Considering that the average estimated physiological concentrations for ARIH1 , UBE2D3 , and UBE2R1/2 ( Table 2 ) were all quite close to the Km values of these enzymes for SCF ( Table 1 ) , we reasoned that relatively subtle changes in ARIH1 , UBE2D3 , and UBE2R1/2 levels may result in substantial differences in both the fraction of substrate converted into ubiquitylated product as well as the lengths of the poly-ubiquitin chains . To test for this , single encounter ubiquitylation reactions were assembled with SCFFBW7 , Cyclin E peptide , and where ARIH1 , UBE2D3 , and UBE2R2 levels were assayed either up to four times higher or lower than their estimated physiological concentrations . Very little product was observed for any of the reactions when the enzyme levels were assayed at either one-fourth or one-half the estimated cellular concentrations ( Figure 4 ) . Substantial mono-ubiquitylation of Cyclin E peptide occurred for both ARIH1 and UBE2D3 when assayed at the estimated physiological concentrations , and introduction of UBE2R2 to these reactions also resulted in the formation of poly-ubiquitin chains onto product containing four or more ubiquitins ( Figure 4 and Figure 4—figure supplement 1 ) . Increasing ARIH1 levels to either twice or four times the estimated cellular concentration resulted in both increased conversion of substrate to product containing at least one ubiquitin , as well as substantial poly-ubiquitin chains on product ( up to five ubiquitins in the latter case ) . Increasing UBE2D3 levels also resulted in greater product formation , although to a lesser extent than for ARIH1 . The UBE2D3 product was also primarily mono-ubiquitylated . Increasing the UBE2R2 concentration also led to some product formation , all containing very long poly-ubiquitin chains . Finally , combining either ARIH1 or UBE2D3 with UBE2R2 and assaying at levels two or four times greater than the estimated physiological levels resulted in greater amounts of product containing much longer poly-ubiquitin chains . Thus , subtle changes in the concentrations of ARIH1 , UBE2D3 , and UBE2R2 led to significant changes in both the amount of substrate converted to product as well as the lengths of the poly-ubiquitin chains on the product . Since modest changes in enzyme levels , particularly UBE2R2 , resulted in substantial changes in poly-ubiquitin chains appended to substrate , we reasoned that a 2-fold reduction in UBE2R1/2 levels might result in the stabilization of SCF substrates in the cell . Surprisingly , the stabilities of the SCFFBW7 substrates p27 and CYCLIN E were comparable by cycloheximide chase in either wild-type ( WT ) or UBE2R1 knock out HEK293T Flp-In T-Rex ( 293T-FiTx cells; Figure 5—figure supplement 1a ) . Note that ablation of UBE2R1 protein did not significantly affect UBE2R2 mRNA levels , ruling out a dosage compensation effect ( Figure 5—figure supplement 1b ) . Since the above results may be reconciled by the presence of UBE2R2 protein that might compensate for loss of UBE2R1 protein , CRISPR/Cas9 technology was used to disrupt both UBE2R1 and UBE2R2 loci in HEK 293T cells . Multiple clones were isolated that contained frameshift indels that caused total ablation of UBE2R1 and/or UBE2R2 protein , as demonstrated by immunoblotting with antibodies with high specificity to either UBE2R1 or UBE2R2 ( Figure 5—figure supplements 2 and 6b ) . Remarkably , the steady state levels of three SCF substrates , β-CATENIN , CYCLIN E , and p27 , were not significantly enriched in three independent UBE2R1/2 knockout cell lines in comparison with three control lines ( Figure 5a , b ) . Cycloheximide chase analysis performed on the UBE2R1/2 double knockout cells demonstrated that p27 protein degradation was indistinguishable in comparison with control cells ( Figure 5c and Figure 5—figure supplement 3a ) . The degradation of IκBα in response to TNFα treatment in control or UBE2R1/2 double knockout cells was also indistinguishable ( Figure 5d and Figure 5—figure supplement 3b ) . Finally , since Cdc34 is required for the G1-S cell cycle transition in budding yeast , flow cytometry was used to estimate the percentage of cells in each phase of the cell cycle in control versus UBE2R1/2 double knockout cells but no differences were found ( Figure 5e and Figure 5—figure supplement 4 ) . These results demonstrated that both UBE2R1 and UBE2R2 are entirely dispensable for the degradation of at least some SCF substrates and progression through the cell cycle . To explain how p27 , CYCLIN E , β-CATENIN , and IκBα become ubiquitylated in the absence of UBE2R1/2 , we considered two potential hypotheses: ( 1 ) ARIH1 and/or UBE2D1/2/3 are sufficient for both initiation as well as modest poly-ubiquitin chain elongation ( as evidenced by the kinetic results here ) ; and ( 2 ) a different E2 may complement UBE2R1/2 activity in the cell . The first hypothesis predicts that ARIH1 and/or UBE2D1/2/3 might become more essential in the absence of UBE2R1/2 , whereas the second hypothesis suggests that one or more other E2s may become essential in the absence of UBE2R1/2 . To address this question in an unbiased manner , we performed genome-wide CRISPR knockout screens in the NALM-6 pre-B cell lymphoma line with the previously reported EKO sgRNA library ( Bertomeu et al . , 2018 ) to identify genes that exhibit synthetic lethality with the loss of UBE2R1 and/or UBE2R2 . Cell populations were first transduced with individual sgRNAs targeting either UBE2R1 , UBE2R2 , or both UBE2R1 and UBE2R2 , as well as the AAVS1 locus and a non-targeting control sgRNA ( Figure 6a and Figure 6—figure supplement 1; for screen details , see Materials and methods ) . Each population was then transduced with the EKO library pool and propagated for 14 days , followed by determination of sgRNA frequencies by next generation sequencing and calculation of gene-level scores by the RANKS algorithm ( Bertomeu et al . , 2018 ) . Differential RANKS scores for every gene in each experimental screen were obtained by subtraction of the averaged RANKS scores of the two control screens . Strikingly , the UBE2G1 gene scored as the top synthetic lethal interactor in both replicates of the UBE2R1/2 double knockout screen but did not score in either of the UBE2R1 or UBE2R2 single knockout screens ( Figure 6b–e ) . Moreover , out of the five experimental screens analyzed , the only statistically significant hit ( FDR < 0 . 05 ) was UBE2G1 in the UBE2R1/2 double knockout screen . The three-way genetic interaction between UBE2R1 , UBE2R2 and UBE2G1 was validated by population-level knockout of UBE2G1 in UBE2R1/2 single and double knockout clones in the NALM-6 parental cell line ( Figure 6—figure supplement 2 ) . We did not observe any other significant genetic interactions with loss of UBE2R1/2 , including with ARIH1 , UBE2D3 or other SCF components . These genetic screen data suggested that UBE2G1 uniquely buffers the loss of UBE2R1/2 , and in particular that the elongation activity of ARIH1 and UBE2D3 are unable to substitute for the three dedicated elongation E2 enzymes . To determine whether depletion of UBE2G1 may affect the stability of SCF substrates , UBE2R1/2 double knockout 293T cell lines were treated with siRNAs targeting UBE2G1 . p27 levels were modestly increased in populations treated with non-targeting siRNAs but were strongly increased in UBE2R1/2 double knockout cells treated with UBE2G1-targeting siRNA ( Figure 7a , b ) . Stabilization of CYCLIN E protein was also observed in UBE2R1/2 double knockout cells treated with UBE2G1-targeting siRNA . To determine whether these observations might extend to additional CRLs , we assessed the CRL2VHL substrate HIF1α in control versus UBE2R1/2 knockout cell lines . Similar to p27 , HIF1α was heavily stabilized upon knockdown of UBE2G1 in UBE2R1/2 double knockout cells ( Figure 7a , b ) . These results demonstrate the functional redundancy between UBE2R1/2 and UBE2G1 across the CRL family . We then determined whether UBE2G1 can support SCF-mediated substrate ubiquitylation in vitro . UBE2G1 itself had no measurable chain initiation activity , but exhibited substantial chain elongation activity against either mono-ubiquitylated β-Catenin or Cyclin E peptides in the presence of SCFFBW7 or SCFβTRCP , respectively ( Figure 7—figure supplement 1a ) . The Km of UBE2G1 was estimated as 1 . 30 ± 0 . 2 μM from multi-turnover Michaelis-Menten kinetics in the presence of mono-ubiquitylated Cyclin E peptide substrate ( Figure 7—figure supplement 1b , c and Table 1 ) . We then tested the ability of UBE2G1 to participate in a hand-off reaction with either ARIH1 or UBE2D3 . Maximal chain elongation on substrate was observed only when ARIH1 or UBE2D3 were assayed in the presence of UBE2G1 ( Figure 7c , d ) . Finally , the pre-steady state kinetics of UBE2G1-catalyzed ubiquitin transfer to the Cyclin E peptide yielded rates of ubiquitin transfer of 1 ± 0 . 1 sec−1 under conditions where UBE2G1 levels were sufficient to saturate SCF ( Figure 7e , Figure 7—figure supplement 1d , and Table 3 ) . In terms of enzyme efficiency ( kobs/Km ) , the UBE2R2 efficiency ( 1 . 3 × 108 M−1 sec−1 ) was 173-fold greater than for UBE2G1 ( 7 . 7 × 105 M−1 sec−1 ) for the first ubiquitin transfer to mono-ubiquitylated substrate , and 17-fold greater for the subsequent transfer . Thus , while UBE2R2 is far more efficient during catalysis of ubiquitin transfer to substrate on SCF , UBE2G1 activity is nevertheless sufficient to poly-ubiquitylate substrates that have been primed by either UBE2D1/2/3 or ARIH1 . Collectively , these results demonstrated that UBE2G1 can act as a dedicated elongation E2 for SCF complexes , and thereby buffer cells against the loss of UBE2R1 and UBE2R2 . Our studies using in vitro enzyme kinetics and genetic approaches have uncovered new complexities in the mechanism of poly-ubiquitin chain initiation and elongation by human SCF ubiquitin ligases . First , comparison of the ARIH1- or UBE2D3-catalyzed ubiquitin chain initiation demonstrates that the rates are not substantially affected by enzyme concentrations that range from physiological to saturating levels . Second , while the rates of ubiquitin chain initiation by UBE2R2 when saturating for SCF are comparable to those of ARIH1 and UBE2D3 , at physiological levels UBE2R2 clearly cannot support ubiquitin chain initiation or ubiquitylated product formation . Third , the rate of chain initiation by UBE2D3 for β-Catenin peptide and SCFβTRCP is some 50-fold faster than for reactions containing SCFFBW7 and Cyclin E peptide , indicating that UBE2D3 has a strong preference for some substrates and/or substrate receptors . Finally , UBE2R1/2 function is genetically buffered by UBE2G1 , such that loss of UBE2R1/2 does not cause overt proliferation defects in cell line models , whereas the simultaneous loss of all three E2s leads to inviability . Consistently , UBE2G1 supports chain elongation activity against primed mono-ubiquitylated substrates in vitro . Our results suggest that human SCF-catalyzed substrate ubiquitylation has evolved additional regulatory layers compared to the simpler yeast SCF system ( Figure 8 ) . In S . cerevisiae , Cdc34 appears both necessary and sufficient for SCF function since Ubc4/5 ( the UBE2D1/2/3 orthologs ) have no in vitro activity with reconstituted yeast SCF and cannot compensate for the loss of CDC34 in cells . While it is surprising that Cdc34 homologs UBE2R1 and UBE2R2 appear to be dispensable in human cell lines , this difference may in part be explained by the respective concentrations of Cdc34 and UBE2R1/2 in yeast versus human cells . The concentration of Cdc34 in the yeast nucleus has been estimated at approximately 10 μM and 2- to 3-fold lower in the cytoplasm ( Kleiger et al . , 2009a ) . Based on the kinetic results here , these levels would be sufficient for robust chain initiation and elongation . In contrast , UBE2R1/2 protein levels in several human cell lines are at least an order of magnitude lower than in yeast , and well-below the low micromolar value of the Km of UBE2R2 for SCF bound to an unmodified substrate ( Table 1 ) . On the other hand , the Km of UBE2R2 for SCF bound to a mono-ubiquitylated substrate is approximately 0 . 3 μM , similar to the physiological concentration of UBE2R1/2 in tissue culture cells , thus explaining why UBE2R1/2 is still able to promote chain elongation on primed substrate in vitro . A possible exception to this interpretation may be our observation that a fraction of UBE2R1 is concentrated into foci in which protein levels may be higher ( Figure 5—figure supplement 5 ) , and thus potentially capable of initiating localized ubiquitylation reactions with CRL ligases . Multiple E3s contain sequence motifs that may drive the formation of membraneless organelles such as nuclear speckles and Cajal bodies ( Hughes et al . , 2018 ) . In particular , the CUL3 substrate receptor SPOP forms such structures in cells in the presence of its substrates ( Bouchard et al . , 2018 ) . Additional experiments are necessary to determine whether SCF or other CRLs and their substrates are co-located within UBE2R1 foci , and whether UBE2R2 or UBE2G1 also form foci . Our genome-wide genetic screen suggests that CRL function is more buffered in human cells compared to yeast . Aside from UBE2G1 , at least at the level of resolution of our screen , we detected no other significant genetic interactions with loss of UBE2R1/2 , including ARIH1 and UBE2D3 . This result implies that the elongation activity of ARIH1 and UBE2D3 are insufficient to compensate for the combined loss of UBE2R1/2 and UBE2G1 . In contrast , conditional alleles of CDC34 exhibit many dozens of well-documented synthetic lethal interactions in yeast ( Oughtred et al . , 2019 ) . We note that while the UBE2R1/2 double mutant is viable in two different transformed cell lines , UBE2R1/2 function may still be essential or at least important in other cell lines . For instance , the reduction of UBE2R1 protein levels in U2OS cells resulted in partial stabilization of IKBα ( Wu et al . , 2010 ) and UBE2R1/2 function may still be required in an organismal context , an issue that remains to be addressed . Why did SCF-catalyzed substrate ubiquitylation evolve separate chain initiating versus elongating enzymes , and why are multiple initiators and elongating E2s necessary ? One likely reason is that this separation of function affords additional opportunities to diversify the control of SCF ligase activity . For instance , since ARIH1 is active only in the presence of neddylated CRL complexes , and SCF complexes tend to be activated only when bound to substrate ( Emberley et al . , 2012; Enchev et al . , 2012; Fischer et al . , 2011; Pierce et al . , 2013 ) , ARIH1 activity is held in check unless it is in the presence of activated SCFs . At least in vitro , saturating levels of UBE2R1/2 have substantial activity with un-neddylated SCF , such that the ARIH1-catalyzed hand-off mechanism may prevent unwanted auto-ubiquitylation of E3s not bound to substrate . A further layer of control appears to lie at the poly-ubiquitin chain initiation step , based on E2 specificity for particular substrates and/or substrate receptors . Thus , while UBE2D3 was less efficient than ARIH1 at chain initiation onto Cyclin E peptide substrate , the rate of the first ubiquitin transfer to β-Catenin peptide ( 5 sec−1 ) was 50-fold greater than with Cyclin E peptide , and 25-fold faster than ARIH1 for the same reaction . This effect may be explained by the affinity of the E2-E3 interaction and competition between different initiating enzymes . For instance , we observed that the Km of UBE2D3 for SCFFBW7 is significantly higher than for SCFβTRCP . Similarly , ARIH1 but not UBE2D3 supports in vitro ubiquitylation of CRY1 bound to SCFFbxL3 ( Scott et al . , 2016 ) . The use of multiple E2s and initiating enzymes evidently allows more elaborate specificity across the SCF substrate repertoire . The slower ubiquitin transfer rate of UBE2G1 compared to UBE2R1/2 , and the weak elongation activity of ARIH1 and UBE2D3 may have biological relevance . For instance , delayed substrate ubiquitylation and degradation is important during mitotic exit in yeast ( Rape et al . , 2006 ) and in addition the proteasome is capable of recognizing substrates modified by different chain lengths and/or on different lysine residues ( Lu et al . , 2015 ) . Whether such effects are important for UBE2G1-mediated degradation of particular CRL substrates remains to be determined . Finally , the combinatorial plasticity of SCF chain initiation and elongation reactions may have important practical considerations in drug discovery . Recently , bivalent small molecule ligands ( variously referred to as PROTACS , IMiDs , or SNIPERs ) have been developed to target non-cognate substrates to E3 enzymes , and thereby eliminate proteins that contribute to tumorigenesis and other disease states ( Paiva and Crews , 2019 ) . This promising E3-based therapeutic approach , termed event-driven pharmacology ( Lai and Crews , 2017 ) , will be enabled by a more precise understanding of E3 catalytic mechanisms ( Scudellari , 2019 ) . To date , bivalent or ‘molecular glue’ type ligands have been developed for several CRL complexes including scaffolds that direct neo-substrates to the CUL4CRBN , CUL4DCAF15 , CUL2VHL and CUL3KEAP1 enzymes ( Paiva and Crews , 2019 ) . It seems likely that particular combinations of chain initiating and elongating enzymes will preferentially modify neo-substrates in a context-specific fashion . For example , UBE2G1 has recently been shown to mediate chain elongation of neomorphic substrates that are bridged to CRL4CRBN by thalidomide analogs ( Lu et al . , 2018; Patil et al . , 2019; Sievers et al . , 2018 ) . Many other details remain to be elucidated regarding how SCF and other CRLs function with a suite of ubiquitin-modifying enzymes . Do different combinations of ARIH1 or UBE2D3 with UBE2R1/2 or UBE2G1 produce different products that alter the kinetics of substrate degradation ? What is the structural basis that underpins initiation versus elongation activities ? How do the localized dynamics of E2 , ARIH1 and SCF interactions in vivo relate to function ? Are there yet additional E2s or E3s that can also function with CRLs under different circumstances ? Answers to these questions will further our understanding of this important and fascinating enzyme system . HEK293T Flp-In T-Rex ( 293T-FiTx ) HEK293T/17 , HeLa , and MRC5 cells were all grown in DMEM ( 4 . 5 g/L glucose ) supplemented with 10% fetal bovine serum , 4 mM L-Glutamine , 100 units/mL Penicillin , 100 μg/mL Streptomycin , and 10 μg/mL Ciprofloxacin in a standard tissue culture incubator with 5% carbon dioxide . NALM-6 cells were grown in RPMI 1640 medium supplemented with 10% fetal bovine serum . The NALM-6 cell line was provided by Stephen Elledge ( Harvard Medical School ) . Authenticity was determined by whole genome sequence analysis of the parental line and the doxycycline inducible Cas9 clone used for CRISPR screens . Both cell lines matched the previously reported NALM-6 sequence and chromosome complement . Both lines were verified as mycoplasma negative before experiments were initiated . All other cell lines were purchased directly from the ATCC ( see Supplementary file 1 - key resources table ) prior to beginning these experiments , and were grown in the presence of Ciprofloxacin to eliminate potential mycoplasma contamination . All expression constructs used in this study have been listed in the key resources table ( Supplementary file 1 ) . Expression constructs that had been generated specifically for this study include human ( His ) 8-NEDD8 , human ( His ) 6-ubiquitin , ( His ) 6-no lysine ( K0 ) human ubiquitin , and UBE2G1 . Detailed protocols for the expression and purification of split-n-co-express human K720R and neddylated CUL1-RBX1 ( Li et al . , 2005 ) , full-length human βTRCP2 ( FBW11 ) -SKP1 complex ( Scott et al . , 2016 ) , human FBW7 ( 263-C-terminus ) -SKP1 complex ( Scott et al . , 2016 ) , human UBE2L3 and ARIH1 ( Scott et al . , 2016 ) , UBE2R2 ( Hill et al . , 2018 ) , UBE2D3 ( Hill et al . , 2018 ) , human APPBP1-UBA3 , human UBC12 ( Huang and Schulman , 2005 ) , human UBE1 ( E1 ) ( Scott et al . , 2016 ) , and mono-ubiquitylated β-Catenin peptide ( Hill et al . , 2018 ) have been described in detail elsewhere . The ( His ) 8-NEDD8 construct was synthesized by Integrated DNA Technologies and cloned into pET-11b . The vector was transformed into Rosetta ( DE3 ) cells and cultured in LB medium containing 10 g/L dextrose at 37° C . Bacterial cells were collected by centrifugation at 5000 x g for 10 min and the media exchanged for LB without dextrose just prior to induction of protein expression with 0 . 4 mM IPTG at 24° C for 4 hr . Cells were harvested by centrifugation at 5000 x g for 10 min , and the cell pellets washed in 1x PBS prior to drop-freezing in liquid nitrogen and storage at −80° C . The cells were lysed by sonication in nickel wash buffer containing 30 mM Tris-HCl ( pH 7 . 5 ) , 250 mM NaCl , 20 mM imidazole , 1 mM β-mercaptoethanol , 5% glycerol , and 1% Triton X-100 . Lysates were centrifuged at 21 , 000 x g for 1 hr and incubated with Ni-NTA agarose resin for 1 hr with gentle rotation . The resin was washed several times with standard nickel wash buffer , followed by elution in buffer containing 300 mM imidazole , 200 mM NaCl , and 50 mM Na-HEPES ( pH 8 . 0 ) . The eluate was then spin filtered and immediately loaded onto a Superdex 75 gel filtration column that had been pre-equilibrated in storage buffer containing 30 mM Tris-HCl ( pH 7 . 5 ) , 100 mM NaCl , and 10% glycerol . Note that the eluate was not concentrated prior to gel filtration to avoid protein precipitation . All fractions containing NEDD8 protein were combined , concentrated to approximately 200 μM , and sub-aliquoted into single-use volumes before drop-freezing in liquid nitrogen and long-term storage at −80° C . Typical protein yield was 0 . 5 mg per 1L of cell culture . The calculated extinction coefficient was 2 , 560 M−1cm−1 . The ( His ) 6-ubiquitin construct was synthesized by IDT and cloned into pET-11b . Following transformation of the plasmid into the Rosetta ( DE3 ) bacterial strain , cells were cultured in LB medium containing 10 g/L dextrose at 37° C , followed by centrifugation at 5000 x g for 10 min to exchange the media for LB lacking dextrose . Protein expression was then induced by adding 0 . 4 mM IPTG followed by incubation at 30° C for 4 hr . Cells were harvested as described above for NEDD8 protein . To initiate protein purification , the cell pellets were first lysed in nickel wash buffer containing 30 mM Tris-HCl ( pH 7 . 5 ) , 250 mM NaCl , 20 mM imidazole , 5% glycerol , and 0 . 1% IGEPAL . Lysates were centrifuged at 21 , 000 x g for 1 hr , the precipitates discarded , and the lysates incubated with Ni-NTA agarose resin for 1 hr and gentle rotation . The resin was washed several times with standard nickel wash buffer , followed by elution in buffer containing 300 mM imidazole , 200 mM NaCl , and 50 mM Na-HEPES ( pH 8 . 0 ) . The eluate was then concentrated , spin filtered , and subjected to gel filtration on a Superdex 75 column buffered in storage buffer ( see above ) . Typical protein yield was 6 mg per 1L of cultured cells . The calculated extinction coefficient was 1 , 280 M−1cm−1 . A plasmid for TEV protease expression was purchased from Addgene ( Plasmid #8827 ) . Bacterial cells were grown as described ( Tropea et al . , 2009 ) . The cell pellets were lysed by sonication in buffer containing 50 mM sodium phosphate ( pH 8 . 0 ) , 200 mM NaCl , 10% glycerol , and 25 mM imidazole . Lysates were centrifuged at 21 , 000 x g for 1 hr and incubated on Ni-NTA agarose resin for 1 hr with gentle rotation . The resin was washed several times with lysis buffer , then eluted in a buffer containing 30 mM Tris-HCl ( pH 8 . 0 ) , 300 mM imidazole , 200 mM NaCl , and 10% glycerol . After elution , 1 mM EDTA and 5 mM DTT were added to the eluate , which was then concentrated , spin filtered , and subjected to gel filtration on a Superdex 75 column that had been pre-equilibrated in storage buffer . Typical protein yield was 0 . 5 mg per 1L of lysed cells . The calculated extinction coefficient was 32 , 290 M−1cm−1 . The ( His ) 6-K0-ubiquitin construct was synthesized by IDT with the amino acid sequence for human ubiquitin . It was cloned into pET-11b , and expressed and purified in an identical manner as ( His ) 6-ubiquitin ( see above ) . Typical protein yield was 3 mg per 1L of cell culture . The calculated extinction coefficient was 1 , 280 M−1cm−1 . An expression construct for UBE2G1 was first obtained from Addgene ( Plasmid #15790 ) , and then sub-cloned into pGEX-4T1 including a TEV protease cleavage site between GST and the UBE2G1 N-terminus . Expression and purification were carried out identically to that of UBE2D3 ( as above ) . Typical protein yield was 1 mg per 1L of cell culture . The calculated extinction coefficient was 29 , 500 M−1cm−1 . Mono-ubiquitylated Cyclin E peptide substrate was generated by incubating 130 μM ( His ) 6-ubiquitin , 0 . 25 μM human E1 , 2 μM UBE2D3 , 600 nM K720R CUL1-RBX1 , 600 nM FBW7 ( 263-C-terminus ) -Skp1 complex , and 100 μM cyclin E peptide overnight at 20° C . The reaction was quenched the following morning by addition of DTT to a final concentration of 10 mM DTT , and product was isolated by three successive rounds of gel filtration on a Superdex 75 column that had been pre-equilibrated in storage buffer . The yield of purified mono-ubiquitylated Cyclin E peptide was approximately 5% of the starting peptide . The calculated extinction coefficient was 2 , 560 M−1cm−1 . Reactions were setup by first adding E1 to a mixture already containing reaction buffer ( 30 mM Tris-HCl ( pH 7 . 5 ) , 100 mM NaCl , 5 mM MgCl2 , 2 mM DTT and 2 mM ATP ) and ubiquitin and incubating for 1 min ( see Supplementary file 2 for concentrations ) . The mixture was then divided into ten individual Eppendorf tubes , followed by the addition of a 2-fold dilution series of ARIH1 ( UBE2L3 was kept constant for all reactions in the titration series and was always in excess of ARIH1 ) , UBE2D3 , Ube2R2 , or UBE2G1 . After 2 min of incubation , SCFβTRCP or SCFFBW7 complexes were added , centrifuged briefly , and then initiated by addition of 32P-labeled β-catenin or Cyclin E peptide substrate ( unmodified or mono-ubiquitylated ) . Reactions were quenched in 2x SDS-PAGE buffer ( 100 mM Tris-HCl ( pH 6 . 8 ) , 20% glycerol , 30 mM EDTA , 4% SDS , and 4% beta-mercaptoethanol ) , ensuring that reactions containing the highest concentration of E2 or ARIH1 had converted no more than 20% of substrate into product . Each reaction was performed in duplicate , and time points were resolved on 18% polyacrylamide SDS-PAGE gels . Autoradiography was performed using a Typhoon 9410 Imager and Image Quant software ( GE Healthcare ) . Percent conversion was determined by dividing the total product signal ( e . g . any species that migrated slower than substrate ) by the entire lane signal , and then estimating the velocity by first normalizing for substrate and enzyme concentrations ( multiplying by substrate concentration and dividing by the SCF concentration ) and then dividing by the time of incubation . The data were fit to the Michaelis-Menten equation , velocity = kcat[S] ( [KM + [S] ) , where [S] represents the substrate concentration and KM is the Michaelis constant , using nonlinear curve fitting ( Prism 8 software ) . HEK 293T/17 , 293T-FiTx , HeLa or MRC5 cells were grown in isotopically heavy ( R6 and K8 ) SILAC medium for a minimum period of 10 cell doublings . Cells were harvested from 10 cm dishes , resuspended in 1 ml PBS , filtered through 40 µm mesh and counted using a CEDEX HiRES automated cell counter ( Roche ) , which determines average cell number and cell diameter from 20 technical replicates . Cells were lysed in 500 µl of lysis buffer ( 8 M Urea , 100 mM ammonium bicarbonate , 5 mM TCEP ) that was spiked with a master mix of purified recombinant proteins ( ARIH1 , UBE2D3 , UBE2R1 , UBE2R2 , CUL1 and SKP1; for concentrations and details see Table 2—source data 1 ) . Lysates were then sonicated for 30 s ( 1 s on/off ) at 30% maximum efficiency on a Branson Sonifier and subsequently incubated at 23°C for 20 min . Lysates were then clarified by centrifugation at 26 , 000 x g for 10 min and the protein concentration was determined via A280 on a nanodrop instrument . Subsequently , 100 µg of total protein were alkylated via iodoacetamide followed by digest for 4 hr at 1 µg/30 µg LysC and o/n at 1 µg/30 µg trypsin at 23C and 550 rpm in a temperature-controlled shaker . The digested samples were acidified with 50% formic acid ( final concentration 5% ) , diluted 1:1 in 0 . 2% formic acid and then desalted using C18 cartridges ( SPE 50 mg/mL C18 Hypersep column Thermo #60108–390 ) . C18 eluates were lyophilized and resuspended in 2% Acetonitrile and 0 . 2% formic acid Buffer A and approximately 300 ng of total peptide were subjected to SRM MS analysis on a QTRAP 6500 ( SCIEX ) under conditions previously described in Reitsma et al . ( 2017 ) . Data was analyzed using Skyline ( MacLean et al . , 2010 ) , RStudio , and Excel . The obtained heavy to light ratios were used to calculate estimates of cellular concentrations ( for details , see Table 2—source data 1 ) . Reactions were assembled in two separate mixtures: an E1/E2 mix that contained excess unlabeled peptide ( tube 1 ) and an SCF-32P-labeled substrate mix ( tube 2; see Supplementary file 2 for concentrations ) . Following addition of E2 and/or ARIH1 to tube 1 already containing reaction buffer ( 30 mM Tris-HCl ( pH 7 . 5 ) , 100 mM NaCl , 5 mM MgCl2 , 2 mM DTT and 2 mM ATP ) and ubiquitin , each mix was incubated for at least 8 min while being loaded into separate sample loops on a KinTek RQF-3 quench flow instrument . Reactions were initiated by bringing the two mixes together in drive buffer ( 30 mM Tris-HCl ( pH 7 . 5 ) , 100 mM NaCl ) , and then quenched at various time points in reducing 2x SDS-PAGE loading buffer ( 100 mM Tris-HCl ( pH 6 . 8 ) , 20% glycerol , 30 mM EDTA , 4% SDS , and 4% beta-mercaptoethanol ) . Each reaction was performed at least in duplicate , and time points were resolved on 18% polyacrylamide SDS-PAGE gels . Autoradiography was performed using a Typhoon 9410 Imager and Image Quant software ( GE Healthcare ) . Each product species was quantified as a fraction of the total signal of its respective lane . The rates of ubiquitin transfer were determined by fitting to analytical closed-form solutions ( Pierce et al . , 2009 ) using Mathematica . Reactions were assembled identically to those for the pre-steady state reactions ( see Supplementary file 2 for the concentrations of the reaction constituents ) . After an 8 min incubation , the E1/E2 mix was pipetted into the SCF/substrate mix initiating the reaction , which was then briefly vortexed and quenched in reducing 2x loading SDS-PAGE buffer after 10 s . The reactions were performed in duplicate , and resolved and quantified in the same manner as the quench flow reactions above . WT 293T-FiTx cells were seeded onto 6-well tissue culture plates to 80% confluency , and then transfected with 0 . 5 μg of a pX330 vector containing a small guide sequence targeting exon 2 of UBE2R1 as well as single stranded DNA oligos ( Supplementary file 1 ) using Lipofectamine 3000 . The DNA oligo contained homology arms adjacent to the predicted Cas9 cut site as well as a 50 base pair insert that disrupts protein translation by the introduction of 4 stop consecutive stop codons into the UBE2R1 reading frame . After 48 hr , detached cells were removed by washing with DPBS , followed by treatment of the attached cells with trypsin solution . Detached cells were diluted in growth media , counted and diluted to obtain single cells in each well of a 96-well plate . After 1 week , each well was observed under a microscope to verify single colony formation . After 2 weeks , colonies were exposed to trypsin , and half of the cells were used to seed wells in a 48-well plate , whereas the other half were lysed using QuickExtract solution . Genomic DNA was used in PCR reactions with primers adjacent to the PAM site ( Supplementary file 1 ) that were analyzed by standard agarose gel electrophoresis and product visualization by staining with ethidium bromide . Incorporation of the oligo into the genome resulted in an increase of PCR product sizes by 50 base pairs ( Figure 5—figure supplement 6a ) . Additional PCR primers were generated that also contained restriction sites which allowed for cloning of the PCR product into pGex-4T1 vector and DNA sequencing to confirm correct incorporation of the oligo sequence into the genomic one . Disruption of UBE2R1 protein expression was also verified by immunofluorescence microscopy ( Figure 5—figure supplement 5 ) . WT or Ube2R1 knockout 293T-FiTx cells were seeded onto 6-well tissue culture plates with approximately 0 . 5 × 106 cells per well . Within 24 hr , the growth media was removed , the cells were washed once with DPBS , and fresh growth media containing 100 μg/mL cycloheximide was added . At each time-point , cells were washed once with DPBS , followed by preparation of lysate by the introduction of 50–100 μL RIPA buffer and extraction of the cells from the plate . The lysates were briefly sonicated followed by centrifugation . A small sample of lysate was withdrawn for estimation of the amount of protein in each sample by the BCA assay ( Thermo Fisher Scientific ) , followed by the addition of an equal volume of 2x SDS-PAGE loading buffer ( 100 mM Tris-HCl ( pH 6 . 8 ) , 20% glycerol , 30 mM EDTA , 4% SDS , and 4% beta-mercaptoethanol ) . Equal amounts of lysate were loaded onto 4–20% Tris-Glycine SDS-PAGE gels and subject to electrophoresis . Denatured proteins were transferred to nitrocellulose membranes , followed by blocking in TBS-T containing 10% non-fat milk and overnight incubation with primary antibody at 4° C and light agitation ( a 1:2000 dilution was used for anti-CYCLIN E antibody , and 1:1000 for anti-p27 antibody ) . Immunoblots were washed three times with TBS-T , followed by incubation with the appropriate secondary antibodies ( 1:3000 dilution ) . After three additional washes with TBS-T , the immunoblots were exposed to enhanced chemiluminescence reagent . Detection was performed by exposure of the immunoblot to x-ray film . Control proteins were detected with anti-α-TUBULIN ( 1:2 , 000 ) and anti-UBE2R1 ( 1:5 , 000 ) . Cells were seeded onto 6-well tissue culture plates at 60–80% confluency . After 72 hr , cells were treated with trypsin solution to remove them from the tissue culture plates . Half of the cells were used for immunoblotting analysis , and the other half were processed for mRNA extraction using the RNeasy mini kit ( Qiagen ) . The mRNA levels were quantified using a nanodrop spectrophotometer , and equal amounts were used for cDNA synthesis using the SuperScript III First-Strand Synthesis System for RT-PCR ( Invitrogen ) . The cDNA samples were included in PCR reactions containing SsoAdvanced Universal SYBR Green Supermix ( BioRad ) and primers corresponding to amplicons for UBE2R1 , UBE2R2 , or GAPDH as a control ( Supplementary file 1 ) . PCR reactions were performed on a Bio-Rad CFX96 Real-Time PCR Detection System . The relative expression ratios were calculated according to the Pfaffl method . UBE2R1/2 knockout cell lines were obtained by transfecting HEK 293T cells ( ATCC ) with pSpCas9 ( BB ) −2A-Puro ( pX459 ) plasmids encoding sgRNA sequences targeting either UBE2R1 or UBE2R2 . UBE2R1/2 double knockout clones were generated by sequential knockout of either UBE2R1 then Ube2R2 or vice versa . Two sgRNA sequences were synthesized targeting either locus ( Supplementary file 1 ) and inserted into the BbsI restriction site of pX459 . The transfected cell population was selected in puromycin for 2 days , and surviving cells were then cloned by serial dilution . Cell clones were expanded and then screened at the protein level by immunoblotting of total cell lysates with isoform specific antibodies ( Figure 5—figure supplement 2 ) . Clones that showed a loss of protein expression were verified at the DNA level by sequencing . Flanking sequences of the Cas9 target site were amplified by PCR , followed by Sanger sequencing . If the resulting sequence traces showed overlapping sequences ( suggestive of heterozygosity ) , the PCR products were cloned into TOPO TA plasmids , and DNA isolated from individual colonies were sequenced until the DNA sequences for both alleles were obtained . Knockout rate at the protein level was 7 out of 11 clones screened for UBE2R1 clones , 8 out of 11 for UBE2R2 clones , 11 out of 17 for UBE2R1 knockout followed by UBE2R2 knockout , and 7 out of 7 for UBE2R2 knockout followed by UBE2R1 . DNA sequencing of the Cas9 target site was carried out on four protein-negative clones for each category and all proved to harbor biallelic indels causing protein termination . Control or UBE2R1/2 double knock-out 293T cells were seeded onto 10 cm tissue culture plates at 10% confluency and were incubated under standard tissue culture conditions for 48 hr . For the analysis of p27 , CYCLIN E , and β-CATENIN proteins , cells were washed with DPBS , followed by the introduction of 1x SDS-PAGE loading buffer directly to the plate . Following brief sonication , samples were boiled for 5 min , and centrifuged at maximum speed using a table top microcentrifuge . Equivalent volumes of lysate were loaded onto 4–20% Tris-Glycine SDS-PAGE gels and subjected to electrophoresis . Immunoblotting was performed as described above . A 1:2000 dilution was used for anti-β-CATENIN antibody ( the same dilutions were used for p27 and CYCLIN E as described above ) . Additional cellular proteins include: anti-α-TUBULIN ( 1:5 , 000 ) , anti-UBE2R1 ( 1:5 , 000 ) , anti-ARIH1 ( 1:500 ) , anti-Ube2R2 ( 1:2000 ) , and anti-UBE2D3 ( 1:5 , 000 ) . For quantitation of the steady-state protein levels , immunoblots were processed as described above and incubated in primary antibody overnight . Blots were then washed three times with TBS-T and incubated with StarBright B700 ( BioRad ) secondary antibodies ( 1:3000 ) for approximately 1 hr . Blots were incubated at room-temperature in the dark , followed by 4 to 6 washes with TBS-T . Immunoblots were imaged on a BioRad ChemiDoc MP gel imaging system . Quantitation of the protein levels was accomplished in Image Lab ( BioRad ) by first correcting for protein loading using the α-TUBULIN levels present in each lane . The corrected values for protein levels were first averaged across all of the lanes and then used to normalize the total signal for each protein in each lane . Cells were seeded onto 6-well plates at approximately 60% confluency late in the day and incubated overnight . Cells were then treated with 100 μg/mL cycloheximide and time-points were collected at 0 , 1 , 3 and 6 hr . Cells were washed once with DPBS , and collected as described above . Immunoblots with quantitation were performed as above . Control or UBE2R1/2 double knockout 293T cells were seeded onto 6-well tissue culture plates with approximately 0 . 5 × 106 cells per well in starvation media ( DMEM supplemented with 0 . 1% fetal bovine serum , 400 μg/mL Bovine Serum Albumin , 4 mM L-Glutamine , 100 units/mL Penicillin and 100 μg/mL Streptomycin ) overnight . The next day , 100 μg/mL cycloheximide was added to the media . After 5 min , TNFα was added to each well ( 50 ng/mL final ) . At the indicated time-points , the cells were washed once with 2 mL of DPBS , followed by the addition of 1x SDS-PAGE loading buffer directly to the plate and immediate collection of the lysate into Eppendorf tubes . Lysates were briefly sonicated , followed by boiling for 5 min , and centrifugation in a table top microcentrifuge at the maximum setting for 2 min . Equal amounts of sample were loaded to 4–20% Tris-Glycine SDS-PAGE gels , followed by quantitative immunoblotting as described above ( the anti-IκBα antibody was diluted 1:1 , 000 ) . WT or UBE2R1/2 double knockout cells were seeded onto 6-well tissue culture plates at approximately 60% confluency and incubated for 48 hr under typical tissue culture conditions . Cells were then trypsinized and grown on a 10 cm dish for an additional day . A minimum of 300 , 000 cells were collected by centrifugation at 1500 x g for 5 min . Cell pellets were resuspended in 1 mL of cold PBS and then recollected by centrifugation as described above . Cell pellets were resuspended in 100 µL cold PBS and fixed by adding 900 µL cold ethanol dropwise while gently vortexing . Fixed cells were stored at 4° C for a minimum of 24 hr and then centrifuged as above to remove ethanol . Cells were re-suspended in 300 µL of propidium iodide staining solution containing a 1:1:1 ratio by volume of propidium iodide ( 150 µg/mL ) , Triton X-100 ( 0 . 1% ) , and Ribonuclease A ( 1 mg/mL ) , transferred to round bottom tubes ( Falcon ) and incubated in the dark at room temperature for 20 min before being placed on ice to be analyzed . Flow cytometry was performed on a BD FACSCalibur flow cytometer with BD CellQuest Pro software for the acquisition , a 488 nm laser for the excitation , and measuring the emission in the FL2 detector with a 585/42 nm emission filter . The samples were run at a low flow rate and 20 , 000 events were collected . The cell cycle analysis was determined by histogram plots of the fluorescent signal in the FL2-A verses counts using FlowJo 7 . 6 . 5 software ( Tree Star ) . sgRNAs for Cas9-mediated generation of UBE2R1 and UBE2R2 single and double knockout populations were drawn from the previously described Extended Knockout ( EKO ) sgRNA library ( Bertomeu et al . , 2018 ) , and are also represented in an earlier sgRNA library ( Wang et al . , 2014a ) . Additional sgRNAs that targeted the AAVS1 locus and GFP were used as inert targeting sequences to control for the effects of multiple lentiviral infections . The sgRNA targeting sequences used are shown in Supplementary file 1 . To allow iterative infections with up to three series of sgRNA lentiviral constructs selectable under different antibiotics , the previously described plasmid pLX-sgRNA 2X BfuA1 ( Bertomeu et al . , 2018 ) was modified to change the cloning cassette and the selection marker . A 1032 bp AAVS1 insert filler was amplified with oligos ( Supplementary file 1 ) from pLX-sgRNA 2X BfuA1 and cloned into pLX-sgRNA 2X BfuA1 digested with BfuA1 by Gibson assembly to create a BsiW1-based cloning cassette ( both Hygromycin and Neomycin resistance genes are cut by BfuA1 ) . The resulting plasmid was cut with BspE1 and EcoR1 to remove the blasticidin resistance gene and re-assembled by Gibson assembly with either the Hygromycin or Neomycin resistance genes ( Supplementary file 1 ) to generate respectively pLX-sgRNA 2X BsiW1 AAVS1 Hygro and pLX-sgRNA 2X BsiW1 AAVS1 Neo . The sgRNA cassettes were cloned into these vectors by PCR amplification and Gibson assembly as previously described ( Bertomeu et al . , 2018 ) . NALM-6 knockout populations were generated by lentiviral infection with UBE2R1#1 , UBE2R1#2 and AAVS1 sgRNA constructs in pLX-sgRNA 2X BsiW1 Neo followed by selection with G418 ( 800 µg/mL ) for 7 days or until uninfected controls were inviable . Subsequently , the UBE2R1#1 , UBE2R1#2 and AAVS1 infected populations were re-infected with UBE2R2#1 , UBE2R2#2 or GFP sgRNA lentiviruses in pLX-sgRNA 2X BsiW1 Hygro , followed by selection with Hygromycin B ( 200 µg/mL ) for 7 days ( or until uninfected controls were inviable ) to generate populations for double knockout generation . Finally , each dual guide cell population was transduced with a lentiviral vector ( Wong et al . , 2016 ) that expressed Cas9 from the constitutive EF1α promoter and selected on Zeocin ( 200 µg/mL ) for 5 days or until uninfected controls were inviable . Populations were verified for UBE2R1 and/or UBE2R2 knockout after 10 and 18 days by immunoblot with antibodies against UBE2R1 ( Santa Cruz ) and UBE2R2 ( Santa Cruz ) at recommended dilutions and detection by SuperSignal West Femto chemiluminescent substrate ( Thermoscientific; Figure 6—figure supplement 1 ) . Aliquots of each AAVS1/GFP control , UBE2R1/GFP , AAVS1/UBE2R2 and UBE2R1/UBE2R2 knockout population were frozen until the start of the genome-wide EKO screens . To initiate screens , thawed aliquots of each population genotype ( UBE2R1#1 + GFP + Cas9; UBE2R2#2 + AAVS1 + Cas9; UBE2R1#2 + GFP + Cas9; UBE2R2#2 + AAVS1 + Cas9; UBE2R1#1 + UBE2R2#1 + Cas9; UBE2R1#2 + UBE2R2#2 + Cas9; AAVS1 + GFP + CAS9 negative control ) were expanded and then infected with the EKO sgRNA library at an MOI of 0 . 36 ( Bertomeu et al . , 2018 ) and a clonal representation of 1 , 700 cells per sgRNA , selected for 6 days with blasticidin ( 10 µg/mL ) , and allowed to grow in the absence of antibiotic selection for an additional 14 days . Cell cultures were harvested at day 0 ( immediately post-blasticidin treatment ) and day 14 for each screen . Genomic DNA was extracted , sgRNAs amplified by two rounds of PCR and high-throughput sequencing performed on an Illumina HiSeq 2000 or a NextSeq 500 instrument in multiplexed format , as previously described ( Bertomeu et al . , 2018 ) except that an internal primer within the blasticidin resistance gene was used in the first PCR step to avoid amplification of sgRNA cassettes initially used to target UBE2R1/2 . Total read counts for each screen ranged between 14 . 8 and 26 . 9 million reads . High through-put sequence data can be found at the GEO repository: https://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE136175 . Sequences were aligned using Bowtie 2 . 2 . 5 ( Langmead and Salzberg , 2012 ) with default parameters other than the ‘--norc’ option . sgRNA read counts from all day 0 samples were summed together to generate consensus day 0 sgRNA levels , which were used as the control for scoring each screen . Each sample was analyzed using RANKS ( Bertomeu et al . , 2018 ) with default parameters and using the non-targeting sgRNA control set that is contained in the EKO library . To control for the effects of genotype-independent fitness defects , we averaged the RANKS scores from the two control screens and subtracted these scores from those of each knockout screen . The resulting differential RANKS scores represent the genotype-specific fitness effects of the culminative gene indels in the pooled library . P-values and FDR values were estimated using a custom-generated control distribution modeling the effects of subtracting scores . RANKS scores for different replicates of a given genetic background were then averaged together . We note that UBE2G1 scored as the 108th highest ranked synthetic lethal gene deletion ( RANKS score = −1 . 9 ) in one of the two UBE2R2 single knockout backgrounds . Since UBE2R2 and UBE2R1 have a high degree of sequence similarity at the DNA level , we hypothesized that the sgRNA in question might have had an off-target effect on UBE2R1 with a lower but nevertheless detectable efficiency . We compared the UBE2R2 sgRNA sequence ( CGACCTCTACAACTGGGAGG ) to all potential sgRNAs with an associated PAM site that target RefSeq genes and identified one site in UBE2R1 with only two mismatched bases near the beginning of the sequence ( CGATCTATACAACTGGGAGG ) , which is where mismatches are known to have a relatively small effect on Cas9 recognition and nuclease efficiency ( Wang et al . , 2014a ) . It is therefore likely that UBE2R1 was fully or partially deleted in a subset of cells in the UBE2R2 population generated with this sgRNA , explaining why a weak genetic interaction with UBE2G1 was observed . The other UBE2R2-targeting sgRNA did not have any predicted off-target cleavage sites with up to two mismatches and correspondingly UBE2G1 did not score strongly in the screen with this sgRNA . The genetic interaction between UBE2R1 , UBE2R2 and UBE2G1 was validated in NALM-6 cells by generation of clonal UBE2R1/2 double knockout cell lines . Populations of a clone that expressed Cas9 under doxycycline-inducible control ( NALM-6 #20; Bertomeu et al . , 2018 ) were generated by lentiviral infection with sgRNA constructs for each of GFP , UBE2R2#1 and UBE2R2#2 in pLX-sgRNA 2X BsiW1 Hygro , followed by selection with 200 µg/mL Hygromycin B . Cell populations were subsequently infected with lentiviral constructs for AAVS1 , UBE2R1#1 and UBE2R1#2 in pLX-sgRNA 2X BsiW1 Neo , followed by selection with 800 µg/mL G418 . Control , single knockout and double knockout populations were then induced with 2 µg/mL doxycycline for 5 days , after which clonal knockout cell lines were generated by serial dilution single cell cloning . Clones were confirmed for loss of UBE2R1 and/or UBE2R2 protein expression by immunoblot ( Figure 6—figure supplement 2a , b ) . Each cell line was then transduced with lentiviral constructs for two different sgRNAs that target the UBE2G1 locus ( Supplementary file 1 ) . After selection with blasticidin , UBE2R1 , UBE2R2 or UBER1/2 double knockout clones that expressed the UBE2G1 sgRNAs were treated with 2 µg/mL doxycycline to induce generation of indels in the UBE2G1 locus . Following 6 days of doxycycline induction , single cells were seeded in 96-well plates by serial dilution and monitored for colony growth . After 21 days , individual colonies for each genotype were assessed for cell number and viability by phase contrast microscopy ( Figure 6—figure supplement 2c ) . Control or UBE2R1/2 double knockout cells were seeded at ~60–80% confluency . The next day , cells were transfected with 25 pmol of siRNA ( Dharmacon ) for either UBE2G1 or a nontargeting control ( Supplementary file 1 ) . After a 48 hr incubation period , the cells were treated with trypsin and seeded onto 10 cm dishes . Cells were incubated for an additional 24 hr and then harvested for protein analysis 3 days post transfection . Lysates were prepared and immunoblotting was performed as above ( detection by both x-ray film and fluorescent secondary antibodies ) . The anti-HIF1α antibody was diluted 1:500 , and the anti-UBE2G1 antibody was diluted 1:250 . WT 293T-FiTx or 293T-FiTx UBE2R1Δ were seeded onto 35 mm glass bottom plates and incubated for 24–36 hr prior to fixation in 4% paraformaldehyde for 3 min . Cells were then washed three times in DPBS for 3 min , quenched in 50 mM NH4Cl for 5 min , washed again , and permeabilized in 0 . 1% Triton X-100 for 30 min . Primary antibody ( anti-UBE2R1; Abcam ) was incubated overnight at 4° C , which was then repeatedly washed off in PBS-T , first every 5 min for an hour , and again every 10 min for an additional hour . Secondary antibody ( Alexa Fluor 488 ) was incubated for 1 hr at room temperature , repeatedly washed in PBS-T first every 5 min for two hours , and again every 10 min for an additional two hours . The cells were then mounted using Permount with DAPI , and imaged using a Nikon A1D confocal microscope . The images were taken as maximum projections of z-stacks ( 0 . 5 μm ) .

Proteins are the molecules that perform most of the tasks that keep cells alive , but often they need to be removed . If human cells lose control over protein degradation it can result in diseases such as cancer or neurodegenerative disorders . The enzymes responsible for tagging proteins for destruction are called ubiquitin ligases . Drugs that hijack ubiquitin ligases to tag disease-causing proteins have been successfully used to treat a cancer called multiple myeloma . Encouragingly , human cells have over 600 ubiquitin ligases and most have not yet been tested as therapeutic targets . A clear understanding of how these enzymes work in human cells could therefore lead to new therapies for conditions such as cancer . To tag a protein for degradation , ubiquitin ligases transfer a small protein ( ubiquitin ) from a ubiquitin-carrying enzyme to the target protein . In yeast , this process is relatively simple , since in many cases there is a one-to-one relationship between each ubiquitin ligase and its ubiquitin-carrying enzyme partner . In human cells , on the other hand , the process seems to be more complex . The biggest family of ubiquitin ligases in humans is the cullin-RING ligase family , and the number of partner ubiquitin-carrying enzymes for this family remains unknown , as are the effects of different interactions between members of the family and different ubiquitin-carrying enzymes . Now , Hill et al . have used biochemical assays to measure the activities of four ubiquitin-carrying enzymes that partner with cullin-RING ligases . They found that while certain proteins can be tagged for degradation by different combinations of the cullin-RING ligases and the ubiquitin-carrying enzymes , others display preferences for specific ubiquitin-modifying enzymes partnering up . The experiments also revealed that one of the ubiquitin-carrying enzymes tested was active at high concentrations , but could not tag proteins when assayed at concentrations closer to those found in the cell . Finally , Hill et al . genetically removed two of the ubiquitin-carrying enzymes and showed that , unlike in yeast , a third ubiquitin-carrying enzyme could compensate for their loss , a redundancy that makes the system robust . These results show that the human cullin-RING ligases can interact with multiple ubiquitin-carrying enzyme partners . Controlled protein degradation affects every major activity in human cells , and a good understanding of the mechanisms that regulate this process can help researchers better understand many biological processes . Additionally , these findings are relevant to the development of therapies trying to use ubiquitin ligases to remove faulty proteins .

2019

Robust cullin-RING ligase function is established by a multiplicity of poly-ubiquitylation pathways

Tumor-initiating cells ( TIC ) are dynamic cancer cell subsets that display enhanced tumor functions and resilience to treatment but the mechanism of TIC induction or maintenance in lung cancer is not fully understood . In this study , we show the calcium pathway transcription factor NFATc2 is a novel regulator of lung TIC phenotypes , including tumorspheres , cell motility , tumorigenesis , as well as in vitro and in vivo responses to chemotherapy and targeted therapy . In human lung cancers , high NFATc2 expression predicted poor tumor differentiation , adverse recurrence-free and cancer-specific overall survivals . Mechanistic investigations identified NFATc2 response elements in the 3’ enhancer region of SOX2 , and NFATc2/SOX2 coupling upregulates ALDH1A1 by binding to its 5’ enhancer . Through this axis , oxidative stress induced by cancer drug treatment is attenuated , leading to increased resistance in a mutation-independent manner . Targeting this axis provides a novel approach for the long-term treatment of lung cancer through TIC elimination . Lung cancer results from mutations induced by DNA adducts , free radicals and reactive oxygen species ( ROS ) generated during tobacco smoking and chronic inflammation ( Acharya et al . , 2010; Okumura et al . , 2012; Houghton , 2013 ) . Due to late presentation and the lack of effective long term therapy , it has a high mortality and new treatment approaches are needed to improve patient outcomes . Recent research has shown the cellular landscape of a cancer is heterogeneous . Cells showing aberrant expressions of various molecules have enhanced propensities for survival , tumorigenicity , drug resistance , designated as cancer stem cells or tumor-initiating cells ( TIC ) . In some cancers , constitutive activities of inherent embryonic or developmental pathways for stem cell renewal are involved in TIC maintenance , such as the WNT/β-catenin pathway in colonic adenocarcinomas ( AD ) . In tissues with slow cell turnover , mechanisms that elicit cell plasticity and stemness properties during tissue response to intracellular and extracellular stresses are involved ( Valent et al . , 2012; Visvader and Lindeman , 2012; Beck and Blanpain , 2013 ) . For the adult lung , stem cell niches or their physiological regulatory mechanisms are ill-defined , and molecular programs sustaining lung TIC are still elusive . Intracellular free calcium is at the hub of multiple interacting pathways activated by extracellular and/or intrinsic stimulations , e . g . EGFR , endoplasmic reticulum and mitochondrial stresses , etc . ( Roderick and Cook , 2008; Prevarskaya et al . , 2010; Zhao et al . , 2013; Déliot and Constantin , 2015 ) , raising the possibility stress signals transduced by calcium pathway mediators could be involved in the induction of TIC phenotypes . In cancers of the breast , pancreas , colon and melanoma , the calcium signaling transcription factor nuclear factor of activated T-cells ( NFAT ) has been shown to contribute to malignant properties including cell invasion , migration , survival , proliferation , stromal modulation and angiogenesis ( Werneck et al . , 2011; Gerlach et al . , 2012; Qin et al . , 2014 ) . However , information on its role in TIC phenotypes , especially drug resistance , is limited and details of the molecular pathways linking calcium signaling to TIC induction have not been reported . In this study , we demonstrated the parlor NFATc2 supports tumorigenicity , cell survival , motility and drug resistance of human lung AD . Amongst essential factors of pluripotency , SOX2 is an NFATc2 target upregulated through its 3’ enhancer , while SOX2 couples to a 5’ enhancer of ALDH1A1 mediating overexpression . NFATc2 induces ALDH+ subsets and ALDH+/CD44+-TIC and enhances drug resistance by ROS scavenging through the NFATc2/SOX2/ALDH1A1 axis . Our study reports a novel lung TIC maintenance pathway which links micro-environmental stimulation to induction of stemness phenotypes , evasion of cell death and enhancement of drug resistance . NFATc2 could be an important target in treatment strategies aiming at disruption of lung TIC . By analyzing transcripts expression , we observed NFATc2 was significantly overexpressed in human primary NSCLC compared to normal lung ( Figure 1A ) . Using IHC , high level activated NFATc2 expression with intense and widespread nuclear staining were detected in 41 of 102 ( 40 . 2% ) excised primary NSCLC , while 61 ( 59 . 8% ) showed low expression with weak nuclear and/or cytoplasmic staining in isolated or small clusters of tumor cells ( Figure 1B , C ) . In normal lung epithelium , NFATc2 was expressed in the bronchiolar stem cell compartment of basal reserve cells while differentiated bronchiolar cells or alveolar pneumocytes were negative ( Figure 1D ) . Using log rank test and Kaplan-Meier survival analysis , we showed tumors with high level NFATc2 expression had significantly shorter recurrence-free survival ( RFS ) and cancer-specific overall survival ( OS ) ( Figure 1E , F ) . High NFATc2 expression significantly predicted poor tumor differentiation , advanced tumor stage and TNM stage ( Table 1A ) . Multivariate Cox regression analysis further showed high NFATc2 , late pathological stage , age and smoking history were independent prognostic indicators for shorter OS , while high NFATc2 and advanced pathological stage were predictive for shorter RFS ( Table 1B , C ) . The results indicated NFATc2 expression was associated with repressed tumor differentiation and adverse patient survivals . 10 . 7554/eLife . 26733 . 003Figure 1 . NFATc2 was overexpressed in human NSCLC and predicted poor survivals . ( A ) NFATc2 expression analyzed by qPCR in human NSCLC and corresponding normal lung . P: Wilcoxon test . p=0 . 0003 . ( B–C ) NFATc2 expression analyzed by IHC , showing representative areas of high NFATc2 scores with strong nuclear staining in the majority of cancer cells ( B ) , or low NFATc2 scores with weak nuclear and cytoplasmic staining ( C ) , respectively . ( D ) NFATc2 expression in normal bronchial epithelium by IHC , showing nuclear NFATc2 staining in scattered bronchiolar reserve/stem cells of the basal layer ( arrows ) . For B-D: Scale bars , 50 µm . ( E–F ) Kaplan Meier survival curves by log-rank tests on 102 resected primary NSCLC stratified by NFATc2 expression levels for recurrence-free survival ( RFS ) ( E ) , and overall survival ( OS ) ( F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26733 . 00310 . 7554/eLife . 26733 . 004Figure 1—source data 1 . Statistical analyses for Figure 1A . DOI: http://dx . doi . org/10 . 7554/eLife . 26733 . 00410 . 7554/eLife . 26733 . 005Table 1 . Clinico-pathological correlation of NFATc2 in NSCLC patients . DOI: http://dx . doi . org/10 . 7554/eLife . 26733 . 005A . Clinico-pathological correlation of NFATc2 in NSCLC patientsNFATc2Clinico-pathological variablesLowHighP valueGender Female22110 . 328 Male3930Age ( Years ) ≤6536240 . 961 >652517Smoking history Non-smoker32240 . 545 Smoker2917Differentiation Well to moderate45210 . 019* Poor1620Histologic type Adenocarcinoma42240 . 357 Squamous cell carcinoma1110 Others87Tumor Stage T1-T253240 . 001* T3-T4817Lymph node metastasis Absent43250 . 318 Present1816Pathological ( TNM ) stage Stage I36140 . 014* Stage II-IV2527B . Multivariate COX regression analysis for RFSVariablesP valueHazard Ratio ( HR ) 95 . 0% CI† of HR NFATc20 . 0371 . 9051 . 039–3 . 494 TNM stage0 . 0012 . 0351 . 347–3 . 075C . Multivariate COX regression analysis for OSVariablesP valueHazard ratio ( HR ) 95 . 0% CI† of HR NFATc20 . 0022 . 8241 . 462–5 . 457 TNM stage0 . 0121 . 8271 . 140–2 . 927 Age0 . 012 . 331 . 224–4 . 432 Smoking history0 . 0092 . 4161 . 251–4 . 665Statistical tests: c2; *: P<0 . 05†: Confidence Interval . Statistics: COX regression analysis . If NFATc2 supports tumor-initiating phenotypes , it is expected to be expressed at a higher level in TIC compared to non-TIC . To address this notion , marker-independent TIC surrogates comprising tumorspheres raised from 4 lung AD cell lines cultured in non-adherent , stem cell conditions ( Liu et al . , 2007; Shi et al . , 2015; Sun et al . , 2015 ) were compared with non-TIC growing in monolayers . TIC showed higher NFATc2 expression by Western blot ( Figure 2A ) , while transcripts of NFATc2 and its target FASL were also significantly upregulated ( Figure 2—figure supplement 1A ) . Luciferase reporter assays also showed significantly higher NFAT activities in spheres isolated from H1299 and A549 cells ( Figure 2—figure supplement 1B ) . Furthermore , TIC selected by the lung TIC markers ALDH+/CD44+ from HCC827 and the patient-derived lung cancer cell lines , HKUCL2 and HKUCL4 , showed higher NFATc2 expression than the ALDH-/CD44- non-TIC counterpart ( Figure 2B ) ( Liu et al . , 2013a ) . Using another lung TIC marker , CD166high , for TIC isolation from HCC827 , NFATc2 was also shown to be upregulated ( Figure 2C ) ( Zhang et al . , 2012 ) . 10 . 7554/eLife . 26733 . 006Figure 2 . NFATc2 NFATc2 was overexpressed in lung TIC and mediated TIC properties in vitro . ( A–C ) Expression of NFATc2 analyzed by Western blot , in TIC isolated as tumorspheres compared with non-TIC from cells in monolayers ( A ) ; TIC isolated as ALDH+/CD44+ subset compared with ALDH-/CD44- subset ( B ) ; TIC isolated as CD166high subset compared with the CD166low subset ( C ) . ( D ) NFATc2 expression by Western blot in cells with stable NFATc2 knockdown , overexpression , or knockout , respectively . ( E ) BrdU proliferation assay of HCC827 cells with NFATc2 knockdown . ( F ) Cell cycle analysis of HCC827 cells with NFATc2 knockdown . ( G–I ) Tumorsphere formation and serial passage assays , in HCC827 cells after stable NFATc2 knockdown ( G ) or knockout ( H ) , or in PDCL#24 cells with NFATc2 knockdown ( I ) . ( J–K ) Tumorsphere formation and serial passage assays in cells with stable NFATc2 over-expression , including A549 cells ( J ) and H1299 cells ( K ) . ( L ) Tumorsphere formation assay in HCC827 cells with or without treatment with 1 µM of CSA or 5 µM of FK506 . ( M–P ) Cell migration and invasion assays in cells with stable NFATc2 knock down ( M and N ) or over-expression ( O and P ) . For G-P: *p<0 . 05 **p<0 . 01 , comparison with control by t-test . Error bar indicates the mean ±SD for at least three independent replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 26733 . 00610 . 7554/eLife . 26733 . 007Figure 2—figure supplement 1 . NFATc2 was up-regulated in tumorspheres . ( A ) Expression of NFATc2 and its target FASL analyzed by qPCR in TIC isolated as tumorspheres compared to the monolayers of non-TIC . ( B ) NFAT luciferase reporter activity in TIC isolated as tumorspheres compared to non-TIC monolayer controls . *p<0 . 05 **p<0 . 01 , comparison with control by t-test . Error bar indicates the mean ±SD for three independent replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 26733 . 00710 . 7554/eLife . 26733 . 008Figure 2—figure supplement 2 . NFATc2 knockdown did not affect cell cycle progression of HCC827 cells . Representative views of cell cycle distribution of HCC827 cells with or without stable NFATc2 knockdown analyzed by flow cytometry . DOI: http://dx . doi . org/10 . 7554/eLife . 26733 . 008 For functional studies , NFATc2 was silenced by 2 shRNA sequences ( shNFATc2-A and -B ) in 2 lung cancer cell lines with high basal expression ( HCC827 , PDCL#24 ) , and ectopically expressed in 2 cell lines with relatively low de novo expression ( A549 , H1299 ) ( Figure 2D ) . Knockout using CRISPR/CAS9 and gRNA targeting NFATc2 ( gNFATc2 ) was also performed on HCC827 ( Figure 2D ) . As shown by BrdU proliferation assay and cell cycle assay , NFATc2 knockdown did not significantly affect the proliferation and cell cycle distribution of HCC827 cells ( Figure 2E and F , Figure 2—figure supplement 2 ) . However , abrogation of NFATc2 significantly reduced 60–70% of tumorspheres in all the cell models and inhibited tumorspheres renewability for 2 consecutive generations ( Figure 2G–I ) , while overexpression significantly augmented tumorspheres in both A549 and H1299 ( Figure 2J–K ) . To demonstrate the actions of NFAT on TIC-related phenotypes are mediated through calcium signaling , we blocked calcium-mediated NFAT activation by disrupting the upstream calcineurin/NFAT dephosphorylation complex using the potent and specific inhibitors cyclosporin A ( CSA ) and FK506 , respectively . Treatment with both CSA and FK506 significantly inhibited sphere formation of HCC827 cells ( Figure 2L ) . Transwell assays for cell motility showed silencing NFATc2 significantly reduced the migration and invasion ability of both HCC827 and PDCL#24 cells ( Figure 2M–N ) , while the opposite effects were rendered by NFATc2 overexpression in A549 and H1299 cells ( Figure 2O–P ) . Subcutaneous xenograft models showed NFATc2 knockdown significantly reduced tumor sizes and retarded growth rates of HCC827 and PDCL#24 , respectively ( Figure 3A–B ) . Using IHC , NFATc2 expression was detected in xenografts established from HCC827 and PDCL#24 knockdown cells ( Figure 3C ) , indicating successful tumorigenesis preferentially involved cells that had selected away from NFATc2 knockdown . In contrast , NFATc2 overexpression in A549 significantly augmented tumor growth ( Figure 3D ) . To evaluate TIC frequencies in vivo , limiting dilution assays were performed by subcutaneous transplantation of serially decreasing numbers of tumor cells in nude mice . NFATc2 knockdown led to significantly reduced xenograft incidence and TIC frequency of HCC827 ( Figure 3E , Figure 3—figure supplement 1A ) . Reciprocal effects were observed with NFATc2-overexpression , ( Figure 3F , Figure 3—figure supplement 1B ) . Together , the data supported NFATc2-mediated in vivo tumorigenesis . 10 . 7554/eLife . 26733 . 009Figure 3 . NFATc2 regulated tumorigenesis in vivo . ( A–B ) 1 × 104 of HCC827 cells ( A ) , and PDCL#24 cells ( B ) , respectively , were subcutaneously inoculated into the flanks of SCID mice , and tumor volumes were monitored . Representative tumor images and tumor growth curves are shown . **p<0 . 0001 , comparison with respective control by two-way ANOVA . Error bar indicates the mean ±SD of tumor volumes of mice as indicated . ( C ) NFATc2 expression by IHC in xenografts generated from HCC827 or PDCL#24 cells , respectively , with or without NFATc2 knockdown . Tumor cells at the tumor/stroma interface ( arrows ) showed stronger NFATc2 expression , possibly due to micro-environmental induction . Scale bars , 50 µm . ( D ) 1 × 104 of A549 cells with or without NFATc2 overexpression were subcutaneously injected into SCID mice , and tumor volumes were monitored . Representative tumor images and tumor growth curves are shown . **p<0 . 0001 , comparison with control by two-way ANOVA . Error bar indicates the mean ±SD of tumor volumes of mice as indicated . ( E–F ) Limiting dilution assay in vivo . Indicated numbers of HCC827 cells ( E ) , and A549 cells ( F ) were subcutaneously inoculated into SCID mice , and the tumor incidence and latency were monitored for 3 months . The TIC frequency and P values were calculated using the L-Calc software ( Stemcell Tech , Vancouver , Canada , http://www . stemcell . com ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26733 . 00910 . 7554/eLife . 26733 . 010Figure 3—source data 1 . Statistical analyses for Figure 3A , B and D . DOI: http://dx . doi . org/10 . 7554/eLife . 26733 . 01010 . 7554/eLife . 26733 . 011Figure 3—figure supplement 1 . NFATc2 regulated in vivo tumorigenesis . ( A–B ) Xenografts of HCC827 cells with stable NFATc2 knockdown ( A ) , and of A549 cells with NFATc2 over-expression ( B ) used for limiting dilution assays . DOI: http://dx . doi . org/10 . 7554/eLife . 26733 . 011 The effect of NFATc2 on chemotherapy response was first investigated using cisplatin chemotherapy on PDCL#24 , a KRAS V12D mutant lung AD cell line raised from a local male chronic smoker . Upon NFATc2 inhibition , sensitization to cisplatin with statistically significant reduction of IC50 was observed compared to control cells ( Figure 4A ) . Similarly , NFATc2 knockout as well as CSA and FK506 treatment also sensitized HCC827 to cisplatin with reduction of IC50 ( p<0 . 01 ) ( Figure 4—figure supplement 1A , Figure 4B ) . In contrast , overexpressing NFATc2 in A549 increased cisplatin resistance with significantly elevated IC50 ( p<0 . 01 ) ( Figure 4C ) . In vivo , mice bearing PDCL#24 xenografts treated with cisplatin alone showed 1 . 68 fold tumor shrinkage compared with vehicle control . With additional stable NFATc2 knockdown , tumor shrinkage was enhanced to 3 . 15 and 2 . 2 fold relative to respective vehicle controls ( p<0 . 01 ) ( Figure 4D ) , while tumor growth rate was also retarded ( Figure 4—figure supplement 1B ) . Besides cisplatin , NFATc2 knockdown significantly increased paclitaxel sensitivity of HCC827 cells ( Figure 4—figure supplement 2A ) . A549 cells induced for cisplatin resistance by chronic progressive drug exposure ( A549 CR ) showed elevated IC50 with NFATc2 upregulation compared to parental A549 cells ( Figure 4E–F ) . With NFATc2 suppression , A549 CR was re-sensitized with return of IC50 to around the pre-induction level ( p<0 . 01 ) ( Figure 4F ) . In line with this observation , induction of H1299 for paclitaxel resistance also caused NFATc2 upregulation ( Figure 4—figure supplement 2B ) . 10 . 7554/eLife . 26733 . 012Figure 4 . NFATc2 promoted resistance to cytotoxic and targeted therapy . ( A ) Effect of NFATc2 knockdown on cisplatin response of PDCL#24 cells by MTT assay . ( B ) Effect of CSA or FK506 treatment on cisplatin response of HCC827 cells by MTT assay . ( C ) Effect of NFATc2 overexpression on cisplatin response of A549 cells by MTT assay . *p<0 . 05 , **p<0 . 01 versus control by t-test . ( D ) In vivo effect of NFATc2 knockdown on cisplatin response of PDCL#24 xenografts . 1 × 106 of PDCL#24 cells were subcutaneously inoculated into the flanks of Nude mice . Nude mice bearing subcutaneous xenografts were randomly separated into two groups and treated with intraperitoneal injections of cisplatin ( 4 mg/kg every three days ) or saline control , respectively . Xenografts were photographed and histograms of tumor volumes were compared to vector and no-treatment controls . *p<0 . 05 , **p<0 . 01 by t-test . Error bar indicates the mean ±SD of tumor volumes of five mice . ( E ) NFATc2 expression by Western blot in A549 and corresponding cells with induced cisplatin-resistance ( A549 CR ) with or without NFATc2 knockdown . ( F ) Effects of NFATc2 knockdown on cisplatin sensitivity by MTT assays in A549 and A549 CR cells . ##p<0 . 01 , versus vector control of parental cells , **p<0 . 01 versus CR Sh-Ctrl by t-test . ( G–H ) Dose response curves of gefitinib treatment by MTT assays of HCC827 cells with NFATc2 knockdown ( G ) , or knockout ( H ) . ( I ) Dose response curves of gefitinib treatment by MTT assays of HCC827 cells in the presence of CSA or FK506 for 72 hr . For G-I: **p<0 . 01 versus control by t-test . ( J ) Effects of NFATc2 stable knockdown on response of HC827 xenografts to gefitinib . 1 × 106 of HCC827 cells were subcutaneously inoculated into the flanks of Nude mice . Nude mice bearing subcutaneous xenografts were randomly separated into two groups and treated with gefitinib ( 25 mg/kg/day by oral gavage ) or 1% Tween 80 as control . *p<0 . 05 , **p<0 . 01 by t-test . Error bar indicates the mean ±SD of tumor volumes of six mice . ( K and L ) NFATc2 expression with or without NFATc2 knockdown by Western blot ( K ) , and NFAT activity by luciferase reporter assay ( L ) , in HCC827 parental and gefitinib-resistant ( GR ) cells . ( M ) Gefitinib sensitivity of HCC827GR cells treated with CSA or with NFATc2 knockdown analyzed by MTT assays . **p<0 . 01 versus control by Student’s t-test . For all MTT assays , error bar indicates mean ±SD for at least three replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 26733 . 01210 . 7554/eLife . 26733 . 013Figure 4—source data 1 . Statistical analyses for Figure 4D and J , Figure 4—figure supplement 1B and 3A . DOI: http://dx . doi . org/10 . 7554/eLife . 26733 . 01310 . 7554/eLife . 26733 . 014Figure 4—figure supplement 1 . NFATc2 promoted cancer cell resistance to cisplatin treatment . ( A ) Cisplatin sensitivity of HCC827 cells with NFATc2 knockout analyzed by MTT assay . **p<0 . 01 , comparison with control by t-test . Error bar indicates the mean ±SD for three independent replicates . ( B ) Growth curve showing in vivo cisplatin response of PDCL#24 xenografts with or without NFATc2 knockdown . ***p<0 . 0001 versus respective vehicle groups , ###p<0 . 0001 versus cisplatin treated Sh-Ctrl group by two-way ANOVA . Error bar indicates the mean ±SD of tumor volumes of five mice . DOI: http://dx . doi . org/10 . 7554/eLife . 26733 . 01410 . 7554/eLife . 26733 . 015Figure 4—figure supplement 2 . NFATc2 promoted cancer cell resistance to paclitaxel treatment . ( A ) Paclitaxel sensitivity analyzed by MTT assay in HCC827 cells with NFATc2 knockdown . **p<0 . 01 , comparison with control by t-test . Error bar indicates the mean ±SD for three independent replicates . ( B ) NFATc2 expression analyzed by Western blot in H1299 parental and paclitaxel resistant ( TR ) cells . DOI: http://dx . doi . org/10 . 7554/eLife . 26733 . 01510 . 7554/eLife . 26733 . 016Figure 4—figure supplement 3 . NFATc2 promoted cancer cell resistance to gefitinib treatment . ( A ) Growth curve showing in vivo gefitinib response of HCC827 xenografts with or without NFATc2 knockdown . Nude mice bearing subcutaneous xenografts derived from 1 × 106 cells were randomly assigned to two groups and treated with gefitinib or 1% Tween 80 as control . Gefitinib ( 25 mg/kg/day by oral gavage ) was administered for 2 cycles of 5 treatment days followed by 2 rest days per week . ( B ) Effects of NFATc2 stable knockdown on response of HC827 xenografts to short term gefitinib treatment . Nude mice bearing subcutaneous xenografts derived from 2 . 5 × 106 cells were randomly assigned to 2 groups and treated with gefitinib ( 25 mg/kg/day by oral gavage ) or 1% Tween 80 as control for 5 consecutive days . Tumors were then allowed to grow for three weeks without treatment . ***p<0 . 0001 versus respective vehicle control , #p<0 . 05 , ##p<0 . 01 versus cisplatin treated Sh-Ctrl group by two-way ANOVA . ( C ) Images of xenografts in ( B ) and histograms of tumor volumes compared to vector and no-treatment controls . **p<0 . 01 by t-test . Error bar indicates the mean ±SD of tumor volumes of the mice cohorts . DOI: http://dx . doi . org/10 . 7554/eLife . 26733 . 016 HCC827 is a lung AD cell line known to harbor an activating EGFR exon19 deletion which sensitizes it to tyrosine kinase inhibitor ( TKI ) therapy . To investigate whether NFATc2 contributes to targeted therapy resistance , NFATc2 was stably inhibited by shRNA knockdown or CRISPR knockout . This led to significantly reduced IC50 for the TKI gefitinib ( Figure 4G , H ) . Similarly , co-treatment with either CSA or FK506 , respectively , significantly sensitized HCC827 cells to gefitinib ( Figure 4I ) . In vivo , although both groups of NFATc2 knockdown mice showed higher folds of tumor shrinkage ( 3 . 57 fold , 4 . 73 fold , respectively ) after 2 weeks of gefitinib treatment compared to scramble control ( 3 . 38 fold ) ( Figure 4J ) , the added effects of NFATc2 inhibition were very modest and the effects on the growth curve was not clear-cut ( Figure 4—figure supplement 3A ) . Using an alternative model , we investigated the effects of NFATc2 inhibition on response to short term gefitinib treatment for 5 days which allowed tumor recovery from the pronounced effect of gefitinib . In mice with NFATc2 knockdown , tumor regrowth was observed in only 3 and 2 mice of the sh-NFATc2-A and sh-NFATc2-B groups , respectively . In contrast , all mice in the control group showed tumor regrowth . The NFATc2-inhibited xenografts showed more effective tumor inhibition ( 5 . 44 fold , 39 . 73 fold , respectively ) compared to the control group ( 3 . 54 fold ) ( p<0 . 01 ) and differences in tumor volumes on the growth curve were statistically significant ( Figure 4—figure supplement 3B–C ) . In HCC827 induced for gefitinib resistance ( HCC827 GR ) , NFATc2 was upregulated and NFAT promoter activities were increased compared to parental cells ( Figure 4K–L ) . Upon CSA inhibition or sh-NFATc2 knockdown , respectively , re-sensitization to gefitinib with significantly reduced IC50 resulted ( Figure 4M ) . Integrating the in vitro and in vivo data of various combinations of multiple cell lines and cancer drug treatments , the enhancing effect of NFATc2 on drug resistance to cytotoxic and targeted therapy was demonstrated . To understand the molecular mechanism through which NFATc2 mediates cancer cell stemness and drug resistance , we hypothesize NFATc2 might be linked to the pluripotency machinery through its regulatory action . Indeed , analysis of 4 lung AD cell lines showed transcripts of the major stemness factors SOX2 , OCT4 and NANOG were significantly elevated in tumorspheres compared to monolayers ( Figure 5—figure supplement 1 ) . Genetic inhibition of NFATc2 in HCC827 and PDCL#24 led to consistent SOX2 repression with the highest magnitude of change compared to the other 2 factors ( p<0 . 01 ) ( Figure 5A–B ) , while all 3 were significantly upregulated on NFATc2 ectopic expression ( Figure 5C–D ) . Corresponding changes were shown at the protein level ( Figure 5E ) . Further , inhibition of calcineurin activity either by inhibitors ( Figure 5F–G ) , or by siRNA transient knockdown of one of its subunits PPP3R1 ( Figure 5H–I ) consistently down-regulated SOX2 expression in both HCC827 and PDCL#24 cells . On the other hand , NFATc1 has been reported to be a transcriptional regulator of SOX2 in pancreatic cancer ( Singh et al . , 2015 ) . To study whether it is also involved in regulating SOX2 , expression in lung AD , NFATc1 was transiently knocked down by siRNA in HCC827 and PDCL#24 cells , which did not result in consistently reduced SOX2 expression ( Figure 5J–K ) , indicating NFATc2 , rather than NFATc1 , was involved in the regulation of SOX2 expression in lung AD . Together , the data suggested SOX2 is a major stemness target of the calcineurin/NFATc2 axis . 10 . 7554/eLife . 26733 . 017Figure 5 . NFATc2 regulated SOX2 expression . ( A–D ) Pluripotency genes expressions analyzed by qPCR in HCC827 ( A ) , PDCL#24 ( B ) , A549 ( C ) , and H1299 cells ( D ) with NFATc2 knockdown or overexpression . ( E ) Effects of stable NFATc2 knock-down , knockout or overexpression on SOX2 expression in respective lung cancer cells by Western blot analysis . ( F–G ) Pluripotency genes expression analyzed by qPCR in HCC827 ( F ) , and PDCL#24 cells ( G ) treated with CSA or FK506 , respectively , for 24 hr . ( H–I ) Effects of transient knockdown of PPP3R1 on pluripotency gene expressions analyzed by qPCR in HCC827 ( H ) and PDCL#24 cells ( I ) . ( J–K ) Effects of transient knockdown of NFATc1 on pluripotency gene expressions analyzed by qPCR in HCC827 ( J ) and PDCL#24 cells ( K ) . *p<0 . 05 , **p<0 . 01 versus control by t-test . Error bar indicates the mean ±S . D . for at least three independent replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 26733 . 01710 . 7554/eLife . 26733 . 018Figure 5—figure supplement 1 . NFATc2 knockdown did not affect cell cycle progression of HCC827 cells . Expression of pluripotency factors in tumorspheres was analyzed by qPCR and normalized to monolayer non-TIC of indicated cell lines . *p<0 . 05 , **p<0 . 01 versus control by t-test . Error bar indicates the mean ±S . D . for three replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 26733 . 018 To further delineate the molecular mechanism of SOX2 regulation by NFATc2 , we screened , in silico , the genomic sequences spanning 5 kb up- and downstream of the SOX2 transcription start site ( TSS ) , which identified 4 regions encompassing multiple conserved NFAT binding sequences ( Figure 6—figure supplement 1A ) . Alignment with ChIP-seq data of A549 cells retrieved from public databases showed significant overlap at loci of H3K27Ac occupancy with regions 2 and 3 , respectively ( Figure 6A ) . Luciferase assays confirmed these regions are active transcriptional regulatory regions ( Figure 6—figure supplement 1B ) . Further evaluation with respective SOX2 luciferase reporter revealed transcriptional activities were mediated by sites 1 , 2 , 4 , and 5 ( Figure 6B ) . Using H441 lung cancer cell line with NFATc2 transient overexpression , we observed only sites 1 , 4 and 5 showed statistically significant increased reporter activities while those of sites 4 and 5 were reciprocally abolished by CSA treatment ( Figure 6C ) . Finally , site directed mutagenesis of NFAT motifs ( GGAAA to GACTA ) prevented reporter activities of sites 4 and 5 only ( Figure 6D ) , and the findings were supported by data from A549 and H1299 cells ectopically expressing NFATc2 , respectively ( Figure 6—figure supplement 1C , D ) . Notably , sequence homology analysis showed SOX2 sites 4 and 5 are highly conserved across different mammalian species ( Figure 6E ) . Thus , the data suggested NFATc2 was highly likely to regulate SOX2 expression through binding to 3’ enhancers at sites 4 and 5 . For validation , NFATc2 ChIP-qPCR assays were performed using A549 with NFATc2 upregulation , which showed statistically significant enrichment of sites 4 and 5 sequences compared to vector control ( Figure 6F ) . In HCC827 cells , sites 4 and 5 sequences were significantly enriched by anti-NFATc2 antibody compared to IgG control . Conversely , these sequences were significantly reduced upon NFATc2 knockout in HCC827 , compared to their endogenous levels in control cells , indicating de novo physical binding of NFATc2 to SOX2 at sites 4 and 5 ( Figure 6G ) . Together , the data showed NFATc2 upregulates SOX2 by binding to its 3’ enhancer region at around 3 . 2 kb ( site 4 ) and 3 . 6 kb ( site 5 ) from the TSS , respectively . 10 . 7554/eLife . 26733 . 019Figure 6 . NFATc2 regulated tumor functions through trans-activating SOX2 expression . ( A ) Genome browser view of NFAT binding sites and H3K27Ac marks ( lowest panel ) on SOX2 regulatory regions ( regions 2 and 3 indicated in Figure 6—figure supplement 1 ) analyzed in A549 cells . ( B ) Transcriptional activities of sites 1–5 by dual luciferase reporter assays in H441 cells . ( C ) Transcriptional activities of the indicated putative NFAT binding sites by respective luciferase reporters in H441 cells with transient NFATc2 over-expression , with or without CSA treatment . ( D ) Effects of site-directed mutagenesis of the indicated putative NFAT binding sequences by respective luciferase reporter assays in H441 cells with transient NFATc2 overexpression . For B-D , *p<0 . 05 , **p<0 . 01 versus control by t-test . Error bar indicates the mean ±S . D . for at least three independent replicates . ( E ) Alignment of sites 4 and 5 genomic sequences showing highly homologous regions ( gray ) in different mammalian species , with putative NFAT binding sites highlighted in red . ( F–G ) ChIP–qPCR assays of NFATc2 binding to the indicated SOX2 sites in A549 cells with or without stable NFATc2 overexpression ( F ) , or HCC827 cells with or without NFATc2 knockout ( G ) . #p<0 . 05 , ##p<0 . 01 versus IgG control , m **p<0 . 01 versus vector control by t-test . Error bar indicates the mean ±S . D . for at least three independent replicates . ( H ) Correlation of immunohistochemical expressions of NFATc2 and SOX2 in 92 moderately to poorly differentiated human lung adenocarcinoma by χ2-test . Pearson R , Pearson correlation coefficient . ( I ) Correlation of mRNA levels of SOX2 and NFATc2 in a panel of lung AD cell lines analyzed by q-PCR and Pearson correlation test . ( J ) Expression of NFATc2 and SOX2 in A549 cells with or without NFATc2 overexpression and SOX2 stable knockdown by Western blot . ( K–L ) Effect of SOX2 knockdown on tumorsphere formation ( K ) , cell migration and invasion ability ( L ) , of A549 cells with NFATc2 overexpression . *p<0 . 05 , **p<0 . 01 versus control by t-test . Error bars indicate the mean ±SD for at least three independent replicates . ( M ) In vivo tumorigenicity of A549 cells with NFATc2 overexpression and SOX2 knockdown by subcutaneous inoculation of 1 × 104 cells in SCID mice . **p<0 . 0001 versus control by two-way ANOVA . Error bar indicates the mean ±SD of tumor volumes of six mice . ( N ) Effect of NFATc2 knockdown on SOX2 expression in A549 CR cells analyzed by immunoblot . ( L ) Effect of SOX2 knockdown on cisplatin sensitivity by MTT assay of A549 cells with NFATc2 overexpression . ##p<0 . 01 , versus vector control , **p<0 . 01 versus RFP-NFATc2_Sh-Ctrl by t-test . Error bar indicates the mean ±SD for at least three independent replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 26733 . 01910 . 7554/eLife . 26733 . 020Figure 6—source data 1 . Statistical analyses for Figure 6F and I . DOI: http://dx . doi . org/10 . 7554/eLife . 26733 . 02010 . 7554/eLife . 26733 . 021Figure 6—figure supplement 1 . NFATc2 regulated SOX2 expression through binding to 3’ regulatory regions . ( A ) Computational prediction of NFAT binding sites ( marked in red ) on 5’ and 3’ SOX2 regulatory regions ( Region 1 to 4 ) . TSS: transcription start site . ( B ) Transcriptional activities of the respective SOX2 regions 1–4 of H441 cells analyzed by luciferase reporter assays . *p<0 . 05 , **p<0 . 01 versus control by Student’s t-test . Error bars indicate the mean ±SD for at least three independent replicates . ( C–D ) Luciferase reporter activities of mutant or wild-type SOX2 reporters of A549 ( C ) , or H1299 cells ( D ) , with or without NFATc2 stable overexpression . *p<0 . 05 , **p<0 . 01 , comparison with RFP; # p<0 . 05 , ##p<0 . 01 , wild type versus mutant in RFP-NFATc2 cells by t-test . Error bars indicate the mean ±SD for at least three independent replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 26733 . 021 The clinical significance of NFATc2/SOX2 coupling was further assessed in human lung cancers . To avoid the confounding effect of SOX2 gene amplification in squamous cell carcinoma ( SCC ) , and to focus on tumors with demonstrated involvement of NFATc2 , we performed IHC analysis on 92 moderately to poorly differentiated AD . A significant correlation between SOX2 and NFATc2 expressions was observed ( Figure 6H ) . In lung AD cell lines , NFATc2 and SOX2 transcript expressions were also positively correlated ( Figure 6I ) . Together , the data supported NFATc2 upregulates SOX2 in clinical and cultured lung AD . Next , we evaluated whether NFATc2-induced SOX2 upregulation was functionally relevant for its role in sustaining TIC . Using A549 transduced for NFATc2 overexpression , SOX2 suppression by 2 shSOX2 sequences led to significantly reduced tumorspheres formation ( Figure 6J–K ) , and cell motility ( Figure 6L ) . In vivo , the xenografts enhanced by NFATc2 overexpression were also abrogated by SOX2 knockdown ( Figure 6M ) . Similar to NFATc2 , SOX2 was also upregulated in A549 induced for cisplatin resistance ( A549 CR ) but on NFATc2 knockdown , SOX2 levels were repressed ( Figure 6N ) , suggesting NFATc2/SOX2 coupling was functionally active in resistant cancer cells . Moreover , while NFATc2 overexpression induced cisplatin resistance of A549 cells , SOX2 silencing restored sensitivity to a level comparable to that of the control cells ( Figure 6L ) . Overall , the data indicated NFATc2 induces TIC , cancer initiating phenotypes and drug resistance through upregulating SOX2 expression . We have shown NFATc2 was upregulated in ALDH+/CD44+-TIC , suggesting NFATc2 might regulate this TIC population ( Figure 2B ) . With NFATc2 knockdown or knockout in HCC827 , ALDH+/CD44+-TIC were significantly reduced ( Figure 7A , B ) . Consistent changes were observed for PDCL#24 with NFATc2 knockdown ( Figure 7—figure supplement 1A ) . Further , inhibition of calcineurin by CSA and FK506 , respectively , also significantly reduced ALDH+/CD44+-TIC ( Figure 7C ) . Conversely , in both A549 with NFATc2 overexpression and in A549 CR cells , ALDH+/CD44+-TIC proportions were increased ( Figure 7D and Figure 7—figure supplement 1B ) . Breakdown analysis showed the trend of changes that were more consistent with those of the ALDH+ but not CD44+ population , suggesting ALDH might be the main target of NFATc2 . 10 . 7554/eLife . 26733 . 022Figure 7 . ALDH1A1 was a target of NFATc2/SOX2 coupling . ( A–D ) Effects on ALDH+ , CD44+ and ALDH+/CD44+ cell populations by flow cytometry analysis of HCC827 with NFATc2 knockdown ( A ) , NFATc2 knockout ( B ) , or NFATc2 inhibition by CSA or FK506 ( C ) , and of A549 cells with NFATc2 overexpression ( D ) . ( E–H ) Effects on ALDH1A1 mRNA expression by qPCR analysis of cancer cells with NFATc2 knockdown ( E ) , NFATc2 inhibition by CSA or FK506 ( F ) , NFATc2 up-regulation ( G ) , or of A549 with induced cisplatin resistance ( H ) . ( I ) Effects of SOX2 knockdown on ALDH+ , CD44+ and ALDH+/CD44+ cell populations by flow cytometry in A549 with NFATc2-overexpression . ( J ) Expression of SOX2 and ALDH1A1 transcripts in A549 cells with NFATc2 overexpression and SOX2 knockdown . ( K ) Representative images of A549 xenografts with or without NFATc2 overexpression immunohistochemically stained for NFATc2 , SOX2 and ALDH1A1 , respectively . Scale bars , 50 µm . ( L ) Conserved SOX2 binding sequences ( ATTCA ) at ALDH1A1 enhancer region by ChIP-seq of PDCL#24 cells , aligned with homologous mammalian sequences and H3K27Ac peaks of A549 cells from published databases . ( M ) Detection of endogenous SOX2 binding to ALDH1A1 sites by ChIP–qPCR analysis in PDCL#24 cells . ( N ) Luciferase reporter activities at sites 1 and 2 of ALDH1A enhancer region by dual luciferase reporter assay in A549 cells with SOX2 overexpression . ( O–P ) Effects of transient ALDH1A1 suppression on A549 with upregulated NFATc2 with respect to invasion and migration ( O ) , and cisplatin sensitivity . ##p<0 . 01 , versus RFP_Scramble by t-test . ( P ) . ( Q ) Correlation between ALDH1A1 and SOX2 expressions by IHC in human lung adenocarcinomas by χ2-test . *p<0 . 05 , **p<0 . 01 versus control by t-test . Error bar indicates the mean ±S . D . for at least three independent replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 26733 . 02210 . 7554/eLife . 26733 . 023Figure 7—source data 1 . Statistical analyses for Figure 7A , B , I and M . DOI: http://dx . doi . org/10 . 7554/eLife . 26733 . 02310 . 7554/eLife . 26733 . 024Figure 7—figure supplement 1 . NFATc2 regulated ALDH activity . ( A–C ) Flow cytometry analysis of ALDH/CD44 distribution in PDCL#24 cells with stable NFATc2 knockdown ( A ) , A549 cells with NFATc2 overexpression ( B ) , or A549CR compared to parental cells ( C ) . *p<0 . 05 , **p<0 . 01 versus control by t-test . Error bars indicate the mean ±SD for at least three independent replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 26733 . 02410 . 7554/eLife . 26733 . 025Figure 7—figure supplement 2 . NFATc2 regulated ALDH1A1 . ( A ) ALDH+ proportions analyzed by flow cytometry in A549 cells with NFATc2 overexpression and transient ALDH1A1 knockdown . ( B ) mRNA level of ALDH1A1 analyzed by qPCR in HCC827 cells with NFATc2 knockout . ( C ) Luciferase reporter activities for site1 and 2 of A549 cells with or without NFATc2 overexpression . **p<0 . 01 versus control by t-test . Error bars indicate the mean ±SD for at least three independent replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 26733 . 02510 . 7554/eLife . 26733 . 026Figure 7—figure supplement 3 . Effect of siALDH1A1 on ALDH1A1 expression . Analysis of ALDH1A1 expression by qPCR in A549 cells with ectopic NFATc2 expression and ALDH1A1 knockdown . **p<0 . 01 versus control by t-test . Error bar indicates the mean ±S . D . for 3 replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 26733 . 02610 . 7554/eLife . 26733 . 027Figure 7—figure supplement 4 . Effect of NFATc2/SOX2 on β-catenin activity . Expressions of total β-catenin , active β-catenin ( non-phosphorylated ) , and phosphorylated β-catenin ( p-β-catenin ) analyzed by immunoblot in A549 with or without NFATc2 overexpression and SOX2 knockdown . DOI: http://dx . doi . org/10 . 7554/eLife . 26733 . 027 ALDH1A1 is the most frequent and important ALDH isozyme reported in lung cancer TIC ( Ucar et al . , 2009; Tomita et al . , 2016 ) . Although ALDH1 is marketed as the major subtype contributing to ALDH activities detected by the ALDEFLUORTM assay , cross-reactivity with other isoforms cannot be excluded . Hence , to explore the part contributed by ALDH1A1 , we abrogated ALDH1A1 in A549 engineered to overexpress NFATc2 , which led to significantly suppressed aldefluor activities ( Figure 7—figure supplement 2A ) . ALDH1A1 expression was suppressed in NFATc2 knockdown or knockout cells , as well as in cells treated with CSA or FK506 , respectively ( Figure 7E–F , Figure 7—figure supplement 2B ) . Conversely , ALDH1A1 was up-regulated in NFATc2 overexpressing cells and A549 CR cells ( Figure 7G–H ) , indicating NFATc2 regulates ALDH1A1 expression which contributes to the majority of ALDH positivity in ALDH+/CD44+-TIC . Further analysis of the role of NFATc2/SOX2 coupling in ALDH1A1 regulation showed silencing SOX2 in NFATc2-overexpressing A549 cells consistently prevented the increase of ALDH+ and ALDH+/CD44+ subpopulations only ( Figure 7I ) . Specifically , the expected upregulation of ALDH1A1 was also abolished ( Figure 7J ) . In addition , as analyzed by IHC , xenografts derived from NFATc2-overexpressing A549 cells showed corresponding upregulation of SOX2 and ALDH1A1 ( Figure 7K ) . Together , the data strongly supported NFATc2/SOX2/ALDH1A1 form a regulatory axis in lung cancer . To examine the mechanism of SOX2 in ALDH1A1 regulation , ChIP-seq analysis was performed to identify SOX2 binding sequences in PDCL#24 cell line . Two loci of peak signals encompassing the SOX2 consensus binding motif ( ATTCA ) were identified in the 5’ region of the ALDH1A1 gene at around 27 kb from the TSS ( site 1 and 2 , Figure 7L ) . Bioinformatics analysis showed these loci were located at chr9:75 , 595 , 820–75 , 595 , 832 and chr9:75 , 602 , 860–75 , 602 , 872 which overlap with H3K27Ac occupancy deposited in public ChIP-seq database of A549 cells as well as mammalian conserved sequences ( Figure 7L ) . On the other hand , no SOX2 binding motif was identified in regions flanking other ALDH isoform genes . Together , the findings suggested the presence of an accessible and highly conserved chromatin region encompassing putative SOX2-binding motifs 5’ to the ALDH1A1 gene . To confirm , ChIP-q-PCR assay using PDCL#24 cells was performed which showed anti-SOX2 antibodies significantly enriched both sequences 1 and 2 ( Figure 7M ) . To study their transcriptional regulatory role , luciferase reporter constructs for sites 1 and 2 were co-transfected with SOX2 expression plasmids into A549 cells which yielded significantly enhanced reporter activities compared to control cells ( Figure 7N ) . Compatible results were obtained for A549 with NFATc2 overexpression ( Figure 7—figure supplement 2C ) . To evaluate whether ALDH1A1 is a functionally relevant target , ALDH1A1 was knocked-down by siRNA in NFATc2-overexpressing A549 cells ( Figure 7—figure supplement 3 ) . This led to significant suppression of cell motility ( Figure 7O ) and cisplatin sensitization ( Figure 5N ) . Furthermore , moderately to poorly differentiated human lung AD showed statistically significant positive correlation between SOX2 and ALDH1A1 expressions by IHC staining ( Figure 5O ) . Collectively , the data supported ALDH1A1 is a functional target of regulation through NFATc2/SOX2 coupling . Alleviation of oxidative stress induced by chemotoxicity promotes cancer cell survival and mediates drug tolerance . Thus , in A549 CR cells , intracellular ROS levels were significantly lower compared to parental A549 cells ( Figure 8A ) . To investigate for possible relation between NFATc2 and ROS modulation , NFATc2 was silenced by knockdown or knockout which led to increased ROS levels ( Figure 8B–D ) . As shown in Figure 8E , NFATc2 depletion sensitized PDCL#24 cells to cisplatin treatment , but the addition of the reducing agent NAC reversed cisplatin IC50 to above the control level dose-dependently . Reciprocally , the enhanced resistance of A549 by NFATc2 overexpression was reversed by oxidative stress induced by the glutathione inhibitor BSO ( Figure 8F ) , consistent with the suggestion that drug resistance by NFATc2 is effected through ROS attenuation . Similarly , ROS regulation also supported other tumor phenotypes mediated by NFATc2 . For example , tumorspheres suppression by NFATc2 knockdown was restored by NAC dose-dependently but in control cells , no significant changes were induced even in the presence of additional NAC ( Figure 8G–H ) . Likewise , cell migration and invasion efficiencies inhibited by NFATc2 depletion were reversed by NAC ( Figure 6I ) . To further address the involvement of SOX2 coupling and ALDH1A1 , we showed suppression of ROS by NFATc2-overexpression in A549 cells were reversed by silencing SOX2 or ALDH1A1 , respectively ( Figure 8J–K ) . Together , the data suggested NFATc2/SOX2/ALDH1A1 form a functional axis in the homeostatic regulation of an optimal level of ROS for in vitro tumorigenicity , cell motility , and mediation of drug resistance . 10 . 7554/eLife . 26733 . 028Figure 8 . NFATc2 regulated TIC properties through ROS suppression . ( A ) ROS levels detected by flow cytometry in A549 and A549 CR cells . ( B–C ) ROS levels in HCC827 cells ( B ) and PDCL#24 cells ( C ) with or without NFATc2 stable knockdown . ( D ) ROS levels in HCC827 cells with or without NFATc2 knockout . ( E–F ) Cisplatin sensitivity expressed as IC50 by MTT assays of NFATc2-silenced PDCL#24 cells treated with increasing doses of NAC ( E ) , or NFATc2-overexpressing A549 cells treated with the oxidizing agent BSO ( F ) , respectively . *p<0 . 05 , **p<0 . 01 versus vector control without REDOX reagents; ##p<0 . 01 versus the corresponding treatment control by t-tests . Error bar indicates the mean ±S . D . for three independent replicates . ( G–H ) Effects of increasing doses of NAC on tumorsphere formation ability of HCC827 ( G ) cells and PDCL#24 cells ( H ) . *p<0 . 05 , **p<0 . 01 versus corresponding treatment controls , ##p<0 . 01 versus vector control by t-test . Error bar indicates the mean ±S . D . for three independent replicates . ( I ) Effects of increasing doses of NAC on cell migration and invasion of HCC827 cells with NFATc2 down-regulation by 2 sh-RNA knockdown sequences . ( J–K ) ROS levels in NFATc2-overexpressing A549 cells with stable SOX2 ( J ) or transient ALDH1A1 ( K ) knockdown . *p<0 . 05 , **p<0 . 01 versus respective control by t-test . Error bar indicates the mean ±S . D . for at least three independent replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 26733 . 02810 . 7554/eLife . 26733 . 029Figure 8—source data 1 . Statistical analyses for Figure 8A–D , J and K . DOI: http://dx . doi . org/10 . 7554/eLife . 26733 . 029 The elucidation and disruption of TIC maintenance pathways offer the opportunity to eliminate the most resilient cancer cells and improve treatment outcome . Many studies have demonstrated cell populations expressing high levels of specific markers such as ALDH , CD44 , CD166 , CD133 , etc . are enhanced for a multitude of tumor phenotypes , with tumorigenicity and drug resistance being clinically the most important . Constitutive stem cell programs or stress-induced pathways are the main TIC sustaining mechanisms but details of their regulation are still elusive . NFAT is a family of transcription factors with the calcium-responsive parlors NFATc1 , -c2 , -c3 and -c4 being expressed in a tissue-dependent manner . In this study , using NFATc2 depletion and overexpression models of multiple lung cancer cell lines in TIC-defining functional assays ( Pattabiraman and Weinberg , 2014 ) , as well as clinical evidence from excised human lung cancers , we showed NFATc2 mediates TIC phenotypes . In vitro cell renewability was demonstrated by tumorspheres passaged for consecutive generations and and in vivo tumorigenicity tumorigenicity was illustrated by the limiting dilution assay . In clinical tumors , high level NFATc2 segregated with impaired tumor differentiation , advanced pathological stage , shorter recurrence-free and overall survivals in NFATc2-positive NSCLC , suggesting NFATc2 mediates the more primitive and aggressive tumor phenotypes . In the literature , only one research group has reported NFATc2 expression in 52% of 159 lung cancers and similar to our findings , high expression was associated with late tumor stage and poor survival . In their studies , supportive evidences for cell proliferation , invasion and migration were demonstrated by cell models but the in vivo role of NFATc2 and , in particular , its effects on TIC , drug response or mechanisms of action were not addressed ( Chen et al . , 2011; Liu et al . , 2013b ) . We have also evaluated public data sets on NFATc2 mRNA expression in lung cancer but a correlation with adverse patient outcome was not found ( data not shown ) . Since NFATc2 is expressed by tumor infiltrating leukocytes , specific conclusions cannot be reached using this approach . To identify the TIC sustaining mechanism of NFATc2 , we hypothesized NFATc2 might be coupled to the core pluripotency factors SOX2 , NANOG and/or OCT4 , whose aberrant activities , if present , would be most suitable to orchestrate multifaceted cancer propensities through extensive transcriptional and epigenetic reprograming . Using multiple analyses of cancer cells and tumorspheres , we observed SOX2 was the most consistently altered factor with the highest magnitude of change when NFATc2 expression or calcineurin activity were manipulated . SOX2 is an important oncogene for squamous cell carcinomas ( SCC ) of the lung and other organs through SOX2 locus amplification at 3q26 ( Hussenet et al . , 2010; Lu et al . , 2010; Boumahdi et al . , 2014 ) . For lung AD , although SOX2 is often expressed at high levels and predicts adverse survivals ( Chou et al . , 2013 ) , distinct genetic mechanisms have not been identified , indicating regulatory or signaling aberrations might be involved . To evaluate the effects of NFATc2 on SOX2 expression , we avoided the potentially confounding element of SCC and focused on moderate to poorly differentiated AD where NFATc2 is shown to play an important prognostic role . Indeed , SOX2 expression is significantly correlated with that of NFATc2 in this group of human lung cancer . Functionally , NFATc2/SOX2 coupling contributes to tumor behavior as depletion of SOX2 in cell lines with NFATc2 overexpression led to significant suppression of TIC phenotypes . Hence , clinical and experimental evidences support NFATc2 impedes tumor differentiation and negatively affects patient outcome through coupling to SOX2 in lung AD . We observed NFATc2 binds to the 3’ enhancer region of SOX2 at around 3 . 2 kb and 3 . 6 kb from TSS , effecting functional TIC enhancement and supporting the direct involvement of NFATc2 in stemness induction . In a recent study of a transgenic mouse model of pancreatic ductal adenocarcinoma with KRAS G12VD mutation and p53 heterozygous inactivation , Singh et al reported an analogous mechanism involving another NFAT family protein , NFATc1 , which acts as a coactivator and transcriptional regulator of SOX2 ( Singh et al . , 2015 ) . They showed NFATc1 played a permissive role for tumor dedifferentiation and expression of epithelial–mesenchymal transition ( EMT ) genes , while p53 disruption was essential for tumorigenesis , suggesting under the appropriate genetic context , NFATc1 activation transduces EMT through SOX2 upregulation . The authors proposed since chronic inflammation is a known etiology of pancreatic AD , NFATc1 might be a crucial factor involved in the progression of this cancer . On the other hand , we have shown NFATc1 inhibition does not result in SOX2 suppression , indicating NFATc2 is likely to be preferentially involved in lung cancers . Our data also distinguish a different SOX2 enhancer region accessible to NFATc2 which functions not only in the presence of KRAS mutation ( A549 , PDCL#24 ) but also in EGFR mutant ( HCC827 ) cancers . Chronic inflammation is an important etiological mechanism of lung cancer through release of ROS and free radicals from alveolar macrophages and neutrophils . While many studies modeling carcinogenetic mechanisms of tobacco toxicity or chronic obstructive pulmonary diseases have featured the NF-κB pathway as a major mediator , our findings on the effects of NFATc2 on TIC induction might add to this repertoire . In fact , NFATc2 and NF-κB share highly similar DNA binding domains but differ in the upstream activators , with NF-κB being stimulated by cytokine receptors and inflammatory molecules while NFATc2 is downstream of calcium signaling reputed as a stress response integrator , illustrating the multiple mechanisms through which inflammation-induced carcinogenesis might be initiated in the lung . We have shown the ALDH+/CD44+ fraction of lung cancer cells , despite being the smallest subset , demonstrates the highest tumorigenic capacity compared to counterpart subsets ( Liu et al . , 2013a ) . Evaluation of the role of NFATc2/SOX2 coupling in inducing this cell fraction revealed only ALDH but not CD44 showed consistent changes upon NFATc2 suppression or overexpression , respectively , and more specifically , ALDH1A1 was identified as a major functional target . Recent reports have shown β-catenin can directly regulate ALDH1A1 ( Condello et al . , 2015 ) , and since SOX2 can upregulate β-catenin ( Yang et al . , 2014 ) , this could infer SOX2 might indirectly regulate ALDH1A1 through β-catenin . We thus assessed β-catenin activity by western blot in A549 cells with NFATc2 overexpression and SOX2 knockdown , but no significant alternation of total or activated β-catenin levels was observed in these cells and the possible mechanism of indirect ALDH1A1 upregulation through β-catenin was not supported ( Figure 7—figure supplement 4 ) . On the other hand , computational screening did not detect significant and conserved SOX2-binding motifs within the proximal promoter region of ALDH1A1 , but a more distant locus at 27 kb upstream from its TSS was shown to be a probable response region which we confirmed by CHIP-seq , CHIP-qPCR and luciferase reporter assays , respectively . This illustrates a distal enhancer is involved in the trans-regulation of ALDH1A1 expression , which has not been reported before . Current views on cancer stem cells suggest TIC is unlikely to be a single population with unique identifiers; instead , cell plasticity might induce variant TIC populations through dynamic response mechanisms , enabling cancer cells to meet the requirements of a complex micro-environment . In this connection , since we have only demonstrated the NFATc2/SOX2/ALDH1A1 axis , further investigation for the mechanisms of CD44 regulation in ALDH+/CD44+-TIC is needed . Acquired drug resistance is mediated through complex genetic and molecular mechanisms ( Holohan et al . , 2013 ) . We have shown NFATc2 augments cancer resistance with more prominent effects on tumors treated by cytotoxic chemotherapy . Response to targeted therapy is dominated by addiction to aberrant signaling from the mutant EGFR , and the effects of NFATc2 are modest by comparison . Nevertheless , as shown by the xenograft model of short term gefitinib treatment , complementary NFATc2 inhibition might be considered for patients receiving sub-therapeutic treatment regimes , e . g . , due to the occurrence of serious skin or liver side effects . Adaptive antioxidant response to alleviate oxidative stress from ROS surge during systemic therapy is one of the most important mechanisms of drug resistance ( Zhang et al . , 2011 ) , suggested to be accentuated in cancer stem cells ( Diehn et al . , 2009; Achuthan et al . , 2011; Ishimoto et al . , 2011; Chang et al . , 2014 ) . Indeed , we have observed in multiple cell lines with induced resistance to chemotherapy or targeted therapy , NFATc2 was upregulated , while ROS was maintained at a lower level compared to parental cells . Changes in TIC phenotypes induced by NFATc2 up- or down-regulation were correspondingly restored by the redox reagents BSO or NAC , respectively , suggesting ROS scavenging is an important mechanism of drug resistance and other TIC properties mediated by NFATc2 . In line with this suggestion , it has been reported in adult immortalized bronchial epithelial cells , NFAT can be upregulated in response to inflammatory and carcinogenic stimulation such as benzo- ( a ) -pyrene and heavy metals ( Huang et al . , 2001; Ding et al . , 2007; Cai et al . , 2011 ) , leading to ROS-induced COX2 pathway signaling and enhanced cell survival ( Ding et al . , 2006 ) . However , whether ROS is in turn suppressed by NFATc2 through a negative feedback mechanism has not been reported . Our findings supplement this information and show NFATc2 facilitates ROS scavenging , and further implicates this effect is mediated by ALDH1A1 through SOX2 coupling . This is consistent with findings of other studies on the role of ALDH1A1 in drug resistance through repressing ROS level ( Singh et al . , 2013; Raha et al . , 2014; Mizuno et al . , 2015 ) . In summary , this study demonstrates the calcium signaling molecule NFATc2 enhances functional characteristics associated with cancer stemness phenotype . Our data reveal a novel mechanism of SOX2 upregulation in lung cancers through enhancer binding by NFATc2 . The NFATc2/SOX2/ALDH1A1 axis contributes to drug resistance by mediating a negative feedback mechanism for ROS scavenging and restoration of redox homeostasis . This study employed a candidate-based approach and the involvement of other potential NFATc2 and SOX2 targets in cancer phenotypes is not addressed . Nevertheless , the findings implicate NFATc2 is a potential therapeutic target for sequential or combination therapy of lung cancer that aims to eliminate TIC . Established cell lines ( H1993 , HCC1833 , H358 , H1650 , H2228 , H1299 , H1437 , H1975 , H23 , H2122 , HCC827 , HCC78 , A549 , H441 , and BEAS-2B ) were obtained from ATCC . HCC366 and HCC78 were kindly provided by Dr . J . Minna ( University of Texas Southwestern Medical Center . Dallas ) . All cell lines were kept as frozen aliquots upon receipt and only the first 20 passages were used in experiments . Patient-derived cell lines ( HKULC1 , HKULC2 , HKULC3 , HKULC4 , PDCL#24 , and FA31 ) were raised from resected primary lung cancers or malignant pleural effusions and only the 1st to 10th passages were used for study ( Lam et al . , 2006; Liu et al . , 2013a ) . Cancer cells were maintained in RPMI-1640 ( Invitrogen , Carlsbad , CA ) with 10% FBS ( Invitrogen , Carlsbad , CA ) . BEAS-2B were cultured in Keratinocyte-SFM ( Invitrogen , Carlsbad , CA ) . Gefitinib , paclitaxel or cisplatin-resistant ( -GR , TR or –CR , respectively ) cells were generated by chronic exposure of cancer cells to stepwise increased doses of the respective drugs . All procured cell lines used in this study were free of mycoplasma contamination and were authenticated using the AmpFlSTR Identifiler PCR Amplification Kit for short tandem repeat profiling according to the manufacturer’s instruction ( Thermo Fisher Scientific , Waltham , MA ) . None of the cell lines used in this study were included in the list of commonly misidentified cell lines maintained by the International Cell Line Authentication Committee . Small interfering RNA ( siRNA ) with pre-designed sequences targeting human NFATc1 , PPP3R1 , ALDH1A1 and scramble siRNA were from Sigma-Aldrich ( St Louis , MO ) . pGL3-NFAT luciferase ( 17870 ) , two shRNA sequences targeting SOX2 , pLKO . 1 Sox2 3HM a ( 26353 ) and pLKO . 1 Sox2 3 hr b ( 26352 ) , the negative control vector pLKO . 1-puro ( 1864 ) , the envelope vector pMD2 . G ( 12259 ) and packaging vector psPAX2 ( 12260 ) were purchased from Addgene ( Cambrige , MA; http://www . addgene . org ) . The pLKO . 1-lentiviral shRNA with different inserts specifically targeting NFATc2 were purchased from Sigma-Aldrich ( TRCN0000016144 , TRCN0000230218 ) . Human full length NFATc2 were amplified by PCR , and the RFP-NFTAc2 plasmids were generated by cloning the sequences into PCDH-CMV-MCS-EF1-COPRFP vector ( SBI , Mountain View , CA ) . For luciferase reporter construction , SOX2 regulatory regions were amplified by PCR from human genomic DNA and cloned into pGL3 ( Promega ) to generate the SOX2-luc constructs . Primers used for genomic DNA amplification were listed in Supplementary file 1A . Site directed mutagenesis of the consensus NFAT binding site ( GGAAA to GACTA ) were performed using QuikChange ( Stratagene ) . Lentiviral shRNA was produced by transfecting the shRNA , envelope and packaging vectors into 293 T cells using lipofectamine 2000 ( Invitrogen , Carlsbad , CA ) . Viruses were harvested after 48 hr of transfection followed by infection of target cells for 72 hr . Cells stably expressing shRNA were selected using puromycin ( Sigma-Aldrich ) for 14 days after 72 hr of viral infection . RFP-NFATc2 lentiviral particles were produced and transduced into target cells using Lenti Starter kit ( SBI , Mountain View , CA ) according to manufacturer’s instructions . RFP-positive cells stably over-expressing NFATc2 and SOX2 were selected by FACS using BD Aria ( BD Biosciences ) . LentiCas9-Blast and lentiGuide-Puro were purchased from Addgene ( Cambrige , MA; http://www . addgene . org ) . The gRNA targeting NFATc2 was designed using Zifit ( http://zifit . partners . org/ZiFiT/ ) and listed in Supplementary file 1A . The annealed gNFATc2 oligonucleotides were cloned into lentiGuide-Puro . Lenti-viral cas9 and lenti-viral gNFATc2 were generated by transfecting lentiCas9-Blast or lenti-viral gNFATc2 together with pMD2 . G and psPAX2 , respectively , into 293FT cells by lipofectamine 2000 according to published protocols ( Sanjana et al . , 2014 ) . After infection of lenti-viral cas9 , cells stably expressing Cas9 were selected using Blasticidin ( Sigma-Aldrich ) for 10 days . HCC827-Cas9 cells were further infected with lenti-gNFATc2 virus for 72 hr , and cells stably expressing gNFATc2 were selected using puromycin ( Sigma-Aldrich ) for 14 days . ALDH activity was analyzed by the Aldefluor kit ( Stem Cell Technologies ) according to manufacturer’s instructions . CD44 expression was stained by anti-CD44-APC ( BD Pharmingen ) as previously described ( Liu et al . , 2013a ) . Flow cytometry was performed using FACS Canto II ( BD Biosciences ) and data were analyzed using FlowJo ( Tree star ) . RFP positive cells with NFATc2 over-expression were isolated by FACS using BD Aria ( BD Biosciences ) . Sorted cells were re-analyzed after collection to ensure a purity of >95% . Non-viable cells were identified by propidium iodide inclusion . Cells were harvested , washed once in PBS and fixed in 1 ml cold 70% ethanol for 1 hr at 4°C . The fixed cells were washed twice with PBS . Then 50 µl of RNase A solution ( 100 µg/ml ) and 200 µl of propidium iodide ( 50 µg/ml ) were added to the cell pellet . Cells were incubated at room temperature for 10 min . Fluorescence was measured by flow cytometry ( FACS Canto II Analyzer , BD Biosciences ) and data were analyzed using FlowJo ( Tree star ) . BrdU assay was performed using BrdU cell proliferation assay kit ( Cell Signaling , Beverly , MA ) according to manufacturer’s instructions . Briefly , 5000 cells were seeded in a 96-well plate and incubated overnight followed by adding 10 µM BrdU and incubation for 12 hr . After the medium was removed , cells were fixed by 100 µl/well of fixing solution for 30 min at room temperature . BrdU was detected by 100 µl/well of 1X detection antibody solution followed by 1X HRP-conjugated secondary antibody solution . Then , 100 µl TMB substrate was added to each well and incubated at room temperature for 30 min followed by 100 µl of stop buffer . The absorbance was read at 450 nm using a plate spectrophotometer . Five hundred cells were seeded in an ultra-low plate ( Costar ) and cultured in cancer stem cell medium ( RPMI-1640 medium supplemented with 20 ng/mL FGF , 20 ng/mL EGF , 40 ng/mL IGF and 1X B27 ( Invitrogen , Carlsbad , CA ) for 14 days . Tumorspheres were harvested , dissociated with trypsin , re-suspended in RPMI-1640 , and 500 cells were seeded again for second passage using the same stem cell culture conditions . The migration and invasion assays were performed using Corning Transwell . Both chambers were filled with RMPI-1640 medium , and the lower chamber was supplemented with 10% FBS . For the migration assay , 5 × 104 cells were seeded into the upper chamber and allowed to migrate for 24 hr . For the invasion assay , the upper chamber was first coated with Matrigel ( BD Pharmingen ) ; 1 × 105 cells were seeded and allowed to invade for 24 to 36 hr . Cells that migrated or invaded to the lower surface of the transwells were fixed with methanol and stained with crystal violet . Cell densities were photographically captured in three random fields . The dye on the transwell membrane was dissolved by 10% acetic acid , transferred to a 96 well plate , and the dye intensity was measured by a plate spectrophotometer at 570 nm . Drug sensitivity was tested by MTT assays . 6000 cells per well were seeded into 96-well plates and incubated for 24 hr at 37°C , followed by exposure to gefitinib ( Selleckchem Houston , TX ) , paclitaxel ( Sigma-Aldrich , St Louis , MO ) , or cisplatin ( Sigma-Aldrich , St Louis , MO ) at various concentrations for 72 hr with or without CSA ( Selleckchem , Houston , TX ) , NAC or BSO ( Sigma-Aldrich , St Louis , MO ) . Subsequently , Thiazolyl Blue Tetrazolium Bromide ( MTT ) ( Sigma-Aldrich , St Louis , MO ) was added and the mixture was incubated at 37°C for 4 hr . The absorbance was read at 570 nm using a plate spectrophotometer . The drug response curve was plotted and IC50 was calculated using nonlinear regression model by GraphPad Prism 7 . 0 . Total RNA was isolated using RNAiso Plus reagent ( Takara , Mountain View , CA ) and complementary DNA ( cDNA ) was generated using PrimeScript RT Reagent Kit ( Takara , Mountain View , CA ) according to the manufacturer’s instructions . Gene mRNA levels were analyzed by quantitative RT-PCR ( qPCR ) ( 7900HT , Applied Biosystems , Carlsbad , CA ) and SYBR green ( Qiagen , Hilden , Germany ) detection . Average expression levels of RPL13A and beta-2-microglobulin ( B2M ) were used as internal controls . Primers were listed in Supplementary file 1B . Cells were harvested and lysed on ice by lysis buffer [50 mM Tris HCl pH 7 . 4 , 1% Triton X-100 , 1 mM EDTA , 150 mM NaCl , 0 . 1% SDS , with freshly added 1:50 Phosphatase Inhibitor Cocktail 2 ( Sigma ) , 1:50 Protease Inhibitor Cocktail ( Sigma ) ] for 30 min . The cell lysate was then centrifuged at 13 k rpm for 20 min at 4°C to remove cell debris . The protein amount was quantified by the Dc Protein Assay ( Bio-Rad ) . Cell lysates were resolved by 6–10% SDS-PAGE and then transferred onto PVDF membranes ( Millipore ) . Primary antibodies including SOX2 ( 1:1000 ) , NFATc2 ( 1:1000 ) , β-catenin ( 1:1000 ) , non p-β-catenin ( 1:1000 ) , p- βcatenin ( 1:1000 ) or ACTIN ( 1:1000 ) ( Cell Signaling , Beverly , MA ) , respectively , where appropriate , were added . After overnight incubation , the membrane was washed with PBS and then incubated with the anti-rabbit secondary antibody . Target proteins on the membrane were visualized on X-ray films using ECL Plus Western Blotting Detection Reagents ( Amersham , Buckinghamshire , UK ) . ChIP assay was performed using the Magna ChIPTM A kit ( Millipore , Billerica , MA ) according to manufacturer’s instructions . Briefly , cells were sonicated and lysed after protein/DNA cross-linking by 1% formaldehyde for 10 min . The crosslinked complex was immuno-precipitated by anti-NFATc2 antibody or control rabbit IgG ( Cell Signaling , Beverly , MA ) bound to protein A magnetic beads . After overnight incubation at 4°C , the complex was eluted and DNA was purified . The immune-precipitated DNA was quantified by qPCR using primer sequences designed to detect specific regulatory regions listed in Supplementary file 1B . ChIP assay was performed using the EZ-Magna ChIP A/G Chromatin Immunoprecipitation Kit ( Millipore , 17–10086 ) according to manufacturer’s instructions . Cells were cultivated and treated with 1% formaldehyde to crosslink protein and DNA . Cell lysate was sonicated to reduce the DNA length to 100 to 500 bp . The DNA-protein fragments were then incubated with 10 ug SOX2 antibodies ( Abcam ) and magnetic beads coated with protein A/G to form DNA-protein-antibody complex . The DNA was isolated and purified by Spin column and sent for commercial ( BGI ) library construction and sequencing using Illumina Hi-Seq platforms . Sequence reads were aligned to Human Reference Genome ( hg19 ) using Bowtie ( Langmead et al . , 2009 ) . Model-based analysis of ChIP-Seq ( MACS ) was used for peaks identification by comparing ChIP sample over input sample with default parameters ( Zhang et al . , 2008 ) . The 5’- and 3’- flanking regions ( −5000 to +5000 bp ) of SOX2 were scanned for NFAT binding sequences using PWMSCAN ( Levy and Hannenhalli , 2002 ) . The significance of the predicted sites was evaluated statistically using a permutation-based method and comparison with occurrence of the motif in background genomic sequences of intergenic regions . Phylogenetically non-conserved binding sites were filtered ( Li et al . , 2010 ) . Cells were transfected with luciferase reporters , expression plasmids and pRL-TK vector using lipofectamine 2000 ( Invitrogen , Carlsbad , CA ) . Luciferase activities were measured by using the Dual-Luciferase Reporter Assay System ( Promega ) . All animal experiments were performed after approval by the Animal Ethics Committee , the University of Hong Kong according to issued guidelines . Briefly , different numbers of cells mixed with an equal volume of matrigel ( BD Pharmingen ) were injected subcutaneously at the back of 6 week old severe combined immunodeficiency ( SCID ) mice or Ncr-nu/nu-nude mice . Tumor sizes were monitored every 3 days using digital vernier calipers , and tumor volumes were calculated using the formula [sagittal dimension ( mm ) ×cross dimension ( mm ) 2]/2 and expressed in mm3 . Cells with or without respective treatments were washed with PBS and stained with 1 µM of the ROS probe CellROXTM Deep Red ( Lift Technologies ) for 30 mins according to manufacturer’s instructions . Fluorescence was measured by flow cytometry ( FACSCanto II Analyzer , BD Biosciences ) and data were analyzed using FlowJo ( Tree star ) . Surgically resected primary human NSCLC and corresponding normal lung tissues were collected prospectively in the Queen Mary Hospital , University of Hong Kong . Tissue collection protocols were approved by the Joint Hospital and University Institutional Review Board and written informed consents from patients were obtained . Fresh tissues were snap-frozen within 45–60 min after vascular clamping and kept in −70°C until use . Adjacent tumor tissues were fixed in 4% neural buffered formalin for 24 hr and processed into formalin fixed , paraffin embedded ( FFPE ) tissue blocks . Tumor classification and differentiation grading was according to the WHO classification of lung tumors , 2004 . Tumor typing and pathological staging was performed by a qualified anatomical pathologist ( MPW ) . Clinical parameters and outcomes were charted from hospital records in consultation with relevant clinicians . Tissue microarrays were constructed using at least 5 cores of tissue from different representative tumor areas and 1 core of corresponding normal lung from each case . Tumor cores were randomly arranged in the microarray to prevent positional bias during recording of IHC results . De-paraffinized tissue microarray sections ( 5 µm ) were subjected to antigen retrieval using microwave heating at 95°C in 1 mM EDTA buffer , pH 8 . 0 . Endogenous peroxidase was quenched with 3% hydrogen peroxide for 10 min . Blocked sections were labeled with primary antibodies against NFATc2 ( 1:50 dilution , Cell Signaling ) , SOX2 ( 1:200 dilution , Cell Signaling ) and ALDH1A1 ( 1:1000 dilution , Abcam ) overnight at 4°C . Anti-rabbit HRP-labeled polymer ( DAKO ) was used as a secondary antibody . Color detection was performed by liquid DAB +substrate chromogen system ( DAKO ) . Protein expression levels were semi-quantitatively analyzed using an automated image capturing and analysis system ( Aperio ) . NFATc2 expression level was scored according to the extent and intensity of nuclear staining in the tumor cells only and expression in the cytoplasm , stromal or inflammatory cells was excluded from evaluation . The intensity was graded as 1 , 2 , or 3 according to whether nuclear staining was absent or weak , moderate , or strong , respectively . The staining extent was graded as 1 , 2 , or 3 according to whether expression was observed in scattered individual cells , aggregates of 5 or more but <19 cells , or sheets of 20 or more cells . The products of the 2 grades were then computed , and cases with scores of 4 and above were counted as high level expression . Data were analyzed by SPSS ( version 16 . 0; SPSS Inc . , Chicago , IL , USA ) , GraphPad Prism 7 . 0 or Excel ( Microsoft , Redmond , WA , USA ) software packages and shown as mean ±standard deviations ( s . d . ) . Differential expression between paired tumor/normal tissues were analyzed by Wilcoxon text . Differences between groups were analyzed by t test for continuous variables . Differences between growth curves of xenograft model were analyzed by two-way ANOVA . Correlation between NFATc2 and SOX2 mRNA level were analyzed by Pearson correlation test . Correlation between NFATc2 , SOX2 , ALDH1A1 expressions and clinicopathological variables in lung cancers were analyzed by the χ2-test . Association between NFATc2 expression and overall survival and recurrence-free survival were analyzed by the Kaplan–Meier method with log-rank test . Multivariate survival analyses were performed by Cox regression model . Two-sided p values < 0 . 05 were considered as being statistically significant .

Cancer develops when cells become faulty and start to grow uncontrollably . They eventually form lumps or tumors , which may spread to surrounding tissues or even to other areas in the body . One of the reasons why cancer treatment remains a challenge is that there are over 200 types of cells in the body , and there are a lot of moments in the life cycle of a cell when things could go wrong . Researchers have shown that many cancers , including lung cancer , are not only extremely different from patient to patient , but also display great differences between cancer cells within the same tumor . Increasing evidence suggest that these differences may be caused by a type of cells called tumor initiating cells , or TICs for short . These TICs behave like stem cells and can renew themselves or mature into different types of cells . They are thought to help cancers grow and spread , and even make them resistant to treatments . Previous research has shown that in many types of cancer , the protein NFATc2 helps cancer cells to grow and spread . Until now , however , it was not known if NFATc2 is also important in TICs in lung cancer . Using human lung cancer cell lines and animal models , Xiao et al . show that the protein NFATc2 stimulates the stem-cell like behavior of TICs . The results showed that TICs had higher levels of the NFATc2 protein than other lung cancer cells that were not TICs . Tumors with higher levels were also more aggressive . When NFATc2 was removed from the cells , they formed smaller tumors and were more sensitive to drug treatment compared to cancer cells with NFATc2 . Further experiments revealed that NFATc2 helped to increase the levels of a protein called Sox2 , which gives cells the ability to renew or develop into different cell types . Together , these two proteins stimulated the production of another protein that was already known to play a crucial role in TIC maintenance . A better understanding of the mechanisms regulating TICs in lung cancer will help scientists tackle new questions about how this cancer progresses and resists to therapy . In the longer-term , combining classic cancer treatments with new therapeutic strategies targeting NFATc2 could make treatments for lung cancer patients more effective .

2017

NFATc2 enhances tumor-initiating phenotypes through the NFATc2/SOX2/ALDH axis in lung adenocarcinoma

The replicative lifespan ( RLS ) of a cell—defined as the number of cell divisions before death—has informed our understanding of the mechanisms of cellular aging . However , little is known about aging and longevity in symmetrically dividing eukaryotic cells because most prior studies have used budding yeast for RLS studies . Here , we describe a multiplexed fission yeast lifespan micro-dissector ( multFYLM ) and an associated image processing pipeline for performing high-throughput and automated single-cell micro-dissection . Using the multFYLM , we observe continuous replication of hundreds of individual fission yeast cells for over seventy-five generations . Surprisingly , cells die without the classic hallmarks of cellular aging , such as progressive changes in size , doubling time , or sibling health . Genetic perturbations and drugs can extend the RLS via an aging-independent mechanism . Using a quantitative model to analyze these results , we conclude that fission yeast does not age and that cellular aging and replicative lifespan can be uncoupled in a eukaryotic cell . Aging is the progressive decrease of an organism’s fitness over time . The asymmetric segregation of pro-aging factors ( e . g . , damaged proteins and organelles ) has been proposed to promote aging in mitotically active yeast and higher eukaryotes ( Bufalino et al . , 2013; Erjavec et al . , 2008; Henderson et al . , 2014; Katajisto et al . , 2015 ) . In budding yeast , asymmetric division into mother and daughter cells ensures that a mother cell produces a limited number of daughters over its replicative lifespan ( RLS ) ( Mortimer and Johnston , 1959 ) . Aging mother cells increase in size , divide progressively more slowly , and produce shorter-lived daughters ( Mortimer and Johnston , 1959; Bartholomew and Mittwer , 1953; Kennedy et al . , 1994 ) . Mother cell decline is associated with asymmetric phenotypes such as preferential retention of protein aggregates , dysregulation of vacuole acidity , and genomic instability ( Henderson et al . , 2014; Aguilaniu et al . , 2003; Saka et al . , 2013 ) . By sequestering pro-aging factors in the mothers , newly born daughters reset their RLS ( Henderson et al . , 2014; Aguilaniu et al . , 2003; Kwan et al . , 2013 ) . These observations raise the possibility that alternate mechanisms may be active in symmetrically dividing eukaryotic cells . The fission yeast Schizosaccharomyces pombe is an excellent model system for investigating RLS and aging phenotypes in symmetrically dividing eukaryotic cells . Fission yeast cells are cylindrical , grow by linear extension , and divide via medial fission . After cell division , the two sibling cells each inherit one pre-existing cell tip ( old-pole ) . The new tip is formed at the site of septation ( new-pole ) . Immediately after division , new growth is localized at the old-pole end of the cell . Activation of growth at the new-pole cell tip occurs ~30% through the cell cycle ( generally halfway through G2 ) . This transition from monopolar to bipolar growth is known as new end take-off ( NETO ) ( Mitchison and Nurse , 1985; Sveiczer et al . , 1996; Martin and Chang , 2005 ) . Prior studies of fission yeast have yielded conflicting results regarding cellular aging . Several papers reported aging phenotypes akin to those observed in budding yeast ( e . g . , mother cells become larger , divide more slowly , and have less healthy offspring as they age ) ( Erjavec et al . , 2008; Barker and Walmsley , 1999 ) . However , a recent report used colony lineage analysis to conclude that protein aggregates are not asymmetrically distributed , and that inheriting the old cell pole or the old spindle pole body during cell division does not lead to a decline in cell health ( Coelho et al . , 2013 ) . However , this report tracked the first 7–8 cell divisions of microcolonies on agar plates and thus could not observe the RLS of single cells ( Coelho et al . , 2013 ) . The controversy between these studies may partially stem from the difficulty in tracking visually identical cells for dozens of generations . Replicative lifespan assays require the separation of cells after every division . This is traditionally done via manual micro-dissection of sibling cells on agar plates , a laborious process that is especially difficult and error-prone for symmetrically dividing fission yeast . Extrinsic effects related to using a solid agar surface may confound observations made under these conditions ( Mei and Brenner , 2015 ) . Finally , recent work using high-throughput microfluidic devices to study individual budding yeast and bacterial cells ( Lee et al . , 2012; Crane et al . , 2014; Wang et al . , 2010; Liu et al . , 2015; Jo et al . , 2015; Nobs and Maerkl , 2014; Tian et al . , 2013; Huberts et al . , 2014; Minc and Chang , 2010 ) has shown that large sample sizes are needed to truly capture cellular lifespan accurately – populations less than ~100 cells do not reliably estimate the RLS ( Huberts et al . , 2014 ) . Here , we report the first high-throughput characterization of both RLS and aging in fission yeast . To enable these studies , we describe a microfluidic device—the multiplexed fission yeast lifespan microdissector ( multFYLM ) —and a software analysis suite that capture and track individual S . pombe cells throughout their lifespan . Using this platform , we present the first quantitative replicative lifespan study in S . pombe , settling a long-standing controversy regarding whether this organism undergoes replicative aging ( Erjavec et al . , 2008; Barker and Walmsley , 1999; Coelho et al . , 2013; Minois et al . , 2006 ) . The RLS of fission yeast is substantially longer than previously reported ( Erjavec et al . , 2008; Barker and Walmsley , 1999 ) . Remarkably , cell death is stochastic and does not exhibit the classic hallmarks of cellular aging . Despite the lack of aging phenotypes , rapamycin and Sir2p overexpression both extend RLS , whereas increased genome instability decreases RLS . These results demonstrate that RLS can be extended without any aging phenotypes , a fact that has implications for the proposed mechanism of lifespan extension by these interventions . Using these results , we also describe a quantitative framework for analyzing how stochastic and age-dependent effects contribute to the experimentally measured replicative lifespan . This framework will be broadly applicable to future replicative lifespan studies of model organisms . We conclude that fission yeast dies primarily via a stochastic , age-independent mechanism . Replicative lifespan assays require the separation of cells after every division . This is traditionally done via manual micro-dissection of sibling cells , a laborious process that is especially difficult for symmetrically dividing fission yeast . We recently developed a microfluidic platform for capturing and immobilizing individual fission yeast cells via their old cell tips ( Spivey et al . , 2014 ) . However , our first-generation device could only observe a single strain per experiment , required an unconventional fabrication strategy , and suffered from frequent cell loss that ultimately shortened the observation time . To address these limitations , we developed a multiplexed fission yeast lifespan microdissector ( multFYLM , Figure 1 ) , along with a dedicated software package designed to streamline the analysis and quantification of raw microscopy data . Single cells are geometrically constrained within catch channels , preserving the orientation of the cell poles over multiple generations ( Figure 1A–B ) . The cells divide by medial fission , thereby ensuring that the oldest cell pole is retained deep within the catch channel . If new-pole tips are loaded initially , then these outward-facing tips become the old-pole tips after the first division . 10 . 7554/eLife . 20340 . 003Figure 1 . A multiplexed fission yeast lifespan microdissector ( multFYLM ) . ( A ) Illustration of multFYLM ( gray ) loaded with fission yeast ( orange ) . Blue arrows represent the media flow though the multFYLM . ( B ) Left: Fission yeast cells initially grow from the old-pole end ( magenta ) . After new end takeoff ( NETO ) , growth begins at the new-pole end ( green ) . Right: multFYLM permits tracking of the old-pole cell , as well as its most recent siblings . ( C ) Optical image of a multFYLM showing six independent subsystems . Arrows indicate direction of media flow . Scanning electron micrographs of ( D ) a multFYLM subsystem and ( E ) a single catch channel . The channel is long enough to accommodate the old-pole cell , as well as the most recent new-pole sibling . ( F ) White-light microscope image of a row of catch channels loaded with cells . ( G ) Time-lapse images ( H ) and single-cell traces of a replicating cell . The old-pole ( magenta ) is held in place while the new-pole ( green ) is free to grow . The old-pole of the most recent sibling ( black ) extends until it is removed by flow into the central trench ( after ~2 hr 30 m ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20340 . 00310 . 7554/eLife . 20340 . 004Figure 1—figure supplement 1 . Schematic of the multiplexed fission yeast lifespan microdissector ( multFYLM ) . ( A ) The parallel subunits are designed to have the same fluidic resistance . This ensures similar flow rates through all devices when equal pressure is applied . ( B ) Detail showing the arrangement of central and side trenches of a single subunit . ( C ) Detail showing arrangement of the catch channels relative to the central and side trenches . ( D ) Detail showing the dimensions of one catch channel . DOI: http://dx . doi . org/10 . 7554/eLife . 20340 . 00410 . 7554/eLife . 20340 . 005Figure 1—figure supplement 2 . Loading and retention of cells in the multFYLM . ( A ) Loading efficiency of each subunit in the multFYLM . Error bars: S . D . of the mean of three loading experiments . ( B ) Retention efficiency of cells in the multFYLM under constant media flow . Three representative experiments shown , with the associated number of loaded cells . We observed retention efficiencies as high as 99% over 140 hr . ( C ) Image of a single field of view of one subunit . White triangles mark captured cells . Scale bar: 25 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 20340 . 00510 . 7554/eLife . 20340 . 006Figure 1—figure supplement 3 . Experimental apparatus and image processing workflow . ( A ) Schematic of experimental apparatus . Blue lines represent the flow of media , red lines represent temperature control , and black lines represent computer-controlled mechanical processes . A syringe pump delivers a constant flow of fresh media to the multFYLM . ( B ) FYLM Critic software workflow . Image stacks are first registered to remove jitter associated with moving the microscope stage . The user then selects cells of interest with a graphical user interface ( GUI ) , and the software generates kymographs . Kymographs are annotated via a semi-automated GUI with the user helping to trace the growth of each cell . The software then quantifies the trace data and exports associated white-light and fluorescence intensities for downstream processing in MATLAB . DOI: http://dx . doi . org/10 . 7554/eLife . 20340 . 006 The multFYLM consists of six closely spaced , independent microfluidic subsystems , with a total capacity of 2352 cells ( Figure 1C , Figure 1—figure supplement 1 ) . To ensure equal flow rates throughout the device , each of the six subsystems is designed to have the same fluidic resistance . Each subsystem consists of eight parallel rows of 49 cell catch channels ( Figure 1D ) . The catch channel dimensions were optimized for loading and retaining wild type fission yeast cells ( Figure 1E , Figure 1—figure supplement 2; ( Spivey et al . , 2014 ) . The eight rows of cell catch channels are arranged between a large central trench ( 40 µm W x 1 mm L x 20 µm H ) and a smaller side trench ( 20 µm W x 980 µm L x 20 µm H ) . Cells are drawn into the catch channels by flow from the central trench to the side trenches via a small ( 2 µm W x 5 µm L x ~12 µm H ) drain channel ( Figure 1F ) . We did not monitor all 2352 catch channels during our experiments , but we routinely filled >80% of the 224 monitored catch channels in each of the six microfluidic sub-systems ( Figure 1—figure supplement 2A ) . A constant flow of fresh media supplies nutrients , removes waste , and ensures that cells are stably retained for the duration of the experiment . Once loaded , up to 98% of the cells could be retained in their respective catch channels for over 100 hr ( Figure 1—figure supplement 2B–C , Video 1 ) . As a cell grows and divides , the old-pole cell is maintained by flow at the end of the catch channel , while the new-pole cells grow towards the central trench . Eventually , the new-pole cells are pushed into the central trench , where they are washed out by constant media flow ( Video 1 ) . This allows for continuous , whole-lifetime observation of the old-pole cell as well as its newest siblings ( Figure 1G–H ) . 10 . 7554/eLife . 20340 . 007Video 1 . Operation of the multFYLM . Time-lapse imaging of fission yeast in a single field of view of the multFYLM for approximately 140 hr . Scale bar is 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 20340 . 007 An inverted fluorescence microscope with a programmable motorized stage allowed the acquisition of single cell data with high-spatiotemporal resolution ( Figure 1—figure supplement 3A ) . Following data collection , images are processed through the image analysis pipeline ( see Supplemental Methods ) , which measures the length and fluorescence information for each cell at each time point ( Figure 1—figure supplement 3B ) . These data are parsed to determine the cell’s size , division times , growth rates , lifespan , and fluorescence information ( if any ) . Cells within the multFYLM grew with kinetics and morphology similar to cells in liquid culture and showed growth and NETO rates ( Figure 2—figure supplement 1 , Supplementary file 1A ) similar to those reported previously ( Coelho et al . , 2013; Nobs and Maerkl , 2014; Forsburg and Rhind , 2006 ) . In sum , the multFYLM allows high content , continuous observation of individual fission yeast cells over their entire lifespans . We used the multFYLM to measure the fission yeast RLS ( Figure 2 ) . From these data , we plotted the replicative survival curve ( Figure 2A ) and determined the survival function , S ( g ) , which describes the probability of being alive after generation g . Using the survival data , we also computed the hazard function , λ ( g ) =−dS ( g ) dg*S ( g ) −1 , which describes the instantaneous risk of death after cell division g ( Figure 2B ) . The hazard rate ( also called the death rate ) can be calculated for any generational age using this function . Surprisingly , the fission yeast survival curve did not fit the traditional aging-dependent Gompertz model , ( Gompertz , 1825; Greenwood , 1928; Wilson , 1993 ) , which describes survival and hazard in terms of a generation-dependent ( i . e . , aging ) and a generation-independent term ( Equation ( 2 ) in Materials and methods ) . The RLS data were best described by a single exponential decay , corresponding to a generation-independent hazard rate . Strikingly , the hazard rate does not increase as the replicative age increases; instead , it remains steady at an average ~2% chance of death per cell per generation . For comparison , we also analyzed the survival data and hazard function for budding yeast ( S . cerevisiae ) grown in a microfluidic device ( Jo et al . , 2015 ) . As expected , the budding yeast hazard function increases with each generation and fits the aging-dependent Gompertz model . Thus , the replicative age , g , strongly influences the probability of death in budding yeast , but not in fission yeast . 10 . 7554/eLife . 20340 . 008Figure 2 . The fission yeast replicative lifespan ( RLS ) . ( A ) Survival curves for wild-type S . pombe ( black ) and wild-type S . cerevisiae ( brown , data from Jo et al . , 2015 ) ; both were grown in microfluidic microdissection devices . Numbers indicate the average lifespan . Red lines are a fit to a Gompertz ( S . cerevisiae ) and exponential decay ( S . pombe ) survival models . Shading indicates 95% confidence interval ( C . I . ) . Dashed blue line: 50% survival . ( B ) Hazard rate curves for the data shown in ( A ) . The hazard rate increases dramatically with increased replicative age for S . cerevisiae but not for S . pombe . DOI: http://dx . doi . org/10 . 7554/eLife . 20340 . 00810 . 7554/eLife . 20340 . 009Figure 2—figure supplement 1 . Health of cells in the multFYLM . ( A ) Histogram of the doubling times and ( B ) division lengths of wild-type S . pombe ( h- 972 ) observed in the multFYLM . The doubling time was 2 . 05 ± 0 . 45 hr ( mean ± S . D . , n = 13 , 453 cells ) . The mean length at division was 16 ± 2 . 2 µm . ( C ) Four representative examples of new end take-off ( NETO ) . Each graph shows the normalized increase in the length of a randomly selected cell over a normalized time comprising one generation . ( D ) Doubling time did not generally change with age . Horizontal error bars are the SD of the mean time at division for a given generation . Number of cells used ranged from n = 517 at generation 1 , to n = 147 at generation 53 . The number of cells declines due mostly to cell death during the experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 20340 . 00910 . 7554/eLife . 20340 . 010Figure 2—figure supplement 2 . Survival and hazard curves for wild type fission yeast isolates . ( A ) Survival curves of several wild-type strains . The replicative half-life is indicated in parenthesis . Red lines: exponential decay fit to the data . For the two particularly long-lived strains NCYC132 and JB760 , we could only estimate a lower bound on the mean replicative lifespan . ( B ) Hazard functions of strains shown in ( A ) . Inset: hazard rates normalized to the wild-type ( h- 972 ) hazard rate ( error bars indicate 95% CI ) . The constant hazard rates illustrate that none of the WT populations observed are aging . DOI: http://dx . doi . org/10 . 7554/eLife . 20340 . 010 The RLS is defined by the number of generations at which 50% of the starting cells are dead . We measured an RLS of 39 . 2 generations ( 95% C . I . 38 . 6–39 . 8 , n = 440 , Figure 2A and Supplementary file 1 ) for the laboratory strain h- 972 grown in rich media ( Leupold , 1970 ) . To explore the effect of different genetic backgrounds on replicative lifespan , we determined survival curves , hazard functions , and replicative half-lives for three additional strains , including two wild fission yeast isolates with distinct morphologies and population doubling times ( Jeffares et al . , 2015 ) . Most laboratory fission yeast strains have three chromosomes , but the recently identified strain CBS2777 has a fourth chromosome that originated via a series of complex genomic rearrangements ( Brown et al . , 2014 , 2011; Rhind et al . , 2011 ) . CBS2777 had the shortest RLS of 22 . 7 generations ( 95% CI 21 . 9–23 . 6 generations ) , consistent with its aberrant genome and altered genome maintenance ( Brown et al . , 2014 ) . In contrast , strains NCYC132 and JB760 showed greater longevity with estimated RLS of >50 and >70 generations , respectively ( Figure 2—figure supplement 2 , Supplementary file 1B ) . These RLS correspond to a hazard rate that is ~5 fold lower than strain h- 972 ( Figure 2—figure supplement 2B ) . The longevity of strain NCYC132 is consistent with an earlier manual microdissection study performed on solid media ( Coelho et al . , 2013 ) . Each strain’s survival curve was best described by an exponential decay function , resulting in a generation-independent hazard function . Thus , an aging-independent replicative lifespan is a feature of diverse fission yeasts isolates , and likely the entire species . Prior RLS studies have identified three common phenotypes associated with cellular aging ( Henderson et al . , 2008; Piper , 2006 ) . In budding yeast , aging mother cells increase in cell size , progressively slow their doubling times , and produce daughters with decreased fitness ( Mortimer and Johnston , 1959; Bartholomew and Mittwer , 1953; Kennedy et al . , 1994 ) . Whether older fission yeast cells also undergo similar aging-associated phenotypes remains unresolved . To answer these outstanding questions , we examined time-dependent changes in morphology , doubling times , and sibling health in individual fission yeast cells as they approached death . Quantitative observation of cell length revealed two classes of dying cells: the majority ( 72%; n = 234 ) died prior to reaching the normal division length ( <16 . 2 µm , Figure 3A–B ) . The remaining cells ( 28%; n = 92 ) showed an elongated phenotype and exceeded the average division time by more than three-fold . In S . pombe , cell length is strongly correlated with division time and cell length , indicating that cell cycle checkpoints were likely dysregulated in these dying cells ( Mitchison and Nurse , 1985; Sveiczer et al . , 1996; Wood and Nurse , 2015; Hachet et al . , 2012 ) . Despite these differences in length at death , most cells had normal doubling times throughout their lifespan ( Figure 3C–D ) . Next , we investigated whether there were any morphological changes in the generations preceding death . The majority of cells retained wild type cell lengths and doubling rates until the penultimate cell division . In the two generations immediately preceding death , there was a mild , but statistically significant change in the distributions of cell lengths ( Figure 3E ) and doubling times ( Figure 3F ) . However , we did not observe any predictable trends for individual cells , arguing against a consistent pattern of age-related decline . In sum , S . pombe dies without any aging-associated morphological changes . 10 . 7554/eLife . 20340 . 011Figure 3 . Fission yeast does not show signs of aging . ( A ) Images of cells showing a short ( top ) and long ( bottom ) phenotype at death . Triangles indicate the old-pole , new-pole , and the new-pole of the previous division as in Figure 1G . Scale bar: 10 µm . ( B ) Histogram of cell length at death . The birth length was 8 . 3 ± 1 . 5 µm ( mean ± st . dev . , n = 326 ) and the length at division was 16 ± 2 . 2 µm ( n = 326 ) . ( C ) Cells dying with either short ( top ) or long ( bottom ) phenotype have normal length and ( D ) doubling times prior to death , as indicated by five representative cells . ( E ) Distribution of length and ( F ) doubling time at division in the five generations preceding death . Cells were post-synchronized to the time of death . The black bar shows the median value for each generation ( n > 290 cells for all conditions ) . Sequential generations were compared using the Kolmogorov-Smirnov test ( * for p<0 . 05 , ** for p<0 . 01 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20340 . 011 We next determined the fate of the last sibling produced by dying cells . This is possible because the last siblings of dying cells also remain captured in the catch channels ( Figure 4A ) . These are siblings that are produced by the final division of the old-pole cells . We categorized these siblings into three classes: ( i ) those that died without dividing , ( ii ) those that divided once and died , and ( iii ) those that divided two or more times ( Figure 4A , Videos 2–4 ) . The fate of the new-pole siblings was independent of the replicative age of the old-pole cells at death ( Figure 4B ) . Most new-pole siblings of a dying cell ( 66% , n = 135 ) never divided and typically did not grow , suggesting that the underlying cause of death was distributed symmetrically between the two sibling cells . Similarly , new-pole siblings that divided once and died ( 14% , n = 29 ) also typically did not grow . old-pole cells that died while hyper-elongated were more likely to have these unhealthy offspring ( Figure 4C–D ) . As elongated cells generally indicate the activation of a DNA damage checkpoint ( Furnari et al . , 1997 ) , our findings suggest that genome instability in these cells may be the underlying cause of death in both the old cell and its most recent sibling . In new-pole siblings that divided two or more times ( 20% , n = 40 ) , doubling times were indistinguishable from rapidly growing , healthy cells ( 2 . 1 ± 0 . 7 hr , n = 208 ) . These healthy new-pole cells were typically born from old-pole cells that did not elongate during their terminal division ( Figure 4C–D ) . These observations suggest that the cause of death was contained within the old-pole cell , thereby reducing the likelihood of death in the sibling . Taken together , our observations suggest that in 76% of cases , cell death impacts both sibling cells . However , in 24% of cells , death is localized to just one of the two siblings . Importantly , there was no correlation between the replicative age of the old-pole cell and survival probability of the last sibling cell . These observations are in stark contrast to S . cerevisiae , where aging factors are generally sequestered to younger mother cells ( Henderson et al . , 2014; Zhou et al . , 2014 ) . However , older S . cerevisiae mothers produce larger and slower-dividing daughter cells . Thus , we do not observe any aging-dependent outcomes for the fate of the new-pole sibling , further confirming that S . pombe does not age . 10 . 7554/eLife . 20340 . 012Figure 4 . Analysis of siblings born during the last division of a dying cell . ( A ) Last new-pole sibling continued to divide ( top ) , died after one division ( middle ) or died without dividing ( bottom ) . ( B ) The distribution of last-sibling phenotypes as a function of the old-pole replicative age ( n = 245 ) . Error bars are st . dev . measured by bootstrap analysis . ( C ) The length at division and ( D ) the doubling time of the new-pole siblings . Vertical black lines and gray bars show the mean and standard deviation of the total cell population . DOI: http://dx . doi . org/10 . 7554/eLife . 20340 . 01210 . 7554/eLife . 20340 . 013Video 2 . New-pole sibling phenotypes . Time-lapse images of a cell where the last division produces a healthy new-pole cell that divides more than once . Scale bar is 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 20340 . 01310 . 7554/eLife . 20340 . 014Video 3 . New-pole sibling phenotypes . Time-lapse images of a cell where the last division produces an unhealthy new-pole cell that divides once before dying . Scale bar is 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 20340 . 01410 . 7554/eLife . 20340 . 015Video 4 . New-pole sibling phenotypes . Time-lapse images of a cell where the last new-pole cell does not divide . Scale bar is 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 20340 . 015 The histone deacetylase Sir2p modulates lifespan and aging in a wide variety of organisms from yeasts to mice ( Wierman and Smith , 2014; Donmez and Guarente , 2010; Ganley and Kobayashi , 2014 ) . For example , sir2 deletion ( sir2Δ ) in budding yeast reduces the replicative lifespan by ~50% ( Jo et al . , 2015; Kaeberlein et al . , 1999 ) , whereas Sir2p overexpression can increase the RLS by up to 30% ( Kaeberlein et al . , 1999 ) . The S . pombe genome encodes three Sir2p homologs , one of which shares a high degree of sequence similarity and biochemical functions with the budding yeast Sir2p ( Shankaranarayana et al . , 2003; Freeman-Cook et al . , 2005 ) . Although S . pombe does not show aging phenotypes , we next investigated whether Sir2p still modulates the RLS . Deletion of Sir2p ( sir2Δ ) reduced the RLS by 15% ( 33 . 4 ± 0 . 8 generations; n = 329 ) relative to the wild type parental strain ( 39 . 2 ± 0 . 6 generations , n = 440; Figure 5A ) . These results are consistent with prior observations that wild type and sir2Δ cells had similar growth rates when cultured without stressors ( Erjavec et al . , 2008 ) . In contrast , constitutive 2-fold sir2 overexpression ( sir2OE , Figure 5—figure supplement 1 ) increased the RLS over 50% , with a mean replicative lifespan of >60 generations ( n = 301; Figure 5A ) . Cells overexpressing Sir2p have a 2-fold lower hazard rate ( total risk of death ) than wild-type cells ( Figure 5B and Figure 5—figure supplement 2 ) . In sum , overexpression of Sir2p increases the RLS of S . pombe , but does not affect aging . 10 . 7554/eLife . 20340 . 016Figure 5 . Sir2p and rapamycin extend replicative lifespan . ( A ) Replicative lifespans and ( B ) hazard rates of strains normalized to the mean RLS and hazard rate of wild-type ( h- 972 ) strain . Error bars: 95% C . I . on an exponential decay fit to the experimental survival curve . ( C ) Distribution of normalized doubling time in the five generations preceding death for Sir2p overexpression cells and ( D ) wild-type cells treated with 100 nM rapamycin . ( E ) Distribution of normalized length at division in the five generations preceding death for Sir2p overexpression cells and ( F ) wild-type cells treated with 100 nM rapamycin . Black bars show the median value . Sequential generations were compared using the Kolmogorov-Smirnov test ( * for p<0 . 05 , ** for p<0 . 01; n > 144 cells for all conditions ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20340 . 01610 . 7554/eLife . 20340 . 017Figure 5—figure supplement 1 . Sir2p expression levels . ( A ) sir2 mRNA transcript levels as measured by qPCR . The sir2Δ strain was included as a negative control . Error bars: normalized SEM of at least three replicates . ( B ) Sir2p protein levels as detected by an anti-Sir2p antibody . Actin is included as loading control . Sir2p-eGFP migrates slower than endogenous Sir2p . ( C ) Single cell fluorescence imaging confirms that Sir2p-eGFP is expressed and localizes to the nucleus . Scale bar: 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 20340 . 01710 . 7554/eLife . 20340 . 018Figure 5—figure supplement 2 . The effect of Sir2p and rapamycin on the RLS of fission yeast . ( A ) The replicative half-life is indicated in parenthesis . A lower bound is reported for the long-lived Sir2p overexpression strain . Red lines represent an exponential decay fit to the curve . ( B ) Hazard functions of strains shown in ( A ) . The constant hazard rates illustrate that none of the WT strains are aging . DOI: http://dx . doi . org/10 . 7554/eLife . 20340 . 018 The target of rapamycin ( TOR ) pathway is the central regulator of cell growth and aging in eukaryotes ( Johnson et al . , 2013; Loewith and Hall , 2011 ) . Rapamycin inhibits TOR proteins in eukaryotes ( Kunz et al . , 1993; Brown et al . , 1994 ) , reducing cell growth by altering translation initiation and repressing ribosome biogenesis . Rapamycin increases the RLS of budding yeast ( Medvedik et al . , 2007 ) , and also increases the longevity of worms , flies and mice ( Robida-Stubbs et al . , 2012; Bjedov et al . , 2010; Harrison et al . , 2009 ) . However , the effect of rapamycin on fission yeast RLS has not been reported . The addition of 100 nM rapamycin to the flow medium increased the RLS by 41% ( 55 . 3 generations , 95% CI 53 . 1–57 . 6 , n = 184; Figure 5A ) . As with Sir2p overexpression , treating cells with rapamycin increased their longevity by reducing the aging-independent hazard rate ( Figure 5B ) . Importantly , both rapamycin and Sir2p overexpression did not affect the lengths or doubling times of cells as they approached death . The cell length at division remained constant for the five generations preceding cell death , whereas the cell doubling time increased only in the last generation before death ( Figure 5C–F ) . These results are consistent with an aging-independent mechanism for RLS extension in S . pombe . Ribosomal DNA ( rDNA ) is a repetitive , recombination-prone region in eukaryotic genomes . Defects in rDNA maintenance arise from the illegitimate repair of the rDNA locus and have been proposed to be key drivers of cellular aging ( Ganley and Kobayashi , 2014; Kobayashi , 2008 ) . Fission yeast encodes ~117 rDNA repeats on the third chromosome , and incomplete rDNA segregation can activate the spindle checkpoint ( Toda et al . , 1984; Maleszka and Clark-Walker , 1993 ) . If unresolved , this can result in chromosome fragmentation and genomic instability ( Win et al . , 2005 ) . Indeed , we observed that 28% of dying cells were abnormally long , suggesting activation of a DNA-damage checkpoint ( Figure 3B ) . Aberrant rDNA structures are processed by the RecQ helicase Rqh1p and mutations in the human Rqh1p homologs are implicated in cancer and premature aging-associated disorders ( Croteau et al . , 2014; Ellis et al . , 1995; Yu et al . , 1996; Kitao et al . , 1999; Takahashi et al . , 2011 ) . Therefore , we sought to determine whether rDNA instability and loss of Rqh1p can contribute to stochastic death . We visualized segregation of rDNA in cells using the nucleolar protein Gar2 fused to mCherry . Gar2 localizes to rDNA during transcription and has been used to monitor rDNA dynamics in live cells ( Win et al . , 2005 ) . The strain expressing Gar2-mCherry at the native locus divided at approximately the same length and rate as wild-type cells ( Figure 6—figure supplement 1 ) . Dividing cells mostly had equal Gar2 fluorescence in both sibling cell nuclei ( Figure 6A–B ) . However , 7% of cells ( n = 88/1182 ) showed rDNA segregation defects that appear as multiple rDNA loci , asymmetric fluorescence distributions , or rDNA bridges ( Figure 6A ) . The rDNA mis-segregation was similar to mis-segregation of a fluorescent marker on chromosome I ( 7% of cells; n = 20/278 ) and chromosome II ( 5% of cells; n = 19/370 ) ( Figure 6—figure supplement 2 ) . Multi-punctate rDNA loci were nearly always lethal , whereas asymmetric rDNA segregation and rDNA bridges were lethal in 40% of cells ( n = 31/78; Figure 6B ) . Importantly , 40% of cells ( n = 56/141 ) showed rDNA defects immediately preceding death . In dying cells , rDNA instability was elevated relative to mis-segregation of chromosome I ( 16% of cells; n = 15/95 ) and chromosome II ( 13% of cells; n = 18/137 ) ( Figure 6—figure supplement 2 ) . Loss of Rqh1p caused higher frequencies of spontaneous rDNA defects , a shorter RLS ( 5 . 3 generations , 95% CI 5 . 0–5 . 7 ) , and a >7 fold higher hazard rate ( Figure 6—figure supplement 3 ) . rDNA defects were even more prevalent in dying rqh1Δ cells ( Figure 6—figure supplement 3 ) . We next tested the influence of Sir2p over-expression or addition of rapamycin on the short RLS of rqh1Δ cells . The addition of rapamycin showed a mild RLS extension , whereas the rqh1Δ sir2OE strain was extremely short lived ( Figure 6—figure supplement 3 ) . These data suggest that the elevated rDNA instability cannot be rescued by Sir2p and is only partially rescued by rapamycin . In summary , the longevity of fission yeast is promoted by rDNA stability . 10 . 7554/eLife . 20340 . 019Figure 6 . Ribosomal DNA ( rDNA ) instability is highly correlated with cell death . ( A ) Images of cells exhibiting rDNA instability , as reported by gar2-mCherry , which binds to rDNA . ( B ) Likelihood of cell death following one of the defects observed in ( A ) . ( C ) Dying cells exhibited elevated rDNA defects ( outer ring ) relative to healthy dividing cells ( inner ring ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20340 . 01910 . 7554/eLife . 20340 . 020Figure 6—figure supplement 1 . Strains expressing gar2-mCherry maintain wild-type replication rates . ( A ) Histogram of doubling times of a strain expressing gar2-mCherry . The doubling time is 2 . 83 ± 0 . 87 hr ( mean ± st . dev . ; n = 1547 ) . ( B ) Histogram of length at division . The length is 18 . 0 ± 2 . 3 µm ( mean ± st . dev . ; n = 1547 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20340 . 02010 . 7554/eLife . 20340 . 021Figure 6—figure supplement 2 . Live imaging of chromosome mis-segregation rates . ( A ) Chromosome loci were labeled with a 112-tetO repeat . TetR-tdTomato binds this array and allows live cell imaging of chromosome dynamics . Schematic adapted from Petrova et al . ( 2013 ) . ( B ) Images of cells with chromosome I and chromosome II labeled with the tetO-TetR array . ( C ) Chromosome mis-segregation defects were characterized as in Figure 6 . Norm: equal separation of fluorescent foci between siblings , Multi: multiple foci observed in single sibling , Uneq: unequal fluorescence between siblings , Bridge: foci remain connected post-division . Dying cells showed 2 . 3-fold elevated chromosome I and ( D ) 2 . 6-fold elevated chromosome II mis-segregation defects ( outer ring ) relative to healthy dividing cells ( inner ring ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20340 . 02110 . 7554/eLife . 20340 . 022Figure 6—figure supplement 3 . Characterization of the RLS of an rqh1Δ strain . ( A ) Survival curve of WT ( black , 95% CI: 38 . 6–39 . 8 generations ) , rqh1Δ ( green , 95% CI: 5 . 0–5 . 7 ) , rqh1Δ + 100 nM rapamycin ( blue , 95% CI: 6 . 2–7 . 2 ) , and rqh1Δ sir2OE ( orange ) . The extremely short lifetime of the rqh1Δ sir2OE strain precluded an accurate fit . ( B ) Hazard curves for the strains shown in ( A ) . The number of cells , RLS , hazard rate , and goodness of fit parameters are summarized in Supplemental file 1 . ( C ) The incidence of nucleolar defects for all divisions ( center ring ) and preceding death ( outer ring ) in rqh1Δ strain . DOI: http://dx . doi . org/10 . 7554/eLife . 20340 . 022 Here , we report the first study of fission yeast RLS in a high-throughput microfluidic device ( Figure 1 ) . The multFYLM provides a temperature-controlled growth environment for hundreds of individual cells for up to a week ( >75 generations ) , facilitating tens of thousands of micro-dissections . We also developed an image-processing pipeline for quantitative phenotypic analysis of individual cell lineages . The blueprints for multFYLM , as well as the image analysis pipeline are both available via GitHub ( see Materials and methods ) . We note that microfluidics-based lifespan measurements are free from potential extrinsic effects ( e . g . , secretion of small molecules onto the solid agar surface ) which have been proposed to confound observations of RLS assays using manual microdissection ( Mei and Brenner , 2015 ) . Using the multFYLM , we set out to determine whether fission yeast undergoes replicative aging . Taken together , our data provide three sets of experimental evidence that fission yeast does not age . First , the RLS survival curves and corresponding hazard rates are qualitatively different between budding and fission yeasts ( Figure 2 ) . The budding yeast RLS is best described by a Gompertz model that includes both an age-dependent as well as an age-independent survival probability . In contrast , the fission yeast RLS is best described by a single exponential decay . This corresponds to a hazard function λ ( g ) that is constant with each generation , as would be expected for an organism that does not age . Second , we observed that cell volumes and division rates did not change until the penultimate cell division ( Figure 3 ) . Third , after the death of a cell , the health of the surviving sibling was also independent of age ( Figure 4 ) . These conclusions are also broadly consistent with a recent report that observed robust , aging-independent growth of the symmetrically dividing E . coli in a microfluidic device ( Wang et al . , 2010 ) . Our results highlight a critical consideration when interpreting replicative lifespan curves . The experimentally determined RLS does not necessarily report on whether a population of cells is aging . This is because the RLS is affected by a combination of both age-independent and age-dependent biological processes—cell may die randomly without aging . A mathematical model is required to further parse the relative contributions of aging-independent and aging-dependent mechanisms . The Gompertz function is an excellent model for understanding the replicative lifespan survival curve ( Bansal et al . , 2015 ) . In this model , the replicative lifespan is dependent on just two parameters: ( i ) an age-independent hazard rate , and ( ii ) an age-dependent hazard rate , which determine how rapidly mortality increases as a function of the cell’s age ( Kirkwood , 2015 ) . Together , these two parameters describe a population’s likelihood of death at any age . Figure 7 illustrates that the experimentally observed RLS is dependent on both of these terms . An increased RLS is not sufficient to conclude that cells age more slowly . Indeed , our data indicate that a decreased RLS can also be achieved by increasing the age-independent hazard rate . To further illustrate this point , we include a Gompertz model analysis of the RLS of S . cerevisiae grown in a microfluidic device ( Jo et al . , 2015 ) and our own data from S . pombe . For all of our data , changes in RLS can be completely accounted for by changes in the age-independent hazard rate . In contrast , budding yeast have age-dependent hazard rates an order of magnitude higher than their age-independent rates ( Figure 7 ) . We anticipate that the development of the multFYLM , as well as a quantitative analysis framework , will continue to inform comparative RLS studies in model organisms . 10 . 7554/eLife . 20340 . 023Figure 7 . The replicative lifespan is an incomplete reporter of cellular aging . RLS contours were generated from experimentally determined ranges of Gompertz coefficients using Equation ( 7 ) . Fission yeast strains examined in this study and budding yeast from an analogous study ( Jo et al . , 2015 ) were plotted on the chart based on the coefficient values from either a Gompertz or exponential decay fit . In all cases , the 95% CI of the coefficients was smaller than the marker size . DOI: http://dx . doi . org/10 . 7554/eLife . 20340 . 023 What are the molecular mechanisms of stochastic cell death in a symmetrically dividing eukaryote ? Strikingly , the RLS of S . pombe can be extended by up to 41% via rapamycin treatment and Sir2p over-expression without influencing aging-related phenotypes ( Figure 5 ) . This suggests that rapamycin and Sir2p both reduce the onset of sudden death . To our knowledge , this is the first evidence these interventions can extend the RLS in an organism that does not age . This suggests that both Sir2p and rapamycin may partially affect the RLS of aging organisms like S . cerevisiae via an aging-independent mechanism . Additional studies will be required to parse out the aging-dependent and aging-independent mechanisms of RLS extension . For a subset of cells , genomic instability due to aberrant segregation of the rDNA locus may be one cause of death ( Saka et al . , 2013; Ganley and Kobayashi , 2014 ) . Consistent with this hypothesis , rDNA segregation defects were highly elevated relative to general chromosome mis-segregation in wild type cells immediately prior to death ( Figure 6 ) . Ablating Rqh1p , a helicase that promotes rDNA stability and replication fork progression ( Stewart et al . , 1997; Murray et al . , 1997; Doe et al . , 2000 ) , further increased rDNA defects while drastically reducing the RLS , independent of age . Recent studies also indicate that protein aggregation may be a second major cause of sudden death in rapidly dividing fission yeast cells ( Coelho et al . , 2013 , 2014 ) . These two mechanisms of death need not be mutually exclusive . Future studies will further define the molecular mechanisms of aging-independent death in fission yeast . Master structures were fabricated using SU-8 3005 photoresist for the first layer and SU-8 2010 photoresist for the second layer , following standard photolithography techniques described in the product literature ( Microchem , Westborough , MA ) . Photoresist was spun onto 100 mm-diameter , test-grade , P-doped silicon wafers ( ID# 452 , University Wafers ) . Photoresist thickness was 20–30 µm for the second layer ( conduits , central and side trenches ) , and 5–6 µm for the first layer ( catch and drain channels ) . Custom chrome on quartz photomasks ( Compugraphics ) were manufactured from designs created using the freeware integrated circuit layout editor Glade ( www . peardrop . co . uk ) or with OpenSCAD ( www . openscad . org ) . A Suss MA-6 Mask Aligner ( Suss MicroTec Lithography GmbH ) was used for mask alignment and photoresist exposure . multFYLM construction is described in more detail at Bio-protocol ( Jones Jr et al . , 2018 ) . Approximately 25 g of polydimethylsiloxane ( PDMS , Sylgard 184 , Dow Corning ) was mixed at a weight ratio of 10:1 polymer:hardener , placed on a rotator for >30 min , then centrifuged to remove bubbles . A tape barrier was applied to the edge of a silicon wafer bearing the multFYLM master structures , and 13 g of PDMS was poured onto the wafer , covering the surface . The wafer with PDMS was degassed for ~10 min ( at 60–70 cmHg vacuum in a vacuum chamber ) to remove additional bubbles , then placed in a 70°C oven for 15–17 min . The wafer was intentionally removed while the PDMS was still tacky to improve adhesion of the microfluidic connectors ( nanoports ) . To create a source and drain interface for the device , nanoports ( N-333–01 , IDEX Health and Science , Oak Harbor , WA ) were applied to the surface of the PDMS over the ends of the master structure of the multFYLM , then an additional 14 g of PDMS was poured over the surface , with care taken to avoid getting liquid PDMS inside the nanoports . The wafer with PDMS and nanoports was then degassed >10 min to remove additional bubbles , and then returned to the 70°C oven for 3 hr to cure completely . The cured PDMS was then removed whole from the wafer after cooling to room temperature . A multFYLM with nanoports was trimmed from the cured PDMS disk to approximately 15 × 25 mm . A 1 mm diameter biopsy punch ( Acu-punch , Accuderm ) was used to make holes connecting the bottom of each nanoport to the conduits on the bottom surface of the PDMS . The multFYLM was then placed in isopropanol and sonicated for 30 min , removed , and placed to dry on a 70°C hotplate for 2 hr . Borosilicate glass coverslips ( 48 × 65 mm #1 , Gold Seal ) were concurrently prepared by cleaning for >1 hr in a 2% detergent solution ( Hellmannex II , Hellma Analytics ) , then rinsing thoroughly in deionized water and isopropanol before drying for >2 hr on a 70°C hotplate . Cleaned multFYLM and coverslips were stored in a covered container until needed . To bond , the multFYLM and a coverslip were placed in a plasma cleaner ( PDC-32G , Harrick Scientific ) and cleaned for 20 s on the ‘high’ setting with oxygen plasma from ambient air ( ~21% oxygen ) . The multFYLM and glass coverslip were then gently pressed together to form a permanent bond . To make the microfluidic interface , PFA tubing ( 1512L , IDEX ) , with a 1/16” outer diameter was used to connect a 100 mL luer lock syringe ( 60271 , Veterinary Concepts , www . veterinaryconcepts . com ) to the multFYLM , and to construct a drain line leading from the multFYLM to a waste container . The tubing was connected to the multFYLM nanoports using 10–32 coned nuts ( F-332–01 , IDEX ) with ferrules ( F-142N , IDEX ) . A two-way valve ( P-512 , IDEX ) was placed on the drain line to allow for better flow control . For longer experiments , two syringes could be loaded in tandem , connected with a Y-connector ( P-512 , IDEX ) . The valve and Y-connector were connected to the tubing using ¼−28 connectors ( P-235 , IDEX ) with flangeless ferrules ( P-200 , IDEX ) . The source line tubing was connected to the syringe with a Luer adapter ( P-658 , IDEX ) . The syringe ( s ) was/were then placed in a syringe pump ( Legato 210 , KD Scientific ) for operation . A 2 µm inline filter ( P-272 , IDEX ) was placed upstream of the multFYLM to reduce the chance of debris clogging the multFYLM during operation . As soon as possible after plasma bonding ( usually within 15 min ) , the multFYLM was placed on the microscope stage and secured , then 15 µL of cells suspended in YES media with 2% BSA were gently injected into the source nanoport . The drain and source interface tubing ( with drain valve closed ) were then connected to the drain and source nanoports . The syringe pump was then started at 40 µL min−1 , and the drain valve was opened to allow flow to start . The syringe pump was adjusted between 20–60 µL min−1 until flow was established , and catch channels were observed to be filling with cells , then a programmed flow cycle was established: 1–5 min at 50 µL min−1 followed by 10–14 min at 5 µL min−1 ( average flow rate , 0 . 5–1 . 2 mL h−1 ) . Images were acquired using an inverted microscope ( Nikon Eclipse Ti ) running NIS Elements and equipped with a 60X , 0 . 95 NA objective ( CFI Plan Apo λ , Nikon ) and a programmable , motorized stage ( Proscan III , Prior ) . The microscope was equipped with Nikon’s ‘Perfect Focus System’ ( PFS ) , which uses a feedback loop to allow consistent focus control over the multiple-day experiments . Images were acquired approximately every two minutes using a scientific-grade CMOS camera ( Zyla 5 . 5 , Andor ) . To improve contrast during image analysis , images were acquired both in the plane of focus and 3–4 µm below the plane of focus . Temperature control was maintained using an objective heater ( Bioptechs ) and a custom-built stage heater ( Omega ) calibrated to maintain the multFYLM at 30–31°C . Fluorescence images were acquired with a white light LED excitation source ( Sola II , Lumencorp ) and a red filter set ( 49004 , Chroma ) . Fluorescent images were acquired concurrently with white light images , but only every four minutes . All strains were propagated in YES liquid media ( Sunrise Scientific ) or YES agar ( 2% ) plates , except where otherwise noted . A complete list of strains is reported in Supplemental file 1 . Deletion of sir2 in the wild-type h- 972 strain was completed by first generating a PCR product containing KANMX4 flanked by SIR2 5’ and 3’ untranslated regions . Competent IF30 cells were then transformed with the PCR product and selected on YES +G418 agar plates . Deletion was confirmed by PCR . A SIR2 over-expression plasmid was generated via Gateway cloning ( Life Technologies ) . First , SIR2 was integrated into the pDONR221 plasmid to yield pIF133 . SIR2 was then swapped from pIF133 to pDUAL-GFH1 ( Riken ) , yielding plasmid pIF200 . pIF200 expresses SIR2 under the NMT1 promoter , and also labels the protein C-terminally with a eGFP-FLAG-His6 tags . Plasmid pIF200 was then transformed into strain IF140 for integration at the leu1 locus to generate clones IF230 and IF231 ( Matsuyama et al . , 2006 ) . Integration was checked by PCR and expression was determined by RT-qPCR . Wild-type h- 972 and Gar2-mCherry fusion protein expressing strains ( Megan King ) were transformed with a PCR product containing HPHMX6 ( hygromycin resistance ) flanked by rqh1 5’ and 3’ untranslated regions . Transformants were selected on YES+hygromycin agar plates , then confirmed by PCR . Chromsome I and II reporter strains were kindly provided by Christian Haering ( strains CH2774 and CH3245 ) ( Petrova et al . , 2013 ) . Strains IF235 ( sir2Δ ) , IF140 and IF230 ( sir2 over-expression; sir2OE ) were grown in synthetic complete media ( Sunrise Scientific ) to OD 1 . 0 , and whole cell extracts were prepared by trichloroacetic acid precipitation with bead lysis ( Keogh et al . , 2006 ) . These cell extracts were combined with loading buffer , boiled for 5 min at 95°C , and resolved via a 4–20% TGX SDS-PAGE gel ( Bio-Rad ) . Proteins were transferred to a PVDF membrane , and then blocked with PBS Odyssey blocking buffer ( Li-Cor , Inc . ) for one hour with shaking at 22°C . The membrane was incubated overnight at 4°C while shaking in phosphate buffered saline +0 . 05% Tween-20 ( PBST ) containing 1:1000 dilution of anti-β actin antibody ( mouse , ab8224 , Abcam ) and 1:500 dilution of an S . pombe-specific anti-Sir2p antibody ( rabbit , generously provided by Dr . Allshire ) ( Buscaino et al . , 2013 ) . The membrane was washed once in PBST , and then agitated for 4 hr at 22°C with 1:10 , 000 dilution of goat anti-mouse IgG and 1:10 , 000 dilution of goat anti-rabbit IgG ( IRDye 680RD and IRDye 800CW , respectively , Li-Cor , Inc . ) . The membrane was washed three times with PBST and imaged on an Odyssey CLx dual-channel imaging system ( Li-Cor , Inc . ) . To analyze our single-cell data , we developed FYLM Critic—a Python package for rapid and high-content quantification of the time-lapse microscopy data ( github . com/finkelsteinlab/fylm ) . The FYLM Critic pipeline consists of two discrete stages: ( i ) pre-processing the raw microscope images to correct for stage drift , and ( ii ) quantification of cell phenotypes ( e . g . , length , doubling time , fluorescence intensity and spatial distribution ) . Several image pre-processing steps were taken to permit efficient quantification . First , the angles of the four solid sections of PDMS on either side of the catch channels were measured relative to the vertical axis of the image , and a corrective rotation in the opposite direction was applied to all images in that field of view . This resulted in the long axis of each cell being completely horizontal in all subsequent analyses . The multFYLM was too large to be imaged in its entirety within a single field of view , so the microscope stage was programmed to move to eight different subsections of the device in a continuous cycle . Due to the imperfect motion of the microscope stage , subsequent images of the same field of view randomly translated by a few micrometers relative to the previous image at the same nominal location . Images were aligned to the first image of the sequence using a cross-correlation algorithm ( Guizar-Sicairos et al . , 2008 ) . The exact location of each catch channel was manually specified , but only if a cell had already entered the channel at the beginning of the experiment . In order to analyze cell lengths and phenotypes , kymographs were produced for each catch channel and then annotated semi-manually . To produce the kymographs , a line of pixels was taken along the center of each catch channel , starting from the center of the ‘notch’ and extending into the central trench . This kymograph captured information along the long axis of the cells . The time-dependent one-dimensional ( 1D ) kymographs for each catch channel were stacked together vertically , producing a two-dimensional ( 2D ) kymograph of each cell’s growth and division . The out-of-focus image was used for this process as the septa and cell walls were much more distinct , as described previously ( Nobs and Maerkl , 2014 ) . The position of the old-pole and the leftmost septum were then manually annotated using a simple point-and-click interface to identify each cell division . Once all single-cell kymographs were annotated , this annotation was used to calculate a cell length for each time point . The final status of the cell ( whether it died , survived to the end of the experiment , or was ejected ) was determined by the user . Further analysis of the cell length , growth rates , and fluorescence intensities was performed in MATLAB ( Mathworks ) . The local minima of the growth curves were used to determine division times and the replicative age of each cell . Ejections and replacement of individual cells within the catch channels may confound our whole-lifespan tracking . Therefore , we optimizing cell loading , retention and imaging conditions to minimize the loss of individual cells throughout the course of each experiment . First , ejection of individual cells was minimized over the course of ~100 hr ( Figure 1—figure supplement 2B ) . Second , each cell was imaged once every two minutes , minimizing the amount of time during which a cell could conceivably be ejected and replaced by another from the central trench . If a cell were indeed replaced within our two-minute frame rate , we would expect an abrupt change in the captured cell length ( indicated a re-loading event ) . All kymographs were carefully monitored for such events , and in the rare cases where this occurred , were discarded from the final analysis . In summary , our analysis ensures that cell ejection or re-capture is mitigated by both efficient cell capture and by analysis of the single-cell datasets . Kaplan-Meier ( Kaplan and Meier , 1958 ) survival function estimates were calculated in MATLAB . Due to the low number of lost cells ( Figure 1—figure supplement 2B ) and the highest total number of total cells ( n ≥ 100 for all experiments ) , the survival curves did not include cells that were lost during the course of the experiment ( i . e . , there was no right censoring ) . Our analysis is consistent with similar studies in S . cerevisiae ( Crane et al . , 2014 ) . The survival function estimates were fitted with either a simple exponential decay function ( 1 ) S ( g ) =e−α*g or the Gompertz survival function ( Gompertz , 1825; Greenwood , 1928 ) ( 2 ) S ( g ) =eα/β ( 1−eβg ) Where S is the fraction of the initial population surviving at generation g . The coefficients α and β are >0 and <1 , and are further defined for the hazard function λ ( g ) below . The plotted survival data were weighted at 1/S ( g ) to increase the influence of older cells on the fit . The hazard function λ ( g ) ( the probability of cell death during a given generation ) can be derived from the survival function: ( 3 ) λ ( g ) =−S . ( g ) S ( g ) where S . is the first derivative of S . For a survival curve fit to an exponential decay function , the corresponding hazard function is a constant: ( 4 ) λ ( g ) =α For a survival curve fit to the Gompertz function , the hazard function simplifies to: ( 5 ) λ ( g ) =α∙eβg Where the coefficient α amplitude-scales λ ( g ) , and so has an age-independent influence on the hazard function , while the coefficient β time-scales λ ( g ) , and so provides a compact method to describe the age-dependent increase in the probability of cell death ( Figure 7 ) . Note that as β→0 , Equation ( 5 ) approaches Equation ( 4 ) . Mean replicative lifespan ( RLS ) can be calculated by setting S ( g ) =12 , and solving for g ( Simms , 1946 ) . For exponential decay , RLS is simply the half-life of the decay function: ( 6 ) RLS = ln ( 2 ) /α For the Gompertz function , RLS is: ( 7 ) RLS=ln ( 1+ ( βln ( 2 ) α ) ) β All fitting and statistical data analysis were performed in MATLAB . Briefly , datasets were first tested for normality using the Anderson-Darling method , and parametric tests were used when possible . The fitting of all data was performed in MATLAB using either the nonlinear least squares method or least squares method . Determination of whether a survival curve was fit to an exponential decay or Gompertz function was made primarily by selecting the fit with the highest adjusted r2 value .

As the cells in our bodies age , their ability to carry out their normal processes also degrades . Ultimately , this causes tissues to deteriorate . How rapidly a cell ages depends on the genes encoded in its DNA , and can also be affected by certain drug treatments . Cells reproduce by dividing to form two new cells . One common approach used to study cellular aging is to follow how genetic modifications and drug treatments alter how many offspring a cell produces before it dies . So far , most of these studies have been performed using budding yeast cells , which reproduce by dividing asymmetrically to form two differently sized cells . Little is known about aging in cells that divide symmetrically . Another type of yeast , called fission yeast , divides symmetrically . Furthermore , many of the processes used by fission yeast cells to repair and replicate DNA are the same as those used in human cells . To study aging in fission yeast , Spivey , Jones et al . developed a “microfluidic” device and software tools that can image and track a large number of cells across their entire lifespans ( which last for up to a week ) . Unexpectedly , it appears that fission yeast do not age . Although cells died , their deaths were random and their chance of dying did not increase as the cells got older . Treating the cells with a drug called rapamycin lengthened their average lifespan , as did stabilizing their DNA . Treatments that impaired the ability of the cells to repair their DNA reduced the yeast's lifespan . However , in all cases , a cell’s chance of death stayed constant as the cells got older . One conclusion that could be drawn from the work of Spivey , Jones et al . is that cellular aging may be intrinsic to cells that divide asymmetrically – so budding yeast ages whilst fission yeast does not . Future studies will investigate this in more detail; for example , by looking at how “pro-aging factors” are segregated between dividing cells .

2017

An aging-independent replicative lifespan in a symmetrically dividing eukaryote

Whether retrieval still depends on the hippocampus as memories age or relies then on cortical areas remains a major controversy . Despite evidence for a functional segregation between CA1 , CA3 and parahippocampal areas , their specific role within this frame is unclear . Especially , the contribution of CA3 is questionable as very remote memories might be too degraded to be used for pattern completion . To identify the specific role of these areas , we imaged brain activity in mice during retrieval of recent , early remote and very remote fear memories by detecting the immediate-early gene Arc . Investigating correlates of the memory trace over an extended period allowed us to report that , in contrast to CA1 , CA3 is no longer recruited in very remote retrieval . Conversely , we showed that parahippocampal areas are then maximally engaged . These results suggest a shift from a greater contribution of the trisynaptic loop to the temporoammonic pathway for retrieval . While it is a consensus that the hippocampus is engaged for the retrieval of recent memory , its involvement in retrieving more remote memories is still controversial . Some studies showing temporally-graded retrograde amnesia following hippocampal damage in humans have provided evidence that memory becomes hippocampal-independent as it ages and is ultimately ‘stored’ in the cortex ( Alvarez and Squire , 1994: “the system consolidation theory” ) . An alternative theory posits that the hippocampus is not only engaged in retrieving recent memories but that it also supports cortical areas in retrieving more remote memories ( Nadel and Moscovitch , 1997: the ‘multiple trace theory’ ) . The hippocampal subfields CA1 and CA3 are functionally segregated and quite a few studies have investigated their role in recent memory , for example in the retrieval of contextual fear conditioning ( Tanaka et al . , 2014; Wheeler et al . , 2013; Remaud et al . , 2014; Daumas et al . , 2005; Rampon et al . , 2000; McHugh and Tonegawa , 2009; Hunsaker et al . , 2009; Goshen et al . , 2011; see for reviews: Nakazawa et al . , 2004; Kesner , 2007 ) . However , it is unclear whether they contribute to a similar extent to the retrieval of more remote memories . A recent lesion study in humans indicates that CA1 participates to the retrieval of early and very remote memories ( over 30 years-old ) in addition to its well-established role for recent memories ( Bartsch et al . , 2011 ) . In contrast , the role of CA3 in the retrieval of remote memory has not been yet studied in humans and has rarely been investigated in animals for memories older than a month , a period of time roughly equivalent to four years in humans based on life expectancy . Hence , very little is known about the extent to which CA3 contributes to the retrieval of remote and very remote memory . Nonetheless , computational studies suggest that CA3 would contribute to retrieving memories , at least recent and early remote memory , via processes involving the completion of memory representations based on partial or degraded features of these representations ( the ‘pattern completion’ theory: Rolls et al . , 1997 ) . However , the memory for single features and/or details that could serve as retrieval cues degrades over time ( Wiltgen and Silva , 2007; Wiltgen et al . , 2010 ) . Hence , it raises the question whether CA3 would also contribute to the retrieval of very remote memories or whether its contribution depends on the age of the memory trace . In addition , a critical involvement of the parahippocampal region in the retrieval of recent and early remote memory is supported by a handful of animal studies and by recent reports of flat retrograde amnesia in patients with damage extending from the hippocampus to other medial temporal lobe areas . However , these studies did not investigate very remote memory traces while the contribution of cortical areas to the retrieval of memory is thought to be maximal at this stage . ( Squire and Alvarez , 1995; Gusev et al . , 2005; Bucci et al . , 2002; Burwell et al . , 2004; Izquierdo et al . , 1997 ) . Moreover , despite evidence for a functional segregation between parahippocampal areas in terms of spatial/non-spatial information processing and/or of the mediation of distinct memory processes ( Mishkin et al . , 1983; Eichenbaum et al . , 2007; Beer et al . , 2013 ) , no study has yet dissociated the specific contribution of the perirhinal ( PER ) and postrhinal cortices ( POR ) nor that of the lateral and medial entorhinal cortices ( LEC , MEC ) to the retrieval of remote memory . Here , we address these issues using a contextual fear conditioning task by testing whether CA3 contributes to a comparable extent as CA1 to the retrieval of recent ( 1 day and 1 week-old ) , early remote ( 1 month-old ) and very remote memories ( 6 months or 1 year-old ) , the latter corresponding roughly to 20 and 40 years-old memories in humans based on life expectancy which are among the delays the most investigated in studies focusing on very remote memory ( Figure 1A ) . In addition , we investigated the extent to which the PER , the POR , the LEC and the MEC differentially contribute to the retrieval of these memories . Because performing in-vivo electrophysiology recording simultaneously in six distant brain areas remains a major challenge and using a lesion/inactivation approach in adjacent brain areas in mice was unlikely to yield the spatial resolution necessary to tease apart the specific function of each area , a high-resolution molecular imaging technique ( e . g . to the cellular level ) was employed . This imaging technique is based on the detection of the expression of the immediate-early gene Arc which has been especially linked to plasticity processes and cognitive demands and has been recently used for mapping cognitive processes in the medial temporal lobe ( Guzowski et al . , 1999; Nakamura et al . , 2013; Sauvage et al . , 2013; Figure 2B–G ) . 10 . 7554/eLife . 11862 . 003Figure 1 . Contextual fear-conditioning task and memory performance . ( A ) Schematic description of the task . Mice were exposed to a conditioning chamber they explored freely for 2 . 5 min after which they received a single 1 mA footshock and were returned to their home-cage after 3 . 5 additional min . Memory for the association of the footshock with the conditioning context was tested after either 1 day , 1 week , 1 month , 6 months or a year by re-exposing the conditioned mice to the conditioning chamber and measuring freezing levels ( n=4 mice per delay; n=20 total ) . Upon completion of the task , mice were sacrificed and their brain processed for imaging . ‘No-shock’ age-matched control groups were exposed to the exact same experimental conditions but did not receive any footshock ( n=4 mice per delay; n=20 total ) . Of note , additional age-matched home-cage control groups were generated to control for Arc baseline expression ( n=4 mice per delay; n=20 total; not shown here ) . ( B ) Differences in freezing indices between ‘shock’ and ‘no-shock’ groups ( e . g . normalized freezing indices ) in function of the age of the memory trace . All shocked groups showed stronger freezing than their no-shock controls at test as shown by differences in freezing levels significantly higher than 0 ( a 55 . 9 ± 3 . 4% increase in average ) . Importantly , freezing levels were similar across delays , suggesting that memory strength did not significantly differ as memory aged . ( C ) Normalized freezing index induced by exposure either to the conditioning chamber or to a new context . Two additional age-matched groups ( ‘new context’ groups ) were conditioned and normalized freezing levels were evaluated as mice were exposed at test to a new context instead of the conditioning chamber one day or one year after conditioning ( n=4 mice per delay; n=8 total ) . Conditioned mice reexposed to the conditioning chamber froze in average 36 ± 4 . 7% more than the conditioned mice tested in the new context , demonstrating that the freezing levels were specific to the conditioning chamber . Error bars are mean ± SEM . ‘o’ indicate a significant comparison to 0 at p<0 . 05; asterisks a significant effect of the context at p<0 . 05 for ‘*’ and p<0 . 01 for ‘***’ . DOI: http://dx . doi . org/10 . 7554/eLife . 11862 . 00310 . 7554/eLife . 11862 . 004Figure 2 . Imaging brain activity in the medial temporal lobe ( MTL ) . ( A ) Location of counting frames for CA1 , CA3 , the perirhinal ( PER ) , the lateral entorhinal ( LEC ) , the postrhinal ( POR ) and the medial entorhinal cortices ( MEC ) . Task-induced Arc nuclear RNAs’ expression was detected on three nonconsecutive brain slices for each target area covering approx . 800 microns . ( B-G ) Cells activated during the retrieval of one day–old and one year-old memory traces in CA1 , CA3 and the LEC: Arc positive cells ( red arrowheads ) and an exemplar of non-activated cells Arc negative cells ( white arrowheads ) . Cell nuclei are labeled in blue with DAPI . ( B , C ) CA1 is engaged independently of the age of the memory trace while ( D , E ) activity levels in CA3 are negligible for very remote memories . ( F-G ) The parahippocampal areas , for example the LEC , are maximally engaged during the retrieval of very remote memories . DOI: http://dx . doi . org/10 . 7554/eLife . 11862 . 004 Freezing levels upon re-exposure to the conditioning context were high and did not significantly differ between delays , indicating that the memory for the footshock/context association was strong and that its strength was comparable between delays whether the memory was recent or remote ( comparisons to 0: all ps<0 . 015 , in average a 55 . 9 ± 3 . 4% increase in freezing compared to age-matched ‘no-shock’ groups; F ( 4 , 15 ) =0 . 26 , p=0 . 897; Figure 1B ) . Of note , statistical analysis of the data with or without normalization , e . g . direct analysis of the% resting time at test , yield the same results . Importantly , this result shows that a difference in memory strength across delays cannot account for the differences in the pattern of activation reported in the present study . In addition , we tested that the freezing behavior observed during re-exposure was specific to the conditioning context by exposing two additional conditioned groups to a new context either 1 day or 1 year after conditioning . In this case , freezing levels were strikingly lower than those observed by re-exposing the mice to the conditioning chamber , demonstrating that the freezing behavior was indeed specific to the conditioning context ( one day: conditioning context: 52 . 4 ± 2 . 5% vs new context: 19 . 8 ± 5 . 9%; one year: conditioning context: 62 . 5 ± 8 . 4% vs new context: 21 . 9 ± 8 . 385%; both ts>3 . 9 and ps<0 . 007; Figure 1C ) . Of note , a time-dependent generalization of the freezing behavior to a similar context has been reported in some studies testing subjects 15 to 42 days after conditioning i . e . for the retrieval of early remote memory ( Winocur et al . , 2007; Wiltgen et al . , 2010; Wiltgen and Silva , 2007 ) as well as a generalization to different contexts depending on experimental conditions ( multiple footshocks , preexposure to conditioning context etc . . ) ( Wang et al . , 2009; Kitamura et al . , 2012; Biedenkapp and Rudy , 2007 ) . No such generalization was seen in our study one year after conditioning ( Figure 1C ) as the new testing context was designed to be as different as possible from the conditioning context ( in shape , color , cues , heights and width , odor , background noise , lightning and texture ) , the protocol did not include preexposure to the conditioning context and only one footshock was delivered . Similar findings were also observed in studies investigating the retrieval of memory at remote delays ( 50 days and 16 months , respectively; Anagnostaras et al . , 1999; Gale et al . , 2004 ) . An alternative way to express conditioned fear responses is to report the% resting time at test . However , as mentioned in the material and methods , the conditioning baseline of one of the ten groups studied ( the one year-shock group ) was slightly lower than that of the other shock groups ( F ( 4 , 15 ) =3 . 37 , p=0 . 037; post-hoc Tukey: 1 year vs . 1 day p=0 . 041; vs . 1 month or 6 months: p=0 . 079 and p=0 . 071 , respectively; importantly though , no significant shock vs no-shock effect was found at this delay: t ( 6 ) =1 . 97; p=0 . 131 ) . Since a similar pattern was found at test ( shock groups: F ( 4 , 15 ) =6 . 85 , p<0 . 005 , post hoc Tukey: 1 year vs . all other delays ps<0 . 010 , except 6 months p=0 . 068 ) focusing solely on the% resting time at test could have led to a misinterpretation of the test data since the lower freezing index at test is carried by a lower baseline at conditioning . Therefore , freezing levels discussed in the manuscript have been normalized by the conditioning baseline for each group . Nevertheless , for the sake of transparency , we also performed the analysis on non-normalized freezing levels ( e . g . on% resting time at test ) . This analysis led to the same conclusion as the analysis of the normalized data: shocked animals display significantly higher freezing levels than the no-shock groups at all delays , reflecting a successful memory for the context-footshock association ( two-way ANOVA with ‘shock’ and ‘delay’ as factors and ‘% resting time at test’ as dependent variable: shock effect: F ( 1 , 30 ) =275 . 9 , p<0 . 001; delay effect: F ( 4 , 30 ) =15 . 147 , p<0 . 001; interaction shock x delay effect: F ( 4 , 30 ) =1 . 21 , p=0 . 328; all ts>5 . 25 , ps<0 . 012 ) . First , we tested if CA3 and CA1 were differentially recruited as memory aged and especially if CA3 was engaged for the retrieval of very remote memories . Statistical analysis of the task-induced Arc expression revealed that , the contribution of CA3 to the retrieval of memory is restricted to that of recent and early remote memory traces ( e . g . to up to 1 month-old memories ) while it was negligible for very remote memories ( 6 months- and 1 year-old memory traces; Figure 3A ) . Indeed , CA3 was no longer engaged for memory retrieval at this time point ( F ( 4 , 15 ) =26 . 73 , p<0 . 001; Tukey posthoc: recent or early remote traces vs very remote traces: all ps<0 . 028; comparisons to 0: up to 1 month-old traces: ps<0 . 010 , 6 months and 1 year old traces: n . s ) . Importantly , this was only the case for CA3 as revealed by a significant region by delay interaction and further statistical analysis showing that , in contrast to CA3 , CA1 was engaged for all delays and overall more activated than CA3 ( CA1 and CA3 comparisons to 0: all ps<0 . 033; area effect: F ( 1 , 30 ) =62 . 74 , p<0 . 001; delay effect: F ( 4 , 30 ) =20 . 15 , p<0 . 001; region by delay interaction effect: F ( 4 , 30 ) =5 . 84 , p=0 . 001; post-hoc analysis: CA1 vs CA3: all ts>3 . 89 and ps<0 . 030 but for 6 months t=2 . 36 , p=0 . 099 and 1 week n . s; Figure 3A ) . In addition , within-area analysis showed that activation levels in CA1 were higher one day after conditioning compared to all other delays but comparable between the remaining delays ( F ( 4 , 15 ) = 9 . 46 , p=0 . 001; Tukey posthoc: one day higher than any other delay: all ps<0 . 041 but for 1 month p=0 . 090; one week to one year: all ps>0 . 083 but 1 month vs 6 months: p=0 . 038 ) . Of note , analysis of the task-induced Arc expression with or without normalization yields the same results . This new finding of a time-limited involvement of CA3 for the retrieval of memory was further supported by significant correlations between memory performance and Arc levels in CA3 for recent and early remote memories , which failed to reach significant for very remote memory traces ( recent and early remote: R2=0 . 47 , p<0 . 001; very remote: R2=0 . 06 , p=0 . 351; Figure 3B–C ) while correlations were significant for CA1 independently to the age of the memory trace ( R2s . >0 . 40 , ps<0 . 008; Figure 3D , E ) . The lack of involvement of CA3 in the retrieval of very remote memories was further strengthened by the fact that levels of activation in CA3 did not differ between the memory-impaired and the memory-intact animals that were tested one year after conditioning ( intact vs impaired: freezing index: t ( 6 ) = 4 . 32 , p=0 . 005 , Arc expression: t ( 6 ) =1 . 08 , p=0 . 321 , comparison to 0: n . s; see Figure 4A and B ) . In a striking contrast and supporting a crucial role of CA1 in the retrieval of very remote memories , CA1 was not recruited in the memory-impaired group as opposed to the memory–intact group ( intact vs impaired: t ( 6 ) =3 . 38 , p=0 . 015; comparisons to 0: memory impaired: t ( 3 ) =0 . 08 , p=0 . 936; memory intact: t ( 3 ) =6 . 69 , p=0 . 007; see Figure 4B ) . Of note , these results are unlikely due to an age- related decline specific to CA3 since baseline Arc expression was comparable between and within CA1 and CA3 over time as shown by the absence of significant ‘area’ x ‘delay’ interaction , main area effects and posthocs analysis over time failing to reach significance ( area effect: ( F ( 1 , 30 ) = 2 . 30; p=0 . 14 ) , delay effect: F ( 4 , 30 ) =0 . 77;p=0 . 55 , area by delay interaction effect: ( F ( 4 , 30 ) =40; p=0 . 81; posthocs: delay effects: CA1: F ( 4 , 20 ) = 0 . 63 , p=0 . 65; CA3: F ( 4 , 20 ) =0 . 52 , p=0 . 72 ) and because Arc expression elicited by a maximal electroconvulsive shock , a standard positive control for this type of study , does not significantly differ between CA1 and CA3 ( Nakamura et al . , 2013 ) . Thus , these findings suggest that the decline of expression seen in CA3 is specific to the task . These results show for the first time a clear functional segregation of CA3 and CA1 in the retrieval of memory over time , with CA3 contributing only to the retrieval of recent and early remote memories and CA1 being recruited independently of the age of the memory trace . 10 . 7554/eLife . 11862 . 005Figure 3 . Activity patterns in CA1 and CA3 over time and correlations with memory performance . ( A ) CA3’s contribution to memory retrieval depends on the age of the memory trace , not CA1’s , as CA3’s was no longer significantly recruited for the retrieval of very remote memories . ( B , C ) CA3’s activity levels were predictive of memory performance only for the retrieval of recent and early remote memories while D , E ) CA1’s activity was independently of the age of the memory trace , Error bars are mean ± SEM . ‘o’ indicate a significant comparison to 0 at p<0 . 05; asterisks a t-test at p<0 . 05 for ‘*’ and at p<0 . 01 for ‘***’ . DOI: http://dx . doi . org/10 . 7554/eLife . 11862 . 00510 . 7554/eLife . 11862 . 006Figure 4 . Memory performance of ‘memory-intact’ and ‘memory- impaired’ mice tested one year after conditioning and corresponding MTL patterns of activity . ( A ) Behavioral performance: ‘memory-impaired’ mice froze significantly less than ‘memory-intact’ mice at test , reflecting impaired memory retrieval in this group . ( B ) Activity patterns in CA1 and CA3: in contrast to mice that successfully retrieved the footshock-context association one year after conditioning , CA1 was not recruited in those that had impaired memory . In addition , activity levels in CA3 were comparable between the two groups , underlining the critical role of CA1 in the retrieval of very remote memories . ( C ) Activity patterns in parahippocampal areas: parahippocampal areas of the ‘memory-impaired’ mice were recruited to much lesser extent than those of ‘memory-intact’ animals . Since this reduced activation did not lead to successful memory retrieval , this result suggests that CA1 might play a role as important as the cortical areas in the retrieval of very remote memories in memory-intact mice . Error bars are mean ± SEM . ‘o’ indicate a significant comparison to 0 at p<0 . 05; asterisks a t-test at p<0 . 05 for ‘*’ and at p<0 . 01 for ‘***’ . DOI: http://dx . doi . org/10 . 7554/eLife . 11862 . 006 Second , we investigated the extent to which parahippocampal areas contribute to the retrieval of recent , remote and very remote memory traces . All parahippocampal areas participated qualitatively to a comparable extent to retrieving memory over time as they were all maximally engaged for the retrieval of the oldest traces . In addition , the PER and the LEC were more recruited than the POR and the MEC specifically for the oldest memory traces ( PER vs LEC , PER vs POR , POR vs MEC and LEC vs MEC: delay effects: Fs ( 4 , 30 ) >18 . 89 , ps<0 . 001; within-area analysis; LEC , PER and POR: all Fs ( 4 , 15 ) >6 . 69 , ps<0 . 003; tukey posthocs: recent or early remote memories versus very remote memories: all ps<0 . 006; MEC: all ps<0 . 047 but for 1 month vs . 1 year: n . s; area effects: PER vs POR and LEC vs MEC: Fs ( 1 , 30 ) >10 . 91 , ps<0 . 002; comparisons for 6 months and one year: ts ( 3 ) >3 . 77; ps<0 . 034 while PER vs LEC and POR vs MEC: n . s; interactions effects: PER vs POR and LEC vs MEC: Fs ( 4 , 30 ) >2 . 78 , ps<0 . 045 while PER vs LEC and POR vs MEC: n . s ) . In addition , all areas were activated across all delays ( comparisons to 0: all ts>3 . 52 , ps<0 . 039; Figure 5A , B ) at the exception of the MEC and the POR for recent memories and the PER one week after conditioning ( all ts<2 . 22 , ps>0 . 113 ) . However , this latter result might reflect a slightly larger SEMs for these time points rather than a true functional difference between brain areas since comparisons of activity levels between the LEC and the MEC and between the PER and the POR for the recent memory time points did not yield any significant differences ( ts ( 3 ) <1 . 98; ps>0 . 142 ) . Of note , analysis with or without normalization of the task-induced Arc expression yields the same results . Interestingly , parahippocampal areas of animals showing impaired memory when tested one year after conditioning were also significantly activated ( see Figure 4C ) , albeit less than those of animals whose memory was intact ( comparisons to 0: memory intact: ts ( 3 ) >3 . 821 , ps<0 . 032; memory impaired: ts ( 3 ) >3 . 929 , ps<0 . 029 , except MEC: t ( 3 ) =2 . 65 , p=0 . 077; intact vs . impaired: LEC and PER: ts ( 6 ) >5 . 58 , ps<0 . 001; POR: t ( 6 ) = 2 . 27 , p=0 . 063; MEC: t ( 6 ) =1 . 67 , p=0 . 146 ) . This result might suggest that , in the memory-impaired group , a partial memory representation had been consolidated and stored in the cortical areas , but that a weak activation of these areas alone does not suffice for successful retrieval . Thus , in summary , no qualitative differences were observed in the patterns of activation of the parahippocampal areas over time . However , the PER and LEC , which receive stronger connections from the amygdala than the MEC and the POR , were more activated than the MEC and the POR during the retrieval of the most remote memories ( 6 months- and one-year old traces ) , a time point which coincides with the epoch at which cortical areas are thought to play a strong role in the retrieval of memories . This result suggests a segregation of the parahippocampal areas in terms of memory processes rather than in terms of spatial or non-spatial information processing at this time point . In addition , these results further underline the critical role of CA1 in the retrieval of very remote memories as activation of the parahippocampal areas alone appeared not to be sufficient for a successful retrieval of the memory trace at this time point . 10 . 7554/eLife . 11862 . 007Figure 5 . Activity patterns in the parahippocampal areas over time and apparent ( and possibly misleading ) over-time shift from the involvement of the hippocampus to the involvement of the parahippocampal region in memory retrieval . All parahippocampal areas were maximally engaged for the most remote memories and patterns of activity were comparable between the ( A ) LEC and PER and ( B ) the MEC and POR . Furthermore , in line with the existence of stronger projections from the amygdala to the PER and LEC than to the POR and MEC , and a more important role of the cortical areas for the most remote memories than for more recent ones , the levels of activity in the PER and the LEC were higher than those in the POR and MEC during the retrieval of very remote memories , providing further support to an emerging theory according to which the parahippocampal areas might be segregated in terms of memory types/processes rather than in terms of information content ( spatial versus spatial information; Eichenbaum et al . , 2007; of note , for the sake of clarity , significant area differences between graphs A and B are not shown ) . ( C ) Contribution of the hippocampal CAs and parahippocampal region to memory retrieval over time: when CA1 and CA3 activity levels are not dissociated , the overall contribution of the hippocampus to the retrieval of very remote memories is largely underestimated when compared to that of CA1 shown in Figure 3A . Even in this case though , the hippocampus is still significantly recruited at all times . In contrast , overall activity of the parahippocampal region is comparable to that of any of the parahippocampal areas , with a maximal activation during the retrieval of very remote memories ( see Figure 5A and B ) . These results underline the need of dissociating CA1 and CA3 activity patterns to better understand the contribution of the hippocampus to the retrieval of memory over time . Importantly , when these contributions are not dissociated , the apparent over-time shift remains relative at most ( and not absolute ) since both areas are significantly activated at all delays . Error bars are mean ± SEM . ‘o’ indicate a significant comparison to 0 at p<0 . 05; asterisks a t-test at p<0 . 05 for ‘*’ and at p<0 . 01 for ‘***’ . DOI: http://dx . doi . org/10 . 7554/eLife . 11862 . 007 Also , to focus on the fear response reflecting the context-footshock association ( and not that contributed by other parameters ) , data presented in Figures 3A , 4B-C , 5A-C represent the task-induced Arc expression observed in the shock groups from which was subtracted that of the ‘no-shock’ groups , which were subjected to the exact same experimental conditions but could not associate the context with an aversive experience ( i . e . the footshock ) . Nevertheless , for the sake of transparency , the same analyses were also performed on the task-induced Arc expression found in the ‘shock’ and ‘no-shock’ groups . These analyses led the same outputs . In the hippocampus , activity levels in the CA1 of the conditioned mice remained high at all delays while CA3 was no longer recruited for the retrieval of memories that were 6 months or older . Three-way ANOVA with ‘shock’ , ‘delay’ and ‘area’ as factors and task-induced Arc expression as dependent variable showed that task-induced Arc expressions differed between CA1 and CA3 in function of the delay imposed and whether animals had been conditioned or not ( shock effect: F ( 2 , 90 ) =519 . 74 , p<0 . 001; delay effect: F ( 4 , 90 ) =3 . 15 , p=0 . 018 , area effect: F ( 1 , 90 ) =359 . 1 , p<0 . 001; shock x delay effect: F ( 8 , 90 ) =9 . 525 , p<0 . 001; interaction shock x area effect: F ( 2 , 90 ) , p<0 . 001; interaction area x delay effect: n . s; interaction shock x area x delay effect: F ( 8 , 90 ) =3 . 63 , p<0 . 001 ) . Moreover , further separate two-way ANOVAs with ‘shock’ and ‘delay’ as factors on activity levels and t-tests between shock and no-shock groups in CA1 or CA3 that showed task-induced Arc expression in CA1 were higher in the conditioned groups than the no-shock groups for all delays , with the activation being the highest for the one day delay ( CA1: shock effect: F ( 1 , 30 ) =246 . 88 , p<0 . 001; delay effect: F ( 4 , 30 ) =1 . 79 , n . s; interaction shock x delay effect: F ( 4 , 30 ) =9 . 79 , p<0 . 001; t-tests all ts>2 . 68 , ps<0 . 036; one way ANOVA on CA1 activity levels in ‘shock’ groups: F ( 4 , 15 ) =10 . 65 , p<0 . 001 , post-hoc Tukey: 1 day vs . all other delays ps<0 . 032; 6 months vs . 1 month p=0 . 046 , all other n . s ) while task-induced Arc expressions were higher in shock than in no-shock groups only for recent and early remote memories in CA3 ( shock effect: F ( 1 , 30 ) =83 . 38 , p<0 . 001; delay effect: F ( 4 , 30 ) =7 . 82 , p<0 . 001; interaction shock x delay effect: F ( 4 , 30 ) =13 . 35 , p<0 . 001; t-tests shok vs . no shock: recent and early remote memories ( t ) s>6 . 14 , ps<0 . 001; very remote memories ( t ) s<1 . 59 , ps>0 . 161 ) ; one way ANOVA on CA3 activity levels in ‘shock’ groups: F ( 4 , 15 ) =18 . 83 , p<0 . 001; post-hoc Tukey: 1 year vs . recent and early remote memories: ps<0 . 001; 1 year vs . 6 months: n . s; 6 months vs . 1 day and 1 week ps<0 . 010; 6 months vs . 1 month p=0 . 076 , all other n . s ) . In contrast , task-induced Arc expression in the ‘no-shock’ groups did not significantly differ accross delays in CA1 or CA3 ( one way ANOVAs: CA1 F ( 4 , 15 ) =1 . 92 , ns; CA3 F ( 4 , 15 ) =0 . 548 , n . s ) . In the parahippocampal region , the PER , LEC , MEC and POR are more recruited for the retrieval of 6 months-old and older memories . For the LEC and the PER , a three way-ANOVA with ‘shock’ , ‘delay’ and ‘area’ as factors and task-induced Arc expression as dependent variable showed that the level of the recruitment of both areas is comparable and depends on the age of the memory trace and wether mice received a footshock or not ( shock effect: F ( 2 , 90 ) =465 . 9 , p<0 . 001: delay effect: F ( 4 , 90 ) =37 . 75 , p<0 . 001 , area effect: n . s; interaction effects: shock x area , area x delay and shock x area x delay: n . s; interaction shock x delay effect: F ( 8 , 90 ) =34 . 46 , p<0 . 001 ) . These results were confirmed by two-way ANOVAs performed separately on LEC and PER activity levels ( LEC: shock effect: F ( 2 , 45 ) =235 . 86 , p<0 . 001; delay effect: F ( 4 , 45 ) =17 . 18 , p<0 . 001; interaction shock x delay effect: F ( 8 , 15 ) =16 . 98 , p<0 . 001; PER: shock effect: F ( 2 , 45 ) =230 . 04 , p<0 . 001; delay effect: F ( 4 , 45 ) =21 . 39 , p<0 . 001; interaction shock x delay effect: F ( 8 , 15 ) =17 . 82 , p<0 . 001 ) . Finally , further within group one-way ANOVAs in the LEC or the PER revealed that the delay at which these areas were the most recruited in the conditionined mice were 6 months and one year , while such an effect was not observed in the no-shock groups ( LEC: shock groups: F ( 4 , 15 ) =30 . 14 , p<0 . 001 , post-hoc Tukey: 6 months and 1 year vs . all other delays: ps<0 . 001 , all other ps: n . s . ; no-shock groups: F ( 4 , 15 ) =0 . 875 , n . s . ; PER: shock groups: F ( 4 , 15 ) =45 . 98 , p<0 . 001 , post-hoc Tukey: 6 months and 1 year vs . all other delays ps<0 . 001; 1 month vs . all other delays ps<0 . 030 , all other ps: n . s . ; no-shock groups: F ( 4 , 15 ) =1 . 15 , n . s . ) The patterns of activity in the MEC and the POR were comparable and linked to the age of the memory trace and wether mice were conditionined or not ( three way-ANOVA: shock effect: F ( 2 , 90 ) =174 . 51 , p<0 . 001: delay effect: F ( 4 , 90 ) =9 . 49 , p<0 . 001 , area effect: F ( 1 , 90 ) =2 . 05 , n . s; interaction effects: shock x area , area x delay and shock x area x delay: n . s , interaction shock x delay effect: F ( 8 , 90 ) =10 . 94 , p<0 . 001 ) . Results that were confirmed by two-way ANOVAs performed separately on MEC and POR activity levels ( MEC: shock effect: F ( 2 , 45 ) =66 . 35 , p<0 . 001; delay effect: F ( 4 , 45 ) =3 . 45 , p<0 . 015; interaction shock x delay effect: F ( 8 , 15 ) =4 . 31 , p<0 . 001; POR: shock effect: F ( 2 , 45 ) =117 . 22 , p<0 . 001; delay effect: F ( 4 , 45 ) =6 . 98 , p<0 . 001; interaction shock x delay effect: F ( 8 , 15 ) =7 . 34 , p<0 . 001 ) . Lastely , further within group one-way ANOVAs performed separately on the MEC and the POR activity levels revealed that the delays at which these areas were the most recruited in conditioned mice were the 6 months and one year delays , while such an effect was not observed in animals that did not received a shock at conditioning ( MEC: shock groups: F ( 4 , 15 ) =6 . 32 , p<0 . 005 , post-hoc Tukey: 6 months vs . all other delays: ps<0 . 028; 1 year vs . 1 day p=0 . 049; 1 year vs . 1 week and 1 month p=0 . 088 and 0 . 080 , respectively; no-shock groups: F ( 4 , 15 ) =0 . 167 , ns; POR: shock groups: F ( 4 , 15 ) =18 . 41 , p<0 . 001 , post-hoc Tukey: 6 months and 1 year vs . all other delays ps<0 . 007; all other ps: ns; no-shock groups: F ( 4 , 15 ) =0 . 304 , ns ) To compare more readily our results to the vast majority of studies in humans which do not dissociate the contribution of CA1 from that of CA3 to memory retrieval , nor that of the different parahippocampal areas , we also analyzed the overall activity of the hippocampal CAs and that of the parahippocampal region by averaging CA1 and CA3 activity levels on the one hand and those of the parahippocampal areas on the other hand ( Figure 5C ) . Activity in the hippocampus was higher than that of the parahippocampal region for the retrieval of recent and early remote memories while an inversed relationship emerged for the retrieval of very remote memories ( area effect: F ( 1 , 30 ) =9 . 984; p=0 . 004; delay effect: F ( 4 , 30 ) =1 . 539; p=0 . 216; delay x area interaction effect: F ( 4 , 30 ) =61 . 629 , p<0 . 001; post-hoc analysis: recent and early remote memory: ts>4 . 21 , ps<0 . 024; very remote memories: ts>6 . 07 ps<0 . 009 ) . Importantly , even under this condition , e . g . when the contribution of CA1 was not dissociated from that of CA3 , the hippocampus was significantly recruited independently of the age of the memory trace ( comparisons to 0: all ts>3 . 54 ps<0 . 038 ) in line with the multiple trace theory . Furthermore , in this case , the hippocampus appeared to contribute mainly to retrieving recent and early remote memories ( F ( 4 , 19 ) =16 . 21 , p<0 . 001; Tukey posthocs: recent or early remote memories vs very remote memories: all ps<0 . 024 ) . Conversely , the parahippocampal region was maximally activated for retrieving very remote memories ( F ( 4 , 19 ) =52 . 51 , p<0 . 001; recent or early remote memories vs very remote memories: all ps<0 . 001 ) but was also recruited for recent and early remote memories ( comparisons to 0: ts ( 3 ) >3 . 79; ps<0 . 032 ) . In conclusion , this analysis underlines the fact that , if CA1 and CA3’s contribution are not dissociated , the hippocampal contribution to the retrieval of very remote memories might be largely underestimated . Nevertheless , even in this case , the over-time shift from an engagement of the hippocampus to that of an engagement of the parahippocampal region in retrieving memory is only relative and not absolute , given that the hippocampus is still significantly activated even for the retrieval of very remote memories . Investigating memory retrieval over such an extended period of time enabled us to provide the first evidence of a time-limited contribution of CA3 to the retrieval of memory while , in striking contrast , CA1 remained involved regardless of the age of the memory trace . In addition , parahippocampal areas were substantially more involved in retrieving very remote memory traces than more recent ones and their activity patterns appeared to reflect memory demands rather than information content . To our knowledge , no study has investigated the role of CA3 in retrieving remote memory traces in humans and only one studied early remote memory in animals ( e . g . up to one month-old traces which roughly corresponds to a 4 year-old memory trace in humans based on life expectancy; Gusev et al . , 2005 ) . Thus , the contribution of CA3 to the retrieval of memory traces as old as what is considered to be a remote/very remote memory trace in humans ( in the order of decades ) is not known . For this reason , we studied 6 months-and one year- old memories traces in mice , which roughly correspond to 20 and 40 year-old memory traces in humans . Doing so enabled us to report for the first time the lack of engagement of CA3 in the retrieval of very remote memories . This time-limited role of CA3 was supported by a lack of recruitment during the retrieval of the most remote memory traces ( 6 months- and one year-old ) while CA3 was strongly activated for recent and early remote memories ( up to 1 month-old traces ) . This result was confirmed by a significant correlation between CA3 activity levels and memory performance during the retrieval of the recent and early remote memories , which did not hold for very remote traces . The approach we have adopted does not allow for the characterization of the specific process by which CA3 fails to aid in retrieving memory . However , since computational studies and studies that have limited their investigations to recent and early remote memory traces have speculated that CA3’s contribution to memory retrieval would increase as memory ages because of an increasing demand on the completion of representations using partial or degraded features of these stored representations ( the ‘pattern completion theory’: Rolls et al . , 1997 ) , we speculate that the lack of engagement of CA3 in retrieving very remote memories stems from a failure to ‘pattern complete’ memory representations . Indeed , by this time , these very old memory representations might have degraded to such an extent that it is impossible to use any specific features as an efficient retrieval cue . This hypothesis would go along with reports demonstrating that details of memory representations are lost as memory ages ( Wiltgen and Silva , 2007; Wiltgen et al . , 2010 ) and will require further investigations to be tested . In striking contrast to the limited involvement of CA3 , we found an ongoing involvement of CA1 in memory retrieval . This result is in agreement with the only two studies specifically investigating the role of CA1 in remote memory retrieval . Indeed , transient inhibition of cell firing in CA1 in mice using optogenetics drastically impaired fear memory retrieval , one or twenty eight days after conditioning ( Goshen et al . , 2011 ) . Likewise , a recent fMRI study in humans showed impairments in autobiographic and episodic memory for traces even older than 30 years-old following focal CA1 lesions ( Bartsch et al . , 2011 ) . Our result of a selective involvement of CA1 in the retrieval of very remote memories at this time point is supported by the lack of a recruitment of this area in the memory-impaired group tested one year after conditioning as opposed to a strong activation in the memory intact group and the fact that CA3 was not recruited in either group ( Figure 4B ) . Importantly , using a high resolution molecular imaging technique , we bring the first evidence that the persistent engagement of the hippocampus in memory retrieval does not involve CA3 . This result , in turn , suggest that hippocampal patterns of activation observed in previous human fMRI studies investigating the retrieval of very remote memory , might essentially be driven by CA1 , but could not be specifically identified as such because of the limited spatial resolution of standard fMRI techniques ( Ryan et al . , 2001; Rekkas and Constable , 2005; Bonnici et al . , 2012 ) . Moreover , the fact that only CA1 is activated for these late time points suggests a preferential involvement of the temporoammonic pathway over that of the trisynaptic loop for the retrieval of very remote memories . Robust evidence for this hypothesis is not yet available for very remote memories , but lesions of the temporoammonic pathway in rats specifically impaired early remote memory ( 28 days-old ) in rats tested in a Morris watermaze task , while recent memory ( one day-old ) was spared . This result suggests that direct entorhinal cortex inputs to CA1 contribute to the retrieval of early remote memory , possibly by further consolidating memories that were not consolidated after 28 days ( Remondes and Schuman , 2004 ) . In the present study , because we investigated a much larger time-window and found that CA1 was still activated in addition to the parahippocampal cortical areas , but not CA3 , we speculate that the temporoammonic pathway is not only important for the consolidation of memories but also for the retrieval of already consolidated memories . Furthermore , parallel to the disengagement of CA3 in the retrieval of very remote memories , we found a maximal involvement of the parahippocampal areas . The hypothesis of a functional segregation of the parahippocampal region between a 'PER-LEC stream' and a 'POR-MEC stream' is essentially based on studies investigating the role of the PER during the retrieval of recent memory , while much fewer have addressed the specific role of the POR , the LEC and the MEC within this frame and during the retrieval of early remote memories . In addition , no study has yet investigated their contribution for very remote memories , despite the belief that the contribution of cortical areas to memory retrieval would be the largest for the most remote memories . Our results for recent and early remote time-points match published data according to which , for example , the PER and the POR are recruited to a similar extent for contextual fear conditioning and the LEC and MEC contribute to a similar extent to memory retrieval ( Bucci et al . , 2002; Burwell et al . , 2004; Izquierdo et al . , 1997; Goshen et al . , 2011 ) . In addition , we show for the first time that the contribution of the PER , the POR , the LEC and the MEC is indeed the largest for the retrieval of very remote memories . Patterns of activation were comparable between the LEC and the PER and between the MEC and the POR . Interestingly though , LEC and PER were more activated than the MEC and the POR during the retrieval of very remote memory traces . This result appears to rather reflect the type of memory process the areas were engaged into ( fear conditioning ) rather than the content of information processed ( contextual information ) , supporting a view that has recently emerged from the literature ( Eichenbaum et al . , 2007; Beer et al . , 2013 ) . Indeed , the amygdala which plays a crucial role in fear conditioning , sends stronger projections to the PER/LEC than to the MEC/POR ( Pitkänen et al . , 2000 ) and higher activity levels in the POR and MEC than in the PER and LEC would have been expected if information content was primarily reflected in the pattern of activation observed . Altogether , our data show a preferential role of the parahippocampal region for the retrieval of very remote memories as opposed to that of early remote and recent memories , matching the pattern commonly reported for the prefrontal cortex ( Frankland and Bontempi , 2005 ) . In summary , our results provide the first evidence of a disengagement of CA3 for the retrieval of very remote memories and identify CA1 as the main player at this time point . These results are consistent with the view that at least the hippocampal subfield CA1 is involved in retrieving memory independently of the age of memory trace , possibly by providing support to cortical areas that include the parahippocampal areas . Of note , averaging activity levels of the hippocampal areas on the one hand and those of the parahippocampal areas on the other hand revealed that both the hippocampus and the parahippocampal region were activated independently of the age of the memory trace ( Figure 5C ) . Hence , the gradual “transfer” of the memory trace from the hippocampus to the cortex is unlikely to be an all-or-nothing phenomenon , but seems to go along with a shift in the relative contribution of the two regions as the memory trace ages , and possibly a shift in the contribution of the networks they belong to . More specifically , our results suggest that the trisynaptic loop could play a preponderant role in retrieving early memories as reflected by an involvement of CA3 , CA1 and the parahippocampal areas at this stage , while the retrieval of very remote memories would require an heavier contribution of the temporoamonic pathway as indicated by the lack of recruitment of CA3 and a strong activation of CA1 and the parahippocampal areas . 8–12 weeks old male C57BL/6 mice ( n=72 total ) bred at the Ruhr-Universität Bochum were used . The age of the animals at the term of the experiment varied between 8 and 70 weeks . Animals were housed in groups of 2–3 animals , kept under reversed 12-hr light/dark cycle ( 8:00 a . m . light off; 8 . 00 p . m . light on ) and tested during their active phase . Access to food and water was ad libitum and all procedures were approved by the Ruhr-Universität Bochum Institutional Animal Use Committee and the LANUV ( 84-02 . 04 . 2013 . A419 ) To reduce type 1 error and keep type 2 errors low , only a priori hypotheses were tested , e . g . statistical analysis were restricted to the comparisons of task-induced brain activity in and between CA1 and CA3 , PER and LEC or POR , MEC and POR or LEC , the hippocampal CAs and the parahippocampal region . For these comparisons two-way ANOVAs with 'area' and 'delay' as factors and 'task-induced Arc expression' as the dependent variable were used . In addition , one way ANOVAs followed by tukey post-hocs were performed for within-areas analysis and independent paired t-tests for between-areas comparisons .

There are two schools of thought about what role the hippocampus – a region of the brain – plays in memory . Some neuroscientists think that it is involved in retrieving all memories . Others believe that its contribution is restricted to the retrieval of recent memories , while a neighboring part of the brain called the parahippocampal region takes over to retrieve older memories . The hippocampus contains two distinct areas called CA1 and CA3 , which have recently been suggested to have , at least partially , separate roles . For example . previous studies have shown that CA3 plays an important role in processes that tend to be less efficient as time goes by . However , it remains unclear whether CA1 and CA3 contribute equally to the retrieval of recent and older memories . Lux et al . addressed this question by observing brain activity in mice as they retrieved recent and older memories . The experiments show that both areas of the hippocampus are involved in retrieving recent memories , but that only the CA1 area is involved in the retrieval of older memories . The parahippocampal region is much more active during the retrieval of older memories than recent ones . These findings clarify the role of the hippocampus in memory by showing that it is involved in the retrieval of both recent and older memories . The next steps will be to better understand how the CA1 and CA3 areas contribute to memory and to pin point the specific molecular mechanisms these regions rely on to do so .

2016

Imaging a memory trace over half a life-time in the medial temporal lobe reveals a time-limited role of CA3 neurons in retrieval

The actomyosin cytoskeleton is a primary force-generating mechanism in morphogenesis , thus a robust spatial control of cytoskeletal positioning is essential . In this report , we demonstrate that actomyosin contractility and planar cell polarity ( PCP ) interact in post-mitotic Ciona notochord cells to self-assemble and reposition actomyosin rings , which play an essential role for cell elongation . Intriguingly , rings always form at the cells′ anterior edge before migrating towards the center as contractility increases , reflecting a novel dynamical property of the cortex . Our drug and genetic manipulations uncover a tug-of-war between contractility , which localizes cortical flows toward the equator and PCP , which tries to reposition them . We develop a simple model of the physical forces underlying this tug-of-war , which quantitatively reproduces our results . We thus propose a quantitative framework for dissecting the relative contribution of contractility and PCP to the self-assembly and repositioning of cytoskeletal structures , which should be applicable to other morphogenetic events . In many developmental and cellular contexts , actin filaments construct complex and highly dynamic structures to accomplish cell shape changes such as in migration and cytokinesis ( Munjal and Lecuit , 2014 ) . Correct positioning of the actin filaments is essential . In polarized migrating cells , actin flows posteriorly and becomes associated with myosin II at the trailing edge to propel the cell forward ( Cramer , 2010 ) . In cytokinesis of vertebrate cells , the equatorial ring is established in many incidences by a cortical flow of actin filaments ( Bray and White , 1988 ) driven by myosin contractility and is concentrated at the equator to ensure correct cell division ( Cao and Wang , 1990; DeBiasio et al . , 1996; Mayer et al . , 2010 ) . Despite the importance of the proper positioning of the actin cytoskeleton , our understanding of how cell polarity contributes to the organization of the cytoskeleton , and vice versa , is still incomplete . In Caenorhabditis elegans early embryogenesis , a flow of cortical myosin and F-actin towards the anterior pole carries PAR polarity proteins , which in turn modulate the actomyosin dynamics ( Munro et al . , 2004; Mayer et al . , 2010 ) . Emerging evidence also point to a role for the Wnt/planar cell polarity ( PCP ) pathway in modulating cytoskeleton dynamics through its key mediators , Rho GTPases , which exert effects on actin polymerization and myosin contractility ( Schlessinger et al . , 2009 ) , although the mechanisms underlying this cross-talk remain obscure . On the other hand , in vitro experiments on reconstituted cytoskeletal structures ( Surrey et al . , 2001 ) , as well as recent mathematical models ( Kruse et al . , 2005; Hannezo et al . , 2015 ) suggest that actomyosin gels could have the properties to self-assemble , but the applicability of these findings to in vivo situations is not yet clear . Therefore , the interplay between self-assembly and polarity signals that organize the cytoskeleton remains largely unexplored . The Ciona notochord is a transient embryonic structure , which is composed of 40 post-mitotic cells that are arranged in a single file after convergent/extension ( C/E ) . Following C/E , the coin-shaped cells undergo continuous elongation along the anterior–posterior axis ( Cloney , 1964; Miyamoto and Crowther , 1985; Jiang and Smith , 2007; Dong et al . , 2009 ) , acquiring a drum shape ( Figure 1A ) . Our previous studies show that an actomyosin contractile ring is present in the basal equator ( Dong et al . , 2011 ) and produces a circumferential constriction . The force generated by the constriction is transmitted three dimensionally from the basal cortex towards anterior and posterior lateral domains through an incompressible cytoplasm , driving notochord cell elongation ( Dong et al . , 2011; Sehring et al . , 2014 ) ( Figure 1B , C ) . The actomyosin ring is maintained by a bi-directional cortical flow and is under constant turnover in a manner remarkably similar to that of the cytokinetic ring during cell division . The position of contractile rings influences notochord cell shape and elongation . For example , in α-actinin mutants , the ring cannot maintain its position at the equator , and consequently , the cells fail to elongate but acquire an asymmetric shape ( Sehring et al . , 2014 ) . However , the mechanism of positioning the contractile ring in the equator of the notochord cells is unknown . This question is also of crucial relevance to our understanding of cytokinesis , where the position of the actomyosin ring is critical for the cells to divide properly ( Sedzinski et al . , 2011 ) and to direct the distribution of cell-fate determinants correctly ( Clevers , 2005; Gómez-López et al . , 2014 ) . 10 . 7554/eLife . 09206 . 003Figure 1 . Establishment and relocation of anterior basal cortical actin filaments . ( A ) Ciona embryos at 16 . 5 and 23 . 5 hr post fertilization ( hpf ) . Following cell intercalation , notochord cells at 16 . 5 hpf are coin-shaped ( one is highlighted in the insert ) . At 23 . 5 hpf , cells are cylindrically elongated , and a circumferential constriction is present midway between the two poles ( red arrowheads in insert ) . ( B ) Notochord cells are labeled with Lifeact-mEGFP ( green ) for actin and Anillin-mCherry ( red ) for the nucleus . Red arrowheads indicate the equatorial constrictions; yellow brackets outline the circumferential actin rings at the equatorial region . ( C ) A diagram of an elongating notochord cell at the onset of lumen formation with the nomenclature used in this paper . Small dark green arrows indicate the bi-directional cortical flow of actin filaments contributing to the construction of the actin ring . ( D ) Notochord cells labeled with Lifeact-mEGFP ( green ) for actin and Anillin-mCherry ( red ) for the nucleus . At the start of intercalation ( 11 . 5 hpf ) , actin is evenly distributed in the cell boundaries ( white arrows ) . During cell intercalation , basal cortical actin patches ( white arrowheads ) appear adjacent to the anterior lateral domain . The actin patches begin to fuse next to the anterior pole of the cells ( yellow arrowheads ) . The intensity was measured at positions of arrowheads . Vertical green bars indicate lateral domains . ( E ) Notochord cells expressing Lifeact-mEGFP for actin . These images are from Video 1 . After cell intercalation , basal cortical actin patches ( arrowheads ) continue to fuse , forming a circumferential ring next to the anterior lateral domain , which subsequently relocates to the equator , as cells elongate . ( F ) Mean distances between the anterior lateral domain and the cortical actin ring ( black ) , and the posterior lateral domain and the cortical actin ring ( red ) during cell elongation ( n = 7; error bars = SEM ) . ( F′ ) Mean ring width over time ( n = 7; error bars = SEM ) . ( G , H ) Blebbistatin inhibits relocation of anterior basal cortical actin filaments and cell elongation . Notochord cells are labeled with mCherry-UtrCH for actin . Embryos are either treated with DMSO ( G ) or incubated in blebbistatin ( H ) for 120 min . Anterior to the left in all images . Scale bars in A represent 50 μm; in inserts , 20 μm; in B–E , G , H represent 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 09206 . 00310 . 7554/eLife . 09206 . 004Figure 1—figure supplement 1 . Establishment and relocation of anterior basal cortical myosin . Notochord cells expressing mCherry-MRLC ( red ) and Lifeact-mEGFP for actin ( green ) . In coin-shaped cells , cortical basal myosin is enriched adjacent to the anterior pole ( white arrowheads ) . The myosin ring colocalizes with actin ( cyan arrowheads ) and relocates towards the equator ( yellow arrowheads ) during cell elongation . Anterior to the left . Scale bars , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 09206 . 004 In addition , notochord cells acquire a subtle yet stable anterior/posterior ( A/P ) polarity: nuclei in all but the most posterior cell become localized at the posterior pole of the cell , while the classical PCP protein prickle is localized at the anterior pole of the cell during the notochord cell elongation ( Jiang et al . , 2005; Newman-Smith et al . , 2015 ) . Whether the PCP pathway contributes to the process of cell elongation and whether and how PCP components affect contractile ring formation and positioning remains mysterious . In this study , we investigated the processes of actin ring formation and found , through dynamic imaging , physical modeling , as well as comparative and genetic analyses that the actomyosin contractility and PCP pathway work antagonistically to achieve a robust localization of the cytoskeleton . To analyze the development of the equatorial actin ring , we followed the expression of actin markers Lifeact-mEGFP and mCherry-UtrCH in notochord cells from the onset of C/E . Both Lifeact-mEGFP and mCherry-UtrCH bind to endogenous actin without interfering with its dynamics ( Burkel et al . , 2007; Riedl et al . , 2008 ) and were shown previously to have the same localization pattern as the endogenous protein in Ciona intestinalis ( Sehring et al . , 2014 ) . Before C/E ( 11 . 5 hpf , hours post fertilization ) , actin was uniformly localized in the basal cortex , with high concentration at notochord cell–cell contacts ( arrows in Figure 1D ) . During cell intercalation ( 16 . 5 hpf ) , when the cells start to align into a column , the low and evenly distributed actin in the basal cortex was replaced by patches of cortical actin accumulations close to the anterior pole ( white arrowheads in Figure 1D ) . With the alignment of the cells into a column ( 17 hpf ) , these actin patches connected into an actin ring spanning the entire circumference of the basal domain next to the anterior pole of the coin-shaped cell ( yellow arrowheads and intensity graph in Figure 1D ) . Time-lapse recordings revealed that this early anterior ring observed in coin-shaped cells is the precursor of the equatorial ring ( Video 1 ) . While the cells elongated , the ring relocated from the anterior pole to the equator of the cells ( white arrowheads in Figure 1E ) . To ascertain if the relocation of the cortical actin ring was a continuous movement , we measured the distance between the edges of the ring and the two cell poles over time . While the length between the ring and the posterior pole grew only minimally , the length between the anterior pole and the ring increased steadily , until the distance from the ring to both poles was similar ( Figure 1F ) . Concomitantly with ring relocation and cell elongation , the ring width increased ( Figure 1F′ ) . Myosin filaments ( labeled with myosin regulatory light chain fused to mCherry , mCherry-MRLC ) also formed a circumferential ring in the anterior basal cortex at the end of C/E . Similarly to the actin ring , the myosin ring subsequently relocated to the equator ( Figure 1—figure supplement 1 ) , where it remained for the rest of the elongation process . 10 . 7554/eLife . 09206 . 005Video 1 . Relocation of cortical actin during cell elongation . Notochord cells expressing Lifeact-mEGFP were recorded every min . Frame rate , 13fps . DOI: http://dx . doi . org/10 . 7554/eLife . 09206 . 005 Treatment with myosin II ATPase inhibitor blebbistatin at the onset of cell elongation prevented both cell elongation and the posterior relocation of the actin ring; instead , the actin ring remained at the anterior end of the cells ( Figure 1G , H ) . This dependence of actin ring migration on myosin II activity prompted us to ask if long-term maintenance of the ring at the equator is also dependent on myosin II activity . We thus analyzed the effect of blebbistatin on cells that had already elongated substantially , and whose circumferential actin ring had been positioned at the equator ( Figure 2A ) . At this stage , notochord cells in drug-treated embryos ceased to elongate ( Figure 2A , A′ ) . Surprisingly , the equatorial actin ring present at the start of the treatment ( white arrowheads in Figure 2A ) was lost; instead , we observed an anterior accumulation of actin ( yellow arrowheads in Figure 2A′ ) , similar to the anterior concentration of actin in coin-shaped cells . The effect of blebbistatin is reversible ( Figure 2B , B′ ) : notochord cells were able to elongate significantly following a 60-min wash , and the cortical actin ring returned from the anterior edge of the cells to the equator . 10 . 7554/eLife . 09206 . 006Figure 2 . Shifting of equatorial actin filaments upon blebbistatin treatment . Notochord cells are labeled with mCherry-UtrCH or mCherry-hActin for actin . ( A , A' ) The equatorial actin ring ( white arrowheads ) in early elongating cells ( A , 20 . 5 hpf ) is relocated to the anterior pole ( yellow arrowheads ) after 70 min blebbistatin treatment ( A' ) . ( B , B' ) The anterior relocation of the actin ring and inhibition of cell elongation after 60-min blebbistatin treatment ( B ) is reversed by a 60 min wash ( B' ) . ( C ) At 23 hpf , the elongated notochord cell has a broad equatorial actin ring ( white arrowhead ) that is associated with a prominent constriction ( red arrowhead ) . After 45-min blebbistatin treatment , the ring is shifted to the anterior pole ( yellow arrowhead ) , whereas the constriction is not . ( D ) Mean distance between the anterior lateral domain and the middle of the ring ( red ) , and the mean ring width ( black ) over time . While the ring shifts toward the anterior pole , indicated by the decrease of the anterior basal domain , its width stays relatively constant ( shaded in gray ) . After the ring reaches the anterior lateral domain , the width decreases . n = 5; error bars = SEM . ( E , E' ) Kymograph of the shifting actin ring based on Video 2 . Individual filaments ( arrows ) from the anterior and posterior edge move towards the center of the ring ( the equator , indicated by the dashed line ) , which itself is shifting anteriorly . After a certain time , the equatorial bound movement of filaments becomes parallel to the movement of the ring ( arrowheads ) . A diagram of the different movements is shown in E' . ( F ) Angles of single filament movement at specific times with respect to the center of the ring . Red ( positive values ) , movement from posterior towards the center; yellow ( negative values ) , movement from anterior towards the center . n = 2–7; error bars = SEM . ( G ) Fluorescence recovery after photobleaching in cells expressing mCherry-hActin . The entire actin ring region was bleached . Recovery is significantly slower in blebbistatin-treated cells . Control , n = 4; blebbistatin , n = 7; p = 0 . 012 . Solid lines indicate a single exponential fit for the control ( red curve ) with a turnover time of 90 ± 3 s , and a double exponential fit for the blebbistatin treatment ( green curve ) , with a fast fraction with the same turnover as the control , and a slow fraction ( f = 70% ± 4% ) with a turnover time of 7 . 8 ± 0 . 6 min . ( H ) Time-lapse frames of Video 3 showing the anterior movement of the equatorial ring ( 1 ) , its disappearance , and the emergence of a second ring ( 2 ) at the equator . Red arrowheads indicate the circumferential constriction . ( I ) The kymograph illustrates the close succession of the second ring to the first ring . The red dot indicates the time when the first ring disappears at the anterior lateral domain , and the second ring begins to appear . ( J ) Time-lapse frames from Video 4 showing the emergence of a second ( 2 ) and third ( 3 ) ring with long-term blebbistatin treatment . Anterior to the left in all panels . Scale bars , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 09206 . 00610 . 7554/eLife . 09206 . 007Figure 2—figure supplement 1 . Effect of blebbistatin treatment . ( A ) Cell length does not significantly change during ring shifting . N = 10; error bars = SEM . ( B ) Kymograph illustrating the directed movement of filaments toward the shifting ring equator ( arrows ) and the parallel filaments after prolonged incubation ( arrowheads ) . ( C ) Dislodgment of constriction and actin ring from each other and the cell equator . The green graph shows the distance between the middle of the first ring and the cell equator . Measurement stopped when the ring reached the anterior lateral domain . The constriction stayed equatorial ( black graph ) . Between 26 and 39 min , no constriction could be observed . At 40 min , a new constriction emerged posteriorly and shifted anteriorly . The second ring ( blue graph ) is not associated with the new constriction . DOI: http://dx . doi . org/10 . 7554/eLife . 09206 . 00710 . 7554/eLife . 09206 . 008Figure 2—figure supplement 2 . Talin localization at the equator is not affected by lower contractility . Notochord cells are labeled simultaneously with Lifeact-mEGFP for actin ( A , A' ) and mCherry-talinA I/LWEQ ( B , B' ) . Both actin and talin are enriched in the equatorial cortex before the blebbistatin treatment ( A , B ) . After 60-min blebbistatin treatment , actin ring is shifted to the anterior pole ( white arrowhead in A' ) , whereas talin remains at the equator ( white arrowheads in B' ) . Anterior to the left in all panels . Scale bars , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 09206 . 008 To characterize the anterior shift of the equatorial actin ring triggered by blebbistatin treatment , we recorded the actin dynamics during drug treatment in elongated cells possessing an equatorial constriction and a broad actin ring . High-speed time-lapse recordings revealed a migration of the cortical actin ring from the equator to the anterior pole ( Figure 2C , C′; Video 2 ) , mirroring exactly the reverse sequence of migration of the ring towards the center in normal development . The front of the ring moved at a velocity of 91 . 8 ± 10 . 5 nm/min ( n = 10 ) . The speed of the shift was blebbistatin dose dependent; halving the blebbistatin concentration reduced the velocity significantly to 27 ± 0 . 5 nm/min ( n = 10; p < 0 . 0001 ) . The ring width stayed constant when it shifted anteriorly ( Figure 2D ) . Remarkably , the encounter of the ring with the anterior lateral domain did not bring the movement to a halt . Instead , after the ring contacted the lateral domain ( red line in Figure 2D ) , the posterior edge of the ring continued to move anteriorly at an increasing speed of 191 . 6 ± 21 . 4 nm/min ( n = 10 ) , so that the width of the ring narrowed ( Figure 2D ) , until the entire ring disappeared at the position of the lateral domain ( Video 2 ) . The cell length did not significantly change during this process ( Figure 2—figure supplement 1A; n = 10 ) . We also examined the dynamic localization of talin , an actin-binding protein that bridges actin filaments and the adhesion apparatus at the cleavage furrow of dividing cells ( Sanger et al . , 1994; Critchley , 2009; Kanchanawong et al . , 2010 ) , and normally colocalized with the cortical actin ring at the equator in Ciona notochord ( Sehring et al . , 2014 ) ( Figure 2—figure supplement 2B ) . Live imaging showed that talin concentrated at the cell equator slightly after the ring had been established centrally ( data not shown ) , suggesting that talin actually responds to cortical repositioning rather than driving it . Upon blebbistatin treatment of already established central rings ( Figure 2—figure supplement 2A ) , whereas the cortical actin was shifted to the anterior pole ( Figure 2—figure supplement 2A′ ) , talin lagged behind at the equatorial position ( Figure 2—figure supplement 2B′ ) . This result suggests not all components of the ring are shifted by blebbistatin . 10 . 7554/eLife . 09206 . 009Video 2 . Shifting of equatorial actin filaments upon blebbistatin treatment . Notochord cells expressing mCherry-UtrCH were recorded every 15 s . Frame rate , 13fps . DOI: http://dx . doi . org/10 . 7554/eLife . 09206 . 009 The actin ring in elongating notochord cells is highly dynamic and consists of circumferential filaments that flow to the equator from both sides of the ring ( Sehring et al . , 2014 ) . The kymograph unveiled a persistence of this dynamics within the ring in blebbistatin-treated embryos: circumferentially oriented actin filament bundles ( arrows in Figure 2E; further examples are shown in Figure 2—figure supplement 1B ) from the anterior and posterior edges of the ring continued to flow towards the equator , which itself was shifting anteriorly ( dashed line in Figure 2E ) . In the first 15 min of blebbistatin treatment , the filament bundles moved at a velocity of 7 . 62 ± 0 . 78 nm/s ( n = 18 ) , which was significantly slower than the velocity in control cells ( DMSO-treated cells ) ( 29 . 92 nm/s , n = 10; p < 0 . 001 ) , where the filaments moved towards a stationary equator . Continued exposure to blebbistatin further reduced filament movement . Within the next 15 min , the velocity decreased significantly to 3 . 92 ± 0 . 9 nm/s ( n = 22; p = 0 . 004 ) . The deceleration of the filament bundles resulted in a flattening of their trajectories in the kymograph . Measurement of the angles of filament bundles revealed a reduction from 39° to 3° within 30 min of blebbistatin treatment ( Figure 2E′ , F ) . After 31 . 1 ± 1 . 36 min ( n = 20 ) , directed movement of filament bundles towards the center of the ring ( the moving equator ) ceased , and only filament bundles moving nearly parallel to the direction of the shifting ring could be observed ( yellow arrowheads , Figure 2E ) . After the ring had reached the anterior lateral domain , no pronounced filaments were detectable . There is no statistical difference in either velocities or angles between anterior- and posterior-directed filament bundles at any time . To analyze if there was still actin turnover within the ring at a time point when no prominent filament bundles were visible in a kymograph , or if it was a static ring shifting anteriorly , we performed fluorescence recovery after photobleaching ( FRAP ) experiments on notochord cells expressing mCherry-hActin . We bleached the entire equatorial ring in cells of control embryos , or the entire shifting ring in cells of embryos treated with blebbistatin for at least 30 min . For control embryos , the recovery curve was well fitted by a single exponential ( Figure 2G ) , yielding a characteristic turnover rate of 90 s ± 3 s . In the shifting ring with blebbistatin treatment , fluorescence also recovered , indicating a persistent cortical flow of actin elements from outside the ring towards its center . Interestingly , the dynamics could no longer be fitted by a single exponential , as it contained two characteristic recovery times: a fast one , very similar to the control , and a much slower one of 7 . 8 ± 0 . 6 min ( Figure 2G ) . A rough estimate of the flow-induced duration to close a bleached segment of 3 µm , assuming a mean velocity of 30 nm/s , is 100 s , which cannot be distinguished from the turnover time . However , after blebbistatin treatment , the mean velocity drops to roughly 6 nm/s ( see Figure 6E ) , leading to a duration of 8 min , which is very similar to the measured slower recovery time . This further indicates a requirement of myosin motor activity for a strong cortical flow of actin elements , while local polymerization and depolymerization is still present and represents 30% of the recovery and appears to be very slightly affected by the blebbistatin treatment . These observations together reveal a persistence of fast inner-ring dynamics , albeit a bit slower because of decreased actin velocity , superimposed on a slow global shifting of the ring towards the anterior edge upon blebbistatin treatment . Prolonged treatment with blebbistatin ( >1 hr ) led to the complete disappearance of the actin ring . Surprisingly , when the first ring was nearly gone , a new actin ring formed ( Figure 2H , Video 3 ) . A kymograph generated from the time-lapse reveals that the two rings overlapped in time only transiently: at the time point the first ring disappeared , the second ring emerged ( Figure 2I ) . The position , shape , and size of the second rings were often less precise and less sharp than the first rings ( Figure 2H , J ) . While the first ring disappeared after 106 . 67 ± 6 . 06 min of blebbistatin treatment ( n = 6 ) , the second ring had a significantly shorter dwelling time of only 61 . 33 ± 7 . 81 min ( p = 0 . 001 ) before it disappeared at the anterior edge of the cells . Intriguingly , we consistently observed the dislodging of the actin rings from the morphological constriction ( Figure 2H , arrowheads; Figure 2—figure supplement 1C ) , instead , the moving actin rings were associated with a morphological bulge at the basal domain ( yellow arrowhead in Figure 2C′ ) , and cells formed a new circumferential constriction ( red arrowheads in Figure 2C′ , H ) in the wake of the shifting ring . 10 . 7554/eLife . 09206 . 011Video 3 . Disappearance of the original actin ring and emergence of a new ring at the equator upon long-term blebbistatin treatment . Notochord cells expressing mCherry-UtrCH were recorded every 15 s . Frame rate , 13fps . DOI: http://dx . doi . org/10 . 7554/eLife . 09206 . 011 In cells that survived prolonged treatment of blebbistatin with relatively normal morphology , we observed the emergence of a third ring at the equator , after the disappearance of the second ring at the anterior lateral domain ( Figure 2J; Video 4 ) . In Video 4 , the second and third ring appeared at 151 and 238 min of treatment , respectively . 10 . 7554/eLife . 09206 . 012Video 4 . Emergence of a second and third actin ring at the equator upon long-term blebbistatin treatment . Notochord cells expressing mCherry-UtrCH were recorded every 3 min . Frame rate , 13fps . DOI: http://dx . doi . org/10 . 7554/eLife . 09206 . 012 As notochord cells are planarly polarized , and ring migration is always unidirectional in wild-type embryo , this prompted us to examine the effect of A/P polarity on actin ring migration . A conspicuous feature of the A/P polarity is the posterior localization of the nucleus , which is regulated by prickle , a core PCP component . The Ciona savignyi mutant aimless carries a deletion in prickle , resulting in a loss of A/P polarity manifested in the randomized localization of nucleus in addition to an earlier convergent extension defect ( Jiang et al . , 2005 ) . We first explored the possibility that the posterior nucleus might influence the direction of ring movement mobilized by blebbistatin . To this end , we examined the notochord of the ascidian Halocynthia roretzi , which follows a remarkably similar early developmental process as in Ciona ( Figure 3A , B ) , except at the cell elongation stage , the nuclei are positioned in the center of the cells ( asterisks in Figure 3B ) . A conspicuous circumferential constriction is present at the equator , which is colocalized with a cortical ring of actin ( Figure 3C ) and activated myosin ( Figure 3D ) , whose activity is essential for notochord cell elongation ( Figure 3E , F ) . We next treated embryos with elongating notochord cells with blebbistatin for 3 hr . Similar to what was observed in Ciona , the actin ring was shifted invariably anteriorly ( Figure 3G ) , indicating that the position of the nucleus does not influence the dynamic behavior of the ring . We thus used aimless embryos to explore the role of a compromised A/P polarity on the repositioning of the ring and its shifting upon blebbistatin treatment , independent from its influence on nuclear position . In wild-type C . savignyi embryos , elongated notochord cells possess a circumferential actin ring at the cell equator and a posterior nucleus ( Figure 4A ) . Blebbistatin treatment shifts the ring towards the anterior pole ( Figure 4B ) , mirroring exactly the events in C . intestinalis . In aimless embryos , the intercalation of notochord cells ( outlined in Figure 4C ) is impaired , except in the posterior region where the cells often align into a single file ( Jiang et al . , 2005 ) . In these cells , the loss of A/P polarity is evident by the random position of the nuclei ( Figure 4C insert ) . In 92% of mock-treated aimless cells in this region , the actin ring is positioned at the equator ( Figure 4C insert ) , showing that equatorial ring positioning is independent from PCP . However , reducing contractility in polarity-deficient mutant cells led to a dramatically different phenotype: after 60-min blebbistatin treatment , no unidirectional migration towards the anterior was observed . The direction of the shift was randomized and independent of the localization of the nucleus: examples for anterior nucleus and ring , anterior nucleus and posterior ring , posterior nucleus and ring , and posterior nucleus and anterior ring could be found ( Figure 4D , D' ) . In 56% of the cells , the ring was still positioned at the equator , and in the rest , the ring migrated either to the anterior or posterior side ( 18% and 26% ) ( n = 57 ) ( Figure 4E ) . These results imply that A/P polarity by prickle is not necessary for the establishment of an equatorial actin ring; however , it is instrumental for the direction of movement of the ring upon blebbistatin treatment . 10 . 7554/eLife . 09206 . 013Figure 3 . Circumferential actin rings are shifted anteriorly in Halocynthia roretzi notochord cells with centrally localized nuclei . ( A , B ) Halocynthia notochord cells elongate from coin-shaped ( A ) to drum-shaped ( B ) . A circumferential constriction appears at the equator of the cylindrical cell ( arrowhead ) . The nucleus ( asterisk ) is localized in the center of each cell . ( C , D ) Cortical F-actin ( arrow in C ) and MRLC ( arrow in D ) accumulate at the equatorial region of the basal domain . ( E , F ) Notochord elongation ( E , DMSO-treated ) is abolished in Halocynthia embryos treated with 100 μM blebbistatin for 16 hr ( F ) starting at the onset of cell elongation ( 27 hpf ) . ( G ) Circumferential actin rings ( white arrowheads ) are shifted anteriorly after 3-hr blebbistatin treatment at 31 hpf ( 13°C ) . Similar to what is observed in Ciona notochord cells ( Figure 2C' , H ) , the shifted ring is associated with a circumferential bulge ( white arrowheads ) , whereas the constriction is located posterior to the ring ( red arrowheads ) . These results indicate ( 1 ) a conservation of the equatorial actomyosin contractile mechanism to drive notochord cell elongation in Halocynthia , ( 2 ) that the position of the ring does not influence the position of the nucleus , ( 3 ) the position of the nucleus does not influence the position of the ring , and ( 4 ) the nucleus position does not affect the direction of the ring shift . Anterior to the left in panels A–B , G . Scale bars in E and F , 50 μm; in all others , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 09206 . 01310 . 7554/eLife . 09206 . 014Figure 4 . Anterior shifting of the actin ring is disrupted in the prickle mutant aimless . ( A–D′ ) Ciona savignyi embryos are stained with phallacidin for actin and DAPI for nuclei . The actin ring ( white arrowheads ) is positioned at the equator in control notochord cells ( A ) and is shifted anteriorly by 60 min blebbistatin treatment ( B ) , whereas the posterior localization of nucleus ( n ) is not affected by blebbistatin . Red arrowhead indicates the constriction . ( C ) Notochord in an aimless embryo ( outlined by dashed line ) with impaired cell intercalation in the anterior region , and fully intercalated and significantly elongated cells in the posterior region . The actin ring in these cells is localized at the equator , but the nucleus is placed in a random position ( insert ) . ( D , D' ) 60-min blebbistatin treatment mislocalizes the actin ring in aimless embryos . ( E ) Distribution of actin rings in mock-treated ( black; n = 53 ) and blebbistatin-treated ( gray; n = 57 ) aimless cells . Anterior to the left in all panels . Scale bars , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 09206 . 014 In order to probe quantitatively these findings , we follow the theory of active gels ( Kruse et al . , 2005; Prost et al . , 2015 ) to develop a very simple biophysical model of the actomyosin cortex as a viscous contractile gel , undergoing steady turnover ( see Appendix 1: Physical modeling of the Ciona cortical flows for details ) . Such models generically predict spontaneous accumulations of actomyosin . Indeed , if the actomyosin concentration is slightly higher in a given region , then the contractile stress is also locally higher compared to the surroundings . Because of this initially small imbalance , surrounding actomyosin fibers flow towards the accumulation , making it even denser and even more contractile ( Recho et al . , 2013 ) . This self-reinforcing loop , which concentrates actomyosin in a single spot at the cortex with filaments flowing towards the ring , is resisted by their depolymerization and effective diffusion ( Figure 5A ) , which favor a homogenous cortex . 10 . 7554/eLife . 09206 . 015Figure 5 . Verifications of the model assumptions and fitting of parameters . ( A ) Sketch of our model . Contractility destabilizes an initially homogenous cortex into a central ring , whereas PCP-driven preferential anterior polymerization localizes the ring on the edge . ( B ) Measurement of the angle between lateral and basal membrane during the elongation ( 2 . 5 increase ) and ring migration ( 1 . 5 increase ) process , which indicates their relative tensions ( n > 15 for each time point ) . Basal tension increases with time . σ¯ is the basal contractility , γ is the lateral contractility and θ is the angle between lateral and basal membranes . We have the geometric relation σ¯ cos⁡θ=γ . ( C ) PIV analysis of cortical flows in late stage embryos . ( D ) Linear negative correlation between local actin intensity and velocity gradients , as extracted from PIV . Actin intensity and velocities have been rescaled in the dimensionless units described in the main text and in Appendix 2: rescaling . ( E ) Comparison between intensity and velocity profiles and our theoretical predictions ( data extracted from C ) . The velocity field is rescaled by the average velocity . DOI: http://dx . doi . org/10 . 7554/eLife . 09206 . 015 The cell cortical layer of actomyosin is modeled as a thin axisymmetric layer of length L with principal axis z = −L/2 ( anterior side ) …L/2 ( posterior side ) . Because the thickness of the cortex is two orders of magnitude below cell size , we can use a thin film lubrication approximation and do not resolve the cortex radial direction . Therefore , all equations are invariant in the radial and orthoradial direction and the model is one dimensional . The simplified chemo-mechanical problem then consists of three equations equipped with appropriate boundary conditions: • Conservation of the actomyosin density ρ through a classical reaction-drift-diffusion process reads: ( 1 ) ∂tρ+∂z ( ρν ) −D∂zzρ=ρ0−ρτ , where ν denotes the actomyosin velocity . We have denoted τ the turnover time , D an effective diffusion coefficient of actomyosin accounting for non local turnover and ρ0 the target density . Boundary conditions prescribe the actin fluxes J=ρν−D∂zρ at z = −L/2 and L/2 . Force balance neglecting friction with the extracellular matrix reads ( 2 ) ∂zσ=0 , where σ is the mechanical stress in the actomyosin meshwork . The neighboring cells impose a given residual mechanical stress σ¯ on the cell boundaries . In the absence of friction , as indicated by Equation ( 2 ) , the mechanical stress is homogeneous and σ ( z ) =σ¯ . • At last , we prescribe the constitutive behavior of the gel as ( 3 ) σ=η∂zν+χρ . We have denoted χ the contractility arising from myosin motors , and η the viscosity of the gel . We set σ¯=χρ0 , such that the value of the residual stress at low contractility , in the absence of boundary fluxes ( J ( ±L2 ) =0 ) , ensures a homogenous distribution of actin as observed experimentally before elongation ( 16 . 5 hpf , Figure 1A ) . Indeed , ρ=ρ0 , ν=0 , and σ=σ¯ is then a trivial solution of Equations ( 1–3 ) . For the problem to be fully specified , we still need to impose the values of the boundary fluxes of actin J ( ±L2 ) . To begin , in order to expose the role of contractility only , we first assumed that they vanish . It should be noted that the actomyosin flux J encompasses both actin filament velocity , and an effective diffusive flux arising from actin polymerization ( see Appendix 1: the model ) . Therefore , a vanishing total flux does not necessarily entail a vanishing velocity . Then , a linear stability analysis of Equations ( 1–3 ) predicts a threshold of actomyosin contractility ( χρ0 ) c=η ( 1τ+4π2DL2 ) above which , the homogeneous cortex loses stability and a mechanically stable central ring forms even in the absence of external signaling cues ( see Appendix 3: steady states , and Figure 1 in Appendix 1: the model for the stability diagram and details on the boundary conditions ) . We can then interpret that the driving force positioning the ring at the equator is the contractility increase during the process of ring migration . Before finalizing the model , we check two of its key assumptions that ( 1 ) contractility increases smoothly as ring migration proceeds , and ( 2 ) the velocity gradient of actin filaments towards the center should depend linearly on the local contractility , indicated by local actomyosin concentration ( as seen in Equation 3 ) . To address the first assumption , we measure the angle between the lateral and basal side , from embryos at various stages ( n > 20 angles for each embryo ) ( Figure 5B ) . As shown recently ( Maître et al . , 2012 ) , this angle θ=arccos ( σ¯γ ) reflects a force balance between the tensions of the basal ( σ¯ ) and lateral ( γ ) surfaces , and therefore can be used as a proxy for tension changes . Interestingly , the angle decreased smoothly and continuously during elongation and ring migration , indicative of an increased basal tension relative to lateral tension ( Figure 5B ) , by roughly a factor of 2 . 5 during the elongation process , and 1 . 5 during the ring migration process . We assume at first order approximation that lateral tension is constant and set in the model that ( σ¯ ) increases by a factor 1 . 5 . As we shall show later , such an increase enables a good fit to all of the available data ( see Appendix 4: rough estimates of model parameters , and Appendix 5: model predictions for further details ) . To test the second assumption , we performed PIV analysis on high frequency movies of actin flows ( Figure 5C ) , to measure local velocity gradients . When plotting these as a function of local actin concentration , we found a robust negative linear correlation ( Figure 5D ) , which shows that flows are driven by differences in actomyosin concentration , validating quantitatively Equation 3 of our model . From the slope of the correlation , as well as from the characteristic of actin bundles measured above in the kymographs , we could extract the ratio ( χ/η ) of contractility and viscosity . Finally , we fixed the last parameters of our model through FRAP experiments ( τ = 90 s ) and through the width of the ring intensity profile ( D = 2 . 10−12 m2/s ) ( see Appendix 4: rough estimates of model parameters for the details of parameter fitting ) . Under these assumptions and with these parameters , a central ring spontaneously forms at the center of the cell when contractility increases above the critical threshold ( χρ0 ) c , and contractility is self-sufficient to maintain the ring structure through actomyosin flows . However , the experimental data show the existence of a stage where the ring is positioned at the anterior side and also underlines the importance of PCP in the repositioning of actin rings . Our final model therefore incorporates polarity in the model in the simplest way possible: by assuming that it creates a small preferential polymerization of actin at the anterior side , that is , there is a non-zero flux at the anterior side , different from the flux at the posterior side . We considered that the flux on the posterior side was still zero , that is , J ( −L/2 ) =F and J ( L/2 ) =0 . We showed in Figure 7 in Appendix 5: model prediction that lifting this constraint does not qualitatively change our results , as the key parameter is the difference between anterior and posterior flux , but not their respective magnitude and our system is locally robust with respect to this type of perturbation . Assuming the existence of a preferential polymerization at one boundary due to PCP was supported by the well-studied link between PCP and actin polymerization ( Wallingford and Habas , 2005 ) . In particular , Disheveled ( Dsh ) , one of the core member of the PCP pathway has been shown to activate key actin regulators such as Rho and Rac ( Tahinci and Symes , 2003; Wallingford and Habas , 2005 ) , as well as Daam1 , a member of the formin protein family ( Kida et al . , 2007; Gao and Chen , 2010 ) . It should be noted that as we are treating the actomyosin gel as a single species ( with the assumption that bipolar filaments performing the contractile power stroke co-localize with actin ) , assuming that PCP localizes myosin anteriorly , as reported in Newman-Smith et al . , 2015 , would yield the same qualitative results . With these final boundary conditions , the dynamical system Equations ( 1–3 ) predicts a transition ( which is now smooth , see Figure 1 in Appendix 1: the model ) between two mechanically stable states of the actin ring: a central position if the contractility χ is large enough , and an anterior position when χ is small enough and the polarity-induced actin flux F dominates . We set the value of F using the filament velocity order of magnitude as well as the experimental actin density profiles when the contractility is impaired ( blebbistatin experiments , see Appendix 5: model predictions ) . To verify the model , we then numerically integrated Equations ( 1–3 ) in order to calculate the steady state of a central actin ring , using the parameters deduced above . We found that our model can reproduce quantitatively very well both the velocity and intensity profiles in the entire actin cortex at the late stages of elongation ( Figure 5E ) . Interestingly , we predicted , and observed in the PIV , a rather large , non-zero value for the actin filament velocity at the posterior edge of the cell , which is compensated in the model by actin polymerization on the side , since the total flux is zero . Moreover , we note a remaining , albeit small bias in ring position towards the anterior side , as in the data , showing that an anterior-imbalance remains at the later stages , both in simulations and in the data . Next , we wished to compare the dynamics of ring migration predicted by the model to the experimental data , both in the case of normal ring migration towards the center , and blebbistatin-induced ring migration towards the anterior side . We simulated the effect of the linear increase in contractility , deduced from Figure 5B , and followed the appearance and migration towards the center of a ring . After 110 min , we then simulated a blebbistatin treatment as an exponential decrease in contractility , following previous experimental results on traction force decrease with blebbistatin ( Lam et al . , 2012; details in Appendix 5: model predictions ) , which predicted a rapid migration back towards the anterior pole . Figure 6A displays a kymograph of the numerical integration . Next , we compared model predictions and experimental data ( average over 7 cells for normal ring migration , and 5 cells for blebbistatin treatment ) . It should be noted that in our model , the migration speed of the ring is controlled by the dynamics of contractility changes . This is in agreement with our experimental observation that the velocity is dependent on the dose of blebbistatin used , but also with the fact that the velocity of the ring migration is an order of magnitude smaller than the velocity of the actin filament bundles . 10 . 7554/eLife . 09206 . 010Figure 6 . Comparison between theory and experiments on the dynamics of actin rings during migration and blebbistatin treatment . All theoretical curves are extracted from the same parameter set ( see Appendix 5: model predictions for details on non dimensional values and parameters ) . ( A ) Kymograph of actin intensity during central ring migration ( left part ) and during blebbistatin treatment starting at 110 min . The color code indicates local actin intensity . ( B ) Comparison of the model and experimental anterior and posterior lateral domains during normal development ( data taken from the average of 5 cells ) and blebbistatin treatment at 110 min ( data taken from the average of 7 cells ) . The y-axis indicates the position of the anterior and posterior border ( defined as 50% of the ring maximal intensity ) . ( C ) Actin ring width vs with cell length , throughout cell elongation . The thick line is our theoretical prediction . The black dots are the measured data ( n = 7; error bars = SEM ) . ( D ) Theory-to-experiment comparison of actin intensity profiles during central ring migration: 0 min ( red ) , 30 min ( green ) , 60 min ( blue ) , 80 min ( purple ) . ( E ) Theory-to-experiment comparison of filament velocity following blebbistatin treatment . ( F ) Theory-to-experiment comparison of actin intensity profiles after blebbistatin treatment: 0 min ( red ) , 15 min ( green ) , 30 min ( blue ) , 45 min ( purple ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09206 . 01010 . 7554/eLife . 09206 . 016Figure 7 . PCP participates in force balance to reposition actin rings . ( A ) Left: effect of a slow , 1 . 5-fold linear increase in contractility , for polarity-deficient mutants ( no preferential flux on the edges ) . The ring forms directly at the center . Right: ring positioning for random uncoordinated polarity ( preferential flux on the anterior , full red line , preferential flux on the posterior , dashed red line , and equal flux on anterior and posterior , full blue line ) . ( B , C ) Localization of Flag-Dsh ( B ) , and Myc-Pk ( C ) in notochord cells . At early stages ( 19 and 20 . 8 hpf ) , Flag-Dsh localizes at the basal surface . At 20 . 8 hpf , it concentrates at the equator ( white arrow ) . Subsequently , it shifts to both lateral surfaces , with a preference for anterior side of cells ( yellow and blue arrows in B ) . Myc-Pk localizes at the anterior lateral surface of the cell at early stages and gradually concentrates to the center of anterior lateral surface ( white arrow in C ) . ( D ) Myosin contractility antagonizes PCP to position a dynamic actin cytoskeleton . Anterior to the left . Scale bars , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 09206 . 016 Moreover , a simple order of magnitude calculation shows that this decrease in the velocity of actomyosin explains the FRAP recovery curves ( Figure 2G ) . Indeed , in control cells , a rough estimate of the time needed to close a bleached segment of 3 µm through flows , given a mean velocity of 30 nm/s , is 100 s , which cannot be distinguished from the turnover time . Therefore , we only see a single-exponential recovery . However , after blebbistatin treatment the mean velocity drops to roughly 6 nm/s ( see Figure 6E ) , leading to a time of 8 min for flow-induced recovery , which is strikingly similar to the time extracted from the FRAP analysis of 7 . 8 min . There are then two time scales for recovery , one linked to turnover , unaffected by blebbistatin , and one linked to flows , which is dramatically compromised by blebbistatin . We first examined the dynamics of normal ring migration to the center and verified that our dynamical prediction for the positions of the anterior and posterior sides of the ring matched very well the experimental data ( Figure 6B ) . We also compared theoretical and experimental profiles of actin intensity during ring migration towards the center ( Figure 6C ) , and observed again a good agreement between the two: first an exponential ring profile at the anterior pole , then detachment of a ring of broad thickness during migration , and then refinement of a thinner central actin ring . Moreover , a rather intriguing feature of our model is that it predicts the size of the actin ring should first increase linearly with the length of the cell , before showing a plateau region for large cell sizes ( Figure 6C ) . This is because ring formation is a spontaneous , self-organizing phenomenon that depends on boundary conditions , and therefore on the size of the cell . We measured experimentally actin ring width during cell elongation , and found it indeed increases linearly with the length of the cell , in very good quantitative agreement with our model , with all parameters having already been fixed above . In cytokinesis , it was proposed that such a scaling could allow cytokinetic time to be independent of cell size as observed in C . elegans embryos ( Turlier et al . , 2014 ) . Our model provides a natural explanation to how this might be implemented in a simple manner in vivo . Next , we turned to the blebbistatin treatment . The only free parameter is the timescale of contractility decrease ( over a time of 15 min ) , which we fitted through ring position shifting , keeping all other parameters constant . This yielded a good theory-experiment agreement for the shifts of both the anterior and posterior sides of the ring ( Figure 6B ) . Having fixed the one parameter for the blebbistatin treatment , we sought to test our model further via second independent measurement . We plotted the predicted decay of actin velocity from our simulations and compared it to the experimental data measured above . Again , we found an excellent quantitative agreement between predictions and data ( Figure 6E ) . Finally , we examined the spatial profiles of actin intensity during blebbistatin treatment and migration towards the anterior edge . This recapitulated in reverse the normal migration sequence and showed good qualitative agreement between theory and experiments . To verify the role of PCP in the model , we performed the simulation under the conditions of no polarity and random polarity ( Figure 7A ) : for low polarity and increased contractility ( aimless mutant ) , the ring formed directly in the center of the cell; for low contractility and low , uncoordinated polarity between neighboring cells ( aimless mutant with blebbistatin ) , the actin ring did not have any strong forces positioning it , and was therefore positioned randomly , depending on the strength of the respective anterior and posterior boundary fluxes ( see Appendix 5: model predictions for further details ) . We next turned back to our data to find experimental clues for this preferential polymerization of actin at the anterior side . As noted earlier , Dsh is a core member of the PCP pathway ( Keys et al . , 2002 ) , and we found that it initially accumulated at the basal membrane , but was relocated to the lateral domains , with a preference for the anterior side ( 28% ± 9% enrichment compared to the posterior side , n = 11 cells ) ( Figure 7B ) , mirroring the localization of Prickle protein at this stage ( Figure 7C ) ( Jiang , et al . , 2005; Newman-Smith , et al . , 2015 ) . Interestingly , this relocation occurred roughly at the same time as the appearance of the anterior ring ( between 19 and 20 hpf ) . This is fully in line with our prediction of a PCP-driven preferential polymerization at the anterior side , which could be mediated by Dsh . Ascidians have the simplest notochord in the Chordate phylum . It consists of merely 40 cells arranged in a single file at a transient stage of development . Yet , emerging evidence suggests that the Ciona notochord possesses an intriguing array of complexity , including differential gene expression , asymmetrical cell lineages , desynchronized morphogenetic behavior ( Reeves et al . , 2014 ) , and an A/P polarity which manifests in the posterior localization of nuclei in the first 39 cells ( Jiang et al . , 2005 ) . The current study reveals hitherto unknown aspects of A/P polarity in dynamical properties of actomyosin cortex . An actin ring forms at the anterior side of cells as core PCP proteins such as Disheveled relocate there , and then migrates to the equator as contractility increases , where it contributes to cell elongation . Interestingly , planar polarized contractile flows were reported during Drosophila germ-band extension ( Rauzi et al . , 2010 ) . The ring movements that we observe are very robust , with surprisingly little fluctuation , given the highly dynamic nature of the ring . Here , we could disentangle the contributions of contractility and PCP to this robust positioning , by a combination of drug and genetic experiments . We show that decreasing contractility via blebbistatin reverses the normal developmental sequence , causing an anterior shift of the equatorial ring ( Figure 2 ) , which can be observed consistently throughout the elongation stage , but not in polarity mutants . This suggests a tug-of-war between contractility and cellular polarity: sufficiently large myosin contractility antagonizes the Prickle/Dishevelled-mediated anterior bias and can self-organize the actin ring in the center of the cell purely by physical forces , as shown by our theoretical analysis ( Figure 7A ) . Indeed , the equatorial ring in the notochord cells , despite its resemblance to the equatorial ring in cytokinesis , is established in a spindle-independent manner . A similar phenomenon has been reported in the asymmetrical cell division of Drosophila neuroblasts ( Cabernard et al . , 2010 ) , which also uses cortical polarity signals to regulate the furrow . During longer blebbistatin treatments , we observe complex oscillations of the cortex , with a second ring usually emerging at the equator . This suggests the existence of additional signals for ring formation and from a theoretical perspective , underlines the need to go beyond our simple one-component model , for instance , in order to take into account the dynamics of adhesion complexes as well . Indeed , our model is the simplest one can write for active gels , and it assumes for instance that all the rheological coefficients are constant , while in principle some , for example , viscosity , could depend on the state of the gel , and on blebbistatin concentration in the drug treatment ( Stirbat et al . , 2013 ) . Nevertheless , the fact that we can capture well the experimental data with such a simple model suggests that such coupling effects are probably of secondary importance for the observed phenomenon . It should be noted that , in this study , we have considered polarity as a constant , externally fixed parameter and examined its effect on actomyosin localization . In fact , a recent study in the same system ( Newman-Smith et al . , 2015 ) suggested an active feedback of myosin in core PCP protein localization . The next logical step would therefore be to incorporate this feedback loop to our formalism . Earlier reports have noted movement of constrictions in isolated amphibian cells , especially the neural plate cells , which form local circumferential constriction rings traveling the length of the cells in successive waves ( Holtfreter , 1946 ) . Traveling constrictions have also been observed in leukocytes during cell shape changes and migration ( Senda et al . , 1975; Haston and Shields , 1984; Shields and Haston , 1985; Bornens et al . , 1989 ) . Taken together , our data and theoretical analysis suggest a novel and generic framework through which PCP and contractility can interact to modify the forces acting on actomyosin rings and therefore their positioning ( Figure 7D ) . Adult C . intestinalis were obtained from Station Biologique de Roscoff , France and maintained in running filtered seawater . For fertilization , gametes from several individuals were surgically removed and mixed . Fertilized eggs were dechorionated with 1% sodium thioglycolate and 0 . 05% protease E as described by Mita-Miyazawa et al . ( 1985 ) , followed by five washes with UV-treated seawater . Embryos were cultured at 13°C . Adult C . savignyi were collected at the Santa Barbara harbor ( Santa Barbara , CA ) and maintained in running seawater . Fertilization and dechorionation were performed as described above for C . intestinalis . For the aimless mutant , spawning was controlled by light conditions . Embryos were cultured at 15°C . Adult H . roretzi were collected near the Asamushi Research Center for Marine Biology ( Aomori , Japan ) and the Otsuchi International Coastal Research Center ( Otsuchi , Japan ) , and kept in tanks . Spawning was controlled by temperature and light conditions . Spawned eggs were fertilized with a suspension of non-self sperm . Embryos were cultured in Millipore-filtered seawater containing 50 μg/ml of streptomycin and kanamycin at 11°C . Expression constructs in this study have been described previously: mCherry-UtrCH , lifeact-mEGFP , mCherry-MRLC and mCherry-hActin ( Dong et al . , 2011 ) ; mCherry-tropomyosin ( Sehring et al . , 2014 ) ; Flag-Dsh and Myc-Pk ( Jiang et al . , 2005 ) . Electroporation was modified after previously published protocol ( Corbo et al . , 1997 ) . Plasmid DNA ( 80 μg in 80 μl ) was mixed with 400 μl 0 . 95 M mannitol in 4-mm cuvettes . 320 μl dechorionated fertilized eggs were added and electroporated with a Gene Pulser Xcell System ( BIO-RAD ) , using a square pulse protocol ( 50 V and 15 ms per pulse ) . After electroporation , embryos were cultured at 13°C . Blebbistatin dissolved in DMSO ( Calbiochem , 203389 ) was used at a final concentration of 100 μM . Control embryos were treated with DMSO . For recovery experiments , embryos were washed 10× after blebbistatin treatment . All blebbistatin experiments were repeated at least three times . C . savignyi and Halocynthia embryos were fixed with 4% formaldehyde in seawater for 1 hr at room temperature ( RT ) , washed 3 times with PBS , and stained with 5 units/ml BODIPY-FL phallacidin ( Invitrogen , B607 , Carlsbad , CA ) in PBS containing 0 . 2% Triton X-100 for 2 hr at RT . After 3 washes for 10 min each with PBS , C . savignyi embryos were counterstained with DAPI , and transitioned through an isopropanol series with 30 s steps: 70% , 85% , 95% , and two times 100% isopropanol . For myosin staining in Halocynthia , fixed embryos were blocked with 0 . 1% BSA in PBT overnight at 4°C , followed by incubation with Rabbit anti Ser19 myosin antibodies ( Cell Signaling Technology , 3671 ) ( 1:50 ) overnight at RT . After 2 × 40 min washes with PBT , embryos were incubated with Alexa594 anti-rabbit secondary antibodies ( Invitrogen , A11011 ) overnight at RT , washed 3 times in PBT and counterstained with 5 units/ml BODIPY-FL phallacidin in PBS . Localization of Dsh and Pk in C . savignyi followed previously published procedure ( Jiang et al . , 2005 ) . C . intestinalis embryos were observed under a Leica TCS SP5 confocal laser-scanning microscope ( CLSM ) equipped with a 40X oil-immersion objective ( NA 1 . 2 ) . If necessary , embryos were sedated using 0 . 2% MS222 ( Sigma , A5040 ) . Halocynthia embryos were analyzed with a BX61 CLSM ( Olympus ) . C . savignyi embryos were imaged on an Olympus Fluoview 1000 CLSM using a 40 × 1 . 3 NA objective . Images were processed and analyzed with ImageJ . All fluorescent images shown are maximum projections . The ring width was measured on projections by drawing a line at the right angle to the lateral domains and across the ring . Kymographs were compiled using the kymograph plug in in ImageJ . Statistical parameters were determined using SigmaPlot software ( Systat Software Inc . ) . Significance of differences was calculated using Student's t tests . FRAP experiments were performed with the Leica TCS SP5 . The whole cortical actin ring of notochord cells was bleached by using maximum laser power at 561 nm for an empirically determined number of iterations to achieve bleaching throughout the full thickness of the cortical signals . After bleaching , images were taken at regular intervals ( between 4 . 5 and 9 s ) with the same laser at 30% laser intensity . The halftime of recovery was calculated by measuring the signal intensity in the region of interest ( ROI ) over time . The raw data were corrected for background noise and image acquisition bleaching . The intensity was plotted as a function of time and the half-time of recovery , t1/2 , was extracted .

Animal cells can move , and cell movements are particularly important during the early stages of development , when the developing embryo rapidly changes shape . These movements depend on a network of fibers made up of a protein called actin . Just like an animal's skeleton , this network provides an internal scaffold for the cell and supports the cell's movements . Another protein called myosin works closely with actin and acts as a motor that drives these movements . To study how cellular movements contribute to development , scientists often turn to simple , tube-like sea animals called sea squirts . These strange-looking creatures are distant relatives of humans and other animals with a spinal cord . All of these related creatures develop a long , rod-shaped structure called the notochord during the earliest stages of their development that is critical for forming the nervous system . Studying the development of the notochord is easier in sea squirts than other animals because the sea squirt's notochord is made up of just 40 cells arranged in a single file . Now , Sehring et al . provide new details about the forces that shape the notochord cells in sea squirts . The experiments used microscopes and fluorescent markers to see what happens as the cells elongate to form the rod-like notochord , which stretches from the front to the back of the animal . This revealed that a ring made of actin and myosin forms near the front end of each cell and then migrates to the middle of the cell , stretching it along the way . Sehring et al . then treated the cells with a drug that blocks myosin's motor-like ability . In these cells , the actin–myosin ring remains stuck in the front of the cell . Treating the cells at a later stage , that is , when the rings had already arrived at the center , led to the central rings moving back to the front end of the cell . Furthermore , when a mutant sea squirt that had cells without a distinct front or back was treated with the myosin-blocking drug , the ring ended up at random places in the cells . Together , these results suggest that the placement of the actin–myosin ring is determined by a tug-of-war between the pull of the front end of the cell and the myosin-driven force of the ring itself . These findings reveal a general rule that can determine the position of cell's inner skeleton . Future studies will ask how the front end of the cell attracts and pulls the ring , and what this means for tissue growth and organ formation .

2015

Assembly and positioning of actomyosin rings by contractility and planar cell polarity

Voltage-gated potassium ( Kv ) channels enable potassium efflux and membrane repolarization in excitable tissues . Many Kv channels undergo a progressive loss of ion conductance in the presence of a prolonged voltage stimulus , termed slow inactivation , but the atomic determinants that regulate the kinetics of this process remain obscure . Using a combination of synthetic amino acid analogs and concatenated channel subunits we establish two H-bonds near the extracellular surface of the channel that endow Kv channels with a mechanism to time the entry into slow inactivation: an intra-subunit H-bond between Asp447 and Trp434 and an inter-subunit H-bond connecting Tyr445 to Thr439 . Breaking of either interaction triggers slow inactivation by means of a local disruption in the selectivity filter , while severing the Tyr445–Thr439 H-bond is likely to communicate this conformational change to the adjacent subunit ( s ) . Enzymes and catalytic proteins have evolved to balance the thermodynamic challenges of stability and substrate throughput ( Shoichet et al . , 1995 ) . Ion channels , for instance , must efficiently interconvert between open , closed and inactivated states to regulate ionic flux across biological membranes . Even small alterations in their function that change the rates of isomerization between states can underlie inherited or acquired diseases ( Hille , 2001 ) . Voltage-gated potassium ( Kv ) channels are tetrameric membrane proteins with a central ion-conducting pore domain , surrounded by four voltage-sensor domains ( VSDs ) , which tightly regulate the conductive state of the pore domain ( Figure 1A ) . The pore domain consists of two transmembrane helices ( S5–S6 ) connected by a re-entrant pore helix , which forms the selectivity filter . Kv channels negatively regulate conductance after channel opening through a process termed inactivation . The rate and extent of inactivation exhibit considerable isoform-dependent differences , which are reflected in the physiological contributions of these channels to cellular excitability in neuronal and cardiac tissues ( Bean , 2007; Smith et al . , 1996; Spector et al . , 1996; Sanguinetti and Tristani-Firouzi , 2006; Aldrich et al . , 1979 ) . Inactivation can be described by two kinetically and mechanistically distinct processes termed fast ( or ‘N-type’ ) inactivation and slow ( or ‘C-type’ ) inactivation ( Hoshi et al . , 1991; Kurata and Fedida , 2006; Hoshi and Armstrong , 2013 ) . While the former results from a channel peptide docking within the open cytoplasmic entrance to the permeation pathway ( Hoshi et al . , 1990 , 1991; Zhou et al . , 2001 ) , the latter is assumed to involve highly cooperative local conformational changes near the selectivity filter , a notion supported by electrophysiological , structural and computational approaches ( Lopez-Barneo et al . , 1993; Yellen et al . , 1994; Liu et al . , 1996 , 1997; Starkus et al . , 1997; Kiss et al . , 1999; Cordero-Morales et al . , 2006 , 2007; Peng et al . , 2007; Cuello et al . , 2010a , b; Cordero-Morales et al . , 2011; Ostmeyer et al . , 2013; van der Cruijsen et al . , 2013 ) . However , and despite this available data , fundamental questions persist regarding the precise molecular determinants that mediate the rate of slow inactivation , the basis for cooperativity and the relationship between slow inactivation and the structural integrity of the selectivity filter ( Hoshi and Armstrong , 2013 ) . An ongoing challenge is to elucidate how these proteins precisely time the entry into the slow inactivated state in the presence of a sustained voltage stimulus . 10 . 7554/eLife . 01289 . 003Figure 1 . The aromatic cuff is part of a highly conserved region in potassium channels . ( A ) Top view of a Kv channel based on the structure of the tetrameric Kv1 . 2/2 . 1 chimera ( PDB 2R9R; individual subunits are colored in gray , cyan , green and yellow , respectively ) . The inset shows a magnified view of the side chains that form the ‘aromatic cuff’: Trp434 , Trp435 and Tyr445 ( by numbering in Shaker potassium channels ) . Note the backbone carbonyls are shown for Tyr445 to highlight their role in the coordination of potassium ions ( gray circle ) ; ( B ) Sequence alignment of the pore helix and the selectivity filter of various potassium channels: Shaker ( GI:288442 ) , Kv1 . 1 ( GI:119395748 ) , Kv2 . 1 ( GI:84570020 ) , Kv3 . 1 ( GI:298603 ) , Kv4 . 1 ( GI:8272404 ) , Kv5 . 1 ( GI:24418476 ) , Kv6 . 1 ( GI:24418479 ) , Kv7 . 1 ( GI:6166005 ) , Kv8 . 1 ( GI:7657289 ) , Kv9 . 1 ( GI:219520418 ) , and KCSA ( GI: 61226909 ) . Side chains constituting the aromatic cuff are highlighted in gray ( see above ) and all positions studied here are indicated using their numbering in Shaker potassium channels ( these residues correspond to Trp362 , Trp363 , Ser367 , Thr369 , Thr370 , Tyr373 and Asp375 in the Kv1 . 2/Kv2 . 1 [voltage-gated potassium channel isoforms 1 . 2 and 2 . 1] chimera crystal structure [PDB 2R9R] ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01289 . 003 The Kv channel pore domain contains intermeshed aromatic side chains , an arrangement termed the ‘aromatic cuff’ that is located at the extracellular end of the selectivity filter and pore helix ( Figure 1 ) . This highly conserved region has long been suggested to play a part in the stability of the pore and likely slow inactivation ( Doyle et al . , 1998; Larsson and Elinder , 2000; Kurata and Fedida , 2006 ) , yet the dynamic rearrangements during inactivation have remained poorly resolved given that many side chains within this region are intolerant to replacement , likely due to the large chemical and steric changes produced by standard mutagenesis . Here , we bypass this experimental hurdle by employing subtle synthetic derivatives of naturally occurring side chains in combination with concatenated subunits to probe the inter- and intra-subunit atomic determinants that control the onset and cooperativity of slow inactivation in Kv channels . Crystallographic data demonstrate the close physical proximity of highly conserved Asp and Trp side chains within the same subunit of potassium channels ( Figure 2A ) ( Doyle et al . , 1998; Long et al . , 2005 ) . Mutations at these positions ( Asp447 and Trp434 by Shaker numbering , which is used throughout the manuscript ) elicit drastic changes in channel function , including effects on inactivation ( Perozo et al . , 1993; Molina et al . , 1997 , 1998; Yang et al . , 1997; Loots and Isacoff , 2000; Loboda et al . , 2001; Cordero-Morales et al . , 2011 ) . However , functional studies have not demonstrated the chemical basis of the interaction between these two side chains . One previously proposed ( Cordero-Morales et al . , 2011; Hoshi and Armstrong , 2013 ) , but untested possibility is that a H-bond is formed between the hydrogen on the indole nitrogen of Trp434 and the carboxylate moiety of Asp447 , and disrupting this H-bond would promote conformational changes associated with slow inactivation . If correct , then a side chain such as Glu ( with altered stereochemistry , but a conserved negatively charged carboxylate ) , might weaken the interaction but not abolish it completely . By contrast , the virtually isosteric but uncharged Asn side chain should not form a significant H-bond with Trp434 and should thus result in dramatically accelerated inactivation . Indeed , the Asp447Glu mutation leads to a rapid and complete decrease in ionic current ( Molina et al . , 1997 , 1998 ) , and this loss in conductance can be drastically slowed by addition of extracellular tetraethyl ammonium ( TEA ) . This deceleration of current decay in the presence of extracellular TEA is a hallmark of slow inactivation and can serve to discriminate this process from other phenotypes , such as fast inactivation or generic protein dysfunction ( Grissmer and Cahalan , 1989; Choi et al . , 1991; Molina et al . , 1997 ) . The charge-neutralizing Asp447Asn mutation resulted in gating currents only , generally interpreted to reflect an inactivation phenotype too rapid to resolve ( Figure 2B and Table 1 ) ( Hurst et al . , 1996; Yang et al . , 1997; Loots and Isacoff , 1998 ) . Conversely , manipulations of the putative H-bonding partner Trp 434 should cause complimentary effects . For example , if the hydrogen on the indole nitrogen of Trp434 does , indeed , play a critical role in slow inactivation , then removing it should drastically increase the rate of inactivation . However , there are no naturally occurring Trp derivatives that can faithfully test this hypothesis without perturbing the local structure . Other ( smaller ) aromatic side chains markedly alter channel function: Tyr in position 434 speeds up slow inactivation ( and is sensitive to TEA , Figure 2—figure supplement 1 ) , while Phe results in gating currents only ( Figure 2C; Table 1 ) ( Perozo et al . , 1993; Yang et al . , 1997; Cordero-Morales et al . , 2011 ) . Thus , while structural data suggests a possible H-bond between Trp434 and Asp447 , the available functional data cannot definitely discriminate between the roles of side chain size and/or volume or hydrogen bonding ability being the major determinant of slow inactivation at position 434 . We therefore employed synthetic derivatives of Trp to test the hypothesis that a H-bond between Trp434 and Asp447 is a rate-controlling interaction for slow inactivation . If this H-bond is a critical determinant of the rate of slow inactivation , predicted outcomes are that strengthening the interaction should decelerate inactivation , while weakening the putative H-bond would be expected to increase the rate of inactivation . To this end , we introduced F4-Trp ( Figure 2D ) , a fluorinated Trp derivative that increases the acidity of the hydrogen on the indole nitrogen ( Deutsch and Taylor , 1987 , 1989 ) , while leaving side chain size and hydrophobicity intact . This slowed the rate of slow inactivation by threefold ( Figure 2E ) . However , this slowing is preceded by an initial faster inactivating component in the Trp434TAG + Trp trace ( gray trace in Figure 2D ) , which may stem from a small degree of nonspecific incorporation of ( endogenous ) amino acids other than the one ligated to the tRNA ( in this case Trp ) . This likely produces an underestimate of the true slowing of inactivation through fluorination as all naturally occurring amino acids other than Trp in position 434 lead to accelerated slow inactivation . As a chemical complement , we sought to incorporate 2-amino-3-indol-1-yl-propionic acid ( Ind , Figure 1F ) ( Lacroix et al . , 2012 ) , a synthetic isosteric Trp derivative in which the indole hydrogen cannot act as a H-bond donor . As co-injection of Trp434TAG cRNA and Ind-coupled tRNA alone did not yield measurable ionic currents , Wild type cRNA was mixed with Trp434TAG cRNA and Ind-coupled tRNA . However , even in the presence of WT ( Wild type ) subunits , the resulting heteromeric channel population displayed a 70-fold increase in inactivation rate ( Figure 2G ) , demonstrating that breaking the proposed Asp447–Trp434 H-bond in the absence of steric perturbation leads to substantially accelerated inactivation . Similar to our recordings with Trp434TAG + Trp alone , we observed an initial fast inactivating component with the Trp434TAG + Trp: WT mix ( gray trace in Figure 2F ) , possibly indicating a small degree of nonspecific incorporation of endogenous amino acids in position 434 . However , this also is unlikely to have a major impact on our results because , first , the resulting fast component is a minor component only visually discernible during the initial phase ( less than 1 s ) of the depolarization and , second , the overall measured effect is likely more affected ( slowed ) by the presence of the WT subunits that were necessary to obtain ionic currents with Ind-containing subunits . We thus conclude that slow inactivation is amenable to the simple atomic ‘push/pull’ of a single H-bond , and manipulation of the strength of this H-bond generates a spectrum of inactivation rates , that can be tuned to occur faster or slower than what is observed in WT channels . 10 . 7554/eLife . 01289 . 004Figure 2 . Evidence for an intra-subunit H-bond between Asp447 and Trp434 . ( A ) Structure of a Kv1 . 2/2 . 1 chimera ( 2R9R ) pore region demonstrating the physical proximity of Asp447 and Trp434 ( Shaker residue numbering ) ; ( B ) Chemical structures of side chains at position 447 and ( normalized ) representative currents for Asp447 ( WT ) , Asp447Glu and Asp447Asn ( −80 mV to +20 mV in 10 mV increments ) . Inset for Asp447Glu shows the traces recorded for a pulse to +20 mV in the absence and presence of 30 mM TEA ( see Figure 2—figure supplement 1 for further details ) ; ( C ) Chemical structures of side chains at position 434 and ( normalized ) representative currents for Trp434 ( WT ) , Trp434Tyr and Trp434Phe ( −80 mV to +20 mV , 10 mV increments ) . Inset for Trp434Tyr shows the traces recorded for a pulse to +20 mV in the absence and presence of 30 mM TEA ( see Figure 2—figure supplement 1 for further details ) . Unlike the Trp434Phe mutant the Asp447Asn mutant remained nonconducting on the Thr449Val background as shown in Figure 2—figure supplement 2; the vertical scale bar indicates 1 μA; ( D ) / ( F ) Left panels: model for the Asp447–Trp434 pair ( based on PDB 2R9R ) for either Trp and F4-Trp ( D ) or Trp and Ind ( F ) ; Right panels: normalized sample currents for WT , Trp434TAG + Trp or Trp434TAG + F4-Trp ( D ) or WT , Trp434TAG + Trp: WT or Trp434TAG + Ind: WT ( F ) . The initial faster decay of the gray traces in ( D ) and ( F ) may indicate that a very small fraction of channels has incorporated amino acids other than the one ligated to the tRNA , see text for details . Note that for the experiments in ( F ) and ( G ) WT cRNA was mixed with Trp434TAG cRNA to account for the fact the Trp434TAG + Ind alone did not yield measurable ionic currents ( see ‘Materials and methods’ for details ) . Note the different time scales in ( D ) and ( F ) ; ( E ) / ( G ) Averaged inactivation time constants for the constructs shown in ( D ) and ( E ) , respectively . Single exponentials were fit to the entire ( 45 s ) depolarizations in ( D ) , while in ( F ) the time constants for Ind were determined by fitting only the initial 500 ms of the depolarization ( 5 s for WT and Trp434TAG + Trp: WT ) ; Note the logarithmic scaling in ( G ) ; *p<0 . 05 ( WT vs mutants ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01289 . 00410 . 7554/eLife . 01289 . 005Figure 2—figure supplement 1 . Tuning the Asp447–Trp434 intra-subunit H-bond . ( A ) / ( B ) Recordings with Asp447Glu ( A ) and Trp434Tyr ( B ) in the absence ( black ) or presence of 30 mM TEA ( −80 mV to +20 mV , 10 mV increments ) , clearly demonstrating their sensitivity to extracellular TEA , a hallmark of slow inactivation . The insets show the inactivation time constants at + 20 mV for WT and Asp447Glu or Trp434Tyr in the absence or presence of TEA; single exponentials were fit to the rst 100 ms ( Asp447Glu control ) or the entire length of the depolarization ( all other currents shown above ) ; *p=0 . 05 ( WT vs mutant ) , **p=0 . 05 ( control vs TEA for mutants ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01289 . 00510 . 7554/eLife . 01289 . 006Figure 2—figure supplement 2 . Unlike Trp434Phe , Asp447Asn remains nonconducting on the Thr449Val background . ( A ) / ( B ) Representative currents of Trp434Phe ( A ) and Asp447Asn ( B ) on either the WT background ( black ) or the Thr449Val background ( red ) ( −80 mV to +20 mV , 10 mV increments; the vertical scale bar indicates 1 μA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01289 . 00610 . 7554/eLife . 01289 . 007Table 1 . Mutants that result in gating currents only show very similar gating charge-voltage ( QV ) relationsDOI: http://dx . doi . org/10 . 7554/eLife . 01289 . 007ConstructV1/2 ( mV ) ZnTrp434Phe−52 . 9 ± 0 . 95 . 2 ± 0 . 36Asp447Asn−53 . 4 ± 0 . 95 . 3 ± 0 . 36Thr439Val−51 . 0 ± 1 . 45 . 5 ± 0 . 35Tyr445Ala−50 . 3 ± 2 . 15 . 3 ± 0 . 44Displayed are the values for the midpoints ( V1/2 ) , the amount of gating charge ( Z ) of the QVs ( derived from the OFF gating currents ) and the number of experiments conducted ( n ) . Slow inactivation is highly cooperative ( Ogielska et al . , 1995; Panyi et al . , 1995; Yang et al . , 1997 ) and the single mutation , Trp434Phe , in the first of four concatenated subunits accelerates slow inactivation ( Yang et al . , 1997 ) . We reasoned that the phenotype obtained by breaking the putative H-bond through manipulations at Trp434 should be mimicked by mutations at the complementary Asp447 site . Indeed , the Asp447Glu mutation in the first of four concatenated Shaker subunits accelerated inactivation rates , similar to those obtained from Trp434Phe concatemers ( Figure 3A , B ) . Together , these data support the notion that the strength of the Asp447–Trp434 intra-subunit H-bond is directly correlated with slow inactivation rate , suggesting that breaking of this interaction is an intrinsic timing mechanism that tightly regulates Kv channel activity . 10 . 7554/eLife . 01289 . 008Figure 3 . Concatemers support the notion of an intra-subunit H-bond between Asp447 and Trp434 . ( A ) Concatemer structure and ( normalized ) representative currents ( 5 s pulses from −20 mV to +20 mV , 10 mV increments ) for WT , Trp434Phe and Asp447Glu concatemers , respectively . The vertical scale bars indicate 2 μA . The insets show recording from the same cells in the presence of 30 mM TEA; ( B ) Averaged inactivation rates ( logarithmic scaling ) over different voltages for the constructs shown in ( A ) . Note that for the Trp434Phe and Asp447Glu concatemers only the first 2 s of the depolarization were fit with a single exponential . To avoid a potential bias of this approach , we have also analyzed the time to half-maximal current for all constructs . Importantly , this approach yielded similar results , see Figure 3—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 01289 . 00810 . 7554/eLife . 01289 . 009Figure 3—figure supplement 1 . Comparison of different metrics to determine the rate of inactivation . The decay of ionic current of WT , Trp434Phe and Asp447Glu concatemers was either fit with a single exponential ( colored symbols , reproduced from Figure 3B ) or quantfied by analyzing the time to half-maximal current ( empty symbols ) . Displayed is the data for depolarizations to +20 mV; note that long ( 20 s ) depolarizations were necessary to determine the time to half-maximal current for WT concatemers . DOI: http://dx . doi . org/10 . 7554/eLife . 01289 . 009 Structural evidence suggests that Trp435 ( Figure 4A ) forms an inter-subunit H-bond via its hydrogen on the indole nitrogen with the Tyr445 hydroxyl ( Doyle et al . , 1998; Larsson and Elinder , 2000; Kurata and Fedida , 2006 ) , and therefore substitution of Tyr or Phe for Trp435 would be expected to disrupt this H-bond , and potentially accelerate inactivation ( as observed for aromatic substitutions of the adjacent Trp434 residue ) . However , while the Trp435Ala mutation produced non-functional channels ( as suggested by the absence of ionic or gating currents ) , Tyr and Phe substitutions at position 435 resulted in WT-like slow inactivation rates ( Figure 4B , C ) , ruling out a role for Trp435 H-bonding in slow inactivation . However , the Tyr445Phe mutation results in a mix of gating current and ionic current , with markedly accelerated slow inactivation ( Harris et al . , 1998 ) ( a phenotype antagonized by TEA ) ( Figure 4D , Figure 4—figure supplement 1 ) . Furthermore , Tyr445Ala channels exhibited gating currents akin to Trp434Phe channels ( Figure 4D; Table 1 ) ( Heginbotham et al . , 1994 ) . Interestingly , crystallographic data ( Doyle et al . , 1998; Long et al . , 2007 ) place the Tyr445 hydroxyl within 3 Å of the hydroxyl moiety of a conserved Thr or Ser side chain ( Thr439 in Shaker , Figure 1B ) , raising the intriguing possibility of an uncharacterized inter-subunit interaction between Tyr445 and Thr439 . Consistent with this possibility , the Thr439Val mutant exhibited exclusively gating currents ( Figure 4E ) , while the Thr439Ser mutation resulted in a modest threefold faster inactivation than observed for WT channels , likely indicating a minor role for the Thr439 methyl group in slow inactivation ( Figure 4E , Figure 4—figure supplement 3 ) . Further , and unlike the Trp434Phe mutation ( Kitaguchi et al . , 2004 ) , introducing Tyr445Ala , Tyr445Val or Thr439Val on the Thr449Val background did not slow inactivation to an extent that ionic currents could be observed ( Figure 4—figure supplement 3 ) . While these data point to an inter-subunit H-bond between Tyr445 and Thr439 , they do not inform on the cooperativity between individual subunits or whether this interaction contributes to the same extent as the Asp447–Trp434 H-bond . However , we speculated that breaking this inter-subunit H-bond may have a more pronounced effect when introduced in only a single subunit , compared to the intra-subunit Trp434–Asp447 H-bond . Consistent with this possibility , either the Tyr445Ala or the Thr439Val mutation in the first of four concatenated Shaker subunits ( Figure 5A ) had similar phenotypes , with a clearly biphasic inactivation phenotype composed of fast ( around 50 ms ) and WT-like slow ( around 3 s ) components ( Figure 5B ) . The fast component was affected by TEA , implicating a slow inactivation mechanism ( Figure 5—figure supplement 1 ) . The sizable gating currents at hyperpolarized potentials ( Figure 5—figure supplement 2 ) suggest that either mutation ( one per concatenated tetramer ) reduces the ratio of ionic current to gating charge at a given voltage , an effect that would arise if a significant portion of channels rapidly adopt a non-conducting conformation . To further test this possibility , the pore blocker agitoxin II ( Eriksson and Roux , 2002; Banerjee et al . , 2013 ) was used to assay the gating currents as a metric for normalization of the number of channels present in the cell , and thus permitting an estimate of the relative reduction in ionic current in the mutant concatemers relative to WT concatemers . Indeed , we found the ratio of ionic current to gating charge to be significantly reduced in both mutant concatemers ( Figure 5C ) , suggesting that a sizable proportion of channels rapidly enter an inactivated state upon depolarization . This behavior is further illustrated in Figure 5D , where currents from Tyr445Ala or Thr439Val concatemers were normalized to WT ( by gating charge ) , thus emphasizing the very rapid and near-complete inactivation in Tyr445Ala and Thr439Val concatemers . These experiments establish a previously unidentified inter-subunit H-bond between Thr439 and Tyr445 that controls slow inactivation in Kv channels . 10 . 7554/eLife . 01289 . 010Figure 4 . An inter-subunit H-bond connects Tyr445 with Thr439 , not Trp435 . ( A ) Structure of a Kv1 . 2/2 . 1 chimera ( 2R9R ) pore region demonstrating the physical proximity of Tyr445 to both Thr439 and Trp435 on the adjacent subunit ( Shaker residue numbering ) . Note that the position equivalent to position 439 in Shaker ( Thr439 ) is a serine in the Kv1 . 2/2 . 1 chimera; ( B ) Chemical structures of side chains at position 435 and representative currents for Trp435 ( WT ) , Trp435Tyr and Trp435Phe ( −80 mV to +20 mV , 10 mV increments ) . The vertical scale bar indicates 1 μA; ( C ) Representative normalized currents for a 45 s depolarization to +20 mV for Trp435 ( WT ) , Trp435Tyr and Trp435Phe . Inset shows average inactivation time constants for the constructs shown in ( B ) ( single exponential fit over the entire duration of the depolarization ) ; ( D ) / ( E ) Chemical structures and ( normalized ) representative currents for different side chains in position 445 ( D ) and 439 ( E ) , respectively ( −80 mV to +20 mV , 10 mV increments ) . The vertical scale bars indicate 1 μA , note the different time scales . Inset for Tyr445Phe shows the traces recorded for a pulse to +20 mV in the absence and presence of 30 mM TEA ( see Figure 4—figure supplement 1 for further details ) . The Tyr445Ala and Thr439Val mutants remained nonconducting on the Thr449Val background as shown in Figure 4—figure supplement 2 . The inactivation time constant ( τ ) for Thr439Ser was 948 ± 30 ms compared to 3247 ± 186 ms for WT channels ( see Figure 4—figure supplement 3 for further details ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01289 . 01010 . 7554/eLife . 01289 . 011Figure 4—figure supplement 1 . Tuning the Tyr445–Thr439 inter-subunit H-bond . ( A ) Recordings with Tyr445Phe in the absence ( black ) or presence of 30 mM TEA ( green ) ( −80 mV to +20 mV , 10 mV increments ) , clearly demonstrating the sensitivity to extracellular TEA , a hallmark of slow inactivation; ( B ) Inactivation time constants at + 20 mV for WT and for Tyr445Phe in the absence or presence of TEA; single exponentials were fit to the entire 200 ms of the depolarization; *p<0 . 05 ( WT vs Tyr445Phe ) , **p<0 . 05 ( control vs TEA for Tyr445Phe ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01289 . 01110 . 7554/eLife . 01289 . 012Figure 4—figure supplement 2 . Tyr445Ala , Tyr445Val and Thr439Val remain nonconducting on the Thr449Val background . ( A ) / ( B ) Representative currents of Tyr445Ala and Tyr445Val ( A ) and Thr439Val ( B ) on either the WT background ( black ) or the Thr449Val background ( red ) ( −80 mV to +20 mV , 10 mV increments; the vertical scale bar indicates 1 μA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01289 . 01210 . 7554/eLife . 01289 . 013Figure 4—figure supplement 3 . The methyl moiety of Thr439 may play a minor role in inactivation . ( A ) GVs for Thr439Ser ( V1/2 Thr441Ser: −28 . 4 ± 0 . 3 mV; vs V1/2 for WT: −23 . 5 ± 0 . 9 mV ) ; ( B ) Representative currents for WT and Thr439Ser ( 5 s pulses from −20 mV to +20 mV , 10 mV increments ) ; ( C ) Averaged inactivation rates at +20 mV for WT and Thr439Ser , respectively; *p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 01289 . 01310 . 7554/eLife . 01289 . 014Figure 5 . Breaking the Tyr445–Thr439 inter-subunit H-bond results in rapid inactivation . ( A ) Concatemer structure and representative currents ( 5 s pulses from −20 mV to +20 mV , 10 mV increments ) for WT , Tyr445Ala and Thr439Val concatemers , respectively . The vertical scale bars indicate 2 μA . The insets compare normalized currents in response to a +20 mV step for WT and Tyr445Ala or Thr439Val concatemers over a short ( 200 ms ) time scale ( see Figure 5—figure supplement 1 for details on TEA sensitivity; see Figure 5—figure supplement 2 for gating currents at hyperpolarized potentials with Tyr445Ala and Thr439Val concatemers ) ; ( B ) Averaged inactivation time constants over different voltages for the constructs shown in ( A ) , with rates for Tyr445Ala and Thr439Val concatemers split into fast and slow components ( single exponentials were fit to the first 50 ms and the remainder of the depolarization , respectively ) . Note the logarithmic scaling; ( C ) Ratio of maximal ionic current to gating charge ( both recorded at +20 mV ) for WT , the Tyr445Ala and the Thr439Val concatemers ( gating currents were recorded in the presence of 10 μM agitoxin II , not shown ) ; *p<0 . 05 ( WT vs mutants ) ; ( D ) Comparison of ionic currents recorded at +20 mV normalized to the amount of gating charge recorded from WT concatemers and the Tyr445Ala ( left panel ) and the Thr439Val ( right panel ) concatemers , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 01289 . 01410 . 7554/eLife . 01289 . 015Figure 5—figure supplement 1 . Characterizing the Thr439Val and Tyr445Ala concatemers . ( A ) / ( C ) Representative recordings from Thr439Val ( A ) and Tyr445Ala ( C ) concatemers in the absence ( black ) or presence ( green ) of 30 mM TEA ( −20 mV to +20 mV , 10 mV increments ) , clearly demonstrating their sensitivity to extracellular TEA , a hallmark of slow inactivation; ( B ) / ( D ) Averaged inactivation time constants at +20 mV for Thr439Val ( B ) and Tyr445Ala ( D ) concatemers ( single exponentials were fit to the first 50 ms of the control , and to the entire 200 ms of the depolarization in the presence of TEA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01289 . 01510 . 7554/eLife . 01289 . 016Figure 5—figure supplement 2 . Pronounced gating currents at hyperpolarized potentials in Tyr445Ala and Thr439 Val concatemers . ( A ) / ( B ) Representative currents of Tyr445Ala concatemers ( A ) and Thr439Val concatemers ( B ) showing signicant gating currents before the onset of ionic currents at around −40 mV ( 50 ms depolarizations from a holding potential to the indicated test potential ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01289 . 016 Thr441 and Thr442 are highly conserved amongst Kv channels and are favorably located at the junction of selectivity filter and pore helix ( Figure 6A ) for a possible role in pore stability and/or slow inactivation . We aimed to compare the relative contribution of Thr441 and Thr442 to slow inactivation with more extracellular structural elements of the selectivity filter . Interestingly , mutations here produce vastly different outcomes ( depending on the amino acid substitution ) , including loss-of-function , alterations in open state stability , and the appearance of subconductance states with diminished selectivity ( Yool and Schwarz , 1991; Heginbotham et al . , 1994; Zheng and Sigworth , 1997 ) . Consistent with these reports , we found that valine substitutions a 441 and 442 had severe consequences: while Thr442Val displayed a non-expressing phenotype ( Figure 6C ) ( Zheng and Sigworth , 1997 ) , the Thr441Val mutation resulted in voltage-dependent currents both in the inward and the outward direction , as well as reduced potassium selectivity ( Figure 6B , Figure 6—figure supplement 1 ) , suggesting a significant perturbation of the local structure . However , Thr441Ser channels displayed a WT-like GV whereas Thr442Ser channels displayed a modest left-shifted activation relationship , yet both had inactivation behaviors similar to WT channels ( Figure 6B , C , Figure 6—figure supplement 2 ) , supporting the notion that the hydroxyl moieties at positions 441 and 442 support normal pore function ( in addition to the proposed role of the Thr442 methyl group in ion binding ( Rossi et al . , 2013 ) ) . However , despite the disruptive phenotypes observed in homotetrameric Thr441Val or Thr442Val channels , single subunit mutations had minimal effects in the background of a concatenated tetramer ( Figure 6D , E ) . Thus , hydroxyl removal at either 441 or 442 produces channel phenotypes that are very mild compared to perturbations within the aromatic cuff , and importantly , their effects are not propagated to other subunits in the channel tetramer , suggesting they are unrelated to the cooperative mechanism of slow inactivation . We believe this is an important finding as it demonstrates that severe functional consequences of mutations at the selectivity filter are not necessarily linked to changes in slow inactivation . 10 . 7554/eLife . 01289 . 017Figure 6 . Thr441 and Thr442 are critical to channel function but not inactivation . ( A ) Structure of a Kv1 . 2/2 . 1 chimera ( 2R9R ) pore region highlighting the positions of Thr441 and Thr442 at the bottom of the selectivity filter ( Asp447 and Trp434 are shown for reference; all by Shaker numbering ) ; ( B , C ) Chemical structures and ( normalized ) representative currents for different side chains in position 441 ( B ) and 442 ( C ) , respectively ( −80 mV to +20 mV , 10 mV increments for WT , Thr441Ser and Thr442Ser; −200 mV to +150 mV , 10 mV increments for Thr441Val and a single pulse to +20 mV for Thr442Val ) . Note that recordings for Thr441Val were conducted from a holding potential of 0 mV with no leak subtraction . The vertical scale bars indicate 1 μA . See Figure 6—figure supplement 1 for details on the loss of potassium selectivity of Thr441Val . GVs and inactivation behavior of Thr441Ser and Thr442Ser are shown in Figure 6—figure supplement 2; ( D ) Concatemer structure and representative currents ( 5 s pulses from −20 mV to +20 mV , 10 mV increments ) for WT , Thr441Val and Thr442Val concatemers , respectively . The vertical scale bars indicate 2 μA . The small inward tail current for Thr441Val could indicate reduced potassium selectivity; ( E ) Averaged inactivation time constants over different voltages for the constructs shown in ( D ) ; similar results were obtained with longer ( 20 s ) depolarizations , see Figure 6—figure supplement 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 01289 . 01710 . 7554/eLife . 01289 . 018Figure 6—figure supplement 1 . The Thr441Val mutation results in a loss of potassium selectivity . Current vs voltage plot for the Thr441Val mutant ( n = 9 ) . The inset shows that the current reverses around −40 mV . DOI: http://dx . doi . org/10 . 7554/eLife . 01289 . 01810 . 7554/eLife . 01289 . 019Figure 6—figure supplement 2 . Hydroxyl moieties are critical to channel function in positions 441 and 442 . ( A ) / ( D ) GVs for Thr441Ser ( A ) and Thr442Ser ( V1/2 Thr441Ser: −19 . 7 ± 1 . 5 mV; V1/2 Thr442Ser: −45 . 6 ± 0 . 6 mV; vs V1/2 for WT: −23 . 5 ± 0 . 9 mV ) ; ( B ) / ( E ) Representative currents for WT and Thr441Ser ( B ) and WT and Thr442Ser ( E ) , respectively ( 5 s pulses from −20 mV to +20 mV , 10 mV increments ) ; ( C ) / ( F ) Averaged inactivation rates at +20 mV ( fit with a single exponential ) for WT and Thr441Ser ( C ) and WT and Thr442Ser ( F ) , respectively; *p≤0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 01289 . 01910 . 7554/eLife . 01289 . 020Figure 6—figure supplement 3 . Comparison of inactivation time constants using different pulse durations . Averaged inactivation time constants obtained from WT , Thr441Val and Thr442Val concatemers by fitting the current decay during 5 s ( colored symbols , reproduced from Figure 6E ) or 20 s ( empty symbols ) depolarizations to +20 mV with a single exponential . DOI: http://dx . doi . org/10 . 7554/eLife . 01289 . 020 Despite intense experimental scrutiny for almost 25 years , the molecular and atomic origin ( s ) of the ability of Kv channels to enter a non-conducting conformation in the presence of a sustained ( voltage ) stimulus has remained enigmatic . A major challenge with addressing the contribution of individual side chains in the selectivity filter and pore helix to slow inactivation is that many amino acids lack naturally occurring analogs that allow subtle manipulation without dramatic disruption of the overall structure of this critical protein region . Furthermore , since slow inactivation is tightly coupled to ion occupancy in the selectivity filter , it has been difficult to distinguish direct effects on the mechanistic underpinnings of slow inactivation from indirect effects arising from changes in the structural integrity of the selectivity filter . Here , we overcome this hurdle by employing subtle synthetic analogs of naturally occurring amino acids and by introducing isolated mutations in single subunits . When using concatenated Shaker constructs to introduce single mutations in a fourfold symmetric channel , it is crucial to confirm that the concatenated subunits do not form functional channels that vary in their stoichiometry from that predicted from the cloning strategy . Albeit possible ( McCormack et al . , 1992; Hurst et al . , 1995 ) , we believe the constructs used here assemble correctly for three reasons . First , the Thr439Val concatemers and the Tyr445Ala concatemers showed almost complete inactivation over a period of only 200 ms ( when corrected for the amount of gating charge ) . Second , we observed very low ratios of Imax to Qmax for Thr439Val concatemers and Tyr445Ala concatemers . Both scenarios are not compatible with the idea of a significant WT-only channel subpopulation; Lastly , Sigworth and co-workers have used the same concatemers and successfully demonstrated that channels containing only a single mutated subunit generally assemble in the correct stoichiometry ( Yang et al . , 1997 ) . We conclude that the ( vast majority of ) concatemers assemble correctly , although we cannot ultimately rule out a small subpopulation of channels with WT-like slow inactivation . Further , we employed fluorinated Trp derivatives , which have been used extensively to probe electrostatic ( cation-pi ) interactions ( Dougherty , 1996 ) between Trp side chains and organic cations as fluorination allows a step-wise dispersion of the electronegative surface potential of aromatic side chains ( Pless and Ahern , 2013 ) . As such , our finding that F4-Trp in position 434 significantly slows channel inactivation could be interpreted as a result of a cation-pi interaction at Trp434 that is being diminished by fluorination . However , if this were true , Ind , a synthetic amino acid which lacks H-bonding ability , should have no effect on channel inactivation as it is isosteric and isoelectric to the native Trp side chain . By contrast , we observe a substantial increase in the rate of slow inactivation with Ind in position 434 , a result not compatible with the notion of an energetically significant cation-pi interaction at Trp434 . We thus conclude that it is the ability of the indole nitrogen to participate in a H-bond that regulates the strength of the intra-subunit interaction between Trp434 and Asp447 . Together , our experimental approaches provide strong evidence for two H-bonds that are critical for slow inactivation of Kv channels: one that confers stability within an individual subunit ( the Trp434–Asp447 interaction ) , and a second that stabilizes the relative orientation of two adjacent subunits ( the Tyr445–Thr439 interaction ) ( Figure 7 ) . Although disruption of the Trp434–Asp447 interaction has profound effects on slow inactivation , breaking of the Tyr445–Thr439 interaction elicits more functionally significant phenotypes . We surmise that these comparatively more severe phenotypes seen when disrupting the Tyr445–Thr439 pair arise from its location at the inter-subunit interface , but cannot exclude the possibility that this difference arises from the fact that the Tyr445 backbone carbonyl is also directly involved in coordinating permeant ions . Furthermore , the Tyr445–Thr439 interaction is , to our knowledge , the first evidence for an inter-subunit interaction contributing to slow inactivation , possibly providing an explanation for the observed subunit cooperativity during slow inactivation . However , although previous studies have suggested evidence for both constriction ( Baukrowitz and Yellen , 1996; Liu et al . , 1996 , 1997 ) and dilation ( Hoshi and Armstrong , 2013 ) of the selectivity filter , the data here is not definitive in distinguishing these models of slow inactivation . 10 . 7554/eLife . 01289 . 021Figure 7 . A network of inter- and intra-subunit H-bonds regulates slow inactivation . ( A ) The left panel shows a top view of pore helix and selectivity filter ( based on the Kv1 . 2/2 . 1 chimera structure ( 2R9R ) ; individual subunits are colored in gray , cyan , green and yellow , respectively ) . Note the backbone carbonyls are shown for Tyr445 to highlight their role in the coordination of potassium ions ( gray circle ) . The center panel highlights the two proposed H-bonds: Thr439–Tyr445 ( inter-subunit , red oval , ) and Asp447–Trp434 ( intra-subunit , blue oval ) , all by Shaker numbering . The panel on the right compares the averaged inactivation time constants over a range of voltages for different concatemers ( data reproduced from Figure 3 and Figure 5; note that for Tyr445Ala and Thr439Val only the fast components are displayed ) . Note the arrows pointing to the respective interactions in the model in the center . DOI: http://dx . doi . org/10 . 7554/eLife . 01289 . 021 The notion that side chains critical to slow inactivation cluster around the ‘aromatic cuff’ ( formed between the extracellular end of the selectivity filter and the pore helix ) is further supported by the marked differences between side chains at the outer vs the inner end of the selectivity filter and pore helix: only those located around the ‘aromatic cuff’ result in notable effects on slow inactivation that are propagated to the entire channel ( Figures 2–5 ) , while those residing in the middle or lower section of the selectivity filter do not affect slow inactivation ( see Figure 6 for positions 441 and 442; see ( Heginbotham et al . , 1994 ) for position 443 ) . Overall , the results point towards an intriguing molecular explanation for the mechanism of slow inactivation: upon depolarization and channel opening , the stability of the channel open state is proportional to the strength of two H-bonds that regulate entry into slow inactivation , thus endowing Kv channels with an intrinsic timing mechanism that tightly regulates their biological activity . During a sustained voltage stimulus , channels experience a sequential breaking of the Trp434–Asp447 and Tyr445–Thr439 H-bonds and given the relative arrangement of their hydroxyl moieties this would likely result in an anti-clockwise swivel movement of the Tyr445 backbone carbonyl away from the permeation pathway , ultimately disrupting the coordination and occupancy of potassium ions at the outer end of the selectivity filter . Such a scenario would lead to mutual repulsion between the Tyr445 backbone carbonyls of the remaining three subunits ( Almers and Armstrong , 1980; Hoshi and Armstrong , 2013 ) , further lowering filter-occupancy at its outer mouth . The resulting strain could trigger a cascade of disrupted H-bonds critical to inactivation near the extracellular end of the selectivity filter in all subunits , ultimately resulting in a fully inactivated channel . Shaker IR ( Inactivation Removed by deletion of amino acids 6–46 ) cDNA in pBSTA was used as the parent clone unless stated otherwise ( note that Cys301 and Cys308 were present ) . For mutations , standard site-directed mutagenesis was employed in combination with automated sequencing to confirm successful incorporation of mutations . For experiments with concatenated Shaker tetramers ( Yang et al . , 1997 ) , mutations were introduced in the first of the four subunits only: the first subunit was subcloned into pBSTA with SacI and XbaI , and standard site-directed mutagenesis was used to introduce mutations followed by sequence verification . Next , the mutated construct was subcloned ( with SacI and XbaI ) back into the parent concatemer . For electrophysiology experiments , Stage V-VI Xenopus oocytes were prepared , and injected with cRNA transcribed with the T7 mMessage mMachine kit ( Ambion , Austin , TX ) as previously described ( Pless et al . , 2013 ) . Oocytes were incubated at 18°C and all recordings were conducted within 12–72 hr after injection . The fluorinated Trp derivative F4-Trp ( 4 , 5 , 6 , 7-F4-Trp ) was purchased from Asis Chem ( Watertown , MA ) and the Trp analog 2-Amino-3-indol-1-yl-propionic acid ( Ind ) was synthesized as described previously ( Lacroix et al . , 2012 ) . The principle of the in vivo nonsense suppression methodology is outlined elsewhere ( Pless and Ahern , 2013 ) . In short , nitroveratryloxycarbonyl ( NVOC ) was used to protect the amine of the synthetic amino acid , while the carboxyl group was activated as the cyanomethyl ester for coupling to the dinucleotide pdCpA ( Dharmacon , Lafayette , CO ) . The resulting product was stored in DMSO at −80°C before enzymatic ligation to a modified ( G73 ) Tetrahymena thermophila tRNA , which was synthesized using an oligonucleotide by Integrated DNA Technologies ( Coralville , IA ) as a template . The NVOC protection group of the aminoacylated tRNA-UAA was removed directly prior to co-injection with the cRNA by UV irradiation for 8 min at 400 W . In a typical experiment , 10–80 ng of tRNA-UAA and 25–50 ng of cRNA were co-injected in a 50 nl vol . In control experiments , the tRNA coupled to pdCpA ( without an appended synthetic amino acid ) was co-injected with the Trp434TAG cRNA . The control did not yield currents larger than for uninjected oocytes , ruling out significant levels of non-specific amino acid incorporation , or re-charging of the tRNA with endogenous amino acids . Note that incorporation of Ind at position 434 did not result in measurable ionic currents . This was expected given the results with the conventional Trp434Tyr and Trp434Phe mutants , as side chains in position 434 with no propensity to contribute to a H-bond ( such as Ind ) were shown to result in gating currents only . However , the nonsense suppression method is generally not efficient enough to establish current levels of necessary magnitude to resolve gating currents , even for sites with exceptionally high incorporation efficiency ( Pless et al . , 2011 , 2013 ) . We thus co-injected WT Shaker cRNA with the Trp434TAG cRNA for incorporation of Ind . Despite the high degree of cooperativity between subunits ( Figure 3 and Figure 5 ) , this likely results in slower and less complete slow inactivation than would be expected for Ind-containing subunits alone and the 70-fold increase in inactivation rate ( Figure 2 ) is likely to be an underestimate of the real acceleration of inactivation induced by the Ind side chain in position 434 . Two electrode voltage-clamp recordings were conducted with an OC-725C voltage clamp ( Warner , Hamden , CT ) in standard Ringers solution ( in mM ) : 116 NaCl , 2 KCl , 1 MgCl2 , 0 . 5 CaCl2 , 5 HEPES ( pH 7 . 4 ) . TEA ( St . Louis , MO ) and agitoxin-II ( Alomone Labs , Jerusalem , Israel ) were dissolved in Ringers and stored at −20°C until use . Glass microelectrodes with resistances of 0 . 1–1 MΩ were backfilled with 3 M KCl . Currents were acquired using leak subtraction and from a holding potential of −80 mV , unless stated otherwise . To obtain conductance-voltage ( GV ) relationships , isochronal tail current amplitudes were plotted vs the depolarizing pulse potential . All data = mean ± SEM; Student’s t test was used to determine statistically significant differences . To obtain gating charge-voltage ( QV ) relationships , the total area of the off gating charge was plotting against the depolarizing pulse potential .

Proteins are made from long chains of smaller molecules , called amino acids . These chains twist and bend into complex three-dimensional shapes , and sometimes two or more chains , or ‘subunits’ , are packed into a protein . These shapes are often held together by hydrogen bonds between some of the amino acids . Moreover , since the shape of a protein defines its function , some proteins must be able to switch between different shapes to function properly . Ion channels are proteins that form pores through cell membranes , allowing ions to flow in and out of the cell . Potassium ion channels , which are found in neurons and heart muscle cells , have four subunits that move to open or close the central pore in response to various signals . The closing of the channels can be ‘fast’ or ‘slow’ . When the channels are closed quickly ( called fast inactivation ) , a small part of the protein ‘plugs’ the pore from the inside of the cell . However , the mechanism behind slow inactivation remained obscure . It was thought to involve hydrogen bonds between some of the bulky amino acids that are found at the edge the pore . However , testing this hypothesis—by replacing these amino acids with alternatives that cannot form hydrogen bonds—was tricky because none of the 20 naturally occurring amino acids were alike enough to be suitable replacements . Now , Pless et al . have overcome this limitation by using synthetic amino acids that form hydrogen bonds that are stronger or weaker than those formed by the amino acids they are replacing . The results suggest that two types of hydrogen bond keep the pore open: one is a bond between two amino acids in the same subunit , and the other is an inter-subunit bond between amino acids in neighbouring subunits . Pless et al . suggest that opening the channel causes small movements that gradually weaken , and eventually break , these bonds in one of the four subunits . Specific amino acids within the pore are then free to twist and—via a cascade of similar movements in the other three subunits—block the pore and halt the flow of ions . As such , these networks of hydrogen bonds act as pre-set breaking points allowing channels to close , even in response to continued stimulation . Since regulated potassium channel activity underpins healthy neurons and heart muscles; understanding what controls their inactivation rate may lead to new approaches to tune their activity and treatments for important diseases .

2013

Hydrogen bonds as molecular timers for slow inactivation in voltage-gated potassium channels

Cerebral malaria ( CM ) can be classified as retinopathy-positive or retinopathy-negative , based on the presence or absence of characteristic retinal features . While malaria parasites are considered central to the pathogenesis of retinopathy-positive CM , their contribution to retinopathy-negative CM is largely unknown . One theory is that malaria parasites are innocent bystanders in retinopathy-negative CM and the etiology of the coma is entirely non-malarial . Because hospitals in malaria-endemic areas often lack diagnostic facilities to identify non-malarial causes of coma , it has not been possible to evaluate the contribution of malaria infection to retinopathy-negative CM . To overcome this barrier , we studied a natural experiment involving genetically inherited traits , and find evidence that malaria parasitemia does contribute to the pathogenesis of retinopathy-negative CM . A lower bound for the fraction of retinopathy-negative CM that would be prevented if malaria parasitemia were to be eliminated is estimated to be 0 . 93 ( 95% confidence interval: 0 . 68 , 1 ) . Cerebral malaria ( CM ) is responsible for a substantial proportion of the approximately 500 , 000 annual malaria deaths and 2 , 000 , 000 severe malaria cases ( WHO , 2014 , WHO , 2015 ) . CM is defined by the World Health Organization ( WHO ) as unarousable coma with circulating malaria ( Plasmodium ) parasitemia and no known non-malaria causal explanation ( WHO , 2000 ) . Based on the presence or absence of malaria-specific retinal changes , CM can be classified as retinopathy-positive ( Ret+ ) or retinopathy-negative ( Ret- ) ( Lewallen et al . , 1999; Beare et al . , 2006 ) . Ret- CM is a common and devastating condition – 40% of CM cases in our cohort were Ret- and of these , 12% died and 10% developed neurological problems ( Table 1 ) . Autopsy data show that children dying of Ret+ CM have a high degree of sequestration of parasitized red blood cells in cerebral vasculature ( Taylor et al . , 2004 ) , considered the pathological hallmark of CM . The pathogenesis of Ret- CM has been a puzzle , in particular the role of malaria parasitemia ( Postels and Birbeck , 2011 ) . In the only autopsy study among children dying with CM , that we are aware of ( Taylor et al . ( 2004 ) with follow-up results in Milner et al . ( 2015 ) and Barrera et al . ( 2015 ) ) , among children for whom retinopathy was assessed , 41 of 42 children dying with Ret+ CM had substantial cerebral sequestration of parasitized red blood cells in the cerebral microvasculature ( defined as ≥23% of cerebral capillaries had sequestration ) and mostly lacked other identified potential causes of death besides the malaria parasitemia , whereas all 15 children dying with Ret- CM lacked substantial cerebral sequestration ( <23% of cerebral capillaries had sequestration ) and mostly had non-malarial etiologies of death ( see Appendix 1 for causes of death ) ; these numbers update those in Taylor et al . ( 2004 ) to include patients enrolled after 2004 . Incidental malaria parasitemia is common in people living in areas of high malaria transmission . Therefore , it is possible that at least some children with Ret- CM have a non-malarial etiology of coma and an incidental ( asymptomatic ) malaria parasitemia . The pathogenesis of Ret-CM is therefore unclear , and the role of malaria parasitemia in the etiology of the acute illness is unknown ( Postels and Birbeck , 2011 ) . Our aim of the research presented here was to assess the contribution of acute malaria infection in the pathophysiology of Ret- CM . 10 . 7554/eLife . 23699 . 003Table 1 . Characteristics of study participants at admission , Means ± SD for continuous variables . The proportions of missing data are shown in Appendix 1 . There are 3704 community controls , but their characteristics are not shown because only their genotypes and not their clinical characteristics were collected . Bold denotes p-value less than 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 23699 . 003Retinopathy Positive CMRetinopathy Negative CMNon-Malaria Hospital Controlsp-value , Ret + vs . Ret -p-value , Ret + vs . Controlsp-value , Ret – vs . ControlsNumber of participants438288204Female50%52%43%0 . 540 . 110 . 05Age ( months ) 40 ± 2644 ± 3046 ± 300 . 100 . 050 . 53Mid-upper arm circumference ( cm ) 14 . 9 ± 1 . 615 . 0 ± 1 . 714 . 8 ± 1 . 80 . 720 . 530 . 39Weight ( kg ) 12 ± 413 ± 513 ± 60 . 370 . 290 . 74Height ( cm ) 90 ± 1691 ± 1791 ± 200 . 300 . 401 . 00Temperature ( °C ) 38 . 6 ± 1 . 238 . 4 ± 1 . 437 . 7 ± 1 . 50 . 03<0 . 001<0 . 001Febrile ( Temperature≥ 37 . 5°C ) 81%77%56%0 . 23<0 . 001<0 . 001Pulse rate – beats/minute152 ± 26148 ± 24139 ± 280 . 06<0 . 001<0 . 001Respiratory rate – breaths/minute47 ± 1545 ± 1345 ± 150 . 120 . 170 . 93Liver size – cm below costal margin2 . 0 ± 1 . 91 . 5 ± 1 . 91 . 1 ± 1 . 7<0 . 001<0 . 0010 . 04Spleen size – cm below costal margin1 . 7 ± 2 . 11 . 6 ± 2 . 10 . 9 ± 1 . 60 . 56<0 . 001<0 . 001Deep breathing33%25%30%0 . 030 . 590 . 18Blantyre Coma Score: 0 1 2 3 4 514% 35% 49% 1% 0% 0%19% 38% 43% 0% 0% 0%28% 40% 23% 3% 0% 5%0 . 010 . 080 . 90CSF opening pressure – mm of water176 ± 75152 ± 82176 ± 990 . 0010 . 960 . 07Hematocrit -- %19 . 8 ± 6 . 928 . 2 ± 7 . 528 . 1 ± 9 . 6<0 . 001<0 . 0010 . 86Platelets81 , 220± 67 , 219161 , 600± 124 , 747248 , 400± 162 , 287<0 . 001<0 . 0010 . 86Malaria parasitemia – parasites/mm3230 , 500± 321 , 924180 , 500± 280 , 6763 , 619± 28 , 9170 . 03<0 . 001<0 . 001White blood cells13 , 040± 916313 , 020± 892313 , 930± 95440 . 970 . 290 . 31Lactate – mmol/liter8 . 6 ± 5 . 07 . 3 ± 4 . 45 . 5 ± 3 . 90 . 05<0 . 0010 . 007Blood glucose – mmol/liter6 . 1 ± 3 . 96 . 8 ± 4 . 47 . 6 ± 5 . 30 . 03<0 . 0010 . 05CSF white cell count – % ≥ 516%20%24%0 . 310 . 060 . 37Blood culture positive for pathogen4%2%14%0 . 51<0 . 001<0 . 001HIV positive18%17%15%0 . 910 . 590 . 66OutcomesDischarge outcome: Full recovery Neurological Sequalae Died69% 10% 21%78% 10% 12%57% 15% 28%0 . 0030 . 01<0 . 001 Figure 1 depicts three possible pathways to clinically-defined ( WHO-defined ) CM ( Postels and Birbeck , 2011 ) . One pathway is to Ret+ CM for which there is evidence that malaria parasites play a primary role . As noted above , at autopsy , Ret+ CM is associated with the sequestration of parasitized red cells in the cerebral microvasculature ( Taylor et al . , 2004 ) . Compared to children with Ret- CM , those who are Ret+ have increased concentrations of P . falciparum HRP2 , a parasite-produced protein reflecting total body parasite burden ( Seydel et al . , 2012 ) . Ocular funduscopic findings in Ret+ CM mirror the microvascular pathology observed on fluorescein angiography ( MacCormick et al . , 2015 ) and are correlated with the severity of sequestration in both the retina and the brain at autopsy ( Barrera et al . , 2015 ) . Two pathways to Ret- CM , pathway ( a ) and pathway ( b ) are depicted in Figure 1 . As noted above , in patients dying with Ret- CM , the cerebral microvasculature does not have substantial sequestered parasitized erythrocytes ( <23% of cerebral capillaries have sequestration ) , plasma concentrations of HRP2 are decreased , and a variety of non-malarial causes of death have been identified ( Taylor et al . , 2004 ) . For Ret- CM , one potential pathway is asymptomatic parasitemia and another illness that is sufficient , in and of itself , to produce coma ( pathway ( a ) in Figure 1 ) . Another potential pathway is parasitemia leading to uncomplicated malaria illness ( e . g . , fever ) combined with a second insult ( innate or acquired ) , resulting in coma ( pathway ( b ) in Figure 1 ) ; the two hits ( symptomatic malaria+ innate or acquired second factor ) result in the clinical syndrome of Ret- CM . A key unanswered question about the pathogenesis of Ret- CM is , are malaria parasites incidental to coma ( only pathway ( a ) exists ) or do they play a role in the pathogenesis of Ret- CM ( pathway ( b ) exists ) ( Bearden , 2012; Postels and Birbeck , 2011 ) ? 10 . 7554/eLife . 23699 . 004Figure 1 . Potential pathways to clinically-defined cerebral malaria and genetic bottle necks . There are three potential pathogenetic routes to WHO-defined cerebral malaria ( CM ) . The first , shown in red , is the classical pathway: a malaria infection evolves into retinopathy-positive ( Ret+ ) CM . The second and third possibilities produce retinopathy-negative ( Ret- ) CM . In ( a ) the coma is entirely the result of another etiology and the malaria parasitemia is incidental . In ( b ) , the coma is a product of the interaction between the malaria parasitemia and an additional cause ( or causes ) of coma . Sickle cell trait is underrepresented in patients with Ret+ and Ret- cerebral malaria ( CM ) because of the bottleneck at the transition between 'malaria infection' ( asymptomatic malaria ) and 'malaria disease' ( uncomplicated malaria ) . Blood group O is underrepresented in patients with Ret+ CM , but not in those with Ret- CM . Taken together , the results for sickle cell trait and blood group O suggest that some Ret- CM cases occur through pathway ( b ) ( because sickle cell trait is underrepresented in Ret- CM ) and that malaria parasites contribute to the pathogenesis of these cases , and that sickle cell trait reduces the pathogenetic potential of malaria infection for Ret- CM but do not provide evidence that blood group O reduces the pathogenetic potential of malaria infection for Ret- CM . DOI: http://dx . doi . org/10 . 7554/eLife . 23699 . 004 Whether malaria parasites play a role in the pathogenesis of Ret- CM could in principle be tested by a randomized experiment . For example , Smith ( 2007 ) considered a hypothetical blood-stage malaria vaccine that reduces parasite density by 50%; the vaccine would reduce malaria illness but not the incidence of parasitemia . If such a vaccine existed , then a way to test whether malaria parasites are pathogenetic in Ret- CM would be to randomize a large number of children to either ( i ) the blood-stage vaccine or ( ii ) placebo . If malaria parasites are never pathogenetic in Ret- CM , then we would expect no difference in Ret- CM because the blood stage vaccine would fail to prevent the cause of the development of the Ret- CM whereas if malaria parasites are sometimes pathogenetic in Ret- CM in a way that requires the development of uncomplicated malaria illness , then the blood stage vaccine would prevent the development of Ret- CM in some cases . Such an experiment is not currently feasible because no blood-stage vaccine has reached a Phase III trial ( Miura , 2016 ) and even if an effective blood-stage vaccine was developed , the experiment would require a huge sample size to have power to detect a change in Ret- CM rates . Though a randomized experiment with a blood-stage malaria vaccine that would have power to detect a difference in Ret- CM rates is not currently feasible , nature provides traits that protect against malaria illness in a random way through genetic inheritance . The general approach of using genetic variation to construct natural experiments is called Mendelian randomization ( Smith and Ebrahim , 2003 ) . The sickle cell trait ( HbAS ) – inheritance of one abnormal allele of the betaglobin gene – protects against symptomatic malaria ( Modiano et al . , 2001; Taylor et al . , 2012; Williams et al . , 2005; Willcox et al . , 1983 ) . Thus , a person who inherits one abnormal allele of the betaglobin gene has an antimalarial biochemical protection provided by nature . A second inherited trait which affects susceptibility to malaria illness is blood group O ( BGO ) , which protects against CM compared to groups A , B or AB ( Cserti and Dzik , 2007; Malaria Genomic Epidemiology Network , 2014 ) . Analogous to the randomized trial described above , possession vs . lack of a malaria protective trait assigns children to an arm in which some malaria illness is prevented vs . not prevented . For possession vs . lack of a trait to be fully analogous to the blood stage vaccine randomized trial described above , possession of the trait cannot affect malaria parasitemia incidence just like assignment to the blood-stage vaccine arm in the randomized trial does not affect malaria parasitemia incidence . If the trait affects malaria parasitemia incidence , then it could decrease the Ret- CM rate not because it decreases coma but because the WHO definition of Ret- CM requires malaria parasitemia; the natural experiment induced by the trait would then be biased for assessing the effect of the trait on coma in the same way that a randomized trial is biased if the treatment could affect the measurement of the outcome ( WHO-defined Ret- CM ) without affecting the true outcome of interest ( coma without malarial retinopathy ) . For both BGO and HbAS , it is plausible that the traits do not protect against malaria parasitemia incidence as systematic reviews have not found consistent evidence for protection ( Uneke , 2007; Taylor et al . , 2012 ) ; we will assume no protection for our main analysis but do sensitivity analyses that allow for protection . Under the assumptions that a trait does not affect malaria parasitemia incidence and the trait is randomly assigned , then if the trait decreases the probability of developing both Ret+ and Ret- CM , this suggests that malaria parasites contribute to the pathogenesis of both conditions . If the trait decreases the probability of developing Ret+ CM but not Ret- CM , this would suggest that malaria parasites are pathogenetic for Ret+ CM but are either not pathogenetic for Ret-CM or the trait affects an aspect of disease not causal to the development of Ret- CM . Using data gathered from 1996 to 2007 in a study of CM pathogenesis in Blantyre , Malawi ( Taylor et al . , 2004; Seydel et al . , 2015 ) as well as the MalariaGEN consortium ( Malaria Genomic Epidemiology Network , 2008 ) , we compared children with CM to two types of controls – ( 1 ) community controls; ( 2 ) hospital controls , children who were admitted to the Paediatric Research Ward with a known non-malarial cause of illness – meningitis , non-malarial anemia or other non-malaria illness . Table 1 shows admission characteristics of the hospitalized participants . The Blantyre coma score is statistically significantly higher in Ret+ CM patients than Ret- CM patients , but we do not regard the difference as clinically significant . In general , the patients with Ret+ CM were more severely ill than those with Ret- CM ( higher lactate , more deep breathing and a higher chance of death ) . The malaria illness is more severe in Ret+ CM than Ret- CM patients ( lower platelet count and more anemia ) . The higher opening CSF pressures in Ret+ CM patients compared to Ret- CM patients suggests a higher proportion of Ret+CM patients have increased brain volume . Comparing CM cases to non-malaria controls , as expected , laboratory abnormalities associated with malaria infection ( e . g . low hematocrit and platelet count ) were more frequent in the CM cases . For each trait t ( t=HBAS or blood group O ( BGO ) ) , we test the null hypothesis ( H0t ) that the trait frequency is the same in controls and true Ret- CM cases vs . the alternative hypothesis ( Hat ) that the trait frequency is higher in controls than true Ret- CM cases . Here , true Ret- CM refers to Ret- CM measured without error; in the actual data , retinopathy status may be measured with error and this measurement error is taken into account in the inferences ( Materials and methods ) . Under a model in which the trait does not affect other potential contributors to Ret- CM besides malaria and does not affect which CM cases are admitted to the Paedeatric Research Ward ( as compared to dying before reaching the Ward or recovering before being referred to the Ward ) , the null hypothesis H0t implies that malaria parasitemia is an incidental finding in children with Ret- CM and/or the trait affects an aspect of disease not causal to development of Ret- CM , while the alternative hypothesis Hat implies that malaria parasitemia is necessary for some Ret- CM cases and the trait reduces the pathogenetic potential of malaria infection for Ret- CM . Note that if either H0HbAS or H0BGO is false , this implies that malaria parasitemia plays a pathogenetic role in at least some Ret- CM cases . We calculated HbAS and BGO proportions in study subjects and made inferences about odds ratios ( Table 2 ) and tested H0HbASand H0BGO . For both HbAS and BGO , non-malaria hospitalized controls did not differ from community controls ( HbAS p-value=0 . 86; BGO p-value=0 . 83 ) ; therefore , subsequent analyses combined the control groups . Controls had a higher proportion of HbAS than true Ret- CM patients ( odds ratio: 14 . 33 , 95% CI: 3 . 21 , 257 . 24 ) and true Ret+ CM patients ( odds ratio: 1223 . 22 , 95% CI: 9 . 87 , ∞ ) . For BGO , the controls were comparable to true Ret- CM patients ( odds ratio: 1 . 03 , 95% CI: 0 . 83 , 1 . 29 ) but higher than true Ret+ CM patients ( odds ratio: 1 . 23 , 95% CI: 1 . 01 , 1 . 50 ) . There is strong evidence to reject H0HbAS ( p-value<0 . 0001 ) but not H0BGO ( p-value=0 . 79 ) ; these results are insensitive to different plausible assumptions about the false discovery rate and false omission rate for malarial retinopathy ( Appendix 1 ) . Taken together , these tests suggest that malaria parasitemia is pathogenic for a proportion of Ret- CM cases . Sickle cell trait protects against Ret- CM , but blood group O does not . 10 . 7554/eLife . 23699 . 005Table 2 . The top panel displays sickle cell trait ( HbAS ) proportions in retinopathy-positive ( Ret+ ) cerebral malaria ( CM ) , retinopathy-negative ( Ret- ) CM and control groups . The bottom panel displays ABO blood group gene proportions in Ret+ CM , Ret- CM and control groups . The last two rows of each panel display the odds ratios comparing controls to true Ret+ and true Ret- CM groups , which account for the fact that there is measurement error in observed retinopathy status ( false discovery rate = 0 . 07 and false omission rate = 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23699 . 005Ret+ CMRet- CMNon-malaria hospital controlsCommunity controlsSample size4382871923657HbAS*018175HbAA4372861843482Proportion of HbAS0 . 003 . 042 . 048Odds ratio ( 95% CI ) Non-malaria hospital controls vs . community controls0 . 87 ( 0 . 36 , 1 . 78 ) Controls vs . true Ret- CM14 . 33 ( 3 . 21 , 257 . 24 ) Controls vs . true Ret+ CM1223 . 22 ( 9 . 87 , ∞ ) Ret+ CMRet- CMNon-malaria hospital controlsCommunity controlsSample size4332861993543Blood Group O175135961739Blood Group A , B or AB2581511031804Proportion of Blood Group O . 404 . 472 . 482 . 491Odds ratio ( 95% CI ) Non-malaria hospital controls vs . community controls0 . 97 ( 0 . 72 , 1 . 30 ) Controls vs . true Ret- CM1 . 03 ( 0 . 83 , 1 . 29 ) Controls vs . true Ret+ CM1 . 23 ( 1 . 01 , 1 . 50 ) * HbAS ( sickle cell trait ) means that that the person has one normal and one abnormal copy of the hemoglobin beta gene . HbAA means the person has two normal copies of the hemoglobin beta gene . In Materials and methods , we formulate a sufficient-component cause model ( Rothman , 1976 ) based on Figure 1 and describe how to make inferences about the fraction of Ret- CM cases that are due to pathway ( b ) in Figure 1 , i . e . , the malaria parasitemia attributable fraction of Ret- CM ( the fraction of Ret- CM cases that would be prevented if malaria parasitemia were eliminated [Benichou et al . , 1998] ) . The fraction itself cannot be estimated without strong biological assumptions ( Greenland and Robins , 1988 ) , but a lower bound can be estimated under plausible assumptions . Table 3 shows inferences for this lower bound under the main model that assumes the traits do not protect against malaria parasitemia incidence and sensitivity analyses . Under the main model , the lower bound is estimated to be . 93 with 95% confidence interval ( . 68 , 1 ) . For the sensitivity analyses , although the systematic review of Taylor et al . ( 2012 ) found no consistent evidence that HbAS reduces malaria parasitemia incidence , some studies reviewed found protection and we consider sensitivity analyses that allow for a small amount of protection ( 10% ) and the largest amount of protection found in all the studies reviewed ( 41% ) ( Ntoumi et al . , 1997 ) . Also , the sensitivity analyses vary the false discovery rate and false omission rate between 0 and the upper bound estimated in Materials and methods . Under all scenarios considered , we found evidence for a substantial contribution of malaria parasites to the pathogenesis of Ret- CM with lower 95% confidence bounds ranging from . 37 to . 77 and point estimates for the lower bound ranging from . 86 to . 95 . 10 . 7554/eLife . 23699 . 006Table 3 . Inferences for lower bound on malaria parasitemia attributable fraction of Ret- CM ( fraction of Ret- CM cases that would be prevented if malaria parasitemia were to be eliminated ) under the sufficient-component cause model based on Figure 1 presented in Materials and methods . Inferences under the main model and sensitivity analyses that vary the effect of HbAS on malaria parasitemia incidence rate , the false discovery rate ( FDR ) and the false omission rate ( FOR ) for malarial retionopathy . DOI: http://dx . doi . org/10 . 7554/eLife . 23699 . 006Effect of HbAS on malaria parasitemia incidence rateFDRFORLower bound on malaria parasitemia attributable fraction of Ret- CM Estimate ( 95% CI ) Main ModelNo Effect . 07 . 05 . 93 ( . 68 , 1 ) Sensitivity AnalysesReduce 10% . 07 . 05 . 92 ( . 64 , 1 ) Reduce 41% . 07 . 05 . 88 ( . 46 , 1 ) No Effect . 30 . 11 . 94 ( . 75 , 1 ) Reduce 10% . 30 . 11 . 94 ( . 72 , 1 ) Reduce 41% . 30 . 11 . 91 ( . 58 , 1 ) No Effect0 . 11 . 92 ( . 62 , 1 ) Reduce 10%0 . 11 . 91 ( . 58 , 1 ) Reduce 41%0 . 11 . 86 ( . 37 , 1 ) No Effect . 300 . 95 ( . 77 , 1 ) Reduce 10% . 300 . 94 ( . 74 , 1 ) Reduce 41% . 300 . 91 ( . 61 , 1 ) No Effect00 . 92 ( . 66 , 1 ) Reduce 10%00 . 92 ( . 63 , 1 ) Reduce 41%00 . 87 ( . 44 , 1 ) We have studied a natural experiment that alters the level of malaria illness and found evidence that children with genetic traits associated with resistance to malaria illness are underrepresented in admissions with both Ret+ and Ret- CM ( HbAS ) or in admissions with Ret+ CM only ( BGO ) . Figure 1 shows a model of how HbAS and BGO affect Ret- and Ret+ CM . HbAS protects children with malaria parasitemia from developing uncomplicated malaria illness ( e . g . , fever ) and severe malaria illness ( Taylor et al . , 2012 ) . By contrast , current evidence suggests that BGO has no effect on developing uncomplicated malaria illness ( Uneke , 2007 ) , but does inhibit the cytoadherence of parasitized red blood cells to endothelial cells in the microcirculation , e . g . , by affecting rosetting ( Rowe et al . , 2007 ) and physical properties of the red cell membrane ( Méndez et al . , 2012 ) , thereby preventing severe malaria illness ( Cserti and Dzik , 2007; Migot-Nabias et al . , 2000; Uneke , 2007 ) . Consistent with this current evidence , in Figure 1 , HbAS prevents Ret- CM through pathway ( b ) , which involves two components ( uncomplicated malaria illness + other illness ) , by preventing one of the components , uncomplicated malaria illness , whereas BGO does not protect against Ret- CM through pathway ( b ) because it does not prevent uncomplicated malaria illness ( Figure 1 ) . In Figure 1 pathway ( b ) , Ret- CM results from an interaction between malaria illness and other illness . While the interactions between malaria parasites and other pathogens are incompletely understood and not well investigated , there is evidence for interaction ( Obaro and Greenwood , 2011; Mallewa et al . , 2013; Postels and Birbeck , 2011 ) . In studies of sepsis and malaria , the effect of microvascular parasite sequestration on the integrity of the gut mucosa is thought to allow bacterial seeding into the blood stream and hence bacteremia ( Scott et al . , 2011 ) . Another example is that following antigenic challenge from a Plasmodium falciparum candidate vaccine , children coinfected with schistosomiasis had lower acquired specific immune responses than those not infected ( Diallo et al . , 2010 ) . Another way in which malaria parasitemia could play a causal role in Ret- CM besides pathway ( b ) in Figure 1 is that the parasitemia could be the sole cause of coma . Taylor et al . ( 2004 ) ’s autopsy study results suggest that among children dying from Ret- CM , malaria parasitemia is typically not the sole cause of death ( see Causes of Death in Autopsy study in Appendix 1 ) , but it is possible that in survivors from Ret- CM , malaria parasitemia is the major contributor to acute illness . Whether malarial retinopathy is present or absent in CM could be affected by factors such as the child’s genome , the parasite’s genome and the child’s previous exposures to malaria ( Postels and Birbeck , 2011 ) . The protection provided against Ret- CM by HbAS but not BGO could be explained by an interaction between the factor ( s ) affecting whether malarial retinopathy is present and BGO , e . g . , a parasite genotype which causes malarial retinopathy to be absent could interfere with the mechanism by which BGO provides protection against severe malaria . Our study has several limitations . We only considered the genetic variants of sickle cell trait and blood group because these were the only statistically significant ( p< . 05 ) protective variants in the Malawi sample on which malarial retinopathy was measured , but future work could further test the model in Figure 1 by looking at additional genetic variants that have been found to affect the risk of malaria in other sites in Africa ( Malaria Genomic Epidemiology Network , 2014 ) . Our study only addresses CM pathogenesis in children . Adult and pediatric CM have important clinical differences and our results may not be generalizable to adults with Ret- CM . Detection of the presence or absence of malarial retinopathy was determined by several ophthalmologists over the 11 year duration of data collection , the sensitivity of detection of retinal changes may have varied between practitioners . Malarial retinopathy was determined on the basis of ophthalmoscopy alone , but since the time of our study , techniques such as optical coherence tomography that may increase sensitivity have been developed ( Joshi et al . , 2017 ) . We measure malarial retinopathy at the time of admission , but patients are admitted at different points on the disease trajectory . Malarial retinopathy can change over time; it doesn’t resolve during the 2–4 days of hospitalization but can become worse , which is a poor prognostic sign . Although there are limitations in our study to the sensitivity with which malarial retinopathy is measured , even if a moderate number of Ret- CM cases should be Ret+ CM cases in Table 2 , e . g . , if the false omission rate is . 5 and the false discovery rate is 0 , there is still strong evidence that HbAS has a protective effect for Ret- CM ( p-value=0 . 007; odds ratio: 6 . 81 , 95% CI: ( 1 . 50 , ∞ ) ) . Our main analysis assumed that possession of HbAS or BGO does not affect malaria parasitemia incidence; however , sensitivity analyses showed that our results were not sensitive to plausible violations of this assumption . Our analysis assumed that HbAS and BGO do not have selection effects on which CM cases are admitted to the Paediatric Research Ward as opposed to dying before reaching the Paedieatric Research Ward or being cured before needing to be referred to the ward; another interpretation of our study is that instead of studying ‘cerebral malaria’ as such , we have studied , ‘cerebral malaria mild enough to make it to the ward but severe enough not to recover without being taken to the ward . ’ Some of the non-malaria hospital controls had malaria parasitemia; since children can sometimes develop severe malaria with relatively low levels of parasitemia , it is possible that malaria contributed to the illness in these control participants . Furthermore , older non-malaria hospital controls may have a selection bias for HbAS since they have survived to an older age ( Ackerman et al . , 2005 ) . Our results are robust to considering only the community controls ( Appendix 1—table 4 ) which do not suffer from these potential biases of the non-malaria hospital controls . In summary , we studied a natural experiment that alters the level of malaria illness experienced by children through genetic variation and found evidence that malaria parasitemia is on the causal pathway to a substantial proportion of Ret- CM cases . Our approach of using genetically inherited traits to study CM pathogenesis could be adapted to illuminate the pathogenesis of other diseases . The study sample includes children with WHO-defined CM from 1997 to 2007 who were admitted to the Paediatric Research Ward at Queen Elizabeth Central Hospital , a tertiary referral center and teaching hospital in Blantyre , Malawi . The WHO definition of CM requires coma , circulating Plasmodium falciparum parasites , and no other cause of coma evident either by history or physical examination . Patients were enrolled during the rainy season , the time of annual peak incidence of CM . All patients were treated with intravenous quinine as standard antimalarial therapy . Enrollment in the study required explicit written consent from the parent or guardian . There were 947 enrolled patients . Malarial retinopathy was assessed on 726 of these patients and we subsequently limited our analysis to these 726 patients . To assess retinal changes , direct and indirect ophthalmoscopy were performed by an ophthalmologist well versed in findings typical of malarial retinopathy . We consider two types of controls in the study . The first type are community controls – 3704 children intended to be representative of the populations to which the cases belonged . Community controls were cord blood samples from Queen Elizabeth Central Hospital , mostly from the newborn nursery . The second type of controls are 194 patients who were admitted during the study period to the Paediatric Research Ward at Queen Elizabeth Central Hospital with a non-malarial cause of illness – meningitis , non-malarial anemia or other non-malaria illness . Two sources of bias in case-control studies are hidden bias ( unmeasured confounding ) and selection bias ( Rosenbaum , 1987 , Rosenbaum , 2002 ) . Hidden bias occurs when there are unmeasured variable ( s ) that are associated with the exposure and the outcome . Selection bias occurs when control subjects are selected based on a variable that is affected by the exposure . Different control groups might be subject to different amounts of hidden bias and selection bias . If two such control groups have similar exposure rates , this provides evidence against hidden bias and selection bias ( Rosenbaum , 1987 ) . Further , if the case group has a different exposure rate than both control groups and the control groups have similar exposure rates , this provides stronger evidence that the exposure has a causal effect on the outcome , rather than the results being biased by hidden bias or selection bias , as compared to a study with a single control group ( Rosenbaum , 1987 ) . Community controls are less likely to suffer from selection bias than the hospital controls . Community controls were randomly selected from the same population as the cases . In contrast , hospital controls could suffer from selection bias if the genetic variant of interest was associated with a non-malaria illness -- this type of selection bias is known as Berkson’s bias ( Westreich , 2012; Berkson , 1946 ) . Although the hospital controls might suffer from more selection bias than the community controls , the hospital controls have the potential advantage that they would reduce hidden bias if there was population stratification such that the genetic variant was associated with an unmeasured population stratification feature ( e . g . , housing conditions ) that caused both malaria and non-malaria illness . Participants were genotyped using the Sequenom iPLEX MassARRAY platform ( Malaria Genomic Epidemiology Network , 2014 ) . The genotyping included 55 SNPs for which there were previously reported associations with severe malaria . Two of the SNPs were found to be statistically significantly ( p< . 05 ) associated with cerebral malaria in Malawi -- HBB rs334 , which encodes the sickle cell trait , and ABO rs8176719 , which encodes the blood type ( O vs . A , B or AB ) ; see Figure 1 in Malaria Genomic Epidemiology Network ( 2014 ) . These two SNPs were used in our analysis . For a given genetically inherited trait , we consider two competing hypotheses for testing whether malaria parasitemia plays a pathogenic role in Ret- CM . We assume a biological model under which the trait does not affect other illnesses that cause Ret- CM and under which the trait does not have a selection effect on which CM cases are admitted to the Paedieatric Research Ward vs . dying before reaching the Paediatric Research Ward or being cured before needing to be referred to the ward . We use data on 65 patients at Queen Elizabeth Central Hospital in Blantyre , Malawi who were each examined by two ophthalmologists ( Beare et al . , 2002 ) . Malarial retinopathy is diagnosed if any of the following six signs are present: retinal hemorrhages ( RH ) , macular whitening ( MW ) , foeval whitening ( FW ) , peripheral whitening ( PW ) , vessel changes ( VC ) and capillary whitening ( CW ) . RH , VC and CW require little observer judgement if they are seen , but they may not be seen , depending on the degree of pupillary dilation and the presence/absence of spontaneous eye movements; for these consider the specificity to be 1 ( Beare et al . , 2002 ) . Identifying MW , FW and PW requires more experience on the part of the observer; MW and FW are more reproducible , and PW is less so ( Beare et al . , 2002 ) . We first consider the false discovery rate . To estimate an upper bound on the false discovery rate , we assumed that if RH , VC or CW was detected by either ophthalmologist , there was true malarial retinopathy , but if MW , FW or PW was detected without RH , VC or CW , then it was a false discovery; this is likely an overestimate since in some cases where MW , FW or PW was detected without RH , VC or CW , there is likely true malarial retinopathy that was missed by RH , VC and CW . There were 39 patients diagnosed by ophthalmologist 1 with malarial retinopathy , 33 of whom had RH , VC or CW by at least one ophthalmologist and there were 36 patients diagnosed by ophthalmologist 2 with malarial retinopathy , 33 of whom had RH , VC or CW by at least one ophthalmologist . To find a conservative upper bound on the false discovery rate , we consider ophthalmologist 1 and found the 95% Wilson binomial confidence interval ( Wilson , 1927 ) based on 6 out of 39 false discoveries , resulting in an 95% confidence interval of ( 0 . 07 , 0 . 30 ) , so resulting in an upper bound of 0 . 30 for the false discovery rate . To find a point estimate for the false discovery rate , we assume that there was true malarial retinopathy if RH , VC or CW was detected by either ophthalmologist or if FW was detected by both ophthalmologists or if MW was detected by both ophthalmologists since FW and MW were found to be reproducible by Beare et al . , 2002 , and then we averaged the resulting point estimates for the false discovery rate for the two ophthalmologists ( ( 4/39 + 1/36 ) /2 ) , to obtain an estimate of . 07 . We next consider the false omission rate . To estimate an upper bound on the false omission rate , we estimate an upper bound on the false omission rate if we were to only use RH , VC and CW to diagnose malarial retinopathy . The actual false omission rate is likely to be at least as small because we also diagnose malarial retinopathy if there’s any of MW , FW or PW and we think that the majority of time when we find MW , FW or PW but not RH , VC and CW , there is true malarial retinopathy , whereas a false omission will be relatively rare . We assume that the false positive rate for RH , VC and CW is 0 and use a model similar to model 1 in Nedelman ( 1988 ) . Specifically , let ψ denote the prevalence of true malarial retinopathy and ζ the false negative probability that an ophthalmologist will fail to detect RH , VC or CW in a child with true malarial retinopathy . Assume that we have two independent ophthalmologists . Then , the probability that both ophthalmologists will detect at least one of RH , VC or CW is ψ ( 1−ζ ) 2 , the probability that one but not the other ophthalmologist will detect RH , VC or CW is 2ψζ ( 1−ζ ) and the probability that both ophthalmologists will detect none of RH , VC or CW is ψζ2+1−ψ . The false omission rate is FOR=ψζ1−ψ+ψζ . We found a point estimate of 0 . 05 for the false omission rate and a 95% confidence interval ( using the percentile bootstrap ) of ( 0 , . 11 ) . Thus we take . 11 as an estimated upper bound for the false omission rate and . 05 as a point estimate for the false omission rate , recognizing that it is likely to be an upwardly biased point estimate . We will test the null hypothesis H0t:λt . r−=λt . co ( the probability of the malaria resistance trait t is the same in true Ret- CM cases as controls ) versus the alternative hypothesis Hat:λt . r−<λt . co ( the probability of the malaria resistance trait t is lower in true Ret- CM cases than controls ) . To test these hypotheses , we will estimate the parameters under the null and alternative hypotheses by maximum likelihood and use the generalized likelihood ratio test with 1 degree of freedom ( Rice , 2007 ) . The maximum likelihood estimation takes into account the assumed false discovery rate and false omission rate for malarial retinopathy detection . We will form confidence intervals for the odds ratios of the trait among controls vs . true Ret- CM patients ( [{λt . co/ ( 1−λt . co ) }/{λt . r−/ ( 1−λt . r− ) }]and the odds ratio of the trait among controls vs . true Ret+ CM patients [{λt . co/ ( 1−λt . co ) }/{λt . r+/ ( 1−λt . r+ ) }] by inverting the generalized likelihood ratio test . We will use Fisher’s exact to test whether the odds ratio of the trait differs among the community controls vs . the non-malaria illness controls and construct a 95% confidence interval for this odds ratio . The malaria parasitemia attributable fraction for coma among Ret- CM cases is the fraction of coma that would be prevented if malaria parasitemia were to be eliminated among Ret- CM cases . We formulate a model using the sufficient-component cause framework ( Rothman , 1976 ) and estimate the malaria parasitemia attributable fraction for Ret- CM using this model . A sufficient cause for a disease is a set of conditions that inevitably produces the disease . We assume that Ret+ CM can be represented by one sufficient cause: malaria parasitemia + factors that lead the malaria parasitemia to develop into uncomplicated malaria illness ( e . g . , lack of immunity ) + factors that lead the uncomplicated malaria illness to further progress to complicated malaria illness with coma and malarial retinopathy ( e . g . , genetic complexity of the malaria infection that overwhelms a child’s ability to control the infection ) . We assume that Ret- CM can be represented by two sufficient causes: ( a ) malaria parasitemia + another illness that is sufficient , in and of itself , to produce coma without malarial retinopathy; ( b ) malaria parasitemia + factors that lead the malaria parasitemia to develop into uncomplicated malaria illness + second insult ( innate or acquired ) that combined with the uncomplicated malarial illness leads to a coma without malarial retinopathy but which would not in and of itself be sufficient to produce coma . These two sufficient causes ( a ) and ( b ) for Ret- CM correspond to pathways ( a ) and ( b ) to Ret- CM in Figure 1 . Let p be the proportion of Ret- CM from sufficient cause ( b ) , which is the malaria parasitemia attributable fraction for coma among Ret- CM cases . Let rt , p be the factor by which the trait t multiplies the risk of malaria parasitemia ( i . e . , relative risk of malaria parasitemia for individuals with trait t compared to individuals without trait t ) , rt , u be the factor by which the trait multiplies the risk of factors that lead the malaria parasitemia to develop into uncomplicated malaria illness conditional on having malaria parasitemia , rt , c be the factor by which the trait multiplies the risk of factors that lead uncomplicated malaria illness to further progress to complicated malaria illness with coma and malarial retinopathy , rt , a be the factor by which the trait multiplies the risk of another illness that is sufficient , in and of itself , to produce coma without malarial retinopathy in the presence of malaria parasitemia and rt , s be the factor by which the trait multiplies the risk of a second insult that combined with uncomplicated malaria illness leads to a coma without malaria retinopathy but which would not in and of itself be sufficient to produce coma . We assume that rt , u , rt , c≤1 , i . e . , that the trait has no effect or a beneficial effect in preventing uncomplicated and complicated malaria . Let λt be the proportion of trait t in the population . For a rare disease , this proportion is approximately the proportion of trait t among the controls , λt≈λt . co . Ret+ CM and Ret- CM are both rare diseases – the vast majority of malaria infections do not progress to cerebral malaria; subsequently we assume λt=λt . co . We then haveλt . r+=rt , prt , urt , cλt . cort , prt , urt , cλt . co+1−λt . coλt . r−= ( 1−p ) rt , prt , aλt . cort , prt , aλt . co+1−λt . co+prt , prt , urt , sλt . cort , prt , urt , sλt . co+1−λt . co For our main model , we make the following assumptions: ( i ) rt , p=1 -- the trait has no effect on malaria parasitemia; ( ii ) rt , a=1 -- the trait has no effect on illnesses that are sufficient in and of themselves to produce coma without malarial retinopathy in the presence of malaria parasitemia and ( iii ) rt , s=1 -- the trait has no effect on an insult that combined with uncomplicated malaria illness leads to a coma without malaria retinopathy but which would not in and of itself be sufficient to produce coma . We relax the assumption that rt , p=1 in sensitivity analyses . Under assumptions ( i ) - ( iii ) , ( 1 . 1 ) λt . r+=rt , urt , cλt . cort , urt , cλt . co+1−λt . coλt . r−= ( 1−p ) λt . co+prt , uλt . cort , uλt . co+1−λt . co The parameters λt . co , λt . r+ , λt . r− can be identified from the data and thus Equation ( 1 . 1 ) involve two equations in three unknowns ( p , rt , u , rt , c ) . The p is minimized ( and hence the malaria parasitemia attributable fraction for coma among Ret- CM cases , is minimized ) by letting rt , c=1 , specifically the minimizing p solves λt . r−= ( 1−p ) λt . co+pλt . r+ . Let plb , t represent this lower bound on p based on trait t , which is a function of ( λt . co , λt . r+ , λt . r− ) ; let plb , t' represent this lower bound on p based on another trait t' , which is a function of ( λt' . co , λt' . r+ , λt' . r− ) and let plb=min ( plb , t , plb , t' ) be the lower bound on p based on both traits , which is a function of ( λt . co , λt . r+ , λt . r− , λt' . co , λt' . r+ , λt' . r− ) . We estimate plb by maximum likelihood and form a 95% confidence interval by inverting the generalized likelihood ratio test for plb . We conduct sensitivity analyses that allow for rt , p<1 . Maintaining the assumptions ( ii ) rt , a=1 and ( iii ) rt , s=1 , we have that the lower bound on p solvesλt . r−= ( 1−p ) rt , pλt . cort , pλt . co+1−λt . co+pλt . r+ The supplemental file rcode_for_paper . R contains R ( R Development Core Team , 2016 ) code for replicating the analyses in our paper .

Malaria is a life-threatening disease caused by a parasite that is transferred between people by infected mosquitoes . Most infected individuals suffer flu-like symptoms , but in rare cases malaria can affect the brain , resulting in brain damage , coma or death . The World Health Organization defines a person as suffering from cerebral malaria if the person is in a coma , has malaria parasites in his or her blood , and has no known alternative cause of the coma . Patients suffering from cerebral malaria are categorized based on whether they have damage to the back of the eyes known as retinopathy . It had previously been found that children who died of “retinopathy-positive” cerebral malaria ( i . e . those who had retinopathy ) had malaria parasites stuck in small vessels in their brains , which likely caused the coma . By contrast , children who died of “retinopathy-negative” cerebral malaria lacked this parasitic condition , and often also had other infections that can cause a coma , such as meningitis or sepsis . Because hospitals in many of the areas most affected by malaria often lack the ability to identify what – other than malaria – caused a coma , it was not clear whether malaria parasites influence how retinopathy-negative cerebral malaria develops . People with certain genetic variants – such as those that underlie sickle cell disease – are protected against the symptoms of malaria infections , and so these variants should also protect against cerebral malaria cases caused by the parasites . Small et al . therefore looked through data that had been collected over several years from people who had been admitted to a hospital in Malawi for cerebral malaria . This revealed that the genetically inherited sickle cell trait is highly protective against retinopathy-negative ( as well as retinopathy-positive ) cerebral malaria . Therefore , malaria parasites do play a role in a substantial proportion of cases of retinopathy-negative cerebral malaria . Although Small et al . provide evidence that malaria parasites play a role in retinopathy-negative cerebral malaria , they may not be the only cause of the coma . In the future , the absence of retinopathy could be used as a sign to look for additional factors that contribute to the coma . Currently , all cerebral malaria patients are treated in the same way . Understanding how malaria parasites interact with other illnesses to produce a coma could lead to the development of targeted treatment plans for retinopathy-negative patients .

2017

Evidence from a natural experiment that malaria parasitemia is pathogenic in retinopathy-negative cerebral malaria

Candida albicans hyphae can reach enormous lengths , precluding their internalization by phagocytes . Nevertheless , macrophages engulf a portion of the hypha , generating incompletely sealed tubular phagosomes . These frustrated phagosomes are stabilized by a thick cuff of F-actin that polymerizes in response to non-canonical activation of integrins by fungal glycan . Despite their continuity , the surface and invaginating phagosomal membranes retain a strikingly distinct lipid composition . PtdIns ( 4 , 5 ) P2 is present at the plasmalemma but is not detectable in the phagosomal membrane , while PtdIns ( 3 ) P and PtdIns ( 3 , 4 , 5 ) P3 co-exist in the phagosomes yet are absent from the surface membrane . Moreover , endo-lysosomal proteins are present only in the phagosomal membrane . Fluorescence recovery after photobleaching revealed the presence of a diffusion barrier that maintains the identity of the open tubular phagosome separate from the plasmalemma . Formation of this barrier depends on Syk , Pyk2/Fak and formin-dependent actin assembly . Antimicrobial mechanisms can thereby be deployed , limiting the growth of the hyphae . Candida albicans is a commensal fungus that colonizes the epithelial surfaces of 30–70% of healthy individuals ( Perlroth et al . , 2007 ) . However , in immune-compromised individuals , C . albicans can cause invasive , life-threatening disease . The mortality rate for infected patients is 46–75% , with candidiasis classified as the fourth most common nosocomial bloodstream infection ( Brown et al . , 2012 ) . Invasive candidiasis is correlated with a switch of C . albicans from its yeast form to a hyphal form , a shift that can be induced in vitro by nutrient deprivation among other cues ( reviewed in Sudbery , 2011 ) . In vivo , C . albicans hyphae are capable of invading epithelium and endothelium; in addition C . albicans is capable of forming recalcitrant biofilms and inducing inflammation ( Sudbery , 2011 ) . These conditions activate host defense mechanisms for the control and clearance of C . albicans , mounted predominantly by phagocytic cells of the innate immune system . Phagocytes can effectively sense , internalize and kill invasive C . albicans . Accordingly , impairment of the phagocytic response , e . g . by elimination of macrophages and neutrophils , is associated with disseminated candidiasis ( reviewed in Netea et al . , 2015 ) . Phagocytic cells possess receptors that bind the C . albicans cell wall and trigger uptake of the fungus into a phagosome . The C . albicans cell wall is composed mostly ( 80–90% ) of polysaccharides , containing ≈ 60% β- ( 1 , 3 ) and - ( 1 , 6 ) glucans , and ≈ 40% O- and N-linked mannans ( Ruiz-Herrera et al . , 2006 ) . As such , the main non-opsonic phagocytic receptors for C . albicans are the C-type lectin family of receptors , including Dectin1 , the mannose receptor , and DC-SIGN ( reviewed in Hardison and Brown , 2012 ) . The phagosome typically matures rapidly after closure , evolving into an acidic , degradative and microbicidal compartment . Acquisition of antimicrobial properties by this compartment depends on its ability to accumulate and retain toxic compounds , including reactive oxygen species ( ROS ) . Superoxide produced by the NADPH oxidase undergoes dismutation into hydrogen peroxide in the acidic luminal environment generated by the V-ATPase , which additionally favors the catalytic activity of various hydrolases . Transporters such as NRAMP-1 , that antagonize microbial growth by depleting the phagosome of nutrients , also depend on phagosomal H+ for the extrusion of metal ions . Unlike most other microbes , C . albicans presents a distinct problem for phagocytes . The hyphal form of C . albicans can grow at a rate of 18 . 8 μm hr−1 ( GOW and Gooday , 1982 ) , quickly exceeding the size of the phagocytes themselves . The challenge is greatest for macrophages , which migrate to infection sites later than the polymorphonuclear cells , and thus encounter growing hyphae ( reviewed in Erwig and Gow , 2016 ) . Despite being remarkably plastic , macrophages have difficulty engulfing the much larger C . albicans hyphae , an impasse that no doubt contributes to the pathogenesis of candidiasis . The aim of the current study was to examine the dynamic and complex process of C . albicans phagocytosis by macrophages . We found that attempts to engulf large hyphae result in the formation of incomplete ( frustrated ) phagosomes , which nevertheless segregate a section of the hypha , preferentially exposing it to microbiostatic products . The mechanism and fungal components underlying the formation of the diffusion barrier established by the phagocyte when generating the frustrated phagosome was analyzed using a combination of imaging , pharmacological and genetic approaches . To optimize the phagocytosis of C . albicans , which has a cell wall rich in β-glucans ( Gow et al . , 2011 ) , we used RAW 264 . 7 macrophages stably expressing the Dectin1 receptor ( RAW-Dectin1; Esteban et al . , 2011 ) . Yeast or hyphal forms of C . albicans expressing BFP ( Candida-BFP; Strijbis et al . , 2013 ) were used as targets to facilitate their visualization . Under the conditions used to generate them , C . albicans hyphae were considerably longer ( >15 μm ) than the macrophages ( 8–10 μm in diameter ) . After 1 hr of co-incubation with the macrophages the yeast form was fully engulfed ( Figure 1A ) , while a significant number of hyphal C . albicans were only partially internalized ( 68 . 5% ± 4 . 5 , while 31 . 5% ± 4 . 6 were fully internalized; 1019 events from 12 independent experiments ) , which was verified using fluorescent concanavalin A to label exposed hyphae ( Figure 1B ) . This was similar to the frustrated engulfment of >20 μm C . albicans hyphae reported earlier ( Lewis et al . , 2012 ) . Transmission electron microscopy confirmed that most hyphae were only partially internalized ( Figure 1C ) and , in addition , revealed the existence around the neck of the frustrated phagosome of a low-contrast structure seemingly devoid of membrane-bound organelles ( Figure 1C , inset ) , previously interpreted by Strijbis et al . , 2013 as accumulated actin . Indeed , this region corresponded to an actin-rich cuff-like structure ( Figure 1D ) ; F-actin was so highly accumulated at the cuff that the remainder of the cellular actin could only be visualized when images were overexposed ( Figure 1D , inset ) . Note that the remainder ( i . e . the base ) of the frustrated phagocytic cup was virtually devoid of F-actin . 3D visualization verified the continuous accumulation of F-actin around the neck of the tubular phagosomes lining individual hyphae and its sharp delineation of the intracellular and extracellular portions of the fungus ( Figure 1E , F , G , H and Video 1 ) . This actin cuff was observed for RAW-Dectin1 cells engulfing C . albicans hyphae up to 100 μm in size ( data not shown ) , and occurred in 96 . 3% ± 1 . 9 of the partially internalized hyphae ( 674 events analyzed in 12 independent experiments ) . These data support published accounts of actin cuff-like structures seen during the phagocytosis of various filamentous targets ( García-Rodas et al . , 2011; Gerisch et al . , 2009; Heinsbroek et al . , 2009; Prashar et al . , 2013; Strijbis et al . , 2013 ) . The occurrence of frustrated phagocytosis with formation of a pronounced actin cuff was not unique to the RAW-Dectin1 cell line; similar features were seen when murine or human primary macrophages were confronted with C . albicans hyphae ( Figure 1—figure supplement 1A and B , respectively ) . The actin cuff was remarkably stable , lasting for at least 90 min without contracting ( Figure 1I ) . Nevertheless , the actin composing these structures undergoes measurable turnover ( treadmilling ) , since the cuffs underwent gradual disassembly when the cells were treated with latrunculin A , which scavenges actin monomers ( last two panels , Figure 1I ) . These long-lasting yet dynamic cuffs identify the frustrated phagocytic cups generated by macrophages attempting to eliminate C . albicans hyphae . We proceeded to probe the receptors whose signaling could potentiate the formation of the actin cuff . Because C-type lectin signaling contributes importantly to C . albicans phagocytosis ( de Turris et al . , 2015; Tafesse et al . , 2015; Xu et al . , 2009 ) , we analyzed whether Dectin1 accumulated in the membrane at sites where cuffs were evident . Remarkably , while Dectin1 was clearly concentrated in patches elsewhere along the frustrated phagocytic cup , it was poorly detectable by immunostaining near the actin cuff ( ratio cuff: cup 0 . 60 ± 0 . 04; n = 30 p<0 . 0001; Figure 2A and inset ) . The failure to detect accumulation of Dectin1 at these sites was not attributable to masking of the exofacial epitope , possibly resulting from tight apposition to the hyphae , because similar results were obtained when the receptors were tagged with emerald fluorescent protein and visualized directly in live cells ( ratio cuff: cup 0 . 56 ± 0 . 04; n = 15 , p<0 . 0001; Figure 2B and inset ) . In epithelial and endothelial cells , host E- or N-cadherin , respectively , have been reported to contribute to C . albicans internalization ( Moreno-Ruiz et al . , 2009 ) . This process involved the recruitment of α- and β-catenins and activation of the Arp2/3 pathway for actin nucleation . In agreement with these reports , we observed E-cadherin and β-catenin accumulation at sites of where C . albicans hyphae were being internalized by epithelial A431 cells , with particular accumulation at sites where actin polymerized ( Figure 2—figure supplement 1 ) . We considered whether a similar mechanism was responsible for the formation of actin cuffs by macrophages . However , neither E-cadherin nor β-catenin was detectable in RAW-Dectin1 cells or in primary human macrophages by immunoblotting ( Figure 2C ) or by immunofluorescence ( not illustrated ) . Under comparable conditions , robust signals were obtained when probing A431 cells ( Figure 2C ) . When expressed heterologously in macrophages E-cadherin-GFP was found to line the surface membrane , but was absent from the phagocytic cup ( Figure 2D ) , while β-catenin-GFP was largely soluble and did not accumulate at the cuff ( Figure 2E ) . Thus , E-cadherin and β-catenin are unlikely to mediate phagocytosis of C . albicans in macrophages . Nevertheless , low levels of expression of these proteins ( below the level of detection of our assays ) or other cadherins may have mediated the internalization . This possibility was assessed by treating the cells with EDTA , which chelates the Ca2+ known to be required for ligand binding by cadherins ( reviewed in Brasch et al . , 2012 ) . As shown in Figure 2F , omission of Ca2+ had no effect on actin cuff formation in C . albicans-infected RAW-Dectin1 cells . Actin can also be tethered to the phagocytic cup via integrins ( Freeman et al . , 2016 ) . Integrins can be directly or indirectly involved in the phagocytosis of opsonized particles , apoptotic cells and a variety of other targets ( reviewed in Dupuy and Caron , 2008 ) and link with actin filaments via talin and vinculin ( reviewed in Shattil et al . , 2010 ) . However , canonical integrin activation and ligand binding require divalent cations ( reviewed in Leitinger et al . , 2000 ) , and would therefore be inhibited by their chelation with EDTA . Moreover , actin cuffs formed normally in CALDAG-GEF1−/− macrophages ( Figure 3—figure supplement 1 ) , consistent with the notion that cuff formation was independent of canonical activation of integrins , which involves Rap1 ( reviewed in Hogg et al . , 2011 ) . There is , however , one atypical instance where integrin activation can occur in the absence of divalent cations . The α chain of the integrin complement receptor 3 ( CR3 , also referred to as Mac1 ) , is unique in that it contains a lectin-like domain ( LLD ) that binds carbohydrates in a divalent cation-independent manner ( Thornton et al . , 1996 ) . The LLD is separate from the I-domain –the conventional ligand-binding domain of integrins ( reviewed in Ross , 2002 ) – and , interestingly , binds fungal β-glucan ( Ross et al . , 1985; Vetvicka et al . , 1996 ) . We therefore proceeded to test whether CR3 , which consists of αM ( CD11b ) and β2 ( CD18 ) subunits , is present in the region of the actin cuff . As illustrated in Figure 3 , both CD11b and CD18 accumulated in the region of the actin cuff in RAW-Dectin1 cells that had partially internalized C . albicans hyphae ( CD11b ratio cuff: cup 4 . 75 ± 0 . 29; n = 30 , p<0 . 0001; CD18 ratio cuff: cup 4 . 79 ± 0 . 28; n = 30 , p<0 . 0001; Figure 3A , B and insets ) . Moreover , talin , vinculin and paxillin were also localized to the cuff ( Figure 3C , D and insets; Figure 3—figure supplement 1E and inset ) , as was HS1 , the homologue of cortactin in leukocytes ( Figure 3E and inset ) . Like cortactin , HS1 is thought to regulate actin nucleation and branching ( Daly , 2004 ) . The preceding findings support a model whereby ligation of β-glucan by the LLD causes outside-in activation of CR3 directly ( O'Brien et al . , 2012; Vetvicka et al . , 1996 ) , or in conjunction with Dectin1 signaling ( Huang et al . , 2015; Li et al . , 2011 ) , resulting in Arp2/3-dependent actin nucleation . This model was tested using the M1/70 antibody , which binds to CD11b between its β-propeller and thigh domains ( residues 614–682; Osicka et al . , 2015 ) and effectively blocks the binding of CR3 to β-glucan ( Xia et al . , 1999 ) . Cells pretreated with M1/70 failed to show accumulation of CR3 around partially internalized C . albicans hyphae , and their ability to form actin cuffs was markedly impaired ( Figure 3H ) ; actin cuffs were much less prominent or missing altogether when CR3 was blocked ( Figure 3F versus G ) . The number of fully internalized C . albicans did not differ between conditions ( Figure 3H ) . We concluded that binding of the CR3 integrin to C . albicans was critical for the establishment of long-enduring actin cuffs observed during frustrated phagocytosis of the hyphae . Dectin1 and CR3 both bind β-glucans ( Brown and Gordon , 2001; Brown et al . , 2002; Ross et al . , 1985; Vetvicka et al . , 1996 ) , and have been reported to cooperate during phagocyte responses to fungal pathogens ( Huang et al . , 2015; Li et al . , 2011 ) . Dectin1 has also been reported to cooperate with TLR2 , TLR4 ( Ferwerda et al . , 2008; Netea et al . , 2006; Netea et al . , 2002 ) and mannose receptors ( Astarie-Dequeker et al . , 1999; Bain et al . , 2014; Lewis et al . , 2012; McKenzie et al . , 2010; Netea et al . , 2006 ) in the recognition of C . albicans . We therefore sought to clarify the receptors and ligands involved in actin cuff formation . Untransfected RAW 264 . 7 cells express negligible levels of Dectin1 ( Brown et al . , 2003; Esteban et al . , 2011; Taylor et al . , 2004 ) , providing a means to assess the contribution of this receptor to actin cuff formation . As shown in Figure 4A , RAW 264 . 7 cells rarely formed actin cuffs compared to RAW-Dectin1 cells , suggesting that initial engagement of the hyphae by Dectin1 was essential . The requirement for Dectin1 in C . albicans phagocytosis ( Marakalala et al . , 2013; Taylor et al . , 2007 ) could be bypassed when the hyphae were serum-opsonized , enabling opsonin receptors to establish the initial contact with the fungus ( Figure 4A ) . Thus , while not accumulating in the region of the cuff , Dectin1 binding to the hyphae ( which is evident by its accumulation in the frustrated phagocytic cup; Figure 2A and B ) is required for the subsequent activation of F-actin polymerization by CR3 . We also studied cooperativity by using soluble ligands to competitively block defined receptors , and scoring the frequency of actin cuff formation ( Figure 4B ) . Soluble mannan , a ligand for mannose receptor , had no effect on actin cuff formation by RAW-Dectin1 cells . Accordingly , we did not find mannose receptors in the membrane lining the actin cuff ( data not shown ) . Laminarin , a soluble β-glucan ligand for Dectin1 ( Brown and Gordon , 2001; Brown et al . , 2002 ) impaired phagocytosis and actin cuff formation when present prior to and during phagocytosis , but not if added after the hyphae had adhered to the RAW-Dectin1 cells ( Figure 4B ) . These findings support the notion that Dectin1 , but not mannan receptors , cooperate with CR3 to generate the actin cuffs . C . albicans cell wall components include β- ( 1 , 3 ) -glucans , β- ( 1 , 6 ) glucans , O- and N-linked mannans and chitin ( Netea et al . , 2008; Ruiz-Herrera et al . , 2006 ) . These can contribute to the recognition of C . albicans by phagocytes ( reviewed in Netea et al . , 2008 ) , and potentially also to actin cuff formation . To clarify the contribution of individual wall components we used gene replacement and conditional expression ( GRACE ) strains ( Roemer et al . , 2003 ) with specific depletion targeting chitin , mannan , and β ( 1 , 6 ) -glucan biosynthetic pathways upon incubation with doxycycline ( Table 1; O'Meara et al . , 2015 ) . Repression of pathways involved in chitin , mannan and β ( 1 , 6 ) -glucan synthesis using doxycycline did not affect actin cuff formation ( Figure 4C , D and data not shown ) , implying that these components are dispensable . We next assessed the role of β ( 1 , 3 ) -glucan through pharmacological inhibition of Fks1 with caspofungin ( Douglas et al . , 1997 ) , as genetic depletion of Fks1 results in defects in hyphae formation ( Ben-Ami et al . , 2011 ) . Remarkably , the ability to form actin cuffs was greatly reduced in C . albicans grown and allowed to form hyphae in the presence of caspofungin ( Figure 4E and F ) . The inhibitory effect of caspofungin on actin cuff formation was dose-dependent ( Figure 4F ) , reaching ≈ 80% at 5 ng mL−1 caspofungin , a dose that reduced the β ( 1 , 3 ) -glucan content of the wall by 55 . 3% , as assessed by aniline blue staining . Actin cuff formation around caspofungin-treated hyphae could not be rescued by serum opsonization ( Figure 4—figure supplement 1 ) , suggesting that β ( 1 , 3 ) -glucan is the ligand that promotes actin cuff assembly via CR3 . Interestingly , Aspergillus fumigatus hyphae ( routinely exceeding 80 µm in length ) were also able to illicit actin cuff formation by RAW-Dectin1 cells ( Figure 4G ) . A . fumigatus hyphae , while displaying some unique cell wall components compared to C . albicans hyphae , also have cell wall-associated β ( 1 , 3 ) -glucan ( Erwig and Gow , 2016 ) . We concluded that ligation of fungal β ( 1 , 3 ) -glucan by CR3 is required for actin cuff formation during frustrated phagocytosis of long hyphae . Despite the paucity of Dectin1 and mannose receptors ( Figure 2A and B ) , phosphotyrosine was markedly concentrated at the cuff ( Figure 5A ) , possibly as a consequence of CR3 activation . While there is disagreement over the requirement of Src-family kinases ( SFKs ) for the interaction of phagocytes with fungal targets ( Elsori et al . , 2011; Herre et al . , 2004; Le Cabec et al . , 2002; Mansour et al . , 2013; Underhill et al . , 2005 ) , there is evidence that Syk , as well as Pyk2 and Fak , two related tyrosine kinases , participate in CR3-mediated phagocytosis ( Li et al . , 2006; Paone et al . , 2016; Zhao et al . , 2016 ) . The contribution of individual kinases to the tyrosine phosphorylation was explored next . Phosphorylated SFKs accumulated along the frustrated phagocytic cup ( Figure 5A ) where Dectin1 was also found ( Figure 2A and B ) , but were not particularly enriched in the region of the actin cuff ( ratio cuff: cup 0 . 98 ± 0 . 03; n = 17 , p=0 . 61 ) . SFK inhibition by PP2 following adherence of the hyphae to RAW-Dectin1 cells had no effect on actin cuff formation ( Figure 5E ) . In contrast , the phosphorylated ( active ) forms of Pyk2 and Fak were enriched solely at the actin cuff ( pPyk2 ratio cuff: cup 23 . 69 ± 1 . 20; n = 46 , p<0 . 0001 , pFak ratio cuff: cup 22 . 56 ± 1 . 01; n = 34 , p<0 . 0001; Figure 5C and D ) . Moreover , inhibition of Pyk2/Fak activity by PF573228 following adherence of the hyphae to the cells abolished actin cuff formation , with no effect on internalization ( Figure 5E ) . Also , as reported by Strijbis et al . , 2013 , we observed phosphorylation of Syk with accumulation at the actin cuff ( ratio cuff: cup 21 . 55 ± 1 . 75; n = 22 , p<0 . 0001; Figure 5—figure supplement 1 ) . As expected , inhibition of Syk by piceatannol after C . albicans adherence blocked actin cuff formation ( Figure 5E ) . These data provide evidence that , along with Syk , Pyk2/Fak play a role in the interaction between macrophages and C . albicans , and are important for actin cuff formation during frustrated phagocytosis of hyphae . Interestingly , the interaction of Pyk2 with β2 integrins activates Vav1 ( Gakidis et al . , 2004; Kamen et al . , 2011 ) , a GEF for Rho-family GTPases that is also essential for the phagocytosis and control of C . albicans by macrophages ( Strijbis et al . , 2013 ) . Accordingly , Rac1 and/or Cdc42 were seemingly involved in the marked polymerization of actin at the cuff . This was indicated by the recruitment of PAK ( PBD ) , a biosensor of the active ( GTP-bound ) form of these GTPases ( Benard et al . , 1999 ) , that accumulated at the cuff to levels ≥4 fold higher than along the cup . F-actin accumulation at the cuff was sensitive to the formin inhibitor SMI-FH2 , but not to the Arp2/3 inhibitor CK-666 ( Figure 5G ) . Together , these data suggest that activation of Syk and Pyk2/Fak by CR3 leads to activation of Rho-family GTPases , culminating in formin-mediated actin assembly , a process akin to focal adhesion formation ( reviewed in Vicente-Manzanares et al . , 2005 ) . Phospholipids undergo striking changes during the course of conventional phagocytosis . PtdIns ( 4 , 5 ) P2 that is normally found in the plasma membrane is converted to PtdIns ( 3 , 4 , 5 ) P3 at sites of receptor engagement , and is subsequently degraded by lipases and phosphatases , becoming undetectable in sealed phagosomes . PtdIns ( 3 , 4 , 5 ) P3 can be detected for up to a minute following sealing , but then disappears abruptly as PtdIns ( 3 ) P appears; the latter is detectable on early phagosomes for about 10–15 min ( reviewed in Levin et al . , 2015 ) . These drastic switches are thought to reflect and possibly dictate the identity and developmental stage of the maturing phagosome . It has been observed that the frustrated tubular phagosomes of heat-killed filamentous Legionella pneumophila are accompanied by a sharp separation of plasmalemmal and phagosomal phosphoinositide species ( Naufer et al . , 2018; Prashar et al . , 2013 ) . Additionally , atypical phosphoinositide dynamics can occur in sealed phagosomes containing filamenting C . albicans ( Heinsbroek et al . , 2009 ) or during CR3-mediated phagocytosis of opsonized targets ( Bohdanowicz et al . , 2010 ) . Therefore we analyzed the phosphoinositides in frustrated phagosomes of C . albicans hyphae . We used the genetically-encoded fluorescent biosensor PLCδ-PH-GFP to monitor the distribution of PtdIns ( 4 , 5 ) P2 . Remarkably , while PtdIns ( 4 , 5 ) P2 was present as expected in the surface membrane facing the extracellular milieu , it was undetectable in the invaginated section that constituted the frustrated phagosome ( Figure 6A ) . In stark contrast , PtdIns ( 3 , 4 , 5 ) P3 –which was visualized using AKT-PH-GFP– was found solely in the open phagosomal cup ( Figure 6B ) , where it co-existed with PtdIns ( 3 ) P , detected using the PX-GFP sensor ( Figure 6C ) . In addition to the localization of PtdIns ( 3 , 4 , 5 ) P3 in the cup reported in a previous collaborative study ( Strijbis et al . , 2013 ) , we detected additional enrichment of PtdIns ( 3 , 4 , 5 ) P3 in the actin cuff region ( ratio cuff: cup 1 . 39 ± 0 . 10; n = 30 , p=0 . 0006 ) . In contrast , PtdIns ( 3 ) P was comparatively excluded from the actin cuff ( ratio cuff: cup 0 . 823 ± 0 . 05; n = 30 , p=0 . 0025 ) . The segregation of these phosphoinositides persisted for the duration of our observations ( up to 90 min after frustrated phagosome formation; not illustrated ) . The sharp boundary between the PtdIns ( 4 , 5 ) P2-rich surface membrane and the tubular membrane endowed with PtdIns ( 3 , 4 , 5 ) P3 and PtdIns ( 3 ) P coincided with the location of the actin cuff , suggesting that the latter may function as a diffusion barrier . However , the restricted localization of the phosphoinositides may have resulted from the strategic positioning of synthetic ( i . e . kinases ) and degradative ( i . e . phosphatases or lipases ) enzymes . To more definitively assess the existence of a diffusion barrier , we analyzed the distribution and dynamics of molecules that do not undergo rapid metabolic transformation , including lipid-anchored and transmembrane proteins , which had been reported to segregate in frustrated phagosomes . As shown in Figure 6D , LC3 –a small protein covalently linked to PtdEth– was found in the frustrated phagosome ( Kanayama and Shinohara , 2016; Martinez et al . , 2015; Sprenkeler et al . , 2016; Tam et al . , 2016 ) , yet did not reach the surface membrane . Similarly , both wild-type Rab7 ( Figure 6E ) and constitutively-active Rab7 ( not illustrated ) are confined to the frustrated phagosomal tube and partially excluded from the actin cuff ( Rab7 ratio cuff: cup 0 . 68 ± 0 . 05; n = 30 , p<0 . 0001 ) , as was LAMP1 ( ratio cuff: cup 0 . 59 ± 0 . 03; n = 30 , p<0 . 0001; Figure 6F ) , a late-endosomal/lysosomal membrane-spanning glycoprotein . The exclusion from the actin cuff was better appreciated by 3D visualization of LAMP1 ( Figure 6G , H and Video 2 ) . Because metabolic conversion to other species could not account for the segregation of the latter probes to the invaginated section of the membrane , we considered it more likely that restricted diffusion accounted for the observations . It was nevertheless possible that molecules like LC3 , Rab7 or LAMP were inserted through fusion into the tubular part of the membrane , where they could conceivably remain immobile . To exclude this possibility , we assessed their mobility measuring fluorescence recovery after photobleaching ( FRAP ) . The constitutively-active form of Rab7 , Rab7 ( Q67L ) , was used for these experiments; because this variant is unable to exchange nucleotides , it does not associate stably with GDI and remains membrane associated ( Méresse et al . , 1995 ) , eliminating the confounding effects of fluorescence recovery from a cytosolic pool . Rapid recovery was observed following photobleaching of a ≈3 µm spot within the phagosomal cup . In four independent experiments , half-maximal recovery was attained after 3 . 3 s ( Figure 7B ) . Similar analyses were performed using GFP-tagged LAMP1 ( Figure 7A , B ) , which also recovered within seconds ( t1/2 = 7 . 9 sec ) . Between 75–80% of the fluorescence was recovered in both instances , implying that the majority of the Rab7 ( Q67L ) and LAMP1 molecules were mobile . The retention of Rab7 ( Q67L ) and LAMP1 in the cup for many minutes despite their ability to move laterally in the plane of the membrane implies that they are unable to cross the junction with the surface membrane . The existence of a diffusion barrier was confirmed by expressing the N-terminal domain of Lyn ( Lyn11 ) tagged with GFP . This region of the protein becomes myristoylated and palmitoylated , targeting it to the plasma membrane and , to a lesser extent , to early endosomes . Following frustrated phagocytosis of hyphae , Lyn11-GFP is found both at the membrane and in the phagosomal cup , where its density is lower , likely because of dilution caused by insertion of unlabeled endomembranes . We analyzed comparatively small phagosomes to enable photobleaching of Lyn11-GFP in the entire cup ( Figure 7C ) . Strikingly , the fluorescence of the cup failed to recover , despite the persistence of abundant Lyn11-GFP in the adjacent plasmalemma . In three independent experiments only 19% of the original fluorescence reappeared , possibly via fusion with Lyn11-GFP-containing early endosomes . Failure to recover was not attributable to immobility of Lyn11-GFP in the membrane , which displayed very fast and nearly complete recovery following photobleaching ( Figure 7D , E ) . These data confirm that the region of the actin cuff acts as a lateral diffusion barrier , separating the inner leaflet of the plasma membrane from that of the open phagocytic cup . It is noteworthy that while the barrier curtails the diffusion of lipids and proteins anchored to lipids on the inner leaflet of the membrane , exofacial lipids and lipid-associated proteins readily traverse the junction between the membrane and the tubular phagosome . This was demonstrated by incorporation of rhodamine-labeled PtdEth to the surface membrane following stabilization of the frustrated phagosome . The labeled lipid , which inserts into the outer leaflet of the plasmalemma , reached the entire membrane of the frustrated phagocytic cup within ≈5 min ( Figure 7—figure supplement 1A ) . Similarly , fluorescent cholera toxin B subunit , which binds to exofacial ganglioside GM1 , promptly entered the phagocytic cup ( Figure 7—figure supplement 1B ) . Thus , the actin-dependent diffusion barrier selectively restricted the mobility of components of the inner leaflet , including transmembrane proteins , while exofacial lipids remained able to traverse the junction . How is the diffusion barrier generated ? We speculated that the molecular crowding resulting from tight clustering of integrins and their ancillary proteins could restrict the diffusion of membrane-associated components across the cuff . To test this possibility , we investigated whether sufficient molecular crowding could be generated to exclude other membrane components from regions of integrin clustering . To this end , we used antibody-induced cross-linking , a strategy shown earlier to induce the formation of CR3 patches on the plasma membrane ( Fukushima et al . , 1996; Pavan et al . , 1992; Zhou et al . , 1993 ) . Whether exclusion could be induced by molecular crowding was assessed analyzing the distribution of CD2-CD45-GFP ( Figure 8B ) , a transmembrane protein having a short , 7 nm ectodomain ( Cordoba et al . , 2013 ) . As shown in Figure 8A , prior to cross-linking both CD2-CD45-GFP and CR3 were distributed diffusely throughout the membrane , overlapping extensively at the resolution of the confocal microscope . The CD2-CD45-GFP fluorescence intensity in CR3-positive regions compared to the average CD2-CD45-GFP fluorescence intensity of the entire plasma membrane averaged 0 . 69 ± 0 . 01 ( 585 CR3-positive regions in 20 cells from three different experiments ) . After antibody treatment , CR3 clustered into large , dense patches . Strikingly , CD2-CD45-GFP was largely ( 81% ) and significantly ( p>0 . 0001 ) excluded from such patches , where the fluorescence was only 0 . 13 ± 0 . 01 of the plasmalemmal average ( measured in 472 CR3 patches in 15 cells from three experiments ) . Importantly , the exclusion was not alleviated by treatment with latrunculin A , the fluorescence of the patches averaging 0 . 12 ± 0 . 01 of the plasmalemmal average ( measured in 445 patches in 18 cells from three experiments ) , implying that the actin cytoskeleton is not involved in the domain segregation . CD2-CD45-GFP exclusion did not differ between these two conditions ( p=0 . 66 ) . We concluded that integrins could be sufficiently clustered to exclude other membrane components . By forming a continuous and thick ring around the neck of the frustrated phagosome , the molecular crowding of clustered integrins could generate a diffusional barrier . While actin is not essential to constrain the diffusion across patches of antibody-aggregated integrins , it is nevertheless required to maintain the integrins clustered in response to the glucan during frustrated phagocytosis . As such , an intact actin cuff is required to establish and maintain the barrier to phosphoinositides or transmembrane proteins . This was validated in cells that had formed a stable frustrated phagosome around C . albicans hyphae and were then treated with latrunculin A , which was shown earlier ( Figure 1I ) to cause gradual disassembly of the cuff . PtdIns ( 4 , 5 ) P2 –which in untreated cells is excluded from the phagocytic cup ( Figures 6A and 9A ) – gained access to the entire cup when actin was disassembled by latrunculin ( Figure 9B ) . The PtdIns ( 4 , 5 ) P2 present in the cup , expressed relative to the plasmalemma , increased 4 . 88 times after latrunculin treatment ( Figure 9C ) . Conversely , LAMP1 –that is restricted to the cup in untreated cells ( Figures 6F and 9D ) – was able to reach the surface membrane following treatment with latrunculin ( Figure 9E ) . After latrunculin treatment , the ratio of LAMP1 present in the cup decreased 3 . 99 times ( Figure 9F ) . Clearly , while clustering of CR3 is sufficient to form a diffusional barrier ( Figure 8 ) , the actin cuff formed during phagocytosis of C . albicans hyphae likely contributes to the stability of the barrier between CR3 and C . albicans β ( 1 , 3 ) -glucans , presumably by maintaining integrins in their active conformation ( Kaizuka et al . , 2007; Lavi et al . , 2007; Lavi et al . , 2012 ) during frustrated phagocytosis . Despite remaining unsealed , frustrated phagosomes acquired markers of endosomes and lysosomes , implying that they had undergone at least partial maturation . It was therefore conceivable that the cells established the diffusion barrier in an effort to generate a microbicidal compartment , despite their inability to form a sealed vacuole . Acidification of the lumen , secretion of antimicrobial enzymes and peptides and deployment of the NADPH oxidase are among the principal mechanisms used by leukocytes to eliminate pathogens . We first tested the ability of frustrated phagosomes to generate and maintain an acidic lumen , using the fluorescent acidotropic dye LysoBrite Red dye . As expected , the dye accumulated in lysosomes; however , it was never found to concentrate inside the frustrated phagosome ( Figure 10A ) , suggesting that vacuolar ATPases are not functional on its membrane and/or that the junction separating the lumen from the extracellular milieu is permeable to H+ . That the latter interpretation is correct was suggested by determinations of permeability of the junction using dextrans of varying size . For these experiments lysosomes were loaded with either 10 kDa or 70 kDa fluorescent dextran and then exposed to C . albicans hyphae . The dextrans were delivered into fully formed ( sealed ) phagosomes , where they were clearly retained ( Figure 10—figure supplement 1A ) . The smaller ( 10 kDa ) dextran , however , was not detectable inside frustrated phagosomes; the reduced overall staining of the cells ( cf . main panel and inset in Figure 10B ) suggests that secretion of lysosomes did occur , but that the dextran must have escaped the confines of the frustrated phagosome . In contrast , the 70 kDa dextran was readily visible along the frustrated phagosome , implying that its diffusion into the external medium was limited . Thus , a size-selective filter determined the extent to which solutes were retained within the frustrated phagosome . The cut-off of this filter must be greater than ≈50 kDa , because cathepsin D , a globular protein of ≈28 kDa , managed to escape the frustrated phagosome ( Figure 10D ) , yet was routinely detected in sealed phagosomes ( Figure 10—figure supplement 1B ) . Therefore , the incomplete phagocytic cup formed around partially internalized hyphae would be expected to have limited degradative capacity towards C . albicans . These data are in accord with the findings of ( Prashar et al . , 2013 ) that showed frustrated L . pneumophila phagosomes to retain large molecular weight dextrans , but not protons or lysosomal enzymes , despite acquisition of the V-ATPase and fusion with lysosomes . Though unable to retain luminal macromolecules over extended periods of time , the partial barrier to diffusion at the mouth of the frustrated phagosome , together with the geometrical constraint posed by the length and narrowness of the luminal space , are expected to delay the exit of molecules secreted into the phagosome . Rapidly reacting molecules may therefore be able to exert microbicidal/microbiostatic effects under these circumstances . Such is the case of reactive oxygen species produced by the NADPH oxidase . Indeed , we were able to detect preferential deposition of formazan , a product of the reaction of superoxide with nitroblue tetrazolium ( NBT ) , inside frustrated phagosomes ( Figure 10E ) . Heat-killed or paraformaldehyde-killed C . albicans hyphae were utilized for these experiments , eliminating the need to account for superoxide production by live C . albicans ( see Materials and methods ) . Because the frustrated phagosome appeared to retain some antimicrobial function , we assessed the effect of the frustrated phagosome environment on the fate of partially internalized hyphae . We could not detect significant loss of viability of the partially internalized C . albicans , as assessed by propidium iodide staining . We reasoned that the antimicrobial effectors may not suffice to kill the fungus , yet their effects may manifest as an observable change in the rate of hyphal extension , which can average 0 . 31 µm min−1 on serum agar ( GOW and Gooday , 1982 ) . When measured in RPMI medium without serum ( wthout macrophages present ) C . albicans hyphae grew at a rate of 0 . 22 µm min−1 ±0 . 03 . Remarkably , the extension rate of partially internalized hyphae , which displayed an actin cuff , was significantly reduced ( 0 . 11 µm min−1 ±0 . 01 ) . This reduced growth rate was not different ( p=0 . 742 ) to that of fully internalized hyphae ( 0 . 108 µm min−1 ±0 . 011 ) . In the same experiments , neighboring C . albicans hyphae not in contact with macrophages grew at a rate of 0 . 198 µm min−1 ±0 . 014 , indistinguishable from that measured in the absence of macrophages . Therefore , while small molecular weight contents can eventually diffuse out of the frustrated phagosome , they are nevertheless retained sufficiently to limit the growth of partially internalized C . albicans . This microbiostatic effect on partially internalized C . albicans hyphae could be ablated by blocking macrophage CR3 with the M1/70 antibody ( see Figure 3F , G and H ) before phagocytosis ( Figure 10—figure supplement 2 ) , reiterating the importance of CR3 ligation to β ( 1 , 3 ) -glucan for the generation and maintenance of this atypical phagocytic environment . Most of the microbicidal and degradative properties of the phagosome depend on the release and containment of lysosomal hydrolases , antimicrobial peptides and reactive oxygen species in close proximity to the internalized microorganism . However , when phagocytes are faced with exceptionally large targets , their internalization can become retarded or frustrated altogether . The inability to complete phagocytosis , as in the case of long asbestos fibers ( Donaldson et al . , 2010 ) or bacterial biofilms ( Costerton et al . , 1999; Scherr et al . , 2014; Thurlow et al . , 2011 ) can potentiate harmful inflammation . C . albicans hyphae can attain lengths of ≥50 μm ( GOW and Gooday , 1982 ) , overwhelming the comparatively diminutive phagocytes that are unable to ingest them whole . Accordingly , attempts to internalize such hyphae are frustrated and inflammatory in nature ( Branzk et al . , 2014; Goodridge et al . , 2011; Lewis et al . , 2012; Rosas et al . , 2008 ) . Nevertheless , our study shows that macrophages endeavor to seal the frustrated phagocytic compartment , in an effort to maximize their antimicrobial effect and minimize the release of inflammatory agents . To this end , they generate de novo a strikingly effective diffusion barrier by a process that involves activation of integrins that induce the formation of a thick F-actin cuff at the neck of the tubular phagosomes . The formation of actin-rich structures was reported previously during infection of macrophages with C . albicans ( García-Rodas et al . , 2011; Heinsbroek et al . , 2009; Strijbis et al . , 2013 ) and other rod/filament shaped microbes ( Gerisch et al . , 2009; Prashar et al . , 2013 ) , but neither their mechanism of assembly nor their functional significance were fully understood , which motivated our studies . As expected , Dectin1 –the major phagocytic receptor for fungal β-glucan ( Brown and Gordon , 2001; Brown et al . , 2002; Taylor et al . , 2007 ) – was present along the phagocytic cup , lining the internalized portion of the hyphae . Indeed , the RAW-Dectin1 cell line used for some of our studies was created to allow efficient internalization of fungal zymosan ( Esteban et al . , 2011 ) , and has been used to study C . albicans-macrophage interactions ( Strijbis et al . , 2013 ) . Dectin1 signaling leads to robust production of reactive oxygen species by the NADPH oxidase in response to fungal ligands ( Brown et al . , 2002; Goodridge et al . , 2011; Underhill et al . , 2005 ) , accounting for our observation that superoxide is detected within the frustrated phagocytic cup . Remarkably , however , Dectin1 did not concentrate in the region of the membrane adjacent to the actin cuff where the diffusion barrier was established . Instead , integrin αMβ2 ( CR3 , or CD11b/CD18 ) was found to accumulate in this region . Engagement of CR3 at the cuff is required for the formation of the underlying actin cuff , a process likely mediated by talin and vinculin , which were also found accumulated at the site . The entire assembly appears central to the establishment of the diffusion barrier , which is lost when blocking CR3 binding with the M1/70 antibody and also when latrunculin is used to disassemble actin filaments . CR3 is unique amongst integrins in that its α domain contains a lectin-like domain ( LLD ) capable of binding fungal β-glucan ( Ross et al . , 1985; Vetvicka et al . , 1996 ) . This LLD , located at the membrane-proximal C terminus ( between residues 943–1047; Lu et al . , 1998 ) is distinct from the traditional ligand-binding I domain . Importantly , the LLD can bind β-glucan in a Ca2+-independent manner while the integrin is in the inactive , bent conformation ( Thornton et al . , 1996 ) . Binding of glucan has been shown to induce a semi-active conformation of the integrin that is predicted to facilitate outside-in signaling ( O'Brien et al . , 2012; Vetvicka et al . , 1996 ) . We found that Dectin1 did not have a direct role in actin cuff formation , being instead required for adhesion and the initiation of phagocytosis . Interestingly , when Dectin1 expression is low , the deposition of opsonins contained in serum –including complement and possibly also anti-Candida antibodies– suffice to engage CR3 and promote actin cuff formation directly activating this process , as predicted from earlier observations ( Boxx et al . , 2010; Kozel et al . , 1987; Vetvicka et al . , 1996 ) . Dectin1 signaling can initiate inside-out activation of CR3 ( Li et al . , 2011 ) . However , conventional Rap1-dependent inside-out signaling mediated by CalDAG-GEF1 was dispensable for actin cuff formation , as were divalent cations . Therefore , it is likely that CR3 binds fungal β-glucan in a manner that does not require inside-out activation by Dectin1 . We showed that mannan , a ligand for the mannose receptor , had no effect on actin cuff formation . The utilization of a curated set of C . albicans GRACE strains confirmed that mannan was dispensable and further excluded chitin and β ( 1 , 6 ) -glucan as ligands for cuff formation . Importantly , caspofungin inhibition of β ( 1 , 3 ) -glucan synthesis blocked formation of the cuff , in a dose dependent manner . These observations are in accord with involvement of the LLD of CR3 , which was previously demonstrated to ligate β-glucan ( Mueller et al . , 2000; Thornton et al . , 1996; Vetvicka et al . , 1996 ) . Clearly , β-glucan is also a ligand for Dectin 1 ( Brown and Gordon , 2001; Brown et al . , 2002; Palma et al . , 2006 ) , and initial engagement of this receptor is required for formation of actin cuffs around unopsonized hyphae . However , β ( 1 , 3 ) -glucan appears to play a distinct role in CR3-mediated actin cuff formation , as the effects of caspofungin could not be rescued by serum opsonization , with the caveat that caspofungin treatment may have affected complement deposition on C . albicans ( Boxx et al . , 2010; Kozel et al . , 1987 ) , although we regard this as unlikely because the fungal cell wall is rich in other polysaccharides and proteins that can serve to attach complement . Our observations implicate clustered CR3 as an initiator of actin polymerization and a key constituent of the diffusion barrier . The signals mediating this effect include activation of Syk , which had been reported earlier ( Strijbis et al . , 2013 ) , and also of Pyk2 and Fak . The latter related kinases were enriched at the cuff and dual inhibition by PF573228 blocked cuff formation . Along with activated Syk , Pyk2 and Fak have been shown to interact with β2 integrins , including CR3 ( Duong and Rodan , 2000; Fernandez and Suchard , 1998; Han et al . , 2010; Hildebrand et al . , 1995; Kamen et al . , 2011; Mócsai et al . , 2002; Raab et al . , 2017; Rubel et al . , 2002; Wang et al . , 2010; Yan and Novak , 1999 ) . Pyk2 , in particular , is required for paxillin and Vav1 activation during integrin engagement and CR3-mediated phagocytosis ( Kamen et al . , 2011 ) . Paxillin was proposed to act as a scaffold , bridging integrin-initiated complexes with Rho-GTPases ( Deakin and Turner , 2008 ) , and Vav1 , previously identified as important for the phagocytosis of C . albicans ( Strijbis et al . , 2013 ) , also links β2 integrins to the activation of Cdc42 , Rac1 and RhoA ( Gakidis et al . , 2004 ) . These signaling events appear conserved during the frustrated phagocytosis of C . albicans hyphae , as we detected paxillin and active Rac1/Cdc42 at the actin cuff , and Vav1 was found to be enriched in actin cuff-like structures around C . albicans ( Strijbis et al . , 2013 ) . The activation of the Rho family GTPases is linked to both formin- and Arp2/3- mediated actin dynamics . The actin cuffs formed around C . albicans hyphae were singularly sensitive to SM1-FH2 –and therefore dependent on formin-mediated linear actin polymerization– and not to inhibitors of the Arp2/3 complex that promotes branched actin polymerization . Interestingly , Rac1 and Cdc42 interact with actin-nucleating formins of the mDia family ( Lammers et al . , 2008 ) . Collectively , our findings support a model whereby CR3 initiates signaling through Syk and Pyk2/Fak , leading to activation of Vav1 and Rho GTPases , culminating in formin-dependent actin nucleation . Our studies show that the integrin/actin cuff separates the open phagocytic cup from the plasma membrane , appearing to act as a boundary that segregates distinct and mobile membrane domains . There was a clear segregation of phosphoinositides between the plasma membrane ( PtdIns ( 4 , 5 ) P2 ) and the open cup ( PtdIns ( 3 , 4 , 5 ) P3 and PtdIns ( 3 ) P ) . In principle , such separation could stem from the differential and strategic localization of kinases and/or phosphatases in the two membranes and in the junctional complex . However , we also observed slowly convertible ( LC3 ) or non-convertible lipid-anchored proteins ( constitutively-active Rab7 ) and transmembrane proteins ( LAMP1 ) to be retained in the phagocytic cup , unable to reach the surface membrane . FRAP studies confirmed that these molecules were freely mobile within the phagocytic cup , pointing to the integrin/actin cuff as the diffusion barrier . While antibody-induced clustering of CR3 was sufficient to form an actin-independent diffusional barrier , actin was required to maintain the barrier function of the cuff during phagocytosis of hyphae . Actin can stabilize integrins in their active conformation ( Kaizuka et al . , 2007; Lavi et al . , 2007; Lavi et al . , 2012 ) , and such stabilization is likely required to maintain the cuff during extended frustrated phagocytosis . Actin-dependent diffusional barriers have been invoked in other systems ( Golebiewska et al . , 2011; Nakada et al . , 2003; Prashar et al . , 2013 ) , although the pickets that anchor the cytoskeletal fence and restrict the diffusion of membrane components had not been previously identified . We also analyzed the functional consequences of the establishment of the cuff and diffusion barrier . By segregating the two domains , the barrier enabled the open phagocytic cup to undergo an atypical maturation , despite the fact that scission from the surface membrane never occurred . This enabled targeting and activation within the frustrated phagosome of the NADPH oxidase , which generated toxic superoxide in the immediate vicinity of the portion of the hypha that had been engulfed . Thus , a crucial means for C . albicans control during infection ( Sasada and Johnston , 1980; Brothers et al . , 2013 ) remains operational in the frustrated phagosomes . The observation that the tubular phagosomes were rich in LAMP1 , a prototypic lysosomal marker , suggests that lysosomal hydrolases must have been secreted also into the phagosomal lumen . These were , however , not well retained by the phagosomes because , despite the tight seal that separated the inner leaflet of the membrane , the junction separating the aqueous compartments ( i . e . the lumen from the extracellular space ) was not perfectly tight . While 70 kDa dextran was retained within the phagosome , 10 kDa dextran was not , resembling the findings in frustrated L . pneumophila phagosomes ( Prashar et al . , 2013 ) and indicating the establishment of a sieve that excluded molecules with a hydrodynamic radius greater than ≈ 6–8 nm ( Nicholson and Tao , 1993 ) . Cathepsin family members ( radius ≈ 2 . 4 nm; Fazili and Qasim , 1986 ) and similarly-sized hydrolases would therefore eventually escape the lumen . Nevertheless , because a partial seal is formed , their rate of loss might be slowed , allowing hydrolases and other antimicrobial molecules to act on the partially internalized C . albicans hyphae before exiting the open cup . Fast-acting antimicrobial agents , like ROS , released in close proximity to the target would be expected to be at least partially effective . Consistent with this hypothesis , partially internalized C . albicans hyphae exhibited a reduced growth rate compared to external hyphae . Importantly , this growth restriction was abolished upon antibody blockade of CR3 and loss of actin cuff formation , presumably a result of increased leakage of phagosome contents . In this case , agents such as ROS , would not reach sufficient concentration to manifest the microbiostatic effect . However , leakage of phagosomal contents or ROS does not explain the failure of the frustrated phagosomes to kill the fungus , because C . albicans yeast and hyphae survive also within fully sealed phagosomes . Based on the preceding considerations , we hypothesize that the integrin/actin cuff is generated and maintained by macrophages as a means to sustain antimicrobial functions in open tubular phagosomes formed around C . albicans hyphae and possibly other targets . It is tempting to speculate that the unique conditions established by the diffusion barrier might provide additional benefits to the phagocyte . In dendritic cells , decreased phagosomal proteolysis associated with reduced phagosome acidification protects antigens for enhanced presentation ( Mantegazza et al . , 2008; Rybicka et al . , 2012; Savina et al . , 2006 ) . In addition , the frustrated yet maturing phagosome may enable activation of endomembrane Toll-like receptors ( TLRs ) . TLR3 and TLR9 both localize to intracellular compartments and recognize C . albicans nucleic acids and chitin , respectively , contributing to a protective cytokine response to the fungus ( Nahum et al . , 2011; Wagener et al . , 2014 ) . Interestingly , Dectin1 can collaborate with plasmalemmal TLRs ( TLR2 or TRL4 ) to enhance signaling and augment cytokine production ( Ferwerda et al . , 2008; Netea et al . , 2006; Underhill , 2007 ) and a similar synergy may apply to endomembrane TLRs . Indeed , Dectin1 recognition is required for TLR9 localization to the C . albicans phagosome and TLR9-dependent gene expression ( Khan et al . , 2016 ) . Thus , the unique structure described here may play an important role in the control of fungal infection and possibly also in the management of biofilms and other large targets by phagocytes . Mammalian expression vectors were obtained from the following sources: Emerald-Dectin1 ( plasmid #56291; Addgene , Cambridge , MA ) , PAK-PBD-YFP ( Srinivasan et al . , 2003 ) , E-cadherin-GFP ( plasmid #67937; Addgene ) , β-catenin-GFP ( plasmid #16071; Addgene ) , Talin-GFP ( Franco et al . , 2004 ) , AKT-PH-GFP ( Marshall et al . , 2001 ) , PLCδ-PH-GFP ( Botelho et al . , 2000 ) , PX-GFP ( Kanai et al . , 2001 ) , LC3-GFP ( Kabeya et al . , 2000 ) , Rab7-GFP ( Bucci et al . , 2000 ) , Rab7 ( Q67L ) -RFP ( D’'Costa et al . , 2015 ) , Lamp1-GFP ( Martinez et al . , 2000 ) , Lyn11-GFP ( Teruel et al . , 1999 ) , cathepsin D-RFP ( Yuseff et al . , 2011 ) , LifeAct-RFP or -GFP ( Riedl et al . , 2008 ) , F-tractin-GFP ( Belin et al . , 2014 ) , CD2-CD45-GFP ( Cordoba et al . , 2013 ) . Primary antibodies were purchased from the following vendors: HA ( catalogue #MMS-101P; Covance , Princeton , NJ ) , pTyrosine ( catalogue #05–321; EMD Millipore , Billerica , MA ) , pFAK-Y397 ( catalogue #3283S; Cell Signaling , Berverly , MA ) , pPYK2-Y402 ( catalogue #3291S; Cell Signaling ) , pSFK-Y418 ( catalogue #44660G; Invitrogen , Carlsbad , CA ) , pSYK-Y525/526 ( catalogue #2771S; Cell Signaling ) , Talin ( catalogue #T3287; Sigma-Aldrich , St . Louis , MO ) , Vinculin ( catalogue #MAB3574; EMD Millipore ) , HS1 ( catalogue #4557S; Cell Signaling ) , LAMP1 ( catalogue # 1D4B-s , Developmental Studies Hybridoma Bank , Iowa City , IA ) , actin ( catalogue #A4700; Sigma-Aldrich ) , CD11b ( catalogue #557394; BD Biosciences , Franklin Lakes , NJ ) , CD18 ( catalogue #557437; BD Biosciences ) , rat IgG2B isotype control ( catalogue #MAB0061; R and D systems , Minneapolis , MN ) , paxillin ( catalogue #P13520; Transduction Laboratories , Lexington , KY ) , GAPDH ( catalogue #MAB374; EMD Millipore ) , E-cadherin ( catalogue #610181; BD Biosciences ) , β-catenin ( catalogue #610153; BD Biosciences ) . Unconjugated and Alexa488 , Cy3 , Cy5 , HRP-conjugated secondary antibodies against mouse , goat , rat , rabbit IgGs were obtained from Jackson ImmunoResearch Labs ( West Grove , PA ) . A list of all C . albicans strains tested is provided in Table 1 . C . albicans strain SC5314 expressing BFP ( Candida-BFP; Strijbis et al . , 2013 ) was grown at 30°C in YPD ( BD Biosciences ) . C . albicans cell wall mutant strains were obtained from the GRACE collection of tetracycline-repressible mutant strains ( O'Meara et al . , 2015; Roemer et al . , 2003 ) . Depletion of target gene expression was achieved by adding 0 . 5 μg mL−1 doxycycline ( DOX ) to the growth medium . To induce hyphae of C . albicans , overnight cultures were subcultured 1:1000 in RPMI-1640 medium and incubated at 37°C for 1–3 hr , as indicated in the text . In some cases , caspofungin ( Sigma-Aldrich ) was used to pharmacologically inhibit β ( 1 , 3 ) -glucan synthesis . C . albicans overnight cultures were subcultured into RPMI-1640 containing 10 , 5 , 2 . 5 , 1 . 25 and 0 ng mL−1 caspofungin for 2 hr at 30°C . Cultures were then moved to 37°C for 1 hr to induce hyphae in the presence of caspofungin . To measure the effect of caspofungin on β ( 1 , 3 ) -glucan levels , C . albicans hyphae-infected wells were stained with 0 . 05% aniline blue ( EVANS et al . , 1984; Lee et al . , 2016 ) overnight . A . fumigatus strain AF293 ( clinical isolate ) was grown on YPD agar ( Bioshop , Burlington , ON ) plates at 30°C . Conidia were harvested in PBS containing 0 . 01% Tween-80 . For experiments , resuspended conidia were diluted 1:10 in RPMI-1640 containing 0 . 01% Tween-80 , and allowed to form hyphae at 30°C overnight . Hyphae were then washed twice with PBS 0 . 01% Tween-80 , and diluted 1:10 or 1:100 into RPMI-1640 containing 0 . 01% Tween-80 . The RAW 264 . 7 cell line was obtained from and authenticated by the American Type Culture Collection ( ATCC , Manassas , VA ) . The RAW-Dectin1-LPETG-3xHA cell line ( RAW-Dectin1 ) was provided by Dr . Karin Strijbis and authenticated for Dectin1-HA expression and Dectin1-mediated phagocytic ability by flow cytometry ( Esteban et al . , 2011 ) . Prior to experimentation , these cell lines were validated in our laboratory by assessing their morphology , phagocytic ability and expression of plasma membrane markers . RAW 264 . 7 and RAW-Dectin1 cells were grown in RPMI-1640 medium containing L-glutamine ( MultiCell , Wisent , St . Bruno , QC ) and 10% heat-inactivated fetal calf serum ( FCS; MultiCell , Wisent ) , at 37°C under 5% CO2 . The A431 cell line was obtained from and authenticated by the American Type Culture Collection ( ATCC ) . Prior to experimentation , this cell line was revalidated by assessing its expression of plasma membrane markers , and responsiveness to epidermal growth factor ( EGF ) . A431 cells were grown in DMEM medium containing L-glutamine ( MultiCell , Wisent ) and 10% heat-inactivated FCS , at 37°C under 5% CO2 . All cell lines tested negative for mycoplasma contamination by DAPI staining . Bone marrow-derived macrophages ( BMDM ) were obtained from the femoral bones of CALDAG-GEF1−/− ( Bergmeier et al . , 2007 ) or +/+ ( C57BL/6 ) mice , and differentiated for 5–7 days in DMEM containing L-glutamine , 10% heat-inactivated FCS , 100 U mL−1 penicillin , 100 μg mL−1 streptomycin , 250 ng mL−1 amphotericin B ( MultiCell , Wisent ) and 10 ng mL−1 mM-CSF ( PeproTech , Rocky Hill , NJ ) , at 37°C and 5% CO2 . To obtain M2 human monocyte-derived macrophages , peripheral blood mononuclear cells were isolated from the blood of healthy donors by density-gradient separation with Lympholyte-H ( Cedarlane , Burlington , ON ) . Human monocytes were then separated by adherence , and incubated in RPMI-1640 containing L-glutamine , 10% heat-inactivated FCS , 100 U mL−1 penicillin , 100 μg mL−1 streptomycin , 250 ng mL−1 amphotericin B and 25 ng mL−1 hM-CSF ( PeproTech ) for 7 days . Mammalian cell lines or primary cells were seeded on 18 mm coverslips in 12-well plates at 2 × 105 cells mL−1 . For infections with C . albicans , the medium was aspirated from the wells and replaced with 1 mL of C . albicans that had been induced to form hyphae . Plates were centrifuged for 1 min at 1500 rpm , then incubated with the following cell types at 37°C and 5% CO2 for phagocytosis to proceed: RAW-Dectin1: 1 hr; BMDM: 15 min; human M2 macrophages: 20 min; A431 cells: 3 hr . In some cases , 30 min prior to infection , C . albicans hyphae were opsonized in human serum to promote the deposition of complement , although deposition of donor-specific anti-Candida antibodies may have also occurred , further favoring phagocytosis . Where indicated , cells were treated with either vehicle or 1 μM Latrunculin A ( Sigma-Aldrich ) after 1 hr Candida-BFP infection , or pretreated 30 min with 4 mM EDTA ( Bioshop ) , followed by infection with Candida-BFP in the presence of EDTA . For inhibition of actin polymerization or kinases , after 30 min incubation with Candida-BFP , monolayers were treated with either vehicle , 50 μM CK-666 ( Calbiochem , La Jolla , CA ) , 10 μM SMI-FH2 ( Calbiochem ) , 10 μM PP2 ( Calbiochem ) , 50 μM piceatannol ( Sigma Aldrich ) , or 50 μM PF573228 ( Tocris , Oakville , ON ) , for 30 min . Following infection , monolayers were washed three times with PBS and fixed with 4% paraformaldehyde ( PFA ) . In some cases , wells were treated with various fluorescent reagents before or after phagocytosis , as described below . For infections with A . fumigatus , RAW-Dectin1 cells were incubated with 1 mL diluted A . fumigatus hyphae , and plates centrifuged for 5 min at 1500 rpm . Plates were incubated for 2 hr at 37°C and 5% CO2 for phagocytosis to proceed . RAW-Dectin1 cells were pretreated 10 min and infected in the presence of 100 mM L-cysteine ( Sigma-Aldrich ) to prevent gliotoxin-mediated inhibition of phagocytosis ( Schlam et al . , 2016 ) . Following infection , monolayers were washed three times with PBS and fixed with 4% paraformaldehyde ( PFA ) . In some cases , wells were treated with various fluorescent reagents before or after phagocytosis , as described below . After phagocytosis , external C . albicans were labeled for 20 min at room temperature using a solution of 5 μg mL−1 fluorescent conjugated concanavalin A ( ThermoFisher Scientific , Waltham , MA ) . To stain actin filaments , cells were permeabilized 5 min with 0 . 1% Triton X-100 and incubated 30 min with a 1:1000 dilution of fluorescent phalloidin ( Thermofisher Scientific ) or acti-stain ( Cytoskeleton , Inc . , Denver , CO ) . In some cases , C . albicans and A . fumigatus were stained with 10 μg mL−1 calcofluor white ( Fluorescent Brightener 28; Sigma-Aldrich ) . For transient transfection , RAW-Dectin1 cells were plated on 18 mm glass coverslips at a concentration of 2 × 105 cells mL−1 16–24 hr prior to experiments . FuGENE HD ( Promega , Madison , WI ) transfection reagent was used according to the manufacturer’s instructions . RAW-Dectin1 cells were transfected at a 3:1 ratio using 1 . 5 μL FuGENE HD and 0 . 5 μg DNA per well , and used for experiments 16 hr after transfection . In some cases , DNA transfections were performed using the Neon transfection system ( Life Technologies , Carlsband , CA ) according to the manufacturer’s protocol . RAW-Dectin1 cells were lifted , washed and resuspended to a concentration of 4 × 106 cells mL−1 and 100 μL of the suspension were mixed with 5 μg DNA . Electroporation was done using a single 20 ms pulse of 1750 V . Cells were then immediately transferred to RPMI-1640 containing L-glutamine and 10% heat-inactivated FCS , before seeding on coverslips at concentration of 2 × 105 cells mL−1 . Cells were used for experiments 16 hr after electroporation . After phagocytosis , fixation and concanavalin A staining ( as indicated ) , monolayers were permeabilized in PBS containing 0 . 1% Triton X-100 for 5 min and blocked in PBS containing 5% skim milk and 0 . 1% Triton X-100 for 30 min at room temperature . Samples were incubated with primary antibodies for 30 min at room temperature . Primary antibody dilutions were: HA ( 1:1000 ) , pTyrosine ( 1:100 ) , pFAK-Y397 ( 1:100 ) , pPYK2-Y402 ( 1:100 ) , pSRC-Y418 ( 1:100 ) , pSYK-Y525/526 ( 1:100 ) , talin ( 1:500 ) , vinculin ( 1:500 ) , HS1 ( 1:250 ) , LAMP1 ( 1:20 ) , actin ( 1:100 ) , E-cadherin ( 1:100 ) , β-catenin ( 1:100 ) , CD11b ( 1:100 ) , CD18 ( 1:100 ) , paxillin ( 1:100 ) . After rinsing with PBS , samples were incubated 30 min at room temperature with Alexa488 , Cy3 or Cy5-conjugated secondary antibodies at a 1:10 , 000 dilution . Where indicated , fluorescent phalloidin at a 1:1000 dilution was included with secondary antibodies . Samples were rinsed and viewed in PBS by confocal microscopy . Cells were grown in six well plates at a concentration of 4 × 105 cells per well . After infections , wells were lysed in Laemmli buffer ( Bio-Rad , Mississauga , ON ) . Samples were run on a 7% SDS-PAGE gel for separation and the gel was transferred to a polyvinylidene difluoride ( PVDF ) membrane . Membrane was blocked in PBS containing 5% skim milk and 0 . 05% Tween-20 for 30 min at room temperature , followed by primary antibody staining for 1 hr at room temperature , in blocking buffer . Primary antibodies dilutions: E-cadherin ( 1:10 , 000 ) , β-catenin ( 1:1000 ) , GAPDH ( 1:20 , 000; loading control ) . After washing membrane in PBS containing 0 . 05% Tween-20 , samples were incubated 30 min at room temperature with HRP-conjugated secondary antibodies at a 1:3000 dilution . Blots were visualized using the ECL Prime Western Blot detection reagent ( GE Healthcare , Mississauga , ON ) on an Odyssey Fc ( LI-COR , Lincoln , NE ) . 5 μg Rhodamine-PtdEth ( L-α-phosphatidylethanolamine-N- ( lissamine rhodamine B sulfonyl ) ammonium salt; Avanti Polar Lipids , Inc . , Alabaster , AL ) was dried under N2 and resuspended in 10 μL methanol . After vortexing , 900 μL 3 mg mL−1 bovine serum albumin ( BSA ) was added . This was then diluted 1:1 in cold serum-free RPMI-1640 containing 25 mM HEPES ( HPMI; MultiCell , Wisent ) . Following phagocytosis , the medium was aspirated and replaced with 500 μL of the prepared Rhodamine-PtdEth/HPMI , and incubated at 4°C for 10 min . The adherent cells were rinsed three times with cold HPMI , heated to 37°C 10 min and imaged live . Following phagocytosis , cells were rinsed three times with cold HPMI . Cholera toxin subunit B , Alexa488 conjugate ( Thermofisher Scientific ) was added to a final concentration of 1 μg mL−1 and cells incubated at 4°C for 10 min . The adherent cells were rinsed three times with cold HPMI , heated to 37°C 10 min and viewed live . Following phagocytosis , acidic intracellular compartments were stained using a 1:5000 dilution of the acidotropic LysoBrite Red dye ( AAT Bioquest , Sunnyvale , CA ) for 5 min at 37°C . Monolayers were rinsed three times with PBS , placed in HPMI , and imaged live . RAW-Dectin1 cells were pulsed overnight with 20 μg mL−1 Alexa647-conjugated 10 kDa dextran or 25 μg mL−1 tetramethylrhodamine-conjugated 70 kDa dextran . Cells were washed and incubated with Candida-BFP , as described above , for 1 hr . Following phagocytosis , monolayers were rinsed three times with PBS , placed in HPMI , and viewed live . This assay required PFA-inactivated Candida-BFP hyphae , as metabolically active C . albicans reduced nitroblue tetrazolium ( NBT ) to formazan , confounding the results . Candida-BFP hyphae were made as described above , followed by fixation in 8% PFA for 20 min . PFA-inactivated hyphae were rinsed and used to infect RAW-Dectin1 cells for 1 hr , in the presence of 0 . 5 mg mL−1 NBT ( Sigma-Aldrich ) . Following infection , cells were washed three times with PBS and fixed with 4% PFA . Formazan precipitate , created in response to superoxide anion produced by phagosomal NADPH oxidase , was visualized by bright field microscopy . To block CR3 , adherent RAW-Dectin1 cells were incubated with 10 μg blocking antibody to CD11b ( monoclonal M1/70 ) or rat IgG2B isotype control ( MAB0061; R and D systems ) for 30 min at room temperature . After warming to 37°C , cells were incubated with Candida-BFP hyphae as described above . Following phagocytosis , monolayers were rinsed and fixed in 3% PFA as mentioned above . After permeabilization and blocking , described above , blocking antibody was detected using a Alexa488-conjugated secondary antibody against rat IgG ( 1:10 , 000; Jackson ImmunoResearch Labs ) , and viewed by confocal microscopy . For sugar blocking experiments , adherent RAW-Dectin1 cells were pretreated 30 min with 100 μg mL−1 mannan ( Sigma Aldrich ) or laminarin ( Sigma Aldrich ) , followed by infection with C . albicans in the continued presence of each sugar . In the case of laminarin , Candida-BFP hyphae were allowed to adhere to RAW-Dectin1 cells 10 min , followed by the addition of 100 μg mL−1 laminarin for the remainder of the 1 hr infection . RAW-Dectin1 cells transiently transfected with CD2-CD45-GFP were washed with PBS and cooled to 10°C in 1X HBSS ( MultiCell , Wisent ) . To crosslink surface CR3 , cells were sequentially incubated with: ( 1 ) 5 μg mL−1 anti-CD18 primary ( monoclonal M18/2 ) ; ( 2 ) goat anti-rat secondary and ( 3 ) donkey anti-goat tertiary antibodies , for 30 min each at 4°C , in PBS containing Ca2+ , Mg2+ and 0 . 1% glucose . Labeled cells were then incubated 30 min at 37°C in the presence of 30 μM DYNGO 4a , to patch crosslinked CR3 without its internalization . Following this treatment , cells were treated 10 min with or without 1 μM latrunculin A , in the presence of 1 μM N-ethylmaleimide to prevent internalization of crosslinked CR3 following actin depolymerization . Monolayers were fixed in 3% PFA , and extracellular CR3 was labeled using a fluorescent anti-goat antibody . Plasma membrane clusters of CR3 were analyzed in Volocity software for exclusion of CD2-CD45-GFP . Exclusion was calculated from confocal slices as the ratio of the GFP intensity per pixel in the CR3-positive patches to the average intensity in the plasma membrane . RAW-Dectin1 cells were incubated with Candida-BFP hyphae for 10 min to allow adherence , then fixed in 4% PFA after 0 , 10 , 20 , 30 , 40 , 50 or 60 min . Following phagocytosis , the cells were fixed in 3% PFA as above , external Candida-BFP were labeled with fluorescent concanavalin A , and permeabilized and stained with fluorescent phalloidin , as described above . Samples were imaged by confocal microscopy , and the length of individual Candida-BFP hypha measured in μm . Hyphal extension rate was calculated by linear regression analysis in GraphPad Prism software ( GraphPad Software , Inc . , La Jolla , CA ) . Confocal images were acquired using a Yokogawa CSU10 spinning disk system ( Quorum Technologies Inc . , Guelph , ON ) or a Leica SP8 laser scanning system ( Leica ) . Images were acquired using a 63x/1 . 4 NA oil objectives or 25x/0 . 8 NA water objective ( ZEISS , Germany ) , as indicated , with an additional 1 . 5x magnifying lens . For live experiments , cells were maintained at 37°C using an environmental chamber ( Live Cell Instruments , Korea ) . Routine analyses and colocalizations were done using Volocity software ( Perkin Elmer , Woodbridge , ON ) . 3D data visualization was done using Imaris ( Bitplane , Concord , MA ) software . For colocalization analyses , Volocity software was used to calculate positive product of the differences of the mean ( Li et al . , 2004 ) channels , which were then overlayed on merged images for visualization . For fluorescent intensity calculations , background subtracted intensities per unit area for expressed fluorescent protein constructs or endogenous proteins ( immunofluorescence ) were measured in Volocity software . Ratios were calculated comparing relative intensities in the actin cuff compared to phagocytic cup , or phagocytic cup compared to membrane , as indicated in the text . FRAP experiments of GFP or RFP-tagged proteins , transiently expressed in RAW-Dectin1 cells , were conducted on an A1R point-scanning confocal system ( Nikon Instruments , Japan ) . For FRAP , Candida-BFP hyphae-infected cells were imaged in HPMI at 37°C . Images were acquired using a 60x/1 . 4 NA oil objective ( Nikon ) , 1 . 2-AU pinhole , resonant scanning mode , and 16x line averaging . For a complete 2 min FRAP acquisition at 1 . 9 fps , after 5 s of initial imaging , a region of interest 3 μm in diameter was bleached for 1 . 06 s using the 405 laser at 100% power , followed by imaging for fluorescence recovery . Images were exported and analyzed for fluorescence intensity using Volocity software . After background subtraction , fluorescence intensity units were normalized ( see Figure 6 legend ) using Microsoft Excel software , and transformed to a 0–1 scale , to correct for differences in bleaching depth and allow for comparison of up to 30 individual FRAP curves per condition . Graphpad Prism software was used to fit the FRAP curves to a single exponential , plotted as fractional recovery over time . Candida-BFP hyphae-infected RAW-Dectin1 cells were washed with cold PBS , and fixed with 2% glutaraldehyde in 0 . 1 M sodium cacodylate buffer , pH 7 . 3 . For improved specimen preparation , samples were then subjected to a zymolyase digestion protocol ( Bauer et al . , 2001 ) designed to weaken the fungal cell wall and allow for adequate structural preservation . Samples were then post-fixed in 1% osmium tetroxide in 0 . 1 M sodium cacodylate buffer , pH 7 . 3 , dehydrated in a graded ethanol series followed by propylene oxide , and embedded in Quetol-Spurr resin . Ninety nm sections were cut on a Leica Ultracut ultramicrotome and stained with uranyl acetate and lead citrate . Samples were imaged on a FEI Tecnai 20 transmission electron microscope , equipped with an AMT 16000 digital camera .

Billions of microorganisms live on , and in , the human body . Known as the human microbiome , most of these microscopic hitchhikers are harmless . But , for people with a compromised immune system , common species can sometimes cause disease . For example , the yeast Candida albicans , which colonises between 30 and 70% of the population , is normally harmless , but can switch to a disease-causing version that makes branching structures called hyphae . These hyphae grow fast , piercing and damaging the tissues around them . Immune cells called macrophages usually engulf invading microbes . These cells recognise sugars on the outside of C . albicans , and respond by wrapping their membranes around the yeast , drawing the microorganism in , and sealing it into closed structures called phagosomes . Then , the macrophages fill the phagosomes with acid , enzymes and destructive chemicals , which breaks the yeast down . Yet , C . albicans hyphae grow larger than macrophages , making them difficult to control . Maxson et al . have now tracked the immune response revealing how macrophages try to control large hyphae . The immune cells were quick to engulf C . albicans in its normal yeast form , but the response slowed down in the presence of hyphae . Electron microscopy revealed that the large structures were only partly taken in . Rather than form a closed phagosome , the macrophages made a cuff around the middle of the hypha , leaving the rest hanging out . The process starts with a receptor called CR3 , which detects sugars on the outside of the hyphae . CR3 is a type of integrin , a molecule that sends signals from the surface to the inside of the immune cell . A network of filaments called actin assemble around the hypha , squeezing the membrane tight . The macrophage then deploys free radicals and other damaging chemicals inside the closed space . The seal is not perfect , and some molecules do leak out , but the effect slows the growth of the yeast . When a phagosome cannot engulf an invading microbe , a state that is referred to as being “frustrated” , the leaking of damaging chemicals can harm healthy tissues and lead to inflammation and disease . These findings reveal that macrophages do at least try to form a complete seal before releasing their cocktail of chemicals . Understanding how the immune system handles this situation could open the way for new treatments for C . albicans infections , and possibly similar diseases related to “frustrated engulfment” ( such as asbestos exposure , where asbestos fibers are also too large to engulf ) . However , one next step will be to find out what happens to partly engulfed hyphae , and how this differs from the fate of fully engulfed yeast .

2018

Integrin-based diffusion barrier separates membrane domains enabling the formation of microbiostatic frustrated phagosomes

In social groups , infections have the potential to spread rapidly and cause disease outbreaks . Here , we show that in a social insect , the ant Lasius neglectus , the negative consequences of fungal infections ( Metarhizium brunneum ) can be mitigated by employing an efficient multicomponent behaviour , termed destructive disinfection , which prevents further spread of the disease through the colony . Ants specifically target infected pupae during the pathogen’s non-contagious incubation period , utilising chemical ‘sickness cues’ emitted by pupae . They then remove the pupal cocoon , perforate its cuticle and administer antimicrobial poison , which enters the body and prevents pathogen replication from the inside out . Like the immune system of a metazoan body that specifically targets and eliminates infected cells , ants destroy infected brood to stop the pathogen completing its lifecycle , thus protecting the rest of the colony . Hence , in an analogous fashion , the same principles of disease defence apply at different levels of biological organisation . Pathogen replication and transmission from infectious to susceptible hosts is key to the success of contagious diseases ( Schmid-Hempel , 2011 ) . Social animals are therefore expected to experience a greater risk of disease outbreaks than solitary species , because their higher number of within-group interactions will promote pathogen spread ( Nunn and Altizer , 2006; Schmid-Hempel , 2017; Alexander , 1974 ) . As a consequence , evolutionary immunology predicts that traits mitigating this cost , such as detecting sick conspecifics and using that information to prevent self-infection , should have been selected for in group-living animals as an essential adaptation to social life ( Hamilton , 1987; Ezenwa et al . , 2016 ) . Social animals , including lobsters , tadpoles , mice and mandrills , can use conspicuous disease-associated changes in the physical appearance , behaviour and chemical odour of conspecifics to identify sick group members ( Shakhar and Shakhar , 2015; Arakawa et al . , 2011; Lopes , 2014; Bozza , 2015 ) . Upon detection , healthy animals usually respond by interacting with sick conspecifics less or avoiding them completely ( Poirotte et al . , 2017; Kiesecker et al . , 1999; Behringer et al . , 2006; Anderson and Behringer , 2013 ) . In addition , they may prophylactically increase the expression of their immune defences in preparation for a potential immune challenge ( Hernández López et al . , 2017 ) , and a similar phenomenon is observed in animals and plants in response to chemicals released by wounded conspecifics ( Heil and Silva Bueno , 2007; Peuß et al . , 2015 ) . Because there are downsides to being socially excluded , sick animals may , under some circumstances , attempt to hide their illness ( Lopes , 2014; Lopes et al . , 2012 ) . However , signalling infection to others should be adaptive if ( i ) this elicits care from conspecifics that improves the chance of recovery ( Hart , 1990 ) or ( ii ) if the infection is otherwise likely to spread and infect kin ( Shakhar and Shakhar , 2015 ) . This is because individuals can gain indirect fitness by enhancing the propagation of shared genes , present in relatives , into the next generation ( Hamilton , 1964 ) . Hence , in a closely related social group , an animal that warns its relatives if it is sick is likely to have a greater inclusive fitness than an animal that does not , since fewer of its kin will fall sick and suffer reductions in fitness ( Shakhar and Shakhar , 2015 ) . Hence , in closely related social groups , there may be selection for both the detection of illnesses by healthy group members ( Curtis , 2014 ) and the advertisement of an animal’s disease status by sick individuals themselves ( Shakhar and Shakhar , 2015 ) . In the social insects ( termites and ants and the social bees and wasps ) , colonies are typically single families , comprised of a queen and her daughters , the workers ( Queller and Strassmann , 2003 ) . They typically have an irreversible reproductive division of labour , with the two castes being highly interdependent: the queens are morphologically specialised for reproduction and cannot survive without the assistance of the workers; conversely , the workers cannot reproduce , but gain fitness indirectly by raising the queen’s offspring ( Queller and Strassmann , 2003 ) . Consequently , social insect societies have become indivisible , reproductive units , where natural selection acts on the colony instead of its individual members ( Bourke , 2011; West et al . , 2015 ) . This has parallels to the evolution of complex multicellular organisms , that is metazoan bodies , where sterile somatic tissue and germ line cells form an indivisible reproducing body . Hence , social insect colonies are often termed ‘superorganisms’ and their emergence is considered a major evolutionary transition ( Bourke , 2011; West et al . , 2015; Wheeler , 1911; Boomsma and Gawne , 2017 ) . Since evolution favours the survival of the colony over its members , selection has resulted in a plethora of cooperative and altruistic traits that workers perform to protect the colony from harm ( Hamilton , 1987; Bourke , 2011; Cremer et al . , 2007; Shorter and Rueppell , 2012; Rosengaus et al . , 2011; Boomsma et al . , 2005; Cremer et al . , 2017 ) . In particular , social insects have evolved physiological and behavioural adaptations that limit the colony-level impact of infectious diseases , which could otherwise spread easily due to the intimate social interactions between colony members ( Cremer et al . , 2007; Rosengaus et al . , 2011; Boomsma et al . , 2005; Cremer et al . , 2017; Stroeymeyt et al . , 2014; Meunier , 2015 ) . These defences are performed collectively by the workers to form an emergent layer of protection known as social immunity that , like the immune system of a body , protects the colony from invading pathogens ( Cremer et al . , 2007; Rosengaus et al . , 2011; Cremer et al . , 2017; Cremer and Sixt , 2009 ) . Our understanding of how social immunity functions is based mostly on the behaviours social insects perform to prevent infection in contaminated colony members , referred to as sanitary care ( Cremer et al . , 2007; Rosengaus et al . , 2011; Tragust et al . , 2013a; Wilson-Rich et al . , 2009 ) . In ants , sanitary care involves grooming and the use of antimicrobial secretions to mechanically remove and chemically disinfect pathogens , reducing the likelihood that pathogen exposure leads to the development of an infection ( Tragust et al . , 2013a; Graystock and Hughes , 2011 ) . During sanitary care , protection of the colony is aligned with the protection of the individual colony member . In contrast , this is not the case if an individual is infected with a contagious disease , since they risk infecting the rest of the colony if the pathogen spreads ( Cremer et al . , 2017 ) . Hence , social immunity is characterised by a care-kill dichotomy , where colony members should be cared for when possible but sacrificed if necessary , both of which benefit the colony ( Cremer et al . , 2017; Cremer and Sixt , 2009 ) . Although the kill-component is an unique feature of social immunity that is not present in the disease defence repertoire of other forms of sociality ( e . g . non-superorganismal family groups , communal breeders or aggregations [Cremer et al . , 2017] ) , it is rarely studied in comparison to the care-component . Hence , what happens when sanitary care fails and a pathogen successfully infects a colony member , with the consequent potential to create an epidemic , remains poorly understood ( but see [Rothenbuhler , 1964; Ugelvig et al . , 2010; Tragust et al . , 2013b] ) . To effectively protect the colony , we predict that infections should be detected quickly and accurately by nestmates , in order to overcome the pathogen before it has time to replicate and produce propagules that can infect others . Once detected , the colony should respond by eliminating the infection as effectively as possible , by preventing any further opportunities for disease transmission . This is especially pertinent in the ants and termites , because their sedentary and territorial lifestyle makes it likely infectious corpses are encountered again , even if they are removed from the colony ( Boomsma et al . , 2005; Cremer et al . , 2017; Schmid-Hempel , 1998 ) . For example , previous studies have shown that fungus-infected ants become highly contagious to nestmates after death and can cause epidemics that result in colony collapse ( Hughes et al . , 2002; Loreto and Hughes , 2016 ) . To address the above gaps in our knowledge of social immunity , we investigated how ants detect and respond to infections . To that end , we exposed the immobile brood of the invasive garden ant , Lasius neglectus , to a generalist fungal pathogen , Metarhizium brunneum . When the infectious conidiospores of this fungus come into contact with insect cuticle , they attach , germinate and penetrate the host cuticle within 24–48 hr to cause internal infections ( Vestergaard et al . , 1999; Walker and Hughes , 2009 ) . During a short , non-infectious incubation period following infection , the fungus goes through a single-cell blastospore stage , which lasts 1–2 days . Once the host has died , the fungus enters a saprophytic mycelial phase , then grows out of the corpse , 1–3 days later , releasing millions of new infectious conidiospores , in a process called sporulation ( Hughes et al . , 2002; Deacon , 2006 ) . Previous work found that brood infected with Metarhizium is removed from the brood chamber ( Ugelvig et al . , 2010; Tragust et al . , 2013b ) , however , it is unknown how the ants then respond to the infection . Hence , in this study , we performed a series of behavioural and chemical experiments to test how ants detect and prevent infected brood from causing a systemic colony infection . We exposed ant pupae to one of either three dosages of Metarhizium conidiospores or a sham control . We observed that ants tending pathogen-exposed pupae prematurely removed the pupae from their cocoons in a behaviour we termed ‘unpacking’ , whereas control pupae were left cocooned ( Figure 1A–B , Video 1; Cox proportional hazards regression: likelihood ratio test ( LR ) χ2= 55 . 48 , df = 3 , p<0 . 001; hazard ratios ( x greater chance of unpacking compared to control ) : low dose = 18 , medium = 53 , high = 111; post-hoc comparisons: control vs . low , p=0 . 004; low vs . medium , p=0 . 006; medium vs . high = 0 . 024; all others , p=0 . 001 ) . Unpacking occurred between 2 and 10 days after pathogen exposure , but sooner and more frequently at higher conidiospore dosages ( Figure 1B ) . As unpacking was a belated response to pathogen exposure and we were unable to remove any conidiospores from the majority of the cocoons or the unpacked pupae ( number of colony forming units [mean ± 95% CIs]: cocoons = 0 . 6 ± 0 . 9; pupae = 0 . 1 ± 0 . 35; Figure 1—figure supplement 1 ) , we concluded that the ants were not performing unpacking to simply dispose of contaminated cocoons . Instead , we postulated that unpacking was a response to successful infection of the pupae . At the time of unpacking , the majority of pupae were still alive ( i . e . had an active dorsal aorta pulse; Figure 1—figure supplement 2 ) and fungal outgrowth had not yet occurred ( Figure 1F ) . Hence , to test if the ants were reacting to early-stage infections , we removed both unpacked and non-unpacked pathogen-exposed cocooned pupae from the ants and incubated them under optimal conditions for fungal outgrowth . We found that , on average across the conidiospore dosages , 85% of unpacked pupae harboured infections that sporulated in the absence of the ants . In contrast , only 25% of non-unpacked pupae were infected ( Figure 1—figure supplement 3; generalised linear model [GLM]: overall LR χ2 = 26 . 48 , df = 5 , p<0 . 001; cocooned vs . unpacked pupae: LR χ2 = 18 . 5 , df = 1 , p=0 . 001; conidiospore dose: LR χ2 = 0 . 42 , df = 2 , p=0 . 81 ) . We therefore concluded that the ants were detecting and unpacking pupae with lethal infections during the asymptomatic incubation period of the pathogen’s lifecycle . At this time point , the fungus growing inside the pupae is non-infectious and essentially no viable conidiospores are leftover from the pathogen exposure on their cuticle ( Figure 1—figure supplement 1 ) , so there is very little risk of the ants contracting the disease whilst unpacking the pupae . Next , we filmed ants presented with pathogen-exposed pupae and compared their behaviour before and after unpacking . Prior to unpacking , we observed the typical sanitary care behaviours reported in previous studies ( Graystock and Hughes , 2011; Tragust et al . , 2013a; Hughes et al . , 2002; Reber et al . , 2011; Okuno et al . , 2012 ) . Namely , the ants groomed the pupae ( Figure 1C ) , which has the dual function of removing the conidiospores and applying the ants’ antimicrobial poison ( Tragust et al . , 2013a ) . In L . neglectus , the poison is mostly formic acid and is emitted from the acidopore at the abdominal tip , where the ants actively suck it up and transiently store it in their mouths until application during grooming . Additionally , the ants can spray their poison directly from the acidopore; yet , this behaviour is rarely expressed during sanitary care ( about once every 28 hr; Figure 1D[Tragust et al . , 2013a] ) . However , after unpacking , we observed a set of behaviours markedly different to sanitary care ( Figure 1A , Video 1 ) . The ants sprayed the pupae with poison from their acidopore approx . 15-times more frequently than during sanitary care ( ~13 times/d; Figure 1D; generalised linear mixed model [GLMM]: LR χ2 = 17 . 04 , df = 1 , p<0 . 001 ) , and grooming duration doubled ( Figure 1C; linear mixed effects regression [LMER]: LR χ2 = 145 . 26 , df = 1 , p<0 . 001 ) . Given that there was no fungus to remove at the time of unpacking , the increase in grooming probably functioned solely to apply poison from the oral store ( Tragust et al . , 2013a ) . Furthermore , the ants repeatedly bit the pupae to make perforations in their cuticles and to remove their limbs ( Figure 1E; GLMM: LR χ2 = 39 . 44 , df = 1 , p<0 . 001 ) . Together these three behaviours resulted in the death of the pupae and left their corpses heavily damaged and coated in the ants’ poison ( Figure 1G , Figure 1—figure supplements 2 and 4 ) . Accordingly , we named the combination of unpacking , grooming , poison spraying and biting ‘destructive disinfection’ , and performed a series of experiments to determine its function . Firstly , we wanted to know how the ants identify internal infections during the pathogen’s non-contagious incubation period , when pupae were still alive and showed no external signs of disease . As ants use chemical compounds on their cuticles to communicate complex physiological information to nestmates ( Leonhardt et al . , 2016 ) , we speculated that infected pupae may produce chemical sickness cues . We washed infected pupae in pentane to reduce the abundance of their cuticular hydrocarbons ( CHCs ) . When pentane-washed pupae were presented to ants , there was a 72% reduction in unpacking compared to both non- and water-washed infected pupae ( Figure 2A; GLM: LR χ2 = 12 . 2 , df = 2 , p=0 . 002; Tukey post hoc comparisons: water-washed vs . non-washed , p=0 . 79; all others , p=0 . 009 ) . As pentane-washed pupae had lower abundances of CHCs ( Figure 2—figure supplement 1 ) , this result indicates that the ants use one or more cuticular compounds to detect the infections . Gas chromatography-mass spectrometry ( GC–MS ) analysis of the solvent wash confirmed that unpacked pupae have distinct chemical profiles compared to non-infected control pupae , whilst cocooned ( non-unpacked ) pathogen-exposed pupae were intermediate ( Figure 2B , Figure 2—figure supplement 2; perMANOVA: F = 1 . 49 , df = 46 , p=0 . 002; post-hoc perMANOVA comparisons: unpacked vs . control , p=0 . 003; unpacked vs . cocooned , p=0 . 79; cocooned vs . control , p=0 . 08 ) . There were no novel compounds present on unpacked or cocooned pathogen-exposed pupae that were not also present on control pupae ( Table 1; Figure 2—figure supplement 2 ) , suggesting that these differences were not caused by odours emitted directly by the fungus , but were of pupal origin . By analysing the chemical profiles of each of the pathogen’s separate developmental stages ( infectious conidiospores , post-infection blastospores , and saprophytic mycelium ) and performing a direct comparison of the fungal compounds to the pupal chemical profiles , we confirmed that there were no fungus-derived peaks in the pupal profiles ( see Materials and methods for more information ) . Most chemical messages in social insects are encoded by quantitative shifts of several compounds ( Leonhardt et al . , 2016 ) . Correspondingly , we found that 8 out of the 24 CHCs identified ( Table 1 ) had higher relative abundances on unpacked pupae compared to control pupae ( Figure 2C–F , Figure 2—figure supplement 2; all Kruskal-Wallis [KW] test statistics and post-hoc comparisons in Table 2 ) . Moreover , four of these CHCs were also present in relatively higher quantities on unpacked pupae compared to the non-unpacked cocooned pupae . Several specific CHCs are therefore probably accumulating on infected pupae over time , eventually reaching an amount that , relative to the other compounds , is sufficient to elicit destructive disinfection . This corresponds to current models of social insect behaviour , where the likelihood of a response depends on stimuli exceeding a certain threshold ( Theraulaz et al . , 1998; Beshers and Fewell , 2001 ) . To investigate the possibility that CHC changes on unpacked pupae are the result of an immune response developed by the host , we injected pupae with β−1 , 3-glucans – polysaccharides that are an integral component of fungal cell walls , including Metarhizium ( Wang and St Leger , 2006 ) . β-Glucans act as highly conserved major pathogen-associated molecular patterns ( PAMPs ) that are recognised by the immune system of invertebrates and can therefore be used to elicit an immune response in the absence of a pathogen ( Brown and Gordon , 2005 ) . We found that , within 2 days of injection , β-glucan , but not saline , caused an increase in the expression of immune genes , namely , an IMD signalling pathway regulator gene ( PGRP-SC2 [Bischoff et al . , 2006] ) and a gene encoding for a protein that recognises and binds to β-glucans to elicit an immune response ( β−1 , 3-GBP [Ma and Kanost , 2000; Gottar et al . , 2006] ) , whilst expression of the gene encoding for the melanisation cascade enzyme phenoloxidase ( proPO [Cerenius and Söderhäll , 2004] ) was unaffected ( Figure 2—figure supplement 3 ) . β-glucan injection also altered the chemical profiles of pupae in a similar way to Metarhizium infection ( Figure 2C–F ) . Two of the four compounds we identified as a potential sickness cue on the unpacked pupae ( Tritriacontadiene and Tritriacontene ) were also increased in abundance within 2 days of injection with β-glucan , whilst there was no such increase in control pupae ( Figure 2—figure supplement 4 ) . These data reveal that some of the changes in pupal chemical profile can be directly linked to a host reaction to an immune elicitor , similar to findings in honeybees ( Richard et al . , 2012; Richard et al . , 2008 ) , mice ( Arakawa et al . , 2011 ) and humans ( Shirasu and Touhara , 2011 ) . We next tested if destructive disinfection prevents pupal infections from replicating and becoming infectious . Pathogen-exposed pupae were kept with groups of ants ( eight ants per pupae per group ) until unpacking . They were then left with the ants for a further 1 or 5 days before being removed and incubated for fungal growth . We compared the number that subsequently sporulated to pathogen-exposed pupae kept without ants . Whilst 88% of pupae contracted infections , destructive disinfection significantly reduced the proportion of pupae that sporulated and hence became infectious ( Figure 3A; GLM: LR χ2 = 40 . 47 , df = 2 , p<0 . 001; Tukey post-hoc comparisons: 1 vs . 5 d , p=0 . 04; all others , p<0 . 001 ) . After only 1 day , the number of destructively disinfected pupae that sporulated decreased by 65% . With more time , the ants could reduce the number of pupae sporulating even further by 95% . Since the pupae were removed from the ants for fungal incubation , we can conclude that destructive disinfection permanently prevents pathogen replication . We repeated this experiment with a smaller number of ants ( three ants per pupae per group ) to investigate how group size influences the success of destructive disinfection . Smaller groups of ants were less efficient than larger ones: although they could still inhibit >90% of pupal infections within 5 days of unpacking , pupae tested for infection after 1 day still sporulated 70% of the time ( Figure 3—figure supplement 1; GLM: LR χ2 = 35 . 23 , p<0 . 001; Tukey post-hoc comparisons: 0 vs . 1 day , p=0 . 2; 0 vs . 5 days , p<0 . 001; 1 vs . 5 days , p=0 . 002 ) . As the effectiveness of destructive disinfection increased with the amount of time the ants had , as well as with the number of ants present , we inferred that there must be a limiting factor affecting the inhibition the pathogen . To study the underlying mechanisms of destructive disinfection , we performed its different components – unpacking , biting and poison spraying – in vitro to test for their relative importance and potential synergistic effects . We simulated unpacking by removing the cocoons of the pupae manually with fine forceps , and the cuticle damage caused by biting using dissection scissors . Previous work establishing the composition of L . neglectus poison ( Tragust et al . , 2013a ) allowed us to create a synthetic version for use in this experiment ( 60% formic acid and 2% acetic acid , in water; applied at a dose equivalent to what ants apply during destructive disinfection; Figure 3—figure supplement 2 ) , with water as a sham control . We then performed these ‘behaviours’ in different combinations in a full-factorial experiment . We found that all three behaviours must be performed in the correct order and interact to prevent pathogen replication ( overview graph showing odds ratios of sporulation in Figure 3B , full data dataset displayed in Figure 3—figure supplement 3; GLM: overall LR χ2 = 79 . 9 , df = 5 , p<0 . 001; interaction between behaviours LR χ2 = 20 . 6 , df = 2 , p<0 . 001; all post-hoc comparisons in Table 3 ) . As in sanitary care , the poison was the active antimicrobial compound that inhibited fungal growth ( Figure 3—figure supplement 3 , Table 3 , Tragust et al . , 2013a; Graystock and Hughes , 2011 ) . However , for the poison to function the pupae had to be removed from their cocoons and their cuticles damaged . Firstly , this is because the cocoon itself is hydrophobic and thus prevents the aqueous poison from reaching the pupae inside ( Figure 3—figure supplement 4 ) . Secondly , as the infection is growing internally at the time of unpacking , the cuticle must be broken in order for the poison to enter the hemocoel of the pupae . This is achieved with the perforations created by the ants biting the pupal cuticle . It is possible that in the wild biting also helps to desiccate the pupae , since high levels of humidity are important for Metarhizium growth ( Doberski , 1981 ) . However , in our experiments , the relative humidity inside the petri dishes was always a stable 95% , so desiccation cannot have played a role in fungal inhibition . As the active antimicrobial component , we concluded that the poison is probably the limiting factor determining whether destructive disinfection is successful . Because the poison has a slow biosynthesis and each ant can only store a limited amount ( Tragust et al . , 2013a; Hefetz and Blum , 1978 ) , it would explain why destructive disinfection was more likely to be successful the longer the ants had to treat the pupae , and as the number of ants increased ( Figure 3A , Figure 3—figure supplement 1 ) . By sharing the task of poison synthesis and application , the ants probably increase their chances of preventing the pathogen becoming infectious . Finally , we investigated the impact of destructive disinfection on disease transmission within a social group . We created mini-nests comprising two chambers and a group of ants ( five ants per group ) . Into one of the chambers we placed an infectious sporulating pupa – simulating a failure of the ants to detect and destroy the infection – or a pupa that had been destructively disinfected , and was thus non-infectious . The ants groomed , moved around and sprayed both types of corpses with poison . In the case of the sporulating pupae , all conidiospores were removed from the corpse by the ants . As in previous studies , sporulating corpses were highly virulent ( Hughes et al . , 2002; Loreto and Hughes , 2016 ) and caused lethal infections that became contagious after host death in 42% of ants ( Figure 4A ) . However , there was no disease transmission from destructively disinfected pupae ( Figure 4A; GLM: LR χ2 = 31 . 32 , df = 1 , p<0 . 001 ) . We therefore concluded that by preventing the pathogen from completing its lifecycle destructive disinfection stops intra-colony disease transmission ( Figure 4B ) . In this study , we show that the superorganismal societies of ants have evolved an efficient mechanism to specifically target and eliminate infections that have established in colony members , before they become contagious . This is achieved through the detection of chemical cues emitted by infected pupae during the non-transmissible incubation period of the pathogen ( Figure 2 ) . The ants then engage in destructive disinfection , a multicomponent behaviour that utilises the ants’ antimicrobial poison , in conjunction with cocoon removal and biting , to prevent pathogen replication within the body of the pupae ( Figure 1 , Figure 3 ) . Ultimately , this prevented the pathogen from completing its lifecycle and infecting new hosts , thereby effectively reducing pathogen fitness to zero ( Figure 4 ) . These findings show that ants do not only avoid , groom and isolate pathogens ( Cremer et al . , 2007; Cremer et al . , 2017 ) but can detect and eliminate infections developing inside the bodies of their nestmates , even before they have shown external disease symptoms . Whilst the role of ant poison as a topical disinfectant by ants and other animals ( i . e . ‘anting’ behaviour in birds ) is well characterised ( Clayton et al . , 2010; Verderane et al . , 2007; Tragust , 2016 ) , its use as an internal disinfectant within the body of others during destructive disinfection is a novel and a rare example of the kill-component of social immunity ( Cremer and Sixt , 2009 ) . Eliminating infected kin to protect the rest of the group , observed in termites and honeybees as well ( Rothenbuhler , 1964; Chouvenc and Su , 2012; Spivak and Gilliam , 1998 ) , requires an unconditional level of altruism that is expected to be absent or at least rare in other forms of sociality ( e . g . aggregations , non-superorganismal family groups and communal breeders [Cremer et al . , 2017] ) , but has parallels to the immune system of the metazoan body ( Cremer and Sixt , 2009 ) . Both the immune system and social immunity have first lines of defences that reduce the risk of infection: pathogens that enter the body are met with mechanical and chemical defences , such as ciliated cells in the lung that move pathogens trapped in mucus out of the body ( Cremer and Sixt , 2009 ) , and in ants , sanitary care plays an analogous role ( Tragust et al . , 2013a; Graystock and Hughes , 2011 ) . However , if a pathogen circumvents these first defences in the body and an infection occurs , the second line of defence is often a targeted elimination of the infected cells . This starts with immune cells detecting an infection and then transporting cell death-inducing and antimicrobial compounds into infected cells by creating pores in their membrane ( Walch et al . , 2014; Kägi et al . , 1994; Chowdhury and Lieberman , 2008 ) . Likewise , our experiments revealed that ants unpack infected pupae and make perforations in their cuticle , enabling the ants to spray their poison directly into the pupae’s body; hence they display mechanisms analogous to immune cell elimination . In both cases , this second line of defence destroys the infected cell/insect , along with the infection , to prevent transmission ( Shore et al . , 1976 ) . Since the loss of both somatic cells and individual insect workers can be tolerated with negligible effects on fitness ( Cremer and Sixt , 2009 ) , these convergent strategies have likely evolved at both the multicellular and superorganismal levels of biological organisation , as an effective way to clear infections and avoid any further damage to the body and colony , respectively . Animals from a variety of taxa are known to identify sick conspecifics based on odour signals ( Bozza , 2015; Poirotte et al . , 2017; Kiesecker et al . , 1999; Anderson and Behringer , 2013; Shirasu and Touhara , 2011; Swanson et al . , 2009 ) , and although it has been hypothesised that ants should also use chemical cues to detect sick colony members , evidence has so far been lacking ( Ugelvig et al . , 2010; Leclerc and Detrain , 2016; Bos et al . , 2012 ) . To our knowledge , we have therefore observed ants using chemical information for the first time to rapidly and accurately target infected individuals . We found that the chemical compounds with increased abundances on infected pupae are distinct from those that induce the removal of corpses in ants ( Diez et al . , 2013; Qiu et al . , 2015; Wilson et al . , 1958 ) , and , like in tapeworm-infected ants ( Trabalon et al . , 2000 ) , are not pathogen-derived . This alteration of the hosts’ chemical profile may arise during infection from the breakdown of hydrocarbons by Metarhizium penetration ( Lin et al . , 2011 ) or after infection due to an immune response affecting the synthesis of specific hydrocarbons ( Richard et al . , 2012; Richard et al . , 2008 ) . The latter is more likely because the ants only display destructive disinfection once the fungus is growing inside the pupae and injection of a fungal immune elicitor in the absence of live pathogen induced similar changes . The four CHCs specifically increased on unpacked pupae are all long-chained CHCs ( carbon chain length C33-35 ) with a low volatility , meaning that the ants have to be close to or touching the pupae to detect them ( Sharma et al . , 2015 ) . As ants keep pupae in large piles , using low-volatility CHCs may be important so that the ants can accurately identify the infected pupae and do not mistakenly destroy healthy ones . Interestingly , two of the four CHCs that were increased on infected pupae , as well as on pupae that were injected with the fungal cell wall component , also had higher abundances on virus-infected honeybees ( Tritriacontadiene [Baracchi et al . , 2012] ) , or their brood when injected with a bacterial cell membrane component ( Tritriacontene [Richard et al . , 2012] ) . This raises the possibility that these hydrocarbons are evolutionarily conserved ‘sickness cues’ in Hymenopteran social insects . Such cues may have evolved into general sickness signals in social insects as they alert their kin to the presence of a developing infection that will harm the colony if it spreads ( Shakhar and Shakhar , 2015 ) . Relying on cues generated by an immune response to detect infections , over pathogen-specific cues , is likely more robust and general , similar to the expression of ‘find-me/eat-me’ signals by infected cells in vertebrate bodies ( Ravichandran , 2010; Grimsley and Ravichandran , 2003 ) , as well as the immune system responding to cell damage signals ( danger signal hypothesis [Matzinger , 2007] ) . Such signals will be selected for in social insects because they can enhance colony fitness ( and hence the indirect fitness of the sick individual ) by preventing a systemic infection . Therefore , altruistic displays of sickness can evolve in superorganisms , even if this results in the destruction of the individual that expresses them ( Cremer et al . , 2017 ) . Our experiments show that destructive disinfection was highly effective and prevented 95% of infections becoming transmissible . Destructive disinfection will thus keep the average number of secondary infections caused by an initial infection low and the disease will die out within the colony ( Schmid-Hempel , 2017 ) . This may explain why infections of Metarhizium and other generalist entomopathogenic fungi like Beauveria , though common in the field ( Reber et al . , 2012; Hughes et al . , 2004a; Cremer et al . , 2008; Keller et al . , 2003 ) , do not seem to cause colony-wide epidemics in ants ( Cremer et al . , 2017 ) , but are more numerous in solitary species that lack social adaptations to resist disease ( Roberts and St Leger , 2004; Shimazu , 1989; Lomer et al . , 1997 ) . Behaviours like destructive disinfection that are able to reduce pathogen fitness to zero could have selected for host manipulation in fungi that specialise on infecting ants , for example Ophiocordyceps and Pandora ( Hughes et al . , 2016; Loreto et al . , 2014; Małagocka et al . , 2017 ) . These fungi manipulate their ant hosts into leaving the nest before they become infectious and our study supports previous suggestions that this may be to avoid social immunity defences , like destructive disinfection , which would prevent them completing their lifecycle inside the nest ( Loreto et al . , 2014 ) . In contrast to specialists , generalist pathogens like Metarhizium infect a broad range of solitary and social hosts , making it less likely that they evolve strategies to escape social immunity defences ( Cremer et al . , 2017; Agosta et al . , 2010 ) . We have also observed destructive disinfection in an unrelated supercolonial population of L . neglectus ( Ugelvig et al . , 2008 ) and another non-supercolonial/non-invasive Lasius species , L . niger ( see Materials and methods for more information ) , suggesting that it may be a common behaviour that has evolved in Lasius , and possibly other ant genera , in response to the constant selection pressure applied by generalist pathogenic fungi ( Cremer et al . , 2017 ) . Future work that investigates how social immunity disrupts typical host-pathogen dynamics will shed light on the co-evolution of pathogens and their social hosts ( Schmid-Hempel , 2017 ) . Destructive disinfection has probably evolved in ants because the removal of corpses from the colony alone does not guarantee that disease transmission is prevented ( Loreto et al . , 2014 ) . Ants are usually sedentary , building nests that remain in the same location until the colony dies ( Boomsma et al . , 2005; Schmid-Hempel , 1998 ) . Additionally , they are highly territorial and forage mostly in the area around their nests , meaning that if they do not clear it of dead and potentially infectious nestmates , they are likely to be reencountered ( Boomsma et al . , 2005; Cremer et al . , 2017 ) . Ants tend therefore to place corpses onto specific midden ( trash ) sites that are located inside or outside near the nest , but these sites are still regularly visited by midden workers ( Verza et al . , 2017; Hart and Ratnieks , 2002; Farji-Brener et al . , 2016 ) . Consequently , although middens likely reduce a colony’s exposure to corpses , they still represent a potential route for disease transmission back into the colony; hence the need to destroy infected corpses rather than simply taking them out of the nest . This is in contrast to honeybees , where corpses are dumped randomly outside of the hive ( Wilson-Rich et al . , 2009 ) . This behaviour is sufficient to prevent disease transmission because honeybees are unlikely to reencounter corpses whilst foraging on the wing ( Spivak and Reuter , 2001 ) . Termites on the other hand cannibalise their dead ( Chouvenc and Su , 2012; Rosengaus and Traniello , 2001 ) . Cannibalism is effective because the termite gut and/or its microbiome neutralises ingested pathogens ( Chouvenc et al . , 2009; Rosengaus et al . , 2014; Rosengaus et al . , 1998 ) and has likely evolved because dead nestmates are a source of valuable nitrogen in their cellulose-base diet ( Rosengaus et al . , 2011 ) . The same selective pressure has driven this suite of independently evolved innovations – the need to eliminate or remove infected individuals early in the infectious cycle – with the ants expressing a particularly complex behavioural repertoire . This seems to be a general principle in disease defence , as cells are also rapidly detected and destroyed shortly after infection to prevent pathogen spread in multicellular organisms ( Cremer and Sixt , 2009 ) . Understanding how natural selection can result in similar traits at different levels of biological organisation and in organisms with different life histories is a central question in evolutionary biology ( Bourke , 2011 ) . Studying the similarities and differences between organismal immunity and social immunity could therefore potentially lead to new insights about how disease defences evolve ( Cremer and Sixt , 2009; Kennedy et al . , 2017 ) . For example , in most social animals , where relatedness is low or altruism is only conditionally expressed ( e . g . by young that eventually disperse to reproduce themselves ) , disease defences tend to rely on self-protective infection avoidance ( Curtis , 2014 ) , mutually expressed sanitary care ( Nunn and Altizer , 2006 ) and herd immunity only ( John and Samuel , 2000 ) . In contrast , the results of our study suggest that equivalent selection pressures during the major evolutionary transitions from unicellularity to multicellularity in the metazoans , and from sociality to superorganismality in the social insects , have resulted in convergent defences that protect multicellular organisms and superorganismal insect societies from systemic disease spread , by ensuring the survival of the whole over its parts ( Cremer and Sixt , 2009 ) . We studied the invasive garden ant Lasius neglectus ( Cremer et al . , 2008 ) that forms large , underground nests in the soil . Populations of this species lack territorial structuring and instead consist of interconnected nests , forming a single supercolony between which there is a constant exchanging of individual ants ( Cremer et al . , 2008 ) . We sampled more than one hundred queens , many thousands of workers and hundreds of brood items from a 320 m2 area of the supercolony in Seva , Spain ( 41°48'32 . 4"N 2°15'43 . 9"E ) , and reared them as stock colonies in the laboratory . All experiments were conducted in plastered petri dishes ( Ø=33 , 55 or 90 mm ) with 10% sucrose solution provided ad libitum and environmental conditions were controlled throughout ( 23°C; 70% RH; 14/10 hr light/dark cycles ) . In addition , we directly measured the humidity of plastered petri dishes without ants for 2 weeks , by embedding a digital relative humidity sensor ( Sensirion , Switzerland ) into the lids of the dishes , finding that the relative humidity inside was always a stable 95% . The animal use protocol was performed in accordance with the IST Austria Ethics Committee guidelines . At present , the committee does not provide a specific approval numbers for invertebrate animal research . Animals used in this study , Lasius neglectus , do not belong to regulated or protected species . As a model pathogen , we used an obligately killing pathogen of Lasius ants , Metarhizium brunneum ( CDP unpublished data; strain MA275 , KVL 03–143 ) . Entomopathogenic Metarhizium fungi occur at high densities in the soil ( up to 5000 conidiospores/g soil [Keller et al . , 2003] ) and on sporulating cadavers ( up to 12 million conidiospores/cadaver [Hughes et al . , 2002 , 2004b] ) and are responsible for natural infections of ants in field populations ( Reber et al . , 2012; Hughes et al . , 2004a ) . Metarhizium and other similar generalist fungi are therefore expected to have applied a persistent and pervasive selection pressure on ants over the course of their evolutionary history ( Boomsma et al . , 2005; Cremer et al . , 2017 ) . Multiple aliquots of conidiospore suspensions were kept in long-term storage at – 80°C . Prior to each experiment , the conidiospores were grown on sabaroud dextrose agar at 23°C until sporulation and harvested by suspending them in 0 . 05% sterile Triton X-100 ( Sigma Aldrich , Austria ) . The germination rate of conidiospore suspensions was determined before the start of each experiment and was >90% in all cases . In addition to fungal conidiospores , we also cultured blastospores and mycelia to obtain the chemical profiles ( see below ) of all stages of the fungi’s lifecycle ( Deacon , 2006 ) . Blastospores were cultured by adding 50 µl of conidiospore suspension ( 109/ml ) to 50 ml of Adámek liquid media ( with Streptomycin sulphate [0 . 005 g/l] and Chloramphenicol [0 . 025 g/l] ) added to inhibit bacterial growth ) in a 300 ml Erlenmeyer flask , which was then incubated ( 72 hr , 200 rpm , 23°C ) ( Adámek , 1965; Kleespies and Zimmermann , 1992 ) . After incubation , the liquid ( which contains the blastospores ) was pre-filtered using a flame-sterilised mesh and sieve and the liquid vacuum-filtered ( 40 µm mesh; Millipore Steriflip [Merck , Germany] ) . The resulting blastospore suspension was then washed ( 5 min , 3000 g , 23°C ) three times in PBS . To produce mycelia , we added 50 µl of 106/ml conidiospore suspension to 100 ml of YPD liquid broth ( Yeast extract Peptone Dextrose with Streptomycin sulphate [0 . 005 g/l] and Chloramphenicol [0 . 025 g/l] ) in a 300 ml Erlenmeyer flask . We incubated the flask ( 5 days , 180 rpm , 27°C ) and vacuum-filtered ( 40 µm mesh; Millipore Steriflip ) the resulting fungal mass to remove the liquid broth . We then washed the mycelial mass three times in autoclaved distilled water . Conidiospores were applied in a suspension of 0 . 05% autoclaved Triton-X 100 at a concentration of 106 conidiospores/ml in all experiments , unless otherwise stated . Throughout the study , we used cocooned worker pupae of approximately the same age , which was determined by assessing the melanisation of the eyes and cuticle . Single pupae were exposed by gently rolling them in 1 µl of the conidiospore suspension using sterile soft forceps . Pupae were then allowed to air dry for 5–10 min before being used in experiments . This exposure procedure resulted in pupae receiving ~1800 conidiospores , of which 5% ( ~95 conidiospore ) passed through the cocoon and came into contact with the pupa inside ( Figure 1—figure supplement 1 ) . In all experiments , pupae were allocated to treatment groups haphazardly . Statistical analyses were carried out in R version 3 . 3 . 2 ( Core Team , 2012 ) and all tests were two-tailed . All General ( ised ) linear and mixed models were compared to null ( intercept only ) and reduced models ( for those with multiple predictors ) using Likelihood Ratio ( LR ) tests to assess the significance of predictors ( Bolker et al . , 2009 ) . We controlled for the number of statistical tests performed per experiment to protect against a false discovery rate using the Benjamini-Hochberg procedure ( α = 0 . 05 ) . Moreover , all post-hoc analyses were corrected for multiple testing using the Benjamini-Hochberg procedure ( α=0 . 05 ) ( Benjamini and Hochberg , 1995; García , 2004 ) . We checked the necessary assumptions of all tests that is by viewing histograms of data , plotting the distribution of model residuals , checking for non-proportional hazards , testing for unequal variances , testing for the presence of multicollinearity , testing for over-dispersion , and assessing models for instability and influential observations . For mixed effects modelling , we used the packages ‘lme4’ to fit models ( Bates et al . , 2014 ) , ‘influence . ME’ to test assumptions ( Nieuwenhuis and Pelzer , 2012 ) , and , for LMERs , ‘lmerTest’ to obtain p values ( Kuznetsova et al . , 2015 ) . All logistic regressions were performed using either generalised linear models ( GLMs ) or generalised linear mixed models ( GLMMs ) , which had binomial error terms and logit-link function . The Cox proportional hazards regression was carried out using the ‘coxphf’ package with post-hoc comparisons achieved by re-levelling the model and correcting the resulting p values ( Ploner and Heinze , 2015 ) . For Kruskal-Wallis ( KW ) tests and subsequent post-hoc comparisons we used the ‘agricolae’ package , which implements the Conover-Iman test for multiple comparisons using rank sums ( de Mendiburu , 2016 ) . For the perMANOVA , we used the package ‘vegan’ and performed pairwise perMANOVAs for post-hoc comparisons ( Oksanen et al . , 2016 ) . All other post-hoc comparisons were performed using the ‘multcomp’ package ( Bretz et al . , 2011 ) . All graphs were made using the ‘ggplot2’ package ( Wickham , 2009 ) . Preliminary studies were performed for all major experiments to determine sample size . No data outliers were detected or removed and all replicate information represents biological replicates . Individual descriptions of statistical analyses are given for all experiments below . To study how ants respond to infections , we exposed pupae to a low ( 104/ml ) , medium ( 106/ml ) or high ( 109/ml ) dose of conidiospores or autoclaved Triton X as a sham control ( sham control , n = 24; all other treatments , n = 25 ) . The pupae were then placed into individual petri dishes with two ants and inspected hourly for 10 hr/days for 10 days . When the ants unpacked a pupa , it was removed and surface-sterilised to ensure that any fungal outgrowth was the result of internal infections and not residual conidiospores on the cuticle . To surface-sterilise , pupae were dipped in 70% ethanol , washed in autoclaved distilled water and submerged in 0 . 05% sodium hypochlorite for 1 min , before being washed three times in autoclaved distilled water ( Lacey and Brooks , 1997 ) . After sterilisation , we transferred the pupae to a petri dish lined with damp filter paper at 23°C and monitored them for 2 weeks for Metarhizium sporulation to confirm the presence of an internal infection ( low dose , n = 8; medium dose , n = 18; high , n = 21 ) . In addition , any cocooned pupae that were not unpacked after 10 d were removed from the ants , surface sterilised and observed for sporulation , as above ( low dose , n = 11; medium dose , n = 4; high , n = 4 ) . Some pupae ( control = 16 , low dose = 6 , medium dose = 3 ) successfully emerged from the cocoon as adult ants and were thus treated as non-unpacked in analyses . We analysed the effect of treatment on unpacking using a Cox proportional hazards model with Firth’s penalised likelihood , which offers a solution to the monotone likelihood caused by the complete absence of unpacking in the sham control treatment . We followed up this analysis with post hoc comparisons ( model factor re-levelling ) to test unpacking rates between treatments ( Figure 1B ) . We compared the number of unpacked and cocooned pupae sporulating using a logistic regression , which included pupa type ( cocooned , unpacked ) , conidiospore dose ( low , medium , high ) and their interaction as main effects . The interaction was non-significant ( GLM: LR χ2 = 5 . 0 , df = 2 , p=0 . 084 ) ; hence , it was removed to gain better estimates of the remaining predictors . Photographs of destructive disinfection were captured ( Nikon D3200 [Nikon , Japan] ) and aesthetically edited ( Adobe Photoshop [Adobe Systems , San Jose , California] ) to demonstrate the different behaviours ( Figure 1A ) . They were not used in any form of data acquisition . We also made representative SEMs of a pupa directly after unpacking and one after destructive disinfection ( 24 hr after unpacking; Figure 1F–G ) . As the pupae were frozen at – 80°C until the SEMs were made , we also examined non-frozen pupae taken directly from the stock colony and confirmed that freezing itself does not cause damage to the pupa ( not shown ) . We determined the number of conidiospores on unpacked pupae ( n = 7 ) and their removed cocoons ( n = 7 ) by placing them into separate vials containing 100 µl autoclaved 0 . 05% Trixton-X 100 . The vials were then shaken for 10 m at 600 RPM ( Vortex Genie 2 [Scientific Industries , Bohemia , New York] ) and the resulting supernatant was plated onto selective medium agar . We counted the number of Metarhizium colony-forming units ( CFUs ) that subsequently grew on the plates after 7 d . As a control , we performed the same experiment on pupae directly after pathogen-exposure . We experimentally unpacked the pupae using sterile ( ethanol wiped ) forceps so that we could examine the number of CFUs present on the pupae ( n = 16 ) and cocoon separately ( n = 16 ) . We analysed the number of CFUs on pupae and cocoons using Mann-Whitney U tests ( Figure 1—figure supplement 1 ) . To observe how the behavioural repertoire of the ants changes between sanitary care and destructive disinfection , we filmed three individually colour-marked ants tending a single pathogen-exposed pupa with a USB microscope camera ( Di-Li 970-O [Di-Li , Germany] ) . To characterise the sanitary care behaviours of the ants , we analysed the first 24 hr of the videos following the introduction of the pupa . To study destructive disinfection behaviours , we analysed the 24 hr period that immediately followed unpacking . Videos were analysed using the behavioural-logging software JWatcher ( Blumstein and Daniel , 2007 ) . For each ant ( n = 15 ) , we recorded the duration of its grooming bouts , the frequency of poison application and the frequency of biting . Grooming duration was analysed using a LMER , having first log-transformed the data to fulfil the assumption of normality ( Figure 1C ) . The frequency of poison spraying and biting ( Figure 1D–E ) were analysed using separate GLMMs with Poisson error terms for count data and logit-link function . We included an observation-level random intercept effect to account for over-dispersion in the poison spraying and biting data ( Harrison , 2014 ) . In all three models , we included petri dish identity as a random intercept effect because ants from the same dish are non-independent . Additionally , a random intercept effect was included for each ant as we observed the same individuals twice ( before and after unpacking ) . We established a protocol to determine whether pupae were dead or alive because it is not generally obvious when death has occurred . To ensure that we examined pupae as soon as possible after unpacking , we checked pathogen-exposed pupae housed with ants every 45 min for 15 hr/d . When unpacking occurred , we either removed the pupa immediately ( n = 33 ) or left it with the ants for a further 24 hr so that they could perform destructive disinfection ( n = 44 ) . To check the numbers of dead and alive pupae at the time point of unpacking and after destructive disinfection , we secured the pupae to glass slides using double-sided tape . The pupae were then gently prodded with a glass capillary whilst being examined under a bifocal microscope ( 10 x magnification; Leica DM 1000 [Leica Biosystems , Germany] ) . If pupae were alive , this resulted in contractions of their dorsal aorta ( Broome et al . , 1976 ) , which is visible through the cuticle of the abdomen . If they were dead , no contractions occurred . Each examination lasted a maximum of 5 min . To confirm that this approach was sensitive , we examined experimentally unpacked pupae taken straight from a stock colony ( n = 10 ) . In all cases , these pupae were alive . They were then frozen at – 80°C for 1 day and examined again after defrosting , when they were all found to be dead . We compared the number of dead pupae at the time point of unpacking to the number that were dead after destructive disinfection using a logistic regression ( Figure 1—figure supplement 2 ) . We included the day of unpacking as a covariate to test if pupae unpacked sooner or later were more or less likely to have already died . As L . neglectus poison has a very high acidity ( Tragust et al . , 2013a ) , we could measure the pH of pupae to determine if ants apply higher amounts poison to pupae during destructive disinfection . We kept a pair of pathogen-exposed or sham control pupae with two ants . When one of the pathogen-exposed pupae in a pair was unpacked , we let the ants perform destructive disinfection for 24 hr ( n = 25 ) . In the control , we experimentally unpacked one pupa in a pair and placed it back with the ants for 24 hr ( n = 17 ) . After 24 hr , we removed the unpacked pupae in both treatments along with their discarded cocoons . At the same time , the second , still cocooned pupae in each pair was removed and experimentally unpacked so that pH measurements were consistent across pupal groups ( pathogen exposed , n = 9; control , n = 16 ) . All pupae and their cocoons were placed into individual vials containing 20 µl of autoclaved distilled water and a sterile glass pestle was used to crush each pupa and cocoon for 60 s . The pH of the resulting pupa/cocoon slurry was measured using a pH electrode meter ( INLAB ULTRA-MICRO , SevenGo PRO pH SG8 pH-meter [Mettler-Toledo , Columbus , Ohio] ) . This gave us an indication of how much poison the ants had applied to each type of pupa ( Figure 1—figure supplement 3 ) . We used a LMER with Tukey post-hoc comparisons to compare the pH measurements of the pupae . Pupa treatment ( pathogen-exposed or control ) , type ( cocooned or unpacked ) and their interaction were included as main effects . Petri dish was included as a random intercept effect as pairs of pupae from the same dish are non-independent . As we used a portion of this dataset in Figure 3—figure supplement 2 , we corrected the overall model p value for multiple testing . We determined whether ants detect infected pupae through potential changes in the pupae’s cuticular chemical profile . We established internal infections in pupae by exposing them to the pathogen and leaving them for 3 days in isolation . In pilot studies , approx . 50% of these pupae were then unpacked within 4 hr of being introduced to ants . After 3 days , pupae were washed for 2 . 5 min in 300 µl of either pentane solvent to reduce the abundance of all CHCs present on the pupae ( n = 28 ) , or in autoclaved water as a handling control ( n = 28 ) . After washing , pupae were allowed to air dry on sterile filter paper . Additionally , non-washed pupae were used as a positive control ( n = 30 ) . Pupae were placed individually with a pair of ants in petri dishes and observed for unpacking for 4 hr . We used GC–MS ( see below for methodology ) to confirm that washing was effective at removing cuticular compounds , by comparing the total amount of chemicals present on pupae washed in pentane to non- and water-washed pupae ( n = 8 per treatment; Figure 2—figure supplement 1 ) . The number of pupae unpacked between the different treatments was analysed using a logistic regression ( Figure 2A ) . As several researchers helped to wash the pupae , we included a random intercept for each person to control for any potential handling effects . Additionally , the experiment was run in two blocks on separate days , so we included a random intercept for each block to generalise beyond any potential differences between runs . The total peak area from the GC–MS analysis was compared between treatments using a KW test with post-hoc comparisons . To confirm that infected pupae had chemical profiles that are different from pathogen-exposed cocooned and control pupae , we exposed pupae to the pathogen or a sham control . Pupae were then isolated for 3 days to establish infections in the pathogen-exposed treatment ( as above ) . Following isolation , pupae were individually placed with ants and observed for unpacking for 4 hr . Unpacked pupae were immediately frozen at – 80°C with the removed cocoons ( n = 13 ) and we also froze cocooned pathogen-exposed pupa that had not yet been unpacked ( n = 10 ) . Furthermore , we froze a pair of control pupae , of which one was cocooned ( n = 12 ) , whilst the other was first experimentally unpacked ( to test if the cocoon affects cuticular compound extraction; n = 12 ) . Cuticular chemicals were extracted from individual pupae and their cocoons in glass vials ( 1 . 8 ml [Supelco , Germany] ) containing 100 µl n-pentane solvent for 5 min under gentle agitation . The vials were then centrifuged at 3000 rpm for 1 min to spin down any fungal conidiospores that might be remaining , and 80 µl of the supernatant was transferred to fresh vials with 200 µl glass inserts and sealed with Teflon faced silicon septa ( both Supelco ) . The pentane solvent contained four internal standards relevant for our range of hydrocarbons ( C27 – C37 ) ; n-Tetracosane , n-Triacontane , n-Dotriacontane and n-Hexatriacontane ( Sigma Aldrich ) at 0 . 5 µg/ml concentration , all fully deuterated to enable spectral traceability and separation of internal standards from ant-derived substances . We ran extracts from the different groups in a randomised manner , intermingled with blank runs containing only pentane , and negative controls containing the pentane plus internal standards ( to exclude contaminants emerging for example from column bleeding ) , on the day of extraction , using GC–MS ( GC7890 coupled to MS5975C [Agilent Technologies , Santa Clara , California] ) . A liner with one restriction ring filled with borosilicate wool ( Joint Analytical Systems , Germany ) was installed in the programmed temperature vaporisation ( PTV ) injection port of the GC , which was pre-cooled to −20°C and set to solvent vent mode . 50 µl of the sample extractions were injected automatically into the PTV port at 40 µl/s using an autosampler ( CTC Analytics , PAL COMBI-xt , , CHRONOS 4 . 2 software [Axel Semrau , Germany] ) equipped with a 100 µl syringe ( Hamilton [Sigma-Aldrich] ) . Immediately after injection , the PTV port was ramped to 300°C at 450 °C/min , and the sample transferred to the column ( DB-5ms; 30 m × 0 . 25 mm , 0 . 25 μm film thickness ) at a flow of 1 ml/min . The oven temperature program was held at 35°C for 4 . 5 min , then ramped to 325°C at 20°C/min , and held at this temperature for 11 min . Helium was used as the carrier gas at a constant flow rate of 3 ml/min . For all samples , the MS transfer line was set to 325°C , and the MS operated in electron ionisation mode ( 70 eV; ion source 230°C; quadrupole 150°C , mass scan range 35–600 amu , with a detection threshold of 150 ) . Data acquisition was carried out using MassHunter Workstation , Data Acquisition software B . 07 . 01 ( Agilent Technologies ) . Analytes were detected by applying deconvolution algorithms to the total ion chromatograms of the samples ( MassHunter Workstation , Qualitative Analysis B . 07 . 00 [Agilent Technologies ) . Compound identification ( Table 1 ) was performed via manual interpretation using retention indices and spectral information , and the comparison of mass spectra to the Wiley 9th edition/NIST 11 combined mass spectral database ( National Institute of Standards and Technologies ) . As the molecular ion was not detectable for all analytes based on electronic ionisation , we in addition performed chemical ionisation on pools of 20 pupae in 100 µl n-pentane solvent with 0 . 5 µg/ml internal standards . The higher extract concentration was needed to counteract the loss in ionisation efficiency in chemical ionisation mode . A specialised chemical ionisation source with methane as the reagent gas was used with the MS , while the chromatographic method was the same as in electronic ionisation mode . Use of external standards ( C7-C40 saturated alkane mixture [Sigma Aldrich] ) enabled traceability of all peaks , and thus comparison to runs of single pupae extracts made in electronic ionisation mode . Modified Kovats retention indices for the peaks in question were calculated based on those standards . To further aid identification , we separated the substances based on polarity using solid phase extraction fractionation . For this purpose , pools of 20 pupae were extracted in 500 µl n-pentane containing 0 . 2 µg/ml internal standard , and separated on unmodified silica cartridges ( Chromabond SiOH , 1 ml , 100 mg ) based on polarity . Prior to use , the cartridges were conditioned with 1 ml dichloromethane followed by 1 ml n-pentane . The entire extraction volume was loaded onto the silica and the eluent ( fraction 1 , highly apolar phase ) collected . A wash with 1 ml pure n-pentane was added to fraction 1 . Fraction 2 contained all substances washed off the silica with 1 ml 25% dichloromethane in n-pentane , and finally a pure wash with 1 ml dichloromethane eluted all remaining substances ( fraction 3 ) . The polarity thus increased from fraction 1 through 3 , but no polar substances were found . All fractions were dried under a gentle nitrogen stream and re-suspended in 70 µl n-pentane followed by vigorous vortexing for 45 s . GC–MS analysis of all fractions was performed in electronic ionisation mode under the same chromatographic conditions as before . To quantify the relative abundances of all compounds found on each pupa , analyte-characteristic quantifier and qualifier ions were used to establish a method enabling automatised quantification of their integrated peak area relative to the peak area of the closest internal standard . For each analyte , the relative peak area was normalised , that is divided by the total sum of all relative peak areas of one pupa , to standardise all pupa samples . Only analytes , which normalised peak area contributed more than 0 . 05% of the total peak area , were included in the statistical analysis . We compared the chemical profiles of the pupae using a perMANOVA analysis of the Mahalanobis dissimilarities between pupae , with post hoc perMANOVA comparisons . Since there was no difference between cocooned and unpacked control pupae we combined them into a single control group for the final analysis ( perMANOVA: F = 1 . 09 , df = 23 , p=0 . 1 ) . We also performed a discriminant analysis of principle components ( Figure 2B ) to characterise the differences between the pupal treatments ( De Moraes et al . , 2014; Jombart et al . , 2010 ) . To identify the compounds that differ between treatments , we performed a conditional random forest classification ( n trees = 500 , n variables per split = 4 ) ( De Moraes et al . , 2014; Strobl et al . , 2009a; Strobl et al . , 2009b ) . Random forest identified nine compounds that were important in classifying the treatment group , of which eight were significant when analysed using separate KW tests ( results for significant compounds in Table 2 ) . We followed up the KW tests with individual post hoc comparisons for each significant compound ( Figure 2C–F , post-hoc comparisons in Table 2 ) . One millilitre aliquots of conidiospore ( 109/ml in 0 . 05% TX ) and blastospore ( 4 × 106/ml in PBS ) suspensions and approx . 500 mg of mycelia ( in 500 µl of autoclaved distilled water ) were washed three times by briefly vortexing and centrifuging the samples ( 5 min , 5500 g ) , discarding the supernatants , and replacing with 1 ml of autoclaved distilled water for the first two washes , and 500 µl for the last wash . One hundred and fifty microliters of the conidiospore and blastospore suspensions and 155 µg of hyphae ( n = 3 for each fungal stage ) were transferred into 1 . 5 ml glass vials ( La-Pha-Pack , Germany ) . All samples were centrifuged ( 2 min , 3000 g ) and dried under a nitrogen stream for 2 hr . Once samples were dry , 200 µl n-pentane containing internal standards ( as above ) was added to the samples , which were vortexed for 2 min . Samples were centrifuged ( 5 min , 5000 g ) and the supernatants transferred into 200 µl glass vials with inserts , and closed with aluminium crimper caps that had a silicone septum ( both La-Pha-Pack ) . Fifty microliters of the samples were injected into a pre-cooled PTV inlet at – 20°C and GC–MS analysis carried out following the above protocol . We determined if there was any overlap between the chemical profiles of pupae and the fungus by comparing the results of the fungal GC–MS analysis to the results of the pupal GC–MS . To that end , all fungal chromatograms were automatically de-convoluted and the mass-spectra of the compound peaks compared to a mass-spectral database , composed of the substances found on the pupae ( Agilent Technologies MassHunter Qualitative Analysis , B . 07 . 00 , 2014 ) . Twenty-seven compounds scored above 70 points , with 82 . 21 being the highest score . To determine if these peaks were identical to the peaks of the pupal samples , we calculated their Kovats retention time indices ( RI ) and compared them to that of the pupal substances . This analysis revealed that none of the fungal compound RIs were overlapping with the pupal compounds , hence confirming that the identified pupal substances are not of fungal origin . We injected β−1 , 3-glucans to test whether the changes in the chemical profile of infected pupae may be caused by an immune stimulation ( Vilcinskas and Wedde , 1997; Unestam and Söderhäll , 1977; Gunnarsson , 1988 ) . Soluble β−1 , 3-glucans were acquired by suspending 5 mg of Zymosan-A ( Saccharomyces cerevisiae cell wall fragments [Sigma-Aldrich] ) in 1 ml of sterile physiological ant saline ( as described in [Aubert and Richard , 2008] ) . The Zymosan suspension was vortexed for 1 hr at 3200 rpm before being centrifuged at 10000 rcf for 5 min . The supernatant that contains the soluble β-glucans ( Vilcinskas and Wedde , 1997 ) was then removed and stored at 4°C until use . As a control we used sterile ant physiological saline ( Aubert and Richard , 2008 ) . Pupae were artificially unpacked from their cocoons ( as above ) and placed gently into a sponge harness . Using fine glass capillaries ( with spike to aid injection; inner diameter = 25 µm [BioMedical Instruments , Germany] ) , a microinjector ( parameters: pi = 120 hPa , ti = 0 . 3 s , pc = 20 hPa [FemtoJet , Eppendorf , Germany] ) and a micromanipulator ( Luigs and Neumann , Germany ) , we injected 46 nl of the β-glucan solution or ant physiological saline through the pupae’s first tergite , into their haemocoel . We cleaned the capillaries between injections using 96% ethanol . Half of the pupae were frozen at – 80°C immediately after injection whilst the remainder were kept alone in individual plaster dishes for a further 48 hr , before then also being frozen . Frozen pupae were then used for molecular and chemical analyses ( below ) . We employed a candidate gene approach to test if a β−1 , 3-glucan injection ( above ) elicits an immune response in pupae , with saline injected pupae as a control ( n = 11 , each for pupae frozen 0 hr and 48 hr after injection , for both saline and β-glucan treatments ) . Total RNA was extracted from pupae using the Maxwell RSC simply RNA tissue kit ( Promega , Madison , Wisconsin ) according to manufacturer’s instructions , with a final elution volume of 60 µl . Reverse transcription was performed using the iScript cDNA synthesis kit ( Bio-Rad , Hercules , California ) as per the manufacturer’s recommendations . Primer sequences were taken from ( Konrad et al . , 2012 ) or developed from cDNA sequence information of L . neglectus ( Table 4 ) . Gene expression analyses of 28S Ribosomal Protein S18a ( used as housekeeping gene , which we had previously found to be stably expressed in pupae ) , Prophenoloxidase ( proPO ) , Peptidoglycan Recognition Protein SC2 ( PGRP-SC2 ) and β−1 , 3-glucan binding protein ( β−1 , 3-GBP ) were performed in 20 μl reaction volumes using KAPA SYBR Fast qPCR master mix ( Kapa Biosystems , Wilmington , Massachusetts ) and 0 . 2 μM each of specific primers ( Sigma-Aldrich ) on a Bio-rad CFX96 real-time PCR detection system . Two microliters of the cDNA sample were added per reaction and each sample was analysed in duplicate or triplicate wells . Each run contained an absolute negative as well as a no reverse transcription control . Primer efficiency was >95% for all primer sets using standard curves of 10-fold dilutions , and primer specificity was monitored based on a melting curve analysis following each run . We used the following program for amplification: 95°C for 5 min , followed by 40 cycles of 10 s of 95°C denaturation and 30 s of 60°C ( 55°C ) annealing/extension . Normalised gene expression values ( the average of technical replicates standardised to the housekeeping gene ) were analysed using Mann-Whitney U tests and the resulting p values were corrected for multiple comparisons . The chemical profiles of β−1 , 3-glucan and saline-injected pupae were analysed using GC–MS , following the above protocols . We tested whether the four CHCs that were increased specifically on unpacked pupae were also increased on pupae 48 hr after injection with either β-glucans or saline . We used LMERs to compare the CHC abundances within treatments , on pupae immediately after injection ( saline , n = 27; β-glucan , n = 26 ) and 48 hr later ( saline , n = 22; β-glucan , n = 22 ) , correcting the p values for multiple comparisons across the CHCs . CHC abundances were square root transformed to a normal distribution . Since this experiment was carried out on 2 separate days , run was included as a random intercept effect to account for any potential , uncontrollable differences . The assumption of homogeneity of variances was violated for the LMER analysing Tritriacontadiene , though visual inspection of the model residuals found this violation to be relatively minor . Still , to test for the robustness of the LMER result , we also analysed Tritriacontadiene using a non-parametric test that does not make any assumptions about data distribution , but is unable to account for the random effect . This test also found a strong , significant difference between the two time points ( Mann-Whitney U test , U = 157 , p=0 . 007 ) ; hence we report the result of the LMER . To test if destructive disinfection prevents Metarhizium from successfully replicating , we kept single pathogen-exposed pupae in petri dishes containing groups of 3 or 8 ants . This allowed us to assess how group size affects the likelihood of fungal inhibition . For the following 10 days , we observed the pupae for unpacking . When a pupa was unpacked , we left it with the ants for a further 1 or 5 days so that they could perform destructive disinfection . This allowed us to assess how the duration of destructive disinfection affects the likelihood of fungal inhibition . The destructively disinfected pupae were then removed and placed into petri dishes on damp filter paper at 23°C ( 8 ants 1 day and 5 days , n = 22 pupae each; 3 ants 1 and 5 days , n = 18 pupae each ) . We did not surface sterilise the pupae as this might have interfered with the destructive disinfection the ants had performed . Removed pupae were observed daily for Metarhizium sporulation for 30 days . To determine how many pupae sporulate in the absence of destructive disinfection , we kept pathogen-exposed pupae without ants as a control and recorded the number that sporulated for 30 d ( n = 25 ) . We compared the number of pupae that sporulated after 1 and 5 days and in the absence of ants using logistic regressions and Tukey post hoc comparisons , separately for the two ant group sizes ( Figure 3A , Figure 3—figure supplement 1 ) . We examined the individual effects of unpacking , biting and poison application on destructive disinfection by performing these behaviours in vitro . Pathogen-exposed pupae were initially kept with ants so that they could perform sanitary care . After 3 days , we removed the pupae and split them up into three groups: ( i ) pupae that we left cocooned , ( ii ) experimentally unpacked and ( iii ) experimentally unpacked and bitten . We simulated the damage the ants achieve through biting by damaging the pupal cuticle and removing their limbs with micro scissors . The pupae were then treated with either synthetic ant poison ( 60% formic acid and 2% acetic acid , in water; applied at a dose equivalent to what ants apply during destructive disinfection; Figure 3—figure supplement 2 ) or autoclaved distilled water as a control , using pressurised spray bottles ( Lacor , Spain ) to evenly coat the pupae in liquid . Spraying was carried out at a distance of 36 cm from the pupae and lasted for 1 s . The pupae were allowed to air dry for 5 min before being rolled over and sprayed again and allowed to dry a further 5 min . All pupae were then placed into separate petri dishes and monitored daily for Metarhizium sporulation ( cocooned + poison , n = 24; unpacked + poison + biting , n = 24; all other treatments , n = 25 ) . The number of pupae sporulating was analysed using a logistic regression with Firth’s penalised likelihood , which offers a solution to the monotone likelihood caused by the complete absence of sporulation in one of the groups ( R package ‘brglm’ [Kosmidis , 2013] ) . Pupal manipulation ( cocooned/unpacked only/unpacked and bitten ) , chemical treatment ( water or poison ) and their interaction were included as main effects ( Figure 3B , Figure 3—figure supplement 3 ) . We followed up this analysis with Tukey post-hoc comparisons ( Table 3 ) . We confirmed that synthetic poison spraying resulted in pupae receiving an amount of poison within the natural range that is applied by ants during destructive disinfection . Pupae taken from a stock colony were experimentally unpacked and sprayed with synthetic poison . We then measured their pH ( all as above; n = 21 ) . To test if synthetic poison spraying was similar to natural ant spraying , we compared their pH to pupae destructively disinfected by ants ( data from Figure 1—figure supplement 3 ) using a Mann-Whitney U test ( Figure 3—figure supplement 2 ) . We adjusted the p value to correct for using this dataset twice ( here and in Figure 1—figure supplement 3 ) . To test if the pupal cocoon limits the amount of the ants’ poison that reaches the pupae inside , we took pupae from a stock colony and sprayed half with synthetic ant poison ( as above; n = 10 ) and left the other half untreated ( n = 10 ) . We then unpacked these pupae and measured their pH ( as above ) . As an additional control , we first experimentally unpacked pupae before spraying them with synthetic poison ( n = 10 ) . We analysed pH pupae using a KW test with post hoc comparisons ( Figure 3—figure supplement 4 ) . We tested the impact of destructive disinfection on disease transmission within groups of ants by keeping them with sporulating pupae or pupae that had been destructively disinfected . Infections were established in pupae ( as above ) and half were allowed to sporulate ( n = 11 ) , whilst the other half were experimentally destructively disinfected ( as above; n = 11 ) . Pupae were then kept individually with groups of five ants in mini-nests ( cylindrical containers [Ø=90 mm] with a second , smaller chamber covered in red foil [Ø=33 mm] ) . Ant mortality was monitored daily for 30 days . Dead ants were removed , surface sterilised ( as above ) and observed for Metarhizium sporulation . The number of ants dying from Metarhizium infections in each treatment was compared using a logistic regression ( Figure 4A ) . Mini-nest identity was included as a random intercept effect as ants from the same group are non-independent . To confirm that our findings are not an idiosyncrasy of our specific study population , we tested whether the destructive disinfection observed in the Seva L . neglectus population is also found in another , genetically distinct supercolonial population ( Ugelvig et al . , 2008 ) and a related ( congeneric ) , non-supercolonial/non-invasive species , Lasius niger . We sampled hundreds of queens and many thousands of workers from a 300 m2 area of L . neglectus in the botanical gardens in Jena , Germany ( 50°55'54 . 6"N 11°35'08 . 4"E ) . The studied L . niger colony was raised from a single founding queen collected after a natural mating flight in Harpenden , UK ( 51°48'48 . 9"N 0°22'51 . 5"W ) and reared in the laboratory for 3 years , by which point it contained several hundred workers . To test if these ants also perform destructive disinfection , we kept two workers with single , pathogen-exposed or control-treated pupae ( following the same protocols as above; Jena supercolony , n = 23 replicates per treatment; L . niger , n = 20 per treatment ) . We observed the ants on a daily basis to record the occurrence of destructive disinfection for 10 d . In both the Jena population and L . niger , no control-treated pupae were destructively disinfected ( proportion ± 95% CIs: Jena = 0 ± 0–0 . 14; L . niger = 0 ± 0–0 . 16 ) , whilst >60% of the pathogen-exposed pupae were destructively disinfected ( proportion ± 95% CIs: Jena = 0 . 61 ± 0 . 41–0 . 78; L . niger = 0 . 95 ± 0 . 76–0 . 99 ) .

Ants live in crowded societies where disease can spread rapidly and take a heavy toll on the community . Ants have a number of ways to prevent these outbreaks before they become a problem . Like many other social species , they practice good hygiene and groom nest mates that have picked up a pathogen , which helps them to recover and to reduce the likelihood of the disease spreading . Unlike other social species , ants appear to have evolved collective disease defence , or social immunity , because their colonies behave like a ‘superorganism’ , in which the society behaves much like a single organism would . Like an individual animal that has an infection , the colony needs to be able to eliminate infections collectively when a nest mate falls ill , to prevent the disease from spreading . To understand how an ant colony protects itself when the care fails and a colony member contracts a lethal infection , Pull et al . infected the brood of the invasive garden ant with a common soil fungus . Using a combination of chemical analyses and behavioural observations , it was shown that the infected pupae emitted a chemical cue , which the tending ants could detect . Using a microscopic camera , Pull et al . found that when the ants sensed the cue , they would unpack the infected pupae from their cocoons and bite them . They then sprayed them with an antiseptic poison , which entered the hole in the pupae’s body , killing both the pupae and the fungus inside , before it had a chance to spread . This process of destructive disinfection may seem like a large sacrifice , but it helps to protect the rest of the colony from a fungus that could lead to much greater damage . The tending ants were acting within the superorganism of the colony much like immune cells act within an individual’s body – honing in on infected cells and destroying them before the pathogen can spread to other cells . This suggests that the ability to detect and destroy harmful elements was necessary for both the evolution of multicellular organisms , and from single animals to superorganisms .

2018

Destructive disinfection of infected brood prevents systemic disease spread in ant colonies

The populations of light-demanding trees that dominate the canopy of central African forests are now aging . Here , we show that the lack of regeneration of these populations began ca . 165 ya ( around 1850 ) after major anthropogenic disturbances ceased . Since 1885 , less itinerancy and disturbance in the forest has occurred because the colonial administrations concentrated people and villages along the primary communication axes . Local populations formerly gardened the forest by creating scattered openings , which were sufficiently large for the establishment of light-demanding trees . Currently , common logging operations do not create suitable openings for the regeneration of these species , whereas deforestation degrades landscapes . Using an interdisciplinary approach , which included paleoecological , archaeological , historical , and dendrological data , we highlight the long-term history of human activities across central African forests and assess the contribution of these activities to present-day forest structure and composition . The conclusions of this sobering analysis present challenges to current silvicultural practices and to those of the future . Central African forests underwent an unequal history of disturbances during the Holocene ( after 10 , 000 yrs BP ) compared with Neotropical forests , which remained relatively stable since the Late Glacial Maximum ( LGM , ca . 13 , 000–10 , 000 yrs BP ) ( Anhuf et al . , 2006 ) . Over the last three millennia , significant changes in the vegetation structure and floristic composition were caused by climate fluctuations ( Maley et al . , 2012; Neumann et al . , 2012; Lézine et al . , 2013 ) . Specifically , a dry event around 2500 ya caused forest fragmentationan event with a more pronounced seasonality occurred around 2500 ya and caused forest fragmentation , and this fragmented forest included patches of savanna ( Maley , 2002 ) . This dry episode stopped around 2500 BP , as evidenced from the Mopo Bai site in the Republic of the Congo , where Poaceae pollen severely dropped from 36% to 13% between 2580 and 2400 BP , which is evidence for a retreat of the savannas to the benefit of the forests ( Bostoen et al . , 2015 ) . After 2000 yrs BP , a relatively wet climate in central Africa favored forest recolonization by light-demanding tree species , with few effects imputable to humans ( Maley et al . , 2012; Lézine et al . , 2013; Brnčić et al . , 2009; Bostoen et al . , 2015 ) . The subsequent climatic variations were less important with little effect on the vegetation ( Oslisly et al . , 2013a ) ; however , human activities are assumed to have increased in importance , particularly during the most recent centuries ( Oslisly et al . , 2013a , 2013b; Willis et al . , 2004; Brnčić et al . , 2007; Greve et al . , 2011 ) . The abundance of direct ( artifacts ) and indirect evidence ( charred oil palm endocarps ) in soils confirms the non-pristine nature of central African forests ( Morin-Rivat et al . , 2014 ) . Human activities in the Holocene , and particularly shifting cultivation , have been invoked to partially explain the low diversity of central African forests ( Parmentier et al . , 2007 ) and the abundance of light-demanding species in the canopy ( White and Oates , 1999; van Gemerden et al . , 2003; Engone Obiang et al . , 2014; Vleminckx et al . , 2014; Biwolé et al . , 2015 ) . The light-demanding species form , in some places , almost pure 0 . 5 to 1 ha stands that mirror the size of traditionally cultivated fields ( van Gemerden et al . , 2003 ) . An example is the Sangha River Interval ( SRI ) in which the vegetation currently forms a ‘corridor’ of old-growth semi-deciduous Celtis forests ( Fayolle et al . , 2014a; Gond et al . , 2013 ) , with local variations caused by the geological substrate or the forest degradation along roads and close to cities ( Fayolle et al . , 2012 ) ( Figure 1 ) . The SRI is a 400-km-wide region , with low endemism between the Lower Guinean and the Congolian subcenters of endemism ( White , 1983 ) . This area , which is between southeastern Cameroon , southern Central African Republic and northern Congo , may have been a savanna corridor 2500 ya ( Maley , 2002 ) . Until the recent studies of Harris ( 2002 ) , and Gillet and Doucet ( 2012 ) , the vegetation in the SRI was under sampled , and whether the origin of this corridor is environmental ( Fayolle et al . , 2012 ) or historical ( Morin-Rivat et al . , 2014 ) remains to be explored . In this study , we assessed the potential impact of historical human activities on central African forests . Specifically , we analyzed the population/age structure of four primary light-demanding timber species across the SRI and examined the synchronism with the paleoenvironmental , archaeological , and historical data in this region ( Figure 1 ) . 10 . 7554/eLife . 20343 . 003Figure 1 . Paleoenvironmental changes and human activities in the Sangha River Interval . The 34 sites with paleoenvironmental data ( fires ) and the 38 dated archaeological sites and discoveries ( pots ) are indicated on a vegetation map modified from Gond et al . ( 2013 ) ( http://www . coforchange . eu/products/maps ) . The seven sites used to monitor tree growth ( trees ) are also indicated ( see Supplementary files 4 , 5 and 6 for site names ) . Brown ( three shades ) : savanna of the Sudano-Guinean domain; orange ( three shades ) : savanna included in dense forest; yellow: savanna-forest edge; purple ( two shades ) : very open forest; blue-green: open semi-deciduous forest; medium green ( three shades ) : dense semi-deciduous forest; dark green ( five shades ) : dense evergreen forest; light green ( two shades ) : open evergreen forest; light blue ( two shades ) : swamp forest and swamp . Map: QGIS 2 . 14 ( http://www . qgis . org ) , CAD: Illustrator CS4 ( https://www . adobe . com ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20343 . 003 The 1 , 765 , 483 inventoried trees were studied at the genus level , and included 176 genera ( Supplementary file 1 ) . The five most represented genera were Celtis ( Ulmaceae ) , Polyalthia ( Annonaceae ) , Strombosia ( Olacaceae ) , Petersianthus ( Lecythidaceae ) , and Manilkara ( Sapotaceae ) . Most of the genera included shade-bearers ( n = 71 genera ) , which were followed by the pioneers ( n = 47 ) , and the non-pioneer light-demanding species ( NPLD , n = 37 ) . We had no information for 21 genera . Regarding leaf phenology , 108 genera were evergreen , versus 50 deciduous . No information was available for 16 genera . Wood density ranged from 0 . 22 g . cm−3 for Ricinodendron ( Euphorbiaceae ) to 0 . 88 g . cm−3 for Bobgunnia ( Fabaceae ) . Mean density was 0 . 58 g . cm−3 . Mean diameters ranged from 31 . 62 cm to 93 . 46 cm in dbh for Meiocarpidium ( Annonaceae ) and Autranella ( Sapotaceae ) , respectively , with a mean for all genera of 47 . 45 cm in dbh . Mean basal area ranged from 0 . 12 m² to 0 . 92 m² for Lasiodiscus ( Rhamnaceae ) and Ceiba ( Malvaceae ) , respectively , with a mean for all genera of 0 . 30 m² . Among the inventoried trees , we identified two groups of genera: ( i ) those that showed a reverse-J shape distribution ( Figure 2 , and Supplementary file 1 ) with many small and young trees ( most of the genera , n = 134 , 76% ) , and ( ii ) those for which distributions deviated from this pattern ( n = 42 , 24% ) , including flat ( e . g . , Baillonella ) and unimodal distributions of diameter . Among these , we identified four primary canopy genera ( i . e . , Erythrophleum and Pericopsis ( Fabaceae ) , Terminalia ( Combretaceae ) , and Triplochiton ( Malvaceae ) ) with unimodal diameter distributions ( Figure 2—figure supplement 1 ) . 10 . 7554/eLife . 20343 . 004Figure 2 . Variation in tree diameter distribution among the 176 genera across the SRI . Projection of the genera and the 10-cm-wide diameter classes in the ordination space defined by the first two axes of a correspondence analysis of the abundance matrix , as defined by 176 genera and 13 diameter classes . The size of the circles is proportional to the square root of the genus abundance . The color of the symbol corresponds to the two groups identified with a clustering analysis ( based on Euclidean distances and an average agglomeration method ) on the species score on the first factorial axis . Genera that showed a reverse-J diameter distribution ( n = 134 ) are indicated in gray and those genera that showed a deviation from the reverse-J distribution ( n = 42 ) in black ( e . g . , Baillonella ) . Black filled circles indicate the four genera that are monospecific in the SRI and used for the age estimations . Diameter distribution of the 10 most abundant genera is shown in addition to that of the four selected genera: Celtis ( gray ) , Polyalthia ( gray ) , Strombosia ( gray ) , Petersianthus ( gray ) , Manilkara ( gray ) , Entandrophragma ( black ) , Terminalia ( black ) , Anonidium ( gray ) , Staudtia ( gray ) , and Macaranga ( gray ) . Statistics: R ( https://www . r-project . org/ ) , CAD: Illustrator CS4 ( https://www . adobe . com ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20343 . 00410 . 7554/eLife . 20343 . 005Figure 2—figure supplement 1 . Distribution of diameters of the four study species in the 22 study sites ( black ) . The solid lines on the average diameter distributions ( gray ) correspond to the fitted parametric diameter distribution ( Weibull ) , which was used to estimate the mode and corresponding age . Ages and dates ( AD ) in the Sangha River Interval were estimated using the mean annual increment of diameter of ( a ) 367 Erythrophleum suaveolens , ( b ) 199 Pericopsis elata , ( c ) 152 Terminalia superba and ( d ) 265 Triplochiton scleroxylon . Age estimations were validated with published tree-ring data for these four species in natural forests in Cameroon ( C ) , the Democratic Republic of Congo ( DRC ) and the Ivory Coast ( IC ) ( Supplementary file 3 ) . Maps: QGIS 2 . 14 ( http://www . qgis . org ) , CAD: Illustrator CS4 ( https://www . adobe . com ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20343 . 00510 . 7554/eLife . 20343 . 006Figure 2—figure supplement 2 . Growth models ( a , c , e and g ) and growth trajectories ( b , d , f and h ) for the four study species based on tree-ring data . Eight growth models ( i . e . , Canham , Gompertz , Verhulst , Power , Power mult , Lognormal , Linear and Mean ) were fitted to the data . Age estimations were obtained with numerical solutions to ordinary differential equations ( ODE ) . Solid lines correspond to the model prediction on the observed range of diameters . Dashed lines correspond to the predictions for small trees ( below the inventory threshold of ≤10 cm ) . For growth models , the color of symbols indicates access to light ( light gray = emergent and dominant trees , Dawkins 5; gray = canopy and codominant trees , Dawkins 4; black = lower canopy and understory dominated trees , Dawkins ≤3 ) . For growth trajectories , age/diameter values based on published tree-ring data are indicated in light gray . Statistics: R ( https://www . r-project . org/ ) , CAD: Illustrator CS4 ( https://www . adobe . com ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20343 . 006 These genera are monospecific in the SRI ( Pericopsis elata , Terminalia superba , Erythrophleum suaveolens , and Triplochiton scleroxylon ) , and share similar functional traits ( i . e . , deciduous , emergent , pioneer light-demanding trees ) . Combined , these four species represented 4 . 3% of the inventoried trees , reaching a maximum of 8 . 62% in one site in Cameroon . The dbh ranged from 10 . 6 cm ( T . superba ) to 151 . 6 cm ( E . suaveolens ) ( Supplementary file 2 ) . The mode of the diameter distribution differed between the four studied light-demanding species , with 65 . 3 cm for P . elata , 69 . 8 cm for T . superba , 72 cm for E . suaveolens , and 90 . 3 cm for T . scleroxylon . Weibull distributions indicated modes comprised between 65 . 3 cm in dbh for P . elata , and 90 . 3 cm in dbh for T . scleroxylon . The modes for T . superba and E . suaveolens were 69 . 5 cm and 72 cm in dbh , respectively . Four studies provided growth and age data , which were based on tree-ring analysis ( Supplementary file 3 ) . We found data for 83 discs ( P . elata = 24; T . superba = 41; T . scleroxylon = 18 ) from four locations in the Democratic Republic of Congo , the Ivory Coast , and Cameroon . Data for E . suaveolens were not available . Mean ring width ranged from 0 . 298 ± 0 . 54 cm for P . elata to 0 . 719 ± 0 . 267 for T . superba . It was 0 . 620 ± 0 . 28 cm for T . scleroxylon . In the study sites , the MAId of the monitored trees ranged from 0 . 44 ± 0 . 033 cm/y for E . suaveolens ( 367 stems ) to 0 . 58 ± 0 . 061 cm/y for the fast-growing T . scleroxylon ( 265 stems ) . It was 0 . 45 ± 0 . 026 cm/y and 0 . 53 ± 0 . 112 cm/y for P . elata ( 199 stems ) and T . superba ( 152 stems ) , respectively ( Supplementary file 2 ) . Results of tree modeling ( Figure 2—figure supplement 2 , and Supplementary file 4 ) indicated that the Canham model was the best model to explain tree growth in E . suaveolens ( BIC = 196 . 6 ) , T . superba ( BIC = 256 . 1 ) , and T . scleroxylon ( BIC = 372 . 1 ) , whereas only the Mean model best explained tree growth in P . elata ( BIC = −99 . 1 ) . The performance of the models remained , however , very low . According to the age data from published tree-ring studies ( Supplementary file 3 ) , we found that estimations based on mean growth were likely to be more reliable than those based on growth models ( Figure 2—figure supplement 2 ) . In particular , the performances of the Canham and Lognormal models were low , as well as , to a lesser extent , that of the unimodal distributions ( sigmoidal growth trajectory ) . Based on mean growth estimates , the age of the canopy trees was only a few centuries , with a mode dated to between 142 and 164 ya , which corresponds to the years AD 1836 and 1858 ( mean AD 1850 ) ( Figure 2 ) . Climate of the last 1000 years was documented by sea surface temperatures ( SSTs ) and the atmospheric dust signal from the marine core ODP 659 , taken off the West African coast , and sediments from Mopo Bai and Goualogou Lake in the Republic of the Congo . Climate oscillated between wet and dry periods ( Figure 3 , and Supplementary file 5 ) . Typically , climate was dry until ca . AD 1200 , between AD 1250 and 1450 , and since AD 1850 , with intermediate wet periods , in particular a long one between ca . AD 1450 and 1850 . 10 . 7554/eLife . 20343 . 007Figure 3 . Chronology of paleoenvironmental changes and human activities in the Sangha River Interval . We compiled data on climate , erosion , vegetation types , light-demanding species and paleofires for the last 1000 years from 34 paleoenvironmental sites and data from 38 dated archaeological sites and discoveries with 63 related radiocarbon dates . The summed probability distribution of the radiocarbon ages showed fluctuations in the signal of human activities through time . Three primary time periods were identified: ( a ) before AD 1300; ( b ) from AD 1300 to 1850; and ( c ) after AD 1850 . Color scales ( four levels ) were assigned depending on the proxy influx on the curve: light = present but rare; light-medium = present; medium = frequent; and dark = very frequent . Abbreviations: AD = Anno Domini ( = calendar dates ) ; C = Cameroon; RC = Republic of the Congo; CAR = Central African Republic; SSTs = Sea Surface Temperatures; C3/C4 plants = woody species ( below −20‰ ) /herbs ( above −20‰ ) ; E . guineensis = the oil palm Elaeis guineensis ( Supplementary files 5 and 6 ) . CAD: Illustrator CS4 ( https://www . adobe . com ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20343 . 007 The erosion curve included data related to grain size and chemical elements from the banks of the Nyong , Boumba , Dja , Ngoko rivers in Cameroon and in the Republic of the Congo , and the Mbaéré valley and the Sadika alluvial fan in the Central African Republic . This signal did not overlap climate data , as erosion was high between ca . AD 1350 and 1950 , with a slight drop dated to between AD 1500 and 1650 . The history of vegetation change derived from δ13C values obtained at the same sites as those documenting erosion . Results indicated two main time periods: a first one until ca . AD 1200 dominated by forest vegetation ( C3 dominant , values >25‰ ) , and a second one from ca . AD 1600 until today dominated by grass cover ( C4 dominant , values <25‰ ) . Pollen data of light-demanding species ( i . e . T . scleroxylon , E . suaveolens , Macaranga spp . , Myrianthus/Musanga type and E . guineensis ) were obtained at Lake Télé , Mopo Bai and Goualogou Lake in the Republic of the Congo , and at sites in the northern Lobaye in the Central African Republic . They were more present between ca . AD 1300 and 1400 , then between ca . AD 1600 and 1850 . Paleofires were documented by macro- and microcharcoal data from Mopo Bai and Goualogou Lake in the Republic of the Congo , and the Lobaye area ( in the Rep . of the Congo and the Central African Republic ) . Indicators of paleofires slightly increased between ca . AD 1300 and 1400 . They were more substantial , however , between ca . AD 1550 and 1850 . Evidence of human activities was identified during two main periods: the first one around AD 1000 , and the second between ca . AD 1400 and 1850 ( Figure 3 , and Supplementary file 6 ) . The main discoveries comprised potsherds associated with settlements , iron slags and tuyères related to iron smelting , or were located in places where salt was exploited ( e . g . Ngoko River ) . Most of the artifacts were found between ca . AD 1600 and 1800 . Focusing on the pottery only , dates were distributed into three periods: ( i ) between ca . AD 800 and 1100 , ( ii ) between AD 1300 and 1600 , and ( iii ) between AD 1700 and 1800 . Smelting activities were documented at a few sites only , especially in the southern Central African Republic ( i . e . Bagbaya , Ngara , and Lingbangbo ) , which were in use during short time periods: between ca . AD 1000–1100 , AD 1300–1400 , AD 1500–1700 , and AD 1700–1900 . The results of the Bayesian analysis of the radiocarbon dates indicated a weak radiocarbon signal until ca . AD 1200 , which increased from ca . AD 1200 ( Figure 3 , and Supplementary file 7 ) . Main peaks were centered on ca . AD 1350 , 1550 , and 1750 . The signal strongly decreased after ca . AD 1800 , with a last small peak around AD 1950 related to the nuclear activities of the mi-twentieth century . Key events emerged within the historical chronology ( Supplementary file 8 ) . Firstly , the Triangular Trade , and particularly the period between AD 1400 and 1600 , profoundly destabilized the area . During the following centuries , the slave-raiding , leaded by the Fulbe people , pushed other populations to flee southward in the forest . The second key event is the beginning of the colonization of Africa , which put a stop to the Fulbe’s activities . The exploration of the SRI that began after AD 1875 , and the permanent presence of the European colonists since then , deeply disturbed the spatial distribution of the local populations , as well as their activities ( e . g . enrolment in the concession companies , education , diseases , etc . ) . During this period , the conflicts that opposed France and Germany ( i . e . the 1870 War , and the First and Second World Wars ) were also transferred to the African territories . Finally , the region experienced a massive rural exodus since the 1930s , which was amplified since the independences ( Cameroon , Republic of the Congo , and Central African Republic the same year: 1960 ) . The reverse-J-shape distribution of diameters , characteristic of most genera , is typical of ‘active’ tree populations with many small and young trees ( Figure 2—figure supplement 1 ) . By contrast , the unimodal distribution of diameters for could represent a generalized limited number of young trees ( i . e . , a lack of regeneration ) and indicate the widespread decline of the tree populations . This type of distribution was characteristic of four primary canopy genera ( i . e . , Erythrophleum and Pericopsis ( Fabaceae ) , Terminalia ( Combretaceae ) , and Triplochiton ( Malvaceae ) ) , which we studied further . Historical factors were previously invoked to explain such distributions for E . suaveolens and T . superba in eastern Cameroon ( Durrieu de Madron and Forni , 1997 ) . Similarly , a unimodal distribution of diameters was reported for the light-demanding timber species Aucoumea klaineana in Gabon , which could not be explained only by demography . Ontogenic variations in growth are well described for tropical tree species , and unimodal growth trajectories are widely reported ( Hérault et al . , 2011 ) . The low performance of the models to estimate tree age is explained by the slow growth of young trees ( dbh ≤10 cm ) and the great uncertainty regarding the time a tree remains in the small diameter classes ( Figure 2—figure supplement 2 , and Supplementary file 4 ) . Indeed , a linear relationship between tree diameter and age is acceptable for tropical tree species of a larger size ( Worbes et al . , 2003 ) . Most suppressed individuals were destined to die , and therefore , only the trees with vigorous growth are able to reach the canopy and could be thus included in this type of analysis . The tree-ring approach , including information for the growth of small trees , remains therefore essential for age estimation ( Worbes et al . , 2003 ) , but studies are only sporadic for central African forests . Based on mean growth estimates , canopy trees in the SRI were aged to only a few centuries , with a mode dated to AD 1850 in average . This age range is consistent with the estimated ages of canopy trees in Nigeria ( van Gemerden et al . , 2003 ) and in Cameroon ( Worbes et al . , 2003 ) . Moreover , the population decline of A . klaineana in Gabon is attributed to a shift in the disturbance regime two to three centuries ago ( Engone Obiang et al . , 2014 ) . The argument for a regional trend is supported by these age estimates and the general pattern we reported across the SRI . We assumed that the unimodal population/age structure of the light-demanding tree species was linked to the recent human history . Specifically , we postulated that the decrease in anthropogenic disturbances and the generalized land abandonment from ca . 165 ya were less favorable to the regeneration of light-demanding tree species ( van Gemerden et al . , 2003; Brnčić et al . , 2007; Greve et al . , 2011; Biwolé et al . , 2015 ) . Additionally , the present-day natural gap size has been shown to be insufficient for the regeneration of most of these species ( van Gemerden et al . , 2003 ) . All proxies converged toward the identical regional history that is divided into three primary periods: ( i ) a dry period between AD 950 and 1300 with almost no human activity recorded; ( ii ) a wet period between AD 1300 and 1850 with large-scale human activities and a high disturbance regime that led to a forest-savanna mosaic; and ( iii ) a forest aging period from AD 1850 to the present . The aging period corresponded to a shift in the disturbance regime that was most likely caused by a depopulation of the forest with the beginning of the European colonization ( Robineau , 1967; Copet-Rougier , 1998; Coquery-Vidrovitch , 1998 ) . The first time period before AD 1300 corresponds to a dry climate , consistent with the higher latitude Medieval Warm Period ( DeMenocal et al . , 2000 ) , with only scarce pollen of pioneer and light-demanding species ( Brnčić et al . , 2007 , 2009 ) . The vegetation was composed of forest tree species according to δ13C values between −30 . 6 and −25 . 8‰ ( Sangen , 2012 ) . In southeastern Cameroon , alluvial records indicate a growing human impact on forests between AD 1000 and 1200 , particularly because of shifting cultivation and the associated increase in erosional processes ( Sangen , 2012; Runge et al . , 2014 ) . Charcoal ( related to natural fires and anthropogenic burning ) in lake sediments and soils were recorded only at the end of this period , which corresponds to the end of the hiatus phase in human activities ( massive depopulation ) previously documented for central Africa ( Oslisly et al . , 2013a , 2013b; Wotzka , 2006 ) and specifically for the SRI ( Morin-Rivat et al . , 2014 ) . Surveys have been carried out , but the poor surface visibility can underestimate true human presence ( see Morin-Rivat et al . , 2016 , for an example of methodology ) . Sometimes no archaeological research have been carried out in certain regions covered by dense forest ( B . Clist , pers . comment ) . Notably , at this time , human populations were only indicated at a few sites that were dedicated to iron metallurgy ( southern CAR near Bangui and Nola and the site of Ngombé in the Rep . of the Congo , approximately AD 1300 ) and to salt exploitation ( Ngoko River , approximately AD 1000 ) . From AD 1300 to 1850 , pollen sequences indicated a relatively wet climatic period . Nevertheless , burning increased , and this burning is attributed to human activities because the moisture content of the vegetation was too high for fires to often occur naturally ( Brnčić et al . , 2007 , 2009; Vennetier , 1963 ) . The Mbaéré valley and the Sadika alluvial fan ( Gadzi-Carnot sandstones in CAR ) recorded intensive erosion and relatively high δ13C values after AD 1200 , indicating forest regression and the formation of a forest-savanna mosaic ( Sangen , 2012; Runge et al . , 2014 ) . In southeastern Cameroon , the anthropogenic erosion culminated at approximately AD 1200–1400 ( Sangen , 2012; Runge et al . , 2014 ) . The decrease in the run-off with an increased rate of sedimentation between AD 1400 and 1600 corresponds to the climatic period of the Little Ice Age ( Brnčić et al . , 2007 , 2009 ) , in combination with an increase in the frequency of El-Niño events between AD 1200 and 1500 ( Sangen , 2012 ) . Since then , despite a more humid period following the Little Ice Age , maximal incidence of human activities have been recorded in the SRI , which opened the forest cover and favored the pioneers . Nonetheless , we must remain cautious regarding the interpretation of the archaeological data , as there is a huge gap of knowledge in the SRI , especially in the area between Souanké and Berberati ( Figure 1 ) . In the state of the art , it is not possible to interpret the spatial distribution of human settlements and activities . In particular , iron-smelting sites are only few , they are concentrated in the southern Central African Republic , and were in use during short periods . The volume of charcoal used and , by extension , the associated deforestation , should have been important for feeding the furnaces , as shown by Pinçon ( 1990 ) . However , the debate is still alive about estimating the volume of wood needed for metallurgy ( Lupo et al . , 2015 ) , compared to the volume of trees logged for shifting agriculture ( Goucher , 1981 ) . The period of ca . AD 1850 to the present marked a decrease in the disturbance regime ( Figure 3 ) . The pollen of naturally grown oil palms and pioneer trees became rare or absent . In southeastern Cameroon and in the CAR , pollen , phytoliths , soil charcoal and δ13C values indicate little disturbance during the past 100–150 years , with the recolonization of the savannas by forest trees ( Runge et al . , 2014; Lupo et al . , 2015 ) . The anthropogenic burning persisted , as indicated by charcoal particles found in sites located along rivers ( Brnčić et al . , 2007 , 2009; Tovar et al . , 2014 ) , which might document either the colluvium of charcoals downslope or the concentration of human activities on the riverbanks . During this period , less evidence of human activities is reported ( Oslisly , 2013b; Morin-Rivat , 2014 ) . In the 1960s , young secondary forests ( i . e . , with Musanga cecropioides ) constituted only 1% of the forest types and were located along the main roads ( Vennetier , 1963 ) . Despite the drying of the 20th century , confirmed by low flow regimes in the primary rivers , the Sangha , Ubangui , Lobaye , and Likwala-aux-Herbes ( Runge and Nguimalet , 2005; Aleman et al . , 2013 ) , and the increase in anthropogenic activities in recent years ( e . g . , mining , industrial logging from the 1970s , burning , and cultivation ) that induced very localized , degraded landscapes ( Laporte et al . , 2007; Sangen , 2012; Gond et al . , 2013 ) , forests apparently extended naturally in central Africa ( Sangen , 2012 ) . Although precise historical information is not available before the mid-19th century for central Africa ( Burnham , 1996; Robineau , 1967 ) ( see Supplementary file 8 for a detailed chronology ) , we observed a drop in the radiocarbon signal between AD 1400 and 1650 ( Figure 3 ) that we assigned to the inland impacts of the Triangular trade in the late 15th century ( Gendreau , 2010 ) . Indeed , between AD 1550 and 1850 , the Fulbe populations coming from northern Cameroon ( Burnham , 1996 ) organized the slave-raiding for Europeans and induced the flight of populations southward into the forest ( Vennetier , 1963 ) , explaining the increase in human presence and activities ( i . e . agriculture and smelting ) in the region . The successive displacements of groups until the 18th century explain the numerous interethnic wars in the Upper-Sangha , for land control and cultural supremacy ( Copet-Rougier , 1998 ) . Based on the large dataset that we gathered , human activities clearly decreased after ca . AD 1850 , which corresponds to the beginning of the regeneration shortage of light-demanding tree populations . In the last decades of the 19th century , Savorgnan de Brazza reported that the SRI was densely populated ( Copet-Rougier , 1998 ) , which seems now unlikely given the low density of human populations ( less than one inhabitant per km² ) . We hypothesize that the European colonization deeply disturbed the spatial organization of the local populations in central Africa , as demonstrated in Gabon ( Pourtier , 1989; Engone Obiang et al . , 2014 ) . Colonization stopped the migrations and the interethnic warfare and forced entire groups to settle along rivers and roads for administrative and commercial purposes ( Vennetier , 1963; Robineau , 1967 ) . However , the process of village redistribution during the colonial times strongly varied from one place to another , according to the settlement of the colonial posts , and the borders between the French and German possessions ( Pourtier , 1989: e . g . of the Fang villages in Gabon ) . Additional factors can also be invoked to explain the emptying of the forests , including the involvement of local populations in the Franco-German conflicts during their respective colonial expansions and the two World Wars , the forced or voluntarily labor in concession companies , the deadly repression of riots , and the increased mortality because of diseases ( e . g . , trypanosomiasis along the Ubangui and the Sangha rivers ) ( Robineau , 1967; Runge and Nguimalet , 2005; Runge , 2008; Runge et al . , 2014 ) . Furthermore , because of the land abandonment caused by the new relationships established between the local peoples and the colonists ( Giles-Vernick , 2000 ) , the Mpiemu tales of the late 19th century relate to the regrowth of the forest . From the 1930s and after the independence ( 1960 ) , the abandonment of the forests was amplified because the access to education contributed to an increase in the rural exodus to the main towns and capitals in a search for valued wage labor in administration or trade ( Vennetier , 1963; Robineau , 1967 ) . From this period , deep demographic disparities emerged between towns and rural areas: most working-age people went to cities ( e . g . , Ouesso , Impfondo and Brazzaville in Congo , Yokadouma and Bertoua in Cameroon , and Berberati and Bangui in CAR ) , while children and the elderly people were left in villages . Thus , less labor force was available for forest clearing and cultivation ( Vennetier , 1963 ) . For the first time in the Sangha River Interval , a convergent body of evidence shows the effect of past changes in the disturbance regime on forest structure and composition . Consistent with previous observations in Nigeria ( White and Oates , 1999; van Gemerden et al . , 2003 ) , in Gabon ( Engone Obiang et al . , 2014 ) , and in southwestern Cameroon ( Biwolé et al . , 2015 ) , the population decline of light-demanding tree species that now dominate the canopy is explained by the decrease in anthropogenic disturbances . Caution is nevertheless required regarding the interpretation of the radiocarbon signal . Large-scale historical events , such as the interethnic wars and the European colonization of Africa , contributed to reduce human pressure on the forest . Former agricultural activities such as shifting cultivation , which were scattered in the forest areas between AD 1300 and 1850 , likely had an indirect positive influence on the regeneration of these species . Past local populations of ‘foragers-horticulturists’ ( Kay and Kaplan , 2015 ) gardened the forest by preserving useful light wooded trees ( e . g . , T . scleroxylon ) or dense wooded trees ( e . g . , P . elata and E . suaveolens ) in the fields during forest clearing and therefore created favorable conditions for their recruitment ( Carrière et al . , 2002 ) . Since ca . AD 1850 , the reduced disturbance regime has apparently hindered the regeneration of most species of light-demanding trees ( Carrière et al . , 2002; van Gemerden et al . , 2003; Willis et al . , 2004; Brnčić et al . , 2007 ) . The current lack of regeneration and the general aging of the populations threaten both their viability and the sustainability of logging ( Hall et al . , 2003; van Gemerden et al . , 2003 ) . Thus , based on these results , a renewed interest in silvicultural practices ( Doucet et al . , 2004 ) that create larger openings in the canopy should be inspired . Complementary liberation , thinning treatments , and population enforcement , may also contribute to maintain these timber species ( Fayolle et al . , 2014a ) . The Sangha River Interval ( SRI ) is a 400-km-wide area in southeastern Cameroon , southern Central African Republic ( CAR ) , and northern Republic of Congo . The extremes that encompass the area are 0°−5° N and 13°−19° E ( Gond et al . , 2013 ) . The climate is humid tropical to equatorial from north to south and from east to west with alternating wet ( May , September-October ) and dry seasons ( December-February , July; Gillet and Doucet , 2012 ) . Mean annual rainfall ranges between 1616 and 1760 mm ( Lomié in Cameroon and Impfondo in the Republic of the Congo; www . climatedata . eu ) . Monthly average temperatures fluctuate around 25°C . The vegetation of the area corresponds to moist forests of the Guineo-Congolian domain ( White , 1983; Gond et al . , 2013; Fayolle et al . , 2014b ) . We used published analyzed forest inventory data ( Fayolle et al . , 2014a ) from 22 sites ( i . e . , forest concessions ) scattered over southeastern Cameroon ( n = 6 ) , southeastern Central African Republic ( n = 6 ) , and northern Republic of Congo ( n = 10 ) ( Supplementary file 1 ) . The forest inventories were conducted between 2000 and 2007 with a systematic sampling ≥1% of the concession area . We used a dataset with 1 , 765 , 483 inventoried trees with a dbh ≥30 cm in 22 sites ( i . e . , forest concessions before exploitation ) that covered six million ha in the SRI ( Fayolle et al . , 2014a ) ( Supplementary file 1 ) . We examined the diameter distribution at the genus level for the entire SRI . All trees ≥30 cm in diameter at breast height ( dbh ) were identified and measured in 0 . 5 ha plots consecutively distributed along parallel and equidistant transects in unlogged forest concessions ( Picard and Gourlet-Fleury , 2008; Réjou-Méchain et al . , 2008; Gourlet-Fleury et al . , 2011; Gond et al . , 2013; Fayolle et al . , 2012 , 2014a ) . The minimum diameter of the trees recorded was 30 cm , which effectively confined our analysis to ( sub ) canopy trees with reduced mortality and less variation in growth rates ( Clark and Clark , 1992 ) . Vernacular names were converted into genus-level scientific names , and the trees were assigned to 10-cm-wide diameter at breast height ( dbh ) classes , with the largest trees ≥150 cm in a single class ( total of 13 classes ) . Diameter distributions were analyzed for a set of 176 of the inventoried genera for which we were confident of the identification ( Fayolle et al . , 2014a ) . To detect the main variation in the diameter distribution among the genera , we performed a correspondence analysis ( CA ) of the genus × diameter matrix followed by a clustering based on Euclidian distances and an average agglomeration method . In this study , we focused on four particular genera that are monospecific in the SRI and had a unimodal distribution ( Figure 2 and Supplementary file 1 ) and for which we had data on their annual increments of diameter ( i . e . , Erythrophleum , Pericopsis , Terminalia , and Triplochiton ) . Details on the diameter distribution of the study species at each study site are shown in Figure 2—figure supplement 1 . Terminalia and Triplochiton are characteristic of semi-deciduous Celtis spp . forest in the SRI ( Fayolle et al . , 2014b ) , whereas Pericopsis is an endangered timber species according to the CITES Red List . We later refer to species only ( i . e . , Erythrophleum suaveolens and Pericopsis elata , Terminalia superba , and Triplochiton scleroxylon ) as they are monospecific in the study area . We gathered age data for the four study species in tropical Africa from published tree-ring studies ( Worbes et al . , 2003; De Ridder et al . , 2013a , 2013b , 2014 ) ( Supplementary file 3 ) to identify the growth models that provided reliable age estimations ( Figure 2—figure supplement 2 and Supplementary files 3 and 4 ) . All trees were measured at dbh ( 130 cm in height ) . In Figure 2—figure supplement 2 , the age/diameter relationships are shown . Repeated diameter measurements of 982 monitored trees of the four study species were obtained on seven trails ( n = 4 in Cameroon; n = 3 in the Republic of the Congo ) used for the permanent monitoring of tree growth ( Picard and Gourlet-Fleury , 2008 ) . We calculated the mean annual increment in diameter ( MAId ) for n = 367 E . suaveolens; n = 199 P . elata; n = 152 T . superba; and n = 264 T . scleroxylon . To account for the ontogenic variation in growth generally identified for tropical tree species ( Hérault et al . , 2011 ) , six growth models ( i . e . , Canham , Gompertz , Verhulst , Power , Power mult , and Lognormal ) relating tree diameter ( DBH ) to growth ( MAId ) were fitted to the growth and diameter data for all study species . Linear and Mean models were additionally fitted for comparison ( Supplementary file 4 and Figure 2—figure supplement 2 ) . We used the Bayesian Information Criterion ( BIC ) for assessing the performance of the models . Ordinary differential equations were solved numerically to obtain the relationship between tree diameter and time ( age ) ( Figure 2—figure supplement 2 ) . We finally estimated the age of trees at the mode of the diameter distribution based on the Mean Annual Increment of diameter ( MAId ) and converted these ages into dates using the inventory date of AD 2000 as the reference date ( Supplementary file 2 ) . We documented the paleoenvironmental changes in the SRI for the last 1000 years ( Supplementary file 5 ) ( Laraque et al , 1998; DeMenocal et al . , 2000; Runge and Fimbel , 2001; Harris , 2002; Runge and Nguimalet , 2005; Brnčić et al . , 2007 , 2009 ;Neumer et al . , 2008; Runge , 2008; Sangen et al . , 2011; Sangen , 2012; Aleman et al . , 2013; Runge et al . , 2014; Tovar et al . , 2014; Lupo et al . , 2015 ) . We acquired paleoenvironmental data from 34 sites , either terra firme , swamp , lake or marine sites , that provided data on the past climate ( SSTs and atmospheric dust signal ) , vegetation ( phytoliths , δ13C , pollen ) and anthropogenic disturbances ( charcoal influxes , alluvial discharges through grain size and chemical elements analyses ) . Site locations are shown in Figure 1 , and the data are synthesized in Figure 3 . The degree of frequency of a proxy was determined regarding all similar curves in the identical study ( e . g . , E . guineensis pollen curve ~ all pollen curves in Brnčić et al . , 2009 ) , and the cutoffs were evenly set from the minimum to the maximum values . We used 63 uncalibrated traditional and accelerator mass spectrometry ( AMS ) radiocarbon dates and two optically stimulated luminescence ( OSL ) dates from 52 archaeological sites and punctual discoveries that covered the last 1000 years ( Supplementary file 6 ) ( Fay , 1997; Lanfranchi et al . , 1998; Brnčić , 2003; Moga , 2008; Meyer et al . , 2009; Oslisly et al . , 2013b; Morin-Rivat et al . , 2014 , 2016 ;Lupo et al . , 2015 ) . A total of 22 published dates from 21 sites in Cameroon , 15 dates from 13 sites in the Republic of the Congo , and 28 dates from 18 sites in the Central African Republic were acquired . The site locations are shown in Figure 1 . The analyses on dates were performed using the OxCal v . 4 . 2 program ( Ramsey , 2013 ) with the IntCal13 atmospheric calibration curve ( Reimer , 2013 ) . All dates were tested using an outlier analysis ( Ramsey , 2009 ) . To provide an estimate of the temporal trends of human activities in the SRI , we performed a summed probability distribution of the 63 available radiocarbon dates calibrated in yrs BP in combination with a Bayesian model ( Bayliss , 2009; Ramsey , 2009 ) ( Figure 3 ) . Chronological Query Language ( CQL ) codes used are indicated in the Supplementary file 7 . After reviewing the historical literature , we selected 12 references that illustrate key dates and events from the beginning of the 15th century to the present , which influenced directly or indirectly human populations in the SRI ( Supplementary file 8 ) ( Vennetier , 1963; Robineau , 1967; Kaspi , 1971; Burnham , 1996; Copet-Rougier , 1998; Coquery-Vidrovitch , 1998; Freed , 2010; Giles-Vernick , 2000; Manning and Akyeampong , 2006; Laporte et al . , 2007; Gendreau , 2010; Stock , 2013 ) . All cited localities are indicated in Figure 1 .

The world’s forests contain trillions of trees . Some of those trees require more light than others to mature , and certain species can only grow to reach the forest canopy if they have access to sunlight throughout their whole life . Central Africa is home to the second largest tropical rainforest in the world . Previous studies showed that few young trees of light-demanding species were growing to replace the old trees in this forest . As a result this population is aging and at risk of disappearing , which is a major concern . Many light-demanding tree species in the Central African forest are cut down for their valuable timber . However , if young trees do not grow to replace the mature ones that are logged , even logging operations that follow national and international environmental rules cannot guarantee the sustainability of these trees . As such , Morin-Rivat et al . set out to understand what changed in the Central African forest in the past to stop the regeneration of the light-demanding trees . The analyses focused on four species classified as light-demanding trees in part of Central Africa called the northern Congo Basin . Most of the trees in these species were about 165 years old . This was the case even though the different species grow at different rates , and it means that they all grew from young trees that settled in the middle of the 19th century . So what was it that changed after this period to stop this population of light-demanding trees in the Central African forest from regenerating ? By combining information from a number of datasets and historical records , Morin-Rivat et al . arrived at the following conclusion . Before the mid-19th century , many people lived in the forest and their activities created clearings that turned the forest into a relatively patchy landscape . However from about 1850 onwards , when Europeans started to colonize the region , people and villages were moved out of the forests and closer to rivers and roads for administrative and commercial purposes . Moreover , many people were killed in conflicts or died because of newly introduced diseases , which also led to fewer people in the forest . As a result , the forest became less disturbed . With fewer clearings , fewer light-demanding trees would have had enough access to sunlight to grow to maturity . The findings of Morin-Rivat et al . show that disturbance is needed to maintain certain forest habitats and tree species , including light-demanding species of tree . As common logging operations do not create openings large enough to guarantee that such species will be able to establish themselves naturally , complementary treatments are needed . These might include selectively logging mature trees around young members of light-demanding species , or planting threatened species .

2017

Present-day central African forest is a legacy of the 19th century human history

Long distance transport in plants occurs in sieve tubes of the phloem . The pressure flow hypothesis introduced by Ernst Münch in 1930 describes a mechanism of osmotically generated pressure differentials that are supposed to drive the movement of sugars and other solutes in the phloem , but this hypothesis has long faced major challenges . The key issue is whether the conductance of sieve tubes , including sieve plate pores , is sufficient to allow pressure flow . We show that with increasing distance between source and sink , sieve tube conductivity and turgor increases dramatically in Ipomoea nil . Our results provide strong support for the Münch hypothesis , while providing new tools for the investigation of one of the least understood plant tissues . Vascular systems allow organisms to distribute resources internally by bulk flow and thus to overcome size limitations set by diffusion . In plants , the evolution of vascular tissues enabled the development of trees and forests and was accompanied by a major increase in the productivity of terrestrial ecosystems . Plant vascular tissues are of two types: the xylem allows water to be pulled from the soil to maintain the hydration of leaves surrounded by air , while the phloem distributes the products of photosynthesis throughout the plant , allowing non-photosynthetic structures , such as roots , to be formed . The current hypothesis for phloem transport dates to 1930 when Ernst Münch proposed that transport through the phloem results from osmotically generated differences between the pressure in sources ( e . g . leaves ) and sinks ( e . g . roots ) , and occurs without any additional input of energy along the transport path ( Münch , 1930 ) . The Münch hypothesis has gained wide acceptance based to a large extent on its simplicity and plausibility , rather than on experimental evidence . A fundamental issue bedeviling the pressure flow hypothesis is the long standing question of whether sieve tubes have sufficient hydraulic conductivity to allow sugars to be transported from leaves to roots in the largest and longest of plants . Increasing distances between sources and sinks would appear to require greater pressure to drive flow , but measurements suggest lower source turgor in trees compared to smaller , herbaceous plants ( e . g . , crops or weeds; Turgeon , 2010 ) . A number of authors have emphasized the difficulty of accommodating the Münch hypothesis in trees , leading to the proposal of re-loading mechanisms via relays ( Lang , 1979; Aikman , 1980 ) . However , it is important to note that such challenges derive in large part from mathematical models of phloem transport in which the values of one or more of the key variables are unknown or poorly constrained ( Tyree et al . , 1974; Pickard and Abraham-Shrauner , 2009; Thompson and Holbrook , 2003; Hölttä et al . , 2009 ) . Phloem transport ceases immediately when sources and sinks become disconnected , which is always the case when preparing sieve tubes for in vitro studies . In situ studies on the other hand are challenging , especially in large plants . Thus , despite its importance , phloem transport and resource allocation is the least understood major process in plant function . Over the last years we have developed in situ methods to measure with high precision all of the parameters needed to quantify laminar flow through passive microtubes as described by the Hagen-Poiseuille equation ( 1 ) U=kηΔpL , , where U is the flow velocity , Δp is the pressure differential , L is the tube length , η is the sap viscosity , and k is the conductivity of the tube . Previously , methods only existed for two of the five parameters: length , which is the distance between source and sink , and velocity , which can be determined using radioisotopes ( Babst et al . , 2005 ) , magnetic resonance imaging ( Mullendore et al . , 2010 ) , or dye tracking ( Savage et al . , 2013 ) . We have recently introduced methods to measure the missing parameters including new EM preparation protocols and mathematical modeling to quantify sieve tube hydraulic resistance ( Froelich et al . , 2011; Mullendore et al . , 2010; Jensen et al . , 2012 ) ; a novel micro-capillary pressure probe to determine pressure in small and sensitive cells such as sieve tubes ( Knoblauch et al . , 2014 ) ; and protocols for in situ observation and flow velocity measurements in individual tubes ( Froelich et al . , 2011 ) . A new method for determining phloem sap viscosity in vivo using fluorescence lifetime imaging of the fluorophore 2-NBDG is described . Here we present a study on phloem flow relevant parameters that tackles the major question on the pressure flow hypothesis . Our data provide strong support for the Münch hypothesis as a unifying mechanism of long distance transport in plants . The results also call into question current understanding of downstream events such as the “high pressure manifold model” of phloem unloading and carbohydrate delivery to sinks . The toolset described here will allow for detailed research on phloem transport and resource allocation , processes critical for food security and improvement of bioenergy crops , as well as understanding ecosystem ecology and the global carbon cycle . We first acquired baseline parameters from 7 . 5 m long plants ( Figure 1 ) and measured anatomical parameters at locations 1 m , 4 m , and 7 m ( Figure 1 E-J ) . The acquisition of geometrical parameters requires the highest accuracy because minor changes in tube geometry have a large impact on the calculated conductivity . This is the main reason why current models have to cope with large uncertainties . In addition , the natural variation of tube parameters requires the acquisition of large data sets . For this study , more than 1000 SEM and confocal images were taken , >100 , 000 sieve plate pore numbers were counted ( in ≥10 plates per data point ) , and >15 , 000 sieve plate pore diameters ( n ≥ 380 per data point ) , >1500 sieve element diameters ( n ≥ 10 per data point ) , and >800 sieve element lengths ( n ≥ 10 per data point ) were measured . Changes in sieve tube geometry including sieve element radius ( Figure 1E ) , sieve element length ( Figure 1F ) , sieve plate pore number ( Figure 1G ) and sieve plate pore radius ( Figure 1H ) caused a slight increase in sieve tube specific conductivity ( see appendix for specifics on conductivity calculations ) towards the base of the 7 . 5 m plant ( Figure 1I ) of around 1 µm2 . Measurement of the total phloem area from high resolution confocal images of stem cross sections ( Figure 1—figure supplement 1A ) showed similar external and internal phloem areas during primary growth ( Figure 1—figure supplement 1B ) , but a much larger increase of external phloem area during secondary growth . 10 . 7554/eLife . 15341 . 003Figure 1 . Geometrical parameters of a moderate sized morning glory plant . ( A ) A 7 . 5 m long morning glory plant with source leaves along the stem . ( B-D ) Scanning electron micrographs of sieve plates at 1 m ( B ) , 4 m ( C ) , and 7 m ( D ) from the base of the stem ( as indicated by red arrows in B–D ) . E–J ) Cell geometrical data were collected separately for internal ( red ) and external ( blue ) phloem . Average sieve element radius ( E; n ≥ 10 per data point ) , length ( F; n ≥ 10 per data point ) , pore number ( G; n ≥ 10 per data point ) , and pore radius ( H; n > 150 per data point ) result in a conductivity of ~1 µm2 for the external phloem but an increasing conductivity along the stem for the internal phloem ( I ) . The total phloem area ( J ) is , however , significantly higher for the external phloem . Error bars show standard deviation . Scale bars in B–D =10 µmDOI: http://dx . doi . org/10 . 7554/eLife . 15341 . 00310 . 7554/eLife . 15341 . 004Figure 1—source data 1 . Source data of sieve tube geometrical parameters for Figure 1 and Figure 3—figure supplement 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 15341 . 00410 . 7554/eLife . 15341 . 005Figure 1—figure supplement 1 . Stem anatomy of morning glory plants . ( A ) Staining of morning glory stem cross sections at 4 m with a mixture of calcofluor white and aniline blue allowed visualization of sieve plates in the bicollateral bundles of external ( blue dashed arrow ) and internal ( green dashed arrow ) phloem , which are easily discernible from primary ( yellow arrow ) and secondary xylem ( green arrow ) as well as sclerenchyma ( white arrow ) . ( B ) Internal and external phloem areas are highlighted in white in stem cross sections ( from left to right ) at 7 m , 4 m , and 1 m from the base of the stem . Scale bars = 1 mmDOI: http://dx . doi . org/10 . 7554/eLife . 15341 . 005 The basipetal flow velocity on the stem was determined by application of 11CO2 to one source leaf approximately 3 m from the shoot base ( see Materials and methods section ) . The average flow velocity was 123 ± 13 µm/s ( n = 3; Figure 2 ) . 10 . 7554/eLife . 15341 . 006Figure 2 . Phloem flow relevant parameters in a medium sized morning glory plant . An illustration summarizing the findings in a medium sized plant with leaves attached along the entire length of the stem . Cell geometrical data were taken at 1 m , 4 m , and 7 m ( blue lines ) along the stem . Resulting conductivities are indicated in green . Source sieve tube turgor measurements ( red ) were taken in the main vein of leaves along the stem axis , sink turgor ( black ) in root tips , and average flow velocity was measured using 11CO2 labeling ( orange ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15341 . 00610 . 7554/eLife . 15341 . 007Figure 2—figure supplement 1 . In situ viscosity measurements . ( A ) 2-NDBG is loaded into the phloem in situ and observed by confocal microscopy in the midrib of a mature leaf in morning glory . The image reveals the location of sieve elements ( SE ) , companion cells ( CC ) sieve plates ( solid arrows ) and sieve element plastids ( dashed arrows ) . ( B ) A corresponding color coded fluorescence lifetime map reveals the relatively low viscosity of phloem sap of 1 . 7 mPas ( blue ) in contrast to the much higher viscosity above 5 mPas of the cell wall , membrane , and nucleus areas . ( C ) Calibration curve of 2-NBDG lifetime versus viscosity for aqueous sucrose solutions at temperature T = 298 K . A strong lifetime change between one and 10 mPas renders 2-NDBG a good probe for intracellular viscosity measurements . n ≥ 21 for each data point . Error bars show standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 15341 . 00710 . 7554/eLife . 15341 . 008Figure 2—figure supplement 2 . In situ sieve tube turgor measurements . Two frames extracted from Video 1 showing in situ pico gauge pressure measurements . A sieve tube ( bright green ) translocates distantly applied fluorescent dye . The black arrow demarcates a sieve plate . The red arrow points to the location of the water oil interface before impalement into the cell . The turgor pressure of the sieve tube results in a compression of the pico gauge filling oil , indicated by the movement of the meniscus interface ( blue arrow ) . Inflow of fluorescent dye into the water phase of the pico gauge ( yellow arrow ) provides evidence , that the measurement occured in the sieve tube . DOI: http://dx . doi . org/10 . 7554/eLife . 15341 . 00810 . 7554/eLife . 15341 . 009Figure 2—figure supplement 3 . Symplastic phloem unloading in the root tip of morning glory plants . Carboxyfluorescein-diacetate is a phloem mobile fluorophore that can be loaded into the phloem of a leaf of a morning glory plant . Once the dye enters a cell , the diacetate residue is cleaved and the fluorophore carboxyfluorescein is generated which is membrane impermeant . The dye is then translocated with the phloem sap into sinks . ( A , B ) Since the dye can only exit cells symplastically , the spread of the dye shown in the fluorescence micrograph ( B ) within the root tip ( A ) indicates symplastic unloading . DOI: http://dx . doi . org/10 . 7554/eLife . 15341 . 009 Currently , phloem sap viscosity has not been directly measured , but only estimated based on the concentrations of extracted phloem sap contents ( Thompson and Holbrook , 2003 ) . Recently , a group of dyes has been identified that can be loaded into the phloem ( Knoblauch et al . , 2015 ) . We tested carboxyfluorscein diacetate ( CFDA ) , 8-Hydroxypyrene-1 , 3 , 6-trisulfonic acid acetate ( HPTSA ) , Esculin , 2- ( N- ( 7-Nitrobenz-2-oxa-1 , 3-diazol-4-yl ) Amino ) -2-Deoxyglucose ( 2-NDBG ) , and carboxytetraethylrhodamine glucoside ( CTER ) for potential molecular rotor properties which allow the measurement of solute viscosity due to viscosity dependent changes in fluorescence decay times determined with fluorescence lifetime imaging ( FLIM ) . FLIM calibrations confirmed that 2-NBDG is a molecular rotor . To determine the phloem sap viscosity in situ , sieve elements were loaded with 2-NBDG and the decay of fluorescence lifetime was measured to generate a color coded viscosity map of the phloem ( Figure 2—figure supplement 1 ) . The average fluorescence lifetime of 2-NBDG in source phloem in morning glory was 1 . 366 ns ( ± 0 . 037 ns , n = 8 ) , which corresponds to a viscosity of 1 . 7 mPas or a sucrose concentration of about 18% . This value is within the range of estimates based on solute concentrations in phloem sap extracts of various plant species ( Jensen et al . , 2013 ) . Using these parameters ( σ , = 1 . 7 mPas , k = 1 µm2 , U = 123 µm/s ) we estimated that a pressure differential of 0 . 21 MPa would be required to drive flow through a 1 m long tube of the determined anatomy . To evaluate if such a pressure differential exists between source and sink tissue , we measured phloem pressure in situ . 10 . 7554/eLife . 15341 . 010Video 1 . In situ sieve tube turgor pressure measurement using a pico gauge . Real time video of a pico gauge measurement . The fluorescence of a distantly loaded sieve tube provides evidence for tube intactness and transport . The oil – water interface in the pico gauge moves backwards when the pico gauge is impaled into the sieve tube , indicating compression of the oil volume . Inflow of fluorescent dye into the pico gauge tip provides evidence that the measurement occured in the sieve tube . DOI: http://dx . doi . org/10 . 7554/eLife . 15341 . 010 Because the phloem is embedded in a thick tissue layer , in situ experiments have to be carried out by removing cortical tissue until the sieve tubes are exposed , but care must be taken to not injure the tubes . The preparation was achieved by making cortical hand sections with fresh razor blades as described earlier ( Knoblauch and van Bel , 1998 ) . Other methods such as laser ablation or sectioning by micromanipulators appear more elegant , but turned out to cause massive artifacts and flow stoppage . The section has to be as accurate as a twentieth of a millimeter in order to be useful for measurements of sieve tubes . This reduces the success rate to below 30% even for very experienced investigators . If sieve tubes were injured , the leaf was removed and the plant was allowed to recover for at least 24 hr . If a section appeared successful , upstream application of phloem mobile fluorescent dyes was used to verify intactness by monitoring translocation of the dye through the tube ( Video 1 , Figure 2—figure supplement 2 ) . The preparation required the tissue to be covered with an artificial medium , which could have had an influence on sieve tube turgor . However , the low resistance between sieve elements provided a buffer for local disturbance . If the medium resulted in a local increase in turgor , an artificial source would have been generated and flow would reverse towards the leaf and no fluorochromes would have arrived at the site of observation . If preparation decreased turgor , a local sink would have been induced and flow would have been towards the area of preparation from both sides . The applied fluorochromes would eventually have arrived at the site of observation but would not have passed by . In all cases , turgor measurements were conducted on tubes showing quickly increasing fluorescence downstream of the site of preparation indicating that the applied medium did not disturb turgor pressure inside the sieve tube . Sieve tube turgor in source leaves ( Figure 2 ) at different levels along the stem averaged 1 . 08 MPa ( ± 0 . 13 MPa , n = 5 ) . The anatomy of sinks ( e . g . roots ) does not permit direct measurements in sieve tubes . However , since phloem unloading in root tips follows a symplastic path ( Figure 2—figure supplement 3; Wright and Oparka , 1997; Patrick , 1997; 2013 ) the pressure in root cortical cells cannot be ( significantly ) higher than in sink sieve tubes . In situ turgor measurements in the cortex of the root elongation zone revealed a sink turgor of 0 . 59 MPa ( ± 0 . 11 MPa; n = 5 ) . Therefore the measured pressure differential of 0 . 49 MPa could account for a pressure flow since the distance from a sink to the closest source did not exceed 2 m and a pressure gradient of 0 . 21 MPa/m is required to drive the flow . Having established the baseline parameters in a moderate-sized plant , we investigated if the parameters scale with increasing transport distance , a treatment we achieved by increasing the length of the stem without source leaves ( Figure 3; Figure 3—figure supplement 1 ) . In response to this treatment , the plant must change conductivity , pressure , velocity , or viscosity ( or any combination of those ) to maintain sink assimilate delivery if a passive pressure flow is the mode of translocation . Over the course of 5 months all newly developing tissues ( leaves , side shoots ) below the top 4 m were pruned daily . Phloem pressure was measured in situ ( Video 1 ) in the main vein of the lowermost leaf throughout the growth period until the single stem of the plant was 17 . 5 m long with about 14 m of stem free of source leaves ( Figure 4 ) . At this time point , we measured phloem flow velocity by 11CO2 application and used micro PET scanning to confirm phloem translocation direction ( Figure 4—figure supplement 1 ) . Afterwards , the tissue was harvested , cell geometrical data were collected at 1 m increments along the stem , and tube conductivity was estimated ( Figure 3 A-F ) . 10 . 7554/eLife . 15341 . 011Figure 3 . Geometrical parameters of a large morning glory plant with partially defoliated stem . Geometrical data of a 17 . 5 m long morning glory plant after 5 months growth with daily removal of developing side branches and flowers as well as removal of source leaves below the top 4 m . ( A ) Total phloem area at different locations along the shoot . Plotting phloem area versus distance indicates that only the external phloem ( blue ) increases in area significantly ( internal phloem , red ) . ( B–F ) Cell geometrical data for sieve element radius ( B; n ≥ 10 per data point ) , sieve element length ( C; n ≥ 10 per data point ) , sieve plate pore number ( D; n ≥ 10 per data point ) , and sieve plate pore radius ( E; n ≥ 380 per data point ) reveal that sieve tube conductivity ( F ) increases with the length of the transport pathway . Please see Figure 3—figure supplement 3 for a comparison of the parameters and standard deviations between the moderate sized foliated morning glory ( Figure 1 ) and the partially defoliated large morning glory plant . DOI: http://dx . doi . org/10 . 7554/eLife . 15341 . 01110 . 7554/eLife . 15341 . 012Figure 3—source data 1 . Source data of sieve tube geometrical parameters for Figure 3 and Figure 3—figure supplement 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 15341 . 01210 . 7554/eLife . 15341 . 013Figure 3—figure supplement 1 . Habitus and anatomy of a partially defoliated large morning glory plant . 17 . 5 m long morning glory plant ( left ) after 5 months growth with daily removal of developing side branches and flowers as well as removal of source leafs below the top 4 m . Confocal images of cross sections ( right ) at the indicated location ( in m ) from the shoot base with highlighted phloem area . Scale bar = 2 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 15341 . 01310 . 7554/eLife . 15341 . 014Figure 3—figure supplement 2 . Anatomical adaptation to increase sieve tube conductivity . ( A ) Scanning electron micrograph of a stem cross section at 4 m distance from the stem base of the 17 . 5 m long morning glory plant , showing the anatomical adaptation of sieve plates to increasing demand by changing plate anatomy from simple to compound with steep plate angles to increase plate surface area ( compare Figure 1B–D ) . ( B , C ) Confocal micrographs of tangential sections through the phloem at 15 m ( B ) and 5 m ( C ) from the base of the stem stained for cellulose ( blue ) with calcofluor white and callose ( red ) with aniline blue . Youger tissue in close proximity to sinks has mainly simple sieve plates in a 90 degrees angle , while tissue distantly located from the closest sink generated compound sieve plates in steep angles . Scale bars: A = 50 µm; B , C = 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 15341 . 01410 . 7554/eLife . 15341 . 015Figure 3—figure supplement 3 . Comparison of geometrical parameters between a small foliated ( compare Figure 1; green = external phloem , black = internal phloem ) and a large , partially defoliated ( Figure 3; blue = external , red = internal ) morning glory plant . Error bars show standard deviation . The significantly larger sieve plate pore radius ( C ) in the external phloem of the large plant results in a several times higher conductivity ( E ) . Changes in other geometrical parameters ( A , B , D ) have less impact on the tube conductivity . The conductivity is relatively low in the small foliated plant , because conductivity scales as 1/r2 when the SE radius r is large . In combination with an increase in conducting area ( F ) , the bulk of phloem transport during secondary growths occurs through the external phloem . DOI: http://dx . doi . org/10 . 7554/eLife . 15341 . 01510 . 7554/eLife . 15341 . 016Figure 4 . Phloem flow relevant parameters in a morning glory plant with increasing leafless stem length . An illustration summarizing the findings in large morning glory plants after artificial increase of source-to-sink transport distance achieved through continuous partial defoliation . Leaves were maintained only on the upper four meters of the plant and pressure was measured throughout the growth period . Plants with a short distance between the leaves and the roots maintain a relatively low sieve tube turgor pressure ( crimson ) in the source phloem in the range of 0 . 7–0 . 8 MPa , but the pressure scales with increasing length ( black ) of defoliated stem and the conductivity increases ~5 fold ( green ) compared with the 7 . 5 m long plant shown in Figure 1 and 2 . Flow velocity ( blue ) was measured by 11C application . DOI: http://dx . doi . org/10 . 7554/eLife . 15341 . 01610 . 7554/eLife . 15341 . 017Figure 4—figure supplement 1 . In situ measurements on large morning glory plants . ( A ) In situ turgor measurement using pico gauges and a fixed stage remote controlled fluortescence microscope . ( B ) A large morning glory plant during Micro PET scan experiment setup . ( C ) PET scan image of the lowermost leaf provides evidence that transport occurs through the petiole only towards the root . No label was detected in the upper stem towards the shoot tip ( location of stem indicated by white lines ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15341 . 01710 . 7554/eLife . 15341 . 018Figure 4—figure supplement 2 . Morning glory stem growth . The helical growth of morning glory stems to wrap around objects in order to climb leads to an increase in the sieve tube length compared to straight stems . DOI: http://dx . doi . org/10 . 7554/eLife . 15341 . 018 Source sieve tube turgor pressure was 0 . 75 ( ± 0 . 05 n = 3 ) MPa in young plants and increased to more than 2 . 2 MPa at the end of the experiment ( Figure 4 ) when the distance from the lowest leaf to the base of the stem was 14 m with a transport distance of 17 . 5 m when applying a correction factor of 1 . 25 to account for helical stem growth ( Figure 4—figure supplement 2 ) . Despite the high pressure generated in source tissue , the pressure differential ( assuming sink turgor of 0 . 59 MPa ) would have allowed photoassimilates to be transported only 3 . 7 m at a measured flow velocity of 256 µm/s , if the tube geometry in the long plants was the same as in the smaller plants . Instead , sieve tube conductivity increased 5-to-6 fold compared to fully foliated young plants ( Figure 3F ) mainly due to a significant increase in average sieve plate pore radius ( Figure 3E , Figure 3—figure supplement 3 ) . Taking into account the much higher conductivity , the estimated maximum transport distance is 20 . 35 m , which is in good agreement with the actual stem length , especially given that transport distance in the roots is not included . The change in sieve tube conductivity appears to be largely a response to the increased distance between source leaves and the sink ( roots ) as the average sieve tube conductivity of a 13 m long morning glory plant with source leaves all along its stem was only 2 . 8 µm2 ( Figure 5 ) . Tilting of the sieve plate , and formation of compound instead of simple sieve plates increases the sieve plate area to allow larger pores ( Figure 3—figure supplement 2 ) . Measurement of the total phloem area from confocal micrographs ( Figure 3A; Figure 3—figure supplement 1 , Figure 3—figure supplement 3 ) indicated a significant increase of the external phloem area compared to plants with leaves along their entire stem , suggesting that the continuous extension of the distance between source and sink induces cambial activity to produce new sieve tubes with higher conductivity . Interestingly , despite a slight increase in tube conductivity , the conducting area of the internal phloem actually decreases ( Figure 3—figure supplement 3F ) . This is attributed to the restricted space when secondary xylem and phloem is formed by the internal cambium . The significant reduction of the pith diameter in older tissue indicates the space limitations for internal secondary phloem . Therefore , the impact of the internal phloem on whole phloem transport decreases with increasing secondary growth . 10 . 7554/eLife . 15341 . 019Figure 5 . Anatomical adjustments to growth conditions . An illustration summarizing the findings in a large morning glory plant with leaves along the length of the stem . In contrast to partially defoliated plants , the conductivity remains relatively low , likely due to the shorter distance from source leaves to sink tissue . DOI: http://dx . doi . org/10 . 7554/eLife . 15341 . 01910 . 7554/eLife . 15341 . 020Figure 5—source data 1 . Source data of sieve tube geometrical parameters for Figure 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 15341 . 020 These studies on morning glory show that when the transport length is increased , sieve tube conductivity and pressure increase accordingly . The observations provide strong support for pressure driven long distance transport . In this study we address the major question about phloem transport in angiosperms: do phloem parameters scale in accord with the Münch hypothesis ? We found that large morning glory vines exhibited a significant increase in the conductivity of sieve tubes in the external phloem and that this increase was primarily due to larger sieve plate pores . In addition to changes in sieve tube structure , the pressure in sieve tubes scaled with the distance between leaves and roots , such that morning glory plants with greater distance separating leaves and roots had significantly higher source turgor pressures . In summary , plants actively adjust flow relevant parameters to accommodate a passive pressure driven mass flow . After almost a century , we finally have experimental evidence that addresses the key challenge to the pressure flow hypothesis . Although many cellular aspects , such as the function of most p-proteins and sieve element plastids , remains unclear , the overall transport concept proposed by Münch is supported . Prior to our study , many of the parameters governing phloem transport were poorly quantified . Thus , an additional important outcome of the study is a toolkit' for determining all of the parameters affecting flow . This means , that our work opens the door to new studies of phloem function . Foremost among the questions to be answered are how plants control the distribution of resources needed to support assimilate storage and the growth of new tissues . Phloem unloading fills the reservoirs of the most important human food sources ( primary sources such as fruits , roots , tubers , cereals , and through consumption also other sources such as meat ) but our understanding of the unloading mechanism is similarly rudimentary . A currently popular modification of the Münch hypothesis that also accounts for phloem unloading is the high pressure manifold model ( Patrick , 2013 ) . This model proposes maintenance of high hydrostatic pressures throughout the sieve tube system with minimal pressure differences from source to sink , but high differences between sieve elements and surrounding cells in the unloading zone ( Figure 6A ) . Our data , however , suggest that tube resistance will consume most of the energy provided by the source pressure to allow flow at the measured velocities ( Figure 6B ) . Most importantly and contrary to the high pressure manifold model , small morning glory plants generate only low pressures sufficient for flow , but not additional large margins to maintain high pressure differentials in the unloading zone . This suggests that at least in the species investigated , either plasmodesmal conductance in the unloading zone is very high for passive unloading , or additional energy , e . g . in form of active unloading processes , is required . A better understanding of short and long distance flow patterns and assimilate distribution in plants will be necessary to resolve these important questions . 10 . 7554/eLife . 15341 . 021Figure 6 . Phloem pressure gradients in relation to phloem unloading . Schematic drawing of a single source leaf ( green ) loading assimilates into the phloem ( blue ) and unloading through plasmodesmata ( black ) into a single root tip ( yellow ) . ( A ) Independent of the plant size the pressure manifold model proposes nearly uniform high pressure along the stem , but large differences between sieve elements and surrounding cells in the unloading zone where the steepest gradients are found . ( B ) Our data suggest that tube resistance in the stem of morning glory plants consumes most of the pressure gradient , contrary to the high pressure manifold model . DOI: http://dx . doi . org/10 . 7554/eLife . 15341 . 021 Organisms are structures whose integrity depends on active and passive flows . While active flows are dominated by proteins in the form of cytoskeletal elements or membrane transporters , passive flow is dependent on cell geometry and cellular connections in orchestration with concentration gradients and osmotic potentials ( which are again generated by active flows ) . The lack of precise geometrical data of cellular connections is therefore a major problem , as this information forms the basis for modeling long distance flow patterns of assimilates , signaling molecules , and other substances . Large-scale efforts to collect cell geometrical data ( a field one may call quantitative anatomy or 'Anatomics' ) at a resolution sufficient to extract sieve tube connections , diameters , length as well as plasmodesmal frequencies etc . throughout the plant are needed . Integration of physiological and molecular data ( e . g . cell surface specific distribution of membrane transporters etc . ) into a 'virtual model plant could lead to a major step forward in rationally designing crop plants to meet future societal and environmental needs . To accomplish this methods for tissue preparation , automated acquisition , and reconstruction need to be combined , but the technology is available . Due to the long growth period needed to attain large stem axes , plants were started in the greenhouse in February in pots at 23°C , with 60 to 70% relative humidity , and a 14/10 hr light/dark period ( daylight plus additional lamp light [model #PL 90 ( PL Lighting Systems , Ontario , Canada ) ] with a minimum irradiance of 150 µE m-2 s-1 . The plants were moved outside at the end of May . The plants were grown on ropes to allow climbing and easy transport . Morning glory is a fast growing vine , but highly susceptible to pests . Primarily spider mites and aphids quickly grow in high numbers and have devastating impacts on the plants . In order to prevent potential effects of systemic pesticides on the plant’s physiology , we chose to remove pests manually . Usually every day , but at least 4 days per week we checked every leaf and the stem of each plant and removed pests by spraying them off with a fine mist of water . All developing side branches and flower buds were pruned at the same time to maintain one single stem . Due to higher abundance of predators , pest infestation was much lower outside compared to in the greenhouse . Plants were transferred from the growth site to the lab just before the experiment was conducted . Phloem transport has been shown to be very stable and that diurnal changes have little or no impact on phloem transport velocities , likely due to sufficient starch pools even when photosynthesis capacity changes ( Windt et al . , 2006 ) . Therefore we did not expect negative effects of movement of the plants . At the main vein of the lowermost source leaf , a small paradermal section to remove cortical tissue and to expose sieve elements was made with a fresh razor blade by hand . The section was immediately covered with phloem recovery medium consisting of 10 mM KCl , 10 mM NaCl , and 0 . 2 mM EDTA in unbuffered distilled water . Since the sieve tube to be measured must be the uppermost uninjured cell , the section was checked under the microscope . If sieve tube damage was observed , i . e . the cut was too deep , the leaf was removed and the plant was brought back to the growth site and a new attempt was performed after a recovery period of at least one day . If the section appeared useful in that the sieve tubes were well visible and appeared uninjured , a phloem mobile dye ( CDFA or Esculin , Sigma-Aldrich , St . Louis , MO ) was injected locally into the leaf apoplast about 5 cm upstream of the preparation site by pushing the dye solution through the stomata with the blunt end of a 3 ml syringe . About 2 cm2 were filled with dye to provide sufficient fluorochromes for visualization after phloem loading . After filling up the apoplast , the phloem at the preparation site was observed for arrival of the fluorochromes which usually took 20–30 min . Translocating , intact tubes exposed to the surface were measured by impaling pico-gauges into the tube and monitoring the compression of the pico-gauge filling oil . Pico gauge production and pressure calculations were performed as described in detail in ( Knoblauch et al . , 2014 ) . Sink root pressure was measured in root tips of morning glory plants grown in the greenhouse under standard conditions ( see above ) . The plant was removed from its pot with the soil still attached to the roots and transferred into a plastic bag to keep roots moist . A small hole was cut into the bag , a single root was pulled out and covered with moist paper towels to maintain humidity . The root tip was immersed in water , and held in place by using a glass slide as weight . Measurements were carried out using pico gauges , produced by method A as described in ( Knoblauch et al . , 2014 ) . About 2–3 min were given for temperature adjustments of the pico gauges to the surroundings . A Sutter Instruments ( Novato , CA ) model MPC 200 micromanipulator was used to control movement of the pico gauges . To take measurements , cortex cells of the root were impaled with a pico gauge and measurements were recorded at approximately 6 frames per second using a Leica DFC 450 camera , mounted to a Leica DM LFSA microscope equipped with a HCX Plan APO 40x lens . ( Leica , Wetzlar , Germany ) . Data processing was conducted as described earlier ( Knoblauch et al . , 2014 ) . In situ viscosity measurements were conducted by loading 2-NBDG into source tissue of intact morning glory plants by pressure injection of 2-NBDG solution through the stomata using the blunt end opening of a 5 ml syringe . Fluorescence was observed downstream of the application site ( Knoblauch and van Bel , 1998 ) . Measurements were carried out on a Leica SP8 SMD white light laser system equipped with a pico quant FLIM attachment . The dye was excited with the 470 nm line of the supercontinuum laser and emission light was collected between 490 nm and 550 nm . Carbon-11 , as 11CO2 , was generated by the 14N ( p , α ) 11C nuclear transformation ( Ferrieri and Wolf , 1983 ) , on an 19 MeV cyclotron ( EBCO , Richmond , British Columbia , Canada ) . The 11C was administered to a single leaf as 11CO2 gas as a 30 sec pulse in continuously streaming air in a leaf cuvette with PAR 750 µmol m-2 s-1 , as previously described ( Babst et al . , 2013 ) . Leaf fixation , carbon export from the leaf , and 11C-photoassimilate transport velocity were monitored in real-time using a detector built into the leaf cuvette , and two detectors shielded with collimated lead and positioned to detect radioactivity from the stem . After the pulse of 11CO2 , plants were incubated in place with continuous airflow through the leaf cuvette for 2–3 hr to allow sufficient time for transport of 11C-photoassimilate through the stem . Transport velocity was calculated as the distance between the stem detectors divided by the transit time of the 11C radioactivity between the first and second stem detectors . Positron Emission Tomography ( PET ) imaging was performed by positioning the stem and petiole within , and the load leaf outside of , the field of view of a microPET R4 ( Concorde Microsystems , Knoxville , TN , USA ) approximately 2 hr after 11CO2 was administered . Data was acquired from stem and petiole radioactivity emissions for 10 min , and a 3-dimensional image was reconstructed as a single frame using microPET Manager 2 . 3 . 3 . 0 software ( Siemens Medical Solutions USA Molecular Imaging; Knoxville , TN , USA ) . To obtain cell geometrical data for the number of sieve plate pores per plate , the radius of pores , the radius of the tube , and the thickness of the plate , morning glory stem segments of approximately 5 cm length were cut and immediately immersed into 70% ethanol at –20°C to prevent callose formation on the sieve plate . The tissue was freeze substituted for at least 1 week . After that , longitudinal- and cross-sections of approximately 1 mm length were taken and immersed in a medium containing 0 . 5% proteinase K , 8% triton X 100 at pH 8 ( Mullendore et al . , 2010 ) . The tissue was digested at 55°C for 2–6 weeks with weekly changes of the digestion medium to remove the symplast . After digestion of the symplast the tissue was freeze dried and sputter coated . 40 SEM images per sample point were taken with a FEI ( Hillsboro , Or ) Quanta 200 FEG SEM to measure and quantify cell geometrical data . Measurements were performed with imageJ software as described in ( Mullendore et al . , 2010 ) . To collect data on the average length of sieve elements , longitudinal hand sections of freeze substituted tissue were taken with a razor blade , stained with calcofluor white and aniline blue , and images were taken with a Leica SP8 confocal microscope at excitation wavelength 405 nm and emission collection at 420–480 nm for calcofluor white , and 550–600 nm for aniline blue fluorescence . Measurements were performed with Leica LAS software . Primary leaves in morning glory seedlings were loaded with Carboxyfluorescein diacetate as described previously ( Wright and Oparka , 1997 ) . After 1 hr seedlings were removed from the pots , the soil was washed off the primary root and fluorescence was monitored with a I3 filter block on a Leica DMLFSA microscope and images were taken with a DFC-300 camera .

Plants use energy from sunlight to make sugars in a process called photosynthesis . Most photosynthesis takes place in the leaves and so much of the sugar needs to be transported to other parts of the plant , such as fruits or roots . The sugars are transported by phloem tubes , which form a system that spans the entire plant . In 1930 , a German scientist called Ernst Münch proposed a hypothesis for how phloem tubes move sugars and other molecules around the plant . He proposed that the loading of these molecules into phloem tubes in the leaves or other “source" tissues makes the fluid inside the vessels more concentrated so that water is drawn into the phloem from neighboring “xylem” vessels . This creates pressure that pushes the fluid along the phloem tube towards the fruit , roots and other “sink” tissues . In the sink tissues the sugars are consumed , which reduces their concentration in the phloem and the pressure . Overall , this results in the flow of sugars and other molecules from where they are produced to where they are most needed . However , this hypothesis is still largely untested because it has proved difficult to carry out experiments on phloem . Detaching the source tissues from the sink tissues stops the flow of fluid so only experiments in whole plants can provide meaningful data . Knoblauch et al . have now developed new methods to study phloem in an ornamental plant called morning glory . The experiments show that plants can alter the shape of phloem vessels and the pressure within the vessels to allow them to transport sugars and other molecules over different distances . These findings strongly support the Münch hypothesis and make other alternative hypotheses seem unlikely . Furthermore , the methods developed by Knoblauch et al . will allow others to further investigate phloem transport . New findings in this area may allow plant biologists to direct the flow of sugars and other molecules towards particular plant tissues to improve the nutritional quality of food crops in the future .

2016

Testing the Münch hypothesis of long distance phloem transport in plants

Three amino acid loop extension homeodomain transcription factors ( TALE HD TFs ) act as life cycle regulators in green algae and land plants . In mosses these regulators are required for the deployment of the sporophyte developmental program . We demonstrate that mutations in either of two TALE HD TF genes , OUROBOROS or SAMSARA , in the brown alga Ectocarpus result in conversion of the sporophyte generation into a gametophyte . The OUROBOROS and SAMSARA proteins heterodimerise in a similar manner to TALE HD TF life cycle regulators in the green lineage . These observations demonstrate that TALE-HD-TF-based life cycle regulation systems have an extremely ancient origin , and that these systems have been independently recruited to regulate sporophyte developmental programs in at least two different complex multicellular eukaryotic supergroups , Archaeplastida and Chromalveolata . Developmental processes need to be precisely coordinated with life cycle progression . This is particularly important in multicellular organisms with haploid-diploid life cycles , where two different developmental programs , corresponding to the sporophyte and gametophyte , need to be deployed appropriately at different time points within a single life cycle . In the unicellular green alga Chlamydomonas , plus and minus gametes express two different HD TFs of the three amino acid loop extension ( TALE ) family called Gsm1 and Gsp1 ( Lee et al . , 2008 ) . When two gametes fuse to form a zygote , these two proteins heterodimerise and move to the nucleus , where they orchestrate the diploid phase of the life cycle . Gsm1 and Gsp1 belong to the knotted-like homeobox ( KNOX ) and BEL TALE HD TF classes , respectively . In the multicellular moss Physcomitrella patens , deletion of two KNOX genes , MKN1 and MKN6 , blocks initiation of the sporophyte program leading to conversion of this generation of the life cycle into a diploid gametophyte ( Sakakibara et al . , 2013 ) . Similarly , the moss BEL class gene BELL1 is required for induction of the sporophyte developmental program and ectopic expression of BELL1 in gametophytic tissues induces the development of apogametic sporophytes during the gametophyte generation of the life cycle ( Horst et al . , 2016 ) . In mosses , therefore , the KNOX and BEL class life cycle regulators have been recruited to act as master regulators of the sporophyte developmental program , coupling the deployment of this program with life cycle progression . P . patens KNOX and BEL proteins have been shown to form heterodimers ( Horst et al . , 2016 ) and it is therefore possible that life cycle regulation also involves KNOX/BEL heterodimers in this species . The filamentous alga Ectocarpus has emerged as a model system for the brown algae ( Cock et al . , 2015; Coelho et al . , 2012 ) . This alga has a haploid-diploid life cycle that involves alternation between multicellular sporophyte and gametophyte generations ( Figure 1A ) . A mutation at the OUROBOROS ( ORO ) locus has been shown to cause the sporophyte generation to be converted into a fully functional ( gamete-producing ) gametophyte ( Figure 1B ) ( Coelho et al . , 2011 ) . This mutation therefore induces a phenotype that is essentially identical to that observed with the P . patens mkn1 mkn6 double mutant , but in an organism from a distinct eukaryotic supergroup ( the stramenopiles ) , which diverged from the green lineage over a billion years ago ( Eme et al . , 2014 ) . Here we identify mutations at a second locus , SAMSARA , that also result in conversion of the sporophyte generation into a gametophyte . Remarkably , both OUROBOROS and SAMSARA encode TALE HD TFs and the two proteins associate to form a heterodimer . These observations indicate that TALE-HD-TF-based life cycle regulatory systems have very deep evolutionary origins and that they have been independently recruited in at least two eukaryotic supergroups to act as master regulators of sporophyte developmental programs . The ORO gene was mapped to a 34 . 5 kbp ( 0 . 45 cM ) interval on chromosome 14 using a segregating family of 2000 siblings derived from an ORO x oro cross and a combination of amplified fragment length polymorphism ( AFLP ) ( Vos et al . , 1995 ) and microsatellite markers . Resequencing of the 34 . 5 kbp interval in the oro mutant showed that it contained only one mutation: an 11 bp deletion in exon six of the gene with the LocusID Ec-14_005920 , which encodes a TALE homeodomain transcription factor . ( Figure 1C ) . A visual screen of about 14 , 000 UV-mutagenised germlings identified three additional life cycle mutants ( designated samsara-1 , samsara-2 and samsara-3 , abbreviated as sam-1 , sam-2 and sam-3 ) . The sam mutants closely resembled the oro mutant in that gamete-derived parthenotes did not adopt the normal sporophyte pattern of development but rather resembled gametophytes . Young , germinating individuals exhibited the wavy pattern of filament growth typical of the gametophyte and , at maturity , never produced unilocular sporangia ( the reproductive structures where meiosis occurs; Figure 1A ) , a structure that is uniquely observed during the sporophyte generation ( Figure 2A–C; Figure 2—figure supplement 1 ) . Moreover , the sam mutants exhibited a stronger negative phototrophic response to unilateral light than wild type sporophytes ( Figure 2D ) , a feature typical of gametophytes ( Peters et al . , 2008 ) that was also observed for the oro mutant ( Coelho et al . , 2011 ) . Genetic crosses confirmed that the sam mutants were fully functional ( i . e . gamete-producing ) gametophytes and complementation analysis indicated that the mutations were not located at the same genetic locus as the oro mutation ( Supplementary file 1 ) . Interestingly , hybrid sporophytes that were heterozygous for the sam mutations failed to produce functional unilocular sporangia . Wild type unilocular sporangia contain about a hundred haploid meio-spores produced by a single meiotic division followed by several rounds of mitotic divisions , whereas unilocular sporangia of SAM/sam heterozygotes never contained more than four nuclei indicating that abortion was either concomitant with or closely followed meiosis ( Figure 2F ) . This indicated either a dominant effect of the sam mutations in the fertile sporophyte or abortion of the sporangia due to arrested development of the two ( haploid ) meiotic daughter cells that carried the mutant sam allele . Note that no meiotic defects were observed in heterozygous sporophytes carrying the oro mutation . Ectocarpus sporophytes produce a diffusible factor that induces gametophyte initial cells or protoplasts of mature gametophyte cells to switch to the sporophyte developmental program ( Arun et al . , 2013 ) . The oro mutant is not susceptible to this diffusible factor ( oro protoplasts regenerate as gametophytes in sporophyte-conditioned medium ) indicating that ORO is required for the diffusible factor to direct deployment of the sporophyte developmental pathway ( Arun et al . , 2013 ) . We show here that the sam-1 mutant is also resistant to the action of the diffusible factor . Congo red staining of individuals regenerated from sam-1 protoplasts that had been treated with the diffusible factor detected no sporophytes , whereas control treatment of wild type gametophyte-derived protoplasts resulted in the conversion of 7 . 5% of individuals into sporophytes ( Figure 2E , Supplementary file 2 ) . Therefore , in order to respond to the diffusible factor , cells must possess functional alleles of both ORO and SAM . The Ectocarpus genome contains two TALE HD TFs in addition to the ORO gene . Resequencing of these genes in the three sam mutants identified three genetic mutations , all of which were predicted to severely affect the function of Ec-27_006660 ( Figure 2G ) . The identification of three disruptive mutations in the same gene in the three independent sam mutants strongly indicates that these are the causative lesions . Ec-27_006660 was therefore given the gene name SAMSARA ( SAM ) . ORO and SAM transcripts were most abundant in gametes ( Figure 3A ) , consistent with a role in initiating sporophyte development following gamete fusion . Interestingly , transcripts of both ORO and SAM were detected in both male and female gametes indicating that gametes of both sexes carry both ORO and SAM proteins . This situation therefore appears to differ from that observed in Chlamydomonas where GSP1 and GSM1 are expressed uniquely in the plus and minus gametes , respectively ( Lee et al . , 2008 ) . Whilst we cannot rule out the possibility that post-transcriptional regulatory processes result in ORO and SAM exhibiting sex-specific patterns of gamete expression , genetic evidence also supports a bi-sexual pattern of expression , at least for ORO , because complementation was observed when both male and female strains carrying the oro mutation were crossed with wild type strains ( Supplementary file 1 ) . This would not be expected if the ORO protein were supplied to the zygote uniquely by the male or the female gamete . Quantitative PCR experiments demonstrated that sporophyte and gametophyte marker genes ( Peters et al . , 2008 ) were down- and up-regulated , respectively , in sam mutant lines ( Figure 3B ) , as was previously demonstrated for the oro mutant ( Coelho et al . , 2011 ) . To investigate the genetic mechanisms underlying the switch from the gametophyte to the sporophyte program directed by the ORO and SAM genes , we characterised the gene expression networks associated with the two generations of the Ectocarpus life cycle . Comparative analysis of RNA-seq data for duplicate cultures of wild type sporophytes and wild type gametophytes grown under identical conditions ( libraries GBP-5 and GBP-6 and libraries GBP-7 and GBP-8 in Supplementary file 9 , respectively ) identified 1167 genes that were differentially regulated between the two generations ( 465 upregulated in the sporophyte and 702 upregulated in the gametophyte; Supplementary file 3 ) . The predicted functions of these generation-biased genes were analysed using a system of manually-assigned functional categories , together with analyses based on GO terms and KEGG pathways . The set of generation-biased genes was significantly enriched in genes belonging to two of the manually-assigned categories: ‘Cell wall and extracellular’ and ‘Cellular regulation and signalling’ and for genes of unknown function ( Figure 3C , Supplementary file 3 ) . Enriched GO terms also included several signalling- and cell wall-associated terms and terms associated with membrane transport ( Figure 3D , Supplementary file 4 ) . The gametophyte-biased gene set was enriched for several cell signalling KEGG pathways whereas the sporophyte-biased gene set was enriched for metabolic pathways ( Figure 3E , Supplementary file 5 ) . We also noted that the generation-biased genes included 23 predicted transcription factors and ten members of the EsV-1–7 domain family ( Macaisne et al . , 2017 ) ( Supplementary file 3 ) . The latter were significantly enriched in the sporophyte-biased gene set ( χ2 test p=0 . 001 ) . Both the sporophyte-biased and the gametophyte-biased datasets were enriched in genes that were predicted to encode secreted proteins ( Fisher's Exact Test p=2 . 02e−8 and p=4 . 14e−6 , respectively; Supplementary file 3 ) . Analysis of GO terms associated with the secreted proteins indicated a similar pattern of enrichment to that observed for the complete set of generation-biased genes ( terms associated with signalling , cell wall and membrane transport; Supplementary file 4 ) . Figure 3C illustrates the relative abundances of manually-assigned functional categories represented in the generation-biased genes predicted to encode secreted proteins . The lists of differentially expressed genes identified by the above analysis were used to select 200 genes that showed strong differential expression between the sporophyte and gametophyte generations . The pattern of expression of the 200 genes was then analysed in parthenotes of the oro and sam mutants and of a third mutant , immediate upright ( imm ) , which does not cause switching between life cycle generations ( Macaisne et al . , 2017 ) , as a control . Figure 3F shows that mutation of either ORO or SAM leads to upregulation of gametophyte generation genes and down-regulation of sporophyte generation genes , consistent with the switch from sporophyte to gametophyte phenotypic function . Moreover , oro and sam mutants exhibited similar patterns of expression but the patterns were markedly different to that of the imm mutant . Taken together with the morphological and reproductive phenotypes of the oro and sam mutants , this analysis supports the conclusion that ORO and SAM are master regulators of the gametophyte-to-sporophyte transition . HD TFs that act as life cycle regulators or mating type determinants often form heterodimeric complexes ( Banham , 1995; Horst et al . , 2016; Hull et al . , 2005; Kämper et al . , 1995; Lee et al . , 2008 ) . The ORO and SAM proteins were also shown to be capable of forming a stable heterodimer using an in vitro pull-down approach ( Figure 4 ) . Deletion analysis indicated that the interaction between the two proteins was mediated by their homeodomains . Analysis of sequence databases indicated that all brown algae possess three HD TFs , all of the TALE class , including orthologues of ORO and SAM ( Figure 5A , Supplementary file 6 ) . Comparison of brown algal ORO and SAM orthologues identified conserved domains both upstream and downstream of the HDs in both ORO and SAM ( Figure 5B , C ) . These domains do not correspond to any known domains in public domain databases and were not found in any other proteins in the public sequence databases . In particular , we did not detect any clear similarity with HD-associated domains that have been shown to be deeply conserved across eukaryotic TALE HD TFs ( Bürglin , 1997; Joo et al . , 2018 ) but we cannot rule out the possibility that the ORO and SAM proteins possess highly diverged versions of these domains . The HD was the only domain that was common to both the ORO and SAM proteins ( Figure 5 ) . To identify more distantly-related orthologues of ORO and SAM , we searched a broad range of stramenopile TALE HD TFs for the presence of characteristic ORO and SAM protein domains . Only one non-brown-algal protein , from the raphidophyte Heterosigma akashiwo , possessed similarity to these domains , allowing it to be classed tentatively as an ORO orthologue ( gene identifier 231575mod; Figure 5A , C , Supplementary file 6 ) . The transcriptome of this strain also included a truncated TALE HD TF transcript similar to SAM but more complete sequence data will be required to confirm orthology with SAM ( gene identifier 296151; Figure 5A , Supplementary file 6 ) . This analysis allowed the origin of ORO to be traced back to the common ancestor with the raphidophytes ( about 360 Mya; Brown and Sorhannus , 2010 ) but the rate of divergence of the non-HD regions of ORO and SAM precluded the detection of more distantly related orthologues . An additional search based on looking for TALE HD TF genes with intron positions corresponding to those of ORO and SAM did not detect any further orthologues ( Figure 5—figure supplement 1 ) . The analysis presented here demonstrates that two TALE HD TFs , which are capable of forming a heterodimer , are required for the deployment of the sporophyte program during the life cycle of the brown alga Ectocarpus . The parallels with life cycle regulation in the green lineage , where TALE HD TFs have also been shown to regulate deployment of the sporophyte program ( Horst et al . , 2016; Sakakibara et al . , 2013 ) , are striking . Knockout of the KNOX class TALE HD TF genes MKN1 and MKN6 in Physcomitrella patens result in conversion of the sporophyte generation into a functional gametophyte ( Sakakibara et al . , 2013 ) , essentially the same phenotype as that observed with Ectocarpus oro or sam mutants despite the fact that more than a billion years of evolution separate the two lineages ( Eme et al . , 2014 ) and that the two lineages independently evolved complex multicellularity . The similarities between life cycle regulators in the two eukaryotic supergroups suggests that they are derived from a common ancestral system that would therefore date back to early eukaryotic evolution . The ancient origin of this life cycle regulatory system is further supported by the fact that distantly-related homeodomain or homeodomain-like proteins act as mating type factors in both fungi and social amoebae ( Hedgethorne et al . , 2017; Hull et al . , 2005; Nasmyth and Shore , 1987; Van Heeckeren et al . , 1998 ) . Moreover , in Basidiomycetes these proteins regulate multiple aspects of sexual development including the formation of filaments , basidia and spores indicating recurrent recruitment as developmental regulators ( Banham , 1995; Hull et al . , 2005; Kämper et al . , 1995 ) . It has been proposed that the ancestral function of homeodomain-based life cycle regulators was to detect syngamy and to implement processes specific to the diploid phase of the life cycle such as repressing gamete formation and initiating meiosis ( Perrin , 2012 and references therein ) . With the emergence of complex , multicellular organisms , it would not have been surprising if additional processes such as developmental networks had come under the control of these regulators as this would have ensured that those developmental processes were deployed at the appropriate stage of the life cycle ( Cock et al . , 2014 ) . Indeed , it has been suggested that modifications to homeodomain-based regulatory circuits may have played an important role in the emergence of sporophyte complexity in the green lineage ( Bowman et al . , 2016; Lee et al . , 2008 ) . Key events may have included the replacement of the Gsp1-like class of BELL-related1 genes with alternative ( true BEL-class ) proteins and diversification of both the true BEL-class and the KNOX-class TALE HD TFs . In particular , the emergence and subfunctionalisation of two KNOX subfamilies early in streptophyte evolution is thought to have facilitated the evolution of more complex sporophyte transcriptional networks ( Furumizu et al . , 2015; Sakakibara et al . , 2013 ) . In the brown algae , ORO and SAM also function as major developmental regulators but , in this lineage , the emergence of a multicellular sporophyte has not been associated with a marked expansion of the TALE HD TF family . However , there does appear to have been considerable divergence of the ORO and SAM protein sequences during brown algal evolution , perhaps reflecting the evolution of new functions associated with multicellular development and divergence of the sporophyte and gametophyte developmental programs . Heterodimerisation appears to be a conserved feature of brown algal and green lineage TALE HD TFs ( Figure 4 and Lee et al . , 2008 ) despite the lack of domain conservation . However , in Ectocarpus heterodimerisation involves the ORO and SAM HDs whereas in Chlamydomonas , it is the KNOX1 and KNOX2 domains of Gsm1 that interact with the C-terminal region of Gsp1 ( which includes the HD , Ala and DE domains ) . In Chlamydomonas , the Gsp1 and Gsm1 proteins are carried specifically by plus and minus gametes , respectively , so that dimerisation of the two proteins allows the organism to detect syngamy and therefore the transition from a haploid to a diploid state . Based on transcript detection ( Figure 3A ) and genetic analysis ( Supplementary file 1 ) , ORO and SAM do not appear to exhibit sex-specific patterns of expression in gametes . This also appears to be the case in P . patens , where both the class 2 KNOX proteins MKN1 and MKN6 and the BEL proteins BELL1 and BELL2 are expressed in the egg ( Horst et al . , 2016; Sakakibara et al . , 2013 ) . It is not known whether these proteins are also expressed in the sperm , although BELL1 appears not to be ( Horst et al . , 2016 but see Ortiz-Ramírez et al . , 2017 ) . Taken together , these observations suggest that novel mechanisms may lead to the activation of TALE HD TF life cycle regulators in groups that have evolved complex multicellularity . In P . patens , the glutamate receptor GLR2 may be a component of such a mechanism ( Ortiz-Ramírez et al . , 2017 ) . It is perhaps not unexpected that the recruitment of TALE HD TFs to act as master regulators of complex developmental programs should be associated with a modification of the regulation of these systems themselves . Moreover , modified regulation of these TALE HD TFs may have had advantages in terms of life cycle flexibility . For example , in Ectocarpus , it would not be possible to deploy the sporophyte program in parthenogenetic gametes if gamete fusion was strictly required to create an ORO-SAM heterodimer . Interestingly , diploid sporophytes heterozygous for sam mutations exhibited abortive development of unilocular sporangia at a stage corresponding to the meiotic division of the mother cell . At first sight it might seem surprising that a gene should play an important role both directly following the haploid to diploid transition ( initiation of sporophyte development ) and at the opposite end of the life cycle , during the diploid to haploid transition ( meiosis ) . However , these phenotypes make more sense when viewed from an evolutionary perspective , if the ORO SAM system originally evolved as a global regulator of diploid phase processes . There is now accumulating evidence for an ancient role for HD TFs in life cycle regulation in both the bikont and unikont branches of the eukaryotic tree of life ( Hedgethorne et al . , 2017; Horst et al . , 2016; Hull et al . , 2005; Lee et al . , 2008; Sakakibara et al . , 2013 and this study ) . We show here that these systems have been adapted to coordinate life cycle progression and development in at least two multicellular eukaryotic lineages ( land plants and brown algae ) . The recruitment of TALE HD TFs as sporophyte program master regulators in both the brown and green lineages represents a particularly interesting example of latent homology , where the shared ancestral genetic toolkit constrains the evolutionary process in two diverging lineages leading to convergent evolution of similar regulatory systems ( Nagy et al . , 2014 ) . The identification of such constraints through comparative analysis of independent complex multicellular lineages provides important insights into the evolutionary processes underlying the emergence of complex multicellularity . One particularly interesting outstanding question is whether HD TFs also play a role in coordinating life cycle progression and development in animals ? Analysis of the functions of TALE HD TFs in unicellular relatives of animals may help provide some insights into this question . Sporophyte-conditioned medium , gametophyte-conditioned medium and protoplasts were produced as previously described ( Arun et al . , 2013 ) . Protoplasts were allowed to regenerate either in sporophyte-conditioned medium supplemented with osmoticum or in gametophyte-conditioned supplemented with osmoticum as a control . Congo red staining was used to distinguish sporophytes from gametophytes ( Arun et al . , 2013 ) . At least 60 individuals were scored per treatment per experiment . Results are representative of three independent experiments . The oro mutation has been shown to behave as a single-locus , recessive , Mendelian factor ( Coelho et al . , 2011 ) . AFLP analysis was carried out essentially as described by Vos et al . ( 1995 ) . DNA was extracted from 50 wild type and 50 oro individuals derived from a cross between the outcrossing line Ec568 ( Heesch et al . , 2010 ) and the oro mutant Ec494 ( Coelho et al . , 2011; Supplementary file 1 ) . Equal amounts of DNA were combined into two pools , for bulk segregant analysis . Pre-selective amplification was carried out with an EcoRI-anchored primer and an MseI-anchored primer , each with one selective nucleotide , in five different combinations ( EcoRI +T/MseI +G; EcoRI +T/MseI +A; EcoRI +C/MseI +G; EcoRI +C/MseI +A; EcoRI +A/MseI +C ) . These reactions were diluted 1:150 for the selective amplifications . The selective amplifications used an EcoRI-anchored primer and an MseI-anchored primer , each with three selective nucleotides , in various different combinations . The PCR conditions for both steps were 94°C for 30 s , followed by 20 cycles of DNA amplification ( 30 s at 94°C , 1 min at 56°C and 1 min at 72°C ) and a 5 min incubation at 72°C except that this protocol was preceded by 13 touchdown cycles involving a decrease of 0 . 7°C per cycle for the selective amplifications . PCR products were analysed on a LI-COR apparatus . This analysis identified two flanking AFLP markers located at 20 . 3 cM and 21 . 1 cM on either side of the ORO locus . For 23 ( 12 oro and 11 wild type ) of the 100 individuals , no recombination events were detected within the 41 . 4 cM interval between the two markers . Screening of these 23 individuals ( 11 wild type and 12 oro ) with the microsatellite markers previously developed for a sequence-anchored genetic map ( Heesch et al . , 2010 ) identified one marker within the 41 . 4 cM interval ( M_512 ) and located the ORO locus to near the bottom of chromosome 14 ( Cormier et al . , 2017 ) . Fine mapping employed a segregating population of 2000 individuals derived from the cross between the outcrossing line Ec568 and the oro mutant line ( Ec494 ) and an additional 11 microsatellite markers within the mapping interval ( Supplementary file 7 ) designed based on the Ectocarpus genome sequence ( Cock et al . , 2010 ) . PCR reactions contained 5 ng of template DNA , 1 . 5 μl of 5xGoTaq reaction buffer , 0 . 25 units of GoTaq-polymerase ( Promega ) , 10 nmol MgCl2 , 0 . 25 μl of dimethyl sulphoxide , 0 . 5 nmol of each dNTP , 2 pmol of the reverse primer , 0 . 2 pmol of the forward primer ( which included a 19-base tail that corresponded to a nucleotide sequence of the M13 bacteriophage ) and 1 . 8 pmol of the fluorescence marked M13 primer . The PCR conditions were 94°C for 4 min followed by 13 touch-down cycles ( 94°C for 30 s , 65–54°C for 1 min and 72°C for 30 s ) and 25 cycles at 94°C for 30 s , 53°C for 1 min and 72°C for 30 s . Samples were genotyped by electrophoresis on an ABI3130xl Genetic Analyser ( Applied Biosystems ) followed by analysis with Genemapper version 4 . 0 ( Applied Biosystems ) . Using the microsatellite markers , the oro mutation was mapped to a 34 . 5 kbp ( 0 . 45 cM ) interval , which contained five genes . Analysis of an assembled , complete genome sequence for a strain carrying the oro mutation ( strain Ec597; European Nucleotide Archive PRJEB1869; Ahmed et al . , 2014 ) together with Sanger method resequencing of ambiguous regions demonstrated that there was only one mutation within the mapped interval: an 11 bp deletion in the gene with the LocusID Ec-14_005920 . The sequence of the 34 . 5 kbp mapped interval containing the ORO gene ( chromosome 27 , 5463270–5497776 ) in the wild type Ectocarpus reference strain Ec32 included one short region of uncertain sequence 1026 bp downstream of the end of the ORO open reading frame . The sequence of this region was completed by PCR amplification and Sanger sequencing and confirmed by mapping Illumina read data to the corrected region . The corrected ORO gene region has been submitted to Genbank under the accession number KU746822 . Comparison of the reference genome ( strain Ec32 ) supercontig that contains the SAM gene ( sctg_251 ) with homologous supercontigs from several independently assembled draft genome sequences corresponding to closely related Ectocarpus sp . strains ( Ahmed et al . , 2014; Cormier et al . , 2017 ) indicated that sctg_251 was chimeric and that the first three exons of the SAM gene were missing . The complete SAM gene was therefore assembled and has been submitted to Genbank under the accession number KU746823 . Total RNA was extracted from wild-type gametophytes and partheno-sporophytes ( Ec32 ) and from sam-1 ( Ec374 ) and sam-2 ( Ec364 ) partheno-gametophytes using the Qiagen RNeasy Plant mini kit and any contaminating DNA was removed by digestion with Ambion Turbo DNase ( Life Technologies ) . The generation marker genes analysed were Ec-20_001150 and Ec-26_000310 ( sporophyte markers ) , and Ec-23_004240 and Ec-21_006530 ( gametophyte markers ) , which are referred to as IDW6 , IDW7 , IUP2 and IUP7 respectively , in Peters et al . ( 2008 ) . Following reverse transcription of 50–350 ng total RNA with the ImPro-II TM Reverse Transcription System ( Promega ) , quantitative RT-PCR was performed on a LightCycler 480 II instrument ( Roche ) . Reactions were run in 10 µl containing 5 ng cDNA , 500 nM of each oligo and 1x LightCycler 480 DNA SYBR Green I mix ( Roche ) . The sequences of the oligonucleotides used are listed in Supplementary file 8 . Pre-amplification was performed at 95°C for 5 min , followed by the amplification reaction consisting of 45 cycles of 95°C for 10 s , 60°C for 30 s and 72°C for 15 s with recording of the fluorescent signal after each cycle . Amplification specificity and efficiency were checked using a melting curve and a genomic DNA dilution series , respectively , and efficiency was always between 90% and 110% . Data were analysed using the LightCycler 480 software ( release 1 . 5 . 0 ) . A pair of primers that amplified a fragment which spanned intron 2 of the SAM gene was used to verify that there was no contaminating DNA ( Supplementary file 1-table supplement 8 ) . Standard curves generated from serial dilutions of genomic DNA allowed quantification for each gene . Gene expression was normalized against the reference gene EEF1A2 . Three technical replicates were performed for the standard curves and for each sample . Statistical analysis ( Kruskal-Wallis test and Dunn's Multiple Comparison Post Test ) was performed using the software GraphPad Prism5 . RNA for RNA-seq analysis was extracted from duplicate samples ( two biological replicates ) of approximately 300 mg ( wet weight ) of tissue either using the Qiagen RNeasy plant mini kit with an on-column Deoxyribonuclease I treatment or following a modified version ( Peters et al . , 2008 ) of the protocol described by Apt et al . ( 1995 ) . Briefly , this second protocol involved extraction with a cetyltrimethylammonium bromide ( CTAB ) -based buffer and subsequent phenol-chloroform purification , LiCl-precipitation , and DNAse digestion ( Turbo DNAse , Ambion , Austin , TX , USA ) steps . RNA quality and concentration was then analysed on a 1 . 5% agarose gel stained with ethidium bromide and a NanoDrop ND-1000 spectrophotometer ( NanoDrop products , Wilmington , DE , USA ) . Between 21 and 93 million sequence reads were generated for each sample on an Illumina Hi-seq2000 platform ( Supplementary file 9 ) . Raw reads were quality trimmed with Trimmomatic ( leading and trailing bases with quality below three and the first 12 bases were removed , minimum read length 50 bp ) ( Bolger et al . , 2014 ) . High score reads were aligned to the Ectocarpus reference genome ( Cock et al . , 2010; available at Orcae; Sterck et al . , 2012 ) using Tophat2 with the Bowtie2 aligner ( Kim et al . , 2013 ) . The mapped sequencing data was then processed with HTSeq ( Anders et al . , 2014 ) to obtain counts for sequencing reads mapped to exons . Expression values were represented as TPM and TPM >1 was applied as a filter to remove noise . Differential expression was detected using the DESeq2 package ( Bioconductor; Love et al . , 2014 ) using an adjusted p-value cut-off of 0 . 05 and a minimal fold-change of two . Genes that were differentially expressed in the gametophyte- and sporophyte generations were identified using duplicate RNA-seq datasets for whole gametophytes ( GBP-5 and GBP-6 , Supplementary file 9 ) and whole sporophytes ( GBP-7 and GBP-8 , Supplementary file 9 ) that had been grown in parallel under identical culture conditions . Heatmaps were generated using the Heatplus package for R ( Ploner , 2015 ) and colour schemes selected from the ColorBrewer project ( http://colorbrewer . org ) . The entire set of 16 , 724 protein-coding genes in the Ectocarpus Ec32 genome were manually assigned to one of 22 functional categories ( Supplementary file 10 ) and this information was used to determine whether sets of differentially expressed genes were enriched in particular functional categories compared to the entire nuclear genome ( χ2 test ) . Blast2GO ( Conesa and Götz , 2008 ) was used to detect enrichment of GO-terms associated with the genes that were consistently up- or downregulated in pairwise comparisons of the wild type gametophyte , the sam mutant and the oro mutant with the wild type sporophyte . Significance was determined using a Fisher exact test with an FDR corrected p-value cutoff of 0 . 05 . Sub-cellular localisations of proteins were predicted using Hectar ( Gschloessl et al . , 2008 ) . Sets of secreted proteins corresponded to those predicted to possess a signal peptide or a signal anchor . Gametophytes carrying oro or sam mutations did not exhibit any obvious phenotypic defects , despite the fact that both genes are expressed during this generation ( although SAM expression was very weak ) . In P . patens , GUS fusion experiments failed to detect expression of KNOX genes in the gametophyte but RT-PCR analysis and cDNA cloning has indicated that KNOX ( and BEL ) transcripts are expressed during this generation ( Champagne and Ashton , 2001; Sakakibara et al . , 2013; Sakakibara et al . , 2008 ) . However , no phenotypes were detected during the haploid protonema or gametophore stages in KNOX mutant lines ( Sakakibara et al . , 2013; Sakakibara et al . , 2008; Singer and Ashton , 2007 ) and the RT-PCR only amplified certain regions of the transcripts . Consequently , these results have been interpreted as evidence for the presence of partial transcripts during the gametophyte generation . To determine whether the ORO and SAM transcripts produced in Ectocarpus were incomplete , RNA-seq data from male and female , immature and mature gametophytes was mapped onto the ORO and SAM gene sequences . This analysis indicated that full-length transcripts of both the ORO and SAM genes are produced during the gametophyte generation ( Figure 3—figure supplement 1 ) . Pull-down assays were carried out using the MagneGSTTM Pull-Down System ( Promega , Madison , WI ) by combining human influenza hemagglutinin ( HA ) -tagged and glutathione S-transferase ( GST ) fusion proteins . In vitro transcription/translation of HA-tagged ORO proteins was carried out using the TNT Coupled Wheat Germ Extract System ( Promega , Madison , WI ) . GST-tagged SAM proteins were expressed in Escherichia coli . Protein production was induced by adding IPTG to a final concentration of 2 mM and shaking for 20 hr at 16°C . After the capture phase , beads were washed four times with 400 μL of washing buffer ( 0 . 5% IGEPAL , 290 mM NaCl , 10 mM KCl , 4 . 2 mM Na2HPO4 , 2 mM KH2PO4 , at pH 7 . 2 ) at room temperature . Beads were then recovered in SDS-PAGE loading buffer , and proteins analysed by SDS-PAGE followed by ClarityTM chemiluminescent detection ( Biorad , Hercules , CA ) . The anti-HA antibody ( 3F10 ) was purchased from Roche , and the anti-GST antibody ( 91G1 ) from Ozyme . Searches for homeodomain proteins from additional brown algal or stramenopile species were carried out against the NCBI , Uniprot , oneKP ( Matasci et al . , 2014 ) and iMicrobe databases and against sequence databases for individual brown algal ( Saccharina japonica , Ye et al . , 2015; Cladosiphon okamuranus , Nishitsuji et al . , 2016 ) and stramenopile genomes ( Nannochloropsis oceanica , Aureococcus anophagefferens , Phaeodactylum tricornutum , Thalassiosira pseudonana , Pseudo-nitzschia multiseries ) and transcriptomes ( Vaucheria litorea , Heterosigma akashiwo ) using both Blast ( Blastp or tBlastn ) and HMMsearch with a number of different alignments of brown algal TALE HD TF proteins . As the homeodomain alone does not provide enough information to construct well-supported phylogenetic trees , searches for ORO and SAM orthologues were based on screening for the presence of the additional protein domains conserved in brown algal ORO and SAM proteins . As intron position and phase was strongly conserved between the homeoboxes of ORO and SAM orthologues within the brown algae , this information was also used to search for ORO and SAM orthologues in other stramenopile lineages . However , this analysis failed to detect any additional candidate ORO or SAM orthologues . These observations are consistent with a similar analysis of plant homeobox introns , which showed that intron positions were strongly conserved in recently diverged classes of homeobox gene but concluded that homeobox introns were of limited utility to deduce ancient evolutionary relationships ( Mukherjee et al . , 2009 ) . GenomeView ( Abeel et al . , 2012 ) was used together with publically available genome and RNA-seq sequence data ( Nishitsuji et al . , 2016; Ye et al . , 2015 ) to improve the gene models for some of the brown algal TALE HD TFs ( indicated in Supplementary file 6 by adding the suffix ‘mod’ for modified to the protein identifier ) . All the stramenopile species analysed in this study possessed at least two TALE HD TFs , with some species possessing as many as 14 ( Supplementary file 6 ) . Note that genomes of several diverse stramenopile lineages outside the brown algae were predicted to encode proteins with more than one HD ( Supplementary file 6 ) . It is possible that these proteins have the capacity to bind regulatory sequences in a similar manner to heterodimers of proteins with single HDs . Multiple alignments were generated with Muscle in MEGA7 ( Tamura et al . , 2011 ) . Phylogenetic trees were then generated with RAxML ( Stamatakis , 2015 ) using 1000 bootstrap replicates and the most appropriate model based on an analysis in MEGA7 . Domain alignments were constructed in Jalview ( http://www . jalview . org/ ) and consensus sequence logos were generated with WebLogo ( http://weblogo . berkeley . edu/logo . cgi ) . Intrinsic disorder in protein folding was predicted using SPINE-D ( Zhang et al . , 2012 ) , low complexity regions with SEG ( default parameters , 12 amino acid window; Wootton , 1994 ) and secondary structure with PSIPRED ( Buchan et al . , 2013 ) . The conserved domains that flank the homeodomains in the ORO and SAM proteins share no detectable similarity with domains that are associated with TALE HDs in the green ( Viridiplantae ) lineage , such as the KNOX , ELK and BEL domains . Interestingly , both the ORO and SAM proteins possess regions that are predicted to be highly disordered ( Figure 5B ) . Intrinsically disordered region are a common feature in transcription factors and the flexibility conferred by these regions is thought to allow them to interact with a broad range of partners ( Niklas et al . , 2015 ) , a factor that may be important for master developmental regulators such as the ORO and SAM proteins .

Brown algae and land plants are two groups of multicellular organisms that have been evolving independently for over a billion years . Their last common ancestor is thought to have existed as a single cell; then , complex multicellular organisms would have appeared separately in each lineage . Comparing brown algae and land plants therefore helps us understand the rules that guide how multicellular organisms evolve from single-celled ancestors . During their life cycles , both brown algae and land plants alternate between two multicellular forms: the gametophyte and the sporophyte . The gametophyte develops sexually active reproductive cells , which , when they merge , create the sporophyte . In turn , spores produced by the sporophyte give rise to the gametophyte . Specific developmental programs are deployed at precise points in the life cycle to make either a sporophyte or a gametophyte . Two proteins known as TALE HD transcription factors help to control the life cycle of single-celled algae related to land plants . Similar proteins are also required for the sporophyte to develop at the right time in land plants known as mosses . This suggests that , when multicellular organisms emerged in this lineage , life cycle TALE HD transcription factors were recruited to orchestrate the development of the sporophyte . However , it was not clear whether TALE HD transcription factors play equivalent roles in other groups , such as brown algae . To address this question , Arun , Coelho et al . examined two mutants of the brown alga Ectocarpus , which produce gametophytes when the non-mutated alga would have made sporophytes . Genetic analyses revealed that these mutated brown algae carried changes in two genes that encode TALE HD transcription factors , indicating that these proteins also regulate the formation of sporophytes in brown algae . Taken together , the results suggest that TALE HD transcription factors were originally tasked with controlling life cycles , and then have been independently harnessed in both land plants and brown algae to govern the formation of sporophytes . This means that , regardless of lineage , the same fundamental forces may be shaping the evolutionary paths that lead to multicellular organisms . Proteins similar to TALE HD transcription factors also regulate life cycles in other groups such as fungi and social amoebae , which indicates that their role is very ancient . It now remains to be explored whether such proteins control life cycles and developmental programs in other multicellular organisms , such as animals .

2019

Convergent recruitment of TALE homeodomain life cycle regulators to direct sporophyte development in land plants and brown algae

N-Methyl-D-aspartate receptors ( NMDA-Rs ) are ion channels that are important for synaptic plasticity , which is involved in learning and drug addiction . We show enzymatic targeting of an NMDA-R antagonist , MK801 , to a molecularly defined neuronal population with the cell-type-selectivity of genetic methods and the temporal control of pharmacology . We find that NMDA-Rs on dopamine neurons are necessary for cocaine-induced synaptic potentiation , demonstrating that cell type-specific pharmacology can be used to dissect signaling pathways within complex brain circuits . N-Methyl-D-aspartate receptors ( NMDA-Rs ) are glutamate-gated ion channels that are critical for the regulation of synaptic functions in the central nervous system , such as synaptic plasticity ( Malenka and Nicoll , 1993; Collingridge et al . , 2004 ) . NMDA-R dependent synaptic plasticity plays an important role in learning . This includes learning that can also have maladaptive consequences , for example sensitization of drug-related behaviors ( Kalivas and Alesdatter , 1993; Ungless et al . , 2001 ) . However , because NMDA-Rs are expressed in most cell types in the brain ( Conti et al . , 1997; Verkhratsky and Kirchhoff , 2007 ) , it is a considerable challenge to selectively assess the importance of NMDA-R mediated synaptic plasticity in specific cell types . The functional contribution of NMDA-Rs to physiology and behavior can be examined using either genetic or pharmacological methods . Small molecule antagonists allow rapid blockade of NMDA-Rs , which has been critical for analyzing synaptic plasticity processes that operate over short timescales . However , existing pharmacological agents are not cell type-selective . Alternatively , genetic methods can be used , typically involving Cre-recombinase ( Cre ) -mediated excision of a loxP flanked exon in Grin1 , which encodes a critical NMDA-R subunit that is required for channel function . This allows cell type-specific ablation of NMDA-R function by crossing to a mouse line with cell type-selective expression of Cre or by spatially targeted injection of Cre-expressing viral vectors . A major limitation is that genetic methods are not fast; they typically disrupt NMDA-Rs over timescales ranging from a week to the lifetime of an animal , during which time substantial compensatory effects in circuit function are observed ( Engblom et al . , 2008; Zweifel et al . , 2008 ) . Here , we describe an approach to combine the cell type-selectivity of genetic methods with the temporal control of pharmacology by targeting a small molecule NMDA-R antagonist to molecularly defined neuronal cell types ( Figure 1A ) . 10 . 7554/eLife . 10206 . 003Figure 1 . Cell type specific pharmacology for NMDA-R inhibition . ( A , B ) Strategy for cell type-specific targeting of a masked MK801 molecule to a defined subpopulation of neurons in brain tissue that transgenically express an unmasking enzyme . The masked MK801 enters every cell but MK801 is unmasked in only those cells that transgenically express the enzyme porcine liver esterase ( PLE ) . ( B ) The intracellularly liberated MK801 can block NMDA-Rs in the PLE-expressing neurons . Yellow triangle , glutamate . ( C ) Synthesis of CM-MK801 . ( a ) chloromethylchlorosulfate; ( b ) NaI; ( c ) 4-hydroxy-3-nitrobenzaldehyde , 44% for a-c; ( d ) NaBH4 , 48%; ( e ) 4-nitrophenyl chloroformate , 65%; ( f ) MK801 , 32% . ( D ) Enzymatic hydrolysis of CM-MK801 by PLE is followed by spontaneous 1 , 6-elimination to liberate MK801 . DOI: http://dx . doi . org/10 . 7554/eLife . 10206 . 003 MK801 is a potent non-competitive antagonist of NMDA-Rs that use-dependently blocks the channel in the open ( glutamate-bound ) state ( Wong et al . , 1986; Woodruff et al . , 1987 ) . Furthermore , MK801 can bind and block the NMDA-R channel ion conductance pore from the intracellular compartment ( Berretta and Jones , 1996; Bender et al . , 2006 ) . To develop a cell type-specific pharmacological strategy to rapidly block NMDA-Rs in subpopulations of neurons within brain tissue , we synthesized an inert masked MK801 derivative that could be unmasked inside cells by transgenically expressed porcine liver esterase ( PLE ) ( Figure 1B ) ( Tian et al . , 2012 ) . A carboxymethylpropyl ester group has been used previously in brain tissue to mask a fluorophore and a kinesin inhibitor , and extensive analysis showed that this group was stable to endogenous neuronal esterases but was cleaved in neurons transgenically expressing PLE ( Tian et al . , 2012 ) . Because MK801 contains a secondary amine that is critical for NMDA-R antagonism , we used a 4-hydroxy-3-nitrobenzyl carbamate ( Gesson et al . , 1994 ) to link MK801 and a carboxymethylpropyl ester to generate CM-MK801 ( Figure 1C ) . Enzymatic cleavage of the CM ester inside neurons expressing PLE leads to spontaneous 1 , 6-elimination of the vinylogous hemiaminal , liberating MK801 ( Figure 1D ) . Three critical properties of the CM-MK801/PLE system must be evaluated for cell type-specific pharmacology: 1 ) CM-MK801 should be efficacious and completely block NMDA-Rs in PLE+ neurons , 2 ) PLE− neurons should not unmask CM-MK801 , and 3 ) unmasked MK801 should be confined to PLE+ neurons without affecting adjacent PLE− neurons . To examine the cell type selectivity of this reagent , we measured NMDA-R excitatory postsynaptic currents ( EPSCs ) in brain slices from the mouse cerebral cortex . In utero electroporation ( IUE ) of the bicistronic plasmid pCAG::PLE-IRES-mCherry into the developing mouse brain allowed PLE and mCherry co-expression in only a subset of cortical layer 2/3 pyramidal neurons . Consistent with previous work , PLE expression was well tolerated in cortical neurons ( Tian et al . , 2012 ) . We isolated NMDA-R EPSC responses for cortical layer 4→layer 2/3 ( L4→L2/3 ) projections by recording neurons in L2/3 while electrically stimulating presynaptic neurons in L4 during constitutive pharmacological block of other ionotropic glutamate receptors ( AMPA and kainate receptors ) and GABA receptors . Brain slices were incubated with CM-MK801 ( 5 µM ) , and L4→L2/3 NMDA-R EPSCs were recorded using the perforated patch configuration ( necessary to avoid dilution of enzyme-liberated MK801 by diffusion into the patch pipette that was observed with whole cell recordings ) in mCherry-expressing ( PLE+ ) and adjacent non-expressing ( PLE− ) neurons ( Figure 2A , B ) . The NMDA-R EPSC amplitude ( Figure 2C ) and mean NMDA-R charge transfer ( QNMDA: mean integrated current , Figure 2E ) decreased in a use-dependent manner in PLE+ neurons to a level similar to treatment with the competitive NMDA-R antagonist D- ( − ) -2-amino-5-phosphonopentanoic acid ( AP5 , 100 µM ) ( Figure 2E–G; QNMDA inhibition: CM-MK801 , 82% ± 4; AP5 , 84% ± 4; n = 11; unpaired t test , p = 0 . 75 ) . This shows that CM-MK801 can completely block NMDA-R currents in neurons expressing the PLE transgene . In contrast , in the presence of CM-MK801 , PLE− neurons that were adjacent to PLE+ neurons showed scant reduction of QNMDA ( Figure 2D , F ) . The slight QNMDA reduction in PLE− neurons exposed to CM-MK801 was not significantly different than QNMDA in neurons in the absence of CM-MK801 ( QNMDA inhibition: PLE−/CM-MK801: 16% ± 8 . 9 , n = 6; PLE+: 33% ± 8 . 6 , n = 4; PLE−: 21% ± 6 . 3; ANOVA , F2 , 13 = 2 . 3 , p = 0 . 15 ) , thus QNMDA reduction in PLE− neurons was due to modest NMDA-R synaptic rundown ( Rosenmund and Westbrook , 1993 ) and not an effect of MK801 in PLE− neurons . These results show that CM-MK801 can be targeted specifically to neurons expressing a transgenic esterase and that CM-MK801 is selective , where MK801 , once liberated , is confined to the PLE-expressing cell population . 10 . 7554/eLife . 10206 . 004Figure 2 . Cellular selectivity of CM-MK801/PLE ester/esterase pair . ( A , B ) Schematic diagrams of the experimental procedure . Perforated patch voltage clamp recordings of NMDA-R synaptic currents were made on both ( A ) PLE/mCherry-expressing ( PLE+ , red ) layer 2/3 ( L2/3 ) cortical neurons and ( B ) adjacent neurons lacking PLE ( PLE− , gray ) while electrically stimulating presynaptic neurons in layer 4 ( L4 ) in the presence of CM-MK801 . ( C , D ) Treatment of the brain slices with CM-MK801 ( 5 µM ) during L4→L2/3 synaptic stimulation showed gradual use-dependent reduction of the NMDA-R EPSC amplitude in ( C ) PLE+ but not in ( D ) PLE− neurons , indicating that the CM-MK801 was converted to MK801 selectively in PLE+ neurons without spilling over to adjacent PLE− neurons . Subsequent addition of the competitive NMDA-R antagonist , AP5 , suppressed the NMDA-R EPSC in PLE− neurons but not in PLE+ neurons , which were fully blocked by CM-MK801 . Electrical stimuli were delivered every 15 s . ( E , F ) Overlaid NMDA-R averaged excitatory postsynaptic currents ( EPSCs ) from ( E ) PLE+ and ( F ) PLE− neurons showing the initial response ( black ) , responses after electrical stimulation ( late , blue ) and the response in the presence of AP5 ( green ) . Electrical stimulation artifact reduced for clarity . ( G ) Grouped data for NMDA-R EPSC charge transfer ( QNMDA ) inhibition for PLE+ ( n = 14 ) and PLE− ( n = 17 ) neurons during CM-MK801 treatment and subsequent exposure to AP5 . For comparison , dotted line shows QNMDA inhibition in PLE+ and PLE− neurons in the absence of CM-MK801 , which is due to modest synaptic rundown . Data is represented as mean and error bars indicate s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 10206 . 004 An application of cell type-specific pharmacology for NMDA-R is to examine synaptic plasticity in molecularly defined neurons . One problem that has been difficult to address with existing tools is the role of NMDA-Rs in cocaine-induced synaptic plasticity . Cocaine blocks dopamine reuptake , which elevates extracellular concentrations . Earlier work demonstrated that cocaine and elevated dopamine leads to long-term synaptic potentiation of AMPA receptor-type ( AMPA-R ) synaptic currents in dopamine ( DA ) neurons in the ventral tegmental area ( VTA ) ( Ungless et al . , 2001; Argilli et al . , 2008 ) ( Figure 3A ) , a key neuronal cell type associated with reward and addiction . Further analysis demonstrated that synaptic potentiation involved signaling through dopamine receptor 5 , new protein synthesis , and insertion of AMPA-R subunits ( Argilli et al . , 2008 ) . These experiments also showed that NMDA-R blockade , by systemic application of MK801 , prevented synaptic potentiation in DA neurons ( Ungless et al . , 2001 ) ( Figure 3A ) . However , it has been challenging to establish the functional importance of NMDA-Rs specifically in DA neurons for cocaine-induced synaptic AMPA-R potentiation , as opposed to a possible indirect process involving NMDA-Rs on other cell types . To disrupt this process specifically in DA neurons , studies using dopamine neuron-specific knockout of Grin1 were performed but , in several instances , these were confounded by compensatory effects in which synaptic AMPA-Rs were constitutively potentiated in DA neurons in the absence of cocaine ( Engblom et al . , 2008; Zweifel et al . , 2008 ) . Even virally mediated deletion of Grin1 showed these compensatory effects within 1 week ( Zweifel et al . , 2008 ) . In contrast , another study found that use of tamoxifen-activiated Cre-ER for selective Grin1 ablation in DA neurons prevented a cocaine-induced shift in the rectification of AMPA-Rs after 1 week ( Engblom et al . , 2008 ) , which indicated that dopamine neuron NMDA-Rs were required for cocaine-induced plasticity . 10 . 7554/eLife . 10206 . 005Figure 3 . Cell type-selective blockade of cocaine-induced synaptic plasticity in dopamine neurons . ( A ) Schematic diagram of cocaine-induced synaptic plasticity . Cocaine increases extracellular dopamine which leads to NMDA-R dependent upregulation of synaptic AMPA ( α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid ) receptors . It is unclear if this is through NMDA receptors on DA neurons or through NMDA-R on other cell types that may have a subsequent effect on AMPA-R activity in DA neurons . ( B ) Confocal images of ventral tegmental area ( VTA ) brain slices from dopamine transporter Cre ( Slc6a3Cre ) mice transduced with Cre-dependent viruses expressing either PLE and mCherry ( left ) or EGFP ( right ) on opposite sides of the VTA . Scale , 50 μm . ( C ) Experimental protocol where VTA-containing brain slices were incubated with CM-MK801 for 40 min , followed by addition of cocaine for 10 min and washout . 3–5 hr later , miniature excitatory postsynaptic AMPA-R currents ( mEPSCs ) in DA neurons expressing either GFP ( PLE− ) or mCherry ( PLE+ ) were recorded . Inset , images of ( left ) patch pipette recording from dopamine neuron ( right ) expressing mCherry and PLE . Scale , 5 μm . ( D ) For brain slices treated with CM-MK801 , AMPAR-mediated mEPSC amplitude was increased in PLE− VTA DA neurons following cocaine treatment . Synaptic potentiation was blocked in PLE+ neurons treated with cocaine , and the mEPSC amplitude was similar to that in PLE+ or PLE− neurons that were not exposed to cocaine . ( E ) mEPSC frequency was unaffected by cocaine or CM-MK801 . ( F ) Resting neuronal membrane potential ( Vm ) , input resistance ( Rin ) , and membrane capacitance ( Cm ) were not changed by expression of PLE , or exposure to cocaine and CM-MK801 . Data is represented as mean and error bars indicate s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 10206 . 005 We sought to investigate cocaine-induced synaptic plasticity by rapid blockade of NMDA-Rs with the temporal control of pharmacology and the cell-type selectivity of genetics by directing MK801 to DA neurons using cell type-specific pharmacology . For this , we co-expressed PLE and the fluorescent protein mCherry in DA neurons using a Cre-dependent recombinant adeno-associated viral vector , rAAV2/1-Synapsin::FLEX-rev-PLE-2a-mCherry ( 2a: ribosomal skip sequence [Donnelly et al . , 2001] ) that was targeted unilaterally into the VTA of Slc6a3Cre mice , a mouse line that selectively expresses Cre-recombinase in DA neurons ( Zhuang et al . , 2005 ) . In the same mice , control ( PLE− ) DA neurons were labeled using a Cre-dependent virus expressing EGFP that was targeted to DA neurons in the VTA on the other side of the brain . To investigate the involvement of dopamine neuron NMDA receptors , we adopted a previously described brain slice model that was sufficient to recapitulate the processes underlying cocaine-mediated synaptic AMPA-R potentiation in DA neurons ex vivo ( Argilli et al . , 2008 ) . VTA brain slices containing DA neurons expressing PLE/mCherry and EGFP on opposite sides of the brain ( Figure 3B ) were incubated with CM-MK801 ( 5 µM ) in artificial cerebrospinal fluid ( ACSF ) followed by a short exposure to cocaine ( 5 µM in ACSF , 10 min ) or vehicle and afterwards the brain slices were transferred to ACSF with CM-MK801 for 3–5 more hours . After this induction period , miniature excitatory postsynaptic AMPA-R currents ( mEPSCs ) were recorded from DA neurons , which could be isolated by blocking inhibitory GABA-Rs and voltage gated sodium channels . As shown previously in VTA-containing brain slices exposed to cocaine ( Argilli et al . , 2008 ) , DA neurons showed significantly elevated potentiation of mEPSC amplitudes in PLE−/GFP+ DA neurons in the presence of CM-MK801 , which further demonstrates the lack of activity for CM-MK801 in PLE− neurons . Strikingly though , for PLE+/mCherry+ DA neurons in CM-MK801 and cocaine , mEPSC amplitudes were not elevated and were similar to PLE+ or PLE− DA neurons in CM-MK801 that had not been exposed to cocaine . ( Figure 3D; Coc+/PLE− , 19 . 9 ± 0 . 9 pA , n = 17; Coc+/PLE+ , 12 . 7 ± 0 . 8 pA , n = 14; Coc−/PLE− , 13 . 9 ± 0 . 6 pA , n = 12; Coc−/PLE+ , 12 . 8 ± 0 . 7 pA , n = 15; ANOVA F3 , 57 = 19 . 0 , p < 0 . 001 ) . The frequency of mEPSCs in DA neurons was not affected by CM-MK801 or cocaine ( Figure 3E , ANOVA F3 , 57 = 0 . 30 , p = 0 . 82 ) . In addition , the membrane properties of DA neurons after CM-MK801 exposure were indistinguishable in PLE+ or PLE− cells treated or untreated with cocaine , further indicating that CM-MK801 is not perturbative to resting neuronal properties . Therefore , using cell type-specific pharmacology in an ex vivo preparation , these experiments demonstrate that NMDA-Rs in DA neurons are necessary for cocaine-induced potentiation of synaptic AMPA-R currents in this cell type ex vivo , which is consistent with a prior report that examined this problem using Cre-ER in vivo ( Engblom et al . , 2008 ) . Taken together , these experiments demonstrate the effectiveness of a cell type-specific pharmacology approach to selectively block NMDA-Rs in a molecularly defined neuron population using an esterase–ester pair . Cell type-specific pharmacology for activating genetically engineered ion channels has been described ( Slimko et al . , 2002; Magnus et al . , 2011 ) , and here we demonstrate rapid cell type-specific blockade of a native ion channel with a small molecule . Our data indicates that MK801 was liberated only in PLE+ but not in neighboring PLE− neurons , showing that this approach can be used to block NMDARs with cellular specificity . The absence of a neighbor effect with CM-MK801/PLE is consistent with experiments targeting MK801 to individual neurons in a patch pipette ( Bender et al . , 2006 ) . We applied this to DA neurons to further examine the role for NMDA-Rs in cocaine-induced synaptic potentiation , which has been investigated in several pharmacological ( Ungless et al . , 2001; Argilli et al . , 2008 ) and genetic studies ( Engblom et al . , 2008; Zweifel et al . , 2008 ) . Our experiments using cell type-specific pharmacology provide additional evidence for the necessity of NMDA-Rs in cocaine-induced long-term synaptic potentiation in DA neurons ex vivo . One limitation of the approach presented here is that the aqueous solubility of CM-MK801 ( 5–10 μM ) make it less suitable for direct injection into the brain because of the requirement for elevated levels of co-solvents , such as dimethyl sulfoxide ( DMSO ) ( >0 . 1% ) , which resulted in VTA-containing brain slices that were unsuitable for recordings . Because of this , it was not possible for us to determine if AMPA-R potentiation occurs after inactivation of NMDA-R in vivo using the chemogenetic method described here . Therefore , additional improvements to cell type-specific pharmacological techniques are needed to facilitate manipulations in mammalian brains in vivo . However , because functional NMDA receptors are widely expressed throughout the mammalian nervous system , ex vivo approaches to rapidly disrupt NMDA-R signaling with cell type specificity have broad utility . NMDA-Rs are expressed in both neurons and glia ( Conti et al . , 1997; Verkhratsky and Kirchhoff , 2007 ) , and selective genetic access to these cell classes in the brain could allow for dissection of their relative role in synaptic function . Moreover , neurons have been classified into many different molecularly defined cell types , which are accessible through a growing resource of Cre-expressing mouse lines ( Gong et al . , 2007 ) . In combination with Cre-dependent viral approaches ( Atasoy et al . , 2008 ) , this will allow the role of NMDA-Rs to be selectively evaluated in any molecularly defined neuron population in manner similar to what we demonstrated for DA neurons . In addition , the relative roles of presynaptic and postsynaptic NMDA-Rs has been extensively analyzed in brain slices ( Corlew et al . , 2008; Rodriguez-Moreno and Paulsen , 2008 ) , which PLE/CM-MK801 may facilitate because it offers a new method for selectively blocking NMDA-Rs in defined subpopulations of neurons . Finally , the strategy applied here for selective delivery of MK801 can be extended to other small molecules for use in brain tissue ( Tian et al . , 2012 ) , allowing for an approach that combines chemistry , molecular genetics , and neurobiology to dissect cell signaling pathways in specific cell types in the central nervous system . As previously described ( Saito and Nakatsuji , 2001 ) , timed pregnant female C57Bl/6NCrl mice were purchased from Charles River laboratories and E15 embryos were electroporated following pressure microinjection ( Picrospritzer ) into the right lateral ventricle of a DNA mixture ( 0 . 5 µg/µl , approximately 200 nl; pCAG::PLE-IRES-mCherry ) encoding PLE and a mCherry signal marker . Postoperative analgesia was provided ( buprenorphine was administrated intraperitoneally at a dose of 0 . 1 mg/kg along with ketoprofen administrated subcutaneously at a dose of 5 mg/kg ) . Electroporated mice were sacrificed between postnatal days 14–21 for brain slice recordings from transfected layer 2/3 cortical pyramidal neurons . As previously described ( Yang et al . , 2011 ) , animals were deeply anesthetized with isofluorane , decapitated and the brains removed in ice cold sectioning solution . The brains were then mounted on a stage using Krazy Glue , and coronal brain slices ( 300 µm thickness ) were cut using a cooled tissue slicer ( Vibratome , 2000 ) . Slices were prepared in chilled cutting solution containing the following ( in mM ) : 110 choline chloride , 2 . 5 KCl , 1 . 25 mM NaH2PO4 , 2 CaCl2 , 7 MgSO4 , 25 D-glucose , 3 . 1 Na-pyruvate , and 11 . 6 Na-L-ascorbate , aerated with 95% O2 / 5% CO2 . Brain slices were transferred to the chamber containing ( in mM ) : 119 NaCl , 25 NaHCO3 , 11 D-glucose , 2 . 5 KCl , 1 . 25 MgCl2 , 2 CaCl2 , and 1 . 25 NaH2PO4 , aerated with 95% O2/5% CO2 at 34 °C for 30 min and moved to room temperature until transferred to a recording chamber on the stage of an Olympus ( Tokyo , Japan ) BW51XI microscope . Electrophysiological recordings were performed as described previously ( Yang et al . , 2011 ) using a multiple 700B amplifier ( Molecular Devices ) . Slices were transferred to the recording chamber on the stage of an Olympus BW51XI microscope and perfused at a rate of 1–2 ml/min with ACSF ( 28–30°C ) containing the following ( in mM ) : 119 NaCl , 25 NaHCO3 , 11 D-glucose , 2 . 5 KCl , 1 . 25 MgCl2 , 2 CaCl2 , and 1 . 25 NaH2PO4 , aerated with 95% O2/5% CO2 , also 5 µM MK-801 . Perforated patch recordings were made on either PLE/mCherry expressing neurons expressing neurons or adjacent non-expressing neurons using electrodes with tip resistances of 4–5 MΩ immediately after circulating the ACSF with CM-MK801 . 40 to 50 min later , ACSF was switched to 0 Mg2+ in the presence of picrotoxin ( 50 µM ) and CNQX ( 10 µM ) . The intracellular solution for voltage-camp recordings contained the following ( in mM ) : 125 CsCl , 5 NaCl , 10 HEPES , 0 . 6 EGTA , 4 Mg-ATP , 0 . 3 Na2GTP , 10 lidocaine N-ethyl bromide ( QX-314 ) and Gramicidin , pH 7 . 35 and adjusted to 290 mOsm . Gramicidin ( Sigma , St . Louis , MO , United States ) was dissolved in dimethylsulfoxide ( Sigma ) ( 1–2 mg/ml ) then diluted in the pipette filling solution to a final concentration of 0 . 2–5 µg/ml . The holding potential for voltage-clamp recordings was—70 mV , and responses were digitized at 10 kHz . NMDAR-EPSCs were elicited by 0 . 1 ms electrical stimuli ( 100–300 µA ) every 15 s with an isolated pulse stimulator ( WPI ) using a concentric bipolar electrode ( FHC , CBAEC75 ) placed at layer 4 of cortex . 2-amino-5-phosphonovaleric acid ( 100 µM , AP5 ) was added in the end of most experiments . Experimental techniques were similar to those reported previously ( Argilli et al . , 2008 ) . 9–10 days after viral injection , mice expressing Cre recombinase in DA neurons ( Slc6a3Cre mice ) were deeply anaesthetized with isofluorane and decapitated . Coronal brain slices ( 200 µm ) containing VTA were prepared in chilled cutting solution containing ( in mM ) : 110 choline chloride , 2 . 5 KCl , 1 . 25 mM NaH2PO4 , 2 CaCl2 , 7 MgSO4 , 25 D-glucose , 3 . 1 Na-pyruvate , and 11 . 6 Na-L-ascorbate , aerated with 95% O2 / 5% CO2 . Brain slices were incubated at 34 °C for 30 min in ACSF containing ( in mM ) : 119 NaCl , 25 NaHCO3 , 11 D-glucose , 2 . 5 KCl , 1 . 25 MgCl2 , 2 CaCl2 , and 1 . 25 NaH2PO4 , aerated with 95% O2/5% CO2 and then transferred to the chamber at room temperature containing ACSF with CM-MK801 ( 5 µM ) for 40 min . Cocaine ( 5 µM ) dissolved in ACSF or ACSF alone were added to the chamber . After 10 min , the brain slices were rinsed twice and transferred to another chamber containing ACSF and CM-MK801 . Electrophysiological recordings were performed 3–5 hr later . The brain slices were transferred to a recording chamber on the stage of an Olympus ( Tokyo , Japan ) BW51XI microscope . Dopamine neurons were identified by fluorescence of either mCherry ( via rAAV-Synapsin::FLEX-rev-PLE-2a-mCherry virus ) or EGFP ( via rAAV-CAG::FLEX-rev-EGFP virus ) and hyperpolarization-induced currents ( Ih ) , and then visually targeted with infrared gradient contrast optics Whole-cell recordings were made on DA neurons using a Multiclamp 700B amplifer ( Molecular Devices , Sunnyvale , CA , United States ) . Neurons were voltage-clamped at −70 mV . mEPSCs were recorded from DA neurons in the presence of picrotoxin ( 50 µM ) and tetrodotoxin ( TTX , 1 µM ) . Responses were digitized at 10 kHz . Series resistances were less than 30 MOhm . The intracellular solution for voltage-clamp recordings contained the following ( in mM ) : 125 CsCl , 5 NaCl , 10 HEPES , 0 . 6 EGTA , 4 Mg-ATP , 0 . 3 Na2GTP , and 10 lidocaine N-ethyl bromide ( QX-314 ) , pH 7 . 35 and 290 mOsm . For detection of mEPSCs , we used template matching ( Clampfit , Molecular Devices ) followed by visual inspection . Statistical tests were performed in Microsoft Excel or SigmaPlot . Data is represented as mean ± s . e . m . , and error bars indicate s . e . m . Mice expressing Cre recombinase in DA neurons ( Slc6a3Cre mice ) ( P21-P24 ) were anaesthetized with isoflurane and placed into a stereotaxic apparatus ( David Kopf Instruments , Tujunga , CA , United States ) . The skull was exposed via a small incision and two small holes were drilled on either side of the midline for viral injection . A pulled-glass pipette with 20–40 µm tip diameter was inserted into the brain and two injections ( 60 nl ) of the rAAV2/1-Synapsin::FLEX-rev-PLE-2a-mCherry or rAAV-CAG::FLEX-rev-EGFP virus were made at coordinates around the VTA ( coordinates , bregma: −2 . 5 mm; midline: ±0 . 5 mm; skull surface: −4 . 5 mm and −5 . 0 mm ) . Mice were returned to their home cage typically for 9 or 10 days to recover and for expression of PLE and mCherry or EGFP . A micromanipulator ( Narishige ) was used to control injection speed at 30 nl/min , and the pipette was withdrawn 15 min after the final injection . Preparation of chloromethyl 1-methylcyclopropanecarboxylate ( 1 ) ; 1-methyl cyclopropane-1-carboxylic acid ( 4 . 0 g , 40 . 0 mmol ) was added to mixture of potassium carbonate ( 22 . 1 g , 160 mmol ) , tetra butyl ammonium hydrogen sulfate ( 1 . 36 g , 4 . 0 mmol ) , water ( 50 ml ) , and dichloromethane ( DCM ) ( 40 ml ) at room temperature . After stirred for 10 min , solution of chloromethyl chlorosulfate in DCM ( 40 ml ) was slowly added and stirred for 5 hr . The reaction mixture was diluted with 150 ml of water and extracted with DCM ( 3 × 100 ml ) , combined organic layer was washed with saturated brine ( 100 ml ) , dried over MgSO4 , filtered , and concentrated to an oil . The oil was filtered through a pad of silica gel with 150 ml of DCM and concentrated to give crude chloromethyl 1-methylcyclopropanecarboxylate ( 3 . 8 g , 64% ) . 1H NMR , δH ( CDCl3 , 400 MHz ) 5 . 70 ( 2H , s ) , 1 . 33 ( 3H , s ) , 1 . 33 ( 2H , dd ) , 0 . 79 ( 2H , dd , J = 5 . 46 , ) ; m/z ( ES+ ) found: [M + H]+ , 149 . Preparation of iodomethyl 1-methylcyclopropanecarboxylate ( 2 ) ; sodium iodide ( 11 . 5 g , 77 . 0 mmol ) was added to solution of chloromethyl 1-methylcyclopropanecarboxylate ( 3 . 8 g , 25 . 7 mmol ) in acetone ( 25 ml ) and stirred for 3 hr at 45°C . Reaction mixture was filtered , washed with acetone , and concentrated under vacuum . The residual oil was dissolved in diethyl ether ( 100 ml ) , washed with aqueous sodium bicarbonate and aqueous sodium thiosulfate , dried over MgSO4 , filtered , and concentrated to give crude iodomethyl 1-methylcyclopropanecarboxylate ( 4 . 6 g , 75% ) 1H NMR , δH ( CDCl3 , 400 MHz ) 5 . 92 ( 2H , s ) , 1 . 31 ( 3H , s ) , 1 . 29 ( 2H , dd , J = 7 . 10 ) , 0 . 74 ( 2H , dd , J = 6 . 88 , ) ; m/z ( ES+ ) found: [M + H]+ , 240 . Preparation of ( 4-formyl-2-nitrophenoxy ) methyl 1-methylcyclopropanecarboxylate ( 3 ) ; added solution of iodomethyl 1-methylcyclopropanecarboxylate ( 4 . 6 g , 19 . 2 mmol ) in acetonitrile ( 50 ml ) to silver oxide ( 13 . 7 g , 57 . 5 mmol ) in acetonitrile ( 100 ml . ) After stirring for 10 min at 0°C , solution of 4-hydroxy-3-nitro-benzaldehyde ( 4 . 0 g , 24 . 0 mmol ) in acetonitrile ( 100 ml ) was added dropwise and stirred over night at 0°C . The reaction mixture was filtered through Celite then concentrated under vacuum to give crude ( 4-formyl-2-nitrophenoxy ) methyl 1-methylcyclopropanecarboxylate ( 4 . 9 g , 91% ) 1H NMR , δH ( CDCl3 , 400 MHz ) 9 . 98 ( 1H , s ) , 8 . 35 ( 1H , d , J = 2 . 04 ) , 8 . 10 ( 1H , dd , J = 8 . 68 ) , 7 . 42 ( 1H , d , J = 8 . 68 ) , 5 . 92 ( 2H , s ) , 1 . 29 ( 3H , s ) , 1 . 30 ( 2H , m ) , 0 . 79 ( 2H , t , J = 3 . 76 , ) ; m/z ( ES+ ) found: [M + H]+ , 280 . Preparation of ( 4- ( hydroxymethyl ) -2-nitrophenoxy ) methyl 1-methylcyclopropanecarboxylate ( 4 ) ; NaBH4 was added to a stirred solution of ( 4-formyl-2-nitrophenoxy ) methyl 1-methylcyclopropanecarboxylate ( 4 . 9 g , 17 . 5 mmol ) in chloroform ( 60 ml ) and isopropanol ( 30 ml ) at 0°C . Mixture was stirred at 0°C for an hour then at room temperature overnight . The reaction mixture was diluted with 150 ml of water and extracted with DCM ( 3 × 100 ml ) . The combined organic layer was washed with saturated brine ( 100 ml ) , dried over MgSO4 , filtered , and concentrated to residue oil . Purified by column chromatography ( 50% Ethyl acetate: hexane ) to afford ( 4- ( hydroxymethyl ) -2-nitrophenoxy ) methyl 1-methylcyclopropanecarboxylate ( 2 . 35 g , 48% ) 1H NMR , δH ( CDCl3 , 400 MHz ) 7 . 85 ( 1H , d , J = 2 . 12 ) , 7 . 55 ( 1H , dd , J = 7 . 92 ) , 7 . 42 ( 1H , d , J = 7 . 00 ) , 5 . 82 ( 1H , s ) , 4 . 73 ( 2H , s ) , 1 . 31 ( 3H , s ) , 1 . 27 ( 2H , dd , J = 6 . 86 ) , 0 . 76 ( 2H , dd , J = 6 . 9 , ) ; m/z ( ES+ ) found: [M + H]+ , 282 . Preparation of ( 2-nitro-4- ( ( ( ( 4-nitrophenoxy ) carbonyl ) oxy ) methyl ) phenoxy ) methyl 1-methylcyclopropanecarboxylate ( 5 ) ; A solution of 4-nitrophenyl chloroformate ( 3 . 4 g , 16 . 7 mmol ) in THF ( 10 ml ) was added to a mixture of ( 4- ( hydroxymethyl ) -2-nitrophenoxy ) methyl 1-methylcyclopropanecarboxylate ( 2 . 35 g , 8 . 35 mmol ) and triethylamine ( 4 . 7 ml , 33 . 4 mmol ) in THF ( 40 ml ) and was stirred at 0°C . The reaction mixture was stirred in the dark at room temperature overnight . The reaction mixture was diluted with 150 ml of water and extracted with ethyl acetate ( 3 × 100 ml ) , the combined organic layer was washed with saturated brine ( 100 ml ) , dried over MgSO4 , filtered , and concentrated to residue oil . Purified by column chromatography ( 50% ethyl acetate: hexane ) to afford ( 2-nitro-4- ( ( ( ( 4-nitrophenoxy ) carbonyl ) oxy ) methyl ) phenoxy ) methyl 1-methylcyclopropanecarboxylate ( 4 . 8 g , 65% ) 1H NMR , δH ( CDCl3 , 400 MHz ) 8 . 26 ( 2H , d , J = 9 . 24 ) , 7 . 92 ( 1H , d , J = 2 . 20 ) , 7 . 62 ( 1H , dd , J = 8 . 6 ) , 7 . 36 ( 2H , d , J = 9 . 24 ) , 7 . 28 ( 1H , d , J = 8 . 6 ) 5 . 82 ( 2H , s ) , 5 . 26 ( 2H , s ) , 1 . 29 ( 3H , s ) , 1 . 26 ( 2H , dd , J = 6 . 92 ) , 0 . 75 ( 2H , dd , J = 6 . 74 ) ; m/z ( ES+ ) found: [M + H]+ , 447 . Preparation of CM-MK-801 ( 6 ) ; added and N , N-diisopropylethylamine ( 13 μl , 72 nmol ) to mixture of ( 2-nitro-4- ( ( ( ( 4-nitrophenoxy ) carbonyl ) oxy ) methyl ) phenoxy ) methyl 1-methylcyclopropanecarboxylate ( 10 mg , 22 nmol ) and ( + ) -MK 801 hydrogen maleate ( 8 mg , 24 nmol ) in DMF ( 0 . 5 ml ) at room temperature and stirred overnight . The reaction mixture was diluted with 100 ml of water and extracted with ethyl acetate ( 3 × 100 ml ) , combined organic layer was washed with saturated brine ( 100 ml ) , dried over MgSO4 , filtered , and concentrated . Purified by prep-HPLC ( water: acetonitrile ) to afford CM-MK-801 ( 4 mg , 32% ) 1H NMR , δH ( CDCl3 , 400 MHz , temperature: 330K ) 7 . 52 ( 1H , br s ) , 7 . 20 ( 1H , br d , J = 7 . 44 ) , 7 . 15 ( 2H , m ) , 7 . 02 ( 3H , m ) , 6 . 94 ( 2H , m ) , 6 . 90 ( 2H , m ) , 6 . 73 ( 1H , m ) , 5 . 63 ( 2H , s ) , 5 . 30 ( 1H , d , J = 5 . 52 ) , 4 . 96 ( 2H , dd , J = 33 . 8 ) , 3 . 46 ( 1H , d , J = 16 . 86 ) , 2 . 13 ( 3H , s ) , 1 . 17 ( 3H , s ) , 1 . 27 ( 2H , dd , J = 6 . 84 ) , 0 . 59 ( 2H , dd , J = 6 . 7 , ) ; m/z ( ES+ ) found: [M + H]+ , 529 .

Learning is critical to survival for humans and other animals . The learning process is regulated by receptors on the surface of brain cells called N-Methyl-D-aspartate receptors ( or NMDA receptors for short ) . These receptors help to strengthen signals between brain cells , which allows a new concept or action to be learned . However , it has been difficult to pin down how the role of NMDA receptors selectively in specific types of brain cells . While drugs can be used to quickly block NMDA receptors throughout the brain , it is hard to target drugs to a specific cell type . Also , genetic engineering can be used to selectively knock out NMDA receptors in certain types of brain cells , but these techniques are too slow , and can take weeks or even a lifetime to work . Now , Yang et al . have developed a clever way to combine an NMDA-blocking drug and genetic engineering to study NMDA receptors' responses to cocaine in specific brain cells . This approach involved first creating an inactive form of an NMDA-blocking drug that can only becomes active when it is processed by an enzyme that is normally produced in pigs' livers . Next , living mouse brain cells , including some that were engineered to express the pig enzyme , were exposed to the drug in the laboratory . The drug blocked the NMDA receptors on brain cells that expressed the enzyme , but not the receptors on nearby brain cells that lacked the enzyme . This occurred even though all the cells produced NMDA receptors and all were exposed to the drug . NMDA receptors have been known to play an important role in cocaine addiction for more than 20 years . Drugs like cocaine can co-opt the normally healthy learning process involving NMDA receptors and lead to a maladaptive form of learning that is commonly called addiction . Cocaine strengthens signals between brain cells causing the behaviors associated with using cocaine to become deeply ingrained and difficult to change . Yang et al . used cell type-specific targeting of a drug that blocks NMDA receptors to observe what happened in cocaine-exposed brain cells with , or without , working NMDA receptors . As expected , the experiments showed that cocaine didn't strengthen brain signals in cells without working NMDA receptors . Specifically , the experiments showed that NMDA receptors on a type of brain cell that release a pleasure-inducing chemical called dopamine are necessary for cocaine–induced synaptic plasticity . The combination technique developed by Yang et al . will likely be used by other scientists to further study the role of NMDA receptors in specific brain cells during addiction and normal brain activity .

2015

Cell type-specific pharmacology of NMDA receptors using masked MK801

Cardiac pumping depends on the morphological structure of the heart , but also on its subcellular ( ultrastructural ) architecture , which enables cardiac contraction . In cases of congenital heart defects , localized ultrastructural disruptions that increase the risk of heart failure are only starting to be discovered . This is in part due to a lack of technologies that can image the three-dimensional ( 3D ) heart structure , to assess malformations; and its ultrastructure , to assess organelle disruptions . We present here a multiscale , correlative imaging procedure that achieves high-resolution images of the whole heart , using 3D micro-computed tomography ( micro-CT ) ; and its ultrastructure , using 3D scanning electron microscopy ( SEM ) . In a small animal model ( chicken embryo ) , we achieved uniform fixation and staining of the whole heart , without losing ultrastructural preservation on the same sample , enabling correlative multiscale imaging . Our approach enables multiscale studies in models of congenital heart disease and beyond . Congenital heart disease ( CHD ) , which manifests as a morphologically defective heart , affects about 1% of newborn babies , and remains the primary cause of non-infectious children mortality in the developed world ( Gilboa et al . , 2010; van der Linde et al . , 2011 ) . While CHD mortality rates have been dramatically reduced in recent years thanks to advances in surgical practice and interventional technologies ( Gilboa et al . , 2010; Czosek and Anderson , 2016 ) , CHD patients continue to be at an increased risk of developing heart failure at a much younger age than the general population ( Gilljam et al . , 2019 ) . Despite early indicators of success , heart failure continues to take the lives of young children with CHD: 10% to 25% of newborns with a critical heart defect do not survive the first year , and 44% do not survive to 18 years of age ( Oster et al . , 2013; Stout et al . , 2016 ) . This unfortunate trend points to cardiac deficiencies in CHD that are not yet understood ( Sugimoto et al . , 2015 ) . Imaging approaches have been employed to study cardiac function in normal and CHD hearts ( Prakash et al . , 2010 ) . In humans , ultrasound-based echocardiography , which can image the in vivo motion of cardiac walls and measure blood flow velocities within the heart , is used to diagnose fetal CHD in utero and assess CHD severity ( Arya et al . , 2014; Jatavan et al . , 2016 ) . After the baby is born , echocardiography is useful to monitor cardiac function before and after interventions to repair CHD . Magnetic resonance imaging ( MRI ) is also employed to precisely diagnose malformations and monitor cardiac function in CHD ( Sreedhar et al . , 2005; Ntsinjana et al . , 2011 ) . In heart development animal research , moreover , optical coherence tomography ( OCT ) and echocardiography are used with avian and mammalian models of CHD . OCT , like ultrasound , is a non-invasive technique that can image heart motion and measure blood flow velocities within the heart . OCT resolution ( <5 µm ) is ideal for early avian and mouse embryos during tubular-heart developmental stages ( Larina et al . , 2009; Syed et al . , 2011; Larina et al . , 2011; Peterson et al . , 2017; Jenkins et al . , 2007 ) . For later stages of heart development echocardiography is used in mice and chick ( Grune et al . , 2018; Damen et al . , 2017; Midgett et al . , 2017a ) . In small animal research , moreover , due to the small size of developing hearts , functional imaging techniques are frequently complemented with micro-CT or histology to more accurately phenotype the heart ( Midgett et al . , 2017a; Butcher et al . , 2007; Gregg and Butcher , 2012 ) . The structural ( morphological or ‘geometrical’ ) characteristics of heart malformations , including changes in cardiac wall architecture , have been extensively studied . However , the mechanisms that lead to an anomalous cardiac architecture in CHD and the clinical consequences of it are unknown . Recent studies have revealed an abnormal orientation of myocardial cells ( the heart muscle cells ) within CHD hearts ( Garcia-Canadilla et al . , 2019; Stephenson et al . , 2018; Garcia-Canadilla et al . , 2018 ) . Myocardial cells are elongated , cylindrical-like cells , that contract along their long axis . In a normal heart , myocardial cells arrange in sheet-like layers with their long axes in parallel to each other , forming a helical pattern ( Gilbert et al . , 2007; Omann et al . , 2019 ) . This highly organized pattern ensures that cardiac contraction occurs along specific directions , so that the heart can effectively eject blood into the pulmonary and systemic circulations . Perturbations of the normal myocardial architecture affect cardiac contractility and compliance . This is both because the tissue can no longer contract on the very specific directions and patterns that optimize cardiac function , but also because cardiac contraction force may be diminished . Therefore , changes in myocardial architecture ( with respect to normal ) frequently lead to heart tissues with impaired contractility and reduced compliance . Newly developed contrast episcopic microscopy and synchro micro-CT imaging techniques , enable non-destructive analyses of banked human fetal and neonatal hearts with CHD ( Stephenson et al . , 2018; Garcia-Canadilla et al . , 2018 ) . These emergent studies are revealing myocardial disarray in CHD ( with respect to their normal counterparts ) that very likely affect cardiac function before and after surgical repair . In addition to changes in the myocardial organization , the subcellular contraction machinery of myocardial cells ( e . g . the myofibrils that contract the cell; and the mitochondria that provide energy for contraction ) may also be compromised in CHD , affecting heart function . For instance , a reduced number of myofibrils per myocardial cell will reduce the contractile force that the cell can generate , leading to cardiac dysfunction . The extent to which the cells of malformed hearts exhibit deficiencies is unknown ( Garcia-Canadilla et al . , 2019; Garcia-Canadilla et al . , 2018; Sanchez-Quintana et al . , 1999 ) . This is in part due to limitations of existing technologies that have not achieved precise multiscale mapping to decipher the association between structural and cellular deficiencies in the heart and beyond . We describe here a proof-of-concept correlative multiscale procedure that combines imaging of whole heart morphology and its subcellular organization ( ultrastructural organelle architecture ) . Our multiscale procedure uses micro computed tomography ( micro-CT ) imaging to capture heart morphology at micrometer resolution , and scanning electron microscopy ( SEM ) to capture cardiac tissue ultrastructure at nanometer resolution . Current SEM technologies allow for three-dimensional ( 3D ) imaging , enabling reconstruction and quantification of ultrastructural features within a tissue volume ( Rennie et al . , 2014; Midgett et al . , 2017b; Hussain et al . , 2018 ) . Among 3D SEM methods , we have selected serial block-face SEM ( SBF-SEM ) for ultrastructural imaging , as it allows 3D imaging of relatively large volumes ( sample size 40 × 60×40 µm3 ) . The methodology we present herein improves upon previous protocols by achieving uniform staining of a relatively large heart sample ( 4-5 mm wide ) , circumventing micro-CT X-ray penetration issues , and allowing sample screening and selection prior to full SEM sample preparation . Our multiscale imaging , further , enables mapping of structural and ultrastructural heart features . As proof of concept , we applied our developed multiscale imaging procedure to two embryonic chick hearts . These hearts were collected at stages corresponding to about 5–6 months of human fetal development , when the heart is already formed but maturing in preparation for birth/hatching . We imaged: ( 1 ) a control heart with no structural defects and ( 2 ) a heart with tetralogy of Fallot ( TOF ) , a combination of structural heart malformations found in humans ( Bailliard and Anderson , 2009; Wiputra et al . , 2018 ) . Our results show the morphology and ultrastructure of these two hearts , and emphasize the need for a multiscale approach to deepen our understanding of CHD and enable the development of effective strategies to combat heart failure in CHD . To achieve multiscale imaging , we followed a four-step protocol ( see Figure 1; details in Materials and methods ) . Briefly , in Step 1 the heart was excised , homogeneously fixed and stained for micro-CT . Initial staining followed a modified ferrocyanide-reduced osmium–thiocarbohydrazide–osmium ( ROTO ) protocol ( Hua et al . , 2015; Malick and Wilson , 1975; Willingham and Rutherford , 1984 ) typically used for electron microscopy ( EM ) sample preparation . Three-dimensional micro-CT images ( 10 μm resolution ) confirmed uniform ROTO staining of the whole heart and provided morphological cardiac details . At this time , the heart samples were stored until further processing , enabling selection of specific samples for full processing based on micro-CT scans . Step 2 finalized the preparation of the whole heart for SBF-SEM by post-staining the hearts with uranyl acetate and lead aspartate and then embedding them in a resin block . Uniform staining was confirmed on a semithin ( 250 nm ) section of the block , which also determined ultrastructural quality and enabled registration to micro-CT images . In Step 3 , a slab of the sample was cut and sectioned around a specific region of interest ( ROI , ~500×500 µm2 ) from which sub-ROIs for 3D SBF-SEM imaging ( ~40×60 μm2 ) were selected . In Step 4 , 3D SBF-SEM datasets were acquired ( 10 nm lateral resolution and 40 nm depth resolution ) . We obtained 3D micro-CT images of the whole heart , featuring both external and internal structures at 20 µm resolution ( Figure 2; Step 1 in Figure 1 ) . Despite the relatively large dimensions of the heart ( 5-6 mm long ) , tissue contrast was uniform across the heart walls and septae and allowed us to visualize the microstructural details of the heart chambers , valves , and great arteries . Micro-CT images guided selection of two hearts for subsequent SBF-SEM imaging and analysis . We selected: ( 1 ) a normal heart and ( 2 ) a heart exhibiting TOF malformation . In a normal heart , blood in the left ventricle ( LV ) and right ventricle ( RV ) is separated by an interventricular septum; the pulmonary valve and pulmonary artery connect to the RV , which pumps blood to the lungs; and the aortic valve and aorta connect to the LV , which pumps blood to the body . TOF is characterized by a combination of four defects: ( i ) ventricular septal defect , which is a hole in the interventricular septum; ( ii ) overriding aorta , a change in the position of the aorta such that it sits in the middle of the two ventricles , on top of the ventricular septal defect; ( iii ) pulmonary stenosis or atresia , a narrowing or closure of the pulmonary artery or pulmonary valve; and ( iv ) RV hypertrophy , a thickening and enlargement of the RV wall . RV hypertrophy in TOF , however , develops over time as the stenosis of the pulmonary artery increases pressure in the RV after birth ( Bailliard and Anderson , 2009 ) and was not present in the heart examined in this study ( see Figure 3 for a comparison of the selected normal and TOF hearts ) . The TOF heart analyzed here featured supravalvular pulmonary stenosis , a ventricular septal defect , and an overriding aorta . The right ventricle was enlarged compared to the control heart ( Figure 3 ) . Further , the TOF heart was missing the left branch of the pulmonary artery . In humans , this rare condition , called unilateral absence of a pulmonary artery , is known to occur in conjunction with TOF or cardiac septal defects ( Reading and Oza , 2012 ) . From micro-CT images of the two hearts , we quantified cardiac structural features for illustration purposes . It is important to mention , however , that these quantifications only pertain to the two hearts compared in this study , and thus are not necessarily representative of normal nor TOF heart populations . For the two hearts compared , we found that the LV volume was about the same in the normal , control ( CON ) heart and TOF hearts . However , in the CON heart the RV volume was 40% smaller than the LV volume ( RV volume/LV volume ~0 . 6 ) , while in the TOF heart RV volume was 30% larger than the LV volume ( RV volume/LV volume ~1 . 3 ) . Overall the volume of the TOF RV was about twice as big as the CON RV , suggesting functional impairment of the TOF RV . We chose to characterize the ultrastructural architecture of the selected hearts at approximately the transverse section at which the heart width ( from LV to RV wall ) is maximal , also referred to as the equatorial plane . Images of semithin cross-sections for each heart ( Figure 4 , top row; corresponding to Step 2 in Figure 1 ) show that staining was uniform , indicating successful stain penetration through the heart tissues . Please note that the transverse section of the TOF heart was below the ventricular septal defect and thus exhibits a continuous septum . SEM images of semithin sections guided selection of ROIs from each heart . For this study , we selected two ROIs within each heart LV , denoted by ROI A and ROI B . After cutting the sample , the ROIs were first imaged with SBF-SEM at low resolution ( 65–80 nm lateral resolution; Step 3 in Figure 1 ) , from which sub-ROIs were further selected ( see Figure 4 ) . These sub-ROIs ( ~40×60 μm2 ) were imaged at 10 nm lateral resolution , and we acquired 800–1000 images in depth ( with ~40 nm of depth distance , thus 32 to 40 μm in depth ) . These high-resolution images showed the conservation of ultrastructural features ( see Figure 5 ) . Images exhibited continuous nuclear membranes , intact mitochondria , and defined myofibrils , indicating that we achieved both appropriate and uniform fixation and staining of the hearts with our protocol . SBF-SEM image stacks provided 3D volumetric reconstructions of sub-ROIs . While volumetric image resolution was not isotropic ( 10 nm lateral resolution versus 40 nm depth resolution ) , ultrastructural features could be visualized from any angle of view within the reconstructed images ( see Figure 6 ) . Thus SBF-SEM images allowed us to visualize the orientation and organization of nuclei , myofibrils , and mitochondria ( among other features ) in heart tissue samples . To more easily visualize and quantify cardiac ultrastructure , we segmented ( delineated ) from SBF-SEM images the cell nuclei , myofibrils , mitochondria , and the extracellular space . To this end , we used a combination of deep learning algorithms and tools available on the Dragonfly 4 . 1 software ( Object Research Systems , Quebec , Canada ) . When independently tested against carefully annotated images ( two each from the CON and TOF hearts , region A ) , the segmentation accuracy from the deep learning algorithm was at least 90% for myofibrils , 94% for mitochondria , and 98% for nuclei . Additional manual segmentation ‘clean up’ was thus required to improve the accuracy of organelle depictions . These segmentations , however , reveal the detailed 3D ultrastructural architecture of the heart wall ( e . g . see Figure 7 ) . Ultrastructural quantifications of the two hearts were performed for illustration purposes . To control the accuracy of segmentations for quantification , we used subsets of the full SBF-SEM datasets by selecting images and regions within images from the complete dataset . From these selected image regions , we quantified the percentage of the cell occupied by nuclei , myofibrils , and mitochondria . That is , we quantified selected organelle density within cells . We found that , for our two hearts , the density of nuclei , myofibrils and mitochondria was similar between the CON and TOF hearts ( see Figure 8A ) . We also quantified the percentage of the image regions occupied by extracellular space . The TOF heart exhibited more extracellular space than the CON heart ( see Figure 8B ) , which was also consistent with a visual inspection of segmented extracellular space images ( see Figure 8C ) . Visualization of segmentations of the entire sub-ROIs from region A of the CON and TOF hearts revealed a slightly different orientation of myocardial cells between samples ( Figure 9 ) . Quantification of myofibril elliptical and transmural angle orientations ( see Figure 10A ) revealed that in the CON heart the myofibrils in the left ventricle were oriented at around 45° in the radial direction with respect to the circumferential directions ( θ angle or transmural angle ) with a standard deviation of approximately 32° . The myofibril elliptical angle , defined here as the angle between the circumferential and longitudinal directions ( Φ angle ) was around 70° , and exhibited a wide dispersion , with a standard deviation of 36° ( Figure 10B ) . In contrast , the myofibrils in the TOF heart were oriented almost circumferentially in the transmural plane , with an average transmural θ angle of 7° ( 20° standard deviation ) ; while the elliptical angle Φ was around 15° with a standard deviation of 15° . Thus in the TOF heart , at the sample location , myofibrils run closer to the circumferential direction than in the CON heart ( see Figure 10C ) . It is important to notice that these quantifications and differences pertain only to the two hearts under consideration ( and samples within the heart ) and do not reflect population trends . A more detailed study including more animals is needed to infer population differences . Researchers have used EM techniques , including SEM , for decades to visualize the organization of organelles within cells ( Malick and Wilson , 1975; Karnovsky , 1965 ) . In the heart , studies using EM have revealed the ultrastructural architecture of mature myocardial cells ( Hussain et al . , 2018; Pinali and Kitmitto , 2014 ) , and elucidated the maturation of myocardial ultrastructure during embryonic development ( Fischman , 1967; Wainrach and Sotelo , 1961; Manasek , 1970 ) . A few studies , moreover , have determined changes in ultrastructure due to pathophysiological conditions in the mature heart ( Sanchez-Quintana et al . , 1999; Pinali et al . , 2015; Holzem et al . , 2016 ) . Properly preparing samples for EM requires meticulous protocols that aim at preserving the ultrastructure of the tissue under study ( e . g . intact cell and nucleus membranes , mitochondria and their crystae , myofibrils and z-disks ) . Because the ultrastructural features analyzed are at the nanometer scale , samples used for EM are typically very small ( <1 mm3 ) , which facilitates proper sample preparation . In preparing samples , portions of the heart are typically excised , carefully prepared for imaging ( fixed and stained ) ( Rennie et al . , 2014; Mukherjee , 2016 ) , and then imaged with an EM modality ( Rennie et al . , 2014; Pinali and Kitmitto , 2014 ) . Using this procedure , however , finding the ultrastructure associated with a specific microscopic feature or malformation site can be daunting . Moreover , in micro-dissecting tissue samples , the myocardial organization within the heart may be lost ( Gilbert et al . , 2007 ) . Our methodology enables correlative microscopy in a way that allows precise identification and mapping of portions of the heart to their ultrastructure . Whole animals and organs have been scanned with computed tomography ( CT ) , a 3D X-ray imaging modality , to determine the internal and external structure of organs , including the heart . For small animal models , micro-CT , a high-resolution CT , is typically employed ( Midgett et al . , 2017a; Butcher et al . , 2007 ) . To enhance contrast and thus resolution , prior to micro-CT imaging excised tissue samples are stained ( Midgett et al . , 2017a ) . Preparing samples for micro-CT imaging requires good and uniform penetration of the stain , intended to preserve and contrast the tissue microstructure ( e . g . heart morphology , heart chambers and valves ) . Micro-CT can then reveal subtle and overt malformations in the heart and its microstructure ( Midgett et al . , 2017a ) . For example , using micro-CT cardiac images it is possible to visualize ventricular septal defects or translocations of the great arteries , but also wall and septum thickness , and differentiate trabecular from compact myocardium ( Midgett et al . , 2017a ) . The discrepancy in the scale at which we acquire micro-CT and SBF-SEM images introduces fundamental differences in the requirements for sample preparation . Due to diverse fixation and staining protocols ( Midgett et al . , 2017a; Hopkins et al . , 2015 ) , tissues prepared for micro-CT or other microstructural imaging ( e . g . histology ) cannot typically be simultaneously processed for SBF-SEM ( and other EM modalities ) , thereby restricting the ability to analyze both the microstructural and ultrastructural characteristics within the same tissue sample . Our correlative micro-CT/SBF-SEM procedure can reveal the sub-cellular architecture associated with specific pathological or malformed regions of the heart found from micro-CT images . Applications combining micro-CT and EM technologies have recently begun to emerge , for example ( Karreman , 2017; Morales et al . , 2016; Sengle et al . , 2013 ) . However , several challenges remain in applying these methods to correlative , multiscale imaging of a relatively large organ like the heart ( even the heart of a small animal ) . To achieve both EM and micro-CT high-quality imaging of the same sample , existing protocols have capitalized on heavy metal contrast in small tissue samples , which are fully processed prior to EM and micro-CT imaging , for example ( Genoud et al . , 2018; Karreman , 2017 ) . However , achieving the uniform staining necessary for optimal imaging with both micro-CT and SEM becomes progressively challenging with increasing tissue sample size . This is mainly due to difficulties in achieving uniform and fast fixation ( that preserves the ultrastructure ) , and uniform stain penetration ( both for post-fixation purposes and to enhance contrast for SEM and micro-CT imaging ) . Modifications to the classic ROTO protocols ( Hua et al . , 2015; Malick and Wilson , 1975; Willingham and Rutherford , 1984 ) for EM tissue preparation have been quite successful in achieving strong and uniform staining of relatively large samples ( Deerinck et al . , 2010; Tapia et al . , 2012 ) . However , acceptable staining was typically only up to a depth of 200 μm , and more recently 500 μm ( Hua et al . , 2015 ) in dense brain tissues . In an attempt to stain whole brains for EM reconstruction of synapses , Mikula et al . developed the brain-wide reduced-osmium staining with pyrogallol-mediated amplification ( BROPA ) protocol ( Mikula and Denk , 2015 ) for 3D SBF-SEM ( no other heavy metals were used ) . While the protocol is compatible with both micro-CT and 3D SBF-SEM imaging , preparing a whole mouse brain ( about 8–10 mm in diameter ) using BROPA required 2–3 months . A fast BROPA protocol ( fBROPA ) was later developed and used to prepare whole brains from zebrafish in about 4 days ( Genoud et al . , 2018 ) . Zebrafish brains , however , are significantly smaller than mouse brains ( diameter of 1 . 1 mm vs 8–10 mm , respectively ) . For our hearts , we needed to achieve stain penetration of a relatively large sample ( 4–5 mm wide ) and a fast preparation protocol was also desired . We found that preparing the heart for SBF-SEM and stopping the protocol after initial ROTO staining ( 1 day processing ) , was compatible with micro-CT and later SBF-SEM full sample processing and imaging . Further , full sample preparation compatible with SBF-SEM required and additional 3 days , making the entire heart processing around 4 days . To our knowledge , this is the first time that correlative micro-CT/SBF-SEM imaging is applied to the heart . Sample preparation for SBF-SEM required strong , immediate fixation to preserve the ultrastructure of the heart tissue . We used a modified Karnovsky’s fixative with equal parts glutaraldehyde and paraformaldehyde . The paraformaldehyde rapidly penetrated and temporarily stabilized the tissue , and the slower-penetrating glutaraldehyde , a superior cross-linker , more permanently preserved the tissue sample ( Karnovsky , 1965 ) . An obvious difficulty was to achieve uniform fixation of the whole heart sample . Homogenous fixation was achieved by perfusing fixative into the heart prior to excision and then promptly immersing the heart in fixative after excision , allowing the fixative to simultaneously penetrate the heart through the tissue’s internal and external surfaces . The hearts were then post-fixed with osmium tetroxide , a lipid cross-linker , which fully stabilized membrane structures while enhancing contrast for micro-CT and SBF-SEM imaging . For our heart samples , we could achieve uniform stain penetration using a variation of the ROTO protocol with extended staining timing ( about 30% increase; see Materials and methods ) . The extended timing was sufficient for our hearts , even considering small size variations . We expect , however , that further time increases would apply to larger heart samples ( for instance if we image embryos at a more advanced developmental stage , or other species are considered ) . For the chick embryos studied here , sample preparation after ROTO was adequate for micro-CT imaging of the whole heart ( see Figure 2 ) . For 3D SBF-SEM images , heart samples were further stained with a combination of heavy metals . This is because SEM images are acquired via the detection of secondary and backscattered electrons that are emitted as the tissue is scanned with a high energy beam of primary electrons ( Cazaux , 2012 ) . Soft biological tissues , like the heart muscle , yield few backscattered electrons and need to be stained with heavy metals , which readily produce secondary and backscattered electrons . Heavy metal stains interact with specific ultrastructural components , and therefore combinations of stains are frequently used in a single sample . The application of osmium tetroxide , used in ROTO protocols , served as the first application of a heavy metal stain in the SBF-SEM preparation . The osmium tetroxide , which interacts with lipids in membranes and vesicles , both post-fixed and stained tissues . Further staining with uranyl acetate stained lipids and proteins , and lead aspartate stained proteins and glycogens . Together , these heavy metal stains fully preserved and contrasted ultrastructural details , as evidenced by high-resolution SBF-SEM images ( see Figure 5 ) . For whole-heart samples , we found that staining with uranyl acetate and lead aspartate rendered resin-embedded hearts opaque to micro-CT imaging ( data not shown ) . Our multiscale multimodality imaging procedure overcame these difficulties by performing micro-CT imaging after ROTO ( see above ) but prior to the lead and uranium staining steps , such that correlative 3D micro-CT and 3D SBF-SEM could be implemented . This initial modified ROTO post-fixing and staining provided tissue contrast for micro-CT while also ensuring that the sample was preserved and stabilized before final SBF-SEM sample preparation and imaging ( see Figure 1 ) . In addition , ROTO post-fixing enabled screening , storing and selection of hearts before final SBF-SEM processing . This feature of our multiscale approach is advantageous for several reasons . At this early processing point , hearts could be stored for relatively long periods ( at least 2 weeks , but potentially months/years ) before further processing . This allows researchers to prepare and screen by micro-CT a large number of hearts , and then select only those hearts of interest ( e . g . with a specific malformation ) for further analysis with SBF-SEM . Not only does this approach permit banking of samples , but it also saves considerable time and resources by avoiding full sample preparation of hearts that are not useful . This is important for our application , in which at most 60% of treated hearts develop structural malformations , and the nature and severity of defects vary among individual hearts . Since we cannot accurately classify malformations until they are scanned with micro-CT , being able to use the micro-CT as both a screening tool and as a navigational tool for later correlative microscopy is invaluable to CHD studies . Due to the size of the hearts and slow diffusion ( penetration ) of heavy metals into the tissue , achieving uniform staining during further SBF-SEM tissue processing was not straightforward . Initial iterations of the procedure ( data not shown ) resulted in a strong gradient of staining into the heart tissue , a manifestation of poor stain penetration ( Genoud et al . , 2018; Mikula and Denk , 2015 ) . Application of microwave steps to enhance diffusion was not helpful , although perhaps optimizations of those steps could achieve improved results . We found , however , that following a Renovo Neural , Inc protocol ( see Materials and methods ) after ROTO and before embedding the sample in a resin capsule allowed the samples to achieve homogeneous staining . Uniform stain penetration throughout the heart , was apparent from semithin transverse sectional images of the heart and low-resolution images of ROIs ( see Figure 4 ) . The uniform contrast and resolution of ultrastructural details provided further evidence of uniform staining and proper tissue fixation ( see Figure 5 ) . While the focus of this study was to demonstrate homogeneous staining and fixation , our procedure enables correlative microscopy . One way to achieve accurate localization of ultrastructures within the heart structure , is to register semithin transverse sections to micro-CT images , and then SBF-SEM images to the semithin images ( as done in Figure 4 ) . Because sectioning of samples for SBF-SEM imaging is done after micro-CT images are acquired , sectioning is guided by the images of the whole heart , facilitating the selection of regions of interest . Registration can then be performed among the images themselves . This could be done directly , or by adding fiduciary markers in the resin/heart to facilitate image alignment . Our procedure allows imaging of ultrastructure at several regions of interest within the heart , enabling extensive ultrastructural mapping . We explored some possible analysis and quantification strategies enabled by our multiscale imaging procedure . We acknowledge that results from this study are very preliminary: Analysis of more heart samples is needed to reach conclusions applicable to CHD . In the future , combining echocardiography , which can acquire in vivo images of the heart for functional analysis ( including blood flow ) ( Midgett et al . , 2017a ) , together with 3D micro-CT and SEM , can reveal functional as well as detailed microstructural and ultrastructural characteristics of normal versus CHD hearts . Further , a combination of segmentation , quantification and other refined methods to interrogate images ( at the functional , microstructural and ultrastructural levels ) will elucidate similarities and differences between normal and malformed hearts , possibly informing therapeutic treatment strategies . To be biologically meaningful , however , these studies need to include more animals . The quantifications and comparisons presented here for one control and one TOF heart ( thus n = 1 ) pertain only to these two hearts , and are presented as an illustration of possible ways of extracting information from the proposed multiscale imaging method . Micro-CT images reveal the microstructural characteristics of hearts . Not only could we classify the hearts based on phenotype ( normal vs TOF ) , but cardiac characteristics , such as heart size and wall thickness could be visualized and quantified . For example , it was noted from our analysis that the RV of the TOF heart exhibited a larger volume than that of the CON heart examined here ( see Figure 3 ) . During fetal life , the lungs are not functional , and blood to the lungs is shunted to the systemic circulation through the ductus arteriosus ( Kiserud , 2005 ) ( a pair of ducti in chick [Dzialowski , 2018] ) . The RV hypertrophy characteristic of TOF , develops over time after the baby is born ( Iacobazzi et al . , 2016 ) . RV hypertrophy , therefore , may not be present at the fetal stages of heart development examined in this study and was not observed in our TOF heart . In the future , it would be interesting to determine whether the trait observed in this study is preserved among TOF fetal hearts , and if so under which conditions and how it affects the RV wall ultrastructure ( not examined here ) . To be meaningful , however , such study requires a larger number of heart samples , and is outside of the scope of this paper . Another difference between the two hearts was that the ventricles of the TOF heart exhibited less dense tissue and a more extended trabecular architecture compared with the control heart . This is consistent with reduced myocardial compaction in TOF ( Jenni et al . , 1999; Weiford et al . , 2004; Finsterer et al . , 2017 ) . The heart trabeculae is characterized as a ‘spongy’ or porous tissue that develops inside the heart ventricles , and in our samples was evident from semithin transverse heart sections ( Figure 4 ) , but could also be approximately quantified as the extracellular portion of the images ( Figure 8 ) . It has been shown that the heart trabecular architecture is sensitive to blood flow conditions during development ( Sedmera et al . , 1999 ) , and thus a disrupted trabecular architecture may be a characteristic of many CHD hearts due to their anomalous flow characteristics during fetal stages ( Wiputra et al . , 2018 ) . However , the trabecular and myocardial architecture can also exhibit variations from heart to heart ( Gilbert et al . , 2007 ) , therefore further analysis with a larger sample size is required before we can make conclusions related to TOF . We noticed differences in SBF-SEM image sharpness , which are attributable to excessive ‘charging’ ( accumulation of static charge on a sample’s surface ) when scanning the TOF heart ( Figure 5 ) . The increased charging in our TOF heart is linked to its trabeculation , which features larger and more numerous void regions filled with free resin . To address this problem during imaging , we slightly shortened the dwell time when acquiring SBF-SEM images from the TOF sample . In the future , we could embed silver particles in the resin to increase sample conductivity and enhance image quality ( Genoud et al . , 2018 ) . Nevertheless , the image quality of both the TOF and CON hearts was sufficient to appreciate ultrastructural details ( Figure 5 ) . While 2D EM images have been invaluable in deciphering ultrastructural features of myocardial cells and tissues , 3D images can unravel more details in the spatial organization of the ultrastructural architecture ( Hussain et al . , 2018; Pinali et al . , 2015 ) . As an example , segmentation and quantification of the 3D data revealed that myofibril alignment was slightly different between our TOF and control hearts ( Figure 10 ) . This is perhaps because the ROIs from the two hearts are not exactly corresponding , or due to the more extended trabecular architecture of the TOF heart , and warrants further investigation . For myocardial alignment quantification , it is also important to arrest the heart consistently ( in diastole as done here , or systole ) as myocardial cell orientation changes over the cardiac cycle ( Omann et al . , 2019; Sonnenblick et al . , 1967 ) . 3D SBF-SEM images also revealed a greater proportion of endocardial cells in TOF heart tissues than in control tissues , such that volumetric studies not focusing on myocardial cells show reductions in the myofibril density of the TOF heart ( data not shown ) . When the analysis was focused exclusively on myocardial cells , however , we could not find any differences in the density of myofibrils or mitochondria ( Figure 8 ) . Outside the scope of this paper , but relevant to the comparison of normal versus CHD hearts , would be an extensive analysis of left and right ventricular wall microstructure and ultrastructure , including the distribution of lipid droplets , glycogen , and mitochondria with respect to the myofibrils . In addition , studies of the ultrastructural organization within myocardial , endocardial , fibroblast and conduction cells , in normal and CHD hearts , would be relevant to decipher the impact of CHD on cell and cardiac function . While outside the scope of this paper , future studies that include more animals should focus on elucidating ultrastructural cardiac differences in animal models of CHD as such differences can inform clinical studies and impact the lives of children and adults with congenital heart defects . Our proposed multiscale imaging methodology could certainly enable such studies . Previous studies have detailed the congenital heart anomalies associated with outflow tract banding ( OTB ) Gessner , 1966; Clark et al . , 1984; Midgett et al . , 2014 used in this study to induce TOF . Those studies found a spectrum of congenital heart defects after OTB , which originated from abnormalities in the outflow tract ( conotruncal defects ) . These abnormalities included increased separation between the aortic and mitral valve annuli ( altered aortic-mitral valve continuity ) , ventricular septal defects , abnormal position of the aorta , including TOF and double outlet right ventricle ( DORV ) , in which both the aorta and pulmonary trunks emerge from the right ventricle . Meanwhile , neural crest cell ablation also leads to conotruncal anomalies in the chick embryo ( Hutson and Kirby , 2003; Kirby et al . , 1983 ) . Cardiac neural crest cells are required for normal heart development ( in the chick and mouse ) , and ablation of these cells leads to persistent truncus arteriosus ( PTA ) , characterized by lack of separation of the aorta and pulmonary trunk , but also to TOF and DORV . Likewise , diverse genetic anomalies are also associated with conotruncal heart defects ( Srivastava and Olson , 2000; Rugonyi , 2016 ) . However , the mechanisms by which anomalous genes , neural crest cell ablation , and altered hemodynamics lead to conotruncal defects may differ , and these differences may impact myocardial and myofiber orientation and maturation . Multiscale imaging studies could reveal the impact of diverse interventions on cardiac microstructure and ultrastructure , both when the same or different phenotypes are obtained . This in turn could contribute to our understanding of the underpinnings of CHD and their functional and structural consequences . Direct application of the proposed multiscale imaging method to human hearts is limited . The method presented here is destructive , and thus can only be applied to human samples of deceased individuals . Moreover , the larger size of the human heart will lead to difficulties in attaining homogeneous fixation and staining . To circumvent fixation and stain homogeneity issues , increased timing for diffusion of fixative and heavy metal stains is certainly a possibility as is microwave steps ( to accelerate diffusion ) . In addition , changes in processing , such as sectioning of hearts after micro-CT to facilitate diffusion of heavy metal stains and control imaging are also possible . As presented , using our methods , multiscale imaging of human hearts is limited . Our correlative , multiscale imaging procedure allowed us to acquire detailed micro-CT images of an entire embryonic chicken heart ( see Figure 2 ) , followed by ultrastructural 3D SBF-SEM images from the same heart ( see Figures 5 and 6 ) . The described approach allows the correlation of microstructural and ultrastructural architecture in selected regions of the heart . This is important when studying CHDs , as each malformation phenotype may be different and therefore may need to be analyzed separately to fully appreciate multiscale effects and to understand how phenotypes affect cardiac architecture at disparate levels . Furthermore , similar phenotypes that result from diverse insults ( e . g . hemodynamics , neural crest cell ablation ) could also lead to dissimilar cardiac ultrastructure and function . Importantly , multiscale studies can be used to decipher the imprints that early alterations in the environment in which the heart is growing have on cardiac formation and function . Other potential applications to CHD ( and beyond ) are determinations of extracellular matrix organization/disorganization , cardiac fibrosis , glycogen distribution and myxomatous degeneration of valve tissues in response to different insults and aging . The multiscale imaging approach presented here therefore could enable animal studies to inform how human cardiac anomalies , even when repaired , could subsequently lead to increased cardiac dysfunction and heart failure . For patients with CHD , such studies may further reveal associated pathologies in cardiac tissues that , if not properly treated , may have devastating implications for survival and long term cardiac health . While our multiscale imaging approach was implemented and optimized using embryonic chicken hearts , we expect it will be straightforward to adapt it for use in mouse and other small animal models of cardiac malformations . It will be advantageous to use complementary models , as typically genetic insults are studied using mouse models , while environmental perturbations are studied using avian models . Heart dimensions in those species ( mouse and chicken ) are very similar , and we anticipate that tissue processing will not differ significantly . Slight increases in heart size may just require an increase in protocol staining times . Extending the approach to different species and models of congenital heart disease will likely enable us to understand in detail the similarities and differences between cardiac defects , and the underpinnings of malformations that result from genetic and environmental insults . Our research used chicken embryos . According to the US National Institutes of Health ( NIH ) Office of Laboratory Animal Welfare ( ILAR News 1991; 33 ( 4 ) :68–70 ) , the NIH’s ‘Office for Protection from Research Risks has interpreted ‘live vertebrate animal’ to apply to avians only after hatching . ’ Our Institutional animal care and use committee ( IACUC ) follows NIH interpretation . Therefore , chicken embryos are not considered animals and our research did not require approval . Incubator logs in the lab were monitored daily to ensure there were no eggs near the hatching time of 21 incubation days . Nevertheless , we used the minimum possible number of embryos to achieve our goals . Our multiscale approach was implemented and optimized using fully formed embryonic chicken hearts ( heart length ~ 5–6 mm ) , and applied to a chick animal model of congenital heart disease . Chicken embryos were prepared as described previously ( Midgett et al . , 2017a ) . Briefly , fertilized white Leghorn chicken eggs were incubated blunt end up at 38°C and 80% humidity for approximately 3 days ( to Hamburger and Hamilton ( HH ) stage HH18 [Hamburger and Hamilton , 1992] ) . Control and treatment interventions were then performed as described below and the embryos were re-incubated for an additional 9 days ( to HH38 , when the heart has four chambers and valves ) . Two embryonic hearts were included in this study: ( 1 ) a control , normal heart; and ( 2 ) a malformed heart with tetralogy of Fallot ( TOF ) . TOF was achieved by performing outflow tract banding ( OTB ) at HH18 , wherein a 10–0 nylon suture was passed under the mid-section of the heart outflow tract and tied in a knot ( band tightness 38% ) . The band was removed from the outflow tract ~ 24 hr after placement ( HH24 ) , and then the embryo was allowed to develop to HH38 . The control heart was obtained by passing a 10–0 nylon suture under the heart outflow tract without knotting it , and subsequently allowing the embryo to develop to HH38 . Embryo hearts were collected at HH38 for multiscale imaging . At HH38 , embryonic whole hearts were excised and fixed as follows . The chest cavity was opened and the pericardial sac around the heart gently removed with forceps . Each heart was arrested by injecting 500 µL of chick ringer solution containing 60 mM KCl , 0 . 5 mM verapamil , and 0 . 5 mM EGTA ( Tobita et al . , 2005 ) into the left ventricle through the heart’s apex . Hearts were then immediately perfused with ~2 mL of ice-cold ( 0°C ) modified Karnovsky’s fixative ( 2 . 5% Glutaraldehyde and 2 . 5% PFA in PBS ( pH 7 . 4 ) ) through the same injection site . All perfusions were performed with a 21 gauge needle . A transfer pipette was used to quickly apply ~ 1 mL of fixative to the heart’s exterior to ensure uniform fixation of the heart tissue . Next , the heart great vessels were cut with small spring scissors and hearts were placed in 1 . 5 mL fixative and stored at 4°C until further processing . In order to enable both whole-heart micro-CT imaging and subsequent SBF-SEM imaging of regions of interest ( ROIs ) , we processed fixed hearts for micro-CT using the initial portion of a Renovo Neural , Inc ( Cleveland , USA ) protocol ( Mukherjee , 2016 ) designed for SBF-SEM imaging ( see Figure 1 , Step 1 ) . Each heart was placed in a 5 mL glass scintillation vial and we used 3 mL of solution per vial for each incubation/wash . First , the fixed hearts were washed in 0 . 1M Sodium Cacodylate ( pH 7 . 4 ) for 20 min with 4 exchanges of fresh buffer . Next , the hearts were incubated in 0 . 1% ( w/v ) of tannic acid in 0 . 1M Sodium Cacodylate ( pH 7 . 4 ) for 15 min at room temperature . Samples were then washed in 0 . 1M Sodium Cacodylate ( pH 7 . 4 ) for 20 min with 4 exchanges of fresh buffer . Since the reducing agents used in subsequent steps ( modified ROTO protocol ) were light-sensitive , the sample vials were covered in aluminum foil from this point on . The whole hearts were post-fixed in 2% ( v/v ) Osmium Tetroxide ( OsO4 ) and 1 . 5% ( w/v ) Potassium Ferricyanide ( K₃[Fe ( CN ) ₆] ) in distilled water ( dH2O ) for 2 hr at room temperature on a rotating platform . The samples were then extensively washed in dH2O for 20 min with four exchanges of fresh dH2O . Next , the samples were immersed in 0 . 1% ( w/v ) Thiocarbohydrazide ( TCH ) solution in dH2O , placed in an oven , and incubated for 40 min at 60°C . This step was followed by another four exchanges of fresh dH2O over 20 min . Samples were then immersed in a 2% ( v/v ) OsO4 solution in dH2O for 2 hr at room temperature on a rotating platform . Finally , the hearts were washed extensively in dH2O over 20 min with four exchanges of fresh water . Each heart was stored in dH2O at 4°C until imaged by micro-CT . This preparation provided excellent contrast for micro-CT scans ( see Results ) . Micro-CT images of whole hearts were acquired to assess the cardiac structure . We acquired high-resolution ( ~10 μm ) 3D scans of each heart using a Caliper Quantum FX Micro-CT system ( Perkin-Elmer , CLS140083 ) with 10 mm field of view , 140 μA current , 90 kV voltage , and a scan time of 3 min . We used the Amira 6 . 0 software platform ( FEI Company ) or Dragonfly 4 . 1 software ( Object Research Systems , Quebec , Canada ) to visualize these scans and identify cardiac defects . Hearts were then stored in double distilled water at 4°C until further processing . Please note that at this step in the processing ( cardiac tissues fixed and post-fixed with OsO4 ) water does not damage the tissues . After whole hearts were imaged with micro-CT , sample preparation of selected hearts for 3D SBF-SEM imaging was finished ( see Figure 1; Step 2 ) , following the Renovo Neural , Inc protocol . In large samples , like the whole hearts described in this manuscript , it is necessary to extend most of the staining steps . Failure to extend the timing of staining resulted in a heterogenous stain distribution throughout the tissue ( in our early iterations of the procedure ) . In our final , optimized procedure , we incubated the samples in 1% ( w/v ) aqueous uranyl acetate for 24 hr at 4°C , after which they were washed in dH2O for 30 min with 6 exchanges of fresh dH2O . We then incubated the samples in lead aspartate for 30 min at 60°C . The samples were then extensively washed in dH2O for 20 min with 4 exchanges of dH2O . Dehydration steps were done in a series of acetone-dH2O mixtures ( 50 , 75 , 85 , 95 , and 100% ) ; each step was repeated twice for 5 min at room temperature . The whole heart sample was then embedded in an epoxy ( Epon 812 ) resin for further manipulation and SBF-SEM sample preparation . The first infiltration step was done for 1 hr at room temperature in a mixture of 1:1 ( v/v ) acetone:epon followed by a 1:3 ( v/v ) acetone:epon incubation for 1 hr at room temperature . The hearts were subsequently incubated overnight in pure ( 100% ) epon on a rotating platform . The following day the epoxy solution was exchanged four times , each time with 30 min incubation steps at room temperature . Samples were polymerized at 60°C for 48 hr in a conventional oven , leading to a whole heart sample embedded in an Epon block . Using the micro-CT images as reference , the Epon-embedded heart blocks were sectioned to reach a selected short axis ( transverse ) section using a diamond-wire jewelry saw . For this study we selected the mid cardiac transverse section , at a plane where the heart is wider ( the equatorial plane ) . After this step , a semithin section ( 250 nm ) was obtained using an ultramicrotome and mounted on a silicon chip previously glow discharged for 1 min at 15 mA ( PELCO easyGlow , Ted Pella ) . Semithin section images were used to confirm the area of interest as well as to check for both the ultrastructural quality of the sample and the success of the staining procedure ( see Figure 1; Step 2 ) . This step is crucial since the SBF-SEM imaging requires samples with extremely good contrast . Semithin sections were imaged on a Teneo Volume Scope in low vacuum mode using a VS-DBS backscattered electron detector and the MAPS software ( FEI Company ) . Imaging conditions used were 2 . 5 kV and 0 . 2 mA , dwell 3–5 µs . In some cases , the samples imaged using this method needed to be coated with a thin ( 5–8 nm ) layer of carbon to minimize charging artifacts induced by the electron beam . The same diamond-wire jewelry saw was then utilized to generate a slab ( ~1 . 5 mm ) from the sample ( see Figure 1; Step 3 ) . The slabs were sectioned into smaller ROIs , which were subsequently mounted on Microtome stub SEM pins ( Agar Scientific 61092450 ) using H20E Epo-Tek silver epoxy ( Ted Pella 16014 ) and cured overnight at 60°C in a conventional oven . The resulting small blocks were then trimmed using a Trim90 diamond knife ( Diatome ) to generate a pillar of 500 × 500 µm2 . The block was then coated with 20 nm of gold using a Leica ACE 600 unit . In the last step of our multiscale imaging procedure , 3D SBF-SEM images of sub-ROIs selected from the mounted sample were acquired ( see Figure 1; Step 4 ) . 3D image acquisition was done on a Teneo Volume Scope SBF-SEM in low vacuum mode ( 50 Pa ) using a VS-DBS backscattered detector . Images were acquired at a lateral resolution of 10 nm/pixel and image sets included 800–1000 serial sections ( with each section thickness measuring 40 nm in the z axis ) . SBF-SEM data sets were approximately 40 μm × 60 μm×32–40 μm . All registration , of SBF-SEM data was performed with Amira 6 . 0 ( FEI Company ) . First , complete image stacks ( 800–1000 slices ) from each ROI and sub-ROIs were automatically aligned to generate a continuous 3D volume . Next , a non-local means filter was applied to every 2D slice in order to improve the signal-to-noise ratio . Due to slight differences in the intrinsic properties of the tissue , sections from the TOF heart appeared slightly lighter compared to the control heart . We adjusted the intensity of the TOF sections during post-processing to match that of the control heart . To better appreciate ultrastructural differences between the two hearts , we used Dragonfly 4 . 1 software to segment and quantify SBF-SEM images . We segmented: cell nuclei , mitochondria , myofibrils and the extracellular space ( this later one to allow quantification of relative organelle volume within cells ) . The segmentation used a combination of tools in Dragonfly , including deep learning algorithms . Briefly , we employed a six-level U-Net deep learning model ( Ronneberger et al . , 2015 ) implemented in Dragonfly to perform an initial segmentation of nuclei , mitochondria , and myofibrils within imaged cells . Training sets required for the deep learning model were obtained initially through manual segmentations of a few selected images from the image stacks . The training set was later augmented by applying the segmentation deep learning algorithms to other ( selected ) images from the set , follow by thorough manual cleaning . For each training session , the model was run for 50 epochs with a patch size of 128 pixels . Dragonfly automatically divides the training sets into training and validation regions , so that training ( and further inclusion of training images ) continued until the reported accuracy ( from validation regions ) was > 98% with a loss < 0 . 06 . Images were then segmented with the trained model , and segmentations further refined both using Dragonfly automated tools , such as morphological operations , and manual clean up using painter tools available in Dragonfly . Unlike organelles , the extracellular space was easily recognized by intensity levels , and thus simply segmented based on its intensity , followed by clean up using both automatic and manual tools in Dragonfly . For visualization purposes , we segmented a whole dataset from the control heart and an approximately corresponding dataset from the TOF heart ( region A , Figure 9 ) . The total volume of the dataset was 40 × 60 × 32 μm3 . For quantification purposes , we segmented and further curated smaller portions of the data sets from two corresponding regions of the control and the TOF hearts ( regions A and B ) . Quantifications of extracellular space and nuclei were done from 17 evenly spaced images from the 800 image datasets of region A; and 21 evenly spaced images from the 1000 image datasets of region B ( thus every 50th image was used for these segmentations ) . For quantification of mitochondria and myofibrils , we cropped images ( n ≥ 10 ) so that we could focus on smaller regions , allowing us to manually improve the accuracy of segmentations in a more tractable manner and focusing on myocardial regions . Quantifications were performed based on segmented images . We quantified , from each image or image portion , the total surface area ( ST ) , and the surface area occupied by extracellular space ( SE ) , nuclei ( SN ) , mitochondria ( SMit ) and myofibrils ( SMyo ) . We then computed the fraction of the total surface area occupied by the extracellular space ( SE/ST ) ; and the fraction of the cell occupied by organelles ( nuclei , mitochondria and myofibrils ) , computed as the ratio of organelle surface ( Si , with i = N , Mit , Myo ) to the cell surface ( Si/ ( ST - SE ) ) . Because quantifications were performed from different portions ( n ≥ 10 ) of the dataset , average and standard deviations were calculated to represent quantifications for the dataset ( control or TOF hearts , regions A and B ) . Myofibril orientation ( angle ) quantifications were also performed . To this end , we used the myofibril segmentations in the transmural image plane ( x-y ) , split them into individual segmentations ( each myofibril was separately assessed using the multi-ROI functions in Dragonfly ) and quantified the transmural angle of each myofibril , as the angle with respect to the x-direction , θ angle . Because the x-direction was approximately parallel to the wall boundary , θ = 0° represents myofibrils that are oriented parallel to the wall or in circumferential direction , and θ = 90° are myofibrils oriented towards the ventricle lumen in the radial direction . Angular quantifications were plotted as histograms showing frequency ( number of myofibrils ) versus an angular range . For quantifications in the x-z plane ( a plane approximately parallel to the heart wall ) , we re-sampled the images acquired in the x-y plane , to obtain images in the x-z plane , and then performed the quantifications as before . We quantified the elliptical orientation angle of each myofibril ( x-z plane ) with respect to the x-direction , Φ angle . Thus Φ = 0° , represents myofibrils that are oriented circumferentially around the ventricle , and θ = 90° are myofibrils oriented longitudinally along the ventricular wall ( see Figure 10 ) . Results were plotted as histograms , from which elliptical and transmural angular orientations were further characterized using average and standard deviation data .

The heart is our hardest-working organ and beats around 100 , 000 times a day , pumping blood through a vast system of vessels to all areas of the body . Specialized heart cells make the heart contract rhythmically , enabling it to work efficiently . Contractile molecules inside these cells , called myofibrils , align within the heart cells , and heart cells align to each other , so that the heart tissue contracts effectively . However , when the heart has defects or is diseased this organization can be lost , and the heart may no longer pump blood efficiently , leading to sometimes life-threatening complications . For example , around one in a hundred newborn babies suffer from congenital heart defects , and despite medical advances , these conditions remain the main cause of non-infectious mortality in children . Many cases of congenital heart disease are diagnosed before a baby is born during an ultrasound scan . However , these scans , as well as subsequent diagnostic tools , lack the precision to detect problems within the heart cells . Now , Rykiel et al . used two complementary imaging techniques known as micro-computed tomography and scanning electron microscopy to acquire pictures of the whole heart as well as of the organization inside the heart cells . This made it possible to capture the structure of the heart tissue at both micrometer ( the whole heart ) and nanometer resolution ( the inside of the cells ) , and to study what happens within the heart and its cells when the heart has a defect . Rykiel et al . tested the imaging technology on the hearts of chicken embryos , at stages equivalent to a five to six-month-old human fetus , and compared a healthy heart with a heart with a defect called tetralogy of Fallot . They found that the tissues in the heart with a defect had a sponge-like appearance , with increased space in between cells . Moreover , the myofibrils of the heart with a defect were aligned differently compared to those in the normal heart . More research is needed to fully understand what happens when the heart has a defect . However , the imaging technology used in this study offers the possibility of examining the heart at an unprecedented level of detail . This will deepen our understanding of how structural heart defects arise and how they affect the pumping of the heart , and will give us clues to design better treatments for patients with heart defects and other heart anomalies .

2020

Multiscale cardiac imaging spanning the whole heart and its internal cellular architecture in a small animal model

We performed a genetic analysis of sRNA abundance in flag leaf from an immortalized F2 ( IMF2 ) population in rice . We identified 53 , 613 , 739 unique sRNAs and 165 , 797 sRNA expression traits ( s-traits ) . A total of 66 , 649 s-traits mapped 40 , 049 local-sQTLs and 30 , 809 distant-sQTLs . By defining 80 , 362 sRNA clusters , 22 , 263 sRNA cluster QTLs ( scQTLs ) were recovered for 20 , 249 of all the 50 , 139 sRNA cluster expression traits ( sc-traits ) . The expression levels for most of s-traits from the same genes or the same sRNA clusters were slightly positively correlated . While genetic co-regulation between sRNAs from the same mother genes and between sRNAs and their mother genes was observed for a portion of the sRNAs , most of the sRNAs and their mother genes showed little co-regulation . Some sRNA biogenesis genes were located in distant-sQTL hotspots and showed correspondence with specific length classes of sRNAs suggesting their important roles in the regulation and biogenesis of the sRNAs . Small RNAs ( sRNAs ) are non-coding RNAs mainly 18–30 nt in length that regulate a wide range of biological processes in eukaryotic organisms ( Carthew and Sontheimer , 2009; Axtell , 2013 ) . According to their origin , sRNAs can be grouped into two major types: hpRNAs that are derived from single-stranded precursors with a hairpin structure ( such as microRNAs [miRNAs] ) and short interfering RNA ( siRNAs ) that are derived from double-stranded RNA precursors such as heterochromatic small interfering RNAs ( hc-siRNA ) and trans-acting siRNAs ( ta-siRNA ) . There has been an explosion of interest in recent years in studies of miRNAs and siRNAs on their identification , biogenesis , and functioning in diverse biological processes . In plants , sRNAs function in regulating growth , development ( Juarez et al . , 2004; Zhu and Helliwell , 2011 ) , architecture ( Jiao et al . , 2010; Miura et al . , 2010 ) , yield ( Zhang et al . , 2013 ) , and response to biotic and abiotic stresses ( Lu et al . , 2008; Shukla et al . , 2008 ) . Such regulations are usually achieved by mediating endogenous mRNA cleavage and decay , DNA methylation of source and target loci , and chromatin modification and transcriptional silencing ( Arikit et al . , 2013 ) . Although the sRNAs differ in length , sequences , and functions , the pathways for their biogenesis and functioning from precursor transcription , processing , maturation , and action are relatively conserved , which involve the activities of a number of enzymes including RNA polymerase II ( Pol II ) , RNA-dependent RNA polymerases ( RDRs ) , Dicer-like proteins ( DCLs ) , and Argonautes ( AGOs ) ( Chen , 2009; Ghildiyal and Zamore , 2009 ) . It is also known that the abundance of sRNA species can be highly variable among individuals within a species , and the possible regulatory role of such quantitative difference has been assumed ( He et al . , 2010; Groszmann et al . , 2011b ) . It is not known whether the quantitative variation of sRNA species between genotypes is related to the biological machinery , as quantitative variation of sRNAs has not been assayed at the population level and their genetic control has yet to be elucidated . The recently developed expression quantitative trait locus ( eQTL ) analysis has provided an approach for determining the genetic control of the expression level of a gene , including cis- and trans-eQTLs , as well as epistatic effects ( Becker et al . , 2012 ) . This approach can also be applied to the genetic analysis of quantitative variation of sRNAs by regarding the abundance of the sRNAs in the population as quantitative traits . Once the QTLs are identified , subsequent studies can be pursued very much the same way as the analysis of genes and regulatory networks underpinning phenotypic QTLs ( Xing and Zhang , 2010 ) . There have been several studies focusing on the genetic regulation of known and validated miRNAs and small nucleolar RNAs in specific tissues/cells from samples of the human population ( Borel et al . , 2011; Gamazon et al . , 2012; Parts et al . , 2012; Civelek et al . , 2013; Jin and Lee , 2013; Siddle et al . , 2014 ) . These studies detected a number of cis- and trans-miQTLs that could also influence the expression of the mRNA targets , which may be associated with phenotype difference . Here , we performed a whole genome QTL analysis of the entire sRNA kingdom consisted of 18-nt to 26-nt sRNAs from flag leaf of rice using an experimental genetic population . The analysis revealed features of the genetic controls of sRNA abundance showing both commonality and distinction with their precursor transcripts . It was also shown that the abundance of sRNAs is probably related to proteins constituting the machinery for sRNA biogenesis and functioning . The genetic materials consisted of 98 hybrids obtained by paired crosses of 196 recombinant inbred lines ( RILs ) derived by single seed descent from a cross between Zhenshan 97 and Minghui 63 , the parents of Shanyou 63 that was the most widely cultivated rice hybrid in the 1980s and 1990s and still used with reduced area in recent years in China . Because the genetic composition of these crosses resembled individuals in an F2 population , they were referred to as an immortalized F2 ( IMF2 ) population ( Hua et al . , 2002 , 2003 ) , and each hybrid in the population was hereafter referred to as an IMF2 for ease of description . An sRNA library was constructed using RNA extracted from flag leaf at the day of full expansion for each IMF2 , and two biological replicates were obtained for each of the two parental lines and their F1 hybrid , producing a total of 104 libraries . The read size of the raw sequencing data obtained using Illumina Hiseq2000 varied from 3 nt to 44 nt ( Figure 1—figure supplement 1 ) . Sequences of 18–26 nt in length that appeared to be the more abundant than others were kept for the analyses after filtering out low-quality reads and eliminating ones matching tRNAs , rRNAs , snRNAs , and snoRNAs ( Figure 1—figure supplement 2 ) . The numbers of resulting reads varied from 14 . 52 million to 27 . 73 million per library ( 2 . 08 billion in total ) with non-redundant reads ranging from 3 . 00 million to 6 . 86 million per library ( 0 . 52 billion in total ) ( Supplementary file 1 in Dryad [Wang et al . , 2015] , Table 1 , Figure 1—figure supplement 2 ) . The 24 nt sRNAs were the most numerous in both redundant and distinct reads ( Figure 1—figure supplement 2 ) . 10 . 7554/eLife . 03913 . 003Table 1 . Average number of reads obtained for the 104 libraries from flag leaves of the IMF2 population and the parental linesDOI: http://dx . doi . org/10 . 7554/eLife . 03913 . 003Quality filtering*Size filtering†NcRNA filtering‡Genome mapping§Total#Redundant reads22 , 864 , 06722 , 318 , 92819 , 968 , 5669 , 816 , 9331 , 020 , 961 , 021Distinct reads5 , 106 , 6545 , 035 , 0734 , 986 , 1842 , 499 , 23353 , 613 , 739*Reads after filtering out low-quality reads . †Reads of 18–26 nt in length . ‡Reads after eliminating ones matching tRNAs , rRNAs , snRNAs , and snoRNAs . §Reads mapped to the SNP-replaced reference genomes of the parents with unique locations allowing no mismatch . #The total reads of 104 libraries used to identify s-traits . We mapped the reads to the SNP-replaced reference genomes of the parents ( ‘Materials and methods’ ) , with unique location allowing no mismatch . The reads could be divided into four categories: ( 1 ) approximately 82 . 28% of the sRNAs had identical sequences between parents , ( 2 ) 0 . 71% of the sRNAs had SNPs between the two parents , ( 3 ) 8 . 35% were only mapped to the Zhenshan 97 genome , and ( 4 ) 8 . 66% were specifically mapped to the Minghui 63 genome ( Supplementary file 2 in Dryad [Wang et al . , 2015] ) . A total of 53 , 613 , 739 unique sRNA sequences including ones with SNPs between the parents were identified by combining all the reads from 104 libraries ( Table 1 ) . Approximately 84 . 7% of sRNAs were found in no more than 5 IMF2s , while only 0 . 13% were present in all 98 IMF2s ( Figure 1A ) . 10 . 7554/eLife . 03913 . 004Figure 1 . The distribution of sRNAs and s-traits in different genomic regions across the IMF2 population . ( A ) The distribution of sRNAs across the IMF2 population . ( B ) The distribution of sRNAs aligned to the genic region , 2-kb upstream , and 500-bp downstream of annotated genes , as well as intergenic regions . TE: transposons; NTE: non-transposon genes . ( C ) The percentages of sRNAs of different sizes . ( D ) The distribution of sRNAs of different sizes in different genomic regions . ( E ) The distribution of sRNAs of different sizes in different portions of genic regions . ( F ) The percentages of s-traits of different sizes . ( G ) The distribution of s-traits aligned to the genic region , 2-kb upstream , and 500-bp downstream of annotated genes , as well as intergenic regions . The legend is the same as in ( B ) . ( H ) The distribution of s-traits of different size in different genomic regions . The legend is the same as in ( D ) . ( I ) The distribution of s-traits of different sizes in different portions of genic regions . The legend is the same as in ( E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03913 . 00410 . 7554/eLife . 03913 . 005Figure 1—figure supplement 1 . The number of sRNA reads of different lengths in the raw sequencing data of the IMF2 population . The number of sRNAs with specific length for all the IMF2s is shown as box plot . DOI: http://dx . doi . org/10 . 7554/eLife . 03913 . 00510 . 7554/eLife . 03913 . 006Figure 1—figure supplement 2 . Processing of raw sRNA reads to genome-mapped reads . The number of sRNAs of specific length in the IMF2 population is shown as box plot . DOI: http://dx . doi . org/10 . 7554/eLife . 03913 . 00610 . 7554/eLife . 03913 . 007Figure 1—figure supplement 3 . Distribution of sRNAs in the upstream and downstream of genic regions . The distribution of the distance between sRNAs and transcription start site ( TSS ) ( 2-kb upstream ) ( A ) and transcription termination site ( TTS ) ( 500-bp downstream ) ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03913 . 00710 . 7554/eLife . 03913 . 008Figure 1—figure supplement 4 . The distribution of coefficients of variation ( CV ) for 165 , 797 s-traits . DOI: http://dx . doi . org/10 . 7554/eLife . 03913 . 00810 . 7554/eLife . 03913 . 009Figure 1—figure supplement 5 . The distributions of the expression values across all the IMF2s for 9 randomly selected s-traits . DOI: http://dx . doi . org/10 . 7554/eLife . 03913 . 009 The distributions of the sRNAs in different portions of the genome were not random . Approximately 51 . 1% of sRNAs originated from the 2-kb upstream and the genic regions of non-transposon genes ( Figure 1B ) . In addition , sRNAs occurred in high frequencies near the transcription start sites compared to other regions of the promoters ( Figure 1—figure supplement 3 ) . About 34 . 4% of sRNAs originated from intergenic regions , which account for no more than 20% of the whole genome length ( Figure 1B ) . About 43 . 9% of sRNAs were 24 nt , while 12 . 1% were 21 nt ( Figure 1C ) . Different species of sRNAs differed in their origins of genomic regions . The genic regions of non-transposon genes held the highest number of 21 nt sRNAs , while 24 nt sRNAs were most common in the intergenic regions ( Figure 1D ) . sRNAs of 21 nt in genic regions were mostly derived from the exon of non-transposon genes , while 24 nt sRNAs in genic regions were mainly from the intron of non-transposon genes ( Figure 1E ) . The 24-nt sRNAs mainly consist of endogenous heterochromatic siRNAs ( Axtell , 2013 ) and tend to be produced from repeats and transposable elements as well as intergenic regions ( Chen , 2009 ) . The 24-nt sRNAs are components of the epigenome that target the homologous genomic regions for de novo DNA methylation through RNA-directed DNA methylation to maintain genome stability by transcriptional gene silencing ( Groszmann et al . , 2011a , 2013; Pooggin , 2013 ) . Next , the abundance of each sRNA in a library was normalized to number of reads per millions ( RPM ) ( ‘Materials and methods’ ) to quantify the sRNA levels in each library , which were subject to analyses and comparisons . An sRNA with RPM value ≥0 . 6 was regarded as expressed , and an sRNA identified as expressed in more than 25 of the 98 IMF2s was regarded as an sRNA expression trait ( s-trait ) . In this way , a total of 165 , 797 s-traits were recovered ( Supplementary file 3 in Dryad [Wang et al . , 2015] ) . A total of 87 . 9% of s-traits were 24 nt ( Figure 1F ) . The s-traits mainly originated from the intergenic regions , 2-kb upstream , and genic regions of non-transposon genes ( Figure 1G ) . Although the 2-kb upstream and the genic regions of non-transposon genes contained almost equal number of sRNAs ( Figure 1B ) , the number of s-traits derived from the 2-kb upstream of non-transposon genes was more than twice of the number of s-traits in genic regions of non-transposon genes ( Figure 1G ) . The distribution of different species s-traits in different genomic regions was similar to that of sRNAs ( Figure 1D , H ) . Although sRNAs of various lengths from genic regions were mostly from exons of non-transposon genes ( Figure 1E ) , the s-traits of different sizes were mainly derived from introns of non-transposon genes ( Figure 1I ) . The coefficients of variation of the 165 , 797 s-traits ( Figure 1—figure supplement 4 ) showed a very wide range of distribution of the sRNA levels in the population . We selected 9 s-traits randomly from all the 165 , 797 s-traits , and the distributions of the expression values for most of these s-traits across the IMF2 population were more or less normal ( Figure 1—figure supplement 5 ) . To test whether s-traits originated from the same gene were transcribed together , the correlation coefficients between the expression values of s-traits originated from the same gene were calculated . A total of 3 , 739 , 873 correlation coefficients involving 15 , 078 genes and 96 , 914 s-traits were obtained , 2 , 593 , 503 ( 69 . 3% ) and 507 , 528 ( 13 . 6% ) of which were contributed by 2278 s-traits from genes LOC_Os03g01360 and 1008 s-traits from LOC_Os07g01240 , respectively ( Figure 2 ) . The correlation coefficients naturally fell into three classes: strong negative correlations , no correlation , and strong positive correlations with obvious dividing points at −0 . 3 and 0 . 3 ( Figure 2A ) , which were far above 0 . 25 , the threshold for statistical significance at p < 0 . 05 determined by simulation data . While 85 . 8% of these strong correlations ( positive or negative ) were contributed by s-traits originated from LOC_Os03g01360 ( Figure 2B ) , most of s-traits originated from gene LOC_Os07g01240 showed low ( or no ) correlations with each other ( Figure 2C ) . Although the expression values of s-traits originated from the same regions of the remaining 15 , 076 genes in the whole genome were mostly slightly positively correlated , the correlation coefficients were significantly distinct from that of the simulation of random data ( Mann–Whitney U test , p-value <2 . 2e-16 , Figure 2D ) , indicating various degrees of correlations . We found that the correlations were affected by the sizes of s-traits ( Figure 2—figure supplement 1 ) . Expression correlations between s-traits of the same size were stronger than that between s-traits of different sizes . In particular , the expression correlations between 21-nt sRNAs and 25-nt sRNAs from the same genes were very weak . The correlations between overlapped s-traits were slightly stronger than that between s-traits not overlapping with each other ( Figure 2—figure supplement 2A , B ) . The correlations between s-traits from the same transposons were slightly different from that of s-traits from the same non-transposons genes ( Figure 2—figure supplement 2C , D ) . Most of the s-traits originating from LOC_Os03g01360 were significantly correlated ( positive or negative ) with each other ( Figure 2—figure supplement 3 ) , and the correlations between different s-traits were mainly dependent on their genomic positions rather than their sizes , which were mostly 21 , 22 , and 24 nt , accounting for 40 . 2% , 34 . 4% , and 19 . 4% of the 2278 s-traits . This gene could be separated as three parts , the 2 kb-upstream , 5′ UTR , the first intron and the first coding DNA sequence ( CDS ) as the first part , the most of the second intron comprising the second part , and the second CDS and part of the second intron as the third part ( Figure 2—figure supplement 3 ) . s-traits within each part were positively correlated with each other , while s-traits from the first part were mostly negatively correlated with those from the other two parts . The correlations between the expression levels of s-traits from the second parts and s-traits from the third parts were mostly positive . Thus , although very strong positive and negative correlations were found between sRNAs from the same genic regions of specific genes , this was not a common phenomenon . Taken together , this correlation analysis indicated that most of the sRNAs derived from the same genes were correlated with each other to various extents , implying that nearby sRNAs were possibly jointly transcribed . Consecutively overlapped sRNAs were found from introns of specific genes and intergenic regions , indicating that sRNAs were not necessarily jointly transcribed from the genes . 10 . 7554/eLife . 03913 . 010Figure 2 . Expression correlations between s-traits originating from the same genes . ( A ) The distribution of expression correlations between s-traits originating from the same genes . ( B ) The distribution of expression correlations between s-traits originating from LOC_Os03g01360 . ( C ) The distribution of expression correlations between s-traits originating from LOC_Os07g01240 . ( D ) Comparison of the distribution of expression correlations for the simulation data and the true data excluding s-traits from LOC_Os03g01360 and LOC_Os07g01240 . DOI: http://dx . doi . org/10 . 7554/eLife . 03913 . 01010 . 7554/eLife . 03913 . 011Figure 2—figure supplement 1 . Expression correlations between s-traits of different sizes originating from the same genes . The s-traits originated from same genes were grouped by their sizes . The expression correlations between s-traits from pairs of groups were calculated , and results with more than 2000 correlations are shown in each case . The numbers on top of each plot represent the sizes of two groups of s-traits . DOI: http://dx . doi . org/10 . 7554/eLife . 03913 . 01110 . 7554/eLife . 03913 . 012Figure 2—figure supplement 2 . Expression correlations between s-traits originating from different regions of the same mother genes . ( A ) The distribution of expression correlations between s-traits originating from introns of the same mother genes . ( B ) The distribution of expression correlations between s-traits originating from 2-kb upstream of the same mother genes . ( C ) The distribution of expression correlations between s-traits originating from 2-kb upstream and s-traits from introns of the same mother genes . TE: transposons; NTE: non-transposon genes . ( D ) The distribution of expression correlations between s-traits originating from exons and s-traits from introns of the same mother genes . DOI: http://dx . doi . org/10 . 7554/eLife . 03913 . 01210 . 7554/eLife . 03913 . 013Figure 2—figure supplement 3 . The expression correlations between s-traits originating from LOC_Os03g01360 . The structure of LOC_Os03g01360 is shown by the bars in the left and bottom of the plot . Each point represents the expression correlation between a pair of s-traits . DOI: http://dx . doi . org/10 . 7554/eLife . 03913 . 01310 . 7554/eLife . 03913 . 014Figure 2—figure supplement 4 . Distribution of sRNA cluster sizes . ( A ) Distribution of the sizes of all 80 , 362 sRNA clusters . ( B ) Distribution of the sizes of sRNA clusters not longer than 10 kb . ( C ) Distribution of the sizes of sRNA clusters not longer than 1 kb . ( D ) Distribution of the sizes of sRNA clusters not longer than 100 bp . DOI: http://dx . doi . org/10 . 7554/eLife . 03913 . 01410 . 7554/eLife . 03913 . 015Figure 2—figure supplement 5 . Expression correlations between s-traits originating from the same sRNA clusters . ( A ) The distribution of expression correlations between s-traits originating from the same sRNA clusters . ( B ) The distribution of expression correlations between s-traits originating from chr03-261504-273325 . ( C ) The distribution of expression correlations between s-traits originating from chr07-136331-141317 . ( D ) Comparison of the distribution of expression correlations between s-traits originating from the same mother genes and s-traits from the same sRNA clusters . ( E ) Comparison of the distribution of expression correlations for the simulation data and the true data excluding s-traits from chr03-261504-273325 and chr07-136331-141317 . DOI: http://dx . doi . org/10 . 7554/eLife . 03913 . 015 As many sRNAs may be derived from a single primary transcript , we defined sRNA clusters to quantify sRNA expression ( Castel and Martienssen , 2013 ) , based on our data of all 104 libraries . An sRNA island was defined as a genomic region composed of consecutive genomic positions matched by at least 30× sRNA read coverage . Nearby sRNA islands resided within 1000 nt from each other were merged and regarded as the same sRNA cluster . As a result , 80 , 362 sRNA clusters were recovered with an average size of 1927 bp ( ranging from 51 bp to 47 , 021 bp ) ( Figure 2—figure supplement 4 , Supplementary file 4 in Dryad [Wang et al . , 2015] ) . A total of 46 , 293 ( 57 . 6% ) sRNA clusters overlapped with genic regions , 23 , 143 ( 28 . 8% ) of which were entirely derived from genic regions . In addition , the distances between another 19 , 164 ( 23 . 8% ) sRNA clusters and the genic regions were smaller than 1 kb . The abundance of each sRNA cluster was normalized using DEseq ( ‘Materials and methods’ ) ( Anders and Huber , 2010 ) . An sRNA cluster with normalized expression value ≥6 was regarded as expressed , and an sRNA cluster expressed in more than 25 of all the 98 IMF2s was regarded as an sRNA cluster expression trait ( sc-trait ) . A total of 50 , 139 ( 62 . 4% ) sRNA clusters were determined as sc-traits , 34 , 762 of which were expressed in all of the 98 IMF2s . We next calculated the correlation coefficients between the expression values of s-traits originating from the same sRNA clusters to infer whether they were transcribed together . A total of 4 , 282 , 587 correlation coefficients involving 21 , 571 sRNA clusters and 157 , 994 s-traits were obtained , 72 . 5% of which were contributed by s-traits from clusters chr03-261504-273325 and chr07-136331-141317 ( Figure 2—figure supplement 5 ) . These two sRNA clusters corresponded to the two genes , LOC_Os03g01360 and LOC_Os07g01240 , mentioned above . Of all the correlations , 2 , 776 , 296 were relatively strong ( above 0 . 3 or below −0 . 3 based on a threshold of 0 . 25 determined by simulation data with p <0 . 05 ) . However , 81 . 5% of these strong correlations were contributed by the 2280 s-traits from cluster chr03-261504-273325 ( Figure 2—figure supplement 5B ) . A total of 1008 s-traits originated from cluster chr07-136331-141317 , most of which were only slightly correlated with each other ( Figure 2—figure supplement 5C ) . The distribution of correlation coefficients of s-traits from the same sRNA clusters was quite similar to that of s-traits from the same genes ( Figure 2—figure supplement 5D ) . Although the s-traits originating from the same region of the remaining 215 , 69 sRNA clusters in the whole genome were only slightly positively correlated , the correlation coefficients were significantly different from that of simulated random data ( Mann–Whitney U test , p-value <2 . 2e-16 , Figure 2—figure supplement 5E ) . Thus , again very strong positive and negative expression correlations were only observed for sRNAs from certain sRNA clusters , while most sRNAs originating from the same sRNA clusters were only slightly positively correlated with each other . mRNA sequencing was conducted using the same samples as in the sRNA sequencing with one biological replicate for the IMF2 population and two biological replicates for each of the two parental lines and their F1 hybrid , producing a total of 104 libraries . After removal of low-quality sequencing data , reads were mapped to the Nipponbare reference genome using TopHat ( Figure 3—figure supplement 1 , Supplementary file 5 in Dryad [Wang et al . , 2015] ) ( Trapnell et al . , 2009 ) . Most of the reads were mapped to the CDS and UTR regions of non-transposon genes ( Figure 3—figure supplement 2 ) . The abundance of each mRNA in a library was normalized to the number of fragments per kilobase of transcript per million mapped fragments ( FPKM ) using Cufflinks ( Trapnell et al . , 2010 ) to quantify the expression levels of mRNAs in each library . An mRNA with FPKM value ≥1 was regarded as expressed; 35 , 819 ( 54 . 2% ) mRNAs were expressed in none of the 98 IMF2s , while 16 , 849 ( 25 . 5% ) were expressed in all the 98 IMF2s . An mRNA identified as expressed in more than 25 of the 98 IMF2s was regarded as an e-trait . In this way , a total of 24 , 987 e-traits were obtained from a total of 66 , 123 mRNAs in the whole genome . For ease of description , we hereafter refer to genic sequences including both exons and introns that encoded the s-traits as ‘mother genes’ . To assess whether sRNAs and their mother genes were transcribed together , the correlation coefficients of the expression values of s-traits and their mother genes were calculated . A total of 77 , 640 correlation coefficients were obtained , involving 9968 mRNAs and 56 , 185 s-traits . Among all these correlations , 13 , 749 ( 17 . 7% ) were strong correlations ( above 0 . 3 or below −0 . 3 , based on a threshold of 0 . 23 determined by simulation data with p < 0 . 05 ) . However , 45 . 6% of these strong correlations were again contributed by gene LOC_Os03g01360 ( Figure 3A ) . Correlations involving LOC_Os07g01240 were mostly weak ( Figure 3B ) . The correlation coefficients involving rest of the genes in the whole genome were centered around zero ( Figure 3C ) . 10 . 7554/eLife . 03913 . 016Figure 3 . Correlations between the s-traits and the transcripts of their mother genes . ( A ) Correlations between s-traits and e-traits derived from LOC_Os03g01360 . ( B ) Correlations between s-traits and e-traits derived from LOC_Os07g01240 . ( C ) Correlations between s-traits and e-traits originating from the same genes excluding the above two loci . ( D ) Correlations between s-traits and different e-traits derived from LOC_Os03g01360 . DOI: http://dx . doi . org/10 . 7554/eLife . 03913 . 01610 . 7554/eLife . 03913 . 017Figure 3—figure supplement 1 . Processing of mRNA sequencing data from raw sequencing reads to genome-mapped reads . The read number of all IMF2s in each step is shown as box plot . DOI: http://dx . doi . org/10 . 7554/eLife . 03913 . 01710 . 7554/eLife . 03913 . 018Figure 3—figure supplement 2 . The distribution of mRNA sequencing reads in different genomic regions . DOI: http://dx . doi . org/10 . 7554/eLife . 03913 . 018 In-depth analysis of s-traits from gene LOC_Os03g01360 revealed that the expression values of s-traits from the first part of LOC_Os03g01360 were positively correlated with two transcripts , LOC_Os03g01360 . 1 and LOC_Os03g01360 . 4 , and negatively correlated with the other LOC_Os03g01360 . 2 ( Figure 3D ) . On the other hand , the expression values of most s-traits from the third part of LOC_Os03g01360 were negatively correlated with LOC_Os03g01360 . 1 and LOC_Os03g01360 . 4 but positively correlated with LOC_Os03g01360 . 2 ( Figure 3D ) . Whereas both negative and positive expression correlations were found between s-traits from the second part of LOC_Os03g01360 and these three transcripts ( Figure 3D ) . Thus , although strong expression correlations , either positive or negative , were observed between specific sRNAs and their mother gene , the expression levels of most sRNAs were not correlated with their mother genes , suggesting independent transcription of sRNAs and mRNAs . QTL analysis was performed for the 165 , 797 s-traits based on the ultrahigh-density SNP map , which contained 1556 recombination events and was composed of 1568 bins with an average size of 238 kb ( ranging from 6 kb to 7947 kb ) ( Figure 4—figure supplement 1 , Figure 4—figure supplement 2 , Supplementary file 6 in Dryad [Wang et al . , 2015] , ‘Materials and methods’ ) ( Xie et al . , 2010; Yu et al . , 2011 ) , using composite interval mapping ( CIM ) in R/qtl with 1000 permutations ( Haley and Knott , 1992; Broman and Speed , 2002; Manichaikul et al . , 2009 ) . With a false discovery rate ( FDR ) set at 5% , a total of 70 , 858 sQTLs were recovered for 66 , 649 s-traits ( Supplementary file 7 in Dryad [Wang et al . , 2015] ) . These sQTLs were classified as local-sQTLs and distant-sQTLs , which were also referred to as cis- and trans-QTLs , respectively , in previous studies ( Wang et al . , 2010 , 2014 ) , according to their locations relative to the s-traits . A local-sQTL indicated the existence of local functional polymorphism ( s ) that could influence the abundance of the s-trait , and a distant-sQTL meant the s-trait expression variation in the population was controlled by regulatory element ( s ) distant from the s-trait precursor sequence . For defining a local-sQTL , we adopted a 1 . 5 LOD-drop support interval of the corresponding sQTL or no more than 250 kb from the closest marker in a recombination sparse region . Otherwise , it was regarded as a distant-sQTL ( Yvert et al . , 2003; Morley et al . , 2004; Keurentjes et al . , 2007 ) . Such classification resulted in 40 , 049 local-sQTLs ( shown in the diagonal of Figure 4 ) and 30 , 809 distant-sQTLs ( off-diagonal of Figure 4 ) ( Supplementary file 7 in Dryad [Wang et al . , 2015] ) . Expression variations of s-traits explained by local-sQTLs were substantially higher than that by distant-sQTLs ( Figure 4—figure supplement 3 ) . The LOD values for 80 . 4% of the local-sQTLs and 34 . 5% of distant-sQTLs were larger than 10 . 10 . 7554/eLife . 03913 . 019Figure 4 . sQTLs for the 66 , 649 s-traits . The color key shows the LOD value . X-axis , the physical position of sQTLs along the genome of 12 chromosomes . Y-axis , the physical position of s-traits . QTLs with LOD value <5 are not included in the presentation . DOI: http://dx . doi . org/10 . 7554/eLife . 03913 . 01910 . 7554/eLife . 03913 . 020Figure 4—figure supplement 1 . The distribution of the sizes of 1568 bins of the genetic map . DOI: http://dx . doi . org/10 . 7554/eLife . 03913 . 02010 . 7554/eLife . 03913 . 021Figure 4—figure supplement 2 . The genotypes of 98 IMF2s across the 1568 bins . Each row represents an IMF2 , while each column represents a bin . Different color indicates different genotype . The 12 chromosomes were separated by black vertical lines . DOI: http://dx . doi . org/10 . 7554/eLife . 03913 . 02110 . 7554/eLife . 03913 . 022Figure 4—figure supplement 3 . The LOD values and expression variations explained by sQTLs ( R2 ) for local- and distant-sQTLs . ( A ) The LOD values for local-sQTLs . ( B ) The LOD values for distant-sQTLs . ( C ) Expression variation explained by local-sQTL . ( D ) Expression variation explained by distant-sQTL . DOI: http://dx . doi . org/10 . 7554/eLife . 03913 . 02210 . 7554/eLife . 03913 . 023Figure 4—figure supplement 4 . scQTLs for the 20 , 049 sc-traits . The color key shows the LOD value . X-axis shows the physical position of scQTLs along the genome . Y-axis shows the physical position of sc-traits . QTLs with LOD value <5 are not included in the presentation . DOI: http://dx . doi . org/10 . 7554/eLife . 03913 . 02310 . 7554/eLife . 03913 . 024Figure 4—figure supplement 5 . The LOD values and expression variations explained by scQTLs ( R2 ) for local- and distant-scQTLs . ( A ) The LOD values for local-scQTLs . ( B ) The LOD values for distant-scQTLs . ( C ) Expression variation explained by local-scQTL . ( D ) Expression variation explained by distant-scQTL . DOI: http://dx . doi . org/10 . 7554/eLife . 03913 . 02410 . 7554/eLife . 03913 . 025Figure 4—figure supplement 6 . eQTLs for the 6123 e-traits . The color key shows the LOD value . X-axis shows the physical position of eQTLs along the genome . Y-axis shows the physical position of e-traits . QTLs with LOD value <5 are not included in the presentation . DOI: http://dx . doi . org/10 . 7554/eLife . 03913 . 02510 . 7554/eLife . 03913 . 026Figure 4—figure supplement 7 . The LOD values and expression variations explained by eQTLs ( R2 ) for local- and distant-eQTLs . ( A ) The LOD values for local-eQTLs . ( B ) The LOD values for distant-eQTLs . ( C ) Expression variation explained by local-eQTL . ( D ) Expression variation explained by distant-eQTL . DOI: http://dx . doi . org/10 . 7554/eLife . 03913 . 02610 . 7554/eLife . 03913 . 027Figure 4—figure supplement 8 . Examples of s-traits expressed in only one of the two parents ( Zhenshan 97 ) . The expression value ( reads per millions [RPM] ) of each s-trait in the IMF2 population is shown as box plot . IMF2s with different genotypes as determined by the genotype of the bin harboring these s-traits are indicated by different colors . DOI: http://dx . doi . org/10 . 7554/eLife . 03913 . 027 Next , we performed QTL analysis for 50 , 139 sc-traits based on the same genetic map as in sQTL analysis using CIM in R/qtl with 1000 permutations . A total of 22 , 263 scQTLs including 11 , 476 local-scQTLs and 10 , 787 distant-sQTLs were recovered for 20 , 049 sc-traits utilizing the same definitions of local- and distant-QTLs ( Figure 4—figure supplement 4 , Supplementary file 7 in Dryad [Wang et al . , 2015] ) . Again , the LOD values of local-scQTLs and expression variations explained by local-scQTLs were much higher than distant-scQTLs ( Figure 4—figure supplement 5 ) . We also performed QTL analysis for the 24 , 987 e-traits based on the same genetic map , using the same algorithm and parameters as in the QTL analysis of s-traits and sc-traits . With an FDR set at 5% , a total of 6423 eQTLs were recovered for 6123 e-traits ( Supplementary file 7 in Dryad [Wang et al . , 2015] ) . These eQTLs were classified as local-eQTLs and distant-eQTLs according to the same rule applied in the sQTL analysis . As a result , 2964 local-eQTLs ( shown in the diagonal of Figure 4—figure supplement 6 ) and 3459 distant-eQTLs ( off-diagonal of Figure 4—figure supplement 6 ) were detected ( Supplementary file 7 in Dryad [Wang et al . , 2015] ) . Expression variations of mRNAs explained by local-eQTLs were significantly higher than that by distant-eQTLs ( Figure 4—figure supplement 7 ) . Local-QTLs regulating the expression of large numbers of traits were observed for s-traits and sc-traits but not for e-traits ( Figure 4 , Figure 4—figure supplement 4 and Figure 4—figure supplement 6 ) , probably due to the consecutive transcription of sRNAs and sRNA clusters , which was distinct from the transcription of mRNAs . In the extreme case , a total of 1660 s-traits were regulated by a single local-sQTL ( Bin358 ) on chromosome 3 ( Supplementary file 8 in Dryad [Wang et al . , 2015] ) . Each of another five local-sQTLs ( Bin359 , Bin1183 , Bin969 , Bin286 , and Bin166 ) explained the expression variations of more than 200 s-traits . The local-scQTL ( Bin1183 ) with the largest effect regulated the expression of 93 sc-traits . Each of another 29 local-scQTLs was responsible for the expression variations of more than 30 sc-traits . In comparison , a total of 24 e-traits were regulated by Bin3 , which is the local-eQTL explaining the expression variations of the largest number of e-traits . Distant-QTLs regulating many traits across all chromosomes were very common for s-traits and sc-traits but relatively rare for e-traits . A total of 15 bins were identified as distant-sQTLs , each of which regulated more than 200 s-traits; in one case , 514 s-traits were regulated by a single distant-sQTL ( Bin1209 ) . We observed three distant-scQTLs ( Bin634 , Bin635 , and Bin636 ) , each of which regulated more than 100 sc-traits , and another 21 distant-scQTLs , each of which explained the expression variations of more than 30 sc-traits . Whereas in case of mRNAs , only 9 bins were each responsible for the expression of more than 30 e-traits . We further inspected the QTL results of 22 , 142 s-traits expressed ( RPM >0 ) in only one of the two parents . In all , 11 , 194 and 10 , 948 s-traits were expressed only in the genome of Zhenshan 97 and Minghui 63 , respectively . A total of 15 , 037 QTLs , 11 , 781 of which were local-sQTLs , were obtained for 14 , 139 of these s-traits . This proportion was significantly higher than the whole genome average level ( Fisher's exact test , p value <2 . 2e-16 ) , implying that the expression variations of these s-traits were mostly controlled by local polymorphisms . Examples of s-traits expressed in only one of the two parents were shown in Figure 4—figure supplement 8 . We next identified bins enriched with traits or QTLs based on the relative densities of the distributions in the genome ( ‘Materials and methods’ ) . 40 bins were identified as hotspots for s-traits ( Supplementary file 8 in Dryad [Wang et al . , 2015] , Figure 5 ) . A total of 132 and 129 bins were identified as local-sQTL and distant-sQTL hotspots , respectively ( Supplementary file 8 in Dryad [Wang et al . , 2015] , Figure 5 ) . 21 bins were both s-trait and local-sQTL hotspots , while 64 bins were both local-sQTL and distant-sQTL hotspots ( Supplementary file 8 in Dryad [Wang et al . , 2015] , Figure 5 ) ; 123 were local-scQTL hotspots and 149 distant-scQTL hotspots ( Supplementary file 8 in Dryad [Wang et al . , 2015] ) . In all , 76 bins were local-QTL hotspots and 80 bins were distant-QTL hotspots for both s-traits and sc-traits . On the other hand , 116 and 126 bins were identified as local- and distant-eQTL hotspots . The identified hotspots of sQTLs and scQTLs were quite similar but distinct from that of eQTLs ( Figure 5 ) . Totally , 33 bins were local-QTL hotspots and 36 bins were distant-QTL hotspots for both s-traits and e-traits . Consecutive distant-QTL hotspots in adjacent genomic regions could explain the expression variations of a large number of s-traits , sc-traits , or e-traits ( Supplementary file 8 in Dryad [Wang et al . , 2015] ) . Four consecutive distant-QTL hotspots on chromosome 4 , 5 , 6 , and 9 were identified for both s-traits and sc-traits ( Figure 5A , B ) . The genomic positions of consecutive distant-eQTLs hotspots were quite distinct from that of sQTL or scQTLs ( Figure 5 ) . A cluster of the consecutive distant-eQTL hotspots ( chr04: 1 . 713 Mb–2 . 495 Mb ) was located close to one end of chromosome 4 , while another cluster of consecutive distant-QTL hotspots ( chr04: 21 . 997 Mb–23 . 342 Mb ) for both s-traits and sc-traits was about 10 Mb distant from the other end of chromosome 4 . A cluster of consecutive distant-eQTL hotspots ( chr07: 28 . 058 Mb–28 . 557 Mb ) was located close to one end of chromosome 7 , while no apparent consecutive distant-QTL hotspots for s-traits and sc-traits were on chromosome 7 . A total of 12 bins ( Bin1199–Bin1210 ) on chromosome 9 were identified as consecutive distant-QTL hotspots for both s-traits and sc-traits but not for mRNA e-traits . Only the consecutive distant-QTL hotspot on chromosome 6 ( chr06: 2 . 847 Mb–4 . 023 Mb ) was responsible for the expression variations of s-traits , sc-traits , and e-traits . 10 . 7554/eLife . 03913 . 028Figure 5 . The distribution of traits and QTLs in the 1568 bins . ( A ) s-traits and sQTLs . ( B ) sc-traits and scQTLs . ( C ) e-traits and e-QTLs . The 1568 bins are arranged from left to right according to their genomic position . The width of bar represents the size of the bin . The chromosome identifiers are labeled below each plot . Bins in red color were trait or QTL hotspots . Locations for a number of interesting regions like sRNA biogenesis genes , Bin358 ( LOC_Os03g01360 ) and Bin 969 ( LOC_Os07g01240 ) , are also indicated , in addition to the hotspots . DOI: http://dx . doi . org/10 . 7554/eLife . 03913 . 02810 . 7554/eLife . 03913 . 029Figure 5—figure supplement 1 . DNA sequence polymorphisms of sRNA biogenesis genes between Zhenshan 97 and Minghui 63 . ( A ) SNPs in sRNA biogenesis genes between Zhenshan 97 and Minghui 63 . ( B ) Indels in sRNA biogenesis genes between Zhenshan 97 and Minghui 63 . DOI: http://dx . doi . org/10 . 7554/eLife . 03913 . 02910 . 7554/eLife . 03913 . 030Figure 5—figure supplement 2 . Expression correlations between sRNA biogenesis genes in distant-sQTL hotspots and s-traits regulated by these distant-sQTLs . ( A ) Expression correlations between s-traits regulated by Bin632–Bin637 and the e-traits in these bins . ( B ) Expression correlations between s-traits regulated by Bin632–Bin637 and the e-traits of OsRDR2 . ( C ) Expression correlations between s-traits regulated by Bin1199–Bin1210 and the e-traits in these bins . The e-traits of OsDCL2b and others e-traits are represented by different colors . ( D ) Expression correlations between s-traits regulated by Bin632–Bin637 and the e-traits of OsDCL2a . DOI: http://dx . doi . org/10 . 7554/eLife . 03913 . 030 The s-trait hotspot with the highest s-trait density was Bin969 ( chr07: 0 . 121 Mb–0 . 155 Mb ) , holding 1025 s-traits , of which 66 . 6% obtained at least one sQTL ( Figure 5A , Supplementary file 7 in Dryad [Wang et al . , 2015] ) . Bin969 was also a local-sQTL hotspot with the third highest local-sQTL density . The local-sQTL hotspot with the highest local-sQTL density was Bin358 ( chr03: 0 . 266 Mb–0 . 368 Mb ) , which held 2338 s-traits , and an overwhelming majority of these s-traits ( 2245 ) had at least one sQTL ( Figure 5A , Supplementary file 8 in Dryad [Wang et al . , 2015] ) . sRNAs are usually generated by activities of Dicer-like proteins , Argonautes , and RNA-dependent RNA polymerases ( RDRs ) ( Kapoor et al . , 2008; Axtell , 2013 ) . In rice , there are 43 transcripts of 32 genes for sRNA biogenesis including eight encoding Dicer-like proteins ( OsDCLs ) , 19 Argonautes ( OsAGOs ) , and 5 OsRDRs ( Supplementary file 9 in Dryad [Wang et al . , 2015] ) ( Kapoor et al . , 2008 ) . In this study , 26 transcripts for 20 genes were detected ( FPKM ≥1 ) in at least one of the parents . Six transcripts of five genes ( OsDCL2a , OsDCL2b , OsAGO1d , OsAGO13 , and OsAGO18 ) were significantly differentially expressed between Zhenshan 97 and Minghui 63 ( Supplementary file 9 in Dryad [Wang et al . , 2015] ) . Comparison of DNA sequences between Zhenshan 97 and Minghui 63 revealed that 23 of the 32 sRNA biogenesis genes had polymorphisms of either SNPs or indels ( Figure 5—figure supplement 1 , Supplementary file 9 in Dryad [Wang et al . , 2015] ) . We compared the genomic positions of these sRNA biogenesis genes and distant-sQTL hotspots . OsRDR2 ( in Bin636 and Bin637 ) was located in the consecutive distant-sQTL hotspots ( Bin632–Bin637 ) on chromosome 4 regulating 1416 s-traits . OsDCL2b ( in Bin1205 ) was located in the consecutive distant-sQTL hotspots ( Bin1199–Bin1210 ) on chromosome 9 ( Figure 5A , Supplementary file 10 in Dryad [Wang et al . , 2015] ) , which regulated the expression variation of 3759 s-traits . OsDCL2a was located in Bin448 , which regulated the expression of 213 s-traits , although this bin was not identified as a distant-sQTL hotspot due to its large size ( 1 . 254 Mb ) . However , four neighboring bins ( Bin449–Bin451 , Bin453 ) of Bin448 were identified as distant-sQTL hotspots ( Supplementary file 8 in Dryad [Wang et al . , 2015] ) . Eight adjacent bins ( Bin734–Bin741 ) on chromosome 5 were found to harbor 364 distant-sQTLs , and OsDCL1c was in Bin738 , which contained 33 distant-sQTLs ( Supplementary file 8 in Dryad [Wang et al . , 2015] ) . The expression correlations between all the e-traits in the distant-sQTL hotspots ( Bin632–Bin637 ) on chromosome 4 and all the s-traits whose expression variations were controlled by QTLs in these bins were calculated . Most of these correlations including those between the e-traits of OsRDR2 and all the s-traits were weak ( Figure 5—figure supplement 2A , B ) . For s-traits regulated by QTLs in the consecutive distant-sQTL hotspots ( Bin1199–Bin1210 ) on chromosome 9 ( Figure 5A ) , most of the correlations between the s-traits and the e-traits in these bins were weak , while the correlations between s-traits and the e-traits of OsDCL2b were relatively strong ( Figure 5—figure supplement 2C ) . About 79 . 7% , 17 . 4% , and 1 . 2% of the s-traits whose expression levels were regulated by the consecutive distant-sQTL hotspots harboring OsRDR2 on chromosome 4 were 24 nt , 23 nt , and 22 nt , respectively . On the other hand , 51 . 0% of the s-traits regulated by the consecutive distant-sQTL hotspots harboring OsDCL2b on chromosome 9 were 22 nt while 35 . 8% were 24 nt . For s-traits regulated by distant-QTLs in Bin448–Bin453 ( containing OsDCL2a ) , 40 . 8% of them were 21 nt and 24 . 4% were 22 nt and 24 . 1% were 24 nt . Five local-eQTLs and six distant-eQTLs for nine transcripts of these sRNA biogenesis genes were identified ( Supplementary file 10 in Dryad [Wang et al . , 2015] ) . We also compared the genomic positions of these eQTLs with the distant-sQTL hotspots . A distant-eQTL explaining the expression variation of OsDCL2a was located in Bin1207 , which was in the consecutive distant-sQTL hotspots ( Bin1199–Bin1210 ) on chromosome 9 ( Supplementary file 10 in Dryad [Wang et al . , 2015] ) . The expression correlations between s-traits regulated by QTLs in these bins and the e-traits of OsDCL2a were mostly strong ( Figure 5—figure supplement 2D ) . The functional annotations and DNA polymorphisms between the two parents for all the genes in the distant-sQTL hotspots harboring OsRDR2 , OsDCL2b , and OsDCL2a were listed in Supplementary file 11 in Dryad ( Wang et al . , 2015 ) . The distant-sQTL hotspots on chromosome 5 ( Bin770–Bin777 ) and 6 ( Bin850–Bin860 ) were also surveyed . Approximately 98 . 4% and 90 . 5% of all the s-traits regulated by the distant-sQTL hotspots on chromosome 5 and chromosome 6 were 24 nt . The functional annotations of genes in the distant-sQTL hotspots on chromosome 5 were investigated , and no apparent relationship to the biogenesis of sRNA was found , implicating the existence of unknown mechanisms of sRNA biogenesis and/or regulation ( Supplementary file 12 in Dryad [Wang et al . , 2015] ) . A gene annotated as RNA Pol II subunit Rpb7 was found in the distant-sQTL hotspots on chromosome 6 ( Supplementary file 12 in Dryad [Wang et al . , 2015] ) . Study in yeast reveals that Rpb7 has a specific role in pre-siRNA transcription ( Djupedal et al . , 2005 ) . To test whether the expression of s-traits originating from the same mother gene was regulated by the same genetic mechanism , the genomic positions of QTLs for s-traits originated from the same mother gene were compared with each other . S-traits originated from LOC_Os03g01360 and LOC_Os07g01240 were not included in this calculation . We calculated correlations for the following comparisons: 97 , 040 pairs of s-traits that were both controlled by local-sQTLs ( Figure 6A ) , 36 , 123 pairs of s-traits that were both controlled by distant-sQTLs ( Figure 6B ) , and 51 , 081 pairs of s-traits in which one was controlled by local-sQTL and the other one by distant-sQTL ( Figure 6C ) . For a pair of s-traits originated from the same gene controlled by local-sQTL , the correlation coefficients between them were mostly strong ( Figure 6A ) . The correlation coefficients between s-traits controlled by the same or nearby distant-sQTLs were also strong ( Figure 6B ) , while the correlation coefficients between s-traits controlled by distant-sQTLs distant from each other were relatively weak ( Figure 6B ) , which represented the majority ( 75 . 0% ) of the s-traits pairs . Most of the correlations between the s-traits of local- and distant-regulation were also quite high ( Figure 6C ) . For comparison , we also calculated correlations for all pairs of s-traits from the entire genome not necessarily the same genes that were regulated by the same sQTLs; their expression correlations were also mostly strong , either positive or negative ( Figure 6D ) , indicating that the strong correlations between the s-traits had common genetic basis . 10 . 7554/eLife . 03913 . 031Figure 6 . Genetic co-regulation between s-traits originating from the same mother genes . ( A ) Expression correlations between pairs of s-traits originating from the same mother genes and both regulated by local-sQTLs . ( B ) Expression correlations between pairs of s-traits originating from the same mother genes and both regulated by distant-sQTLs . Green: distance between the two distant-sQTLs was within two 2 Mb; red: distance between the distant-sQTLs was larger than 2 Mb . ( C ) Expression correlations between pairs of s-traits from the same mother genes with one regulated by distant-sQTL and another one regulated by local-sQTL . ( D ) Expression correlations between pairs of s-traits from the whole genome regulated by distant-QTLs located in ≤2 Mb regions . DOI: http://dx . doi . org/10 . 7554/eLife . 03913 . 031 To infer the commonality of the genetic controls of the expression variations for sRNAs and their mother genes , we compared the map positions of the QTLs for sRNAs and their mother genes . Again s-traits and e-traits originating from LOC_Os03g01360 and LOC_Os07g01240 were not included in this calculation . We calculated correlations for the following comparisons: 4018 s-traits and 1084 corresponding mother genes that were both controlled by local-QTLs ( Figure 7A ) , 1721 s-traits and 695 mother genes both regulated by distant-QTLs ( Figure 7B ) , 1840 s-traits controlled by local-sQTLs and 544 corresponding mother genes regulated by distant-eQTLs ( Figure 7C ) , and 1110 s-traits and 582 mother genes controlled by distant-sQTLs and local-eQTLs , respectively ( Figure 7D ) . The expression correlations between s-traits and their mother genes that were both controlled by local-QTLs or the same and/or nearby distant-QTLs were mostly strong ( Figure 7A , B ) . On the other hand , the correlations between s-traits and their mother genes that were regulated by different/distant-QTLs were relatively weak ( Figure 7B–D ) . For all the pairs of e-traits and s-traits regulated by the same QTL position with the s-traits not necessarily from the genomic region of the e-traits , their expression correlations were mostly strong ( Figure 7E ) , again indicating a common genetic basis . 10 . 7554/eLife . 03913 . 032Figure 7 . Genetic co-regulation between s-traits and their mother genes . ( A ) Expression correlations between s-traits and their mother genes that were both controlled by local-QTLs . ( B ) Expression correlations between s-traits regulated by distant-sQTLs and their mother genes also controlled by distant-eQTLs . Green , the distance between distant-QTLs for s-traits and the mother genes is within 2 Mb . Red , the distance between distant-QTLs is larger than 2 Mb . ( C ) Expression correlations between s-traits regulated by local-sQTLs and their mother genes controlled by distant-eQTLs . ( D ) Expression correlations between s-traits regulated by distant-sQTLs and their mother genes controlled by local-eQTLs . ( E ) Expression correlations between s-traits and e-traits from the whole genome regulated by distant-QTLs located in ≤2 Mb . DOI: http://dx . doi . org/10 . 7554/eLife . 03913 . 032 There were three genotypes at each polymorphic locus in the IMF2 population: AA , Aa , and aa , where A represents the allele from one parent and a is the allele from the other parent . This allowed estimating of additive and dominant genetic effects for each of the sQTLs . Additive effect for sQTL was half of the sRNA expression difference between the two homozygotes , and dominant effect was the difference between the heterozygote and the average of the two homozygotes . Of all the s-traits analyzed , s-traits regulated by 35 , 235 ( 49 . 7% ) of the sQTLs exhibited higher expression levels in the Zhenshan 97 genotype than Minghui 63 genotype ( Supplementary file 13 in Dryad [Wang et al . , 2015] ) . We identified sQTLs with significant dominance effects using h-test ( 1000 permutations , p < 0 . 05 ) ( Huang et al . , 2006; Zhou et al . , 2012 ) . The analysis detected 23 , 018 sQTLs ( 32 . 5% ) with significant dominant effects , of which the majority ( 71 . 8% ) showed negative dominance such that the levels of the sRNAs were lower in heterozygotes than the mid-parent values ( Supplementary file 13 in Dryad [Wang et al . , 2015] ) . Moreover , 3842 ( 5 . 4% ) sQTLs exhibited negative overdominance , in which sRNA levels of the heterozygotes were lower than both homozygotes . Approximately 92 . 4% of the sQTLs showing significant overdominance effects , either positive or negative , were distant-sQTLs . The genetic effects of scQTLs were quite similar to that of sQTLs . Of all the 6913 scQTLs exhibiting significant dominant effects , 5140 ( 74 . 3% ) showed negative dominance ( Supplementary file 13 in Dryad [Wang et al . , 2015] ) . A total of 1167 scQTLs ( 5 . 2% ) exhibited negative overdominance , 89 . 3% of which were distant-scQTLs . We took a closer look at the two loci that contributed extremely high numbers of sRNAs . Approximately 97 . 4% ( 2278 ) of the s-traits produced by Bin358 were encoded by the 2-kb upstream and genic region of the locus LOC_Os03g01360 , where the sRNA sequences showed extensive and consecutive overlapping ( Figure 8A ) . LOC_Os03g01360 did not have homologs in Arabidopsis but had a homolog in maize with unknown function ( http://rice . plantbiology . msu . edu/cgi-bin/ORF_infopage . cgi ? orf=LOC_Os03g01360 ) . Three isoforms of this gene were significantly differentially expressed as measured in FPKM between Zhenshan 97 and Minghui 63 ( Figure 8—figure supplement 1A ) . The sRNAs of Zhenshan 97 were mainly derived from the second and the third part of LOC_Os03g01360 , while sRNAs of Minghui 63 originated from all the three parts with the expression level from the third part lower than that of sRNAs of Zhenshan 97 ( Figure 8A ) . The expression levels of LOC_Os03g01360 . 1 and LOC_Os03g01360 . 4 in the genome of Minghui 63 were higher than Zhenshan 97 ( Figure 8—figure supplement 1A ) . On the contrary , LOC_Os03g01360 . 2 was higher expressed in Zhenshan 97 than in Minghui 63 . The expression of all the three transcripts in the hybrid was down-regulated compared with the mid-parent value ( Figure 8—figure supplement 1A ) . We checked the cytosine methylation levels in this region , and the genomic DNA showed high cytosine methylation , especially from the start to the third exon ( Figure 8A , ‘Materials and methods’ ) . This high methylation might be due to the enrichment of sRNAs resulting in RNA-directed DNA methylation . In the differentially expressed region of sRNAs between Zhenshan 97 and Minghui 63 , such as the third exon according to the annotated gene model , Minghui 63 had lower methylation and higher mRNA transcript level than Zhenshan 97 , suggesting a negative correlation between the transcript level and methylation . The expression variation of three isoforms of LOC_Os03g01360 in the IMF2 population was controlled by local-eQTLs ( Supplementary file 7 in Dryad [Wang et al . , 2015] ) . At least one sQTL each was detected for 2200 of the 2278 s-traits located in the gene LOC_Os03g01360 region , of which 2181 ( 84 . 5% ) were local-sQTLs ( Supplementary file 7 in Dryad [Wang et al . , 2015] ) . The expression variation of different transcripts of LOC_Os03g01360 probably corresponded to sRNAs from specific parts of the genic region and is regulated by the same genetic factor . The majority of distant-sQTLs regulating the expression of s-traits originating from LOC_Os03g01360 was for 21 and 22 nt s-traits and was attributed to Bin442–Bin453 on chromosome 3 . 10 . 7554/eLife . 03913 . 033Figure 8 . The expression levels of sRNA , mRNA , and the DNA methylation levels of the hybrid and its parents in the region of LOC_Os03g01360 and LOC_Os07g01240 . ( A ) The structure of LOC_Os03g01360 is shown in the top panel , with black color representing exons and red color indicating introns and UTRs . The number of mRNA reads is shown in blue color . The sRNAs expression level is shown in orange . The DNA methylation level is displayed in the bottom three panels . ZS , Zhenshan 97; MH , Minghui 63; HY , Hybrid . ( B ) The structure of LOC_Os07g01240 is shown in the top panel , with black color representing exons and green color indicating introns and UTRs . The meaning of different colors in the other panels is the same as in ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03913 . 03310 . 7554/eLife . 03913 . 034Figure 8—figure supplement 1 . The expression level ( fragments per kb per million [FPKM] ) of the three transcripts of LOC_Os03g01360 ( A ) and the two transcripts of LOC_Os07g01240 ( B ) in Zhenshan 97 , Minghui 63 , and the hybrid . DOI: http://dx . doi . org/10 . 7554/eLife . 03913 . 034 The sRNAs in Bin969 were almost entirely derived from the intron of LOC_Os07g01240 , of which 617 ( 61 . 2% ) were 21-nt and 22-nt sRNAs and 280 ( 27 . 8% ) were 24-nt sRNAs ( Figure 8B ) . LOC_Os07g01240 had a homolog in Arabidopsis with unknown function and in Maize annotated as uncharacterized GPI-anchored protein ( http://rice . plantbiology . msu . edu/cgi-bin/ORF_infopage . cgi ? orf=LOC_Os07g01240 ) and was reported to modulate rice leaf rolling by regulating the formation of bulliform cells ( Xiang et al . , 2012 ) . In all , 677 s-traits mapped 841 sQTLs , of which 514 were local-sQTLs ( Supplementary file 7 in Dryad [Wang et al . , 2015] ) . However , no significant expression difference of the two isoforms of LOC_Os07g01240 was detected between the parents ( Figure 8—figure supplement 1B ) , and thus no eQTL for the e-traits was detected . The majority of the 327 distant-sQTLs was for 21 and 22 nt s-traits and was again attributed to Bin442–Bin453 on chromosome 3 . Some of the distant-sQTLs for 24 nt s-traits were attributed to Bin770–Bin775 on chromosome 5 . The expression correlation between e-traits of LOC_Os07g01240 and s-traits originating from LOC_Os07g01240 was weak ( Figure 3B ) . This result was similar to that of Tong et al . ( 2013 ) who showed that intronic mRNAs had a distinct expression pattern from their host genes , suggesting that intronic sRNAs might have genetic regulatory mechanisms independent of their mother genes . The variation of sRNA abundance in a population depends on a range of factors: the levels of transcription of the precursors that is influenced by local elements mostly residing in the promoter regions of the sRNA itself; the mother gene or long non-coding RNAs where the sRNA is produced; polymorphisms in sequence influencing their biogenesis; and variation in distant regulatory factors located distantly from where the sRNA is generated . Our results indicated that almost one third of the detected local-sQTLs exhibited much larger effects than distant-sQTLs , consistent with the studies for eQTL of mRNA expression ( Kliebenstein , 2009 ) , suggesting that local elements are a major regulatory factor for sRNA variation . Local-QTLs regulating the expression of large numbers of traits were observed for s-traits but not for e-traits , resulting from the enrichment of s-traits in the region produced by the extensive and consecutive overlapping transcripts from the mother genes . Hotspots were found for both distant-QTLs of mRNA e-traits and sRNA s-traits , both of which could influence the expression variations of a large number of traits . Our results showed that quantitative variation for a large portion of sRNAs transcribed from genic regions did not share the genetic control of the expression with the corresponding mother genes , although by common sense those sRNAs have to be transcribed along and regulated by the promoters of the mother genes . Similar results were reported in previous studies of miRNAs and their host genes , which were regulated or processed independently from their respective regulatory elements ( Siddle et al . , 2014 ) . The results of Monteys et al . ( 2010 ) and Ozsolak et al . ( 2008 ) suggested that the transcription of sRNAs may also be regulated by their own promoters , independent of the mother genes . The widespread occurrence of such independent genetic controls between the sRNAs and the mother genes suggested that independent transcription of the sRNAs might be a common phenomenon , rather than only special circumstances . Moreover , the large number of distant-sQTLs that were not collocated with the distant-eQTLs for the corresponding mother genes supported the independent regulatory basis of the transcription of the sRNAs . Our results showed that some of the sRNA biogenesis genes such as DCLs , AGOs , and RDRs were probably responsible for the quantitative variation of a large number of sRNAs . OsRDR2 was found in a region of consecutive distant-sQTL hotspots explaining the expression variation of many sRNAs , most of which were 24 nt . This was also a region of consecutive distant-scQTL hotspots with the highest number of distant-scQTLs . OsDCL2b was also found in consecutive distant-sQTL hotspots regulating the expression of sRNAs , most of which were 22 and 24 nt . Although OsDCL2a and OsDCL1c were not in distant-sQTL hotspots due to lack of local recombination in that region , they were also in bins regulating the expression of a large number of s-traits . This is in accordance with the reports that RDRs function on the upstream of DCLs in the process of sRNA biogenesis ( Chapman and Carrington , 2007 ) . Studies in sRNA pathways in Arabidopsis showed that DCL2 is responsible for the synthesis of 22 nt or 24 nt siRNAs , while RDR2 functions in the production of endogenous 24 nt siRNAs and the conversion of ssRNA template into dsRNAs that serve as substrates for DCLs ( Arikit et al . , 2013 ) , which is in good agreement with the sQTLs found in this study . However , it should also be noted that these genes were associated with s-QTLs for only a small portions of the s-traits , while the quantitative variation of the sRNA abundance for majority of the s-traits was independent of the sRNA biogenesis genes . One of the most interesting finding perhaps concerns the sRNAs from the two loci , LOC_Os03g01360 and LOC_Os07g01240 , and their regulation patterns . Although both loci produced thousands of sRNAs , which were the most numerous in the genome , they showed sharp contrast in where the sRNAs were generated and the regulatory mechanisms with which the sRNAs were produced . sRNAs from LOC_Os03g01360 mostly originated from the 2-kb upstream and genic regions and were tightly co-regulated with each other and also with the mother gene . By contrast , sRNAs from LOC_Os07g01240 were mostly produced in the intronic region and loosely co-regulated with each other and not co-regulated with the mother gene; in fact , the s-trait variation and sQTLs were detected even without the expression variation of the mother gene . Since neither of the genes has been previously identified as related to the production and function of sRNAs , their roles in sRNA biogenesis warrant further investigation . Another noticeable finding is the widespread negative dominance of sRNA levels detected in the IMF2 population such that heterozygotes had lower level of the sRNAs than the means of the two homozygotes ( negative partial dominance ) or the lower homozygote ( negative overdominance ) . Such negative dominance was also observed in comparative sRNA profiling of hybrids relative to the parents in crosses of rice and Arabidopsis ( Groszmann et al . , 2011b , 2013; He et al . , 2013 ) , revealing a predominant-negative regulation of siRNA expression in hybrids ( He et al . , 2010; Chodavarapu et al . , 2012 ) . Our genetic analysis revealed that the sQTLs showing negative dominance of the sRNA levels are mostly distant-sQTLs , indicating regulation by distant elements likely at the transcriptional level . Studies of transcript levels of genes ( mRNAs ) in the hybrid against the parents in the same rice cross also revealed more negative dominance than positive dominance , also indicating down-regulation in the hybrid relative to the parents ( Huang et al . , 2006 ) . How the down-regulation of these two classes of transcripts was related to each other and how the two classes of down-regulation are related to the hybrid performance present great challenges for future studies . The plant materials consisted of an IMF2 population of 98 hybrids produced by paired crosses of 196 RILs ( Supplementary file 14 in Dryad [Wang et al . , 2015] ) , the two parental lines Zhenshan 97 and Minghui 63 and their hybrid . The plants were grown under normal agricultural conditions at experimental farm of Huazhong Agricultural University in the rice-growing season ( May to September ) in Wuhan , China . A flag leaf at the day of full expansion from each of three random plants per replicate was harvested between 17:00 and 18:00 for library construction . Total RNA was isolated from the leaf tissue using TRIzol reagent ( Invitrogen , Waltham , Massachusetts ) , the process to convert total RNA into template suitable for high throughput DNA sequencing of mRNA-seq and sRNA-seq libraries using Sample Preparation Kit ( Illumina , San Diego , California ) followed the manufacturer's guide . Bisulfite sequencing ( WGBS ) libraries of the parents and the hybrid were made from genomic DNA isolated from the same leaf tissues as used for RNA-seq libraries . Sequencing was performed on Illumina HiSeq 2000 at BGI ( http://www . genomics . cn/index , Shenzhen , China ) ( 50 bp single end for mRNA-seq and 100 bp pair end for BS-seq ) . Poor-quality reads were removed using fastq_quality_filter in FASTX-Toolkit ( http://hannonlab . cshl . edu/fastx_toolkit/ ) with parameters q = 20 and p = 85 . Reads shorter than 18 nt or longer than 26 nt were excluded from further analysis . Reads mapped to rice tRNA , rRNA , snRNA , snoRNA obtained from fRNAdb ( http://www . ncrna . org/frnadb/ ) , NONCODE , GtRNAdb ( http://lowelab . ucsc . edu/GtRNAdb/Osati/ ) , and Rfam ( ftp://ftp . sanger . ac . uk/pub/databases/Rfam/11 . 0/ ) were also removed . There were 2 , 305 , 391 high-quality SNPs between Nipponbare ( Oryza sativa ssp . japonica ) and Zhenshan 97 and Minghui 63 ( Oryza sativa ssp . indica ) in the whole genome , of which 971 , 883 were the specific SNPs between Zhenshan 97 and Minghui 63 ( data from http://211 . 69 . 128 . 148/rice/ ) . We corrected the SNP sites of the Nipponbare reference genome ( http://rice . plantbiology . msu . edu , The MSU Rice Database release 7 . 0 ) according to the sequences of Zhenshan 97 and Minghui 63 to reconstruct ‘SNPs-replaced reference genomes’ for the two parents . For sRNA analysis , all filtered sRNA-seq reads from Zhenshan 97 libraries were specifically mapped to the SNPs-replaced Zhenshan 97 genome and reads from Minghui 63 libraries were mapped to SNPs-replaced Minghui 63 genome . The sRNAs from heterozygous materials of F1 and IMF2 were simultaneously mapped to the Zhenshan 97- and Minghui 63-replaced genomes . Bowtie ( Langmead et al . , 2009 ) was used to align short reads to each genome at unique genome location with no mismatch allowed . The reads from heterozygous materials could be divided into three groups: reads specifically mapped to Zhenshan 97 , reads only mapped to Minghui 63 genome , and reads mapped to both Zhenshan 97 and Minghui 63 at the same position . In the third class , the sequences of the sRNAs from a monomorphic site were indistinguishable between the parents and thus obtained one count in quantitation of the abundance , whereas the two counts of the sRNAs from a polymorphic site were put together . The expression level of an sRNA in a specific library was defined as the number of this sRNA divided by the total number ( in millions ) of genome-mapped sRNAs in this library , which was designated as ‘RPM’ . The R package DESeq ( Anders and Huber , 2010 ) was used to quantify sRNA cluster expression level . The number of sRNA reads in each sRNA cluster for each sample was calculated and integrated as a count table , with each line representing an sRNA cluster and each column representing a sample . Then , the effective library size for each sample was estimated using the ‘estimateSizeFactors’ function in the DESeq package . Each column of the count table was divided by the corresponding library size to get the normalized read count , which was regarded as the expression level of the sRNA cluster . The removal of poor-quality reads for mRNA-seq reads was done in the same way as sRNA analysis . The sequences from 104 libraries were mapped to the O . sativa ssp . japonica ( cv . Nipponbare ) version 7 reference genome using TopHat ( Trapnell et al . , 2009 ) with default parameters . Cufflinks ( Trapnell et al . , 2010 ) were utilized to estimate gene expression levels according to the Nipponbare version 7 reference annotation . For bisulfite sequencing , trimmomatic ( Lohse et al . , 2012 ) was used to remove low-quality reads . Bismark ( Krueger and Andrews , 2011 ) was performed to align bisulfite-treated reads to the SNP-replaced genomes , allowing no mismatch in the seed of 40 nucleotides and up to two good alignments . This bisulfate mapping tool aims to find unique alignment through running four alignment processes simultaneously ( Krueger and Andrews , 2011 ) . Then , the de-duplication tool provided by Bismark was applied to remove potential PCR duplicates . Methylation calls were extracted for every single cytosine analyzed depending on its context ( CpG , CHG , or CHH ) . The ultrahigh-density bin map constructed by genotyping the RILs with population sequencing ( Xie et al . , 2010; Yu et al . , 2011 ) was used . The 1568 bin genotypes of each cross in the IMF2 population were deduced from the parental genotypes ( Supplementary file 6 in Dryad [Wang et al . , 2015] , Figure 4—figure supplement 2 ) . CIM in R/qtl ( Haley and Knott , 1992; Broman and Speed , 2002; Manichaikul et al . , 2009 ) was employed to map QTLs with 1000 permutations . Additive and dominant effects were decided by ‘effectscan’ function in R/qtl . Variation explained by the QTL was determined using the linear QTL model as described by Yu et al . ( 2011 ) . The same genetic map and program parameters were used in the QTL analysis for s-traits , sc-traits , and e-traits . The density of s-trait and sQTL was defined as the number of s-traits and sQTLs in each bin divided by the bin size ( Mb ) , respectively . The density of sc-traits , scQTLs , e-traits , and eQTLs was calculated in the same way . Bins with s-trait density larger than three times of the whole genome average level were defined as s-trait hotspots , while bins with sQTL density higher than six times of the whole genome average level were designated as sQTL hotspots . The process of identification of e-traits and eQTL hotspots was the same as that of s-traits and sQTLs . The definition of scQTL hotspots was identical to sQTL hotspots , while sc-traits hotspots were defined as bins with sc-traits density higher than the twice of the whole genome average .

Genes within the DNA of a plant or animal contain instructions to make molecules called RNAs . Some RNA molecules can be decoded to make proteins , whereas others have different roles . A single gene often contains the instructions to make both protein-coding RNAs and non-coding RNAs . Molecules called small RNAs ( or sRNAs ) do not code for proteins . Instead , sRNAs can control protein-coding RNA molecules or chemically alter the DNA itself; this allows them to perform many different roles in living organisms . In plants , for example , these molecules affect how the plant grows , the shapes and structures it forms , and how likely it is to survive challenges such as drought and diseases . Often different plants of the same species have different amounts of sRNAs , but the reasons for this remain unclear . Now , Wang , Yao et al . have made use of a technique called ‘expression quantitative locus’ analysis to look at how sRNAs in rice plants are controlled by additional information encoded within DNA . The analysis identified over 53 million sRNA molecules from a population of rice plants . Many of these sRNAs varied in their abundance between different plants within the population . Wang , Yao et al . also found many thousands of individual instructions within the DNA of the rice that can either increase or reduce the abundance of their associated sRNA . Some of the abundant sRNAs were influenced by instructions within their own genes; some were influenced by instructions from other genes; and some were influenced by both . Wang , Yao et al . also found that the control of protein-coding RNAs was not necessarily related to the control of sRNAs encoded by the same gene . Further work is now needed to identify which specific DNA sequences regulate the abundance of sRNA molecules in plants and other organisms .

2015

Genetic basis of sRNA quantitative variation analyzed using an experimental population derived from an elite rice hybrid

Understanding the initiation and progression of pancreatic ductal adenocarcinoma ( PDAC ) may provide therapeutic strategies for this deadly disease . Recently , we and others made the surprising finding that PDAC and its preinvasive precursors , pancreatic intraepithelial neoplasia ( PanIN ) , arise via reprogramming of mature acinar cells . We therefore hypothesized that the master regulator of acinar differentiation , PTF1A , could play a central role in suppressing PDAC initiation . In this study , we demonstrate that PTF1A expression is lost in both mouse and human PanINs , and that this downregulation is functionally imperative in mice for acinar reprogramming by oncogenic KRAS . Loss of Ptf1a alone is sufficient to induce acinar-to-ductal metaplasia , potentiate inflammation , and induce a KRAS-permissive , PDAC-like gene expression profile . As a result , Ptf1a-deficient acinar cells are dramatically sensitized to KRAS transformation , and reduced Ptf1a greatly accelerates development of invasive PDAC . Together , these data indicate that cell differentiation regulators constitute a new tumor suppressive mechanism in the pancreas . Pancreatic ductal adenocarcinoma ( PDAC ) has a dismal prognosis , with a 5-year survival rate around 5% and a cure rate approaching zero . The most up-to-date chemotherapy regimens extend life only minimally ( Ryan et al . , 2014 ) , and patients undergoing resection of ostensibly local tumors almost invariably succumb to recurrent disease . While this observation suggests that PDAC is usually metastatic at the time of diagnosis , recent studies suggest that tumors require over 20 years to evolve from precancerous pancreatic intraepithelial neoplasia ( PanIN ) to invasive carcinoma ( Yachida et al . , 2010 ) . Thus , in principle , there is a large window of time for effective and early detection , prevention , and treatment , provided appropriate methods are in place . Therefore , defining the cell type of origin and characterizing the process of PanIN-PDAC evolution within the physiologic context of key risk factors ( e . g . , chronic pancreatitis , type 2 diabetes , or genetic cancer syndromes [Ryan et al . , 2014] ) is crucial to finding effective therapies . The vast majority of human PanINs and PDAC contain activating mutations in the KRAS oncogene , which have been shown in mice to represent driver mutations for PanIN initiation , maintenance , and progression to PDAC ( reviewed in Pasca di Magliano and Logsdon , 2013 ) . Notably , the progression of PanINs to PDAC is accompanied by additional mutations in tumor suppressor genes , such as INK4A , CDKN2A , TRP53 ( commonly referred to as P53 ) , and DPC4/SMAD4 ( Ryan et al . , 2014 ) . While the initiation and progression of PDAC has understandably been difficult to study in human patients or to model in human tissue ( Boj et al . , 2015 ) , much has been learned from the ‘KC’ mouse model in which a Cre-inducible oncogenic Kras allele ( KrasLSL-G12D ) causes focal PanIN formation when activated universally in the pancreas ( Aguirre et al . , 2003; Hingorani et al . , 2003; Murtaugh , 2014 ) . Until recently , PDAC was primarily thought to originate from pancreatic ductal cells because of the cancer's duct-like epithelial phenotype . However , recent studies indicate that PanINs ( De La et al . , 2008; Habbe et al . , 2008; Kopp et al . , 2012 ) and PDAC ( Ji et al . , 2009 ) can be initiated by activating oncogenic KrasG12D expression specifically within mature acinar cells , while KrasG12D activation in adult duct cells or centroacinar cells has little or no effect . Interestingly , even in the KC mouse model , where embryonic Cre recombinase activity directs KrasG12D expression to nearly every cell of the mature pancreas , only a small number of acinar cells eventually give rise to PanINs . The mechanism by which most acinar cells remain refractory to KrasG12D-mediated transformation has not been elucidated . An attractive hypothesis is that the factors that induce and maintain acinar cell differentiation state play a crucial role in inhibiting the acinar cell reprogramming step that serves to initiate PDAC formation and progression ( Rooman and Real , 2012; Bailey et al . , 2014; Murtaugh , 2014 ) . Consistent with acinar cells as the cell of origin in PDAC , and acinar cell identity being a protective mechanism against KrasG12D-mediated transformation , recent genome-wide association studies identified PDAC risk-associated single-nucleotide polymorphisms in the non-coding region of the gene encoding the acinar differentiation transcription factor NR5A2 , also known as LRH-1 ( Petersen et al . , 2010 ) . These findings have been confirmed in mouse studies , where pan-pancreatic loss of Nr5a2 significantly sensitizes pancreatic cells to KRAS-induced PanIN initiation . Additionally , pancreatic Nr5a2 is necessary to regenerate the acinar compartment following caerulein-induced pancreatitis ( Flandez et al . , 2014; von Figura et al . , 2014b ) . These studies begin to define how acinar cell differentiation programs may act as an important defense in a progressively severe sequence of events: loss of the mature acinar phenotype , PanIN initiation , and formation of PDAC . In adult pancreata , NR5A2 maintains acinar cell identity by cooperating with the acinar-specific pancreas-specific transcription factor 1 ( PTF1 ) complex , which has binding motifs upstream of essentially all acinar differentiation products , such as Cpa1 , Cela1 , and Cel ( Holmstrom et al . , 2011 ) . The central specificity component of PTF1 is the cell type-restricted basic helix-loop-helix protein , PTF1A ( also known as p48 ) . PTF1A plays two distinct roles during pancreatic organogenesis . First , it is necessary for the growth and morphogenesis of the early pancreatic epithelium , working to impart multipotency and second , its upregulation and lineage-specific interaction with RBPJL promotes acinar differentiation and regulates acinar cell-specific gene expression in adulthood ( Krapp et al . , 1998; Rose et al . , 2001; Kawaguchi et al . , 2002; Masui et al . , 2007 , 2010; Holmstrom et al . , 2011 ) . Homozygous mutations in human PTF1A that disrupt its function or expression cause pancreatic agenesis , supporting its role in pancreas development ( Sellick et al . , 2004; Weedon et al . , 2014 ) . The severity of this phenotype , however , precludes analysis of PTF1A function in mature human acinar cells . Importantly , in the adult pancreas , PTF1A drives its own expression and that of other PTF1 components via a positive autoregulatory loop ( Masui et al . , 2008 ) . Consistent with the central role of this transcription factor in defining and maintaining acinar cell identity , we have shown that PTF1A is downregulated in acinar cells transformed by KrasG12D and Notch activation ( De La et al . , 2008 ) . Beyond these observations , however , a definitive role of PTF1A in regulating the pathogenesis of PDAC and other adult pancreatic pathology has not yet been described . Based on the studies described above , we hypothesized that loss of PTF1A is a necessary and sufficient step in acinar cell reprogramming , the initiation of PanINs , and the progression of PDAC . In this study , we demonstrate that downregulation of PTF1A is a decisive and rate-limiting step in acinar-to-ductal metaplasia ( ADM ) , PanIN initiation , and PDAC progression . Our findings suggest that PTF1A acts in a dosage-sensitive manner to safeguard the pancreatic acinar population against both oncogene activity and environmental insults , such as damage caused by pancreatitis . Our study is the first to establish that an endogenous , autoregulatory differentiation program protects mature pancreatic cells from cancer initiation . We have previously demonstrated that Ptf1a expression is lost when activated Notch and KrasG12D work synergistically to reprogram acinar cells into PanINs ( De La et al . , 2008 ) . Given that Ptf1a is a central regulator of acinar cell gene expression , we hypothesized that this transcription factor should also be downregulated when acinar cells are transformed by oncogenic KrasG12D alone , as well as in human PanINs . To test this hypothesis , we activated KrasG12D specifically in acinar cells using a tamoxifen-inducible Cre expressed by the endogenous Ptf1a locus ( Ptf1aCreERT ) ( Kopinke et al . , 2012; Pan et al . , 2013 ) . Like the widely used Ptf1aCre allele ( Kawaguchi et al . , 2002 ) , Ptf1aCreERT is a ‘knock-in/knock-out’ allele , and therefore , these mice are functionally heterozygous for Ptf1a . We induced KrasG12D expression at 6 weeks of age and harvested pancreata 9 months later . While most acini appeared histologically normal and resistant to KRAS-mediated transformation ( Figure 1A ) , there was intermittent PanIN formation throughout the pancreas ( Figure 1B ) , as previously reported ( Kopp et al . , 2012 ) . By immunohistochemistry ( IHC ) , normal acinar cells in these tissues exhibited robust nuclear PTF1A ( Figure 1C ) ; however , PTF1A was strongly decreased or absent in all acinar-derived PanIN lesions ( Figure 1D ) . To extend these studies to human pancreatic cancer initiation , we stained pathological specimens ( n = 4 ) containing both normal acinar tissue ( Figure 1E ) and PanIN lesions ( Figure 1F ) . As observed in the KrasG12D mouse model , normal acini exhibited a strong PTF1A nuclear signal ( Figure 1G ) , but PTF1A was largely absent from epithelial cell nuclei within PanINs ( Figure 1H ) . In a small fraction of human PanINs , low levels of PTF1A were observed in a subset of epithelial cells ( Figure 1—figure supplement 1 ) . Residual PTF1A expression is consistent with the finding that approximately one-third of human PDAC samples express low levels of acinar-specific genes ( Collisson et al . , 2011 ) . 10 . 7554/eLife . 07125 . 003Figure 1 . PTF1A is downregulated in PanINs from mice and humans . ( A , B ) H&E staining of normal acinar and pancreatic intraepithelial neoplasia ( PanIN ) tissue of Ptf1aCreERT; KrasLSL-G12D pancreata . ( C , D ) PTF1A immunohistochemistry ( IHC ) of mouse acinar and PanIN tissue . ( E , F ) H&E staining of human acinar and PanIN tissue . ( G , H ) PTF1A immunostaining of normal acinar and PanIN tissue of human . Scale bar: 25 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 07125 . 00310 . 7554/eLife . 07125 . 004Figure 1—figure supplement 1 . PTF1A expression in rare epithelial cells of human PanINs . ( A ) IHC for PTF1A from a KrasG12D mouse pancreas 9 months after TM administration ( 0 . 17 mg/g ) . ( B–E ) IHC for PTF1A on human pathology samples . Black arrows indicate normal acinar cells expressing PTF1A; green arrows highlight PanIN epithelial cells that do not express PTF1A; red arrows indicate rare PanIN epithelial cells that retain trace PTF1A expression . Scale bars: ( A–C ) 25 μm; ( D ) 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07125 . 004 In order to determine whether PTF1A downregulation was a functionally important step in PanIN initiation , or a side effect of acinar cell transformation itself , we used an inducible system to delete Ptf1a both in the absence and presence of oncogenic KrasG12D . In this model , we combined the Ptf1aCreERT allele , which does not express PTF1A protein , with a ‘floxed’ Ptf1a allele , to generate Ptf1a conditional knock-out ( cKO ) mice of the genotype Ptf1aCreERT/lox . We also crossed KrasLSL-G12D onto this Ptf1a cKO background . Negative control littermates were Ptf1a heterozygous ( Ptf1aCreERT/+ ) without oncogenic Kras . An additional control group , representing baseline PanIN initiation in the presence of wild-type PTF1A , consisted of Ptf1aCreERT/+; KrasLSL-G12D littermates ( henceforth referred to as KrasG12D mice ) . All inducible-Cre mice also contained a R26REYFP reporter ( Srinivas et al . , 2001 ) , which allowed monitoring of the frequency of Cre-mediated recombination and lineage-tracing of the fate of recombined acinar cells . Table 1 summarizes the genotypes of mice used throughout this study; Figure 2—figure supplement 1 schematically depicts the alleles in each genotype . 10 . 7554/eLife . 07125 . 005Table 1 . Nomenclature of mouse mutants used in this studyDOI: http://dx . doi . org/10 . 7554/eLife . 07125 . 005Short-hand notationPtf1a allelesKras alleleReporter alleleControlPtf1aCreERT/+–R26REYFP/+Ptf1a cKOPtf1aCreERT/lox–R26REYFP/+KrasG12DPtf1aCreERT/+KrasLSL-G12D/+R26REYFP/+Ptf1a cKO; KrasG12DPtf1aCreERT/loxKrasLSL-G12D/+R26REYFP/+cKO , conditional knock-out . In initial studies , 6- to 8-week-old mice were administered tamoxifen ( TM ) at 0 . 17 mg/g body weight , and pancreata were harvested 9 months later ( Figure 2A ) . Compared with control samples , Ptf1a cKO pancreata exhibited intermittent ADM throughout the pancreas ( Figure 2B , C ) . Metaplastic ‘ductules’ of Ptf1a cKO expressed Cytokeratin-19 ( CK19 ) , similar to normal ducts of control; however , Ptf1a cKO ductules appeared more dilated than control ducts ( Figure 2F , G ) . Ptf1a cKO ductules also expressed the duct cell-restricted transcription factor SOX9 ( Figure 2J , K ) , indicating a shift from an acinar to a duct-like differentiation state ( Kopp et al . , 2012 ) . However , these metaplastic ductules did not have the histological morphology of PanINs ( Figure 2C ) , nor did they stain positively for the PanIN-specific markers Claudin-18 ( CLDN18 ) by IHC ( Westmoreland et al . , 2012 ) ( Figure 2O ) or acidic mucins by Alcian Blue histochemistry ( Hingorani et al . , 2003; Kopp et al . , 2012 ) ( Figure 2S ) . Interestingly , ADM in Ptf1a cKO mice was associated with no or scant inflammatory infiltrates , and the surrounding areas did not stain positively with Sirius Red ( Figure 2W ) , a histochemical stain that highlights fibrotic collagen matrix ( Neuschwander-Tetri et al . , 2000 ) . 10 . 7554/eLife . 07125 . 006Figure 2 . Loss of Ptf1a promotes acinar-to-ductal metaplasia and sensitizes acinar cells to KRAS-mediated transformation . ( A ) Mice of indicated genotypes were administered TM ( 0 . 17 mg/g ) to induce recombination , and sacrificed 9 months later . ( B–E ) H&E staining of pancreata from mice of indicated genotypes . ( F–M ) IHC for the duct markers CK19 and SOX9 , indicating upregulation in both acinar-to-ductal metaplasia ( ADM ) and PanINs . ( N–Q ) IHC for the PanIN marker , CLDN18 , highlighting intermittent PanIN formation in KrasG12D mice and widespread lesion development in Ptf1a conditional knock-out ( cKO ) ; KrasG12D . ( R–U ) Alcian Blue staining , indicating PanIN lesions in KrasG12D and Ptf1a cKO; KrasG12D pancreata . ( V–Y ) Sirius Red staining , highlighting local and widespread fibrosis in KrasG12D and Ptf1a cKO; KrasG12D mice , respectively . Scale bars: ( B–E ) 200 µm; ( F–I ) 200 µm; ( J–Q ) 25 µm; ( R–U ) 500 µm; ( V–Y ) 200 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 07125 . 00610 . 7554/eLife . 07125 . 007Figure 2—figure supplement 1 . Schematic of mouse alleles used in this study . Schematic representations of the alleles present in the genotypes referred to , in shorthand , as control , Ptf1a cKO , KrasG12D and Ptf1a cKO; KrasG12D ( see Table 1 for additional details ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07125 . 00710 . 7554/eLife . 07125 . 008Figure 2—figure supplement 2 . Ptf1aCreERT deletion efficiency following tamoxifen treatment . 6–8-week-old mice of indicated genotypes were administered according to low- or high-dose regimens ( 1 × 0 . 17 mg/g or 0 . 25 mg/g , respectively ) , and pancreata were harvested 2 weeks ( A–E ) or 3 days ( F–J ) after the last dose . ( A–D ) Immunofluorescence for amylase ( red ) and the Cre reporter R26REYFP ( green ) on pancreata from low-TM treated mice of the indicated genotypes . For Ptf1a cKO; KrasG12D pancreata , efforts were made to find histologically normal areas to provide an accurate quantification of Cre-mediated recombination . ( E ) The proportion of EYFP expression among amylase+ acinar cells was quantified for all genotypes . No significant difference was noted between any groups ( n = 3–6 per genotype ) . ( F–H ) Immunofluorescence for PTF1A ( red ) , the Cre reporter EYFP ( green ) , and DAPI ( blue ) in Ptf1aCreERT/lox; KrasG12D; R26REYFP/+ mice 3 days after no TM administration ( A ) , low dose TM ( B ) , or high dose TM ( C ) . ( I ) Quantification of the percentage of EYFP+ ( green ) and Ptf1a+ ( red ) pancreatic cells in each indicated treatment group 3 days after final TM administration ( n = 3 per group ) . ( J ) Quantification of total EYFP+ acinar cells that no longer express Ptf1a ( green ) or retain Ptf1a protein expression ( red ) 3 days following low and high TM treatment ( n = 3 per group ) . Scale bars: ( A–D ) 100 μm , ( F–H ) 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07125 . 008 We next tested if inactivation of Ptf1a sensitized acinar cells to oncogenic KRAS-mediated transformation and PanIN initiation . While intermittent PanIN formation was observed in KrasG12D mice ( Figure 2D ) , pancreata from Ptf1a cKO; KrasG12D mice were uniformly composed of extensively distributed PanINs embedded in fibrotic stroma , with almost no remaining normal acinar tissue ( Figure 2E ) . PanINs in both KrasG12D and Ptf1a cKO; KrasG12D mice were positive for the duct marker Cytokeratin-19 ( Figure 2H , I ) and the duct-cell transcription factor SOX9 ( Figure 2L , M ) , as well as the PanIN markers CLDN18 ( Figure 2P , Q ) and Alcian Blue acidic mucin staining ( Figure 2T , U ) . Interestingly , only the Ptf1a cKO; KrasG12D pancreata exhibited abundant Sirius Red staining , indicating widespread fibrotic injury ( Figure 2X , Y ) . Taken together , these data indicate that loss of Ptf1a sensitizes acinar cells to ADM and dramatically increases their susceptibility to oncogenic KRAS transformation and PDAC initiation . Given the severity and robustness of PanIN formation in Ptf1a cKO; KrasG12D mice 9 months after TM administration , we next determined if loss of Ptf1a had a more acute effect on acinar cell transformation . To address this issue , 6- to 8-week-old mice were administered TM ( 0 . 17 mg/g ) and pancreata were harvested 2 or 6 weeks thereafter ( Figure 3A ) . To ensure that Cre-mediated recombination rates were comparable between genotypes , we determined the percentage of acinar cells expressing the R26REYFP reporter at 2 weeks post-TM administration . We found similar acinar recombination rates of 21–25% between genotypes ( Figure 2—figure supplement 1 ) . As the efficiency of Cre-mediated recombination can vary between different target loci ( Liu et al . , 2013 ) , we additionally compared the extent and distribution of PTF1A ablation to that of R26REYFP activation . 3 days after TM administration ( 0 . 17 mg/g ) , there was a ∼20% decrease in the number of PTF1A+ cells detected by immunofluorescence ( Figure 2—figure supplement 2 ) . Importantly , the majority ( ∼75% ) of EYFP+ cells were PTF1A-negative at this dose of TM ( Figure 2—figure supplement 2 ) , indicating that activation of EYFP provides an approximate surrogate for deletion of Ptf1a . 10 . 7554/eLife . 07125 . 009Figure 3 . Loss of Ptf1a is a rate-limiting step in PanIN initiation . ( A ) Mice of specified genotypes were administered 0 . 17 mg/g body weight TM to induce Cre-mediated recombination and were sacrificed either 2 or 6 weeks later . ( B–E ) H&E staining of pancreata from mice of indicated genotypes 2 weeks after TM administration . ( F–I ) IHC for the ductal transcription factor SOX9 , indicating upregulation in ADM and PanINs of Ptf1a cKO; KrasG12D pancreata . ( J–M ) Alcian blue staining , indicating PanIN lesions in Ptf1a cKO; KrasG12D pancreata . In panel ( M ) , green arrow indicates an Alcian Blue+ lesion , while red arrows indicate ADM that is Alcian Blue-negative . ( N ) Quantification of the genotype-dependent PanIN burden; Ptf1a cKO; KrasG12D pancreata possessed significantly more PanINs at 2 weeks post-TM than KrasG12D mice ( p < 0 . 01 ) . ( O–R ) H&E staining of pancreata from mice of indicated genotypes 6 weeks after TM administration . ( S–V ) Alcian Blue staining , highlighting PanIN lesions in pancreata from KrasG12D mice and Ptf1a cKO; KrasG12D mice . ( W ) Quantification of PanINs at 6 weeks post-TM . Ptf1a cKO; KrasG12D pancreata had ∼15-fold more Alcian Blue+ PanINs at this time point than KrasG12D ( p < 0 . 0001 ) . Scale bars: ( B–E ) 200 µm; ( F–I ) 100 µm; ( J–M ) 500 µm; ( O–R ) 200 µm; ( S–V ) 500 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 07125 . 00910 . 7554/eLife . 07125 . 010Figure 3—figure supplement 1 . Microenvironmental remodeling in Ptf1a cKO; KrasG12D pancreata . Pancreata were harvested 2 weeks following TM administration ( 0 . 17 mg/g ) to mice of the indicated genotypes . ( A–D ) Immunofluorescence for the PanIN marker CLDN18 ( red ) , and the leukocyte marker CD45 ( green ) , revealing association of leukocytes with PanINs in Ptf1a cKO; KrasG12D . ( E–H ) IHC for α-SMA , highlighting the activation of pancreatic stellate cells in Ptf1a cKO; KrasG12D pancreata . ( I–L ) Sirius Red staining , highlighting widespread fibrosis in Ptf1a cKO; KrasG12D . Scale bars: ( A–D ) 50 μm , ( E–L ) 200 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07125 . 01010 . 7554/eLife . 07125 . 011Figure 3—figure supplement 2 . Acinar-ductal reprogramming in 3D culture . ( A , B ) Brightfield images of acinarderived ductal cysts from ( A ) KrasG12D and ( B ) Ptf1a cKO; KrasG12D mice , after 3 days of culture in collagen gel . ( C ) Quantification of the maximum diameter of acinar-derived cysts from mice of indicated genotypes ( n = 3 mice per genotype ) . For each mouse , >10 randomly selected fields were photographed , and each cyst within the image was measured at its maximum diameter . Scale bars are 400 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07125 . 011 Interestingly , this level of Ptf1a deletion alone did not produce ADM or other histologically detectable effects at 2 weeks post-TM , compared to control mice ( Figure 3B , C ) . While KrasG12D pancreata exhibited few or no PanINs at this time point , there was widespread induction of ADM , leukocyte infiltration , fibrosis , and PanIN initiation in Ptf1a cKO; KrasG12D pancreata ( Figure 3D , E and Figure 3—figure supplement 1 ) . We further confirmed that acinar-derived ADM and PanINs were being reprogrammed to a duct-like fate based on expression of the ductal transcription factor SOX9 . While only normal ducts expressed SOX9 in control pancreata , PanINs and ADM in Ptf1a cKO; KrasG12D were SOX9+ at 2 weeks post-TM ( Figure 3F–I ) . These data are consistent with a recent study indicating that Sox9 is necessary but not sufficient for the earliest stages of mouse PanIN initiation ( Kopp et al . , 2012 ) . To quantify lesion burden , we stained pancreata from all genotypes with Alcian Blue to highlight acidic mucin-rich PanINs ( Figure 3J–M ) . Following a counting procedure established in our lab ( De La et al . , 2008 ) , we observed a ∼sixfold increase in the frequency of PanINs in Ptf1a cKO; KrasG12D mice compared to mice expressing KrasG12D alone ( Figure 3N ) . This is likely an underestimation of overall phenotypic change , since ADM , which precedes PanIN formation , does not stain with Alcian Blue . Based on histological inspection , ADM is widespread in Ptf1a cKO; KrasG12D mice at 2 weeks post-TM , but negligible in KrasG12D pancreata ( Figure 3D , E , L , M ) . Initiation and progression of PDAC involves interactions between KRAS-active epithelial cells and their stromal microenvironment , with local inflammation being commonly associated with more rapid tumorigenesis ( Gukovsky et al . , 2013 ) . We observed that PanINs developing after 2 weeks in Ptf1a cKO; KrasG12D mice , identified by CLDN18 staining , were consistently surrounded by CD45+ leukocytes , indicating interactions between transformed epithelial cells and inflammatory cells ( Figure 3—figure supplement 1 ) . Because activation of pancreatic stellate cells is a hallmark of PDAC , we assessed the activation state of these cells using the marker α-smooth muscle actin ( SMA ) . While SMA-positive cells were observed around blood vessels in pancreata of all genotypes , lobules of Ptf1a cKO; KrasG12D pancreata affected by ADM and PanIN initiation exhibited widespread SMA+ fibroblasts surrounding ADM and PanIN lesions ( Figure 3—figure supplement 1 ) . Staining with the fibrosis marker Sirius Red confirmed that PanIN-associated fibroblasts of Ptf1a cKO; KrasG12D pancreata were actively secreting collagenous matrix ( Figure 3—figure supplement 1 ) , indicating activation of stellate cells only 2 weeks after Cre-mediated recombination . The activation of stellate cells and fibrotic phenotype observed in Ptf1a cKO; KrasG12D pancreata ( Figure 3—figure supplement 1 ) is likely a reaction to the high level of acinar cell transformation rather than a direct reaction to Ptf1a deletion itself , as Ptf1a cKO pancreata with ADM do not stain with Sirius red ( Figure 2V ) . In order to determine the acinar cell-intrinsic consequences of Ptf1a deletion , we used a 3D culture system in which acini can undergo metaplasia into ductal cysts in response to mutant Kras or EGF receptor ( EGFR ) ligand stimulation , without the influence of other cell types ( Means et al . , 2005; Ardito et al . , 2012 ) . To induce widespread , acinar-specific Kras activation and/or Ptf1a deletion , we treated mice with three daily doses of tamoxifen at 0 . 25 mg/ml , a treatment paradigm that we found to drive widespread recombination ( see below ) . Acinar cell clusters from control , Ptf1a cKO , KrasG12D , and Ptf1a cKO; KrasG12D were isolated at 3 days after the final TM dose , prior to the appearance of any histological abnormalities , and embedded in a collagen matrix , as previously described ( Means et al . , 2005; Ardito et al . , 2012 ) . Neither control nor Ptf1a cKO acinar clusters underwent spontaneous cyst conversion , in the absence of added growth factors , implying that loss of Ptf1a is not sufficient for acinar-ductal reprogramming . As expected , acini of both genotypes generated CK19+ ductal cysts in response to the EGFR ligand TGFα ( data not shown ) . By contrast , KrasG12D activation was sufficient for generation of acinar-derived cysts; importantly , Ptf1a cKO; KrasG12D acini formed significantly larger cysts than those derived from KrasG12D pancreata ( Figure 3—figure supplement 2 ) . These results are consistent with our in vivo data and suggest that acinar cell loss of Ptf1a enhances KRAS-mediated transformation independent of effects on the stromal microenvironment . A generally similar synergy between KrasG12D and Ptf1a cKO was observed in vivo at the 6-week post-tamoxifen time point . Ptf1a cKO pancreata remained histologically unchanged compared to control , as at 2 weeks post-TM , while intermittent PanIN-1 lesions were observed in KrasG12D pancreata ( Figure 3O–Q ) . Ptf1a cKO; KrasG12D pancreata , by contrast , were completely overrun by PanINs at this time point ( Figure 3R ) , most of which stained positively with Alcian Blue ( Figure 3V ) . Quantifying PanIN lesions by Alcian Blue staining , we observed a >15-fold increase in Ptf1a cKO; KrasG12D compared to mice expressing KrasG12D alone ( Figure 3W ) . As we did not score more than one lesion per individual anatomic lobule , to avoid double-counting large or discontinuous lesions , this number likely underestimates the overall PanIN burden in Ptf1a cKO; KrasG12D pancreata given the likelihood of multiple initiation events per lobule . Altogether , the dramatic acceleration of PanIN development upon Ptf1a deletion suggests that downregulation of this TF is a rate-limiting step for KRAS-driven pancreatic tumorigenesis . As we were surprised that a moderate level of acinar cell recombination ( ∼25% ) failed to produce an overt , short-term phenotype in Ptf1a cKO pancreata ( Figure 3 ) , we tested if more pervasive deletion of Ptf1a would produce a more robust reprogramming phenotype . Control and Ptf1a cKO mice were administered a higher dose of TM ( 0 . 25 mg/g ) by oral gavage on three consecutive days ( a net 4 . 5-fold higher dose than previously ) and were harvested 2 weeks later ( Figure 4A ) . Quantification of EYFP+ acinar cells following this TM regimen demonstrated a recombination frequency of ∼65% ( Figure 4—figure supplement 1 ) . Additionally , we quantified the number of PTF1A-deficient acinar cells at 3 days after the final TM gavage , and found that only ∼15% of all pancreatic cells retained nuclear PTF1A , compared with ∼82% in TM-untreated controls ( Figure 2—figure supplement 2 ) . As with low-dose TM , described above , the majority ( >90% ) of EYFP+ cells were PTF1A-negative at 3 days post-TM , confirming that EYFP expression highlights acinar cells deleted for Ptf1a ( Figure 2—figure supplement 2 ) . The apparently greater extent of PTF1A ablation , relative to EYFP activation , may imply the existence of Ptf1a-deleted cells within the EYFP-negative population; such an observation would be consistent with previous evidence of locus-specific Cre deletion efficiencies ( Liu et al . , 2013 ) . 10 . 7554/eLife . 07125 . 012Figure 4 . Widespread loss of Ptf1a promotes rapid acinar-to-ductal metaplasia . ( A ) Control and Ptf1a cKO mice were administered TM ( 0 . 25 mg/g ) on three consecutive days and sacrificed following a 2-week chase period . ( B ) Pancreas mass , measured as a percent of body weight , was significantly decreased in Ptf1a cKO mice 2 weeks after TM administration . ( C , D ) Immunofluorescence for the acinar enzyme carboxypeptidase A1 ( CPA1 ) ( red ) and Cre reporter R26REYFP ( green ) . Nuclei are labeled with DAPI ( blue ) . Inset highlights EYFP+ , CPA1-negative acinar cells forming duct-like structures in Ptf1a null pancreata . ( E ) Quantification of CPA1 expression by EYFP+ ( Cre-recombined ) cells in control and Ptf1a cKO pancreata ( control n = 3 , Ptf1a cKO n = 4 , p < 0 . 01 ) . ( F , G ) H&E staining of control and Ptf1a cKO pancreata 2 weeks after high-dose TM administration . ( H , I ) IHC for the duct marker CK19 highlighting areas of ADM in Ptf1a cKO pancreata . ( J , K ) Immunofluorescence for PTF1A ( red ) and the Cre reporter R26REYFP ( green ) . White arrow indicates an EYFP+ cell expressing PTF1A in control; white arrowheads indicate non-recombined PTF1A+ cells; yellow arrowhead indicates a recombined , PTF1A-negative cell undergoing metaplasia in Ptf1a cKO . ( L , M ) Immunofluorescence for the duct transcription factor SOX9 ( red ) and the Cre reporter R26REYFP ( green ) . Insets highlight restricted expression of SOX9 in controls and upregulation of SOX9 within EYFP+ acinar cells of Ptf1a cKO . Scale bars: ( C , D ) 100 µm , ( F–I ) 200 µm , ( J , K ) 50 µm , ( L , M ) 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 07125 . 01210 . 7554/eLife . 07125 . 013Figure 4—figure supplement 1 . Cre-mediated recombination rates following high-dose tamoxifen treatment . ( A ) 6–8 week old mice were administered 0 . 17 mg/g TM on three consecutive days , and pancreata were harvested 1 week later . ( B , C ) Immunofluorescence for amylase ( red ) and the Cre reporter R26REYFP ( green ) in TM-treated pancreata of the indicated genotypes . ( D ) The proportion of EYFP expression within amylase+ acinar cells was quantified for all genotypes . No significant difference was observed between groups ( n = 3–4 per genotype ) . Scale bar: 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07125 . 01310 . 7554/eLife . 07125 . 014Figure 4—figure supplement 2 . Loss of Ptf1a promotes pancreatic epithelial transdifferentiation and proliferation . ( A–C ) Immunofluorescence for CK19 ( red ) , the Cre reporter Rosa26EYFP ( green ) , and CD45 ( white ) in Ptf1a cKO pancreata , 2 weeks after high-dose TM administration . Caerulein-treated Ptf1a cKO pancreas included as control for pancreatitis ( B ) . Enlarged boxed area highlights EYFP+ acinar cells expressing CK19 , surrounded by CD45+ leukocytes . ( D , E ) Immunofluorescence for E-cadherin ( green ) , cleaved Caspase-3 ( red ) and DAPI ( blue ) in control and Ptf1a cKO pancreata . ( F ) Pancreatic lymph node , positive control for cleaved Caspase-3 staining . ( G , H ) Immunofluorescence for E-cadherin ( green ) , Ki67 ( red ) and DAPI ( blue ) . ( I ) Quantification of Ki67+/E-cadherin+ cells per 40× field ( n = 3 per group , p < 0 . 05 ) . Scale bars: ( A–F ) 100 μm , ( G–H ) 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07125 . 01410 . 7554/eLife . 07125 . 015Figure 4—figure supplement 3 . High-dose tamoxifen administration does not induce pancreatitis . ( A–D ) Lowand ( A′–D′ ) high-magnification H&E staining of wild-type pancreata , fixed 24 hr after indicated treatment . ( E–H ) Immunofluorescence for CK19 ( red ) , CD45 ( green ) and DAPI ( blue ) , highlighting inflammatory cells in the exocrine pancreas . Scale bars: ( A–D ) 200 μm , ( A′–D′ ) 100 μm , ( E–H ) 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07125 . 015 2 weeks following high-dose TM , Ptf1a cKO pancreata were less than half the mass of their control counterparts ( Figure 4B ) . Immunofluorescence revealed that while nearly all EYFP+ acinar cells expressed the acinar marker carboxypeptidase A1 ( CPA1 ) in controls , this marker was lost from approximately 15% of EYFP+ cells in Ptf1a cKO tissues , indicating loss of the normal differentiation state ( Figure 4C–E ) . Histologically , Ptf1a cKO pancreata exhibited extensive acinar disorganization and dilation as well as sporadic upregulation of CK19 within acinar structures , suggestive of early stages of ADM ( Figure 4F–I ) . CK19+ acinar cells ( defined by EYFP co-expression ) were consistently surrounded by CD45+ leukocytes ( Figure 4—figure supplement 2A–C ) , consistent with an intimate association between metaplasia and inflammatory cell recruitment ( Liou et al . , 2013; Murtaugh and Keefe , 2015 ) . Nonetheless , Ptf1a cKO pancreata did not exhibit a general pancreatitis phenotype ( Figure 4—figure supplement 2A–C ) nor did they exhibit a detectable increase in epithelial cell apoptosis ( Figure 4—figure supplement 2D–F ) . In addition , we found that treatment of wild-type mice with high-dose TM was not sufficient to induce pancreatic inflammation ( Figure 4—figure supplement 3 ) , suggesting that the stronger phenotype of high-dose Ptf1a cKO mice , relative to low-dose , was not due to stimulation of ADM by non-specific tissue damage . Loss of PTF1A was accompanied by upregulation of SOX9 by the majority of EYFP+ cells , indicating partial reprogramming to a duct-like state ( Figure 4J–M ) . Surprisingly , we also observed a significant ( ∼fourfold ) increase in the fraction of Ki67+ epithelial cells in Ptf1a cKO pancreata compared with control , suggesting that loss of PTF1A results in deregulation of proliferation as well as differentiation ( Figure 4—figure supplement 2G–I ) . Taken together , these data indicate that Ptf1a is required to maintain acinar gene expression and quiescence , as well as prevent metaplasia to a duct-like state , potentially by inhibiting upregulation of SOX9 . In order to investigate further the mechanism of ADM after loss of Ptf1a , we performed RNA-seq on whole pancreata from three control and three Ptf1a cKO mice , each of which received three doses of TM ( 0 . 25 mg/g ) to induce maximal recombination 2 weeks prior to RNA extraction . Initial analysis of RNA-seq data sets by edgeR ( Robinson et al . , 2010 ) , setting a false discovery rate ( FDR ) threshold of 0 . 05 , identified significant changes in expression of over 3000 total genes ( Figure 5A ) . Consistent with our immunostaining ( Figure 4 ) , among the most significantly downregulated mRNAs were Ptf1a ( 18 . 4-fold ) and Cpa1 ( 5 . 45-fold ) , while Sox9 was significantly upregulated ( 4 . 61-fold ) in Ptf1a cKO pancreata ( Figure 5A ) . Additional downregulated mRNAs included a wide variety of digestive enzymes and other secreted proteins characteristic of the exocrine acinar phenotype , consistent with the long-standing hypothesis that they are directly regulated by PTF1A ( Rose et al . , 2001; MacDonald et al . , in preparation ) . 10 . 7554/eLife . 07125 . 016Figure 5 . Ptf1a suppresses fibroinflammatory pathways and oncogenic KRAS associated gene signatures . ( A ) Volcano plot showing differentially expressed genes ( false discovery rate [FDR] <0 . 05; gray ) in Ptf1a cKO pancreata , relative to control . Individual genes are labeled and highlighted in black . Genes belonging to signatures characteristic of RAS dependency , classical and exocrine-like pancreatic ductal adenocarcinoma ( PDAC ) are highlighted in pink , green , and blue , respectively . Table below indicates p-values from binomial test for enrichment of gene signatures within up- or down-regulated genes . ( B ) Gene Set Enrichment Analysis ( GSEA ) enrichment plots of differentially expressed genes between Ptf1a cKO and control indicating positive enrichment of RAS dependency and classical PDAC signatures and negative enrichment of exocrine-like PDAC signature genes . ( C , D ) Ingenuity Pathway Analysis ( IPA , Qiagen Redwood City , www . qiagen . com/ingenuity ) was used to identify differentially expressed pathways and upstream regulators in Ptf1a cKO pancreata . ( C ) Heat map of pathways that are significantly increased and decreased upon Ptf1a deletion . ( D ) Heat map of upstream pathways and regulators predicted to drive the observed changes in gene expression . Color scale is indicative of the -log p-value ( significance ) . All analyses are based on a ±2 . 0-fold expression threshold . Full details of the data set and analyses can be found in the supplementary data files . DOI: http://dx . doi . org/10 . 7554/eLife . 07125 . 016 Given our finding that loss of Ptf1a strongly potentiates KRAS-induced PanIN initiation ( Figures 2 , 3 ) , we analyzed the expression of genes previously implicated in KRAS signaling and PDAC development . Interestingly , tumor suppressors classically associated with PDAC , such as p53 ( Trp53 ) , Cdkn2a/Ink4a , Pten , Brca2 , and Smad4 , were not significantly downregulated in the absence of Ptf1a ( data not shown ) , leading to the notion that the susceptibility of Ptf1a cKO pancreata to KRAS involves a novel mechanism distinct from canonical tumor suppression pathways . By contrast , we found that two acinar-specific transcription factors previously implicated in suppressing PanIN development , Bhlha15 ( commonly referred to as Mist1 ) and Nr5a2 ( Shi et al . , 2009b; Flandez et al . , 2014; von Figura et al . , 2014b ) , were downregulated in Ptf1a cKO mice , consistent with PTF1A acting at or near the top of a regulatory hierarchy responsible for maintaining acinar identity and suppressing tumorigenesis ( Figure 5A ) . In human cell lines derived from pancreatic and other cancers , dependence on KRAS signaling correlates with expression of specific gene signatures , including genes whose activity is required to sustain RAS activity and malignancy ( Singh et al . , 2009; Loboda et al . , 2010 ) . We found that previously identified RAS dependence genes were significantly enriched , by binomial test , among mRNAs upregulated in Ptf1a cKO tissue ( Figure 5A ) . Within this signature were some of the most highly upregulated mRNAs in our data set , such as Tspan1 ( 32 . 4-fold increase ) , Slc1a2 ( 62 . 2-fold increase ) , Fut2 ( 41 . 4-fold increase ) , and Egr1 ( 6 . 2-fold increase ) ( Supplementary files 1 , 2 ) . The preferential upregulation of RAS dependency genes in Ptf1a cKO was confirmed by Gene Set Enrichment Analysis ( GSEA ) ( Subramanian et al . , 2005 ) ( Figure 5B ) , and suggests that loss of PTF1A results in a phenotypic shift toward a KRAS-permissive phenotype . RAS dependency is characteristic of human PDAC cell lines and primary tumors with a ‘classical’ , duct-enriched gene expression profile ( Collisson et al . , 2011 ) . We find that the classical PDAC signature is also preferentially upregulated upon Ptf1a deletion , while the distinct ‘exocrine-like’ PDAC signature , largely comprising acinar-specific secreted proteins , is downregulated ( Figure 5A , B ) . These results therefore strongly suggest not only that PTF1A maintains acinar differentiation , including expression of genes marking an acinar-like subset of human PDAC , but also that PTF1A suppresses an alternative gene expression program that facilitates KRAS signaling activity . To identify biological pathways that were activated or attenuated by Ptf1a deletion , we analyzed this RNA-seq data set using Qiagen's Ingenuity Pathway Analysis ( IPA , QIAGEN Redwood City , www . ingenuity . com ) ( Thomas and Bonchev , 2010; Kramer et al . , 2014 ) . We analyzed canonical pathways using three different thresholds of gene expression ( 1 . 5-fold up/downregulation , 2 . 0-fold up/downregulation , and 3 . 0-fold up/downregulation; Supplementary files 3–5 ) . At an upregulation threshold of 2 . 0 , deletion of Ptf1a significantly affected over 300 pathways , several of which have an established role in PDAC initiation . These included T-helper cell-signaling pathways ( McAllister et al . , 2014 ) , stellate-cell activation and fibrosis ( Sherman et al . , 2014 ) , and epidermal growth factor ( EGF ) signaling ( Ardito et al . , 2012; Navas et al . , 2012 ) ( Figure 5C ) . A general ‘pancreatic adenocarcinoma signaling’ pathway was also upregulated , consisting primarily of genes involved in PI-3-kinase and JAK/STAT signaling . We also used IPA Upstream Regulator Analysis to predict upstream signaling mediators that could explain the changes in gene expression within our data set ( Kramer et al . , 2014 ) . The predicted upregulated mediators were consistent across multiple expression thresholds and included TNF-α , TGF-β , IL-1β , NFκB , and the SWI/SNF component Smarca4/Brg1 ( Figure 5D ) . All of these signaling pathways have been implicated in PDAC initiation and progression ( Bardeesy et al . , 2006; Adrian et al . , 2009; Khasawneh et al . , 2009; Maniati et al . , 2011; Daniluk et al . , 2012; Maier et al . , 2013; Gore et al . , 2014; von Figura et al . , 2014a ) . Thus , we propose that loss of Ptf1a alters cell state at multiple levels , ultimately promoting gene expression and signaling activities that are supportive of KRAS transformation . Among the upstream mediators activated in the Ptf1a cKO model are TNF-α and NFkB , both of which promote ADM and inflammation in pancreatitis and amplify KRAS activity in pancreatic tumorigenesis ( Maniati et al . , 2011; Daniluk et al . , 2012; Huang et al . , 2013; Maier et al . , 2013; Sendler et al . , 2013 ) . As Ptf1a deletion upregulates other pathways characteristic of pancreatic injury , such as stellate-cell activation , TGF-β signaling , and dendritic cell maturation ( Bedrosian et al . , 2011; Erkan et al . , 2012 ) , we were interested to determine if loss of Ptf1a would sensitize acinar cells to injury-induced reprogramming even without oncogenic KRAS . To test this hypothesis in vivo , we deleted Ptf1a via high-dose TM administration ( three doses of 0 . 17 mg/g ) , which induced a recombination rate of ∼65% ( Figure 4—figure supplement 1 ) . At 1 week post-TM , acute pancreatitis was induced by two consecutive days of treatment with the secretagogue caerulein , as previously described ( Jensen et al . , 2005; Keefe et al . , 2012 ) , and pancreata were harvested 1 week later ( Figure 6A ) . As a control for caerulein injections , additional TM-treated Ptf1a cKO and control mice were administered saline vehicle alone . As previously reported , control mice recovered from caerulein treatment and were indistinguishable from saline-injected controls after 1 week ( Figure 6B–D ) . In contrast , Ptf1a cKO mice subjected to caerulein-induced pancreatitis exhibited widespread acinar atrophy , persistent inflammation , fibrotic stroma , and the appearance of mucinous metaplastic structures ( Figure 6E , F ) . These abnormal ductules were Alcian Blue-reactive , similar to PanINs ( Figure 6G ) , although staining for the PanIN-specific markers CLDN18 and MUC5AC was observed in only rare and isolated lesions ( Figure 6—figure supplement 1A , B ) . Consistent with the overall distorted histology ( Figure 6E ) and atrophy ( Figure 6H ) of caerulein-treated Ptf1a cKO mice , no normal amylase+ acinar clusters could be detected in these pancreata , in contrast to controls ( Figure 6I–L ) . Acinar-derived EYFP+ cells in caerulein-treated Ptf1a cKO pancreata were instead integrated within CK19+ duct-like structures , suggesting that pancreatitis synergizes with loss of Ptf1a to cause a rapid loss of acinar gene expression and complete reprogramming to a duct-like fate ( Figure 6I–L ) . 10 . 7554/eLife . 07125 . 017Figure 6 . Ptf1a is necessary for acinar cell regeneration and suppression of dysplasia following induced pancreatitis . ( A ) 6- to 8-week-old control and Ptf1a cKO mice were administered three doses of TM ( 0 . 17 mg/g ) on consecutive days . 1 week later , mice were administered eight hourly injections of caerulein or saline vehicle , on two consecutive days . Mice were sacrificed 1 week following caerulein treatment . ( B–E ) H&E staining on control and Ptf1a cKO pancreata ( n = 4–5 per group ) 1 week following caerulein treatment . ( F ) H&E stain highlighting a PanIN-like lesion in caerulein-treated Ptf1a cKO . ( G ) Alcian Blue-positive lesions in caerulein-treated Ptf1a cKO . ( H ) Relative pancreas size , measured as a percent of body weight , among treatment groups ( n = 4–5 per group , p < 0 . 01 ) . ( I–L ) Immunofluorescence for amylase ( red ) , CK19 ( white ) , and the Cre reporter R26REYFP ( green ) , in pancreata of control and Ptf1a cKO treated with saline or caerulein . EYFP+ cells of caerulein-treated cKO have downregulated amylase and contribute to CK19+ PanIN-like structures . Scale bars: ( B–E ) 200 µm , ( I–L ) 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 07125 . 01710 . 7554/eLife . 07125 . 018Figure 6—figure supplement 1 . Mucinous metaplasia associated with hyperactive MEK-ERK signaling in caerulein-treated Ptf1a cKO pancreata . ( A , B ) IHC for the PanIN markers Claudin-18 ( A ) and MUC5ac ( B ) in caerulein-treated Ptf1a cKO with corresponding positive and negative controls . Green arrows indicate weakly Claudin-18 positive or Muc5ac positive lesions and red arrows indicate Claudin-18/Muc5ac negative metaplasia . ( C–F ) IHC for phosphorylated-ERK ( p-ERK ) on control and Ptf1a cKO pancreata 1 week following caerulein treatment . Enlarged boxed area highlights mucinous metaplasia-like lesions from caerulein-treated Ptf1a cKO pancreata with strong nuclear p-ERK signal . Scale bars: 200 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07125 . 018 As our findings in Ptf1a cKO; KrasG12D mice indicate that loss of PTF1A enhances the transforming activity of mutant KRAS , we were interested to determine if development of mucinous metaplasia involved enhanced signaling through endogenous RAS . The MEK-ERK pathway is a major regulator of KRAS-induced acinar reprogramming ( Collins et al . , 2014 ) , and we found that nearly all metaplastic lesions of caerulein-treated Ptf1a cKO mice exhibited robust nuclear phospho-ERK staining ( Figure 6—figure supplement 1C–F ) . Phospho-ERK was undetectable in saline-treated Ptf1a cKO mice , or control mice under either treatment . Taken together , these data demonstrate that PTF1A is necessary for acinar-cell redifferentiation and resolution of tissue injury following acute pancreatitis . In the absence of PTF1A , a persistent inflamed microenvironment may have tumor promoter-like activity , enhancing KRAS-MEK-ERK signaling to induce transformation ( Gukovsky et al . , 2013; Murtaugh , 2014 ) . The above studies rely on genetic deletion of Ptf1a , a process without clear parallel in human disease: somatic mutations of PTF1A are not observed in human PDAC , according to the Catalogue of Somatic Mutations in Cancer ( COSMIC ) database ( cancer . sanger . ac . uk ) . PTF1A is more likely to be downregulated by an epigenetic mechanism , for example , via attenuation of the positive autoregulatory loop by which PTF1A maintains its own expression and that of its partner transcription factors ( Masui et al . , 2008 ) . Impaired expression of PTF1-network components , lowering the threshold for KRAS-mediate reprogramming and transformation , might explain the dosage-sensitive requirement for Nr5a2 in preventing PanIN formation ( Flandez et al . , 2014; von Figura et al . , 2014b ) . To determine if the role of Ptf1a itself is dosage-sensitive , we generated mice of the ‘KC’ genotype , using the Pdx1-Cre driver to activate KrasLSL-G12D throughout the pancreas ( Aguirre et al . , 2003; Hingorani et al . , 2003; Murtaugh , 2014 ) , and which were either heterozygous for a germ line deletion of Ptf1a ( Pdx1-Cre; KrasLSL-G12D; Ptf1aΔ/+ ) or remained homozygous Ptf1a wild type . We harvested pancreata at 1 month of age , at which time PanIN formation is usually minimal in KC mice , and quantified PanIN burden by Alcian Blue staining . Mice heterozygous for Ptf1a had increased PanINs at this early stage , compared to Ptf1a+/+ littermates ( Figure 7A–C ) . This result is consistent with a dosage-sensitive function for PTF1A , such that reduced levels or activity already begin to destabilize acinar differentiation in the face of oncogenic insults . 10 . 7554/eLife . 07125 . 019Figure 7 . Ptf1a heterozygosity increases the frequency of PanINs and allows for rapid progression of PDAC . ( A ) Quantification of PanINs in pancreata from 1-month-old Pdx1-Cre; KrasG12D and Pdx1-Cre; KrasG12D; Ptf1a∆/+ mice ( n = 5 per genotype , p < 0 . 05 ) . ( B , C ) Representative Alcian Blue and Eosin staining from 1-month-old mice of indicated genotypes . ( D ) Kaplan–Meier analysis from KPC mice ( Pdx1-Cre; KrasG12D; p53lox/+; Ptf1a+/+ , blue line ) and KPC; Ptf1a∆/+ mice ( red line ) ( Log-Rank test p < 0 . 01 ) . ( E–H ) H&E staining on tumors from both KPC and KPC:Ptf1a∆/+ mice at low and high magnification . ( G , H ) Arrows indicate ductile epithelial cells and arrowheads indicate areas of necrosis . ( I , J ) IHC for CK19 on tumor specimens from mice of indicated genotypes . Scale Bars: ( B , C ) 500 µm , ( E , F ) 500 µm , ( G–J ) 200 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 07125 . 01910 . 7554/eLife . 07125 . 020Figure 7—figure supplement 1 . Liver metastases in KPC mice heterozygous for Ptf1a . ( A ) Gross image of a Pdx1-Cre; KrasG12D; Ptf1a+/− mouse harboring metastatic pancreatic cancer . White box highlights large liver metastasis . ( B ) Dissected lobe of the liver with pancreatic liver metastasis . ( C ) IHC for CLDN18 confirmed liver metastases in a subset of Pdx1-Cre; KrasG12D; p53+/−; Ptf1a+/− mice . ( D ) Gross image of Pdx1-Cre; KrasG12D; p53+/−; Ptf1a+/− mouse harboring metastatic PDAC . White box highlights liver metastasis . DOI: http://dx . doi . org/10 . 7554/eLife . 07125 . 020 In humans , increased PanIN burden in early life is associated with familial risk of PDAC , suggesting that mutations driving genetic predisposition to PDAC act at the level of tumor initiation ( Brune et al . , 2006; Shi et al . , 2009a ) . We therefore hypothesized that decreased Ptf1a dosage would promote cancer susceptibility by increasing the rate of PanIN initiation . Therefore , we utilized the well-characterized ‘KPC’ model of mouse PDAC in which heterozygous loss of p53 ( official gene symbol Trp53 ) is added to the Pdx1-Cre; KrasLSL-G12D genotype ( Hingorani et al . , 2005; Rhim et al . , 2012 ) . As above , KPC mice ( Pdx1-Cre; KrasLSL-G12D; p53lox/+ ) were generated on either Ptf1a+/+ or Ptf1aΔ/+ backgrounds , and animals were monitored for tumor-free survival . The results of Kaplan–Meier analysis showed that Ptf1a-heterozygous KPC mice developed PDAC much earlier than Ptf1a+/+ counterparts ( Figure 7D , Log-rank test , p < 0 . 01 ) . We observed prominent metastases to the liver in 3/9 Ptf1aΔ/+ KPC mice , but none in Ptf1a+/+ KPC controls ( Figure 7—figure supplement 1 ) . Importantly , despite the earlier onset of PDAC in KPC mice with Ptf1a heterozygosity , once tumors arose they were histologically indistinguishable between genotypes ( Figure 7E–H ) . They contained classical features of human PDAC , including abundant fibrotic stroma surrounding CK19+ epithelial cells ( Figure 7I , J ) and substantial areas of necrosis . We therefore conclude that decreased Ptf1a gene dosage sensitizes pancreata to early KRAS-mediated PanIN initiation and rapid progression to PDAC . Previously , we and others established that acinar-to-ductal reprogramming is a necessary step in PanIN initiation ( De La et al . , 2008; Habbe et al . , 2008; Kopp et al . , 2012 ) . Several recent studies extended these findings , demonstrating that several genes required for PanIN and PDAC development appear to act at the level of acinar cell reprogramming ( Heid et al . , 2011; Ardito et al . , 2012; Kopp et al . , 2012; Baer et al . , 2014; Wu et al . , 2014; Zhang et al . , 2014 ) . Here , we demonstrate that the loss of a principal regulator of acinar cell identity , PTF1A , is sufficient to prompt rapid and extensive acinar-to-ductal metaplasia even in the absence of other exocrine insults ( Figure 4 ) . Additionally , we demonstrate that Ptf1a-deficient acinar cells are extremely sensitive to oncogenic transformation , as they undergo rapid and robust KRAS-mediated PanIN formation ( Figures 2 , 3 ) . Deletion of Ptf1a alone at moderate frequency ( ∼25% ) did not produce detectable histological changes in the pancreas over the course of 2–6 weeks ( Figure 3 ) . By contrast , we observed rapid de-differentiation of Ptf1a cKO acinar cells generated under a high-TM dose regimen that produced >65% deletion ( Figure 4 ) . This is an important finding regarding potentially non-cell autonomous protective mechanisms working to offset PanIN/PDAC initiation . We propose two linked hypotheses: first , when Ptf1a is lost from individual cells , other acinar-specific transcription factors that prevent reprogramming and co-regulate PTF1 target genes , such as NR5A2 and BHLHA15/MIST1 , are sufficient to maintain a differentiated phenotype over the short term . Second , we propose that the persistent differentiation of Ptf1a cKO acinar cells is promoted by interactions with neighboring Ptf1a WT cells , producing a phenomenon similar to the ‘community effect’ in embryonic development ( Gurdon et al . , 1993 ) . However , with increasing TM-driven deletion , the fraction of Ptf1a WT acinar cells passes a tipping point , the community effect cannot be sustained , and ductal metaplasia is correspondingly rapid . At a molecular level , the protective effect of wild-type acinar cells could be mediated by their ability to dampen local inflammation , suggested by the upregulation of fibroinflammatory pathways in our RNA-seq analyses ( Figure 5 ) . This is also suggested by our finding that Ptf1a cKO pancreata exhibit sustained inflammation after acute injury , including the conversion of acinar cells to PanIN-like , Alcian Blue+ ductule structures ( Figure 6 ) . These resemble tubular complexes observed in human and mouse chronic pancreatitis ( Bockman et al . , 1982; Strobel et al . , 2007 ) , suggesting that dysregulation of PTF1A expression or function might be involved in the etiology of this disease and its well-known contribution to PDAC risk . We additionally demonstrate that inflammation and loss of Ptf1a synergize to drive sustained activation of the MEK-ERK pathway , a major effector of oncogenic and endogenous KRAS . Going forward , it will interesting to test whether MEK inhibitors are able to prevent acinar cell reprogramming in the context of chronic pancreatitis and/or decrease the risk of chronic pancreatitis progressing to PDAC . Given the dramatic effects of Ptf1a deletion on transformation and inflammation , it will be important to determine which genes in our RNA-seq data set are directly suppressed or activated by PTF1A . It has been previously established that PTF1A regulates a network of transcription factors controlling acinar-specific gene expression ( Masui et al . , 2008 , 2010 ) . Among these are Bhlha15/Mist1 and Nr5a2 , both downregulated in the Ptf1a cKO condition ( Figure 5 ) , and both previously shown to inhibit PanIN development ( Shi et al . , 2009b; Flandez et al . , 2014; von Figura et al . , 2014b ) . Of these three genes , only Ptf1a is indispensable for acinar cell differentiation ( Krapp et al . , 1998; Pin et al . , 2001; Kawaguchi et al . , 2002; Holmstrom et al . , 2011; von Figura et al . , 2014b ) , and it will be of interest to determine the relative rank of these factors as suppressors of cancer initiation and progression , and their epistatic relationship . It will also be useful to understand how KRAS , together with inflammatory and other insults , is capable of downregulating the expression and/or function of PTF1-network components during tumor initiation . Of note , recent studies indicate that oncogenic KRAS induces specific pathways dedicated to silencing tumor suppressor genes ( Wajapeyee et al . , 2013; Serra et al . , 2014 ) ; a similar process may drive downregulation of Ptf1a and its partners during acinar cell reprogramming . Because loss of Ptf1a strongly potentiated KRAS-mediated transformation ( Figures 2 , 3 ) , we hypothesized that PTF1A inhibits KRAS-signaling activity in some capacity . Here , we demonstrate that loss of Ptf1a leads to upregulation of genes associated with KRAS-dependency in human cancer cells ( Singh et al . , 2009; Loboda et al . , 2010 ) . Future investigations should therefore move forward to test if different subtypes of human PDAC exhibit different extents of PTF1A repression , and whether variation in PTF1A expression within human PDAC correlates with KRAS-dependency or disease prognosis . Recent studies have classified ∼1/3 of pancreatic cancers as ‘exocrine-like’ , and several genes that are under Ptf1a control contribute to this signature ( Figure 5A , B ) ( Collisson et al . , 2011 ) . Unfortunately , human PTF1A was not present on the microarray used in that study; nonetheless , their data suggest that PTF1A and its transcriptional targets are retained at low levels in some , but not all , cases of PDAC . Consistent with these previous reports , we found that sparse epithelial cells in human PanIN lesions retain nuclear PTF1A ( Figure 1—figure supplement 1 ) . In addition to supporting the contention that human PanINs and PDAC arise from mature acinar cells , these findings suggest that low levels of persistent PTF1A , held in check by epigenetic rather than genetic mechanisms , may be available for therapeutically targeted restoration . The fact that removal of a single allele of Ptf1a accelerates mouse PDAC development ( Figure 7D ) suggests that even incomplete inhibition of human PTF1A could promote acinar transformation and subsequent tumorigenesis . Our results suggest that PTF1A restoration provides an indirect approach to target KRAS-dependency in pancreatic cancer , inhibiting this currently ‘undruggable’ , although ubiquitous , cancer-driving mutation ( Pasca di Magliano and Logsdon , 2013 ) . Additionally , our data suggest that PTF1A restoration may reduce inflammatory pathways that feed forward to synergize with oncogenic KRAS ( Maniati et al . , 2011; Daniluk et al . , 2012; Maier et al . , 2013 ) . Future studies will focus on genetic PTF1A gain-of-function approaches to determine if sustained PTF1A expression can prevent and/or reverse acinar-to-ductal reprogramming , PanIN initiation and PDAC progression . In summary , we show that acinar cell differentiation , maintained through PTF1A , suppresses multiple oncogenic pathways associated with PDAC initiation and progression . Our data suggest that PTF1A functions as a nodal point in PDAC initiation by maintaining acinar-cell gene expression , suppressing KRAS function , and resisting inflammation . The antagonism between KRAS and the pro-acinar transcription factor network captures , at a genetic level , the tension between differentiation and malignant transformation that has long been hypothesized to exist in cancer ( Harris , 1990 ) . Loss of normal differentiation and reprogramming of cell fate appear to occur during initiation of a diverse array of tumor types ( Blanpain , 2013 ) . Our results , for the first time , demonstrate that this process is rate-limiting for cancer development , thus , constituting a novel mechanism of tumor suppression . The mouse PanIN-PDAC model provides a new experimental system to relate genetic changes in cancer , such as KRAS mutation , to epigenetic changes such as PTF1A downregulation . Furthermore , understanding how PTF1A function is subverted during pancreatic cancer initiation , and whether its reactivation could suppress or reverse tumor development , may yield novel approaches to prevention and treatment . Experimental mice of the following genotypes have been previously described: Ptf1aCreERT ( Ptf1atm2 ( cre/ESR1 ) Cvw [Kopinke et al . , 2012; Pan et al . , 2013] ) , Pdx1-Cre ( Tg ( Pdx1-cre ) 89 . 1Dam [Gu et al . , 2002] ) , KrasLSL-G12D ( Krastm4Tyj [Hingorani et al . , 2003] ) , p53lox ( Trp53tm1Brn [Marino et al . , 2000] ) , and R26REYFP ( Gt ( ROSA ) 26Sortm1 ( EYFP ) Cos [Srinivas et al . , 2001] ) . The Ptf1alox allele ( Ptf1atm3Cvw ) was generated using homologous recombination in mouse ES cells at the Vanderbilt Transgenic Mouse/Embryonic Stem Cell Shared Resource . The 5′ and 3′ loxP sites were placed 1 . 7 kb upstream and 2 kb downstream of the Ptf1a transcriptional start site , respectively . Full details will be provided elsewhere ( Wright et al . , in preparation ) . Mice with a germ line deletion allele of Ptf1a , Ptf1aΔ were generated by crossing Ptf1alox to the ubiquitous early deletor line Sox2-Cre ( Tg ( Sox2-cre ) 1Amc [Hayashi et al . , 2003] ) . To activate CreERT-mediated recombination , mice were administered tamoxifen ( Sigma , St . Louis , MO ) dissolved in corn oil , via oral gavage at doses indicated in the text . All mouse experiments were carried out according to institutional and NIH guidelines . All human pathological specimens were de-identified before their use . The utilization of these human specimens is therefore not considered human subject research under the US Department of Human and Health Services regulations and related guidance ( 45 CDR Part 46 ) . Paraffin embedded specimens were sectioned ( 6 µm ) and IHC was performed for PTF1A , as described below . Samples were analyzed by NMK , MPB , and LCM . After euthanasia , pancreata were dissected in ice-cold phosphate-buffered saline solution ( PBS ) , separated into multiple fragments , and processed for both frozen and paraffin sections as previously described ( De La et al . , 2008; Keefe et al . , 2012; Kopinke et al . , 2012 ) . Briefly , tissues were fixed for paraffin embedding in zinc-buffered formalin ( Z-FIX; Anatech , Battle Creek , MI ) , room temperature overnight , or 4% paraformaldehyde/PBS , 4°C 1–2 hr , followed by processing into Paraplast Plus ( McCormick Scientific ) or Tissue-Tek O . C . T . compound ( Sakura Finetek , Torrance , CA ) . Paraffin and frozen sections were cut at thickness of 6 µm and 8 µm , respectively , and collected sequentially across multiples slides , with ∼100-µm spacing between individual sections on a single slide . IHC and immunofluorescence followed our established procedures ( De La et al . , 2008; Keefe et al . , 2012; Kopinke et al . , 2012 ) , including high-temperature antigen retrieval ( Vector Unmasking Solution; Vector Laboratories , Burlingame , CA ) prior to staining paraffin sections . Primary antibodies utilized in this study are listed in Table 2 . Secondary antibodies , raised in donkey ( Jackson Immunoresearch , West Grove , PA ) , were used at 1:250 dilution . Vectastain reagents and diaminobenzidine ( DAB ) substrate ( Vector Laboratories ) were used for IHC . Slides stained by immunofluorescence were counterstained with DAPI and mounted in Fluoromount-G ( Southern Biotech ) , and photographed on an Olympus IX71 microscope , using MicroSuite software ( Olympus America , Waltham , MA ) . Images were processed in Adobe Photoshop , with exposure times and adjustments identical between genotypes and treatment groups . 10 . 7554/eLife . 07125 . 021Table 2 . Primary antibodies used in this studyDOI: http://dx . doi . org/10 . 7554/eLife . 07125 . 021AntigenSpeciesSourceCatalog #DilutionAmylaseSheepBioGenesis0480-01041:1000Cleaved-caspase-3RabbitAbcamAB23021:1000Cd45RateBioScience14-0451-821:2000Claudin-18RabbitInvitrogen7001781:2000Cpa1GoatR&D SystemsAF27651:1000Cytokeratin-19RatDevelopmental Studies Hybridoma Bank–1:50Cytokeratin-19RabbitAbcamAB1334961:5000GFPChickenAves Labs Inc . GFP-10101:5000Ki67MouseBD Biosciences5506091:500Muc5acMouseNeoMarkers45M11:500Ptf1aRabbitChris Wright , Vanderbilt University–1:5000Ptf1aGoatChris Wright , Vanderbilt University–1:5000Phospho-ERK1/2 ( T202/Y204 ) RabbitCell Signaling91011:1000Sox9RabbitMilliporeAB55351:1000α-SMARabbitAbcamAB325751:2000SMA , smooth muscle actin . For Alcian Blue staining , paraffin sections were washed 5 min in 3% acetic acid , followed by a 10–12 min incubation in staining solution ( 1% Alcian Blue in 3% acetic acid ) , and extensive washing in 3% acetic acid . Sirius Red staining was performed on frozen sections that were fixed for 1 hr in Bouin's fixative at 55°C . Specimens were subsequently washed in dH2O and stained for 1 hr in Picro-Sirius Red ( American MasterTech , Lodi , CA ) . Following staining , specimens were rinsed in 0 . 5% acetic acid , dehydrated and equilibrated into xylene , and mounted with Permount . The entire tissue area of Alcian Blue/eosin-stained slides was photographed , at 4× original magnification , followed by photomerging ( Adobe Photoshop ) and surface area measurement using ImageJ software ( NIH ) . Alcian Blue+ PanIN lesions were counted manually under the microscope , and PanIN burden calculated as the total number of Alcian Blue+ lesions per cm2 surface area . As described in the text , metaplastic lesions that did not stain with Alcian Blue were not counted in the quantification . To avoid double-counting of potentially large and tortuous lesions , no more than one lesion was scored within an anatomically distinct pancreatic lobule ( De La et al . , 2008 ) . Acinar cultures were established according to previous publications ( Kurup and Bhonde , 2002; Means et al . , 2005; Ardito et al . , 2012 ) . Briefly , dorsal pancreata were minced in Hank's buffered saline solution and digested sequentially in 0 . 02% trypsin ( 5 min , 37°C ) and 1 mg/ml collagenase P ( Roche Applied Science , Mannheim , Germany; 15 min , 37°C ) , following filtration to eliminate undigested material , and repeated washing to eliminate debris and dead cells , acinar cell clusters were embedded in rat tail collagen gels ( Corning , Corning , NY ) , and cultured in Waymouth's medium ( Life Technologies , Carlsbad , CA ) supplemented with 1% fetal bovine serum , 0 . 4 mg/ml soybean trypsin inhibitor , and 1 µg/ml dexamethasone . Cultures were fixed and imaged after 5 days . To quantify cyst size , we randomly selected >10 fields per mouse , imaged , and quantified the maximal diameter of each transformed cyst using ImageJ . In order to quantify R26REYFP labeling , 10–12 randomly selected 20× fields per specimen ( taken across multiple sections ) were photographed . Using ImageJ software ( NIH ) , cells co-expressing EYFP with the acinar differentiation markers Amylase or CPA1 were detected by additive image overlay of their staining with DAPI and anti-GFP , and counted using the Analyze Particles function , as described previously ( Keefe et al . , 2012; Kopinke et al . , 2012 ) . To ensure counting accuracy , random images were spot checked by manual counting using Adobe Photoshop . All calculations were performed in Microsoft Excel and the results are reported as the mean ± standard deviation ( error bars ) . p-values were determined by two-tailed , unpaired t-tests performed in Excel or Graphpad Prism 6 . Total RNA was isolated from pancreata of 4- to 5-month-old Ptf1a cKO mice ( Ptf1aCreERT/lox ) and their corollary controls ( Ptf1aCreERT/+ ) , 2 weeks after TM treatment ( 3 days , 0 . 25 mg/g/day ) , using the guanidine thiocyanate protocol previously described with minor modifications ( MacDonald et al . , 1987 ) . Individual RNA-Seq libraries were prepared from 5 μg of pancreatic RNA from three control and three Ptf1a cKO mice with the Illumina True-seq protocol by the UT Southwestern Genomic Core . The sequence data sets from an Illumina HISEQ2500 contained 50-nucleotide uniquely aligned single-end reads of 25 . 1 , 26 . 4 , and 25 . 8 million for the control samples and 29 . 7 , 24 . 7 , and 26 . 7 million for the Ptf1a cKO RNA samples ( Tophat2 ) ( Kim et al . , 2013 ) . Genes with differential expression were derived using edgeR ( Robinson et al . , 2010 ) , with the default trimmed mean of M-values ( TMM ) trim settings of 30% for Mg and 5% for Ag and an FDR cut-off of <0 . 05 . The volcano plot of differentially expressed genes was generated using R ( http://www . r-project . org/ ) with the log2 fold change ( FC ) plotted against the FDR ( −log10 ) ( Supplementary files 1 , 2 ) . Gene signatures of RAS dependency ( Singh et al . , 2009; Loboda et al . , 2010 ) , classical and exocrine-like PDAC ( Collisson et al . , 2011 ) , were mapped to orthologous mouse genes via HomoloGene ID . The RAS dependency signature combines gene lists from two separate studies ( Singh et al . , 2009; Loboda et al . , 2010 ) , comprising 264 genes with only five in common . For GSEA ( Subramanian et al . , 2005 ) , we analyzed signature enrichment within the entire Ptf1a cKO RNA-seq data set , ordered by log2 FC relative to control , using the GSEA desktop application ( http://www . broadinstitute . org/gsea/index . jsp ) . To identify regulatory pathways altered upon Ptf1a deletion , significantly increased and decreased genes were analyzed by IPA ( QIAGEN , Redwood City , CA , www . ingenuity . com ) at expression thresholds of 1 . 5- , 2 . 0 , - and 3 . 0-fold ( Supplementary files 3–5 ) . In order to obtain an accurate comparison between enriched and downregulated pathways , we used the Comparison Analysis function from expression data filtered at a gene expression threshold of ±2 . 0-fold . Heat maps were generated according to the −log p-values given by the IPA software using the comparison analysis function and were constructed in R ( http://www . r-project . org/ ) . We induced acute pancreatitis by i . p . injection of caerulein ( Bachem , Torrance , CA ) , 0 . 1 µg/g , eight times daily over two consecutive days , as previously ( Jensen et al . , 2005; Keefe et al . , 2012 ) . Negative controls were injected with saline vehicle alone . Pancreata from all caerulein- or saline-treated mice were harvested 1 week following the final injection and processed as described above . KPC mice ( of the genotype Pdx1-Cre; KrasG12D; p53lox/+ ) and KPC mice with Ptf1a heterozygosity ( Pdx1-Cre; KrasG12D; p53lox/+; Ptf1aΔ/+ ) were aged until they exhibited lethargy or distress as determined by the authors ( NMK and LCM ) and the in-house veterinary staff , or until the detection of a firm abdominal mass by palpation . The presence of PDAC was confirmed by histological analysis in consultation with a surgical pathologist ( MB ) . At sacrifice , all mice were thoroughly inspected for liver metastases . Survival analysis was performed in GraphPad Prism ( Version 6 ) and p-values were calculated using a Log-rank test .

Pancreatic cancer is one of the most lethal forms of cancer , with fewer than 20% of people surviving for longer than twelve months after diagnosis . Two types of genetic mutation play important roles in pancreatic cancer . First , genes called oncogenes can be activated by mutations to drive unscheduled cell division . Second , the genes for tumor suppressors—proteins that prevent cells from dividing when they should not—can be switched off due to other mutations . Together , these mutations cause cells to over-proliferate and disrupt the structure of the pancreas . In a healthy pancreas , several different cell types perform various roles: acinar cells produce proteins that digest food , ductal cells carry these proteins to the intestine , and β cells produce insulin . Certain proteins are responsible for telling each of these cells what tasks to perform , which defines their so-called differentiation state . The protein PTF1A is crucial for establishing the differentiation state of acinar cells . In the most common form of pancreatic cancer , acinar cells are reprogrammed to become ductal cells . Moreover , pancreatic cancer cells contain much lower levels of PTF1A than normal pancreatic cells . To explore the connection between PTF1A and pancreatic cancer , Krah et al . deleted the gene for PTF1A in mice . This led to acinar cells being reprogrammed to become ductal cells . Additionally , when an oncogene mutation was activated at the same time as the gene for PTF1A was deleted , Krah et al . observed the rapid formation of large numbers of malignant pancreatic tumors in the mice . PTF1A therefore protects against pancreatic cancer by acting as a tumor suppressor and keeping acinar cells in their healthy , differentiated state . Unlike other tumor suppressors , however , PTF1A levels are reduced in cancer cells by a mechanism that does not involve a genetic mutation . Therefore , a future challenge is to determine how the amount of PTF1A protein is reduced , and in the longer term , to explore if it is possible to reverse cancer progression by forcing cancer cells back into their original differentiation state .

2015

The acinar differentiation determinant PTF1A inhibits initiation of pancreatic ductal adenocarcinoma

Possible options in a decision often organize as a hierarchy of subdecisions . A recent study concluded that perceptual processes in primates mimic this hierarchical structure and perform subdecisions in parallel . We argue that a flat model that directly selects between final choices accounts more parsimoniously for the reported behavioral and neural data . Critically , a flat model is characterized by decision signals integrating evidence at different hierarchical levels , in agreement with neural recordings showing this integration in localized neural populations . Our results point to the role of experience for building integrated perceptual categories where sensory evidence is merged prior to decision . Coffee with Jules ( if so , which cafe ? ) or cinema with Jim ( if so , what movie ? ) ? A recent study by Lorteije and colleagues investigated how perceptual mechanisms implement such hierarchically-structured decisions that fill up our daily lives ( Lorteije et al . , 2015 ) . Monkeys performed saccades to one of 4 possible targets based on the information provided at the primary branching and at secondary branching points ( the 'correct' and 'incorrect' branching points ) leading to the targets ( Figure 1A ) . Patterns of responses convincingly indicated that monkeys can integrate information from all branching points in parallel . But does the decision space mimic the hierarchical organization , with parallel decisions going on ( about Jules/Jim , coffee place , movie ) ( Figure 1B ) , or does it directly compare final options ( with Jules at Moe's vs . Sicario with Jim vs . etc . ) ( Figure 1C ) ? Lorteije and colleagues report two behavioral and two neural effects that they argue speak unanimously in favor of the former hierarchical model against the latter non-hierarchical ‘flat model’ of decision-making . We show in contrast that all four effects can equally ( and more parsimoniously ) be explained by the flat model of decision-making , which is also perfectly compatible with the new effects from the same dataset described in the companion paper by Zylberberg and colleagues ( Zylberberg et al . , 2017 ) . 10 . 7554/eLife . 16650 . 003Figure 1 . Hierarchical vs . flat models of perceptual decision during a hierarchically-structured visual task . ( A ) Structure of the task . At each trial monkeys must detect the correct option out of four possible responses based on the visual information provided at the primary branching point L1 and at two secondary branching points L2 and L2’ . Visual information consists of segments of flickering luminosity at the start of each branch ( color segments in our depiction; visual samples changing every 50 ms for a total period of 1000 ms ) . Animals must make a saccade towards the final point that passes through branches of maximal luminosity; that is , must decide based on information scattered across the visual field . Each of the four responses is categorized as TT , TD , DT or DD depending on whether it corresponds to a target T ( correct branch ) or distractor D ( incorrect branch ) at first and second branching points . Detailed description of the task can be found in [Lorteije et al . , 2015] . ( B ) Hierarchical decision model of perception . In the hierarchical model , parallel decision processes run at each branching point ( L1 , L2 and L2’ ) and are integrated into a motor response at a later stage . It can be implemented as a race model composed of three races , each with two possible sub-choices . ( C ) Flat decision model of perception . In the flat model , the decision space is composed of the four possible final responses , so for each response the animal must sum the information provided at the corresponding primary and secondary branches ( here depicted by the sum of the two luminosity signals ) . ( D ) Implementation of the flat decision model . ( Left panel ) Four units coding for the four possible responses integrate information from both L1 and L2 branches leading to that response , as represented by the pattern of connections from sensory units ( coding for instantaneous sensory value ) . Connectivity include self-excitation as well as homogeneous cross-inhibition between all units . Black and red arrows indicate respectively excitatory and inhibitory connections . ( Right panel ) Simulation for one trial , depicting the activity of each unit across time during perceptual integration . Activity is bounded to positive values ( rectification ) . When activity of one unit reaches the decision threshold , the related response is selected . DOI: http://dx . doi . org/10 . 7554/eLife . 16650 . 003 First , stimulus difficulty at the primary branch ( L1 ) was shown to have no influence on the performance at the secondary branch ( L2 ) , a result that was replicated by a race-model implementation of hierarchical decision-making , but not of the flat decision-making . However , this result seems a particular feature of their specific choice of implementation of the race model ( Vickers , 1979; Drugowitsch et al . , 2014 ) , which included neither inhibition between option representations , nor activation rectification ( i . e . enforcing non-negative unit activation ) , two classical ingredients of race models ( Usher and McClelland , 2001; Tsetsos et al . , 2012; Churchland and Ditterich , 2012 ) . Inhibition between option representations and rectification may underlie the well-known reduction of choice-related neural activity when more choice alternatives are provided ( Churchland and Ditterich , 2012 ) . We simulated a race model implementation of the flat model with both rectified activation , cross-inhibition and self-excitation , whereby each of the four options competed during accumulation of evidence ( Figure 1D , see Material and methods ) . Self-excitation models the instability present at the initial point of the race in attractor models of decision-making ( Roxin and Ledberg , 2008 ) and can be at the heart of well-known urgency signals to speed up decisions ( Drugowitsch et al . , 2012 ) . Noise in the model did not scale with stimulus intensity , in line with recent results suggesting that noise in perceptual accumulation tasks is associated with the accumulation rather than with the sensory process ( Drugowitsch et al . , 2016 ) , and as it is typically assumed in drift-diffusion models of decision making ( Gold and Shadlen , 2007 ) . We used a non-absorbing decision threshold ( i . e . activity after the decision as not bounded ) , but found the same results for simulations with an absorbing bound . Parameters were tuned to reproduce qualitatively the proportion for each of the four possible response types , as well as the impact of each sample in the stream and the psychometric curve for both L1 and L2 decision ( Figure 2—figure supplement 2A–D ) . The impact of L1 difficulty onto L2 for the flat model indeed disappeared when we used this common implementation of the race model ( Figure 2—figure supplement 2E ) . In the companion paper ( Zylberberg et al . , 2017 ) , Zylberberg and colleagues question the generality of this finding , arguing that it may not be compatible with the pattern of short reaction times they provide in a new analysis for the same monkey dataset . However , our mathematical analysis ( Appendix 1 ) shows that a clear influence of L1 difficulty onto L2 performance emerges principally when L1 difficulty strongly modulates reaction times , and thus the time of integration of L2 evidence . Short reaction times with low modulation by task difficulty as described in the companion article are thus perfectly compatible with lack of influence of L1 difficulty onto L2 performance . This result was indeed reproduced in a simulation where the threshold was lowered to produce shorter reaction times , and the boundary was eliminated for the first 10 samples ( Figure 2A–F ) . This choice of the threshold implements a form of time-dependent , collapsing bound ( Churchland and Ditterich , 2012; Drugowitsch et al . , 2012 ) and it is consistent with the minimum viewing time imposed on monkeys ( Figure 2 ) . Without such time-dependent bound , the model can produce either a large proportion of premature responses or very large mean reaction times . The discrepancy between results from these simulations with those performed in the companion paper ( who do find dependency of L2 performance on L1 difficulty with short reaction times ) may emerge from early responses ( <500 ms ) in easy trials that are present in their simulations , but prohibited in our simulations by the absence of decision boundary for the first 10 samples . 10 . 7554/eLife . 16650 . 004Figure 2 . Behavioral properties of the flat race model reproduce original monkey data . The figures presented here correspond to a parameter set adjusted to reproduce reaction times ( see Introduction ) . ( A ) Response type for the flat model . We simulated a simple race model with four possible options that compete for selection , based on a standard implementation of race models [Drugowitsch et al . , 2014] . The histogram shows the distribution of each of the four response types ( TT , TD , DT , DD ) for each level of trial difficulty from simulations of the race implementation of the flat model . ( B ) Influence of luminance fluctuations on the L1 decisions . Weights for each of the target ( orange ) and distractor ( blue ) samples across a trial for L1 decisions , as measured with logistic regression . The model reproduces the primacy effect observed in monkey behavioural data , whereby earlier samples have larger influence on L1 choices than later samples . Shades indicating 95% confidence intervals ( barely visible ) are nearly collapsed to the main line . Weights have been normalized . ( C ) Influence of luminance fluctuations on the L2 decisions . Legend same as B . ( D ) Psychometric curve . The curve represents the probability that the right segment was selected at the first ( L1 , red curve ) and second ( L2 , light blue curve ) level depending on the strength of the evidence in favor or right vs . left path at the corresponding branching point . Steeper curve indicates better performance at the primary than secondary branches . ( E ) Influence of L1 difficulty onto L2 performance . Psychometric curve for L2 decision was computed separately for easy and difficult L1 trials . Unlike the flat model implemented in the original study , we find no difference in L2 performance ( permutation test; p>0 . 4 ) , in accordance with observed animal behavior . The null effect emerges because the minimum accumulation time before reaching a decision impedes early response even in the presence of strong evidence in L1 ( see Supp . Material ) . When inhibition was removed , a significance interaction was recovered ( Figure 2—figure supplement 1A ) . ( F ) Mean reaction times . Reaction times for the 3 types of difficulty level ( E: easy . I: intermediate , D:difficult ) reproduce those reported in the companion paper by Zylberberg and colleagues ( Zylberberg et al . , 2017 ) . ( G ) Influence of L2 and L2’ difficulty onto L1 choice . Psychometric curve is plotted separately for trials where decision is easier at the left than right secondary branch ( green curve ) , and where decision is easier at right than left secondary branch ( red curve ) . Inset represents the distribution of difference in evidence between the two branches . The effect emerges because strong evidence at one secondary branch will bias the race towards selecting the corresponding final option , thus appearing as a bias in L1 decision towards selecting the primary branch leading to this secondary branch . Information provided at secondary branches biases choice at L1 towards selecting the branch that leads to easier secondary branch , as observed in animal behavior . Lower panel represents the difference between the two psychometric curves . ( Right Panel ) : Influence of difficulty of L2 on the L1 decision across time , as estimated from the logistic regression . Grey shades indicate 95% confidence intervals . L1 choice is affected by difference in perceptual difficulty between L2 and L2’ branches in early visual samples . DOI: http://dx . doi . org/10 . 7554/eLife . 16650 . 00410 . 7554/eLife . 16650 . 005Figure 2—figure supplement 1 . Reciprocal influence of L1 and L2 evidence and decisions . ( A ) Influence of L1 difficulty on L2 performance for model without inhibition . The analysis of interaction between L1 difficulty and L2 performance was repeated in a control network without inhibition ( α=β=0 ) . Threshold is set to six to yield similar level of performance and primacy effect; all other parameters remain unchanged . When removing inhibition performance for difficult L1 trials was degraded comparer to easy L1 trials ( permutation test , p<10−4 ) , exactly as observed in the original simulations of Lorteije and colleagues . DOI: http://dx . doi . org/10 . 7554/eLife . 16650 . 00510 . 7554/eLife . 16650 . 006Figure 2—figure supplement 2 . Behavioral and neural effects for a parameter set yielding longer reaction times than in monkey data . In this original simulation set , commented in Zylberberg et al . , decision threshold is large and there is no minimal viewing time before reaching a decision ( see Methods ) . All qualitative behavioural ( panels A-F ) as well as neural ( panels G-H ) effects are qualitatively similar to those reported in main text ( Figures 2–3 ) and in agreement with experimental findings of [Lorteije et al . , 2015] . DOI: http://dx . doi . org/10 . 7554/eLife . 16650 . 006 The second interesting observation from monkey behavior reported by Lorteije and colleagues was that L1 decisions were biased towards the branch that leads to the easier L2 decision . The effect could only be explained in the hierarchical model by invoking an extra modulatory signal passed on from L2 to L1 that must be carefully tuned ( and was not implemented in the race model of [Lorteije et al . , 2015] ) . By contrast , such effect is readily accounted for by the flat model , as a strong signal at a secondary branch will boost the chances of selecting the corresponding final option , and thus bias towards selecting the L1 path leading to this option . Indeed , the effect was reproduced in the simulations ( Figure 2G ) . This relates to one important advantage of the flat model over the hierarchical one: the flat model , by integrating the strength of evidence from each level , takes optimally into account uncertainty in each of the decisions , leading to more accurate responses . By contrast , the hierarchical model integrates decisions at each level independently of the level of evidence in support of each decision . This can only be amended by ad hoc mechanisms such as that proposed by Lorteije and colleagues , that probably do not scale up adequately when more than two options are available at each level . At the physiological level , neural groups in visual cortex integrated information conveyed at the primary and secondary branching points leading to the path in their receptive field . Selection signals were shown to first differentiate options at the level of secondary branches irrespective of whether the branch is the correct or incorrect one ( L2 or L2' branching ) , and subsequently grow larger for the L2 than for L2' branching . In the hierarchical model , this can only be explained by referring to a modulating signal once decision is reached in L1 that differentially modulates the L2 race signals corresponding to the selected and non-selected L1 branches . In the flat model , such dynamics emerges naturally with activation rectification , because activation corresponding to the two incorrect options in the incorrect L1 branch vanish as L1 signals favor the two alternative options , and so their difference is also reduced ( Figure 3A ) . Indeed , activation rectification plays a role in the dynamics of recorded neural responses , that reach a floor value ( 0 ) in the final part of the integration period in non-selected branches ( Figure 4 of [Lorteije et al . , 2015] ) . Finally , an interaction between L2 and L2' selection signals was reported: clear evidence in favor of one of the two options in L2 reduced the selection signal in L2' , and vice versa . In the hierarchical model , this would require complex modulatory signals passing on from L2 to L1 to L2' . By contrast , both effects are observed in the flat model ( Figure 3B ) : strong evidence for one option will decrease the activation of all other three options through inhibition , increase their chance of collapsing at the zero boundary , and thus reduce the selection signal in the opposite secondary branch . 10 . 7554/eLife . 16650 . 007Figure 3 . Signal properties of the flat race model reproduce original monkey neural data . ( A ) Amplification of L2 signals by L1 choice . Decision signals for L2 ( resp . L2’ ) , depicted here in cyan ( resp . dark blue ) curves , correspond to the difference between the activation of the units coding for the target and distractor in that branch , i . e . TT-TD ( resp . DT-DD ) . Lower panel represents the difference between the two selection signals , i . e . ( TT-TD ) - ( DT-DD ) . While the two decision signals first increase in parallel , the selection signal rapidly turns much larger for the correct than for the incorrect branches , due to the inhibition between activation of options across different branches . Dynamics closely resembles that observed in neural multi-unit activity in V4 . S . e . m . are too small to be visualized . ( B ) Cross-talk between L2 and L2’ selection signals . Selection signal for L2 was reduced when the first visual samples in L2’ provide clear information in favour of one option ( left panel , blue curve for lowerL2' evidence , green curve for higher L2' evidence ) . The effect readily emerges in the flat model because of inhibition between options across the different branches . The converse effect of L2 difficulty impacting L2’ selection signal was also observed ( right panel ) , and both effects reproduce observation from monkey visual cortex . S . e . m . are too small to be visualized . The horizontal bar indicates samples with significant activity difference ( t-test , p<0 . 05 ) . ( C ) Selection signals integrate L1 level of evidence . Unit activity is modulated by evidence at both L1 and L2 levels . Mean activity in each four units ( classified as TT , TD , DT , DD ) for 4 quartiles of strength of L1 evidence . TT and TD units show positive modulations with L1 evidence , while DT and DD units display negative modulations , consistent with the fact that larger L1 evidence biases competition towards the two options in the correct L1 branch . Right panel displays absolute value of Spearman’s rho correlation between strength of evidence and mean activity for all four branches . DOI: http://dx . doi . org/10 . 7554/eLife . 16650 . 007 Overall we show that all four observations that were taken as evidence in favor of the hierarchical model could be accounted for with a standard race-model implementation of the flat model . These observations were robust and did not rely on fine tuning of the parameters . Table 1 summarizes how each of these observed properties depend on the features of the model . Note that all these features are classical constituents of race models and not ad hoc mechanisms . Overall the flat model provides a more parsimonious explanation of the data , as it does not require appending modulatory signals between parallel decisions as the hierarchical model does . 10 . 7554/eLife . 16650 . 008Table 1 . Each of reported behavioral and neural effects and the associated features from the flat decision model required to display such effects . DOI: http://dx . doi . org/10 . 7554/eLife . 16650 . 008EffectRequired model featuresIndependence of L2 from L1 difficultySignal-independent noise , low dependence of RTs on L1 difficultyBias of L1 choice by L2 difficultynoneAmplification of L2 signals by L1 choiceActivation rectificationCross-talk between L2 and L2' signalsActivation rectification These behavioral and neural effects reported in the original study provide however no univocal evidence in favor of either model . One fundamental difference between the two is indeed that selection signals in the flat model mix evidence from both L1 and L2 branches , while the hierarchical model predicts unmixed selection signals ( the influence across branches can only occur posterior to decision ) . [At this point , an important distinction has to be made in the flat model between localized activations , which indeed mix evidence from both branches , and selection signals at L2 , extracted by looking at the difference between two units in the same L1 branch , and which as shown above are largely insensitive to L1 signals] . In the companion paper , Zylberberg et al . now provide data showing that selection signals in at least 3 out of 4 L2 branches mixes evidence from both L1 and L2 branches ( their Figure 4 ) . We believe this observation is most compatible with the flat model and by itself rules out the hierarchical model that relies on complete neural segregation of integration of L1 and L2 evidence ( although the possibility remains that these level-mixing integrative neural signals are completely non-causal in monkey decisions ) . Our simulations reproduce these effects: signals in TT is larger for stronger L1 evidence , while signals in DT and DD are weaker ( p<10−9 , Figure 3C; see Also Appendix 2 ) . As pointed out by Zylberberg and colleagues , our simulations also display a positive modulation in TD signals , unlike the null effect found in monkey data . However , despite a single source of inhibition , the modulation was not equally strong in all four branches: indeed , it was by far the weakest precisely in the TD branch ( Figure 3C ) . This weaker effect may have explained the lack of significance in monkey data . Moreover , modulation in TD could be reduced or abolished if the flat model implied stronger inhibition between options related to the same L1 branch ( TT-TD and DT-DD ) than between options related to distinct L1 branches ( e . g . TT-DT , TT-DD ) . Stronger inhibition between local circuits is indeed a general pattern of cortical connectivity ( Douglas and Martin , 2004; Lund et al . , 2003 ) . Zylberberg and colleagues produced a last analysis of the original dataset , showing that in the flat model errors in L1 are associated with higher sensitivity at L2 , whereas monkeys display no such effect . While we acknowledge that this result challenges the current implementation of the flat model , it is at this point equally unknown whether the hierarchical model proposed by the authors could avoid this feature ( the reward maximization introduced in the hierarchical model to account for the influence of L2 bias onto L1 choice may produce the same interaction ) . It is unclear whether selection signals recorded in visual cortex are generated locally or reflect feedback process from higher regions where integration of evidence takes place . The flat model is arguably more consistent with the latter hypothesis , as it implies integration from and inhibition across distant locations in visual field , which are more typically associated with higher cortical regions ( Wimmer et al . , 2015 ) . In any case , these selection signals certainly represent neural markers for a specific integration process along one subbranch . One important limitation of flat models is they require to learn high-order representations of the environment that can appropriately integrate evidence from all relevant sensory sources , i . e . the structure of connectivity described in Figure 1D may take time to be acquired ( Garrard et al . , 1997; McClelland and Rumelhart , 1985 ) . Here , monkeys performed ~60 , 000 trials each , possibly long enough to learn appropriate global representations linking all visual cues relevant to each response . When such time is not afforded , the system may rely on less efficient strategies such as a hierarchical model . Indeed , the comparison of flat and hierarchical models in this visual task sheds new light to the ancient debate of whether information from distinct sources is integrated before or after the decision stage , analogous respectively to the flat and hierarchical strategies . In multimodal integration such as object localization or motion detection , different modalities provide complementary cues about common objects or features of the environment ( object motion , phoneme identity , etc . ) , so over the lifetime the brain can learn the appropriate crossmodal representations and integrate bimodal information directly over those representations . Indeed , in such context both sources of information are merged prior to decision ( Körding and Wolpert , 2004 ) . By contrast , when subjects must detect the presence of either a visual or an auditory cue , requires to mapping arbitrarily two distinct unimodal signals into a single response , the relevant crossmodal representations are not formed , and therefore integration could by default only be formed following unimodal decisions , as it has been found experimentally ( Otto and Mamassian , 2012 ) . One important hypothesis we make is that the level of integration could strongly depend on experience , gradually switching from post-decision ( i . e . hierarchical ) to pre-decision ( i . e . flat ) integration . Finally , an important distinction has to be made between the hierarchical vs . flat nature of the integration process , and the serial vs . parallel nature of the integration process . While the former has to deal with the level and the structure of the representations at which integration takes place before reaching a decision , the latter corresponds to whether evidence from different sources can be integrated at the same time in these representations . The flat model is perfectly compatible with serial integration of evidence , for example if attentional constraints limit the capacity to process evidence from distant visual locations , such as when subjects must integrate from three distinct levels ( Zylberberg et al . , 2012 ) . In summary , despite premature conclusions , we think that the experimental framework developed by Lorteije and colleagues opens a new promising venue for understanding the level at which evidence is accumulated and individual perceptual decisions are taken in ecological settings . We present here the equations governing the flat race model of decision-making used for simulations . Our implementation of the flat model departs from the one in [Lorteije et al . , 2015] in the following aspects , which are typically subsumed in many standard implementations: ( 1 ) added noise is constant instead of being proportional to signal strength , ( 2 ) there is a positive loop ( i . e . negative leak ) in the integration process and mutual inhibition between competitors and ( 3 ) activities cannot be negative ( i . e . activation rectification ) . The variable xti represents the activity of the activation unit i ( from 1 to 4 ) at sample t , and evolves according to:Δxti=I0+αxt−1i−β∑j≠ixt−1j+Iti+N ( 0 , σ ) where I0 is the constant input , α is the auto-excitation term ( negative if leaking , positive if positive loop providing bistability ) , β is the inhibition strength , Iti is the sensory evidence at sample t , and the last term represents white noise of variance σ . Inhibition is homogeneous and all-to-all between activations units , that is , each unit inhibits equally the other unit from the same branch and the two units from the alternative branch . Sensory evidence integrates information from the primary and secondary branching points:Iti=k1ati+k2bti k1 and k2 represent the sensitivity to evidence at level 1 and 2 respectively , ati represents the information provided at level 1 in favor of the corresponding branch ( i . e . the difference between instantaneous luminosity in that branch and luminosity in the alternative branch at sample t ) , bti represents the information provided at level 2 in favor of the corresponding branch . We enforce non-negative values for unit activity by using rectification:xti=max ( xt−1i+Δxti , 0 ) All units are initiated from the same starting point x0i=0 . 5 , and the decision is reached whenever any of the units reaches threshold K ( Figure 1C ) . If no decision is reached after presentation of all samples , the response corresponding to the unit with highest final activity is selected . Parameters are: input strength at primary branching point k1 = 0 . 016 , at secondary branching points k2 = 0 . 012 , auto-excitation α =0 . 1 , inhibition β = 0 . 07 , constant input I0= 0 . 5 , decision threshold = 50 , initial values x0i= 0 . 5 , white noise variance = 1 . We chose parameters to reproduce qualitatively the monkey behavioral and neuronal data . We simulated 100 , 000 trials , which is the same order of magnitude as used in the original experiment . All analyses performed on simulated data strictly reproduced those realized in the original experiment . Results from these analyses are presented in Figure 2—figure supplement 2 . In the simulation set of Figure 2 , designed to reproduce the pattern of reaction times described in the companion paper , we lowered the decision threshold to 30 to produce shorter reaction times . We introduced a minimum integration time of 10 samples ( i . e . no boundary for the first 10 samples ) , consistent with the minimal time of 500 ms after stimulus onset monkeys had to wait before performing a responses saccade . We also changed inhibition to 0 . 05 and auto-excitation to 0 . 12 . All other parameters were unchanged . A matlab script for all simulations and analyses is available at http://bit . ly/hyafilmoreno2017 .

Should you go for coffee with Jules , or go to the movie theater with Jim ? Both options require you to make additional decisions , for example , which café would you go to , or what movie could you see ? Many of our day-to-day decisions have multiple layers of sub-decisions embedded within them that are not necessarily independent . Our opinions of the cafés in town and the movies showing at the theater may influence our decision over whom to spend the afternoon with . In 2015 , researchers at the Netherlands Institute for Neuroscience performed experiments in macaques to try to work out how the brain makes these decisions . The monkeys learned to choose between two visual stimuli ( decision 1 ) . The outcome of decision 1 determined whether the animals then had to make decision 2 or decision 3 . The results suggested that the monkeys initially made all three comparisons independently and in parallel , before combining the evidence to select their overall strategy . This process is referred to as hierarchical decision-making . In the analogy above , one would compare the relative merits of Jules versus Jim , café A versus café B , and a horror movie versus a comedy at the same time before deciding what to do . Hyafil and Moreno Bote have now reanalyzed the data published in 2015 using new computer simulations . This second analysis suggests the results are in fact more consistent with an alternative model of decision-making called a flat model , in which the brain compares all of the final options simultaneously ( Jules + café A; Jules + café B; Jim + horror movie; Jim + comedy ) before choosing between them . Making decisions by comparing the final outcomes becomes easier as the brain learns through experience to associate stimuli that often occur together . Hyafil and Moreno Bote hypothesize that in response to a new situation , the brain may sometimes start off by using hierarchical decision-making before switching to a more accurate flat model as experience allows . In response to the findings of Hyafil and Moreno Bote , the researchers who conducted the work reported in 2015 have also reanalyzed the original data , and carried out a new experiment in human volunteers . They argue that the flat model provides a poor fit to the original data and struggles to explain the new data . Future studies can build on these conflicting findings by further exploring the limits of parallel decision-making , which may help us to understand how the brain is able to make multiple decisions while keeping the future consequences in mind .

2017

Breaking down hierarchies of decision-making in primates

‘Normal’ genomic DNA contains hundreds of mismatches that are generated daily by the spontaneous deamination of C ( U/G ) and methyl-C ( T/G ) . Thus , a mutagenic effect of their repair could constitute a serious genetic burden . We show here that while mismatches introduced into human cells on an SV40-based episome were invariably repaired , this process induced mutations in flanking DNA at a significantly higher rate than no mismatch controls . Most mutations involved the C of TpC , the substrate of some single strand-specific APOBEC cytidine deaminases , similar to the mutations that can typify the ‘mutator phenotype’ of numerous tumors . siRNA knockdowns and chromatin immunoprecipitation showed that TpC preferring APOBECs mediate the mutagenesis , and siRNA knockdowns showed that both the base excision and mismatch repair pathways are involved . That naturally occurring mispairs can be converted to mutators , represents an heretofore unsuspected source of genetic changes that could underlie disease , aging , and evolutionary change . Species survival depends on the faithful replication of genetic information which is monitored and maintained by a number of complex and interacting DNA repair pathways ( Modrich , 2006; Cannavo et al . , 2007; Cortázar et al . , 2007; Hsieh and Yamane , 2008; Kunz et al . , 2009; Robertson , 2009; Liu et al . , 2010; Jacobs and Schar , 2012; Jiricny , 2013 ) . Continual DNA repair is required to correct the thousands of genetic lesions that occur daily due to just the inherent chemical lability of DNA ( Atamna et al . , 2000; Barnes and Lindahl , 2004 ) . For example , the susceptibility of C ( and its methylated derivative ) to spontaneous hydrolytic deamination daily generates hundreds of U/G and T/G mismatches respectively , ( Barnes and Lindahl , 2004 ) and could explain why C is the most frequent source of single nucleotide substitutions in mammals ( Hwang and Green , 2004 ) . Error-free base excision repair ( BER ) can correct the naturally occurring U/G and T/G mismatches ( reviewed in Cortázar et al . , 2007; Hegde et al . , 2008; Robertson , 2009; Jacobs and Schar , 2012 ) . The basic reaction involves removal of the base that is paired with G by anyone of several glycosylases to generate an abasic site that is cleaved on its 5′ side by the APE1 endonuclease . The resulting 3′ end is extended by insertion of dCMP by the high fidelity polymerase β coincident with its hydrolysis of the 5′-phosphodeoxyribose that had been generated by the glycosylase . This step is followed by sealing the resulting single stranded break ( SSB ) by DNA ligase III with the assistance of the scaffolding protein XRCC1 . However , in lymphoid ( B ) cells the U/Gs that are generated by activation-induced cytidine deaminase ( AID , a member of the AID/apolipoprotein B mRNA editing enzyme , catalytic polypeptide-like ( APOBEC ) family of cytidine deaminases , Conticello , 2008 ) on transient single-stranded DNA regions produced during transcription , are prone to several mutagenic processes that enhance diversification of immunoglobulins ( somatic hypermutation , SHM , Martomo and Gearhart , 2006; Teng and Papavasiliou , 2007; Peled et al . , 2008 ) . One of these involves processing U/G mismatches , or BER products thereof , by a non-canonical application of the mismatch repair ( MMR ) pathway . Normally , MMR ( Modrich , 2006; Hsieh and Yamane , 2008; Jiricny , 2013 ) is a high fidelity process that operates post-replicatively on the nascent DNA strand to remove mismatches that have escaped the proof-reading activity of high fidelity replicative DNA polymerases . Essential components of this pathway include the heterodimer MutSα ( MSH2 and MSH6 ) , which recognizes mismatches , the heterodimer MutLα ( MLH1 and PMS2 ) , which accesses the mismatch-containing nascent strand in a reaction mediated by proliferating cell nuclear antigen ( PCNA , a multipurpose replication clamp , e . g . , Moldovan et al . , 2007; Lee and Myung , 2008 ) . PCNA also activates a latent endonuclease in MutLα ( Kadyrov et al . , 2006; Pluciennik et al . , 2010 ) , which provides entry points for the EXO1 nuclease that excises the mismatch-containing nascent strand to expose the replication template for re-copying by a high fidelity DNA polymerase , such as polymerase δ . In B cells , non-canonical MMR can expose single stranded regions at U/G-containing sites unrelated to DNA replication that can serve as a template for DNA repair . However , the high fidelity DNA polymerase is replaced by the error-prone polymerase η ( mediated by mono-ubiquitylated PCNA ) . Thus , MMR is subverted to an error-prone process that contributes to SHM . Recently , elements of non-canonical MMR have been recapitulated in vitro ( Schanz et al . , 2009; Peña-Diaz et al . , 2012 ) , and the latter study also showed that extracts of non-lymphoid mammalian cells can also process U/G mismatches by a non-canonical MMR process . In addition , these non-lymphoid cells when stressed in vivo by the alkylating agent , N-methyl-N′-nitro-N-nitrosoguanidine ( MNNG ) , generated more mutations in MMR proficient than deficient cells , thereby implicating MMR in the mutagenic process but presumably using an AID-independent mechanism ( Hsieh , 2012 ) . As DNA homeostasis would seem to require a continual state of DNA repair , its involvement in error-prone processes even at a low frequency would have important implications for the mutational mechanisms that could underlie evolution , aging , and disease . Of interest in this regard is the mechanism that produces the enhanced mutation rate that characterizes certain tumors , which has been termed the ‘mutator phenotype’ ( Bielas et al . , 2006; Venkatesan et al . , 2006 ) . Recent examples of such a mutator phenotype are the high mutation rates of the C of TpCpN ( or its complement , G of NpGpA ) that can accompany the progression of some cancers ( Nik-Zainal et al . , 2012; Roberts et al . , 2012 ) . TpC-preferring members of the AID/APOBEC family of C deaminases , particularly APOBEC3B ( A3B ) , mediate these mutations ( Burns et al . , 2013; Leonard et al . , 2013; Roberts et al . , 2013; Taylor et al . , 2013 ) , which can occur in strand-coordinated clusters . These deaminases prefer single stranded DNA and an important issue is how the single stranded APOBEC substrate is generated . Experiments using yeast as a model system showed that the transitory single stranded regions that arise at replication forks or during double strand break repair can accumulate such mutations upon chronic alkylation of DNA ( Roberts et al . , 2012 ) . These mutations were also observed at experimentally introduced double strand breaks coincident with overexpression of AID/APOBEC deaminases ( Taylor et al . , 2013 ) . Here , we determined directly whether repair of the naturally occurring T/G and U/G mismatches would be mutagenic to flanking DNA in mammalian cells that were not stressed by genotoxic agents . We introduced these mismatches ( and other mispairs and lesions ) into an SV40-episome that can replicate in human cells . Numerous studies have shown that processing lesions harbored by such episomes ( as well as those introduced in the SV40 virion ) faithfully captured the DNA repair repertoire and mutational environment of its host ( e . g . , Hare and Taylor , 1985; Seidman et al . , 1985; Brown and Jiricny , 1987; Choi and Pfeifer , 2005; Terai et al . , 2010; Pathania et al . , 2011; Qi et al . , 2012 ) . We found that the introduced mispairs were invariably corrected . However , their repair generated mutations in normal flanking DNA at statistically higher rates than the no mismatch control , but only if the episomes were passed through mammalian cells . Regardless of the lesion , most of the mutations involved the C of TpCpN , and shared other features of the aforementioned mutations that occur in some cancers . siRNA knockdowns showed that the TpC-preferring deaminases , particularly A3B mediated the mutagenic effect , and chromatin immunoprecipitation ( ChIP ) showed that A3B could access the episome in a mismatch dependent way . siRNA knockdowns also showed that components of both the BER and MMR pathways are involved in generating the single-stranded APOBEC substrate and that two factors that have diverse roles in various aspects of DNA metabolism , ataxia telangiectasia and Rad3-related protein ( ATR ) and PCNA ( Moldovan et al . , 2007; Flynn and Zou , 2011 ) modulate the mutagenic effect . The APOBEC deaminases are relatively ubiquitous in various tissues ( Refsland et al . , 2010 ) , as is the potential for generating single-stranded templates from BER processed lesions . These would not only include the mispairs generated from C and methyl-C by their spontaneous deamination as we found here , but potentially also those involving other normal metabolites of methyl-C ( e . g . , Guo et al . , 2011; Wu and Zhang , 2014 ) and the thousands of BER substrates that arise daily from other naturally occurring degradative processes that affect DNA ( Atamna et al . , 2000; Barnes and Lindahl , 2004 ) . The implications of our findings in light of these issues are discussed . Figure 1A shows the SV40-based episome ( shuttle vector ) ( Parris and Seidman , 1992 ) into which we inserted a mismatch region ( MM1 ) . Both strands of MM1 contain two sites for a different pair of single strand restriction enzymes , which facilitate the exchange of either strand with an exact complement to generate a no mismatch ( 0 MM ) control or with an oligonucleotide that would generate a mismatch or other lesions ( Figure 1B , ‘Materials and methods’ , Figure 1—figure supplements 1 and 2 ) . Supplementary file 1 lists the oligonucleotides used and their corresponding numbers are given in the figures . After passage in mammalian cells we screened the episomes for mutations by blue/white selection in E . coli ( Figure 1B ) and determined the DNA sequence of the reporter cassette of the episomes from all the white colonies . 10 . 7554/eLife . 02001 . 003Figure 1 . Procedure to determine mutagenic effect of DNA repair . ( A ) Episome FM1: purple double line with square , supF promoter region ( Pr ) ; red rectangle , supF gene; gray rectangle , reporter region; green box , mismatch region 1 ( MM1 ) ; yellow rectangle , bar code; dark blue rectangle , pBR327 origin ( ori , FM1 replication in E . coli ) ; black square and contiguous heavy black arrow , SV40 origin-promoter and T-antigen coding sequence respectively ( FM1 replication in human cells ) . We refer to this episome as FM1 because the supF gene is upstream of the MM1 . Not shown , the ApR gene , which renders E . coli resistant to ampicillin . ( B ) Steps to generate lesion-containing vectors and procedure for determining the mutagenic effect of DNA repair in mammalian cells: Mismatch region with nicking sites ( vertical arrows ) is digested with a single strand restriction enzyme on the top ( or bottom ) strand , and the nicked strand is removed by hybridization to a 5′ biotin ( blue diamond ) labeled complementary DNA ( red ) . The hybrid is then tethered to a streptavidin ( green polygon ) -coated magnetic bead ( gray oval ) . The purified gapped episome is reconstituted by ligation to its perfect complement or an oligonucleotide that contains one or more mismatches to generate vectors with a top ( or bottom ) strand lesion or its corresponding 0 MM control . These vectors are transferred to mammalians cells , harvested after 48 hr and subjected to blue/white screening . The procedure is described in detail in the ‘Materials and methods’ . DOI: http://dx . doi . org/10 . 7554/eLife . 02001 . 00310 . 7554/eLife . 02001 . 004Figure 1—figure supplement 1 . Monitor the gapping and reconstitution of episomes by KpnI digestion . Summary of the steps in the construction of the lesion-containing episomes shown in Figure 1B . Gel electrophoresis shows the loss and restitution of the KpnI site between the Nt/b . BbC1 nicking restriction enzyme sites during the gapping and reconstitution of the episomes with various oligonucleotides ( lanes 2–7 ) . The reconstituted plasmids ( lanes 4–7 ) yield essentially the same restriction pattern as the starting vector ( lane 2 ) , whereas the gapped plasmid ( lane 3 ) yields only traces of the 3 . 8 kb and 1 . 3 kb KpnI fragments seen with FM1 . DOI: http://dx . doi . org/10 . 7554/eLife . 02001 . 00410 . 7554/eLife . 02001 . 005Figure 1—figure supplement 2 . Monitor the presence of mismatch generating oligonucleotides by AatII digestion . The mismatches in some oligonucleotides eliminated an AatII restriction site in the mismatch region . In these cases , we could monitor the reconstitution of the episome by its resistance to AatII digestion . Lanes 1–3 , AatII and StuI double digestion of FM1 reconstituted with a 0 MM oligonucleotide or with ones that contained either 2 T/G or 2 U/G mismatches . The 0 MM control yields 3 . 4 kb and 1 . 6 kb AatII and StuI fragments ( lane 1 ) , but the mismatch containing plasmids yield only a 5 . 0 kb StuI fragment ( lane 2 , 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02001 . 005 Figure 2A shows that the repair of different types of mismatches or lesions - T/G , 5-hydroxymethyl-U ( hmU ) /G , U/G , or an abasic site opposite a G , ( ab ) /G – induced significantly more mutations in the reporter region than we found with the 0 MM control . HmU is a byproduct of the enzymatic demethylation of methyl-C ( Bhutani et al . , 2011; Guo et al . , 2011; Wu and Zhang , 2014 ) and abasic sites are generated during BER ( e . g . , Robertson , 2009; Jacobs and Schar , 2012 ) . For convenience , we refer to both mismatches and ab/G sites as lesions . Mutation frequency is the percent ( % ) white colonies ( per total screened ) that contained undeleted episomes . We did not consider deletions because most were missing all or part of the reporter region . These deletions had resulted from our initial method of vector preparation and were essentially eliminated by its subsequent modification ( Figure 2—figure supplement 1 and ‘Materials and methods-Vector preparation’ ) . The few percent that persisted were unrelated to either the type or even presence of an introduced DNA lesion ( see ‘Materials and methods-Data acquisition and analysis’ ) . Finally , no mutated episomes were obtained if they were passed directly into E . coli ( 3 deleted episomes /42 , 000 colonies screened , results not shown ) . 10 . 7554/eLife . 02001 . 006Figure 2 . Mutagenic effect of DNA repair . ( A ) Dot plots of the repair-induced mutation frequency ( number of white colonies with undeleted episomes/total—see ‘Materials and methods’ and text ) as a function of the indicated lesion . Each dot represents a separate trial of a given type of lesion that was present in 1 , 2 , or in the case of T/G , 3 copies in various positions ( and thus sequence contexts ) in the mismatch region . See Figure 2—figure supplement 2; Supplementary file 1 . The red horizontal lines indicate the mean , and p values were calculated by the Fisher exact test on pairwise 2 × 2 contingency tables of the total number of non-mutant colonies ( blue ) and total number of white colonies that contained undeleted episomes . We applied a Bonferroni correction for multiple comparisons by multiplying the p values by the number of comparisons for the top or bottom strand lesions . The diagrams above the dot plots depict the reporter cassette . The green triangle indicates the location of the introduced lesion , and the arrow indicates the 5′ single-stranded break that would be generated by BER . ( B ) The fate of the introduced mismatches in the MM region present in the mutated plasmids ( pooled according to the type and strand location of the lesions ) . The percentages of lesions restored to C/G ( blue ) or converted to T/A ( orange ) are shown along with the numbers of introduced lesions . ( C ) Configurations of episomes FM1 , M1F and FM1_R , and bar graphs of the mutation frequency as a function of the indicated lesion introduced on the top or bottom strand ( pooled according to the type and strand location of the lesions ) . Means ( ±SEM ) are shown from at least two independent experiments . The bar graphs shown for FM1 are the same data as the dot plots in panel A . The distances between the lesions and the cis 3′ end of the reporter are given—the diagrams of the episomes are not to scale . DOI: http://dx . doi . org/10 . 7554/eLife . 02001 . 00610 . 7554/eLife . 02001 . 007Figure 2—figure supplement 1 . Minimal manipulation of reconstituted episomes minimizes the generation of deletions . Frequency of the supF-deleted colonies from the two-step ( ligation of the bar code region followed by Bstz17I digestion ) or a one-step ( ligation only ) procedure for reconstituting the episomes . See ‘Materials and methods’ for more details . DOI: http://dx . doi . org/10 . 7554/eLife . 02001 . 00710 . 7554/eLife . 02001 . 008Figure 2—figure supplement 2 . The magnitude of repair-induced mutagenesis differs with the type of lesion but the number or nucleotide context of a given lesion does not materially alter its mutagenic effect . The mutation frequency ( number of white colonies with undeleted plasmids/total ) as a function of the indicated lesions on the top or bottom strand of pFM1 . Means ( ±SD ) from at least two independent experiments are shown . Note the difference in the scales between the various bar graphs . The number of oligonucleotide ( oligo # ) used to reconstitute the shuttle vector is given across the top each bar graph and the sequences of the reconstituting oligonucleotides are listed in Supplementary file 1 . The nucleotide context and number of each type of legion in oligonucleotides 6 , 19 and 9 are identical . p-values were calculated using the Fisher exact test . DOI: http://dx . doi . org/10 . 7554/eLife . 02001 . 00810 . 7554/eLife . 02001 . 009Figure 2—figure supplement 3 . The mutagenic effect is not affected by the C+G content of the mismatch region . Mutation frequency as a function of the indicated lesions on the top strand of vectors FM1 , FM2 and FM3 with MM regions of different C+G content . Means ( ±SD ) from at least two independent experiments are shown . The data were pooled with respect to lesion type . DOI: http://dx . doi . org/10 . 7554/eLife . 02001 . 00910 . 7554/eLife . 02001 . 010Figure 2—figure supplement 4 . The mutagenic effect of single strand break repair supersedes that of T/G repair . Left panel , mutation frequency as a function of the indicated T/G or T/G with a SSB . The top strand of the FM1 episome was reconstituted with either phosphorylated or non-phosphorylated oligonucleotides that would produce one or more T/G mismatches without or with a SSB . Means ( ±SD ) from at least two independent experiments are shown and the number of the oligonucleotide ( oligo # ) used to reconstitute the shuttle vector is given across the top each bar graph . The right panel summarizes the repair outcome for each condition ( i . e . , 1 , 2 and 3 T/Gs without or with an SSB ) in the MM region of the mutated episomes . The numbers of introduced lesions and their percent restoration to C/G or conversion to T/A are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 02001 . 010 The relative mutagenic effect of repairing a given type of lesion on a given strand ( top or bottom ) was more or less indifferent to its number ( up to three ) , context ( i . e . , CpG or non-CpG ) or position ( s ) in the MM region ( Figure 2—figure supplement 2; Supplementary file 1 ) . Thus , the mutagenic effects of repairing the various configurations of T/G , hmU/G and U/G shown in Figure 2A can be recapitulated by a given set of these lesions ( i . e . , 2 T/Gs , 2 hmU/Gs or 2 U/Gs at the same positions in the MM region , the uppermost bar graph of Figure 2—figure supplement 2 ) . 10 . 7554/eLife . 02001 . 011Figure 3 . Mutational spectrum induced by DNA repair of top strand lesions of FM1 . Dot plots of the fraction of total mutations contributed by each base ( squares ) and to which base it was mutated ( mutational fate , circles ) induced by the restoration of T/G ( panel A ) or U/G ( panel B ) to C/G on the top strand of FM1 ( pooled according to the type of the lesions ) . In cases where the base has mutated to transitions and transversions at the same frequency the circles overlap . The beginnings of the promoter ( pro ) , supF and MM regions are indicated . The promoter bases are highlighted in yellow and the MM bases in gray . The supF bases that are highlighted with the same colors indicate the complementary stem-encoding regions . The inverted black triangles indicate the locations of the mismatches in the pooled data set ( see text ) . The numbers of sequences and mutations are shown at the top of each panel . As the repair of multiple lesions on a given episome could generate a singly mutated plasmid , the number of mutated sequences could be less the number of introduced lesions ( Figure 2B ) . ( C ) Fraction of bases ( in terms of top strand sequence ) that were mutated in response to the repair ( i . e . , either restored to C/G or , with respect to T/G , converted to T/A ) of the indicated lesions on the top strand of FM1 ( pooled according to the type of lesion ) . ( D ) Dinucleotide context of the mutated base with respect to its 5′ or 3′ base . Fractions of the mutated base induced by the restoration of top strand lesions as a function of its dinucleotide context are shown ( pooled from all the lesions restored to C/G ) . The faint purple bars show the fractions of all 16 dinucleotides in the reporter region . Means ( ±SEM ) are from eighteen alignments ( 343 sequences , 478 mutations ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02001 . 01110 . 7554/eLife . 02001 . 012Figure 3—figure supplement 1 . Dinucleotide context of mutations induced by the restoration of the indicated lesion generated by reconstituting the FM1 episome on the top strand with the indicated oligonucleotides . The 5′ and 3′ dinucleotide context of the mutated base induced by restoration of the indicated lesions introduced by the oligonucleotides shown in the uppermost bar graph of Figure 2—figure supplement 2 ( each generate 2 lesions at the same sites in the MM region ) and oligonucleotide #16 , which introduces a single ab/G mismatch . The number of the oligonucleotide ( oligo # ) used to reconstitute the shuttle vector is given on the right side . The faint purple bars show the fractions of all sixteen dinucleotides in the reporter region . DOI: http://dx . doi . org/10 . 7554/eLife . 02001 . 01210 . 7554/eLife . 02001 . 013Figure 3—figure supplement 2 . Dot plots of mutations induced by the restoration of bottom strand T/G or U/G on FM1 . Dot plots of the fraction of the pooled total mutations contributed by each base ( squares ) and to which base it was mutated ( mutational fate , circles ) induced by the restoration of lesions to C/G that had been introduced on the bottom strand of FM1–T/G ( top panel ) or U/G ( bottom panel ) . In cases where the base has mutated to transitions and transversions at the same frequency the circles overlap . The beginnings of the promoter ( pro ) , supF and MM regions are indicated . The promoter bases are highlighted in yellow and the MM bases in gray . The supF bases that are highlighted with the same colors indicate the complementary stem-encoding regions . The inverted black triangles indicate the locations of the mismatches in the pooled data set . The numbers of sequences and mutations are shown at the top of each panel . DOI: http://dx . doi . org/10 . 7554/eLife . 02001 . 01310 . 7554/eLife . 02001 . 014Figure 3—figure supplement 3 . Mutational spectra induced by the repair of lesions on the bottom strand of FM1 . Fraction of bases ( in terms of top strand sequence ) that were mutated in response to the repair of the indicated lesions on the bottom strand on FM1 . The data were pooled with respect to lesion type . DOI: http://dx . doi . org/10 . 7554/eLife . 02001 . 01410 . 7554/eLife . 02001 . 015Figure 3—figure supplement 4 . Dinucleotide context of mutations induced by the restoration of bottom strand lesions on FM1 . Fractions of the mutated base induced by the restoration of bottom strand lesions as a function of its dinucleotide context are shown ( pooled from all the lesions that were restored to C/G ) . The faint purple bars show the fractions of all 16 dinucleotides in the reporter region . Means ( ±SEM ) are shown from 13 alignments ( 197 sequences , 293 mutations ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02001 . 01510 . 7554/eLife . 02001 . 016Figure 3—figure supplement 5 . Dinucleotide context of mutations induced by the conversion of top or bottom strand T/G to T/A . Top and bottom panels show respectively the dinucleotide context of the mutations recovered from the conversion of T/G to T/A on the top strand or bottom strand of FM1 . The faint purple bars show the fractions of all 16 dinucleotides in the reporter region . The data are pooled from all the T/Gs on the top or bottom strand that are converted to T/A . DOI: http://dx . doi . org/10 . 7554/eLife . 02001 . 01610 . 7554/eLife . 02001 . 017Figure 3—figure supplement 6 . Most mutations induced by SSB repair involve the C of repair template TpCs . Top and bottom panels show respectively the dinucleotide context of the mutated base induced by the repair of a SSB on the top strand of FM1 or the bottom strand of M1F . For top strand lesions , mutations involve the G of GpA , the complement of the C of TpC ( see Figure 4 main text ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02001 . 01710 . 7554/eLife . 02001 . 018Figure 3—figure supplement 7 . The 0 MM mutational signature is consistent with the mutational spectra of repairing a mixture of top and bottom strand SSBs . The dinucleotide context of the mutations recovered from the 0 MM controls reconstituted on the top strand displays features of the dinucleotide contexts induced by both top and bottom strand SSBs ( compare with top and bottom panels of Figure 3—figure supplement 6 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02001 . 01810 . 7554/eLife . 02001 . 019Figure 3—figure supplement 8 . Trinucleotide context of mutated G induced by the restoration of top strand lesions on FM1 . The trinucleotide context of the G mutations induced by restoration of top strand lesions to C/G in Figure 3D ( pooled from all the lesions that are restored to C/G ) . Means ( ±SEM ) are from 18 alignments ( 437 mutations ) . The faint purple bars show the fractions of the 16 trinucleotide contexts of G in the reporter region . DOI: http://dx . doi . org/10 . 7554/eLife . 02001 . 019 The top or bottom strand location of the lesions differentially affected their mutagenic effect: repair of top strand T/Gs generated ∼fourfold more mutated episomes than the 0 MM control ( respective mean percent mutation frequency 0 . 107 vs 0 . 025 ) . HmU/G repair induced ∼sevenfold more mutations ( 0 . 182% ) , and U/G or ab/G repair induced ∼30-fold more mutations ( 0 . 726% and 0 . 78% respectively ) than the 0 MM control . These differences were statistically significant ( legend , Figure 2A ) . The bottom strand location reduced the mutagenic effect of repairing U/G or ab/G , but not of T/G or hmU/G . Repair of bottom strand T/G was ∼twofold more mutagenic than on the top strand: 0 . 23% ± 0 . 013 vs 0 . 11% ± 0 . 011 ( mean ± standard error , p<10−7 , Fisher exact test ) . The differences in mutagenesis were not due to differences in repair efficiency . Figure 2B shows that , except for T/G , essentially all of the top and bottom strand lesions were restored to C/G . T/G was converted to T/A about 25% of the time . This would result from using the T-containing strand as the repair template ( see section , ‘The mutational spectrum , sequence context , and fate of the bases mutated in response to DNA repair are consistent with APOBEC-mediated mutagenesis’ ) . The mutated plasmids represented repair of 426 T/G , 116 hmU/G , 293 U/G and 90 ab/G mismatches ( combined top and bottom lesions , Figure 2B ) , and the percent restoration to the C/G base pair was: 72 . 3 , 99 , 100 and 100 respectively . The values for the restoration of lesions in the non-mutated episomes ( isolated from blue colonies ) , were 75% of 111 T/G , 95 . 8% of 24 hmU/G , 99% of 105 U/G , and 100% of 7 ab/G mismatches . Thus , the ratio of restoring T/G to C/G or converting it to T/A was independent of repair-induced mutagenesis . Conversion of T/G mismatches to T/A had also been observed with T/G-containing SV40 virion DNA ( e . g . , Hare and Taylor , 1985; Brown and Jiricny , 1987 ) . The mutagenic effect was also independent of the G+C content of the mismatch region; 69% G+C for FM1 , 39% for FM2 , and 53% for FM3 ( Figure 2—figure supplement 3 ) . Thus , mutagenesis was not affected by the presumed stability of the MM region helix . The distance between the cis 3′ end of the reporter region and the lesion is <50 bp for top-strand , but ∼5 kb for bottom-strand lesions in FM1 ( top panel Figure 2C ) . This difference accounts for their different mutagenic effects because relocating the MM region upstream of the reporter region ( M1F ) switches these distances ( now <50 bp for bottom—but ∼5 kb for top-strand lesions respectively ) , and it also switches the extent of their respective mutagenic effects ( cf . top and middle panels of Figure 2C ) . Reversing the orientation of the entire reporter cassette ( the DNA between the EcoRI and NheI sites [Figure 1A] to produce FM1_R [Figure 2C , lower panel] ) has the same effect . These results also show that strand-specific mutagenesis is neither due to transcription nor DNA replication effects ( i . e . , transcribed vs non-transcribed strand , and leading vs lagging strand , e . g . , Fijalkowska et al . , 1998; Hanawalt and Spivak , 2008 ) imposed on the reporter region by the episome backbone . BER would be recruited to T/G , hmU/G and U/G mismatches ( Robertson , 2009; Jacobs and Schar , 2012 ) . Removal of the mismatched T , hmU , or U by a glycosylase and cleavage of the DNA 5′ of the ensuing ( or introduced ) abasic site ( ab ) , and modification of the 3′ ends ultimately generates a single strand break ( SSB , arrow immediately 5′of the green triangle Figure 2A , also see Figure 4A ) that could be extended in the 5′ to 3′ direction by either a single ( short patch ) or several ( long patch ) nucleotides . Neither is error prone nor would BER remove the DNA strand on the 5′ side of the lesion ( green triangle , Figure 2A ) , which would expose its complement as the template for the error-prone process that would generate mutations in the reporter region . However , components of the MMR pathway can hijack U/G-BER intermediates and generate such gapped substrates ( e . g . , Kadyrov et al . , 2006; Schanz et al . , 2009; Pluciennik et al . , 2010; Peña-Diaz et al . , 2012 ) and we show later in the paper that components of both BER and MMR are involved in the mutagenesis . 10 . 7554/eLife . 02001 . 020Figure 4 . Mutagenic process and mutational outcome . ( A ) Graphical representation showing the expected repair template strands ( red ) for the mutagenic repair of top strand ( left panel ) or bottom strand ( right panel ) lesions that were restored to C/G . BER would generate a nick 5′ to the lesions ( black arrow ) . Subsequent exposure of the respective bottom or top strand as a single-stranded repair template would render TpC susceptible to deamination by APOBEC to TpU . The U could be subject to several different processes that would eventually register as a G mutation on top strand GpA ( left panel ) or as a C mutation on top strand TpC ( right panel ) . See text for more details . ( B ) Identity and fate of the mutated bases ( as percentages of the total mutations ) in terms of top strand sequence induced by repair that either restored lesions to C/G ( left quadrants ) or converted T/G to T/A ( right quadrants , gray box ) for top strand ( upper two quadrants ) and bottom strand ( lower two quadrants ) lesions on FM1 . The numbers of mutations ( n ) are given for each category . The data are pooled according to the fate and the strand location of the lesions . DOI: http://dx . doi . org/10 . 7554/eLife . 02001 . 020 The most right hand bar graphs in the upper and middle panels in Figure 2C also show that a preformed SSB ( i . e . , one did not result from BER activity on the introduced mismatches ) can also induce mutagenesis in flanking DNA . Here , we replaced respectively the top or bottom strands of FM1 , and bottom or top strands of M1F with non-5′ phosphorylated versions of the 0 MM control oligonucleotide . The preformed SSB <50 bp from the cis 3′ end of the reporter region in the top strand of FM1 or bottom strand of M1F generates the same mutagenic effect as U/G-ab/G at these positions . We also obtained this result when using non-5′ phosphorylated versions of T/G-mismatch producing oligonucleotides ( Figure 2—figure supplement 4 ) . Therefore , processing this nick bypasses T/G repair , and generates a substrate that is susceptible to mutagenesis . However , unlike the other lesions , a preformed SSB at ∼5 kb from the cis 3′ end of the reporter region ( bottom and top strand respectively for FM1 and M1F ) has little if any mutagenic effect . This difference could reflect the fact that these SSBs would be substrates for a SSB repair ( SSBR ) pathway that would recruit proteins different from those at the SSB generated during BER ( Caldecott , 2008 ) . We determined the fraction of total mutations contributed by each base of the reporter and mismatch regions , and the identity of the base to which it was mutated ( i . e . , its mutational fate ) , and report these results in terms of the top strand sequence . ( Supplementary file 2 lists all the mutations for each mutated episome . ) The pooled data for top strand T/G or U/G that were restored to C/G are shown in Figure 3A , B respectively . The mutations are distributed throughout the cassette , mostly on G , irrespective of the type of lesion ( summarized in Figure 3C ) . Mutations of either G or C would equally affect the integrity of the supF tRNA stems and potentially produce a non-functional tRNA . Thus , the skewed mutational pattern toward G does not likely reflect an ascertainment bias . In addition , the sequence context of ∼80% of the G mutations is GpA , but with less specificity for the 5′ base ( Figure 3D ) . Figure 3—figure supplement 1 shows that essentially the same mutational spectrum was induced by restoring each type of top strand lesion—that is , T/G , hmU/G , U/G or ab/G to C/G . Thus a similar mutagenic process is recruited during repair of any of these lesions . The restoration of top strand lesions to C/G would employ the bottom strand as the repair template . Thus , the preponderance of mutations on the G of GpA indicates that its complement on the repair template , the C of TpC , is targeted by the mutagenic process that is recruited during repair . This preference implicates the TpC-preferring APOBEC family members of single strand-specific cytidine deaminases . The left panel of Figure 4A outlines how this could occur during the restoration of top strand lesions to C/G after the repair process was initiated by BER at the lesion ( green triangle ) , and subsequent resection of the lesion-containing strand by MMR as described in the previous section . Deamination of the C of TpC would generate TpU , and as discussed by others ( Lange et al . , 2011; Nik-Zainal et al . , 2012; Peña-Diaz et al . , 2012; Roberts et al . , 2012; Burns et al . , 2013; Leonard et al . , 2013; Roberts et al . , 2013; Taylor et al . , 2013 ) further processing of the U could generate a number mutational outcomes of the original C ( and its complement G ) . Faithful copying of the U in the repair template would produce an A/U pair . Subsequent replication would result in a G to A transition ( complementary C to T transition ) . On the other hand , the A/U base pair could also be a substrate for a U glycosylase , which would remove the U and generate an abasic site . Subsequent replication of this strand by a high fidelity DNA polymerase would generally insert an A opposite the abasic site–that is , the ‘A rule’ , ( Strauss , 2002 ) . The fact that the most frequent mutational outcome at GpA was a G to A transition is consistent with the above two mechanism of generating an A opposite the original C of TpC ( Figure 4B , upper left quadrant ) . On the other hand , replication across the abasic site by various other DNA polymerases could insert a T or C opposite the abasic site and generate G to T or C transversions ( Lange et al . , 2011 ) , which we also found ( Figure 4B , upper left quadrant ) . As a consequence approximately equal amounts of transitions and transversions result from repair-induced mutagenesis , an outcome that was also observed for APOBEC-mediated mutations in some tumors ( Burns et al . , 2013; Leonard et al . , 2013 ) . All the data shown in Figure 4B are the pooled mutational fates induced by repair of all the lesions , and similar results were found for each type of lesion ( results not shown ) . The right hand side of Figure 4A illustrates the steps when bottom strand lesions are restored to C/G . In this case , the top strand serves as the repair template , but because we show all the mutations in terms of the top strand sequence the final mutational outcome is basically the complement of the results generated from the repair of top strand lesions . Figure 3—figure supplement 2 shows the fate and mutational spectra ( summarized in Figure 3—figure supplement 3 ) induced by repair of bottom strand lesions . Now , most of the mutations are on C of TpC when the lesions are restored to C/G ( Figure 3—figure supplement 4 ) , and , as the lower left quadrant of Figure 4B shows , the most common mutational outcome involves C to T transitions and with about an equal amount of transversions , to A and G . The right side of Figure 4B ( gray box ) shows the mutational outcome of converting top or bottom strand T/Gs to T/As . In these cases , the top or bottom strand respectively serves as the repair templates ( indicated in red font , right side Figure 4B ) , and the mutational outcomes parallel those that found for the restoration of bottom or top strand lesions to C/G wherein the top or bottom strands respectively were used as the repair template . Thus , when a top strand T paired with a bottom strand G is converted to T/A , most of the mutations are on C ( Figure 3C , T/G_con ) , ∼80% of which involve the C of TpC with almost no specificity for the 3′ base ( top panel , Figure 3—figure supplement 5 ) . However , when a bottom strand T/G is converted to T/A , most mutations are on the G of GpA ( Figure 3—figure supplement 3–T/G_con , Figure 3—figure supplement 5 , bottom panel ) . Approximately , equal amounts of G and C mutations are generated from the 0 MM controls ( most left hand bar graphs , Figure 3C , Figure 3—figure supplement 3 ) . These mutational spectra and their dinucleotide sequence contexts resemble what would result from repair of a mixture of top and bottom SSBs ( Figure 3—figure supplement 6 , Figure 3—figure supplement 7 ) , and are thus consistent with that expected from repairing trace amount of random nicks on both strands . APOBEC-mediated mutations in some cancer cells showed a strong preference for the C of TpCpW ( W is T or A ) ( Roberts et al . , 2013 ) . The trinucleotide context for opposite strand G mutations would be ApGpA and TpGpA . Although the trinucleotide context of about half of our G mutations was ApGpA , the others were mostly CpGpA and not TpGpA , which not surprising as the latter trinucleotide is represented only once in the reporter region ( Figure 3—figure supplement 8 ) . These results indicate that the APOBECs do not strongly discriminate between the nucleotides that are 3′ to TpC ( readily apparent in Figure 3—figure supplement 4 ) . A lack of discrimination on 3′ nucleotides was also shown in vitro for APOBEC3B ( Burns et al . , 2013 ) , and reflected in the mutational effect of APOBECs on HIV ( Bishop et al . , 2004; Doehle et al . , 2005 ) , and in this case the very low preference toward TpCpG ( opposite strand CpGpA ) , reflected the strong bias against CpG in the HIV genome ( van der Kuyl and Berkhout , 2012 ) . Clustered mutations can be characteristic of APOBEC-mediated TpC ( GpA ) mutations in tumors ( Nik-Zainal et al . , 2012; Roberts et al . , 2012 , 2013; Burns et al . , 2013; Taylor et al . , 2013 ) and also for AID-induced mutations in lymphoid cells ( e . g . , Peled et al . , 2008 ) . Figure 3 and Supplementary file 2 showed that the reporter/mismatch region ( ∼300 bp ) of some episomes contained more than one mutation . Thus , mutations can occur in a concerted fashion , which is the case for 25–30% of the episomes ( Figure 5A , B ) . This figure also shows that the mutational spectra and the dinucleotide contexts of the clustered and unclustered mutations induced by repair of top strand lesions did not differ , and all exhibited the mutational preference for the complement of the C of TpC , that is , the G of GpA . However , the spectra and dinucleotide context of clustered and unclustered mutations induced by repair of bottom strand lesions differed . In contrast to the clustered mutations , which had the same mutational APOBEC signature as the top strand lesions ( Figure 5B ) , the unclustered mutations show less of a preference towards mutating the C of TpC , most evident for those induced by restoration of T/G to C/G ( cf . lower 2 panels , Figure 5A and the lower panels labeled unclustered , Figure 5B ) . Perhaps , the generation of unclustered mutations with a non-strictly TpC mutational signature in the reporter region is facilitated by the ∼5 kb distance between its cis 3′ end and the lesion ( top panel , Figure 2C ) . 10 . 7554/eLife . 02001 . 021Figure 5 . Mutational spectra and dinucleotide context of clustered and unclustered mutations . ( A ) Fraction of the bases ( in terms of top strand sequence ) that were mutated in response to the repair of the indicated lesions on the top ( top two panels ) or bottom ( bottom two panels ) strand of FM1 for the clustered ( more than one mutations per sequence ) or unclustered ( one mutation per sequence ) . The numbers of the mutated sequences and the mutations are shown below each bar graph . p-values were calculated using the Fisher exact test . ( B ) Dinucleotide context of the clustered and unclustered mutations with respect to its 5′ or 3′ base . The faint purple bars show the fractions of all sixteen dinucleotides in the reporter region . The number and percentage of clustered and unclustered mutations are shown . p-values were calculated using the Fisher exact test . The data are pooled according to the type and strand location of the lesion . DOI: http://dx . doi . org/10 . 7554/eLife . 02001 . 021 The results in the foregoing two sections strongly implicated the involvement of APOBEC deaminases in repair-mediated mutagenesis . We examined this possibility in two ways: first , after determining the repertoire of APOBEC deaminases in the HeLa cells used for the foregoing experiments , we examined the effect of their knockdown by siRNA on repair-induced mutagenesis . Then , we used ChIP to determine if APOBEC could access the episome in response to an introduced lesion . As APOBEC-mediated mutagenesis occurred in response to DNA repair , we used siRNA to knock down selected components of different DNA repair pathways and two proteins that would be expected to process or respond to the lesions that we introduced . For these experiments , we treated the reconstituted episomes briefly with the 5′–3′ exonuclease− Klenow polymerase just before transfection into the mammalian cells to remove traces of gapped plasmids ( see Figure 8—figure supplement 1 , the legends to Figures 8 and 9 , ‘Materials and methods’ ) . 10 . 7554/eLife . 02001 . 025Figure 8 . siRNA knockdowns of DNA repair pathways reduce mutagenesis . ( A ) Effect of the knockdowns of BER or MMR proteins . The mutation frequency is plotted as a function of the indicated lesion on the top strand of FM1 for control siRNA ( siCtrl ) and the indicated siRNA transfections . Means ( ±SEM ) are shown from at least two independent experiments and the number of oligonucleotide ( Oligo # ) used to reconstitute the shuttle vector is given below each bar graph . p-values were calculated using the Fisher exact test . These experiments were carried out with Klenow treated reconstituted episomes ( see text and ‘Materials and methods’ ) . ( B ) Efficiency of DNA repair protein knockdowns . Western blots for the whole-cell lysates of HeLa cells transfected with siRNAs against various components of the BER and MMR pathways were performed with the antibodies against these proteins and α-T ( α-Tubulin , loading control ) . The western blot results were quantified with Quantity One software ( Bio-Rad , Hercules , CA ) and normalized to the amount of α-Tubulin to calculate the efficiency of the siRNA knockdowns . DOI: http://dx . doi . org/10 . 7554/eLife . 02001 . 02510 . 7554/eLife . 02001 . 026Figure 8—figure supplement 1 . Klenow treatment reduces the gapped contaminants to undetectable levels . The reconstituted episomes were incubated under Klenow reaction conditions without ( lanes 3–5 ) or with ( lanes 6–8 ) the 3′-5′ exo− Klenow polymerase ( see ‘Materials and methods’ ) , purified , and then restricted with KpnI . Klenow treatment reduces the level of gapped plasmids to undetectable levels . DOI: http://dx . doi . org/10 . 7554/eLife . 02001 . 02610 . 7554/eLife . 02001 . 027Figure 8—figure supplement 2 . Knockdown of DNA repair proteins does not affect the mutational signature . The 3′ dinucleotide context of the mutated bases read out on the top strand that were induced by restoration of the lesions ( pooled from the T/G and U/G mismatches ) to C/G for the indicated knockdowns . DOI: http://dx . doi . org/10 . 7554/eLife . 02001 . 02710 . 7554/eLife . 02001 . 028Figure 9 . Effect of PCNA and ATR on repair-induced mutagenesis . ( A ) Effect of PCNA and ATR knockdowns . The left panel shows the efficiency of PCNA , ATR ( or both ) knockdowns . The right panel shows the effect of these knockdowns on mutation frequency induced by the repair of each lesion . ( B ) Rescue of the PCNA knockdown . The left panel shows the expression of endogenous PCNA ( empty vector ) , or exogenously expressed wild type PCNA ( PCNA_wt ) , or siRNA-resistant PCNA ( PCNA_R ) in the absence of siRNA , or in the presence of control siRNA ( siCtrl ) , or siRNA against PCNA ( siPCNA ) . The right panel shows the effects of these conditions on the mutation frequency induced by the repair of the indicated lesions—the cross hatched bar graphs show the corresponding results from panel A . ( C ) Rescue of the ATR knockdown . The left panel shows the expression of endogenous ATR ( empty vector ) , or exogenously expressed wild type ATR ( ATR_wt ) , or siRNA-resistant ATR ( ATR_R ) , or siRNA-resistant-kinase defective ATR ( ATR_KDR ) in the absence of siRNA , or in the presence of control siRNA ( siCtrl ) or siRNA against ATR ( siATR ) . The right panel shows the effects of these conditions on the mutation frequency induced by the repair of each lesion and the cross hatched bar graphs show the corresponding results from panel A . For all panels means ( ±SEM ) are shown from at least two independent experiments and the number of oligonucleotide ( Oligo # ) used to reconstitute the shuttle vector is given below each bar graph . These experiments were carried out with Klenow treated reconstituted episomes ( see text and ‘Materials and methods’ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02001 . 02810 . 7554/eLife . 02001 . 029Figure 9—figure supplement 1 . Knockdown of PCNA or ATR does not affect the mutational signature . The 3′ dinucleotide context of the mutated bases read out on the top strand induced by restoration of the lesions ( pooled from the T/G and U/G mismatches ) to C/G for the indicated knockdowns . The faint purple bars show the fractions of the 16 dinucleotides in the reporter region . DOI: http://dx . doi . org/10 . 7554/eLife . 02001 . 029 Here , we showed that repair of normally occurring mismatches ( i . e . , those that arise from the inherent chemical instability of DNA ) can be mutagenic to flanking DNA . Figure 10 summarizes our results , which suggest that APOBEC-mediated mutagenesis induced by DNA repair acts downstream of the DNA repair pathways that generate the obligatory single-stranded substrates for TpC preferring APOBEC deaminases ( Conticello et al . , 2007 ) . siRNA knockdowns ( bold font depicts the factors that we tested ) indicate that components of both BER and MMR are required to generate the deaminase substrates from both T/G and U/G , and that PCNA and ATR could modulate the availability of these substrates to the deaminases . With respect to U/G and T/G , we depict the interaction of the BER and MMR pathways in the framework of that proposed from elegant biochemical studies that showed that MMR can hijack the 3′ end of the nick that would result from a U/G-BER intermediate and generate a single stranded region 5′ of the lesion that could serve as an APOBEC substrate ( Kadyrov et al . , 2006; Schanz et al . , 2009; Pluciennik et al . , 2010; Peña-Diaz et al . , 2012 ) . The cited in vitro studies from the Modrich group also showed that MMR could directly access T/G mismatches in a PCNA-dependent reaction , which we illustrate on the right hand side of Figure 10 . In such cases , MMR could load on either strand of the T/G containing episome and lead to resection of either the bottom or top strand of the duplex . However , we can only currently score mutagenic events that occur on the 5′ side of the T/G wherein resection of the top strand would expose the supF reporter region . Thus , our results suggest an apparent 3′ bias for the MMR-mediated resection . We are now implementing experiments to examine the induction of mutations 3′ of the T/G . Although the sensitivity of U/G-induced mutagenesis to the proximity of this lesion to the cis 3′ end of the reporter region is consistent with the foregoing biochemical results , some of our findings are at odds with the biochemical model—for example , the apparent inhibitory effect of PCNA on U/G-repair induced mutagenesis ( knockdown increases its mutagenic effect , Figure 9 ) vs the requirement for PCNA in the MMR hijacking of BER ( Kadyrov et al . , 2006; Pluciennik et al . , 2010 ) . While these paradoxical results could be reconciled by considering differences between threshold concentrations of PCNA required for its various functions ( e . g . , Pluciennik et al . , 2010 ) , our results do not directly address the mechanistic details of the interaction of PCNA and the U/G-repair generated APOBEC substrate . Furthermore , while we depict PCNA and ATR converging on a common pathway downstream of the hijacked BER intermediate , our only rationale for doing so are the results in Figure 9A , which show that their simultaneous knockdown cancels the effects of their separate knockdowns . Likewise , while the effects of knocking down various BER and MMR components , as well as the ATR and PCNA knockdowns distinguish the T/G- and U/G-repair intermediates , the mechanistic relationships between these pathways or factors , and the properties of the T/G repair intermediate that distinguish it from the U/G intermediate are not clear . The fact that the extent of T/G-induced mutagenesis is more sensitive to the knockdown of MMR than BER suggests that MMR can generate an APOBEC-sensitive substrate independent of ( i . e . , in addition to ) that mediated by BER ( right hand side of Figure 10 ) . Also , that an APOBEC-vulnerable substrate can be generated from either strand of a T/G mismatch—either restoration of T/G to C/G or conversion to T/A can induce mutagenesis—and that it is sensitive to PCNA knockdown ( Figure 9 ) are consistent with studies in vitro that showed that MMR components can access either strand of T/G containing DNA in a PCNA-dependent reaction ( Pluciennik et al . , 2010 ) . Given the relative insensitivity of the mutagenic effect of T/G to its distance from a cis 3′ end of reporter region , we tentatively propose that a migrating D-loop can be mounted on either strand of the episome to provide a top or bottom strand deaminase substrate ( Figure 10 ) . Numerous recent studies support a role for the TpC preferring APOBEC C deaminases , especially APOBEC3B , in generating large numbers of mutations ( i . e . , the ‘mutator phenotype’ [Bielas et al . , 2006; Venkatesan et al . , 2006] ) that characterize the progression , and perhaps initiation , of cancers ( Nik-Zainal et al . , 2012; Roberts et al . , 2012; Stephens et al . , 2012; Burns et al . , 2013; Leonard et al . , 2013; Roberts et al . , 2013; Taylor et al . , 2013 ) . Less certain is the source of the single-stranded DNAs that are the preferred substrates for the APOBEC deaminases . Here , we directly showed that A3B can access the T/G and U/G repair intermediates generated in our mammalian model system ( Figure 6C ) and generate the mutational signatures induced in some cancers ( Figure 6B , Figure 6—figure supplement 1 ) . The hundreds of T/Gs and U/Gs that are generated daily by spontaneous hydrolytic deamination respectively of methyl-C and C would provide a continual source of potential APOBEC substrates . This source could also be potentially augmented by the thousands of other BER-processed lesions that arise daily due to additional spontaneous degradative processes ( Atamna et al . , 2000; Barnes and Lindahl , 2004 ) . Although our findings showed that DNA repair of T/G , U/G , and other lesions can generate APOBEC deaminase substrates , the composition of the repair intermediates that both renders them susceptible to APOBEC-mediated deamination and accounts for their distinct mutagenic properties are unknown . Components of both the BER and MMR pathways that are normally recruited to these lesions can also engage in complex interactions with factors involved in both DNA metabolism and other cellular functions ( e . g . , Cortázar et al . , 2007; Kovtun and McMurray , 2007; Jacobs and Schar , 2012; Peña-Diaz and Jiricny , 2012; Peña-Diaz et al . , 2012 ) . Additionally , the availability of APOBEC deaminases per se may not be sufficient to support repair-induced mutagenesis ( Figure 7 , also observed in certain cancers , Roberts et al . , 2013 ) . Thus some of the factors required to either generate APOBEC substrates from repair-intermediates , or recruit the deaminases to them , might not be present in all cells . Given the fairly ubiquitous distribution of ABOPBEC deaminases in various cell types and tissues ( Refsland et al . , 2010 ) , determining such factors will be important for evaluating the extent to which DNA repair intermediates can be accessed by APOBEC-mediated mutators , and whether DNA repair could be vulnerable to other error-prone processes as well . Such alternate processes might contribute to the distinctive mutational signatures of the unclustered mutations induced by repair of bottom strand lesions ( Figure 5A , B ) . Therefore , an important next step is to characterize the composition of the mutagenic and non-mutagenic repair intermediates that assemble on the various mismatches . An advantage of generating these repair intermediates using preformed mispairs on a defined DNA as done here , is that they could be isolated from cells by selection for either the nucleic acid or a presumed protein component ( e . g . , Dèjardin and Kingston , 2009; Wu et al . , 2011 ) , or by the sequential application of each . The latter approach should enhance the chances of isolating genuine mutagenic intermediates , a prerequisite for more definitive analysis of their functional and biochemical properties . Repair-associated mutagenesis might not only contribute to the genetic changes that underlie various disease states ( e . g . , cancer , as discussed above ) but also aging and evolutionary change ( as has been earlier suggested , e . g . , Huttley et al . , 2000; Walser and Furano , 2010 ) , which in some instances might result from normal physiological processes . For example , we recently found that the repair of 5-carboxy-C/G , an intermediate of the physiological demethylation of methyl-C ( Bhutani et al . , 2011; Wu and Zhang , 2014 ) , can also be mutagenic . Thus , the normal cycling of the epigenetic methyl-C mark could contribute to the high mutation rate of regulatory sequences thought to contribute both disease processes ( Maurano et al . , 2012 ) and evolutionary novelty ( Wittkopp and Kalay , 2012 ) . We constructed various versions of a previously described SV40-based shuttle vector ( Seidman et al . , 1985 ) as described in detail in the next section . Our major modification was the introduction of a mismatch ( MM ) region immediately 3′ of the bacterial supF tRNA gene to generate the prototypical FM1 vector ( Figure 1A ) . SupF tRNA suppresses an amber mutation in lacZamb bacteria ( ß-galactosidase minus ) , which permits blue/white screening to assess the activity of the supF gene ( loss of activity due to mutations in either its promoter region or coding sequence produces a white colony ) . We also introduced a bar code of eight random nucleotides to uniquely identify each clone ( previously named the signature element , Parris and Seidman , 1992 ) . In our initial experiments some episomal plasmids suffered large deletions that included all or part of the reporter cassette , which we found was due to over zealous attempts to remove episomes that lack the bar code by digestion with BstZ17I , which is unique to this region . However , we found that this step was not necessary and eliminating it largely eliminated the deletions ( Figure 2—figure supplement 1 ) . Nonetheless , all of the mutational frequencies reported here are the percentage of white colonies containing undeleted plasmids—‘Materials and methods—Data acquisition and analysis’ . The design of the MM region follows the method of Hou et al . ( 2007 ) , which facilitates removal and replacement of the top or bottom strand of a DNA duplex and is illustrated for the top strand in Figure 1B and Figure 1—figure supplement 1 . After nicking the MM region with the New England Biolabs ( NEB , Ipswich , MA ) enzymes , Nt . BbvCI ( top strand ) or Nb . BbvCI ( bottom strand ) , the nicked strand was removed by incubation with its biotinylated complement followed by removal of the hybrid with streptavidin-magnetic beads . The gapped vectors were purified by extraction with phenol/chloroform and EtOH precipitation . A 100-fold excess of 5′-phosphorylated control ( perfect complement , 0 MM ) or oligonucleotides that would generate T/G , hmU/G , U/G or ab/G mismatches ( lesions ) when paired to the gapped vector were added , and the mixture was heated for 5 min at 95°C , then for 4 hr at 40°C , and after slow cooling to room temperature incubated overnight with T4 DNA ligase ( NEB ) . Supplementary file 1 lists the oligonucleotides and their numbers are indicated in the figures . We monitored the efficiency of gapping and the reconstitution of the gapped plasmids by following the loss and restoration of the KpnI site between the nicking sites ( Figure 1—figure supplement 1 ) . For those mismatches that would eliminate an AatII restriction site in the mismatch region , we also monitored the reconstitution of the mismatch-containing episomes by their resistance to digestion by this enzyme ( Figure 1—figure supplement 2 , as has been done by others , e . g . , Peña-Diaz et al . , 2012 ) . Using the T4 DNA ligase from NEB , but not from other sources , eliminated the need for purifying closed circular plasmids prior to their use ( Walser , J-C and Aschrafi , A , unpublished observations ) . However , the reconstituted episomes contained traces of gapped plasmids ( Figure 1—figure supplement 1 ) , which were inconsequential except when we knocked down various proteins involved in DNA repair ( i . e . , selected components of the BER and MMR pathways , or PCNA and ATR; see Figures 8–10 ) . Most of these knockdowns increased the mutagenic effect of the 0 MM control ( and presumably also that of the lesion-containing plasmids , results not shown ) . We reasoned that perhaps the trace amounts of gapped episomes are not readily repaired when various repair enzymes are depleted and thereby provide ready-made substrates for the APOBEC deaminases . This idea was supported by our finding that knockdowns of any of the tested repair proteins increased the mutagenesis induced by non-reconstituted gapped plasmids ( results not shown ) . A brief treatment of the reconstituted vectors in the ligation reaction with the 5′–3′ exo− Klenow nuclease just prior to their transfection into mammalian cells reduced the amount of gapped plasmids to undetectable amounts ( Figure 8—figure supplement 1 ) , eliminated the increase of the mutagenic effect of the 0 MM control during siRNA knockdowns of repair proteins , but did not materially affect the extent of the mutagenesis induced by the 0 MM control or the lesion-containing episomes in experiments with the siRNA controls—cf . the siCtrl for 0 MM , T/G or U/G in Figures 8 and 9 with those in Figures 2 , 6 and 7 ( for HeLa_JM ) . The Klenow treatment consisted of a 10-min incubation at 37°C of 6 μl of the ligation reaction with 3 μl of a solution that contained 0 . 2 unit of Klenow polymerase ( NEB , M0212 ) , 3 × NEBuffer 2 and 100 μM of dNTP . Cells were seeded in a six-well plate at a density of 3 × 105 per well . After 24 hr , we added 100 µl of serum-free DMEM that contained 6 µl each of the overnight ligation mixture ( 1 µg of reconstituted plasmid ) and 6 µl of Fugene 6 . After 48 hr , the plasmids were extracted from the cells with Wizard Plus SV Miniprep kit ( Promega , Madison , WI ) , digested with DpnI ( removes un-replicated input plasmid ) and electroporated into E . coli MBM7070 ( lacZuag_amber ) , which were grown on LB plates containing 100 µg/ml ampicillin , 1 mM IPTG and 0 . 03% Bluo-gal ( Invitrogen/Life Technologies , Grand Island , NY ) . After incubation at 37°C overnight and at room temperature for another day ( for maximal color development ) , the plasmids from white colonies were collected and analyzed for deletions by PCR ( primer: F4914 , 5′CCAGCGTTTCTGGGTGAGCA3′/R250 , 5′TTTTTGTGATGCTCGTCAGG3′ , which respectively flank the EcoRI and NheI sites that encompass the reporter cassette , Figure 1A ) . The mutation frequency is the number of white colonies that contained undeleted plasmids/total number of colonies . Directly electroporating reconstituted plasmids into E . coli generated no white colonies with undeleted vectors ( 3 deleted vectors /42 , 000 screened , results not shown ) . Thus , the mutagenic effect of DNA repair requires passage through mammalian cells . For siRNA knockdown , we added 20 pmol siRNA ( in a total volume of 400 µl serum-free DMEM that contained 6 µl Lipofectamine RNAiMAX , Invitrogen/Life Technologies ) to cells immediately after they were plated . To rescue the siRNA knockdowns , we added 1 µg of the siRNA-resistant plasmid in 400 µl serum-free DMEM that contained 3 µl Lipofectamine LTX , Invitrogen/Life Technologies 24 hr after the siRNA was added , and these cells were transfected 24 hr later with the reconstituted plasmids . We changed the media before each addition . We extracted total RNA with the PureLink RNA Mini Kit , Ambion/LifeTechnologies according to the provided instructions and synthesized cDNA with ProtoScript II Reverse Transcriptase ( NEB ) . RNA ( 2 . 5 μl @ 400 ng/μl ) and 1 . 5 μl random hexamer DNA primers ( 100 μM ) were heated at 65°C for 10 min and immediately put on ice . The reverse transcription mixture containing 4 μl 5 × RT buffer , 2 μl 10 × DTT , 1 μl 10 mM dNTP , 1 μl RNAse inhibitor ( SUPERaseIn , 20 U/μl , Ambion/LifeTechnologies ) , 1 μl ProtoScript II Reverse Transcriptase ( 200 U/μl ) and 7 μl nuclease-free water . The reactions were incubated at 25°C for 5 min , 42°C for 1 hr and then 80°C for 5 min to inactivate the enzyme . The reactions were diluted with 80 μl of nuclease-free water and qPCR was performed as described in Refsland et al . ( 2010 ) and Burns et al . ( 2013 ) . The target site of the human apobec3b siRNA is located in the 3′ UTR of the native mRNA . Thus , the APOBEC3B expression plasmid , A3B , which contains only the apobec3b protein coding sequence , is resistant to apobec3b siRNA . A catalytic deficient version of the plasmid , A3B_Cat , was made by introducing the mutations E68A and E255Q ( Burns et al . , 2013 ) . The PCNA gene was amplified from Mammalian Gene Collection—Human ( ATCC , Manassas , VA MGC-8367 ) with primers ( PCNA_BAMHI_F , 5′TTCAAAGGATCCCGTTCGCCCGCTCGCTCTGAGGCT3′/OTB7_R , TTTTTGTTTGCAAGCAGCAGATTAC ) , digested with BamHI and XhoI and then ligated to BamHI/XhoI digested pcDNA3 . 1 ( + ) ( Invitrogen ) to generate the PCNA-expressing plasmid PCNA_WT . The siRNA target site in PCNA_WT was changed to 5′AATAGAAGACGAGGAGGGT3′ to generate the RNAi-resistant plasmid , PCNA_R . The ATR expression plasmids ( ATR_WT and ATR_KD ) were gifts from Dr Stephen J Elledge ( Cortez et al . , 2001 ) and the siRNA target site was changed to 5′TACCCGTCTTCTCAGAATAGCTGCA3′ to generate RNAi-resistant plasmids ATR_R and ATR_KDR . All site-directed mutagenesis were performed with Agilent , Santa Clara , CA , QuikChange II Kits . Control siRNA ( no target in mammalian cells , R0017 ) was purchased from Abnova , Walnut , CA . siRNA against atr ( 5′AACGAGACTTCTGCGGATTGCAGCA3′ ) and msh2 ( 5′AATCTGCAGAGTGTTGTGCTTAGTA3′ ) were supplied by Invitrogen ( Stealth RNAi siRNA duplex ) . SiRNA against xrcc1 ( M-009394-01-0005 ) , ape1 ( M-010237-00-0005 ) , mlh1 ( J-003906-10-0005 ) , pcna ( D-003289-10-0005 , M-003289-02-0005 for rescue ) , apobec3b ( J-017322-08-0005 ) , apobec3c ( J-013711-08-0002 ) and apobec3f ( J-019039-17-0005 ) were purchased from Dharmacon RNAi Technologies / GE Healthcare , Fairfield , CT . Antibodies were purchased from the following sources: antibodies to ATR ( ab10312 ) , MSH2 ( ab52266 ) –Abcam , Cambridge , MA; to PCNA ( #2586 ) , XRCC1 ( #2735 ) , APE1 ( #4128 ) , MLH1 ( #3515 ) –Cell Signaling Technology , Boston , MA; to HA ( HA-7 , H3663 ) ; FLAG ( Ctrl IgG for ChIP , F7425 ) –Sigma , St . Louis , MO . ChIP assays were performed according to a published protocol ( Carey et al . , 2009 ) . HeLa cells were seeded in a 6-cm dish plate at a density of 8 × 105 and after 24 hr , 1 µg of the A3B-3HA expressing plasmid in 400 µl serum-free DMEM that contained 3 µl Lipofectamine LTX was added . After another 48 hr , 200 µl of solution of serum-free DMEM that contained 12 µl each of the overnight ligation mixture ( 2 µg of mismatch/DNA lesion-containing plasmids ) and 12 µl of Fugene 6 was added to the cells and after 4 hr cells were harvested for ChIP assay . The qPCR primers for detecting the SupF region are: SupF_F , 5′GGGGCGAAAACTCTCAAGGATCTTACCGCTG3′ and SupF_R , 5′GGGATCCGGGTATTGAATTTCGGCCGTG3′; for the T antigen region are: 189LT_F , 5′CCAGCCATCCATTCTTCTATGTCAGCAGAGCC3′ and 189LT_R , AAGAACAGCCCAGCCACTATAAGTACCATGAA . HeLa_JM was provided by Dr John Moran ( University of Michigan ) , and HeLa_KU by Dr Karen Usdin ( NIH ) , 2102ep ( a human embryonal carcinoma cell line ) was provided by Dr Tom Fanning , HEK293 was provided by Dr Roland Owens ( NIH ) , and PA-1 ( an ovarian teratocarcinoma cell line ) and 143B ( an osteosarcoma cell line ) were purchased from ATCC . Unless otherwise indicated all the experiments were carried out with HeLa_JM cells . Each reconstituted vector was introduced into various cells in two or more independent transfections for a total of ∼500 distinct transfections . The replicated plasmids isolated from each transfection were subject to one or more independent trials of blue/white screening . The plasmids from all the white clones were screened for deletions by PCR and all those with undeleted plasmids from a given transfection were sequenced . More than 99% of the undeleted plasmids contained at least one mutation in the reporter region ( data not shown ) . The ratio of white colonies with mutated undeleted plasmids to the total number of colonies examined in the blue/white trials for a given transfection is taken as the mutation frequency . We aligned the sequences from each transfection to its relevant starting sequence using SEAVIEW ( Galtier et al . , 1996 ) . Overall , we determined the DNA sequences of ∼2600 undeleted plasmids , about 5% of which contained small indels in the reporter region that could not be detected by PCR screening . We excluded these sequences from further analysis because their frequency was not correlated with either the type of introduced lesion or even the presence of one . We grouped the rest ( ∼2500 ) into 155 distinct alignments . We also determined the DNA sequences of ∼250 blue clones , grouped into 38 alignments . We parsed these alignments and computed mutational data using custom Unix , Perl , and R scripts ( R Foundation for Statistical Computing , http://www . R-project . org ) . We also used R for statistical analysis as indicated in the legends to the Figures or text . We determined the fate of the introduced DNA lesions , and the frequency with which each base of the reporter and mismatch regions was mutated , and to which base it was mutated ( i . e . , its mutational fate ) . We determined the mutagenic effect caused by the repair of each type of lesion ( T/G , hmU/G , U/G , ab/G ) located on the top or bottom strand from the pooled results of each lesion class irrespective of their location , number , or context within the mismatch region of a given vector as these variables did not materially affect the mutagenic effect of their repair ( Figure 2—figure supplement 2 ) .

The inherent chemical instability of the four bases that are found in DNA leads to our genetic material being damaged on a daily basis . The sequence of these bases codes the genetic instructions necessary for all cellular functions , so damaged bases must be efficiently recognized and accurately repaired . The base excision repair pathway carries out these functions . However , there are some circumstances in which random changes to the genetic code can be beneficial . In immune cells , for example , these changes enhance the diversity of antibodies generated to fight bacteria and viruses . In immune cells , a second repair pathway—the mismatch repair pathway—hijacks the base excision repair pathway . This gives enzymes belonging to the APOBEC family access to the DNA that is undergoing repair , and these enzymes change cytosine bases to uracil bases . Subsequent processing steps can lead to different bases substituted for the original cytosine . The recent discovery that APOBEC enzymes are abundant in other types of cells raised the possibility that these enzymes could be significant source of mutations in the DNA of cells where such mutations are not welcome . To explore this possibility Chen et al . deliberately introduced a number of mutations ( that are normally repaired by the base excision repair pathway ) into non-immune human cells and observed what happened . The mutations were repaired , but the number of mutations in neighboring bases increased by a statistically significant amount . In particular , most of these mutations involved a cytosine base that was preceded by a thymine base . Chen et al . also showed that both APOBEC and the mismatch repair pathway are involved , as is the case for the mutations caused by APOBEC enzymes in immune cells . Similar APOBEC mutations are known to be involved in cancer . The model system developed by Chen et al . not only shows that normally error-free DNA repair can be involved in generating these mutations , but also used to obtain a better understanding of these processes and thereby provide new insights in cancer biology .

2014

Repair of naturally occurring mismatches can induce mutations in flanking DNA

Channelrhodopsin-2 ( ChR2 ) has quickly gained popularity as a powerful tool for eliciting genetically targeted neuronal activation . However , little has been reported on the response kinetics of optogenetic stimulation across different neuronal subtypes . With excess stimulation , neurons can be driven into depolarization block , a state where they cease to fire action potentials . Herein , we demonstrate that light-induced depolarization block in neurons expressing ChR2 poses experimental challenges for stable activation of specific cell types and may confound interpretation of experiments when ‘activated’ neurons are in fact being functionally silenced . We show both ex vivo and in vivo that certain neuronal subtypes targeted for ChR2 expression become increasingly susceptible to depolarization block as the duration of light pulses are increased . We find that interneuron populations have a greater susceptibility to this effect than principal excitatory neurons , which are more resistant to light-induced depolarization block . Our results highlight the need to empirically determine the photo-response properties of targeted neurons when using ChR2 , particularly in studies designed to elicit complex circuit responses in vivo where neuronal activity will not be recorded simultaneous to light stimulation . Optogenetics is a powerful emergent technology for investigating complex patterns of synaptic connectivity and circuit properties that underlie physiology and behavior . By selectively expressing and activating light-gated ion channels or light-driven ion pumps , optical control of neural circuitry can be accomplished with high spatial and temporal resolution ( Boyden et al . , 2005; Li et al . , 2005; Nagel et al . , 2005 ) . To date , many variants of light-sensitive proteins are available for use towards the activation or inhibition of excitable cells ( Lin , 2011; Rein and Deussing , 2012 ) . The most commonly used variant , channelrhodopsin-2 ( ChR2 ) has become increasingly valuable in studies that exploit cell type-specific excitation . Isolated from the green algae , Chlamydomonas reinhardtii , ChR2 is a seven-transmembrane-domain , nonselective cation channel that is directly activated by blue light when bound with the chromophore all-trans-retinal ( Hegemann et al . , 1991; Lawson et al . , 1991; Takahashi et al . , 1991; Nagel et al . , 2003 ) . Absorption of photons by the chromophore induces conformational changes in the channel that allow for the passage of cations across the cell membrane in which it is expressed ( Boyden et al . , 2005 ) . This direct photo-gating makes ChR2 useful for manipulating neuronal activity in vitro ( Boyden et al . , 2005; Li et al . , 2005; Zhang and Oertner , 2007 ) and in vivo ( Markram et al . , 2004; Li et al . , 2005; Nagel et al . , 2005; Petreanu et al . , 2007; Douglass et al . , 2008; Huber et al . , 2008; Mahoney et al . , 2008; Ayling et al . , 2009; Baier and Scott , 2009; Guo et al . , 2009; Liu et al . , 2009; Zhu et al . , 2009; Katzel et al . , 2011; Schultheis et al . , 2011; Boyd et al . , 2012; Britt et al . , 2012; English et al . , 2012; Beltramo et al . , 2013; Chen et al . , 2013; Chiu et al . , 2013; Owen et al . , 2013 ) . Coupled with genetically-targeted expression , ChR2 affords exquisite functional control of specific neuronal subpopulations . Many studies to date that utilized ChR2 for in vivo manipulation of behavior and/or circuit function have largely focused on the activation of principal excitatory neurons , particularly in the cortex ( Petreanu et al . , 2007; Huber et al . , 2008; Ayling et al . , 2009; Boyd et al . , 2012; Britt et al . , 2012; Beltramo et al . , 2013; Chen et al . , 2013 ) . As the experimental applications of ChR2 move to include biophysically diverse interneurons ( Markram et al . , 2004; Katzel et al . , 2011; Schultheis et al . , 2011; English et al . , 2012; Chiu et al . , 2013; Owen et al . , 2013 ) , a fuller understanding of its possibilities and limitations becomes essential . Although ChR2 expression , trafficking , and activation has been achieved in most types of nervous tissue ( Li et al . , 2005; Nagel et al . , 2005; Bi et al . , 2006; Schroll et al . , 2006; Adamantidis et al . , 2007; Zhang et al . , 2007; Douglass et al . , 2008; Mahoney et al . , 2008; Baier and Scott , 2009; Guo et al . , 2009; Han et al . , 2009; Liu et al . , 2009; Zhu et al . , 2009; Gourine et al . , 2010; Hagglund et al . , 2010; Diester and et al , 2011; Figueiredo et al . , 2011; Sasaki et al . , 2012; Ljaschenko et al . , 2013 ) , little consideration has been given to the kinetics that constrain light-induced firing properties in different neuronal subtypes . Factors that affect the degree of neuronal photostimulation , including intrinsic differences in firing dynamics , membrane properties , and channel composition differ among neuronal cell types . These properties render certain neuronal populations more susceptible to depolarization block , which is generally induced by excessive cation influx , prolonged membrane depolarization , and/or the inability to repolarize , resulting in the failure to support continuous firing of action potentials . In vitro electrophysiological recordings have shown the ability of ultrafast microbial opsins to inactivate neurons via excessive photo-induced currents , where regular-spiking excitatory neurons showed a greater tendency to enter depolarization block than fast-spiking interneurons ( Mattis et al . , 2012 ) . Similarly , recent studies in primary hippocampal cultures have explored the effect of different ChR2 variants on neuronal firing properties in response to differing light intensity and pulse duration ( Lin et al . , 2013 ) . Such reports characterizing channelrhodopsin behavior highlight the need for empirically determining optimal light stimulation parameters for translation in vivo , as many factors including pulse duration , light intensity , ChR2 variant , method of expression , and targeted cell type may all affect a cell’s response to light activation . To date , little attention has been given to the extent and specificity of depolarization block subsequent to ChR2-mediated excitation . Using multiple ChR2 mouse lines that expressed the common ChR2 variant ( H134R ) , we show both ex vivo and in vivo that light-induced neuronal firing can be accompanied and/or followed with periods of latency in various neuronal subtypes . Further , we show that interneurons are particularly vulnerable to this effect . In particular , at a fixed moderate frequency of light stimulation , increasing pulse width resulted in transient activation and protracted silencing of regular-spiking interneurons , but excitatory principal/pyramidal cells or subsets of fast-spiking neuronal populations were more resistant to depolarization block . Our data highlight that especially for interneurons , it is important to empirically define stimulation parameters to elicit desired effects . In particular , careful considerations must be taken when applying in vivo optogenetic approaches in animal studies , being mindful of the time intervals chosen for light pulses used to activate neurons as temporal changes in the photo stimulus duration may result in variable or unpredictable response patterns . Additionally , we found that the duration of a light pulse beyond a certain point , independent of frequency , fails to result in higher spike rates in neurons expressing ChR2 , but rather silences them . Together , our data recommend the use of short-width light pulses when using ChR2 for the activation of interneuron cell types , while slightly longer pulse widths may be desirable for consistent spiking of larger excitatory cell types . Finally , our findings also support the possibility of exploiting this effect to investigate the physiological phenomena that underlie depolarization block , or targeted neuronal silencing . To date , little consideration has been given to the kinetics that constrain light-induced firing properties in different neuronal subtypes . To test for differential effects of ChR2-mediated light stimulation between different classes of neurons , we first performed whole cell electrophysiological recordings in acute brain slices from several classes of commonly studied interneurons and principal excitatory cells that express the ChR2-EYFP fusion protein . The first type of interneuron we investigated has been characterized by its expression of corticotropin-releasing hormone ( CRH ) ( Taniguchi et al . , 2011 ) . CRH serves diverse physiological roles in the nervous system as a hormonal regulator of stress ( Bale and Vale , 2004 ) , feeding ( Richard et al . , 2000 ) , and reproduction ( Laatikainen , 1991 ) , among others , and is highly expressed in neurons of the hypothalamus ( Cusulin et al . , 2013 ) , interneurons of the olfactory bulb ( Huang et al . , 2013 ) , and sparsely throughout the cortex . To investigate the photo response properties of these cells , we crossed male Crh-Cre+/− animals with conditional female ROSA26-lox-stop-lox-ChR2-EYFP animals ( Gt ( ROSA ) 26Sortm32 . 1 ( CAG-COP4*H134R/EYFP ) Hze/J , herein referred to as ROSA26LSL-ChR2-EYFP ) . Within the main olfactory bulb , Crh-Cre+/−; ROSA26LSL-ChR2-EYFP animals expressed the ChR2-EYFP fusion protein selectively in CRH-positive interneurons of the external plexiform layer ( EPL ) ( Figure 1A ) . To investigate the photo-response kinetics of these cells , we performed whole-cell recordings from CRH-expressing EPL interneurons ( N = 10 cells ) to graded light-on/light-off stimulation parameters ( Figure 1B ) . For each parameter , we used a single stimulation train consisting of 20 light pulses at 20 Hz . Initial stimulation parameters varied only with respect to ‘time on’ pulse width so as to determine the effect of light pulse duration on ChR2-mediated neuronal excitation , with fixed light intensity and frequency , which was kept at ≈40 mW/mm2 and 20 Hz , respectively . For olfactory bulb CRH interneurons , we observed that short duration ( 5–25 ms ) light pulses ( ‘time on’ ) were optimal for sustaining consistent neuronal firing , with a highest average firing rate elicited at 10 ms pulse widths ( Figure 1C ) . In contrast , increasing the duration of a light pulse to more than 25 ms , such that ‘time on’ pulse duration outlasted the interval between pulses ( ‘time off’ ) , reliably decreased firing rates , and completely silenced neurons at pulse durations of ≥49 ms ‘time on’ ( Figure 1D ) . This phenomenon is not likely due to desensitization of the ChR2 itself , as stepwise current injections independent of ChR2-mediated light activation mirror silencing observed in neurons from increased light pulse duration ( Figure 1E , F ) . Interestingly , CRH-expressing EPL interneurons exhibit high probability ( >80% ) co-expression with the fast-spiking interneuron marker , parvalbumin ( PV ) , and are capable of achieving high firing rates when stimulated by light ( Huang et al . , 2013 ) . Our present results suggest that CRH-expressing EPL interneurons exhibit a functional heterogeneity in their electrical responses , at least in a subset of neurons , highlighting the limitation of using broad neuronal markers ( e . g . , parvalbumin ) as being indicative of a general firing response pattern ( i . e . , fast-spiking ) that may not be generalizable from one brain region ( e . g . , cortex ) to another ( e . g . , olfactory bulb ) . 10 . 7554/eLife . 01481 . 003Figure 1 . Effects of light pulse duration on CRH-expressing interneurons of the olfactory bulb . ( A ) Crh-Cre+/−; ROSALSL-ChR2-EYFP mice express the ChR2-EYFP fusion protein in CRH-expressing interneurons of the external plexiform layer ( EPL ) of the main olfactory bulb ( scale bar , 0 . 5 mm ) . Inlay represents zoomed image of ChR2-expressing interneurons of the EPL ( scale bar , 100 μM ) . ( B ) Firing responses of ChR2-expressing neurons were recorded for seven different stimulation parameters . Each light stimulation parameter consists of a single train comprised of 20 light pulses ( ≈40 mW/mm2 ) at 20 Hz . Pulse width is the only condition that varies among the seven parameters . Parameter 1–1 ms pulse width/49 ms intervals , Parameter 2–5 ms pulse width/45 ms intervals , Parameter 3–10 ms pulse width/40 ms intervals , Parameter 4–25 ms pulse width/25 ms intervals , Parameter 5–40 ms pulse width/10 ms intervals , Parameter 6–45 ms pulse width/5 ms intervals , Parameter 7–49 ms pulse width/1 ms intervals . ( C ) Robust firing of a CRH-expressing EPL interneuron in response to brief light pulses ( 20 Hz , 10 ms pulse width ) . ( D ) Prolonged light pulse duration ( 20 Hz , 49 ms pulse width ) leads to depolarization block in CRH interneurons . ( E ) Moderate current injection ( 60 pA ) elicits regular firing of ChR2-expressing CRH interneurons . ( F ) High current injection ( 160 pA ) results in depolarization block of ChR2-expressing CRH interneurons . DOI: http://dx . doi . org/10 . 7554/eLife . 01481 . 003 To further characterize this phenomenon across other cell types , we next performed whole-cell electrophysiological recordings from two types of choline acetyltransferase ( ChAT ) -expressing interneurons in a mouse line that expressed ChR2-EYFP under the control of the ChAT promoter ( Zhao et al . , 2011 ) . Cholinergic signaling has been implicated in reward , learning , and addiction ( Tapper et al . , 2004; Maskos et al . , 2005; De Biasi and Dani , 2011 ) , and optogenetic tools are now being broadly applied to characterize the neural mechanisms underlying these diverse behaviors ( Witten et al . , 2010; Ren et al . , 2011 ) . Chat-ChR2 transgenic mice show expression of ChR2-EYFP in cholinergic neurons of the brain , with dense expression in the medial habenula ( MHB ) ( Figure 2A ) and throughout the striatum ( Figure 2B ) . Like CRH-expressing interneurons of the olfactory bulb , ChAT-positive MHB interneurons ( Figure 2C , N = 9 ) and striatal interneurons ( Figure 2D , N = 15 ) exhibited robust and consistent firing with short duration ( ≤25 ms ) light pulses , and dramatic decreases in firing rates with pulse width durations greater than 40 ms ( Figure 2E , F ) . Again , this effect recapitulates the firing response to stepwise current injections ( Figure 2G–J ) . 10 . 7554/eLife . 01481 . 004Figure 2 . Effects of light pulse duration on ChAT-expressing interneurons of the MHB and striatum . ( A ) Chat-ChR2 transgenic mice display ChR2-EYFP expression in the medial habenula ( MHB ) ( scale bar , 1 mm ) and ( B ) striatum ( scale bar , 1 mm ) . Inlays show zoomed images of respective ChR2-expressing MHB or striatal interneurons ( scale bars , 100 μM ) . Regular firing of ChR2-expressing ChAT interneurons of the ( C ) MHB and ( D ) striatum in response to brief light pulses ( 20 Hz , 10 ms pulse width ) . Prolonged light pulse duration ( 20 Hz , 49 ms pulse width ) leads to depolarization block in ( E ) MHB and ( F ) striatal ChAT interneurons . Moderate current injection results in steady firing of ChR2-expressing ( G ) MHB ( 50 pA ) and ( H ) striatal ( 80 pA ) ChAT interneurons . High current injection results in depolarization block of ChR2-expressing ( I ) MHB ( 150 pA ) and ( J ) striatal ( 300 pA ) ChAT interneurons . DOI: http://dx . doi . org/10 . 7554/eLife . 01481 . 004 While the photocurrent responses of ChAT-expressing interneurons were largely uniform and comparable , somatostatin ( SST ) -positive interneurons demonstrated greater diversity of responses . SST-expressing interneurons have been characterized as a heterogeneous population , both morphologically and electrophysiologically ( Markram et al . , 2004; Wang et al . , 2004; Ma et al . , 2006 ) . Distinct morphologies could be observed of SST interneurons throughout all layers of the cortex . Whereas those that resembled Martinotti cells made up a subset of the population ( Figure 3A , B , white arrows ) , others resembled multipolar basket cells with varicose axonal projections ( Figure 3C , D , open arrows ) , or appeared as small ovoid bitufted/bipolar cells ( Figure 3E ) . To investigate the effects of light on these cells , we crossed male Sst-Cre+/− animals with floxed conditional female ROSA26LSL-ChR2-EYFP animals to target ChR2-EYFP expression to SST-positive interneurons throughout the cerebral cortex ( Figure 3F ) . Our recordings validate previous reports ( Wang et al . , 2004; Ma et al . , 2006 ) of electrophysiological heterogeneity between SST cortical interneurons , with one subset of SST interneurons exhibiting regular-spiking neuronal activity and another exhibiting faster-spiking dynamics . The former subset showed photostimulation responses consistent with those seen in CRH and ChAT interneurons ( Figure 3G–J , N = 9 ) . Interestingly , faster-spiking SST interneurons ( N = 9 ) increased their firing rates with longer light pulse duration ( Figure 3K , L ) . Furthermore , current injections ( Figure 3M , N ) recapitulated this result . This type of heterogeneity , even within a subclass of interneuron characterized by subtype-specific patterns of gene expression , urges further caution when designing light stimulation parameters so as to avoid potential effects of light-induced depolarization block , and to optimize for consistent and reliable neuronal firing among targeted populations of neurons . 10 . 7554/eLife . 01481 . 005Figure 3 . Effects of light pulse duration on a heterogeneous SST interneuron population . ( A–E ) Sst-Cre+/−; ROSA26LSL-tdTomato animals express the tdTomato reporter in SST-positive interneurons , which display heterogeneous morphologies ( scale bars , 50 μM ) . ( F ) Sst-Cre+/−; ROSALSL-ChR2-EYFP mice display diffuse expression of ChR2-EYFP throughout the cortex ( scale bar , 1 mm ) . Inlay shows zoomed image of ChR2-expressing SST cortical interneurons ( scale bar , 100 μM ) . ( G ) Steady firing of a regular-spiking ChR2-expressing SST cortical interneuron in response to brief light pulses ( 20 Hz , 10 ms pulse width ) . ( H ) Prolonged light pulse duration ( 20 Hz , 49 ms pulse width ) leads to depolarization block in regular-spiking SST cortical interneurons . ( I ) Moderate current injection ( 30 pA ) leads to steady firing of regular-spiking SST cortical interneurons expressing ChR2 . ( J ) High current injection ( 100 pA ) results in depolarization block of regular-spiking ChR2-expressing SST cortical interneurons . ( K ) Steady firing of a fast-spiking SST cortical interneuron in response to brief light pulse stimulation ( 20 Hz , 10 ms pulse width ) . ( L ) Prolonged light pulse duration ( 20 Hz , 49 ms pulse width ) leads to robust firing in fast-spiking SST cortical interneurons . ( M ) Current injection ( 120 pA ) leads to steady firing of fast-spiking SST interneurons . ( N ) High current injection ( 500 pA ) results in robust firing of fast-spiking SST cortical interneurons . DOI: http://dx . doi . org/10 . 7554/eLife . 01481 . 005 Because light intensity may also affect ChR2-mediated spiking , we recorded firing dynamics in response to increasing pulse widths with different light intensities ( ≈10 , 20 , 30 , and 40 mW/mm2 ) . Though overall average firing rates scaled with increasing light intensities across all interneuron cell types tested , increased pulse duration was sufficient to drive neurons into depolarization block regardless of light intensity ( Figure 4A–D ) . Fast-spiking SST interneurons , however , exhibited increased and sustained firing with increasing pulse duration at all four light intensities ( Figure 4E ) . These data are consistent with decreased averaged spike amplitudes over the duration of the stimulus train in response to increased pulse widths ( Figure 4F–I ) , with the exception of fast-spiking SST interneurons , which showed only modest reductions in spike amplitudes over time ( Figure 4J ) . While interneurons generally exhibited high spike probabilities with short light pulses ( ≤25 ms ) , prolonged light pulses dramatically increased spike failure ( Figure 4K ) . With long pulses , regular-spiking interneurons characteristically fired an initial action potential or a short burst of action potentials , followed by prolonged depolarization block . Fast-spiking SST interneurons exhibited higher spike fidelity than other interneurons tested , even with long pulses ( Figure 4K ) . 10 . 7554/eLife . 01481 . 006Figure 4 . Firing dynamics of diverse interneurons in response to varying light pulse duration . Average firing rates of ChR2-expressing ( A ) CRH ( N = 5–10 cells/intensity from 4 animals ) , ( B ) ChAT MHB ( N = 4–9 cells/intensity from 5 animals ) , ( C ) ChAT striatal ( N = 6–15 cells/intensity from 5 animals ) , ( D ) regular-spiking SST ( N = 5–9 cells/intensity from 4 animals ) , and ( E ) fast-spiking SST interneurons ( N = 4-9 cells/intensity from 4 animals ) in response to variable light intensity and pulse duration ( 20 Hz ) . ( F–J ) Interneuron amplitudes , normalized to spike onset , in response to increasing pulse width . ( K ) Spike probabilities of various interneuron cell types in response to increasing pulse widths . Data points represent averages ± SEM . OB = Olfactory Bulb , MHB = Medial Habenula . DOI: http://dx . doi . org/10 . 7554/eLife . 01481 . 006 Due to size , typical excitatory neurons have low membrane resistance . This property makes excitatory neurons more resistant to depolarization block , as changes in voltage across the membrane require higher levels of current input . However , membrane channel composition promotes regular-spiking activity in these neurons , and thus likely establishes innate firing limits . Previous work has shown that increasing the frequency of light stimulation by various activating opsins results in decreased spike probability ( Mattis et al . , 2012 ) , consistent with known firing properties of pyramidal neurons and fast-spiking PV interneurons . To test the response kinetics of light-induced firing in excitatory cell types , we next performed whole-cell recordings in brain slices from Thy1-ChR2 mice that selectively express ChR2-EYFP in mitral/tufted cells of the main olfactory bulb ( Arenkiel et al . , 2007 ) ( Figure 5A ) , as well as in layer V cortical pyramidal neurons ( Figure 5B ) . In contrast to the observed inhibition reported in regular-spiking interneurons , these populations of excitatory cells were more resistant to prolonged light pulse-induced depolarization block . Increased light pulse duration elicited rapid , sustained neuronal firing in mitral cells ( N = 14 ) ( Figure 5C , D ) , whereas short pulse widths ( <25 ms ) appeared less effective than slightly longer pulse durations ( 25–49 ms ) at eliciting robust spiking , likely due to low membrane resistance . This effect appeared to plateau at light pulse durations ≥40 ms , even with 5 s continuous light pulse stimulation ( data not shown ) , though at this duration , 1 of 5 cells tested exhibited depolarization block , while the remaining cells did not . Pyramidal cells were consistently driven at pulse widths up to 40 ms ( Figure 5E , F ) , but were more susceptible to depolarization block than mitral cells at longer pulse durations . Though reduced light intensities attenuated average firing rates in mitral cells , consistent upward trends in firing were observed at all four light intensities , even at extended pulse durations ( Figure 6A ) . Cortical pyramidal cells were more resistant to depolarization block than other interneurons tested , but also tended to display characteristic silencing beyond 40 ms ( Figure 6B ) . Similar to fast-spiking SST interneurons , mitral cells displayed modest amplitude reductions over the duration of a stimulus train during prolonged light pulses ( Figure 6C ) . Pyramidal cells exhibited slight amplitude reductions up to 40 ms pulse widths before drastic reductions were observed at longer durations ( Figure 6D ) . Interestingly , while brief light pulses were better at promoting higher spike fidelity in pyramidal cells , they were often not sufficient for eliciting robust firing in mitral cells , resulting in reduced spike probability ( Figure 6E ) . Notably , extra spikes , particularly with longer pulses , were observed in all cell types . In fact , minimal pulse widths for all cell types tested were determined to have sub-millisecond response kinetics ( Figure 6F ) , likely contributing to the observation of spike doublets and triplets when neurons were stimulated with long pulses . Together , these data show that different neuronal subtypes respond with drastically different response kinetics when stimulated with ChR2 , and that the proper photostimulation parameters should be determined to elicit the desired firing output in target neurons . 10 . 7554/eLife . 01481 . 007Figure 5 . Principal excitatory cell types are less susceptible to light-induced depolarization block . Thy1-ChR2 transgenic mice display ChR2-EYFP expression in ( A ) excitatory mitral cells of the main olfactory bulb ( scale bar , 0 . 5 mm ) —note that ChR2 expression is also observed throughout the granule cell layer of the olfactory bulb due to axon collaterals from mitral/tufted cells and ChR2-expressing centrifugal inputs from the piriform cortex–and ( B ) layer V cortical pyramidal neurons ( scale bar , 1 mm ) . Inlays display zoomed images of ChR2-expressing olfactory bulb mitral cells ( scale bar , 100 μM ) or cortical pyramidal neurons ( scale bar , 200 μM ) . ( C ) Mitral cells display steady firing in response to brief light pulses ( 20 Hz , 10 ms pulse width ) and ( D ) enhanced firing in response to prolonged light pulse duration ( 20 Hz , 49 ms pulse width ) . ( E ) Steady firing of ChR2-expressing pyramidal cells in response to brief light pulse stimulation ( 20 Hz , 10 ms pulse width ) and ( F ) prolonged light pulse duration ( 20 Hz , 40 ms pulse width ) . OB = Olfactory Bulb . DOI: http://dx . doi . org/10 . 7554/eLife . 01481 . 00710 . 7554/eLife . 01481 . 008Figure 6 . Firing dynamics of principal cell types in response to varying light pulse duration . Average firing rates of ( A ) mitral cells ( N = 8–14 cells/intensity from 4 animals ) and ( B ) cortical pyramidal cells ( N = 6 cells/intensity from 3 animals ) in response to variable light intensity and increasing pulse width ( 20 Hz ) . ( C ) Mitral cell and ( D ) pyramidal cell amplitudes , normalized to spike onset , in response to increasing pulse width . ( E ) Spike probabilities of mitral cells and cortical pyramidal neurons in response to increasing pulse widths . ( F ) Minimal pulse widths required to elicit single action potentials in various neuron populations . Data points represent averages ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 01481 . 008 The potential for unknowingly silencing neurons that express ChR2 is greatest when firing properties of targeted neurons are not recorded simultaneously with light exposure . This is often the case in optogenetic manipulations used to elicit behaviors in freely behaving animals . Our ex vivo data suggest that studies of ChR2-expressing interneurons may require empiric determination of the pulse duration that elicits the desired activation of neurons . To validate these findings in vivo , we next performed extracellular recordings in anesthetized animals that express ChR2 in SST-expressing cortical interneurons , olfactory bulb mitral cells , and cortical pyramidal neurons , and measured their firing activity in response to differing light pulse durations . Consistent with observations made in slices , regular-spiking SST interneurons exhibited robust silencing due to prolonged light pulse duration ( Figure 7A ) . Though in vivo paired recordings were not made , presumptive pyramidal cell firing was effectively inhibited by ChR2-expressing SST interneurons activated by short light pulse durations , while pyramidal cell firing returned to baseline upon depolarization block of SST interneurons ( Figure 7B ) . The opposite effect was observed in fast-spiking SST cortical interneurons . Increasing light pulse duration onto these interneurons led to enhanced firing ( Figure 7C ) and enhanced inhibition of presumptive cortical pyramidal cells with increasing pulse duration ( Figure 7D ) . Verification of ChR2 expression in stimulated neurons in vivo was assumed by evaluating latencies to photo-induced spike activity ( median 1 . 8 ms for presumptive regular-spiking SST interneurons , N = 7 , and median 6 ms for presumptive fast-spiking SST interneurons , N = 3 ) . While this is only suggestive that recorded neurons expressed ChR2 , whole cell recording data on SST cells that expressed ChR2 revealed that photostimulation results in firing of recorded neurons within a similar range ex vivo ( 3–8 ms ) as observed latencies in vivo . These data are consistent with the assumption that in vivo recordings were made from ChR2-expressing neurons . 10 . 7554/eLife . 01481 . 009Figure 7 . Effects of light pulse duration on ChR2-expressing neurons in vivo . Increasing light pulse duration onto ( A ) a presumptive regular-spiking SST cortical interneuron ( median latency to spike = 1 . 8 ms , N = 7 cells from 5 animals ) leads to depolarization block and ( B ) disinhibition of a presumptive cortical pyramidal cell . Increasing light pulse duration onto ( C ) a presumptive fast-spiking SST cortical interneuron ( median latency to spike = 6 ms , N = 3 cells from 5 animals ) results in enhanced interneuron firing and ( D ) subsequent inhibition of a presumptive cortical pyramidal cell . In contrast to regular-spiking interneurons , increasing light pulse duration onto excitatory ( E ) mitral cells and cortical pyramidal cells enhance average firing rate with increasing pulse width . OB = Olfactory Bulb . DOI: http://dx . doi . org/10 . 7554/eLife . 01481 . 009 Interestingly , mitral cells ( Figure 7E ) displayed enhanced firing rates in response to incrementally increased light pulse durations with no apparent obstruction from depolarization block , though short pulse durations were not as effective at eliciting robust neuronal firing as longer pulses , consistent with what we observed previously ( Figure 6E ) . However , cortical pyramidal cells ( Figure 7E ) showed lower firing rates than mitral cells , consistent with ex vivo recordings ( Figure 6B ) , suggesting that pyramidal neurons exhibit some sensitivity to depolarization block at durations longer than 40 ms , but perhaps not as susceptible as previously noted interneurons . The use of ChR2 in animal studies is fast becoming a mainstay method to drive neuronal excitability in vivo ( Li et al . , 2005; Nagel et al . , 2005; Arenkiel et al . , 2007; Wang et al . , 2007 ) . Here , we show that targeted expression of ChR2 in specified cell populations of the mammalian brain can lead to depolarization block as a result of prolonged light-induced hyperexcitability . In particular , we demonstrate that the duration of light pulse stimulation used to activate ChR2-expressing neurons is critical to induce consistent and reliable firing of action potentials . Using several mouse lines to drive ChR2 expression in a neuronal subtype-specific manner , we show prominent ex vivo and in vivo silencing of regular-spiking interneuron cell types . In addition , we show that excitatory cell types and fast-spiking interneurons are more resistant to prolonged light-induced depolarization block . Given its nature as a cation-permeable membrane channel , ChR2-mediated depolarization block is likely due to excessive cation influx into a targeted neuron , resulting in prolonged membrane depolarization . If depolarization block were known to occur in a uniform fashion across interneuron subpopulations , the use of a narrow range of stimulation parameters could reliably avoid spike failures . Currently , however , pulse duration parameters vary widely in the growing number of optogenetic studies , and the use of prolonged light pulses has been employed in optogenetic applications used to drive neuronal dynamics in vivo ( Daou et al . , 2013; Liske et al . , 2013; Tabuchi et al . , 2013 ) . Though many optogenetic studies to date have employed short-width pulse parameters for in vivo manipulations , this concern is not inconsequential and should be considered when designing studies that employ ChR2 . The physiological relevance of depolarization block is supported by data that show depolarization block can be achieved under normal physiological synaptic function in vivo ( Bianchi et al . , 2012 ) , yet light-induced stimulation of a population of ChR2-expressing neurons is likely to far exceed synaptic activity produced under normal physiological states . Furthermore , and consistent with our data from SST-expressing interneurons , specific populations of seemingly homogeneous neurons in the same brain region may exhibit dramatically different electrophysiological properties , such that some percentage of those neurons exhibit a greater tendency to enter into depolarization block more readily than others ( Unal et al . , 2012 ) . Furthermore , specific biophysical properties of neurons , such as the distribution and number of voltage-gated sodium channels , will influence how susceptible a given class of neurons is to depolarization block to excessive photostimulation ( Tucker et al . , 2012 ) . Similarly , heterogeneity of transgene expression across different driver lines presents another dimension in which cells that express ChR2 may be variably susceptible to block . Likewise , viral-mediated expression of ChR2 with the use of variable-strength promoters is often a more potent method to strongly express ChR2 in cells of interest compared to transgenic models , but this too will lead to further variability . To reduce variability in photo-induced firing responses , it is important to empirically define for each neuronal subset the stimulation parameters that reliably and reversibly activate neurons of interest , before adopting stimulation parameters for in vivo testing . In vivo optogenetic studies in awake , behaving animals commonly use light stimulation to elicit and measure behavioral or complex physiological changes rather than a change in membrane potential or firing rate of the manipulated neuron ( Adamantidis et al . , 2007; Aponte et al . , 2011; Liu et al . , 2012; Shabel et al . , 2012; Tan et al . , 2012; van Zessen et al . , 2012 ) . The facile use of this technology in these contexts lends itself to confound when the appropriate light stimulation parameters are not employed . This stands in contrast to in vivo optophysiology in anesthetized animals , or ex vivo recordings in tissue slices where firing properties of the cells of interest are directly recorded in response to light exposure . However , when even these experimental preparations are used to study the responses of postsynaptic partners of stimulated neurons , the same concerns apply . In either case , interpretation of the output measure is susceptible to corruption when the cause of changes observed can be attributed to the block of activity in the neuron of interest rather than its excitation . Therefore , it is important to optimize stimulation conditions for targeted activation of a population of neurons . For experiments that require prolonged neuronal activation , the use of step-function opsins ( Berndt et al . , 2009; Diester and et al , 2011 ) might be warranted . Interestingly , these concerns might also be used advantageously . Consistent with our data , it may be possible to design strategies using ChR2-mediated optical stimulation for the explicit study of depolarization block in which it is thought to subserve a number of neurophysiological processes . Recent studies have highlighted a role for depolarization block in complex neuronal activity ( Marder et al . , 1996; Grace et al . , 1997; McIntyre et al . , 2004; Ullah and Schiff , 2010; Bianchi et al . , 2012 ) . For example , states of depolarization block may be a significant factor for information processing in certain classes of neurons ( Marder et al . , 1996; Dovzhenok and Kuznetsov , 2012 ) . Interestingly , a long-standing hypothesis regarding the therapeutic action of antipsychotic drugs commonly used in the treatment of schizophrenia features depolarization block in subsets of dopamine neurons after long-term drug treatment ( Grace et al . , 1997; Boye and Rompre , 2000; Valenti et al . , 2011 ) . Similarly , depolarization block has been proposed as a mechanism to explain the therapeutic benefit of deep brain stimulation ( McIntyre et al . , 2004 ) , a method used in the treatment of a variety of movement disorders such as Parkinson’s disease . Moreover , persistent sodium currents and consequent depolarization block are thought to facilitate the generation of electrographic seizures ( Bikson et al . , 2003; Ziburkus et al . , 2006; Ullah and Schiff , 2010 ) . With a precise model to control neuronal excitability , or to purposefully induce depolarization block , these phenomena may be topics for future investigation . Optogenetics affords the ability to mark , map , and manipulate brain cells and circuits with previously unimaginable power and precision . Although the technology to probe brain circuits is rapidly evolving at a breakneck pace , a detailed understanding of the cells being targeted for optogenetic studies remains limited . Our data highlight the need to empirically determine the optimal photostimulation parameters best suited for the cell types being investigated , since as a field we are still learning the possibilities and limitations of optogenetic manipulations . Animals were treated in compliance with the US Department of Health and Human Services and Baylor College of Medicine IUCAC guidelines . Chat-ChR2 ( Zhao et al . , 2011 ) and Thy1-ChR2 mice ( Arenkiel et al . , 2007; Wang et al . , 2007 ) have been previously described . Crh-Cre+/− ( Crhtm1 ( cre ) Zjh ) ( Taniguchi et al . , 2011 ) and floxed conditional ROSA26 ChR2-EYFP female mice ( Gt ( ROSA ) 26Sortm32 . 1 ( CAG-COP4*H134R/EYFP ) Hze/J ) were obtained from Jackson Laboratories . Crh-Cre+/−; ROSA26LSL-ChR2-EYFP mice were generated by crossing Crh-Cre+/− male mice with homozygous floxed conditional ROSA26LSL-ChR2-EYFP female mice . Sst-Cre+/−; ROSA26LSL-ChR2-EYFP mice were generated by crossing male Sst-Cre+/− ( Ssttm2 . 1 ( cre ) Zjh/J ) mice with conditional ROSA26LSL-ChR2-EYFP female mice . Sst-Cre+/−; ROSA26LSL-tdTomato animals were generated by crossing male Sst-Cre+/− mice with conditional ROSA26LSL-tdTomato ( B6 . Cg-Gt ( ROSA ) 26Sortm14 ( CAG-tdTomato ) Hze/J ) female mice . Animals were deeply anesthetized using isoflurane and were transcardially perfused with PBS followed by 4% paraformaldehyde ( PFA ) . Brains were dissected and postfixed in 4% PFA for 1 hr at room temperature or overnight at 4°C . Brains were coronally sectioned at 50 μm ( olfactory bulb ) or 100 μm ( forebrain ) using a Compresstome ( Precisionary Instruments , San Jose , CA ) . Slices were mounted with Vectashield mounting medium ( Vector Laboratories , Burlingame , CA ) and detection of EYFP or tdTomato expression was performed using a Leica M205-FA for low-magnification images , and a Leica TCS SPE confocal microscope under a 20X objective for higher magnification images . Coronal brain slices ( 300 μm ) were prepared from 3- to 6-week-old animals for all genotypes tested . The slices were embedded in low melting point agarose and sectioned into ice-cold oxygenated ( 5% CO2 , 95% O2 ) dissection buffer ( in mM: 87 NaCl , 2 . 5 KCl , 1 . 6 NaH2PO4 , 25 NaHCO3 , 75 sucrose , 10 glucose , 1 . 3 ascorbic acid , 0 . 5 CaCl2 , 7 MgCl2 ) , recovered for 15 min at 37°C in oxygenated artificial cerebrospinal fluid ( ACSF ) ( in mM: 122 NaCl , 3 KCl , 1 . 2 NaH2PO4 , 26 NaHCO3 , 20 glucose , 2 CaCl2 , 1 MgCl2 , 305-310 mOsm , pH 7 . 3 ) , and acclimated at room temperature for 10 min before performing electrophysiological recordings . Borosilicate glass electrodes ( Sutter Instruments , Novato , CA ) were used for whole cell patch clamp recordings . Electrodes were pulled with tip resistance between 3–8 MΩ , and filled with internal solution ( in mM , 120 K-gluconate , 5 KCl , 2 MgCl2 , 0 . 05 EGTA , 10 HEPES , 2 Mg-ATP , 0 . 4 Mg-GTP , 10 creatine phosphate , 290–300 mOsm , pH 7 . 3 ) . During recordings , coronal brain slices were placed in a room temperature chamber mounted on an Olympus upright microscope ( BX50WI ) and perfused with oxygenated ACSF . Cells were visualized under differential interference contrast imaging . Data were obtained via a Multiclamp 700B amplifier , low-pass Bessel-filtered at 4 kHz , and digitized on computer disk ( Clampex , Axon Instruments ) . Excitation light was from a BLM-Series 473 nm blue laser system ( Spectra Services , Ontario , NY ) , which was controlled by digital commands from Clampex to trigger photostimulation . The firing rate for each cell recorded for each stimulation parameter was counted manually for the full stimulus train and then averaged and plotted using GraphPad Prism statistical software ( GraphPad Software Inc , La Jolla , CA . ) . Light-evoked spike amplitudes were measured for the duration of the full stimulus train . Amplitudes were normalized to the initial light-evoked action potential for each respective trace , and temporally corresponding amplitudes were averaged for each cell type for a given pulse duration parameter . Spike probabilities were calculated by counting successful light-evoked spikes for each trace of a corresponding stimulus parameter and averaging success rates across cells for each cell type tested . Minimal pulse widths were determined by photostimulating neurons starting with short ( sub-millisecond ) to high ( 1 ms ) pulse widths . Criteria for consideration of minimal pulse width was that a given pulse width was capable of eliciting ≥70% fidelity ( i . e . , a minimum of seven light-evoked action potentials out of 10 pulses ) . Minimal pulse widths were then averaged across cells for each cell type tested . For in vivo recordings , animals were injected IP with ketamine ( 150 μg/g body weight ) , followed by sustained delivery of 0 . 3% isoflurane with oxygen to the animal . For olfactory bulb recordings , the dorsal surface of the olfactory bulb was carefully exposed so as to not damage the pia or underlying brain tissue . For extracellular recordings from the cortex , a small area of the skull overlying the cortical region of interest was removed to expose the underlying brain tissue . For light stimulation and electrophysiological recordings , optrodes made from fiber optics and 1 . 0 MΩ extracellular tungsten recording electrodes ( Microprobe Inc . , Gaithersburg , MD ) were used . A blue laser source ( CrystaLaser , Reno , NV ) was controlled by a Master-8 ( AMPI , Israel ) , and guided to either the olfactory bulb or cortex by focusing light onto fused silica fiber optics . Extracellular recordings were amplified by a Model 1800 AC amplifier ( A-M systems , Carlsborg , WA ) , digitized using a CED Power 1401 mk II ( Cambridge Electronic Design , Cambridge , England ) , and processed using Spike2 acquisition software ( Cambridge Electronic Design , Cambridge , England ) . For in vivo recordings , ChR2 expression was suggested by short latency to spike . Median latencies were calculated using MATLAB software . In addition to latencies , mitral cell identity can be determined by characteristic firing that is coupled to respiration . Thus , ChR2 expression can be indirectly determined by monitoring changes in respiratory-coupled firing responses when a ChR2-expressing cell is activated by light .

The brain is a highly complex structure composed of trillions of interconnecting nerve cells . The pattern of connections between these cells gives rise to the various brain circuits that govern how the brain functions . Understanding how the brain is wired together is important for determining how ‘faulty circuits’ contribute to various neurological disorders . New optogenetic technique tools allow neuroscientists to turn on specific neurons simply by shining light on them . These techniques involve genetically manipulating the organisms so that their neurons express proteins that are activated when they are exposed to light of a particular wavelength . However , it is important to understand the limitations of this approach—including the possibility that the light might actually turn off some neurons—when using it to study animal behavior . Now , Herman , Huang et al . show that shining light pulses for long durations onto neurons expressing a light-activated protein called channelrhodopsin-2 causes the neurons to become silenced rather than activated . Moreover , certain types of neurons , called interneurons , are more susceptible to this effect—termed ‘depolarization block’—than the other types of neurons . Researchers need to be mindful of this effect when channelrhodopsin-2 is used in optogenetic experiments to study the behavior of living animals . However , this silencing property could be useful in experiments that investigate situations in which depolarization block is thought to contribute to brain function and health: such as in the treatments of schizophrenia and Parkinson’s disease .

2014

Cell type-specific and time-dependent light exposure contribute to silencing in neurons expressing Channelrhodopsin-2

The mechanisms that maintain the functional heterogeneity of stem cells , which generates diverse differentiated cell types required for organogenesis , are not understood . In this study , we report that Trithorax ( Trx ) actively maintains the heterogeneity of neural stem cells ( neuroblasts ) in the developing Drosophila larval brain . trx mutant type II neuroblasts gradually adopt a type I neuroblast functional identity , losing the competence to generate intermediate neural progenitors ( INPs ) and directly generating differentiated cells . Trx regulates a type II neuroblast functional identity in part by maintaining chromatin in the buttonhead ( btd ) locus in an active state through the histone methyltransferase activity of the SET1/MLL complex . Consistently , btd is necessary and sufficient for eliciting a type II neuroblast functional identity . Furthermore , over-expression of btd restores the competence to generate INPs in trx mutant type II neuroblasts . Thus , Trx instructs a type II neuroblast functional identity by epigenetically promoting Btd expression , thereby maintaining neuroblast functional heterogeneity . Stem cells employ several strategies to generate the requisite number of diverse differentiated cell types required for organ development and organ homeostasis in higher eukaryotes ( Franco and Müller , 2013; Kohwi and Doe , 2013 ) . One such strategy involves stem cells changing their temporal identities . For example , neuroblasts sequentially express distinct temporal-identity transcription factors , allowing them to generate diverse differentiated cells in the fly embryonic ventral nerve cord ( Isshiki et al . , 2001; Pearson and Doe , 2003 ) . Another strategy involves maintaining a functionally heterogeneous pool of tissue-specific stem cells . Studies in flies and vertebrate systems show that functionally heterogeneous stem cells directly contribute to the generation of diverse cell types during hematopoiesis , gut homeostasis , and brain development ( Barker et al . , 2007; Bello et al . , 2008; Boone and Doe , 2008; Bowman et al . , 2008; Graf and Stadtfeld , 2008; Copley et al . , 2012; Franco et al . , 2012; Marianes and Spradling , 2013 ) . Numerous patterning mechanisms have been described to explain how the fates of distinct stem cells within a developing organ become specified , but how their functional heterogeneity is maintained throughout the lifespan of an organism remains completely unknown . The central complex of the insect brain is comprised of an intricate network of neurons and glia that process a vast number of environmental inputs essential for daily life ( Boyan and Reichert , 2011; Boyan and Williams , 2011 ) . All differentiated cell types in the central complex arise from repeated rounds of self-renewing asymmetric divisions of type I and type II neuroblasts , which are molecularly and functionally distinct ( Bello et al . , 2008; Boone and Doe , 2008; Bowman et al . , 2008 ) ( Figure1—figure supplement 1 ) . In every asymmetric division , a type I neuroblast always generates a precursor cell ( ganglion mother cell or GMC ) that divides once to produce two differentiated cells . By contrast , every asymmetric division of a type II neuroblast invariably leads to the generation of an immature INP that acquires an INP functional identity during maturation . An INP undergoes 5–8 rounds of asymmetric division to regenerate and generate a GMC with each division ( Homem et al . , 2013 ) . Thus , the ability to generate INPs functionally distinguishes these two types of neuroblasts . Type II neuroblasts uniquely express the ETS transcription factor Pointed P1 ( PntP1 ) ( Zhu et al . , 2011; Xiao et al . , 2012 ) . Mis-expression of PntP1 can induce a type II neuroblast functional characteristic in a type I neuroblast ( Zhu et al . , 2011 ) . However , the physiological function of PntP1 in the maintenance of a type II neuroblast functional identity remains unclear . The pnt locus encodes at least three distinct alternatively spliced transcripts . Thus , it is formally possible that multiple isoforoms of Pnt or a yet unknown mechanism function to maintain a type II neuroblast functional identity . Epigenetic mechanisms such as the methylation of histone H3 Lysine 4 ( H3K4 ) play central roles in specifying cell type identities during development ( Lim et al . , 2009; Ang et al . , 2011; Schuettengruber et al . , 2011; Shilatifard , 2012; Yang et al . , 2012 ) . The evolutionarily conserved SET1/Mixed-lineage leukemia ( MLL ) complexes catalyze the methylation of H3K4 and maintain the target gene loci in a transcriptionally active state ( Miller et al . , 2001; Roguev et al . , 2001; Krogan et al . , 2002 ) . The fly genome encodes three orthologs of the SET1/MLL protein , Trx , Trithorax-related ( Trr ) , and dSet1 . Similar to their mammalian counterparts , Trx , Trr , or dSet1 can each assemble functionally active complexes by binding to Absent , small , or homeotic discs 2 ( Ash2 ) , Retinoblastoma binding protein 5 ( Rbbp5 ) , and will die slowly ( Wds ) ( Wu et al . , 2008; Ardehali et al . , 2011; Mohan et al . , 2011 ) . Functionally , Trr or dSet1 regulates global mono- or tri-methylation of H3K4 respectively . In contrast , Trx appears to selectively regulate the expression of the Hox genes through the methylation of H3K4 ( Breen and Harte , 1993; Yu et al . , 1995 ) . However , little is known about the targets of Trx beyond the Hox genes . Here , we report that Trx maintains the type II neuroblast functional identity by regulating the transcription of btd during fly larval brain neurogenesis . Type II neuroblasts mutant for trx or genes encoding the core components of the SET1/MLL complex display a type I neuroblast marker expression profile and generate GMCs instead of INPs . These results indicate that Trx maintains a type II neuroblast functional identity by regulating the transcription of specific target genes . We identified a direct downstream target of Trx , Btd , that plays an important role in the maintenance of a type II neuroblast functional identity . btd mutant type II neuroblasts adopt a type I neuroblast functional identity and directly generate GMCs instead of INPs . Conversely , type I neuroblasts over-expressing btd assume a type II neuroblast functional identity and generate INP progeny . Most importantly , over-expression of btd restores the competence of trx mutant type II neuroblasts to generate INPs . Thus , we conclude that Trx functions to epigenetically maintain Btd expression in type II neuroblasts , thereby maintaining neuroblast functional heterogeneity in the larval brain . Analyses of gene transcription in mutant larval brains enriched with type I or type II neuroblasts led us to hypothesize that differential regulation of gene expression contributes to neuroblast functional heterogeneity ( Carney et al . , 2012 ) ( Komori and Lee , unpublished observation ) . Because the trx gene contributes to cell fate maintenance in a variety of developmental processes , we tested whether it is required for maintaining neuroblast heterogeneity . We induced GFP-marked mosaic clones derived from single wild-type or trx mutant type I or II neuroblasts and assessed the identities of cells in the clones by examining the expression of cell fate markers in a time-course study ( Figure 1—figure supplement 1 ) . Identical to wild-type neuroblasts , trx mutant type I neuroblasts maintained the expression of Deadpan ( Dpn ) and Asense ( Ase ) and the cytoplasmic localization of Prospero ( Pros ) , but lacked PntP1 expression ( Dpn+Ase+PntP1−Proscytoplasmic ) ( Table 1 , data not presented ) . In addition , both wild-type and trx mutant type I neuroblasts were always surrounded by GMCs ( Dpn−Ase+Prosnuclear ) ( data not presented ) . Thus , Trx is dispensable for the maintenance of a type I neuroblast functional identity . While all wild-type type II neuroblasts displayed a Dpn+Ase−PntP1+Pros− marker expression profile in all stages examined , trx mutant type II neuroblasts progressively altered their marker expression profile ( Figure 1A–D , Table 1 ) . Strikingly , almost all trx mutant type II neuroblasts in 72-hr clones displayed a type I neuroblast marker expression profile ( Figure 1B–D; Table 1 ) . These data strongly suggest that trx mutant type II neuroblasts adopt a type I neuroblast identity . 10 . 7554/eLife . 03502 . 003Table 1 . Summary of the marker expression profile in various genetic backgroundsDOI: http://dx . doi . org/10 . 7554/eLife . 03502 . 003GenotypeNeuroblast typeDpnAsePros*PntP1wild-typeI+++−wild-typeII+−−+Trx−/−I+++−Trx−/−II+++−Rbbp5−/−I+++−Rbbp5−/−II+++−btd−/−I+++−btd−/−II+−−+‘+’ indicates detected marker expression whereas ‘−’ indicates lack of marker expression . ‘*’ indicates basal asymmetric localization at the basal cortex in mitotic neuroblasts . 10 . 7554/eLife . 03502 . 004Figure 1 . trx mutant type II neuroblasts display characteristics of a type I neuroblast . Key for all figures: all clones are outlined in yellow . Wild-type type II neuroblasts or mutant type I neuroblasts ( Dpn+Ase−Pros−; white arrow ) ; Ase− immature INPs ( Dpn−Ase−Pros−; white arrowhead ) ; Ase+ immature INPs ( Dpn−Ase+Pros−; yellow arrow ) ; INPs ( Dpn+Ase+erm-lacZ+Proscytoplasmic; yellow arrowhead ) ; GMC generated by INPs ( Ase+erm-lacZ+Prosnuclear; orange arrow ) ; wild-type type I neuroblasts or mutant type II neuroblasts ( Dpn+Ase+Proscytoplasmic; magenta arrow ) ; GMC generated by wild-type type I neuroblasts or mutant type II neuroblasts ( Ase+Pros+erm-lacZ−; magenta arrowhead ) . Single asterisks indicate a statistically significant ( p-value <0 . 05 ) difference between the marked genotype and the control genotype in the same bar graph , as determined by the Student's t-test . n . s . indicates that the difference is statistically insignificant . NB: neuroblast . ( A–D ) trx mutant type II neuroblasts progressively acquire a type I neuroblast identity . ( A–B ) In the 72-hr GFP-marked clone , a wild-type type II neuroblast displays a Dpn+Ase− marker expression profile whereas a trx mutant type II neuroblast displays a Dpn+Ase+ expression profile . Scale bar , 10 μm . ( C ) Three-dimensionally reconstructed images of type II neuroblasts clones of the indicated genotypes . Scale bar , 10 μm . ( D ) The frequency of trx mutant type II neuroblasts displaying a type I neuroblast maker expression profile ( PntP1−Ase+ ) . N = 10 per time point . ( E–H ) trx mutant type II neuroblasts lose the ability to generate INPs . ( E ) The average number of INPs per staged type II neuroblast clone of the indicated genotype . N = 10 per time point . ( F–G ) In the 72-hr GFP-marked clones , a wild-type type II neuroblast is surrounded by INPs and their GMC progeny identified by erm-lacZ expression . In contrast , a trx mutant type II neuroblast is surrounded by GMCs that are directly derived from neuroblasts and lack erm-lacZ expression . Scale bar , 10 μm . ( H ) The average number of GMCs with or without erm-lacZ expression per type II neuroblast clone of the indicated genotypes . DOI: http://dx . doi . org/10 . 7554/eLife . 03502 . 00410 . 7554/eLife . 03502 . 005Figure 1—figure supplement 1 . A diagram of two distinct neuroblast lineages . A summary of the cell fate marker expression profile in type I and type II neuroblast lineage in the larval brain . NB: neuroblast; GMC: ganglion mother cell; INP: intermediate neural progenitor; imm INP: immature INP . DOI: http://dx . doi . org/10 . 7554/eLife . 03502 . 005 We extended our analyses to examine the identity of progeny directly derived from trx mutant type II neuroblasts . We observed a time-dependent reduction in INPs in trx mutant type II neuroblast clones as compared to identically staged wild-type clones . At 72 hr after clone induction , a control type II neuroblast was surrounded by approximately 20 INPs and 12 INP-derived GMCs that can be unambiguously identified by the expression of an erm-lacZ reporter transgene ( Figure 1C , E–F , H , Figure 1—figure supplement 1 ) . In contrast , an identically staged trx mutant neuroblast was directly surrounded by non-neuroblast progeny that displayed a Dpn−Ase+Prosnuclearerm-lacZ− expression profile identical to GMCs derived from type I neuroblasts ( Figure 1C , G , Figure 1—figure supplement 1 ) . Although trx mutant clones also contained an average of 3 INPs and 4 INP-derived GMCs , these cells were located at the extreme distal end of the clone , consistent with trx mutant type II neuroblasts adopting a type I neuroblast identity following the clone induction ( Figure 1C , E , G–H ) . These data strongly suggest that Trx regulates neuroblast heterogeneity by maintaining a type II neuroblast identity . The competence to generate INPs is a main feature that distinguishes the functional identity of a type II neuroblast from that of a type I neuroblast ( Weng and Lee , 2011; Homem and Knoblich , 2012; Janssens and Lee , 2014 ) . brain tumor ( brat ) and erm function in the immature INP to promote INP identity specification in the type II neuroblast lineage , and the defective specification of an INP identity leads to the formation of supernumerary type II neuroblasts in the brat or erm mutant brain ( Xiao et al . , 2012; Eroglu et al . , 2014; Janssens et al . , 2014; Koe et al . , 2014; Komori et al . , 2014 ) . If trx mutant type II neuroblasts indeed adopt a type I neuroblast functional identity , their progeny should be insensitive to the loss of brat or erm function and generate differentiated cells instead of reverting into supernumerary neuroblasts . A control type II neuroblast clone in the brat mutant brain contained more than 100 supernumerary type II neuroblasts and was devoid of GMCs and neurons ( Figure 2A , E ) . By contrast , a trx mutant type II neuroblast clone in the brat mutant brain contained far fewer supernumerary type II neuroblasts and far more GMCs and neurons as compared to the control clone ( Figure 2A–B , E ) . Similarly , a control type II neuroblast clone in the erm mutant brain contained more than 50 supernumerary type II neuroblasts and few GMCs and neurons ( Figure 2C , E ) . In contrast , a trx mutant type II neuroblast clone in the erm mutant brain contained fewer supernumerary type II neuroblasts but more GMCs and neurons as compared to the control clone ( Figure 2C–E ) . Together , these data strongly suggest that trx mutant type II neuroblasts lost the competence to generate immature INPs . 10 . 7554/eLife . 03502 . 006Figure 2 . trx mutant type II neuroblast directly generates GMCs . ( A–E ) trx is required for the expansion of supernumerary type II neuroblasts in the brat or erm mutant . ( A–D ) Removing trx function suppresses the expansion of supernumerary type II neuroblasts and restores differentiation in the 96-hr brat or erm mutant type II neuroblast clones . Three-dimensionally reconstructed images of the clones are shown to the right . Scale bar , 10 μm . ( E ) The average number of type II neuroblasts per clone of the indicated genotypes . ( F–I ) trx mutant type II neuroblasts exclusively distribute Pros to their progenies to specify GMC identity . ( F–G ) In the 48-hr clones , a wild-type type II neuroblast shows undetectable expression of Pros in telophase , whereas a trx mutant type II neuroblast shows the basal cortical localization of Pros . Scale bar , 10 μm . ( H ) The frequency of wild-type or trx mutant mitotic type II neuroblasts displaying the basal localization of Pros . ( I ) The average number of type I neuroblasts per type II neuroblast clone of the indicated genotypes at 72 hr after clone induction . DOI: http://dx . doi . org/10 . 7554/eLife . 03502 . 006 We directly tested whether trx mutant type II neuroblasts adopt a type I neuroblast functional identity and directly generate GMCs . Pros segregates exclusively into GMCs where it suppresses a type I neuroblast functional identity during asymmetric division of a type I neuroblast , but is undetectable in mitotic type II neuroblasts ( Knoblich et al . , 1995; Spana and Doe , 1995; Choksi et al . , 2006; Bayraktar et al . , 2010 ) . In a telophase trx mutant type II neuroblast , however , Pros localized asymmetrically in the basal cortex and segregated uniquely into the cortex of the future non-neuroblast progeny ( Figure 2F–H ) . Most importantly , removing pros function in trx mutant type II neuroblasts leads to the formation of supernumerary type I neuroblasts ( Figure 2I ) . These data confirm that trx mutant type II neuroblasts adopt a type I neuroblast functional identity and directly generate GMCs . Thus , we conclude that trx regulates neuroblast heterogeneity by maintaining a type II neuroblast functional identity . We assessed whether the histone methylation activity of Trx is required for maintaining a type II neuroblast functional identity . We induced mosaic clones derived from type II neuroblasts carrying the trxZ11 allele , which results in a missense mutation in the SET domain of Trx and reduces the histone methyltransferase activity of the Trx protein ( Smith et al . , 2004; Tie et al . , 2014 ) . Twenty-seven percent of trxZ11 type II neuroblasts assumed a type I neuroblast functional identity as determined by both the expression of a type I neuroblast marker expression profile and the generation of GMCs ( Figure 3A–B ) . This result indicates that the histone methylation activity of Trx is essential for the maintenance of a type II neuroblast functional identity . Trx was co-purified with the core components of the SET1/MLL complex , Ash2 , Rbbp5 , and Wds , from the lysate extracted from S2 cells ( Mohan et al . , 2011 ) . Thus , we tested whether the core components of the SET1/MLL complex are required for maintaining a type II neuroblast identity . Indeed , knocking down the function of ash2 , rbbp5 , or wds individually leads to fewer type II neuroblasts and INPs per brain lobe , identical to reducing trx function ( Figure 3—figure supplement 1A–G ) . Together , these data strongly support our hypothesis that Trx maintains a type II neuroblast functional identity through the SET1/MLL complex via a mechanism dependent of the histone methyltransferase activity . 10 . 7554/eLife . 03502 . 007Figure 3 . Trx and the core components of the SET/MLL complex maintain a type II neurobalst functional identity dependently on their catalytic activity for H3K4 methylaiton . ( A–B ) The function of trx for the H3K4 methylation is required for the maintenance of a type II neuroblast functional identity . ( A–B ) In the 72-hr clones , a trxZ11 mutant type II neuroblast displays a type I neuroblast marker expression profile and directly generates GMCs . Scale bar , 10 μm . Three-dimensionally reconstructed images of the clones are shown to the right . Scale bar , 10 μm . ( C–K ) The function of rbbp5 for the H3K4 methylation is required for the maintenance of a type II neuroblast functional identity . ( C–E , H , J ) In the 96-hr clones , rbbp5null type II neuroblasts display a type I neuroblast marker expression profile and directly generate GMCs . Over-expression of rbbp5FL but not rbbp5SG restores a type II neuroblast functional identity in rbbp5null type II neuroblasts . Three-dimensionally reconstructed images of the clones are shown to the right . Scale bar , 10 μm . ( F ) The frequency of type II neuroblasts of the indicated genotypes displaying the type I or type II marker expression profiles . ( G , I , K ) rbbp5 function is essential for the H3K4 methylation in fly larval brains . Scale bar , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 03502 . 00710 . 7554/eLife . 03502 . 008Figure 3—figure supplement 1 . Decreasing the function of the core components of the SET1/MLL complex leads to a reduction in type II neuroblasts . ( A–E ) Knocking down the function of trx , rbbp5 , wds or ash2 specifically reduces the number of type II neuroblasts per brain lobe . Scale bar , 20 μm . ( F–G ) The average number of type II neuroblasts or INPs per brain lobe of the indicated genotypes after knocking down the function of trx , rbbp5 , wds , or ash2 for 72 hr . DOI: http://dx . doi . org/10 . 7554/eLife . 03502 . 00810 . 7554/eLife . 03502 . 009Figure 3—figure supplement 2 . Generation of the rbbp5null allele and the UAS-rbbp5SG transgene . ( A ) The genomic organization of the rbbp5 locus . The rbbp5null allele was generated via imprecise excision of the P ( EP ) G4226 element , which removes the entire rbbp5 coding region . Yellow squares indicate the coding exons of rbbp5 while blue squares indicate the untranslated regions . The red line indicates the molecular lesion induced by the rbbp5null allele . ( B ) The average number of INPs per clone of the indicated genotypes at 96 hr after clone induction . ( C ) An alignment of the hinge region of the yeast , fly , and human Rbbp5 protein . The amino acid substitutions in the Rbbp5SG transgenic protein are indicated in red . DOI: http://dx . doi . org/10 . 7554/eLife . 03502 . 009 We focus on the Rbbp5 protein , which is essential for eliciting the histone methyltransferase activity of the SET1/MLL complex ( Cao et al . , 2010 ) , to test whether Trx maintains a type II neuroblast functional identity through the SET1/MLL complex . We first generated a null allele of the rbbp5 gene ( rbbp5null ) by excising a transposable P-element inserted at the 5ʹ end from the transcription start site ( Figure 3—figure supplement 2A ) . Mutant analyses confirmed that rbbp5null type II neuroblasts indeed adopt a type I neuroblast functional identity ( Figure 3C–F , Table 1 , Figure 3—figure supplement 2B ) . Thus , a rbbp5null type II neuroblast is phenotypically indistinguishable from a trx mutant type II neuroblast . We next examined the H3K4 methylation pattern in the rbbp5null type II neuroblast . All cells in the clones derived from single rbbp5null type II neuroblast showed undetectable mono- and tri-methylation of H3K4 ( Figure 3G , data not presented ) . This result is consistent with the SET1/MLL complex exerting its regulatory functions through the H3K4 methylation . Most importantly , over-expression of a UAS-rbbp5FL transgene that encodes a full-length Rbbp5 completely restored a type II neuroblast functional identity and significantly restored both the H3K4 mono- and tri-methylation in rbbp5null type II neuroblasts ( Figure 3F , H–I , Figure 3—figure supplement 2B , data not presented ) . By contrast , over-expression of a UAS-rbbp5SG transgene , which encodes a mutant Rbbp5 protein predicted to perturb the histone methyltransferase activity of the SET1/MLL complex ( Figure 3—figure supplement 2C ) ( Cao et al . , 2010 ) , failed to restore a type II neuroblast functional identity and the methylation of H3K4 in rbbp5null type II neuroblasts ( Figure 3F , J–K , Figure 3—figure supplement 2B , data not presented ) . Similarly , type II neuroblasts bearing a strong ash2 mutant allele also adopted a type I neuroblast functional identity and lost most H3K4 methylation based on the same criteria ( data not presented ) . Thus , the histone methyltransferase activity of the SET1/MLL complex is required for the maintenance of a type II neuroblast identity . We conclude that Trx maintains a functional identity of type II neuroblasts through the histone methylation activity of the SET1/MLL complex . Knocking down the function of trr or dset1 drastically reduced the global H3K4 mono- or tri-methylation in type II neuroblasts but had no effects on the maintenance of their functional identity ( Figure 4—figure supplement 1A–J ) . By contrast , removing trx function had no appreciable effects on the global H3K4 pattern in type II neuroblasts ( Figure 4—figure supplement 1K–N ) . These data led us to hypothesize that Trx maintains the type II neuroblast functional identity by regulating a small number of genes that are specifically expressed in the type II neuroblast . We compared gene transcription profiles by using mRNAs isolated from dissected larval brains enriched with type I or II neuroblasts to identify the candidate Trx target genes ( Bowman et al . , 2008; Weng et al . , 2010; Carney et al . , 2012; Haenfler et al . , 2012 ) . pnt and btd were among a small number of genes that were dramatically up-regulated in the mRNAs isolated from larval brains enriched with type II neuroblasts as compared to the mRNAs isolated from larval brains enriched with type I neuroblasts . We confirmed that both pntP1 and btd transcripts were indeed highly enriched in the brain lysate enriched with type II neuroblasts by qRT-PCR ( Figure 4A ) . Furthermore , we detected the binding of Trx to the transcription start site for both the pntP1 and btd transcription units ( Figure 4B , Figure 4—figure supplement 2A ) . In addition , the promoter region of both the pntP1 and btd transcription units also displayed a high level of H3K4 di-methylation , consistent with Trx-maintaining chromatin in an active state in these two loci through the H3K4 methylation ( Figure 4B , Figure 4—figure supplement 2A ) . By contrast , we did not detect Trx binding to the negative control region located 7 . 5 kilobases 3ʹ from the btd transcription unit ( Figure 4B; data not presented ) ( Petruk et al . , 2012 ) . Thus , both pnt and btd are the direct target genes of Trx . 10 . 7554/eLife . 03502 . 010Figure 4 . Btd likely acts downstream of Trx to maintain a type II neuroblast functional identity . ( A–D ) The btd gene is an excellent candidate target of Trx in the type II neuroblast . ( A ) The btd mRNA is highly enriched in the lysate extracted from larval brain enriched with type II neuroblasts . The elav transcript is highly enriched in differentiated neurons . The quantification represents the average of three biological replicates . ( B ) Trx directly binds to the type II neuroblast-specific enhancer element as well as the transcription start site ( TSS ) of the btd gene . The ChIP experiments were performed using the extract isolated from dissected brat mutant brains that are enriched with type II neuroblasts . Quantification of chromatin immunoprecipitated by the indicated antibodies relative to 5% of input . The quantification represents the average of three biological replicates . ( C–D ) An enhancer element from the btd gene is sufficient to induce type II neuroblast-specific expression of a UAS-mCD8::gfp reporter transgene in wild-type brain , while the enhancer activity of btd-Gal4 was reduced in rbbp5null brain . Scale bar , 20 μm . ( E–H ) btd is required for maintaining the functional identity but not the molecular signature of a type II neuroblast . ( E–F ) In the 72-hr clones , btd mutant type II neuroblasts maintain a type II neuroblast marker expression profile and are surrounded by 1–2 immature INP-like cells . Three-dimensionally reconstructed images of the clones are shown below . Scale bar , 10 μm . ( G ) The average number of INPs per clone of the indicated genotypes . ( H ) The average number of GMCs with or without erm-lacZ expression per type II neuroblast clones of the indicated genotypes at 72 hr after clone induction . ( I–J ) The immature INP-like cells generated by btd mutant type II neuroblasts are insensitive to loss of brat function . Removing brat function does not lead to supernumerary neuroblast formation in the 72-hr btd mutant type II neuroblast clones . Scale bar , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 03502 . 01010 . 7554/eLife . 03502 . 011Figure 4—figure supplement 1 . Global H3K4 mono- or tri-methylation is not required for maintenance of a type II neuroblast functional identity . ( A–H ) The core component of the SET1/MLL complex is required for the global methylation of H3K4 . ( A , C , E , G ) Knocking down the function of ash2 or trr leads to global loss of the H3K4 mono-methylation while knocking down the function of dSet1 does not . Scale bar , 10 μm . ( B , D , F , H ) Knocking down the function of ash2 or dSet1 leads to global loss of the H3K4 mono-methylation while knocking down the function of trr does not . ( I–J ) trr and dSet1 are dispensable for the maintenance of type II neuroblasts . ( I–J ) The average number of type II neuroblasts or INPs per brain lobe of the indicated genotypes after knocking down the function of trr or dSet1 for 72 hr . ( K–N ) trx mutant type II neuroblasts do not display appreciable reduction in the global methylation pattern . Scale bar , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 03502 . 01110 . 7554/eLife . 03502 . 012Figure 4—figure supplement 2 . Pnt likely functions to specify an INP identity . ( A ) Trx directly binds to transcription start site ( TSS ) of the pntP1 transcript . Quantification of chromatin immunoprecipitated by the indicated antibodies relative to 5% of input . The quantification represents the average of three biological replicates . The black lines indicate three different pnt transcripts . The magenta lines indicate three UAS-RNAi used to target the common exon of pnt transcripts . ( 1 ) UAS-pntRNAi ( 7171 ) , ( 2 ) UAS-pntRNAi ( TRiP . JF02227 ) , and ( 3 ) UAS-pntRNAi ( TRiP . HMSO1452 ) . ( B–C ) Expression of the UAS-pntRNAi transgene efficiently reduces PntP1 protein expression throughout the type II neuroblast lineage . ( D–E ) Knocking down the function of pnt induces supernumerary neuroblast formation . Scale bar , 10 μm . ( F–G ) The average number of type II neuroblasts per clone of the indicated genotypes . DOI: http://dx . doi . org/10 . 7554/eLife . 03502 . 012 We next tested whether either one of these two genes might regulate a functional identity of type II neuroblasts . pnt: because the pnt locus encodes multiple alternatively spliced transcripts , we assessed the function of pnt in the type II neuroblast by over-expressing three independent UAS-RNAi transgenes targeting two different regions of the same exon shared by all pnt transcripts ( Figure 4—figure supplement 2A ) . All three RNAi transgenes efficiently reduced pnt expression as indicated by a drastic reduction in the PntP1 protein ( Figure 4—figure supplement 2B–C; data not presented ) . Unexpectedly , knocking down the function of pnt in type II neuroblasts led to the formation of supernumerary neuroblasts ( Figure 4—figure supplement 2D–F ) . These results strongly suggest that pnt functions in the immature INP to promote INP identity specification similar to brat and erm . Consistently , heterozygosity of the pnt locus strongly enhanced the supernumerary neuroblast phenotype in the brat or erm hypomorphic brain ( Figure 4—figure supplement 2G ) . In addition , overexpression of pntP1 failed to restore a type II neuroblast functional identity in trx mutant type II neuroblasts ( data not presented ) . Thus , we conclude that pnt functions downstream of trx to specify an INP identity in the immature INP rather than to maintain the type II neuroblast functional identity . btd: a specific antibody against Btd is currently unavailable , and a genomic transgene that carries a BAC clone containing the entire btd locus led to embryonic lethality ( Komori and Lee , unpublished ) . Thus , we determined the spatial expression pattern of the btd gene by examining the expression of a btd-Gal4 transgene containing an enhancer element that was bound by Trx and displayed a high level of the di-methylation of H3K4 located 5 Kb upstream from the btd transcription start site ( Figure 4B ) . The expression of a UAS reporter transgene driven by btd-Gal4 was detected specifically in type II neuroblasts but was undetectable in type I neuroblasts in wild-type brains ( Figure 4C ) . Importantly , the expression of btd-Gal4 was drastically reduced in rbbp5null mutant brains ( Figure 4D ) . Together , these data strongly support our hypothesis that btd is an excellent candidate for functioning downstream of trx to maintain the type II neuroblast functional identity . If Trx maintains a type II neuroblast functional identity by regulating btd transcription , removing btd function should trigger type II neuroblasts to adopt a type I neuroblast functional identity . We assessed the identities of cells in the clones derived from single btd mutant type II neuroblasts by examining cell fate marker expression . btd mutant type II neuroblasts maintained a type II neuroblast marker expression profile in all stages examined , but these clones displayed a time-dependent reduction in INPs ( Figure 4F–G ) . Unlike the control clone , however , INPs in the 72-hr btd mutant clone were always located at the extreme distal end of the clone ( data not presented ) . In these clones , btd mutant type II neuroblasts were surrounded by 1–2 progeny resembling Ase− immature INPs but never Ase+ immature INPs ( Figure 4F ) . Instead , the remaining cells directly adjacent to the btd mutant type II neuroblast displayed a marker expression profile indicative of GMCs and immature neurons that are normally found in the type I neuroblast lineage ( Figure 4F , H ) . These observations prompted us to test whether the progeny of the btd mutant type II neuroblast resembling Ase− immature INPs were indeed functional by examining their dependency on brat function . In the brat mutant type II neuroblast clone , Ase− immature INPs rapidly reverted to supernumerary neuroblasts ( Figure 4I ) ( Xiao et al . , 2012; Komori et al . , 2014 ) . Most importantly , we never detected supernumerary neuroblast formation in the btd , brat double type II neuroblast clone , indicating that the direct progeny of the btd mutant type II neuroblast were insensitive to the loss of brat function ( Figure 4J ) . These data led us to conclude that btd mutant type II neuroblasts generate non-functional Ase− immature INPs that likely adopt an identity of GMCs normally found in the type I neuroblast lineage . Thus , we conclude that Trx most likely maintains the type II neuroblast functional identity through btd . Because btd is necessary for the maintenance of a type II neuroblast functional identity , we tested whether over-expression of btd is sufficient to induce a type II neuroblast functional identity in a type I neuroblast . We induced GFP-marked lineage clones derived from single type I neuroblasts mis-expressing a UAS-btd transgene and assessed the identities of cells in the clones by examining the expression of cell fate markers . In the control clones , type I neuroblasts maintained Ase expression and generated GMCs ( Figure 5A ) . Eighteen percent of type I neuroblasts mis-expressing btd lost Ase expression and generated progeny displaying a marker expression profile that is typically diagnostic of an immature INP or an INP ( Figure 5B , D ) . Another 10% of type I neuroblasts mis-expressing btd generated progeny that resembled immature INPs or INPs by marker expression , but maintained Ase expression ( Figure 5C ) . Thus , we conclude that mis-expression of btd is sufficient to trigger the characteristics that are specific for a type II neuroblast in a type I neuroblast . 10 . 7554/eLife . 03502 . 013Figure 5 . Over-expression of btd is sufficient to instruct a type II neuroblast functional identity in the type I neuroblast . ( A–E ) Over-expression of btd is sufficient to elicit a type II neuroblast functional identity . ( A–D ) In the 72-hr clones , 18% of type I neuroblasts over-expressing btd lose Ase expression and are surrounded by INP-like cells . An additional 10% of these neuroblasts maintain Ase expression despite being surrounded by INP-like cells . Three-dimensionally reconstructed images of the clones are shown to the right . Scale bar , 10 μm . ( E–H ) Progeny of type I neuroblasts over-expressing btd revert back to supernumerary neuroblast in the brat mutant or erm mutant . In the 72-hr clones , removing brat or erm function induces the formation of supernumerary type II neuroblasts derived from the progeny of type I neuroblasts over-expressing btd . Three-dimensionally reconstructed images of clones are shown to the right . Scale bar , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 03502 . 013 We extended our analysis to assess whether mis-expression of btd might endow a type I neuroblast with the functional feature unique to a type II neuroblast—the competence to generate INPs . We reasoned that if a type I neuroblast mis-expressing btd indeed assumes a type II neuroblast functional identity , it should be able to generate immature INPs capable of maturing into an INP , a process critically dependent on the function of brat and erm . While removing brat function had no effects on the identities of progeny derived from control type I neuroblasts , it led to supernumerary type II neuroblast formation in the lineage clones derived from single type I neuroblasts mis-expressing btd ( Figure 5E–F ) . Similarly , removing erm function also led to supernumerary type II neuroblast formation in the lineage clones derived from single type I neuroblast mis-expressing btd while not having any effects on the control type I neuroblast clones ( Figure 5G–H ) . Since brat and erm function specifically in the immature INP to promote an INP identity ( Xiao et al . , 2012; Janssens et al . , 2014; Komori et al . , 2014 ) , these data strongly suggest that mis-expression of btd was sufficient to endow a type I neuroblast with the competence to generate INPs . Thus , we conclude that btd plays an important role in eliciting the functional identity of a type II neuroblast . Finally , we tested whether Trx maintains the type II neuroblast functional identity through btd . Consistent with our hypothesis , 40% of trx mutant type II over-expressing btd regained the characteristics that are specific for a type II neuroblast including loss of Ase expression and the generation of immature INPs and INPs ( Figure 6A–C ) . Furthermore , over-expression of btd also significantly enabled trx mutant type II neuroblasts to generate INPs ( Figure 6D ) . Thus , we conclude that btd is a key downstream target gene of Trx in the maintenance of the type II neuroblast functional identity . 10 . 7554/eLife . 03502 . 014Figure 6 . Over-expression of btd restores a type II neuroblast functional identity in trx mutant type II neuroblasts . ( A–D ) Overexpression of btd reinstates the ability to generate INPs in trx mutant type II neuroblasts . ( A–B ) In the 72-hr clones , while the control trx mutant type II neuroblasts are surrounded by GMCs , trx mutant type II neuroblasts over-expressing btd are surrounded by INP progeny . Three-dimensionally reconstructed images of the clones are shown to the right . Scale bar , 10 μm . ( C ) The neuroblast marker expression profile displayed by type II neuroblasts of the indicated genotypes . ( D ) The average number of INPs per clone of the indicated genotypes . DOI: http://dx . doi . org/10 . 7554/eLife . 03502 . 014 The SET1/MLL complex elicits biological responses by maintaining its target genes in an active state through the methylation of H3K4 ( Shilatifard , 2012 ) . Our data showed that the core components of the SET1/MLL complex is required for the maintenance of the H3K4 methylation in a type II neuroblast and the maintenance of a type II neuroblast functional identity ( Figure 3C–D , F , Figure 3—figure supplement1 ) . Most importantly , over-expression of rbbp5FL , but not rbbp5SG , which encodes a mutant Rbbp5 protein that partially compromises the H3K4 methylation activity of the SET1/MLL complex ( Cao et al . , 2010 ) , restored both H3K4 methylation and a type II neuroblast functional identity in rbbp5 null type II neuroblasts ( Figure 3C–K ) . These results indicate that the H3K4 methylation activity of the SET1/MLL complex is required for maintaining the functional identity of a type II neuroblast . In the fly genome , Trx , Trr , and dSet1 can each bind to the core components of the SET1/MLL complex ( Wu et al . , 2008; Ardehali et al . , 2011; Mohan et al . , 2011 ) . Although the methylation activity of Trx was required for maintaining the type II neuroblast functional identity , removing trx function did not alter the global H3K4 methylation ( Figure 3A–B , Figure 4—figure supplement 1K–N ) . In contrast , knocking down the function of trr or dset1 did not affect the maintenance of a type II neuroblast functional identity despite resulting in the global loss of H3K4 mono- or tri-methylation ( Figure 4—figure supplement 1A–J ) . These data strongly suggest that Trx maintains a type II neuroblast functional identity by regulating H3K4 methylation in specific downstream target loci . The functional identity of a type II neuroblast is defined by the competence of a neuroblast to generate INPs ( Weng and Lee , 2011; Homem and Knoblich , 2012; Janssens and Lee , 2014 ) . Our data indicate Trx plays a central role in maintaining the functional identity of a type II neuroblast by promoting the expression of a small number of genes ( Figures 1 and 4A ) . We identified the btd gene as a critical downstream target of Trx that is both necessary and sufficient for the regulation of the type II neuroblast functional identity ( Figures 4–7 ) . btd encodes a C2H2 zinc finger transcription factor required for proper patterning of the head segment during fly embryogenesis and likely functions as a transcription activator ( Wimmer et al . , 1993; Schöck et al . , 1999 ) . However , the role of Btd in regulating neuroblasts has never been established , and the mechanisms by which Btd elicits biological responses remain unclear . Several possible reasons exist to explain the relatively inefficient nature of eliciting the type II neuroblast functional identity in a type I neuroblast by the mis-expression of btd ( Figure 5 ) . First , certain co-factors might be required for Btd to efficiently activate its target gene transcription , and a lower abundance of these co-factors in type I neuroblasts hinders the functional output of mis-expressed Btd . Second , the epigenetic landscape might be vastly different between the two types of neuroblasts such that mis-expressed Btd may not have access to all of its target genes required to elicit the type II neuroblast functional identity in a type I neuroblast . Lastly , additional transcription factors might function in parallel with Btd to regulate the functional identity of a type II neuroblast . Btd is a highly conserved transcription factor ( Estella and Mann , 2010; MuhChyi et al . , 2013 ) . Future studies to elucidate the mechanisms by which Btd regulates the functional identity of a type II neuroblast will provide critical insight in the regulation of neural stem cell heterogeneity during both invertebrate as well as vertebrate neurogenesis . We identified the pnt gene as another direct downstream target of Trx ( Figure 4A , Figure 4—figure supplement 2A ) . We initially hypothesized that Pnt might function in parallel with Btd to maintain the functional identity of a type II neuroblast . This hypothesis was extremely appealing in light of a previous study demonstrating mis-expression of PntP1 can transform a type I neuroblast into a type II neuroblast ( Zhu et al . , 2011 ) . Unexpectedly , knocking down the function of the pnt gene , which encodes at least three alternatively spliced transcripts , had no effect on the maintenance of the type II neuroblast functional identity , and instead , resulted in the formation of supernumerary type II neuroblasts ( Figure 4—figure supplement 2 ) . This result led us to revise our hypothesis and propose that Pnt functions in the immature INP to specify an INP identity . Consistently , heterozygosity of the pnt locus dominantly enhanced the supernumerary neuroblast in the brat or erm hypomorphic genetic background ( Figure 4—figure supplement 2G ) . These two genetic backgrounds have been used extensively for elucidating the mechanisms that regulate the specification of an INP identity in the immature INP ( Xiao et al . , 2012; Janssens et al . , 2014; Komori et al . , 2014 ) . Furthermore , over-expression of pntP1 failed to restore the functional identity of a type II neuroblast in trx mutant type II neuroblasts ( data not presented ) . Together , these data strongly suggest that pnt mainly functions to specify an INP identity rather than to maintain the type II neuroblast functional identity . Thus , we propose that in addition to maintaining the type II neuroblast functional identity , Trx also functions to promote INP identity specification through pnt ( Figure 7 ) . Strategies that uniquely target the functional properties of cancer stem cells will revolutionize cancer treatments . Cancer stem cells generate a hierarchy of progeny that include cell types directly contributing to the exponential expansion of cancer stem cells ( Magee et al . , 2012 ) . Thus , reprogramming their functional identity to bypass the cell types that directly contribute to the exponential expansion of cancer stem cells should halt further tumor growth . In our study , removing trx function efficiently reduced the number of supernumerary type II neuroblasts , which are proposed to serve as cancer stem cells in several Drosophila brain tumor models ( Caussinus and Gonzalez , 2005; Xiao et al . , 2012; Eroglu et al . , 2014; Janssens et al . , 2014; Koe et al . , 2014; Komori et al . , 2014 ) , and increased the number of differentiated cells in the brat or erm mutant brain ( Figure 2 ) . Similarly , attenuating the competence of type II neuroblasts to generate INPs by removing btd function also efficiently halted the expansion of brat or erm mutant brain tumors ( Figure 4I–J , data not presented ) . Our results strongly support the hypothesis that reprogramming the functional identity of putative cancer stem cells can significantly alter the course of tumorigenesis . As such , understanding the mechanisms that maintain stem cell heterogeneity during normal development might provide novel insight into designing rational therapies to promote switching of cancer stem cells to an alternative , non-cancerous stem cell type . Fly strains used in this study include Oregon R , Ase-Gal4 ( Zhu et al . , 2006 ) , Ase-Gal80 ( Neumüller et al . , 2011 ) , bratDG19310 , bratk06028 and brat11 ( Komori et al . , 2014 ) , erm1 and erm2 ( Weng et al . , 2010 ) , erm-flag ( Janssens et al . , 2014 ) , erm-lacZ and UAS-aPKCCAAX ( Haenfler et al . , 2012 ) , pntΔ88 ( Morimoto et al . , 1996 ) , trxZ11 ( Tie et al . , 2014 ) , and Wor-Gal4 ( Lee et al . , 2006 ) . The following stocks were obtained from the Bloomington Drosophila Stock Center: Elav-GAL4 , Act-FRT-Stop-FRT-GAL4 , ash21 , btdXA , FRT19A , FRT2A , FRT82B , GMR85C07-GAL4 ( Btd-GAL4 ) , hs-flp , P ( EP ) G4226 , pros17 , UAS-pntRNAi ( TRiP . JF02227 ) , UAS-pntRNAi ( TRiP . HMS01452 ) , trxE2 , tubP-Gal80 , tubP-Gal80 ts , UAS-Dcr-2 . D , UAS-mCD8-GFP , and UAS-trrRNAi ( TRiP . JF03242 ) . We obtained the following stocks from the Vienna Drosophila RNAi Center UAS-ash2RNAi ( 100718 ) , UAS-dSet1RNAi ( 40683 ) , UAS-pntRNAi ( 7171 ) , UAS-rbbp5RNAi ( 106139 ) , UAS-trxRNAi ( 108122 ) , and UAS-wdsRNAi ( 105371 ) . UAS-HA-btd , UAS-HA-pntP1 , UAS-rbbp5FL-myc , and UAS-rbbp5SG-myc were generated in this study by cloning the cDNA cloned into p{UAST}attB vector . The transgenic fly lines were generated via ϕC31 integrase-mediated transgenesis ( Bischof and Basler , 2008 ) . The rbbp5 null allele was generated by imprecisely excising the P ( EP ) G4226 element . Clones were induced following previously published methods ( Janssens et al . , 2014 ) . Three-dimensional model of clones was generated using the Mimics software from Materialize , Leuven , Belgium . Confocal images were acquired using a Z-step size of 1 . 5 μm , and the identity of every cell within a clone was determined individually . Larvae brains were dissected in Schneider's medium ( Sigma , St . Louis , MO ) and fixed in 100 mM Pipes ( pH 6 . 9 ) , 1 mM EGTA , 0 . 3% Triton X-100 , and 1 mM MgSO4 containing 4% formaldehyde for 23 min . Larval brains were processed for immunofluorescent staining according to a previously published protocol ( Weng et al . , 2012 ) . Antibodies used in this study include chicken anti-GFP ( 1:2000; Aves Labs , Tigard , OR ) , guinea pig anti-Ase ( 1:1000; Wang H ) , mouse anti-cMyc ( 1:100 Roche , Basel , Switzerland ) , mouse anti-Pros ( MR1A; 1:500; DSHB , Iowa city , IA ) , rabbit anti-Ase ( 1:400 ) , rabbit anti-β-gal ( 1:1000; MP Biomedicals , Santa Ana , CA ) , rabbit anti-H3K4me1 ( 1:500; Abcam , Cambridge , United Kingdom ) , rabbit anti-H3K4me3 ( 1:500; Active motif , Carlsbad , CA ) , rabbit anti-Phospho-Histone-H3 ( Ser10 ) ( 1:1000; EMD Millipore , Billerica , MA ) , rabbit anti-PntP1 ( 1:600; Skeath JB ) , rat anti-Dpn ( 1:2 ) , rat anti-Mira ( 1:500 ) . Secondary antibodies were from Jackson ImmunoResearch Inc . , West Grove , PA . The confocal images were acquired on a Leica SP5 scanning confocal microscope ( Leica Microsystems Inc . , Buffalo Grove , IL ) . To obtain more than 2 × 106 supernumerary type II neuroblasts , we dissected 100 brains from brat mutant larvae aged for 4 days at 33°C in Schneider's medium ( Sigma , St . Louis , MO ) and fixed in 1 . 8% formaldehyde solution for 20 min . We stopped fixation by incubating the lysate with Glycine ( 0 . 25 M ) at room temperature for 4 min and on ice for 10 min . Following fixation , samples were washed with wash buffer ( 1xPBS , 5 mM Tris–HCl pH7 . 5 , 1 mM EDTA ) containing proteinase inhibitors ( Roche , Basel , Switzerland ) and 1 mM PMSF for three times and homogenized in SDS lysis buffer ( 1% SDS , 50 mM Tris–HCl pH8 . 1 , 10 mM EDTA ) to obtain nuclear extracts . The nuclear extracts were disrupted by using a sonicator ( 18 cycles of sonicating for 30 s and interval for 30 s ) . Five percent of the sonicated sample was stored for INPUT . The rest of the sonicated chromatin was incubated with antibodies in ChIP dilution buffer ( 0 . 01% SDS , 1 . 1% Trition X-100 , 1 . 2 mM EDTA , 16 . 7 mM Tris–HCl pH8 . 1 , 167 mM NaCl ) at 4°C overnight . Samples were incubated with Dynal beads ( Life technologies , Grand Island , NY ) at 4°C overnight , washed twice with low salt immune complex wash buffer ( 0 . 1% SDS , 1% TritonX-100 , 2 mM EDTA , 20 mM Tris–HCl pH8 . 1 , 150 mM NaCl ) , twice with high salt immune complex wash buffer ( 0 . 1% SDS , 1% TritonX-100 , 2 mM EDTA , 20 mM Tris–HCl pH8 . 1 , 500 mM NaCl ) , three times with LiCl immune complex wash buffer ( 0 . 25 M LiCl , 1% NP40 , 1% deoxycholate , 1 mM EDTA , 10 mM Tris–HCl pH8 . 1 ) , twice with TE buffer , and then were eluted from beads . Cross-linking of chromatin–protein complex was reverted at 65°C overnight . Samples were treated with RNase A at 55°C for 2 hr and incubated with 2 μg of proteinase K at 45°C for 1 hr . Samples were cleaned up by phenol:chloroform and precipitated by EtOH precipitation . Samples were resuspended in 100 μl of water . 5 μl were used in each qPCR reaction . Antibodies used in this experiment were anti-Trx antibody ( Mazo A ) , anti-H3K4me2 ( 07–030; Millipre , Billerica , MA ) , and rabbit IgG ( ab46540; Abcam ) . The following individual specific primer sets were used for quantitative PCR: btd-E1 , 5ʹ-gttggccattgcgtgtcctgtttc-3ʹ and 5ʹ-gccccgctgcgctctatcca-3ʹ , btd-E2 , 5ʹ-ggattaccgcagacgat-3ʹ and 5ʹ-ggttggccggtggttgagt-3ʹ , btd-TSS , 5ʹ-cagcagcagcagcagcaacagt-3ʹ and 5ʹ-gtcggcccgggtccaagtaa-3ʹ , negative control , 5ʹ-cagcagcagcagcagcaacagt-3ʹ and 5ʹ-gtcggcccgggtccaagtaa-3ʹ , pntP1-TSS , 5ʹ-tttggtgttgttgtttttcttctt , -3ʹ and 5ʹ-acgcgttctgttctgtttt-3ʹ . Another negative control primer set was used in previously published paper ( Petruk et al . , 2012 ) . Total RNA was extracted following the standard Trizol RNA isolation protocol ( Life technologies , Grand Island , NY ) and cleaned by the RNeasy kit ( Qiagen , Venlo , Netherlands ) . First strand cDNA was synthesized from the extracted total RNA using First Strand cDNA Synthesis Kit for RT-PCR ( AMV ) ( Roche , Basel , Switzerland ) . qPCR was performed using ABsolute QPCR SYBR Green ROX Mix ( Thermo Fisher Scientific Inc . , Waltham , MA ) . Data were analyzed by the comparative CT method , and the relative mRNA expression is presented . The following individual specific primer sets were used for quantitative PCR: ase , 5ʹ-agcccgtgagcttctacgac-3ʹ and 5ʹ-gcatcgatcatgctctcgtc-3ʹ , btd , 5ʹ-gcacggacgtacgcacaccaat-3ʹ and 5ʹ-cctcggcggccaataccttct-3ʹ , dpn , 5ʹ-catcatgccgaacacaggtt-3ʹ and 5ʹ-gaagattggccggaactgag-3ʹ , elav , 5ʹ-gcggcgcgtatcccattttcatct-3ʹ and 5ʹ-tggccgcctcatcgtagttggtca-3ʹ , pntP1 , 5ʹ-ggcagtacgggcagcaccac-3ʹ and 5ʹ-ctcaacgcccccaccagatt-3ʹ .

Whereas the majority of cells in the brain are unable to divide to produce new cells , neural stem cells can divide numerous times and have the potential to become many different types of brain cells . However , between these two extremes there is another group of cells called neural progenitors . These cells can give rise to multiple types of neurons but , in contrast to stem cells , they can undergo only a limited number of divisions . Many of the molecular mechanisms by which stem cells give rise to progenitors are similar in mammals and in the fruit fly Drosophila . In the brains of fly larvae , a subset of neural stem cells called type II neuroblasts give rise to ‘intermediate neural progenitors’ , each of which can divide between four and six times . Every division generates a replacement intermediate neural progenitor and a cell called a ganglion mother cell , which divides one last time to produce two brain cells . Thus , intermediate neural progenitors increase the overall output of cells derived from every division of a type II neuroblast . The ability of type II neuroblasts to generate intermediate neural progenitors is important for development . Loss of this ability will result in a shortage of cells , disrupting brain development , while the faulty generation of intermediate neural progenitors will result in the formation of tumors . Now , using Drosophila brain cells cultured in the laboratory , Komori et al . show that an evolutionarily conserved enzyme called Trithorax has an important role in maintaining this ability . Trithorax acts through a protein called Buttonhead . The role of Buttonhead in regulating intermediate neural progenitors has also been identified by Xie et al . Komori et al . show that type II neuroblasts that lack Trithorax activity lose their unique identity and behave as type I neuroblasts , which never generate intermediate neural progenitors . Trithorax maintains the cellular memory of a type II neuroblast by keeping regions of chromatin—a macromolecule made of DNA and proteins called histones—in an active state . These regions contain key genes , such as the gene for Buttonhead . Re-introducing Buttonhead in type II neuroblasts that lack Trithorax activity can reinstate their ability to produce intermediate neural progenitors .

2014

Trithorax maintains the functional heterogeneity of neural stem cells through the transcription factor Buttonhead

Innovations in metazoan development arise from evolutionary modification of gene regulatory networks ( GRNs ) . We report widespread cryptic variation in the requirement for two key regulatory inputs , SKN-1/Nrf2 and MOM-2/Wnt , into the C . elegans endoderm GRN . While some natural isolates show a nearly absolute requirement for these two regulators , in others , most embryos differentiate endoderm in their absence . GWAS and analysis of recombinant inbred lines reveal multiple genetic regions underlying this broad phenotypic variation . We observe a reciprocal trend , in which genomic variants , or knockdown of endoderm regulatory genes , that result in a high SKN-1 requirement often show low MOM-2/Wnt requirement and vice-versa , suggesting that cryptic variation in the endoderm GRN may be tuned by opposing requirements for these two key regulatory inputs . These findings reveal that while the downstream components in the endoderm GRN are common across metazoan phylogeny , initiating regulatory inputs are remarkably plastic even within a single species . While the core regulatory machinery that specifies embryonic germ layers and major organ identity in the ancestor of modern animals is largely conserved in all extant animals , GRN architecture must be able to accommodate substantial plasticity to allow for evolutionary innovation in developmental strategies , changes in selective pressures , and genetic drift ( Peter and Davidson , 2011; Félix and Wagner , 2008 ) . Genetic variation , often with neutral effects on fitness , provides for plasticity in GRN structure and implementation ( Félix and Wagner , 2008 ) . Although studies of laboratory strains of model organisms with a defined genetic background have been highly informative in identifying the key regulatory nodes in GRNs that specify developmental processes ( Davidson and Levine , 2008; Oliveri et al . , 2008; Peter and Davidson , 2017 ) , these approaches do not reveal the evolutionary basis for plasticity in these networks . The variation and incipient changes in GRN function and architecture can be discovered by analyzing phenotypic differences resulting from natural genetic variation present in distinct isolates of a single species ( Milloz et al . , 2008; Nunes et al . , 2013; Phinchongsakuldit et al . , 2004 ) . The endoderm has been proposed to be the most ancient of the three embryonic germ layers in metazoans ( Hashimshony et al . , 2015; Rodaway and Patient , 2001 ) , having appeared prior to the advent of the Bilateria about 600 Mya ( Peterson et al . , 2004 ) . It follows , therefore , that the GRN for endoderm in extant animals has undergone substantial modifications over the long evolutionary time span since its emergence . However , the core transcriptional machinery for endoderm specification and differentiation appears to share common mechanisms across metazoan phylogeny . For example , cascades of GATA-type transcription factors function to promote endoderm development not only in triploblastic animals but also in the most basal metazoans that contain a digestive tract ( Martindale et al . , 2004; Boyle and Seaver , 2008; Boyle and Seaver , 2010; Gillis et al . , 2007; Davidson et al . , 2002 ) . Among the many observations supporting a common regulatory mechanism for establishing the endoderm , it has been found that the endoderm-determining GATA factor , END-1 , in the nematode C . elegans , is sufficient to activate endoderm development in cells that would otherwise become ectoderm in Xenopus ( Shoichet et al . , 2000 ) . This indicates that the role of GATA factors in endoderm development has been preserved since the nematodes and vertebrates diverged from a common ancestor that lived perhaps 600 Mya . To assess the genetic basis for evolutionary plasticity and cryptic variation underlying early embryonic germ layer specification , we have analyzed the well-described GRN for endoderm specification in C . elegans . The E cell , which is produced in the very early C . elegans embryo , is the progenitor of the entire endoderm , which subsequently gives rise exclusively to the intestine . The EMS blastomere at the four-cell stage divides to produce the E founder cell and its anterior sister , the MS founder cell , which is the progenitor for much of the mesoderm ( Sulston et al . , 1983 ) . Both E and MS fates are determined by maternally provided SKN-1 , an orthologue of the vertebrate Nrf2 bZIP transcription factor ( Bowerman et al . , 1992; Bowerman et al . , 1993; Maduro and Rothman , 2002 ) . In the laboratory N2 strain , elimination of maternal SKN-1 function ( through either knockdown or knockout ) results in fully penetrant embryonic lethality as a result of misspecification of EMS cell descendants . In these embryos , the fate of MS is transformed to that of its cousin , the mesectodermal progenitor C cell . E cells similarly adopt a C cell-like fate in a majority , but not all , of these embryos ( Bowerman et al . , 1992 ) . SKN-1 initiates mesendoderm development via the GRN in E and MS cells in part by activating zygotic expression of the MED-1/2 divergent GATA transcription factors ( Maduro et al . , 2007; Maduro et al . , 2001 ) . This event mobilizes a cascade of GATA factors in the E cell lineage that ultimately direct intestinal differentiation ( Maduro and Rothman , 2002; Maduro , 2017; Wiesenfahrt et al . , 2016; McGhee , 2007 ) . This differential requirement for SKN-1 in endoderm ( E ) and mesoderm ( MS ) development is determined by its combinatorial action with triply redundant Wnt , MAPK , and Src signaling systems , which act together to polarize EMS ( Meneghini et al . , 1999; Shin et al . , 1999; Bei et al . , 2002; Thorpe et al . , 1997 ) . MOM-2/Wnt acts through the MOM-5/Frizzled receptor , mobilizing WRM-1/β-catenin , resulting in its cytoplasmic accumulation in the posterior side of EMS . WRM-1 , together with LIT-1/NLK kinase , alters both the nucleocytoplasmic distribution and activity of the Wnt effector POP-1/Tcf ( Thorpe et al . , 1997; Nakamura et al . , 2005; Rocheleau et al . , 1999 ) , converting it from a repressor of endoderm in the MS cell lineage to an activator in the E cell lineage ( Owraghi et al . , 2010; Huang et al . , 2007; Maduro et al . , 2002; Phillips et al . , 2007; Maduro et al . , 2005a; Shetty et al . , 2005 ) . Loss of MOM-2 expression in the laboratory N2 strain results in a partial gutless phenotype , while removal of both MOM-2 and SKN-1 , through either knockdown or knockout , leads to a completely penetrant loss of gut ( Thorpe et al . , 1997 ) , revealing their genetically redundant roles . The regulatory relationship between SKN-1 and POP-1 , the effector of Wnt signaling , shows substantial variation even in species that diverged 20–40 million years ago , suggesting significant evolutionary plasticity in this key node in the endoderm GRN . C . elegans embryos lacking maternal POP-1 always make gut , both in the normal E cell lineage and in the MS cell lineage . However , in embryos lacking both SKN-1 and POP-1 , endoderm is virtually never made , implying that these two factors constitute a Boolean ‘OR’ logic gate . In contrast , removal of either SKN-1 or POP-1 alone in C . briggsae causes >90% of embryos to lack gut , indicative of an ‘AND’ logic gate ( Figure 1A , B ) ( Lin et al . , 2009 ) . In this study , we sought to determine whether the plasticity in regulatory logic of the two major inputs into endoderm development is evident within a single species . The availability of many naturally inbred variants ( isotypes ) of C . elegans that show widespread genomic variation ( Félix and Braendle , 2010; Cook et al . , 2017; Andersen et al . , 2012 ) , provides a genetically rich resource for investigating potential quantitative variation in developmental GRNs . We report here that the requirement for activation of the endoderm GRN by SKN-1 or MOM-2 , but not POP-1 , is highly variable between natural C . elegans isolates , and even between closely related isotypes . Genome-wide association studies ( GWAS ) in isolates from the natural populations and targeted analysis of recombinant inbred lines ( RILs ) , revealed that a multiplicity of loci and their interactions are responsible for the variation in the developmental requirement for SKN-1 and MOM-2 . We found a complex , but frequently reciprocal requirement for SKN-1 and MOM-2 among variants underlying these phenotypes: loci associated with a high requirement for SKN-1 in endoderm development tend to show a more relaxed requirement for MOM-2 and vice-versa . Consistent with this finding , three other endoderm regulatory factors , RICT-1 , PLP-1 , and MIG-5 , show similar inverse relationships between these two GRN inputs . These findings reveal that the activation of the GRN network for a germ layer , one of the most critical early developmental switches in embryos , is subject to remarkable genetic plasticity within a species and that the dynamic and rapid change in network architecture reflects influences distributed across many genetic components that affect both SKN-1 and Wnt pathways . The relationship between SKN-1 and Wnt signaling through POP-1 in the endoderm GRN has undergone substantial divergence in the Caenorhabditis genus ( Lin et al . , 2009 ) . While neither input alone is absolutely required for endoderm specification in C . elegans , each is essential in C . briggsae , which has been estimated to have diverged from C . elegans ~ 20–40 Mya ( Zhao et al . , 2008; Cutter , 2008 ) . In contrast to the C . elegans N2 laboratory strain , removal of either SKN-1 or POP-1 alone results in fully penetrant conversion of the E founder cell fate into that of the mesectodermal C blastomere and of E to MS fate , respectively , in C . briggsae ( Lin et al . , 2009 ) . These findings revealed that the earliest inputs into the endoderm GRN are subject to substantial evolutionary differences between these two species ( Figure 1B ) . We sought to determine whether incipient evolutionary plasticity in this critical node at the earliest stages of endoderm development might be evident even within a single species of the Caenorhabditis genus by assessing their requirement in C . elegans wild isolates and testing whether the quantitative requirements of each input were correlated . Elimination of detectable maternal SKN-1 from the laboratory N2 strain by either a strong ( early nonsense ) chromosomal mutation ( skn-1 ( zu67 ) ) , or by RNAi knockdown , results in a partially penetrant phenotype: while the E cell adopts the fate of the C cell in the majority of embryos , and gut is not made , ~30% of arrested embryos undergo strong gut differentiation , as evidenced by the appearance of birefringent , gut-specific rhabditin granules , or expression of elt-2::GFP , a marker of the developing and differentiated intestine ( Figure 1C–H ) . In our experimental conditions , we found that RNAi of skn-1 in different N2-derived mutant strains gave highly reproducible results: 100% of the embryos derived from skn-1 ( RNAi ) -treated mothers arrest ( n > 100 , 000 ) and 32 . 0 ± 1 . 9% of the arrested embryos exhibited birefringent gut granules ( Figure 2A; Supplementary file 1 ) over many trials by separate investigators . We found that the LSJ1 laboratory strain , which is derived from the same original source as N2 , but experienced very different selective pressures in the laboratory owing to its constant propagation in liquid culture over 40 years ( Sterken et al . , 2015 ) , gave virtually identical results to that of N2 ( 31 . 0% ± s . d 1 . 2% ) , implying that SKN-1-independent endoderm formation is a quantitatively stable trait . The low variability in this assay , and high number of embryos that can be readily examined ( ≥500 embryos per experiment ) , provides a sensitive and highly reliable system with which to analyze genetic variation in the endoderm GRN between independent C . elegans isolates . To assess variation in SKN-1 requirement within the C . elegans species , we analyzed the outcome of knocking down SKN-1 by RNAi in 96 unique C . elegans wild isolates ( Andersen et al . , 2012 ) . Owing to their propagation by self-fertilization , each of the isolates ( isotypes ) is a naturally inbred clonal population that is virtually homozygous and defines a unique haplotype . The reported estimated population nucleotide diversity averages 8 . 3 × 10−4 per bp ( Andersen et al . , 2012 ) , and we found that a substantial fraction ( 29/97 ) of isotypes were quantitatively indistinguishable in phenotype between the N2 and LSJ1 laboratory strains ( Figure 2A , Supplementary file 1 ) . We found that all strains , with the exception of the RNAi-resistant Hawaiian CB4856 strain , invariably gave 100% embryonic lethality with skn-1 ( RNAi ) . However , we observed dramatic variation in the fraction of embryos with differentiated gut across the complete set of strains , ranging from 0 . 9% to 60% ( Figure 2A ) . Repeated measurements with >500 embryos per replicate per strain revealed very high reproducibility ( Figure 2—figure supplement 1 ) , indicating that even small differences in the fraction of embryos generating endoderm could be reproducibly measured . Further , we found that some wild isolates that were subsequently found to have identical genome sequences also gave identical results . Although birefringent and autofluorescent rhabditin granules have been used as a marker of gut specification and differentiation in many studies ( Clokey and Jacobson , 1986; Hermann et al . , 2005 ) , it is conceivable that the variation in fraction of embryos containing this marker that we observed might reflect variations in gut granule formation rather than in gut differentiation per se . We note that embryos from all strains showed a decisive ‘all-or-none’ phenotype: that is , they were either strongly positive for gut differentiation or completely lacked gut granules , with virtually no intermediate or ambiguous phenotypes . A threshold of gene activity in the GRN has been shown to account for such an all-or-none switch in gut specification ( Maduro et al . , 2007; Raj et al . , 2010; Zhu et al . , 1997 ) . This observation is inconsistent with possible variation in gut granule production: if SKN-1-depleted embryos were defective in formation of the many granules present in each gut cell , one might expect to observe gradations in numbers or signal intensity of these granules between gut cells or across a set of embryos . Nonetheless , we extended our findings by analyzing the expression of the gut-specific intermediate filament IFB-2 , a marker of late gut differentiation , in selected strains representing the spectrum of phenotypes observed ( Figure 2B ) . As with gut granules , we found that embryos showed all-or-none expression of IFB-2 . In all cases , we found that the fraction of embryos containing immunoreactive IFB-2 was not significantly different ( Fisher’s exact test , p-values>0 . 05 ) from the fraction containing gut granules , strongly suggesting that the strains vary in endoderm specification per se and consistent with earlier studies of SKN-1 function ( Bowerman et al . , 1992; Maduro et al . , 2007 ) . Although we found that skn-1 ( RNAi ) was 100% effective at inducing embryonic lethality in all strains ( with the exception of the RNAi-defective Hawaiian strain , CB4856 ) , it is conceivable that , at least for the strains that showed a weaker phenotype than for N2 ( i . e . , higher number of embryos specifying endoderm ) , the variation observed between strains was attributable to differences in RNAi efficacy rather than in the endoderm GRN . Indeed , studies with N2 and CB4856 showed that germline RNAi sensitivity is a quantitative trait , involving the Argonaute-encoding ppw-1 gene and additional interacting loci present in some wild isolates ( Tijsterman et al . , 2002; Pollard and Rockman , 2013 ) . To address this possibility , we introgressed the strong loss-of-function skn-1 ( zu67 ) chromosomal mutation into five wild isolates whose phenotypes spanned the spectrum observed ( ranging from 2% of embryos with differentiated gut for MY16% to 50% for MY1 ) ( Figure 2C ) . In all cases , we found that introgression of the allele through five rounds of backcrosses resulted in a quantitative phenotype that was similar or indistinguishable from that observed with skn-1 ( RNAi ) . The phenotypes of the introgressed allele were significantly different ( p-values<0 . 01 ) from that of the parental N2 skn-1 ( zu67 ) strain , except for DL238 , whose skn-1 ( RNAi ) phenotype was indistinguishable from that of N2 . The results obtained by introgression from four of the isotypes ( CX11262 , DL238 , EG4724 and MY1 ) , were not statistically different ( Student t-test , p-values>0 . 05 ) from the corresponding RNAi knockdown results ( Figure 2C ) ( i . e . , the phenotype was suppressed or enhanced relative to N2 in these genetic backgrounds to the same extent as with skn-1 ( RNAi ) ) . However , while the MY16 skn-1 ( zu67 ) strain shifted in the predicted direction ( i . e . , became stronger ) as compared to the N2 strain , it showed a weaker phenotype than was evident by RNAi knockdown , even following eight rounds of introgression . Regardless , diminished RNAi efficacy in MY16 cannot explain the large difference between the skn-1 ( RNAi ) phenotype of N2 and MY16 , as the latter phenotype is , in fact , much stronger , not weaker , than the former . As described below , we identified a modifier locus in the MY16 strain that is closely linked to the skn-1 gene; it therefore seems likely that the N2 chromosomal segment containing this modifier was carried with the skn-1 ( zu67 ) mutation through the introgression crosses , thereby explaining the somewhat weaker phenotype of the introgressed allele in MY16 . We conclude that the extreme variation in skn-1 ( RNAi ) phenotype between the wild isolates tested results from bona fide cryptic variation in the endoderm GRN , rather than differences in RNAi efficacy . We note that the strength of skn-1 ( RNAi ) phenotype does not correlate with phylogenetic relatedness between the strains ( Pagel’s λ = 0 . 42 , p-value=0 . 14 ) . For example , while some closely related strains ( e . g . , MY16 and MY23 ) showed a similar phenotype , other very closely related strains ( e . g . , JU1491 and JU778 ) showed phenotypes on the opposite ends of the phenotypic spectrum ( Figure 3A ) . We also did not observe any clear association between geographical distribution and skn-1 ( RNAi ) phenotype ( Figure 3B ) . These findings suggest that the initiating inputs into the endoderm GRN is subject to rapid intraspecies evolutionary divergence . The switch in the relationship of the SKN-1 and Wnt inputs between C . elegans ( ‘OR’ operator ) and C . briggsae ( ‘AND’ operator ) ( Lin et al . , 2009 ) , and the extensive variation in the requirement for SKN-1 seen across C . elegans isolates , raised the possibility that the quantitative requirement for Wnt components might vary between unique isolates of C . elegans . It has been shown that signaling from Ras pathway varies in different C . elegans wild isolates and hyperactive Wnt signaling can compensate for reduced Ras activity in the vulva signaling network ( Milloz et al . , 2008; Gleason et al . , 2002 ) . Given that removal of the maternal Wnt input also results in a partially penetrant gut defect ( through either knock-out or knockdown of Wnt signaling components ) , it is conceivable that a compensatory relationship may exist between the SKN-1 and Wnt inputs . We investigated this possibility by examining the requirement for the MOM-2/Wnt ligand in the same wild isolates that were tested for the SKN-1 gut developmental requirement . Indeed , we observed broad variation in the requirement for MOM-2/Wnt in activation of the endoderm GRN between isotypes . mom-2 ( RNAi ) of 94 isotypes resulted in embryonic arrest , indicating that , as with skn-1 ( RNAi ) , mom-2 ( RNAi ) was effective at least by the criterion of lethality . Two isotypes , CB4853 and EG4349 , did not exhibit mom-2 ( RNAi ) -induced lethality and were omitted from further analyses . In the affected strains , the fraction of mom-2 ( RNAi ) embryos with differentiated gut varied from ~40% to~99% ( Figure 4A , Supplementary file 1 ) . As with skn-1 ( RNAi ) , the mom-2 ( RNAi ) phenotype of isotypes N2 , JU440 , and JU1213 was further confirmed by immunostaining with IFB-2 ( Figure 4B ) , again demonstrating that birefringence of gut granules is a reliable proxy for endoderm formation for this analysis . To assess whether the observed variation in the mom-2 ( RNAi ) phenotype reflected differences in the GRN or RNAi efficacy , the mom-2 ( or42 ) allele was introgressed into three different genetic backgrounds chosen from the extreme ends of the phenotypic spectrum . mom-2 ( RNAi ) of the laboratory N2 strain resulted in the developmental arrest of embryos . Of those , ~90% contained differentiated endoderm , a result that was highly reproducible . In contrast , the introgression of an apparent loss-of-function allele , mom-2 ( or42 ) , into the N2 strain results in a more extreme phenotype: only ~28% of embryos show endoderm differentiation ( Figure 4C ) ( Thorpe et al . , 1997 ) . While this discrepancy can partly be explained by incomplete RNAi efficacy , it is notable that the penetrance of mom-2 alleles vary widely ( Thorpe et al . , 1997 ) . We observed strain-specific variation in embryonic lethality response to RNAi of mom-2 between the different isotypes . However , we found that the mom-2 ( or42 ) introgressed strains show qualitatively similar effects to those observed with mom-2 RNAi . For example , the mom-2 ( or42 ) allele introgressed into the isotype JU1213 background resulted in a low fraction of arrested embryos with gut ( 5 . 7% ± s . d 2 . 4%; n = 2292 ) , a more extreme effect than was seen with RNAi ( 34 . 0% ± s . d 1 . 5%; n = 1876 ) . This is the strongest phenotype that has been reported for any known mom-2 allele . On the other hand , introgression of the mom-2 mutation gave rise to a significantly higher fraction of embryos with endoderm in isotypes DL226 ( 55 . 2% ± s . d 1 . 2% , n = 1377 ) and PB303 ( 65 . 5% ± s . d 4 . 9% , n = 2726 ) , relative to the laboratory strain N2 ( 29 . 1% ± s . d 3 . 1%; n = 1693 ) , consistent with the RNAi phenotypes ( Figure 4C ) . These findings indicate that the differential requirement for MOM-2 is , at least in part , attributable to genetic modifiers in these strains . As with skn-1 ( RNAi ) , we found no correlation between the mom-2 ( RNAi ) phenotype and phylogenetic relatedness or geographical distribution ( Figure 4—figure supplement 1 ) , suggesting rapid intraspecies developmental system drift . As the MOM-2/Wnt signal is mediated through the POP-1 transcription factor , we sought to determine whether the requirement for POP-1 might also vary between isolates . We found that , while pop-1 ( RNAi ) resulted in 100% embryonic lethality across all 96 RNAi-sensitive isolates , 100% of the arrested embryos contained a differentiated gut ( n > 500 for each isolate scored ) ( results not shown ) . Thus , all isolates behave similarly to the N2 strain with respect to the requirement for POP-1 . These results were confirmed by introgressing a strong loss-of-function pop-1 ( zu189 ) allele into four wild isolates ( N2 , MY16 , JU440 , and KR314 ) ( Figure 4—figure supplement 2 ) . The lack of variation in endoderm specification after loss of POP-1 is not entirely unexpected . As has been observed in a pop-1 ( - ) mutant strain , elimination of the endoderm-repressive role of POP-1 in the MS lineage ( which is not influenced by the P2 signal ) supersedes its endoderm activating role in the presence of SKN-1 . Indeed , the original observation that all pop-1 ( - ) embryos in an N2 background contain gut masked the activating function for POP-1 , which is apparent only in the absence of SKN-1 ( Owraghi et al . , 2010; Maduro et al . , 2002; Maduro et al . , 2005a ) . It is likely that , as with the N2 strain , gut arises from both E and MS cells in all of these strains; however , as we have scored only for presence or absence of gut , it is conceivable that the E lineage is not properly specified in some strains , a possibility that cannot be ruled out without higher resolution analysis . Our results contrast with those of Paaby et al . ( 2015 ) , who reported that RNAi of 29 maternal-effect genes across a set of 55 wild isolates in liquid culture resulted in generally weaker effects on lethality than we observed . This difference is likely attributable to diminished and variable RNAi efficacy in the latter study owing to the different culture methods used ( see Materials and methods ) ( Çelen et al . , 2018; Gomez-Amaro et al . , 2015 ) . To assess this possibility further , we compared our results with those of Paaby et al . ( 2015 ) and found no correlation between the variation in fraction of embryos with gut and the lethality observed in that report with both mom-2 ( RNAi ) and skn-1 ( RNAi ) ( Pearson’s R = 0 . 19 , p=0 . 23; Pearson’s R = 0 . 22 , p=0 . 17 , respectively ) . In addition , Paaby et al . ( 2015 ) found that the genetically divergent strain QX1211 consistently showed weak penetrance across all targeted genes , while under our experimental conditions , QX1211 exhibited a slightly stronger skn-1 ( RNAi ) phenotype ( 25 . 2% vs . 32 . 0% , Fisher’s exact test p-value=0 . 03 ) and a similar mom-2 ( RNAi ) phenotype ( 90% vs . 90% , Fisher’s exact test p-value=0 . 9 ) compared to the N2 strains with fully penetrant lethality in all cases . We sought to examine the genetic basis for the wide variation in SKN-1 and Wnt requirements across C . elegans isolates and to evaluate possible relationships in the variation seen with the SKN-1 and Wnt inputs by performing GWAS using the available SNP markers and map ( Andersen et al . , 2012 ) , adjusting for population structure by using Efficient Mixed-Model Analysis ( EMMA ) ( Figure 5A , B ) ( Kang et al . , 2008; Wang , 2002 ) . This approach identified two significant closely-located positions on chromosome IV that underlie the variation in SKN-1 requirement ( Figure 5A , Table 1 ) . GWAS of the mom-2 ( RNAi ) variation proved more challenging because this phenotype showed a highly skewed distribution ( Shapiro-Wilk’ test W = 0 . 8682 , p-value=1 . 207×10−7 ) ( Figure 5—figure supplement 1 ) . While GWAS did not reveal any genomic regions for the mom-2 ( RNAi ) variation that exceeded an FDR of 5% , we found that the most strongly associated loci for the mom-2 ( RNAi ) phenotype also showed large effects for skn-1 ( RNAi ) ( Figure 5C ) . In particular , we observed substantial overlap in the p-values for individual SNPs from skn-1 ( RNAi ) and mom-2 ( RNAi ) in the central region of chromosome IV ( Figure 5—figure supplement 2 ) , raising the possibility that common genetic factors might underlie these phenotypes . In an effort to narrow in on causal loci underlying the skn-1 ( - ) and mom-2 ( - ) phenotypic variation , and to assess possible relationships between these two GRN inputs , we prepared and analyzed 95 recombinant inbred lines ( RILs ) between two C . elegans isotypes , N2 and MY16 . These strains were chosen for their widely varying differences in requirement for both inputs ( see Materials and methods ) . In contrast to the very low variation seen between multiple trials of each parental strain , analysis of the RNAi-treated RIL strains ( >500 embryos/RIL ) revealed a very broad distribution of phenotypes . We found that , while some RILs showed phenotypes similar to that of the two parents , many showed intermediate phenotypes and some were reproducibly more extreme than either parent , indicative of transgressive segregation ( Rieseberg et al . , 2003 ) . For skn-1 ( RNAi ) , the phenotype varied widely across the RILs , with 1% to 47% of embryos containing gut ( Figure 6A , Supplementary file 2 ) . This effect was even stronger with mom-2 ( RNAi ) , for which virtually the entire possible phenotypic spectrum was observed across a selection of 31 RILs representing the span of skn-1 ( RNAi ) phenotypes . The mom-2 ( RNAi ) phenotypes ranged from RILs showing 3% of embryos with gut to those showing 92% ( Figure 6A ) . In all RILs , skn-1 ( RNAi ) and mom-2 ( RNAi ) resulted in 100% lethality . It is noteworthy that one RIL ( JR3572 , Supplementary file 2 ) showed a nearly completely penetrant gutless phenotype , an effect that is much stronger than has been previously observed for mom-2 ( - ) ( Thorpe et al . , 1997 ) . These results indicate that a combination of natural variants can nearly eliminate a requirement for MOM-2 altogether , while others make it virtually essential for endoderm development . Collectively , these analyses reveal that multiple quantitative trait loci ( QTL ) underlie SKN-1- and MOM-2-dependent endoderm specification . To identify QTLs from the recombinant population , we performed linkage mapping for both phenotypes using interval mapping ( see Materials and methods ) . For skn-1 ( RNAi ) , two major peaks were revealed on chromosomes II and IV ( above 1% FDR estimated from 1000 permutations ) . Two minor loci were found on chromosomes I and X ( suggestive linkage , above 20% FDR ) ( Figure 6B ) . For mom-2 ( RNAi ) , two major independent QTL peaks were found on chromosomes I and II ( above the 5% FDR estimated from 1000 permutations ) . Although the candidate peaks observed on chromosome IV for skn-1 ( RNAi ) did not appear to overlap with those for mom-2 ( RNAi ) , overlap was observed between the chromosomes I and II candidate regions for these two phenotypes ( Figure 6B ) . These QTLs show large individual effects on both phenotypes ( Figure 6C ) . The preceding findings unveiled wide cryptic variation in the requirements for both SKN-1 and MOM-2/Wnt in the endoderm GRN , raising the possibility that the variation affecting the two inputs might be related . Indeed , comparisons of the GWAS and QTL mapping results for skn-1 and mom-2 showed an overlap in candidate QTL regions on chromosome I , II and IV ( Figure 5 , Figure 6 , Figure 5—figure supplement 2 ) , suggesting a possible connection between the genetic basis underlying these two traits . It is conceivable that some genetic backgrounds are generally more sensitive to loss of either input ( e . g . , the threshold for activating the GRN is higher ) and others more robust to single-input loss . Alternatively , a higher requirement for one input might be associated with a relaxed requirement for the other , that is , a reciprocal relationship . As an initial assessment of these alternatives , we examined whether the requirements for SKN-1 and MOM-2 across the strains were significantly correlated . This analysis revealed no strong relationship between the cryptic variation in the requirement for these inputs seen across all the strains ( Spearman correlation R = 0 . 18 , p-value=0 . 07 ) ( Figure 7A ) . This apparent lack of correlation at the level of strains is not unexpected , as many factors likely contribute to the cryptic variation and the comparison reflects the collective effect of all causal loci in the genome of each strain ( Figures 5 and 6 ) . We next sought to examine possible relationships between the two GRN inputs at higher resolution by comparing association of specific genetic regions with the quantitative requirement for each input . We used the available sequencing data for all isotypes tested ( Andersen et al . , 2012 ) and examined the impact of each allele on the skn-1 ( RNAi ) and mom-2 ( RNAi ) phenotypes , correcting for outliers and using a pruned SNP map ( see Materials and methods ) . We found a weak positive correlation ( Pearson’s R = 0 . 21 , p=p value<2 . 2e-16 , Figure 7—figure supplement 1 ) between the allelic effects . One possible explanation for this observation might be that variants across the set of wild isolates may generally influence the threshold for activating the positive feedback loops that lock down gut development ( Raj et al . , 2010; Sommermann et al . , 2010 ) , thereby altering the sensitivity for regulatory inputs into the endoderm pathway . Alternatively , although evidence for variation in germline RNAi sensitivity among C . elegans wild isolates is lacking ( except for CB4856 , which has been omitted from our study ) ( Tijsterman et al . , 2002; Félix et al . , 2011 ) , and we have shown above that variation in SKN-1 and MOM-2 requirement reflects in large part cryptic genetic differences in the endoderm GRN , it remains possible that a fraction of the variation found in the two phenotypes tested is attributable to varying RNAi penetrance , which may underlie the minor positive correlation between skn-1 ( RNAi ) and mom-2 ( RNAi ) phenotypes . In contrast , analysis of the N2/MY16 RILs uncovered a potential reciprocal relationship between the requirements for SKN-1 and MOM-2: we observed a negative correlation between the skn-1 ( RNAi ) and mom-2 ( RNAi ) phenotypes across the genome ( Figure 7B , genome-wide Pearson’s R = −0 . 35 , p=0 . 001 , correcting for LD and outliers as above; correlation without chromosome IV R = −0 . 59 , p<0 . 001 ) ( Figure 7B ) . This finding suggested that at least some quantitative variants result in opposing effects on the requirement for SKN-1 and MOM-2 . While a reciprocal relationship was observed generally across the genome spanning five of the chromosomes , we observed the opposite correlation on chromosome IV ( Pearson’s R = 0 . 83 , p-value=1 . 695×10−5 ) . No correlation was observed for chromosome IV with the wild isolates ( Pearson’s R = 0 . 08 , NS , Figure 7—figure supplement 1 ) . As there is a major QTL on chromosome IV for the SKN-1 requirement and there is substantial overlap in the same region with the GWAS analysis of the skn-1 ( RNAi ) and mom-2 ( RNAi ) phenotypes , we sought to dissect further the relationship between the requirement for MOM-2 and SKN-1 in this region . We created six near-isogenic lines ( NILs ) in which the QTL region for the skn-1 ( RNAi ) phenotype on chromosome IV from N2 was introgressed into the MY16 background , and vice-versa ( Figure 7—figure supplement 2 ) . Control lines were created from the same crosses at the same generation by selecting the original parental region ( e . g . , selecting for the N2 region in an N2 background and MY16 in MY16 background ) . We found that the region affects the skn-1 ( RNAi ) phenotype as expected: the N2 region increased the fraction of gut in an MY16 background , and the MY16 regions decreased this fraction in an N2 background . However , for mom-2 ( RNAi ) , while introgressing the N2 region in MY16 dramatically changed the phenotype ( Figure 7C ) , we found that the MY16 region was not sufficient to alter the phenotype in an N2 background . We created segregant NILs in which one of the genetic markers was lost ( see Materials and methods ) and found that replacing the N2 region with the corresponding MY16 region in all cases results in a stronger mom-2 ( RNAi ) phenotype . However , for the skn-1 ( RNAi ) phenotype six of nine segregants showed the opposite effect: that is , a weaker phenotype ( Figure 7D ) , revealing that when contributing variants were separated by recombination , a reciprocal effect was frequently seen . These observations suggest that complex genetic interactions between variants on chromosome IV might mask the potential reciprocal effects that were observed on the other chromosomes . While the above findings revealed that the relationship between the requirement for SKN-1 and MOM-2 may be complicated by genetic interactions , our results raised the possibility of compensatory relationships between them . To further assess this possibility , we tested other candidate genes that reside in the QTL regions and that have been implicated in endoderm development ( Ruf et al . , 2013; Witze et al . , 2009; Walston et al . , 2004 ) . We found that loss of RICT-1 , the C . elegans orthologue of the human RICTOR ( Rapamycin-insensitive companion of mTOR; Tatebe and Shiozaki , 2017 ) , a component of the TORC2 complex , which has been shown to antagonize SKN-1 function ( Ruf et al . , 2013 ) , results in opposite effects on skn-1 ( - ) and mom-2 ( - ) mutants ( Figure 8A ) . Specifically , while rict-1 ( RNAi ) suppresses the absence of gut in skn-1 ( zu67 ) embryos ( skn-1 ( zu67 ) : 34 . 3% ± s . d 4 . 1% with gut vs . skn-1 ( zu67 ) ; rict-1 ( RNAi ) : 48 . 3% ± s . d 4 . 9%; p=<0 . 001 ) , we found that it enhances this phenotype in mom-2 ( or42 ) mutants ( mom-2 ( or42 ) : 23 . 8% ± s . d 2 . 0%; vs . mom-2 ( or42 ) ; rict-1 ( RNAi ) : 11 . 2% ± s . d 3 . 2%; p<0 . 001 ) . Confirming this effect , a similar outcome was observed when SKN-1 was depleted by RNAi in rict-1 ( ft7 ) chromosomal mutants ( skn-1 ( RNAi ) : 31 . 6% ± s . d 4 . 3% with gut vs . rict-1 ( ft7 ) ; skn-1 ( RNAi ) : 45 . 9% ± s . d 6 . 3%; p<0 . 05 ) ( Figure 8A ) . Similarly , RNAi depletion of PLP-1 , the C . elegans homologue of the Pur alpha transcription factor that has been shown to bind to and regulate the end-1 promoter ( Witze et al . , 2009 ) , reciprocally affects the outcome of removing these two inputs in the same direction: loss of PLP-1 function suppresses the skn-1 ( - ) phenotype ( to 48 . 0% ± s . d 6 . 6% ) , and strongly enhances the mom-2 phenotype ( to 6 . 9% ± s . d 1 . 6% ) . Again , this result was confirmed by RNAi of skn-1 in a plp-1 ( ok2156 ) chromosomal mutant ( Figure 8B ) . Thus , as observed with the effect across the genome with natural variants , we observed a substantial reciprocal effect of both of these genes on loss of SKN-1 and MOM-2 . We also observed a reciprocal effect on the SKN-1 and Wnt inputs with MIG-5/dishevelled , a component of the Wnt pathway that acts downstream of the Wnt receptor ( Walston et al . , 2004 ) ; however , in this case the effect was in the opposite direction as seen for RICT-1 and PLP-1 . Loss of MIG-5 as a result of chromosomal mutation or RNAi leads to enhancement of the skn-1 ( - ) phenotype ( mig-5 ( rh94 ) ; skn-1 ( RNAi ) : 6 . 6% ± s . d 2 . 3%; skn-1 ( zu67 ) ; mig-5 ( RNAi ) : 9 . 4% ± s . d 1 . 4% ) and suppression of the mom-2 ( - ) phenotype ( 88 . 6% ± s . d 4 . 0% ) ( Figure 8C ) . Together , these findings reveal that , as observed with many of the N2/MY16 RILs variants across most of the genome , RICT-1 , PLP-1 , and MIG-5 show opposite effects on the phenotype of removing SKN-1 and MOM-2 , suggesting a trend toward genetic influences that reciprocally influence the outcome in the absence of these two inputs . Quantitative analyses of the wild isolates and RILs revealed that multigenic factors are responsible for the difference in requirement for SKN-1 and MOM-2 between isotypes . Notably , we observed substantial overlap on chromosome IV in the GWAS analyses of the skn-1 and mom-2 requirements in wild isotypes ( Figure 5 , Figure 5—figure supplement 2 ) and on chromosome II from analyses using RILs ( Figure 6B ) . This finding raises the possibility that some QTLs may influence requirement for both inputs into the endoderm specification pathway: as SKN-1 and Wnt converge to regulate expression of the end-1/3 genes , it is conceivable that common genetic variants might modulate the relative strength or outcome of both maternal inputs . However , our findings do not resolve whether these genetic variants act independently to influence the maternal regulatory inputs . Genetic interactions are often neglected in large-scale genetic association studies ( Jenkins et al . , 2009 ) owing in part to the difficulty in confirming them ( Page et al . , 2003 ) . Many studies ( Mackay , 2014; Volis et al . , 2011; Félix , 2007; Barkoulas et al . , 2013 ) , including ours here , showed that epistasis can strongly influence the behavior of certain variants upon genetic perturbation . In addition , selection on pleiotropically acting loci facilitates rapid developmental system drift . Together , epistasis and selection on pleiotropic loci play important roles in the evolution of natural populations ( Duveau and Félix , 2012; Johnson and Porter , 2007; Phillips , 2008; Wagner and Zhang , 2011 ) . Although we did not observe a direct correlation between the skn-1 ( - ) and mom-2 ( - ) phenotypes across the isotypes studied here , we found a negative correlation across much of the genome for the N2 X MY16 RILs ( Figures 6 and 7 ) . Further , while GWAS and QTL analysis of natural and inbred lines , respectively , did not reveal a causal region in chromosome IV for mom-2 ( RNAi ) variation , analysis of NILs results did uncover at least one QTL affecting this phenotype . Moreover , while broad regions of the chromosome showed a positive correlation between the SKN-1 and MOM-2 requirements , isolation of variants in NILs revealed an inverse requirement for these inputs for at least some regions on this chromosome . These results reflect the limitations of genome-wide studies of complex genetic traits: in the case of chromosome IV , several closely linked loci appear to influence both the SKN-1 and MOM-2 requirements . Our findings raise the possibility that the SKN-1 and MOM-2/Wnt inputs might compensate for each other and that genetic variants that enhance the requirement for one of the inputs may often relax the requirement for the other . Such reciprocality could reflect cross-regulatory interactions between these two maternal inputs or could be the result of evolutionary constraints imposed by selection on these genes , which act pleiotropically in a variety of processes . Further supporting this possibility , we identified two genes , rict-1 and plp-1 , that show similar inverse effects on the requirements from skn-1 and mom-2: debilitation of either gene enhances the phenotype of mom-2 ( - ) and suppresses that of skn-1 ( - ) . RICT-1 function extends lifespan in C . elegans through the action of SKN-1 ( Ruf et al . , 2013 ) , and loss of RICT-1 rescues the misspecification of the MS and E blastomeres and lethality of skn-1 ( - ) embryos ( Ruf et al . , 2013 ) , consistent with our finding . We previously reported that PLP-1 , a homologue of the vertebrate transcription factor pur alpha , binds to the end-1 promoter and acts in parallel to the Wnt pathway and downstream of the MAPK signal ( Witze et al . , 2009 ) , thereby promoting gut formation . PLP-1 shows a similar reciprocal relationship with SKN-1 and MOM-2 as with RICT-1 ( Figure 8 ) . Given that PLP-1 binds at a cis regulatory site in end-1 near a putative POP-1 binding site ( Witze et al . , 2009 ) , and that SKN-1 also binds to the end-1 regulatory region ( Zhu et al . , 1997 ) , it is conceivable that this reciprocality might reflect integration of information at the level of transcription factor binding sites . As the architecture of the GRN is shaped by changes in cis-regulatory sequences ( Peter and Davidson , 2011; Davidson and Levine , 2008 ) , analyzing alterations in SKN-1 and Wnt/POP-1 targets among C . elegans wild isolates may provide insights into how genetic changes are accommodated without compromising the developmental output at microevolutionary time scale . MIG-5 , a dishevelled orthologue , functions in the Wnt pathway in parallel to Src signaling to regulate asymmetric cell division and endoderm induction ( Bei et al . , 2002; Walston et al . , 2004 ) . We found that the loss of mig-5 function enhances the gut defect of skn-1 ( - ) and suppresses that of the mom-2 ( - ) , the opposite reciprocal relationship to that of rict-1 and plp-1 , and consistent with a previous report ( Figure 8 ) ( Bei et al . , 2002 ) . These effects were not observed in embryos lacking function of dsh-2 , the orthologue of mig-5 ( data not shown ) , supporting a previous study that showed overlapping but non-redundant roles of MIG-5 and DSH-2 in EMS spindle orientation and gut specification ( Walston et al . , 2004 ) . Recent studies showed that Dishevelled can play both positive and negative roles during axon guidance ( Shafer et al . , 2011; Zheng et al . , 2015 ) . Dishevelled , upon Wnt-activation , promotes hyperphosphorylation and inactivation of Frizzled receptor to fine-tune Wnt activity . It is tempting to speculate that MIG-5 may perform similar function in EMS by downregulating activating signals ( Src or MAPK ) , in the absence of MOM-2 . We hypothesize that compensatory mechanisms might evolve to fine-tune the level of gut-activating regulatory inputs . Successful developmental events depend on tight spatial and temporal regulation of gene expression . For example , anterior-posterior patterning in the Drosophila embryo is determined by the local concentrations of the Bicoid , Hunchback , and Caudal transcription factors ( Rivera-Pomar and Jäckle , 1996 ) . We postulate that SKN-1 and Wnt signaling is modulated so that the downstream genes , end-1/3 , which control specification and later differentiation of endoderm progenitors , are expressed at optimal levels that ensure normal gut development . Suboptimal END activity leads to poorly differentiated gut and both hypo- and hyperplasia in the gut lineage ( Maduro et al . , 2015; Choi et al . , 2017; Maduro , 2015 ) . Hyper- or hypo-activation of Wnt signaling has been implicated in cancer development ( Zhan et al . , 2017 ) , bone diseases ( Jenkins et al . , 2009; Baron and Gori , 2018 ) , and metabolic diseases ( Chen and Wang , 2018; Schinner , 2009 ) , demonstrating the importance of regulating the timing and dynamics of such developmental signals within a quantitatively restricted window . This study revealed substantial cryptic genetic modifications that alter the relative importance of two partially redundant inputs into the C . elegans endoderm GRN , leading to rapid change in the developmental network architecture ( Figure 9 ) . Such modifications may occur through transitional states that are apparent even within this single species . For example , the finding that POP-1 is not required for gut development even in a wild isolate ( e . g . , MY16 ) that , like C . briggsae , shows a near-absolute requirement for SKN-1 may reflect a transitional state between the two species: that is , a nearly essential requirement for SKN-1 but non-essential requirement for POP-1 , an effect not previously seen in either species . In addition , duplicated GATA factors ( the MEDs , ENDs , and ELTs ) and partially redundant activating inputs ( SKN-1 , Wnt , Src , and MAPK ) in endoderm GRN , provide an opportunity for genetic variation to accumulate and ‘experimentation’ of new regulatory relationships without diminishing fitness ( Félix and Wagner , 2008; Schinner , 2009; Frankel et al . , 2010 ) . Redundancy in the regulatory inputs may act to ‘rescue’ an initial mutation and allow for secondary mutations that might eventually lead to rewiring of the network . For example , loss of either MyoD or Myf5 , two key regulators of muscle differentiation in metazoans , produces minimal defects in myogenesis as a result of compensatory relationship between the myogenic factors ( Mohun , 1992 ) . In vertebrates , gene duplication events have resulted in an expansion of Hox genes to a total of >200 , resulting in prevalent redundancy ( Imai et al . , 2001; Manley and Capecchi , 1997; Nam and Nei , 2005 ) . This proliferation of redundant genes provides opportunities for evolutionary experimentation , subsequent specialization of new functions , and developmental system drift ( Nam and Nei , 2005; True and Haag , 2001 ) . In C . elegans , loss of GAP-1 ( a Ras inhibitor ) or SLI-1 ( a negative regulator of EGFR signaling ) alone does not produce obvious defects , while double mutations lead to a multivulva phenotype ( Yoon et al . , 2000 ) . Similar redundant relationships between redundant partners exist in many other contexts in the animal . Notably , the relative importance of Ras , Notch , and Wnt signals in vulva induction differ in various genetic backgrounds ( Milloz et al . , 2008; Gleason et al . , 2002 ) and physiological conditions ( Braendle and Félix , 2008; Grimbert et al . , 2018 ) , resulting in flexibility in the system . While vulval development in C . elegans , when grown under standard laboratory conditions , predominantly favors utilization of the EGF/Ras signaling pathway ( Braendle and Félix , 2008 ) , Wnt is the predominant signaling pathway in the related Pristionchus pacificus , which is ~250 MY divergent ( Zheng et al . , 2005; Tian et al . , 2008 ) . In addition , while Cel-lin-17 functions positively to transduce the Wnt signal , Ppa-lin-17/Fz antagonizes Wnt signaling and instead the Wnt signal is transmitted by Ppa-lin-18/Ryk , which has acquired a novel SH3 domain not present in the C . elegans ortholog ( Wang and Sommer , 2011 ) . Thus , extensive rewiring of signaling networks and modularity of signaling motifs contribute to developmental systems drift ( True and Haag , 2001; Haag et al . , 2018 ) . The broad cryptic variation may drive developmental system drift , giving rise to GRN architectures that differ in the relative strength of the network components . Our finding that the key regulatory inputs that initiate the endoderm GRN show dramatic plasticity is consistent with comparative transcriptomic studies that demonstrate high gene expression variability and divergence during early embryonic stages in fly ( Kalinka et al . , 2010; Gerstein et al . , 2014; Levin et al . , 2016 ) , worm ( Gerstein et al . , 2014; Levin et al . , 2016; Levin et al . , 2012; Zalts and Yanai , 2017 ) , Xenopus ( Levin et al . , 2016; Yanai et al . , 2011; Irie and Kuratani , 2011 ) , zebrafish ( Levin et al . , 2016; Irie and Kuratani , 2011 ) , and mouse ( Levin et al . , 2016; Irie and Kuratani , 2011 ) . Therefore , early developmental events may be highly evolvable , in part due to weak purifying selection on maternal-effect genes ( Cruickshank and Wade , 2008; Cutter et al . , 2019 ) . This is in accordance with the ‘hourglass’ concept of embryonic development ( Kalinka et al . , 2010; Raff , 1996; Domazet-Lošo and Tautz , 2010 ) , in which divergent developmental mechanisms during early embryogenesis converge on a more constant state ( i . e . , a ‘phylotypic stage’ at the molecular regulatory level ) . Indeed , unlike the terminal differentiation factor ELT-2 , upstream MEDs and ENDs genes are present only in closely related Caenorhabditis species ( Gillis et al . , 2007; Maduro , 2015; Coroian et al . , 2006; Maduro et al . , 2005b ) . This is likely attributable either to positive selection during early embryonic and later larval stages or to developmental constraints . Analysis of developmental gene expression in mutation accumulation lines , which have evolved in the absence of any positive selection , showed similarity to the developmental hourglass model of evolvability , consistent with strong developmental constraints on the phylotypic stage ( Zalts and Yanai , 2017 ) . However , they do not rule out the possibility that early and late stages of development might be more adaptive and therefore subject to positive selection . It will be of interest to learn the degree to which the divergence in network architecture might arise as a result of differences in the environment and selective pressures on different C . elegans isotypes . All wild isolates , each with a unique haplotype ( Andersen et al . , 2012 ) , were obtained from the Caenorhabditis Genetics Center ( CGC ) ( see Supplementary file 1 ) . Worm strains were maintained as described ( Brenner , 1974 ) and all experiments were performed at 20°C unless noted otherwise . Feeding-based RNAi experiments were performed as described ( Kamath and Ahringer , 2003 ) . RNAi clones were obtained from either the Vidal ( Rual et al . , 2004 ) or Ahringer libraries ( Kamath et al . , 2003 ) . RNAi bacterial strains were grown at 37°C in LB containing 50 μg/ml ampicillin . The overnight culture was then diluted 1:10 . After 4 hr of incubation at 37°C , 1 mM of IPTG was added and 60 μl was seeded onto 35 mm agar plates containing 1 mM IPTG and 25 μg/ml carbenicillin . Seeded plates were allowed to dry and used within five days . Five to 10 L4 animals were placed on RNAi plate . 24 hr later , they were transferred to another RNAi plate and allowed to lay eggs for four or 12 hr ( 12 hr for skn-1 RNAi and four hours for the other RNAi ) . The adults were then removed , leaving the embryos to develop for an extra 7–9 hr . Embryos were quantified and imaged on an agar pad using a Nikon Ti-E inverted microscope . We chose to perform RNAi on agar plates to maximize sensitivity , robustness , and reproducibility of the assay , as liquid culture RNAi can introduce variability owing to aggregation and settling of bacteria , which affects RNAi efficacy ( Gomez-Amaro et al . , 2015 ) . In addition , performing RNAi on agar plates allowed us to collect large numbers of embryos with which to quantify gut formation ( as described below ) . The embryonic gut cells and nuclei of all cells were stained with MH33 ( mouse anti-IFB-2 , deposited to the DSHB by Waterston , R . H . ) and AHP418 ( rabbit anti-acetylated histone H4 , Serotec Bio-Rad ) respectively . Fixation and permeabilization were carried out as described previously ( Sommermann et al . , 2010 ) . Goat anti-mouse Alexa Fluor 594 and goat anti-rabbit Alexa Fluor 488 secondary antibodies were used at 1:1000 dilution . Gut was scored by the presence of birefringent gut granule in arrested embryos ( Clokey and Jacobson , 1986; Hermann et al . , 2005 ) . For skn-1 ( RNAi ) , the laboratory strain N2 , which shows invariable ~30% of embryos with endoderm , was used as a control for all experiments . To introgress skn-1 ( zu67 ) into wild isolates ( WI ) , males from the wild isolate strains were crossed to JJ186 dpy-13 ( e184 ) skn-1 ( zu67 ) IV; mDp1 ( IV;f ) hermaphrodites . mDp1 is a free duplication maintained extrachromosomally that rescues the Dpy and lethal phenotypes of dpy-13 ( e184 ) and skn-1 ( zu67 ) respectively . mDp1 segregates in a non-Mendelian fashion and animals that have lost the free duplication are Dpy and produce dead offspring . Wild type F1 hermaphrodites that have lost the free duplication , as determined by the presence of 1/4 Dpy progeny in the F2 generation , were selected . 10 single non-Dpy F2 hermaphrodite descendants from F1 animals heterozygous for skn-1 ( zu67 ) ( 2/3 of which are expected to be of the genotype WI dpy-13 ( + ) skn-1 ( + ) /dpy-13 ( e184 ) skn-1 ( zu67 ) were backcrossed to their respective parental wild strain . 10 F3 hermaphrodites were picked to individual plates . Half of the F3 cross progeny are expected to be heterozygous for dpy-13 ( e184 ) skn-1 ( zu67 ) , as evidenced by presence of F4 Dpy progeny that produced dead embryos . Non-Dpy siblings were used to continue the introgression as described . This strategy was repeated for at least five rounds of introgression . The embryonic gutless phenotype in the progeny of the Dpy animals was quantified . Similarly , to introgress pop-1 ( zu189 ) or mom-2 ( or42 ) alleles into wild isolates , JJ1057 pop-1 ( zu189 ) dpy-5 ( e61 ) /hT1 I; him-5 ( e1490 ) /hT1V or EU384 dpy-11 ( e1180 ) mom-2 ( or42 ) V/nT1 [let- ? ( m435 ) ] ( IV;V ) were used , respectively . The mutant strain was crossed to the wild isolates . Non-Dpy F2 animals heterozygous for the chromosomal mutation were selected and backcrossed to their respective parental wild strain for at least four rounds of introgression for pop-1 and seven rounds for mom-2 . The embryonic gutless phenotype in the progeny of the Dpy animals was quantified , as above . All data were analyzed and plotted using R software v 3 . 2 . 3 ( https://www . r-project . org/ ) . GWAS for both phenotypes was performed using C . elegans wild isolates and a previously published SNP map containing 4 , 690 SNPs ( Andersen et al . , 2012 ) with the EMMA R package . P-values were calculated using mixed model analysis ( Kang et al . , 2008 ) ( emma . REML . t ( ) function ) and identity-by-state ( IBS ) kinship matrix to account for population structure . For skn-1 and mom-2 RNAi phenotypic data , a genome-wide permutation-based FDR was also calculated for the EMMA results from 10 , 000 permuted values ( Millstein and Volfson , 2013; Hansen and Kerr , 2012 ) . Phylogenetic trees were constructed from 4690 polymorphisms using R package ‘ape’ ( Paradis et al . , 2004 ) . Neighbor-joining algorithm based on pairwise distances was used . Phylogenetic signal ( Pagel’s λ statistics ) was measured using ‘phylosig ( ) ” function in phytools R package ( Revell , 2012; Ives et al . , 2007 ) . Statistical significance of λ was obtained by comparing the likelihood a model accounting for the observed λ with the likelihood of a model that assumes complete phylogenetic independence . Geographic information for strains were obtained from Andersen et al . ( 2012 ) , available in Supplementary file 1 , together with the corresponding skn-1 ( RNAi ) and mom-2 ( RNAi ) phenotypes . To test for the relationship between mom-2 ( RNAi ) and skn-1 ( RNAi ) phenotypic data , the differences between median phenotypic values for each SNP were calculated independently on a genome-wide level for the wild isolates . In order to correct for LD , SNPs were pruned with PLINK ( http://pngu . mgh . harvard . edu/purcell/plink/ ) ( Purcell et al . , 2007 ) and only a subset of SNPs was used for the correlation analyses . Outliers were removed from the calculations by using z-score with a cutoff of 1 . 96 ( i . e . , 95% of values fall within ±1 . 96 in a normal distribution ) . Recombinant inbred lines ( RILs ) were created by crossing an N2 hermaphrodite and an MY16 male . 120 F2 progeny were cloned to individual plates and allowed to self-fertilize for 10 generations . A single worm was isolated from each generation to create inbred lines . A total of 95 lines were successfully created and frozen stocks were immediately created and kept at −80°C ( Supplementary file 2 ) , prior to DNA sequencing . DNA was extracted using Blood and Tissue QIAGEN kit from worms from each of the RILs grown on four large NGM plates ( 90 × 15 mm ) with OP50 E . coli until starved ( no more than a day ) . Samples were submitted in 96-well plate format at 10 ng/µl < n < 30 ng/µl . GBS libraries were constructed using digest products from ApeKI ( GWCGC ) , using a protocol modified from Elshire et al . ( 2011 ) . After digestion , the barcoded adapters were ligated and fragments < 100 bp were sequenced as single-end reads using an Illumina HiSeq 2000 lane ( 100 bp , single-end reads ) . SNP calling was performed using the GBSversion3 pipeline in Trait Analysis by aSSociation , Evolution and Linkage ( TASSEL ) ( Bradbury et al . , 2007 ) . Briefly , fastq files were aligned to reference genome WS252 using BWA v . 0 . 7 . 8-r455 and SNPs were filtered using vcftools ( Danecek et al . , 2011 ) . Samples with greater than 90% missing data and SNPs with minor allele frequencies ( mAF ) of <1% were excluded from analysis , identifying 27 , 396 variants . Variants identified by GBS pipeline were filtered to match the SNPs present in the parental MY16 strain ( using vcftools –recode command ) , and variants were converted to a 012 file ( vcftools –012 command ) . Single-QTL analysis was performed in R/QTL ( Broman and Sen , 2009 ) using 1770 variants and 95 RILs . Significant QTL were determined using Standard Interval Mapping ( scanone ( ) ‘em’ ) and genome-wide significance thresholds were calculated by permuting the phenotype ( N = 1 , 000 ) . Change in log-likelihood ratio score of 1 . 5 was used to calculate 95% confidence intervals and define QTL regions ( Broman et al . , 2003 ) . SNP data for the RILs and their corresponding phenotypes used in analysis are shown in Supplementary files 2 and 3 . Three N2-derived mutant strains were used to introgress regions from chromosome IV from N2 into the MY16 strain background and vice-versa . For both types of crosses , N2 was always used as the maternal line . The following strains were used: To introgress the N2 region into the MY16 genetic background , hermaphrodites from the N2-derived strains containing genetic markers flanking the genomic region of interest were crossed with MY16 males ( Figure 7—figure supplement 2 ) . After successful mating , 10 F1 heterozygotes were isolated and allowed to self . After 24 hr , the F1 adults were removed from the plate , and the F2 hermaphrodites left to develop to young adults . F2 animals homozygous for the region being introgressed were selected as young adults and crossed with MY16 males . This process was repeated until ten rounds of introgression were completed . These new lines were preserved at −80C . Introgression of MY16 region into an N2 background began with the same initial cross as above . F1 heterozygous males were crossed with N2 hermaphrodites containing phenotypic markers near the region being introgressed ( Figure 7—figure supplement 2 ) . After successful mating , the F1 parents were removed and the F2 generation was left to develop until heterozygous males were visible . F2 heterozygous males were crossed with hermaphrodites from the N2-derived strain . This process was repeated until ten successful introgressions were completed . To homozygose the introgressed MY16 regions , worms were singled and allowed to self until a stable wildtype population was obtained . These new lines preserved at −80C . NILs were genotyped to test for correct introgression of the desired regions by Sanger sequencing of 10 markers spaced along chromosome IV ( carried out by the Centre for Genomics and Proteomics , University of Auckland ) . Upon confirmation of their genetic identity , one NIL was used to further dissect the QTL region by segregating the visual markers ( Dpy and Unc ) .

Two people with the same disease , or who inherit the same genetic mutation , often show different symptoms or respond to medical treatments in different ways . This is because many traits are not the result of a single gene , but of several genes interacting with each other in complex ways to form networks that lead to many possible outcomes . Gene regulatory networks , which control how animals develop , change over evolutionary time to create the vast variety of different species that exist today . However , it is still unclear how mutations in these networks can occur without negatively impacting their activity , or how networks become rewired during evolution . To address these questions , Torres Cleuren et al . studied the gene regulatory network that controls the development of the gut across approximately 100 different strains of Caenorhabditis elegans , a widely studied nematode worm . This involved testing how switching off particular genes affected gut development in embryos of the worm . The experiments revealed that the first steps in the gene regulatory networks that control gut development vary drastically between the different wild strains of C . elegans . For example , in some of the strains , two genes known as skn-1 and mom-2 are essential for gut formation , whereas in others the gut often forms even when these genes are switched off . These results support the idea that some of the genes in the network can compensate for loss of others , explaining how mutations can accumulate without impacting the development of the embryo . The findings of Torres Cleuren et al . provide important insights into how gene regulatory networks can be rewired , with some components accumulating mutations and acquiring new roles , while others stay the same .

2019

Extensive intraspecies cryptic variation in an ancient embryonic gene regulatory network

In the retina , synaptic transmission between photoreceptors and downstream ON-bipolar neurons ( ON-BCs ) is mediated by a GPCR pathway , which plays an essential role in vision . However , the mechanisms that control signal transmission at this synapse and its relevance to behavior remain poorly understood . In this study we used a genetic system to titrate the rate of GPCR signaling in ON-BC dendrites by varying the concentration of key RGS proteins and measuring the impact on transmission of signal between photoreceptors and ON-BC neurons using electroretinography and single cell recordings . We found that sensitivity , onset timing , and the maximal amplitude of light-evoked responses in rod- and cone-driven ON-BCs are determined by different RGS concentrations . We further show that changes in RGS concentration differentially impact visually guided-behavior mediated by rod and cone ON pathways . These findings illustrate that neuronal circuit properties can be modulated by adjusting parameters of GPCR-based neurotransmission at individual synapses . Signaling via G protein coupled receptors ( GPCR ) mediates neurotransmitter actions and shapes many essential neuronal processes including vision , motor control , and cognition ( Wettschureck and Offermanns , 2005 ) . In addition to tremendous diversity in neurotransmitter receptors and the neuronal responses they modulate , GPCR pathways can be tuned in terms of response magnitude , sensitivity , and kinetics , oftentimes creating characteristic signature profiles specifically suited to meet the signaling demands of a neuronal circuit ( Marder , 2012; Nusbaum and Blitz , 2012 ) . However , the molecular mechanisms underlying the formation of specific signaling patterns , their relationship to synaptic processing , and ultimately to behavior remain poorly understood . Signaling in GPCR pathways is initiated as the active receptor stimulates GDP/GTP exchange on G protein α subunits , triggering their dissociation from Gβγ subunits . These dissociated G proteins form active species that regulate the activity of second messenger enzymes , and either directly or indirectly modulate the opening of a number of ion channels ( Cabrera-Vera et al . , 2003; McCudden et al . , 2005 ) . It is now well recognized that the predominant role in controlling GPCR signaling belongs to the Regulator of G protein Signaling ( RGS ) proteins . RGS proteins stimulate GTP hydrolysis thus promoting subunit re-association and resulting in the termination of G protein signaling on a physiological timescale ( Ross and Wilkie , 2000; Hollinger and Hepler , 2002 ) . Studies with knockout mice indicate that the elimination of RGS impacts key parameters of GPCR-driven signaling including sensitivity , temporal characteristics , and response amplitudes ( Chen et al . , 2000; Heximer et al . , 2003; Rahman et al . , 2003; Fu et al . , 2006; Posokhova et al . , 2010 ) . These changes are often paralleled by distinct behavioral changes , for example , motor deficits , alterations in drug sensitivity , memory and reward behavior ( Traynor and Neubig , 2005; McCoy and Hepler , 2009; Xie and Martemyanov , 2011 ) . Interestingly , the effect of RGS protein loss appears to be cell-type specific . For example , the elimination of the dominant RGS in rod photoreceptors profoundly slows deactivation kinetics while minimally affecting response amplitude or sensitivity ( Chen et al . , 2000; Krispel et al . , 2006 ) , whereas in hippocampal pyramidal neurons , moderate changes in response kinetics are accompanied by increased sensitivity ( Xie et al . , 2010 ) . Altogether , these findings suggest that changes in relative RGS activity may filter the GPCR-generated responses , tuning response parameters to control signaling and ultimately behavior . Although intuitive , this hypothesis remains untested , as the constitutive elimination of RGS proteins typically results in an all-or-nothing impact on GPCR pathways , and compensatory changes during development further confound the interpretation of behavioral changes . GPCR pathways play an especially critical role in enabling vision at low light levels . Single-photon responses generated by rod photoreceptors are transmitted to the downstream rod ON-bipolar cells ( ON-BC ) . The high gain and rapid response of visual signals puts exquisite demands on both the timing and sensitivity of signal transmission at this synapse to enable rod reliable vision ( Okawa and Sampath , 2007; Pahlberg and Sampath , 2011 ) . The transient suppression of the glutamate release from photoexcited rods is sensed by the postsynaptic GPCR , mGluR6 ( Snellman et al . , 2008; Morgans et al . , 2010 ) . In darkness , mGluR6 activates a G protein , Gαo , which maintains the TRPM1 effector channels closed . The opening of TRPM1 channels and resulting generation of the depolarizing response requires deactivation of Gαo ( Dhingra et al . , 2000; Sampath and Rieke , 2004 ) . Recent studies have indicated two RGS proteins , RGS7 and RGS11 play central role in this process ( Morgans et al . , 2007; Mojumder et al . , 2009; Cao et al . , 2012; Shim et al . , 2012 ) . In humans , loss of function mutations in mGluR6 or TRPM1 prevent the depolarizing activity of ON-BC and result in congenital stationary night blindness ( Dryja et al . , 2005; Audo et al . , 2009; van Genderen et al . , 2009 ) . Similarly , knockout of mGluR6 ( Masu et al . , 1995 ) , TRPM1 ( Morgans et al . , 2009; Shen et al . , 2009; Koike et al . , 2010 ) , Gαo ( Dhingra et al . , 2000 ) , or RGS7/RGS11 ( Cao et al . , 2012; Shim et al . , 2012 ) in mice also disrupts synaptic transmission between rods and ON-BC , completely abrogating responses of ON-BCs to light flashes . The total loss of signal transmission in these models has largely prevented dissection of the mechanisms that connect the fine-tuning of the GPCR signaling cascade to relevant physiological functions . Furthermore , cone photoreceptors are active under bright illumination and also signal through ON-BC neurons forming a distinct circuit . However , contributions of mGluR6 cascade elements to signal transmission at cone synapses , and their implications for vision , remain poorly understood . In this study we used an inducible system to examine the effect of progressive RGS loss in mature and differentiated retinas on the ability of ON-BC neurons to process signals generated by rod and cone photoreceptors and its impact on visually-guided behavior . We describe how graded reductions in RGS concentration affect ON-BC light-evoked responses and relate these observations to the performance in visually-guided behavioral tasks . Our findings illustrate how changes in RGS activity may tune GPCR response properties to adapt circuit function to specific behavioral demands . While elimination of either RGS7 or RGS11 alone does not substantially alter depolarizing activity of ON-BCs ( Mojumder et al . , 2009; Chen et al . , 2010; Zhang et al . , 2010 ) , simultaneous loss of both proteins completely abolishes the responses of these cells to light flashes ( Cao et al . , 2012; Shim et al . , 2012 ) . In an effort to achieve an intermediate state of RGS expression we studied mice where elimination of one RGS is combined with the haploinsufficiency of the other ( e . g . , Rgs11−/−: Rgs7+/− or Rgs11+/−: Rgs7−/− ) . Evaluation of these mouse models by both electroretinography ( ERG ) and single cell recordings from rod ON-BCs revealed largely normal responses , that showed only minor delays in the onset time of the ERG b-wave ( Figure 1; Table 1 ) . Thus , normal depolarizing activity of ON-BCs is supported by a relatively low concentration of the RGS proteins . 10 . 7554/eLife . 06358 . 003Figure 1 . Effects of RGS haploinsufficiency on ON-BC properties . ( A ) Representative electroretinography ( ERG ) traces of RGS11 knockout ( Rgs11−/− ) , RGS7 knockout ( Rgs7−/− ) alone or in combination with haploinsufficiency of the other ( Rgs11−/− , Rgs7+/− and Rgs11+/− , Rgs7−/− ) . Dashed line shows position of the response peak recorded at the dimmest flash . ( B ) Quantification of ERG b-wave onset timing across increasing light intensities . Some genotypes showed minor delay in the onset of b-wave at moderate to high light intensities ( >1 cd*s/m2 ) . ( C ) Dependence of ERG b-wave amplitude on flash intensity ( >1 cd*s/m2 flashes [t-test , p > 0 . 5 , n = 4–5] ) . Maximal amplitudes or sensitivities of b-waves were not affected in any of the genotypes . ( D ) Analysis of rod ON-BC light responses by single cell recordings . Responses of rod ON-BCs to flashes of light that varied in strength by factors of 2 , from 0 . 5–23 R*/rod . DOI: http://dx . doi . org/10 . 7554/eLife . 06358 . 00310 . 7554/eLife . 06358 . 004Table 1 . ERG b-wave parameters extracted from fitting of scotopic and photopic phases of maximal b-wave amplitudes elicited by varying flash strengthsDOI: http://dx . doi . org/10 . 7554/eLife . 06358 . 004GenotypeI0 . 5 , scotopic ( cd*s/m2 ) Rmax , ( scotopic ) ( µV ) I0 . 5 , photopic ( cd*s/m2 ) I0 . 5 , photopic ( µV ) WT0 . 0009 ± 0 . 0001587 ± 173 . 5 ± 1 . 2457 ± 30RGS11−/−0 . 001 ± 0 . 0001580 ± 444 . 4 ± 1 . 4419 ± 32RGS7−/−0 . 0008 ± 0 . 0001615 ± 272 . 4 ± 1 . 1510 ± 22RGS11−/− 7+/−0 . 0008 ± 0 . 0001628 ± 512 . 5 ± 1 . 4506 ± 25RGS11+/− 7−/−0 . 0007 ± 0 . 0001593 ± 323 . 1 ± 1 . 0439 ± 16 To achieve a greater reduction in RGS concentration we developed a model for inducible RGS7 elimination . A conditional Rgs7 allele ( Rgs7flx/flx ) was introduced in the Rgs11−/− background ( cDKO ) and the mice were further crossed with a ubiquitous driver line expressing inducible Cre-recombinase ( cDKO:Cre+; Figure 2A ) . In the resulting model , RGS elimination can be induced postnatally by the oral administration of tamoxifen . We first characterized changes in RGS7 expression following tamoxifen administration in cDKO:Cre+ mice . Immunostaining of retinal cross-sections revealed a progressive loss of the RGS7-positive signal from the ON-BC synapses in the outer plexiform layer ( Figure 2B ) . A reduction was evident in both the number of RGS7 puncta and staining intensity reaching minimal levels by day 35 . During this timeframe , no changes in the postsynaptic concentration of mGluR6 were detected ( Figure 2B ) . We further quantified the decline in RGS7 protein by quantitative Western blotting . Similar to the immunohistochemical analysis , we detected a progressive decrease in the RGS7-positive band ( Figure 2C ) . Quantitative analysis found immunostaining and Western blotting data to be in good agreement , both showing monophasic decays in the RGS7 levels ( Figure 2D ) . We further analyzed RGS7 concentration upon tamoxifen induction separately for synapses between rods and cones , and their respective ON-BCs ( Figure 2B , D; Figure 2—figure supplement 1 ) . Before tamoxifen administration ( 0 day ) , RGS7 levels in rod and cone ON-BC synapses were comparable , showing no more than a 5 ± 2% difference in staining intensity between the two . The time course of RGS7 loss induced by tamoxifen was also not different between rod and cone ON-BC synapses: half of RGS7 in rod ON-BC was found at 7 . 4 ± 1 . 5 days , vs 7 . 8 ± 1 . 3 days in cone ON-BCs . At day 35 following tamoxifen administration , RGS7 expression was reduced to 18 ± 3% in rod ON-BCs and to 17 ± 3% in cone ON-BC synapses . Importantly , the cytoarchitecture at the ON-BC dendritic tips of cDKO mice remained unaltered despite the RGS7 loss . Postsynaptic accumulations of both TRPM1 and mGluR6 and the alignment of pre- and post- synaptic compartments were both preserved ( Figure 2E; Figure 2—figure supplement 2 ) . Prior to tamoxifen administration both cDKO:Cre- and cDKO:Cre+ mice were indistinguishable from the previously characterized Rgs11−/− background strain in both RGS7 expression and localization ( Figure 3A ) . Furthermore , no changes in RGS7 concentration or its postsynaptic accumulation were detected upon tamoxifen administration in cDKO littermates lacking Cre recombinase expression ( Figure 3A ) . Consistent with these observations , ERG responses recorded in cDKO:Cre- and cDKO:Cre+ mice were indistinguishable from their Rgs11−/− controls and tamoxifen administration failed to alter ERG responses in cDKO:Cre- animals ( Figure 3B ) . Thus , the observed effects are associated specifically with genomic editing of the Rgs7 locus . 10 . 7554/eLife . 06358 . 005Figure 2 . Progressive RGS7 loss in the ON-BC neurons of a conditional mouse knockout model ( cDKO ) . ( A ) Schematic representation of a breeding strategy for the conditional inactivation of RGS7 on Rgs11−/− background . To induce recombination and RGS7 loss mice were administered tamoxifen for 5 days followed by testing . ( B ) Immunohistochemical analysis of RGS7 expression and localization in the outer plexiform layer of mouse retinas following tamoxifen administration . Retina sections were stained with RGS7 ( green ) and mGluR6 ( red ) antibodies from cDKO:Cre+ mice at 12 , 17 , 22 , 28 , and 35 days after the start of tamoxifen administration . Retinas lacking RGS7 ( Rgs7−/− ) were used as a control . Two independent experiments yielding similar data were conducted . Note progressive loss of RGS7 immunoreactivity from postsynaptic puncta denoted by mGluR6 . ( C ) Analysis of changes in RGS7 expression in total retina lysates by Western blotting . cDKO:Cre+ mice were treated with tamoxifen and retinas were collected at indicated time points . To generate retina lysates containing various RGS7 protein content , varying amounts of RGS11 knockout lysates were spiked into lysates isolated from RGS7 and RGS11 double knockout retinas ( DKO ) . To determine the level of RGS7 reduction , a calibration curve was plotted from densities of varying amounts of RGS7 protein and used to determine the protein content in retina extracts obtained from tamoxifen-treated mice . Retinas from three separate animals were used in these experiments . ( D ) Quantification of RGS7 protein content at different time points following tamoxifen treatment . Values representing density of RGS7 band ( red ) and fluorescence intensity in mGluR6-positive puncta determined separately for rods ( green ) and cones ( blue ) were normalized to respective values observed before tamoxifen treatment ( day 0 ) . Averaged RGS7 percentages from western and IHC experiments are fitted with a single exponential decay function . Error bars are SEM values . ( E ) Intact synaptic morphology and retina cytoarchitecture at 28 days post-tamoxifen treatment . Retina cross-section from Rgs11−/− and cDKO:Cre+ mice were co-stained with postsynaptic marker mGluR6 and pre-synaptic marker CtBP2 and their direct apposition was noted . Immunostaining with TRPM1 ( green ) reveals intact morphology of ON-BC , their dendritic branching and distribution of postsynaptic puncta . Figure 2—figure supplement 1: assignment of rod vs cone synapses by immunohistochemistry . Figure 2—figure supplement 2: intact synaptic morphology and retina cytoarchitecture at 28 days post-tamoxifen treatment . DOI: http://dx . doi . org/10 . 7554/eLife . 06358 . 00510 . 7554/eLife . 06358 . 006Figure 2—figure supplement 1 . Assignment of rod vs cone synapses by immunohistochemistry . Cone synapses ( indicated by arrows ) were distinguished by their position in the lower portion of the OPL where they form rows of clusters in apposition to cone pedicles ( counter-stained for β-arrestin ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06358 . 00610 . 7554/eLife . 06358 . 007Figure 2—figure supplement 2 . Intact synaptic morphology and retina cytoarchitecture at 28 days post-tamoxifen treatment . Retina cross-sections from Rgs11−/− and cDKO:Cre+ mice were co-stained with postsynaptic marker mGluR6 and pre-synaptic marker CtBP2 and their direct apposition was quantified ( right ) . Immunostaining with TRPM1 reveals intact morphology of ON-BC , their dendritic branching and distribution of postsynaptic puncta ( left ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06358 . 00710 . 7554/eLife . 06358 . 008Figure 3 . Expression of Cre recombinase without tamoxifen induction and tamoxifen administration without Cre expression do not alter RGS7 expression , localization and light response of mice . ( A ) Immunohistochemical analysis of RGS7 ( green ) and PKCα ( red ) expression and localization in the outer plexiform layer of Rgs11−/− , cDKO:Cre- , and cDKO:Cre+ mouse retinas . ( B ) Representative ERG traces elicited from scotopic ( left ) and photopic ( right ) flashes in mice described above show unaltered retina electrophysiological responses . DOI: http://dx . doi . org/10 . 7554/eLife . 06358 . 008 To obtain a more quantitative picture , we determined the absolute expression levels of RGS7 and RGS11 referencing them to the levels of mGluR6 . Using recombinant protein standards and quantitative Western blotting , we found that RGS7 , RGS11 , and mGluR6 are present at 20 ± 2 , 23 ± 1 , 31 ± 1 fmol per 10 mg of total protein , respectively ( Figure 4 ) . Based on immunohistochemical analysis , RGS11 ( Cao et al . , 2009 ) , RGS7 ( Cao et al . , 2012; Figure 4—figure supplement 1 ) and mGluR6 ( Masu et al . , 1995 ) are found exclusively , in the outer plexiform layer of the retina , indicating that the values that we obtained reflect protein content specifically at the ON-BC synapses . Considering that mGluR6 receptors form obligate dimers that function as a single unit , there is about threefold molar excess of RGS proteins ( ∼43 fmol ) relative to mGluR6 dimer ( ∼15 fmol ) in ON-BCs . In summary , these data illustrate that RGS7 loss beyond its stoichiometry with mGluR6 can be induced postnatally in a graded manner without affecting synaptic morphology or retinal cytoarchitechture . 10 . 7554/eLife . 06358 . 009Figure 4 . Quantification of RGS7 , RGS11 and mGluR6 proteins in the retina . ( A ) Representative Western blots of retina lysates analyzed alongside with recombinant standards . ( B ) Quantification of protein content . A calibration curve was plotted from densities of recombinant protein standards ( closed circles ) and used to determine the protein content ( open circles ) in retina extracts obtained from four separate mice . ( C ) Comparison of absolute RGS7 , RGS11 and mGluR6 protein levels in the retinas . The experiment was performed 3 times . Error bars indicate the SEM . Figure 2—figure supplement 1: distribution of RGS7 immunofluorescence in the retina . DOI: http://dx . doi . org/10 . 7554/eLife . 06358 . 00910 . 7554/eLife . 06358 . 010Figure 4—figure supplement 1 . Distribution of RGS7 immunofluorescence in the retina . Retina cross-sections were stained with anti-RGS7 antibodies as described in the ‘Materials and methods’ . Staining reveals the presence of specific staining only in the outer plexiform layer . DOI: http://dx . doi . org/10 . 7554/eLife . 06358 . 010 To determine how progressive loss of RGS affects signal transmission in the rod pathway , we examined retinal responses to light in dark-adapted live anesthetized mice by ERG . In cDKO:Cre+ mice prior to tamoxifen administration , the delivery of very dim ( scotopic ) flashes , that activate rods but not cones , elicited a robust b-wave , whose generation requires the activity of rod ON-BCs ( Figure 5A ) . These responses were indistinguishable from those observed in Rgs11−/− strain ( Figure 3B ) . Administration of tamoxifen to cDKO:Cre+ mice dramatically changed these responses , progressively reducing their amplitude and slowing their time course ( Figure 5A; Table 2 ) . We next analyzed changes in response properties across a range of flash strengths . As flash strengths increase , the contribution of cones to the overall response becomes appreciable . However , rod- and cone- driven b-waves saturate at distinct light intensities leading to a characteristic biphasic shape of the intensity-response relationship ( Figure 5B ) . Accordingly , the first phase of the curve is commonly considered to consist of relatively pure rod-driven component that can be extracted mathematically ( Figure 5C; Equation 1 ) . Analysis of this scotopic ERG b-wave showed marked time-dependent changes in both maximal response amplitude and sensitivity ( Table 2 ) . At 35 days following tamoxifen administration , the flash responses were barely distinguishable from the baseline seen upon constitutive elimination of both RGS7 and RGS11 , with an approximately10-fold reduction in b-wave maximal amplitude and nearly a 100-fold decrease in its sensitivity . The responses further showed a time-dependent slowing in their onset , as evidenced by a progressive increase in b-waves time-to-peak and visible deceleration of the b-wave slopes ( Figure 5D ) . Importantly , we observed no tamoxifen-induced changes in ERG b-wave characteristics in littermates lacking Cre recombinase ( Figure 3B ) , indicating that these changes are caused specifically by the progressive loss in RGS7 expression . 10 . 7554/eLife . 06358 . 011Figure 5 . The effect of progressive RGS loss on rod ON-BC responses as revealed by scotopic ERG . ( A ) Representative ERG responses elicited with dim flashes ( 0 . 001 cd*s/m2 ) from cDKO:Cre+ mice at 0 , 12 , 17 , 22 , 28 , and 35 days post-tamoxifen administration . ( B ) Maximal ERG b-wave amplitudes plotted against their eliciting flash intensities . The ERG b-waves exhibit biphasic saturation pattern , with first one around 0 . 05 cd*s/m2 , corresponding to activity of rod ON-BC , and a second one around 100 cd*s/m2 , corresponding to activity of cone ON-BC . ( C ) Rod component of the ERG response extracted from fitting the first phase of the response in panel B . ( D ) Analysis of the changes in the onset of ERG b-wave elicited by half-saturating flashes . Time to peak values are plotted as a function of days after tamoxifen treatment . Errors bars are SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 06358 . 01110 . 7554/eLife . 06358 . 012Table 2 . Effect of tamoxifen treatment on scotopic ERG b-wave parametersDOI: http://dx . doi . org/10 . 7554/eLife . 06358 . 012Days post tamoxifenI0 . 5 , scotopic ( cd*s/m2 ) Rmax , scotopic ( µV ) Time to peak ( ms ) at I0 . 500 . 0007 ± 0 . 0001560 ± 10100 ± 6120 . 001 ± 0 . 0002390 ± 10130 ± 5170 . 005 ± 0 . 0001190 ± 8180 ± 10220 . 008 ± 0 . 0003160 ± 5260 ± 21280 . 03 ± 0 . 00170 ± 3280 ± 25350 . 06 ± 0 . 00650 ± 4260 ± 22DKON . A0N . AValues are mean + SEM , n = 5 mice for all timepoints . To determine how the en masse activity of ON-BCs seen in ERGs is set by the responses of single cells , we recorded the light-evoked activity of rod ON-BCs in dark-adapted retinal slices ( Figure 6 ) . Whole-cell patch clamp recordings revealed some variability in ON-BCs responses suggesting that efficiency of tamoxifen-induced recombination varied on a cell-by-cell basis ( Table 3 ) . For example , we noticed that some cells were very susceptible to tamoxifen treatment and completely lost responsiveness to light , perhaps due to stochastic hyperactivation of Cre-recombinase leading to rapid elimination of RGS proteins . Because the proportion of unresponsive cells remained relatively stable ( ∼31–46% ) across different days following tamoxifen administration , and because elimination of RGS proteins result in complete abrogation of ON-BC responses to flashes ( Cao et al . , 2012 ) , these cells were excluded from the analysis . Analysis of the remaining light-sensitive cells revealed a progressive increase in time-to-peak and concomitant decrease in maximal amplitudes following tamoxifen administration ( Figure 6A; Table 3 ) . We next compared the average half-maximal flash strengths across all rod ON-BCs recorded on days 17 , 22 , 28 , and 35 and found their progressive increase , reflecting a reduction in sensitivity . No changes in noise variance were detected , suggesting that reduction in RGS concentration also effectively decreases the signal-to-noise ratio ( Figure 6B; Table 3 ) . Interestingly , the sensitivity reduction determined from patch clamp recordings matched well that obtained by ERG , which summates the responses of all cells across the retina ( Figure 6C ) . In summary , we observed that gradual titration of RGS levels affects kinetics , sensitivity , and amplitude of rod ON-BC light responses . 10 . 7554/eLife . 06358 . 013Figure 6 . Single cell recordings from rod ON-BC recapitulate key changes in average light-evoked responses measured by ERG . ( A ) Responses of rod ON-BCs to flashes of increasing strength . Shown are representative response families for 10 ms flashes ( arrow ) measured at an increasing number of days following tamoxifen administration . Flash strengths yielded 0 . 70 , 1 . 4 , 2 . 9 , 5 . 7 , 11 , and 23 R*/rod for control ( Rgs7+/+:Rgs11−/− ) , 0 . 58 , 1 . 0 , 2 . 2 , 4 . 3 , 8 . 6 and 17 R*/rod for day 17; 1 . 0 , 2 . 2 , 4 . 3 , 8 . 7 , 17 , and 35 R*/rod for day 22 , and 16 , 32 , 65 , 130 , 260 , 520 , and 1000 R*/rod for day 35 . Note with time following tamoxifen administration the progressive slowing and reduction in maximum amplitude of the light-evoked response ( see Table 3 ) . ( B ) Average normalized response-intensity relationships for all cells recorded for control ( day 0 ) as well as days 17 , 22 , 28 , and 35 following tamoxifen administration . Data were sampled at 1 kHz and filtered at 50 Hz . ( C ) Comparison of half-saturation responses from ON-BC measured by single-cell recordings ( top ) ( n = 4–11 ) and ERG ( bottom ) ( n = 5 ) from cDKO:Cre+ mice at different days after tamoxifen administration . Data points plotted are half-maximal values from each tamoxifen time point fitted by sigmoidal dose-response ( variable slope ) equation with the least squares method . The time required for tamoxifen to reduce the sensitivity of ON-BC by half is not different when obtained by ERG ( 28 ± 1 days ) vs single cell patch-clamp ( 27 ± 0 . 3 days ) . Errors bars are SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 06358 . 01310 . 7554/eLife . 06358 . 014Table 3 . Parameters of rod ON-BC light responses by single cell recordingsDOI: http://dx . doi . org/10 . 7554/eLife . 06358 . 014Days post tamoxifenI0 . 5 , ( R*/rod ) *Rmax ( pA ) Noise Variance ( pA ) 2Unresponsive† ( µV ) 01 . 6 ± 0 . 10 ( 4 ) 340 ± 331 . 2 ± 0 . 20/4124 . 8 ± 1 . 6 ( 9 ) 200 ± 522 . 1 ± 0 . 44/13226 . 4 ± 0 . 10 ( 8 ) 200 ± 311 . 6 ± 0 . 25/132869 ± 31 ( 6 ) 130 ± 451 . 3 ± 0 . 15/1135160 ± 34 ( 13 ) 68 ± 142 . 1 ± 0 . 411/24*Mean + SEM , number of cells is indicated in parenthesis . †Unresponsive cells were those where a flash did not evoke a response . Because ON-BCs making synapses with cone photoreceptors utilize a similar mGluR6-TRPM1 signaling pathway for generating depolarizing responses , we examined how changes in RGS concentration affected the light-evoked responses in cone ON-BCs ( Figure 7 , Table 4 ) . Given low abundance of cones in murine retinas and the functional heterogeneity of cone ON-BC that they connect to , we relied exclusively on ERG for these studies . Prior to tamoxifen administration , bright ( photopic ) flashes produced a characteristic two-component ERG waveform in cDKO mice . In contrast , after tamoxifen administration we observed a progressive reduction in the b-wave amplitudes and kinetics without detectable changes in the a-wave ( Figure 7A ) . We performed dose-response studies analyzing the dependence of photopic b-wave amplitude on flash intensity across all treatment groups . As in the case of rods , the component of the response that reflects cone contribution ( Figure 7B ) was derived mathematically from the biphasic response profile ( Figure 5C ) . The resulting traces revealed changes in cone-generated responses ( Figure 7B ) with a progressive decrease in maximal response amplitude and sensitivity . To verify that recorded responses reflected specifically changes in cone ON-BC signaling , we performed ERGs when rod function was suppressed by exposure to background light ( Figure 7C; Figure 7—figure supplement 1 ) . Light-evoked responses collected on this rod-suppressing background showed a progressive decline in b-wave amplitude and kinetics , just as for dark-adapted conditions . Quantification of the parameters derived from the analysis of intensity-response relationships indicated that maximal amplitude , sensitivity , and onset timing of the response similarly declined with time after tamoxifen administration ( Figure 7D , E ) . Together , these results indicate that down-regulation of RGS proteins in the cone synapses lead to changes in cone ON-BC light-evoked responses , that are similar to changes observed in rod synapses . 10 . 7554/eLife . 06358 . 015Figure 7 . The effect of progressive RGS loss on cone ON-BC responses as revealed by photopic ERG . ( A ) Representative dark-adapted ERG responses elicited with photopic ( 100 cd*s/m2 ) flashes from cDKO:Cre+ mice at 0 , 12 , 17 , 22 , 28 , and 35 days post-tamoxifen administration . ( B ) Cone-generated component of the ERG response extracted from fitting the second phase of the response in Figure 5B . ( C ) Representative light-adapted ERG responses elicited with photopic flashes ( 10 cd*s/m2 , left ) and ( 100 cd*s/m2 , right ) . Steady background light of 50 cd/m2 was applied to saturate rods . ( D ) ON-BC dose-response plot of maximal ERG b-wave amplitudes recorded with background light and plotted against their eliciting flash intensities . ( E ) Analysis of the changes in the ERG b-wave onset timing elicited by half-saturating flash intensities on a light background . Errors bars are SEM . Figure 2—figure supplement 1: isolation of the ERG b-wave peak . DOI: http://dx . doi . org/10 . 7554/eLife . 06358 . 01510 . 7554/eLife . 06358 . 016Figure 7—figure supplement 1 . Isolation of the ERG b-wave peak . ( A ) Single traces of light-adapted ERG responses elicited with photopic flashes of 100 cd*s/m2 . ( B ) Oscillatory potentials are removed from traces by a smoothing algorithm . Steady background light of 50 cd/m2 was applied to saturate rods . See details in the ‘Materials and methods’ . DOI: http://dx . doi . org/10 . 7554/eLife . 06358 . 01610 . 7554/eLife . 06358 . 017Table 4 . Effect of tamoxifen treatment on light-adapted photopic ERG b-wave parametersDOI: http://dx . doi . org/10 . 7554/eLife . 06358 . 017Days post tamoxifenI0 . 5 , photopic ( cd*s/m2 ) Rmax , photopicTime to peak ( ms ) at I0 . 505 ± 1204 ± 3135 ± 51210 ± 2111 ± 2242 ± 41723 ± 370 ± 763 ± 62265 ± 465 ± 669 ± 1028190 ± 2955 ± 473 ± 535260 ± 4453 ± 374 ± 6DKON . A0N . AValues are mean + SEM , n = 5 mice for all timepoints . To assess how changing responses in rod ON-BCs affects dim light vision we developed a behavioral water maze assay . The assay measures the ability of mice to locate an escape platform under varying luminance levels . Under photopic conditions , trained wild-type mice reached the visible platform in ∼10 s ( Figure 8A ) . Video rate analysis of tracks indicated that under these conditions mice swim directly to the visible platform taking the shortest path ( Figure 8A ) . In contrast , when the platform was hidden mice randomly swam and appeared to encounter the platform by chance , increasing the time to escape by nearly fivefold . Thus , mouse vision in this test can be evaluated by measuring the time to escape on the platform . 10 . 7554/eLife . 06358 . 018Figure 8 . Evaluation of scotopic vision using water maze behavioral test . ( A ) Development and validation of visually-guided behavioral task for the assessment of mouse vision . Mice were trained to find a randomly placed visible escape platform in a water maze . Their tracks were recorded and used to determine times to escape from water . Under bright photopic light environment , mice with constitutive elimination of RGS7 and RGS11 ( DKO ) find visible ( 7 ± 1 s ) , but not hidden ( 45 ± 4 s ) platform as readily as wild-type mice ( 7 ± 1 s ) . Error bars are SEM ( *p < 0 . 05 , t-test , n = 4–5 ) . Escape latencies thus served as a measure of their visual abilities . The room luminance was then decreased in a step-wise fashion followed by the re-assessment of mouse performance . DKO animals performed significantly worse than wild-type ( WT ) mice in the task at luminance levels ≤0 . 1 cd/m2 ( *p < 0 . 05 , t-test , n = 5 for both groups ) . ( B ) Behavioral performance of the conditional knockout mice ( cDKO ) mice in the visual task upon induction of RGS7 loss by tamoxifen administration . Mice were assessed under pure scotopic conditions equating to luminance of 0 . 005 cd/m2 . Performances of cDKO:Cre+ and cDKO:Cre- littermates were compared across days after tamoxifen treatment . At days 28 and 35 cDKO:Cre+ mice performed comparably to DKO mice , taking 28 ± 4 and 42 ± 7 s ( vs 36 ± 4 s for DKO ) to find the platform ( p > 0 . 05 , One-Way ANOVA , n = 5 for both groups ) . In comparison , performance of cDKO:Cre+ mice lagged that of cDKO:Cre- littermates at day 28 ( 28 ± 4 vs 10 ± 2 s ) and day 35 ( 42 ± 7 vs 14 ± 3 s ) ( *p < 0 . 05 , t-test for each time point , n = 5 for each group ) . ( C ) Elimination of RGS7 or RGS11 along with RGS7 haploinsufficiency does not affect mouse performance in the visual discrimination water maze task . Behavioral performance of WT , Rgs7−/− , and Rgs7+/−:Rgs11−/− mice was assessed in the water maze based behavioral task across various luminance levels covering the photopic and scotopic ranges . Performance was compared among all genotypes at each specific luminance level . DOI: http://dx . doi . org/10 . 7554/eLife . 06358 . 018 To validate the test , we evaluated the performance of the constitutive RGS7/11 double knockouts ( DKO ) , which completely lack an ERG b-wave and thus are considered night blind . Under photopic conditions , the escape latency of DKO mice with the visible platform was no different from that of wild-type animals , indicating that they see the platform normally ( Figure 8A ) . However , when the maze luminance was decreased to the scotopic range , the performance of DKO mice dropped sharply and their escape latencies became statistically equivalent to random encounters with the hidden platform , indicating a complete loss of scotopic vision ( Figure 8A ) . Thus , RGS expression in rod ON-BCs is absolutely required for high sensitivity scotopic vision . We next tested the performance of the cDKO mice under scotopic conditions . Before tamoxifen administration escape times of cDKO:Cre+ mice were substantially shorter than in DKO and no different from these of cDKO:Cre- littermates or WT mice , indicating that their scotopic vision remained intact ( Figure 8B ) . This performance did not decline up to 22 days after the beginning of tamoxifen treatment . However , a marked increase in escape time was observed in cDKO:Cre+ mice at 28 days , and by 35 days they displayed similar difficulty in visually locating the platform as the constitutive DKO animals , indicating that they developed night blindness ( Figure 8B ) . Administration of tamoxifen to RGS7 haploinsufficient or Rgs7−/− mice did not alter the performance compared to wild-type animals at any light levels ( Figure 8C ) . To understand better the relevance of RGS function in ON-BCs to behavior , we also assessed visual function of mice in a virtual environment using the optokinetic reflex ( OKR ) . The optomotor response test is based on the ability of mice to respond reflexively to computer-generated rotating sine-wave gratings , which form a virtual cylinder around them ( Prusky et al . , 2004 ) . Previous studies have established an essential contribution of ON-pathway to visual contrast sensitivity under both scotopic and photopic conditions reflecting the critical role of both rod and cone pathways to this behavior ( Schiller et al . , 1986; Iwakabe et al . , 1997; Kolesnikov et al . , 2011 ) . To assess scotopic contrast sensitivity change in cDKO mice we used the optimal mouse scotopic temporal frequency of 0 . 75 Hz in combination with a moderate stimuli speed of 7 . 5 deg/s ( Umino et al . , 2008 ) at which the contrast sensitivity is close to maximal ( Figure 9A ) . Before tamoxifen administration , both genotypes were indistinguishable in their ability to visualize rotating gratings at their contrast threshold . Tamoxifen treatment severely diminished the performance of cDKO:Cre+ mice without affecting their cDKO:Cre- littermates ( Figure 9A ) . Interestingly , the deterioration of scotopic vision was somewhat resistant to tamoxifen treatment until day 22 . These observations are in a good agreement with the data from the water maze task ( Figure 9B ) . Thus , the loss of RGS affects mouse scotopic vision in a stereotyped manner that does not depend on the paradigm chosen to evaluate behavioral performance . 10 . 7554/eLife . 06358 . 019Figure 9 . Evaluation of scotopic and photopic mouse vision by optomotor task . ( A ) Scotopic ( ∼3 × 10−4 cd/m2 ) contrast sensitivity deficits in mice lacking RGS proteins ( cDKO:Cre+ ) vs cDKO:Cre- mice as a function of time of post-tamoxifen administration ( *p < 0 . 05 , t-test , n = 4 ) . ( B ) Comparison of visual performance of mice in water maze task and optomotor response tests . Mouse vision declined by half after 25 ± 1 days of tamoxifen administration in both experiments . ( C ) The reliance of high speed contrast sensitivity on intact ON-BC function . Deficits in photopic ( 70 cd/m2 ) contrast sensitivity of DKO mice are apparent at stimuli speeds of >12 deg/s ( error bars are SEM; t-test: **p < 0 . 01 , ***p < 0 . 001 , n = 4–5 ) . ( D ) Visual acuity is mainly unaffected by the loss of RGS11 together with RGS7 ( DKO ) as compared to Rgs11−/− controls . ( error bars are SEM . t-test: *p < 0 . 05 , n = 4–5 mice ) . ( E ) Photopic ( 70 cd/m2 ) contrast sensitivity deficits in mice lacking RGS proteins ( cDKO:Cre+ ) vs cDKO:Cre- controls as a function of time of post-tamoxifen administration ( error bars are SEM; t-test: ***p < 0 . 001 , n = 7 ) . ( F ) Comparison of progressive RGS elimination effects on scotopic vs photopic contrast sensitivities of mice . The half-time of contrast sensitivity decline is 25 ± 1 days for scotopic vs 9 . 0 ± 2 days for photopic conditions , respectively . The data for each group are normalized to their respective values upon beginning of the tamoxifen treatment ( day 0 ) and expressed as percentage . DOI: http://dx . doi . org/10 . 7554/eLife . 06358 . 019 Finally , we used the OKR approach to evaluate the effect of gradual RGS loss on photopic vision . The parallel processing of cone-generated signals by different classes of BCs makes signaling through ON-BC not entirely necessary for photopic vision ( e . g . , Figure 8 ) . Therefore , we first determined the contribution of cone ON-BC to photopic ( 70 cd/m2 ) OKR performance using DKO mice lacking ON-BC responses and compared them to control Rgs11−/− littermates that show normal ON-BC function . We found that Rgs11−/− and DKO mice had virtually identical contrast sensitivity for slow stimuli ( up to 12 deg/s ) , whereas at higher stimuli speeds ( Sp ) the DKOs displayed impaired discrimination of low-contrast gratings pattern compared to their Rgs11−/− counterparts ( Figure 9C ) . In contrast , photopic visual acuity of DKO mice remained unaffected across most Sp and showed only a minor decline at the fastest speed of 50 deg/s ( Figure 9D ) . Having established the contribution of the cone ON-BC pathway to overall contrast discrimination , we evaluated the impact of declining RGS concentration on photopic vision . Tamoxifen administration in cDKO:Cre+ mice caused a 2 . 4-fold reduction in photopic contrast sensitivity observed at maximal stimuli speed of 50 deg/s in as early as 12 days ( Figure 9E ) . The average photopic contrast sensitivity reached its lowest level by day 22 , with a nearly a fivefold decline compared with either untreated littermates or a group of cDKO:Cre- controls that also received tamoxifen . This is in stark contrast to scotopic conditions ( ∼3 × 10−4 cd/m2 ) where contrast sensitivity did not start deteriorating until after day 22 following tamoxifen administration , and by day 35 it had reached a sevenfold decline as compared to cDKO:Cre- control mice ( Figure 9F ) . The availability of quantitative information regarding RGS levels at any time point in the analysis allowed us to determine how signal transmission to ON-BCs influences visually-guided behavior with the RGS concentration as a common denominator ( Figures 10 , 11 ) . Thus , we compared how both ON-BC ERG response parameters ( i . e . , b-wave onset , amplitude , and sensitivity ) and behavioral sensitivity varied as a function of the RGS7 concentration . These results reveal that the effect of RGS7 concentration on b-wave onset kinetics occurred with a half-saturating concentration of 37 ± 2% of available protein ( Figure 10A ) . Similarly , a halving of the response amplitude in rod ERG b-wave occurred at RGS7 levels of 39 ± 3% of available protein ( Figure 10B ) , and a twofold decline of scotopic response sensitivity occurred at RGS7 levels of 26 ± 2% ( Figure 10C ) . A similarly sharp dependence on RGS7 levels was observed for the scotopic vision of mice ( Figure 10D ) . Animals showed a precipitous drop in their visual contrast sensitivity once the levels of RGS7 declined below ∼35% , with 26 ± 2% reflecting the half-saturating level . To compare directly changes in rod ON-BC response characteristics to behavioral performance , we calculated RGS protein concentration that produced half-saturating responses across these different measures ( Figure 10E ) . The data reveals that rod-driven ERG response amplitudes , onset and sensitivity are disproportionately affected by changes in RGS concentration . Although measured response parameters are interrelated by their underlying biochemical mechanism , they correlate differentially with the mouse behavioral performance . For example , a threefold reduction in response amplitude and a twofold deceleration of ERG b-wave onset did not appear to have an appreciable effect on scotopic visual performance . Instead , mouse scotopic behavior tightly correlated with the rod ON-BC response sensitivity , albeit compounded by pronounced changes in other response parameters . 10 . 7554/eLife . 06358 . 020Figure 10 . Correlation of RGS levels with key parameters of rod-rod ON-BC synaptic transmission and behavioral performance under scotopic conditions . Dose response relationship between half-maximal values of rod ERG b-wave onset timing ( A ) , maximal response amplitude ( B ) , half-saturating light intensity ( I0 . 5 ) ( C ) and performance in the behavioral task ( D ) were plotted as a function of RGS7 concentration quantitatively determined at each time point following tamoxifen administration . In panels A–D data points are fitted with sigmoidal dose-response ( variable slope ) equation with the least squares method . ( E ) Comparison of changes from different response parameters . RGS7% was converted to RGS7 protein levels using quantitative values determined in Figure 4 . Scotopic behavioral performance is resistant to RGS reduction and tracks most closely with rod ON-BC sensitivity . One-Way ANOVA with Bonferroni's post hoc test reveals that half-maximal values for sensitivity and behavior do not significantly differ from each other ( p > 0 . 05 , n = 5 ) , but differ from those for response kinetics and amplitude ( p < 0 . 05 , n = 5 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06358 . 02010 . 7554/eLife . 06358 . 021Figure 11 . Correlation of RGS levels with key parameters of cone-cone ON-BC synaptic transmission and behavioral performance under photopic conditions . Dose response relationship between half-maximal values of cone ERG b-wave onset timing ( A ) , maximal response amplitude ( B ) , half-saturating light intensity ( I0 . 5 ) ( C ) and performance in the behavioral task ( D ) were plotted as a function of RGS7 concentration quantitatively determined at each time point following tamoxifen administration . In panels A–D data points are fitted with sigmoidal dose-response ( variable slope ) equation with the least squares method . ( E ) Comparison of changes from different response parameters . RGS7% was converted to RGS7 protein levels using quantitative values determined in Figure 4 . Photopic behavioral performance is highly affected by RGS reduction and tracks closely with response amplitude and kinetics . One-Way ANOVA with Bonferroni's post hoc test reveals that half-maximal value for sensitivity differs from all the other parameters ( p < 0 . 05 , n = 5 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06358 . 021 A similar relationship was derived for the cone-driven photopic ERG responses and behavioral sensitivity ( Figure 11 ) . We found that a halving of the photopic ERG b-wave amplitude occurred at an RGS7 concentration of 43 ± 2% of available protein , a twofold effect on onset kinetics occurred at an RGS7 concentration of 38 ± 2% of available protein , and a twofold decline in sensitivity occurred at an RGS7 concentration of 25 ± 2% of available protein . Thus the photopic ERG b-wave amplitude was the response parameter most sensitive to the reduction in RGS7 . However , in contrast to rod-mediated vision , we found that behavioral performance relying on the cone ON pathway was very sensitive to even small decline in RGS7 , losing half of their photopic visual performance with 43 ± 2% of the RGS remaining ( Figure 11D ) . To compare changes in cone ON-BC response characteristics to behavioral performance we plotted the absolute RGS protein concentration that produced half-saturating responses ( Figure 11E ) . We found that photopic vision is more sensitive to changes in the amplitude and kinetics of the cone ON-BC light-evoked responses rather than their light sensitivity . In this study , we used the well-defined rod and cone ON circuits of the retina to examine how changing key parameters of GPCR-mediated responses affected visually-guided behavior . Through genetic manipulations we observed graded alterations in the maximal response amplitudes , timing of response onset , and sensitivity of rod and cone ON-BCs , while testing the effect of these changes on mouse visual performance . We found stark differences between rod- and cone-driven ON circuits with respect to the influence of signal transfer parameters on visually-guided behavior . In the rod circuitry , changes in the timing or amplitude of signal transmission between photoreceptors and ON-BC neurons did not affect behavioral performance in visually-guided tasks that required intact dim vision . However , a change in the sensitivity of transmission at this synapse , albeit aggravated by kinetic deficits , was not tolerated behaviorally and loss of the response sensitivity strictly correlated with the decrease in behavioral performance of the animals . In contrast , cone-mediated vision was sensitive to minor changes in kinetic parameters of signal transmission , causing a pronounced decrease in behavioral performance even when light sensitivity was unaffected . The rod circuit is biased for high sensitivity to detect single photon absorptions close to absolute visual threshold ( Pahlberg and Sampath , 2011 ) , whereas cone pathways operate at higher light levels and require the capacity to extract temporal features from light stimuli ( Rodieck , 1998 ) . Thus , our observations suggest that differential specialization of rod and cone pathways involves adaptation at the first synapse in these respective circuits . Here we find that the parameters of the GPCR-mediated response can be dissociated to provide a hierarchical influence on function . In ON-BCs , decreasing concentrations of two RGS proteins expressed in dendritic tips , RGS7 and RGS11 , sequentially influence the onset timing , amplitude , and sensitivity of mGluR6-mediated responses ( Figure 12 ) . In this pathway , response kinetics and amplitude appeared to be the parameters most sensitive to changes in RGS concentration , whereas sensitivity required a much larger reduction in RGS concentration to produce an impact . There is a window in the RGS concentration range where profound effects on kinetics were not accompanied by appreciable changes in response sensitivity . These observations suggest a mechanism by which GPCRs can generate unique patterns of activity and differentially tune circuit properties based on varying levels of RGS proteins . Thus , regulatory nodes where information processing is mediated by GPCRs may be tuned to extract temporal features at one end of the RGS concentration range , whereas others may serve as sensitivity filters at the other end of the range . This possibility is supported by observations in photoreceptors where manipulation of RGS9 concentration affects temporal characteristics of the response without changes in sensitivity ( Chen et al . , 2000; Krispel et al . , 2006 ) and in hippocampal pyramidal neurons where knockout of RGS proteins changes both response kinetics and sensitivity ( Xie et al . , 2010 ) . Interestingly , levels of multiple RGS proteins have also been shown to fluctuate in response to extracellular cues ( Hurst and Hooks , 2009; Traynor et al . , 2009 ) suggesting that changes in RGS concentration may dynamically control GPCR response properties . 10 . 7554/eLife . 06358 . 022Figure 12 . Model for RGS function in setting ON-BC responses to light . In the dark , mGluR6 activated by high synaptic glutamate concentration produces ample amounts of GTP-bound Go , which closes TRPM1 channels . RGS proteins ( RGS7 and RGS11 ) deactivate Go converting it back to inactive GDP-bound state but the equilibrium is dominated by excess of Go-GTP ensuring no channel activity . Upon light exposure , the activity of mGluR6 is reduced , and reduction in Go-GTP catalyzed by RGS proteins becomes dominant , allowing channels to open . When concentration of RGS proteins decline below a threshold point the speed of Go deactivation becomes rate-limiting to reduce the speed and magnitude of TRPM1 channel opening as well as its sensitivity to the reduction in glutamate concentration . DOI: http://dx . doi . org/10 . 7554/eLife . 06358 . 022 Under native conditions , RGS proteins in ON-BCs appear to be present in substantial excess over the minimal amount needed for sustaining wild-type response properties . Rod ON-BCs tolerate an almost 75% reduction in RGS levels before any measurable effects on response properties could be observed . To understand the basis for this overexpression we compared levels of RGS7 and RGS11 to that of the mGluR6 receptors . We found that RGS proteins are present at approximately threefold stoichiometric excess over mGluR6 dimers . Several recent studies suggest that the signal transduction components of the ON-BC are organized in a macromolecular complex ( Morgans et al . , 2010; Dhingra and Vardi , 2012 ) . Indeed , both RGS7 and RGS11 form physical complexes with mGluR6 ( Cao et al . , 2009 ) and a recently discovered orphan receptor GPR179 ( Orlandi et al . , 2012; Ray et al . , 2014 ) . These interactions are required for proper postsynaptic targeting of RGS7/11 ( Cao et al . , 2009; Orlandi et al . , 2012 ) . The expression of RGS proteins also depends on mGluR6 , and its elimination in mice significantly decreases RGS7 and RGS11 levels ( Cao et al . , 2009 ) . Additionally , mGluR6 and its effector channel TRPM1 are further integrated by a scaffolding protein , nyctalopin ( Cao et al . , 2011; Pearring et al . , 2011 ) . The relatively rapid ( ∼50–100 ms ) response of ON-BCs to flashes of light is unlikely to be supported by the stochastic action of RGS proteins if they were deactivating Gαo in the vicinity of TRPM1 simply by diffusion . Thus , these considerations suggest a model where functionally relevant RGS molecules must be physically integrated into the macromolecular signaling complex . Since RGS proteins are present in the molar excess over the mGluR6 complex , their depletion does not alter the signaling until the level reaches the stoichiometric threshold and RGS is reduced within the macromolecular assembly . Therefore , it seems reasonable to speculate that overexpression of RGS proteins over the stoichiometry with the mGluR6 signaling complex would minimize fluctuation in the kinetics and sensitivity of ON-BC responses , ensuring their reproducibility . In support of this hypothesis we find that titration of RGS proteins below a certain threshold point ( ∼1:1 stoichiometry with mGluR6 complex ) detrimentally affects the signal-to-noise ratio of ON-BC responses and increases their variability . These results additionally suggest that G protein deactivation normally is not a rate-limiting step in the cascade of events that lead to generation of a depolarizing response in ON-BCs . Previous studies in rod photoreceptors that utilize similar G protein cascade for light reception demonstrated that G protein deactivation limits the recovery of a photoresponse . In these neurons , increase in the concentration of RGS protein ( RGS9 ) resulted in acceleration of the response termination kinetics ( Krispel et al . , 2006; Burns and Pugh , 2009 ) . Since a reduction in mGluR6 activity underlies the light-evoked responses in ON-BCs ( compared to an increase in rhodopsin's activity for phototransduction ) , G protein deactivation drives the activation phase of the response and RGS proteins influence its slope . By titrating the RGS concentration in the ON-BCs we found a critical threshold point , below which the onset of the ON-BC depolarization is sensitive to changes in G protein deactivation . The critical threshold point roughly corresponds to ∼25% of normal RGS levels ( contributed by both RGS7 and RGS11 ) in wild-type synapses . Notably , a substantial excess in RGS concentration ( threefold to fourfold ) , and hence G protein deactivation rates above this threshold , has no influence on ON-BC response properties . Thus , by setting the levels of RGS expression above this threshold , mGluR6 signaling in ON-BCs ensures that G protein deactivation is not a rate-limiting step in generation of a depolarizing response . Instead , this rate-limiting reaction could be associated with either TRPM1 channel opening or intrinsic deactivation of the mGluR6 receptor . Determining the identity of this rate limiting process that drives the onset of ON-BC depolarizing response will be an important area for future investigation . Interestingly , we found that RGS proteins have similar impact on response properties of ON-BCs that form synapses with either rod or cone photoreceptors . Given monosynaptic connectivity between rods and rod ON-BCs , and the progressive night blindness associated with RGS loss , the implications of these changes for rod-mediated behavior are rather unequivocal . However , signaling of cones via parallel ON- and OFF- circuits , together with the heterogeneous nature of cone ON-BCs , complicates the interpretation of the dependence of photopic ( cone ) vision on cone-to-cone ON-BC signal transmission . The ability to discriminate changes in luminance across a broad range of stimulation frequencies is characteristic feature of photopic vision in many species , including mice ( Umino et al . , 2008 ) . Previous studies have indicated that pharmacological or genetic blockade of the ON-pathway affects contrast sensitivity in mice and monkeys under photopic conditions ( Schiller et al . , 1986; Iwakabe et al . , 1997 ) . In line with these observations , we find that changes in amplitude and kinetics of the photopic ERG b-wave caused by the loss of RGS expression result in pronounced deficits in the temporal properties of contrast detection for cone vision . Ablation of RGS proteins severely impacted photopic contrast sensitivity for high-speed visual stimuli while leaving visual acuity relatively unaffected . This highlights an essential role that temporal regulation of mGluR6-TRPM1 signaling at the cone-to-cone ON-BC synapse plays for normal photopic vision . Although the details on how RGS proteins may contribute to the responses of individual cone ON-BCs remain to be established , it is possible that differential expression of RGS7 vs RGS11 at cone ON-BC synapses ( Mojumder et al . , 2009 ) may contribute to setting their unique profiles , thus increasing the temporal resolution of cone vision . Generation of mice with the constitutive deletion in Rgs7 ( Rgs7−/− ) ( Cao et al . , 2012 ) , Rgs11 ( Rgs11−/− ) ( Cao et al . , 2008 ) , or both ( DKO ) ( Cao et al . , 2012 ) was previously described . Conditional targeting of RGS7 was achieved by flanking exon 4 with LoxP sites ( Rgs7flx/flx ) ( Cao et al . , 2012 ) . The resulting mice were crossed with Rgs11−/− line to generate cDKO:Cre- strain ( Rgs11−/−: Rgs7flx/flx ) , and further with a Cre driver line ubiquitously expressing tamoxifen-inducible Cre-ERT2 recombinase B6 . Cg-Tg ( CAG-cre/Esr1* ) 5Amc/J to produce cDKO:Cre+ mice ( Rgs11−/−: Rgs7flx/flx: CAG-CreERT2 ) . To induce Cre expression 20 mg/kg tamoxifen ( Sigma , St . Louis , MO ) , dissolved in corn oil and 10% ethanol , was administered by oral gavage with an 18–24 gauge smooth tip needle . Tamoxifen was administered once daily for 5 consecutive days . All procedures were carried out in accordance with the National Institute of Health guidelines and were granted formal approval by the Institutional Animal Care and Use Committees of the Scripps Research Institute ( IACUC protocol number 14-001 ) , Washington University ( IACUC protocol number 20140236 ) , and the University of Southern California ( IACUC protocol number 10890 ) . The generation of sheep anti-TRPM1 is described ( Cao et al . , 2011 ) . Rabbit anti-RGS7 ( 7RC1 ) , was a generous gifts from William Simonds ( NINDDK/NIH ) and the guinea pig anti-mGluR6 antibody was a gift from Dr Takahisa Furukawa ( Osaka University ) . Mouse anti-PKCα ( ab11723; Abcam , Cambridge , MA ) , rabbit anti-RGS7 antibodies ( 07-237; Upstate Biotechnology , Billerica , MA ) , mouse anti-CtBP2 ( 612044; BD Biosciences ) and rabbit anti-cone arrestin ( ab15282; Millipore , Billerica , MA ) were purchased . Recombinant His-tagged RGS7 and RGS11 were co-expressed in Sf9 insect cells together with Gβ5 , via baculovirus-mediated delivery , then purified by Ni-NTA chromatography as described previously ( Martemyanov et al . , 2005 ) . Glutathione S-transferase ( GST ) -tagged mouse mGluR6 C terminus protein ( aa840-871 ) was expressed in Escherichia coli and affinity purified on GSTrap HP column ( GE , Fairfield , CT ) . Protein concentration was determined by bicinchoninic acid ( BCA ) Protein Assay Kit ( Pierce , Carlsbad , CA ) and adjusted to reflect the protein purity determined by densitometry of Coomassie blue-stained gels . Whole retinas were removed from mice and lysed by sonication in ice-cold PBS supplemented with 150 mM NaCl , 1% Triton X-100 , and Complete protease inhibitor tablets ( Roche , Basel , Switzerland ) . Lysates were cleared by centrifugation at 20 , 800×g for 15 min at 4°C . Total protein concentration in the supernatant was measured by using BCA Protein Assay Kit ( Pierce , Carlsbad , CA ) . Supernatants were added with SDS sample buffer ( pH 6 . 8 ) containing 8 M urea and were subjected to 12 . 5% SDS/PAGE . Protein bands were transferred onto PVDF membranes , subjected to Western blot analysis first with primary antibodies against RGS7 , RGS11 or mGluR6 and then with HRP-conjugated secondary antibodies , and detected by using ECL West Pico system ( Pierce , Carlsbad , CA ) . For the quantitative detection of RGS7 we used rabbit anti-RGS7 antibodies ( 07-237; Upstate Biotechnology , Billerica , MA ) . Signals were captured on film and scanned by densitometer , and band intensities were determined by using NIH ImageJ software . Dissected eyecups were fixed for 15 min in 4% paraformaldehyde , cryoprotected with 30% sucrose in PBS for 2 hr at room temperature , and embedded in optimal cutting temperature medium . 12-micrometer frozen sections were obtained and blocked in PT1 ( PBS with 0 . 1% Triton X-100 and 10% donkey serum ) for 1 hr , then incubated with primary antibody in PT2 ( PBS with 0 . 1% Triton X-100 and 2% donkey serum ) for at least 1 hr . After four washes with PBS with 0 . 1% Triton , sections were incubated with fluorophore-conjugated secondary antibodies in PT2 for 1 hr . After four washes , sections were mounted in Fluoromount ( Sigma , St . Louis , MO ) . Images were taken with a Leica SP800 confocal microscope . Quantitative analysis of immunofluorescence from confocal images was performed using Leica software . At each time point ( day: 0 , 12 , 17 , 22 , 28 , 35 ) immunohistochemical staining of cDKO:Cre+ was repeated twice and at least two sections obtained from two individual animals were used for analysis and averaging . Sections were double stained for RGS7 and marker protein mGluR6 , which co-localizes with RGS7 . Positive staining of mGluR6 was used as a reference for the normalization of the recorded fluorescence intensity in the channel containing RGS7 protein . Rod and cone synapses were differentiated by: ( i ) counter staining with PNA and/or b-arrestin that specifically label cone terminals , ( ii ) sublamina position in the outer plexiform layer , ( iii ) specific clustering pattern of synapses at cone terminals . The fluorescence intensity within synaptic puncta was analyzed using line-scan mode and a constant puncta-encircling area , which tightly surrounded the contours of each puncta . Mean intensity ( measured in pixels ) were averaged across ∼20 individual punctas per imaged section , taking three sections per retina , and two to three retinas used per time point . Imaging parameters were the same for all sections and retinas . Values for groups were normalized using pre-tamoxifen RGS7 values as a 100% percentage of protein remaining . RGS7 protein percentages were plotted vs the tamoxifen time course and fitted nicely with a simple decay function in GraphPad Prism 6 . Electroretinograms were recorded by using the UTAS system and a BigShot Ganzfeld ( LKC Technologies , Gaithersburg , MD ) . Mice ( 4–8 wk old ) were dark-adapted ( ≥6 hr ) or light adapted ( 50 cd/m2 , 5 min ) and prepared for recordings by using dim red light . Mice were anesthetized with an i . p . injection of ketamine and xylazine mixture containing 100 and 10 mg/kg , respectively . Recordings were obtained from the right eye only , and the pupil was dilated with 2 . 5% phenylephrine hydrochloride ( Bausch & Lomb , Bridgewater , NJ ) , followed by the application of 0 . 5% methylcellulose . Recordings were performed with a gold loop electrode supplemented with contact lenses to keep the eyes immersed in solution . The reference electrode was a stainless steel needle electrode placed subcutaneously in the neck area . The mouse body temperature was maintained at 37°C by using a heating pad controlled by ATC 1000 temperature controller ( World Precision Instruments , Sarasota , FL ) . ERG signals were sampled at 1 kHz and recorded with 0 . 3 Hz low-frequency and 300 Hz high-frequency cut-offs . Full field white flashes were produced by a set of LEDs ( duration < 5 ms ) for flash strengths ≤2 . 5 cd·s/m2 or by a Xenon light source for flashes > 2 . 5 cd·s/m2 ( flash duration < 5 ms ) . ERG responses were elicited by a series of flashes ranging from 2 . 5 × 10−4 to 2 . 5 × 101 cd·s/m2 in increments of 10-fold . Ten trials were averaged for responses evoked by flashes up to 2 . 5 × 10−1 cd·s/m2 , and three trials were averaged for responses evoked by 2 . 5 × 100 cd·s/m2 flashes . Single flash responses were recorded for brighter stimuli . To allow for recovery , interval times between single flashes were as follows: 5 s for 2 . 5 × 10−4 to 2 . 5 × 10−1 cd·s/m2 flashes , 30 s for 2 . 5 × 100 cd·s/m2 flashes , and 180 s for 2 . 5 × 101 cd·s/m2 flashes . ERG traces were analyzed using the EM LKC Technologies software , Sigma Plot and Microsoft Excel . The b-wave amplitude was calculated from the bottom of the a-wave response to the peak of the b-wave . To isolate the b-wave peak from sharp varying oscillatory potentials , a local regression smoothing algorithm using a bi-square weighting function was used in SigmaPlot ( see Figure 7—figure supplement 1 ) . Traces obtained from DKO , that completely lack the ON-BC triggered b-wave were subtracted from ERG traces recorded in cDKO:Cre+ mice following tamoxifen administration , each at corresponding light intensities . This was done to remove ambiquity of assigning diminishing b-wave due to interference with P1 component . The data points from the b-wave stimulus–response curves were fitted by Equation 1 using the least-square fitting method in GraphPad Prism6 . ( 1 ) R=Rmax , r×I/ ( I+I0 . 5 , r ) +Rmax , c×I/ ( I+I0 . 5 , c ) . The first term of this equation describes rod-mediated responses ( r ) , and the second term accounts primarily for responses that were cone mediated ( usually at flash intensities ≥1 cd·s/m2 for dark-adapted mice; index c ) . Rmax , r and Rmax , c are maximal response amplitudes , and I0 . 5 , r and I0 . 5 , c are the half-maximal flash intensities . Stimulus responses of retina cells increase in proportion to stimulus strength and then saturate , this is appropriately described by the hyperbolic curves of this function . The maximal amplitudes of rod and cone-driven b-wave stimulus–response curves in Tables 1 , 2 are plotted against RGS7 protein concentration and fitted with a simple dose–response curve ( variable slope ) in GraphPad Prism6 . Time to peak of the b-wave was measured as the time from the peak of a-wave to the peak of the b-wave . Recordings of light-evoked currents in rod ON-BCs were performed as described previously ( Okawa et al . , 2010; Cao et al . , 2012 ) . Briefly , mice were dark-adapted overnight and euthanized according to protocols approved by the Institutional Animal Care and Use Committee of the University of Southern California ( Protocol 10890 ) . Retinas were isolated , embedded in a low-gelling temperature agar , and 200 µm slices were made on a vibrating microtome ( Leica VT-1000S ) . Retinal slices were placed on the stage of an upright microscope and superfused with heated Ames Media ( A-1420; 35-37°C ) equilibrated with 5% CO2/95% O2 . Whole-cell voltage clamp recordings ( Vm = −60 mV ) were made while delivering flashes of light that varied in strength from evoking a just-measurable response to those generating a maximal response . Noise variance was computed from the 0 . 2 s preceding the flash . Rod ON-BCs were identified in retinal slices based on the distinctive shape of the cell body and its position in the inner nuclear layer jammed up against the outer plexiform layer . The internal solution for whole-cell recordings consisted of ( in mM ) : 125 K-Aspartate , 10 KCl , 10 HEPES , 5 NMG-HEDTA , 0 . 5 CaCl2 , 1 ATP-Mg , 0 . 2 GTP-Mg; pH was adjusted to 7 . 2 with NMG-OH . Membrane currents were filtered at 300 Hz by an 8-pole Bessel filter , and digitized at 1 KHz . Mouse visual behavior was assessed using a water maze task with a visible escape platform . The method for assessing visual function in this experiment is principled on a Morris water maze ( Morris , 1984 ) and previous reports describing evaluation of mouse vision by a swimming-based task ( Prusky et al . , 2000 ) . Mice are natural swimmers and this task exploits their innate inclination to escape from water to a solid substrate . This task uses an ability of a mouse to see a visible platform with a timed escape from water , as an index of its visual ability . Before testing , mice readily learned to swim to the visible escape-platform and performance usually plateaued at around 10 s within 15 trials for all treated groups . Mice which did not learn the task , for example , performance did not improve or plateau for at least the last three or more consecutive trials or had any visible motor deficits were discarded from the experiment . Visual-guided behavior was tested at 100 , 1 , 0 . 1 , 0 . 01 , and 0 . 001 cd/m2 and timed performances from 20 trials ( four sessions of five trials each ) for each mouse at each light-intensity were averaged . Uniform room luminance settings were stably achieved by an engineered adjustable light-source and constantly monitored with a luminance meter LS-100 ( Konica Minolta , Tokyo , Japan ) . To be certain that we were measuring the mice's visual ability only and not memory , the platform was placed pseudo-randomly in the water tank and all external visual cues were eliminated . Visual acuity and contrast sensitivity of mice was evaluated from optomotor responses using a two-alternative forced-choice protocol ( Umino et al . , 2008; Kolesnikov et al . , 2011 ) . Briefly , the animal was placed on a pedestal surrounded by four computer monitors and observed using infrared-sensitive television camera . For scotopic measurements , a round array of six infrared LEDs mounted above the platform was used to visualize the animal . Mice responded to visual stimuli ( sine-wave vertical gratings presented by computer using staircase paradigm and invisible to the experimenter ) , by reflexively rotating their head in either clockwise or counterclockwise direction . The observer's task was to track the direction of optomotor responses and report it to the computer which determined the correctness of the choice ( Prusky et al . , 2004 ) . Photopic visual acuity was determined as the threshold for spatial frequency ( Fs ) of the stimuli at 100% contrast . Contrast sensitivity was defined as the inverse of obtained contrast threshold values . Photopic responses of constitutive RGS7/11 DKO mice were measured at un-attenuated luminance of monitors ( 70 cd/m2 at the mouse eye level ) , over a range of various Sp , from 3 to 50 deg/s . Fs of stimuli was kept constant at its optimal value of 0 . 128 cyc/deg for all speed regimes thus providing a range of corresponding temporal frequencies ( Ft ) from 0 . 38 to 6 . 4 Hz , according to following equation: Ft = Sp × Fs ( Umino et al . , 2008 ) . Photopic contrast sensitivities of cDKO mice were determined at specified low ( 6 deg/s ) or high ( 50 deg/s ) Sp at 70 cd/m2 luminance level . Scotopic contrast sensitivity of cDKO mice was evaluated at optimal values of Ft ( 0 . 75 Hz ) and Sp ( 0 . 1 cyc/deg ) , and a corresponding stimuli speed of 7 . 5 deg/s . In this mode , the monitor luminance was attenuated to ∼3 × 10−4 cd/m2 by several neutral density film filters that formed a cylinder around the animal . All data were analyzed using independent two-tailed Student t-test , with accepted significance level of p < 0 . 05 .

At the back of the eye , a structure called the retina contains several types of cell that convert light into the electrical signals that the brain interprets to produce vision . Cells called rods and cones detect the light , and then signal to other neurons in the retina that relay this information to the brain . Rods and cones are specialized to respond best to different visual features: cones detect color and can track rapid movement; whereas rods are more sensitive to low light levels and so enable night vision . All rods and cones communicate with particular types of neuron called an ‘ON bipolar cell’: rods send their information to rod-specific ON bipolar cells and cones to cone ON-bipolar cells . To maintain the differences in how visual features are detected , the signals sent by the rod or cone cells need to be tuned separately . Previous studies showed that bipolar cells rely on the action of proteins called RGSs to control how information is passed from rods and cones to ON bipolar cells . However , how the RGS proteins produce their effects is not well understood , and neither is their impact on vision or behavior . Sarria et al . used a genetic approach to create mice that progressively lost RGS proteins from their retina over the course of several weeks . Recording the nerve impulses produced by the bipolar cells as light shone on the retina revealed that RGS depletion affects these neurons in three ways: how sensitive they are to the signals sent by the rod and cone cells , how quickly they respond to a signal , and the size of the electrical response that they produce . Sarria et al . then investigated how these changes affected the behavior of the mice . To test the response of the rod cells , the mice performed tasks in dim light . This revealed that it was only when the sensitivity of the bipolar cells decreased that the mice performed worse . However , in a task involving fast-moving objects that investigated the response of cone cells , only changes to the speed of the response affected vision . Therefore , the RGS protein has different effects on the signals from rod cells and cone cells . These findings will be useful for understanding how different light sensitive cells in the retina communicate their signals to extract important visual features , allowing us to both see well at night and track rapid changes in scenery on a bright sunny day .

2015

Sensitivity and kinetics of signal transmission at the first visual synapse differentially impact visually-guided behavior

Dendritic and axonal arbors of many neuronal types exhibit self-avoidance , in which branches repel each other . In some cases , these neurites interact with those of neighboring neurons , a phenomenon called self/non-self discrimination . The functional roles of these processes remain unknown . In this study , we used retinal starburst amacrine cells ( SACs ) , critical components of a direction-selective circuit , to address this issue . In SACs , both processes are mediated by the gamma-protocadherins ( Pcdhgs ) , a family of 22 recognition molecules . We manipulated Pcdhg expression in SACs and recorded from them and their targets , direction-selective ganglion cells ( DSGCs ) . SACs form autapses when self-avoidance is disrupted and fail to form connections with other SACs when self/non-self discrimination is perturbed . Pcdhgs are also required to prune connections between closely spaced SACs . These alterations degrade the direction selectivity of DSGCs . Thus , self-avoidance , self/non-self discrimination , and synapse elimination are essential for proper function of a circuit that computes directional motion . The geometry of a neuron's dendritic and axonal arbors is believed to be a major determinant of the neuron's role within a circuit . In some cases , the relationship is clear: in sensory systems , for example , the size and shape of a dendritic arbor determine the size and shape of the neuron's receptive field , and the degree of branching within the arbor determines how densely the field is sampled ( Lefebvre et al . , 2015 ) . Other cases are more conjectural , and in very few cases have experiments attempted to make a causal link between particular dendritic geometries and neuronal function . Here , we address this issue by analyzing a retinal direction-selective circuit . The phenomena we investigate are self-avoidance and self/non-self discrimination ( S/NSD ) . In self-avoidance , sibling dendritic branches do not contact each other . Although not all neurons exhibit self-avoidance , this phenomenon has been observed in a variety of systems including sensory neurons of leech ( Hirudo medicinalis; in which the process was first described ) , moth ( Manduca sexta ) , fruit fly ( Drosophila melanogaster ) , worms ( Caenorhabditis elegans ) , and zebrafish ( Danio rerio ) ( Nicholls and Baylor , 1968; Yau , 1976; Kramer and Kuwada , 1983; Kramer and Stent , 1985; Grueber et al . , 2001 , 2003; Liu and Halloran , 2005; Sagasti et al . , 2005; Smith et al . , 2012 ) . Dendrites of olfactory projection neurons and axons of mushroom body neurons also exhibit self-avoidance in Drosophila ( Wang et al . , 2002a; Zhan et al . , 2004; Hattori et al . , 2007 ) . In mammals , self-avoidance has been documented in cerebellar Purkinje cells and some types of retinal horizontal , bipolar , amacrine , and ganglion cells ( Montague and Friedlander , 1991; Wassle et al . , 2009; Lefebvre et al . , 2012; Matsuoka et al . , 2012 ) . Several cell-surface proteins have been implicated in self-avoidance , including Dscam1 , Turtle , Flamingo , LAR-like receptor tyrosine phosphatase , Unc-5 , Unc-6 ( Netrin ) , and Unc-40 ( DCC ) in invertebrates ( Baker and Macagno , 2000; Gao et al . , 2000; Matthews et al . , 2007; Long et al . , 2009; Smith et al . , 2012 ) and Dscam , DscamL1 , Slit , Robo , Sema6A , PlexA4 , PlexA2 , and gamma-Protocadherins ( Pcdhgs ) in mice ( Fuerst et al . , 2008 , 2009; Lefebvre et al . , 2012; Matsuoka et al . , 2012; Sun et al . , 2013; Gibson et al . , 2014 ) . In each case , they appear to act through contact-dependent repellent mechanisms . In some instances , processes of neurons that exhibit self-avoidance do not avoid other neurons of the same type; rather , they overlap extensively with and sometimes even form synapses on each other . Thus , these neurons appear to discriminate between their own processes , which they repel , and those of their neighbors , with which they interact ( Figure 1A ) . This puzzling observation suggests that processes of nominally identical neurons are immune to the repellent forces that act within each other's arbors , a phenomenon that has been called S/NSD ( Zipursky and Grueber , 2013 ) . Of the molecules that mediate self-avoidance , two have also been shown to mediate S/NSD: fly Dscam1 and mouse Pcdhgs ( Hattori et al . , 2007; Hughes et al . , 2007; Matthews et al . , 2007; Soba et al . , 2007; Lefebvre et al . , 2012 ) . While Dscam1 and Pcdhg proteins are not structurally related , they have three properties that allow them to mediate both self-avoidance and S/NSD . First , both are transmembrane recognition molecules with remarkable extracellular diversity . Alternative splicing of the Dscam1 transcripts and alternative promoter choice ( Figure 1B ) plus isoform multimerization of Pcdhgs lead to >10 , 000 recognition units ( Schmucker et al . , 2000; Tasic et al . , 2002; Murata et al . , 2004; Schreiner and Weiner , 2010; Thu et al . , 2014 ) . Second , each Dscam1 and Pcdhg isoform binds homophilically , but does not bind appreciably to other , closely related isoforms ( Wojtowicz et al . , 2004 , 2007; Schreiner and Weiner , 2010; Thu et al . , 2014 ) . Finally , in those cases where tests have been made , each neuron in a population expresses a small randomly selected subset of isoforms ( Neves et al . , 2004; Zhan et al . , 2004; Kaneko et al . , 2006; Miura et al . , 2013; Toyoda et al . , 2014 ) , leading to molecular diversification that , in the case of Drosophila Dscam1 , has been demonstrated to be important for proper patterning of neural circuits ( Hattori et al . , 2009 ) . Together , these observations have led to a model for self-avoidance and S/NSD in which Dscam1- and Pcdhg-mediated homophilic interactions generate signals leading to repulsion . Because all dendrites ( or axons ) of a single neuron display the same set of Dscam1 or Pcdhg isoforms , they exhibit self-avoidance . On the other hand , any individual neuron is unlikely to encounter a neighbor that displays the same combination of isoforms , so the neurons do not repel each other and thus display S/NSD . 10 . 7554/eLife . 08964 . 003Figure 1 . Pcdhg-dependent self-avoidance and self/non-self discrimination in SACs . ( A ) Self-avoiding neurites lack isoneuronal contacts ( repulsion ) but adhere to and can form synapses with neurites of other cells of the same type , displaying self/non-self discrimination ( adhesion ) . ( B ) Schematic of Pcdhg genomic locus and protein product . Distinct Pcdhg isoforms are assembled by splicing one of 22 variable exons , encoding the extracellular and transmembrane portions of the protein , to three constant exons , encoding the intracellular portion of the protein . ( C ) Vertical section of retina stained against ChAT to label all SACs ( gray ) overlaid with cartooned individual OFF and ON SACs ( red ) . OFF SAC cell bodies reside in the inner nuclear layer ( INL ) and ON SAC cell bodies reside in the ganglion cell layer ( GCL ) . SAC neurites reside in the inner plexiform layer . ( D ) En face view of individual dye-filled ON SAC in Pcdhg22 retina . ( E ) Schematic of the retinal direction-selective circuit components and connections . PRs , photoreceptors; BCs , bipolar cells; SACs , starburst amacrine cells; DSGC , direction-selective ganglion cell . Gray stripes indicate OFF and ON direction-selective sublaminae ( S2 and S4 , respectively ) . Green and red arrows indicate directional preferences of DSGCs and SAC dendrites , respectively . ( F–H ) Schematic representation of the effects of changing Pcdhg expression in SACs ( summary from Lefebvre et al . , 2012 ) . SACs from Pcdhg22 retinas ( F ) are posited to express unique subsets of Pcdhgs and thus exhibit both self-avoidance and non-self adhesion . SACs from Pcdhg0 retinas ( G ) express no Pcdhgs and thus do not exhibit self-avoidance . SACs from Pcdhg1 retinas ( H ) all express the same Pcdhg and thus exhibit self-avoidance but not non-self adhesion . Scale bar = 50 μm in C and D . DOI: http://dx . doi . org/10 . 7554/eLife . 08964 . 003 These morphological and molecular analyses of self-avoidance and S/NSD have led to several hypotheses about roles they might play in the function of neurons and neuronal circuits . To our knowledge , however , none of these hypotheses has been tested experimentally . Here , we report such tests , focusing on retinal starburst amacrine cells ( SACs; Figure 1C ) . These neurons have planar , radially symmetric dendritic arbors that exhibit striking self-avoidance ( Figure 1D ) , but they fasciculate and form synapses with neighboring SACs ( Lee and Zhou , 2006 ) , and thus exhibit S/NSD . SACs also provide the principal inhibitory input to ON and ON-OFF direction-selective retinal ganglion cells ( DSGCs ) and are essential for their direction selectivity ( Yoshida et al . , 2001 ) . Elegant structural and functional studies have revealed the principal elements of the underlying mechanism: individual SAC dendrites are inhibitory direction-selective subunits that wire asymmetrically to DSGCs and inhibit these ganglion cells when visual motion is presented along their proximo–distal axis ( Euler et al . , 2002; Fried et al . , 2002; Briggman et al . , 2011; Vaney et al . , 2012 ) . Thus , the preferred direction of motion for the DSGC is opposite , the preferred direction of motion for the SAC dendrites that innervate it ( Figure 1E ) . In addition , SACs form inhibitory synapses onto each other , and it has been suggested that these connections sharpen the directional preference of SAC dendrites and thus the directional preference of the DSGCs that they innervate ( Lee and Zhou , 2006; Enciso et al . , 2010; Taylor and Smith , 2012 ) . We showed recently that Pcdhgs mediate self-avoidance and S/NSD in SACs ( Lefebvre et al . , 2012 ) . Pcdhg-deficient SACs exhibit a dramatic loss of self-avoidance but maintain overlap with neighboring SACs , as if they mistake their own dendrites for those of their neighbors and fail to repel them . In contrast , forcing all SACs to express the same single Pcdhg isoform restores self-avoidance to individual cells but decreases the overlap between neighboring cells , as if they mistake dendrites of these neighbors for their own and repel them ( Figure 1F–H ) . These results lead to three specific hypotheses about circuit function: ( 1 ) in the absence of self-avoidance , SACs will form synapses with themselves ( autapses ) , ( 2 ) when S/NSD fails , SACs will form few synapses with each other , and ( 3 ) loss of self-avoidance or S/NSD will degrade the direction selectivity of DSGCs . Here , we present evidence in support of these hypotheses , thereby providing insights into the functional roles of self-avoidance and S/NSD . We also demonstrate an unexpected role of Pcdhgs in control of synapse elimination . Zheng et al . ( 2004 ) demonstrated the presence of GABAergic synapses between SACs shortly after eye-opening in rabbits . To begin this study , we confirmed that similar connections occur in young mice and asked whether they persist in adults . In addition to releasing GABA , SACs also release acetylcholine , the only retinal neuron to do so ( Hayden et al . , 1980; Famiglietti , 1983 ) , so we used a line that expresses Cre recombinase from the choline acetyltransferase locus to mark and manipulate them selectively ( Chatcre; Rossi et al . , 2011 ) . We mated Chatcre mice to lines that express Cre-dependent fluorescent reporters ( Buffelli et al . , 2003; Madisen et al . , 2010 ) , identified SACs in explants , and recorded from pairs of ON SACs shortly after eye opening ( postnatal day [P] 15–24; eye opening occurs at P14 ) and in young adults ( P40-100 ) ( Figure 2A ) . We refer to wild-type SACs as Pcdhg22 SACs , since they have their full complement of Pcdhgs . In each case , we tested pairs separated by distances varying from 35 to 175 µm; the dendritic radius of SACs in living tissue is ∼100 µm and varies little between P15 and P100 ( Figure 2—figure supplement 1A , B ) . For each pair , we stepped presynaptic SACs from a holding potential ( Vh ) of −70 mV to +20 mV while holding postsynaptic SACs at +30 mV to record inhibitory currents . In the majority of cases , we obtained bidirectional recordings; we found fewer unidirectional connections between neighboring pairs than would be expected by chance ( Figure 2—figure supplement 2D ) . 10 . 7554/eLife . 08964 . 004Figure 2 . SAC–SAC connections in Pcdhg22 and Pcdhg0 retinas . ( A ) Paired recording configuration: SACs at various intercellular distances were targeted for recording in Pcdhg22 ( left ) and Pcdhg0 ( right ) retinas . Imaged are tracings of real SACs . ( B–E ) Presynaptic voltage steps from Vh = −70 to +20 mV ( top ) and examples of currents recorded from both pre- and postsynaptic pairs of SACs that were connected ( middle ) and not connected ( bottom ) in juvenile Pcdhg22 retinas ( B ) , adult Pcdhg22 retinas ( C ) , juvenile Pcdhg0 retinas ( D ) , and adult Pcdhg0 retinas ( E ) . ( F–I ) Scatter plots of intercellular distance vs peak current size in juvenile Pcdhg22 retinas ( F ) , adult Pcdhg22 retinas ( G ) , juvenile Pcdhg0 retinas ( H ) , and adult Pcdhg0 retinas ( I ) . Number of connections tested = 34 , 35 , 37 , and 39 in F–I , respectively . ( J ) Average peak current in connected SAC pairs at P15-24 ( left ) and P40-100 ( right ) . Number of connections recorded = 21 , 9 , 23 , and 20 in juvenile Pcdhg22 retinas , adult Pcdhg22 retinas , juvenile Pcdhg0 retinas , and adult Pcdhg0 retinas , respectively . ( K ) Distance-dependence of SAC–SAC connectivity at in P15-24 animals ( left ) and P40-100 animals ( right ) . Data are shown as mean ± S . E . M . Statistics: n . s . = not significant , *p < 0 . 05 , **p < 0 . 01 . See also Figure 2—figure supplements 1–4 . DOI: http://dx . doi . org/10 . 7554/eLife . 08964 . 00410 . 7554/eLife . 08964 . 005Figure 2—figure supplement 1 . Recording distances and SAC dendritic radii . ( A ) Cumulative histogram of intercellular distances of SAC connections that were tested in juvenile animals ( left ) and adult animals ( right ) . Histogram range is 25–175 μm with bins of 10 μm . Number of connections tested = 34 , 37 , 19 , 35 , 39 , and 13 for P15-24 Pcdhg22 , P15-24 Pcdhg0 , P15-24 Pcdhg1 , P40-100 Pcdhg22 , P40-100 Pcdhg0 SACs , and P40-100 Pcdhg1 SACs , respectively . ( B ) Top: histograms of dendritic radii of Pcdhg22 , Pcdhg0 , and Pcdhg1 SACs . Bottom: average dendritic radii across conditions . These measurements are from living retinas; following fixation , staining , and mounting , the dendritic radius of SACs is ∼25% larger , as reported in many anatomical studies . ( C ) Relationship between intercellular distance and amount of dendritic overlap ( SACs modeled as 100 µm radius circles ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08964 . 00510 . 7554/eLife . 08964 . 006Figure 2—figure supplement 2 . Characterization of SAC–SAC synaptic connections . ( A ) Left: histogram of synaptic latencies in all connected cells . Histogram range is 0–20 ms after presynaptic depolarization , and bin size is 2 ms . Right: same analysis performed on non-connected pairs with histogram range of 0–50 ms and bin size 2 ms . ( B ) Two examples of SAC paired recording showing that application of 50 μM picrotoxin eliminates transmission in these pairs , and thus currents are GABAergic . Presynaptic cells were stepped from Vh = −70 to +20 mV . Postsynaptic cells were held at Vh = +30 mV . ( C ) Average current–voltage relationship of SAC–SAC connections , showing reversal at ECl ( n = 3 pairs ) . ( D ) Monte Carlo simulations of paired SAC recordings to assess the specificity of reciprocal connections . Each gray histogram was generated from 105 simulations using experimentally determined connection probabilities . Experimentally observed values are indicated by cyan arrows . Top: comparison of observed values with simulations for all reciprocal connections recorded in Pcdhg22 retinas . Unidirectional connections are significantly underrepresented in our data set . Bottom: comparison of observed values with simulations for all reciprocal connections recorded across all conditions . Unidirectional connections are significantly underrepresented in our data set , and non-connected pairs are significantly overrepresented in our data set . Statistics: n . s . = not significant , *p < 0 . 05 , **p < 0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 08964 . 00610 . 7554/eLife . 08964 . 007Figure 2—figure supplement 3 . Lamination and spacing of SACs are normal in Pcdhg0 and Pcdhg1 retinas . ( A ) Vertical section of retina stained against ChAT to label all SACs in Pcdhg22 , Pcdhg0 , and Pcdhg1 retinas ( left to right ) . Scale bar = 50 μm . ( B ) Top: En face view of OFF SACs in Pcdhg22 , Pcdhg0 , and Pcdhg1 retinas ( left to right , respectively ) . Scale bar = 100 μm . Bottom: density recovery profile of OFF SACs in Pcdhg22 , Pcdhg0 , and Pcdhg1 retinas ( left ) and total OFF SAC density . ( C ) Same as B but for ON SACs . DOI: http://dx . doi . org/10 . 7554/eLife . 08964 . 00710 . 7554/eLife . 08964 . 008Figure 2—figure supplement 4 . Normal retinal morphology in Pcdhg0 and Pcdhg1 retinas . ( A ) Nuclear label of all retinal neurons ( TO-PRO3 ) showing retinal thickness is similar across conditions . Whole retinal thickness is shown . ( B ) Anti-Chx10 immunostaining to label bipolar cells . Image is cropped to just show INL . ( C ) Anti-AP2 immunostaining to label amacrine cells . Image is cropped to just show INL through GCL . Some retinal blood vessels were also labeled because primary antibody is mouse monoclonal . ( D ) Anti-Brn3a immunostaining to label many retinal ganglion cells . Image is cropped to just show GCL . ( E ) Anti-Calbindin immunostaining to label horizontal cells , some amacrine cells , and some retinal ganglion cells . Image is cropped to just show outer plexiform layer through GCL . All images show examples from P21 Pcdhg22 , Pcdhg0 , and Pcdhg1 vertical retinal sections in parallel ( left to right , respectively ) . Scale bar = 50 μm in all panels . Images are oriented with photoreceptors towards the top and retinal ganglion cells toward the bottom of the page . DOI: http://dx . doi . org/10 . 7554/eLife . 08964 . 008 Stimulation of a SAC elicited an inhibitory current in a neighboring SAC in some but not all pairs tested at P15-24 and P40-100 ( Figure 2B , C ) . Currents occurred with a latency of ∼7 ms and averaged ∼15 pA in connected cells at both ages ( Figure 2J ) . They were blocked by 50 µM picrotoxin and reversed at the chloride reversal potential for our recording solutions ( ∼–70 mV ) , indicating that they were GABAergic and inhibitory ( Figure 2—figure supplement 2A–C ) . Although SAC–SAC connections have a cholinergic component before eye-opening in both rabbits and mice ( Zheng et al . , 2004; Ford et al . , 2012 ) , they exhibited no significant cholinergic component after eye opening ( data not shown ) . The frequency with which SACs were interconnected varied systematically with the distance between their somata and with age . At P15-24 , pairs were over twice as likely to be connected if they were separated by 35–100 µm than if they were separated by 100–175 µm ( Figure 2F , K , left ) . This difference mirrors the inverse relationship of the distance between SACs and the overlap of their dendritic arbors ( Figure 2—figure supplement 1C ) . In contrast , connections were seldom detectable between pairs separated by <100 µm in adults . The frequency of connections between pairs >100 µm apart did not change significantly with age ( Figure 2G , K , right ) , indicating that the decline did not reflect decreased ability to detect connections in older mice . The most parsimonious explanation for this difference is that synapses between closely spaced SACs are eliminated as SACs mature . Next , we asked whether Pcdhgs are required for formation of SAC–SAC synapses . For this purpose , we inactivated all 22 Pcdhgs in SACs using a conditional Pcdhg allele ( Pcdhgflox ) ( Lefebvre et al . , 2008 ) and the Chatcre line . We refer to Pcdhgflox/flox; Chatcre mice as Pcdhg0 and controls ( Pcdhgflox/+ or Pcdhg+/+; Chatcre ) as Pcdhg22 . Restricting mutation to SACs allowed us to analyze roles of Pcdhgs in SACs without the complication of directly affecting other synaptic partners . Moreover , deletion of Pcdhgs leads to excessive cell death in many retinal neuronal populations , but not in SACs ( Lefebvre et al . , 2008 , 2012 ) . As expected , we observed no alterations in the density of SACs or of other retinal cells in Pcdhg0 retinas . We further verified that the laminar position and mosaic spacing of SACs , as well as overall retinal structure , did not differ detectably between Pcdhg22 and Pcdhg0 retinas ( Figure 2—figure supplements 3 , 4 ) . At P15-24 , the number and strength of SAC–SAC connections were similar in Pcdhg22 and Pcdhg0 retinas: in both genotypes , connections were over twice as common in closely spaced pairs than in pairs separated by >100 µm and current sizes did not differ significantly between Pcdhg22 and Pcdhg0 retinas ( Figure 2D , H , J ) . Thus , Pcdhgs are dispensable for formation of SAC–SAC synapses . In adults , in contrast , the pattern of SAC–SAC connectivity differed between Pcdhg22 and Pcdhg0 mice . Synapses between closely spaced SACs were retained in mutants during the period that they were lost from controls ( Figure 2E , G , I , K ) . This loss of proximal connections was selective in that the frequency and size of connections between SACs separated by >100 µm did not differ significantly between Pcdhg22 and Pcdhg0 mice ( Figure 2J , K ) . These results reveal a requirement of Pcdhgs for synapse elimination . If Pcdhg0 SAC dendrites treat other dendrites of the same SAC as if they are dendrites of other SACs , they might form autapses . To test this hypothesis , we adapted a protocol that had been used to elicit autaptic currents in cultured neurons and cortical slices ( Bekkers and Stevens , 1991; Bacci et al . , 2003 ) . We stimulated SACs with brief voltage steps to very positive potentials ( V = +60 mV , 2–4 ms ) , then returned to more negative potentials ( V = −20 mV ) ( Figure 3A ) . We confirmed that this stimulation was able to elicit synaptic release in paired recordings ( Figure 3—figure supplement 1A ) . These stimuli elicited autaptic currents in ∼75% of Pcdhg0 SACs at P21-24 , but in no Pcdhg22 SACs ( Figure 3B , C ) . Autaptic currents resembled SAC–SAC connections in their latencies and rise times , were blockable by application of 50 µM picrotoxin , and averaged ∼20 pA in size ( Figure 3F , G and Figure 3—figure supplement 1B , C ) . We also asked whether autapses are present in adult Pcdhg0 SACs or whether , like synapses between closely spaced SACs in wild-type retina ( see previous section ) , they are progressively eliminated . Autapses persisted into adulthood in Pcdhg0 SACs with sizes and frequency similar to those observed at P21-24 ( Figure 3D–G ) . Thus , one role of Pcdhg-mediated self-avoidance is to prevent formation of autapses . 10 . 7554/eLife . 08964 . 009Figure 3 . Pcdhg0 SACs form autapses . ( A ) SAC autaptic voltage stimulus ( left ) . Single SAC recording configuration in Pcdhg22 ( middle ) and Pcdhg0 ( right ) retinas . ( B–E ) Example currents recorded from SACs in juvenile Pcdhg22 retinas ( B ) , juvenile Pcdhg0 retinas ( C ) , adult Pcdhg22 retinas ( D ) and adult Pcdhg0 retinas ( E ) in response to voltage stimulus shown in A . Arrowheads in C and E points to autaptic currents in SAC from Pcdhg0 retinas that were blocked by 50 μM picrotoxin ( blue trace in C ) . Gray bars indicate depolarization steps to +60 mV ( stimulus artifacts ) that were 2 ms long in both B , C , and D , and 4 ms long in E . The shorter latency in E likely reflects the longer depolarization step . Full traces are shown as insets with enlarged regions outlined in magenta . ( F ) Peak outward currents measured during falling phase recorded current after initial voltage step to +60 mV . Data points are staggered slightly for visual clarity . Number of SACs recorded = 8 , 8 , 7 , and 6 in juvenile Pcdhg22 retinas , juvenile Pcdhg0 retinas , adult Pcdhg22 retinas , and adult Pcdhg0 retinas , respectively . ( G ) Average peak autaptic currents evoked in SACs from Pcdhg0 retinas at P21-24 ( left ) and P40-100 ( right ) at Vh = −20 mV . Data are shown as mean ± S . E . M . See also Figure 3—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 08964 . 00910 . 7554/eLife . 08964 . 010Figure 3—figure supplement 1 . Quantification of autaptic currents . ( A ) Example of paired recording using ‘autaptic’ voltage stimulus , showing that brief depolarization can evoke transmission in pairs of neurons . ( B ) Top: example current recordings of Pcdhg22 ( left ) and Pcdhg0 SACs in response to autaptic voltage stimulus . Bottom: insets from top panel showing averaged responses ( black ) and 4 raw traces in each condition ( gray ) . ( C ) Comparison of synaptic and autaptic properties in SACs showing similarity in latency , rise time , and picrotoxin sensitivity . Data are shown as mean ± S . D . DOI: http://dx . doi . org/10 . 7554/eLife . 08964 . 010 The proposed mechanism for Pcdhg-dependent S/NSD is that the stochastic expression of a subset of Pcdhg isoforms endows each SAC with a unique molecular identity that circumvents Pcdhg-dependent avoidance , allowing neighboring SACs to interact ( Lefebvre et al . , 2012 ) . We hypothesized that if all SACs expressed the same Pcdhg isoform , they would treat dendrites of other SACs as if they were other dendrites of the same SAC , and form few SAC–SAC synapses . To test this idea , we used a mouse line in which a single Pcdhg isoform ( PcdhgC3 ) can be expressed in any cell type in a Cre-dependent manner ( Lefebvre et al . , 2012 ) . We call mice in which SACs expressed only this isoform Pcdhg1 ( Rosa-CAGS-lox-stop-lox-PcdhgC3-mCherry; Chatcre; Pcdhgflox/flox ) . The overall morphology , number , and spacing of SACs , as well as overall retinal structure , were normal in Pcdhg1 mice ( Figure 4A and Figure 2—figure supplements 3 , 4 ) , and SAC dendrites formed a fine plexus within which , despite a decrease in overlap between pairs of neurons ( Lefebvre et al . , 2012 ) , they had ample opportunity to come into close proximity to each other ( Figure 4B ) . We made paired recordings from SACs in Pcdhg1 mice at P15-24 using methods described in Figure 2 ( Figure 4C , D ) . The frequency of SAC–SAC connections in Pcdhg1 mice was ∼20% of that in Pcdhg22 or Pcdhg0 mice ( Figure 4F ) . Similarly , current sizes in connected pairs in Pcdhg1 mice were on average ∼40% of those recorded in Pcdhg22 or Pcdhg0 mice ( Figure 4H ) . Thus , forcing expression of the same Pcdhg isoform in all SACs decreased their connection strength to <10% ( 0 . 2 × 0 . 4 ) of controls . A similar decrease was observed in adult Pcdhg1 retinas ( Figure 4E , G , H ) . We conclude that Pcdhg diversity is required for functional connectivity between neighboring SACs . 10 . 7554/eLife . 08964 . 011Figure 4 . Decreased SAC–SAC connections in Pcdhg1 retina . ( A ) Replacement of all 22 Pcdhgs in SACs with a single Pcdhg isoform ( top ) rescues self-avoidance in individual SACs ( bottom ) . ( B ) Plexus of all SAC dendrites ( stained with anti-ChAT ) in Pcdhg22 ( left ) , Pcdhg0 ( middle ) , and Pcdhg1 ( right ) retinas . ( C ) Presynaptic voltage steps from Vh = −70 to +20 mV ( top ) and examples of currents recorded from both pre- and postsynaptic pairs of SACs that were connected ( middle ) and not connected ( bottom ) in juvenile Pcdhg1 retinas . ( D–E ) Scatter plots of intercellular distance vs peak current size in juvenile ( D ) and adult ( E ) Pcdhg1 retinas . ( F ) Percent of P15-24 recorded SAC pairs that were connected , irrespective of intercellular distance . Number of connections tested = 34 , 37 , and 19 in Pcdhg22 , Pcdhg0 , and Pcdhg1 retinas , respectively . ( G ) Same as F for adult retinas . Number of connections tested = 35 , 39 , and 13 in Pcdhg22 , Pcdhg0 , and Pcdhg1 retinas , respectively . ( H ) Average peak current in connected SAC pairs at all ages . Number of recorded connections = 30 , 43 , and 3 in Pcdhg22 , Pcdhg0 , and Pcdhg1 retinas , respectively . Scale bar = 50 μm in A and 25 μm in B . Data are shown as mean ± S . E . M . Statistics: **p < 0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 08964 . 011 Having found that manipulation of Pcdhg expression affects the ability of SACs to form synapses on their own dendrites or those of other SACs , we asked whether such manipulations affect their ability to receive synapses from bipolar cells or form synapses onto DSGCs . We used visual stimuli based on previous findings that the main visually-evoked excitatory input to SACs is from bipolar cells , and that SACs provide the main inhibitory input to DSGCs ( Figure 5A ) ( Taylor and Wassle , 1995; Vaney et al . , 2012; Helmstaedter et al . , 2013; Hoggarth et al . , 2015 ) . 10 . 7554/eLife . 08964 . 012Figure 5 . Integration of SACs into a direction-selective circuit is Pcdhg-independent . ( A ) Schematic of excitatory and inhibitory synaptic inputs of retinal direction-selective circuit , showing bipolar inputs to SACs ( measured in B and C ) , SAC inputs to DSGCs ( measured in D and E ) , and bipolar inputs to DSGCs ( measured in F and G ) . ( B ) Example excitatory currents ( Vh = −70 mV ) of ON SACs from Pcdhg22 ( black ) , Pcdhg0 ( gray ) , and Pcdhg1 ( red ) retinas evoked by a bright spot flash . ( C ) Average peak current responses to the onset of flash stimulus . Number of SACs recorded is 8 , 9 , and 7 in Pcdhg22 , Pcdhg0 , and Pcdhg1 retinas , respectively . ( D ) Example inhibitory currents ( Vh = 0 mV ) of vDSGCs from Pcdhg22 , Pcdhg0 , and Pcdhg1 retinas evoked by the onset ( left ) and offset ( right ) of a bright spot flash . ( E ) Average peak current responses to the onset ( left ) and offset ( right ) of flash stimulus . Number of vDSGCs recorded is 12 , 13 , and 10 in Pcdhg22 , Pcdhg0 , and Pcdhg1 retinas , respectively . ( F ) Example excitatory currents ( Vh = −70 mV ) of vDSGCs from Pcdhg22 , Pcdhg0 , and Pcdhg1 retinas evoked by the onset ( left ) and offset ( right ) of a 2 s bright spot flash . ( G ) Average peak current responses to the onset ( left ) and offset ( right ) of flash stimulus . Number of vDSGCs recorded is 14 , 11 , and 13 in Pcdhg22 , Pcdhg0 , and Pcdhg1 retinas , respectively . ( H–J ) Dye-filled vDSGCs with OFF and ON arborizations separated ( top and middle , respectively ) in Pcdhg22 ( H ) , Pcdhg0 ( I ) , and Pcdhg1 ( J ) retinas . Bottom panels: Overlay of ON vDSGC dendrites ( green ) with ON SAC dendrites labeled with anti-ChAT antibody ( red ) . Similar co-fasciculation was seen for OFF dendrites . Scale bar = 50 μm . Data are shown as mean ± S . E . M . Spot flashes were displayed for 2 s in each case . See also Figure 5—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 08964 . 01210 . 7554/eLife . 08964 . 013Figure 5—figure supplement 1 . Normal expression , spacing , and number of vDSGCs in Pcdhg0 and Pcdhg1 retinas . ( A ) Anti-GFP immunostaining in HB9-GFP positive retinas . Note that GFP signal is faint in dendrites by P21 but retained in cell body . Image is cropped to show INL through GCL . ( B ) Anti-CART immunostaining in same sections from A to label all populations of DSGCs . ( C ) Merge of panels A and B showing that HB9-GFP positive vDSGCs are positive for CART but not the only CART positive cells in the GCL . Note that CART antibody also labels Tyrosine hydroxylase-positive amacrine cells strongly in IPL sublamina 1 . ( D ) Top: En face view of field of HB9-positive vDSGCs ( inverted contrast ) . Bottom: density recovery profile of HB9-GFP positive vDSGCs in Pcdhg22 , Pcdhg0 , and Pcdhg1 retinas ( left ) and total HB9-GFP positive vDSGC density . All images show examples from P21 Pcdhg22 , Pcdhg0 , and Pcdhg1 retinas in parallel ( left to right , respectively ) . Scale bar = 50 μm in panels A–C and 100 μm in panel D . Images in panels A–C are vertical retinal sections oriented such that photoreceptors are towards the top of the page . DOI: http://dx . doi . org/10 . 7554/eLife . 08964 . 013 To assess bipolar input to SACs , we recorded from ON SACs while holding the cells at Vh = −70 mV and presenting bright spot flashes centered on the soma of the recorded cell . SACs received strong excitatory inputs in Pcdhg22 , Pcdhg0 , and Pcdhg1 mice , with no significant differences among them ( Figure 5B , C ) . Four populations of ON-OFF DSGCs have been described , each tuned to one of the cardinal directions: dorsal , ventral , nasal , and temporal ( Barlow and Hill , 1963; Oyster and Barlow , 1967; Elstrott et al . , 2008 ) . Their physiological properties other than preferred direction are similar , but they exhibit molecular differences that allow them to be marked selectively ( Kay et al . , 2011 ) . To assess SAC input to DSGCs , we used a transgenic line in which DSGCs that prefer motion in the ventral direction express GFP ( HB9-GFP; Trenholm et al . , 2011 ) . We introduced this transgene into the Pcdhg22 , Pcdhg0 , and Pcdhg1 backgrounds , and recorded inhibitory currents ( Vh = 0 mV ) from GFP-labeled DSGCs , which we call vDSGCs here . Sizes of both ON and OFF inhibitory responses to spot flashes were indistinguishable across the three genotypes ( Figure 5D , E ) . Similarly , excitatory spot flash responses in vDSGCs were unaffected by manipulation of Pcdhgs in SACs ( Figure 5F , G ) . Thus , Pcdhg expression in SACs is dispensable for their ability to form and maintain synapses with other cell types . We also assessed the structure of vDSGCs in Pcdhg22 , Pcdhg0 , and Pcdhg1 mice . We filled single cells with fluorescent dye , stained SACs with antibodies to ChAT , and imaged the two cell types . In all conditions , the ON and OFF dendrites of these DSGCs stratified in the ON and OFF SAC plexus , fasciculated with SAC dendrites , and maintained their modest structural asymmetry ( Figure 5H–J ) . Thus , altering Pcdhg expression in SACs had no detectable effect on the morphology of vDSGCs . We also validated that altering Pcdhg expression in SACs did not affect cell number , mosaic spacing , or expression patterns of vDSGCs ( Figure 5—figure supplement 1 ) . We next tested the hypothesis that loss of self-avoidance or S/NSD degrades the information-processing ability of SACs within the direction-selective circuit . To this end , we recorded spikes from vDSGCs while moving a bright bar over their receptive field in 8 different directions . Because vDSGCs are all tuned to a single direction in wild-type mice , we were able to ask whether manipulation of Pcdhgs affects preferred direction as well as the degree of direction selectivity . vDSGCs in Pcdhg22 mice exhibited strong ON and OFF directional responses ( Figure 6A ) as shown previously ( Kim et al . , 2010; Trenholm et al . , 2011; Duan et al . , 2014 ) . We calculated a direction-selective index ( DSI ) for each vDSGC by computing the vector sum of the responses to different directions ( Kim et al . , 2008 ) and calculated both the magnitude of directional responses and the angle of preference ( Figure 6B ) . Direction selectivity of vDSGCs was diminished in both Pcdhg0 and Pcdhg1 retinas but in different ways . In both genotypes , the average magnitude of the DSI vector was significantly decreased with respect to controls ( by ∼50% in Pcdhg0 and ∼35% in Pcdhg1; Figure 6C–G ) . In contrast , responses of vDSGCs in Pcdhg0 retinas exhibited a significantly greater scatter around the ventral axis than those in wild-type retinas , whereas vDSGCs in Pcdhg1 retinas were as precisely tuned to ventral motion as controls ( Figure 6C–F , H and Figure 6—figure supplement 1 ) . This variance likely reflects the contorted morphology of SAC dendrites in Pcdhg0 but not Pcdhg1 retinas . Likewise , the variation between the preferred direction of ON and OFF responses was greater in Pcdhg0 retinas than in either Pcdhg22 or Pcdhg1 retinas , indicating that SAC morphology and connectivity are disrupted independently in the ON and OFF SAC layers ( Figure 6I ) . 10 . 7554/eLife . 08964 . 014Figure 6 . Alteration of Pcdhg expression degrades direction selectivity . ( A ) Spiking responses of vDSGC from adult Pcdhg22 retina to a bright moving bar moving in 8 directions . Polar plot is of peak firing rates in response to bar entering ( ON , green ) and exiting ( OFF , blue ) the receptive field center . Vectors represent vector sum direction-selective indices ( DSIs ) of ON and OFF responses . Surrounding central plots are spike histograms used to make polar plot and calculate DSIs and preferred directions . ( B ) ON ( left , green ) and OFF ( right , blue ) DSI vectors for all recorded DSGCs in Pcdhg22 retina ( n = 28 cells ) . Axes of retina are indicated with compass arrows: D , V , N , and T represent dorsal , ventral , nasal , and temporal . ( C , D ) Same as A and B but from adult Pcdhg0 retinas ( n = 28 cells ) . ( E , F ) Same as A and B but from adult Pcdhg1 retinas ( n = 19 cells ) . ( G ) Mean absolute DSI for all cells recorded , irrespective of which direction they preferred . ( H ) Mean angle deviated from ventral direction for all cells recorded . ( I ) Mean absolute difference between DSI ( left ) and angle of preference ( right ) for all recorded cells . ( J ) Plot of mean ventral projections of DSI vectors . For each recorded vDSGC in J , maximal ON and OFF firing rates in each direction were summed and used to generate a single DSI vector for each cell . Data are shown as mean ± S . E . M . Statistics: n . s . = not significant , *p < 0 . 05 , **p < 0 . 01 , ***p < 0 . 001 . See also Figure 6—figure supplements 1 , 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 08964 . 01410 . 7554/eLife . 08964 . 015Figure 6—figure supplement 1 . ON and OFF direction responses of vDSGCs are similarly blunted when Pcdhg expression in SACs is altered . ( A ) Cumulative histogram of ON and OFF direction-selective indices ( top ) and angle away from ventral ( bottom ) for all recorded vDSGCs in Pcdhg22 , Pcdhg0 , and Pcdhg1 retinas ( n = 28 , 28 , and 19 , respectively ) . Histogram bins are 0 . 05 DSI units and 10° for top and bottom panels , respectively . ( B ) Ratios of excitatory ( left ) and inhibitory ( right ) ON and OFF current sizes ( ventral over dorsal motion ) for all recorded cells in Pcdhg22 , Pcdhg0 , and Pcdhg1 retinas ( Excitatory n = 14 , 10 , and 10 ON and OFF each , respectively; Inhibitory n = 14 , 10 , and 13 ON and OFF each , respectively ) . ( C ) Relative timing of ON and OFF excitation compared to inhibition during ventral motion ( left ) and dorsal motion ( right ) for all recorded cells in Pcdhg22 , Pcdhg0 , and Pcdhg1 retinas ( n = 12 , 8 , and 8 ON and OFF each , respectively ) . ON and OFF responses are shown in green and blue , respectively . Statistics: n . s . = not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 08964 . 01510 . 7554/eLife . 08964 . 016Figure 6—figure supplement 2 . Age-dependent improvement in direction selectivity of vDSGCs requires Pcdhgs . ( A ) Comparison of DSI of juvenile and adult vDSGCs in Pcdhg22 and Pcdhg0 retinas . ( B ) Comparison of deviations from ventral of juvenile and adult vDSGCs Pcdhg22 and Pcdhg0 retinas . Statistics: n . s . = not significant , *p < 0 . 05 , **p < 0 . 0005 . n = 10 and 6 for P15-24 Pcdhg22 and Pcdhg0 vDSGCs , respectively , and n = 28 and 28 for adult Pcdhg22 and Pcdhg0 vDSGCs , respectively . Leading edge ( ON ) and trailing edge ( OFF ) responses were measured for all cells and used as independent data points . DOI: http://dx . doi . org/10 . 7554/eLife . 08964 . 016 To obtain a single measure of how well vDSGCs reported on ventral motion , we projected the directional vectors onto the ventral axis . This gave us a ventral DSI that combined the degree of directional preference and the fidelity of ventral preference for ON and OFF responses together . vDSGCs in Pcdhg22 ( control ) retinas were most ventrally selective , followed by those in Pcdhg1 retinas; vDSGCs in Pcdhg0 retinas were the least selective ( Figure 6J ) . Together , these results demonstrate that manipulating Pcdhg expression in SACs , and thereby attenuating self-avoidance or S/NSD , degrades the direction selectivity of DSGCs . Recently , Sun et al . ( 2013 ) also showed that morphological alterations of SACs disrupt directional tuning of DSGCs . Previous studies have shown that direction-selective responses are already present at eye opening in mice but become more selective with age ( Elstrott et al . , 2008; Yonehara et al . , 2011; Chan and Chiao , 2013; Chen et al . , 2014 ) . We wondered whether this improvement of direction selectivity with age was related to the loss of proximal SAC–SAC connections . To assess this possibility , we recorded from direction-selective responses from vDSGCs at P15-24 in Pcdhg0 mice , which do not go through a developmental change in SAC–SAC connectivity . We confirmed the improved age-dependent directional tuning of DSGCs in control retinas . In contrast , direction selectivity of vDSGCs did not improve in Pcdhg0 retinas ( Figure 6—figure supplement 2 ) . This result is consistent the idea that developmental refinement in SAC–SAC connectivity contributes to age-dependent improvement in direction selectivity . Finally , we sought to explain the degradation of directional selectivity of vDSGCs in Pcdhg0 , and Pcdhg1 retinas ( Figure 6 ) in terms of alterations in SAC connectivity ( Figures 2–4 ) . To this end , we recorded inhibitory and excitatory currents from vDSGCs in the three genotypes in response to bars moving in the null and preferred directions ( dorsal and ventral , respectively ) . As noted previously , the inhibitory currents arise predominantly from SACs , which are genetically altered in mutants , while the excitatory currents arise predominantly from bipolar cells , which are not altered . Studies in mice and rabbits have revealed two key aspects of SAC–DSGC connectivity that lead to direction selectivity ( Fried et al . , 2002; Taylor and Vaney , 2002; Vaney et al . , 2012; Yonehara et al . , 2013; Park et al . , 2014 ) , both of which we confirmed in vDSGCs from Pcdhg22 retinas . First , inhibitory input to DSGCs is greater for movement in the null direction ( dorsal for vDSGCs ) than for movement in the preferred direction ( ventral for vDSGCs ) , whereas excitatory input is similar for movement in both directions ( Figure 7A , J , K ) . Second , excitatory and inhibitory currents recorded from DSGCs arise at the same time when motion is in the null direction , whereas inhibitory currents lag with respect to excitatory currents when motion is in the preferred direction ( Figure 7B , L , M ) . Together , these features allow inhibition to veto excitation in DSGCs more strongly for null motion than for preferred motion . Consequently , net depolarization in DSGCs is largest for motion in the preferred direction . 10 . 7554/eLife . 08964 . 017Figure 7 . Synaptic basis of degraded direction selectivity in Pcdhg0 and Pcdhg1 retinas . ( A ) Example excitatory ( black , Vh = −70 mV ) and inhibitory ( gray , Vh = 0 mV ) currents evoked by leading edge ( ON response ) of bar moving in ventral ( left ) and dorsal ( right ) directions in vDSGC from Pcdhg22 retina . ( B ) Examples of relative timing of excitation and inhibition in same cell from panel A . ( C ) Schematic of inhibitory input to vDSGCs in Pcdhg22 retinas . vDSGCs receive inhibitory input from SAC dendrites with predominately dorsal orientations and directional preferences , setting the null direction of vDSGCs . These SAC dendrites , in turn , receive inhibitory input from SAC dendrites with predominately ventral orientation and preference , suppressing inhibition to vDSGCs during ventral motion through inhibition of inhibition . ( D , E ) Same as A , B but in Pcdhg0 retina . ( F ) Schematic of inhibitory input to vDSGCs in Pcdhg0 retinas . vDSGCs receive inhibitory input from curvilinear SAC dendrites with disrupted orientations and directional preferences , diminishing their ability to set the null direction of vDSGCs . These SAC dendrites , in turn , receive inhibitory input from both parallel and antiparallel SAC dendrites . ( G , H ) Same as A , B but for trailing edge ( OFF response ) in Pcdhg1 retina . ( I ) Schematic of inhibitory input to vDSGCs in Pcdhg1 retinas . vDSGCs receive inhibitory input from SAC dendrites with predominately dorsal orientations and directional preferences , setting the null direction of vDSGCs . These SAC dendrites , however , are no longer inhibited by SAC dendrites with predominately ventral orientation and preference , so their input to vDSGCs during ventral motion is not suppressed . ( J ) Ratio of peak excitatory current sizes evoked in vDSGCs by ventral vs dorsal motion in Pcdhg22 ( black ) , Pcdhg0 ( gray ) , and Pcdhg1 ( red ) retinas . ( K ) Same as J but for inhibitory currents . ( L ) Relative timing of onset of excitation compared to inhibition during ventral motion in Pcdhg22 ( black ) , Pcdhg0 ( gray ) , and Pcdhg1 ( red ) retinas . ( M ) Same as L but during dorsal motion . Data are shown as mean ± S . E . M . Number of recorded vDSGCs = 14 , 10 , and 13 in Pcdhg22 , Pcdhg0 , and Pcdhg1 retinas . Leading edge ( ON ) and trailing edge ( OFF ) responses were measured for all cells and used as independent data points for quantification . Statistics: n . s . = not significant , *p < 0 . 05 , **p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 08964 . 017 We found that both of these contributors to direction selectivity were blunted in Pcdhg0 and Pcdhg1 retinas ( Figure 7D , E , G , H and Figure 6—figure supplement 1 ) . First , inhibitory currents were larger for ventral motion and smaller for dorsal motion in Pcdhg0 and Pcdhg1 retinas than in Pcdhg22 retinas , with no significant change in excitation ( Figure 7J , K ) . The difference from control values was greater for Pcdhg0 than for Pcdhg1 retinas but significant in both . Second , the delay of inhibition in response to preferred motion was less in Pcdhg0 and Pcdhg1 retinas than in Pcdhg22 retinas , with no significant change for movement in the null direction; in this case , Pcdhg0 and Pcdhg1 retinas were equally affected ( Figure 7L , M ) . Thus , the ability of inhibition to veto excitation for preferred motion was greater in Pcdhg0 and Pcdhg1 retinas than in Pcdhg22 retinas . It is likely that the differences in the size and timing of inhibitory currents in vDSGCs from Pcdhg0 and Pcdhg1 retinas result in the changes in spiking observed in Figure 6 . In the Discussion , we suggest a possible explanation for these alterations in terms of perturbations in SAC self-avoidance , S/NSD , and synapse elimination . Loss of Pcdhgs has been shown to have several effects on developing neurons including decreased neuronal survival in retina and spinal cord , decreased synaptic maintenance in spinal cord , decreased dendritic branching in neocortex , and decreased self-avoidance in retina and cerebellum ( Wang et al . , 2002b; Weiner et al . , 2005; Lefebvre et al . , 2008 , 2012; Prasad et al . , 2008; Garrett et al . , 2012 ) . Any of these phenotypes would complicate our attempt to assess roles of self-avoidance and S/NSD in SAC and circuit function . We therefore manipulated Pcdhg expression selectively in SACs and performed a variety of control experiments to assess whether our manipulations affected other aspects of retinal development or function . Our results are as follows: First , SACs are unusual among retinal neurons in that their survival does not depend on Pcdhg expression ( Lefebvre et al . , 2012 ) , and we confirmed that SAC number was unaltered in Pcdhg0 and Pcdhg1 retina . Second , we confirmed ( Lefebvre et al . , 2012 ) that alteration of Pcdhg expression in SACs had no effect on their dendritic length or mosaic spacing . Third , basic electrical properties ( resting membrane potential and input resistance ) of SACs were preserved in Pcdhg0 and Pcdhg1 retinas ( data not shown ) . Fourth , manipulation of Pcdhgs in SACs had no detectable effect on the strength of the inputs they receive from bipolar cells or deliver to DSGCs . Fifth , removing or replacing Pcdhgs in SACs had no detectable effect on cell number or general organization of the retina . Finally , we detected no alteration in the number , spacing , dendritic arbors , molecular markers or electrical properties of ventrally-preferring DSGCs . Thus , although we cannot completely exclude the possibility that Pcdhg manipulation had additional effects , we favor the explanation that alterations in SAC connectivity and circuit function documented here result from perturbation of Pcdhg-dependent self-avoidance , S/NSD , and synapse elimination . Morphological studies led to the idea that self-avoidance serves to optimize coverage of a receptive field by a dendritic arbor , minimizing gaps , and clumps ( Kramer and Kuwada , 1983; Kramer and Stent , 1985; Grueber and Sagasti , 2010 ) . Our physiological studies revealed an additional role of self-avoidance in SACs: it prevents formation of autapses ( Figure 3 ) . In many neuronal types , autapses cannot form because pre- and postsynaptic machinery are confined to axons and dendrites , respectively , which are physically segregated . SACs , in contrast , form dendro-dendritic synapses , and therefore have pre- and postsynaptic specializations intermingled . This situation is not uncommon in the retina and elsewhere in the central nervous system , such as the olfactory bulb ( Murthy , 2011 ) . We suggest that self-avoidance may play similar roles in other such cells . S/NSD is generally viewed as a means of limiting inter-dendritic repulsion to sibling processes , so that neurons of a single type can share territory ( Grueber and Sagasti , 2010; Zipursky and Grueber , 2013; Lefebvre et al . , 2015 ) . In the retina , it additionally allows formation of synapses between SACs . Several types of neurons have been shown to form homotypic connections in cortex and cerebellum ( Pfeffer et al . , 2013; Rieubland et al . , 2014 ) . Since most molecules described to date that mediate self-avoidance are ill-suited to mediate S/NSD , additional mechanisms likely remain to be discovered . In addition , some cell types that connect homotypically may lack robust mechanisms for self-avoidance . Indeed , cortical fast-spiking interneurons , which form homotypic connections , also form autapses ( Bacci et al . , 2003 ) . It is unclear whether these autapses are beneficial to the circuit or whether they are an acceptable cost of homotypic connectivity . Zhou and colleagues previously demonstrated inhibitory SAC–SAC synaptic connections in rabbit retina soon after eye opening , a result we confirmed here for mouse ( Zheng et al . , 2004; Lee and Zhou , 2006 ) . We also discovered two additional features of these connections . First , in mature retina ( >P40 ) , SACs separated by less than 100 µm seldom formed synapses with each other , whereas SACs separated by > 100 µm were connected frequently . Since dendritic overlap is inversely proportional to the distance between SACs , this distance-dependence is not a passive consequence of proximity but instead implies spatial selectivity to SAC–SAC connections . Second , we found that this distance-dependence was absent in immature retinas ( P15-24; eye opening occurs at P14 ) . Thus , connections between closely spaced SACs are selectively lost as the retina matures . We view the loss of proximal SAC–SAC connections as synapse elimination , a process that occurs in many and perhaps most neuronal types ( Kano and Hashimoto , 2009 ) but has not previously been described for SACs . In most cases , synapse elimination was first described physiologically ( Redfern , 1970; Crepel et al . , 1976; Purves and Lichtman , 1980 ) as we have done here . For these cases , morphological confirmation was obtained many years later . We expect this will be the case for SACs as well . Such demonstration will be difficult , however , because SAC dendrites are so thin and densely packed that it is infeasible to map synapses on them by light microscopic methods . Ultrastructural studies using genetic tags or extensive reconstruction at several developmental time points will therefore be needed to decide this issue . Why might connections between closely spaced SACs be counterproductive ? Inhibitory connections between nearby SACs would frequently be made between dendrites with similar directional preferences . The ability of a SAC dendrite to respond to centrifugal motion along its dendrite would thereby decrease , because this motion would lead to inhibition of the dendrite by other SACs . This , in turn , would degrade the direction selectivity of DSGCs ( Taylor and Smith , 2012 ) . In contrast , connections of distant SACs will most frequently be made between dendrites with opposite directional preference; as discussed in the next section , this enhances directional computation . Conversely , might there be a role for connections between closely spaced SACs early in development ? In fact , strong SAC–SAC connectivity is critical for the developing visual system , because it underlies propagation of the retinal waves that pattern the segregation of binocular input in retinorecipient areas such as the superior colliculus and lateral geniculate nucleus ( Ford et al . , 2012; Ackman and Crair , 2014; Burbridge et al . , 2014 ) . Because waves occur before eye-opening , directional selectivity is unimportant . Thus , we suggest that postponing distance-dependent elimination of SAC–SAC connections until after eye-opening allows both the dense connectivity needed for wave propagation and the selective anti-parallel connectivity needed for direction selectivity . Consistent with this view , the direction selectivity of DSGCs increases during the period in which connections between closely spaced SACs are being eliminated . We also found that connections between closely spaced SACs are not eliminated in the absence of Pcdhgs , revealing a novel role for these molecules in neural development . The mechanism of this effect remains to be determined . One attractive possibility is that an uneven distribution of Pcdhgs within SACs might confine synapses to distal portions of dendrites . We have argued that alterations in SAC connectivity in Pcdhg0 and Pcdhg1 retinas documented in the first part of this study ( Figures 2–4 ) result from defects in self-avoidance , S/NSD , and synapse elimination . We now argue that these defects largely explain the degradation in direction selectivity in vDSGCs documented in the second part ( Figures 6 , 7 ) . As described above , SACs contribute to the direction selectivity of DSCGs in two ways . First , inhibitory currents are larger during null motion than preferred motion . The difference in inhibitory currents arises in large part from the geometric arrangement of SAC–DSGC connections: vDGSCs , for example , receive most SAC input from dendrites that respond preferentially to dorsal ( null ) motion ( Briggman et al . , 2011 ) . In addition , anti-parallel inhibitory connections between SACs decrease the currents that these dendrites would otherwise provide during preferred motion ( Figure 7C ) . Together , these processes result in greater net depolarization and therefore spiking for preferred motion than null motion . The number of SAC–SAC connections is markedly decreased in Pcdhg1 retinas ( Figure 7I ) . These connections persist in Pcdhg0 retinas , but their efficacy is decreased because parallel SAC dendrites remain connected and inhibit each other , resulting in decreased inhibitory input from SACs to DSCGs for null motion and decreased antiparallel SAC–SAC inhibition ( and thus increased SAC–DSCG inhibition ) for preferred motion ( Figure 7F ) . The autapses in Pcdhg0 retina would act similarly to synapses between parallel dendrites , since autapsing dendrites are likely to point in similar directions ( see Pcdhg0 SAC image in Figure 3 ) . Second , inhibitory and excitatory currents in DSGCs are nearly simultaneous during null motion , allowing inhibition to veto excitation , whereas inhibition is delayed with respect to excitation during preferred motion , decreasing the power of the veto . A recent computational model argues that the delayed inhibition for preferred motion arises in part because anti-parallel SAC–SAC connections transiently suppress transmitter release from SACs to DSGCs ( Taylor and Smith , 2012 ) . Decreased inhibition , from loss of SAC–SAC connections in Pcdhg1 retinas and decreased efficacy of SAC–SAC synapses Pcdhg0 retinas , would thus be expected to decrease the delay , thereby blunting the response to preferred motion . In summary , the spatial organization of SAC–SAC inhibition and SAC–DSGC inhibition work together to generate a direction-selective output from the retina . When self-avoidance , S/NSD , or synapse elimination is perturbed , SAC–SAC inhibition is rendered less effective and direction selectivity is degraded . Thus , our results demonstrate roles for these Pcdhg-dependent processes in computation of direction selectivity and provide new evidence in support of the hypothesis ( Lee and Zhou , 2006; Enciso et al . , 2010; Taylor and Smith , 2012; Vaney et al . , 2012 ) that SAC–SAC connections play important roles in this computation . Animals were used in accordance with NIH guidelines and protocols approved by Institutional Animal Use and Care Committee at Harvard University . All mice were maintained on a C57BL/6 background . The lines used were reported previously: Pcdhgfcon3 ( Lefebvre et al . , 2008; Prasad et al . , 2008; Lefebvre et al . , 2012 ) , ChatCre ( Rossi et al . , 2011 ) , Thy1-stop-YFP line #15 ( Buffelli et al . , 2003 ) , Mnx1::eGFP ( here called HB9-GFP ) ( Wichterle et al . , 2002; Trenholm et al . , 2011 ) , RC-stop-tdTomato ( Madisen et al . , 2010 ) , and RC-stop-PcdhgC3-mCherry ( Lefebvre et al . , 2012 ) . We generally used ChatCre mice as homozygotes , because we found that this gave earlier and more even Cre activity at P1 , when SAC dendrites are beginning to elaborate . Mice were dark adapted for at least 2 hr prior to euthanasia . Retinas were rapidly dissected under infrared illumination into room temperature , oxygenated ( 95% O2 , 5% CO2 ) Ames medium and placed in a recording chamber on the stage of a custom built electrophysiology set up . Recordings were carried out in the same medium heated to 32–34°C . Fluorescent cells were identified with a brief ( <40 ms ) LED flash , overlaid onto infrared images , and targeted with electrodes . Recordings were made from SACs and vDSGCs using patch electrodes with resistance of 6–8 MΩ and 4–6 MΩ , respectively . For loose patch recordings , electrodes were filled with Ames medium . For intracellular recordings , electrodes were filled with intracellular solution containing the following ( in mM ) : 120 Cs-Methanesulfonate , 10 Na-Acetate , 0 . 2 CaCl2 , 1 MgCl2 , 10 EGTA , 5 CsCl , 2 Mg-GTP , and 0 . 5 Na2-GTP ( pH 7 . 3 ) . Intracellular recording solutions were supplemented with 5 mM QX314-Br for vDSGC voltage clamp recordings and 5 mM TEA-Br for SAC autapse recordings . Paired connections were tested with 200 ms voltage steps from Vh = −70 mV to +20 mV in presynaptic SACs while postsynaptic SACs were held at +30 mV for all current size measurements and at potentials between −70 and +30 in 20 mV increments to establish I–V relationships . Approximately 10 voltage step repetitions were acquired for each pre-post pair and bidirectional measurements were made if recordings were sufficiently stable . Cells were analyzed in a semi-automated fashion and deemed connected using the following criteria: ( 1 ) Average traces had a peak in the first 30 ms after presynaptic stimulus onset that was >2 standard deviations from the baseline established in the 50 ms before stimulus onset , ( 2 ) current deflection was present in ≥ 80% of trials , ( 3 ) peak current had short latency ( <12 ms ) and fast rise time ( 10–90% rise time <4 ms ) . Each recording was checked after the fact for large baseline deviations or other anomalous signals . Autapse recordings were evoked using a brief voltage step from Vh = −70 to +60 mV ( 2–4 ms ) followed by a return to −20 mV . This stimulus activated some intrinsic currents in SAC that decayed in <100 ms . During this decay phase , a large fraction Pcdhg0 SACs exhibited outward currents with synaptic latencies , rise times , and amplitudes that were blockable by the addition of 50 µM picrotoxin and thus autaptic currents . To analyze these recordings , we ( 1 ) fit the first 30 ms of each trace after returning to our holding potential of −20 mV with a double exponential curve , ( 2 ) looked for residuals of the fit >2 standard deviations of the pre-stimulus baseline in order to identify SACs that potentially made autapses , and ( 3 ) applied criteria used to find connected SAC pairs . We could not make reliable measurements of autapses in SACs from retinas younger than P21 due to large inward calcium currents evoked by depolarization . These currents were also apparent at even younger ages ( P8 ) and may therefore be residually present from the ages at which SACs initiate and propagate retinal waves ( see ‘Discussion’ ) . In loose patch spike recordings , action potentials were detected and analyzed using a simple thresholding criterion in MATLAB ( Mathworks , Natick , MA ) . Spike histograms were made with 50 ms bins and used to find peak firing rates . DSIs and preferred directions of individual cells were calculated using the maximal firing rates elicited by moving visual stimuli in 8 directions ( θ = 0o:45o:315o ) and vector sums were calculated as in Kim et al . ( 2008 ) . Light stimuli were presented using a modified DLP projector ( Dell , Round Rock , TX ) suspended underneath the microscope stage with a custom substage lens system focused onto the retinal photoreceptors . Monochrome light was used ( wavelength peak = 405 nm ) at a background intensity 5 × 102 R*/rod/s set using neutral density filters . Visual stimuli were presented at 100:1 positive contrast and patterns generated using Psychophysics Toolbox in MATLAB and are available as Source code 1 . All stimuli were centered on the cell body of recorded neurons . Spot flash stimuli were 300 μm-diameter circles . Moving bars were 1000–1500 μm long and 300 μm wide , traveled at 1000 μm/s , presented moving along their long axis in 8 directions , and rotated by 135° with each presentation . At the speeds we used for our visual stimuli , nonlinear dendritic processes contributing to directional tuning are not observed in HB9-GFP vDSGCs ( Trenholm et al . , 2011 ) . A minimum of 4 repetitions were presented for each stimulus . Electrophysiological recordings were acquired using a Multiclamp 700B Amplifier ( Axon Instruments , Molecular Devices , Sunnyvale , CA ) at 20 kHz . Acquisition was controlled by custom LabView software ( National Instruments , Austin , TX ) and is available as Source code 2 . Data were analyzed using custom written MATLAB software available as Source code 3 and displayed in IgorPro ( Wavemetrics , Portland , OR ) . All statistics were calculated in MATLAB . Pairwise comparisons were made using two-tailed t-test , and multiple samples were compared using one-way analysis of variance . Errors on connection probability were calculated using the variance of the binomial distribution . The specificity of reciprocal connections between neighboring SACs was assessed by comparison with Monte Carlo simulations using recorded connection probabilities . Latencies for paired recordings from SACs and directional voltage-clamp recordings from vDSGCs were measured by fitting the rising phase of each current using a Boltzmann function in IgorPro and finding the intersection this line with the baseline . Latencies of autaptic currents ( after automated detection ) were calculated manually and taken from the beginning of the short voltage steps to +60 mV . SACs and DSGCs were filled through patch electrodes using methods described above . Alexa Fluor 488 hydrazide ( 200 μM ) was added to the intracellular recording solution , and recordings were maintained for ∼20 min in current-clamp mode while maintaining a negative holding potential ( <−60 mV ) . After individual cells were filled , retinas were either imaged live ( to measure SAC dendritic radius ) or immediately placed in fixative and processed for histology . Mice used exclusively for histology were euthanized by intraperitoneal injection of pentobarbital or euthasol and either enucleated immediately or transcardially perfused with Ringer's solution followed by 4% paraformaldehyde ( PFA ) in PBS . Eye cups were removed and fixed in 4% PFA in PBS on ice for 1 hr then rinsed with PBS . Retinas were analyzed as whole mounts or cryosections as described previously ( Lefebvre et al . , 2012 ) . Whole mount retinas were incubated in blocking buffer ( 0 . 5% Triton-X-100 , 5% normal donkey serum in PBS ) for 1–2 hr at room temperature , then incubated for 5–7 days at 4°C with primary antibodies . For cryosections , fixed retinas were incubated with 30% sucrose/PBS for >2 hr ( until they lost buoyancy ) , frozen , and sectioned at 20 μm in a cryostat . Sections were blocked with 5% donkey serum/0 . 5% Triton X-100/PBS for 1–2 hr at room temperature , with primary antibodies overnight at 4°C , and with secondary antibodies for 2 hr at room temperature . Whole mount retinas or sections were mounted onto glass slides using Fluoromount G ( Southern Biotech ) . The following primary antibodies were used: chick anti-GFP ( 1:500 , Abcam ) ; rabbit anti-DsRed ( 1:1000 , Clontech ) ; goat anti-choline acetyltransferase ( ChAT ) ( 1:400 , Millipore ) ; goat anti-VAChT ( 1:1000 , Promega ) ; rabbit anti-Calbindin ( 1:2500 , Swant ) ; rabbit anti-CART ( 1:1000 , Phoenix ) ; mouse anti-Brn3a ( 1:1000 , Millipore ) ; goat anti-Chx10 ( 1:200 , Santa Cruz ) ; and mouse anti-AP2 ( 1:1000 , DSHB ) . Nuclei were labeled with TO-PRO3 ( 1:3000 , Invitrogen ) . Secondary antibodies were conjugated to Alexa Fluor 488 , Alexa Fluor 568 ( Invitrogen ) , or DyLight 649 ( Jackson ImmunoResearch ) and used at 1:1000 . Immunofluorescence samples were imaged using Olympus FV1000 confocal microscope using 488 , 568 , and 647 lasers with a z-step size of 1 . 0 µm . FIJI ( NIH ) was used to analyze confocal stacks and generate maximum intensity projections . ON and OFF dendrites of DSGCs were separated using depths in the inner plexiform layer and corresponding SAC bands . Retinal orientations were maintained throughout .

Nerve cells ( or neurons ) connect to one another to form circuits that control the animal's behavior . Typically , each neuron receives signals from other cells via branch-like structures called dendrites . Each specific type of neuron has a characteristic pattern of branched dendrites , which is different from the pattern of other types of neuron . Therefore , it is reasonable to imagine that the shape of these branches can influence how the neuron works; however , this idea has rarely been tested experimentally . Different processes are known to act together to control the pattern of the branched dendrites . For example , dendrites in some neurons avoid other dendrites from the same neuron . This phenomenon is referred to as ‘self-avoidance’ . In some of these cases , the same dendrites freely interact with the dendrites of neighboring neurons of the same type; this is called ‘self/non-self discrimination’ . It is not clear , however , how these two processes influence the activity of neural circuits . Both self-avoidance and self/non-self discrimination rely on the expression of genes that encode so-called recognition molecules . Kostadinov and Sanes have now altered the expression of these genes in mice to see the effect that disrupting these two phenomena has on a set of neurons called ‘starburst amacrine cells’ that are found at the back the eye . The dendrites of starburst amacrine cells generate signals when objects move across the animal's field of vision . These dendrites then signal to other starburst amacrine cells and to so-called ‘direction-selective ganglion cells’ , which in turn send this information to the brain for further processing . The experiments revealed that these disruptions affected the connections between the dendrites . Starburst amacrine cells that lacked self-avoidance mistakenly formed connections with themselves—as if they mistook their own dendrites for those of other starburst cells . In contrast , neurons that lacked self/non-self discrimination made the opposite mistake , and rarely formed connections with each other—as if they mistook the dendrites of other starbursts for their own . Disruptions to either phenomenon interfered with the activity of the direction-selective ganglion cells . Following on from the work of Kostadinov and Sanes , the next challenges include uncovering how the recognition molecules help with self-avoidance and self/non-self discrimination . It will also be important to examine whether the conclusions based on one type of neurons can be generalized to others that also exhibit these two phenomena .

2015

Protocadherin-dependent dendritic self-avoidance regulates neural connectivity and circuit function

Transcription factors ( TFs ) operate by the combined activity of their DNA-binding domains ( DBDs ) and effector domains ( EDs ) enabling the coordination of gene expression on a genomic scale . Here we show that in vivo delivery of an engineered DNA-binding protein uncoupled from the repressor domain can produce efficient and gene-specific transcriptional silencing . To interfere with RHODOPSIN ( RHO ) gain-of-function mutations we engineered the ZF6-DNA-binding protein ( ZF6-DB ) that targets 20 base pairs ( bp ) of a RHOcis-regulatory element ( CRE ) and demonstrate Rho specific transcriptional silencing upon adeno-associated viral ( AAV ) vector-mediated expression in photoreceptors . The data show that the 20 bp-long genomic DNA sequence is necessary for RHO expression and that photoreceptor delivery of the corresponding cognate synthetic trans-acting factor ZF6-DB without the intrinsic transcriptional repression properties of the canonical ED blocks Rho expression with negligible genome-wide transcript perturbations . The data support DNA-binding-mediated silencing as a novel mode to treat gain-of-function mutations . Transcription factors ( TFs ) operate by entangling their DNA-binding and transcriptional activation or repression functions ( Ptashne , 2014 ) . However , in eukaryotes TF DNA binding and effector activities are typically structurally modular ( Brent , 1985 ) consisting of a DNA-binding domain ( DBD ) controlling the TF topology on genomic targets and an effector domain ( ED ) ( Brent , 1985; Kadonaga , 2004 ) that recruits co-activator or co-repressor complexes ( Malik and Roeder , 2010; Perissi et al . , 2010 ) resulting in either transcriptional activation or repression of gene regulatory networks ( GRNs ) ( Neph et al . , 2012 ) . Engineered TFs mimic the design of natural TFs ( Pavletich and Pabo , 1991; Beerli and Barbas , 2002 ) . To generate target specificity the DBD module is engineered to recognize unique genome sites ( Beerli and Barbas , 2002 ) , whereas the transcriptional activation or repressor properties are conferred by the selection of the ED ( Konermann et al . , 2013 ) . To silence gain-of-function mutations , while studying the features of genomic DNA-TF interactions , here we investigated the hypothesis that engineered DNA-binding proteins without canonical ED activity possess transcriptional repression properties . As a transcriptional repression target we selected the G-protein-coupled Receptor Rhodopsin ( RHO ) gene whose gain-of-function mutations are those most commonly associated with autosomal dominant retinitis pigmentosa ( adRP ) , an incurable form of blindness ( Dryja et al . , 1990 ) . We generated a DNA-binding protein targeted to a cis-regulatory element ( CRE ) of the human proximal RHO promoter region by deconstructing an engineered TF ( synthetic ) composed of a DBD ( ZF6-DNA-binding protein , ZF6-DB ) and the ED ( Kruppel-associated box , KRAB repressor domain , KRAB ) , which we have shown to be effective in repressing specifically the human RHO transgene carried in an adRP mouse model ( Mussolino et al . , 2011a ) . The deletion of the ED resulted in a protein , ZF6-DB targeting 20 base pairs of genomic CRE , here named ZF6-cis , found at -84 bp to -65 bp from the transcription start site ( TSS ) of the human RHO gene ( Figure 1a; Mitton et al . , 2000 ) . Genomic ZF6-cis is without apparent photoreceptor-specific endogenous transcription factor-binding sites ( TFBS; Figure 1a ) , as reported ( Kwasnieski et al . , 2012 ) . To study the CRE features of ZF6-cis that ZF6-DB would interfere with upon binding in the absence of KRAB-mediated co-repressor recruitment , we deleted the 20 bp genomic ZF6-cis sequence and assessed its function by eGFP reporter assay ( Kwasnieski et al . , 2012 ) in living porcine retina by AAV delivery . The 776 bp-long RHO promoter fragment carrying the ZF6-cis deletion upstream of the eGFP reporter gene ( AAV-RHO-cis-del-EGFP ) , after delivery to the porcine retinal photoreceptor , showed loss of eGFP expression compared to the control vector ( AAV-RHO-EGFP ) ( Figure 1b , c ) . This suggests that ZF6-cis CRE is necessary for RHO expression ( at least for the 776 bp region used in the assay ) and that binding of the synthetic ZF6-DB trans-acting counterpart of ZF6-cis , may indeed repress RHO transcription . 10 . 7554/eLife . 12242 . 003Figure 1 . Delivery of ZF6-DB DNA-binding synthetic trans-acting factor targeted to a 20 bp of RHO cis-acting regulatory element ( CRE ) dramatically reduces Rho expression in photoreceptors . ( a ) Schematic representation of the chromosomal location of the RHO locus and its proximal promoter elements indicating the transcription start site ( in green , +1 ) and the location of ZF6-DB binding site ( in red , ZF6-Cis ) and ZF6-DB ( based on Mitton et Al . , 12 ) ; BAT1 , Bovine A/T-rich sequence1; NRE , NRL response element; TBP , TATA box binding protein . ( b ) qReal Time PCR of mRNA levels ( 2^-ΔCT ) on the adult porcine retina 15 days after vector delivery of either AAV8-hRHO-eGFP ( n=2 ) or AAV8-hRHO-cis-del-eGFP ( n=2 ) subretinally administered at a dose of 1x1010 , showed that AAV8-hRHO-cis-del-eGFP resulted in decreased transduction ( about fifty fold ) compared with hRHO . ( c ) Histology confirmed the decrease of eGFP expression in hRHO-cis-del-eGFP injected retina compared with the retina injected with hRHO-eGFP . Scale bar , 50 µm . ( d ) qReal Time PCR of mRNA levels ( 2^-ΔCT ) of adult porcine retina injected subretinally with AAV8-CMV-ZF6-DB ( n=6 ) at a vector dose of 1x1010 genomes copies ( gc ) compared with non-transduced area ( n=7 ) of the same eye 15 days after vector delivery , resulted in robust transcriptional repression of the Rho transcript . pRHO , porcine Rhodopsin; Gnat1 , Guanine Nucleotide Binding Protein1 . ( e ) Rho Immunofluorescence ( green ) histological confocal analysis of AAV8-CMV-ZF6-DB treated porcine retina compared with non-transduced area . Scale bar , 100 um . The treatment with ZF6-DB determined collapse of the outer-segment ( OS ) with apparent retention of nuclei ( stained with DAPI ) in the outer nuclear layer ( ONL ) . ( f ) Immunofluorescence triple co-localization staining of porcine retina shown in ( b ) with Rho ( blue ) , rod specific protein Gnat1 ( green ) and HA ( ZF6-DB , red ) antibodies . White arrows indicate co-localization of both HA-tag-ZF6-DB and Gnat1 rods depleted of Rho , whereas yellow arrows showed residual Rho and Gnat1 positive cells lacking ZF6-DB . A magnification of the triple staining ( box ) is highlighted . Scale bar , 100 µm . OS , outer segment; IS , inner segment; ONL , outer nuclear layer; INL , inner nuclear layer . ( g ) Representative fluorescence-activated cell sorting ( FACS ) of porcine retina 15 days after injections of either AAV8-GNAT1-eGFP ( dose 1x1012 gc ) or co-injection with both AAV8-GNAT1-eGFP and AAV8-CMV-ZF6-DB ( dose of eGFP , 1x1012 gc; ZF6-DB dose 5x1010 gc ) . eGFP positive sorted cells ( AAV8-GNAT1-eGFP ) corresponded to 17 , 3% of the analysed population ( left panel; P2 area , green dots ) , whereas , 22 , 4% of eGFP positive cells in the retina that received both vectors ( AAV8-GNAT1-eGFP and AAV8-CMV-ZF6-DB; right panel; P2 area , green dots ) . ( h ) qReal Time PCR on sorted rods treated with AAV8-GNAT1-eGFP ( n=3 ) and AAV8-CMV-ZF6-DB ( n=3 ) showed a repression of about 85% of total rhodopsin when compared with rods treated with eGFP ( mRNA levels: 2^-ΔCT ) . Error bars , means +/- s . e . m . n =; *p<0 . 05 , **p<0 . 01 , ***p<0 . 001; two-tailed Student’s t test . DOI: http://dx . doi . org/10 . 7554/eLife . 12242 . 00310 . 7554/eLife . 12242 . 004Figure 1—figure supplement 1 . Chromatin Immunoprecipitation ( ChIP ) of ZF6-DB and ZF6-KRAB . ( a ) Alignment of RHODOPSIN proximal promoter between human , mouse and pig DNA sequences; in bold ZF6-cis sequence , underlined the sequence differences present in ZF6-cis bound by ZF6-DB . ( b ) qPCR ChIP analysis of RHO TSS region including ZF6-cis site ) , GNAT1 , ARR3 and tubulin beta proximal promoters controls in transfected HEK293 cells ( n=3 indipendent expermients ) . RHO-specific enrichment is shown on RHO TSS region . ***p<0 . 001; two-tailed Student’s t test . DOI: http://dx . doi . org/10 . 7554/eLife . 12242 . 004 Chromatin immunoprecipitation ( ChIP ) experiments to evaluate binding to the ZF6-cis target genomic sequence showed occupancy by the DNA-binding protein ZF6-DB ( Figure 1—figure supplement 1b ) . To evaluate whether ZF6-DB represses transcription of the RHO gene in a physiological genomic context , we used the porcine retina ( Mussolino et al . , 2011b ) , which shares 19 out of 20 DNA bp with the human genomic ZF6-cis sequence ( Figure 1—figure supplement 1a ) . Subretinal delivery of a low AAV8 vector dose ( 1x1010 genome copies; gc ) of ZF6-DB ( AAV8-CMV-ZF6-DB ) resulted in a 45% decrease of porcine Rho transcript levels at 15 days post-injection ( Figure 1d ) . Immunofluorescence analysis showed depletion of Rho protein and consequent collapse of the rod outer segments in ZF6-DB positive cells ( Figure 1e , f ) . Despite the lack of detectable Rho expression in most of the transduced rods , rows of photoreceptor nuclei were preserved from degeneration at this time point ( Figure 1e ) . To further evaluate the extent of Rho silencing in rod photoreceptors by ZF6-DB , we performed FACS analysis on eGFP-labelled rod cells . Rod cells were isolated from porcine retina that had received a subretinal injection of an AAV vector encoding eGFP under the control of a rod-specific promoter ( human Guanine Nucleotide Binding Protein1 , GNAT1 promoter elements ( Lee et al . , 2010 ) ; AAV8-GNAT1-eGFP; dose 1x1012 gc ) with or without the vector encoding ZF6-DB ( 5x1010 gc ) . Fifteen days after injection , the retina were disaggregated and FACS-sorted . The retina co-transduced with eGFP and ZF6-DB vector showed virtually a 'somatic knock-out' of Rho expression ( ~85% decrease of Rho transcript levels; Figure 1g , h ) . To evaluate genome–wide transcriptional specificity , we analyzed the porcine retinal transcriptome by RNA sequencing ( RNA-Seq ) from retina harvested 15 days after subretinal injection of the AAV8 vector encoding ZF6-DB ( Figure 2 ) . For comparison we used the engineered TF with the ED , KRAB ( AAV8-CMV-ZF6-KRAB ) . The low vector doses delivered to the porcine retina ( 1x1010 gc ) resulted in about twenty-fold lower expression levels of the ZF6-DB and ZF6-KRAB transgenes compared to Crx and Nrl , two retina-specific TFs ( Swaroop et al . , 2010 ) ( Figure 2a ) . Of note , despite these low expression levels , we observed robust Rho transcriptional repression ( Figure 2b ) . We then analyzed the transcriptional perturbation in response to the AAV retinal gene transfer of ZF6-DB by determining the differentially expressed genes ( DEGs ) . Remarkably , in vivo the ZF6-DB protein generated about ten-fold less transcriptional perturbation compared with the ZF6-KRAB protein ( 19 vs . 222 DEGs; Figure 2e ) . Notably , this magnitude of perturbation is twenty five-fold lower than that induced by the ablation of an endogenous rod-specific TF ( NRL , 500 DEGs vs 19 DEGs , ZF6-DBD; [Roger et al . , 2014] ) . Retinal-specific pathway analysis of DEGs showed that ZF6-DB–induced down-regulation is restricted to the Rho biochemical interactor Gnat1 ( Palczewski , 2012 ) , and the up-regulation of 2 genes associated with acute phase inflammatory response , alpha-2-macroglobulin ( A2m ) and glial fibrillary acidic protein ( Gfap ) ( Figure 2c ) . ZF6-KRAB induced the de-regulation of 17 retinal network associated genes ( Figure 2—figure supplement 1 ) . These results suggest that both ZF6-DB and the consequent Rho down-regulation marginally interferes with photoreceptor specific pathways , apart from Gnat1 repression , and that the up-regulation of the inflammatory response genes may be due to the collapse of the retinal scaffold caused by Rho depletion . The intersection of retinal transcriptome changes between ZF6-DB and ZF6-KRAB showed that both drive similar perturbation in the expression of 16 genes , which represent 84% of the entire pool of ZF6-DB DEGs ( Figure 2e ) . Consistently , both ZF6-DB and ZF6-KRAB generated similar functional effects , i . e . concordant up- or down- differential expression of these 16 shared genes ( Figure 2d ) . These results suggest that both ZF6-DB and ZF6-KRAB bind to similar genomic targets . We next studied whether the differential transcriptional repression induced by ZF6-DB and ZF6-KRAB was due to similar biochemical binding properties for the ZF6-cis DNA target . Both ZF6-DB and ZF6-KRAB proteins bind the ZF6-cis RHO DNA target site with similar affinities ( Figure 2—figure supplement 2 ) . These data suggest that , despite the presence of an active canonical repressor domain , ZF6-KRAB generated Rho silencing by DNA binding . Indeed , ZF6-DB , being exclusively a DNA-binding protein identical to the DBD of ZF6-KRAB , showed similar Rho silencing effects but far less retinal transcriptional perturbations . This indicates that the specificity is conferred by both the engineered design of the DNA-binding on a genome-specific target ( Beerli and Barbas , 2002 ) and the lack of the ED . In addition , this finding supports the notion that that ZF6-cis CRE is necessary for Rho expression genome-wide . 10 . 7554/eLife . 12242 . 005Figure 2 . Photoreceptor delivery of ZF6-DB resulted in reduced genome-wide transcript perturbations . ( a ) RNA-Seq expression levels ( Mean Normalized Counts ) comparison between 2 endogenous TFs ( Crx and Nrl ) and the expression levels resulting from transduction of AAV8-CMV-ZF6-DB and AAV8-CMV-ZF6-KRAB , 15 days after retinal delivery ( AAV8-CMV-ZF6-DB n= 6; AAV8-CMV-ZF6-KRAB n= 4 and 7 controls , non-transduced area ) . ( b ) Rho and rod Gnat1 and Cone Arrestin 3 expression levels in treated and control retina . ( c ) Ingenuity Pathway Analysis of DEGs after ZF6-DB AAV delivery in porcine retina showed a network of 13 genes . The 2 phototransduction genes RHO and GNAT1 are shown in green ( down-regulated ) whereas the 2 genes associated with primary inflammatory response network , A2M and GFAP , are up-regulated ( red ) . ( d ) Transcriptional activation and repression concordances among Log Fold Changes of the genes in common ( Swaroop et al . , 2010 ) between ZF6-DB and ZF6-KRAB ( Pearson Correlation Test; PC=0 . 9787; p value << 1x10-5 ) . ( e ) Venn Diagrams , pairwise intersection of the 2 sets of Differentially Expressed Genes ( DEGs ) . An adjusted p value ( False Discovery Rate; FDR ≤ 0 . 1 ) , without filtering on fold change levels , resulted in 19 and 222 DEGs , in ZF6-DBD and ZF6-KRAB treated retina , respectively . The intersection resulted significant by hypergeometric test ( p value << 1x10-5 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12242 . 00510 . 7554/eLife . 12242 . 006Figure 2—figure supplement 1 . Ingenuity Pathway Analysis on DEGs of ZF6-KRAB treated retina . ( a ) Delivery of AAV8-CMV-ZF6-KRAB resulted in up-regulation of 10 genes associated with inflammatory responses ( red , up-regulation ) and the down-regulation ( green ) of 7 genes associated with the rod phototransduction cascade . ( b ) the ZF6-DB pathway analysis ( Figure 2 ) is reported for comparison . DOI: http://dx . doi . org/10 . 7554/eLife . 12242 . 00610 . 7554/eLife . 12242 . 007Figure 2—figure supplement 2 . Determination of the binding constants of ZF6-KRAB and ZF6-DB . ( a ) Gel mobility shift titrations of ZF6-KRAB and ZF6-DB with the hRHO 65 bp oligonucleotide ( see 'Materials and methods' ) . ( b ) In the saturation binding experiments the nanomolar concentration of specific binding data was plotted against of nanomolar increasing concentration ( 130 , 135 , 145 , 150 , 165 , 170 , 175 , 180 , 190 , and 200 nM , respectively ) and ( 145 , 150 , 170 , 175 , 195 , 210 , 220 , 225 , 240 , and 250 nM , respectively ) of DNA ligand and Scatchard analysis of the gel shift binding data . The ratio of bound to free DNA is plotted versus the nanomolar concentration of bound DNA in the reaction mixture . The ZF6-KRAB and ZF6-DB apparent dissociation constants ( Kd ZF6-KRAB = 108 . 00 ± 11 . 78 nM ( R2 = 0 . 97 ) and Kd ZF6-DB = 41 . 94 ± 3 . 45 nM , R2 = 0 . 96 ) , respectively ) were determined . ( c ) combination of the a and b panels . DOI: http://dx . doi . org/10 . 7554/eLife . 12242 . 007 To test the functional activity of ZF6-DB in an adRP animal model , we used the RHO-P347S mouse ( Li et al . , 1996 ) , which harbors the P347S RHO human mutant allele including the 20 DNA base pairs of the human genomic ZF6-cis sequence , whereas the murine Rho promoter sequence lacks the ZF6-Cis target ( Figure 1—figure supplement 1a ) ( Li et al . , 1996 ) . Therefore , human-specific P347S RHO silencing by ZF6-DB ( which does not affect murine Rho expression ) may result in preservation of retinal function ( Mussolino et al . , 2011a ) . Strikingly , AAV8 vector delivery of ZF6-DB resulted in significantly higher functional protection in both the A- and B-wave components of electroretinography ( ERG analysis ) compared to ZF6-KRAB and AAV-GFP controls ( Figure 3a ) . In addition , injection of either ZF6-KRAB or ZF6-DB in C57Bl/6 wild type retina did not produce detectable functional detrimental effects ( Figure 3b ) . Thus , the higher ERG responses observed in ZF6-DB- compared to ZF6-KRAB-treated P347S mice should be further investigated . 10 . 7554/eLife . 12242 . 008Figure 3 . ZF6-DB DNA-binding protein preserves retinal function of the P347S adRP mouse model . ( a ) Electroretinography ( ERG ) analysis on P347S mice mice subretinally injected at post natal day 14 ( PD14 ) with AAV8-CMV-ZF6-DB ( n=10 ) , AAV8-CMV-ZF6-KRAB ( n=10 ) , or AAV8-CMV-eGFP ( n=10 ) and analysed at P30 . Retinal responses in both scotopic ( dim light ) and photopic ( bright light ) showed that both A- and B-waves amplitudes , evoked by increasing light intensities , were preserved in both AAV8-CMV-ZF6-DB and ZF6-KRAB compared to eGFP control ( b ) A- and B-wave are shown for injected C57Bl/6 mice with ZF6-DB ( n=4 ) , ZF6-KRAB ( n=4 ) and eGFP ( n=4 ) , independently . No functional impairment is observed for each construct . Error bars , means +/- s . e . m . *p<0 . 05 , **p<0 . 01 , ***p<0 . 001; two-tailed Student’s t test . DOI: http://dx . doi . org/10 . 7554/eLife . 12242 . 008 To determine the therapeutic potential of DNA-binding-mediated silencing , we carried out the silencing-replacement strategy ( Kiang et al . , 2005 ) by coupling ZF6-DB with RHO replacement ( human RHO , hRHO CDS ) in order to complement Rho transcriptional repression in porcine retina ( Figure 4—figure supplement 1 ) . To achieve simultaneous photoreceptor transduction of both ZF6-DB and hRHO , we cloned the two expression cassettes into a single vector ( DNA-binding repression and replacement , DBR-R , construct; Figure 4a ) . The key variables to achieve highly differential expression required for balanced simultaneous RHO repression and replacement are the vector dose and promoter strength . Indeed , ZF6-DB ( ~200 counts; RNA-seq expression levels ) generated a decrease of about 100 , 000 Rho RNA-seq counts after transduction ( ~250 , 000 counts in controls vs . ~150 , 000 after treatment; Figure 2a , b ) . Thus , to ensure high and rod-specific hRHO replacement , we opted for a high vector dose and the strength of the GNAT1 promoter ( Figure 1h; [Lee et al . , 2010] ) . To decrease ZF6-DB expression levels at high vector dose , while keeping rod-specificity , we both shortened the human RHO promoter and deleted the 5’ sequence of the ZF6-DB target ZF6-Cis ( Figure 4—figure supplement 2 ) . We used 1x1012 gc of vector of DBR-R ( AAV8-RHOΔ-ZF6-DB-GNAT1-hRHO ) to administer to porcine retina . As an internal control the contralateral eye received the previously used ZF6-DB vector ( AAV8-CMV-ZF6-DB; at 1x1010 gc; Figure 1 ) . AAV8-CMV-eGFP was co-administrated to label the transduced area . Administration of the DBR-R vector resulted in rod-specific transcriptional repression of the porcine Rho ( 38% ) and in concomitant replacement of the exogenous hRHO ( 68% ) , as assessed by transcripts , protein expression levels , and integrity of photoreceptor outer segments ( Figure 4b–d and Figure 4—figure supplement 3 ) . Notably , Gnat1 transcript and protein ( data not shown ) levels demonstrated complementation , supporting a secondary down-regulation of Gnat1 associated with Rho repression ( Figure 4b ) . 10 . 7554/eLife . 12242 . 009Figure 4 . DNA-binding repression-replacement ( DBR-R ) of Rho in the porcine retina . ( a ) AAV8-RHOΔ-ZF6-DB-GNAT1-hRHO DBR-R construct features , including the two expression cassettes , RHOΔ-ZF6-DB encoding for both the DNA-binding repressor ZF6-DB ( orange ) , and GNAT1-hRHO for human RHO for replacement ( blue ) . The size ( kb ) of the construct is indicated as a bar . ( b ) qReal Time PCR , mRNA levels ( 2^-ΔCT ) 2 months after vector delivery of either AAV8-CMV-ZF6-DB ( DBR; orange bars ) or AAV8-RHOΔ-ZF6-DB-GNAT1-hRHO ( DBR-R , blue bars ) and non-transduced controls ( green bars ) . pRho , porcine Rhodopsin; Gnat1 , Guanine Nucleotide Binding Protein1; Arr3 , Arrestin 3; hRHO , human Rhodopsin . The result is representative of two independent experiments . Error bars , means +/- s . e . m . ; *p<0 . 05 , **p<0 . 01 , ***p<0 . 001; two-tailed Student’s t test . ( c ) Western blot analysis on the retina showed in b , c and d . ( d ) Immunofluorescence double staining with Rho ( green ) and HA-ZF6-DB ( red ) antibodies . Left panel , non-transduced control retina; middle panel , AAV8-CMV-ZF6-DB treated retina; left panel , AAV8-RHOΔ-ZF6-DB-GNAT1-hRHO DBR-R treated retina . OS , outer segment; IS , inner segment; ONL , outer nuclear layer; INL , inner nuclear layer; scale bar , 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 12242 . 00910 . 7554/eLife . 12242 . 010Figure 4—figure supplement 1 . Outline of the DNA-binding repressor-replacement ( DBR-R ) strategy . The DNA-binding protein , ZF6-DB , for transcriptional silencing of RHO is coupled to replacement , R , in the same AAV vector ( left hexagon ) to ensure simultaneous transduction of photoreceptors . The DNA-binding protein ZF6-DB operates on the regulatory region of the RHO promoter , in an allele and mutation independent manner i . e . ZF6 DNA binding on the endogenous promoter represses RHO transcription irrespectively of the mutated and WT alleles , preventing RHO expression ( mutated RHO , green; and WT , blue ) . This strategy is designed to overcome the high heterogeneity of RHO mutations . Highlighted as a 'chromosome zoom-in' [Genome Browser] the ZF6-DB ( orange squares ) bound to the regulatory DNA-target sequence . Exogenously AAV vector delivered RHO ( blue ) for replacement is shown in the black box . DOI: http://dx . doi . org/10 . 7554/eLife . 12242 . 01010 . 7554/eLife . 12242 . 011Figure 4—figure supplement 2 . Strength and tissue specificity of RHOΔ promoter elements in murine retina . ( a ) qReal Time PCR , mRNA levels ( 2^-ΔCT ) on the adult murine retina 15 days after vector delivery of either AAV8-hRHO short-eGFP or AAV8-hRHO-s-ΔZF6-eGFP subretinally administered at a dose of 1x109 , showed that AAV8-hRHO-s-ΔZF6-eGFP resulted in decreased transduction ( about ten fold ) compared with hRHO long . Error bars , means +/- s . e . m . ; *p<0 . 05 , **p<0 . 01; two-tailed Student’s t test . ( b ) Histology demonstrated maintenance of rod-specific expression by AAV8-RHOΔ-eGFP . Scale bar , 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 12242 . 01110 . 7554/eLife . 12242 . 012Figure 4—figure supplement 3 . Cone morphological integrity after DNA-binding repression-replacement ( DBR-R ) subretinal delivery . Rod-specific expression of DBR-R ( AAV8-RHOΔ-ZF6-DB-GNAT1-hRHO ) 2 month after vector delivery . Immunofluorescence double staining with human cone Arrestin 3 ( hCAR; green ) and HA-ZF6-DB ( red ) antibodies , showed rod specific transduction . Left panel , non-transduced control retina; right panel , AAV8-RHOΔ-ZF6-DB-GNAT1-hRHO DBR-R treated retina . OS , outer segment; IS , inner segment; ONL , outer nuclear layer; INL , inner nuclear layer . Scale bar , 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 12242 . 012 In this study , we showed that photoreceptor genomic binding of a 20 bp-long DNA sequence by a synthetic DNA-binding protein dramatically reduces Rho expression . The combination of Rho transcriptional silencing and the restricted transcriptome perturbation induced by ZF6-DB , without the intrinsic repression activity contained in an ED , indicate that the local binding to ZF6-cis per se is the determinant of transcriptional repression , whereas the high specificity observed may result by both DNA-binding specificity ( biochemical affinity ) of ZF6-DB and the rod genomic context . The transcriptional repression mechanism of ZF6-DB binding likely relies on the interference occurring between TFs and local DNA sequence features within the RHO proximal promoter region ( Mitton et al . , 2000 ) , which we showed here to be necessary to control Rho expression at the genomic level . The lack of known TFBSs and the low level of expression of ZF6-DB expressed in the photoreceptors , which was twenty-fold below the levels of photoreceptor specific TFs ( Crx and Nrl ) , suggest that the molecular determinant of silencing may not be the simple displacement of key RHO TFs ( Mao et al . , 2011 ) . We propose a model in which the molecular features of the DNA ( loop , twisting , bending , for instance ) may contribute to Rhodopsin transcriptional output . In this context , the DNA may be envisaged as not being exclusively the source of storage of functional information ( protein coding and non-coding transcripts ) or an inert DNA-binding protein harbor ( i . e . positional information for TFs DNA binding ) , but also as an intrinsically active operator of the transcriptional function . This contribution of DNA is supported by both the transcriptional repression upon the ZF6-cis deletion of 20 bp of DNA and intereference in trans by using a synthetic DNA binding protein that , not being encoded by the genome , may occupy a protein-free portion of the genome . It follows that in terms of signaling , the information source , the DNA , generates an output signal the RNA and eventually a protein ( TF ) , whose final output functional activity is completed back by the DNA . Thus the information source , the DNA , becomes also integral part of the signaling ( Rhodopsin transcriptional output ) . In this vision , to act ( interfere ) upstream the DNA , an external function is necessary , which is , in this study , the synthetic ZF6-DB carried by the vector . From a therapeutic prospective , a relevant property of ZF6-DB DNA binding interference is the high rate of transcriptional silencing observed after in vivo gene transfer , which is consistent with canonical TFs mode of action ( Kiang et al . , 2005 ) . DNA binding interference via ZF6-DB in transduced retina generated 45% Rho transcriptional repression , which reached 85% when rods were sorted , supporting its use for diseases requiring correction of a large number of affected cells , such as adRP and other Mendelian disorders due to gain-of-function mutations . Furthermore , Rho transcriptional silencing and its complete RHO replacement support in principle the use of the DBR-R constructs for treatment of any RHO mutation including those caused by a dominant negative mechanism ( Mao et al . , 2011 ) . However , the 38% silencing and replacement observed may not yet be sufficient to achieve therapeutic efficacy/benefit in patients with adRP . Therefore , further development of this proof-of-concept will include optimization of the design of the silencing and replacement double construct ( tuning the strength of the promoter elements ) , vector selection and dose , and the surgical approach . In conclusion , in vivo retinal gene transfer of an AAV vector ( Doria et al . , 2013; Liang et al . , 2001; Scatchard , 1949 ) carrying a 22 kDa orthogonal ( Surace et al . , 2005 ) and gene-extrinsic noise-resistant ( the permissive rod photoreceptor cell-specific environment; [Li et al . , 2011] ) synthetic protein acts as a transcriptional repressor , which results in a potent and specific silencing of the Rho gene upon binding to an essential Rho DNA element . The ZF6-DNA-binding domain ( NΔ96 deletion mutant , ZF6-DB ) was amplified by PCR from AAV2 . 1 CMV-ZF6-KRAB ( Mussolino et al . , 2011a ) using primers ZF6-DBfw ( TTGCGGCCGCATGATCGATCTGGAACCTGGCG ) and ZF6-DBrv ( AAGCTTTCAAGATGCATAGTCT ) . The PCR product was digested using NotI and HindIII restriction enzymes and cloned in pAAV2 . 1 . The hGNAT1 promoter was synthetized by Eurofins MWG based on Lee et al . 2010 adding the 5’UTR . The fragment was cloned in pAAV2 . 1 using NheI and NotI restriction enzymes . The human Rhodopsin CDS was amplified by PCR from human retina cDNA using the hRHOfw ( GCGGCCGCATGAATGGCACAGAAGGCCC ) e hRHOrv ( AAGCTTTTAGGCCGGGGCCACCTG ) primers and the PCR fragment was digested using NotI and HindIII restriction enzymes and cloned in pAAV2 . 1 plasmid under the control of hGNAT1 promoter . The human rhodopsin short promoter ( hRHO-short- ( s ) , 164 bp from the transcription starting site ( TSS ) + 5’UTR ) , the human rhodopsin long promoter ( hRHO-long , 796 bp from the TSS + 5’UTR ) , the human rhodopsin long promoter mutated of the ZF6-cis ( hRHO-cis-del , 776 bp from the TSS lacking the bases -82 -62 from the TSS ) and the human rhodopsin muted promoter ( hRHO-s-ΔZF6 , lacking the bases -84 -77 from the TSS ) were generated by gene synthesis of Eurofins MWG and cloned in pAAV2 . 1 using NheI and NotI restriction enzymes . For the generation of DBR-R plasmid the Eurofins MWG synthetized the expression cassette RHOΔ-ZF6-DB-bGHpolyA ( bovine growth hormone polyA ) that we cloned in pAAV2 . 1 hGNAT1-hRHO using NheI restriction enzyme . AAV vectors were produced by the TIGEM AAV Vector Core , by triple transfection of HEK293 cells followed by two rounds of CsCl2 purification ( Auricchio et al . , 2001 ) . For each viral preparation , physical titers [genome copies ( GC ) /mL] were determined by averaging the titer achieved by dot-blot analysis ( Doria et al . , 2013 ) and by PCR quantification using TaqMan ( Applied Biosystems , Carlsbad , CA , USA ) . All procedures were performed in accordance with institutional guidelines for animal research and all of the animal studies were approved by the authors . P347S+/+ animals ( Mussolino et al . , 2011a; Li et al . , 1996 ) were bred in the animal facility of the Biotechnology Centre of the Cardarelli Hospital ( Naples , Italy ) with C57Bl/6 mice ( Charles Rivers Laboratories , Calco , Italy ) , to obtain the P347S+/- mice . Intraperitoneal injection of ketamine and medetomidine ( 100 mg/kg and 0 . 25 mg/kg respectively ) , then AAV vectors were delivered sub-retinally via a trans-scleral transchoroidal approach as described by Liang et al . ( Liang et al . , 2001 ) . Eleven-week-old Large White ( LW ) female piglets were utilized . Pigs were fasted overnight leaving water ad libitum . The anesthetic and surgical procedures for pigs were previously described ( Mussolino et al . , 2011b ) . AAV vectors were inoculated sub-retinally in the avascular nasal area of the posterior pole between the two main vascular arches , as performed in Mussolino et al . ( Mussolino et al . , 2011b ) . This retinal region is crossed by a streak-like region that extends from the nasal to the temporal edge parallel to the horizontal meridian , where cone density is high , reaching 20 , 000 to 35 , 000 cone cells mm2 . Each viral vector was injected in a total volume of 100 µl , resulting in the formation of a subretinal bleb with a typical ‘dome-shaped’ retinal detachment , with a size corresponding to 5 optical discs . DNA fragments encoding the sequence of the engineered transcription factors and ZF6-KRAB , to be expressed as maltose-binding protein ( MBP ) fusion were generated by PCR using the plasmids pAAV2 . 1 CMV-ZF6-KRAB and pAAV2 . 1 CMV-ZF6-DB as a DNA template . The following oligonucleotides were used as primers: primer 1 , ( GGAATTCCATATGGAATTCCCCATGGATGC ) and primer 2 , ( CGGGATCCCTATCTAGAAGTCTTTTTACCGGTATG ) , for ZF6-KRAB primer 3 , ( GGAATTCCATATGCTGGAACCTGGCGAAAAACCG ) and primer 4 , ( CGGGATCCCTATCTAGAAGTCTTTTTACCGGTATG ) for ZF6-DB . All the PCR products were digested with the restriction enzymes NdeI and BamH1 and cloned into NdeI BamH1-digested pMal C5G ( New England Biolabs , Ipswich , MA ) bacterial expression vector . All the plasmids obtained were sequenced to confirm that there were no mutations in the coding sequences . The fusion proteins were expressed in the Escherichia coli BL21DE3 host strain . The transformed cells were grown in rich medium plus 0 . 2% glucose ( according to protocol from New England Biolabs ) at 37°C until the absorbance at 600 nm was 0 . 6–0 . 8 , at which time the medium was supplemented with 200 µM ZnSO4 , and protein expression was induced with 0 . 3 mM isopropyl 1-thio-β-D-galactopyranoside and was allowed to proceed for 2 hr . The cells were then harvested , resuspended in 1X PBS ( pH 7 . 4 ) , 1 mM phenylmethylsulfonyl fluoride , 1 µM leupeptin , 1 µM aprotinin , and 10 µg/ml lysozyme , sonicated , and centrifuged for 30 min at 27 , 500 relative centrifugal force . The supernatant was then loaded on amylose resin ( New England Biolabs ) according to the manufacturer’s protocol . To remove the MBP from the proteins , bound fusion proteins as cleaved in situ on the amylose resin with Factor Xa ( 1 unit/20 µg of MBP fusion protein ) in FXa buffer ( 20 mM Tris , pH 8 . 0 , 100 mM NaC1 , 2 mM CaC12 ) for 24–48 hr at 4°C and collected in the same buffer after centrifugation at 500 relative centrifugal force for 5 min . The supernatant containing the protein without the MBP tag was then recovered . The affinity binding costant of proteins for hRHO proximal promoter sequence was measured by a gel mobility shift assay by performing a titration of the proteins with the oligonucleotides . The purified proteins were incubated for 15 min on ice with hRHO 65 bp duplex oligonucleotide in the presence of 25 mM Hepes ( pH 7 . 9 ) , 50 mM KCl , 6 . 25 mM MgCl2 , 1% Nonidet P-40 , 5% glycerol . After incubation , the mixture was loaded on a 5% polyacrylamide gel ( 29:1 acrylamide/bisacrylamide ratio ) and run in 0 . 5 TBE at 4°C ( 200 V for 4 hr ) . Protein concentration was determined by a modified version of the Bradford procedure . After electrophoresis , the gel was stained with the fluorescent dyes SYBR Green I Nucleic acid gel stain ( Invitrogen , Carlsbad , CA ) to visualize DNA . 2 . 5 µM of the ZF6-KRAB protein was incubated with increasing concentrations ( 130 , 135 , 145 , 150 , 165 , 170 , 175 , 180 , 190 , and 200 nM , respectively ) of the duplex hRHO 65 bp , an apparent higher protein concentration ( 2 . 5 µM ) was required likely because not all the protein sample was correctly folded . In the case of ZF6-DB , 1 . 5 µM of the protein was incubated with increasing concentrations ( 145 , 150 , 170 , 175 , 195 , 210 , 220 , 225 , 240 , and 250 nM , respectively ) of the duplex hRho 65 bp . Scatchard analysis of the gel shift binding data was performed to obtain the Kd values ( 25 ) . All numerical values were obtained by computer quantification of the image using a Typhoon FLA 9500 biomolecular imager ( GE Healthcare Life Sciences ) . RNAs from tissues were isolated using RNAeasy Mini Kit ( Qiagen , Germany ) , according to the manufacturer protocol . cDNA was amplified from 1 μg isolated RNA using QuantiTect Reverse Transcription Kit ( Qiagen ) , as indicated in the manufacturer instructions . The PCRs with cDNA were carried out in a total volume of 20 μl , using 10 μl LightCycler 480 SYBR Green I Master Mix ( Roche , Switzerland ) and 400 nM primers under the following conditions: pre-Incubation , 50°C for 5 min , cycling: 45 cycles of 95°C for 10 s , 60°C for 20 s and 72°C for 20 s . Each sample was analysed in duplicate in two-independent experiments . Transcript levels of pig retinae were measured by quantitative Real Time PCR using the LightCycler 480 ( Roche ) and the following primers: pRho_forward ( ATCAACTTCCTCACGCTCTAC ) and pRho_reverse ( ATGAAGAGGTCAGCCACTGCC ) , pGnat1_forward ( TGTGGAAGGACTCGGGTATC ) and pGnat1_reverse ( GTCTTGACACGTGAGCGTA ) , pArr3_forward ( TGACAACTGCGAGAAACAGG ) and pArr3_reverse ( CACAGGACACCATCAGGTTG ) . humanRho_forward ( TCATGGTCCTAGGTGGCTTC ) , humanRho_reverse ( ggaagttgctcatgggctta ) and eGFP_forward ( ACGTAAACGGCCACAAGTTC ) and eGFP_reverse ( AAGTCGTGCTGCTTCATGTG ) . All of the reactions were standardized against porcine Actβ using the following primers: Act_Forward ( ACGGCATCGTCACCAACTG ) and Act_reverse ( CTGGGTCATCTTCTCACGG ) . Frozen retinal sections were washed once with PBS and then fixed for 10 min in 4% PFA . Sections were immerse in a retrieval solution ( 0 , 01 M citrate buffer , pH 6 . 0 ) and boiled three times in a microwave . After the blocking solution ( 10% FBS , 10% NGS , 1% BSA ) was added for 1 hr . The primary antibody mouse anti-HA ( 1:300 , Covance ) was diluted in a blocking solution and incubated overnight at 4°C . The secondary antibody ( Alexa Fluor® 594 , anti-mouse 1:1000 , Molecular Probes , Invitrogen , Carlsbad , CA ) has been incubated for 1 hr . Vectashield ( Vector Lab Inc . , Peterborough , UK ) was used to visualize nuclei . Frozen retinal sections were permeabilized with 0 . 2% Triton X-100 and 1% NGS for 1 hr , rinsed in PBS , blocked in 10% normal goat serum ( NGS ) , and then incubated overnight at 4°C with rabbit human cone arrestin ( hCAR ) antibody , kindly provided by Dr . Cheryl M . Craft ( Doheny Eye Institute , Los Angeles , CA ) diluted 1:10000 in 10% NGS . After three rinses with 0 . 1 M PBS , sections were incubated in goat anti-rabbit IgG conjugated with Texas red ( Alexa Fluor 594 , anti-rabbit 1:1000 , Molecular Probes , Invitrogen , Carlsbad , CA ) for 1 hr followed by three rinses with PBS . Frozen retinal sections were permeabilized with 0 . 1% Triton X-100 , rinsed in PBS , blocked in 20% normal goat serum ( NGS ) , and then incubated overnight at 4°C in a mouse anti-1D4 rhodopsin antibody diluted 1:500 in 10% NGS . After three rinses with 0 . 1 M PBS , sections were incubated in goat anti-mouse IgG conjugated with Texas red ( Alexa Fluor® 594 , anti-mouse 1:1000 , Molecular Probes , Invitrogen , Carlsbad , CA ) for 1 hr followed by other three rinses with PBS . Sections were photographed using either a Zeiss 700 Confocal Microscope ( Carl Zeiss , Oberkochen , Germany ) or a Leica Fluorescence Microscope System ( Leica Microsystems GmbH , Wetzlar , Germany ) . Triple-immunostaining for anti-HA , anti-GNAT1 , and anti-Rhodopsin antibody . Frozen retinal sections were washed once with PBS and then fixed for 10 min in 4% PFA . Sections were immerse in a retrieval solution ( 0 , 01 M sodium citrate buffer , pH 6 . 0 ) and boiled three times in a microwave . After the blocking solution ( 10% FBS , 10% NGS , 1% BSA ) was added for 1 hr . The two primary antibody mouse anti-HA ( 1:300 , Covance ) and rabbit GαT1 ( Santacruz Biotechnology ) , were diluted in a blocking solution and incubated overnight at 4°C . The secondary antibodies ( Alexa Fluor 594 , anti-mouse 1:800 , Molecular Probes , and Alexa Fluor 488 , anti-rabbit 1:500 , Molecular Probes , Invitrogen , Carlsbad , CA ) have been incubated for 1 hr , followed by three rinses with PBS . After the slides were incubated in blocking solution ( 10% NGS ) for 1 hr and then incubated O . N . with primary antibody mouse- 1D4 ( 1:500 , Abcam ) . The secondary antibodies ( Alexa Fluor 405 , anti-mouse 1:200 , Molecular Probes , Invitrogen ) . Sections were photographed using a Leica Fluorescence Microscope System ( Leica Microsystems GmbH , Wetzlar , Germany ) . Western blot analysis was performed on retinae , which were harvested . Samples were lysed in hypotonic buffer ( 10 mM Tris-HCl [pH 7 . 5] , 10 mM NaCl , 1 , 5 mM MgCl2 , 1% CHAPS , 1 mM PMSF , and protease inhibitors ) and 20 µg of these lysates were separated by 12% SDS-PAGE . After the blots were obtained , specific proteins were labeled with anti-1D4 antibody anti-Rhodopsin-1D4 ( 1:1000; Abcam , Cambridge , MA ) , and anti-β-tubulin ( 1:10000; Sigma-Aldrich , Milan , Italy ) antibodies . For ChIP experiments , HEK293 cells were transfected by CaCl2 with pAAV2 . 1 CMV-ZF6-KRAB , pAAV2 . 1 CMV-ZF6-DB or pAAV2 . 1 CMV-eGFP . The cells are harvested after 48 hr . ChIP was performed as follow: cells were homogenized mechanically and cross linked using 1% formaldehyde in PBS at room temperature for 10 min , then quenched by adding glycine at final concentration 125 mM and incubated at room temperature for 5 min . Cells were washed three times in cold PBS 1X then cells were lysed in cell lysis buffer ( Pipes 5 mM pH 8 . 0 , Igepal 0 . 5% , KCl 85 mM ) for 15 min . Nuclei were lysed in nuclei lysis buffer ( Tris HCl pH8 . 0 50 mM , EDTA 10 mM , SDS 0 . 8% ) for 30 min . Chromatin was shared using Covaris s220 . The shared chromatin was immunoprecipitated over night with anti HA ChIP grade ( Abcam , ab 9110 , Cambridge , MA ) . The immunoprecipitated chromatin was incubated 3 hr with magnetic protein A/G beads ( Invitrogen , Carlsbad , CA ) . Beads were than washed with wash buffers and DNA eluted in elution buffer ( Tris HCl pH 8 50 mM , EDTA 1 mM , SDS 1% ) . Then Real Time PCR was performed using primers on rhodopsin TSS , hRHOTSSFw ( TGACCTCAGGCTTCCTCCTA ) and hRHOTSSRv ( ATCAGCATCTGGGAGATTGG ) , trasducin 1 TSS , hGNAT1TSSFw ( CAGCCCTGACCCTACTGAAC ) and hGNAT1TSSRv ( CAACCGCTGACTCTGCACT ) , arrestin 3 TSS , hArr3TssFw ( CCTGCTGTGCACATAAGCTG ) and hArr3TssRv ( CGTGTCCCACTCCAATCTCT ) , and β-tubulin TSS , hTUBTSSFw ( TCCTGTACCCCCAAGAACTG ) and hTUBTSSRv ( GCTGCAAAATGAAGTGACGA ) . Co-injected porcine retina with AAV8-CMV-ZF6-DB ( dose 5x1010 gc ) and AAV8-GNAT1-eGFP ( dose 1x1012 gc ) were disaggregated using Papain Dissociation System ( Worthington biochemical corporation ) following the manufacturers protocol . Dissociated retinal cells were analysed using BD FACSAria at IGB ( Institute of Genetic and Biophysic “A . Buzzati-Traverso” ) FACS Facility and sorted , dividing eGFP positive cells ( rods ) from eGFP negative fraction . The method is as described ( Surace et al . , 2005 ) . Brifley , mice were dark reared for 3 hr and anesthetized . Flash electroretinograms ( ERGs ) were evoked by 10-ms light flashes generated through a Ganzfeld stimulator ( CSO , Costruzione Strumenti Oftalmici , Florence , Italy ) and registered as previously described . ERGs and b-wave thresholds were assessed using the following protocol . Eyes were stimulated with light flashes increasing from −5 . 2 to +1 . 3 log cd*s/m2 ( which correspond to 1×10−5 . 2 to 20 . 0 cd*s/m2 ) in scotopic conditions . The log unit interval between stimuli was 0 . 3 log from −5 . 4 to 0 . 0 log cd*s/m2 , and 0 . 6 log from 0 . 0 to +1 . 3 log cd*s/m2 . For ERG analysis in scotopic conditions the responses evoked by 11 stimuli ( from −4 to +1 . 3 log cd*s/m2 ) with an interval of 0 . 6 log unit were considered . To minimize the noise , three ERG responses were averaged at each 0 . 6 log unit stimulus from −4 to 0 . 0 log cd*s/m2 while one ERG response was considered for higher ( 0 . 0−+1 . 3 log cd*s/m2 ) stimuli . The time interval between stimuli was 10 s from −5 . 4 to 0 . 7 log cd*s/m2 , 30 s from 0 . 7 to +1 log cd*s/m2 , or 120 s from +1 to +1 . 3 log cd*s/m2 . a- and b-waves amplitudes recorded in scotopic conditions were plotted as a function of increasing light intensity ( from −4 to +1 . 3 log cd*s/m2 , Figure 3 ) . The photopic ERG was recorded after the scotopic session by stimulating the eye with ten 10 ms flashes of 20 . 0 cd*s/m2 over a constant background illumination of 50 cd/m2 . The 17 libraries were prepared using the TruSeq RNA v2 Kit ( Illumina , San Diego , CA ) according to manufacturer’s protocol . Libraries were sequenced on the Illumina HiSeq 1000 platform and in 100-nt paired-end format to obtain approximately 30 million read pairs per sample . Sequence Reads were trimmed using Trim Galore ! software ( v . 0 . 3 . 3 ) , that trims low-quality ends and removes adapter from reads , using a Default Phred score of 20 . To obtain a precise estimation of this yet uncharacterized tissue , the 17 libraries were aligned against the full transcriptome for Sus scrofa ( Pig ) as provided by ENSEMBL ( SusScrofa 10 . 2 . 73 ) . The GTF included the sequences for the 20 canonical chromosomes plus 4563 scaffolds , and counted 30 . 567 transcripts plus the sequences for the 3 exogenes used in the analysis ( the 2 TRs and eGFP ) . Alignment was performed with RSEM ( v . 1 . 2 . 11 ) ( Li et al . , 2011 ) with default parameters . The resulting expected counts ( the sum of the posterior probability of each read coming from a specific transcript over all reads ) were used for subsequent analysis . The dataset was composed of 17 samples and 25 . 325 genes , divided in 3 experimental groups: 7 Controls , 4 ZF6-KRAB-treated , 6 ZF6-DB-treated . We analyzed the data following the standard Differential Expression Analysis Pipeline with DESeq2 R/Bioconductor package ( v . 1 . 8 . 1 ) ( Love et al . , 2014 ) , filtering and normalizing all libraries together . We filtered low tag counts retaining those which had 1 CPM in at least 3 samples . We fitted a unique Generalized Linear Model ( GLM ) with 1 factor and 3 levels ( Control , ZF6-KRAB-treated , ZF6-DB-treated ) . Differentially expressed genes were obtained out of the 2 contrasts ( each treatment compared to the controls ) , an adjusted pvalue ( FDR ) of less than or equal to 0 . 1 was considered significant . We observed the expected upregulation of the exogenous genes used for the treatment ( ZF6-KRAB , ZF6-DB , eGFP ) and for further evaluations we didn’t take into account their differential expression . The 16 genes in common between the ZF6-DB and ZF6-KRAB DEGs were tested for functional concordance using Pearson product moment correlation coefficient ( cor . test , R package stats v . 3 . 2 . 0 ) ( Huber et al . , 2015; R Core Team , 2015 ) . Two named numeric vectors , one for each condition , containing the Log fold changes values of the 16 genes were tested for association with cor . test default function , method = 'Pearson' . A manually curated list of Human Gene IDs including representative Retinal Markes and a subset of Retina Disease Genes ( Daiger BR et al . , 1998 ) was used to show the interference power of the 2 TRs with the overall regulatory circuitry . The Human IDs were used to retrieve their homolog Porcine genes , if present . Genes with duplicated homolog in the Sus scrofa genome were included in the list ( genes tagged with_1 in Figure 2 ) . All the analyses , except for the reads quality filtering , alignment and expression estimates , were performed in the R statistical environment ( v . 3 . 2 . 0 ) ( Huber et al . , 2015; R Core Team , 2015 ) . Plots were generated with ggplot2 R/Bioconductor package ( v . 1 . 0 . 1 ) ( Wickham , 2009 ) . Data are presented as mean ± Error bars indicate standard error mean ( SEM ) . Statistical significance was computed using the Student’s two-sided t-test and p-values <0 . 05 were considered significant . No statistical methods were used to estimate the sample size and no animals were excluded .

Proteins called transcription factors bind to sections of DNA known as regulatory elements to activate or deactivate nearby genes . In animals , transcription factors typically have two sections: a “DNA-binding domain” that attaches to DNA , and an “effector domain” that is responsible for interacting with other proteins to regulate the gene’s expression . Rhodopsin is a gene that encodes the instructions needed to make a light-sensitive protein in the eyes of humans and other animals . Botta et al . have now used this gene as an example to investigate whether proteins that contain a DNA-binding domain – but not an effector domain – can repress gene expression . The experiments show that only a small section of the regulatory elements in the human Rhodopsin gene is actually required for the gene to be expressed . Botta et al . designed an artificial protein – referred to as ZF6-DB – that is able to bind to this section of DNA . The binding of ZF6-DB to this short DNA section was sufficient to switch off a Rhodopsin gene in living pig cells , and , unlike conventional transcription factors , seemed to have minimal impact other genes . Next , Botta et al . used a virus to insert both the gene that encodes ZF6-DB and a normal copy of Rhodopsin into pigs . In these animals , ZF6-DB switched off the existing copy of Rhodopsin , but not the inserted copy so the cells produced a working form of the light-sensitive protein . Further experiments were carried out in mice that have both a faulty version and a normal copy of the Rhodopsin gene . ZF6-DB switched off the faulty Rhodopsin gene , which allowed the normal Rhodopsin gene to work without any interference from the faulty copy . Mutations in Rhodopsin can cause an eye disease that leads to severe loss of vision in humans . These new findings could now guide future efforts to develop treatments for people with this condition . It will also be important to investigate how ZF6-DB binds to the regulatory elements in the Rhodopsin gene and whether a similar strategy could be used to alter the expression of other genes .

2016

Rhodopsin targeted transcriptional silencing by DNA-binding

A longstanding question is how influenza virus evolves to escape human immunity , which is polyclonal and can target many distinct epitopes . Here , we map how all amino-acid mutations to influenza’s major surface protein affect viral neutralization by polyclonal human sera . The serum of some individuals is so focused that it selects single mutations that reduce viral neutralization by over an order of magnitude . However , different viral mutations escape the sera of different individuals . This individual-to-individual variation in viral escape mutations is not present among ferrets that have been infected just once with a defined viral strain . Our results show how different single mutations help influenza virus escape the immunity of different members of the human population , a phenomenon that could shape viral evolution and disease susceptibility . Infection of humans with influenza virus elicits potent neutralizing antibodies targeting the viral hemagglutinin ( HA ) protein . These antibodies provide long-lasting immunity against the viral strain that elicited them ( Couch , 1975; Davies et al . , 1982; Couch and Kasel , 1983; Yu et al . , 2008; Krammer , 2019 ) . Unfortunately , the effectiveness of this immunity against future strains is rapidly degraded by viral antigenic evolution ( Bedford et al . , 2014; Smith et al . , 2004 ) , such that the typical human is infected by influenza virus roughly every 5 years ( Couch and Kasel , 1983; Kucharski et al . , 2015; Ranjeva et al . , 2019 ) . Classic studies of this evolutionary process demonstrated that it is easy to experimentally select mutant viruses that escape neutralization by individual monoclonal antibodies ( Yewdell et al . , 1979; Laver et al . , 1979; Webster and Laver , 1980; Gerhard et al . , 1981 ) . But these same studies also found that monoclonal antibodies target a variety of non-overlapping regions on HA , such that no single viral mutation can escape a mix of antibodies targeting distinct regions ( Yewdell et al . , 1979; Laver et al . , 1979; Webster and Laver , 1980; Caton et al . , 1982 ) . This work therefore posed a perplexing question: given that human immunity is polyclonal , how does influenza virus evolve to escape all the myriad antibody specificities in human sera ? Two primary explanations have been suggested for this conundrum . One explanation proposes that polyclonal immunity simply selects mutations that provide generalized neutralization resistance by increasing receptor avidity , without necessarily abrogating antibody binding ( Yewdell et al . , 1986; Hensley et al . , 2009 ) . Another explanation proposes that human immunity is sufficiently focused on one epitope in HA that single viral mutations can appreciably reduce binding by the mix of antibodies in polyclonal sera . This latter explanation has been shown to be true for the immunity of certain individuals against H1N1 influenza virus ( Li et al . , 2013; Linderman et al . , 2014; Huang et al . , 2015; Davis et al . , 2018 ) . It is also supported by mass spectrometry studies showing that HA-binding antibodies after vaccination are often dominated by relatively few ‘clonotypes’ ( Lee et al . , 2016; Lee et al . , 2019 ) , and studies with ferret sera suggesting that mutations to a small number of HA residues near the receptor binding pocket can lead to large antigenic changes ( Koel et al . , 2013 ) . Of course , how polyclonal immunity exerts selection on influenza virus may depend on the serum in question . However , although several studies have selected individual viral mutants with enhanced resistance to neutralization by human sera ( Davis et al . , 2018; DeDiego et al . , 2016; Li et al . , 2016 ) , there have been no systematic analyses of the immune selection exerted by sera from different individuals . Here , we harness mutational antigenic profiling ( Doud et al . , 2017; Doud et al . , 2018 ) to map how all amino-acid mutations to HA affect neutralization of H3N2 influenza virus by human serum . We show that human serum can select single mutations that reduce viral neutralization by over an order of magnitude . Although the escape mutations usually occur in a similar region on HA’s globular head , there is remarkable person-to-person variation in their antigenic effect: we identify mutations that reduce neutralization by >10-fold for one individual’s serum but have little effect for another individual’s serum . Our work suggests that person-to-person variation in the fine specificity of anti-influenza immunity may play a major role in shaping viral evolution and disease susceptibility . Prior work studying immune selection from polyclonal sera has used escape-mutant selections , which simply isolate individual mutant viruses with reduced neutralization sensitivity . As a more comprehensive alternative , we recently developed mutational antigenic profiling ( Doud et al . , 2017; Doud et al . , 2018 ) . As illustrated in Figure 1 , this approach quantifies how every amino-acid mutation to HA affects viral neutralization . Here , we perform mutational antigenic profiling using mutant virus libraries with the HA from A/Perth/16/2009 ( Lee et al . , 2018 ) , which was the H3N2 component of the influenza vaccine from 2010 to 2012 ( WHO , 2019 ) . To validate mutational antigenic profiling for these viral libraries , we first performed selections with four monoclonal antibodies isolated from humans who had received the 2010–2011 trivalent influenza vaccine ( Zost et al . , 2017; Henry Dunand et al . , 2015 ) . These four antibodies were 037-10036-5A01 , 034-10040-4C01 , 028-10134-4F03 , and 041-10047-1C04 , hereafter referred to as ‘5A01’ , ‘4C01’ , ‘4F03’ , and ‘1C04’ , respectively . Based on binding assays to a small HA mutant panel , two antibodies target near the receptor-binding pocket , while two antibodies target lower on HA’s head ( Zost et al . , 2019 ) . For each antibody , we performed at least two replicates of mutational antigenic profiling using independently generated virus libraries and an antibody concentration such that <10% of the library retained infectivity after neutralization ( Figure 2—figure supplement 1 ) . Note that the magnitude of the measured immune selection depends on the strength of antibody selection ( Doud et al . , 2017 ) , so while heights of letters in our maps of immune selection ( i . e . logo plots like the one at right of Figure 1 ) can be compared within a given map , y-axis scales are not directly comparable across maps . Importantly , the selected mutations were consistent across the biological replicates with independently generated mutant virus libraries ( Figure 2—figure supplement 2 ) , confirming that our approach systematically maps the antigenic effects of all mutations rather than simply selecting one-off viral mutants . The strongly selected mutations occurred mostly at sites that are classically categorized as antigenic region B ( Supplementary file 1; Wiley et al . , 1981; Wu and Wilson , 2017 ) . However , although the antibodies all targeted the same antigenic region of HA , they selected different escape mutations ( Figure 2A ) . These antibody-to-antibody differences in escape mutations were validated by traditional neutralization assays ( Figure 2B ) . For instance , K160T reduces neutralization by antibody 5A01 but has no effect on neutralization by antibody 4C01 despite the fact that both antibodies target a similar region of HA ( Figure 2E ) . The two antibodies that we expected to target lower on HA’s head indeed selected mutations in this portion of HA , mostly at sites classically categorized as antigenic regions C , D , and E ( Supplementary file 1 ) . But again the specific mutations selected by each antibody differed ( Figure 2C , D , F ) . Overall , these results demonstrate that mutational antigenic profiling accurately maps antibody selection on the Perth/2009 HA , and underscore the observation that different antibodies targeting an apparently similar epitope often have distinct escape mutations ( Doud et al . , 2017; Dingens et al . , 2019 ) . We next mapped immune selection from polyclonal human sera . We reasoned that it would be most informative to examine serum collected during the time when the Perth/2009 virus was circulating in the human population . We therefore screened the neutralizing activity of sera collected from 16 healthy adults between 2008 and 2010 , and identified four sera that completely neutralized virus with the wild-type Perth/2009 HA at a dilution of ≥1:40 ( which is the HAI titer traditionally assumed to reduce risk of infection; Hobson et al . , 1972 ) . Notably , this pre-screening means that all the sera that we characterized had high neutralizing activity , with inhibitory concentrations 50% ( IC50s ) ranging from ~1:800 to ~1:3000 ( Supplementary file 2 ) . No information on influenza vaccination or infection history was available for these four individuals , who ranged in age from 21 to 65 years old at the time of collection . We performed mutational antigenic profiling using serum dilutions chosen so that ~5% of the mutant virus library survived neutralization , although the exact percentage varied slightly among sera and replicates ( Figure 3—figure supplement 1 ) . We targeted this amount of neutralization to impose strong selection while still allowing identification of mutations that might mediate only partial escape . Despite being polyclonal , each serum strongly selected single escape mutations ( Figure 3A ) . The selected mutations occurred at a relatively small number of sites , predominantly in the portion of HA classically categorized as antigenic region B ( Figure 3A , C and Supplementary file 1 ) . The sites of selection were reproducible across three biological replicates using independently generated mutant virus libraries ( Figure 3—figure supplement 2 ) . To validate the mutations identified by the antigenic profiling , we performed neutralization assays on viruses carrying individual mutations ( Figure 3B ) . In all cases , the most strongly selected mutation in the profiling had a large antigenic effect in a neutralization assay . For three individuals ( the 21-year-old , 64-year-old , and 65-year-old ) , the strongest escape mutation had a >10-fold effect on neutralization , and for the remaining individual ( the 53-year-old ) the strongest escape mutation had a ∼5-fold effect ( Figure 3B ) . Amazingly , the antigenic effect of the strongest escape mutant from three of the polyclonal sera was comparable to that of strongest escape mutant from the monoclonal antibody 5A01 ( compare Figure 2B to Figure 3B ) . For most of the sera , the strongest escape mutation completely escaped neutralization at the serum concentration used in the mutational antigenic profiling ( see dashed lines in Figure 3B ) . These results show that single mutations can dramatically reduce viral sensitivity to neutralization by polyclonal human sera . However , the exact mutations that mediated escape differed markedly across sera ( Figure 3A ) . For instance , F193D escaped viral neutralization by the serum from the 21-year-old individual by >10-fold—but had no effect for the serum from the 64-year-old ( Figure 3B ) . Similarly , F159G escaped neutralization by the 64-year-old’s serum but had minimal effect on neutralization by the 53-year-old ( Figure 3B ) . Figure 3A , B shows numerous other examples of such person-to-person variation in viral escape mutations . In fact , the only viral mutation that had a consistent antigenic effect across individuals is K160T , which moderately enhanced neutralization resistance to all four sera . This serum-to-serum variation suggests that we are mapping serum-specific antigenic mutations rather than avidity mutants that generally enhance neutralization resistance ( Yewdell et al . , 1986; Hensley et al . , 2009 ) . It also shows that serum-escape mutations in the same classically defined antigenic region can have widely different effects across individuals . To confirm that our serum-escape maps are replicable over short timeframes during which there are not expected to be large changes in underlying immunity , we examined a second serum sample from the 53-year-old individual collected two months after the sample mapped in Figure 3 . There is no indication that the individual was infected or vaccinated during these 2 months ( which spanned from January to March of 2009 ) , and an absence of immune exposure is further supported by the fact that there was no rise in serum neutralizing titer during this time ( Supplementary file 2 ) . Figure 4 shows that the maps of immune selection were highly similar for the two samples , consistent with the expectation that the specificity of the serum was stable over the 2 months . Because vaccination can shape anti-influenza serum responses ( Fonville et al . , 2014; Zost et al . , 2017; Levine et al . , 2019 ) , we next performed mutational antigenic profiling with sera collected from four individuals pre-vaccination and 1 month post-vaccination . These individuals , who ranged in age from 25 to 49 years , received the 2015–2016 vaccine , for which the H3N2 component was A/Switzerland/9715293/2013 ( WHO , 2019 ) . Because these sera were collected 6 years after 2009 from individuals vaccinated with an antigenic successor of Perth/2009 , we anticipated that it might be more difficult to select escape mutations in our Perth/2009 library , since the sera is likely to have immunity both to Perth/2009-like viruses and their naturally occurring antigenic drift variants . The serum from one individual ( the 25-year-old ) strongly selected escape mutants in the Perth/2009 library even pre-vaccination ( Figure 5A ) . This individual’s serum also had the most potent pre-vaccination neutralizing activity against Perth/2009 , with an IC50 of ∼1:800 ( Figure 5B ) . The mutation most strongly selected by the pre-vaccination serum ( F159G ) reduced neutralization by over an order of magnitude ( Figure 5B ) . Vaccination of this individual substantially enhanced the serum’s potency , dropping the IC50 to ∼1:10 , 000—but F159G continued to have the largest effect on neutralization , still reducing the IC50 by over an order of magnitude ( Figure 5A , B ) . The second most potent pre-vaccination serum ( that of the 29-year-old , which had an IC50 of ∼1:150 ) also perceptibly selected an antigenic mutation ( F159G ) , but the effect of the mutation was much smaller ( Figure 5A , B ) . Vaccination enhanced the serum potency by ∼20-fold , and also slightly shifted its specificity ( Figure 5A , B ) . Prior to vaccination , F159G but not K144E had a small antigenic effect , whereas the finding was reversed post-vaccination . Notably , both F159 and K144 are mutated in the Switzerland/2013 virus relative to Perth/2009 . However , the fact that no mutations greatly alter the IC50 of this individual’s serum suggest that it is not narrowly focused on any specific epitope on the Perth/2009 HA . The remaining two individuals ( the 48- and 49-year-old ) had low serum potency before vaccination ( Figure 5B ) , such that we were unable to exert sufficient immune selection to map escape mutations from their pre-vaccination sera ( Figure 5A and Figure 5—figure supplement 1 ) . Vaccination enhanced the serum potency of both individuals ( Figure 5B ) . For the 48-year-old , this enhanced neutralizing activity was strongly focused , with the K189D mutation decreasing the IC50 of the post-vaccination serum by an order of magnitude ( Figure 5A , B ) . But for the 49-year-old , the enhanced neutralizing activity was not associated with increased selection of any HA mutations ( Figure 5A ) , suggesting that either the vaccine response was not dominated by any single HA specificity , or that it targeted epitopes that are not tolerant of viral escape mutations . These results show that it is possible to select strong escape mutants even from serum collected well after the Perth/2009 virus circulated from individuals vaccinated with an antigenic successor of this virus . In some cases vaccination shifts the specificity of the serum , whereas in other cases it primarily boosts existing specificities , consistent with prior work using other approaches ( Fonville et al . , 2014; Lee et al . , 2016; Lee et al . , 2019; Henry et al . , 2019 ) . As in the previous section , the strong escape mutations are predominantly but not exclusively in antigenic region B of HA ( Figure 5C and Supplementary file 1 ) —but again the effects of specific mutations vary markedly among sera . The antigenicity of influenza viral strains is currently characterized mostly using sera from ferrets that have been infected just once with a single viral strain ( Smith et al . , 2004 ) . In contrast , humans have complex exposure histories that may influence their antibody response ( Li et al . , 2013; Linderman et al . , 2014; Andrews et al . , 2015b; Gostic et al . , 2016; Cobey and Hensley , 2017 ) . To compare selection from the sera of humans and singly-infected ferrets , we performed mutational antigenic profiling using sera from five ferrets . Three ferrets were infected with the Perth/2009 viral strain at the University of Pittsburgh . The other two ferrets were infected at the World Health Organization ( WHO ) Collaborating Centre in Melbourne , one with Perth/2009 and one with A/Victoria/361/2011 ( the immediate antigenic successor of Perth/2009 in the influenza vaccine; WHO , 2019 ) . We used ferret sera from different labs and viral strains to sample across factors that might affect serum specificity . In stark contrast to the person-to-person variation of the human sera , the maps of immune selection were very similar for all post-infection ferret sera ( Figure 6A ) . This was true even for the ferret infected with Victoria/2011 rather than Perth/2009 virus: visual comparison of Figure 6A and Figure 3A suggests that the differences between ferrets infected with these two antigenically distinct H3N2 viruses are smaller than the differences among nearly all the human sera . To statistically confirm this observation , we calculated the beta diversity of the site-level selection for each group of serum using the Simpson index ( Jost , 2007 ) . Higher beta diversity indicates that the sera in a group show more variation in the HA sites where they exert selection . This analysis confirmed that the ferret sera ( beta diversity of 1 . 12 ) were more consistent in the sites where they exerted selection than the human sera from either 2009–2010 or post-vaccination from 2015 ( beta diversity of 1 . 38 and 1 . 60 , respectively ) . The strongest selection from the ferret sera focused on sites 189 and 193 , with mutations K189D and F193D tending to have a ∼5 to 10-fold effect on neutralization ( Figure 6B ) . Prior work has also noted that these two sites are important for antigenic recognition by ferret sera ( Koel et al . , 2013 ) . Mutations at a handful of additional sites ( such as 142 , 144 , and 222 ) were also modestly selected by some ferret sera ( Figure 6A , B ) . Although the ferret sera was similar to the human sera in focusing mostly on antigenic region B of HA ( Figure 6C and Supplementary file 1 ) , there were notable differences in the specific sites where mutations were selected . The site where mutations were most strongly selected by all the ferret sera ( site 189 ) had an appreciable antigenic effect for only one of the human sera . The other site where mutations were consistently and strongly selected by the ferret sera ( site 193 ) had an antigenic effect for less than half the human sera . The converse was also true: many sites of mutations strongly selected by human sera ( such as 157 , 159 , and 160 ) had little antigenic effect for the ferret sera . These findings highlight major differences in the fine specificity of immune focusing between ferrets that have been infected with a single viral strain and humans with complex exposure histories . The preceding sections show that polyclonal sera often select single antigenic mutations—a phenomenon more typically associated with monoclonal antibodies . To test how a single antibody can contribute to the selection of escape mutations by polyclonal serum , we spiked an antibody that targets lower on HA’s head into serum that selects mutations at the top of HA’s head . We spiked the antibody into the serum at three concentrations . In all cases , we kept the serum at a concentration where ∼4% of the viral library survived neutralization by serum alone . The three antibody concentrations were chosen such that in the presence of antibody alone , the percents of the viral library that survived neutralization were ~25% ( low concentration ) , ~5% ( mid concentration ) , and ~2% ( high concentration ) . In the actual selections with the mix of serum and antibody , there was combined selection from the serum and antibody , so that the percents of the library that survived neutralization were lower than those due to the serum or antibody alone . Specifically , the percents of the library that survived selection with the mix were 1 . 7% ( low antibody ) , 0 . 16% ( mid antibody ) , and 0 . 015% ( high antibody ) ; see Figure 7—figure supplement 1 for details . These selections therefore span a range in which the antibody is a modest versus dominant contributor to the overall neutralizing activity of the mix . Even when the serum was spiked with the lowest concentration of antibody , the mutational antigenic profiling showed signals of selection at sites targeted by the antibody ( Figure 7A ) . At the middle antibody concentration , selection from the antibody exceeded that from the serum , although both were still apparent ( Figure 7A ) . At the highest antibody concentration , selection at antibody-targeted sites completely overwhelmed selection at serum-targeted sites . These results do not imply a loss of serum activity when antibody is added , but rather indicate that at high concentration the antibody completely dominates the neutralizing activity of the mix , and so exerts all the measurable immune selection . But the dominance of a few sites in one of our antigenic maps does not imply that there are no antibodies targeting other sites—once a mutation escapes the dominant antibody specificity , mutations at other more weakly targeted sites would begin to have measurable antigenic effects . In other words , our maps show the relative effects of different mutations on viral neutralization , not the absolute effect of any given mutation on antibody binding . In addition , there could be non-additive effects if antibodies sterically hinder each other or mutations induce conformational changes in HA . Part of the reason the antibody so readily dominates the selection in Figure 7A may be that the strongest serum-escape mutations ( F159G and F193D ) actually sensitize the virus to neutralization by the antibody ( Figure 7B ) , a result reminiscent of earlier work showing synergistic effects of mixing antibodies ( Lubeck and Gerhard , 1982; Webster and Laver , 1980 ) . The loss of antibody selection at certain sites when serum is present ( e . g . 220 and 259 in Figure 7A ) also suggest possible synergistic effects of mixing antibody and serum . Overall , these spike-in experiments show that a single antibody can dominate selection if it comprises half or more of a serum’s overall potency . Of course , these experiments cannot reveal if the focused selection from human sera is actually driven by a single antibody versus a collection of antibodies targeting a similar epitope . If our mutational antigenic profiling reflects real antigenic pressure on human influenza virus , then we would expect to observe changes at sites in HA that our experiments mapped as being under selection . To test if this is true , we examined the natural evolution of human H3N2 influenza HA since 2007 at sites with clear signals of antigenic selection from at least one human serum ( these are the 16 sites shown in the logo plots of Figure 3A and Figure 5A ) . There has been substantial amino-acid evolution ( new variants reaching ≥ 5% frequency ) at nine of the sites strongly selected by the human sera ( Figure 8 ) . Across all sera that strongly selected mutations , the single largest-effect mutations occurred at five sites: 144 , 157 , 159 , 189 , and 193 ( Figure 3A and Figure 5A ) . During the evolution of human H3N2 influenza virus , new amino-acid variants have reached >10% frequency at all these sites ( Figure 8 ) . Notably mutations at each of these five sites are selected by only some of the human sera that we examined . The one HA mutation with an antigenic effect across almost all human sera we examined was K160T , which consistently caused a moderate increase in neutralization resistance ( Figure 3A , B ) . This mutation introduces a N-linked glycosylation motif near the top of HA’s head . K160T appeared among natural human H3N2 viruses in late 2013 and is now present in the majority of circulating viruses ( Figure 8 ) . In addition , K160T and mutations at three of the sites strongly selected by just some sera ( 144 , 159 , and 193 ) distinguish the 3C2 . A and 3C3 . A clades that currently circulate in humans ( Bedford and Neher , 2018 ) . Although the number of sera that we have characterized is too small to draw conclusions about population-level immunity , it is likely that evolutionary selection on a viral mutation depends both on how many individuals have immunity that is affected by the mutation and the magnitude of the antigenic effect in those individuals . Notably , the amino-acid that emerges at a site in nature is not always the one that our experiments map as being under the strongest immune selection . There are several possible reasons: First , our experiments probe all amino-acid mutations but natural evolution is mostly limited to single-nucleotide changes . For instance , L157D is the largest-effect mutation at site 157 ( Figure 3 ) but L157S appears in nature ( Figure 8 ) —likely because S but not D is accessible from L by a single-nucleotide change . Second , natural evolution selects for viral replication and transmission as well as immune escape . Finally , the context of a mutation may influence its effect: for instance , F159Y has spread in nature but was not selected in our experiments , perhaps because in nature this mutation co-occurred with secondary changes such as N225D ( Chambers et al . , 2015; Jorquera et al . , 2019 ) . Importantly , it is now feasible to also make high-throughput measurements of how mutations to H3N2 HA affect viral replication ( Lee et al . , 2018 ) and epistatic interactions among sites ( Wu et al . , 2018 ) , so it may eventually be possible to build evolutionary models that incorporate all these factors ( Luksza and Lässig , 2014; Neher et al . , 2016 ) . We have mapped the selection that polyclonal human sera exert on all amino-acid mutations to a H3N2 influenza virus HA . Many sera are highly focused , and select single viral mutations that reduce neutralization by over an order of magnitude . This basic finding that human serum can be focused is consistent with other recent work ( Li et al . , 2013; Linderman et al . , 2014; Huang et al . , 2015; Davis et al . , 2018 ) . But because our experiments mapped selection on all mutations , we were able to systematically quantify the extent of focusing and compare its targets across sera . A striking result is that the targets of serum selection vary widely from person to person . Viral mutations that greatly reduce neutralization by one individual’s serum sometimes have no effect for another individual’s serum , which is instead affected by different mutations . Our results suggest a model in which a virus with a single mutation could be strongly favored in pockets of the human population with immunity focused on that site . Does this type of selection actually drive influenza virus evolution ? Many of the mutations mapped in our experiments are at sites that have recently substituted in nature , suggesting that selection from focused human sera may be relevant to viral antigenic drift . However , it is important to note that we performed our selections using near-neutralizing concentrations of relatively potent human sera . There is evidence that sub-neutralizing serum concentrations more often select generalized resistance mutations that increase receptor avidity ( Yewdell et al . , 1986; Hensley et al . , 2009 ) . Both types of selection may occur in nature , and understanding their roles in shaping influenza virus evolution will require comparing the serum focusing of individuals with the viral strains that actually infect them . The sera that we profiled predominantly focus on the portion of HA classically categorized as antigenic region B , consistent with other studies reporting that this region is immunodominant in recent H3N2 strains ( Popova et al . , 2012; Chambers et al . , 2015; Broecker et al . , 2018 ) . However , our results highlight the limitations of subdividing HA into broad antigenic regions . These regions were originally defined based on the idea that a mutation selected by one antibody targeting a region often also abrogated binding of other antibodies targeting that region ( Webster and Laver , 1980; Wiley et al . , 1981; Caton et al . , 1982 ) . But our results show that different mutations in the same antigenic region can have very different effects across sera . The specificity of human sera is therefore more fine-grained than the classical categorization of HA into antigenic regions . It is important to note that our experiments only mapped in vitro immune selection exerted by neutralizing anti-HA antibodies in serum . During actual human infections , antibodies to HA can also exert selection by non-neutralizing mechanisms such as antibody-dependent cellular cytoxicity ( He et al . , 2016; Vanderven et al . , 2018 ) , activation of complement ( Rattan et al . , 2017 ) , and inhibition of the viral neuraminidase ( Kosik et al . , 2019; Chen et al . , 2019 ) . In addition , antibodies in serum are mostly of the IgG class , but IgA antibodies predominate in the upper airway mucosa where human influenza virus typically replicates ( Gould et al . , 2017 ) . But despite these caveats , many studies have shown that neutralizing serum activity is a strong correlate of protection against influenza ( Hobson et al . , 1972; Wagner et al . , 1987; Krammer , 2019 ) , so mapping selection from such serum provides an important if incomplete picture of the selection that human immunity imposes on the virus . Why is the neutralizing activity of human polyclonal sera often so focused ? One possibility is that very few distinct antibodies are responsible for the anti-HA activity of any given serum . Indeed , multiple studies have shown that relatively few clonotypes comprise the majority of the anti-HA serum repertoire of a single individual ( Andrews et al . , 2015a; Lee et al . , 2016; Lee et al . , 2019 ) . Our findings further suggest that either most of these clonotypes target the same sites in HA , or that the serum’s neutralizing activity is due to very few of the clonotypes . In support of the second idea , recent work has shown that as an individual ages , much of the antibody response is diverted to non-neutralizing epitopes because sites targeted by potently neutralizing antibodies change during viral evolution , while non-neutralizing epitopes are more often conserved ( Henry et al . , 2019; Ranjeva et al . , 2019 ) . Prior work has shown that exposure history shapes human immunity to influenza ( Li et al . , 2013; Andrews et al . , 2015b; Gostic et al . , 2016; Cobey and Hensley , 2017 ) , and variation in exposure history is a plausible cause of the person-to-person variation we observe in the antigenic effects of viral mutations . A role for exposure history is further suggested by the lack of variation in immune selection among ferrets that have been experimentally infected just once with a defined viral strain . However , our results do not formally distinguish whether the differences between human and ferret sera are due to exposure history or species-to-species variation in immunodominance hierarchies ( Liu et al . , 2018; Angeletti et al . , 2017; Fonville et al . , 2016; Altman et al . , 2015 ) —rigorously addressing this question will require mapping the immune selection exerted by children that have been infected with influenza virus just once . In either case , the differences between human and ferret sera underscore the shortcomings of using singly infected ferrets as the primary tool for characterizing viral antigenicity , as has previously been shown for H1N1 ( Li et al . , 2013 ) . Overall , our work uses a new approach to show that polyclonal human sera can strongly select single antigenic mutations in HA , and that the effects of these mutations vary greatly person to person . Expanding this approach should shed further light on the causes of this variable focusing , and its implications for viral evolution and disease susceptibility . Deep sequencing data are on the Sequence Read Archive under BioSample accession numbers SAMN10183083 and SAMN11310465 ( viral libraries selected with monoclonal antibodies ) , SAMN10183146 ( monoclonal antibody selection controls ) , SAMN11310372 ( viral libraries selected with human sera ) , SAMN11310371 ( viral libraries selected with ferret sera ) , SAMN11341221 ( viral libraries selected with human sera spiked with monoclonal antibody ) , and SAMN11310373 ( serum selection controls ) . The computer code used to analyze the data and generate the paper figures is available on GitHub ( Lee and Bloom , 2019; copy archived at https://github . com/elifesciences-publications/map_flu_serum_Perth2009_H3_HA ) . Key processed data are tracked in this GitHub repository as detailed in the repository’s README file , and are also available in the Supplementary files for this manuscript . All sites are referred to in H3 numbering unless indicated otherwise . The signal peptide is in negative numbers , the HA1 subunit in plain numbers , and the HA2 subunit in numbers denoted with ‘ ( HA2 ) . ' Sequential 1 , 2 , … numbering of the Perth/2009 H3 HA can be converted to H3 numbering simply by subtracting 16 for the H1 subunit , and subtracting 345 for the HA2 subunit . A file mapping sequential 1 , 2 , … numbering to H3 numbering is at https://github . com/jbloomlab/map_flu_serum_Perth2009_H3_HA/blob/master/data/H3renumbering_scheme . csv . The ‘wild-type’ Perth/2009 HA used in this study is that from Lee et al . ( 2018 ) . This HA has two amino-acid mutations ( G78D and T212I ) relative to the most common Perth/2009 HA sequences in Genbank ( there are several variants in Genbank ) . These two mutations improve viral growth in cell culture ( Lee et al . , 2018 ) . The sequence of HA with these mutations is provided in Lee et al . ( 2018 ) and at https://github . com/jbloomlab/map_flu_serum_Perth2009_H3_HA/blob/master/data/Perth09_HA_reference . fa . Unless otherwise indicated , when this paper refers to the ‘Perth/2009 HA’ it means this HA . As described below , all other genes in this virus were derived from A/WSN/1933 , a H1N1 strain that grows efficiently in cell culture and was heavily adapted from its original human isolate predecessor by passaging in mouse brain in the lab ( Stuart-Harris , 1939; Sun et al . , 2010 ) . The monoclonal antibodies in Figure 2 are described in Zost et al . ( 2019 ) , Zost et al . ( 2017 ) and Henry Dunand et al . ( 2015 ) . These antibodies were isolated from peripheral blood mononuclear cells of human donors 7 days post-vaccination with the 2010–2011 influenza vaccine containing the A/Victoria/210/2009 strain as the H3N2 component , using the approach in Smith et al . ( 2009 ) . The VH and VL chains were amplified using single-cell RT-PCR , and cloned into human IgG expression vectors . To produce the monoclonal antibodies , 293 T cells were transfected with plasmids encoding the heavy and light chains , and the antibodies were purified using protein A/G affinity purification . The human sera in Figure 3 came from the Infectious Disease Sciences Biospecimen Repository at the Vaccine and Infectious Disease Division ( VIDD ) of the Fred Hutchinson Cancer Research Center . These sera were collected in Seattle , WA , from healthy prospective bone marrow transplant donors , who provided written consent for the use of their sera in research , and the sera were banked at –20°C . We used neutralization assays to screen sera collected from 16 individuals from 2008 and 2010 , and selected the four that neutralized virus with the Perth/2009 HA at ≥ 1:40 dilution . No information was available on the influenza vaccination or infection status of these individuals . The human sera in Figure 5 were collected at the Wistar Institute , Philadelphia , PA , from four individuals pre-vaccination and four weeks post-vaccination with the 2015–2016 Northern Hemisphere influenza vaccine . The H3N2 component of this vaccine was A/Switzerland/9715293/2013 , which is an antigenic successor of Perth/2009 by several vaccine updates: Perth/2009 was the Northern Hemisphere vaccine strain in 2010–2012 , A/Victoria/361/2011 was the strain in 2012–2014 , A/Texas/50/2012 was the strain in 2014–2015 , and Switzerland/2013 was the strain in 2015–2016 . The Wistar Institute IRB approved the study collecting these samples . The dates of collection of the human sera are rounded to the nearest year in this paper . The serum samples denoted ‘ferret-Pitt-#' were collected from three ferrets pre-infection and again 23 days post-infection at the University of Pittsburgh . Each ferret was infected with Perth/2009 ( H3N2 ) virus that carries the exact same HA sequence used for the studies in this paper . Ferret studies were performed under an approved IACUC protocol ( #16077170 ) at the University of Pittsburgh , an AAALAC accredited facility . The serum samples denoted ‘ferret-WHO*’ were collected from two ferrets post-infection at the World Health Organization Collaborating Centre in Melbourne , Australia . One ferret was infected with the Collaborating Centre’s version of the Perth/2009 H3N2 virus , and the second ferret was infected with the Collaborating Centre’s version of the A/Victoria/361/2011 H3N2 strain . The sera were collected in bleeds performed 14 days after intranasal infection with virus . Experiments were conducted with approval from the University of Melbourne Biochemistry and Molecular Biology , Dental Science , Medicine , Microbiology and Immunology , and Surgery Animal Ethics Committee , in accordance with the NHMRC Australian code of practice for the care and use of animals for scientific purposes . All sera were treated with receptor-destroying enzyme ( RDE ) and heat-inactivated before use in our experiments , using a protocol adapted from Zost et al . ( 2017 ) . The RDE treatment was to ensure that the viral libraries would not bind to residual sialic acids present in the serum . One vial of lyophilized RDE II ( Seiken , Cat No . 370013 ) was first resuspended in 20 mL PBS . We then incubated 100 μL of serum with 300 μL of RDE solution at 37°C for 2 . 5 hr . We next heat-inactivated the serum and RDE by incubating at 55°C for 30 min . Finally , we centrifuged the serum at 20 , 000 x g for 20 min to pellet any precipitated material , collected the supernatant , aliquoted , and stored at –80°C . Neutralization assays were performed using influenza viruses carrying GFP in the PB1 segment . The ‘wild-type’ Perth/2009 viruses were generated using reverse genetics with the following plasmids: pHW-Perth2009-HA-G78D-T212I ( Lee et al . , 2018 ) , pHH-PB1flank-eGFP ( which encodes GFP on the PB1 segment; Bloom et al . , 2010 ) , and pHW181-PB1 , pHW183-PA , pHW185-NP , pHW186-NA , pHW187-M , and pHW188-NS ( which encode genes from A/WSN/1933 for the other six segments; Hoffmann et al . , 2000 ) . The individual amino-acid mutant viruses tested in the neutralization assays were generated from the same plasmids except the indicated mutation was introduced into the pHW-Perth2009-HA-G78D-T212I plasmid . The ‘syn’ mutant shown in Figure 3 has a synonymous F193F mutation . To generate the viruses , we transfected a co-culture of 4×105 293T-CMV-PB1 ( Bloom et al . , 2010 ) and 0 . 5×105 MDCK-SIAT1-CMV-PB1-TMPRSS2 ( Lee et al . , 2018 ) cells with the eight reverse genetics plasmids and the pHAGE2-EF1aInt-TMPRSS2-IRES-mCherry-W plasmid ( Lee et al . , 2018 ) . The transfections were performed in D10 media ( DMEM , supplemented with 10% heat-inactivated FBS , 2 mM L-glutamine , 100 U of penicillin per milliliter , and 100 μg of streptomycin per milliliter ) . Each well received a transfection mixture of 100 μL DMEM , 3 μL BioT transfection reagent , and 250 ng of each plasmid . We changed the media in each well to 2 mL IGM ( Influenza Growth Media , consisting of Opti-MEM supplemented with 0 . 01% heat-inactivated FBS , 0 . 3% BSA , 100 U of penicillin per milliliter , 100 μg of stremptomycin per milliliter , and 100 μg of calcium chloride per milliliter ) eight hours post-transfection . At approximately 56 hr post-transfection , transfection supernatants were harvested , clarified by centrifugation at 2000 xg for 5 min , aliquoted , and frozen at −80°C . To titer the GFP-carrying viruses , we plated 1×105 MDCK-SIAT1-CMV-PB1-TMPRSS2 cells per well in 12-well plates in IGM . We infected cells with dilutions of viral supernatant four hours after plating . At 16 hr post-infection , we chose wells that appeared to show 1% to 10% of cells that were GFP-positive , and determined the exact fraction of GFP-positive cells by flow cytometry to calculate the titer of infectious particles per μl . The neutralization assays were performed using these GFP-expressing viruses as described previously ( Hooper and Bloom , 2013; Doud et al . , 2018; see also Bloom and Lee , 2019; copy archived at https://github . com/elifesciences-publications/flu_PB1flank-GFP_neut_assay ) . All neutralization curves represent the mean and standard error of three replicate curves run on the same 96-well plate . The curve fits are Hill-like curves with the bottom and top constrained to zero and one , and were fit using the neutcurve Python package ( https://jbloomlab . github . io/neutcurve/ ) , version 0 . 3 . 0 . Detailed information on the curve fitting is available on the project GitHub repository at https://github . com/jbloomlab/map_flu_serum_Perth2009_H3_HA/blob/master/results/notebooks/analyze_neut . md . The curve-fit parameters ( e . g . , IC50s ) for all of the neutralization curves shown in this paper are in Supplementary file 2 . The dashed vertical lines in Figure 3B , Figure 4B , Figure 5 , and Figure 6 indicate the average concentration of serum used across the replicates . For mutational antigenic profiling of some replicates of antibodies 4C01 and 1C04 , we used the exact same mutant virus libraries described in Lee et al . ( 2018 ) . Specifically , we used those mutant virus libraries for the replicate labeled as ‘lib2’ for 4C01 and all replicates of 1C04 in Figure 2—figure supplement 2 . However , the titers of the viral libraries from Lee et al . ( 2018 ) were not sufficient for all the mutational antigenic profiling , so for most of the experiments in the current paper we generated new triplicate stocks of mutant virus libraries using influenza reverse-genetics ( Hoffmann et al . , 2000 ) . This required re-generating the mutant plasmids and mutant virus libraries as described below . These libraries have similar properties to those described in Lee et al . ( 2018 ) , but since they are not exactly the same we re-validated them by deep sequencing the mutant plasmid and mutant virus libraries . The full analysis of this deep sequencing along with plots showing all relevant statistics about the libraries is in Supplementary file 5 . To create the new mutant plasmid libraries , we used two rounds of codon mutagenesis ( Bloom , 2014; Dingens et al . , 2017 ) to introduce all of the possible codon mutations into the Perth/2009 HA in the pHH21 backbone ( Neumann et al . , 1999 ) . The plasmid libraries were generated independently in biological triplicate , starting from independent preps of the wildtype plasmid , using the protocol described in Lee et al . ( 2018 ) . The mutant amplicons were then cloned at high efficiency into the pHH21 vector using digestion with BsmBI , ligation with T4 DNA ligase , and electroporation into ElectroMAX DH10B competent cells ( Invitrogen 18290015 ) . We obtained >4 million transformants for each replicate library . We then scraped the plates , expanded the cultures in liquid LB + ampicillin at 37°C for 3 hr with shaking , and maxiprepped the cultures . Sanger sequencing of 29 randomly chosen clones showed an average mutation rate of 1 . 6 codon mutations per clone ( Supplementary file 5 ) . To generate the mutant virus libraries by reverse genetics , we transfected 40 wells of six-well plates for each library . Each well contained a co-culture of 5×105 293T-CMV-PB1 and 0 . 5×105 MDCK-SIAT1-TMPRSS2 cells in D10 media . We transfected with 250 ng each of pHH-mutant-HA library ( or wildtype control ) , the pHAGE2-EF1aInt-TCmut-P09-HA Perth/2009 HA protein expression plasmid ( which expresses the HA protein from wild type Perth/2009 HA ) , the pHW18* series of plasmids ( Hoffmann et al . , 2000 ) for all non-HA viral genes , and pHAGE2-EF1aInt-TMPRSS2-IRES-mCherry-W . The sequence of the pHAGE2-EF1aInt-TCmut-P09-HA protein expression plasmid is in Supplementary file 6 . We changed the media in each well to 2 mL IGM 8 hr post-transfection . At 45 hr post-transfection , the transfection supernatants were harvested , clarified by centrifugation at 2 , 000 xg for 5 min , aliquoted , frozen at −80°C , and titered in MDCK-SIAT1-TMPRSS2 cells . The titers were 2543 , 3162 , 1000 , and 4739 TCID50 per microliter for the three library replicates and the wild-type control , respectively . We passaged 1 . 125 x 106 TCID50 of the transfection supernatants for each library at an MOI of 0 . 005 TCID50 per cell . We did this by plating 3 x 106 MDCK-SIAT1-TMPRSS2 cells per layer of five 5-layered 875 cm2 flasks ( Corning , 353144 ) in D10 media , and allowed the cells to grow for 24 hr , at which time they were at ∼9 x 106 cells per layer . We then removed the D10 media from each flask , washed with 50 mL PBS and replaced the media with 130 mL per flask of an inoculum of 1 . 73 TCID50 of virus per microliter in IGM . At 3 hr post-infection , we replaced the inoculum with fresh IGM for each replicate . We then collected virus supernatant at 42 hr post-infection , clarified the supernatant by centrifuging at 2000 xg for 5 min , aliquoted , froze at −80°C , and titered in MDCK-SIAT1-TMPRSS2 cells . The titers were 1000 , 14677 , and 6812 TCID50 per microliter for the three library replicates , respectively . The mutant plasmids and mutant viruses were then deep sequenced as in Lee et al . ( 2018 ) to demonstrate that there was good coverage of mutations in both as described in Supplementary file 5 . These libraries were used for all the mutational antigenic profiling except for the subset of antibody replicates mentioned at the beginning of this subsection . As described above , all viral propagation used MDCK-SIAT1-CMV-TMPRSS2 cells . These cells were tested for Mycoplasma at the Fred Hutch Research Cell Bank core , and confirmed to be Mycoplasma negative . We performed the mutational antigenic profiling using the basic process described in Doud et al . ( 2017 ) . For each serum and for the serum-antibody spike-in experiments , we performed three biological replicates of mutational antigenic profiling each using an independently generated mutant virus library . For each antibody , we performed either two or three biological replicates each using an independently generated mutant virus library as indicated in Figure 2—figure supplement 2 . The reason that we only performed two biological replicates for some antibodies is that the noise is less for antibodies than sera . For the mutational antigenic profiling , we diluted each virus library to 106 TCID50 per mL , and incubated the virus dilution with an equal volume of antibody and/or serum at the intended dilution at 37°C for 1 . 5 hr . The dilutions for all samples are in Supplementary file 5 , and account for the initial 1:4 dilution of serum during the RDE treatment . These dilutions were generally chosen with the goal of having 1% to 10% of the virus library survive the serum or antibody treatment . We infected between 2×105 and 4×105 MDCK-SIAT1-TMPRSS2 cells with virus-antibody or virus-serum mix . At 15 hr post-infection , we extracted RNA from the cells , and then reverse-transcribed and PCR amplified full-length HA as in Lee et al . ( 2018 ) . We then deep sequenced these HAs using a barcoded-subamplicon sequencing strategy to ensure high accuracy . This general sequencing approach was first applied to viral deep mutational scanning by Wu et al . ( 2014 ) . The exact approach we used is described ( Doud and Bloom ( 2016 ) , with the primers for Perth/2009 given in Lee et al . ( 2018 ) ; a more general description of the approach is at https://jbloomlab . github . io/dms_tools2/bcsubamp . html . The sequencing was performed using 2 x 250 nucleotide paired-end reads on Illumina HiSeq 2500’s at the Fred Hutchinson Cancer Research Center Genomics Core . To estimate the overall fraction of virions in the library surviving immune selection , we used qRT-PCR against the viral NP and GAPDH , as described in Doud et al . ( 2017 ) . Briefly , we made duplicate 10-fold serial dilutions of each virus library to create a standard curve of infectivity . We then performed qPCR for the standard curve of infectivity as well as each library-selected sample . A linear regression line relating the logarithm of the viral infectious dose in the standard curve to the difference in Ct values between NP and GAPDH was used to interpolate the fraction surviving for each selection . The measured percent surviving for each library are in Supplementary file 4 , and are also plotted in figure supplements for each figure showing mutational antigenic profiling results . The deep sequencing data were analyzed using dms_tools2 ( Bloom , 2015 ) version 2 . 4 . 16 , which is available at https://jbloomlab . github . io/dms_tools2/ . Briefly , we first determined the counts of each codon at each site in both the immune-selected and mock-selected samples . These counts are at https://github . com/jbloomlab/map_flu_serum_Perth2009_H3_HA/tree/master/results/renumbered_codoncounts . These counts were then processed to compute the differential selection on each amino-acid mutation at each site , which is our measure of immune selection . The differential selection statistic is described in Doud et al . ( 2017 ) ( see also https://jbloomlab . github . io/dms_tools2/diffsel . html ) , and represents the log enrichment of each mutation relative to wildtype in the immune-selected sample versusa mock-selected control . The differential selection values for each replicate are at https://github . com/jbloomlab/map_flu_serum_Perth2009_H3_HA/tree/master/results/diffsel . To visualize the differential selection , we took the median across replicates of the differential selection for each mutation at each site—these median values are displayed in all logo plots . The numerical values of these across-replicate median differential selection values are in Supplementary file 7 . The figures in this paper visualize the differential selection in two ways . First , the line plots show the total positive differential selection at each site . Second , the logo plots show the differential selection for each positively selected amino acid at key sites . Note that in both cases , negative differential selection is not shown . These line and logo plots were created using the dmslogo software package ( version 0 . 2 . 3 ) , which is available at https://jbloomlab . github . io/dmslogo/ . We chose which sites to show in the logo plots in the figures by identifying strong or ‘significant’ sites of immune selection for each serum using the approach described in Dingens et al . ( 2019 ) ( excluding the pre-vaccination or pre-infection samples ) . Each figure panel then shows all sites that were ‘significant’ for any sera or antibody in that panel . This ‘significance’ calculation is heuristic , and involves using robust regression to fit a gamma distribution to all of the positive site differential selection values , equating the p-value to the fraction of the distribution ≥ that site’s differential selection , and then calling ‘significant’ sites that have a false discovery rate ≤ 0 . 05 . The code that performs this analysis is at https://jbloomlab . github . io/dms_tools2/dms_tools2 . plot . html#dms_tools2 . plot . findSigSel . In addition , logo plots for all sites for the across-replicate medians for each serum/antibody are in Supplementary file 8 . A detailed notebook with the code for all of the foregoing analyses along with explanations and many additional plots is at https://github . com/jbloomlab/map_flu_serum_Perth2009_H3_HA/blob/master/results/notebooks/analyze_map . md . To quantify the variation among sites of selection for different groups of sera ( i . e . , the ferret sera , the 2009–2010 human sera , and the 2015 post-vaccination human sera ) , we used the beta diversity statistic from ecology . For each serum in a group , we calculated the fraction of the total positive site differential selection attributable to each site . We then computed the beta diversity of this selection fraction among sites , using the Simpson index to quantify diversity , following the method of Jost ( 2007 ) . Specifically , let pr , s be the total fraction of all positive site differential selection for serum s that is attributable to site r ( so 1 = ∑rpr , s ) . The Simpson concentration index for serum s is then λs=∑r ( pr , s ) 2 . For a group of sera , the gamma diversity index ( total diversity across all sera in the group ) is λγ=∑r⟨pr⟩2 where ⟨pr⟩ is the average of pr , s across all sera s . The corresponding true gamma diversity is Dγ=1/λγ . Likewise , the alpha diversity index is simply the mean of the index for each serum , λα=⟨λs⟩ , and the true alpha diversity is Dα=1/λα . The beta diversity is then simply Dβ=Dγ/Dα . Code that implements this calculation has been added to dms_tools2 ( Bloom , 2015 ) version 2 . 4 . 16 ( see https://jbloomlab . github . io/dms_tools2/dms_tools2 . diffsel . html#dms_tools2 . diffsel . beta_diversity ) . The protein structures are all PDB 4O5N , which is the structure of HA from the A/Victoria/361/2011 ( H3N2 ) strain ( Lee et al . , 2014 ) . The residues are colored by the positive site differential selection values . The visualizations were generated using nglview ( Nguyen et al . , 2018 ) via the Python wrapper package dms_struct ( https://jbloomlab . github . io/dms_struct/ ) . Interactive mybinder instances of the notebooks that can be used to rotate and zoom in on the structures are available at the following weblinks: For Figure 8 , we first identified sites that were under strong or ‘significant’ selection from any of the human serum samples ( excluding the pre-vaccination samples in Figure 5 ) using the approach described above . There were 16 such sites; these are the ones shown in the logo plots in Figure 3A or Figure 5A . For each such site , we then examined the frequency of different amino-acid identities from 2007 to 2019 ( see https://github . com/jbloomlab/map_flu_serum_Perth2009_H3_HA/blob/master/results/notebooks/analyze_natseqs . md ) . This analysis identified nine sites where a new amino-acid identity reached at least 5% frequency . For these nine sites , we then took images of the amino-acid frequencies over time from the Nextstrain website ( https://nextstrain . org/ ) ( Hadfield et al . , 2018; Neher and Bedford , 2015 ) and used them to create Figure 8 .

The human immune system protects the body from repeat attacks by remembering past infections . However , a typical person comes down with the flu every five to seven years . This is because flu viruses rapidly evolve to bypass our defenses . So , after a few years , the viruses look so different that the immune system no longer recognizes them . The immune system recognizes flu viruses by producing proteins known as antibodies , which can bind to the virus and prevent it from infecting cells . Many of these antibodies bind to a protein on the surface of the virus called hemagglutinin , but each anti-flu antibody recognizes only a small region of the protein . This means that to escape recognition by a single antibody , all the virus needs to do is wait for a lucky mutation to change the part of hemagglutinin recognized by that antibody . But humans make many different antibodies . To escape them all , flu viruses would need lots of lucky mutations . So how do flu viruses keep winning the evolutionary lottery ? To answer this question , Lee et al . made all the possible individual mutations to the hemagglutinin protein of a human flu virus . A pool of these viruses was then exposed to the full mix of antibodies present in human serum ( the liquid component of blood ) . Lee et al . then checked which mutations helped the virus survive contact with the antibodies . For most human serum samples , a single mutation was enough to allow the virus to escape most of one person’s anti-flu antibodies . This suggests that the immune response to flu is so focused on a small region of hemagglutinin that a mutation in this region can enable the virus to take a huge step towards evading immune detection . Even more surprising was what happened when Lee et al . looked at serum from different people . A mutation that helped the virus to escape immune detection in one person often had little or no effect on escape from another person’s immunity . In other words , the lucky mutation that the virus needed to escape differed from one person to the next . Every year there are many related flu viruses that infect humans . The results of Lee et al . suggest that people could be susceptible to different forms of the virus . Understanding how flu viruses escape immune detection in different people could help us identify which version of the virus different people are more susceptible to , and perhaps eventually better predict how the virus will evolve and spread .

2019

Mapping person-to-person variation in viral mutations that escape polyclonal serum targeting influenza hemagglutinin

Infection and tissue damage induces assembly of supramolecular organizing centres ( SMOCs ) ) , such as the Toll-like receptor ( TLR ) MyDDosome , to co-ordinate inflammatory signaling . SMOC assembly is thought to drive digital all-or-none responses , yet TLR activation by diverse microbes induces anything from mild to severe inflammation . Using single-molecule imaging of TLR4-MyDDosome signaling in living macrophages , we find that MyDDosomes assemble within minutes of TLR4 stimulation . TLR4/MD2 activation leads only to formation of TLR4/MD2 heterotetramers , but not oligomers , suggesting a stoichiometric mismatch between activated receptors and MyDDosomes . The strength of TLR4 signalling depends not only on the number and size of MyDDosomes formed but also how quickly these structures assemble . Activated TLR4 , therefore , acts transiently nucleating assembly of MyDDosomes , a process that is uncoupled from receptor activation . These data explain how the oncogenic mutation of MyD88 ( L265P ) assembles MyDDosomes in the absence of receptor activation to cause constitutive activation of pro-survival NF-κB signalling . During infections and tissue injury , large oligomeric complexes of proteins are assembled . The complexes represent supramolecular organizing centers ( SMOCs ) , which serve as the principal subcellular source of signals that promote inflammation ( Kagan et al . , 2014 ) . In the Toll-like receptor ( TLR ) pathways , the most commonly discussed organizing center is the MyDDosome , which induces NF-κB and AP-1 activation to drive inflammatory transcriptional responses to infection ( Motshwene et al . , 2009; Lin et al . , 2010; Gay et al . , 2014 ) . Cell death pathways are also regulated by SMOCs , such as the pyroptosis-inducing inflammasomes and the apoptosis- inducing DISC ( Lu et al . , 2014; Wang et al . , 2010; Scott et al . , 2009 ) . A common feature of these organizing centers is their ability to be assembled inducibly during infection or other stressful experiences . Structural analysis has also highlighted similarities in the architecture of SMOCs , in that they assemble into helical oligomers ( Lu et al . , 2014 ) . Currently , it is believed that the purpose of assembling these complexes is to create an activation threshold in the innate immune system , such that all-or-none responses can be induced during infection . Microbes , however , contain various inflammatory mediators of varying potency . It is unclear how the innate immune system can convert this diversity of microbial stimuli into a digital response , yet single cell analysis of NF-κB activation induced by TLRs suggests that such a response is indeed induced . As TLR-dependent inflammation is controlled by the MyDDosome , an interrogation of this organizing center in living cells may provide an answer to the receptor proximal events that promote inflammation . We considered several possibilities of how MyDDosome assembly can be regulated by microbial ligands of diverse inflammatory potency . First , there may be a direct correlation between MyDDosome size and the inflammatory activity of microbial products . Second , there may be a direct correlation between the number of MyDDosomes assembled and inflammatory activity . Finally , the speed of MyDDosome assembly may dictate the inflammatory activity of individual microbial products . In order to dissect these possibilities , quantitative single cell analysis in living macrophages is required . We have used single-molecule fluorescence to visualize MyDDosome formation in living cells and analyze the response to lipopolysaccharide ( LPS ) and a much less active synthetic LPS analogue CRX555 . We virally transduced MyD88-GFP into immortalized MyD88-/- macrophages and imaged the living cells with total internal reflection fluorescence microscopy ( TIRFM ) . MyD88-GFP signaled in both HEK cells and macrophages when over expressed . ( Figure 1—figure supplement 1; Figure 1—figure supplement 3 ) . In unstimulated cells GFP-MyD88 was diffusely distributed at or near the cell membrane , but within 3 min of LPS ( 500 nM ) stimulation macromolecular complexes of MyD88 formed ( Figure 1A ) . To estimate the number of MyD88 molecules in these complexes , we compared the intensity of the complexes to those of surface attached dimeric GFP , under identical illumination conditions . The signal intensity of the complexes was approximately three times that of the GFP dimer . This correlates with the 6 MyD88 molecules seen in the crystal structure of the MyDDosome ( Figure 1B ) . MyDDosomes can either persist , are rapidly internalized ( disappear from the field of view within 30 secs of visualization ) or disappear slowly ( 3 min ) . The slow disappearance of some MyDDosomes suggests that the MyD88 complex is able to disaggregate presumably to terminate signaling ( Figure 1A ) . The formation of ‘super’ MyDDosomes ( 2 MyDDosomes coalescing together ) was seen in cells stimulated with LPS with more of these complexes formed in response to high concentrations of this ligand ( Figure 1G ) . No ‘super’ MyDDosomes were seen in response to CRX555 . These data suggest that the MyDDosome size is likely to be related to signaling efficiency . LPS ( 500 nM ) stimulation also resulted in the rapid formation of many MyDDosome complexes which peaked at 5 min post stimulation within the cell ( p<0 . 05; Figure 1C ) . In response to lower concentrations of LPS ( 50 nM ) or CRX555 fewer MyDDosomes were formed within the same time frame ( p<0 . 05; Figure 1D , Figure 1E , Figure 1—figure supplement 4 and Figure 1—figure supplement 5 ) . Current models of signal transduction by TLR4 , based on structural analysis , suggest that the TLR4/MD2 co-receptor assembles into a hetero-tetramer when bound to immunostimulatory LPS and that this induces dimerization of the cytosolic Toll/interleukin-1 receptor ( TIR ) domains . This complex then acts as a scaffold for the recruitment of downstream signal transducers Mal/TIRAP and MyD88 to form a membrane-associated signalosome . Experimental evidence to support this model , however , is lacking . To investigate this question we developed a single-molecule fluorescence approach to quantify TLR4 dimerization on the cell surface in response to LPS and CRX555 . A HaloTag was added to the C-terminal of TLR4 and the construct transduced into immortalized TLR4-/- macrophages ( iBMMs ) . HaloTagged TLR4 signaled in response to LPS in HEK cell reporter assays ( Figure 2—figure supplement 1 ) . Cells were incubated with HaloTag R110Direct for 30 min to label the HaloTag and , following three wash steps , incubated with or without ligand , placed on glass cover slides and fixed at different time points . The cell membrane was imaged using TIRFM and subject to photobleaching analysis to determine the oligomerization state of the labeled TLR4 molecules present . Labeled monomers of TLR4 photobleach in a single step whilst labeled dimers will photobleach in two steps ( Figure 2A ( ii ) ( iii ) ) . Due to the absence of a good antibody to TLR4 we were unable to determine the efficiency of labeling of TLR4 , so our measurements may underestimate the dimer population but allow us to follow relative changes in the number of TLR4 monomer and dimers . Unstimulated cells showed two different populations of TLR4 complexes present: monomeric ( 78 ± 3% ) and dimeric ( 22 ± 3% ) ( Figure 2B ( i ) ) , indicating the labeling level of TLR4 must be at least greater than 30% . The number of TLR4 dimers on the cell surface is small and stimulating cells with LPS showed an increased number of dimers at 5 min following stimulation , compared to unstimulated cells , which rapidly reduced presumably due to internalization of TLR4 and trafficking to the endosome ( Kagan et al . , 2008 ) ( Figure 2B ( ii ) ) . This trend was clearest at lower levels of LPS because at higher concentrations TLR4 internalization occurred very rapidly . At very low concentrations of LPS or with CRX555 the number of dimers formed was comparable to that seen in unstimulated cells ( Figure 2B ( iii ) ) . We did not observe clusters of TLR4 with any ligand and thus higher order oligomerization of the receptors is unlikely . In contrast to stimulatory LPS , structurally related antagonists such as Eritoran or Rhodobacter sphaeroides Lipid A ( RSLA ) bind to TLR4/MD2 co-receptors but do not induce the formation of hetero-tetrameric complexes in vitro . Consistent with this we find that the number of TLR4 dimers on the surface of macrophages treated with RSLA is significantly lower than that seen in unstimulated cells ( Figure 2B ( iv ) ) . This suggests that RSLA stabilizes the TLR4/MD2 heterodimer and prevents the formation of constitutive tetramers . We next used a TIR domain mutant of TLR4 , Pro712His , that is unable to signal ( Poltorak et al . , 1998 ) or recruit downstream signal transducers such as MyD88 to ask whether assembly of the TLR4/MD2 tetramer occurred in the absence of signal transduction . We compared the numbers of TLR4 monomers and dimers in unstimulated and LPS-stimulated cells transduced with HaloTagged TLR4 Pro712His to the numbers seen for wild-type TLR4 . The ratio of monomer to dimer Pro712His TLR4 was similar to wild type TLR4 in unstimulated cells , but in LPS-stimulated cells it was less than that seen for cells transduced with native HaloTagged TLR4 ( Figure 2B ( v ) ) . This result indicates that LPS binding to the Pro712His receptor stabilises the TLR4/MD-2 heterodimer and inhibits the formation of active tetramers . Dimerisation of the TLR4/MD-2 ectodomain and the TIR domain BB-loop are thus synergistic and both events must occur to produce an active receptor complex . This implies a two step activation mechanism for TLR4 , as illustrated in Figure 2C . Alternatively , failure of the P712H receptor to recruit adaptors could lead to rapid removal of the inactive receptor heterotetramers from the cell surface . Dynamic molecular modeling has suggested that binding of different ligands causes conformational changes in MD2 that may be linked to receptor function ( Paramo et al . , 2013 ) . We analyzed the final relative location of MD2 in the dimeric TLR4 complex by dynamic molecular modeling and discovered that the Lipid A agonist-associated state remained close to the LPS-bound X-ray structure ( Song and Lee , 2012 ) , whereas in the absence of agonist , a shift of up to ~10 Å of MD2 relative to its primary TLR4 partner was observed as it disassembled from the secondary , dimeric TLR4 interface . The conformation of the two TLR4/MD2 heterodimers observed in this modeled structure ( Figure 2D , right hand ) closely resembled the X-ray structures of monomeric TLR4 in the unliganded-mouse and Eritoran antagonist-bound human constructs . ( Song and Lee , 2012 ) Principal component analysis ( PCA ) was then performed on the dynamic trajectories in order to isolate the dominant , collective motions of the dimeric TLR4 chains ( Figure 2D , Figure 2—figure supplement 2 ) . This revealed that in the presence of Lipid A there were minor rotational motions of the ECDs with respect to one another at equilibrium , similar to the ‘ring rotation’ observed with ligand bound TLR8 , but no significant separation of the two C-termini ( Ohto et al . , 2014 ) ( Figure 2—figure supplement 2 , Figure 2—figure supplement 3 ) . In the absence of Lipid A , large lateral fluctuations of the C-termini were observed , similar to the ‘hinge motion’ seen in the TLR8 inactivated homodimer ( Ohto et al . , 2014 ) . Consistent with these observations , in the absence of ligand , the backbone of the heterotetrameric TLR4/MD2 complex was increased by ~50–60% whilst up to 60% of the surface area buried between each TLR4/MD2 heterodimer and its adjacent primary TLR4 partner was lost , compared to the lipid A bound state ( Supplementary file 1 ) , indicative of major conformational changes . Collectively , these data provide further support for a two-step TLR4 activation model where ligand binding results in a structural change to facilitate close apposition of the C-termini of the TLR4 dimer to allow TIR dimerization . To address further the coupling of receptor activation and signal transduction , macrophages were stably transduced with the NF-κB subunit RelA tagged with GFP and a TNFα-reporter construct fused to mCherry ( Sung et al . , 2014 ) and were stimulated with increasing concentrations of the TLR4 agonist LPS or the partial agonist CRX555 ( Stöver et al . , 2004 ) . Single cells were visualized using live confocal microscopy for 24 hr ( Figure 3A ) . The dynamics of RelA-GFP translocation into the cell nucleus and TNFα-mCherry induction were quantified for individual cells ( Figure 3—figure supplement 1 ) . LPS induced rapid translocation of NF-κB into the nucleus and the rate of nuclear translocation increased with the dose of LPS ( Figure 3B ) . The fastest time observed from initial stimulation to peak levels of nuclear NFκB was 20 min , findings that correlate well with electrophoretic mobility shift assays where accumulation of NFκB is seen within minutes after cellular stimulation . ( Kopp and Ghosh , 1995 ) Low concentrations of LPS or stimulation of cells with CRX555 showed delayed translocation of NF-κB into the nucleus coupled to reduced expression of the TNFα-mCherry reporter construct ( Figure 3B and C ) . Statistical analysis of the individual parameters of the single cell signal transduction assays shows that TNFα-mCherry production was significantly correlated to the speed of NF-κB translocation into the nucleus ( Figure 3D and E ) . The peak timing of NFκB translocation also correlates with the magnitude of NFκB translocation and the TNF α-mCherry reporter expression for both LPS and CRX555 stimulated cells . These kinetic data correlated with the kinetics of our single cell signaling assays suggesting that the strength of signal is also partially determined by the rapidity with which a critical number of MyDDosomes form to trigger NF-κB translocation to the nucleus . Here we show that in the absence of ligand , TLR4/MD2 on the surface of living immune system cells is in a dynamic equilibrium with populations of heterodimers and heterotetramers . The binding of LPS likely stabilizes the tetrameric form and initiates conformational changes that lead to signal transduction . In the absence of ligand , the ECDs of the endosomal TLR8 exist as stable , inactive preformed dimers . The binding of small molecule agonists causes an extensive conformational rearrangement within the dimer that brings the C-termini of the TLR8 ECDs closer together ( Tanji et al . , 2013 ) . Our molecular dynamic analysis suggests that TLR4 activation involves a similar process in which the two ECDs tilt and rotate with respect to each other during signal transduction . This also suggests that the crystal structure of heterotetrameric TLR4/MD2 might represent an inactive transition state corresponding to the first step of the concerted activation process ( Figure 2C , [Gay et al . , 2006] ) We also visualize , for the first time , the assembly of the membrane associated MyDDosome signaling scaffold in vivo and show that signaling flux depends on the size and number of these structures . The absence of oligomeric receptor clusters also implies that active TLR4 does not form a stoichiometric post-receptor complex with the MyDDosome . A dimer of the receptor TIR domains is assumed to have two binding sites for MyD88 ( Valkov et al . , 2011; Núñez Miguel et al . , 2007 ) whereas the Myddosome has about 6 MyD88 molecules both in vitro and in vivo . This stoichiometric mismatch taken together with the highly transient nature of the Myddosome ( Figure 1 ) indicates that the activated receptor nucleates the assembly of the higher order MyDDosome structures that associate only transiently with the membrane bound receptor rather than forming a stable signalosome ( Motshwene et al . , 2009; Triantafilou et al . , 2004 ) . This assembly mechanism is similar to that proposed for the pyrin domain of Asc which nucleates assembly of filamentous NLRP3 inflammasomes ( Lu et al . , 2014 ) . Rapid dissociation of the MyDDosome from the receptor is likely to be coupled to TLR4/MD-2 internalisation providing a mechanism for the sequential activation of the MyD88 and TRIF directed signals from the cell surface and endosomes respectively ( Kagan et al . , 2008; Zanoni et al . , 2011 ) . These findings also explain the properties of an oncogenic somatic mutation in MyD88 ( L265P ) commonly found in B-cell lymphomas and other conditions such as Waldenstrom's macroglobulinemia ( Treon et al . , 2012; Ngo et al . , 2011 ) . In these diseases , MyDDosomes assemble spontaneously in the absence of receptor activation causing constitutive activation of NFκB which acts as a pro-survival signal ( Avbelj et al . , 2014 ) . Our results are consistent with another model of cooperative assembly for the Mal/TIRAP adaptor that was proposed recently Ve et al . , 2017 Our data show that partial agonists stimulate the formation of a smaller number of MyDDosomes more slowly than the agonist LPS and that this leads to slower translocation of NF-κB into the nucleus . LPS forms a larger number of MyDDosomes more rapidly resulting in more rapid NF-κB translocation . Surprisingly only a small number of MyDDosomes need to be formed on the cell surface for full cellular signalling to occur and we proposed that the difference between full and partial agonism is determined by MyDDosome number and the speed of their formation . This can simply be controlled by the equilibrium between the agonist or partial agonist for the TLR4 dimers present on the cell surface , with agonists having faster on-rates or slower off rates than the partial agonists that is a higher affinity . This provides a simple explanation of how a graded response is achieved in TLR4 signalling . The small number of MyDDosomes needed for signalling is reminiscent of the situation with the T-cell recptor ( TCR ) . In this case 10 TCRs can activate a T-cell with each triggered TCR rapidly forming a larger signalling complex ( Smith-Garvin et al . , 2009; Irvine et al . , 2002 ) . Both TCR and TLR4 signalling are highly sensitive , show graded responses and occur at the level of the single molecule requiring a mechanism to be in place to prevent inadvertent signalling . For the TCR this is based on the affinity of the ligand presented by the MHC to the receptor , and this provides for discrimination between self and non-self by mechanisms that still need to be elucidated ( Huang et al . , 2013 ) . For TLR4 it seems that receptor dimers need to be stabilised by an agonist in order to signal . Affinity is also the mechanism used by TLR4 to obtain a graded response by stabilising TLR4 dimers to greater or lesser extents . In both TCR and TLR4 signalling a larger signalling complex is formed as a result of a single molecule event leading to significant signal amplification . In summary , our study suggests that TLR4 signalling occurs at the level of single molecules with agonists stabilising low numbers of preformed dimers that then nucleate the formation of a short-lived MyDDosome signalling complex which is then removed from the cell surface . Partial agonists form less MyDDosomes more slowly leading to a smaller overall cellular response . This provides new insights into the mechanism of TLR4 signalling and how it is possible to obtain a graded response to different agonists . The RAW264 . 7-derived reporter cell line which expresses enhanced green fluorescent protein ( EGFP ) -tagged RelA and TNFα promoter-driven mCherry was provided by Dr . Iain D . C . Fraser ( National Institute of Health , MD , USA ) . The cells were maintained in Dulbecco’s Modified Eagle’s Medium ( DMEM; Sigma-Aldrich ) supplemented with 10% ( v/v ) heat-inactivated fetal calf serum ( FCS; Thermo Scientific , Rugby , UK ) , 2 mM L-glutamine ( Sigma-Aldrich ) and 20 mM HEPES ( Sigma-Aldrich ) at 37°C , 5% CO2 . The cells were plated on a 35 mm glass-bottom dish ( Greiner Bio-One ) at a concentration of 1 . 0 × 105 cells in phenol red-free DMEM supplemented with 10% ( v/v ) FCS , 2 mM L-glutamine and 20 mM HEPES , and settled down in an incubator for 8 hr prior to experiment . The dish was mounted on the stage of a confocal microscope ( TCS SP5 , Leica ) and kept at 37°C , 5% CO2 during the experiment in a climate chamber . Live cell imaging was performed immediately after stimulating the cells . Images were sequentially taken on a 40x oil-immersion objective ( NA1 . 25 ) with 2 . 0x zoom every 3 min for 15 hr . The image dimension used was 512 × 512 pixels . The pinhole size was 1 A . U . , and the thickness of the focal plane was 0 . 96 μm . Acquired images were exported as 16-bit TIFF files for analysis . A MATLAB-based automated single cell analysis script was used to automatically assess NF-κB nuclear translocation and TNFα promoter-driven mCherry expression ( Figure 3—figure supplement 1 ) . Lipopolysaccharide ( LPS ) and Rhoderbacter sphaeroides lipid A ( RSLA ) stocks were thawed from storage at −20°C and sonicated for 1 min . The ligand was then diluted to the appropriate concentration in DMEM . CRX555 stored at 4°C was sonicated for 1 min prior to dilution to the required concentration with DMEM . The CHARMM22/CMAP all-atom force field ( http://pubs . acs . org/doi/abs/10 . 1021/jp973084f ) was used with explicit TIP3P waters using GROMACS 5 . 0 . 3 ( Bjelkmar et al . , 2010 ) . Lipid A parameters compatible with CHARMM22 were used , which correctly reproduce structural and dynamic properties of lamellar phases as described ( Paramo et al . , 2015 ) . Starting simulations were setup based on the crystal structure of the LPS-bound , heterotetrameric TLR4/MD-2 ( Park et al . , 2009 ) ( pdb: 3FXI ) . The ligand-bound or ligand-free apo complexes were setup as described previously ( Paramo et al . , 2013 ) . Briefly , each complex was placed in an octahedral unit cell ( dimension ~17 nm ) and solvated with a 0 . 1 M NaCl solution . Energy minimization using steepest descents was performed ( <10 , 000 steps ) to remove steric clashes , and a 1 . 5 ns position-restrained equilibration phase followed . Subsequently , 100 ns production simulations were initiated for each system in the NpT ensemble . Since no experimental structures are presently available for apo , dimeric TLR4 , three production simulation replicas of this system were initiated using different initial random velocities ( Figure 2—figure supplement 2 ) . Equations of motion were integrated using a 2 fs time step with bond lengths constrained via LINCS ( Hess , 2008 ) . Lennard-Jones interactions were smoothly switched off between 1 nm and 1 . 2 nm , and electrostatics were computed using the Particle-Mesh-Ewald algorithm ( Essmann et al . , 1995 ) with a 1 . 2 nm real-space cutoff . Temperature and pressure were coupled using the velocity-rescale thermostat ( Bussi et al . , 2007 ) at 298 K and the Parrinello-Rahman barostat ( Nosé and Klein , 1983; Parrinello and Rahman , 1981 ) at 1 atm , respectively . PCA can be used to remove the high-frequency ‘background’ motions from trajectories simulation trajectories , in order to identify collective , low-amplitude protein dynamics . PCA was performed by calculating and diagonalizing the mass-weighted covariance matrix for the C-alpha atoms of each pair of TLR4 chains in the dimer . The corresponding trajectory was projected onto the first eigenvector , and interpolation between the two extreme projections around the average structure were used to generate porcupine plots within the VMD package ( Humphrey et al . , 1996 ) .

Cells in the immune system have proteins at their surface that detect molecules produced by invading microbes . One of these proteins is Toll-like receptor 4 , TLR4 for short . Once TLR4 is activated , the immune cells form MyDDosomes – intricate complexes made of many different proteins . These structures form a signal that mobilizes the cell to fight the infection . In particular , the complexes set up a chain of events that leads to a gene-regulating protein getting access to the cell’s DNA . There , the protein switches on genes which produce other proteins important for inflammation , one of the body’s most important tools to fight an infection . The activation of TLR4 is thought to be an all-or-nothing mechanism: the receptors are either ‘on’ or ‘off’ . However , different microbial molecules recognized by TLR4 trigger different levels of inflammation , ranging from mild to severe . It remained unclear how an all-or-none response from the frontline receptors could lead to a gradual response from the cell . Here , Latty et al . compare what happens to TLR4 , MyDDosomes and the gene-regulating proteins when living immune cells are stimulated by different doses of two microbial molecules . These agents are both recognized by TLR4 , but they lead to different levels of inflammation . The type of microbial molecule , or their concentration , does not change how TLR4 is activated . Two TLR4 proteins can loosely associate with each together to form a dimer . When they bind a microbial molecule , the dimer becomes more stable . This changes the shape of the TLR4 proteins , which in turn triggers the formation of a scaffold of MyDDosomes . More stable TLR4 dimers are formed when the cells is in contact with a microbial molecule that triggers a strong immune reaction , and possibly when its concentration is higher . Crucially , the different microbial agents and their concentration levels modify how MyDDosomes assemble . By ‘tagging’ each protein in the complex with a fluorescent chemical , Latty et al . can follow its formation as it actually happens . When the cells are stimulated with microbial molecules that provoke a strong inflammation , the MyDDosomes may be bigger , in greater numbers , and form more quickly . In turn , under strong microbial activation , the gene-regulating protein that switches on the immune response genes goes to the DNA faster and in higher numbers . This suggests that the pace of assembly , the size and the number of MyDDosomes control the strength of the immune response . TLR4 is involved in diseases such as cancer or Alzheimer’s disease , where the body has an incorrect inflammation response . Knowing in greater detail the cellular processes activated by TLR4 could help efforts to find new drug targets for these conditions .

2018

Activation of Toll-like receptors nucleates assembly of the MyDDosome signaling hub

The rodent olfactory bulb incorporates thousands of newly generated inhibitory neurons daily throughout adulthood , but the role of adult neurogenesis in olfactory processing is not fully understood . Here we adopted a genetic method to inducibly suppress adult neurogenesis and investigated its effect on behavior and bulbar activity . Mice without young adult-born neurons ( ABNs ) showed normal ability in discriminating very different odorants but were impaired in fine discrimination . Furthermore , two-photon calcium imaging of mitral cells ( MCs ) revealed that the ensemble odor representations of similar odorants were more ambiguous in the ablation animals . This increased ambiguity was primarily due to a decrease in MC suppressive responses . Intriguingly , these deficits in MC encoding were only observed during task engagement but not passive exposure . Our results indicate that young olfactory ABNs are essential for the enhancement of MC pattern separation in a task engagement-dependent manner , potentially functioning as a gateway for top-down modulation . The brain remains plastic throughout life . A dramatic example of neural circuit plasticity during adulthood comes in the form of adult neurogenesis ( Altman and Das , 1965 ) . The subventricular zone ( SVZ ) is one of the two main loci in the rodent brain where adult neurogenesis takes place ( Zhao et al . , 2008 ) . In the SVZ , many thousands of new neurons are produced each day throughout adulthood , and these new neurons migrate through the rostral migratory stream to the olfactory bulb ( Lois and Alvarez-Buylla , 1994 ) , the first olfactory center of the brain . Once in the olfactory bulb , about 95% of the adult-born neurons ( ABNs ) differentiate into granule cells ( GCs ) and the majority of the remaining differentiate into periglomerular cells ( Lledo et al . , 2006 ) , both of which are GABAergic local inhibitory neurons . GCs inhibit mitral cells ( MCs ) , the principal neurons of the olfactory bulb , through their dendrodendritic reciprocal connections ( Isaacson and Strowbridge , 1998; Rall et al . , 1966; Shepherd , 1963 ) . This inhibition of MCs can sparsen odor representations and enhance the signal-to-noise ratio ( Koulakov and Rinberg , 2011; Yokoi et al . , 1995; Yu et al . , 2014 ) . Consistent with this idea , general activation of bulbar inhibitory neurons can accelerate learning ( Abraham et al . , 2010 ) , while suppression of inhibitory neuron activity can increase the excitability of MCs and reduce MC pattern separation ( Gschwend et al . , 2015 ) . Thus , local inhibitory neurons in the olfactory bulb , including ABNs , likely control olfactory perception by providing inhibition onto MCs . As ABNs integrate into local circuits , they display higher levels of morphological and functional plasticity during the first ~8 weeks after their birth compared to their later mature stage ( Kelsch et al . , 2009; Mizrahi , 2007; Nissant et al . , 2009; Sailor et al . , 2016 ) . Furthermore , the spine dynamics , synaptic plasticity , sensory response pattern , as well as survival rate of ABNs are influenced by olfactory experience during this early period ( Alonso et al . , 2006; Lemasson et al . , 2005; Lepousez et al . , 2014; Livneh et al . , 2014; Mouret et al . , 2008; Petreanu and Alvarez-Buylla , 2002; Quast et al . , 2017; Rochefort et al . , 2002; Yamaguchi and Mori , 2005 ) . These unique and plastic features of young ABNs make it likely that they play a unique role in the processing of the complex and dynamic olfactory environment . Indeed , some studies have shown that ABNs are essential for certain olfactory behaviors such as odor discrimination and association reversal learning ( Alonso et al . , 2012; Bath et al . , 2008; Enwere et al . , 2004; Gheusi et al . , 2000; Moreno et al . , 2009; Sakamoto et al . , 2014 ) . However , other studies have found little effects of ABN manipulation on odor discrimination ( Breton-Provencher et al . , 2009; Imayoshi et al . , 2008; Lazarini et al . , 2009 ) . Thus , ABNs are not essential for all olfactory processing . Instead , the inconsistencies between these results raise the possibility that the impact of ABNs may depend on the behavioral context . Consistent with the idea that the functions of ABNs are context-dependent , local inhibitory neurons in the olfactory bulb including ABNs receive abundant glutamatergic centrifugal inputs from higher brain areas such as anterior olfactory nucleus , piriform cortex and entorhinal cortex ( Balu et al . , 2007; Boyd et al . , 2015 , 2012; Chapuis et al . , 2013; Kiselycznyk et al . , 2006; Markopoulos et al . , 2012; Nunez-Parra et al . , 2013; Otazu et al . , 2015; Rothermel et al . , 2014 ) . Inhibitory neurons also express a variety of neuromodulator receptors , which provide additional avenues for top-down modulation ( Castillo et al . , 1999; Devore and Linster , 2012; Ma and Luo , 2012; Moreno et al . , 2012; Rothermel et al . , 2014; Shipley et al . , 1985 ) . Furthermore , anesthesia inactivates feedback fibers and reduces GC activity while increasing MC activity ( Kato et al . , 2012; Rothermel and Wachowiak , 2014 ) . Thus , bulbar inhibitory neurons may function as a mediator of feedback regulation to shape MC odor encoding ( Markopoulos et al . , 2012 ) . Consistent with this notion , it has been reported that cortical feedback can decorrelate MC odor representations ( Otazu et al . , 2015 ) . Similarly , neuromodulatory projections can modulate MC activity ( Rothermel and Wachowiak , 2014 ) and improve perceptual learning ( Ma and Luo , 2012 ) . However , the role of young ABNs in mediating context-dependent modulation of MC activity has not been fully explored . To assess the role of ABNs in the olfactory bulb , we adopted a genetic method to inducibly suppress adult neurogenesis . We reasoned that this chronic ablation allows us to probe for the functions of ABNs which cannot be compensated for by other inhibitory neurons . Behavioral experiments showed that ABN ablation animals were impaired in fine , but not coarse , odor discrimination . Neither suppression of hippocampal ABNs alone nor non-selective ablation of a comparable number of bulbar neurons caused the same behavioral deficit , highlighting the unique importance of young ABNs . Two-photon calcium imaging revealed that the behavioral deficit was accompanied by a decreased separation of MC ensemble responses to similar odorants in ablation animals . This decreased separation was largely due to a reduction in suppressive odorant responses by MCs . Interestingly , this difference in suppressive responses was only observed when animals were actively engaged in the task . To investigate the role of adult neurogenesis in olfactory processing , we adopted the transgenic mouse line Gfap-tk ( Snyder et al . , 2011 ) . In this line , herpes simplex virus Thymidine Kinase ( TK ) is expressed under the Glial Fibrillary Acidic Protein ( GFAP ) promoter , rendering mitotic neural stem cells sensitive to the antiviral drug Valganciclovir ( VGCC ) ( Figure 1A ) . This gives us a means to specifically suppress adult neurogenesis in an inducible manner . We sought to investigate the consequences of eliminating young ABNs that are 8 weeks old and younger . To achieve this , we treated Gfap-tk mice with VGCC , starting 8 weeks prior to the beginning of the behavioral experiments and continuing throughout the duration of the experiments ( Figure 1B ) . Hereafter we refer to these animals as the ‘ablation’ animals . The ablation of adult neurogenesis in the olfactory bulb was nearly complete , shown by post-hoc BrdU labeling ( p<0 . 0001 , Wilcoxon rank sum test , Gfap-tk+ vs . Gfap-tk-; Figure 1C ) as well as immunostaining for Doublecortin , a marker for immature neurons ( Figure 1D ) . Importantly , there was no difference in the density of GFAP-positive astrocytes in VGCC-treated Gfap-tk+ and Gfap-tk- mice in the olfactory bulb ( p=0 . 9911 , Wilcoxon rank sum test; Figure 1E ) nor in the hippocampus dentate gyrus ( p=0 . 1359 , Wilcoxon rank sum test ) , consistent with a previous report ( Snyder et al . , 2011 ) . Another previous report also showed that regeneration of olfactory sensory neurons is intact in this mouse line ( Cummings et al . , 2014 ) . Ablation mice showed no obvious general health impairment , and demonstrated mobility and anxiety levels comparable to control mice in an open field test ( average speed: p=1 . 0000 , center time fraction: p=0 . 7984 , Wilcoxon rank sum test; Figure 1F ) . Equipped with this effective and specific method of inducible adult neurogenesis ablation , we explored how the absence of young ABNs could affect olfactory behavior . We compared the behavioral performance of the ablation animals and littermate controls ( control: n = 22; ablation: n = 23 ) . Both groups were treated identically including VGCC administration , and experimenters were blind to their genotypes during experiments . Mice were trained in a two-alternative-choice olfactory discrimination task under head-fixation . In this task , a certain odorant was delivered in each trial for 4 s , followed by an answer period of 2 s during which mice were required to lick -either the left or right port according to the odorant cue to receive a water reward ( Figure 2A ) . Mice were trained daily , one session per day , and each session consisted of 144 . 4 ± 17 . 4 trials for the control group and 146 . 4 ± 13 . 8 trials for the ablation group . After the initial pre-training period ( Materials and methods ) , mice were trained in a relatively easy discrimination task in which mice were required to discriminate between conspicuously different binary mixtures ( Left lick: 80% heptanal and 20% ethyl-tiglate ( 80H20E ) ; right lick: 20% heptanal and 80% ethyl-tiglate ( 20H80E ) , all mixture percentages are of a total concentration of 100 ppm; Figure 2B ) . Both ablation and control animals achieved expertise in this task ( defined as >80% success rate ) in equivalent durations of training ( number of sessions , control: 4 . 18 ± 0 . 21 , ablation: 4 . 39 ± 0 . 25 , mean ±S . E . M; p=0 . 4523 , Wilcoxon rank sum test; Figure 2C ) . Thus , we conclude that young ABNs are not required for the performance of this easy discrimination task . Given these results , we asked whether finer discrimination would reveal a deficit caused by ABN ablation . To address this question , we devised a difficult discrimination task in which 1 of 8 very similar mixtures , each with slightly varying ratios of ethyl-tiglate and heptanal , was presented in each trial . Four of the eight mixtures signaled left lick trials , while the other four mixtures signaled right lick trials ( Figure 2D ) . After mice achieved expertise in the easy discrimination task , they were trained with this difficult discrimination task over 10 sessions . Although the performance of control animals was initially at chance level , it consistently improved over 10 sessions to achieve a success rate of 0 . 760 ± 0 . 017 in session 10 . In contrast , ablation animals showed slower learning ( comparison of linear regression slopes in individual animals , p=0 . 0100 , Wilcoxon rank sum test ) , and their performance was significantly lower than that of the control animals ( p ( group ) <0 . 0001 , p ( session ) <0 . 0001 , two-way ANOVA; Figure 2E ) . The deficits observed in ablation animals were unlikely due to problems in motivation or licking ability , as both groups had the comparable fractions of answered trials ( control vs . ablation , p ( group ) = 0 . 4101 , two-way ANOVA; Figure 2F ) and the comparable licking rates during reward consumption ( control vs . ablation , p ( group ) = 0 . 1036 , two-way ANOVA; Figure 2G ) . To confirm that mice were performing the task by using odorant stimuli as the cues and not other cues ( such as potential differences in sounds of different odorant valves ) , we performed an additional session after the 10th session . In this test session , all odorants were replaced with the same 50H50E mixture while the contingency between odorant valves and correct lick side was maintained . The performance of both groups dropped to chance level in the test session ( control: p=0 . 7241 , ablation: p=0 . 6925 , t test with chance level ( 0 . 5 ) ; Figure 2E , ‘50:50’ ) , indicating that they were indeed relying on odorants as the cue . As an additional control , we trained a separate cohort of mice ( control: n = 7; ablation: n = 6 ) . After the pre-training with easy 2-odorant discrimination , these mice were trained with another version of 8-odorant discrimination task in which we used eight highly distinct mixture ratios ( ‘easy 8-odorant discrimination’; Figure 2—figure supplement 1A ) . On the first day of training with the easy 8-odorant discrimination task , both control and ablation animals performed similarly , at the level equivalent to the easy 2-odorant discrimination ( control: p=0 . 5350 , ablation: p=0 . 2229 , Wilcoxon rank sum test ) , which was significantly better than the first session of difficult 8-odorant discrimination ( p ( task ) <0 . 0001 , 3-way ANOVA; Figure 2—figure supplement 1B ) . Therefore , the deficits of ablation animals in the difficult 8-odorant discrimination were due to the requirement of fine discrimination and not due to the complexity of 8 different mixtures . In conclusion , mice without young ABNs are impaired in the difficult discrimination task requiring fine odorant discrimination . The results above suggest that ABNs in the olfactory bulb are critical for fine odorant discrimination . However , the dentate gyrus ( DG ) of hippocampus is the other major niche for adult neurogenesis ( Gonçalves et al . , 2016; Ming and Song , 2011 ) , and the Gfap-tk method suppresses adult neurogenesis in both the SVZ and DG . To address this issue , we adopted another transgenic method previously described that could specifically suppress postnatally-born DG neurons ( mGfap-Cre::Slc17a7-LoxP-STOP-LoxP-tetanus neurotoxin ( LSL-TeNT ) , hereafter referred to as ‘DG suppression’ , Slc17a7 is also known as vGlut1 , which is expressed in excitatory DG granule cells but not in inhibitory OB GCs ) ( Sakamoto et al . , 2014 ) . As ABNs are a subset of postnatally-born DG neurons , this DG suppression method targeted a much larger fraction ( more than one third ) of DG neurons ( Figure 3—figure supplement 1a and b ) . Supporting the effective expression of TeNT in a large fraction of DG granule cells , immunostaining signal for VAMP2 in the mossy fiber terminals in CA3 was significantly reduced in the DG suppression mice ( Figure 3—figure supplement 1c ) . These results are consistent with the idea that the ‘DG suppression’ strategy effectively suppressed most , if not all , of postnatally-born DG granule cells including DG ABNs , even though it is difficult to directly demonstrate the inhibition of synaptic release from all DG ABNs . During the same easy and difficult discrimination tasks described in Figure 2 , the DG suppression mice showed comparable performance to the control group in both easy and difficult discrimination tasks ( control vs . DG suppression; easy discrimination: number of sessions to reach expertise , p=0 . 8407 , Wilcoxon rank sum test; difficult discrimination: p ( group ) =0 . 2663 , p ( session ) <0 . 0001 , 2-way ANOVA; mean ±S . E . M . ; Figure 3A , B ) . These results indicate that DG postnatally-born neurons that include ABNs are not essential for fine olfactory discrimination and the behavioral impairment in Gfap-tk+ ablation group was primarily caused by the absence of ABNs in the olfactory bulb . We do note , however , that a formal possibility remains that a small fraction of DG ABNs that may have been spared in the ‘DG suppression’ method may partially contribute to the behavioral deficits of the Gfap-tk+ ablation group . Next we considered two alternative possibilities underlying the behavioral deficit in the ablation animals . First , it is possible that ablation of any inhibitory neurons in the olfactory bulb may lead to similar deficits . Second , olfactory fine discrimination may be particularly sensitive to ABN ablation . To distinguish these possibilities , we sought to ablate a random subset of GCs regardless of their age . To this goal , we first estimated the degree of neuron loss in the ABN ablation animals by quantifying cell density in the granule cell layer ( GCL ) using DAPI labeling . This indicated a 10 . 02% reduction of total cell numbers through ABN ablation by the end of the behavioral tasks ( control: n = 3 , ablation: n = 4; Figure 3C , D ) . To achieve a similar level of neuron ablation without specifically targeting ABNs ( ‘random ablation’ ) , we bilaterally injected a mixture of diluted AAV2/1-CMV-Cre and AAV2/1-EF1a-FLEX-taCaspase3 ( Yang et al . , 2013 ) into the center of the olfactory bulb . Post hoc DAPI staining and quantification ~1 month after injections revealed a 19 . 66% reduction in GCL cell density compared with un-injected animals ( injected ( n = 11 ) vs . un-injected control ( n = 7 ) : p<0 . 0001 , Wilcoxon rank sum test; Figure 3E , F ) , with no significant change in the width of GCL ( injected vs . un-injected control: p=0 . 9499 , Wilcoxon rank sum test; Figure 3G ) . Thus , our random ablation eliminated a larger number of cells than our ABN ablation . Importantly , we found no change in Doublecortin immunostaining , indicating that our random ablation did not affect subsequent adult neurogenesis ( injected ( n = 7 ) vs . uninjected ( n = 7 ) control , p=0 . 9015 , Wilcoxon rank sum test; Figure 3H , I ) . We trained these random ablation animals starting at 1 month after the injections . Random ablation animals exhibited normal performance in the easy discrimination task ( control vs . random ablation , number of sessions to reach expertise , p=0 . 4430 , Wilcoxon rank sum test; mean ±S . E . M . , Figure 3J ) . In the difficult discrimination task , the random ablation group initially showed slower learning than the control group , but they eventually reached the performance level that was statistically indistinguishable from the controls and significantly better than ABN ablation animals ( Sessions 1–10: random ablation ( n = 11 ) vs . control ( n = 22 ) , p=0 . 0034 , random ablation vs . ABN ablation , p<0 . 0001; Sessions 1–5: random ablation vs . control , p=0 . 0039 , random ablation vs . ABN ablation , p=0 . 4405; Sessions 6–10: random ablation vs . control , p=0 . 2028 , random ablation vs . ABN ablation , p<0 . 0001; 2-way ANOVA; mean ±S . E . M . ; Figure 3K ) . Together with the observation that the random ablation eliminated more cells than in ABN ablation , these results support the notion that young ABNs have a privileged role in mediating fine olfactory discrimination . To investigate the neural basis of the impaired discrimination in ABN ablation animals , we monitored the activity of MCs in ablation and control animals using two-photon calcium imaging . We utilized the transgenic mouse line Cdhr1-Cre ( Cdhr1 is also known as Pcdh21 ) , which expresses Cre specifically in the olfactory bulb principal neurons . We injected AAV1-hsyn-FLEX-GCaMP6f in the right olfactory bulb of Gfap-tk+/-::Cdhr1-Cre ( ablation ) or Gfap-tk-/-::Cdhr1-Cre ( control , littermates ) animals to specifically express GCaMP6f in mitral/tufted cells ( Figure 4A ) . After training with the easy discrimination task , these mice were trained with the difficult discrimination task while we imaged the ensemble activity of MCs ( control: n = 12 , ablation: n = 10; Figure 4B , C ) . Individual MCs showed odorant-specific responses with an increase or decrease in GCaMP6f fluorescence ( Figure 4D ) . To quantify the discriminability of the eight mixtures by the MC ensembles , we performed decoder analysis ( Chu et al . , 2016 ) which attempts to decode the odorant on each trial based on the population activity of individual MCs during the odorant period ( Materials and methods ) . If the decoded odorant matched the actual odorant , the trial was scored as correct . We found that decoder accuracy was significantly better than chance ( 0 . 125 ) in both control and ablation groups ( control: p<0 . 001 , ablation: p<0 . 001 , Student’s t-test; Figure 4E ) . However , the decoder accuracy was higher in control animals than in ablation animals ( control vs . ablation , p ( group ) <0 . 05 , two-way ANOVA; Figure 4E ) . These results indicate that MC responses to different mixtures are more ambiguous in ablation animals than in control . Next we asked whether the separation of mitral cell odor representations is sensitive to the similarity of odor mixtures . To address this , we performed a pairwise decoder analysis in which we built a decoder to decode the mixture identity for each pair of the eight mixtures . Here we defined the ‘contrast’ between each pair of mixtures as the difference in the percentage of heptanal ( Figure 4F ) . For example , the contrast between 52H48E and 48 . 5H51 . 5E is 3 . 5 ( =52–48 . 5 ) . We found that within each of the control and ablation groups , there was a positive correlation between pairwise decoder accuracy and the contrast between the mixtures ( control: r = 0 . 1862 , p<0 . 01 , ablation: r = 0 . 3029 , p<0 . 0001 , Pearson correlation; Figure 4G ) . Although the decoder performance of the control group was generally better than the ablation group , the difference was more prominent in mixture pairs with smaller contrasts ( ≤3 ) ( control vs . ablation , pairs with contrasts ≤3: p ( group ) <0 . 001 , pairs with contrasts >3: p ( group ) =0 . 2530 , two-way ANOVA; Figure 4G ) . These results suggest that ABN ablation causes the separation of MC population responses to similar odorants to be less robust , possibly underlying the behavioral deficits in fine discrimination . Consistent with this notion , the decoder accuracy of individual animals positively correlated with their behavioral performance ( p<0 . 05 , Pearson correlation; Figure 4H ) . We next asked whether the categorical associations with the left or right lick side influenced the MC population responses . To investigate this , we analyzed the decoder performance based on the difference ( ‘contrast’ ) between the two odorants and whether the two odorants were associated with the same or different lick sides ( Figure 4—figure supplement 1 ) . The decoder performance was better when the two odorants were more distinct . However , the decoder performance was not affected by whether the two odorants were associated with the same or different lick sides ( control: p=0 . 4727 , ablation: p=1 . 0000 , same vs . different associations in the bin ‘contrast = 2–3’ , Wilcoxon rank sum test; Figure 4—figure supplement 1 ) . Thus , in our dataset , we found no evidence that MC responses categorized odorants based on the associated choices . To investigate the basis for the decreased decoder accuracy in ABN ablation animals , we analyzed the responses of individual mitral cells to the eight mixtures . We quantified two measures; the first is the fraction of MCs that responded to at least one mixture , and the second is the fraction of responsive MC-odorant pairs out of all MC-odorant pairs . We found that the fraction of MCs responsive to at least one mixture and the fraction of responsive MC-odorant pairs were both consistently lower in ablation animals compared to control ( control vs . ablation , fraction of cells: p<0 . 001 , fraction of cell-odorant pairs: p<0 . 01 , two-way ANOVA; Figure 5A ) . As a MC can respond to an odorant with increased or decreased activity , we next quantified excitatory and suppressive responses separately . This analysis showed that the excitatory response fraction was not significantly affected by ABN ablation ( control vs . ablation , fraction of cells: p=0 . 0813 , fraction of cell-odorant pairs: p=0 . 6039 , two-way ANOVA; Figure 5B ) . Instead , the decreased responses in ablation animals were primarily due to decreases in suppressive responses ( control vs . ablation , fraction of cells: p<0 . 0001 , fraction of cell-odorant pairs: p<0 . 0001 , two-way ANOVA; Figure 5C ) , suggesting that the net effect of ABNs on MC ensembles is inhibitory . The decrease in suppressive but not excitatory responses of MCs in ablation animals raises the possibility that the decreased suppressive responses may underlie the reduced decoder accuracy in ablation animals . Therefore we explored the relationships of excitatory and suppressive MC responses with decoder accuracy and behavior . We found that the fraction of total ( excitatory and suppressive ) responses positively correlates with decoder accuracy ( fraction of cells: p<0 . 01 , fraction of cell-odorant pairs: p<0 . 01 , Pearson correlation ) and behavioral performance ( cells: p<0 . 01; cell-odorant pairs: p<0 . 05; Figure 6A , B ) . When we only included excitatory responses , however , this relationship was not significant ( decoder accuracy , cells: p=0 . 2089 , cell-odorant pairs: p=0 . 1272; performance , cells: p<0 . 05 , cell-odorant pairs: p=0 . 1459; Figure 6C left and Figure 6D left ) . Instead , the fraction of suppressive responses significantly correlated with both decoder accuracy and behavioral performance ( decoder accuracy , cells: p<0 . 01 , cell-odorant pairs: p<0 . 01; performance , cells: p<0 . 05 , cell-odorant pairs: p<0 . 05; Figure 6C right and Figure 6D right ) . Together these results suggest that young ABNs are essential for high levels of suppressive responses of MCs , which significantly contribute to odorant discriminability by MC population responses . Inhibitory neurons in the olfactory bulb , including ABNs , are major targets of extensive glutamatergic and neuromodulatory projections from higher brain areas . These centrifugal projections are suggested to be sensitive to brain states ( Gilbert and Sigman , 2007; Rothermel and Wachowiak , 2014 ) . Therefore it is tempting to hypothesize that the impact of ABN functions is sensitive to behavioral states such as task engagement . We reasoned that , if this is the case , the differences in the MC responses of control and ablation animals described above would be less pronounced when the mice were not engaged in the task . To test this idea , we performed a new experiment in which another cohort of mice were passively exposed to odorants ( control passive: n = 10; ablation passive: n = 7 ) . Except for the lack of task engagement , all the other conditions were kept identical to the task condition , including VGCC treatment , water restriction , odorant stimulation protocol , and odorant identity ( 4 sessions of 2 ‘easy discrimination’ odorants followed by 10 sessions of 8 ‘difficult discrimination’ odorants ) . Imaging was performed during the passive experience of the eight difficult discrimination odorants that are identical to the task condition . Strikingly , in this passive condition , the fractions of MCs showing excitatory and suppressive responses were no longer statistically distinguishable between ablation and control animals ( control passive vs . ablation passive , fraction of cells: total , p=0 . 6162; excitatory , p=0 . 7625 , suppressive , p=0 . 2620; fraction of cell-odorant pairs: total , p=0 . 2043; excitatory , p=0 . 8691 , suppressive , p=0 . 0625 , two-way ANOVA; Figure 7A–C ) . These results suggest that MC responses , mainly suppressive responses , are increased in a task engagement-dependent manner . We further tested this notion with a linear regression model ( Materials and methods ) . The interaction term for the genotype ( control vs . ablation ) and the condition ( task vs . passive ) was statistically significant for suppressive ( fraction of cells: p=0 . 0101; fraction of cell-odorant pairs: p=0 . 0239 ) , but not excitatory ( fraction of cells: p=0 . 1162; fraction of cell-odorant pairs: p=0 . 7509 ) responses . This result indicates that the effect of task engagement on suppressive , but not excitatory , responses is significantly larger in control animals than in ablation animals , supporting a role for young ABNs in enhancing suppressive responses in a task engagement-dependent manner . Based on these observations , we propose that behavioral states strongly modulate MC activity and in particular suppressive responses , facilitating olfactory discrimination during task engagement . Importantly , this state-dependent enhancement of suppressive responses during task engagement requires young ABNs . Previous studies on the effect of ABN ablation on olfactory behavior have reported inconsistent results . These discrepancies may be due to differences in ablation methods as well as task demands . In this study , we adopted an inducible , genetic ablation method and we confirmed that this method almost completely eliminated ABNs . We and others also have found no evidence of non-specific effects on other cell types ( Cummings et al . , 2014; Snyder et al . , 2011 ) . Thus , this method can ablate young ABNs with high specificity and efficiency , allowing us to investigate its consequences on olfactory behavior and odor representations . We established an olfactory discrimination two alternative choice task with two levels of difficulty . The operant and symmetric nature of the task allowed us to focus on the discrimination ability of individual animals , as opposed to spontaneous discrimination or asymmetric go/no-go tasks in which motivational states are difficult to control . Ablation mice were perfectly capable of discriminating conspicuously different odorants in this task , indicating that young ABNs are not necessary for basic odor processing , consistent with many previous studies . For difficult discrimination , we used eight binary mixtures of similar ratios applied pseudorandomly in each trial . Previous studies showed that tasks involving multiple similar odorants ( Rinberg et al . , 2006; Uchida and Mainen , 2003 ) delivered randomly ( Zariwala et al . , 2013 ) are more difficult than two-odorant tasks . This difficult condition revealed a robust impairment of ABN ablation animals in odor discrimination . Importantly , our ablation method affects adult neurogenesis in both SVZ and hippocampus DG . However , DG-only suppression of ABNs did not produce the deficits in olfactory discrimination . Therefore , we conclude that young ABNs in the olfactory bulb are essential for fine discrimination of odorants . Ablation of any neurons in the olfactory bulb may be expected to lead to deficits in olfactory behaviors . We find that ABNs less than 10–12 weeks old constitute ~10% of all GCs , consistent with a previous report ( Imayoshi et al . , 2008 ) . When we ablated a larger fraction ( ~20% ) of GCs randomly without regard to their age , majority of which were presumably mature GCs , the behavioral impairment was much more subtle . These results support the notion that young ABNs have a unique role in fine olfactory discrimination . However , we acknowledge an important caveat of the random ablation experiments in that the ablation was probably concentrated closer to the injection site and largely spared periglomerular cells , while the Gfap-tk method targeted ABNs in both GC and glomerular layers . To explore the potential neural basis underlying the behavioral impairment of ABN ablation animals , we used two-photon imaging to record the activity of populations of MCs ( Kato et al . , 2012 ) during the task performance , and analyzed data within the entire 4 s stimulus period . We are aware of the previous reports demonstrating the importance of finer time-scale dynamics of MC responses ( Abraham et al . , 2004; Resulaj and Rinberg , 2015; Rinberg et al . , 2006; Uchida and Mainen , 2003; Wachowiak , 2011 ) , which is not accessible with the temporal resolution of our approach . In fact , in certain reaction time tasks , the responses within the first 100 ms of odorant onset are sufficient for discrimination . However , we argue that it is unlikely that the responses within the first 100 ms of odorant onset explain the entirety of odor representations important for odor perception . This may especially be the case in conditions such as the task used here in which mice are not encouraged to react as quickly as possible . Furthermore , our approach affords a unique opportunity to record the activity of a few dozens of MCs longitudinally , allowing us to assess MC ensemble coding in ablation and control animals . We performed MC calcium imaging during the difficult discrimination task to explore the potential neural basis underlying the behavioral impairment in ablation animals . This experiment revealed that the behavioral impairment in ablation animals accompanied a reduced separation of representations of similar odorants by MC ensembles as shown by the decoder analysis . Moreover , we found that the reduced separation of odor representations in ablation animals involved a preferential reduction of MC suppressive responses . The degree of reduction is related to the decoder performance , as the fraction of MC suppressive responses significantly correlated with the decoder accuracy and behavioral performance , supporting the importance of MC suppressive responses . Considering that ABNs exert inhibitory modulation onto MCs , these results together suggest that the ablation of ABNs caused a reduction in MC suppressive responses , which in turn affected the discriminability of MC ensembles . In normal animals , excitatory inputs from sensory neurons combined with local inhibitory control would allow MCs to respond to odorants in both excitatory and suppressive manners ( Yokoi et al . , 1995 ) . The bidirectionality of responses effectively increases the dynamic range of MC responses and would contribute to an enhanced separation of representations of similar odors . Consistently , a modeling study simulated the effect of adult neurogenesis on the olfactory bulb circuitry with excitatory sensory inputs , local GC inhibition and MC outputs and predicted that a constant arrival and activity-dependent survival of ABNs are sufficient to separate MC representations of very similar odorants in an experience-dependent manner ( Cecchi et al . , 2001 ) . A continuous recruitment of new ABNs allows the bulb to adapt to changes in the olfactory environment . Our current results indicate that such a mechanism is particularly sensitive to behaviorally significant experience such as engagement in difficult discrimination . A prediction of our model is that MC suppressive responses and discriminability would be less affected ( more similar to control animals ) in the random ablation animals than in the ABN ablation animals . We made a great effort to test this possibility , but our effort to image MC activity in random ablation animals has so far been unsuccessful . Therefore this remains a question to be addressed in the future . We note that the decoder performance was relatively stable throughout imaging , in contrast to our recent report in which the decoder performance improved during difficult discrimination learning and correlated with behavioral choice on a trial-by-trial basis ( Chu et al . , 2016 ) . The apparent discrepancy likely stems from the fact that Chu et al . investigated changes of representations of novel odorants over time , while in the current study , mice had already been familiarized with the odorants , albeit at different mixture ratios , during the easy discrimination task prior to imaging . Despite the stable decoder accuracy in our current study , the degree of separation of mitral cell responses predicted the final behavioral performance , which is consistent with a previous report ( Gschwend et al . , 2015 ) . It is also noteworthy that we did not detect an effect of categorical associations with lick sides on the degree of pattern separation ( Figure 4—figure supplement 1 ) . One interpretation of these results is that the olfactory bulb performs pattern separation based on the statistics of the olfactory environment ( Chu et al . , 2016 ) and not associations . The downstream areas ( e . g . cortex ) may be able to perform sensory-motor associations more efficiently when olfactory bulb outputs are more decorrelated . Another implication of these results is that there are likely multiple circuit origins of behavioral errors . When MC discriminability is a limiting factor for behavior , MC decoder accuracy may correlate with behavioral choice on each trial . However , in the current task , the downstream areas ( e . g . olfactory cortex ) and their ability to associate MC outputs with behavioral choice may be a main source of behavioral variability . We found that the abundance of MC suppressive responses is highly sensitive to behavioral states . MC suppressive responses are increased during wakefulness compared to anesthetized states , and task engagement further enhances suppressive responses . This is consistent with a previous report stating that suppressive responses in MCs became more prominent during task engagement as opposed to passive exposure ( Fuentes et al . , 2008 ) . Our results further extend these findings and demonstrate that the task engagement-dependent enhancement of suppressive responses is facilitated by young ABNs . During task engagement , ABN ablation animals have fewer suppressive responses than control animals . This difference in suppressive responses was absent during passive exposure . The sensitivity to task engagement may explain the findings from a previous study that broad GC inactivation has only mild effects on MC responses under anesthesia and passive wakefulness ( Fukunaga et al . , 2014 ) . It is known that ABNs receive centrifugal synaptic and neuromodulatory inputs from multiple brain areas , and these inputs can vary depending on brain states . Thus , the state-dependence of the functional role of ABNs can be better appreciated considering previously reported phenomena that synaptic inputs onto developing ABNs within different dendritic compartments formed in a sequential manner ( Kelsch et al . , 2008 ) , with the formation of centrifugal inputs preceding local dendrodendritic inputs ( Whitman and Greer , 2007 ) . It has been shown that the survival rate of ABNs is sensitive to sensory experiences , which then are reflected by neuronal activities . Therefore , we postulate that ABNs that are strongly activated by centrifugal inputs may have a higher chance to survive , which can explain the state-dependent requirement of ABNs for MC suppressive responses that we have observed . The activity of inhibitory circuits has been shown to be sensitive to behavioral states in various brain areas . Intracellular recordings from excitatory neurons in the primary visual cortex revealed that inhibitory inputs are more prevalent in the awake state than in anesthesia ( Haider et al . , 2013 ) . In the primary auditory cortex , task engagement suppresses sound-evoked responses and sharpens the tuning of excitatory neurons ( Lee and Middlebrooks , 2011; Otazu et al . , 2009 ) . This effect is mediated by subtype-specific modulation of local inhibitory neurons ( Kuchibhotla et al . , 2017 ) . In the olfactory bulb , anesthesia suppresses local inhibitory neurons ( Kato et al . , 2012; Wachowiak et al . , 2013 ) . Together with the current study , these results suggest that state-dependent engagement of inhibitory circuits and suppression of excitatory responses may be a common principle conserved across brain areas . We also note that task engagement does not only affect suppressive responses of MCs . In passive exposure , excitatory responses were also reduced compared to the task condition , although this effect was insensitive to ABN ablation ( Figure 7—figure supplement 1 ) . Therefore , it appears that there are additional , ABN-independent mechanisms modulating MC responses in a state-dependent manner . These probably include various feedback systems , which can modulate MC activity and olfactory behavior ( Castillo et al . , 1999; Chaudhury et al . , 2009; Escanilla et al . , 2010; Kapoor et al . , 2016; Linster et al . , 2001; Ma and Luo , 2012; Nunez-Parra et al . , 2013; Rothermel et al . , 2014 ) . It is likely that some of the functions of these systems are independent of ABNs . Taken together , we propose a model that task engagement increases the dynamic range of MC responses through top-down modulation from higher brain areas , which acts at least partially through young ABNs . Consistent with this notion , inactivation of piriform cortex , a main source of feedback projections to the olfactory bulb , enhances excitatory MC responses ( Otazu et al . , 2015 ) , supporting the inhibitory role of cortical feedback . These dynamics of ABNs may underlie the observations that experience and learning profoundly shape the representations of odorant stimuli by olfactory bulb principal neurons ( Chu et al . , 2016 , 2017; Doucette and Restrepo , 2008; Gschwend et al . , 2015; Kato et al . , 2012; Yamada et al . , 2017 ) . Thus , the olfactory bulb functions as a dynamic , adaptive filter for incoming odorant information depending on behavioral demands , and adult neurogenesis is essential for this adaptive role of the olfactory bulb . All procedures were in accordance with protocols approved by the Institutional Animal Care and Use Committee at UCSD or Kyoto University and guidelines of the National Institute of Health . For all experiments , mice were housed in plastic cages with standard bedding in a room with a reversed light cycle ( 12 hr-12hr ) , and all experiments were performed during the dark period . All experiments except suppression of hippocampal postnatally-born neurons were performed at UCSD . Gfap-tk mice were generous gifts from H . Cameron with ICR background . Cdhr1-Cre mice were originally acquired from RIKEN Brain Research Center and backcrossed at least four generations to C57Bl/6 . Only male mice were used . All littermates were used for experiments , roughly 50% of which were positive for Gfap-tk , and the mice negative for Gfap-tk served as control . The experimenters were blinded to the genotype of each mouse until the end of the experiments . The genotypes were confirmed by both PCR and post hoc Doublecortin immunostaining , which were always consistent with each other ( PCR negative mice always showed Doublecortin signals and vice versa ) . Suppression of hippocampal postnatally-born neurons was performed at Kyoto University . mGfap-Cre mice ( Garcia et al . , 2004 ) were crossed with Slc17a7-LSL-TeNT mice ( Sakamoto et al . , 2014 ) . Both strains were maintained on the C57Bl/6 background . The experimenters were blind to the genotype of each mouse during the experiments , after which double transgenic mice were identified by PCR . Slc17a7-LSL-TeNT single transgenic mice served as control . No behavioral abnormalities were observed in the mGfap-Cre and Slc17a7-LSL-TeNT single transgenic mice . All behavioral tests were carried out with 3-months-old male mice . VGCC ( Genentech ) was dissolved in drinking water at 0 . 63 mg/ml before water restriction , and mixed with powdered food ( Harlan , Indianapolis , IN ) at 0 . 44 mg/g during water restriction , to achieve approximately 0 . 1 mg/g body weight/day . Mice were 10–12 weeks old at the beginning of VGCC treatment . After 6 weeks of continuous VGCC treatment , mice were anesthetized with isoflurane ( 3% induction , 0 . 7–2% maintenance ) and surgeries were performed as previously described ( Kato et al . , 2012 ) . Briefly , a stainless-steel custom headplate was secured onto the skull with cyanoacrylate glue , and an optical glass window ( 1 × 2 mm , oval ) was implanted above the right olfactory bulb craniotomy and was secured by dental cement . To express GCaMP6f in mitral cells , a viral vector containing a Cre-dependent , GCaMP6f-expressing construct ( AAV2 . 1 hsyn-FLEX-GCaMP6f , UPenn Vector Core , 1:11 diluted in saline ) was injected into the craniotomy ( 20 nl / site , four sites , 250 μm depth ) . To ablate a random subset of cells in GCL , a mixture of viruses containing Cre-expressing construct ( AAV2 . 1-CMV-PI-Cre-rBG , UPenn Vector Core , 1:10 dilution in saline ) and Cre-dependent modified Caspase3 ( Yang et al . , 2013 ) , AAV2 . 1-EF1a-FLEX-taCasp3-TEVp , custom prep by UPenn Vector Core , 1:1 dilution in saline ) was injected into the olfactory bulb ( 300 nl or 500 nl , one site , 0 . 75 mm M-L , 0 . 8 mm anterior from the inferior cerebral vein , 1 . 5 mm D-V , injection speed: 100 nl / min ) through a small craniotomy . For all behavioral experiments and a subset of histology experiments , the injections were bilateral . For the other histology experiments , the injections were unilateral and the uninjected hemisphere served as control . To validate the effectiveness of adult neurogenesis ablation , after 6 weeks of continuous VGCC treatment , mice ( six control , six ablation ) were treated with BrdU for three consecutive days , and were sacrificed 7 days later for immunostaining . BrdU powder was dissolved in drinking water at 1 mg/ml to achieve approximately 0 . 2 mg/g body weight/day . 30 μm-thick olfactory bulb coronal sections were prepared with a microtome ( Thermo Fisher ) and mounted on pre-coated slides . Immunostaining was then performed with overnight primary antibody and 2 hr secondary antibody incubation . For BrdU staining , sections were incubated at 37°C in HCl ( 6% in water ) for 30 min , and neutralized by borate acid buffer ( 0 . 5 M ) for 10 min prior to incubation with the primary antibody . Both primary and secondary antibodies were diluted in blocking buffer ( 0 . 3% TritonX-100 , 1% serum from the same species as secondary antibody , 0 . 1% bovine serum albumin , 0 . 1 M ph7 . 4 PBS ) . BrdU: primary ( rat , AbD serotec , Oxford , UK , RRID: AB_10015293 ) , 1:500 , secondary ( goat , Alexa 488 , Thermo Fisher , Waltham , MA , RRID: AB_2534074 ) , 1:1000 . Doublecortin: primary ( goat , Santa Cruz , Dallas , TX ) , 1:400 , secondary ( donkey , Alexa 488 , Thermo Fisher , Waltham , MA , RRID: AB_2534102 ) , 1:1000 . GFAP: primary ( goat , Santa Cruz , Dallas , TX ) , 1:400 , secondary ( same as doublecortin ) . NeuN: primary ( mouse , Millipore , Temecula , CA ) , 1:400 , secondary ( goat , Alexa 488 , Thermo Fisher , Waltham , MA , RRID: AB_2633275 ) , 1:1000 . DAPI: 1:10 , 000 ( Invitrogen , Carlsbad , CA ) for Figure 1C , D , E , and Vectashield mounting medium ( Vector Labs , Burlingame , CA ) for Figure 3C , E . GFAP , BrdU , NeuN and DAPI quantification was performed manually using ImageJ . Representative sections ( ~4 for each animal ) were chosen , and in each section , four rectangle areas were selected for counting , each encompassing the entire depth of the GC layer from dorsal , ventral , medial and lateral sides where signals were relatively homogenous . For GFAP signals , only complete structures containing soma were counted . For BrdU , all clearly visible puncta were included . To measure GCL width , 3–4 coronal sections from the widest segment of each OB were selected , and the distances between the central line of ventricle to the mitral cell layer on both medial and lateral sides were measured using ImageJ , and then averaged . Odorants ( Sigma ) were diluted in mineral oil ( Thermo Fisher , Waltham , MA ) to a calculated vapor pressure of 200 ppm . A custom-built olfactometer mixed saturated odorant vapor 1:1 with filtered , humidified air for a final concentration of 100 ppm . Air flow rate was controlled at 1 L / min by a mass flow controller ( Aalborg , Orangeburg , NY ) . Heptanal and Ethyl-tiglate were selected based on their structural dissimilarity and strong odorant-evoked responses in dorsal olfactory bulb . Water restriction started ~1 week after surgery and 14–18 days prior to the start of behavioral training . Mice were given at least 1 ml of water per day to maintain the body weight ≥80% of the initial value . The behavioral program was controlled by a real-time system ( C . Brody ) . Two lick ports with infrared beam detector were available for left and right licks . A correct trial ( determined by the first lick during the answer period ) was rewarded with ~6 μl of water . Each daily training session consisted of 150 trials unless mice disengaged earlier . Two-photon imaging was performed with a commercial microscope ( B-scope , Thorlabs , Newton , NJ ) with 925 nm laser excitation ( Mai-Tai , Spectra-physics , Santa Clara , CA ) at the frame rate of 26–28 Hz . Each frame was 512 × 512 pixels with the average field of view of 546 × 467 μm . Imaging was performed continuously within each of 4000-frame ( ~44 s ) segments , which were separated by a 6 s inter-segment interval . Trials that overlapped with these intervals were discarded . The average image from the first imaging session was used as a template to identify the same imaging field in the following sessions . The image time series were first processed for full-frame motion correction with a custom program in MATLAB .

Most brain cells or neurons form early in life . Yet , in some parts of the brain , new neurons develop throughout adulthood , in a process called adult neurogenesis . These new neurons are incorporated into existing brain circuits and likely help the brain process information . In rodents , adult neurogenesis produces many new cells in the olfactory bulb , a part of the brain that processes smells . This is likely because the sense of smell is important for the survival of these animals . What these adult-born neurons do and how they aid the rodent’s sense of smell is not clear . Previous studies have had conflicting results about whether these cells help animals distinguish smells and under what circumstances . More studies about how these adult-born neurons become incorporated in the brain and how they aid creatures’ sense of smell could help scientists studying brain diseases . Now , Li et al . show that mice that lack adult-born neurons have a difficult time distinguishing very similar smells . In the experiments , mice were genetically engineered to suppress the formation of new neurons in adult animals . These mice lacking adult-born neurons and typical mice were trained to do tasks that require them to distinguish similar or very different scents . While the animals were completing these tasks , Li et al . used a technique called two-photon calcium imaging to see what was happening in cells in the olfactory bulb . The experiments revealed altered neuron activity in the genetically engineered animals compared with normal ones when they were trying to distinguish similar smells . Yet there was no difference when the mice distinguished very different scents . This suggests that adult-born cells are important for mice working to distinguish scents . The mechanisms at work in the mice may be the same ones that help humans distinguish wines or perfumes . Learning more about how new cells form in adult brains could help scientists understand these processes and develop treatments for brain diseases in humans .

2018

Adult-born neurons facilitate olfactory bulb pattern separation during task engagement

The role of mechanical signals in cell identity determination remains poorly explored in tissues . Furthermore , because mechanical stress is widespread , mechanical signals are difficult to uncouple from biochemical-based transduction pathways . Here we focus on the homeobox gene SHOOT MERISTEMLESS ( STM ) , a master regulator and marker of meristematic identity in Arabidopsis . We found that STM expression is quantitatively correlated to curvature in the saddle-shaped boundary domain of the shoot apical meristem . As tissue folding reflects the presence of mechanical stress , we test and demonstrate that STM expression is induced after micromechanical perturbations . We also show that STM expression in the boundary domain is required for organ separation . While STM expression correlates with auxin depletion in this domain , auxin distribution and STM expression can also be uncoupled . STM expression and boundary identity are thus strengthened through a synergy between auxin depletion and an auxin-independent mechanotransduction pathway at the shoot apical meristem . Almost 100 years ago , Thomas D’Arcy Wentworth Thompson proposed that shape of plants and animals could be described as a consequence of the laws of physics ( D’Arcy Thompson , 1917 ) . While this view has been overshadowed by the rise of molecular biology and morphogen-based patterning mechanisms , a growing contribution of mechanics in shape changes is currently emerging . In this framework , an instructing role of mechanical forces has been successfully explored theoretically ( Shraiman , 2005; Aegerter-Wilmsen et al . , 2007 ) and there is today accumulating evidence that many molecular actors involved in development are under both biochemical and mechanical control . In particular , studies on single animal cells show that mechanical forces contribute to the control of cell division and cell polarity , and relevant mechanotransduction pathways have been identified ( e . g . Houk et al . , 2012; Fink et al . , 2011; Thery et al . , 2007 ) . The idea that mechanical stress may also play a role in defining cell identity has also emerged in the past decade ( Engler et al . , 2006 Aliee et al . , 2012; Landsberg et al . , 2009; Farge , 2003; Brunet et al . , 2013 ) . However , this pioneering work is still debated and a role of mechanical signals in cell identity during development remains an open question . Furthermore , all the currently known mechanotransduction pathways involve elements of established biochemical-based transduction pathway ( e . g . Janmey et al . , 2013; Jaalouk and Lammerding , 2009; Orr et al . , 2006 ) . Assuming mechanical signals play an important role in development , one may thus question the added value of mechanical signals in development if their action is so tightly coupled to biochemical signaling . Because their development is slow , iterative and does not involve cell movements , plants are systems of choice to explore the possible contribution of mechanical signals in proprioception , i . e . in channeling growth and identity from intrinsically generated mechanical stresses . Mechanical cues contribute to the emergence of lateral roots in Arabidopsis ( Ditengou et al . , 2008; Richter et al . , 2009 ) , the expression level of the transcription factor PtaZFP2 correlates to the amount of bending in poplar stems ( Coutand et al . , 2009 ) and the expression of the ELA1 gene in a specific cell layer of the developing seed is triggered by mechanical signals that are generated by the growth of the embryo against the endosperm ( Creff et al . , 2015 ) . While these studies suggest that physical forces can contribute to cell identity definition in plants , this remains to be fully demonstrated . A fitting system for this question is the shoot apical meristem ( SAM ) , which contains a plant stem cell niche and controls the formation and identity of all aerial organs . Because cells are glued to each other , differential growth generates mechanical conflicts leading to shape changes . For instance , tissue folding occurs in the boundary domain of the SAM between the slow growing meristem and the fast growing organ . Because the epidermis is load-bearing ( Dumais and Steele , 2000; Kutschera and Niklas , 2007 ) , the boundary domain is characterized by a highly anisotropic mechanical stress; this mechanical stress controls microtubule orientation , which in turn channels growth direction and promotes tissue folding ( Burian et al . , 2013; Hamant et al . , 2008 ) . A contribution of mechanical stress in the polarity of the auxin efflux carrier PIN-FORMED 1 ( PIN1 ) was also proposed suggesting an indirect contribution of mechanical signals in auxin patterns and thus in organogenesis at the shoot apex ( Heisler et al . , 2010; Nakayama et al . , 2012 ) . While the genetic bases of meristem functions are now well documented , a link between these genetic regulators and mechanical signals remains to be identified in the SAM . The homeodomain transcription factor SHOOT MERISTEMLESS ( STM ) is a key regulator of meristem functions and its expression is often considered as the best marker of meristematic identity . In the shoot apical meristem , STM is expressed ubiquitously , with the exception of young primordia where it is down-regulated ( Long et al . , 1996 ) . Interestingly , the STM promoter has previously been reported to be more active in the boundary domains of the SAM ( Heisler et al . , 2005; Leasure et al . , 2009; Kim et al . , 2003; Jurkuta et al . , 2009 ) . In line with this observation , STM expression is regulated by several boundary specific genes ( Aida et al . , 1999; Borghi et al . , 2007; Lee et al . , 2009; Takada et al . , 2001 ) , in part via auxin signaling ( Aida et al . , 2002; Treml et al . , 2005 ) . Because the boundary domain is also a site under mechanical stress , we investigated whether such stresses could act as signals to control STM expression at the shoot apical meristem . Because it is easily accessible , we focus our analysis on the shoot apical meristem ( SAM ) at inflorescence stage; the generated organs are thus floral meristems . We generated a transcriptional fusion pSTM::CFP-N7 with the 5 , 7 kb region upstream of the STM gene ( AT1G62360 ) and observed the presence of an enhanced STM promoter activity in the SAM boundaries , as previously shown ( Heisler et al . , 2005; Leasure et al . , 2009; Kim et al . , 2003; Jurkuta et al . , 2009 ) , n > 20 , Figure 1A ) . To further confirm this result , which was obtained in dissected meristems from greenhouse-grown plants , we also analyzed the CFP signal in meristems from NPA-treated in vitro grown seedlings . In these conditions , polar auxin transport is inhibited and naked meristems are generated ( Grandjean et al . , 2004 ) . When plants were taken off the drug and started to initiate new organs , higher CFP signal was observed specifically in the boundary domain ( Figure 1—figure supplement 1 ) . 10 . 7554/eLife . 07811 . 003Figure 1 . Correlation between pSTM::CFP-N7 expression level and tissue folding at the boundary . ( A ) pSTM::CFP-N7 expression pattern in the SAM . Membranes are labeled with FM4-64 ( white , lower panel ) and pSTM::CFP-N7 expression is shown using the Fire lookup table in ImageJ ( upper panel , n > 20 ) . ( B ) Longitudinal optical sections ( 5 μm thick maximal projection of orthogonal views ) through the middle of five successive boundaries of a representative meristem expressing pSTM::CFP-N7 . Note the increase of pSTM::CFP-N7 signal intensity in the boundary as the crease between organ and meristem becomes deeper . ( C ) Close-ups showing a correlation between pSTM::CFP-N7 signal intensity ( upper panels ) and Gaussian curvature ( lower panels , see Material and methods ) in three successive boundaries of the meristem presented in A . ( D and E ) Quantification of the correlation between pSTM::CFP-N7 signal intensity ( upper panel ) and Gaussian curvature ( lower panel ) in the meristem presented in A ( see Material and methods ) . ( D ) The white outline encloses the cells that are used for the graph presented in E . ( E ) pSTM::CFP-N7 signal intensity is plotted against Gaussian curvature . Values are compared using a bilateral Student test . The same correlation was observed in 5 independent meristems . Scale bars , 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 07811 . 00310 . 7554/eLife . 07811 . 004Figure 1—figure supplement 1 . Time lapse imaging of a meristem recovering from NPA treatment and expressing pSTM::CFP-N7 . i1 and i2 marks the presence of new initia where the CFP-N7 signal decreases and the white arrows points toward a new developing boundary where the CFP-N7 signal increases . Note that t = h corresponds to 24 h after transfer to a NPA free medium . Scale bars: 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 07811 . 00410 . 7554/eLife . 07811 . 005Figure 1—figure supplement 2 . Correlation between pSTM::CFP-N7 expression level and tissue folding at the boundary . ( A ) Correlation between pSTM::CFP-N7 signal intensity and Gaussian curvature in another meristem than the one presented in Figure 1 . The white outline encloses the cells that are used for the graph presented in B ( See Material and methods ) . ( B ) Quantitative measurement of the negative correlation between pSTM::CFP-N7 intensity and Gaussian curvature in the meristem presented in A ( n = 231 cells ) . Successive values are displayed with a Student confidence interval ( α = 0 . 05 ) and compared using a bilateral Student test . Scale bars: 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 07811 . 00510 . 7554/eLife . 07811 . 006Figure 1—figure supplement 3 . pPDF1::CFP-N7 expression pattern in the SAM . Z-projection of a meristem expressing the L1 transcriptional reporter pPDF1::CFP-N7 ( left panel ) . CFP-N7 Signal intensity map of the signal and Gaussian curvature extracted using the level set method and MorphoGraphX ( right panels ) . Scale bars: 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 07811 . 006 When performing live-imaging of the meristem expressing pSTM::CFP-N7 , we found a correlation between STM promoter activity in the boundaries and the progressive formation of a crease in this domain , at least qualitatively ( Figure 1B ) . To assess this quantitatively , we measured the Gaussian curvature of the boundary at different stages of development in parallel with CFP signal intensity . Existing quantification protocols were not adequate: the replica method for instance involves the application of dental resin to make a cast of the meristem ( e . g . Kwiatkowska and Dumais , 2003 ) and this may impact gene expression; the image analysis software MorphoGraphX could provide a mesh surface together with gene expression levels , but some errors in the most curvy parts of the meristem could not be avoided . We thus implemented the level set method ( Sethian , 1999 ) in Python to detect the exact surface of the meristem: a suitable surface obtained by the thresholding of the image is evolved to the smooth surface that is the most accurate representation of the surface contrast in the 3D images . Second , MorphoGraphX was used to mesh the surface ( Kierzkowski et al . , 2012 ) , and to compute the Gaussian curvature taking into account neighborhoods within a 15 μm radius ( Figures 1C and Figure 1—figure supplement 2 ) . Using this pipeline , a negative correlation between CFP signal intensity and Gaussian curvature could be revealed and quantified ( Figure 1C–E , Figure 1—figure supplement 2 ) . As a negative control , and using the same pipeline , no clear correlation between Gaussian curvature and CFP signal could be observed in a line expressing a transcriptional fusion pPDF1::CFP-N7 that exhibits a relatively homogeneous signal in the whole meristem epidermis ( Figure 1—figure supplement 3 ) . To test this correlation further , we also analyzed the expression pattern of STM in the pSTM::ALcR AlcA::GFP line ( named hereafter pBOUND>>GFP ) which contains 4 . 4 kb sequence upstream of the STM gene and has already been described previously ( Laufs et al . , 2004 ) . In this line , expression is restricted to boundaries , thus confirming the correlation between the GFP signal intensity and the extent of curvature at the boundary at least qualitatively ( n > 20 , Figure 2A ) . The correlation could also be observed in meristems from NPA-grown seedlings ( Figure 2—figure supplement 1 ) . In this line , the signal was so clear-cut , that the total area of GFP expressing cells on longitudinal sections could be qualitatively correlated with the formation of a crease ( Figure 2B ) . To quantify this , we generated optical longitudinal section through the middle of each emerging primordia , measured the surface area of expression of pBOUND>>GFP in these sections and plotted it against the angle of tissue folding at the same position ( see Material and methods ) . As in the pSTM::CFP-N7 line , we measured a strong correlation between the area of pBOUND>>GFP expression and the angle of boundary folding measured on the orthogonal sections ( n = 154 , 5 SAMs observed at 5 time points over 48 hr , Figure 2D ) . A similar correlation could be measured in another independent live-imaging experiment ( Figure 2—figure supplement 2 , n = 193 , 5 SAMs observed at 6 time points over 39 h ) . 10 . 7554/eLife . 07811 . 007Figure 2 . Correlation between pBOUND>>GFP expression level and tissue folding at the boundary in WT and bot1-7 . ( A ) pBOUND>>GFP expression pattern in WT ( ecotype WS-4 ) and bot1-7 meristems . Membranes are labeled with FM4-64 ( white ) and pBOUND>>GFP expression is shown using the Fire lookup table in ImageJ . ( B ) Longitudinal sections through the middle of successive boundaries of the meristems presented in A ( 2 μm thick maximal projection of orthogonal views ) . Organ size ( surface area as viewed from the top ) is written in red for each stage . Note the delay in tissue folding and GFP signal expression in bot1-7 when compared to the WT . The white line marks the outer surface of the SAM . ( C ) Quantification of the delay in curvature at the boundary in bot1-7: Folding angle is measured on orthogonal views and organ size is estimated from the measurement of surface area on top views . ( D ) The correlation between the folding angle of the boundary and the area of pBOUND>>GFP expression is maintained in bot1-7 ( both parameters are measured on orthogonal sections; WT: n = 130 from 5 SAM followed during a time lapse of 5 time points during 48 h , bot1-7: n = 79 from 3 SAM followed during a time lapse of 5 time points during 48 h ) . Values are displayed with a Student confidence interval ( α = 0 . 05 ) and compared using a bilateral Student test . Scale bars: 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 07811 . 00710 . 7554/eLife . 07811 . 008Figure 2—figure supplement 1 . Time lapse of a pBOUND>>GFP meristem recovering from NPA treatment . Note the signal induction in old and new boundaries . Scale bar , 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 07811 . 00810 . 7554/eLife . 07811 . 009Figure 2—figure supplement 2 . Correlation between pBOUND>>GFP expression area and tissue folding at the boundary . Quantitative correlation between the folding angle of the boundary and the area of pBOUND>>GFP expression on longitudinal sections from another independent time course as the one presented in Figure 2 ( n = 154 , 5 meristems imaged 5 times over a time course of 48 hr ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07811 . 009 To test the strength of the correlation between STM promoter activity and meristem shape , we next analyzed STM promoter activity in the katanin mutant allele bot1-7 in which the presence of disorganized microtubules leads to the formation of a shallow crease at the boundary ( Uyttewaal et al . , 2012; Bichet et al . , 2001 ) . We reasoned that if STM promoter activity is truly correlated with curvature at the boundary , we should still be able to detect such correlation in the bot1-7 mutant background , although with a delay in time . First , we measured the folding angle with respect to organ size in bot1-7 . As expected , we observed a reduction in folding angle for a given primordium size in bot1-7 when compared to the WT , demonstrating that the relation between tissue folding and organ emergence is impaired in bot1-7: large organs still have shallow boundaries ( Figure 2A , B , C ) . Next , we measured the GFP signal area in the boundaries of the bot1-7 pBOUND>>GFP line as described above . Note that both bot1-7 and pBOUND>>GFP are in the WS-4 ecotype , allowing comparison between mutant and WT backgrounds . Whereas pBOUND>>GFP expression area was still correlated to tissue folding , we also observed a reduction in the area of GFP expression for a given primordium stage , when compared to the WT indicating a delay in pBOUND>>GFP appearance in bot1-7 ( Figure 2D , WT: n = 154 measured on 5 meristems at 5 time points; bot1-7: n = 79 measured on 4 meristems at 5 time points ) . Therefore , modifying the shape of the SAM in bot1-7 did not abolish the correlation between STM promoter activity and curvature , and instead demonstrated that STM promoter activity scales to Gaussian curvature values even when the relation between organ size and boundary shape is affected . Based on these results , STM promoter activity can thus be considered as a read-out of the extent of folding in the boundary of the SAM . Even though we find a correlation between STM expression and tissue folding , it is not clear whether STM promoter activity at the boundary using a reporter line truly recapitulates STM expression in the SAM . To check this , we first crossed the strong stm mutant allele stm-dgh6 ( Aida et al . , 2002 ) with a previously described pSTM::STM-Venus line , whose expression is enhanced in boundaries , as observed in our transcriptional marker lines ( Heisler et al . , 2005; Besnard et al . , 2014; Figure 3A ) . The stm-dgh6 mutant exhibits the typical stm phenotype with two partly fused cotyledons , at an angle divergent from 180° , and a development arrest . At the same stage , the WT can exhibit up to 8 leaves while the stm-dgh6 only displays two old cotyledons ( Figure 3B ) . As observed in other strong stm alleles , stm-dgh6 was able in rare cases to reinitiate organogenesis and generate a lot of vegetative tissues . In our growth conditions , we found that roughly 1out of 30 stm-dgh6 plants managed to grow beyond the cotyledon stage: these plants generated many whorled leaves and a few sterile flowers ( Figure 3C ) . To test whether pSTM::STM-Venus can complement the stm-dgh6 phenotype , we selected 96 plants homozygous for the pSTM::STM-Venus construct and segregating the stm-dgh6 mutation . After genotyping for the presence of the stm-dgh6 mutation , we could not distinguish the WT , stm-dgh6 heterozygotes and stm-dgh6 homozygote plants visually , confirming that pSTM::STM-Venus can fully complement the mutation ( Figure 3D-F , Figure 3—figure supplement 1 ) . Importantly , we checked that STM-Venus was expressed in the stm-dgh6 homozygote background by confocal microscopy . We found that the signal was comparable to that of pSTM::STM-Venus in a WT background , notably with an enhanced signal in boundaries ( Figure 3G ) . In a few plants , we also observed larger stems and meristems , suggesting that pSTM::STM-Venus expression level might be slightly higher than that of the endogenous gene . These results demonstrate that higher STM promoter activity at the boundary in the marker lines reflects the endogenous STM expression pattern . Based on these data , and to ease the reading for the rest of the paper , we will use the wording 'STM expression' when analyzing the fluorescent signal in STM transcriptional reporter lines . 10 . 7554/eLife . 07811 . 010Figure 3 . Organ separation requires STM expression at the boundary . ( A ) Representative expression pattern of the translational fusion pSTM::STM-Venus in a FM4-64 stained meristem showing an increased signal intensity in boundaries . Scale bar , 20 μm . ( B–F ) The translational fusion pSTM::STM-Venus partially rescues the phenotype of the strong mutant allele stm-dgh6 ( n = 5 ) . ( B ) Phenotype of 3-week-old WT and stm-dgh6 plants . Note the absence of postembryonic organs in the mutant . Scale bars , 1 cm . Aerial phenotype of 2-month old stm-dgh6 plants . Scale bar , 1 cm . ( D ) Representative stm-dgh6 pSTM::STM-Venus plant . Scale bar , 1 cm . ( E ) Representative stm-dgh6 pSTM::STM-Venus inflorescence . Scale bar , 1 cm . ( F ) Representative stm-dgh6 pSTM::STM-Venus rosette . Scale bar , 1 cm . ( G ) Representative expression pattern of the translational fusion pSTM::STM-Venus in a stm-dgh6 ( -/- ) meristem showing a similar expression pattern as in the WT . Scale bar , 20 μm . ( H , I ) Homogeneous expression pattern of the translational fusion pSTM::STM-Venus in two independent pSTM::STMamiRNA lines . Scale bar ( microscopy ) , 20 μm . Scale bar ( whole plant ) , 1 cm . ( J ) FM4-64 stained WT meristem ( ecotype Col-0 ) , Gaussian curvature extracted using the level set method and MorphoGraphX , Scale bar ( microscopy ) , 20 μm . Scale bar ( inflorescences ) , 1 cm . ( K ) FM4-64 stained pSTM::STMamiRNA pSTM::STM-Venus meristem , Gaussian curvature extracted using the level set method and MorphoGraphX . Scale bar , 20 μm . Boundaries do not scale to the reduced meristem size; inflorescence phenotype with fusion events . Scale bar , 1 cm . DOI: http://dx . doi . org/10 . 7554/eLife . 07811 . 01010 . 7554/eLife . 07811 . 011Figure 3—figure supplement 1 . Molecular characterization of the stm-dgh6 pSTM::STM-Venus . ( A ) 48 pSTM::STM-Venus plants segregating the stm-dgh6 mutation . ( B ) stm-dgh6 genotyping of the plants shown in A . No obvious difference can be detected between the different lines . DOI: http://dx . doi . org/10 . 7554/eLife . 07811 . 01110 . 7554/eLife . 07811 . 012Figure 3—figure supplement 2 . Molecular characterization of the pSTM::STMamiRNA lines . ( A ) STMamiRNA expression in 2-week-old seedlings by qPCR in three independent pSTM::STMamiRNA lines ( see Material s and methods ) . Values are displayed with a Student confidence interval ( α = 0 . 05 ) and compared using a bilateral Student test . ( B ) STM expression in 2-weeks-old seedlings by qPCR in three independent pSTM::STMamiRNA lines , as in A . DOI: http://dx . doi . org/10 . 7554/eLife . 07811 . 012 To further confirm the biological relevance of higher STM promoter activity at the boundary , we next attempted to down-regulate STM expression at the boundary . Genetic evidence suggests that , in addition to its role in meristem maintenance , STM also contributes to organ separation in the meristem , in conjunction with other boundary expressed genes , such as CUC , KNAT6 , JLO and LOF genes ( Lee et al . , 2009; Aida et al . , 2002; Belles-Boix et al . , 2006; Spinelli et al . , 2011; Borghi et al . , 2007; Aida et al . , 1999 ) . Lack of STM activity can generate organ fusions ( e . g . Endrizzi et al . , 1996 ) and can enhance the fusion phenotype of cuc1 and cuc2 mutant seedlings , revealing a role of STM in boundary formation during the early stages of embryo development ( Aida et al . , 1999; 2002 ) . This function also involves the STM paralog KNAT6 gene , which is expressed in the boundary domain of the meristem in embryos ( Belles-Boix et al . , 2006 ) . Based on this literature , the expression of STM at the boundary is generally thought to control organ separation , but this has not been formally demonstrated . To address this hypothesis , we generated lines expressing an artificial microRNA ( amiRNA ) against STM under the control of the STM promoter . In these lines , STM expression level was not always negatively correlated to STM amiRNA expression level; yet , STM expression was down-regulated in all the lines we analyzed showing that this amiRNA can effectively inactivate STM to some extent ( Figure 3—figure supplement 2 ) . Given the expression pattern in the STM transcriptional marker lines , we reasoned that in these lines , we should primarily inactivate STM in the boundaries and to a lesser extent in the rest of the meristem . To test whether the amiRNA was effective at downregulating STM mRNA accumulation in boundaries , we introgressed the pSTM::STMamiRNA in the pSTM::STM-Venus lines . As expected , the STM-Venus signal was slightly ( Figure 3H ) or more strongly reduced ( Figure 3I ) at the boundary , leading to a homogeneous STM-Venus signal in the meristem . Although meristem size was affected in these lines ( Figure 3J and K ) , meristem termination defects were rare when compared to stm mutants . In particular , many normal and fertile flowers were generated ( Figure 3K ) . We also observed that Gaussian curvature in the boundaries did not scale to meristem size ( Figure 3K ) . Strikingly , major fusion defects were observed in most of the lines we generated , thus showing that organ separation was impaired in these lines ( Figure 3K ) . The presence of wide boundaries ( relative to meristem size ) together with a reduction in meristem size that brings adjacent boundaries next to one another , may cause such fusion events . Because the fusions are revealed much later in development , we cannot completely exclude other scenarios . Note that STM mRNA , protein or amiRNA can move between non-boundary cells and boundary cells . Yet , as a negative impact of the amiRNA on STM-Venus signal could be detected at the boundary , these data rather confirm that STM expression at the boundary has functional implications and further support its role in organ separation . The fact that STM expression correlates with curvature at the boundary can simply be explained by the fact that STM reduces the growth rate at the boundary , leading to tissue folding . KNOX genes can cause alterations in curvature in otherwise smooth surface ( Long and Barton , 1998; Barkoulas et al . , 2008 ) . The impact of STM on crease formation at the boundary could be mediated by inhibiting cell growth in this domain . Consistent with this scenario , KNOX target genes include genes involved in auxin transport as well as genes involved in cell wall synthesis ( Bolduc et al . , 2012 ) . Given the strong correlation between STM promoter activity and tissue folding , we investigated whether a signal related to curvature could add robustness to the STM expression pattern at the boundary . The plant hormone auxin could play such a role . A local auxin peak is one of the earliest marker of organ initiation , and conversely auxin depletion is an early marker of the boundary domain ( de Reuille et al . , 2006; Reinhardt et al . , 2003; Heisler et al . , 2005 ) . As auxin keeps accumulating in the organ , outgrowth goes on and curvature at the boundary increases . Interestingly , auxin and KNOX proteins are known to act antagonistically . Disrupting auxin transport with NPA or in pin1 and pid mutant enhances class I KNOX gene expression ( Scanlon , 2003; Furutani et al . , 2004 ) . Conversely , KNAT1 inactivation can partially rescue the leaf number defects in pin1 , suggesting that the auxin signaling pathway also promotes organ emergence by repressing the expression of KNOX genes ( Hay et al . , 2006 ) . In the boundary , the PIN1-dependent auxin minimum was also shown to promote axillary meristem formation whereas ectopic auxin production in the boundary inhibits axillary meristem formation ( Wang et al . , 2014 ) . This is further consolidated by the analysis of the DII-Venus auxin sensor , which is degraded in the presence of auxin and exhibits a strong signal at the boundary ( Brunoud et al . , 2012; Vernoux et al . , 2011 ) . Consistent with the characteristic localization of PIN1 transporters around the boundary ( Figure 4—figure supplement 1; Heisler et al . , 2005 ) , we also observed that a higher DII-Venus signal in the boundaries correlates with an increase in pSTM::CFP-N7 expression ( Figure 4A ) . Interestingly , while this correlation could also be observed on longitudinal sections through the boundary as it folds ( Figure 4B ) , we also found that this correlation is stronger in the L1 layer: In the L2 layer , pSTM::CFP-N7 expression increased during crease formation , but this was not always correlated with an increase in DII-Venus signal ( Figure 4B ) . 10 . 7554/eLife . 07811 . 013Figure 4 . The DII-Venus and pSTM::CFP-N7 signals largely overlap and can be uncoupled . ( A ) Projection of a representative meristem expressing both pSTM::CFP-N7 and DII-Venus-N7 and stained with FM4-64: both signals are induced in the boundary . ( B ) Orthogonal sections through the middle of the boundaries of the successive primordia of the SAM presented in showing an overlap of both signals , except in the L2 layer . ( C , D ) Overnight treatment with 10 μM of synthetic auxin 2 , 4-D on dissected meristems: ( C ) no effect on pSTM-CFP expression after 2 , 4-D application ( Control: n = 3 , 2-4D treatment: n = 3 ) . White arrows point at new CFP signals in boundaries ( C and D ) ; ( D ) total degradation of DII-Venus after 2 , 4-D application ( Control: n = 11 , 2-4D treatment: n = 12 ) . Scale bars , 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 07811 . 01310 . 7554/eLife . 07811 . 014Figure 4—figure supplement 1 . PIN1 localization in the SAM . Whole-mount PIN1 immunolocalization in the SAM ( n = 6 ) . ( B ) close-up from ( A ) . White arrows point at boundaries where the polarity of PIN1 is strengthened and predicts an auxin depletion in this domain . Scale bars , 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 07811 . 014 So far , our data are thus consistent with a scenario in which a local depletion of auxin leads to the initial induction of STM expression at the boundary . If this scenario were true , the application of exogenous auxin should thus inhibit STM expression at the boundary . To test this hypothesis , we applied the synthetic auxin 2 , 4-D onto the meristem and observed the expression of STM . As expected , DII-Venus levels were dramatically decreased in these conditions ( [Brunoud et al . , 2012]; Figure 4D ) , confirming that auxin could indeed diffuse into the meristem . In the same conditions , STM expression at the boundary was however either unchanged or even slightly increased ( Figure 4C ) . This suggests that , whereas auxin levels negatively correlate with STM expression in the meristem , this correlation is also dispensable: STM expression and auxin depletion can be uncoupled at the boundary . This also shows that the relation between auxin and STM expression is more complex . Next we investigated whether the involvement of mechanical signals in STM expression at the boundary could help clarify these discrepancies . There is growing evidence that mechanical stresses act as instructive signals in parallel to biochemical signals in development . These stresses have notably been involved in the progression of gastrulation in Drosophila embryos ( e . g . Farge , 2003; Lecuit and Lenne , 2007 ) or in the formation of leaves with ruffled edges ( Audoly and Boudaoud , 2003; Nath et al . , 2003 ) . In the meristem , the boundary separates a fast growing organ from the slow growing meristem . Mechanical stresses thus emerge early on from differential growth in the boundary and have been shown to act as instructive signals controlling the behavior of cortical microtubules and PIN1 early on . In turn , this promotes tissue folding that further reinforces the stress pattern at the boundary ( Hamant et al . , 2008; Heisler et al . , 2010 ) . We thus explored a scenario in which mechanical stress may be sufficient to induce and reinforce STM expression at the boundary . To do so , we investigated whether mechanical perturbations can trigger STM expression . Note that mechanical perturbations can be of different nature . Mechanical perturbations induced by wind or touch , are extrinsic and discontinuous; they cause short-term elastic deformations ( e . g . transient bending ) . They can impact gene expression within minutes and induce major developmental responses such as stem thickening , when repeated ( Braam and Davis , 1990 ) . In the context of meristem growth , mechanical ( tensile ) stresses are in contrast intrinsic and continuous , and they cause long-term plastic responses ( growth ) . The origin of such growth-related stresses lies in the presence of high turgor pressure rather than external stimuli . The opposition between these two kind of stresses can be illustrated with the microtubule response: transient pinching does not lead to supracellular microtubule alignments in the meristem , and this has even served as a negative control for the microtubule response to ablations in this tissue ( Hamant et al . , 2008 ) . Nonetheless we cannot exclude the possibility that some elements of the mechanotransduction pathways are shared between these two types of mechanical signals . Here , given our correlation between STM expression and crease formation at the boundary , we designed our mechanical tests to check whether STM expression can be modified by continuous mechanical perturbations . First , we modified the mechanical stress pattern in the SAM using compressions or ablations and we followed the impact on the pSTM::CFP-N7 and pBOUND>>GFP expression patterns . Note that in all these experiments , we used in vitro-grown plants with a naked meristem recovering from a NPA treatment , as plants from the greenhouse need to be dissected to access the SAM , and the wounds may interfere with the mechanical perturbations . Using a microvice , we first induced global compression of the SAM . This is predicted to increase tension in the epidermis , and in the most extreme cases , to induce a bias in maximal stress direction parallel to the blades ( Figure 5A , [Hamant et al . , 2008; Uyttewaal et al . , 2012] ) . Around 8 h after compression , we detected an increase of pSTM::CFP-N7 signal ( n = 8; Figure 5B and Video 1 ) . Similar induction was observed in the pBOUND>>GFP line; although the GFP signal was more variable before compression , a strong induction could nevertheless be observed in the pBOUND>>GFP line after compression ( n = 11; Figure 5C and Video 2 ) . Furthermore , even if the induction of STM expression occurred earlier in the pSTM::CFP-N7 line than in the pBOUND>>GFP line ( Figure 5B and C ) , the overall time frame ( 8 h onwards ) rather suggests an indirect effect of compressions on STM expression . 10 . 7554/eLife . 07811 . 015Figure 5 . STM expression can be induced by mechanical perturbations . ( A to C ) Global compression of meristems with a microvice lead to an increase in STM expression ( arrows indicate the direction of the compression ) . ( A ) Predicted impact of compression on the mechanical stress pattern . ( B ) pSTM::CFP-N7 signal before and after compression in a representative meristem ( n = 8 ) . ( C ) pBOUND>>GFP signal before and after compression in a representative meristem ( n = 11 , red dots correspond to plast auto-fluorescence ) . ( D to F ) Ablation of a small number of cells leads to an increase in STM expression ( white arrows indicate the site of ablation ) . ( D ) Predicted impact of a local ablation on the mechanical stress pattern . ( E ) pSTM::CFP-N7 signal before and after ablation with a needle ( n > 30 ) . ( F ) pBOUND>>GFP signal before and after ablation using a pulsed UV laser ( n > 12 , red dots correspond to plast auto-fluorescence ) . ( G and H ) Isoxaben treatment leads to an increase of pSTM::CFP-N7 signal . ( G ) Representative pSTM::CFP-N7 signal after overnight immersion in water with DMSO ( upper panel ) or in 10 μM isoxaben ( lower panel ) . Note the increased nucleus size after isoxaben treatment , consistent with increased endoreduplication levels . ( H ) Quantifications: CFP signal intensity in 10 nuclei from the central zone of 6 isoxaben-treated meristems and 7 water-treated meristems . Values are displayed with a Student confidence interval ( α = 0 . 05 ) and compared using a bilateral Student test . ( I ) Isoxaben treatment leads to an increase in pBOUND>>GFP signal ( red dots correspond to plast auto-fluorescence ) . ( I , left ) Representative pBOUND>>GFP signal after overnight immersion in water ( n = 10 ) . ( I , right ) Representative pBOUND>>GFP signal after overnight immersion in 5 to 20 μM isoxaben ( n = 20 ) . Scale bars , 20 µm . ( J , K ) Jasmonate does not enhance STM promoter activity ( J ) pSTM::CFP-N7 signal after prolonged incubation in water supplemented with 1/1000 V/V ethanol . ( K ) pSTM::CFP-N7 signal after prolonged incubation in water supplemented with 100 µM jasmonate diluted in ethanol ( 1/1000 V/V ) . Scale bars , 20 µm . ( L-O ) Jasmonate enhances pJAZ10 promoter activity . ( L , M ) Aerial part ( L ) and root ( M ) of 3 week old NPA grown seedlings . pJAZ10::GUS staining after overnight incubation in water supplemented with 1/1000 V/V ethanol ( n = 8 ) . ( N , O ) Aerial part ( N ) and root ( O ) of 3 week old NPA grown seedlings . pJAZ10::GUS staining after overnight incubation in water supplemented with 100 µM jasmonate diluted in ethanol ( 1/1000 V/V ) ( n = 14 ) . Scale bars , 0 . 5 cm . DOI: http://dx . doi . org/10 . 7554/eLife . 07811 . 01510 . 7554/eLife . 07811 . 016Figure 5—figure supplement 1 . pSTM::CFP-N7 induction after ablations of different sizes . ( A ) Z-projections of meristems expressing pSTM::CFP-N7 and labeled with FM4-64 ( white ) before and after ablations of different sizes . ( B ) Quantification of the CFP signal intensity in 10 nuclei around the ablation site ( positive area ) or in the opposite side of the meristem ( control area ) of the three meristems presented in A . Values are displayed with a Student confidence interval ( α = 0 . 05 ) and compared using a bilateral Student test . ( C ) Longitudinal sections ( 5 µm thick maximal projections ) of the ablated meristems presented in A , 20 hr after the ablations revealing that the induction of pSTM::CFP-N7 roughly scales to the ablation size . Scale bars , 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 07811 . 01610 . 7554/eLife . 07811 . 017Figure 5—figure supplement 2 . STM mRNA distribution after ablation in the SAM . ( A ) Whole mount in situ hybridizations using a STM probe in a SAM from WT NPA grown plants . ( B ) Whole mount in situ hybridizations using a STM probe in a SAM from WT NPA grown plants 24 hr ( right and left panels ) and 48 h ( central panel ) after ablation . Red arrows point at the ablation sites . Scale bars , 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 07811 . 017 To check whether STM can also be induced after a more local mechanical perturbation , we next performed ablations of a few cells in the meristem to induce high circumferential stress around the site of ablation ( Figure 5D [Hamant et al . , 2008; Heisler et al . , 2010; Uyttewaal et al . , 2012] ) . Strikingly , we observed a local increase of pSTM::CFP-N7 signal intensity around the ablation site from t = 18 h onwards ( n = 16; Figures 5E and Video 3 ) . The induction also seemed stronger in larger ablations than in smaller ablations , consistent with a greater range of stress perturbation in larger ablations ( Figure 5—figure supplement 1 ) . A similar response was also observed in the pBOUND>>GFP background , again with a delay ( n = 12; Figure 5F and Video 4 ) . While some degree of variability between individual plants at t = 0h could be observed , pBOUND>>GFP expression was also consistently induced after ablation ( Figure 5F ) . We also noticed that with both markers , the induction of STM was not homogeneous around the ablation site . In particular , STM was never induced at the base of meristem , that is in the differentiating cells from the upper part of the stem . This is consistent with the prevailing view from the literature stating that the competence to express STM is limited to meristematic cells sensu stricto , and our data suggest that such a prepattern cannot be overridden by mechanical perturbations . To confirm these results , we analyzed the distribution of STM mRNA by whole mount in situ hybridizations on shoot apices after ablation ( Rozier et al . , 2014 ) . Even though this method is only semi-quantitative , we could detect an asymmetric signal , with a higher intensity near the ablation site , consistent with the results obtained with the pSTM::CFP-N7 and pBOUND>>GFP lines after ablation ( Figure 5—figure supplement 2 , n = 9 ) . Note that ablation may also provide increased accessibility to the probe . Last , to check whether STM expression would be enhanced by wound-induced jasmonate production , we incubated shoot apices with 100 μM jasmonate over night . In these conditions , we did not observe any significant induction of pSTM::CFP-N7 signal in most of the plants ( N = 10/14 , Figure 5J ) , and slight fluctuations in CFP signal in the remaining ones ( N = 4/14 , Figure 5K ) . Based on these results , we cannot rule out completely that jasmonate interferes with STM expression . Yet , as this contrasts with the systematic and steady induction of STM after ablation and with the robust induction of a pJAZ10::GUS reporter by jasmonate ( Figure 5L-O ) , wound-induced jasmonate is not the most likely candidate as a secondary messenger between stress and STM induction . To further confirm an induction of STM expression by mechanical signals , we next modified the mechanical stress level using isoxaben treatments . Isoxaben is a well-known inhibitor of cellulose synthesis; after treatment , walls become mechanically weaker , and thus tensile stress increases ( Heisler et al . , 2010; Uyttewaal et al . , 2012 ) . In these conditions , we could detect a global increase in pSTM-CFP-N7 signal intensity , when compared to that of the control ( measurement on 10 nucleus of the central zone in the control ( n = 6 meristems ) and isoxaben-treated ( n = 7 meristems ) ; Figure 5G and H ) . This response was also confirmed in the pBOUND>>GFP background ( control: n = 10 , isoxaben: n = 20; Figure 5I ) . Note that pBOUND>>GFP was not induced in the entire meristem after isoxaben treatment , further demonstrating that mechanical perturbations cannot override the prepattern in the SAM . Because these different experimental setups have little in common except mechanical stress perturbation , we thus propose that STM expression can indeed be modulated by mechanical stress in the SAM . One may wonder whether such artificial mechanical perturbations could induce the expression of all meristematic genes , and in particular other boundary-expressed genes , as a non-specific stress response . To test this hypothesis , we first focused on PINOID ( PID ) . Like STM in the boundaries , PID has an established role in organ separation and boundary function ( Furutani et al . , 2004; Reinhardt et al . , 2003 ) . In a pPID::PID-YFP line , we observed a more abundant signal at the boundary of the meristem , as in the pSTM::STM-Venus line ( Figure 6A ) . As detailed expression data were lacking in the meristem , we first analyzed the activity of the PID promoter . We generated a pPID::CFP-N7 line and observed a pattern that somewhat echoes that of pSTM::CFP-N7 . In particular , the CFP signal was detected in the entire meristem and was enhanced in the boundary domain ( Figure 6B ) . The signal obtained in a pPID::AlcR ALcA::GFP line ( pPID>>GFP ) further confirmed this result , echoing the pBOUND>>GFP signal ( Figure 6C and D ) . However , after ablation , PID expression was not affected in both the pPID::CFP-N7 and pPID>>GFP lines , in contrast to our results in the pSTM::CFP-N7 and pBOUND>>GFP line ( Figure 6E–G , pPID::CFP-N7: n = 13; pPID>>GFP: n = 7 ) . No induction could also be detected after isoxaben treatment in the pPID>>GFP line ( Figure 6—figure supplement 1; n = 11 ) . Therefore , in our hands PID expression appeared as a negative control for our mechanical perturbations: PID expression at the boundary may not depend on a mechanical signal . Conversely , it suggests that the induction of STM by mechanical stress retains some specificity . 10 . 7554/eLife . 07811 . 022Figure 6 . PINOID promoter activity is not affected by mechanical perturbations . ( A–C ) PINOID expression pattern in representative meristems: a higher expression of PINOID is observed in boundaries . ( A ) Expression pattern of the translational fusion pPID::PID-YFP . ( B ) Expression pattern of the transcriptional reporters pPID::CFP-N7 . ( C ) Expression pattern of pPID>>GFP . ( D ) Orthogonal sections through the middle of the boundaries of the meristems presented in ( A–C ) . ( E–F ) Time lapse of a representative meristems showing the absence of response of pPID::CFP-N7 ( F , n = 13 ) after ablation when compared to the control ( E , n = 6 ) . ( G ) Time lapse of a representative meristems showing the absence of response of pBOUND>>GFP after ablations ( control n = 14 , ablation n = 7 ) . Scale bars , 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 07811 . 02210 . 7554/eLife . 07811 . 023Figure 6—figure supplement 1 . pPID is not significantly induced by isoxaben treatment . Overnight treatment with isoxaben does not lead to an induction of pPID::GFP expression , even 48 h after the first exposition to isoxaben . ( Control , n = 14; isoxaben treatment , n = 11 ) . Scale bars , 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 07811 . 02310 . 7554/eLife . 07811 . 024Figure 6—figure supplement 2 . pCUC1 is not significantly perturbed after an ablation in the SAM . pCUC1::CUC1-GFP expression after an ablation ( arrow ) . No major changes are induced , and the pattern follows the organogenetic pattern instead of consolidating around the ablation site ( n = 11 ) . Scale bar , 20 µm . Drawings illustrate the last time points , with the hatched zone corresponding to the ablated zone . DOI: http://dx . doi . org/10 . 7554/eLife . 07811 . 02410 . 7554/eLife . 07811 . 025Figure 6—figure supplement 3 . pCUC3 is induced after an ablation in the SAM . pCUC3::CFP expression after an ablation . Note the steady induction , when compared to control ( bottom line , n = 20 ) . Scale bar , 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 07811 . 025 To further check whether other boundary expressed transcription factors are induced by mechanical perturbations , we performed ablations in pCUC1::CUC1-GFP and pCUC3::CFP expressing meristems . As ablations are also inducing other signals than mechanical signals , such a test is not sufficient to know whether a gene is induced by mechanical stress . Yet , induction after wounding is a necessary condition for a gene to respond to mechanical perturbation and thus this simple test can help to identify genes whose expression is insensitive to mechanical stress . CUC1 and CUC3 belong to the NAC family of transcription factors and are two well-known regulators of boundary function at the shoot apical meristem ( Aida and Tasaka , 2006 ) . Note that in contrast to CUC3 for which a transcriptional fusion is sufficient to recapitulate boundary expression , CUC1 expression is restricted to the boundary via the post-transcriptional action of miR164 , and thus strict boundary expression of CUC1 can only be observed in a translational fusion line ( Sieber et al . , 2007; Laufs et al . , 2004 ) . After ablations , no significant induction could be observed in the pCUC1::CUC1-GFP line ( Figure 6—figure supplement 2; n = 11 ) . For instance , when comparing the time points before and 24 h after ablations , GFP signal intensities were roughly identical and the GFP pattern could not be easily related to the wound position or shape . Later on , as the meristem recovered from the wound , new boundaries were initiated in which new GFP signal could be observed , again with no relation to the ablation position or shape ( Figure 6—figure supplement 2 ) . Sometimes , the new CUC1-GFP signal even formed a line perpendicular to the wound edge , corresponding to the initiation of the boundary of following organs ( Figure 6—figure supplement 2 , bottom panels ) , in contrast to pSTM::CFP-N7 and pBOUND>>GFP induction that always appear and further consolidates along the wound edge . In other words , the cycle of CUC1 induction at the boundary went on largely undisturbed , independent of the presence of a neighboring ablation . Altogether , this strongly suggests that CUC1 expression at the boundary is also not controlled by a mechanical signal . The response of the pCUC3::CFP line to ablation was completely different than that of the pCUC1::CUC1-GFP line . In the pCUC3::CFP line , we observed a strong and steady induction of CFP signal after ablation in all the analyzed meristems ( n = 20 ) . Interestingly , this induction seemed to appear earlier than in the pSTM::CFP-N7 line , as a strong induction could be detected as early as t = 6 hr after ablation ( Figure 6—figure supplement 3 ) . Although these data may suggest that CUC3 expression is controlled by a mechanical signal , further tests would be required to reach such a conclusion . Nonetheless this demonstrates that boundary expressed transcription factors , even from the same family , are not systematically induced by wounding and , thus , that the mechanical induction of STM retains some specificity . Another issue raised by our mechanical perturbations is the relatively long time responses ( 8 to 24 hr ) for STM to be induced . While this is consistent with the relatively slow timing of organ emergence and boundary folding in Arabidopsis meristems ( Kwiatkowska , 2004; Reddy et al . , 2004; Grandjean et al . , 2004 ) , it strongly suggests that the induction of STM expression after mechanical perturbations is indirect and could involve secondary signals , such as hormones . In line with this hypothesis , mechanical signals in animal development have all been shown to interfere with established biochemical signal transduction pathways ( e . g . [Janmey et al . , 2013; Jaalouk and Lammerding , 2009; Orr et al . , 2006] ) . Auxin signaling could again play such a role at the boundary , as the polarity of the auxin efflux carrier PIN1 depends in part on membrane tension ( Heisler et al . , 2010; Nakayama et al . , 2012 ) . Based on these data , a scenario emerges where mechanical stress would control PIN1 localization at the boundary , which in turn would locally deplete this domain from auxin , leading to a local induction of STM expression . Although our previous results suggest that STM expression at the boundary does not solely rely on auxin depletion ( based on a global auxin treatment , see Figure 4C , D ) , it does not formally exclude the possibility that the contribution of mechanical stress in STM expression at the boundary is mediated by local auxin gradients . In other words , whether mechanical stress impacts STM expression independently of the response of PIN1 to mechanical stress is unknown . First , we checked whether mechanical stress can affect the distribution of auxin in the SAM . To do so , we performed a series of ablations on meristems expressing the DII-Venus sensor . Such treatments induce a local reorientation of PIN1 around the ablation ( Heisler et al . , 2010 ) , i . e . parallel to the new stress pattern , and should lead to the local depletion of auxin . After ablation and despite some variability in the basal level of DII-Venus signal on meristems recovering from NPA treatment ( Figure 7A ) , we indeed consistently observed an induction of DII-Venus signal around the site of ablation ( n = 21; Figure 7B , Video 5 ) . Interestingly , the first effect of mechanical stress on DII-Venus signal was detected around 4 to 8 h after ablation , which is consistent with the timing of the response of PIN1 after ablation ( ca . 4 hr [Heisler et al . , 2010] ) and the subsequent response of pSTM::CFP-N7 after mechanical perturbation ( 8 to 18 hr after compression or ablation respectively , this study ) . The DII-Venus signal was also induced after compression ( Figure 7C ) further matching the response observed in the pSTM::CFP-N7 line . Altogether , this suggests that auxin patterns and mechanical stress fields act synergistically to control STM expression at the boundary . 10 . 7554/eLife . 07811 . 026Figure 7 . STM response to mechanical perturbations can be uncoupled from PIN1-dependent auxin distribution . ( A ) Time lapse of representative meristems from a NPA-grown plant and expressing DII-Venus . From t = h , the plants are not exposed to NPA anymore . ( B ) DII-Venus-N7 signal increases after ablation: Time lapse of a representative meristems from a NPA-grown plant and expressing DII-Venus as in C , after ablation ( Control n = 15 , Ablation n = 21 ) . ( C ) Representative DII-Venus signal before and after compression . An increased signal is usually detected after 4 to 8 hr after compression in the overall meristem ( n = 10 ) . ( D ) Representative pin1-6 meristem expressing DII-Venus-N7 and pSTM::CFP-N7: the presence of CFP signal at the pseudo-boundary does not correlate with DII-Venus-N7 signal anymore ( n = 6 ) . ( E ) Representative pin1-6 meristem expressing pBOUND>>GFP showing the presence of GFP signal in a pseudo-boundary ( n = 7 ) . ( F ) Representative pin1-6 pBOUND>>GFP meristem after ablation: pBOUND>>GFP is induced around the site of ablation ( n = 14 ) . ( G ) Representative pin1-6 DII-Venus-N7 pSTM::CFP-N7 meristem after ablation: pSTM::CFP-N7 is induced around the site of ablation but DII-Venus-N7 is not ( n = 10 ) . Scale bars , 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 07811 . 02610 . 7554/eLife . 07811 . 027Figure 7—figure supplement 1 . pin1-6 DII-Venus-N7 pSTM::CFP-N7 meristem after an ablation in the SAM . Two representative pin1-6 DII-Venus-N7 pSTM::CFP-N7 meristem after ablation ( time-lapse ) : pSTM::CFP-N7 is induced around the site of ablation but not DII-Venus-N7 ( n = 10 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07811 . 02710 . 7554/eLife . 07811 . 028Figure 7—figure supplement 2 . pin1-6 pBOUND>>GFP meristem after an ablation in the SAM . Representative pin1-6 meristems expressing pBOUND>>GFP showing the presence of GFP signal in a pseudo-boundary ( n = 7 ) and an induction of GFP signal after ablation ( n = 14 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07811 . 02810 . 7554/eLife . 07811 . 029Figure 7—figure supplement 3 . STM promoter activity during oryzalin-induced tissue folding in the presence of NPA . ( A ) Top and side views of a pSTM::CFP-N7 meristem grown on NPA and maintained on NPA after the local application of the microtubule depolymerizing drug oryzalin . A bump is induced and a local increase in CFP signal is detected at the pseudo-boundary ( n = 17/22 ) . ( B ) Top and side views of a pBOUND>>GFP meristem grown on NPA and maintained on NPA after the local application of the microtubule depolymerizing drug oryzalin . A bump is induced and a local increase in GFP signal is detected at the pseudo-boundary ( n = 20/21 ) . Scale bars , 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 07811 . 029 Next , we investigated whether the mechanical induction of STM depends on PIN1 . First , we analyzed the DII-Venus signal in the pin1-6 background . Although polar auxin transport in the meristem is largely inhibited , this mutant can generate bulges and sometimes ill-shaped flowers suggesting that compensatory mechanisms can be induced ( Figure 7D and E; [Okada et al . , 1991] ) . Because some degree of organogenesis is present in the pin1 mutant , this allowed us to observe boundary formation in this background . To distinguish this form of organogenesis from what we observed in the WT , we use the word 'pseudo-boundary' in pin1-6 . Despite the variability in DII-Venus signal in the pin1-6 meristems , we could not detect any DII-Venus signal in pin1-6 pseudo-boundaries , suggesting that PIN1 is necessary to deplete auxin from the boundary ( Figure 7D , n = 6 ) . Interestingly , both pBOUND>>GFP and pSTM::CFP-N7 signals were present in the pin1-6 pseudo-boundaries ( Figure 7D and E ) , thus suggesting that PIN1 is not a major player in the correlation between tissue folding and STM expression . To further confirm this result , we performed mechanical perturbations in the pin1-6 DII-Venus pSTM:CFP-N7 background . No significant induction of DII-Venus signal could be detected after ablation in pin1-6 demonstrating that auxin depletion after ablation mainly relies on PIN1 ( n = 11 , Figure 7G; Figure 7—figure supplement 1 ) . However , both pSTM::CFP-N7 and pBOUND>>GFP were systematically induced around the site of ablation ( Figure 7F and G; Figure 7—figure supplement 1 and 2 ) . This demonstrates that the induction of STM after ablation can occur independently of PIN1 . While we cannot rule out the possibility that other auxin carriers substitute for PIN1 in the pin1 background , despite our efforts , the induction of STM around the ablation site could not be clearly correlated with DII-Venus signal intensity , suggesting that the induction of STM after an ablation is not primarily caused by a local reduction in auxin levels . In other words , STM expression and auxin distribution can be partially uncoupled and the correlation between tissue folding and STM expression always remains . Last , we used a recently developed protocol in which tissue folding is triggered by the local application of the microtubule depolymerizing drug oryzalin , in the presence of the auxin transport inhibitor NPA ( Sassi et al . , 2014 ) . In these conditions , a lateral outgrowth can be induced at the periphery of the meristem , mimicking the first stages of organ emergence . In both the pBOUND>>GFP and pSTM::CFP-N7 lines , oryzalin-triggered tissue folding led to a slight increase of GFP and CFP signals respectively , in the young boundaries ( Figure 7—figure supplement 3 ) . Because this induction occurred in the presence of NPA and as oryzalin indirectly modifies the mechanical status of cell walls ( and thus promotes organ emergence ) , this further supports a scenario in which mechanical stress can promote the expression of STM at the boundary , relatively independently of auxin distribution . Because tissue folding is generally associated with the presence of high mechanical stress levels , crease formation is a unique event in development where the biochemical regulation of morphogenesis may also involve a strong contribution of mechanical signals . For instance , the patterning genes Twist and Notail are induced by the deformation of the embryo during gastrulation and epiboly in Drosophila and Zebrafish embryos respectively , through the activation of the β-catenin pathway ( Brunet et al . , 2013; Farge , 2003; Desprat et al . , 2008 ) . This suggests that shape changes are not only the result of biochemical regulation , but are also the source of mechanical signals that further channel morphogenesis via the control of gene expression . Our observations are in line with these conclusions: we report that the expression of STM at the boundary of the shoot apical meristem is correlated with tissue folding and is increased by mechanical stress . This provides a scenario in which biochemical factors , such as auxin , promote differential growth and shape changes in the meristem , which in turn , generate mechanical signals that can impact the expression of some of these regulators . Incidentally , our work also echoes results from Drosophila embryo where tension lines acting as genetic and mechanical boundaries have been revealed in the wing disk and shown to compartment cell identities ( Aliee et al . , 2012; Landsberg et al . , 2009 ) . Here we show that mechanical stress at the boundary contributes to the local expression of STM that is in turn necessary for the separation between organs . Note that tensile stress at the meristem boundary becomes anisotropic before tissue folds , as maximal stress direction is first prescribed by differential growth . In this respect , our results might thus be extended to other tissues like leaves or embryo , where the induction of KNOX genes would be related to differential growth , and thus stress , before tissue folding . Beyond organ separation , the boundary has many other functions . In particular , it is the site where new meristems , called axillary meristems , can be initiated . Interestingly , overexpression of class I KNOX genes is sufficient to induce ectopic meristems ( Chuck et al . , 1996 ) . The boundary also plays a crucial role in plant architecture: this is where auxin is redistributed from the adjacent neighboring organ to the meristematic pool where new organs or new axis need to be initiated ( Heisler et al . , 2005; Reinhardt et al . , 2003; de Reuille et al . , 2006 ) . The central role of the meristem boundary in plants largely explains why this domain has received considerable attention over the past decades , albeit from a molecular genetics point of view mainly ( Aida and Tasaka , 2006; Tian et al . , 2014 ) . The role of mechanical stress in the function and regulation of the boundary domain of the meristem opens the possibility that other functions than organ separation are influenced by mechanical signals . Another question raised in this work is that of the coordination between mechanical and biochemical signals to control development . In animals , well-known elements of biochemical-based transduction pathways are involved in mechanotransduction . For instance , integrin , β-catenin or YAP/TAZ are all involved in transducing mechanical signals ( e . g . Janmey et al . , 2013; Jaalouk and Lammerding , 2009; Orr et al . , 2006 ) . Such intermingling raises the question of whether these signals can really be uncoupled with one another and this may even question the added value of mechanical signals in development . This is what we touched upon by analyzing the coordination between the plant hormone auxin and mechanical stress in the meristem . While we provide further evidence supporting a strong coordination between STM expression and auxin depletion , we also found that the response to mechanical stress can uncouple them . In particular , mechanical perturbations induce STM expression even when auxin transport or level is affected . In addition , mechanical stress does not seem to impact the expression of PINOID , which has been associated with the control of PIN1 polarity and thus auxin patterns in the meristem ( Robert and Offringa , 2008 ) . This is thus a case where two redundant signalling pathways act relatively independently to control the same morphogenetic event . We propose that the absence of a strict coupling between auxin depletion and mechanical stress is a way to add robustness in meristem functions at the boundary . Incidentally , this highlights the added value of mechanical signals in this domain . Alternative cues may be involved in the promotion of STM expression at the boundary . In addition to auxin , STM has been associated with the homeostasis of cytokinins , gibberellins and ethylene . Activation of STM promotes the expression of the cytokinin biosynthesis gene AtIPT7 and cytokinin response factor ARR5 ( Yanai et al . , 2005 ) . Conversely , overproduction or application of cytokinins increase the expression level of STM and can rescue weak stm mutant alleles ( Rupp et al . , 1999; Jasinski et al . , 2005 ) . This interaction with cytokinins is largely shared among the different class I KNOX genes and , consistently , it has been shown that KNAT1 and KNAT6 display redundant functions with STM . Interestingly , the level of cytokinins has been predicted to increase in the boundary , based on the cytokinin activity reporter pTCS signal , and axillary meristem formation has been shown to require cytokinin signalling ( Wang et al . , 2014 ) , consistent with the well-known bushy phenotypes of cytokinin overproducers . Temporally , this cytokinin burst would follow the initial reduction in auxin level at the boundary ( Wang et al . , 2014 ) . Thus cytokinin , in parallel to and after auxin depletion , may very well contribute to STM induction at the boundary . Class I KNOX proteins also reduce gibberellins levels , notably through a direct repression of the GA-20 oxidase gene ( Sakamoto et al . , 2001; Chen et al . , 2004 ) ; conversely gibberellins can rescue the phenotype of KNOX overexpressors ( Hay et al . , 2002 ) . The induction of the gibberellin catabolism gene GA-2 oxidase was also shown to depend on KNOX proteins ( Jasinski et al . , 2005 ) . While both cytokinins and gibberellins are tightly linked to class I KNOX gene expression , the potential interplay with mechanical stress has not been explored . In contrast , other hormones , such as ethylene , have been associated to the plant response to many stresses . Interestingly , ethylene restricts the expression of class I KNOX gene KNAT2 in the shoot meristem to the boundary domain ( Hamant et al . , 2002 ) . Note however that the exact contribution of ethylene in mechanoperception remains debated . For instance , while touch can induce ethylene synthesis , ethylene response mutants still display thigmomorphogenesis in response to touch ( Chehab et al . , 2012 ) ; touch-induced thigmomorphogenesis rather depends on jasmonate ( Chehab et al . , 2012 ) . While certain hormones are rather associated with repetitive elastic deformations ( like touch or wind ) , others are involved in continuous plastic deformations ( i . e . growth and shape changes ) , these may be better candidates for the regulation of STM expression at the boundary . More generally , and as auxin does not seem to act as a secondary messenger of mechanical signals for STM expression , another exciting prospect for this work will be to identify the elements of the mechanotransduction pathway acting on STM at the boundary . The current dissection of the gene regulatory network acting at the boundary ( e . g . [Tian et al . , 2014] ) should help us weigh the putative contributions of these biochemical factors in transducing mechanical signals at the SAM in the future . As illustrated in the art of origami , folding can in principle generate the widest diversity of shapes from simple inputs . In the past decades , there has been tremendous progress in the identification of genes that trigger such events in all living organisms . The robustness of such a complex regulation is often thought to rely on redundancy between different molecular pathways . Here we have investigated whether the shape itself , and its associated mechanics , interferes with gene expression and channels morphogenesis by constraining the possible outputs of the gene regulatory network . If generalized , such a dialog between gene and form could help us understand how reproducible shapes can emerge from a complex gene regulatory network that is susceptible to noise . Because of its essential function in morphogenesis and its impact on developmental robustness , this 'shape to gene' feedback may also have far reaching implications in evolution . The pBOUND>>GFP , DII-Venus , pin1-6 , stm-dgh6 lines were already described in the literature ( Brunoud et al . , 2012; Vernoux et al . , 2011; Laufs et al . , 2004; Dharmasiri et al . , 2005; Vernoux et al . , 2000; Aida et al . , 2002 ) . The pCUC1::CUC1-GFP and pCUC3::CFP lines have recently been described ( Goncalves et al . , 2015 ) . Note that the Mo223::GFP line ( Cary et al . , 2002 ) was also used to test the response of the CUC1 promoter to ablations . The pJAZ10::GUS used as control for the JA treatment has also been previously described ( Mousavi et al . , 2013 ) . The pSTM::CFP-N7 , pPDF1::CFP-N7 and pPID::CFP-N7 lines were generated by fusing either the 5 , 7 kb region upstream of STM ( AT1G62360 ) ATG , the 1456 pb region upstream of PDF1 ( AT2G42840 ) ATG , or the 4 kb region upstream of PID ( AT2G34650 ) ATG to a CFP targeted to the nucleus through a N7 signal in a pH7m34GW vector using the Gateway system and transformed in Col0 ecotype . The pPID>>GFP line was generated by fusing the 4 kb region upstream of PID ( AT2G34650 ) ATG to the Alc-R coding sequence fused to an pAlcA::GFP-ER in a derivate of the pGREEN129 plasmid and transformed into the WS-4 ecotype . The pSTM::STM-Venus line was generated by transforming the plasmid described by Heisler et al . ( Heisler et al . , 2005 ) in Col-0 plants . The STMamiRNA line ( binding to the STM coding sequence though the following sequence: TTAACCACTGTACTTGCGCGA ) was designed and amplified from the pRS300 plasmid following previously described protocol ( http://wmd3 . weigelworld . org/cgi-bin/webapp . cgi ) and fused to the pSTM promoter in a pK7m34GW vector using the Gateway system . The construction was transformed either in Col-0or in the pSTM::STM-Venus line . pBOUND>>GFP and pPID>>GFP expression was induced by an overnight treatment with 70% ethanol vapor following a previously described protocol ( Deveaux et al . , 2003 ) . 'Greenhouse grown plants' were initially grown in short-day conditions ( 8 hr/16 hr light/dark period ) for one month and then transferred to long-day conditions ( 16 h/8 h light/dark period ) . Stems were cut and the SAM was dissected when the inflorescence meristem was visible , i . e . between the appearance of the first flower to the appearance of first silique ( stages 13 to 17 [Smyth , 1990] ) and transferred on a half MS medium with vitamins and 0 . 125 µg/µL of BAP for imaging as already described ( Fernandez et al . , 2010 ) . "In vitro grown plants” were grown in a phytotron in long day conditions on Arabidopsis medium ( Duchefa , Haarlem , the Netherlands ) supplemented with 10 µM NPA to inhibit flower initiation and generate naked meristems . NPA-treated in vitro grown plants were transferred to a medium without NPA as soon as naked meristems were formed as already described ( Grandjean et al . , 2004 ) . Meristems were then imaged from 24 hr to 48 hr after transfer on the NPA-free medium . For in vitro experiments on pin1 meristems , plants were grown in vitro on MS medium without NPA in a phytotron in long day conditions . Dissected meristems and plants grown in vitro were imaged in water using either a LSM700or a LSM780 confocal microscope ( Zeiss , Germany ) to generate stack of optical sections with an interval of 0 . 25 , 1or 2 µm between slices . In some cases , membranes were stained with FM4-64 and the signal in the L1 was extracted using the Merryproj software as described in de Reuille et al . , ( de Reuille et al . , 2005 ) . Longitudinal optical sections of various thicknesses were performed using the ImageJ software . In many cases , the Fire lookup table from ImageJ was used to represent the CFP signal intensity and background signal from raw images was reduced in images , by modulating the minimum level of intensity in ImageJ as shown by the calibration bars on the different pictures . The minimum level was however never above 50 out of 256 in intensity and the same processing was applied on all time points , and on control and assay alike . In some projections from of pSTM::STM-Venus and pSTM::CFP-N7 where the signal was weak , area outside the nucleus emitting in both green and red channels and corresponding to either the auto-fluorescence of the plasts or the FM4-64 signal were automatically removed using the function calculator-plus subtraction of ImageJ . The maps and quantifications of meristem curvature and pSTM::CFP-N7 signal at cellular levels were obtained using the MorphographX software ( www . morphographix . org ) . The curvature maps were generated by plotting the mean Gaussian curvature on either non-segmented or cell-segmented meshes with a neighboring of 15 µm . The maps of pSTM::CFP-N7 signal were generated by blurring the CFP signal with a radius of 10 µm and projecting it on the cellular mesh with a thickness of 10 µm . Note that the oldest organs were not included in the quantifications because of the inability of the mesh to follow the surface of the meristem in very deep boundaries and to extract well the pSTM::CFP-N7 signal in these area . Furthermore , cells at the extremities of the mesh were also not included in the quantifications due to the presence of aberrations of Gaussian curvature on these parts of the mesh . Organ size was quantified using ImageJ: stacks were projected using the function z-stack , a circle was drawn using the freehand selection around each organ viewed from the top and the surface inside this selection was automatically measured . The folding angle of the boundary was measured manually on longitudinal sections taken in the middle of each boundary using the angle tool of ImageJ . For the quantification of the pSTM::CFP-N7 nuclear signal after Isoxaben treatments or after ablations , each nucleus from the central zone ( Isoxaben treatment ) or from the peripheral zone ( around the site or on the opposite site of the ablation ) was isolated manually using the ImageJ software on the slice from the z-stack where its size and intensity was the highest . A circle was manually fitted around each nucleus and the integrated intensity was calculated in this circle . 10 nuclei were measured for each meristem . The quantification of nucleus signal intensity after Isoxaben treatment was performed on one of two sets of individual plants ( see Supplementary file 1 ) . Each experiment was performed on at least two independent sets of plants , and with at least 4 independent plants in each set . In all experiments , the t = X hr time point corresponds to X hours after the beginning of treatment . Controls and assays were analyzed in parallel ( same growth conditions , same imaging conditions ) . The compressions , ablations and isoxaben treatments that were carried out on WT plants were performed on plants previously grown in vitro NPA and transferred in a medium without NPA 0 to 24 hr before the beginning of the experiment . The ablations were performed with a needle or using a pulsed UV laser as already described ( Hamant et al . , 2008; Uyttewaal et al . , 2012 ) . Note that the ablations performed on pin1-6 meristems were performed on plants grown in vitro on NPA-free medium . The isoxaben treatments were conducted by immersing the plants in solutions of 5 to 20 µM of isoxaben overnight ( for 12 to 14 hr , Heisler et al . , 2010; Uyttewaal et al . , 2012 ) . Controls were obtained by water immersion with an equivalent volume of Dimethyl Sulfoxide ( DMSO ) . All these concentrations gave similar results . The presence of isoxaben in the meristem could be confirmed by its impact on meristem , cell and nucleus size . For the auxin treatments , dissected meristems in half MS Vitamins and BAP boxes were immerged in a solution of 10 µM of 2-4D overnight ( for 12 to 14 hr ) . Controls were obtained by immersing the meristem in a solution of equivalent volume of water . Similar experiments were also performed on plants grown in vitroand recovering from an NPA treatment and with IAA and gave similar result ( data not shown ) . The disappearance of the DII-Venus-N7 signal reflected the presence of high auxin levels in treated meristems . For jasmonate treatments on pSTM::CFP-N7 and pJAZ10::GUS , jasmonic acid ( Sigma ) was diluted in 100% Ethanol to make a stock solution . JA was then diluted in water ( 1/1000 ) for the treatments . NPA grown plants in Arabidopsis medium exhibiting naked meristems and recovering from the NPA treatment were fully immersed overnight ( 12 to 14 hr ) in a water solution containing 100 µM of JA or in a similar dilution of ethanol for the control . Oryzalin applications in lanolin paste were carried out as previously described ( Sassi et al . , 2014 ) . To reveal the presence of GUS activity in the pJAZ::GUS plants grown on NPA and treated with jasmonate , the samples were first immersed overnight at 37°C in a GUS solution ( 100 mM Na2HPO4 pH7 , 5 mM K4Fe [CN]6·3H20 , 5 mM K3Fe [CN]6 , 0 . 05% Triton-X-100 , 0 . 5 mg/ml X-Gluc ) before being cleared trough successive ethanol baths ( from 70 to 100% Ethanol ) . Imaging was performed with a Zeiss stereomicroscope Discovery V20 . The whole mount immunolocalizations of PIN1 were performed on WT plants ( WS-4 ecotype ) as previously described ( Besnard et al . , 2013 ) . Imaging was performed with a Zeiss LSM700 microscope . The whole mount in situ hybidization of STM mRNA was performed as previously described ( Rozier et al . , 2014 ) . Imaging was performed with a Zeiss stereomicroscope Discovery V20 . The stm-dgh6 mutant was originally identified in the Versailles T-DNA collection . In this allele the T-DNA is inserted between nucleotide 723 and 724 in the cDNA sequence . To genotype the stm-dgh6 populations , we used the forward primer GGGTAAATACCCCTTTGATGG ( before the T-DNA insertion ) and reverse primer TCGTTCCTTATCCTGAGTTG ( in the following gene ) . This amplified a 2600 bp long fragment in the WT and stm-dgh6 heterozygote , and nothing in the stm-dgh6 homozygote . Conversely , to detect the presence of the T-DNA , we used the forward primer CGTGTGCCAGGTGCCCACGGAATAGT ( LB4 , T-DNA ) and reverse primer GTAGTGACGGCTCCACCAAT ( STM ) . This amplified a 300 bp long fragment in the stm-dgh6 heterozygote and homozygote , and nothing in the WT . The qPCRs for STM and STMamiRNA were performed on 2 week-old seedlings grown on ½ MS medium with kanamycin . The qPCRs on the pSTM::STMamiRNA lines were performed on either homozygous or heterozygous F3 populations . The qPCR on WT plants was performed on plants harboring a kanamycin resistance at homozygous state and growing in the same conditions as the pSTM::STMamiRNA lines . As shown in Figure 3—figure supplement 2 , the expression of the amiRNA ( Primers: STMamiRNA Q5: CACTGGTTATTCACAGGTCGTG , STMamiRNA Q3: CAGTGGTTAATCAAAGAGAATCAATG ) and its effect on STM expression was checked by qPCR ( Primers: qSTM-F: TCGACTTCTTCCTCGGATGACCCA , qSTM-R: TCTCCGGTTATGGAGAGACAGCAA ) . Three independent biological replicates were at least used for each sample . Expression is shown relative to that of TCTP , a stable housekeeping gene .

The bending , stretching or squashing of cells or tissues can be used as a signal to trigger a range of biological responses . However investigating the role of these mechanical signals remains a challenge . This is partly because the forces that trigger the mechanical signals are often short-lived and changeable , and partly because the signals can be difficult to separate from the biochemical responses that they generate . Stem cells present at the tip of the growing shoots in a plant are exposed to mechanical forces . These growing tips are called shoot meristems , and the stem cells they contain create all the aboveground organs of the plant ( stems , leaves and flowers etc . ) . In each meristem , a boundary forms between the slow-growing stem cells at its centre and the fast-growing organs that form around them . Because these plant cells are both stuck together by their cell walls and growing at different rates , strong mechanical stresses are created causing this boundary to fold . A key regulator of the meristem is a protein called STM , but it remains unclear whether mechanical signals are involved in the control of this protein . To investigate this , Landrein et al . tracked where the gene for the STM protein was switch on in shoot meristems in a plant called Arabidopsis , and found that it is highly active at the boundary . Analysing STM in different mutant plants combined with advanced imaging techniques revealed that STM activity correlates with the extent of creasing at this boundary . The STM protein is also required at the boundary to ensure that developing organs separate out . These findings suggest that boundary folding might somehow create signals that activate STM . One candidate signal was the plant hormone called auxin because reduced levels of this hormone were previously associated with boundary formation . However , in further experiments , Landrein et al . ruled out auxin’s involvement in this process . So do mechanical signals activate STM at the boundary ? To test this , the mechanical forces in the meristem were altered by compressing the growing shoot meristems in miniature vices and by killing a few stem cells at the meristem centre . Both of these actions triggered the production of STM in the meristem , consistent with its activity being altered by mechanical stress . Landrein et al . propose that the mechanical regulation identified acts in parallel to auxin signalling , providing robustness to the regulation of gene activity in the shoot meristem . In other words , tissue folding can guide gene expression , via the production of mechanical signals . But how shoot meristem cells respond at a molecular level to mechanical stress awaits future work . Finally , proteins related to STM can be found in all biological kingdoms , including some proteins that regulate important process in animal development . Whether the activity of these related proteins is also regulated by mechanical forces remains to be investigated .

2015

Mechanical stress contributes to the expression of the STM homeobox gene in Arabidopsis shoot meristems

Recent studies demonstrated that mutations in B3GNT1 , an enzyme proposed to be involved in poly-N-acetyllactosamine synthesis , were causal for congenital muscular dystrophy with hypoglycosylation of α-dystroglycan ( secondary dystroglycanopathies ) . Since defects in the O-mannosylation protein glycosylation pathway are primarily responsible for dystroglycanopathies and with no established O-mannose initiated structures containing a β3 linked GlcNAc known , we biochemically interrogated this human enzyme . Here we report this enzyme is not a β-1 , 3-N-acetylglucosaminyltransferase with catalytic activity towards β-galactose but rather a β-1 , 4-glucuronyltransferase , designated B4GAT1 , towards both α- and β-anomers of xylose . The dual-activity LARGE enzyme is capable of extending products of B4GAT1 and we provide experimental evidence that B4GAT1 is the priming enzyme for LARGE . Our results further define the functional O-mannosylated glycan structure and indicate that B4GAT1 is involved in the initiation of the LARGE-dependent repeating disaccharide that is necessary for extracellular matrix protein binding to O-mannosylated α-dystroglycan that is lacking in secondary dystroglycanopathies . Glycosylation is the most abundant and diverse post-translational modification of proteins ( Varki , 2011 ) . The synthesis of complex glycans is catalyzed by the action of over 200 individual glycosyltransferases in humans ( Moremen et al . , 2012 ) . For most of the enzymes studied to date , there is exceptional selectivity for the donor sugar nucleotide and the underlying acceptor glycan as well as stereo- and linkage-specificity for the catalyzed additions ( Moremen et al . , 2012 ) . Mutations in the genes encoding many of these glycosyltransferases have been established as causal for a variety of human diseases including congenital muscular dystrophy , specifically the secondary dystroglycanopathies ( Barresi and Campbell , 2006; Freeze , 2007; Wells , 2013; Praissman and Wells , 2014 ) . At least a dozen enzymes/proteins are involved in the synthesis of O-mannose-initiated glycans that when defective lead to various forms of congenital muscular dystrophy , termed secondary dystroglycanopathies , that range from the phenotypically mild Limb-Girdle to the severe Walker-Warburg muscular dystrophies ( Muntoni et al . , 2011; Mercuri and Muntoni , 2012; Dobson et al . , 2013; Praissman and Wells , 2014 ) . Until recently ( Lommel et al . , 2013; Vester-Christensen et al . , 2013; Winterhalter et al . , 2013 ) , α-dystroglycan , a central component of the dystrophin-glycoprotein complex that serves to connect the cytoskeleton inside the cell with the extracellular matrix outside the cell , was the only well established O-mannoslyated mammalian protein ( Barresi and Campbell , 2006; Endo and Manya , 2006 ) . The proper O-mannosylation of α-dystroglycan is essential for its ability to bind to components of the extracellular matrix including laminin ( Ervasti and Campbell , 1993; Barresi and Campbell , 2006 ) . In recent years , significant progress has been made in defining the multitude of O-mannose structures produced in multicellular animals including partial elucidation of the LARGE-dependent functional glycan structure ( Stalnaker et al . , 2010; Yoshida-Moriguchi et al . , 2010; Hara et al . , 2011; Stalnaker et al . , 2011a; Stalnaker et al . , 2011b; Harrison et al . , 2012; Inamori et al . , 2012 , 2013; Live et al . , 2013; Panin et al . , 2014 ) . Recently , reports in the literature have connected mutations in B3GNT1 with congenital muscular dystrophies ( Buysse et al . , 2013; Czeschik et al . , 2013; Shaheen et al . , 2013 ) and specifically to the defective glycosylation of α-dystroglycan ( Jae et al . , 2013 ) even though no O-mannose glycan structures containing a β3-GlcNAc have yet to be elucidated ( Praissman and Wells , 2014 ) . In 1997 , B3GNT1 ( also known as iGnT ) was reported to be the first successfully cloned β-1 , 3-N-acetylglucosaminyltransferase involved in the synthesis of poly-N-acetyllactosamine ( Sasaki et al . , 1997 ) . Poly-N-acetyllactosamine chains consist of the repeating disaccharide -β3-GlcNAc-β4-Gal- , a structure termed the i-antigen when unbranched ( Ujita et al . , 2000; Zhou , 2003 ) . Poly-N-acetyllactosamine is a prevalent glycan substructure found in N-glycans and O-glycans on proteins whose abnormal levels have been associated with human diseases including cancer ( Ujita et al . , 2000; Zhou , 2003; Ho et al . , 2013; Lu et al . , 2014 ) . The Fukuda group used an expression cloning strategy enriching for plasmids containing cDNA inserts that substantially increased poly-N-acetyllactosamine , as judged by antibody binding , on the surfaces of Namalwa KJM-1 cells ( Sasaki et al . , 1997 ) . Shortly thereafter , work carried out by Hennet's and Sasaki's groups led to the cloning of three other β-1 , 3-N-acetylglucosaminyltransferases , B3GNT2 , 3 , and 4 ( Zhou et al . , 1999; Shiraishi et al . , 2001 ) . However , papers as recently as 2008 continued to use conflicting nomenclature for B3GNT2 by calling it B3GNT1 causing some confusion within the field ( Biellmann et al . , 2008 ) . Importantly , B3GNT2 , B3GNT3 and B3GNT4 were found to share motifs with β3-galactosyl-transferases as well as β3-GalNAc-transferases ( Zhou et al . , 1999; Shiraishi et al . , 2001; Togayachi et al . , 2006 ) . These three enzymes are in the same Carbohydrate-Active Enzymes database ( CAZy ) family and group together based on primary sequence similarity and motifs hypothesized to play a role in forming the β3 linkage ( Zhou et al . , 1999; Shiraishi et al . , 2001; Togayachi et al . , 2006 ) . B3GNT1 lacks these motifs and is in a different CaZY family , interestingly one shared with LARGE and LARGE2 that contain both xylosyl- and glucoronsyl–transferase activity and which functionally modify O-mannosylated glycans ( Inamori et al . , 2012; Goddeeris et al . , 2013; Inamori et al . , 2013 ) . Furthermore , it has been reported that the B3GNT1 enzyme is a binding partner for LARGE in mammalian cells ( Bao et al . , 2009 ) . The combination of data suggesting substantial divergence of B3GNT1 from other B3GNTs and the similarity in primary sequence and disease association with LARGE led us to reexamine the enzymatic activity of B3GNT1 to resolve the apparent inconsistencies and establish its potential enzymatic role in O-mannosylation and secondary dystroglycanopathies . Here we report that B3GNT1 is in fact B4GAT1 ( β-1 , 4-glucuronyltransferase 1 ) that generates the substrate for the LARGE-dependent repeating disaccharide that is required for interaction of O-mannosylated α-dystroglycan with extracellular matrix proteins . A secreted ( lacking the trans-membrane domain ) epitope-tagged form of human B3GNT1 was recombinantly expressed in HEK293F cells , purified , and the epitope tag removed before enzymatic characterization . Purity and enzyme identity was assessed by Coomassie G-250 staining of SDS-PAGE separated protein as well as by shotgun proteomics using reverse-phase liquid chromatography-nanospray tandem mass spectrometry ( LC-MS/MS ) following tryptic digestion ( Figure 1—figure supplement 1 , Supplementary file 1 ) . Incubation of the purified protein with UDP-[3H]GlcNAc and an enzymatically synthesized and purified substrate , Gal-β4-GlcNAc-β-pNP ( Figure 1—figure supplement 2 ) , under previously reported buffering conditions for B3GNTs ( Zhou et al . , 1999; Shiraishi et al . , 2001 ) failed to result in a detectable product as measured by incorporation of radioactivity into substrate ( Figure 1 ) . In order to validate our acceptor sugar and buffering conditions , we expressed , purified , and characterized a secreted epitope-tagged form of human B3GNT2 ( Figure 1—figure supplement 1 , Supplementary file 1 ) . Side-by-side extended ( overnight ) incubations of B3GNT1 and the B3GNT2 enzyme preparations with acceptor and sugar nucleotide showed clear N-acetylglucosaminyltransferase activity toward the acceptor only with B3GNT2 , as assayed by radioactive incorporation of GlcNAc into the N-acetyllactosamine acceptor ( Figure 1 , glycans displayed in symbolic representations [Varki et al . , 2009] ) . 10 . 7554/eLife . 03943 . 003Figure 1 . B3GNT1 does not possess β-1 , 3-N-acetylglucosaminyltransferase activity . ( A ) The β-1 , 3-N-acetylglucosaminyltransferase activity of B3GNT1 towards pNP tagged N-acetyllactosamine was compared to that of B3GNT2 ( reaction scheme presented in ( B ) ) . Incubations were carried out overnight at 37°C in 0 . 1 M MES pH 6 . 5 containing 10 mM MnCl2 , 40 , 000 DPM UDP-[3H]GlcNAc and 2 mM non-radioactive UDP-GlcNAc . pNP-sugars were isolated from sugar nucleotide by reverse-phase C18 spin columns . UDP-[3H]GlcNAc transfer to Gal-β4-GlcNAc-β-pNP was measured by liquid scintillation counting . Averaged results from three independent experiments with error bars indicating standard deviation are shown . GlcNAc-β-pNP was used as a negative control acceptor . ( C ) Relevant sugar code symbols from ‘Essentials of Glycobiology’ as well as other symbols used throughout the paper are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 03943 . 00310 . 7554/eLife . 03943 . 004Figure 1—figure supplement 1 . Coomassie brilliant blue G-250 stained SDS-PAGE gels of purified enzyme samples . The prominent bands appear at the appropriate molecular weights for our constructs . B3GNT1 was cleaved from tag leaving only the catalytic domain while the other enzymes were not cleaved from expression construct elements ( affinity tag and GFP ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03943 . 00410 . 7554/eLife . 03943 . 005Figure 1—figure supplement 2 . Gal-β4-GlcNAc-β-pNP synthesized using bovine B4GALT1 LC-MS and LC-MS/MS spectra of Gal-β4-GlcNAc-β-pNP produced using B4GALT1 . Low m/z fragments include diagnostic GlcNAc oxonium ions at 186 and 168 . DOI: http://dx . doi . org/10 . 7554/eLife . 03943 . 005 Given that B3GNT1 clustered into CAZy family GT49 , the family containing the glucuronyltransferase domain of LARGE , we performed a multiple sequence alignment ( Figure 2A ) . B3GNT1 aligned strongly with LARGE and LARGE2 in the glucuronyltransferase domain DXD-motif region typically involved in metal ion dependent sugar nucleotide binding . This DXD-motif was shown to be necessary for the glucuronyltransferase activity of LARGE and LARGE2 consistent with probable sugar nucleotide binding in this region ( Inamori et al . , 2013 ) . B3GNT2 did not align well with B3GNT1 or the LARGE proteins ( Figure 2A ) . We hypothesized that B3GNT1 might have glucuronyltransferase activity contrary to its proposed enzymatic function and thus carried out a screen using various tagged monosaccharide acceptors . Incubation of B3GNT1 and UDP-GlcA with multiple glycan acceptors resulted in significant transfer only to α- and β-xylopyranosides , as measured by radioactive transfer from UDP-[14C]GlcA ( Figure 2B ) . B3GNT2 did not transfer GlcA to either anomeric form of xylopyranoside tested ( data not shown ) . 10 . 7554/eLife . 03943 . 006Figure 2 . B3GNT1 is a glucuronyltransferase that uses xylose as an acceptor . B3GNT1 is in CAZy family GT49 along with LARGE and LARGE2 . ( A ) Clustal Omega multiple sequence alignment excerpt showing the strong alignment of B3GNT1 with the glucuronyltransferase domain of LARGE and LARGE2 , including the DXD motif shown to be important for activity . B3GNT2 does not align well with the other inputs . ( B ) UDP-[14C]GlcA transfer screen assayed by liquid scintillation counting ( disintegrations per minute–DPM ) , averaged results of three independent experiments with error bars indicating standard deviation . B3GNT1 transfers to xylose in both anomeric configurations . ( C ) LARGE transfers only to α-linked xylose as previously reported . ( D ) LC-MS and LC-MS/MS data showing B3GNT1 transfer to Xyl-α-pNP . The ammonium adduct is the dominant species ( 465 . 136 ) however the protonated species is observable at 448 . 109 . The majority of unlabeled fragment peaks represent losses of water from labeled peaks . ( E ) LC-MS and LC-MS/MS data showing B3GNT1 transfer of UDP-GlcA to Xyl-β-MU . ( F ) LC-MS and LC-MS/MS data showing LARGE transfer to Xyl-α-pNP . Reactions were carried out overnight in 0 . 1 M MES pH 6 . 5 containing 10 mM MnCl2 , 5 mM MgCl2 , 2 mM substrate , 40 , 000 DPM UDP-[14C]GlcA and 2 mM non-radioactive donor . DOI: http://dx . doi . org/10 . 7554/eLife . 03943 . 00610 . 7554/eLife . 03943 . 007Figure 2—figure supplement 1 . Silver stained SDS-PAGE gel of complex formation assay samples . B4GAT1 ( lacking an 8xHis tag ) and LARGE1 ( fused to an 8xHis tag ) were mixed together and incubated for 15 min at 4°C . After incubation , the mixture was bound to Ni-NTA resin , washed three times with 20 mM imidazole containing PBS pH 7 and eluted in 250 mM imidazole in PBS pH 7 . Aliquots were run by SDS-PAGE and silver stained . Evidence for in vitro complex formation was not observed as is seen by the lack of a B4GAT1 band in the elution fraction . DOI: http://dx . doi . org/10 . 7554/eLife . 03943 . 007 We next expressed , purified and characterized a secreted form of human LARGE ( Figure 1—figure supplement 1 , Supplementary file 1 ) whose activity we investigated . Our results confirm previous reports that LARGE transfers GlcA to α-xylopyranosides but not β-xylopyranosides ( Figure 2C , [Inamori et al . , 2012 , 2013] ) . In contrast , B3GNT1 does not appear to be hampered by the stereochemistry of the acceptor sugar ( Figure 2B , C ) . We also confirmed the disaccharide products by LC-MS/MS ( Figure 2D–F ) . In order to investigate whether a complex formed between our soluble enzymes , we incubated the purified forms of the two enzymes together with only the LARGE enzyme being epitope tagged . Following re-purification of LARGE based on the epitope tag , we were unable to detect the presence of B3GNT1 suggesting that the soluble forms of the proteins do not interact in vitro ( Figure 2—figure supplement 1 ) . We performed kinetic analysis of B3GNT1 and LARGE using the α-xylopyranoside as an acceptor and determined only minor differences ( 3 . 0 and 6 . 0 mM , respectively ) in Km values for the acceptor ( Table 1 ) . However , at 2 mM UDP-GlcA , B3GNT1 has a specific activity that is 48 times higher than LARGE for α-xylopyranoside as the acceptor ( Table 1 ) . For B3GNT1 , we were also able to use β-xylopyranoside as an acceptor and determined that the enzyme was more than twice as efficient with this acceptor as opposed to its anomer , α-xylopyranoside , though the Km value for the acceptor ( 4 . 0 mM ) was only slightly altered ( Table 1 ) . Thus , B3GNT1 possesses a significantly higher specific activity , turnover rate and catalytic efficiency than LARGE toward the monosaccharide α-xylopyranoside acceptor and is even more efficient with the β-anomer that LARGE is not able to utilize ( Table 1 ) . 10 . 7554/eLife . 03943 . 008Table 1 . Kinetics at 2 mM UDP-GlcADOI: http://dx . doi . org/10 . 7554/eLife . 03943 . 008EnzymeSubstrateKm ( mM ) kcat ( s−1 ) kcat/Km ( M-1 s−1 ) Specific activity ( pmol/min/μg ) B3GNT1Xyl-α-pNP3 . 00 . 08729130Xyl-β-pNP4 . 00 . 2563380LARGEXyl-α-pNP6 . 00 . 0050 . 842 . 7Xyl-β-pNPNMNMNMNMNM = not measurable ( below limit of detection ) . To further characterize the product of B3GNT1 compared to LARGE , we carried out large-scale transfer reactions with the proven substrates of each enzyme . Disaccharide products were purified using reverse phase C18 HPLC and analyzed by multiple NMR-based experiments ( Figure 3 , Figure 3—figure supplement 1 , Supplementary file 2 ) . Glycosidic linkages and anomeric configurations were determined by NMR ( Bock and Pedersen , 1974; Van de Ven , 1995; Wishart et al . , 1995 ) that clearly demonstrates that the B3GNT1 and LARGE products share the same stereochemistry but different linkages of the terminal glucuronic acid to the underlying xylose ( Figure 3 , Figure 3—figure supplement 1 , Supplementary file 2 ) . From this analysis , we determined that B3GNT1 is a xylopyranoside β1 , 4-glucuronyltransferase that we designate , using standard convention , B4GAT1 and , as previously described , we confirmed that LARGE contains xylopyranoside β1 , 3- glucuronyltransferase activity ( Inamori et al . , 2012 ) . 10 . 7554/eLife . 03943 . 009Figure 3 . B3GNT1 is a β1 , 4-glucuronyltransferase in contrast to LARGE which is a β1 , 3-glucuronyltransferase . 1H–13C HSQC spectrum with peak assignments and structures of the products resulting from the action of B4GAT1 and UDP-GlcA on Xyl-α-pNP ( red ) and LARGE and UDP-GlcA on the same substrate ( blue ) . Cross peaks are at the intersection of the 13C and 1H shifts of the partners in each C-H pair . The changes in the spectrum between the two compounds reflect the differences in the site of the glycosidic linkage formed , and particularly the major differences in the carbon shifts for the Xyl 3 and Xyl 4 sites between the two disaccharides is diagnostic of the change in their participation in the two different respective glycosidic linkages . The β configuration of the GlcA 1 site is confirmed by the 1-bond 1H–13C couplings of 166 Hz for these sites , as well as couplings between the H1 and H2 protons , 7 . 9–8 . 0 Hz . ( data not shown ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03943 . 00910 . 7554/eLife . 03943 . 010Figure 3—figure supplement 1 . NMR determination of the 1 , 4 glycosidic linkage . Section of the 1H–13C HMBC spectrum of the B4GAT1 product of GlcA addition to Xyl-α-pNP with both of the key through bond 1H–13C connections across the glycosidic linkage highlighted , unequivocally identifying the 1 , 4 glycosidic linkage . DOI: http://dx . doi . org/10 . 7554/eLife . 03943 . 010 Since B4GAT1 transfers GlcA to both anomers of xylopyranoside in contrast to LARGE and produces a β1 , 4-linkage as opposed to a β1 , 3-linkage , we sought to determine if the xylosyltransferase domain of LARGE would transfer to products of B4GAT1 . We tested GlcA-β4-Xyl-α-pNP and GlcA-β4-Xyl-β-MU ( products of B4GAT1 ) as well as GlcA-β3-Xyl-α-pNP ( the product of LARGE as a positive control ) , each of which we produced and then purified by reverse phase C18 HPLC before use . Reactions were carried out overnight at 37°C with UDP-Xyl as the donor . Substrates and products were desalted by reverse phase C18 spin columns and then subjected to C18 reverse-phase HPLC . Distinct substrate and product peaks were detected for each of the three potential acceptors tested and the identity of each peak was confirmed by LC-MS/MS ( Figure 4 ) . Hence , LARGE can extend both disaccharide products of B4GAT1 and its own disaccharide product with α3-linked xylose . 10 . 7554/eLife . 03943 . 011Figure 4 . B4GAT1 disaccharide products are substrates for the xylosyltransferase activity of LARGE . S indicates substrate peaks , P indicates product peaks ( xylose added ) . Separation was carried out by isocratic C18 reverse-phase HPLC with absorbance monitoring at 310 nm . Products were confirmed by accurate mass and MS/MS fragmentation shown below with assignment of key characteristic peaks . ( A ) P = Xyl-α3-GlcA-β4-Xyl-α-pNP . ( B ) P = Xyl-α3-GlcA-β4-Xyl-β-MU . ( C ) The product of the sequential dual enzymatic activity of LARGE as a positive control; P = Xyl-α3-GlcA-β3-Xyl-α-pNP . Reactions were carried out overnight in 0 . 1 M MES pH 6 . 5 containing 10 mM MnCl2 , 5 mM MgCl2 , 2 mM substrate and 2 mM UDP-Xyl . DOI: http://dx . doi . org/10 . 7554/eLife . 03943 . 011 Having produced three different trisaccharides with non-reducing end terminal xylose , Xyl-α3-GlcA-β4-Xyl-α-pNP , Xyl-α3-GlcA-β4-Xyl-β-MU ( sequential products of B4GAT1 and xylosyltransferase activity of LARGE ) , and Xyl-α3-GlcA-β3-Xyl-α-pNP ( the product of the bifunctional glucuronyltransferase and xylosyltransferase activities of LARGE ) , we tested each for elongation by B4GAT1 and LARGE . Incubation of the trisaccharides with UDP-GlcA and LARGE produced tetrasaccharide products in all cases , which were observed by HPLC and confirmed by LC-MS/MS ( Figure 5 ) . In sharp contrast , we were unable to produce any observable tetrasaccharide with B4GAT1 using any of the three trisaccharides as acceptors ( Figure 5 ) , though the enzyme was active for transfer to monosaccharide xylopyranosides ( Figure 2 ) . 10 . 7554/eLife . 03943 . 012Figure 5 . Trisaccharides terminating in an a3-xyloside are inefficient substrates for B4GAT1 in contrast to LARGE . Purified trisaccharides were incubated with LARGE or B4GAT1 plus UDP-GlcA overnight and then separated by isocratic reverse-phase C18 HPLC . ( A ) P with a subscript indicates product ( addition of GlcA ) whereas S with a subscript indicates an unmodified trisaccharide substrate . The top chromatogram in each pair shows the result of incubation with B4GAT1 , the bottom shows the result of incubation with LARGE . ( B ) Results were confirmed by MS/MS fragmentation spectra , only product spectra are shown . Reactions were carried out overnight in 0 . 1 M MES pH 6 . 5 containing 10 mM MnCl2 , 5 mM MgCl2 , 2 mM substrate and 2 mM UDP-GlcA . DOI: http://dx . doi . org/10 . 7554/eLife . 03943 . 01210 . 7554/eLife . 03943 . 013Figure 5—figure supplement 1 . B4GAT1 and LARGE do not appear to possess branching activity . Using purified disaccharide products of B4GAT1 or LARGE , we investigated B4GAT1 and LARGE for potential branching activity . Xyl-α-pNP and Xyl-β-pNP were included as controls . Transfer of GlcA by B4GAT1 or LARGE to disaccharides containing a terminal GlcA residue was not detected by radioactive transfer assay . Reactions were carried out overnight in 0 . 1 M MES pH 6 . 5 containing 10 mM MnCl2 , 5 mM MgCl2 , 2 mM substrate , 40 , 000 DPM UDP-[14C]GlcA and 2 mM non-radioactive donor . DOI: http://dx . doi . org/10 . 7554/eLife . 03943 . 013 The inability of B4GAT1 to modify trisaccharides ending with xylose does not rule out the possibility that B4GAT1 may possess a branching activity . We tested whether B4GAT1 or LARGE were capable of transferring glucuronic acid to any of the tagged purified disaccharides GlcA-β4-Xyl-α-pNP , GlcA-β4-Xyl-β-pNP or GlcA-β3-Xyl-α-pNP . We did not detect any transfer of GlcA in these in vitro assays ( Figure 5—figure supplement 1 ) . Thus , none of the GlcA-capped products of the enzymes appear to be substrates for the other enzyme . To further characterize the interplay between LARGE and B4GAT1 , we examined production of polymer by LARGE both with and without B4GAT1 . Incubation of Xyl-α-pNP or Xyl-β-pNP with either B4GAT1 alone , LARGE alone , or B4GAT1 and LARGE with both sugar nucleotide donors resulted in substantially different polymer production time courses as assayed by radioactive transfer ( Figure 6 ) . Time points were taken at 1 , 2 , 14 and 36 hr . B4GAT1 rapidly completes transfer to substrate . ∼1200 DPM represents complete transfer to the 10 nmol of substrate present . LARGE is unable to transfer to Xyl-β-pNP ( Figure 6B ) and shows a significant lag in initial transfer to Xyl-α-pNP before the formation of polymer . In contrast , B4GAT1 and LARGE together demonstrate robust initial transfer activity and formation of polymer using either substrate . Analysis of the synthesized polymer via mass spectrometry shows extensive polymerization is achievable following incubation with both enzymes ( Figure 6—figure supplement 1 ) . These data provide further evidence that LARGE possesses poor initiating activity but works efficiently in polymer formation once B4GAT1 is added for initial transfer of GlcA . 10 . 7554/eLife . 03943 . 014Figure 6 . The slow reaction velocity of LARGE is rescued by addition of B4GAT1 . Reactions were carried out in 0 . 1 M MES pH 6 . 5 with 10 mM MnCl2 , 5 mM MgCl2 , 40 , 000 DPM UDP-[14C]GlcA and 10 mM non-radioactive UDP-GlcA and UDP-Xyl each . Aliquots were removed at the displayed time points , boiled , and processed using RP C18 spin columns to separate untransferred donor from substrate . ( A ) The initial transfer of GlcA by LARGE to Xyl-α-pNP is slow in the absence of B4GAT1 but after transfer of the first GlcA polymerization rates increase to mirror those of LARGE in the presence of B4GAT1 . ( B ) With Xyl-β-pNP as substrate , B4GAT1 , which only possesses glucuronyltransferase activity , transfers a single GlcA per molecule of substrate . LARGE is unable to transfer GlcA to Xyl-β-pNP . Polymerization is only observed with the addition of both enzymes to the reaction mixture . DOI: http://dx . doi . org/10 . 7554/eLife . 03943 . 01410 . 7554/eLife . 03943 . 015Figure 6—figure supplement 1 . B4GAT1 and LARGE combined produce extended GlcA-Xyl polymer chains . B4GAT1 and LARGE were incubated together for 36 hr in 0 . 1 M MES pH 6 . 5 containing 10 mM MnCl2 , 5 mM MgCl2 , 10 mM UDP-GlcA , 10 mM UDP-Xyl and 1 mM Xyl-α-pNP . Mass spectra were acquired on a Bruker microflex and depict polymer formation . Note , due to likely decreased ionization efficiency with increasing polymerization , the sizes of observed peaks are unlikely to represent true abundances of each polymer species in the mixture . DOI: http://dx . doi . org/10 . 7554/eLife . 03943 . 015 O-mannosylation of α-dystroglycan is required for its proper function and when disrupted is a significant cause of a subset of congenital muscular dystrophies referred to as secondary dystroglycanopathies ( Barresi and Campbell , 2006; Freeze , 2007; Muntoni et al . , 2011; Mercuri and Muntoni , 2012; Dobson et al . , 2013; Wells , 2013; Praissman and Wells , 2014 ) . A phosphorylated O-mannose trisaccharide ( core M3 , [Praissman and Wells , 2014] ) attached to α-dystroglycan and its extension by LARGE after poorly defined intermediate biosynthetic steps ( Figure 7 ) has been shown to be directly involved in the required ECM ligand binding activity ( Yoshida-Moriguchi et al . , 2010; Inamori et al . , 2012; Yoshida-Moriguchi et al . , 2013 ) . Here , we have established that B4GAT1 is a β-1 , 4-glucuronyltransferase , not a B3GNT ( Figure 1 ) , with activity toward monomeric α- and β- xylopyranosides ( Figure 2 , Figure 3 ) . We have gone on to show that LARGE can extend the resulting products of B4GAT1 ( Figure 4 ) . Prior reports indicated that B4GAT1 ( B3GNT1 ) activity is necessary for expression of LARGE-dependent functional glycosylation of α-dystroglycan and that B4GAT1 ( B3GNT1 ) and LARGE form a complex ( Bao et al . , 2009; Buysse et al . , 2013; Czeschik et al . , 2013; Shaheen et al . , 2013 ) . We were unable to confirm this latter point using our soluble forms of the enzymes ( Figure 2—figure supplement 1 ) . Since we have shown that the specific activity and catalytic efficiency of B4GAT1 towards the tested monosaccharide xylosides is greater by more than an order of magnitude compared to LARGE catalysis ( Table 1 ) , this suggests that B4GAT1 is an initiating enzyme for LARGE-dependent glycan synthesis . This conclusion is further strengthened by data showing a significant lag in the initial rate for synthesis of or a complete failure to produce polymer in the absence of B4GAT1 ( Figure 6 ) . It also appears that B4GAT1 serves exclusively as a priming enzyme since we could not detect B4GAT1 modification of LARGE extended trisaccharides ( Figure 5 ) . We also demonstrate that neither LARGE nor B4GAT1 is capable of adding GlcA to a disaccharide that has been capped with GlcA suggesting that branching is not occurring ( Figure 5 , Figure 5—figure supplement 1 ) . This leads to the conclusion that there is only one type of repeating disaccharide ( Xyl-α3-GlcA-β3 ) attached to α-dystroglycan that is synthesized by LARGE . This data also serves to highlight the differences in structural isomers by showing both anomeric configuration ( α and β , Figures 2C and 6 ) as well as linkage position ( −3 and −4 , Figure 5 ) of isomeric substrates influences their subsequent ability to be elongated by enzymes ( LARGE and B4GAT1 ) . In sum , our data provide a basis for further attempts to fully elucidate the functional LARGE-dependent O-mannose-initiated structure ( s ) and suggests that at least one of the remaining incompletely characterized genes implicated in functional glycosylation of α-dystroglycan ( FKTN , FKRP , ISPD , and TMEM5 , [Jae et al . , 2013; Wells , 2013] ) likely encodes a xylosyltransferase to provide a substrate for B4GAT1 . Given that B4GAT1 is essential for proper glycosylation of α-dystroglycan ( Bao et al . , 2009; Buysse et al . , 2013; Czeschik et al . , 2013; Shaheen et al . , 2013 ) and that LARGE is incapable of transferring GlcA to a β-xylopyranoside ( Figure 2 , [Inamori et al . , 2012 , 2013] ) or generating a polymer from pNP-β-Xyl without B4GAT1 present ( Figure 6 ) , our findings strongly suggest that the underlying xylose is in a β-linkage on α-dystroglycan . This work also clarifies why deficiencies in this enzyme would be associated with loss of functional glycosylation and laminin binding of α-dystroglycan and be causal for congenital muscular dystrophy ( Bao et al . , 2009; Buysse et al . , 2013; Czeschik et al . , 2013; Shaheen et al . , 2013 ) . Further , our results likely clarify how loss of expression of B4GAT1 ( B3GNT1 ) , similar to loss of expression of LARGE ( de Bernabe et al . , 2009 ) , can lead to loss of laminin-binding α-dystroglycan and promote metastasis in certain cancers ( Bao et al . , 2009 ) . Finally , potential complex interactions between poly-N-acetyllactosamine , as often measured by the i-antigen antibody , and the LARGE-dependent repeating disaccharide , often measured on α-dystroglycan by the IIH6 antibody , need to be reexamined in the context of the newly defined enzymatic activity of B4GAT1 ( B3GNT1 ) ( Ujita et al . , 2000; Bao et al . , 2009; Lee et al . , 2009; Wright et al . , 2012; Yoneyama et al . , 2012; Buysse et al . , 2013; Czeschik et al . , 2013; Shaheen et al . , 2013 ) . For example , B4GAT1 has previously been observed to be in complex with B4GALT1 , an enzyme required for poly-N-acetyllactosamine synthesis ( Lee et al . , 2009 ) . This finding suggested that interaction between these two enzymes , which were thought to generate the repeating disaccharide , potentially influenced poly-N-acetyllactosamine synthesis . In lieu of our findings , this extrapolated conclusion from the interaction data is no longer valid . Instead , the relationship between poly-N-acetyllactosamine synthesis and the functional O-mannose glycan pathway needs to be evaluated , especially given that both have been implicated in cancer metastasis ( Ujita et al . , 2000; Bao et al . , 2009; Lee et al . , 2009; Wright et al . , 2012; Yoneyama et al . , 2012; Buysse et al . , 2013; Czeschik et al . , 2013; Shaheen et al . , 2013 ) . 10 . 7554/eLife . 03943 . 016Figure 7 . Proposed role of B4GAT1 in the O-mannosylation pathway . The molecular link between the phospho-O-mannose trisaccharide synthesized on α-dystroglycan and the LARGE synthesized repeating disaccharide crucial for laminin binding reactivity is still not fully characterized ( represented as [X] ) . B4GAT1 appears to possess a priming activity for LARGE and likely adds to an underlying β-xylose that is added by an as yet undefined glycosyltransferase ( represented with a question mark ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03943 . 016 We propose renaming B3GNT1 according to its defined activity as B4GAT1 ( Figure 1 , Figure 2 , Figure 3 ) . This designation is following in the tradition of the three other known non-proteoglycan glycoprotein glucuronyltransferases that all add GlcA in a β3-linkage ( B3GAT1-3 , [Morita et al . , 2008] ) . Interestingly , the only established glycoprotein glycoslytransferases capable of transferring GlcA in a β4-linkage are the dual-activity exostoses enzymes involved in heparan sulfate ( HS ) proteoglycan polymerization that build the repeating disaccharide ( -GlcNAc-α4-GlcA-β4- ) backbone of the glycosaminoglycan ( Zak et al . , 2002 ) . Similarities exist between HS ( and other proteoglycan ) biosynthesis pathways and the LARGE-dependent functional O-mannose glycan assembly pathway built on α-dystroglycan ( Figure 7 ) . Included in these similarities are that they both contain non-reducing end repeating disaccharides ( GlcA and GlcNAc for HS and GlcA and Xyl for the functional O-Man structure ) , they both contain an acidic glycan ( GlcA ) , there is a copolymerase that has dual–enzymatic activity ( exostoses enzymes for HS and LARGE for the functional O-Man glycan structure ) , both require a specific underlying core structure , and each contains specific priming glycosyltransferases ( EXTLs for HS [Kitagawa et al . , 1999; Kim et al . , 2001] , B4GAT1 for the functional O-Man glycan structure ) that adds one of the sugars found in the repeat to the underlying core structure to initiate elongation . Here we have established that B4GAT1 , previously referred to as B3GNT1 , is a xylopyranoside β1 , 4-glucuronyltransferase that appears to be the priming enzyme for the LARGE copolymerase for building the functional O-Man structure on α-dystroglycan that when defective causes CMD . Tagged monosaccharide glycosides were purchased from Sigma–Aldrich ( St . Louis , MO ) at ≥97% purity as were UDP-GlcA trisodium salt , UDP-GlcNAc disodium salt and UDP-Gal disodium salt . HPLC solvents were also purchased from Sigma–Aldrich ( Chromasolv grade ) . UDP-Xyl was purchased from CarboSource Services ( Athens , GA ) . Mini-protean TGX PAGE gels were purchased from Bio-Rad . UDP-[3H]GlcNAc was purchased from American Radiolabeled Chemicals ( St . Louis , MO ) and UDP-[14C]GlcA was from PerkinElmer ( Waltham , MA ) . The catalytic domains of human B3GNT1 ( amino acid residues 54–415 , UniProt O43505 ) , B3GNT2 ( amino acid residues 35–397 , UniProt Q9NY97 ) , and LARGE ( amino acid residues 91–756 , UniProt O95461 ) were expressed as soluble , secreted fusion proteins by transient transfection of HEK293 suspension cultures ( Meng et al . , 2013 ) . The coding regions were amplified from Mammalian Gene Collection Gerhard , 2004 #4 clones using primers that appended a tobacco etch virus ( TEV ) protease cleavage site ( Phan et al . , 2002 ) to the NH2-terminal end of the coding region and attL1 and attL2 Gateway adaptor sites to the 5ʹ and 3ʹ terminal ends of the amplimer products . The amplimers were recombined via BP clonase reaction into the pDONR221 vector and the DNA sequences were confirmed . The pDONR221 clones were then recombined via LR clonase reaction into a custom Gateway adapted version of the pGEn2 mammalian expression vector ( Barb et al . , 2012; Meng et al . , 2013 ) to assemble a recombinant coding region comprised of a 25 amino acid NH2-terminal signal sequence from the T . cruzi lysosomal α-mannosidase ( Vandersall-Nairn et al . , 1998 ) followed by an 8xHis tag , 17 amino acid AviTag ( Beckett et al . , 1999 ) , ‘superfolder’ GFP ( Pedelacq et al . , 2006 ) , the nine amino acid sequence encoded by attB1 recombination site , followed by the TEV protease cleavage site and the respective glycosyltransferase catalytic domain coding region . Suspension culture HEK293f cells ( Life Technologies , Grand Island , NY ) were transfected as previously described ( Meng et al . , 2013 ) and the culture supernatant was subjected to Ni-NTA superflow chromatography ( Qiagen , Valencia , CA ) . Enzyme preparations eluted with 300 mM imidazole were concentrated to ∼1 mg/ml using an ultrafiltration pressure cell membrane ( Millipore , Billerica , MA ) with a 10 kDa molecular weight cutoff . All reactions for Figure 2 through six were performed in 0 . 1 M MES pH 6 . 5 , 10 mM MnCl2 , 5 mM MgCl2 . Reactions summarized in Figure 1 were performed with omission of MgCl2 , conditions that more closely match those in the original literature ( Sasaki et al . , 1997; Zhou et al . , 1999 ) . Non-radioactive nucleotide sugar donors were included at 2 mM for analytical procedures excluding the polymer production assays in which both UDP-GlcA and UDP-Xyl were included at 10 mM . Nucleotide sugar donors were included at up to 8 mM for preparative scale production of material for NMR or purification for further reactions . Radioactive nucleotide sugar donors were included at approximately 40 , 000 DPM per sample . Substrate concentrations for analytical procedures were kept constant at 2 mM except in the kinetics assays . All incubations were carried out at 37°C . Incubation times for kinetics were set at 2 hr for B4GAT1 and 16 hr for LARGE based on time course curves used to ensure adequate transfer while maintaining the initial rate condition required . All other analytical reactions were performed for 16–18 hr while preparative reactions involving LARGE were carried out for upwards of 24 hr with occasional addition of enzyme due to the low specific activity of LARGE with respect to certain substrates . Enzymatic reactions were stopped by boiling for 5 min , acidified to 0 . 1% TFA , and glycoside acceptors were separated from sugar nucleotide using reverse-phase C18 spin ( The Nest Group , Inc . , Southborough , MA ) columns . Transfer was determined by scintillation counting or HPLC with LC-MS/MS verification of species . For scintillation counting , a PerkinElmer ( Waltham , MA ) Tri-Carb 2910 TR liquid scintillation counter was used with ScintSafe Plus 50% scintillation cocktail under standard settings for the isotope in question . HPLC was carried out on an Agilent ( Santa Clara , CA ) 1100 LC system equipped with variable wavelength absorbance detector set for monitoring pNP and MU derivatives ( 310 nm ) . Quantification was by peak area . Buffer A was 50 mM ammonium formate pH 4 . 3 and buffer B was 20% buffer A in 80% acetonitrile for all assays discussed . Separations were carried out by isocratic elution using a Grace Vydac 218 TP C18 column ( 5 μm particle size , 2 . 1 mm × 150 mm ) . All disaccharide products were separated using 14%B . Trisaccharide and tetrasaccharide separations were noted to be most strongly influenced by the identity of the underlying labeled disaccharide . Hence , all longer chain products and substrates extending GlcA-β4-Xyl-α-pNP were separated at 8%B , all products and substrates extending GlcA-β4-Xyl-β-MU were separated using 10%B , and all products and substrates extending GlcA-β3-Xyl-α-pNP were separated at 5%B . LC-MS and LC-MS/MS of reaction products was performed in positive mode on either a Thermo Fisher ( Waltham , MA ) Orbitrap XL or a Thermo Fisher Orbitrap Fusion utilizing short linear gradients from 0 . 1% formic acid in water to 0 . 1% formic acid in 80% acetonitrile . Full MS was acquired in the Orbitrap for accurate mass determination while glycoside MS/MS fragmentation spectra were obtained in the linear ion trap and assigned manually . Shotgun proteomics was performed on tryptic digests of purified enzyme samples according to our standard protocol on a Thermo Fisher Orbitrap XL ( Porterfield et al . , 2014 ) . Data was searched in Proteome Discoverer 1 . 4 using Sequest HT with the percolator node set at a 1% peptide false-discovery rate and a recent human reference proteome from Uniprot to which common contaminant protein sequences had been added . SDS-PAGE was also carried out according to standard protocols using the Bio-Rad ( Hercules , CA ) Precision Plus Protein Kaleidoscope molecular weight ladder . Gels were Coomassie Brilliant Blue G-250 stained and visualized . Uncleaved LARGE and TEV protease cleaved B4GAT1 were combined at equimolar concentrations in 20 mM HEPES pH 7 with 150 mM NaCl and 10 mM imidazole . Final protein concentration was 0 . 5 mg/ml . After incubation at 4°C for 15 min , the mixture was applied to Novagen Ni-NTA resin , washed three times with 20 mM imidazole in PBS pH 7 and eluted with 250 mM imidazole in PBS pH 7 . The input and collected fractions were analyzed by SDS-PAGE and silver staining ( Porterfield et al . , 2014 ) . Time course experiments were carried out to ensure that the initial rate condition required for Michaelis–Menten kinetic analysis was met . A 2 hr incubation at 37°C was found to be acceptable for B4GAT1 whereas at 16 hr incubation period was judged suitable for LARGE . We compared the activity of B4GAT1 to that of LARGE while varying monosaccharide acceptor concentration . Concentrations ranging from 200 nM to 2 mM were tested to generate typical Michaelis–Menten curves for each enzyme against the substrates Xyl-α-pNP and Xyl-β-pNP . UDP-GlcA was supplied at 2 mM in all reactions . Reactions were terminated by boiling and processed as described above with quantification by HPLC . A Lineweaver-Burk plot was generated in Microsoft Excel and kinetic parameters determined from it . We were unable to detect the presence of product for LARGE at 2 mM donor and acceptor after 16 hr with Xyl-β-pNP . Samples of the products from action of LARGE and B4GAT1 with UDP-GlcA on Xyl-α-pNP were purified as described above , and , for each , ∼1 mg was dissolved in 0 . 5 ml D2O . NMR spectra of each of these samples were obtained on a VNMRS 600 MHz spectrometer with a triple resonance HCN cold probe . 2-Dimensional 1H double quantum COSY , and TOCSY , and 1H-13C HSQC and HMBC spectra ( Van de Ven , 1995 ) were recorded using standard pulse sequences in the Agilent VnmrJ 4 . 3 software and processed in that software . The 1H and 13C peak assignments were made based on analysis of these data using the single and multiple bond connectivities these experiments reveal . The glycosidic linkages were established based on 1H-13C HMBC correlations in both directions across the glycosidic linkages ( Figure 3—figure supplement 1 ) . Anomeric stereochemistry was determined from the 1-bond C-H coupling of the anomeric sites where a value of less than 170 Hz indicates a β-linkage and one of more than 170 Hz indicates an α-linkage ( Bock and Pedersen , 1974 ) . These were further confirmed from the 3JH1 , H2 couplings of the respective sugar residues . 1H chemical shifts were calibrated relative to the HDO signal at 4 . 77 ppm at 25°C and the 13C shifts were then determined by the indirect referencing approach from the proton reference frequency ( Wishart et al . , 1995 ) . Following a protocol developed to significantly enhance spectra acquired for highly acidic carbohydrates ( Jacobs and Dahlman , 2001 ) , Nafion 117 solution was applied to a Bruker MSP 96 ground steel target and allowed to dry . Samples of polymerization reaction mixtures were mixed 1:1 with 2 , 5-DHB resuspended in 50% acetonitrile at 20 mg/ml and spotted on the dried Nafion 117 membrane . Matrix-assisted laser desorption ionization with time-of-flight detection mass spectrometry spectra were acquired using a Bruker Microflex .

Dystroglycan is a protein that is essential for muscles to function correctly , and helps to connect the interior framework of muscle cells with the external matrix of molecules that hold the cells together in the tissue . As is the case for many proteins , dystroglycan must have particular carbohydrate molecules joined to it in order to work correctly . Enzymes called glycosyltransferases assist with the reactions that build the carbohydrates on a protein . Mutations in multiple glycosyltransferases that add carbohydrates to dystroglycan can cause a group of diseases that are characterized by a progressive loss of muscle function , known as congenital muscular dystrophies . Praissman et al . use biochemical experimentation to investigate the role of one of these enzymes , known as B3GNT1 . The enzyme's name is based on a code that describes which carbohydrate it helps to bind to proteins . However , Praissman et al . ( and independently , Willer et al . ) discovered that this enzyme actually works with a different donor and acceptor than previously thought , and so should be called B4GAT1 instead . Praissman et al . propose that the B4GAT1 enzyme starts the process of forming the carbohydrate structures that help muscle cells bind to the muscle tissue matrix . B4GAT1 forms short carbohydrates on the surface of the part of dystroglycan that sits on the surface of cells . These carbohydrates are then extended into longer chains by another glycosyltransferase called LARGE . The results of Praissman et al . suggest that another enzyme is also involved in this process , which will require further studies to identify . Understanding the role of B4GAT1 and other glycosyltransferases that build functionally glycosylated dystroglycan could help to develop treatments for diseases such as muscular dystrophies .

2014

B4GAT1 is the priming enzyme for the LARGE-dependent functional glycosylation of α-dystroglycan

Cardiac progenitor cells ( CPCs ) must control their number and fate to sustain the rapid heart growth during development , yet the intrinsic factors and environment governing these processes remain unclear . Here , we show that deletion of the ancient cell-fate regulator Numb ( Nb ) and its homologue Numblike ( Nbl ) depletes CPCs in second pharyngeal arches ( PA2s ) and is associated with an atrophic heart . With histological , flow cytometric and functional analyses , we find that CPCs remain undifferentiated and expansive in the PA2 , but differentiate into cardiac cells as they exit the arch . Tracing of Nb- and Nbl-deficient CPCs by lineage-specific mosaicism reveals that the CPCs normally populate in the PA2 , but lose their expansion potential in the PA2 . These findings demonstrate that Nb and Nbl are intrinsic factors crucial for the renewal of CPCs in the PA2 and that the PA2 serves as a microenvironment for their expansion . Embryonic cardiac progenitor cells ( CPCs ) , identified from early embryos or differentiating pluripotent stem cells , hold tremendous regenerative potential with their unique capability to expand and differentiate into nearly all cell types of the heart ( Parmacek and Epstein , 2005; Kattman et al . , 2006; Moretti et al . , 2006; Kwon et al . , 2007 ) . Over the past decade , significant progress in developmental cardiology led to the identification of CPC markers and lineages ( Cai et al . , 2003; Kattman et al . , 2006; Moretti et al . , 2006; Kwon et al . , 2009 ) . However , CPCs are highly heterogeneous and it is unknown if they can undergo self-renewal without differentiation . Consequently , understanding the precise mechanisms of CPC self-renewal and maintenance remains a fundamental challenge . Cardiogenesis initiates as the basic helix-loop-helix protein mesoderm posterior 1 ( Mesp1 ) is transiently expressed in the nascent mesoderm during gastrulation ( Saga et al . , 1996 ) . Mesp1+ cells migrate anteriorly and form the first heart field ( FHF ) and second heart field ( SHF ) ( Saga et al . , 2000 ) . The FHF gives rise to the atria and left ventricle ( LV ) , whereas the outflow tract ( OT ) , right ventricle ( RV ) and some of atria are derived from the SHF ( Buckingham et al . , 2005 ) . Before myocardialization , subsets of Mesp1 progeny express CPC markers including Islet1 ( Isl1 ) , fetal liver kinase 1 ( Flk1 ) , Nkx2 . 5 , or myocyte-specific enhancer factor 2c ( Mef2c ) in precardiac mesoderm ( Stanley et al . , 2002; Cai et al . , 2003; Verzi et al . , 2005; Kattman et al . , 2006 ) . Isl1 and Flk1 expression is extinguished as CPCs adopt myocardial fates , but Nkx2 . 5 and Mef2c are continually expressed in cardiomyocytes ( Edmondson et al . , 1994; Tanaka et al . , 1999 ) . While CPCs expressing these markers have similar differentiation potential in vitro ( Kattman et al . , 2006; Moretti et al . , 2006; Wu et al . , 2006 ) , it is unknown if a discrete population of stem cell-like CPCs exist to supply cells for cardiac growth and morphogenesis during development . Numb and Numblike ( Numbl ) —mammalian Numb homologs sharing collinear topology and extensive sequence identity with functional redundancy—are evolutionarily conserved proteins that are required for the self-renewal of neural progenitors and mediate asymmetric cell divisions in various contexts of cell fate decisions ( Zhong et al . , 1997; Petersen et al . , 2002 , 2004; Roegiers and Jan , 2004 ) , but their role in CPC development has not been explored . In the current study , we sought to identify and investigate CPCs affected by Numb and Numbl . By taking combinatorial approaches , we demonstrate that Mesp1+ progenitor-derived Isl1+ Nkx2 . 5− cells renew and expand without cardiac differentiation in the second pharyngeal arch ( PA2 ) and that PA2 serves as their microenvironment during mammalian heart development . Numb is expressed ubiquitously in developing mouse embryos ( Zhong et al . , 1997; Jory et al . , 2009; Figure 1—figure supplement 1 ) . To quantitatively examine the expression of Numb and Numbl in developing CPCs , we used the embryonic stem ( ES ) cell differentiation system that recapitulates early cardiogenesis ( Kattman et al . , 2011; Van Vliet et al . , 2012; Figure 1A ) . Numb levels were relatively low at day 4 , when Mesp1 was induced , but upregulated at day 6 , when Isl1 appeared ( Figure 1B ) . Numbl levels were also increased at day 6 ( Figure 1B ) , implying that Numb and Numbl may have a role in CPC development after Mesp1 induction . To test this possibility , we deleted Numb in Mesp1+ progenitors , the earliest mesodermal progenitors giving rise to the entire heart and vasculature ( Saga et al . , 2000 ) , by crossing Mesp1Cre mice with Numbflox mice ( Zhong et al . , 2000 ) . The deletion did not affect LV formation , but resulted in a hypoplastic OT/RV and PA2 ( Figure 1C , C′ , D , D′ ) . While the phenotype appeared to be confined to the OT , RV , and PA2 , the penetrance was variable . Since Numbl-null mice are viable and fertile , but Numbl can compensate for loss of Numb function ( Petersen et al . , 2002 ) , we extinguished Numb and Numbl expression in Mesp1+ progenitors by deleting Numb in the Numbl-null background ( Petersen et al . , 2002 ) . The resulting Numb and Numbl double knockout ( Numb/Numbl DKO ) embryos showed a hypoplastic OT/RV and PA2 with complete penetrance ( n = 12 embryos , Figure 1E , E′ ) and uniformly lethal by embryonic day ( E ) 10 . 0 . These data suggest that Numb and Numbl are required for the formation of the PA2 and OT/RV . 10 . 7554/eLife . 02164 . 003Figure 1 . Numb and Numbl are required for PA2 and heart development . ( A ) Schema of cardiac differentiation in ES cell system . ( B ) Expression profiles of genes indicated during cardiac differentiation of ES cells . Gene expression was analyzed by qPCR . Data are mean ± SD; n = 4; d , day . ( C–E ) Lateral views of control ( C ) , Mesp1Cre; Numbflox/flox ( Numb KO , D ) , Mesp1Cre; Numbflox/flox; Numbl–/– ( Numb/Numbl DKO , E ) embryos . ( C′–E′ ) Enlargement of boxed areas in ( C–E ) , showing normal , hypoplastic or atrophic PA2 and heart in control ( C′ ) , Numb KO ( D′ ) or Numb/Numbl DKO ( E′ ) embryos , respectively . Pharyngeal arches ( red ) and outflow tract/right ventricle ( green ) are outlined in dashes . Scale bars , 150 μm . hd , head; pa , pharyngeal arch; h , heart; ot , outflow tract; ra , right atrium; rv , right ventricle . DOI: http://dx . doi . org/10 . 7554/eLife . 02164 . 00310 . 7554/eLife . 02164 . 004Figure 1—figure supplement 1 . Numb is ubiquitously expressed in developing embryos . ( A–C ) Mesp1 lineage-traced embryos ( Mesp1Cre; Ai9 ) at E7 ( A ) , E8 ( B ) , or E9 ( C ) . RFP marks Mesp1+ cells and their progeny . ( A′ , A″ , B′ , B″ , C′ , C″ ) Confocal images of corresponding embryos immunostained with Numb ( green ) or RFP ( red ) antibody , showing ubiquitous expression of Numb protein . Section planes are indicated by dotted lines in ( A ) , ( B ) , ( C ) . Boxed areas in ( A′ ) and ( B′ ) are shown in higher magnification in ( A″ ) and ( B″ ) , respectively . Second pharyngeal arch and heart sections are shown in ( C′ ) and ( C″ ) , respectively . Dapi ( blue ) was used to counterstain the nuclei . Scale bars , 10 mm ( A″ ) , 25 mm ( B″ , C′ , C″ ) , 50 mm ( A′ and B′ ) , and 100 mm ( C″ ) . a , anterior; p , posterior; ect , ectoderm; ve , viceral endoderm; hf , head fold; pm , precardiac mesoderm; pa2 , second pharyngeal arch; h , heart . DOI: http://dx . doi . org/10 . 7554/eLife . 02164 . 004 Based on the Numb/Numbl DKO phenotype , we focused on investigating Mesp1 progeny contributing to the PA2 and OT/RV . To visualize Mesp1 progeny affected by Numb and Numbl , we introduced a Cre reporter ( Ai9 ) into the DKO mouse embryos , in which red fluorescent protein ( RFP ) permanently marks Mesp1+ progenitors , and traced the RFP+ cells ( Figure 2A ) . The resulting RFP+ cells showed a near complete deletion of Numb , confirmed by fluorescence-activated cell sorting ( FACS ) and quantitative PCR ( qPCR ) ( Figure 2B ) . Interestingly , control and DKO littermates showed no noticeable morphological defects at E8 . 5 ( 8–12 somite stage ) and were histologically indistinguishable ( Figure 2—figure supplement 1 ) . RFP+ cells were normally expressed the transient CPC marker Isl1 ( Cai et al . , 2003 ) from the primordial PA2 to the bulbus cordis ( BC ) , and there was no significant difference in the number of RFP+ Isl1+ cells or the percentage of RFP+ Isl1+ cells positive for the mitosis marker phospho-histone H3 ( PH3 ) ( Figure 2—figure supplement 1B , B′ , G , G′ , K , Figure 2M ) . The RFP+ Isl1+ cells in PA2 primordia were continuous with the endo- , myo- and peri-cardial cell layers of the heart tube , and expressed the cardiac transcription factors Nkx2 . 5 and Mef2c from the distal BC and the sarcomeric protein α-Actinin from the proximal BC in both control and mutant embryos ( Figure 2—figure supplement 1B–E′ , G–J′ ) . Thus , Numb/Numbl DKO does not appear to affect CPC and heart development at E8 . 5 . 10 . 7554/eLife . 02164 . 005Figure 2 . Deletion of Numb and Numbl depletes Mesp1+ progenitor-derived Isl1+ Nkx2 . 5– cells in PA2 . ( A ) Generation of Numb/Numbl DKO embryos expressing RFP in Mesp1 progeny ( Mesp1Cre; Numbflox/flox; Numbl−/−; Ai9 ) . ( B ) Relative mRNA levels of Numb in RFP+ cells isolated from control and Numb/Numbl DKO embryos . Data are mean ± SD; n = 9 . ( C , C′ , G , G′ ) Lateral views of Mesp1+ cell-traced control ( C and C′ ) or Numb/Numbl DKO ( G and G′ ) embryos analyzed at E9 . 0 . RFP marks Mesp1 progeny . Control embryos show continuous RFP expression from second pharyngeal arch ( PA2 , outlined ) to heart ( asterisk , C′ ) , but the arch ( outlined ) is severely underdeveloped in Numb/Numbl DKO embryos without noticeable RFP expression ( asterisk , G′ ) . ( C″ and G″ ) Frontal views of control ( C″ ) or Numb/Numbl DKO ( G″ ) heart . ( D–F′ ) , ( H–J′ ) , Representative confocal images of transverse sections through PA2 and outflow tract ( OT ) of control ( D–F′ ) and Numb/Numbl DKO ( H–J′ ) embryos . Cutting planes are shown in ( C ) and ( G ) . Internal boundaries of PA2 are outlined in dashes . Dapi ( blue ) was used to counterstain the nuclei . ( K and L ) Representative plots of intracellular staining of RFP-gated cells . Isl1 and Nkx2 . 5 staining during day 5–8 ( K ) and cardiac troponin T ( cTnT ) and Nkx2 . 5 staining at day 8 ( L ) . ( M ) Average number of RFP+ Isl1+ Mef2c−Nkx2 . 5− cells in PA2 section ( 12-micron ) of indicated embryos and stages . Data are mean ± SD; n = 5; *p<0 . 05; ns , not significant . p values were determined using the paired Student t test . Scale bars , 25 μm ( D–F and H–J ) , 150 μm ( C and G ) . hd , head; pa , pharyngeal arch; ot , outflow tract; da , dorsal aorta; fe , foregut endoderm; ra , right atrium; rv , right ventricle; lv , left ventricle; ec , endocardial layer; mc , myocardial layer; pc , pericardial layer . DOI: http://dx . doi . org/10 . 7554/eLife . 02164 . 00510 . 7554/eLife . 02164 . 006Figure 2—figure supplement 1 . Numb/Numbl DKO embryos are grossly normal at E8 . 5 . ( A , A′ , F , F′ ) Lateral views of Mesp1+ cell-traced control ( A and A′ ) or Numb/Numbl DKO ( F and F′ ) embryos analyzed at E8 . 5 . RFP marks Mesp1 progeny . ( B–E′ and G–J′ ) Representative confocal images of transverse sections through primordial PA2 and BC of control ( B–E′ ) and Numb/Numbl DKO ( G–J′ ) embryos ( Cutting planes are shown in A and F ) with indicated markers . Internal boundaries of PA2 are outlined in dashes . Dapi ( blue ) was used to counterstain the nuclei . ( K ) Percentage of PH3+ cells in RFP+ Isl1+ cells in control and Numb/Numbl DKO embryos at E8 . 5 . Data are mean + SD; n = 4; *p<0 . 05; ns , not significant . p values were determined using the paired Student t test . Scale bars , 15 mm ( B–E and G–J ) , 200 mm ( A and F ) . hd , head; ht , heart tube; pa , pharyngeal arch; bc , bulbus cordis; fe , foregut endoderm . DOI: http://dx . doi . org/10 . 7554/eLife . 02164 . 00610 . 7554/eLife . 02164 . 007Figure 2—figure supplement 2 . Mef2c expression in PA2 and OT of control and Numb/Numbl DKO embryos . Scale bars , 25 mm . pa2 , second pharyngeal arch; ot , outflow tract . DOI: http://dx . doi . org/10 . 7554/eLife . 02164 . 00710 . 7554/eLife . 02164 . 008Figure 2—figure supplement 3 . Histogram of RFP + cell induction from Mesp-Cre; Ai9 ES Cells . RFP+ cells are induced from day 2–4 . DOI: http://dx . doi . org/10 . 7554/eLife . 02164 . 008 By E9 . 0 ( 18–22 somite stage ) the RFP+ cells in the PA2 were continuous with the OT in control embryos , but were nearly absent in Numb/Numbl DKO embryos ( Figure 2C–C″ , G–G″ ) . Histological analysis revealed that the PA2 contained a dense cluster of RFP+ Isl1+ cells in control embryos , but they were severely depleted in Numb/Numbl DKO embryos ( Figure 2D , D′ , H , H′ , M ) . This suggests that Numb and Numbl are required for the formation of the RFP+ Isl1+ cell cluster in the PA2 . Intriguingly , the RFP+ Isl1+ cells in the PA2 did not express the reported CPC genes Nkx2 . 5 and Mef2c ( Dodou et al . , 2004; Wu et al . , 2006 ) , but were continuous with the Nkx2 . 5+ Mef2c+ cells in the distal OT and thereafter the α-Actinin+ cells in the proximal OT and the RV ( Figure 2E–F′ , I–J′ , Figure 2—figure supplement 2 ) . This suggested that the transition of the Mesp1 lineage descendant to cardiomyocytes may occur through ( 1 ) Isl1+ Nkx2 . 5−/Mef2c−α-Actinin−cells , ( 2 ) Isl1+ Nkx2 . 5+/Mef2c+ α-Actinin−cells , ( 3 ) Isl1− Nkx2 . 5+/Mef2c+ α-Actinin+ cells . To test this possibility , we rederived ES cells expressing RFP in Mesp1+ cells from Mesp1Cre; Ai9 embryos , induced Mesp1+ precardiac mesoderm ( Kattman et al . , 2011 ) , and analyzed CPC differentiation by FACS . RFP+ cells appeared on day 4 of differentiation and expressed Isl1 on day 5 ( Figure 2—figure supplement 3 , Figure 2K ) . The RFP+ Isl1+ cells started to express Nkx2 . 5 from day 6 and differentiated into cardiac troponin T+ ( cTnT+ ) myocytes ( Figure 2K , L ) . This indicates Mesp1+ progenitors are specified to Isl1+ Nkx2 . 5− CPCs and transition to Nkx2 . 5+/cTnT+ cells in an ES cell system as they differentiate into cardiac cells . Together , these data suggest that Numb and Numbl are required to form a dense population of Mesp1+ cell-derived Isl1+ Nkx2 . 5− cells in the PA2 , which may give rise to Nkx2 . 5+ cardiac cells . While PAs contain SHF progenitors ( Kelly et al . , 2001; Xu et al . , 2004 ) , it is unknown if Mesp1+ cell-derived Isl1+ Nkx2 . 5− cells in the PA2 give rise to cardiac cells . Based on the Numb/Numbl DKO phenotype , cardiac gene expression patterns , and ES cell data , we hypothesized that the Isl1+ Nkx2 . 5− cells expand without differentiation in the PA2 and migrate out of the arch to differentiate into cardiac cells . To test this , we examined proliferating cells in cardiac regions of E9 . 0 embryos by performing whole-mount staining with 5-ethynyl-2′-deoxyuridine ( EdU ) , a nucleoside analog of thymidine . Remarkably , EdU+ cells were concentrated in the PA2 and rarely detected in other cardiac regions including the growing heart ( Figure 3A , B ) . Consistently , about four percent of Mesp1+ cell-derived ( RFP+ ) Isl1+ Nkx2 . 5− cells were PH3+ in the PA2 , whereas PH3+ cells were nearly absent in RFP+ Isl1+/− Nkx2 . 5+ OT and α-Actinin+ RV cells ( Figure 3C ) . Since DNA analogues are commonly used to show actively dividing stem cells and their lineages ( Eisenhoffer et al . , 2008; Snippert and Clevers , 2011 ) , we labeled the proliferating cells by administering a single pulse of EdU , and monitored EdU+ cells in RFP+ cells ( Figure 3D ) . EdU+ cells were first identified in the outer layer of the RFP+ cell cluster in PA2 after 2 hr , but not in the OT/RV ( Figure 3D ) . By 4–8 hr , the RFP+ EdU+ cells were abundant in the PA2 and OT , and eventually found in the RV after 24–48 hr of chase ( Figure 3D ) . RFP+ cells in the PA2 remained EdU+ ( Figure 3D , E ) , implying their renewal in the PA2 . The purdurance of the EdU pulse , determined by a sequential injection of EdU and 5-bromo-2′-deoxyuridine ( BrdU , another thymidine analog ) , was shorter than 4 hr ( Figure 3—figure supplement 1 ) . These data suggest that Mesp1+ cell-derived Isl1+ Nkx2 . 5− cells in the PA2 migrate out of the arch and differentiate into cardiac cells . 10 . 7554/eLife . 02164 . 009Figure 3 . Mesp1+ progenitor-derived Isl1+ Nkx2 . 5– cells expand in PA2 and differentiate into Nkx2 . 5+ heart cells after leaving PA2 . ( A ) Whole-mount view of EdU ( green ) -treated embryo at E9 . 0 . ( B ) Enlargement of the boxed area in ( A ) , showing enrichment or lack of EdU+ cells in the PA2 or heart region , respectively . ( C ) Percentage of RFP+ PH3+ cells in Isl1+ Mef2c−Nkx2 . 5− cells and Isl1+/− Mef2c+ Nkx2 . 5+ cells . Data are mean ± SD; n = 4 . ( D ) EdU pulse experiment . Top , EdU experiment schema . PA2s were also dissected out for ex vivo culture after 2 hr ( ** ) . Bottom , progeny of EdU+ cells at 2 , 4 , 8 , and 32 hr after single pulse of EdU injection at E9 . 0 demonstrating that Mesp1 progeny in PA2 proliferate and migrate to form the OT/RV . ( E ) Quantification of RFP+ EdU+ cell progeny shown in ( D ) . Data are mean ± SD; n = 3 . ( F–F″ ) , Cultured PA2 explants in 2D culture form a sheet of beating cardiomyocytes ( see Video 1 ) . ( G ) EdU-pulsed PA2-derived sheet of beating cells stained with cTnT , EdU , or Nkx2 . 5 ( inset ) . ( H ) Confocal image of RFP+ cells migrated from PA2 stained with cardiac α-actinin ( Act ) . ( I ) Intracellular Ca2+ transients from day 2 and day 8 PA2 explants and adult cardiomyocyte ( CM ) . a . u . , arbitrary unit . ( J ) Cultured PA1 explants form myotube-like cells . Inset shows a magnified view . *p<0 . 05 . D , day; Heart , E10 . 5 embryonic heart; ACTC1 , actin , alpha , cardiac muscle 1 . Dapi ( blue ) was used to counterstain the nuclei . p values were determined using the paired Student t-test . Scale bars , 10 μm ( H ) , 150 μm ( A and B ) , 250 μm ( F–F″ and J ) . pa , pharyngeal arch; h , heart; d-ot , distal outflow tract; p-ot , proximal outflow tract; rv , right ventricle; lv , left ventricle . DOI: http://dx . doi . org/10 . 7554/eLife . 02164 . 00910 . 7554/eLife . 02164 . 010Figure 3—figure supplement 1 . Dual Injection of EdU and BrdU . EdU and BrdU were sequentially injected at E9 . 5 at a 4-hr interval . The resulting embryos were harvested after 4 hr ( 8 hr after EdU injection ) and analyzed by immunostaining . EdU+ cells were mostly negative for BrdU , suggesting the purdurance of EdU is shorter than 4 hr . Some EdU+ BrdU+ cells ( asterisks ) were present in the PA2 , implying their continued proliferation . ot , outflow tract; fe , foregut endoderm; ec , endocardial layer; mc , myocardial layer; pc , pericardial layer . DOI: http://dx . doi . org/10 . 7554/eLife . 02164 . 01010 . 7554/eLife . 02164 . 011Figure 3—figure supplement 2 . Time lapse images of 3-D matrigel PA2 culture . These show growth and migration of Mesp1 progeny ( RFP+ ) in the PA2 ( Video 1 ) . Boxed area is shown in Video 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 02164 . 01110 . 7554/eLife . 02164 . 012Figure 3—figure supplement 3 . Explant culture of PA1 and PA2 dissected from Nkx2 . 5GFP embryo . pa , pharyngeal arch; ot , outflow tract; rv , right ventricle; ra , right atrium; lv , left ventricle . DOI: http://dx . doi . org/10 . 7554/eLife . 02164 . 012 To directly test the potential of RFP+ cells in the PA2 , we labeled proliferating cells with EdU for 2 hr at E9 . 0 , dissected out PA2s from embryos ( asterisks , Figure 3D ) , and cultured the arches in 3D Matrigel culture . RFP+ cells were monitored in real-time with fluorescent time-lapse movies . After 2 days , the cluster of RFP+ cells expanded and exhibited multidirectional migration ( Figure 3—figure supplement 2 ) . The migrated RFP+ cells became contractile by day 4 ( Figure 3—figure supplement 2; Videos 1 and 2 ) . Although the range of their migration was somewhat limited , likely due to the extracellular environment of Matrigel , RFP+ cells continued to expand in the PA2 and differentiated into cardiomyocytes . When cultured in an uncoated dish , RFP+ cells expanded in the PA2 and progressively migrated out of the arch in a unidirectional fashion soon after being attached to the surface . However , no beating activity was observed on day 2 ( Figure 3F ) . They migrated out of the arch progressively soon after being attached to the surface ( Figure 3F ) . The migrated RFP+ cells started to spontaneously contract by day 4 ( Figure 3F´ ) . The expansion and migration of RFP+ cells appear to continue over 2 weeks and the beating area was correspondingly expanded ( Figure 3F″; Video 3 ) . The beating cells were positive for EdU , Nkx2 . 5 , and cTnT/α-Actinin and exhibited a periodic Ca2+ oscillation pattern similar to that of adult cardiomyocytes ( Figure 3G–I ) . To ascertain that the resulting cardiac cells were not derived from contamination from the adjacent OT , we isolated PAs from Nkx2 . 5GFP embryos ( Biben et al . , 2000 ) that express GFP in cardiac cells including the OT , but not in the PAs . Freshly dissected PA2s did not contain GFP+ cells , but cells expressed GFP 2–3 days after migration from the arch ( Figure 3—figure supplement 3 ) . We concluded that RFP+ Isl1+ cells expand in the PA2 and differentiate into cardiac cells as they migrate out of the arch . It is worth noting that PA1 cells also migrated out from the arch , but they remained GFP–and eventually formed myotube-like structures ( Figure 3—figure supplement 3 , Figure 3J ) . This is consistent with the previous finding that pharyngeal mesoderm is also the source of head skeletal muscles ( Nathan et al . , 2008 ) . 10 . 7554/eLife . 02164 . 013Video 1 . 3D-cultured PA2 in Matrigel . This video shows expansion and migration of Mesp1 progeny ( RFP+ ) in ex-vivo cultured PA2 . DOI: http://dx . doi . org/10 . 7554/eLife . 02164 . 01310 . 7554/eLife . 02164 . 014Video 2 . High magnification of inset in Video 1 . This video shows RFP+ beating cardiac myocytes at a migrating edge of ex-vivo cultured PA2 . DOI: http://dx . doi . org/10 . 7554/eLife . 02164 . 01410 . 7554/eLife . 02164 . 015Video 3 . 2D-cultured PA2 . This video shows a beating monolayer of cardiomyocytes derived from RFP+ Mesp1 Progeny in ex-vivo cultured PA2 . DOI: http://dx . doi . org/10 . 7554/eLife . 02164 . 015 To determine if the PA2 affects CPC expansion , we isolated Isl1+ Nkx2 . 5− CPCs from ES cells and cultured them with PA2 cells or heart cells derived from E9 . 0–9 . 5 embryos or without feeders . The CPCs were obtained by differentiating Isl1Cre; Ai9 ES cells ( Uosaki et al . , 2012 ) into Mesp1+ precardiac mesoderm ( Kattman et al . , 2011 ) and purifying RFP+ Isl1+ Nkx2 . 5− cells by FACS at day 5 ( Figure 4A ) . The RFP+ cells spontaneously differentiate and form a sheet of beating cardiomyocytes ( Uosaki et al . , 2012 ) . Strikingly , the co-cultured RFP+ cells formed distinct colonies ( Figure 4A ) , which were never observed in control culture conditions , and their number was greatly increased over time ( Figure 4B ) . The increase was inversely correlated with the appearance of Nkx2 . 5+/cTnT+ cells ( Figure 4C , Figure 4—figure supplement 1 ) , indicating PA2 cells promote expansion of Isl1+ Nkx2 . 5− CPCs and suppress their cardiac differentiation . This is unlikely due to the altered 3D environment as the effect was mimicked by PA2 cell-conditioned medium , but not by embryonic heart cells ( Figure 4A , B , Figure 4—figure supplement 2 ) . 10 . 7554/eLife . 02164 . 016Figure 4 . PA2 cells promote CPC expansion and suppress cardiac differentiation . ( A ) FACS-purification of RFP+ Isl1+ Nkx2 . 5− CPCs induced from ES cell-derived precardiac mesoderm and their culture with no , embryonic PA2 , or embryonic heart feeders . Images show RFP+ cells at day 6 and day 7 . Isl1Cre; Ai9 ES cells were used to purify the CPCs at day 5 of cardiac differentiation , when Nkx2 . 5 is not expressed in Isl1+ CPCs . ( B ) Quantification of numbers of RFP+ cells cultured with no , embryonic PA2 , or embryonic heart feeders . Data are mean ± SD; n = 3 . ( C ) FACS plot of RFP+ Isl1+ Nkx2 . 5− CPCs differentiating into Nkx2 . 5+/cTnT+ cells with no , embryonic PA2 , or embryonic heart feeders , determined at day 8 . ( D ) Time-lapse images of Isl1+ Nkx2 . 5− CPC colony showing cardiac differentiation after removal of PA2 cells , indicated by GFP expression driven by Myh6 promoter . Scale bars , 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 02164 . 01610 . 7554/eLife . 02164 . 017Figure 4—figure supplement 1 . Percentages of Nkx2 . 5+ or cTnT + cells cultured with No , PA2 , or embryonic heart feeders at day 8 . NF , no feeder; PA2 , PA2 co-culture; HT , heart cell co-culture . Data are mean + SD; n = 3; *p<0 . 05 . p values were determined using the paired Student t test . DOI: http://dx . doi . org/10 . 7554/eLife . 02164 . 01710 . 7554/eLife . 02164 . 018Figure 4—figure supplement 2 . PA2 conditioned medium mimics PA2 co-culture . Con , control; CM , PA2-conditioned medium; PA2 , PA2 co-culture . Data are mean + SD; n = 3; *p<0 . 05 . p values were determined using the paired Student t test . DOI: http://dx . doi . org/10 . 7554/eLife . 02164 . 018 To test if the colonies maintain differentiation potential , we established an Isl1Cre; Ai9; Myh6-GFP ES cell line , in which green fluorescent protein ( GFP ) is expressed when cells differentiate into cardiomyocytes ( Ieda et al . , 2010 ) . RFP+ Isl1+ Nkx2 . 5− CPCs were FACS-purified from the line at day 5 of differentiation and co-cultured with PA2 cells to form colonies . In the presence of PA2 cells , the colonies continued to grow without GFP expression . However , when PA2 cells were removed after a week of co-culture , they began to express GFP from day 3 and differentiated into beating cardiomyocytes ( Figure 4D ) . These data suggest that PA2 cells provide a cellular environment for the renewal and expansion of Isl1+ Nkx2 . 5− CPCs and suppress their cardiac differentiation . Although the Cre-mediated conditional deletion revealed a crucial role of Numb and Numbl for the development of Mesp1 progeny in the PA2 , it was unclear if the Numb/Numbl DKO phenotype reflected the intrinsic role of Numb and Numbl . To address this question , we generated mosaic animals lacking Numb and Numbl in Mesp1 lineages ( Figure 5A ) . We first established an Numb/Numbl DKO ES cell line , in which conditional deletion of Numb/Numbl occurs in Mesp1+ cells with resultant expression of RFP , by directly deriving ES cells from the DKO embryos . The ES cells were injected into host blastocysts of UBI-GFP/BL6 mice ( Schaefer et al . , 2001 ) , which ubiquitously express GFP , to generate chimeras . In this system , Numb and Numbl deletion occurs in donor-derived Mesp1+ cells , and the DKO cells are traced by RFP expression . Donor-derived RFP+ cells were formed exclusively within Mesp1 lineages and distinguished from GFP+ host cells and RFP−GFP−donor progeny ( Figure 5B–D ) . No cells were found double positive for RFP and GFP ( Figure 5C , D ) , excluding the possibility of cell fusion , which potentially could rescue the phenotype of DKO cells . Therefore , we used wildtype blastocysts as hosts for further chimera generation . Consistent with the earlier mRNA analysis , Numb was absent in donor-derived RFP+ cells ( Figure 5E ) . 10 . 7554/eLife . 02164 . 019Figure 5 . Generation of Mesp1 lineage-specific somatic cells lacking Numb and Numbl in vivo . ( A ) Scheme for generation of RFP+ Numb/Numbl DKO cells in Mesp1 lineage . In this study , three independent sets of blastocyst injection were carried out and 16 chimeras were obtained from 132 embryos . ( B–E ) Chimeric embryos at E9 . 0 ( B ) generated with GFP+ host cells . Numb/Numbl DKO cells are shown in red . Sections were made transversely through cardiac region ( C–E ) and immunostained with RFP and GFP ( C and D ) or RFP and Numb ( E ) antibodies at corresponding stages . ( F–I ) Lateral views of chimera and sections , focused on PA and heart , showing contribution of RFP+ cells . Major contribution of RFP+ cells causes a phenotype similar to Numb/Numbl DKO embryos ( F and G ) . ( J–M ) Confocal images of heart transverse sections at indicated stages . α-Actinin and PCAM are properly expressed in RFP+ cells ( arrowheads ) . RFP+ area in ( J ) is enlarged in ( K ) . Dapi ( blue ) was used to counterstain the nuclei . Scale bars , 10 μm ( C , D , E , K ) , 100 μm ( B , G , I , J ) , and 200 μm ( L and M ) . a , anterior; p , posterior; hd , head; h , heart; pa , pharyngeal arch; ot , outflow tract; rv , right ventricle; lv , left ventricle; la , left atrium . DOI: http://dx . doi . org/10 . 7554/eLife . 02164 . 01910 . 7554/eLife . 02164 . 020Figure 5—figure supplement 1 . Numb/Numbl DKO cells are normally specified into OT cells . ( A–C ) Confocal images of PA2 sections of E9 . 0 chimera , immunostained with RFP and Isl1 ( A ) , RFP and Mef2c ( B ) or RFP , Isl1 and Mef2c ( C ) antibodies . RFP+ Isl1+ Mef2c+ cells are outlined in white ( C ) . Dapi ( blue ) was used to counterstain the nuclei . Scale bars , 50 mm . ot , outflow tract; pa , pharyngeal arch . DOI: http://dx . doi . org/10 . 7554/eLife . 02164 . 020 The phenotype of DKO chimeras depended on the contribution of the donor RFP+ cells . Major donor contribution caused a phenotype similar to DKO embryos ( 12 . 5% 2/16 , Figure 5F , G ) . In most cases ( 87 . 5% 14/16 ) , the chimeras were indistinguishable from wild-type embryos ( Figure 5H , I ) . Donor RFP+ cells were normally populated in the PA2 and contributed to the OT , cardiomyocytes and endocardial cells ( Figure 5J–M , Figure 5—figure supplement 1 ) , indicating deletion of Numb and Numbl did not affect the migration or cardiac differentiation of CPCs . In control embryos , the Isl1+ Nkx2 . 5– cells expanded dramatically after E8 . 5 in the PA2 ( Figure 2C′ , D , M ) . Proliferating cells , marked by PH3 , were mostly found within the border of the cluster in control embryos , but were depleted in Numb/Numbl DKO embryos ( Figure 6A–C‴ ) . Likewise , the chimeras formed the Isl1+ cell cluster with PH3+ cells in the PA2 ( Figure 6E ) . However , none of the RFP+ Isl1+ Nkx2 . 5− donor cells was found positive for PH3 in the PA2s of chimeric embryos analyzed at E9 . 0–9 . 5 ( n = 10 , Figure 6E–I ) , implying that Isl1+ Nkx2 . 5− cells in the PA2 lose their expansion potential in the absence of Numb and Numbl . There was no evidence of apoptosis in Numb/Numbl DKO cells in the PA2 ( not shown ) . Consistent with the in vivo data , knockdown of Numb and Numbl in ES cell-derived Isl1+ Nkx2 . 5− CPCs , but not in Mesp1+ or Nkx2 . 5+ CPCs , resulted in a significant reduction of cell proliferation ( Figure 6J , Figure 6—figure supplement 1 ) . Conversely , increased levels of Numb in the Isl1+ Nkx2 . 5− CPCs promoted their proliferation ( Figure 6K ) . These suggest that Numb and Numbl are cell-autonomous factors regulating the expansion of Isl1+ Nkx2 . 5− cells in the PA2 . 10 . 7554/eLife . 02164 . 021Figure 6 . Numb and Numbl are required for proliferation of Mesp1+ progenitor-derived Isl1+ Nkx2 . 5– cells in PA2 . ( A–G′ ) Confocal images of PA2 sections of control ( A–C‴ ) , Numb/Numbl DKO ( D ) , and chimeric ( E–G′ ) embryos , immunostained with PH3 , RFP , Isl1 antibodies . Asterisks indicate PH3+ RFP+ Isl1+ ( triple positive ) cells located in outer layers of RFP+ Isl1+ cell population ( outlined , A and E ) . Boxed areas in ( A ) , ( B ) , and ( E ) are shown in higher magnification in ( B ) , ( C–C‴ ) , and ( F–G′ ) , respectively . No PH3+ cells are found in RFP+ cells ( asterisks , F–G′ ) . RFP+ cells were outlined in white ( F–G′ ) . ( H and I ) Percentage of donor-derived Isl1+ Nkx2 . 5− cells in PA2 is shown in ( H ) and number of PH3+ RFP+ Isl1+ cells per 12-micron PA2 section is shown ( I ) . Data are mean ± SD; n = 10; *p<0 . 05 . ( J and K ) Relative percentage of EdU+ cells in ES cell-derived Mesp1+ progenitor , Isl1+ Nkx2 . 5− CPCs or Nkx2 . 5+ CPCs transfected with control or Numb/Numbl DKO siRNA ( J ) or Isl1+ Nkx2 . 5− CPCs transfected with control ( CTV ) or Numb overexpression construct ( CTV-Numb ) ( K ) . Data are mean ± SD; n = 3; *p<0 . 05 . The Mesp1+ , Isl1+ Nkx2 . 5− , or Nkx2 . 5+ cells were FACS-purified from day 4 Mesp1Cre; Ai9 , day 5 Isl1Cre; Ai9 , or day 6 Nkx2 . 5GFP ES cells , respectively . Dapi ( blue ) was used to counterstain the nuclei . p values were determined using the paired Student t test . Scale bars , 10 μm ( B , C , F , G ) . 50 μm ( A , D , E ) . fe , foregut endoderm; pa , pharyngeal arch ec , endocardial layer; mc , myocardial layer; pc , pericardial layer . DOI: http://dx . doi . org/10 . 7554/eLife . 02164 . 02110 . 7554/eLife . 02164 . 022Figure 6—figure supplement 1 . Knockdown efficiency of Numb siRNA and Numbl siRNA . Relative mRNA expression levels of Numb and Numbl in CPCs after transfected with scrambled siRNA ( Con ) , Numb siRNA ( Numb KD ) , or Numbl siRNA ( Numbl KD ) , determined by qPCR . Data are mean + SD; n = 3; *p<0 . 05 . p values were determined using the paired Student t test . DOI: http://dx . doi . org/10 . 7554/eLife . 02164 . 022 Through the use of mouse genetics , lineage-specific mosaicism , embryonic stem cell systems , and ex-vivo organ culture , we have shown the existence of an undifferentiated and expansive population of CPCs and their microenvironment during development ( Figure 7 ) , which may provide a stem cell-niche paradigm in cardiovascular biology . Given that the heart grows rapidly in size and cell number at E8 . 5–10 . 5 , the CPCs may serve as a renewable source to supply the number of cardiac cells needed to sustain the ensuing heart growth . 10 . 7554/eLife . 02164 . 023Figure 7 . Model for Renewal and Niche of Mesp1+ progenitor-derived CPCs . DOI: http://dx . doi . org/10 . 7554/eLife . 02164 . 023 Although highly heterogeneous , CPCs are considered as multipotent cells that are destined to become heart cells . However , it remains unclear if they are capable to self-renew without differentiation . In mice , their precursors are identified as early as E5 . 75 by expression of the T-box transcription factor Eomesodermin in the epiblast ( Russ et al . , 2000 ) and specified to Mesp1+ cells at the onset of gastrulation at E6 . 5 ( Costello et al . , 2011 ) . Mesp1+ cells are further specified to CPCs , vascular progenitors and some of the head mesenchyme that contribute to the entire heart , vasculature and subsets of head muscle cells , respectively ( Saga et al . , 2000 ) . CPCs giving rise to the RV migrate through the OT from the SHF and express the transcription factors Isl1 , Mef2c or Nkx2 . 5 . These factors , however , are also expressed in neighboring cells including pharyngeal ectoderm and endoderm , foregut endoderm , neural progenitors and neural crest cells that are not originated from mesoderm ( Lints et al . , 1993; Edmondson et al . , 1994; Cai et al . , 2003 ) , making it difficult to precisely discern CPCs . Moreover , it is unclear when and where CPCs are specified from Mesp1+ cells or their progeny . Thus , we traced Mesp1 lineages and examined the expression of CPC markers in Mesp1 progeny . Unexpectedly , heart cells rarely proliferated during early cardiogenesis , implying the existence of an alternate cell source for the ongoing growth of the heart . Indeed , we found a proliferative cluster of Mesp1+ cell-derived Isl1+ Nkx2 . 5−cells in the PA2—directly linked to the Isl1+ Nkx2 . 5+ OT of the growing heart—that migrated to the heart and became heart cells . The Isl1+ Nkx2 . 5− CPCs continued to expand in vivo , ex vivo , and in vitro without cardiac differentiation in the PA2 , suggesting that the CPCs may serve as a renewable cell source for the developing heart . This cell renewal system may provide a parsimonious and efficient way to quickly generate a large number of cells used to build the heart during embryogenesis , which can be advantageous over local proliferation of differentiated cardiac cells . Further characterization of the CPCs will be necessary to provide quantitative information on their cellular contribution to the developing heart . Recent studies suggested that subsets of head and cardiac muscles share their progenitors ( Lescroart et al . , 2010 ) . While the progenitor has not been identified , it will be interesting to investigate if this progenitor is derived from Isl1+ Nkx2 . 5− cells . PAs are transient , segmented bulges that appear on the craniolateral side of developing embryos ( Grevellec and Tucker , 2010 ) . Each PA is composed of a mesodermal core and neural crest cell–derived mesenchymal cells that are surrounded by ectoderm outside and endoderm inside . At E8 . 5 , the PA1—positioned most cranially among PAs—is structurally distinct , while the PA2 becomes noticeable after E9 . 0 . Mesp1-derived Isl1+ Nkx2 . 5− cells proliferate in PA2s and initiate expression of Nkx2 . 5+ soon after exiting the arch . This suggests that PA2s function as a microenvironment to maintain the CPCs in an undifferentiated and expanding state and likely contain cells secreting paracrine factors that control the CPC number and fate . In fact , numerous signaling molecules are secreted from PA endoderm , ectoderm , and mesenchyme including Wnts , bone morphogenetic proteins , sonic hedgehog , and fibroblast growth factors ( Rochais et al . , 2009 ) , and dysregulation of these signals is often associated with OT/RV defects ( Frank et al . , 2002; Stottmann et al . , 2004; Washington Smoak et al . , 2005; Kwon et al . , 2007 ) . Thus , it will be important to identify the extrinsic factors and cell types that provide key signals for the CPC maintenance and PA2 development . Although the heart phenotype ( hypoplastic OT/RV ) caused by Numb/Numbl DKO may not result entirely from CPC depletion in the PA2 , our findings together with published literatures suggest that the CPCs in the PA2 might be a major source of cells contributing to the OT/RV . For instance , the SHF—giving rise to the entire OT/RV—is located in PAs ( Kelly et al . , 2001; Rochais et al . , 2009 ) and we found that proliferating cells are present predominantly in the PA1 and PA2 and rarely detected in the rest of the PAs . Our ex-vivo culture further showed that PA2 cells robustly differentiate into cardiac cells , whereas PA1 cells appear to give rise to myotubes without cells expressing Nkx2 . 5 . It is also worth noting that nearly all , if not all , embryos showing severely hypoplastic OT/RV exhibit hypoplastic PA2s ( Srivastava et al . , 1997; Gottlieb et al . , 2002; Cai et al . , 2003; Kwon et al . , 2007 , 2009 ) . Numb and Numbl are highly conserved proteins that participate in cell fate determination by mediating asymmetric division , endocytosis and recycling of specific proteins , ubiquitination and cell migration ( Santolini et al . , 2000; Cayouette and Raff , 2002 ) . Classic studies of Drosophila demonstrated Numb's spatio-temporal segregation to one pole of the mitotic cell as the primary mechanism by which cell fate is determined in single organ precursors ( Uemura et al . , 1989 ) . In mammals , Numb and Numbl are required for the self-renewal of neural progenitors to maintain their number during development; while in the other settings they promote a neuronal fate by neural progenitor specification ( Petersen et al . , 2002 , 2004 ) . Similarly , Numb and Numbl were required for the renewal of CPCs during cardiogenesis , suggesting a conserved role in progenitor maintenance . It is unlikely that disruption of the yolk sac or angiogenesis contributed to the restriction of cardiac growth because there was no discernable difference in the phenotype of embryos or yolk sacs at E8 . 5 . Furthermore , the results from somatic mosaicism demonstrated that Numb/Numbl DKO CPCs were unable to proliferate in normal PA2 environment , suggesting a cell-autonomous role of Numb and Numbl . Curiously , the deletion of Numb and Numbl in CPCs appears to affect the growth of neighboring PA2 cells as well . This suggests that Numb and Numbl may also influence CPC renewal and expansion by regulating PA2 development in a non-cell autonomous manner . Numb and Numbl may affect CPC renewal and expansion via Notch , a conserved transmembrane receptor , given that Numb and Notch are mutually antagonistic ( Schweisguth , 2004 ) . In fact , Notch1 deficiency causes CPC expansion in the OT ( Kwon et al . , 2009 ) . The expansion of CPCs is at least in part mediated by accumulation of the Wnt signaling effector β-catenin that are negatively regulated by membrane Notch ( Kwon et al . , 2011 ) . Membrane Notch appears to require Numb and Numbl for the negative regulation of β-catenin ( Kwon et al . , 2011; Andersen et al . , 2012 ) , suggesting Numb and Numbl may be essential regulators of Notch and Wnt signaling during CPC development . With our current study , it will be necessary to re-examine the roles of Notch and Wnt signals at the level of renewing CPCs in the PA2 . Numb/Numbl DKO mouse embryos were generated by mating Mesp1Cre; Numbflox/+; Numbl−/+ with Numbflox/flox; Numbl−/+; Ai9 mice . Embryos were harvested from E7 . 0–10 . 0 and genotyped as described ( Saga et al . , 1999; Petersen et al . , 2002; Madisen et al . , 2010 ) . Nkx2 . 5GFP mice were used for ex-vivo culture ( Biben et al . , 2000 ) . For ES cell work , ESMesp1−Cre; Ai9 cells ( this work ) , ESMesp1−Cre; Numb flox/flox or flox/+; Numbl−/+ or −/−; Ai9 cells ( this work ) , ESIsl1−Cre; Ai9 cells ( Uosaki et al . , 2012 ) , ESNkx2 . 5−GFP cells ( Hsiao et al . , 2008 ) , and ESIsl1−Cre; Ai9; Myh6−GFP cells ( this work ) were used . ESMesp1−Cre; Ai9 , ESMesp1−Cre; Numb flox/flox or flox/+; Numbl−/+ or −/−; Ai9 or ESIsl1−Cre; Ai9; Myh6−GFP were derived from mice harboring Mesp1Cre; Ai9 , Mesp1Cre; Numbflox/flox or flox/+; Numbl+/− or −/−; Ai9 or Isl1Cre; Ai9; Myh6-GFP , respectively . ES cells were maintained on gelatin-coated dishes in maintenance medium ( Glasgow minimum essential medium with 10% fetal bovine serum and 1000 U/ml ESGRO ( Millipore , Billerica , MA ) , Glutamax ( Life Technologies/Thermo Fisher Scientific K . K . , Waltham , MA ) , sodium pyruvate , MEM non-essential amino acids ) and CPCs were induced with activin A , BMP4 , and VEGF ( R&D Systems , Minneapolis , MN ) ( Kattman et al . , 2011; Cheng et al . , 2013 ) . For explant culture , PA1s or PA2s were carefully dissected from embryos at E9 . 0–9 . 5 . Absence of contamination from the adjacent outflow tract was confirmed by absence of Nkx2 . 5+ cells . The explants were cultured in standard serum free medium supplemented with ascorbic acid at 37°C in 5% CO2 . For 3D cultures , the PA explants were cultured in Matrigel . For PA2 cell co-culture , PA2s were isolated from E9 . 0–9 . 5 Isl1Cre; Ai9 embryos and cultured in gelatin-coated plate with PA media ( DMEM: F-12 , 7 . 5% FBS , 1X penicillin-streptomycin , 1X Glutamax ) . RFP− cells from PA's with significant outgrowth were isolated and further passaged . The PA2-derived or control ( embryonic heart ) cells were plated at a density of 10 , 000–50 , 000 cells/cm2 in multi-well plates and ES cell-derived CPCs were plated on top of the PA2 cells at a density of ∼10 , 000 cells/cm2 in SFD medium . For Numb/Numbl DKO ES cell derivation , 3- to 4-week-old Numbflox/flox; Numbl−/+; Ai9 female mice were super-ovulated and mated with Mesp1Cre; Numbflox/+; Numbl−/+ males . Blastocysts were flushed at E3 . 5 and cultured to establish mouse ES cell lines ( Ying et al . , 2008 ) . Numb/Numbl DKO ES cell lines were identified by genotyping and karyotyped to select suitable lines for the production of chimera . Numb/Numbl DKO ES cells were injected into the UBI-GFP/BL6 or wildtype blastocysts to generate chimera , which were transferred to E0 . 5–1 . 5 pseudo-pregnant recipient mothers . Chimeric embryos were harvested and analyzed at E7 . 5 , 9 . 0 , 14 . 0 . For Numb and Numbl knockdown experiments , Numb/Numbl ON-TARGETplus SMARTpool siRNA ( L-046935/L-046983 ) or scrambled siRNA ( Dharmacon/Thermo Fisher Scientific K . K . ) was used at 100 nM for cell transfection ( Kwon et al . , 2011 ) . For Numb overexpression , full-length Numb cDNA was cloned into CTV vector ( Xiao et al . , 2007 ) and used to increase Numb levels by transfecting in Cre-expressing cells . Cells were transfected with Lipofectamine LTX or Lipofectamine 2000 ( Life Technologies ) in single-cell suspensions . For EdU pulse-tracing experiments , 10 mM EdU was injected at 0 . 075 mg per gram bodyweight intraperitoneally to pregnant mice at E9 . 0 . Embryos were harvested at 2 , 4 , 8 , and 32 hr post EdU injection . We used the Click-it EdU kit ( Life Technologies ) for EdU detection followed by immunostaining with primary and secondary antibodies . For the dual injection experiment , EdU and BrdU ( 300 μl , 10 mM each ) were sequentially injected at a 4-hr interval . Embryos were harvested 4 hr post BrdU injection ( 8 hr of EdU ) , fixed , and sectioned ( 12 μm thickness ) . Antigen retrieval was performed with microwave for 20 min in 10 mM EDTA solution . The section was epimerized with 0 . 2% PBS Triton X for 15 min and stained with the Click-it EdU kit . The resulting sections were washed and incubated with anti BrdU antibody in incubation buffer ( Roche BrdU Labeling and Detection Kit I ) . Anti-Mouse Ig-Fluorescein was used as secondary antibody . For microscopy , embryos were fixed in 4% paraformaldehyde overnight and then 30% sucrose , and then embedded in OCT , sectioned and stained using standard protocols . Antibodies used were: goat α-Islet1 ( 1:200; R&D ) , goat α-PECAM ( 1:200; R&D ) , mouse α-Islet1 ( 1:200; Developmental Studies Hybridoma Bank , Iowa City , IA ) , rabbit α-MEF2c ( 1:200; Cell Signaling Technology , Danvers , MA ) , rabbit α-RFP ( 1:400; Clontech Laboratories , Inc . , Mountain View , CA ) , rabbit α-Numb ( pre-absorbed , 1: 500; Abcam , Cambridge , MA or from Dr . Zhong ) , goat or rabbit anti-β1 integrin ( 1:400; R&D or 1:1000; Abcam ) , mouse α-sarcomeric Actinin ( 1:500; Sigma-Aldrich , St . Louis , MO ) , goat α-Nkx2 . 5 ( 1:20; Santa Cruz Biotechnology , Dallas , Texas ) , mouse α-PH3 ( 1:500; Abcam ) , rabbit α-GFP ( 1:400; Abcam ) , goat α-GFP ( 1:200; R&D ) , rabbit α-caspase 3 ( Abcam ) . Alexa Fluor secondary antibodies ( Life Technologies ) were used for secondary detection and confocal images acquired with a Zeiss LSM 510 Meta confocal microscope using Zen acquisition software . For flow cytometry , cells were dissociated and analyzed with Accuri C6 Flowcytometer ( BD Biosciences , San Jose , CA ) and FlowJo software ( TreeStar , Ashland , OR ) . For intracellular-flow cytometry , cells were stained with indicated antibodies after dissociation as previously described ( Uosaki et al . , 2011 ) . For FACS , dissociated cells were resuspended in PBS containing 0 . 1% FBS , 20 mM Hepes and 1 mM EDTA and sorted with FACSAria II ( BD Biosciences ) and SH800 sorter ( Sony Biotechnology , Japan ) . Time-lapse imaging was done with a BZ9000 All-in-One Fluorescence microscope ( Keyence , Japan ) . PA2 explants were incubated with 3 µM Fura-2 AM ( Invitrogen , Molecular Probes , Carlsbad CA ) for 20 min at 37°C . After washing and de-esterification for 20 min the explants were placed in an imaging chamber and electrical field stimulated at 1 Hz , 37°C , pH 7 . 4 with Ca2+ . The change in intracellular Ca2+ was measured with an inverted fluorescence microscope ( TE2000 , Nikon , Japan ) and Myocam ( IonOptix , Milton , MA ) by Fura-2 AM fluorescence intensity ratio at 340 nm and 380 nm . Differences between groups were examined for statistical significance using the paired Student's t test . A p-value <0 . 05 was regarded as significant . Error bars indicate standard error of the mean .

Human embryos contain cells called ‘cardiac progenitor cells’ that serve as the building blocks to make the heart . Cardiac progenitor cells , or CPCs for short , initially move into areas of the embryo called the first and second heart fields , and then undergo a change to become specific types of heart cells: such as cardiac muscle cells . However , it is not known if CPCs are maintained during the development of the heart . Now , Shenje , Andersen et al . have shown that Numb and Numblike—two proteins that are needed for the development of nerve cells—are also involved in the development of the heart . Mouse embryos without the genes for Numb and Numblike failed to develop hearts normally; and these mutants also had fewer CPCs in the ‘second pharyngeal arch’: a part of the embryo that becomes the sides and front of the neck . Experiments on wild-type mice showed that the CPCs multiplied within this arch , and then changed into specific heart cells as they left this structure . Furthermore , mixing CPCs in a petri dish with cells taken from this arch encouraged the CPCs to multiply without changing into specific cell types . To investigate the importance of these two proteins further , Shenje , Andersen et al . engineered ‘chimeric’ mice in which some CPCs contained the Numb and Numblike genes and other CPCs did not . In most of these chimeric mice , the hearts developed normally , but the CPCs without the Numb or Numblike genes failed to multiply in the second pharyngeal arch . This shows that these genes must be present within an individual CPC to regulate the multiplication of that cell within this arch . By uncovering how problems with the maintenance of CPCs can lead to heart defects—a very common birth defect in humans—this work may lead to new ways to prevent or treat congenital heart disease . Furthermore , identifying the other factors or mechanisms that can allow the long-term maintenance of CPCs in the laboratory will be crucial for research into heart regeneration , and for CPC-based treatments to repair the heart .

2014

Precardiac deletion of Numb and Numblike reveals renewal of cardiac progenitors

The Drosophila genome contains >13000 protein-coding genes , the majority of which remain poorly investigated . Important reasons include the lack of antibodies or reporter constructs to visualise these proteins . Here , we present a genome-wide fosmid library of 10000 GFP-tagged clones , comprising tagged genes and most of their regulatory information . For 880 tagged proteins , we created transgenic lines , and for a total of 207 lines , we assessed protein expression and localisation in ovaries , embryos , pupae or adults by stainings and live imaging approaches . Importantly , we visualised many proteins at endogenous expression levels and found a large fraction of them localising to subcellular compartments . By applying genetic complementation tests , we estimate that about two-thirds of the tagged proteins are functional . Moreover , these tagged proteins enable interaction proteomics from developing pupae and adult flies . Taken together , this resource will boost systematic analysis of protein expression and localisation in various cellular and developmental contexts . With the complete sequencing of the Drosophila genome ( Adams et al . , 2000 ) genome-wide approaches have been increasingly complementing the traditional single gene , single mutant studies . This is exemplified by the generation of a genome-wide transgenic RNAi library ( Dietzl et al . , 2007 ) to systematically assess gene function in the fly or by the documentation of the entire developmental transcriptome during all stages of the fly’s life cycle by mRNA sequencing ( Graveley et al . , 2011 ) . Furthermore , expression patterns were collected for many genes during Drosophila embryogenesis by systematic mRNA in situ hybridisation studies in different tissues ( Hammonds et al . , 2013; Tomancak et al . , 2002; 2007 ) . Particularly for transcription factors ( TFs ) , these studies revealed complex and dynamic mRNA expression patterns in multiple primordia and organs during development ( Hammonds et al . , 2013 ) , supposedly driven by specific , modular enhancer elements ( Kvon et al . , 2014 ) . Furthermore , many mRNAs are not only dynamically expressed but also subcellularly localised during Drosophila oogenesis ( Jambor et al . , 2015 ) and early embryogenesis ( Lécuyer et al . , 2007 ) . Together , these large-scale studies at the RNA level suggest that the activity of many genes is highly regulated in different tissues during development . Since the gene function is mediated by the encoded protein ( s ) , the majority of proteins should display particular expression and subcellular localisation patterns that correlate with their function . However , a lack of specific antibodies or live visualisation probes thus far hampered the systematic survey of protein expression and localisation patterns in various developmental and physiological contexts in Drosophila . Specific antibodies are only available for about 450 Drosophila proteins ( Nagarkar-Jaiswal et al . , 2015 ) , and the versatile epitope-tagged UAS-based overexpression collection that recently became available ( Bischof et al . , 2013; Schertel et al . , 2015 ) is not suited to study protein distribution at endogenous expression levels . Collections of knock-in constructs are either limited to specific types of proteins ( Dunst et al . , 2015 ) or rely on inherently random genetic approaches , such as the large-scale protein-trap screens or the recently developed MiMIC ( Minos Mediated Insertion Cassette ) technology ( Venken et al . , 2011 ) . The classical protein-trap screens are biased for highly expressed genes , and altogether recovered protein traps in 514 genes ( Buszczak et al . , 2007; Lowe et al . , 2014; Morin et al . , 2001; Quiñones-Coello et al . , 2007 ) . The very large-scale MiMIC screen isolated insertions in the coding region of 1862 genes , 200 of which have been converted into GFP-traps available to the community ( Nagarkar-Jaiswal et al . , 2015 ) . Both approaches rely on transposons to mediate cassette insertion and require integration into an intron surrounded by coding exons for successful protein tagging . Thus , about 3 , 000 proteins , whose ORF is encoded within a single exon , cannot be tagged by these approaches ( analysis performed using custom Perl script available here: https://github . com/tomancak/intronless_CDS_analysis ) . Together , this creates a significant bias towards trapping a particular subset of the more than 13 , 000 protein coding genes in the fly genome . Hence , the Drosophila community would profit from a resource that enables truly systematic protein visualisation at all developmental stages for all protein-coding genes , while preserving the endogenous expression pattern of the tagged protein . One strategy to generate a comprehensive resource of tagged proteins is to tag large genomic clones by recombineering approaches in bacteria and transfer the resulting tagged clones into animals as third copy reporter allele as was previously done in C . elegans ( Sarov et al . , 2012 ) . The third copy reporter allele approach was used successfully in Drosophila with large genomic BAC or fosmid clones derived from the fly genome . In Drosophila , it is possible to insert this tagged copy of the gene as a transgene at a defined position into the fly genome ( Venken et al . , 2006 ) . It has been shown that such a transgene recapitulates the endogenous expression pattern of the gene in flies and thus likely provides a tagged functional copy of the gene ( Ejsmont et al . , 2009; Venken et al . , 2009 ) . Here , we introduce a comprehensive genome-wide library of almost 10000 C-terminally tagged proteins within genomic fosmid constructs . For 880 constructs , covering 826 different genes we generated transgenic lines , 765 of which had not been tagged by previous genetic trapping projects . Rescue experiments using a subset of lines suggest that about two thirds of the tagged proteins are functional . We characterised the localisation patterns for more than 200 tagged proteins at various developmental stages from ovaries to adults by immunohistochemistry and by live imaging . This identified valuable markers for various tissues and subcellular compartments , many of which are detectable in vivo by live imaging . Together , this shows the wide range of possible applications and the potential impact this publically available resource will have on Drosophila research and beyond . We aimed to tag all protein coding genes at the C-terminus of the protein , because a large number of regulatory elements reside within or overlap with the start of genes , including alternative promoters , enhancer elements , nucleosome positioning sequences , etc . These are more likely to be affected by a tag insertion directly after the start codon . Signal sequences would also be compromised by an inserted tag after the start codon . This is in agreement with a recently published dataset favouring C-terminal compared to N-terminal tagging ( Stadler et al . , 2013 ) . Additionally , the C-termini in the gene models are generally better supported by experimental data than the N-termini due to an historical bias for 3'-end sequencing of ESTs . Thus , C-terminal tagging is more likely to result in a functional tagged protein than N-terminal tagging , although we are aware of the fact that some proteins will be likely inactivated by addition of a tag to the C-terminus . Moreover , only about 1400 protein coding genes contain alternative C-termini , resulting in all protein isoforms labelled by C-terminal tagging for almost 90% of all protein-coding genes ( analysis performed using custom Perl script available here: https://github . com/tomancak/alternative_CDS_ends ) . In a series of pilot experiments , we tested the functionality of several tagging cassettes with specific properties on a number of proteins ( Figure 1—figure supplement 1 , Table 1 ) . For the genome-wide resource , we applied a two-step tagging strategy , whereby we first inserted a non-functional ‘pre-tagging’ cassette consisting of a simple bacterial selection marker , which is flanked with linker sequences present in all of our tagging cassettes . This strategy enables a very efficient replacement of the ‘pre-tag’ by any tag of interest using homologous recombination mediated cassette exchange in bacteria ( Hofemeister et al . , 2011 ) . As fluorescent proteins and affinity tags with improved properties are continuously being developed , specific clones or the entire resource can be easily re-fitted to any new tagging cassette . For the genome-scale resource , we selected a tagging cassette suitable for protein localisation and complex purification studies , consisting of the 2xTY1 tag as a flexible linker , the superfolder GFP coding sequence ( Pédelacq et al . , 2005 ) , the V5 tag , followed by specific protease cleavage sites ( for the PreScission- and TEV-proteases ) , the biotin ligase recognition peptide ( BLRP ) tag allowing for specific in vivo or in vitro biotinylation ( Deal and Henikoff , 2010; Vernes , 2014 ) , and the 3xFLAG tag ( Figure 1—figure supplement 1 ) . 10 . 7554/eLife . 12068 . 005Table 1 . TransgeneOme constructs and fTRG lines - overview of TransgeneOme constructs generated and verified by sequencing for the different pilot sets and the genome-wide set , including the respective numbers of the transgenic fTRG lines generated . DOI: http://dx . doi . org/10 . 7554/eLife . 12068 . 005Tagged constructs and transgenic linesconstructsverified constructstransgenic lines‘pre-tagging’ - TransgeneOme11257----TY1-sGFP-V5-BLRP-FLAG799- TransgeneOme109959580- pilot set13281328TY1-T2A-sGFPnls-FLAG- pilot set27327330TY1-sGFP-FLAG- pilot set64448351unique constructs2316910711880unique genes112579993826 Of the 13937 protein-coding genes in the dmel5 . 43 genome assembly , 11787 genes ( 84 . 6% ) were covered by a suitable fosmid from the original FlyFos library ( Ejsmont et al . , 2009 ) , extending at least 2 . 5 kb upstream and 2 . 5 kb downstream of the annotated gene model . For picking clones , designing oligonucleotides for recombineering , and for tracking all steps of the transgene engineering process , as well as for providing access to all construct sequences and validation data we used the previously developed TransgeneOme database ( Sarov et al . , 2012 ) , which is available online ( https://transgeneome . mpi-cbg . de ) . For high-throughput tagging of the Drosophila FlyFos clones , we developed an improved version of our previously applied high-throughput , 96-well format liquid culture recombineering pipeline ( Ejsmont et al . , 2011; Sarov et al . , 2012 ) , and we applied it to create a single tagged construct for each gene covered by a suitable fosmid . The high efficiency of recombineering in E . coli allowed for multi-step DNA engineering in 96-well format liquid cultures with single clone selection only at the last step . The pipeline consists of five steps ( Figure 1B ) . First , the pRedFlp helper plasmid containing all genes required for homologous recombination and the Flippase-recombinase ( under L-rhamnose and tetracycline control , respectively ) was introduced into E . coli by electroporation . Second , the ‘pre-tagging’ cassette containing a bacterial antibiotic resistance gene was inserted by homologous recombination with gene-specific homology arms of 50 base pairs . Third , the sGFP-V5-BLRP tagging cassette , including an FRT-flanked selection and counter-selection cassette , was inserted to replace the ‘pre-tagging’ cassette . Since the linker sequences in the ‘pre-tagging’ cassette are identical to the tagging cassette , the tagging cassette was simply excised from a plasmid by restriction digest and no PCR amplification was required . This strongly reduced the risk of PCR-induced mutations in the tagging cassette . Fourth , the selection marker was excised by the induction of Flippase expression . Fifth , the helper plasmid was removed by suppression of its temperature sensitive replication at 37°C ( Meacock and Cohen , 1980 ) and single clones were isolated from each well by plating on selective solid agar plates . All five steps of the engineering pipeline were highly efficient ( between 95 . 8 and 99 . 7% ) , resulting in an overall efficiency of 93 . 6% or 10995 growing cultures ( Figure 1B ) . To validate the sequence of the engineered clones , we developed a new next-generation-sequencing ( NGS ) -based approach ( Figure 1C ) . In short , we pooled single clones from all 96-well plates into 8 rows and 12 columns pools , prepared barcoded mate pair libraries from each pool , and sequenced them using HiSeq2500 ( Illumina ) . The mate pair strategy allowed us to map the otherwise common tag coding sequence to a specific clone in the library and thus to verify the integrity of the tagging cassette insertion in the clones with single nucleotide resolution ( see Materials and methods for details ) . When applied to the final sGFP TransgeneOme collection , we detected no mutations for 9580 constructs ( 87 . 1% ) . 8005 ( 72 . 8% ) of these clones had complete sequence coverage of the tag-coding sequence and thus represent the most reliable subset of the tagged library ( Figure 1C ) . For 1417 of the clones ( 12 . 8% ) , one or more differences to the expected sequences were detected . The most common differences were point mutations , which cluster almost exclusively to the homology regions in the oligonucleotides used to insert the ‘pre-tagging’ cassette . This is suggestive of errors in the oligonucleotide synthesis . Another subset of point mutations clustered around the junctions between the homology arms and the rest of the tagging cassette , indicating an imprecise resolving of the homology exchange reaction in small subset of clones ( Figure 1D ) . Finally , a small group of clones ( 165 ) still contained an un-flipped selection cassette . The NGS results were confirmed by Sanger sequencing of the entire tag-coding sequence for a subset of constructs ( Supplementary file 1 ) . The detailed sequencing results for all clones are available at https://transgeneome . mpi-cbg . de . Taken together , the sGFP TransgeneOme and our pilot tagging experiments resulted in 10711 validated tagged clones , representing 9993 different Drosophila genes . ( Table 1 , Supplementary file 1 ) . The clones are available from Source BioScience as Drosophila TransgeneOme Resource ( MPI-CBG ) ( http://www . lifesciences . sourcebioscience . com/clone-products/non-mammalian/drosophila/drosophila-transgeneome-resource-mpi-cbg/ ) . Moreover , the 'pre-tagged' TransgeneOme library is a versatile resource for generating fosmid clones with arbitrary tags at the C-terminus of the gene models . We next established a pipeline to systematically transform the tagged TransgeneOme clones into flies . To efficiently generate fly transgenic lines , we injected the tagged fosmid constructs into a recipient stock carrying the attP landing site VK00033 located at 65B on the third chromosome using a transgenic nanos-ΦC31 source ( Venken et al . , 2006 ) . For some genes positioned on the third chromosome , we injected into VK00002 located on the second chromosome at 28E to simplify genetic rescue experiments . In total , we have thus far generated lines for 880 tagged constructs representing 826 different genes ( Table 1 , Supplementary file 2 ) . These genes were partially chosen based on results of a public survey amongst the Drosophila community to identify genes for which there is the strongest demand for a tagged genomic transgenic line . 765 ( 87% ) of the newly tagged genes have not been covered by the previous protein-trap projects ( Supplementary file 2 ) , hence , these should be particularly useful for the fly community . From our pilot tagging experiments , we made 51 lines for the 2xTY1-sGFP-3xFLAG tag and 30 lines for the 2xTY1-T2A-sGFPnls-FLAG transcriptional reporter . The majority of the lines ( 799 ) were generated with the versatile 2xTY1-sGFP-V5-Pre-TEV-BLRP-3xFLAG tag , used for the genome-wide resource ( Figure 1—figure supplement 1 , Table 1 ) . The collection of fly lines is called ‘tagged fly TransgeneOme’ ( fTRG ) and all 880 fTRG lines have been deposited at the VDRC stock centre for ordering ( http://stockcenter . vdrc . at ) . To assess whether the tagged fosmids in our transgenic library are functional , we have chosen a set of 46 well-characterised genes , mutants of which result in strong developmental phenotypes . For most cases , we tested null or strong hypomorphic alleles for rescue of the respective phenotypes ( embryonic lethality , female sterility , flightlessness , etc . ) with the tagged fosmid lines . More than two-thirds of the lines ( 31 of 46 ) , including tagged lines of babo , dlg1 , dl , fat , Ilk , LanB1 , numb , osk , rhea , sax , smo and yki rescued the mutant phenotypes ( Figure 2A , Table 2 ) , demonstrating that the majority of the tagged genes is functional . Our rescue test set is biased towards important developmental regulators; 10 of the 15 genes that did not show a rescue are transcription factors with multiple essential roles during development , such as esg , eya , odd , sna and salm . Thus , their expression is likely regulated by complex cis-regulatory regions that may not be entirely covered by the available fosmid clone; for example wing-disc enhancers are located more than 80 kb away from the transcriptional start of the salm gene ( De Celis et al . , 1999 ) . Hence , we expect that a typical gene , which is embedded within many other genes in the middle of the fosmid clone , is more likely to be functional . Together , these data suggest that both the genome-wide tagged construct library and the transgenic fTRG library provide functional reagents that are able to substitute endogenous protein function . 10 . 7554/eLife . 12068 . 006Figure 2 . Functionality tests of the GFP-tagged fTRG lines by genetic complementation analysis . ( A ) Genetic rescue statistics of null/strong mutant alleles for 46 selected fTRG lines . Note that more than two-thirds of the lines show a rescue ( see Table 2 ) . ( B , C ) osk-GFP mRNA ( in yellow ) expressed from fTRG1394 rescues egg-chamber development of an osk null allele ( Jenny et al . , 2006 ) . osk-GFP mRNA enriches in the early oocyte ( B , stage 6 ) and rescues the oogenesis arrest and the DNA condensation defect of the osk mutant ( B’ , yellow arrowhead ) . At stage 10 osk-GFP RNA enriches at the posterior pole ( C ) and produces sufficient protein to ensure proper embryogenesis . osk-GFP mRNA is shown in yellow , DAPI in magenta; scale bars indicate 30 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 12068 . 00610 . 7554/eLife . 12068 . 007Table 2 . Genetic rescue of mutants with the fTRG lines Table listing fTRG lines and respective genetic alleles as well as rescue assays that were used to assess the functionality of the fTRG lines . Note that about two-thirds of the lines rescue the mutant phenotypes . DOI: http://dx . doi . org/10 . 7554/eLife . 12068 . 007GeneChromosomefTRG lineTagRescue ? Rescue assayAlleles , deficiencies used in trans for rescue assayReferenceamos2ndfTRG_2182xTY1-sGFP-V5-preTEV-BLRP-3xFLAGyesantenna size and bristle number rescued to normalamos[3]anterior open ( aop , Yan ) 2ndfTRG_1422xTY1-sGFP-V5-preTEV-BLRP-3xFLAGyesembryonic lethality rescued to viable adultsaop[1] ( BL-3101 ) ; aop[Yan1] ( BL-8780 ) aubergine ( aub ) 2ndfTRG_5812xTY1-sGFP-V5-preTEV-BLRP-3XFLAGyesfemale sterility entirely rescuedaub[HN2] ( BL-8517 ) ; Df ( 2L ) BSC145 ( BL-9505 ) baboon ( babo ) 2ndfTRG_4442xTY1-sGFP-V5-preTEV-BLRP-3xFLAGyeslethality rescued to viable adultsbabo[32] ( BL-5399 ) ; babo[k16912] ( BL-11207 ) bag of marbles ( bam ) 3rdfTRG_32xTY1-sGFP-V5-preTEV-BLRP-3xFLAGyesfemale sterility entirely rescuedbam[delta86]; Df ( 3R ) exel9020Christian Bökel , pers . comm . cactus ( cact ) 2ndfTRG_5162xTY1-sGFP-V5-preTEV-BLRP-3xFLAGyeslethality and female sterility rescuedcact[1]; cact[4]CG321213rdfTRG_922xTY1-sGFP-V5-preTEV-BLRP-3xFLAGyesflightlessness rescuedCG32121 [XZ1] ( X . Zhang and F . S . , unpublished ) ; Df ( 3L ) ED4502 ( BL-8097 ) CG6509 ( dlg5 ) 2ndfTRG_102512xTY1-sGFP-3xFLAGyeslethality rescued to viable adultsCG6509 [KG006748] ( BL13692 ) ; Df ( 2L ) BSC244 ( BL-9718 ) discs large 1 ( dlg1 ) XfTRG_5022xTY1-sGFP-V5-preTEV-BLRP-3xFLAGyesmale lethality rescued to viable adultsDlg1[5] ( BL-36280 ) dorsal ( dl ) 2ndfTRG_292xTY1-sGFP-V5-preTEV-BLRP-3xFLAGyesbristle number rescued to normaldl[1]; dl[4]ebi2ndfTRG_101412xTY1-sGFP-3xFLAGyeslethality rescued to viable adultsebi[CCS-8] ( BL-8397 ) ; ebi[E90] ( BL-30720 ) escargot ( esg ) 2ndfTRG_101702xTY1-sGFP-3xFLAGnolethality not rescuedesg[35Ce-1] ( BL-3900 ) ; esg[35Ce-3] ( BL-30475 ) eyes absent ( eya ) 2ndfTRG_4922xTY1-sGFP-V5-preTEV-BLRP-3XFLAGnolethality not rescuedeya[C0233]; eya[C0275]fat ( ft ) 2ndfTRG_102332xTY1-T2A-nlsGFP-3xFLAGyeslethality rescued to viable adultsft[G-rv] ( BL-1894 ) ; ft[8] ( BL-5406 ) 48 related 2 ( Fer2 ) 3rdfTRG_3342xTY1-sGFP-V5-preTEV-BLRP-3XFLAGyesdefective climbing rescued to wild typeFer2[e03248]Dib et al . , 2014fizzy ( fzy ) 2ndfTRG_102502xTY1-T2A-nlsGFP-3xFLAGnolethality not rescuedfzy[1] ( BL-2492 ) ; fzy[3] ( BL-25143 ) flightless I ( fliI ) XfTRG_4672xTY1-sGFP-V5-preTEV-BLRP-3xFLAGyeslethality or flightlessness rescuedfli[14] ( BL-7481 ) ; fli[3] ( BL-4730 ) Hand2ndfTRG_101632xTY1-sGFP-3xFLAGyessemi-lethality rescued to viable adultsHand[173]hippo ( hpo ) 2ndfTRG_101302xTY1-sGFP-3xFLAGyeslarval lethality rescued to viable adultshpo[KS240] ( BL-25085 ) ; hpo[KC202] ( BL-25090 ) HLH54F2ndfTRG_1532xTY1-sGFP-V5-preTEV-BLRP-3xFLAGyeslethality rescued to viable adultsbHLH54F[598]; Df ( 2R ) Exel7150 ( BL-7891 ) Integrin linked kinase ( Ilk ) 3rdfTRG_4832xTY1-sGFP-V5-preTEV-BLRP-3xFLAGyesembryonic lethality rescued to viable adults ( wing blisters ) Ilk[1] ( BL-4861 ) ; Df ( 3L ) BSC733 ( BL-26831 ) Kinesin heavy chain ( Khc ) 2ndfTRG_102432xTY1-T2A-nlsGFP-3xFLAGyeslethality rescued to viable adultsKhc[8] ( BL-1607 ) ; Khc[1ts] ( BL-31994 ) LanB12ndfTRG_6812xTY1-sGFP-V5-preTEV-BLRP-3XFLAGyeslethality rescued to viable adultsLanB1 [KG03456] ( BL-13957 ) ; Df ( 2L ) Exel7032 ( BL-7806 ) multiple ankyrin repeats single KH domain ( mask ) 3rdfTRG_4862xTY1-sGFP-V5-preTEV-BLRP-3xFLAGyeslethality rescued to viable adultsmask[10 . 22]/ Df ( 3R ) BSC317Barry Thompson , pers . commmidline ( mid ) 2ndfTRG_4902xTY1-sGFP-V5-preTEV-BLRP-3xFLAGnolethality not rescuedmid[B1295]; mid[C2372]numb2ndfTRG_252xTY1-sGFP-V5-preTEV-BLRP-3xFLAGyeslethality rescued to viable adultsnumb[1] ( BL-4096 ) ; Df ( 2L ) 30A-C ( BL-3702 ) odd skipped ( odd ) 2ndfTRG_472xTY1-sGFP-V5-preTEV-BLRP-3xFLAGnolethality not rescuedodd[5] ( BL-5345 ) ; Df ( 2L ) Exel7018 ( BL-7789 ) optomotor-blind-related-gene-1 ( org-1 ) XfTRG_4852xTY1-sGFP-V5-preTEV-BLRP-3xFLAGnomale lethality not rescuedorg-1[OJ487]oskar ( osk ) 3rdfTRG_13942XTY1-SGFP-V5-preTEV-BLRP-3XFLAGyesfemale sterility entirely rescuedosk[A87]/ Df ( 3R ) p-XT103Pabp22ndfTRG_5652xTY1-sGFP-V5-preTEV-BLRP-3XFLAGnolethality not rescuedPabp2[01] ( BL-9838 ) ; Pabp2[55] ( BL-38390 ) patched ( ptc ) 2ndfTRG_822xTY1-sGFP-V5-preTEV-BLRP-3xFLAGyeslethality rescued to viable adultsptc[9] ( BL-3377 ) ; ptc[16] ( BL-35500 ) retina abarrent in pattern ( rap ) XfTRG_12532xTY1-sGFP-V5-preTEV-BLRP-3XFLAGyeslethality rescued to viable adultsrap[ie28]Yuu Kimata , pers . comm . rhea ( Talin ) 3rdfTRG_5872xTY1-sGFP-V5-preTEV-BLRP-3XFLAGyesembryonic lethality rescued to viable adultsrhea[1]; rhea[79]RhoGEF22ndfTRG_5912xTY1-sGFP-V5-preTEV-BLRP-3XFLAGyesembryonic lethality rescued to viable adultsRhoGEF2 [04291]Jörg Großhans , pers . comm . roundabout ( robo ) 2ndfTRG_5672xTY1-sGFP-V5-preTEV-BLRP-3XFLAGnolethality not rescuedrobo[1] ( BL-8755 ) ; robo[2] ( BL-8756 ) saxophone ( sax ) 2ndfTRG_100702xTY1-sGFP-3xFLAGyeslethality rescued to viable adultssax[4] ( BL-5404 ) ; sax[5] ( BL-8785 ) scribbler ( sbb ) 2ndfTRG_4432xTY1-sGFP-V5-preTEV-BLRP-3xFLAGnolethality not rescuedsbb[04440] ( BL-11376 ) ; Df ( 2R ) BSC334 ( BL-24358 ) Sin3A2ndfTRG_5962xTY1-sGFP-V5-preTEV-BLRP-3XFLAGnolethality not rescuedSin3A[08269] ( BL-12350 ) ; Sin3A [B0948]smoothened ( smo ) 2ndfTRG_5992xTY1-sGFP-V5-preTEV-BLRP-3XFLAGyeslethality rescued to viable adultssmo[3] ( BL-3277 ) ; smo[119B6] ( BL-24772 ) snail ( sna ) 2ndfTRG_712xTY1-sGFP-V5-preTEV-BLRP-3xFLAGnolethality not rescuedsna[18] ( BL-2311 ) ; sna[1] ( BL-25127 ) spalt major ( salm ) 2ndfTRG_1652xTY1-sGFP-V5-preTEV-BLRP-3xFLAGnolethality not rescuedsalm[1] ( BL-3274 ) ; Df ( 2L ) 32FP-5 ( BL-29717 ) Target of rapamycin ( Tor ) 2ndfTRG_7132xTY1-sGFP-V5-preTEV-BLRP-3XFLAGnolethality not rescuedTor[deltaP] ( BL-7014 ) ; Df ( 2L ) Exel7055 ( BL-7823 ) traffic jam ( tj ) 2ndfTRG_1632xTY1-sGFP-V5-preTEV-BLRP-3xFLAGnosterility not rescuedtj[PL3] ( BL-4987 ) ; Df ( 2L ) Exel8041 ( BL-7849 ) viking ( vkg ) 2ndfTRG_5952xTY1-sGFP-V5-preTEV-BLRP-3XFLAGnolethality not rescuedvkg[01209] ( BL-11003 ) ; Df ( 2L ) Exel7022 ( BL-7794 ) Unc-89/ Obscurin2ndfTRG_10462xTY1-sGFP-V5-preTEV-BLRP-3XFLAGyesflightlessness rescuedUnc-89[EY15484]yorkie ( yki ) 2ndfTRG_8752xTY1-sGFP-V5-preTEV-BLRP-3XFLAGyeslethality rescued to viable adultsyki[B5]Barry Thompson , pers . comm . To demonstrate the broad application spectrum of our fly TransgeneOme library , we analysed tagged protein expression and subcellular localisation in multiple tissues at various developmental stages . Germline expression in flies differs substantially from somatic expression , requiring particular basal promoters and often specialised 3’UTRs ( Ni et al . , 2011; Rørth , 1998 ) . Therefore , we used ovaries to test the fTRG library and probed the expression of 115 randomly selected lines in germline cells versus somatic cells during oogenesis ( Figure 3A ) . From the 115 lines , 91 ( 79% ) showed detectable expression during oogenesis , with 45 lines being expressed in both , germ cells and the somatic epithelial cells ( Figure 3B , C and Supplementary file 3 ) . 76 ( 66% ) fTRG lines showed interesting expression patterns restricted to subsets of cells or to a subcellular compartment ( Figure 3B-D ) . For example , Tan-GFP is expressed in germline stem cells only , whereas the ECM protein Pericardin ( Prc-GFP ) is concentrated around the neighbouring cap cells , and the transcription factor Delilah ( Dei-GFP ) is specifically localised to the nuclei of somatic stem cells , which will give rise to the epithelial cells surrounding each egg chamber ( Figure 3A , C ) . In early egg chambers , Reph-GFP is expressed in germ cells only , whereas the ECM protein Viking ( Vkg-GFP ) specifically surrounds all the somatic epithelial cells . Interestingly , the transcription factor Auracan ( Ara-GFP ) is only expressed in posterior follicle cells , whereas the putative retinal transporter CG5958 is only detectable in the squamous epithelial cells surrounding the nurse cells ( Figure 3C ) . 10 . 7554/eLife . 12068 . 008Figure 3 . Expression of fTRG tagged proteins in ovaries . ( A ) Schematic overview of oogenesis stages and cell types . ( B ) Summary of the identified expression patterns; see also Supplementary file 3 . ( C ) Selected examples for cell type specific fTRG expression patterns at germarium , early- and mid-oogenesis stages visualised by anti-GFP antibody staining . ( D ) Selected examples of subcellular localisation patterns , highlighting nuclear , cortical and cytoplasmic patterns at different oogenesis stages . GFP is show in green , DAPI in magenta; scale bars indicate 30 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 12068 . 00810 . 7554/eLife . 12068 . 009Figure 3—figure supplement 1 . Co-localisation of fTRG derived tagged proteins with endogenous proteins during oogenesis . ( A , B ) Endogenous untagged Grk protein detected with an anti-Grk antibody localises similarly in wild-type stage 8 oocytes ( A ) , as in fTRG960 oocytes expressing Grk-GFP detected by an anti-GFP antibody ( B ) . Note that both Grk and GFP antibody patterns are indistinguishable ( compare B and B’ ) . ( C , D ) Osk protein detected by an anti-Osk antibody in wild type stage 9–10 oocytes ( C ) compared to Osk antibody labelled protein in an fTRG1394 oocyte expressing Osk-GFP in addition to endogenous Osk ( D ) . Note the co-localisation of anti-Osk and anti-GFP antibodies ( D and D’ ) . Scale bars indicate 30 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 12068 . 00910 . 7554/eLife . 12068 . 010Figure 3—figure supplement 2 . Posttranscriptional regulation of protein expression during oogenesis . ( A , B ) osk-GFP mRNA visualised by an anti-GFP labelled RNA probe ( yellow , DAPI in magenta ) at stage 6 and stage 10 of oogenesis . ( A’ , B’ ) Osk-GFP protein visualised by anti-GFP antibody ( green , DAPI in magenta ) at stage 6 and stage 10 . Note that Osk-GFP protein is not detectable at stage 6 . ( C , D ) corolla-GFP mRNA ( yellow , DAPI in magenta ) at stage 6 and stage 8 . ( E , F ) Corolla-GFP protein ( green , DAPI in magenta ) at stage 6 and stage 8 . Note that Corolla-GFP protein is only detectable at stage 6 but not stage 8 . Scale bars indicate 30 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 12068 . 010 We further investigated the subcellular localisation of the tagged proteins , which revealed a localisation for the RNA helicase l ( 2 ) 35Df to all nuclei , whereas the predicted C2H2-Zn-finger transcription factor Crooked legs ( Crol-GFP ) is restricted to the nuclei of the epithelial cells ( Figure 3D ) . Interestingly , Corolla-GFP is exclusively localised to the oocyte nucleus in early egg chambers . This is consistent with the function of Corolla at the synaptonemal complex attaching homologous chromosomes during early meiosis ( Collins et al . , 2014 ) . In contrast , the uncharacterised homeobox transcription factor E5 ( E5-GFP ) is largely restricted to the nuclei of anterior and posterior epithelial cells ( Figure 3D ) . Apart from nuclear patterns , we found a significant number of cortical localisations , including the well characterised Crumbs ( Crb-GFP ) ( Bulgakova and Knust , 2009 ) and the PDZ-domain containing Big bang ( Bbg-GFP ) ( Bonnay et al . , 2013 ) at the apical cortex of the epithelial cells , the Na+/K+ transporter subunit Nervana 2 ( Nrv2-GFP ) at the lateral epithelial membrane , and the EGF-signalling regulator Star ( S-GFP ) as well as the TGF-β receptor Saxophone ( Sax-GFP ) localised to the cortex or membrane of the germ cells ( Figure 3D and Supplementary file 3 ) . Furthermore , we find a perinuclear enrichment for the uncharacterised predicted NAD-binding protein CG8768 , and oocyte enrichments for the Tom22 homolog Maggie ( Mge-GFP ) ( Vaskova et al . , 2000 ) , the glycosyltransferase Wollknäuel ( Wol-GFP ) ( Haecker et al . , 2008 ) and the TGF-α homolog Gurken ( Grk-GFP ) , the latter with its well-established concentration around the oocyte nucleus ( Neuman-Silberberg and Schüpbach , 1993 ) , where it co-localises with endogenous Grk protein ( Figure 3D and Figure 3—figure supplement 1A , B ) . We did not perform genetic rescue assays for all of lines examined for protein localisation . However , even tagged proteins , for which the rescue assay failed , such as Vkg-GFP , can result in an informative expression pattern ( Figure 3D , Table 2 ) . Obviously , an independent validation of expression patterns established solely based on a tagged transgenic reporters is advisable . To test if genes expressed from the FlyFos system also undergo normal post-transcriptional regulation , we analysed the osk-GFP line , which was recently used as a label for germ granules ( Trcek et al . , 2015 ) . osk mRNA is transcribed from the early stages of oogenesis onwards in the nurse cell nuclei and specifically transported to the oocyte , where it localises to the posterior pole ( Ephrussi et al . , 1991; Kim-Ha et al . , 1991 ) . Only after the mRNA is localised , it is translated from stage 9 onwards ( Kim-Ha et al . , 1995 ) . Indeed , fosmid derived osk-GFP mRNA localises normally during all stages of oogenesis and its translation is repressed during mRNA transport , as Osk-GFP protein can only be detected at the posterior pole from stage 9 onwards ( Figure 3—figure supplement 2A , B ) . osk-GFP also rescues all aspects of an osk null allele ( Figure 2B , C ) and Osk-GFP colocalises with endogenous Osk protein ( Figure 3—figure supplement 1C , D ) . Additionally , we discovered a post-transcriptional regulation for corolla . corolla-GFP mRNA localises to the oocyte at stage 6 and Corolla-GFP protein is transported into the oocyte nucleus . However , despite the presence of the corolla-GFP mRNA at stage 8 , Corolla protein is undetectable , suggesting either a translational block of the RNA or targeted degradation of the protein ( Figure 3—figure supplement 2C-F ) . Taken together , these expression and protein localisation data recapitulate known patterns accurately and identify various unknown protein localisations in various cell types during oogenesis , and thus emphasise the value of the fly TransgeneOme resource . For many genes , the expression patterns at the mRNA level are particularly well characterised during Drosophila embryogenesis ( Hammonds et al . , 2013; Tomancak et al . , 2002; 2007 ) . However , in situ hybridisation techniques on fixed tissues do not visualise the dynamics of expression over time and thus do not allow tracking of the expressing cells during development . As our tagging approach enables live imaging at endogenous expression levels , we set out to test if in toto imaging using the SPIM ( Selective Plane Illumination Microscopy ) technology ( Huisken et al . , 2004 ) can be applied to the fly TransgeneOme lines . We pre-screened a small subset of lines ( Table 3 ) and selected the Na+/K+ transporter subunit Nrv2 , as it shows high expression levels , for long-term time-lapse live imaging with a multi-view dual-side SPIM ( Huisken and Stainier , 2007 ) . During embryogenesis Nrv2 expression was reported in neurons ( Sun et al . , 1999 ) and glial cells ( Younossi-Hartenstein et al . , 2002 ) . Interestingly , we find that Nrv2-GFP is already expressed from stage 11 onwards in most likely all cell types , where it localises to the plasma membrane ( Figure 4A-C ) , similarly to its localisation in ovaries ( Figure 3D ) . The expression level increases during stage 15 in all cells , with a particularly strong increase in the developing central nervous system ( CNS ) labelling the CNS and motor neuron membranes ( Figure 4 , Video 1 , Examine raw data in BigDataViewer ( Fiji -> Plugins -> BigDataViewer -> Browse BigDataServer and enter http: //bds . mpi-cbg . de:8087 ) . These live in toto expression data are consistent with expression data of a recently isolated GFP trap in nrv2 ( Lowe et al . , 2014 ) , thus validating our methodology . 10 . 7554/eLife . 12068 . 011Table 3 . in totoSPIM imaging of fTRG lines in the embryo Table listing the fTRG lines that were imaged in the embryo using Zeiss Lightsheet Z . 1 from multiple angles over time . nrv2 , gsb and gsb-n are discussed in the text . For the remaining lines , we list broad categorisation of the expression detected by SPIM imaging . DOI: http://dx . doi . org/10 . 7554/eLife . 12068 . 011fTRG numberFBgn_idGene symbolSignalEmbryonic expressionMovieBeads58FBgn0001148gsbstrongtissue-specific expressionYesYes71FBgn0003448snailweaktissue-specific expressionYesYes88FBgn0025360Optixmediumtissue-specific expressionYesYes94FBgn0010433atoweaktissue-specific expressionYesYes137FBgn0259685crbmediumtissue-specific expressionYesYes155FBgn0029123SoxNstrongtissue-specific expressionYesYes349FBgn0024294spn43Aastronglate expression , deposited in the cuticleYesYes513FBgn0001147gsb-nmediumtissue-specific expressionYesYes937FBgn0015777nrv2strongubiquitous expression , membrane signalYesYes10 . 7554/eLife . 12068 . 012Figure 4 . Live in toto imaging during embryogenesis with SPIM . ( A-C ) Nrv2-GFP protein is enriched in cell membrane of the epidermis and the CNS of late stage 16 embryos , as shown by a lateral section ( A ) and high magnifications of the posterior epidermis ( B ) and the ventral CNS ( C ) . ( D ) Schemes of the lateral , ventral and transverse optical section views through the embryo shown in ( E-I ) . ( E-I ) Still image from a Nrv2-GFP time-lapse Video with lateral section views on the left , ventral sections in the middle and transverse sections on the right . Note that Nrv2-GFP is first expressed in the developing epidermal epithelial cells ( E , F ) and then becomes enriched in the CNS ( G-I , see Video 1 ) . Scale bars indicate 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 12068 . 01210 . 7554/eLife . 12068 . 013Figure 4—figure supplement 1 . Live Gsb-GFP imaging during embryogenesis with SPIM . ( A , B ) Lateral view of a live stage 10 embryo expressing Gsb-GFP ( green in A ) and Histone2A-mRFPruby ( red in A ) ; anterior is to the left ( see Video 2 ) . ( C-F ) Confocal sections of gsb-positive deuterocerebral proneural domain ( recorded with a regular confocal microscope at a similar position as boxed in B ) . Fate of cells is symbolized by different colours ( blue: epidermal precursor undergoing no further mitosis; purple: epidermal precursor undergoing one mitosis; red: neuroblasts ) . ( C ) and ( D ) show an optical section through the neurectoderm of stage 10 embryo prior to neuroblast delamination; ( E ) and ( F ) display the same region 30 min later , after neuroblasts have delaminated ( stage 11 ) , with a superficial optical section of surface ectoderm ( E ) , and deep section of the neuroblast layer ( F ) . ( G-I ) Optical cross sections ( rotated by 90 degrees ) of a similar embryo as in ( A ) expressing GFP-tagged Gsb ( green ) and Histone-2A-mRFPruby ( red ) at stage 10 ( G ) , early 11 ( H ) and late 11 ( I ) showing neuroblasts delaminating from Gsb-GFP domain . ( J ) Schematic cross section of stage 10 ( left ) and stage 11 ( right ) ectoderm illustrating fate of cells forming part of Gsb-positive pro-neural domain . Scale bars indicate 25 µm ( A , B ) and 10 µm ( G-I ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12068 . 01310 . 7554/eLife . 12068 . 014Figure 4—figure supplement 2 . Live Gsb-n-GFP imaging during embryogenesis with SPIM . ( A , B ) . Ventral view of Gsb-n-GFP expression of a stage 12 embryo during germ-band retraction ( A ) and stage 14 during head involution ( B ) . Note that Gsb-n-GFP remains expressed in neuronal precursors during stage 14 ( Video 3 ) . Scale bars indicate 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 12068 . 014 We wanted to extend our approach beyond highly expressed structural genes towards transcription factors that enable to follow cell lineages in the embryo . For this purpose , we crossed the fTRG line of the homeobox transcription factor gooseberry ( Gsb-GFP ) to H2A-mRFPruby , which labels all nuclei , and recorded a two-colour multi-view dual-side SPIM Video . We find that Gsb-GFP is expressed in the presumptive neuroectoderm of the head region , labelling segmentally reiterated stripe-like domains at stage 10 ( Figure 4—figure supplement 1A , B , Video 2 ) as was described from fixed images ( Gutjahr et al . , 1993 ) . Focusing on the deuterocerebral domain , we could reconstruct the delamination of two neuroblasts , which up-regulate Gsb-GFP while initiating their asymmetric divisions ( Figure 4—figure supplement 1C-F ) . It was possible to individually follow their neural progeny . Gsb-GFP expression also allowed us to directly follow the gradual down-regulation of Gsb-GFP in ectodermal cells that remained at the head surface after neuroblast delamination . As opposed to the neuroblasts , these cells , which give rise to epidermis , did not divide at all , or underwent only one further division ( Figure 4—figure supplement 1G-J , Video 2 ) . 10 . 7554/eLife . 12068 . 015Video 1 . Multi-view SPIM Video of a stage 12 Nrv2-GFP expressing embryo . A stack was acquired every 15 min , lateral , dorsal , ventral and transverse views of the same time points are displayed . From stage 11 onwards Nrv2-GFP is present ubiquitously in the plasma membrane . Later , its expression increases in the CNS , particularly in the neuropil and the motor neurons . Video plays with 7 frames per second . Time is given in hh:mm . Scale bar indicates 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 12068 . 01510 . 7554/eLife . 12068 . 016Video 2 . Lateral head section from a SPIM Video of a stage 10 Gsb-GFP ( green , white in the top Video ) , Histone-2A-mRFPruby ( red ) embryo . A stack was acquired every 7 min . The segmentally re-iterated stripe-like gsb expression domain in the head neuroectoderm is visible . Later , gsb is expressed in ganglion mother cells and nerve cells that are the progeny of gsb expressing neuroblasts . Video plays with 7 frames per second . Time is given in hh:mm . Scale bar indicates 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 12068 . 01610 . 7554/eLife . 12068 . 017Video 3 . Ventral view of a SPIM Video of a stage 6 Gsb-n-GFP embryo . A stack was acquired every 15 min . Gsb-n-GFP is only detectable at the end of germ-band extension . During germ-band retraction , it is expressed in characteristic L-shaped expression domains in the hemi-segments of the trunk . In the late stage embryo , Gsb-n-GFP is present in the neurons of the shortening ventral nerve cord . Video plays with 7 frames per second . Time is given in hh:mm . Scale bar indicates 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 12068 . 017 gsb is in part required for gooseberry-neuro ( gsb-n; also called gsb-d ) expression ( He and Noll , 2013 ) . Notably , the expression of Gsb-n-GFP ( fTRG line 513 ) becomes detectable only at the end of germ-band extension ( end of stage 11 ) in the developing CNS , where it lasts until stage 17 ( Figure 4—figure supplement 2 , Video 3 ) . Again , this is consistent with published immuno-histochemistry data ( Gutjahr et al . , 1993; He and Noll , 2013 ) . We conclude that our fly TransgeneOme library can be used for live in toto imaging , even for transcription factors expressed at endogenous levels . This will be of significant importance for on-going efforts linking the transcription factor expression patterns of embryonic neuroblasts to the morphologically defined lineages that structure the larval and adult Drosophila brain ( Hartenstein et al . , 2015; Lovick et al . , 2013; Pereanu , 2006 ) . Cells and tissues in the embryo are not yet terminally differentiated . To apply our TransgeneOme library to fully differentiated tissues , we decided to study tissues of the adult thorax . We scored expression in the large indirect flight muscles ( IFMs ) , in leg muscles , in the visceral muscles surrounding the gut , in the gut epithelium , the tendon epithelium , the trachea and the ventral nerve cord including the motor neurons . In total , we found detectable expression in at least one tissue for 101 of 121 ( 83 . 5% ) analysed fTRG lines , thus creating a large number of valuable markers for cell types and subcellular structures ( Table 4 , Supplementary file 4 ) . 10 . 7554/eLife . 12068 . 018Table 4 . Summary of adult muscle fTRG expression patterns 54 detected adult muscle localisation patterns ( flight muscle , leg muscle and visceral muscle ) from Supplementary file 4 are summarised . fTRG line number is listed in brackets . DOI: http://dx . doi . org/10 . 7554/eLife . 12068 . 018Thick filamentThin filament / myofibrilM-lineZ-discMuscle attachment siteT-tubules / sarcolemmaDotty pattern / vesicles ( ? ) MitochondriaNucleusNeuro-muscular junctionFln ( 876 , IFM ) Act88F ( 78 , 10028 , IFM ) Prm ( 475 , IFM ) CG31772 ( 63 ) Ilk ( 483 ) Dlg1 ( 502 ) Babo ( 444 ) CG12118 ( 276 ) Adar ( 570 ) Cact ( 516 ) Mf ( Iso-A , G , N , 501 ) Fray ( 125 , 10032 ) Unc-89 ( 1046 ) Kettin ( Sls-Isoform , 569 ) Talin ( rhea , Iso-B , E , F , G , 587 ) Sax ( 10070 ) CG5885 ( 10017 , leg m . ) Blimp-1 ( 10149 ) Veli ( 10125 ) Mhc ( Iso-K , L , M , 500 ) Hsp83 ( 10010 ) Lmpt ( 584 , I-band , leg m . ) β-PS Integrin ( mys , 932 ) CLIP-190 ( 156 ) CG11617 ( 10041 ) Mhc ( Iso-A , F , 519 , leg m . subset & visceral . m . ) TpnC25D ( 1257 , leg m . & visceral m . ) Mask ( 486 , IFM ) Dlg5 ( CG6509 , 10251 , IFM ) CG12391 ( 10036 ) Prm ( 475 , leg m . & visceral m . ) TpnI ( wupA , 925 , leg m . & visceral m . ) Mlp60A ( 709 , leg m . & visceral m . ) Hts ( 585 ) CG17912 ( 10035 ) Mlp84B ( 678 , leg m . & visceral m . ) Mask ( 486 , leg m . ) CG32121 ( 92 ) Talin ( Iso-C , D , 731 , leg m . & visceral m . ) Pyd3 ( 53 ) Dorsal ( 29 , leg m . ) Rho1 ( 31 ) E2F2 ( 115 ) Sc2 ( 79 , 10039 ) Gro ( 21 ) Tango11 ( 699 ) Hand ( 10163 , visceral m . ) Tsc1 ( 59 ) Hb ( 139 , leg m . ) Vps35 ( CG5625 , 57 ) Mnt ( 34 ) P53 ( 84 ) Salm ( 165 ) Vri ( 182 ) The large IFMs are fibrillar muscles , which have a distinct transcriptional program resulting in their distinct morphology ( Schönbauer et al . , 2011 ) . This is recapitulated by the expression of Act88F-GFP , which localises to the thin filaments of IFMs only ( Figure 5A-C ) , whereas Mlp84B-GFP is not expressed in IFMs but at the peripheral Z-discs of leg and visceral muscles only ( Figure 5D-F ) , similar to the published localisation in larval muscle ( Clark et al . , 2007 ) . We find various dotty patterns indicating localisation to intracellular vesicles; a particularly prominent example is Tango1-GFP in the midgut epithelium ( Figure 5G , H , Supplementary file 4 ) . Tango1 regulates protein secretion in S2 cells , where it localises to the Golgi apparatus upon over-expression ( Bard et al . , 2006 ) , suggesting that the pattern described here is correct . We find Par6 with an in analogy to other epithelia expected apical localisation ( Hutterer et al . , 2004 ) in the epithelium of the proventriculus and in trachea ( Figure 5I , J ) , whereas we identified a surprising pattern for the TRP channel Painless ( Tracey et al . , 2003 ) . Pain-GFP is not only highly expressed in motor neurons ( Figure 5K ) but also in a particular set of small cells within the gut epithelium and most surprisingly , in the tendon cells to which the IFMs attach ( Figure 5L , M ) . To clarify the identity of the Pain-GFP positive cell type in the gut , we co-stained with the differentiated enteroendocrine cell marker Prospero ( Ohlstein and Spradling , 2005 ) , however , did not find any overlap with the Pain-GFP positive cells ( Figure 5Q ) . This suggests that the small , likely diploid , Pain-GFP positive cells are either enteroblasts or intestinal stem cells ( ISCs ) , the source of all enterocytes and enteroendocrine cells that build the gut epithelium ( Jiang and Edgar , 2011 ) . At this point , we can only speculate that Pain might be involved in mechanical stretch-sensing in these cell types . We have also tagged various ECM components , with LamininB1 ( LanB1-GFP ) , LamininA ( LanA-GFP ) and BM40-SPARC resulting in the most prominent expression patterns . All three ensheath most adult tissues , particularly the muscles ( Figure 5N , O , Figure 5—figure supplement 1A , B , E , F ) . Interestingly , LanA-GFP and LanB1-GFP also surround the fine tracheal branches that penetrate into the IFMs , whereas BM40-SPARC is only detected around the large tracheal stalk and the motor neurons ( Figure 5P , Figure 5—figure supplement 1C , D , G , H ) . Finally , we also detected prominent neuromuscular junction ( NMJ ) markers; the IκB homolog Cactus shows a distinct pattern on leg muscles , visceral muscles and IFMs , the latter we could confirm by co-staining with the neuronal marker Futsch ( Figure 5—figure supplement 1I-M ) . Interestingly , such a NMJ pattern for Cactus and its binding partner Dorsal has been shown in larval body muscle by antibody stainings ( Bolatto et al . , 2003 ) . Together , these results suggest that our fly TransgeneOme library provides a rich resource for tissue-specific markers in the adult fly that can routinely be used to visualise subcellular compartments in various tissues . 10 . 7554/eLife . 12068 . 019Figure 5 . Expression of fTRG tagged proteins in tissues of the adult thorax . Antibody stainings of the adult thorax with anti-GFP antibody ( green ) and phalloidin ( red ) . ( A-F ) Act88-GFP expression is specific to the IFMs , where it labels the thin filaments ( B ) , whereas Mlp84B specifically labels the Z-discs of leg muscles ( F ) . ( G-J ) Tango1-GFP concentrates in a vesicle-like pattern in the gut epithelium ( H , H' ) , whereas Par6-GFP is highly expressed in trachea ( I ) and the gut epithelium , where it concentrates at the apical membrane , as shown for a cross-section of the proventriculus ( J , J' ) , nuclei are labelled with DAPI ( blue ) . ( K-M ) Pain-GFP expression in the flight muscle motor neurons ( K ) , small cells within the midgut epithelium ( L , L' ) and tendon cells ( M , M' ) . ( N-P ) LanB1-GFP labels the extracellular matrix surrounding the IFMs , the motor neurons and the trachea ( N ) , as well as the visceral muscles ( O ) . Even the finest trachea marked by UV auto-fluorescence ( white ) ( P ) are surrounded by LanB1-GFP ( P’ ) . ( Q ) Pain-GFP positive cells in the midgut do not overlap with Prospero positive nuclei of enterocytes ( in red ) and contain small nuclei , as visualised by DAPI co-stain in white ( Q-Q’’’ ) . Scale bars indicate 100 µm ( A , D , I , K , N ) , 20 µm ( H , J , L , O , Q ) and 5 µm ( B , C , E , F , M , P ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12068 . 01910 . 7554/eLife . 12068 . 020Figure 5—figure supplement 1 . Extracellular matrix and synaptic markers of the adult thorax . ( A-H ) Antibody stainings of the adult thorax with anti-GFP antibody ( green or white in the single colour images ) and phalloidin ( red ) . LanA-GFP and BM-40-SPARC-GFP labels the ECM around motor neurons , visceral muscle and trachea . Note that thin trachea within the IFMs ( marked by UV auto-fluorescence in white ) are surrounded by LanA-GFP but not BM-40-SPARC-GFP ( D , H ) . ( I-M ) Cact-GFP ( green ) shows a distinct localisation in IFMs , leg and visceral muscle reminiscent of a neuromuscular junction pattern . Note the partial co-localisation with the motor neuron marker Futsch ( in red , M , M’ ) , whereas no co-localisation with trachea in IFMs ( in white , M ) . Scale bars indicate 100 µm ( A , E , I ) , 20 µm ( B , C , F , G , L ) , 10 µm ( M ) and 5 µm ( D , H , J , K ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12068 . 020 To further validate the advantages of our TransgeneOme lines to label subcellular structures , we imaged the large IFMs of the same 121 lines at high resolution . We found various markers for the thick filaments ( e . g . the myosin-associated protein Flightin , Fln-GFP ) ( Vigoreaux et al . , 1993 ) , for the myofibrils ( e . g . the protein kinase Fray-GFP ) , the M-lines ( e . g . the titin related protein Unc-89/Obscurin-GFP ) ( Katzemich et al . , 2012 ) , the Z-discs ( e . g . CG31772-GFP ) and the muscle attachment sites ( e . g . Integrin-linked-kinase , Ilk-GFP ) . Furthermore , we identified markers for the T-tubules ( e . g . Dlg1-GFP ) , for different vesicular compartments ( e . g . the TGFβ receptor Baboon-GFP ) and for mitochondria ( CG12118 ) within the IFMs ( Figure 6 , Table 4 , Supplementary file 4 ) . The latter was confirmed by co-labelling the mitochondria with an antibody against the mitochondrial ATPase ( complex V α subunit ) ( Cox and Spradling , 2009 ) ( Figure 6—figure supplement 1 ) . Additionally , we documented the nuclear localisation in IFMs and leg muscles for a variety of fTRG proteins , including the uncharacterised homeodomain protein CG11617 and the C2H2 Zinc-fingers CG12391 and CG17912 ( Figure 6—figure supplement 2A-C , E-G ) ; both of the latter result in flightless animals when knocked-down by muscle-specific RNAi ( Schnorrer et al . , 2010 ) suggesting that these genes play an essential role for IFM morphogenesis or function . Interestingly , the well characterised C2H2 Zinc-finger protein Hunchback ( Hb ) is only localised to leg muscle nuclei , but absent from IFMs suggesting a leg muscle-specific function of Hb ( Figure 6—figure supplement 2G , H ) . 10 . 7554/eLife . 12068 . 021Figure 6 . Subcellular expression patterns in adult flight muscles . Antibody stainings of the adult thorax with anti-GFP antibody ( green or white in the single colour images ) and phalloidin ( red ) . ( A-D ) Localisation to specific myofibrillar sub-regions; Fln-GFP marks the thick filaments ( A , A’ ) , Fray-GFP surrounds the myofibrils with an enrichment at M-lines and Z-discs ( B , B’ ) , Unc-89-GFP marks only M-lines ( C , C’ ) and CG31772-GFP only Z-discs ( D , D’ ) . ( E-H ) Ilk-GFP strongly concentrates at the muscle-tendon attachment sites ( E , E’ ) , Dlg1-GFP labels the T-tubular membranes ( F , F’ ) , Babo-GFP shows a dotty , vesicular pattern ( G , G’ ) and CG12118-GFP displays a mitochondrial pattern ( H , H’ ) . Scale bars indicate 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 12068 . 02110 . 7554/eLife . 12068 . 022Figure 6—figure supplement 1 . CG12118-GFP localises to mitochondria . Co-staining of adult IFMs from fTRG276 ( CG12118-GFP ) with anti-GFP antibody ( green ) and the mitochondrial marker anti-complex V α ( ATPase Vα , in red ) . Note the strong overlap of both signals . Scale bar indicate 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 12068 . 02210 . 7554/eLife . 12068 . 023Figure 6—figure supplement 2 . Nuclear localisations in adult flight muscles . Antibody stainings of the adult thorax with anti-GFP antibody ( green or white in the single colour images ) , phalloidin ( red ) and DAPI ( blue ) . ( A-H ) CG11617-GFP ( A , E ) , CG12391-GFP ( B , F ) and CG17912 ( C , G ) are localised to the nuclei of IFMs and leg muscles , whereas Hb-GFP is only found in leg muscle nuclei ( H ) and not detectable in IFM nuclei ( D ) . Scale bars indicate 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 12068 . 02310 . 7554/eLife . 12068 . 024Figure 6—figure supplement 3 . Alternative splicing into alternative C-termini . Antibody stainings of the adult thorax with anti-GFP antibody ( green or white in the single colour images ) and phalloidin ( red ) . ( A , H ) Gene models of the 3’ end of Mhc ( A ) and rhea ( H ) listing the predicted isoforms; coding exons are shown in pink , 3’-UTRs in yellow boxes . The positions of the GFP tag insertions are marked by green arrows . ( B-G ) The shorter Mhc-GFP isoforms ( Iso K , L , M ) are expressed in IFMs and all leg muscles ( B-D ) , whereas the slightly longer Mhc-GFP isoforms ( Iso A , F , G etc . ) are not detectable in IFMs but present in visceral muscles and a subset of leg muscles ( E-G ) . ( I-N ) The shorter Talin-GFP isoforms ( rhea Iso C , D ) are not detectable at muscle-tendon attachment sites in IFMs ( J , arrowheads ) and leg muscles ( K , arrowhead ) , however do localise to costamers of leg muscles ( K ) . However , the long Talin-GFP isoforms ( rhea Iso B , E , F , G ) do localise to muscle-tendon attachment sites in IFMs ( M ) and leg muscles ( N ) . Scale bars indicate 100 µm ( B , E , I , L ) , 10µm ( G ) and 5 µm ( C , D , F , J , K , M , N ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12068 . 024 However , differences between muscle types are not only controlled transcriptionally but also by alternative splicing ( Oas et al . , 2014; Spletter and Schnorrer , 2014; Spletter et al . , 2015 ) . To investigate if our tagging approach can be used to generate isoform-specific lines , we have chosen two prominent muscle genes , mhc and rhea ( the fly Talin ) , both of which have predicted isoforms with different C-termini ( Figure 6—figure supplement 3A , H ) . Interestingly , we found that Mhc-isoforms K , L , M are expressed in IFMs and all leg muscles , however the predicted Mhc-isoforms A , F , G , B , S , V with the distal STOP codon are selectively expressed in visceral muscle and in a subset of leg muscles , however absent in adult IFMs ( Figure 6—figure supplement 3B-G ) . Even more surprisingly , while the long ‘conventional’ rhea ( Talin ) isoforms B , E , F , G show the expected localisation to muscle attachment sites in IFMs and leg muscles ( Weitkunat et al . , 2014 ) , the short Talin-isoforms C and D do not localise to muscle attachment sites , but are selectively concentrated at costamers of leg muscles ( Figure 6—figure supplement 3I-N ) . Hence , the tagged proteins of our TransgeneOme library are ideally suited to label subcellular compartments and protein complexes , and in some cases they can even distinguish between closely related protein isoforms . Tagging of any protein can in principle affect its localisation pattern in a cell . To investigate the reliability of the observed GFP-tagged protein patterns , we selected eight different fTRG lines with specific GFP patterns in adults described above ( LanA , LanB ( not shown ) , Par6 , Mlp84B , Mhc , Fln , Unc89/Obscurin and Dlg1 ) , for which reliable antibodies against the corresponding proteins were readily available . In all cases , we observed a high degree of overlap between the pattern revealed by staining with anti-GFP antibody and the respective specific antibody staining pattern in the same transgenic line . This suggests that the tagged proteins expressed in the fTRG lines and the endogenous proteins co-localised to a large extent ( Figure 7A-D , I-L ) . To rule out that the tagged protein somehow induces the endogenous protein to adopt an abnormal pattern , we compared the localisation patterns in the fTRG lines ( with the protein specific antibody that visualizes both the tagged and untagged endogenous form of the protein ) to a wild type strain , which only expresses the endogenous protein ( Figure 7E-H , M-P ) . Only in one fTRG line , expressing the highly expressed short Mhc isoforms K , L , M-GFP tagged , we found an abnormally broad anti-Mhc pattern in IFMs , however not in leg muscles compared to wild type ( Figure 7D , H , L , P ) . This pattern is explained by an abnormal myofibril morphology in the IFMs of the fTRG500 line , possibly because of an increased Mhc to actin ratio ( 4 copies vs . 2 copies ) , for which IFMs are particularly sensitive ( Cripps et al . , 1994 ) . 10 . 7554/eLife . 12068 . 025Figure 7 . Co-localisation of fTRG tagged proteins with endogenous proteins . Antibody stainings of the adult thorax from fTRG or wild-type lines with anti-GFP antibody ( green or white in the single colour images ) and antibodies against various fly proteins ( red ) . ( A-D ) Co-localisation of LanA-GFP with anti-Laminin antibody stain around the midgut ( A ) , of Par6-GFP with anti-Par6 at the apical side of the proventriculus epithelium ( B ) and of Mlp84B-GFP as well as Mhc-GFP with anti-Mlp84B and anti-Mhc antibody stain in leg muscles , respectively ( C , D ) . ( E-H ) Adult thoraces from wild-type flies show very similar patterns with the respective antibodies . ( I-L ) Adult IFMs showing the co-localisation of Fln-GFP with anti-Fln antibody staining ( I ) , Unc-89/Obscurin-GFP with anti-Obscurin antibody staining ( J ) , Dlg1-GFP with anti-Dlg1 antibody staining ( K ) and Mhc-GFP with anti Mhc antibody staining ( L ) . ( M-P ) The same antibodies result in very similar patterns in wild-type IFMs apart from the a sharp versus a diffuse Mhc pattern comparing wild-type to Mhc-GFP flies ( L , P ) . ( Q ) Western blots loaded with total protein extract from wild-type , Mlp84B-GFP , Fln-GFP and Dlg1-GFP adult males probed with anti-V5 ( included in the GFP tag ) anti-Mlp84B , anti-Fln and anti-Dlg1 antibodies . Note the about 40 kDa size shift of the tagged proteins in the respective lanes ( marked with green arrow heads ) versus the untagged protein band ( black arrow heads ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12068 . 025 Tagging a protein may modify its turn-over rates and thus its expression levels , despite preserving its localisation . To investigate if our tagging approach generally changes protein levels , we chose three different tagged proteins , expressed in different tissues , for which we had functional antibodies and the expected 40 kDa shift caused by the tag should result in a detectable shift compared to the size of the untagged endogenous protein . We made total protein extracts from adults males and run Western blots . When probing with an antibody against the tag , we find the expected sizes for the IFM-specific Fln-GFP , the leg muscle-specific Mlp84B-GFP and the broader expressed Dlg1-GFP , the latter running in several bands due to splice isoforms ( Figure 7Q ) . Probing the same extracts with specific antibodies against the respective proteins shows that both Fln-GFP and Dlg1-GFP levels are comparable to the endogenous protein , whereas Mlp84B-GFP is expressed at a lower level ( Figure 7Q ) . The latter may be caused by the reduced affinity to the thick filament due to the tag resulting in an unstable protein . Together , these data demonstrate that many of the tagged proteins colocalise with the respective endogenous proteins , as has been observed in other large-scale tagging approaches ( Stadler et al . , 2013 ) , however , exact expression levels can be different from the wild-type level in some cases . Thus , the fTRG library is a valuable collection of strains to study in vivo protein localisation . An attractive application of the fly TransgeneOme library is live in vivo imaging . In the past , we had established live imaging of developing flight muscles in the pupal thorax using over-expressed marker proteins ( Weitkunat et al . , 2014 ) . Here , we wanted to test , if live imaging of proteins at endogenous expression levels is also possible within the thick pupal thorax . We selected six fTRG lines for well established genes and indeed could detect expression and subcellular localisation for all of them using a spinning-disc confocal microscope either at the level of the pupal epidermis or below the epidermis , in the developing flight muscles , or both ( Figure 8 ) . The adducin-like Hts-GFP labels the cytoplasm of fusing myoblasts from 10 to 20 hr APF ( after puparium formation ) and the developing SOPs ( sensory organ precursors ) with a particular prominent concentration in developing neurons and their axons ( Figure 8B-E ) . In contrast , Dlg1-GFP localises to cell-cell-junctions of the pupal epidermis and to a network of internal membranes in the developing IFMs ( Figure 8F-I ) that may resemble developing T-Tubules , for which Dlg1 is a well-established marker ( Razzaq et al . , 2001 ) . Interestingly , the long isoforms of Talin-GFP ( rhea isoforms B , E , F , G ) are largely in the cytoplasm and at the cortex of the epidermal cells , with a marked enrichment in the developing SOPs at 10 to 20 hr APF ( Figure 8J , K ) . Further , Talin-GFP is strongly concentrated at muscle attachment sites of developing IFMs from 24 hr onwards ( Figure 8L , M ) consistent with antibody stainings of IFMs ( Weitkunat et al . , 2014 ) . 10 . 7554/eLife . 12068 . 026Figure 8 . Live imaging of fTRG tagged proteins in living pupal thorax . ( A ) Schematic drawing of a 10–12 hr ( left ) and a 30h pupal thorax ( right ) . The developing epidermis is shown in blue , with the SOP precursors in yellow ( developing neurons in red ) , the differentiating tendons are shown in orange , the myoblasts and muscle fibers in green , and the muscle-tendon junction in red . The schematic positions of the optical sections through epithelium and muscles are indicated with blue and green dotted lines , respectively . ( B-Y ) Live imaging of pupal thoraces at the indicated stages acquired with a spinning disc confocal ( except S and T , which were acquired with a two-photon microscope ) . Blue bars above the image indicate epithelial sections and green bars indicate muscle sections ( as explained in A ) . Hts-GFP is expressed in fusing myoblasts ( B , C ) and strongly in developing SOPs ( D , E ) . Dlg1-GFP labels the epithelial junctions ( F ) , internal muscle structures ( green dots , G ) and an unidentified additional developing epithelium ( yellow dots , H , I ) . Talin-GFP is higher expressed in developing SOPs ( J , K ) and strongly localised to the muscle-tendon junction from 24 hr APF ( red arrowheads , L , M ) . LanB1-GFP localises to the basal side of the developing epithelium ( N ) and surrounds the forming muscle fibers ( green dots , O-Q ) with a slight concentration at the muscle-tendon junction at 30 hr APF ( red arrowheads , Q ) . Act88F-GFP weakly labels the developing epithelium , with a slight concentration in the SOPs until 20 hr APF ( R , S ) and very strongly marks the IFMs from 24 hr onwards ( T , U ) . βTub60D-GFPis expressed in the fusing myoblasts ( V , W ) and also labels the microtubule bundles in the developing muscle fibers ( X , Y ) and hair cells of the developing sensory organs ( light blue arrow heads in X ) . Scale bars indicate 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 12068 . 02610 . 7554/eLife . 12068 . 027Figure 8—figure supplement 1 . Live imaging during pupal development . ( A-F ) Stills from live two-photon imaging of an intact 14 hr APF pupa expressing Act88F-GFP strongly labelling the developing IFMs ( see Video 4 ) . ( G-K ) Stills from a two-colour spinning disc Video expressing Act88F-GFP ( green ) and him-GAL4 , UAS-palm-Cherry ( red ) labelling the myoblasts ( see Video 5 ) . Note the sudden green label of single myoblasts after fusion had occurred ( yellow arrowheads , see Video 5 ) . ( L-Q ) Stills from two-photon Video of an intact 14 hr APF pupa expressing βTub-60D-GFP in fusing myoblasts and developing myofibers ( See Video 6 ) . ( R-V ) Stills from a high resolution two-photon Video of an intact 16 hr APF pupa expressing βTub-60D-GFP . Single myoblast during fusion can be resolved ( See Video 7 ) . Strong microtubule bundles ( red arrow heads ) are visible close to the edges of the splitting myotube ( white dashed lines , R ) ; splitting is complete in ( V ) . Scale bars indicate 50 µm ( A-F , L-Q ) and 10 µm ( G-K ) and ( R-V ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12068 . 027 The dynamics of the extracellular matrix is little described thus far as very few live markers existed . Hence , we tested our LamininB1 fosmid and found that LanB1-GFP is readily detectable within the developing basement-membrane basal to the epidermal cells of the pupal thorax at 10 hr APF ( Figure 8N ) . It also labels the assembling basement-membrane around the developing IFMs from 16 to 30 hr APF without a particularly obvious concentration at the muscle attachment sites ( Figure 8O-Q ) . To specifically visualise the developing IFMs we chose Actin88F , which is specifically expressed in IFMs and a few leg muscles ( Nongthomba et al . , 2001 ) . We find that the Act88F-GFP fTRG line indeed very strongly labels the IFMs from about 18 hr APF but is also expressed in the developing pupal epidermis again with an enrichment in the forming SOPs from 10 to 20 hr APF ( Figure 8R-U ) . The latter is not surprising as Act88F-lacZ reporter has been shown to also label the developing wing epithelium ( Nongthomba et al . , 2001 ) , again suggesting that our fTRG line recapitulates the endogenous expression pattern . Finally , we tested the βTub60D fTRG-line , as βTub60D was reported to label the myoblasts and developing myotubes in embryonic and adult muscles ( Fernandes et al . , 2005; Leiss et al . , 1988; Schnorrer et al . , 2007 ) . Indeed , we detect βTub60D-GFP in fusing myoblasts and the developing IFMs , with particularly prominent label of the microtubule bundles at 24h APF ( Figure 8V-Y ) . In addition , βTub60D-GFP also strongly marks the developing hairs of the sensory organs of the pupal epidermis ( Figure 8X , see also Video 6 ) . In order to test , if the fly TransgeneOme lines and the sGFP-tag are indeed suited for long-term live imaging in pupae , we chose Act88F-GFP and βTub60D-GFP and imaged the developing IFMs for more than 19 hr with a two-photon microscope using an established protocol for over-expressed markers ( Weitkunat and Schnorrer , 2014 ) . For both proteins , we can detect strongly increasing expression after 18 hr APF in the developing IFMs , with Act88F-GFP being restricted to the myotubes and the developing myofibrillar bundles ( Video 4 , Figure 8—figure supplement 1A-F ) whereas βTub60D-GFP also labels the fusing myoblasts and is largely incorporated into prominent microtubule bundles ( Video 6 , Figure 8—figure supplement 1L-Q ) . 10 . 7554/eLife . 12068 . 028Video 4 . Z-projection of a two-photon Video of an about 14 hr APF pupa expressing Act88F-GFP . A stack was acquired every 20 min for 19 hr . Expression of Act88F-GFP increases in the indirect flight muscles dramatically , thus contrast was reduced several times in course of the Video to avoid over-exposure . Video plays with 5 frames per second . Time is given in hh:mm . DOI: http://dx . doi . org/10 . 7554/eLife . 12068 . 02810 . 7554/eLife . 12068 . 029Video 5 . Single plane of a spinning disc confocal Video of an about 14 hr APF old pupa expressing Act88F-GFP ( green ) in the flight muscle myotubes and him-GAL4; UAS-palm-Cherry in the myoblasts . An image stack was acquired every two minutes . Note the newly fused myoblasts acquired the GFP label within a single time interval ( highlighted by green arrows ) . Video plays with 5 frames per second . Time is given in minutes . DOI: http://dx . doi . org/10 . 7554/eLife . 12068 . 02910 . 7554/eLife . 12068 . 030Video 6 . Z-projection of a two-photon Video of an about 14 hr APF pupa expressing βTub60D-GFP . A stack was acquired every 20 min for 25 hr . Note the high expression of βTub60D-GFP in fusing myoblasts and the thick microtubules bundles in the developing flight muscles . Hair cells of the developing sensory organs also show strong expression , however , move out of the Z-stack over time . Video plays with 5 frames per second . Time is given in hh:mm . DOI: http://dx . doi . org/10 . 7554/eLife . 12068 . 03010 . 7554/eLife . 12068 . 031Video 7 . Single plane of a two-photon Video of an about 16 hr APF old pupa expressing βTub60D-GFP in myoblasts and the forming flight muscle myotubes . An image stack was acquired every two minutes for more than 3 hr . Note that single myoblasts can be followed during fusion . Most myoblasts fuse at the center of the myotube , which gradually splits into two myotubes . Video plays with 5 frames per second . Time is given in hh:mm . DOI: http://dx . doi . org/10 . 7554/eLife . 12068 . 031 As photo bleaching was no serious problem in these long Videos , we also recorded Videos at higher time and spatial resolution . We labelled the developing IFMs with Act88F-GFP and the myoblasts with a him-GAL4 , UAS-palm-Cherry and acquired a 3D stack every two minutes using a spinning disc-confocal . This enabled us to visualise single myoblast fusion events in developing IFMs of an intact pupa ( Video 5 , Figure 8—figure supplement 1G-K ) . The six dorsal longitudinally oriented IFMs develop from three larval template muscles to which myoblasts fuse to induce their splitting into six myotubes ( Fernandes et al . , 1991 ) . Using high resolution imaging of the βTub60D-GFP line , we find that most myoblasts fuse in the middle of the developing myotube during myotube splitting , with prominent microtubules bundles located at the peripheral cortex of the splitting myotube ( Video 7 , Figure 8—figure supplement 1R ) . These prominent microtubules bundles are then relocated throughout the entire developing myotube ( Video 7 , Figure 8—figure supplement 1S-V ) . Taken together , these live imaging data suggest that many fTRG lines will be well suited for high resolution live imaging of dynamic subcellular protein localisation patterns in developing Drosophila organs . This will strongly expand the set of live markers available for research in flies . For the proper composition , localisation and in vivo function of most protein complexes endogenous expression levels of the individual components are critical ( Rørth et al . , 1998; Tseng and Hariharan , 2002 ) . Hence , the TransgeneOme library would be an ideal experimental set-up to purify protein complexes from different developmental stages using endogenous expression levels of the bait protein . In principle , all the small affinity tags ( TY1 , V5 , FLAG ) ( Figure 1—figure supplement 1 ) can be used for complex purifications . The presence of precision and TEV cleavage sites even allow two-step purifications . For proof of principle experiments , we selected four tagged proteins as baits: Ilk , Dlg1 , Talin and LanB1 , and analysed two different developmental stages . In each case we homogenised hundred 24 to 48 hr pupae and hundred adult flies per experiment and mixed the cleared lysate with a GFP antibody matrix to perform single step affinity enrichment and mass-spec analysis modified from the QUBIC protocol ( Hein et al . , 2015; Hubner et al . , 2010; Keilhauer et al . , 2014 ) . Each affinity-enrichment was performed in triplicate and intensity profiles of all identified proteins were quantified in a label-free format by running all 30 purifications consecutively on the same Orbitrap mass-spectrometer and analysing the data with the MaxQuant software suite ( Cox and Mann , 2008; Cox et al . , 2014 ) ( Supplementary file 5 ) . Interestingly , enriching Ilk-GFP from both , developing pupae and adult flies , recovered the entire Ilk , PINCH , Parvin , RSU-1 complex ( Figure 9 ) , which had previously been purified in vitro from Drosophila S2 cells ( Kadrmas et al . , 2004 ) and mammalian cells ( Dougherty et al . , 2005; Tu et al . , 2001 ) giving us confidence in our methodology . We also successfully enriched Talin-GFP from pupae or adults , however did not identify an obvious strong and specific binding partner ( Figure 9 , Supplementary file 5 ) . In contrast , we identified Mesh as a novel interactor of Dlg1 from pupae and adult flies . Mesh colocalises with Dlg1 at septate junctions of the embryonic Drosophila midgut , however , a molecular interaction of both proteins was not established ( Izumi et al . , 2012 ) . Finally , we purified the laminin complex by pulling on LanB1 , which recovered LanB2 and LanA roughly stoichiometrically , both from pupae and adult flies , as had been found in cell culture experiments ( Fessler et al . , 1987 ) , showing that extracellular matrix complexes can also be purified from in vivo samples with our methodology . In summary , these data demonstrate that interaction proteomics with the fly TransgeneOme library can confirm known interaction partners and discover novel in vivo complex members , making the system attractive for a variety of biochemical applications . 10 . 7554/eLife . 12068 . 032Figure 9 . Proteomics with fTRG bait proteins . GFP-tagged Ilk , Dlg1 , Talin , and LanB1 , respectively , were affinity-enriched from protein extracts generated from whole pupae ( left ) or adult flies ( right ) using anti-GFP immunoprecipitation . A wild-type fly strain not expressing any GFP-tagged protein served as control . Proteins were quantified using mass spectrometry and the MaxLFQ label-free quantification algorithm in MaxQuant . Selected proteins are visualized by their enrichment factors in individual samples over the control ( or simulated noise level , if not detected in the control ) . Specific interaction partners are characterised by the similarity of the quantitative profiles and co-enrichment with the respective bait proteinsDOI: http://dx . doi . org/10 . 7554/eLife . 12068 . 032 The TransgeneOme resource presented here adds a new powerful component to the arsenal of tools available to the Drosophila research community . It complements the genetic resources for gene disruption and localisation ( Buszczak et al . , 2007; Lowe et al . , 2014; Morin et al . , 2001; Nagarkar-Jaiswal et al . , 2015; Quiñones-Coello et al . , 2007; St Johnston , 2012; Venken and Bellen , 2012; Venken et al . , 2011 ) with a comprehensive genome-scale library that does not suffer the biases of random mutagenesis . Analogously to the powerful MiMIC system ( Nagarkar-Jaiswal et al . , 2015; Venken et al . , 2011 ) , the TransgeneOme resource is versatile and can be adapted to the developments in tag chemistry and to various specialised applications . Although the resource is designed to study behaviour of proteins , it can for example be converted into a toolkit for live imaging of mRNAs . By designing a tagging cassette with an array of MS2 binding sites ( Forrest and Gavis , 2003 ) the existing 'pre-tagged' TransgeneOme can be converted into an MS2-tagged TransgeneOme by a single liquid culture recombineering step in bacteria . However , any new TransgeneOme has to be transformed into flies , and this process still represents a significant bottleneck . We present here an optimised protocol for transgenesis of fosmid size clones into Drosophila melanogaster that was adapted from a previous large-scale transgenesis project ( Venken et al . , 2010 ) . It took three years and four dedicated technicians to generate the 880 fly lines presented in this study . Although , the systematic transgenesis is a continuing process in our laboratories , the value of the TransgeneOme collection is highlighted by the fact that any specific set of genomically tagged gene clones is now available . These can be efficiently transformed by in house transgenesis of Drosophila labs around the world using the optimised protocol presented here . One caveat of designed expression reporters is the necessity to place the tag into a defined position within the gene model . We chose to generate our 'pre-tagged' collection at the most commonly used C-terminus predicted by the gene model , thus labelling most isoforms . In some cases , a tag at the C-terminus will inactivate the protein , however such a reagent can still be useful for visualising the protein , although the result needs to be treated with care . This has been demonstrated for a number of sarcomeric protein GFP-traps , some of which lead to lethality when homozygous , yet result in interesting localisation patterns when heterozygous ( Buszczak et al . , 2007; Morin et al . , 2001 ) . Similarly , we found an interesting localisation pattern for BM40-SPARC-GFP despite being non functional . In some cases the tag may result in a mis-localisation of the tagged protein , compared with the untagged endogenous one , in particular when the tag interferes with protein function . For particular genes , it will be useful to tag differential protein isoforms , which in some cases can be done by tagging alternative C-termini , as shown here for Mhc and rhea . However , tagging a particular isoform requires a very informed construct design , which cannot easily be automated at the genome scale . A functional GFP-tagged gene copy , as present in our fTRG lines , can also serve as a ‘conditional’ allele , when crossed into the mutant background and combined with deGradeFP ( degrade Green Fluorescent Protein ) , an elegant method expressing a nanobody against GFP , coupled with a degradation signal in a tissue- and stage-specific manner ( Caussinus et al . , 2011; Neumuller et al . , 2012 ) . As the expression of the nanobody can be turned on or off , it is also possible to reversibly remove the tagged protein , as recently shown for the GFP-tagged MiMIC lines ( Nagarkar-Jaiswal et al . , 2015 ) . This introduces yet an other level of experimental manipulation , directly controlling protein levels at a given developmental stage . Genome engineering is experiencing a tremendous growth with the introduction of CRISPR/Cas technology , and it will be only a matter of time before a larger collection of precisely engineered fusion proteins at endogenous loci will become available in flies . However to date , such examples are still limited to a few genes ( Baena-Lopez et al . , 2013; Gratz et al . , 2014; Port et al . , 2014; Zhang et al . , 2014 ) , which had been carefully picked and were individually manipulated with custom-designed , gene-specific tools . It remains to be tested which proportion of such engineered loci will be fully functional and thus potentially superior to the fTRG collection . However , having a transgenic third allele copy , as is the case in our TransgeneOme collection , might even be advantageous , if the tagging interferes with protein function , because the TransgeneOme lines still retain two wild-type endogenous gene copies . In some cases , addition of GFP might destabilise the protein , regardless of N- or C-terminal fusion , as recently shown for the Engrailed protein ( Sokolovski et al . , 2015 ) . However , our ability to detect the protein product in the vast majority of our tagged lines argues that this could be a relatively rare , gene-specific phenomenon . Nevertheless , caution should be taken with respect to protein turnover dynamics of any tagged protein . An additional advantage of our transgenic resource , independent of whether or not the gene is tagged , is that it can be used to rescue a classic genetic mutation and thus formally demonstrate that any observed phenotype is caused by the mutation in the studied gene . This cannot easily be done when modifying the endogenous gene copy by a MiMIC insertion or a CRISPR induced mutation . Thus , our resource complements previously published collections of genomic constructs ( Ejsmont et al . , 2009; Venken et al . , 2009 ) . Together , the FlyFos library , the fly TransgeneOme library and the fTRG collection of strains , enable genome-scale examination of expression and localisation of proteins comparable with the high-throughput mRNA in situ screens ( Tomancak et al . , 2002; 2007 ) . Our data for tagged Oscar protein show that fosmid-based reporters can in principle recapitulate all aspects of gene expression regulation at transcriptional and post-transcriptional levels . It will be particularly interesting to combine the spatial expression data of mRNAs with that of proteins . Since many transcripts show subcellular localisation patterns in various developmental contexts ( Jambor et al . , 2015; Lécuyer et al . , 2007 ) , the question arises whether RNA localisation generally precedes localised protein activity . Systematic examination of protein patterns expressed from localised transcripts in systems such as the ovary will provide a genome-scale overview of the extent and functional role of translational control . At the tissue level , the patterns of mRNA expression may be different from the patterns of protein expression , for example due to translational repression in some cells or tissue specific regulation of protein stability , as shown here for the Corolla protein . The combined mRNA and protein expression patterns may therefore uncover a hidden complexity in overall gene activity regulation and the fTRG lines will help to reveal these combinatorial patterns in a systematic manner . The fTRG lines faithfully recapitulate gene expression patterns in ovaries , embryos , larvae , pupae and adults suggesting that they can be used to visualise proteins in every tissue during the life cycle of the fly . This includes adult tissues such as the flight or leg muscles , which thus far had not been subjected to systematic protein expression and localisation studies . However , due to their size and the conservation of the contractile apparatus , these tissues are particularly attractive to study with this new resource . In general , antibody or FISH ( Fluorescent In Situ Hybridisation ) stainings with a single standard anti-tag reagent are easier to optimise , compared to antibody stainings or mRNA in situ hybridisations with gene-specific antibodies/probes . This simplicity makes it possible to explore the expression of the available genes across multiple tissues , as has been done for the rab collection ( Dunst et al . , 2015 ) . Such an approach is orthogonal to the collections of expression data generated thus far , in which many genes were examined systematically but only at particular stages or in certain tissues , i . e . embryos or ovaries ( Jambor et al . , 2015; Lécuyer et al . , 2007; Tomancak et al . , 2007 ) . We are confident that the analysis of regularly studied as well as less explored Drosophila tissues will be stimulated by the fTRG collection . When protein expression levels are sufficiently high , the fusion proteins can be visualised by live imaging approaches in intact animals . It is difficult to estimate the absolute expression levels required for live visualisation , as this depends on the imaging conditions , the accessibility and transparency of the tissue and importantly on the observed protein pattern . A strongly localised protein can result in a very bright local signal , such as Talin or Ilk at the muscle attachment sites or Gsb in the neuroblast nuclei , compared to a protein homogenously distributed throughout the entire cell . Given optimal imaging conditions , we estimate conservatively that about 50% of the tagged proteins can be visualised live , if they are expressed in tissues accessible to live imaging . In particular for tissues , such as the adult legs , antennae or the adult fat body , which are difficult to dissect and stain without losing tissue integrity , these live markers should be enormously beneficial . One important limitation for examining the pattern of protein expression is the accessibility of the tissue of interest for imaging . We have shown that light sheet microscopy can be used to image the dynamics of tagged protein expression throughout embryogenesis . We further demonstrated that two-photon microscopy can be applied to study protein dynamics during muscle morphogenesis in developing pupae . Other confocal or light sheet-based imaging paradigms could be adapted for in totoimaging of living or fixed and cleared specimen from other life cycle stages . Establishing standardized protocols for preparation , staining and imaging of Drosophila stages , isolated tissues and organs will be necessary to realise the full potential of the fTRG collection . Protein interaction data in fly are available from a number of studies ( Formstecher et al . , 2005; Giot et al . , 2003; Guruharsha et al . , 2011 ) . These results were generated using yeast two-hybrid , or over-expression in a tissue culture system , followed by affinity purification and mass spectrometric analysis . Despite high-throughput , these approaches face the problem that the interacting proteins might not be present at the same place within a cell , or not even co-expressed in a developing organism . This is circumvented by affinity purifications of endogenously expressed proteins , which thus far at genome-scale was only reported from yeast ( Gavin et al . , 2002; Ho et al . , 2002; Krogan et al . , 2006 ) . In higher organisms , BAC-based systems , which are closely related to our fosmid approach , elegantly solved these issues , as shown by a recent human interactome study ( Hein et al . , 2015 ) . The collection of transgenic fTRG lines covers currently only about 10% of the available tagged TransgeneOme clones . Expanding the fly collection to include most genes of the genome and importantly characterising the expression of the tagged proteins by imaging in various biological contexts is best achieved by spreading the clones and transgenic lines amongst the community of researchers using Drosophila as a model system . Therefore , all transgenic lines are available from the VDRC stock collection and all TransgeneOme clones from Source Biosciences . Despite the expanding CRISPR-based genome engineering technologies , the fTRG collection will continue to be an important resource for the fly community , in particular , if the full functionality of certain fTRG lines has been demonstrated , as we did here for a selection of important developmental regulators . As with many genome-scale resources it is typically easier to produce them than to fully characterise and exploit their potential . Comprehensive generation of thousands of transgenes and their thorough analysis takes time; it took us 4 years to assemble the collection presented here . Development of protocols and techniques to image these collections of tagged lines and assembling open access databases to share the data needs to continue and will eventually become useful also for the characterisation of resources whose production began only recently . Wangler , Yamamoto and Bellen convincingly argued that the Drosophila system remains an indispensable model for translational research because many essential fly genes are homologs of Mendelian disease genes in humans ( Wangler et al . , 2015 ) . Yet , even after decades of research on fruit flies only about 2 , 000 of the estimated 5 , 000 lethal mutations have been investigated . Resources like ours will therefore provide essential functional information about gene expression and localisation in Drosophila tissues that can serve as a starting point for the mechanistic understanding of human pathologies and their eventual cures . Fosmids were engineered as described previously ( Ejsmont et al . , 2009; 2011 ) , except for the inclusion of the ‘pre-tagging’ step in the genome-wide TransgeneOme set . All tagging cassettes were generated from synthetic DNA and cloned into R6K carrying plasmids , which require the presence of the pir gene product for replication ( Metcalf et al . , 1996 ) . The pir gene is not present in the FlyFos library host strain , thereby ensuring near-complete lack of background resistance in the absence of the correct homologous recombination event . Details of the recombineering steps are as follows ( Figure 1B ) : Step 1 . The E . coli cells containing a FlyFos clone covering the gene locus of interest are transformed with the pRedFlp plasmid , containing the genes necessary for the homologous recombination and the Flp recombinase under independently inducible promoters . Step 2 . Next , a ‘pre-tagging’ cassette carrying an antibiotic resistance gene ( NatR , nourseothricin resistance ) surrounded by regions of homology to all specific tagging cassettes ( Figure 1—figure supplement 1 ) and flanked by gene-specific homology arms is electroporated as linear DNA fragment produced by PCR . By combination of induced ( L-rhamnose ) pRedFlp homologous recombination enzyme action and strong selection with a cocktail of three antibiotics ( one to maintain the fosmid ( chloramphenicol , Cm ) , one to maintain the pRedFlp ( hygromycin , Hgr ) and nourseothricin ( Ntc ) to select for the inserted fragment ) the electroporated linear 'pre-tagging' fragment becomes inserted in front of the STOP codon of the gene of interest . Step 3 . The ‘pre-tagging’ cassette is exchanged for a cassette of the chosen tag coding sequence including an FRT-flanked selection / counter selection marker ( rpsL-neo ) . This cassette is now universally targeting the homologous sequences shared by the tagging and pre-tagging cassettes and is produced in bulk by restriction enzyme-mediated excision from a plasmid . Note that in this way , no PCR-induced mutations can be introduced at this step . Step 4 . Upon Flp induction ( with anhydrotetracycline ) , the rpsL-neo cassette is excised , leaving a single FRT site , positioned in frame after the tag coding sequence . In this way , the endogenous STOP codon and the 3’-UTR of the tagged gene are used . Step 5 . Finally , the recombineering plasmid is removed from the cells containing the engineered fosmids by inhibition of its temperature sensitive origin of replication and release from Hgr selection . The cells are plated on a selective chloramphenicol agar plate , from which a single colony is picked and further validated . For NGS-based validation of the TransgeneOme library single colonies for each TransgeneOme clone were picked into 96-well plates , grown to saturation and the individual wells of all 96-well plates were pooled into 8 row and 12 column pools . Fosmid DNA was isolated from these pools and barcoded mate pair fragment libraries were prepared using the Nextera matePair library preparation chemistry from Illumina . The library was size selected through agarose gel isolation of approximately 3 kb fragments and sequenced on HiSeq 2500 ( Illumina ) , with paired-end read lengths of 100 bp . Adapters and low quality sequences were trimmed with Trimmomatic0 . 32 . ( parameters: ILLUMINACLIP:NexteraPE-PE . fa:2:30:10 LEADING:3 TRAILING:3 SLIDINGWINDOW:4:15 MINLEN:36 ) . To detect un-flipped fosmid sequences ( where the FLP-mediated excision of selection cassette failed ) , the read pairs were mapped with Bowtie2 ( Langmead and Salzberg , 2012 ) against the un-flipped tag sequence and the genome . If any read of the mate of the pair mapped to the un-flipped sequence , while the second mate mapped to the genome consistent with the estimated mate pair insert size of 3000 bp ± 1000 bp , the fosmid was flagged as un-flipped and was not further analysed . To identify mutations in the tag and in the immediate genomic surrounding ( ± 1000 bp ) , the NGS reads were mapped against the fosmid references that included the flipped tag . The Bowtie2 was set to report only hits where both reads of the pairs map concordantly to the insert size in the tag and in the genome ( parameters: -I 2200 -X 3700 --rf --no-discordant --no-unal --no-mixed ) . PCR duplicated read pairs were removed using samtools1 . 1 rmdup ( Li et al . , 2009 ) . Mutations were identified by utilising SNP calling implemented in FreeBayes ( Garrison and Marth , 2012 ) using the standard filters and vcffilter to eliminate reported SNPs with scores < 20 . Finally , in the last step , the information of the row and column pools were compared and summarized using a custom C-program that read the results of the SNP calling and the Bowtie mappings and counted the coverage for each read pair anchored in tag sequence with at least 20 bp . To correct for random PCR or sequencing errors the reported SNPs were compared for the row and column pools of each fosmid and SNPs occurring in both pools with coverage of 3 or more reads were considered as real . Fly stocks were maintained using standard culture conditions . All crosses were grown at 25°C unless otherwise noted . Most of the fly mutant or deficiency strains for the rescue experiments were obtained from the Bloomington Drosophila Stock Center and if located on X or 2nd chromosome crossed together with the respective fTRG line . If the mutant gene was located on the 3rd chromosome , it was recombined with the fTRG insertion . Rescue was generally tested in trans-heterozygotes as indicated in Table 2 . The rescue for 6 genes ( bam , fat , mask , rap , RhoGEF2 and yki ) was done by others , who communicated or published the results ( Table 2 ) . For the rescue of flightlessness a standard flight test was used ( Schnorrer et al . , 2010 ) . Most TransgeneOme fosmid clones were injected into the y[1] , w[*] , P{nos-phiC31int . NLS}X; PBac{y+-attP-3B}VK00033 ( BL-32542 ) . This stock has white eyes and no fluorescent eye markers , which would interfere with screening for the red fluorescent eye marker used in the FlyFos clones ( Ejsmont et al . , 2009 ) . A few fosmid clones were also injected into y[1] , w[*] , P{nos-phiC31int . NLS}X; PBac{y+-attP-3B}VK00002 , with the attP site located on the 2nd chromosome . The osk-GFP fosmid was injected into attP40 . Please note that all fTRG lines contain the strong 3xP3-dsRed marker ( Ejsmont et al . , 2009 ) . This is an eyeless derived promoter fragment resulting in dsRed expression in the developing eye and in the brain . This needs to be taken into account when working with the developing or adult brain . Ovaries: sGFP-protein detection and antibody co-stainings of egg-chambers was done as previously described ( Dunst et al . , 2015 ) . Detection of the oskar-GFP mRNA was performed with a gfp-antisense probe ( Jambor et al . , 2014 ) and co-staining of osk mRNA and Osk protein was done as previously described ( Jambor et al . , 2011 ) using a gfp-antisense probe and a rabbit anti-GFP antibody ( 1:1000 , ThermoFisher ) . Rabbit anti-Osk was used 1:3000 ( gift on Anne Ephrussi ) , mouse anti-Grk was used 1:100 ( DSHB ) . Adult thoraces: Antibody stainings of adult thoraces , including flight , leg and visceral muscles , were done essentially as described for adult IFMs ( Weitkunat and Schnorrer , 2014 ) . Briefly , thoraces from young adult males were fixed for 15 min in relaxing solution ( 20 mM phosphate buffer , pH 7 . 0; 5 mM MgCl2; 5 mM EGTA , 5 mM ATP , 4% PFA ) + 0 . 5% Triton X-100 , cut sagittally with a sharp microtome blade and blocked for 1 hr at room temperature with 3% normal goat serum in PBS-0 . 5% Triton X-100 . The samples were stained with primary antibodies overnight at 4ºC , rabbit anti-GFP 1:2000 ( Amsbio ) ; mouse anti-Futsch 1:100 , mouse anti-Dlg1 clone 4F3 1: 500 , mouse anti-Prospero 1:30 ( all Hybridoma Bank , DSHB ) , mouse anti-ATP5a 1:500 ( Abcam clone 15H4C4 ) , rabbit anti-Fln ( gift of Jim Vigoreaux ) , rabbit anti-Kc cell Laminin H329 1:2000 ( gift of Stefan Baumgartner ) , rabbit anti-Mlp84B 1:500 ( gift of Kathleen Clark ) , mouse anti-Mhc 1:100 ( gift of Judith Saide ) , Mouse anti-Obscurin 1:500 ( gift of Belinda Bullard ) rabbit anti-Par6 1:400 ( gift of Jürgen Knoblich ) , washed and incubated with secondary antibodies coupled to Alexa dyes and rhodamine-phalloidin or phalloidin-Alexa-660 ( all from Molecular Probes ) . After washing , the samples were mounted in Vectashield containing DAPI . Images were acquired with a Zeiss LSM 780 confocal microscope and processed with Fiji ( Schindelin et al . , 2012 ) and Photoshop ( Adobe ) . Protein detection by Western blotting used standard procedures . 15 adult males were homogenised in 200 µl SDS buffer ( 250 mM Tris pH 6 . 8 , 30% glycerol , 1% SDS , 500 mM DTT ) and 5 µl were loaded per lane of a 10% SDS-PAGE gel . The Immobilon membranes ( Millipore ) were blocked with 10% milk powder and incubated with primary antibodies overnight ( mouse anti-V5 1:10 , 000 ( Invitrogen ) , mouse anti-Dlg1 1:10 , 000 , rabbit anti-Mlp84B 1:20 , 000 , rabbit anti-Fln 1:10 , 000 ) . Detection used POD-coupled secondary antibodies ( Jackson labs ) and chemiluminescence ( Millipore ) using a LAS4000 detector system ( FujiFilm ) . SPIM imaging of embryos: De-chorionated embryos of the appropriate age were embedded in 1% low melting point agarose and mounted into a glass capillary . Fluorescent microspheres ( FY050 Estapor microspheres , Merck Millipore; 1:4000 ) were included in the embedding medium for multi-view registration . The embryos were imaged using the Zeiss Lightsheet Z . 1 with a Zeiss 20x/1 . 0 water-immersion Plan Apochromat objective lens with 0 . 8x zoom at 25°C using 488 nm laser set at 4 mW . Five views were imaged using dual-sided illumination with Zeiss 10x/0 . 2 illumination lenses . A mean fusion was applied to fuse both illumination sides after acquisition using the ZEN software ( Zeiss ) . The views were acquired at 72° angles with a stack size of 130 µm and a step size of 1 . 5 µm . Exposure time were 30 ms per slice . Each slice consists of 1920 x 1200 pixels with a pixel size of 0 . 29 µm and a bit depth of 16 bits . The light sheet thickness was 4 µm at the center of the field of view . The embryos were imaged from the onset of GFP expression ( determined empirically ) until late embryogenesis with a time resolution of 15 min . Multi-view processing of the dataset was carried out using the Fiji plugin for multi-view reconstruction ( Preibisch et al . , 2009; Schmied et al . , 2014 ) , which was executed on a high performance computing cluster ( Schmied et al . , 2015 ) . The multi-view reconstruction was followed by multi-view deconvolution ( Preibisch et al . , 2014 ) , for which the images were down sampled by a factor of two . Videos were extracted via the Fiji plugin BigDataViewer ( Pietzsch et al . , 2015 ) . The Gsb-GFP fTRG line was crossed with the H2Av-mRFPruby line ( Fischer et al . , 2004; Preibisch et al . , 2014 ) , the embryos of this cross were imaged using a 40x/1 . 0 water immersion Plan Apochromat lens from Zeiss with 1x zoom at 25°C at 17 . 5 mW of the 488 nm laser and 4 mW of the 561 nm laser . A single angle with dual-sided illumination was imaged . The stack size was 82 . 15 µm with a step size of 0 . 53 µm . Exposure time was 30 ms per slice . Each slice consisted of 1920 x 1920 pixels with a pixel size of 120 nm and a bit depth of 16-bit . The light sheet thickness was 3 . 21 µm at the center of the field of view . The embryos were imaged from early blastoderm onwards until late embryogenesis focusing on the head with a time resolution of 7 min . Imaging of pupae: Staging and live imaging of the pupae were performed at 27ºC . Live imaging of pupae at the appropriate stage was done as described previously ( Weitkunat and Schnorrer , 2014 ) . Briefly , the staged pupa was cleaned with a brush and a small observation window was cut into the pupal case with sharp forceps . The pupa was mounted on a custom-made slide and the opening was covered with a small drop of 50% glycerol and a cover slip . Z-stacks of either single time points or long-term time-lapse Videos were acquired using either a spinning disc confocal microscope ( Zeiss , Visitron ) or a two-photon microscope ( LaVision ) , both equipped with heated stages . Per sample about hundred pupae or adult flies were snap-frozen in liquid nitrogen and ground to a powder . The powder was re-suspended and further processed as described in the quantitative BAC-GFP interactomics protocol ( Hubner et al . , 2010 ) . In brief , 800 µl of lysate per sample were cleared by centrifugation . The cleared lysate was mixed with magnetic beads pre-coupled to anti-GFP antibodies and run over magnetic micro-columns ( both Miltenyi Biotec ) . Columns were washed , and samples subjected to in-column tryptic digestion for 30 min . Eluates were collected and digestion continued overnight , followed by desalting and storage on StageTips . Eluted peptides were analysed with an Orbitrap mass spectrometer ( Thermo Fisher ) . Raw data were analysed in MaxQuant version 1 . 4 . 3 . 22 ( Cox and Mann , 2008 ) using the MaxLFQ algorithm for label-free quantification ( Cox et al . , 2014 ) . Interacting proteins were identified by the similarity of their intensity profiles to the respective baits ( Keilhauer et al . , 2014 ) . Heat maps were plotted in the Perseus module of the MaxQuant software suite .

The fruit fly Drosophila melanogaster is a popular model organism in biological research . Studies using Drosophila have led to important insights into human biology , because related proteins often fulfil similar roles in flies and humans . Thus , studying the role of a protein in Drosophila can teach us about what it might do in a human . To fulfil their biological roles , proteins often occupy particular locations inside cells , such as the cell’s nucleus or surface membrane . Many proteins are also only found in specific types of cell , such as neurons or muscle cells . A protein’s location thus provides clues about what it does , however cells contain many thousands of proteins and identifying the location of each one is a herculean task . Sarov et al . took on this challenge and developed a new resource to study the localisation of all Drosophila proteins during this animal’s development . First , genetic engineering was used to tag thousands of Drosophila proteins with a green fluorescent protein , so that they could be tracked under a microscope . Sarov et al . tagged about 10000 Drosophila proteins in bacteria , and then introduced almost 900 of them into flies to create genetically modified flies . Each fly line contains an extra copy of the tagged gene that codes for one tagged protein . About two-thirds of these tagged proteins appeared to work normally after they were introduced into flies . Sarov et al . then looked at over 200 of these fly lines in more detail and observed that many of the proteins were found in particular cell types and localized to specific parts of the cells . Video imaging of the tagged proteins in living fruit fly embryos and pupae revealed the proteins’ movements , while other techniques showed which proteins bind to the tagged proteins , and may therefore work together in protein complexes . This resource is openly available to the community , and so researchers can use it to study their favourite protein and gain new insights into how proteins work and are regulated during Drosophila development . Following on from this work , the next challenge will be to create more flies carrying tagged proteins , and to swap the green fluorescent tag with other experimentally useful tags .

2016

A genome-wide resource for the analysis of protein localisation in Drosophila

Human speech is one of the few examples of vocal learning among mammals yet ~half of avian species exhibit this ability . Its neurogenetic basis is largely unknown beyond a shared requirement for FoxP2 in both humans and zebra finches . We manipulated FoxP2 isoforms in Area X , a song-specific region of the avian striatopallidum analogous to human anterior striatum , during a critical period for song development . We delineate , for the first time , unique contributions of each isoform to vocal learning . Weighted gene coexpression network analysis of RNA-seq data revealed gene modules correlated to singing , learning , or vocal variability . Coexpression related to singing was found in juvenile and adult Area X whereas coexpression correlated to learning was unique to juveniles . The confluence of learning and singing coexpression in juvenile Area X may underscore molecular processes that drive vocal learning in young zebra finches and , by analogy , humans . The ability to learn new vocalizations is a key subcomponent of language . Complex behaviors such as human speech and birdsong are rarely monogenic in origin , making the attribution of their direct molecular underpinnings a challenge ( Marcus and Fisher , 2003 ) . While language is unique to humans , learned vocal behavior is present in a number of animal taxa . Among laboratory animals , the zebra finch songbird ( Taeniopygia guttata ) is the primary genetic model for vocal learning , and song learning in this species shares numerous parallels with human speech development . For example , both species share corticostriatal loops for producing vocalizations and have direct projections from cortical neurons onto brainstem motor neurons that control the vocal organs , a connection that is lacking or reduced in non-vocal learners ( Lemon , 2008; Jürgens , 2002; Arriaga et al . , 2012; Doupe and Kuhl , 1999; Petkov et al . , 2012 ) . The brains of avian vocal learners contain a distributed corticostriatal network of clustered cells devoted to vocal production learning , commonly referred to as the song control circuit , offering tractable targets for experimental manipulation . Despite their evolutionary distance , humans and zebra finches exhibit shared transcriptional profiles in key brain regions for vocal learning that are unique from surrounding brain areas and from the brains of non-vocal learning species ( Pfenning et al . , 2014 ) . The forkhead box P2 ( FOXP2 ) transcription factor was the first gene shown to be important for vocal learning in both humans and songbirds . Forkhead box proteins are characterized by the presence of DNA-binding FOX domains ( Clark et al . , 1993 ) and FOXP subfamily members form homo- or heterodimers at zinc finger and leucine zipper domains in order to bind DNA . In humans , a heterozygous mutation in the FOX domain of FOXP2 causes a rare heritable speech and language disorder in a cohort known as the KE family ( Vargha-Khadem et al . , 1998; Lai et al . , 2001 ) , potentially by altering the subcellular localization of the molecule ( Vernes et al . , 2006 ) . While the mutation disrupts vocal learning ( Marcus and Fisher , 2003 ) and also vocalization in vocal non-learners ( Chabout et al . , 2016; Castellucci et al . , 2016 ) , multiple FOXP2 isoforms are endogenous to both songbirds and humans , including one that lacks the DNA binding domain ( Teramitsu and White , 2006; Bruce and Margolis , 2002 ) . This truncated variant is referred to as FOXP2 . 10+ because , although it lacks the FOX domain , it retains the dimerization domains plus an additional 10 amino acids that are not found in the full length form ( FoxP2 . FL ) . Consistent with its lack of a FOX domain , in vitro assays of FOXP2 . 10+ indicate that it may regulate other FoxP2 isoforms ( Vernes et al . , 2006 ) . Since it retains the dimerization domain , it has been hypothesized to act as a cytoplasmic sink , binding to other FOXP proteins and preventing their entry to the nucleus and interaction with DNA . Investigation of FoxP2 function in zebra finches has revealed remarkable parallels with humans . Similar FoxP2 expression patterns occur in developing human and zebra finch brains ( Teramitsu et al . , 2004 ) . In zebra finches , knockdown of FoxP2 in the song dedicated striatopallidal nucleus , Area X , during vocal development impaired vocal mimicry of tutor songs ( Haesler et al . , 2007 ) , much as the KE family mutation impairs speech . These observations indicate that functional FoxP2 is necessary for proper vocal learning , an inference supported by work in songbirds ( Haesler et al . , 2007; Heston and White , 2015 ) . The unique organization of song control circuit neurons enabled the discovery that FoxP2 is dynamically downregulated within Area X when zebra finches practice their songs , termed ‘undirected’ ( UD ) singing ( Teramitsu and White , 2006; Miller et al . , 2008; Hall , 1962; Immelmann , 1962; Dunn and Zann , 1996 ) . This decrease in FoxP2 is accompanied by increased vocal variability ( Miller et al . , 2010; Hilliard et al . , 2012a ) , thought to be a form of vocal exploration . Blockade of FoxP2 downregulation impaired birds’ ability to induce variability in their songs . A poor learning phenotype emerged following FoxP2 overexpression ( Heston and White , 2015 ) that was remarkably similar to that observed following FoxP2 knockdown ( Haesler et al . , 2007 ) . Taken together , these results indicate that the dynamic regulation of at least FoxP2 . FL , and thereby the behavior-linked up- and down-regulation of its transcriptional targets , is necessary for the proper learning of vocalizations . No specific role in vocal behavior has yet been attributed to the FoxP2 . 10+ isoform . These observations pinpoint FoxP2 as a molecular entry point to the pathways underlying vocal learning . In adult birds , we previously used Weighted Gene Coexpression Network Analysis ( WGCNA ) to identify thousands of genes regulated by singing specifically in Area X ( Hilliard et al . , 2012a; Langfelder and Horvath , 2008 ) . Since adult zebra finches sing stable , or crystallized , songs , the transcription patterns underlying vocal learning were not identified . Here we conduct a new study with two goals: ( 1 ) Determine whether FoxP2 . 10+ may play a role in vocalization and , ( 2 ) Manipulate FoxP2 isoforms in juveniles to generate a broad range of behavioral and transcriptional states upon which to apply WGCNA and thereby reveal learning-related gene modules . Toward the first goal , overexpression of FoxP2 . 10+ revealed a unique role for this truncated isoform in the acute modulation of vocal variability . Toward the second goal , overexpression of either GFP or one of the two FoxP2 isoforms created three distinct groups of juvenile birds: one that was good at learning and acutely modulating variability ( GFP ) , one that was poor at learning and acutely modulating variability ( FoxP2 . FL ) , and one that was good at learning but injected stability into song ( FoxP2 . 10+ ) . We applied WGCNA to the Area X transcriptome of birds across this behavioral continuum and discovered striatopallidal coexpression patterns that were positively correlated to learning . These learning-related patterns were present in juvenile but not adult Area X . However , singing-driven coexpression patterns in Area X were largely preserved between juveniles and adults , suggesting that: ( 1 ) song production modules are independent of learning state and ( 2 ) the spatiotemporal co-occurrence of both song production and learning-related gene modules in juvenile Area X is fundamental to vocal learning . Adeno-associated viral ( AAV ) constructs were used to drive overexpression of FoxP2 . FL or FoxP2 . 10+ in Area X of developing males ( Figure 1—figure supplement 1 ) . To verify isoform-specific overexpression , we used two riboprobes in in situ hybridization experiments: one antisense to a region common to both transcripts ( mid probe ) and one antisense to a region near the 3’ end of FoxP2 . FL ( 3’ probe; [Teramitsu and White , 2006]; Figure 1A ) . Robust signals beyond endogenous/background levels were observed in the striatopallidum of both hemispheres using the mid probe but only in the hemisphere injected with the FoxP2 . FL construct using the 3’ probe ( Figure 1B ) . These results indicate that each viral construct overexpressed its encoded FoxP2 isoform and was thus suitable for bilateral injection into Area X of juvenile males at 35d . An additional cohort received AAV encoding GFP as a control . We quantified levels of FoxP2 expression at 65d by performing qRT-PCR with a set of primers that amplifies a region common to both transcripts ( Haesler et al . , 2007; Olias et al . , 2014 ) and another set specific to the FoxP2 . 10+ ( see Materials and methods ) . The first primer set indicated that FoxP2 levels were higher in birds injected with either construct relative to control levels . When quantified by the second primer set , we found elevated PCR product only in the animals injected with the FoxP2 . 10+ construct ( Figure 1C ) . No overexpression was detected in the ventral striatopallidum ( VSP; the zebra finch striatum is interspersed with pallidal-like cells and is separate from the pallidum [Reiner et al . , 2004] ) ( Figure 1—figure supplement 2 ) . Taken together , these results indicate that both constructs were effective in elevating levels of their encoded FoxP2 isoform within Area X throughout the 30d experimental period . Overexpression of a tagged form of FoxP2 . 10+ in a human neuronal cell line ( SH-SY5Y ) suggested that FoxP2 . 10+ acts as a posttranslational regulator of FoxP2 . FL through heterodimerization and the formation of cytoplasmic aggresomes ( Vernes et al . , 2006 ) . We thus examined the protein-level distribution of FoxP2 . 10+ and FoxP2 . FL in the finch striatopallidum following overexpression of an N-terminus Xpress tagged FoxP2 . 10+ linked to a GFP reporter ( see Stereotaxic Surgery and Viruses in Materials and methods ) . Transduced cells shared the distinctive FoxP2 . 10+ staining pattern of aggresomes seen previously . In FoxP2+ cells that co-expressed the Xpress tag and GFP reporter , endogenous FoxP2 . FL signal was interspersed among Xpress-positive puncta ( Vernes et al . , 2006 ) ( Figure 1D ) . We previously found that , in unmanipulated birds , two hours of UD singing in the morning is sufficient to decrease Area X FoxP2 mRNA ( as measured by both the mid and 3’ probes ) and protein ( Teramitsu and White , 2006; Miller et al . , 2008 ) . This decrease in FoxP2 was accompanied by an increase in the variability of UD songs , in the form of decreased self-similarity ( see Materials and methods ) , that were sung subsequent to the two hour time-point , a paradigm which we term UD-UD ( Miller et al . , 2010; Hilliard et al . , 2012a ) . In contrast , when birds were distracted from singing for two hours in the morning ( non-singing; NS ) , their subsequent UD songs ( termed NS-UD ) were less variable . Moreover , overexpression of FoxP2 . FL in Area X abolished the increase in vocal variability normally induced by the UD-UD paradigm ( Heston and White , 2015 ) . These observations indicate that downregulation of full length FoxP2 is important for acute vocal variability but we did not directly manipulate FoxP2 . 10+ . Here , we performed similar behavioral experiments to test for the induction of vocal variability and included the FoxP2 . 10+ injected animals ( Figure 2A and B ) . To assess whether UD singing drove an increase in vocal variability , we used the UD-UD paradigm ( see Materials and methods ) and quantified the effect of two hours of UD singing on the coefficient of variation ( CV ) of acoustic features in the subsequent UD songs of ~60d birds overexpressing GFP , FoxP2 . FL , or FoxP2 . 10+ . Results were compared to songs sung by the same birds undergoing the NS-UD paradigm . As predicted , GFP-expressing animals exhibited a negative effect size for most acoustic features , and FoxP2 . FL overexpression diminished these practice-induced changes in vocal variability , replicating our previous findings ( Heston and White , 2015 ) ( Figure 2C ) . Unexpectedly , in animals overexpressing FoxP2 . 10+ , song variability after two hours of UD singing ( UD-UD ) was significantly less than that after two hours of non-singing ( NS-UD ) for syllable duration , amplitude modulation , and Wiener entropy ( Figure 2C ) . Rather than increasing song variability ( as in the GFP group ) or creating a state of equivalent variability ( as in the FoxP2 . FL group ) , UD-UD singing led to markedly invariable songs in the FoxP2 . 10+ birds , suggesting a role for FoxP2 . 10+ in promoting song stability . We also examined variability in the raw acoustic features of NS-UD and UD-UD song and found that expression of either FoxP2 isoform did not dramatically alter variability , indicating that the viral-driven overexpression specifically affected the modulation of variability ( See ‘Acute Modulation of Vocal Variability’ in Materials and methods ) and not its overall level ( Figure 2—figure supplement 1 and Materials and methods ) . Despite its suppressive effect on practice-induced song variability , overexpression of FoxP2 . 10+ did not impair overall vocal learning ( Figure 2D and E ) . As shown by Heston and White ( Heston and White , 2015 ) , FoxP2 . FL birds were capable of changing their songs over the course of the experiment ( data not shown ) but were less able to match their tutors’ songs ( Figure 2D and E ) . These results suggest that the ability to modulate between relatively low and high variability states is important for proper vocal learning . In sum , our viral manipulations generated groups of animals in distinct states of vocal variability and learning . GFP-injected birds learned well and displayed singing-induced variability in the acoustic features of song . FoxP2 . FL birds learned poorly and had no difference in their songs’ acoustic variability following practice . FoxP2 . 10+ birds learned well but seemed to exist in a state where practice drives invariability in vocal acoustics . As such , a broad degree of both learning and variability induction existed across groups ( Figure 2F ) . Next , we used these behavioral metrics as correlates to gene coexpression patterns to interrogate the transcriptional profiles underlying these traits . We used RNA-seq to quantify gene transcription in Area X of 65d juveniles overexpressing GFP , FoxP2 . FL or FoxP2 . 10+ , then used WGCNA to identify gene coexpression modules and link them to song learning . We built an overall network composed from all samples together ( Figure 3A and B ) , as well as construct-specific networks ( Figure 3—figure supplements 1–4 ) . In the overall network ( see Materials and methods ) , 7461 genes formed 21 modules ( Figure 3A and B , Supplementary file 1 ) . We found significant correlations between module eigengenes and the following behaviors: tutor percentage similarity ( i . e . vocal learning: darkred , green , and greenyellow modules ) , number of motifs sung ( i . e . amount of singing: black , orange , darkgreen , royalblue , and blue modules ) , singing-induced acoustic variability ( i . e . variability induction: black , brown , darkgreen , darkgrey , magenta , orange , pink , purple and turquoise modules ) , and motif identity ( i . e . overall vocal variability: darkgrey module ) ( 0 . 00008 < p < 0 . 05; Figure 3B ) . Hereafter , these modules are termed ‘learning-related’ , ‘song-production’ , ‘variability-induction’ and ‘vocal variability’ modules , respectively . We examined all modules whose p-value was ≤0 . 05 and calculated the relationship between module membership and gene significance . ( For definitions of WGCNA and network terms , see Materials and methods: WGCNA and network terminology . For information about significance levels reported here , see Materials and methods: Correlation of behavior to gene expression ) . For most modules , strong correlations were observed for each trait , indicating that the genes most representative of the module’s overall expression profile were those most strongly related to the behavior ( Figure 3C ) . Connectivity is the core gene coexpression network concept and genes with high connectivity have the strongest coexpression relationships across the entire network , indicating greater importance to overall network structure and biological significance . The purple , green , and pink modules contained the most densely interconnected genes ( Figure 3—figure supplement 5 ) , and were correlated to percentage similarity to tutor ( green learning-related module ) or singing-induced variability ( purple and pink variability-induction modules ) ( Figure 3B and D ) . These findings indicate that information about the relationships between gene coexpression and behavior was reflected in the structure of the network: A gene’s relationship to a module or a module’s relationship to the network was predictive of strong behavioral relevance . Therefore , we examined the most well-connected/hub genes within the context of their module ( genes with the greatest intramodular connectivity ) or the entire network ( genes with the greatest whole-network connectivity ) . We discovered that many of these hub genes are known risk genes for human disease . For example , of the 7462 genes in the overall network , Fragile X Mental Retardation 1 ( FMR1 ) had the third highest connectivity and was the most well connected member of the green module ( Supplementary file 1 ) . Deficiency in FMR1 gives rise to Fragile X Syndrome , a genetic disease with a multitude of symptoms including intellectual deficiency and speech and language impairment . To attribute biological meaning to the modules , we calculated a module significance score for the resulting disease , gene ontology , and pathway annotations returned from GeneAnalytics ( Ben-Ari Fuchs et al . , 2016 ) ( See Materials and methods ) . The top five terms for the black song production module ( negatively correlated to the amount of singing ) , the brown variability induction module ( positively correlated to variability induction ) , and green learning-related module ( positively correlated to learning ) are shown in Figure 3E with comprehensive results presented in Supplementary file 2 . Since most modules contain hundreds of genes , prioritizing the ontology terms by the connectivity of their annotated genes allows genes with the greatest network importance ( Figure 3F ) to emphasize the terms with the greatest biological importance ( Figure 3E ) . To validate the specificity of the Area X modules to vocal behavior , we compared the overall Area X network to a network constructed from the adjacent non-song VSP ( Hilliard et al . , 2012a; Feenders et al . , 2008 ) from the same animals . Area X and VSP networks were constructed using the genes that were common to the two , enabling analysis using module preservation functions . We hypothesized that the genes in the Area X song production modules would have no correlation to behavior in VSP since , despite its close proximity and similar cell type composition , the VSP is not similarly linked into song control circuitry ( Person et al . , 2008 ) . Moreover , a body of evidence suggests that the song control circuit evolved as a specialization of existing motor circuitry ( Pfenning et al . , 2014; Feenders et al . , 2008; Barrett , 2012; Oakley and Rivera , 2008 ) . As predicted , no module in the VSP network displayed any correlation to any of the singing or learning behaviors as gene significances using Area X and VSP expression data are markedly different ( Figure 4A , X vs . V ) . We calculated module preservation statistics between the two brain regions and observed that the song production modules were among the most poorly preserved ( Langfelder et al . , 2011 ) across the two networks ( Figure 4B , Supplementary file 3 ) . This result indicates differential connectivity of song production module genes between Area X ( Figure 4C , top ) and VSP ( Figure 4C , bottom ) , further underscoring that Area X is specialized for song . This lack of preservation was not the product of differential gene expression between the two regions ( Figure 4D , top ) but instead reflected altered connectivity among similar genes ( Figure 4D , bottom ) . In striking contrast to the song production modules , the green learning-related module was strongly preserved in VSP ( Figure 4B , Figure 3B ) , indicating a generalized learning-related coexpression state exists in the juvenile striatopallidum that is specialized for singing in Area X . To provide further context for the modules observed in our overall network and how they relate to learned vocalization , we compared them with prior data from adult zebra finch Area X ( Hilliard et al . , 2012a; Hilliard et al . , 2012b ) . Our present network captures a point in zebra finch development when birds are actively learning how to improve their songs whereas in adulthood , the learning process has ended and adult songs are ‘crystallized’ . Contrasts between juvenile and adult networks highlight gene coexpression patterns that change between the two learning states , and inform their molecular underpinnings . Our previous study in adults found multiple modules in Area X that were correlated to singing crystallized songs . We reasoned that if highly similar coexpression patterns were present in juveniles , then they would likely be unrelated to learning . In this case , the capacity to learn a song might be attributable to other genes and/or the relationships between them . To compare across studies , we built two new , age-specific networks composed of genes common to the two original networks , then computed gene significance scores for all genes in both networks . We found a remarkable correlation between gene significances to singing in juveniles and adults ( Figure 5A ) , showing that genes in Area X shared similar relationships to singing , whether it be positive , negative , or nonexistent , independent of the animal’s age and learning state . The replicated discovery of specific sets of song-production genes across studies and ages speaks to the profound effect that singing behavior has on gene transcription profiles within the song-dedicated basal ganglia . We next calculated module preservation across the two studies , which assesses how well the coexpression relationships between genes persist across ages ( Langfelder et al . , 2011 ) . We observed strong to very strong relationships between module preservation and correlation to singing , and genes related to singing clustered together independent of age ( Figure 5B and C , Supplementary file 4 ) . These results indicate that not only are the relationships between genes and singing consistent across ages but those genes’ coexpression patterns are preserved as well . Since singing-driven gene coexpression patterns were similar between juvenile and adult Area X , the capacity to learn vocalizations is not a product of large-scale differences in coexpression of the song production module genes . We therefore looked for any modules that differed between juvenile and adult Area X . We found that the green , greenyellow and darkred learning-related modules that were significantly correlated to tutor similarity in juveniles were poorly preserved in adult Area X ( Figure 5B and C , Supplementary file 4 ) . Irrespective of preservation between juvenile and adult Area X , the genes in song production and learning-related modules were similarly activated by singing ( Figure 5D , top row ) and the ranked gene expression within each module displayed a positive correlation across ages ( Figure 5D , middle row ) . However , only the song production modules showed positive correlations between connectivity in juvenile and adult Area X ( Figure 5D , bottom row ) . These results attribute the difference between juvenile and adult Area X not to differential expression or altered correlation to behavior , but to differential connectivity in adults of modules that are correlated to tutor similarity in juveniles . Our findings suggest that the capacity to alter vocalizations may not reside in the absolute expression level of a given gene but instead the gene’s transcriptional context . For example , FMR1 was poorly connected in the adult network but was positioned as a hub gene in the juvenile network , indicating the gene’s importance during a developmental period when vocalizations are being actively modified but not during their maintenance . In general , genes that were positively correlated with learning and/or had high module membership in the green learning-related module had the greatest decrease in connectivity in adulthood ( Figure 5—figure supplement 1 ) . Above we describe two classes of coexpression modules: ( 1 ) learning-related modules that are preserved throughout the striatopallidum but present only in juveniles , ( 2 ) song production modules that are preserved across age but specific to Area X . Therefore , song production modules and learning-related modules exist simultaneously only in juveniles , and their co-occurrence within Area X may reflect the capacity to dramatically alter vocalizations during sensorimotor learning . Therefore , we hypothesized that interactions between these two modules may drive the vocal learning process . To test this idea using bioinformatics , we examined any genes linked to FoxP2 , whose overexpression drove the broad range of tutor song copying in our animals . The gene with the greatest gene significance to learning was MAPK11 ( Figure 6A and B ) . Interestingly , in Foxp2 heterozygous knockout mice , MAPK11 levels increase , supporting the interaction we observed here ( Enard et al . , 2009 ) . To examine whether MAPK11 could be a target of FoxP2 in the zebra finch , we scanned the MAPK11 gene for sequences corresponding to the FoxP2 binding motif from the JASPAR database ( see Materials and methods ) ( Nelson et al . , 2013; Mathelier et al . , 2016 ) . We found a match with a single base difference beginning 288 base pairs upstream of the zebra finch MAPK11 transcription start site identified in the RefSeq model ( Figure 6C ) . ( Note that the RefSeq model may be incomplete; see MAPK11 annotation note in Materials and methods ) . We then used chromatin immunoprecipitation followed by PCR ( ChIP-PCR ) to test whether or not FoxP2 binds this predicted MAPK11 regulatory region . Chromatin-immunoprecipitation of FoxP2 enriched a MAPK11 fragment of the predicted size and encompassing the putative FoxP2 binding site . Moreover , the sequenced fragment contains the FoxP2 binding motif ( Figure 6D , Figure 6—figure supplement 1 ) . Taken together , these data suggest that birds overexpressing FoxP2 . FL may be limited in their capacity to learn due , at least in part , to FoxP2 regulation of MAPK11 . In line with this , both the FoxP2 . 10+ and GFP animals had higher MAPK11 gene significance scores for tutor similarity than did FoxP2 . FL animals ( Figure 6A ) . A strength of WGCNA is the ‘guilt by association’ approach whereby genes in close network proximity to a gene of interest become candidates for a role in the same biological processes . With this in mind , we used MAPK11 as an entry point to pathways related to vocal learning . We first scanned for genes with high topological overlap with MAPK11 ( e . g . the closest network neighbors to MAPK11 ) . Many of these genes were well-connected members of the green learning module ( Figure 6E ) . One such gene , ATF2 ( formerly known as CREB2 ) , had the fifth highest green intramodular connectivity and third highest whole network connectivity ( Supplementary file 1 ) . ATF2 protein is necessary for proper development of the nervous system ( Reimold et al . , 1996 ) and serves a dual purpose in affecting transcription by binding to cAMP response elements and also by acetylating histones H2B and H4 ( Bruhat et al . , 2007; Kawasaki et al . , 2000 ) . Like FMR1 , ATF2 is poorly connected in the adult network ( Hilliard et al . , 2012a ) . While its role in development of the nervous system has been defined , no specific relationship between ATF2 and learned vocalization has been described . In our network , the ATF2 acetylation target histone H2B sorted into the blue song production module , which is strongly and positively correlated to the act of singing ( Figure 3B , Supplementary file 1 ) and acetylation of histone H2B at lysine five has been linked to learning and memory in rat hippocampus ( Bousiges et al . , 2013 ) . A pathway such as this represents an interaction between a network hub in a learning module ( ATF2 ) and a song production module gene ( histone H2B ) at a developmental time point at which the bird is actively learning its vocalizations . To generalize this strategy , we used the Search Tool for the Retrieval of Interacting Genes/Proteins ( STRING ) database ( Szklarczyk et al . , 2015 ) to identify additional interactions between learning-related network hubs and song production genes in Area X . We submitted genes from the green , greenyellow , and darkred learning-related modules and the black , blue , darkgreen , orange , and royalblue song production modules , then filtered for cross-module interactions and scaled the confidence scores by the average intramodular connectivity of each gene in the interaction . This yielded a ranked list of interactions between genes positively correlated to learning and those correlated to singing , which was prioritized by weighted confidence score to yield the highest confidence interactions between genes with the greatest network importance ( Supplementary file 5 ) . These interactions were plotted as a network with proteins as nodes and interaction scores as edges ( Figure 7 ) . This approach allowed us to not only visualize the confidence in gene interactions but also the local neighborhoods formed by the protein interaction network , emphasizing genes of potentially greater importance in the vocal learning process based on the number of interactions they have . We ranked interactions by four different metrics designed to emphasize or deemphasize gene significance , intramodular connectivity , and differential connectivity in juveniles vs . adults ( see Materials and methods ) . These metrics provide a basis for selecting protein-protein interactions based on the relationship to the genes and their most strongly correlated behavior , the coexpression network importance of the genes , or the change in connectivity between juvenile and adult birds . In using the latter metric , the decreased connectivity of learning-related genes ATF2 and FMR1 in adulthood is accounted for and interactions involving those genes are prioritized . Interactions between ATF2 and IRF2 , DUSP5 , and FOS are among the highest scoring interactions using this metric . All such interactions are presented in Supplementary file 5 . In addition to the overall Area X network presented above , we built and compared construct-specific networks from birds injected with the FoxP2 . FL expressing virus versus those injected with the FoxP2 . 10+ expressing virus versus those expressing GFP ( Figure 3—figure supplements 1–3 ) . This analysis enabled us to assess the level of construct-driven changes in gene coexpression as well as to test for the presence of the learning-related module in the control birds whose FoxP2 levels were unmanipulated . We quantified module preservation between the FoxP2 networks and the GFP network ( Figure 3—figure supplement 4 ) . In both FoxP2 networks , a gradient of module preservation was observed versus the GFP network with both overlapping and significantly different modules observed . Birds in these experimental conditions were siblings , and in some cases from the same clutch , suggesting that the driving effect of network differences is the construct-specific manipulation . The green learning-related module was well-preserved across the three networks . The strong correlation of this module to learning passed false discovery rate correction in the GFP cohort comprised of only seven birds , indicating that the learning-related coexpression pattern observed in the overall network is also present without FoxP2 manipulation . In this study , we overexpressed FoxP2 isoforms or GFP and thereby created a range of song learning and song variability induction ( Figure 2F ) , ideal for transcriptome profiling and WGCNA . We constructed an overall Area X gene network and discovered modules correlated to singing , learning , and vocal variability . The network properties of these modules revealed strong relationships between gene module membership and the behavior ( s ) to which the modules were correlated . To understand how gene coexpression patterns change across the boundary of the sensorimotor critical period for vocal learning , we compared the juvenile Area X overall network constructed here to one previously constructed from adult Area X ( Hilliard et al . , 2012a ) . We had competing hypotheses about whether the inability to learn new songs as an adult is resultant of changes to the song production modules observed in juveniles or associated with some other transcriptional change . Module preservation statistics revealed robust preservation of the juvenile Area X song production modules in the adult network , supporting the latter hypothesis . In striking contrast , the densely interconnected green learning-related module observed in juvenile striatopallidum was poorly preserved in adults , indicating that at least part of the learning-related transcriptome is altered by aging . Further , the green learning-related module was strongly preserved across the construct-specific networks ( Figure 3—figure supplements 1–4 ) and robustly correlated to learning in the GFP network . This latter finding suggests that the coexpression of these genes occurs in non-manipulated birds and is not a byproduct of experimental perturbation of FoxP2 levels . Because we created networks from VSP of the same animals , we could compare how well the Area X modules were preserved in a similar brain region that is unspecialized for song . As in Hilliard et al . ( 2012a ) , Area X song production modules were poorly preserved in VSP in contrast to the strongly preserved green learning-related module . These experiments define juvenile Area X as a nexus wherein the striatopallidal learning-related modules exist in tandem with song production modules . As the brain ages , singing continues to drive transcriptional patterns in Area X but the learning-related patterns are lost ( Figure 8A; Figure 8B ) . Our findings suggest a model for the molecular basis of complex learned vocal behavior as -- not specific genes or coexpression modules -- but rather the spatiotemporal overlap of ‘singing’ and ‘learning’ building blocks . Song control nuclei are proposed to have evolved as specializations of pre-existing motor circuitry ( Pfenning et al . , 2014; Feenders et al . , 2008 ) . A similar principle may thus extend across the songbird telencephalon whereby nonspecialized/learning related and specialized/behavior related coexpression patterns converge to permit sensorimotor learning . Our findings validate prior results in which overexpression of FoxP2 . FL prevented practice-induced changes in song variability and impaired song learning . These results support the hypothesis that behavior-linked cycling of FoxP2 , rather than its absolute level , is critical for vocal learning . In addition , we uncovered singing-induced vocal invariability as a novel behavioral effect of FoxP2 . 10+ overexpression . Despite the poor exploration of motor space induced by FoxP2 . 10+ overexpression , these animals learned their tutors’ songs well , a finding seemingly at odds with motor learning theory where broad exploration of motor space is refined through practice before arriving at an ‘ideal’ precise pattern for execution of the skill ( Kaelbling et al . , 1996; Wu et al . , 2014 ) . A similar phenomenon was observed in a different species of passerine songbird , the Bengalese finch ( Lonchura striata domestica ) , where two hours of UD singing resulted in less variable songs than those sung after two hour of non-singing ( Chen et al . , 2013 ) . In both species , the inability to induce song variability did not affect vocal learning , suggesting that the ability to have relatively low or high variability states in singing are necessary to properly learn a song regardless of whether those differential variability states precede or follow singing . WGCNA identified FMR1 as a gene of great importance in a learning module . FMR1 encodes an RNA-binding protein and therefore its levels could have a profound effect on a number of targets in the network ( Ascano et al . , 2012 ) . FMR1 protein is expressed throughout the zebra finch song control circuit primarily in neurons , and birdsong has been suggested as an interesting model in which to study the gene’s function ( Winograd and Ceman , 2012; Winograd et al . , 2008 ) . Here , we observed a correlative link between FMR1 expression and how well the animal copied its tutor’s song , a novel association that could be reasonably hypothesized given the speech and language phenotype associated with FMR1 deficiency in humans . A key strength of WGCNA is the ability to query the network around genes known to be associated with a trait . FMR1’s close network neighbors included ATF2 which has been associated with learning but has no prior link to vocal behavior . Further investigation into the learning-related modules is likely to reveal pathways fundamental to procedurally learned behavior . To identify those molecules that may interact at this particular developmental time point and brain region , we selected MAPK11 – a likely FoxP2 target ( Enard et al . , 2009 ) and the gene with the greatest significance to learning – to further investigate as an entry point to the pathways underlying learning behavior . Local neighborhood analysis of MAPK11 in the coexpression network revealed high topological overlap with many strongly connected members of green learning-related module , including the hub gene ATF2 . ATF2 is a phosphorylation target of MAPK11 and part of an evolutionarily-conserved pathway for learning and memory ( Guan et al . , 2003 ) . This phosphorylation enhances ATF2 histone-acetyltransferase activity ( Enslen et al . , 1998; Stein et al . , 1997 ) . A known enzymatic substrate of ATF2 is histone H2B ( Kawasaki et al . , 2000 ) , a member of the blue song production module that is positively correlated to singing . To probe for additional protein-protein interactions such as these , we mined the STRING database using song production and learning-related module members , then prioritized the interactions based on the network properties and/or behavioral significance of the input genes . A prioritized list of interactions and a complex network emerged , highlighting genes based on their coexpression network importance and/or the number of protein level interactions in the database ( Figure 7 , Supplementary file 5 ) . While there are differences in overall gene expression between the juvenile and adult brain , the context within which genes express , that is , their connectivity , is drastically altered , especially in the learning-related modules . Changes in connectivity are not necessarily indicative of changes in the absolute level of a gene’s expression , as evidenced by the comparisons between Area X and VSP ( Figure 4D ) or juvenile and adult Area X ( Figure 5D ) , where expression levels correlate positively but connectivity does not . These data support the idea that the coexpression patterns , and thereby the genes’ connectivity and network importance , contribute to the transition from a state of learning to a state of non-learning . In using connectivity as a measure of network importance and protein interaction as a measure of functional biological output , the protein interaction landscape underlying learned vocal behavior shifts across the two developmental time points analyzed here . For example , the local interaction network around green module hub ATF2 ( defined as all those neighbors within two steps and with high confidence of protein interaction ) is composed of well-connected genes in the learning-related and song production modules ( Figure 8C , top ) . Moreover , the connections to learning-related genes are , themselves , inputs to well-connected network hubs . As the juvenile crosses over into adulthood , the connectivity of many of the learning-related genes , like ATF2 , dramatically decreases . As part of the same process , the adjacencies between genes in the interaction network shift such that a connection to a learning-related gene is no longer one with a hub ( Figure 8C , bottom ) . This shift in network importance may present a pattern underlying song maintenance rather than song learning , and potentially the closure of the critical period in which the bird can change its song . To understand the mechanisms underlying the transition between the two learning states , our data highlight the importance of the network position of a gene . To enable vocal plasticity after critical period closure , a goal critically relevant to social and communication disorders , manipulations that coordinate gene expression such that poorly connected genes are reestablished as network hubs are likely required . Tools to accomplish a goal such as this do not yet exist , but the pathways prioritized and presented here provide a framework for teasing out testable components . In sum , we have described the Area X transcriptome at a developmentally significant point in the vocal learning process and provided context for it in terms of aging and brain region specificity . We suggest numerous coexpression and protein level interactions that our data indicate are significant to vocal learning . Due to the large amount of data generated by this study , we provide interactive graphics describing the coexpression and protein interaction networks as a supplement to the figures and tables in the manuscript . These , and the compiled descriptive statistics are hosted at ( https://www . ibp . ucla . edu/research/white/genenetwork . html ) . We encourage exploration of these datasets to confirm or refute their validity and to provide the molecule-to-behavior links suggested herein . All animal use was in accordance with NIH guidelines for experiments involving vertebrate animals and approved by the University of California , Los Angeles Chancellor’s Institutional Animal Care and Use Committee . Birds were selected from breeding pairs in our colony . The experimental timeline is schematized in Figure 2A . Breeding cages that contained candidate experimental birds were placed in sound attenuation chambers along with their parents and siblings when juveniles reached ~20 d , as in Heston and White ( Heston and White , 2015 ) . Chambers were continuously recorded so as to capture tutor song . At 30d , juvenile males were bilaterally injected with AAV1 into Area X to overexpress either FoxP2 . FL , FoxP2 . 10+ , or GFP , then returned to their chambers . At 40d , juvenile males were isolated from all other birds and continuously audio-recorded . At ~60 d , an ‘NS-UD’ experiment was performed according to the methods of Miller et al . , Chen et al . , and Heston et al . ( Heston and White , 2015; Miller et al . , 2010; Chen et al . , 2013 ) to assess the induction of vocal variability . On the ‘NS-UD’ day , for the first two hours after lights-on , birds were distracted by gentle ‘shushing’ if they attempted to sing . ( Those that sang >10 motifs were excluded from that day’s experiment ) . On the ‘UD-UD’ day , birds were allowed to sing UD song for the first two hours after lights-on . The level of variability in songs sung subsequent to those two hours was quantified . At 65d , birds were sacrificed following two hours of UD singing with one exception: In order to assure a broad range of song amounts immediately preceding sacrifice ( and thereby capture a range of singing-induced gene expression ) , we distracted one bird in the GFP group from singing during the two hours preceding sacrifice . A total of 19 birds received stereotaxic injections with AAV ( 7 GFP , 6 FoxP2 . FL , 6 FoxP2 . 10+ ) . Sample size was based on numbers used in Heston and White ( Heston and White , 2015 ) where 5–8 animals per group were sufficient to reveal treatment effects . The authors of the WGCNA R package recommend a minimum of 15 samples for building a network ( https://labs . genetics . ucla . edu/horvath/CoexpressionNetwork/Rpackages/WGCNA/faq . html ) , so we ensured at least five animals in each of the three groups . Countryman EMW or Shure SM93 omnidirectional lavalier microphones were used to continuously record birds from ~20 d until sacrifice ( 65d ) . Sounds were digitized using PreSonus FirePod or PreSonus Audioboxes at a 44 . 1 kHz sampling rate and 24-bit depth . Recordings were managed by SAP 2011 software ( Tchernichovski et al . , 2000 ) . To quantify levels of FoxP2 . FL , we selected a primer pair previously used to quantify FoxP2 knockdown ( Haesler et al . , 2007; Olias et al . , 2014 ) . The forward sequence was 5’-CCTGGCTGTGAAAGCGTTTG-3’ and the reverse was 5’ATTTGCACCCGACACTGAGC-3’ . We designed a primer pair for FoxP2 . 10+ using the NCBI Primer-BLAST tool ( Ye et al . , 2012 ) . The input sequence was FoxP2 . 10+ mRNA CDS ( GenBank accession DQ285023 . 1 ) . The forward primer sequence was 5’-CGCGAACGTCTTCAAGCAAT-3’ and the reverse sequence was 5’-AAAGCAATATGCACTTACAGGTT-3’ . Primer specificity was determined by obtaining a single peak in melting curve analysis and obtaining a single amplicon of predicted size following qPCR . GAPDH forward and reverse primers were 5’-AACCAGCCAAGTACGATGACAT-3’ and 5’-CCATCAGCAGCAGCCTTCA-3’ , respectively . 200 ng of RNA from Area X micropunches was reverse transcribed into cDNA using the Bio-Rad iScript cDNA Synthesis Kit ( Hercules , CA , USA ) . 25 µL qPCR reactions were assembled in MicroAmp Optical 96-Well Reaction Plates ( ThermoFisher Scientific ) . Reaction components were 0 . 5 µL cDNA , 200 nM primers , 12 . 5 µL PowerUp SYBR Green Master Mix ( ThermoFisher Scientific ) , and 10 . 75 uL nuclease-free water . Cycling conditions were 50°C for 2 min , 95°C for 2 min , then 40 cycles of 95°C for 15 s and 60°C for 1 min . A dissociation step of 95°C for 15 s , 60°C for 1 min , 95°C for 15 s , and 60°C for 15 s was then performed . All reactions were run in triplicate and all samples for an individual animal were run together on the sample plate . FoxP2 expression was quantified relative to GAPDH and normalized to the GFP-injected animals using the 2-Δ ΔCT method ( Livak and Schmittgen , 2001 ) . For histological analyses , animals were sacrificed 3–5 days following HSV injection then perfused with warm saline followed by ice cold 4% paraformaldehyde in 0 . 1 M phosphate buffer . Tissue was cryosectioned at 20 µM , thaw-mounted onto glass microscope slides , and stored at −80°C until use . Thawed sections were incubated overnight with goat-anti-FoxP2 ( 1:500; Abcam , Cambridge , UK; [Thompson et al . , 2013] ) and mouse-anti-Xpress ( 1:500; ThermoFisher Scientific , Waltham , MA ) . AlexaFluor 546 donkey-anti-goat ( 1:500 ) and AlexaFluor 405 donkey-anti-mouse ( 1:250 ) secondary antibodies were used to generate anti-FoxP2 and anti-Xpress signals , respectively . Sections were visualized using a Zeiss ( Oberkochen , Germany ) LSM 800 confocal microscope and processed using NIH ImageJ ( Schneider et al . , 2012 ) . The calculation of effect size was performed because it allows for comparison across virus groups instead of a series of paired comparisons within group ( Miller et al . , 2015 ) . The transformation normalizes acoustic features so that any observed changes are viewed in the context of the initial values . We present a hypothetical example in the table below where a change of 50 Hz for two syllables is given a greater weight for a syllable that has an overall lower frequency when using the transformation we applied for our song data: Syllable ASyllable B NSUDRaw delta ( NS-UD ) / ( NS + UD ) NSUDRaw delta ( NS-UD ) / ( NS + UD ) 100 Hz150 Hz50 Hz−0 . 2500 Hz550 Hz50 Hz−0 . 048 Two hours following lights-on at ~65 d , birds were sacrificed by decapitation . Brains were rapidly extracted and frozen on liquid nitrogen , then stored at −80°C until all brains were collected . As in Hilliard et al . ( 2012a ) , tissue micropunches of Area X and VSP were performed . Brains were coronally sectioned on a cryostat at 30 µM until Area X became visible . Area X and outlying VSP were punched using a 20 gauge Luer adapter and stored in RNAlater ( Qiagen , Germantown , MD ) at −80°C until RNA extraction was performed . 30 µM sections were then collected , thaw mounted , and thionin stained for post-hoc validation of punch accuracy . Total RNA extraction was performed as in Hilliard et al . ( 2012a ) . Samples were processed semi-randomly and in parallel with another sequencing project . Tissue punches from both studies were processed in batches of 8 . We used Qiagen RNeasy Micro Kits ( Cat . No 74004 ) following the manufacturer’s protocol and QIAzol as the lysis reagent . An additional wash beyond the manufacturer’s protocol was performed in RW1 and RPE buffers . Final elution volume was 20 µL . Extracted total RNA were stored at −80°C until all RNA extractions were completed . All extractions were completed over the course of two weeks . Total RNA was provided to the UCLA Neuroscience Genomics Core ( UNGC; https://www . semel . ucla . edu/ungc ) where RNA quality was assessed on an Agilent TapeStation ( Agilent Technologies , Santa Clara , California ) . RNA of sufficient quality ( RIN >8 ) was then used to generate cDNA libraries using the Illumina TruSeq Stranded Poly-A Prep Kit ( Illumina , San Diego , CA , USA Cat No 20020594 ) . Libraries for each sample were divided across two lanes and sequenced in a total of 8 lanes using an Illumina HiSeq 2500 in high output mode , generating between 15 and 35 million 50 bp paired-end reads per library . Raw FASTQ files furnished by UNGC were first quality controlled using FASTQC ( http://www . bioinformatics . babraham . ac . uk/projects/fastqc/ ) . FASTQC returned results indicating high quality across all bases in each read in each sample and no adapter contamination was detected , therefore we did not perform any filtration of the reads before alignment . Reads were aligned to the NCBI zebra finch genome assembly 3 . 2 . 4 ( http://www . ncbi . nlm . nih . gov/assembly/524908/ ) and RefSeq annotations using STAR ( Dobin et al . , 2013 ) . Mismatch tolerance was two base pairs . Only uniquely mapped reads were considered in downstream analyses . The featureCounts ( ) function in the Rsubread R package was used to count all reads mapping within exon features , then all exon counts were summed to the gene level so that each gene had a single value of reads mapped to it ( Liao et al . , 2014; Liao et al . , 2013 ) . Gene expression was then quantified by calculation of transcripts per million ( TPM ) . TPM values were log2 transformed and genes with zero variance across samples were removed . We checked for batch effect on average expression resultant of RNA extraction group , RNA extraction experimenter , and across sequencing lanes . No batch effects were observed . We used an iterative process of removing gene expression data from single samples whose expression was >2 . 5 SD of that gene’s expression across all samples , repeating until no samples remained with expression >2 . 5 SD away from the gene’s average expression across all samples . Finally , we calculated the intrasample correlation ( ISC ) and used a hard cutoff of 2 SD away from the group ISC for removal of samples from the study . No sample in any group ( Area X or VSP ) was >2 SD from the group ISC . Data were quantile normalized as the last step . Final data input to WGCNA was 13665 and 13781 genes for Area X and VSP networks , respectively , across 19 total samples . We calculated the soft thresholding power for construction of the WGCNA adjacency matrix using the pickSoftThreshold function in the WGCNA R package at 18 for Area X and 14 for VSP . We then constructed a signed network using the blockwiseModules function in the WGCNA R package . For the Area X network , we used a minimum module size of 100 genes and deepSplit was set equal to four for Area X and two for VSP . Genes were required to have at least a connectivity of 0 . 3 with their module eigengene in order to remain a member of their module and the module ‘core’ ( =minimum module size/3 ) needed to have a minimum eigengene connectivity of 0 . 5 for the module to not be disbanded . All other parameters were set to default . Networks were iteratively constructed with genes in the grey module removed from the expression data after each round of network building and module definition . The networks were considered final after no genes were placed into the grey module . During network construction , FoxP2 was removed , presumably due to the lack of coexpression with other genes in the network resulting from virus-driven overexpression . Therefore , we added FoxP2’s expression data back into the final overall network and it became the only gene in the grey module . Once coexpression modules were defined , we correlated vocal behavior to the module eigengenes . Since the grey module included only a single gene with no significant behavioral correlations , it was excluded from module-trait analyses . WGCNA is a well-established technique for gleaning biologically relevant clusters of coexpressed and functionally related genes from microarray and sequencing data . WGCNA methods and terminology are summarized and defined in numerous manuscripts ( Hilliard et al . , 2012a; Zhang and Horvath , 2005; Dong and Horvath , 2007; Zhao et al . , 2010; Yip and Horvath , 2007; Horvath , 2011 ) . For the sake of convenience , we provide working definitions of network terms that we use throughout the manuscript . Definitions of greater detail are available in the manuscripts cited above . Calculation of gene significance to a trait requires the definition of a single value to which the amount of gene expression in each sample is correlated . Gene significances were calculated for the following traits: Motifs , defined as the number of motifs each animal sang in the two hours following lights-on on the day of sacrifice; Tutor similarity , defined as the percentage similarity between the pupil and its tutor on the day of sacrifice; Variability induction , defined by inserting Wiener entropy CV scores into the equation ( NS-UD ) / ( NS + UD ) from the first twenty syllable renditions sung during the NS-UD experiment performed at ~60 d; Motif identity , defined as the product of the similarity and accuracy scores divided by 100 of the last 20 motifs sung by each bird before sacrifice . Song variability was assessed on the motif level for the purpose of gene significance calculations so as to obtain a single value for each animal . Following network construction , modules were summarized by calculating a module eigengene , defined as the first principal component of the module’s expression data using the moduleEigengenes ( ) function in the WGCNA R package . The relationship between a module and a behavior was assessed by determining the Pearson correlation between the module eigengene and continuous behavioral traits as defined in ‘Song Analysis and Statistics’ , above . Significance was then determined by calculating the Fisher transformation of each correlation using the corPvalueFisher ( ) function in the WGCNA R package . We performed p-value corrections for module-trait correlations using the p . adjust ( ) function with the number of comparisons equal to the number of traits ( 4 ) by the number of modules ( 21; the FoxP2-only grey module was not included for purposes of p-value correction ) . The p-values presented in this manuscript are uncorrected for multiple hypothesis testing but those that pass FDR-correction at p<0 . 05 are indicated . We chose to present uncorrected p-values due to the small sample size used to create the overall network ( n = 19 birds ) . The authors of WGCNA suggest a minimum of 15 samples with >20 preferred ( https://labs . genetics . ucla . edu/horvath/CoexpressionNetwork/Rpackages/WGCNA/faq . html ) . P-value corrections drive nearly all results to insignificance , including well preserved module-trait relationships that are present in adults and survive such corrections due to the larger sample size in that study . We use the significant but uncorrected p-values in this study as a guide toward interesting module-trait relationships , then use the properties of the network to inform the downstream analysis . Our choice of behavioral traits for correlation to the gene network was hypothesis-driven . In addition to the obvious quantification of vocal learning , the comparison for variability induction was planned , as indicated by the fact that we conducted the NS-UD and UD-UD behavioral paradigms ( prior to the bird’s sacrifice ) that led to it . We originally used these paradigms as a method for naturally regulating FoxP2 levels , before we had identified a virus that was effective in doing so . In that study ( Miller et al . , 2010 ) , our prediction was that behavioral conditions that lead to low endogenous FoxP2 in Area X ( namely 2 hr of UD singing ) , would be associated with higher levels of variability . This was indeed the case . We replicated this finding in zebra finches ( Heston and White , 2015 ) but did not observe the same phenomenon in Bengalese finches ( Chen et al . , 2013 ) as noted in our Discussion . The feature highlighted by those studies was Weiner entropy . At the time of this study , annotation of the zebra finch genome is relatively sparse , thus zebra finch gene symbols were converted to their Human Genome Organisation ( HUGO ) Gene Nomenclature Committee ( HGNC ) paralogs , then submitted to GeneAnalytics , a comprehensive tool for the contextualization of gene set data that integrates across multiple databases ( Ben-Ari Fuchs et al . , 2016 ) . Genes with no known human homolog were excluded . Symbols were submitted to the GeneCards GeneAnalytics suite at http://geneanalytics . genecards . org ( Ben-Ari Fuchs et al . , 2016 ) . GeneCards enrichment scores were converted into p-values , which were used as the input to module significance calculations . Module significance of a term was defined as the product of the average module membership for each gene annotated with a term , and one minus the p-value for that term such that the genes with the highest module membership and lowest p-value prioritize the terms ( Hilliard et al . , 2012a ) . Term significance was defined by weighting the module significance score by the gene significance for a given behavioral metric . The FoxP2 consensus binding sequence from the JASPAR database ( Nelson et al . , 2013; Mathelier et al . , 2016 ) was converted into a position-weight matrix ( PWM ) and used to scan the promoter ( defined as the first 1000 base pairs upstream of the transcription start site in the RefSeq models ) for each gene in the zebra finch genome . Putative FoxP2 binding sites were identified using the matchPWM function in the Biostrings R package ( https://bioconductor . org/packages/release/bioc/html/Biostrings . html ) with a minimum hit score of 80% . The MAPK11 region discussed in this manuscript was identified using methods described above . Upon closer inspection of the MAPK11 RefSeq annotation model , we believe the identified region does not lie within the promoter but instead within an intronic region of MAPK11 . There is currently no experimental evidence to verify the RefSeq model’s predicted transcription start site and the Ensembl model for MAPK11 is considerably longer ( 323 residues vs . 285 residues ) due to an expanded N-terminus region . Further , the chicken MAPK11 RefSeq model is 361 residues and contains an N-terminus residue ( MSERGGFYRQELNKTVWEVPQRYQNLTPVGSGAYGSVC ) that maps ~12 kb upstream of the second exon of chicken MAPK11 . This residue does not map to the zebra finch genome , presumably because a gap in the genome exists ~13 kb upstream of zebra finch MAPK11 . The MAPK11 N-terminus peptides of other songbird species ( Bengalese finch , starling , white-throated sparrow , great tit and Tibetan ground-tit ) are highly similar to that in chicken and align to the first exon in chicken MAPK11 . This peptide is found in mice and humans , indicating high conservation . We thus posit that the MAPK11 RefSeq annotation in zebra finch is incomplete on the 5’ end and that we are reporting a binding site internal to MAPK11 and not at the promoter . Chromatin immunoprecipitation ( ChIP ) was performed using ChIP-IT High Sensitivity ( Active Motif , Carlsbad , CA , USA , Cat . No . 53040 ) following the manufacturer’s protocol . Whole brain was isolated from an adult male zebra finch , minced , and crosslinked in a formaldehyde solution . The tissue was homogenized with a hand-held tissue homogenizer for 45 s at 35 , 000 rpm . Following homogenization , the sample was sonicated at 25% amplitude 30 s on , 30 s off , for 10 min . A portion of the sonicate was de-crosslinked and quantified . The sample was split evenly into three tubes . A cocktail of anti-FoxP2 primary antibodies were applied to one sample ( Millipore , Billerica , MA , USA Cat . No . ABE73 , ThermoFisher Scientific Cat . No . 5C11A2 , and Abcam ab16046 ) , IgG in another ( Millipore 12–370 ) , and the third was input DNA . After an overnight incubation , the samples were washed , de-crosslinked and subjected to PCR . The ‘promoter’ sequence for MAPK11 was binned into 100 bp regions for primer construction . MAPK11 primers were as follows: forward 5’- CCCTTTCCCCAAATGGCAGA-3’ and reverse 5’-TATGAGCCTTGCCTTGGAGC-3’ . PCR protocol was performed using DreamTaq PCR Master Mix per manufacturer’s protocol . A PCR protocol was used as follows: ( 1 ) 95°C 1 min , ( 2 ) 95°C 30 s , ( 3 ) 67°C 30 s , ( 4 ) 72°C 1 min , repeat ( 2-4 ) for 40 cycles , ( 5 ) 72°C 10 min . PCR products were run on a 1 . 5% agarose gel in the presence of SYBR Safe to allow visualization of DNA . PCR products were purified ( QIAQuick Gel Extraction Kit ) and sent for sequencing by Laragen , Inc . Reverse primers sent for sequencing are as follows: 5’-TATGAGCCTTGCCTTGGAGC-3’ and 5’-CCTATGAGCCTTGCCTTGGA-3’ . STRING is a comprehensive database of known and predicted protein-protein interactions derived from experimental data , coexpression data , automated text mining , and also pulls information from other interaction databases . STRING accepts gene symbols as input , then mines for interactions between those genes and assigns a confidence score between 0 and 1 based on the evidence in the database for the genes’ interaction . We submitted gene symbols for the human homologs of module members to STRING then operated on the highest confidence interactions ( ≥0 . 9 ) in downstream analyses . Interaction scores were scaled by different metrics to emphasize or deemphasize network position and/or relationship to behavior ( Supplementary file 5 ) . Those metrics are: Network plots presented in this manuscript were constructed using the freely available plotting software , Gephi ( https://gephi . org ) , using edge lists prepared in R and exported in the . GEXF format . We have created interactive versions of many of the network plots in this manuscript ( Figure 3F ) all additional Area X modules ( similar to Figure 3F but not presented in the manuscript ) , and the protein interaction network presented in Figure 7 . They are hosted at our laboratory website ( https://www . ibp . ucla . edu/research/white/genenetwork . html ) along with high resolution static PDF versions . Interactive figures were exported from Gephi using the Sigma . js Exporter plugin ( https://github . com/oxfordinternetinstitute/gephi-plugins ) . In weighted coexpression networks , each node ( i . e . gene ) is connected to every other node in the network , even if the weight of the edge ( i . e . connection ) is zero . Therefore , plots depicting nodes and their edges with other genes become exceedingly complicated and unintuitive if all nodes and edges are included . In an effort to sparsify the networks and present the most salient data , we removed edges and genes from the coexpression networks using the following workflow: first , remove ≤98% of edges , then remove all disconnected nodes , then remove all nodes that are not part of the network’s main component ( e . g . the largest group of connected nodes ) . The remaining nodes and edges were plotted . In this manuscript , we present three types of network plots that look similar but convey different data . The three types are as follows: Raw and processed RNA-seq and behavioral data for each bird are available at the Gene Expression Omnibus ( GEO; https://www . ncbi . nlm . nih . gov/geo/ ) at accession number GSE96843 .

Songbirds , much like in humans , have a critical period in youth when they are best at learning vocal communication skills . In birds , this is when they learn a song they will use later in life as a courtship song . In humans , this is when language skills are most easily learned . After this critical period ends , it is much harder for people to learn languages , and for certain bird species to learn their song . When birds sing every morning , the activity of a gene called FoxP2 drops , which causes a coordinated change in the activity of thousands of other genes . It is suspected that FoxP2 – and the changes it causes – could be a part of the molecular basis for vocal learning . FoxP2 is also known to play a role in speech in humans , and both birds and humans have a long and a short version of this gene . Previous research has shown that when the long version of the gene was altered so its activity would no longer decrease when birds were singing , the birds failed to learn their song . Moreover , humans with a mutation in the long version have problems with their speech . However , until now , it was not known if modifications to the short version had the same effect . Burkett et al . investigated whether there was a noticeable pattern in the effects of FoxP2 before and after the critical period in a songbird . The analysis found that during the critical period , a set of genes changed together as young birds learned to sing . This particular pattern disappeared as the birds aged and the critical period ended . Burkett et al . confirmed that when birds had the long version of FoxP2 altered , they were less able to learn . However , changing the short version of FoxP2 had little effect on learning but led to changes in the birds’ song . The genetic pathways identified in the experiments are known to be present in many different species , including humans . Related pathways have also been found to play a role in non-vocal learning in organisms as distantly related as rats and snails . This suggests that they could be acting as a blueprint for learning new skills . Few treatments for language impairments have been developed so far due to poor understanding of the molecular basis for vocal communication . The findings of this study could help to create new treatments for speech problems in people , such as children with autism or people with mutated versions of FoxP2 .

2018

FoxP2 isoforms delineate spatiotemporal transcriptional networks for vocal learning in the zebra finch

Protective signaling from the leukemia microenvironment leads to leukemia cell persistence , development of resistance , and disease relapse . Here , we demonstrate that fibroblast growth factor 2 ( FGF2 ) from bone marrow stromal cells is secreted in exosomes , which are subsequently endocytosed by leukemia cells , and protect leukemia cells from tyrosine kinase inhibitors ( TKIs ) . Expression of FGF2 and its receptor , FGFR1 , are both increased in a subset of stromal cell lines and primary AML stroma; and increased FGF2/FGFR1 signaling is associated with increased exosome secretion . FGFR inhibition ( or gene silencing ) interrupts stromal autocrine growth and significantly decreases secretion of FGF2-containing exosomes , resulting in less stromal protection of leukemia cells . Likewise , Fgf2 -/- mice transplanted with retroviral BCR-ABL leukemia survive significantly longer than their +/+ counterparts when treated with TKI . Thus , inhibition of FGFR can modulate stromal function , reduce exosome secretion , and may be a therapeutic option to overcome resistance to TKIs . Editorial note: This article has been through an editorial process in which the authors decide how to respond to the issues raised during peer review . The Reviewing Editor's assessment is that all the issues have been addressed ( see decision letter ) . TKIs have revolutionized the treatment of CML and have shown promise in AML , however development of resistance remains a problem . In CML , resistance develops in a minority of patients , and is most often caused by resistance mutations . However , some patients still develop resistance in the absence of known resistance mutations . In contrast , development of resistance in AML is the norm . Inhibitors of mutated FLT3 , which is present in about 30% of AML patients , are initially quite efficacious ( Smith et al . , 2012 ) . However , resistance to FLT3 kinase inhibitors in AML typically develops within a few months . In some cases , resistance is cell-intrinsic and due to secondary mutations in the activating loop of FLT3 that prevent drug binding ( Weisberg et al . , 2009 ) , however , resistance still develops in the absence of these mutations . Within the marrow microenvironment , leukemia cell survival can be mediated by extrinsic ligands that activate alternative survival pathways ( Smith et al . , 2017; Ghiaur and Levis , 2017; Wilson et al . , 2012 ) and over time can lead to development of intrinsic resistance mutations ( Wilson et al . , 2012; Traer et al . , 2012 ) . Bone marrow stromal cells provide a supportive structure and secrete cytokines that contribute to the normal hematopoietic stem cell niche , but can also protect leukemic cells from therapy ( Colmone et al . , 2008; Ayala et al . , 2009 ) . Initial studies into the mechanisms of resistance utilized normal marrow stroma ( Manshouri et al . , 2011 ) , but the stroma can be altered by leukemia , in a manner similar to development of cancer associated fibroblasts in solid tumors ( Paggetti et al . , 2015; Huang et al . , 2015; Huan et al . , 2015 ) . We found that fibroblast growth factor 2 ( FGF2 ) expression is increased in marrow stromal cells during tyrosine kinase inhibitor ( TKI ) therapy and protects leukemia cells ( Ware et al . , 2013; Traer et al . , 2014; Javidi-Sharifi et al . , 2015; Traer et al . , 2016 ) . FGF2 has also been shown to be essential for stress hematopoiesis after chemotherapy ( Itkin et al . , 2012; Zhao et al . , 2012 ) , suggesting that leukemia cells can hijack a normal marrow stress response for their own survival . Despite its important roles in physiology and pathology , several aspects of FGF2 biology remain poorly understood . FGF2 does not have a signal peptide and is not secreted through the canonical secretory pathway . Alternative mechanisms for secretion have been proposed , but how FGF2 is conveyed between two cells remains unclear ( Steringer et al . , 2015; Zacherl et al . , 2015 ) . Additionally , while recombinant FGF2 directly stimulates myeloid colony formation ( Berardi et al . , 1995 ) , there are also reports suggesting that FGF2 can indirectly regulate hematopoiesis by stimulating stromal cells to produce cytokines ( Avraham et al . , 1994 ) . We discovered that FGF2 is largely secreted in extracellular vesicles ( ECVs ) and exosomes from bone marrow stromal cells . ECVs are able to protect leukemia cells from the effects of TKI therapy . Furthermore , autocrine FGF2-FGFR1 activation in marrow stromal cells increases the secretion of FGF2-laden exosomes , indicating that exosome secretion is regulated in part by FGF2-FGFR1 signaling . Inhibition of FGFR1 can reverse this protective stroma-leukemia interaction and restore leukemia cell TKI sensitivity in the marrow niche . The human stromal cell line HS-5 expresses abundant FGF2 , in addition to other soluble cytokines such as IL-5 , IL-8 and HGF ( Roecklein and Torok-Storb , 1995 ) , and conditioned media ( CM ) from HS-5 is highly protective of leukemia cell lines . HS-5 CM was ultracentrifuged at 100 , 000 g to separate soluble proteins ( supernatant , S100 ) from ECVs and larger macromolecules ( pellet , P100 ) . We compared the protective effect of unfractionated CM , S100 , and P100 fractions on the viability of two leukemia cell lines: MOLM14 ( FLT3 ITD+ AML ) and K562 ( CML ) , in the presence of their respective TKIs , quizartinib ( AC220 , a highly selective and potent inhibitor [Zarrinkar et al . , 2009] ) and imatinib ( Figure 1A and B ) . The protective capacity of the S100 fraction was less than unfractionated CM , and protection was enriched in the concentrated P100 ECV fraction ( Figure 1 ) , indicating that a substantial protective component of HS-5 CM is mediated by ECVs . A more extensive profiling of protection is also shown in Figure 1—figure supplement 1 . To determine if ECVs produced by HS-5 cells are internalized by K562 and MOLM14 leukemia cells , K562 and MOLM14 cells were stained with a green lipophilic tracer ( DiO ) and incubated with HS-5 ECVs stained with a red lipophilic tracer ( DiI ) . Analysis by confocal microscopy showed that ECVs are indeed internalized by leukemia cells , although the exact mechanism of internalization is still under investigation ( Figure 1C and Figure 1—video 1 and 2 ) . FGF2 is highly expressed in the HS-5 stromal cell line but the related HS-27 expresses little FGF2 ( Figure 2A; Traer et al . , 2016 ) . We analyzed FGF2 in S100 and P100 fractions of both HS-5 and HS-27 by immunoblot ( Figure 2B ) . Little FGF2 was detected in the soluble protein fraction ( S100 ) , but FGF2 was enriched in ECVs ( P100 ) . Washing the ultracentrifuge tube with detergent liberated even more FGF2 ( detergent wash P100 ) , due to ECVs adhering to the ultracentrifuge tube . To compare FGF2 to other soluble cytokines , HS-5 CM was ultracentrifuged into S100 and ECVs , cytokines quantified by Luminex multiplex assay , and normalized to unfractionated CM ( Figure 2C ) . Pelleted ECVs were resuspended in 10% of the original CM volume , and the P100 bars in Figure 2B thus represent a 10-fold enrichment , although as shown in Figure 2B not all ECVs can be liberated from the ultracentrifuge tube . FGF2 was uniquely enriched in ECVs , whereas soluble cytokines such as stem cell factor , interleukin ( IL ) −6 , IL-8 , etc . were found primarily in the S100 fraction . HS-5 ECVs were further separated into microvesicles , exosomes , and insoluble extracellular matrix proteins ( ECM ) using a sucrose step-gradient to separate by density . FGF2 and cell compartment-specific molecular markers were probed by immunoblot ( Figure 2D ) . FGF2 was most highly enriched in the 15–30% sucrose interface , which also contained the exosome-specific marker CD9 . To determine if FGF2 was bound to the outside of ECVS , or contained within ECVs , proteinase K was used to digest proteins not enclosed by lipid membrane . Recombinant FGF2 , HS-5 or HS-27 ECVs , and intact HS-5 cells were incubated with proteinase K and probed for FGF2 by immunoblot ( Figure 2E ) . Recombinant FGF2 was completely degraded by proteinase K ( * indicates degraded fragments ) but intact FGF2 was detected in both HS-5 ECVs and control HS-5 cells . We repeated this experiment using purified HS-5 exosomes and again observed that a fraction of FGF2 was protected from digestion ( Figure 2F ) . Addition of 0 . 1% Triton X-100 disrupted the lipid membrane and resulted in complete digestion of all protein . We found a similar digestion pattern with the exosomal transmembrane proteins CD9 and tsg101 . We conclude that FGF2 is contained within ECVs and exosomes , however we cannot exclude that FGF2 may also be on the surface since partial FGF2 degradation was noted in intact HS-5 cells , ECVs and purified exosomes ( Figure 2E–F ) . Since HS-5 CM is more protective than HS-27 CM ( Manshouri et al . , 2011; Traer et al . , 2016; Weisberg et al . , 2008 ) , we suspected that ECVs may be more numerous in HS-5 CM . We chose several orthogonal methods to quantify vesicles in CM . First , we used nanoparticle tracking analysis to quantify and compare HS-5 and HS-27 ECVs ( Figure 3A ) . In parallel , we employed the Virocyt Virus Counter , a flow cytometry-based technique developed to detect viruses , which also works well to quantify ECVs ( Figure 3B ) . As a gold standard , negative stain transmission electron microscopy of purified HS-5 and HS-27 exosomes was also used to image and quantify exosomes by counting ( 30–100 nm diameter with cup-shape appearance characteristic for exosomes , Figure 3C ) . Finally , we used sucrose step-gradient fractionation of HS-5 and HS-27 ECVs to compare cell compartment and exosome-specific markers by immunoblot ( Figure 3D ) . Exosomes layer primarily at the 15–30% sucrose interface as indicated by exosomal markers CD9 and tsg-101 , and are increased in HS-5 cells compared to HS-27 . Interestingly the receptor for FGF2 , FGFR1 , was also found to localize preferentially with HS-5 exosomes . With all methods , we consistently observed greater than two-fold excess of vesicles produced by HS-5 compared to HS-27 cells ( see Figure 3—figure supplement 1 for additional data ) . Markers of nucleus ( lamin A/C ) , endoplasmic reticulum ( calreticulin ) and mitochondria ( Bcl-XL ) were located in the 45–60% interface containing larger microvesicles and apoptotic bodies . FGF2 is an autocrine signaling protein for stroma , but recombinant FGF2 also mediates paracrine protection of leukemia cells ( Traer et al . , 2016; Traer et al . , 2014 ) . Thus there are two potential mechanisms by which FGFR inhibition can attenuate protection of leukemia cells in the marrow microenvironment: ( 1 ) FGFR inhibitors block FGF2-mediated paracrine protection at the leukemia cells; and/or ( 2 ) FGFR inhibitors interrupt stromal FGF2-FGFR1 autocrine signaling to reduce secretion of protective FGF2-containing exosomes . To compare the relative effect of FGFR inhibition on autocrine and paracrine signaling , HS-5 cells were pre-treated with the FGFR inhibitor PD173074 ( Mohammadi et al . , 1998; Trudel et al . , 2004 ) for one week prior to collection of CM . CM collected from HS-5 cells pre-treated with PD173074 was significantly less protective than CM from an equal number of untreated HS-5 cells ( Figure 4A ) , providing evidence that interruption of FGF2-FGFR1 signaling affects subsequent protection of leukemia cells . In contrast , addition of PD173074 to untreated HS-5 CM only modestly attenuated protection of MOLM14 cells . We found similar results with K562 cells exposed to imatinib ( Figure 4—figure supplement 1 ) . Purified ECVs from HS-5 CM , which are enriched in FGF2 , were more sensitive to FGFR inhibition ( Figure 1—figure supplement 1 ) , however pre-treatment of HS-5 cells with PD173074 still had the greatest absolute reduction in protection . These results indicate that FGFR inhibitors overcome protection of leukemia cells primarily by directly altering secretion of FGF2-expressing stromal cells , making them significantly less protective . To further evaluate the effects of FGFR inhibition in stromal cells , HS-5 cells were evaluated for viability , morphology , and growth using HS-27 cells as comparison ( low FGF2 ) . HS-5 or HS-27 cells had little reduction in cell viability after 72 hr treatment with PD173074 ( Figure 4B ) , however HS-5 growth slowed dramatically over 15 days ( Figure 4C ) . HS-5 cells exposed to PD173074 changed morphology and became less refractile , larger , and more adherent ( Figure 4D ) . Cell size was quantified using CellProfiler software and PD173074 significantly increased HS-5 cell size ( Figure 4E ) . To evaluate FGF2 and FGFR1 expression in primary leukemia stroma , bone marrow aspirates from a series of leukemia patients were cultured ex vivo and FGF2 and FGFR1-4 expression quantified by RT-PCR ( Figure 4F ) . FGFR1 and FGF2 transcripts were the most highly expressed in primary stroma , and there was a strong positive correlation between FGFR1 and FGF2 expression ( Figure 4G , r2 = 0 . 5683 and p<0 . 0001 on nonparametric correlation ) . This indicates that FGF2 and FGFR1 expression are coordinately regulated in primary marrow stromal cells consistent with activation of an FGF2-FGFR1 autocrine loop . There were nine stromal cultures from AML patients with FLT3 ITD ( indicated with red dots ) , but most of them were newly diagnosed , and based upon our previous data we would not expect increased expression of FGF216 . Similar to our observations in cell lines described above , we also detected FGFR1 and FGF2 in ECVs derived from primary marrow stromal cultures ( Figure 4—figure supplement 2 ) . However , primary marrow stromal cells grow slowly and produce smaller amounts of ECVs , so we were unable to evaluate the effect of FGFR inhibitors on cell morphology , growth , and ECV production with primary marrow stromal cells . Additional characterization of primary stromal cultures is contained in Figure 4—figure supplement 3 . Since FGFR inhibition attenuates HS-5 growth and morphology , we hypothesized that it might also reduce secretion of ECVs . HS-5 cells exposed to graded concentrations of PD173074 and BGJ-398 had a dose-dependent decrease in ECVs measured by Virocyt Virus Counter ( Figure 5A , B ) . Notably , there was a significant decrease in vesicle number as early as 6 hr after drug exposure ( Figure 5—figure supplement 1 ) , suggesting that FGFR inhibition directly affects vesicle production or release . ECVs were also collected from HS-5 and HS-27 cells exposed to PD173074 and analyzed by immunoblot . PD173074 reduced the exosome markers tsg101 and CD9 ( and FGF2 ) but had no effect on ECV production from HS-27 cells ( Figure 5C , similar results with BGJ398 shown in Figure 5—figure supplement 1 ) . Scanning electron microscopy of HS-5 cells revealed abundant budding membrane , whereas the surface of PD173074-exposed cells was smoother , implicating a change in membrane dynamics ( Figure 5—figure supplement 2 ) . To evaluate exosome secretion specifically , sucrose step-gradient fractionation was performed on ECVs from untreated and PD173074 treated HS-5 cells . PD173074 reduced exosomal markers CD9 , tsg101 , and FGF2 in the expected 15–30% interface fraction ( Figure 5D ) . To confirm that decreased exosome secretion is specific for FGFR1 inhibition , HS-5 cells were stably transfected with either a GFP-expressing lentivirus control vector ( GIPZ ) , or doxycycline-induced shRNA targeting FGFR1 . FGFR1 silencing led to a significant reduction in ECVs ( Figure 6C ) . Similar results were obtained with siRNA targeting FGFR1 ( Figure 6—figure supplement 1 ) . siRNA and shRNA constructs targeting FGF2 did not achieve reliable silencing of FGF2 . HS-5 CRISPR/Cas9 knockout of FGFR1 and FGF2 in HS-5 cells were generated , however genetic silencing prevented continued growth . Multiple attempts to make stable deleted cell lines were unsuccessful , likely due to the importance of FGF2-FGFR1 signaling for HS-5 self-renewal and growth ( Bianchi et al . , 2003; Coutu et al . , 2011; Zhou et al . , 1998 ) . That being said , ECVs collected shortly after CRISPR/CAS9 treatment , which resulted in partial silencing of FGF2 or FGFR1 , both demonstrated decreased ECVs by immunoblot and reduced protection of MOLM14 cells ( Figure 6—figure supplement 2 ) . To test the role of FGF2 in ECV production in primary cells , equal numbers of murine stromal cells from Fgf2 +/+ and -/- mice ( Fgf2tm1Doe [Zhou et al . , 1998] ) were treated with PD173074 and ECVs quantified by Virocyt ( Figure 6D ) . Fgf2 +/+ stromal cells secreted significantly more ECVs than -/- , and PD173074 only reduced ECV secretion in +/+ stroma . ECVs from Fgf2 +/+ and -/- mice were also analyzed by immunoblot with similar reduction in ECV proteins from Fgf2 -/- stroma ( Figure 6E ) . To test the role of stromal Fgf2 in an in vivo leukemia model , bone marrow from Fgf2 +/+ mice was retrovirally transfected with BCR-ABL containing GFP as a marker ( Traer et al . , 2012 ) and used to transplant lethally irradiated FGF2 +/+ and -/- mice . This induces a very aggressive disease in mice that is more akin to AML than CML , and TKIs are only effective for a limited duration . Mice were treated with the ABL inhibitor nilotinib 75 mg/kg/day by oral gavage starting on day 14 post-transplant . Mice that were found to have aplastic marrow ( unsuccessful transplantation ) were excluded from analysis since their death was not related to leukemia ( four mice in the Fgf2 +/+ untreated group , one mouse in the Fgf2 -/- untreated group , two mice in the Fgf2 +/+ nilotinib group , and two mice in the Fgf2 -/- nilotinib group ) . The survival curves of the remaining mice are shown in Figure 7A . The cohorts of untreated Fgf2 +/+ and -/- both died rapidly from disease , as expected . Nilotinib significantly increased survival of Fgf2 +/+ and -/- mice compared to untreated mice , but the survival of the nilotinib-treated Fgf2 -/- was also significantly longer than their Fgf2 +/+ counterparts . To ensure equal engraftment of disease in both backgrounds , the blood and bone marrow was analyzed for GFP and found to be similar in both Fgf2 +/+ and -/- mice at time of death ( Figure 7B and Figure 7—figure supplement 1 ) , suggesting that nilotinib was more effective at attenuating disease progression of BCR-ABL leukemia cells in an Fgf2 -/- microenvironment . To directly evaluate the protective effect of ECVs on leukemia progenitor cells , ECVs were isolated from equal numbers of +/+ and -/- primary marrow stromal cells cultured with and without PD173074 treatment . Then , bone marrow from +/+ mice was retrovirally transfected with BCR-ABL and incubated with the ECVs overnight . The cells were washed , plated in methylcellulose with and without imatinib , and colonies counted after 8 days . Imatinib significantly reduced colony formation without ECVs , but ECVs from +/+ stroma almost completely reversed the inhibitory effects of imatinib ( Figure 7C ) . ECVs from PD173074-treated +/+ stroma or -/- stroma were not as protective , suggesting that Fgf2 +/+ stroma more effectively protects BCR-ABL leukemia cells from the effects of kinase inhibition through secretion of protective exosomes . We confirmed the presence of Fgf2 in microvesicles isolated from cultured Fgf2 +/+ mouse stroma ( Figure 6 ) . To confirm that ECVs can be endocytosed by primary cells , lineage-negative hematopoietic progenitor cells were isolated from Fgf2 +/+ mice and stained with a green lipophilic tracer ( DiO ) and incubated with ECVs from Fgf2 +/+ or Fgf2 -/- stromal cells stained with a red lipophilic tracer ( DiI ) . Confocal microscopy confirmed internalization of fluorescently labeled primary stromal ECVs by murine progenitor cells ( Figure 7D and Figure 7—video 1 and 2 ) . The normal hematopoietic microenvironment is altered by leukemia , and can protect leukemia cells from the effects of both chemotherapy and targeted kinase inhibitors ( Yang et al . , 2014; Parmar et al . , 2011; Manshouri et al . , 2011; Traer et al . , 2014; Traer et al . , 2016 ) . Until recently , stromal protection of leukemia cells was thought to be largely mediated by secreted cytokines or through direct contact ( review [Meads et al . , 2009] ) . Here , we show that exosomes from bone marrow stromal cells are transferred to leukemia cells , and protect them from kinase inhibitors . Exosomes have previously been identified as important mediators of malignancy , including recent reports of leukemia exosomes modulating marrow stroma ( Paggetti et al . , 2015; Huan et al . , 2015; Peinado et al . , 2012; Filipazzi et al . , 2012 ) . We found that the reciprocal transfer also occurs , and that marrow stromal exosomes efficiently protect leukemia cells from targeted kinase inhibitors . Along with recent reports that entire mitochondria are transferred between stromal cells and leukemia cells during therapy ( Marlein et al . , 2017; Moschoi et al . , 2016 ) , our data adds to an increasingly complex and intimate relationship between marrow stromal cells and leukemia cells . Indeed , it is almost hard to imagine the leukemia cell in the niche as a separate entity given the direct exchange of organelles , ECVs , cell-cell signaling , and secreted cytokine signaling between stromal and leukemia cells . A better understanding of this relationship is important to develop better ways to eradicate leukemia cells and cure more patients . Isolated reports have previously suggested that FGF2 is contained in ECVs ( Proia et al . , 2008; Choi et al . , 2016 ) , but FGF2 has also been reported to self-assemble into a pore-like structure on the cell membrane and mediate its own translocation with the help of extracellular heparan sulfate ( Steringer et al . , 2015; Schäfer et al . , 2004 ) . Compared to other soluble secreted cytokines , FGF2 was uniquely enriched in ECVs and exosomes ( Figure 2 ) , suggesting that secretion in ECVs is the primary mechanism of FGF2 paracrine signaling from marrow stromal cells . Since FGFR1 is also found on exosomes ( Figure 3D ) , the FGF2-FGFR1 interaction on exosomes may play a direct role in loading FGF2 in exosomes and/or regulate secretion . FGFR inhibitors also increase the amount of FGFR1 protein in stromal cells as measured by immunoblot , consistent with a role in receptor cycling and/or reduced secretion in exosomes . Similar to our observations , epidermal growth factor receptor has been shown to be secreted on ECVs , and secretion is increased after ligand stimulation ( Sanderson et al . , 2008; Perez-Torres et al . , 2008 ) . Likewise , overexpression of oncogenic HER2 in breast cancer cell lines resulted in qualitative differences in microvesicle content ( Amorim et al . , 2014 ) , suggesting a role for activated receptor tyrosine kinases in exosome production and secretion . Receptor-mediated endocytosis is the first step of exosome biogenesis ( Théry et al . , 2002 ) , suggesting that inhibitors of receptor tyrosine kinases may act at this step . How FGFR1 is positioned in the exosome membrane ( inside or out ) , how FGF2 binds FGFR1 in exosomes , and how exosomal FGF2 activates FGFR1 in leukemia cells , are areas of active investigation . FGF2 has been previously implicated in hematologic malignancy progression and development of resistance ( Sato et al . , 2002; Chesi et al . , 2001 ) . Elevated levels of FGF2 have previously been measured in the serum of CML and AML patients ( Aguayo et al . , 2000; Aguayo et al . , 2002 ) , as well as in the bone marrow of AML patients , where it was reported to function as an autocrine promotor of proliferation ( Bieker et al . , 2003 ) . We found that FGF2 expression was increased in CML and AML stroma during the development of resistance to kinase inhibitors , indicating that FGF2 expression is a regulated autocrine growth factor for stroma ( Traer et al . , 2016 ) . This is consistent with the role of FGF2-FGFR1 autocrine expansion of stroma in stress-induced hematopoiesis ( Itkin et al . , 2012; Zhao et al . , 2012 ) and suggests that leukemia cells are able to hijack the FGF2 stress response for survival . The regulation of FGF2-FGFR1 signaling is also supported by the positive correlation in expression of both FGF2 and FGFR1 in a subset of primary AML marrow samples ( Figure 4G ) , indicating that this pathway can be selectively activated . FGFR inhibitors not only inhibit autocrine growth of stroma , but reduce exosome secretion and significantly alter the protective ability of stromal cells ( Figures 4A and 7 ) . Since exosomes contain a complex mixture of proteins , cytokines , lipids and microRNAs ( all of which potentially contribute to leukemia cell protection ) , inhibiting secretion of exosomes is a promising approach to blunting this complex mechanism of resistance . In summary , FGF2 is a regulated autocrine growth factor for marrow stroma that is important in reprogramming the marrow stroma during development of resistance to TKIs . FGF2-FGFR1 activation in marrow stroma leads to increased secretion of exosomes , which are protective of leukemia cells in both in vitro and in vivo models . Given the inevitable development of clinical resistance to TKIs ( FLT3 ITD AML in particular ) , addition of FGFR inhibitors to directly modulate the leukemia niche is a promising approach to improve the durability of response . The human cell line MOLM14 was generously provided by Dr . Yoshinobu Matsuo ( Fujisaki Cell Center , Hayashibara Biochemical Labs , Okayama , Japan ) . The human cell line K562 was obtained from the American Type Culture Collection ( Manassas , VA , USA ) . The human stromal cell lines HS-5 and HS-27a were kindly provided by Dr . Beverly Torok-Storb ( Fred Hutchinson Cancer Research Center , Seattle , WA ) . Cells were maintained in RPMI1640 media supplemented with 10% fetal bovine serum ( FBS ) , 100 U/ml penicillin/100 μg/ml streptomycin , 2 mM L-glutamine , and 0 . 25 μg/ml fungizone ( referred to as R10 ) at 37°C in 5% CO2 . Exosome-depleted FBS was pre-cleared by ultracentrifugation at 100 , 000 g for 2 hr at 4°C . Cell lines were validated by genetic and functional analysis based upon previous reported characteristics . Cell lines were tested monthly for mycoplasma infection and discarded if found to be infected . HS-5 cells grown to 90–100% confluence in 15 cm dishes were washed in 8 ml PBS , and incubated in 12 ml exosome-depleted R10 overnight . The media was collected , cleared of debris ( 2 × 2000 g spin , 10 min ) , and ultracentrifuged at 100 , 000 g for 2 hr at 4°C . The resulting supernatant ( S100 ) was poured off , and 100 ul PBS was added to the ECV pellet ( P100 ) . This was shaken for 4 hr at 4°C at 2000 rpm . P100 was used fresh or stored at −80°C with 10% DMSO . Layers of sucrose ( 60% , 45% , 30% , 15% , 7 . 5% , and 0% ) were carefully pipetted into ultracentrifuge tubes . ECVs were added on top , and the tube ultracentrifuged at 100 , 000 g for 90 min at 4°C . The sucrose interfaces ( 45–60 , 30–45 , 15–30 , 7 . 5–15 , 0–7 . 5 ) were collected with a micropipette , washed in PBS , and pelleted at 100 , 000 g for 2 hr at 4°C . ECVs were quantified by Nanosight LM10 or by Virocyt Virus Counter 3100 , following manufacturers’ protocols . Quizartinib ( AC220 ) was purchased from LC labs ( Woburn , MA , USA ) . Nilotinib , PD173074 and BGJ-398 were purchased from SelleckChem ( Houston , TX , USA ) . Imatinib was purchased from LC labs ( Woburn , MA , USA ) . Recombinant FGF2 was purchased from Peprotech ( Rocky Hill , NJ , USA ) . Viability was assessed with MTS reagent , CellTiter 96 AQueous One Solution Proliferation Assay from Promega Corporation ( Madison , WI , USA ) or by Guava ViaCount flow cytometer assay ( Millipore , Burlington , MA , USA ) . Treated cells were washed in PBS before adding lysis buffer ( Cell Signaling , Danvers , MA , USA ) supplemented with Complete protease inhibitor ( Roche , Indianapolis , IN , USA ) and phosphatase inhibitor cocktail-2 ( Sigma-Aldrich , St . Louis , MO , USA ) . Proteins were fractionated on 4–15% Tris-glycine polyacrylamide gels ( Criterion gels , Bio-Rad ) , transferred to PVDF membranes , and probed with antibodies: FGFR1 , fibronectin ( Cell Signaling , Danvers , MA , USA ) ; CD9 , FGF2 , calreticulin , tsg101 ( Santa Cruz Biotechnology , Dallas , TX , USA ) , CD63 ( Abcam , Boston , MA , USA ) , and actin ( MAB1501 , Millipore , Burlington , MA , USA ) . Stromal CM , S100 and ECVs were lysed with 0 . 1% NP-40 for 30 min , centrifuged 3 , 000 rpm for 10 mins , and 50 μl supernatant was incubated with the magnetic beads overnight and assayed as per manufacturer's instructions ( Luminex Multiplex magnetic beads 30-plex Assay , Life Technologies ) . Bone marrow aspirates were obtained from AML patients after informed consent under the OHSU Institutional Research Board protocol IRB0004422 , and were processed as previously described ( Viola et al . , 2016 ) . After Ficoll , the red cell pellets were incubated with ACK for 30 min on ice to lyse red cells , and plated on 15 cm dishes in MEM-α supplemented with 20% fetal bovine serum ( FBS ) , 100 U/ml penicillin/100 μg/ml streptomycin , 2 mM L-glutamine , and 0 . 25 μg/ml fungizone at 37°C in 5% CO2 . After 48 hr , non-adherent cells were removed and new media was added . This step was repeated after an additional 24 hr . Cells were then incubated for 1–3 weeks with media changes every 7 days , until patchy proliferation became apparent . Cells were trypsinized and replated to facilitate homogenous growth . Cells were expanded over a maximum of 3 passages before use in experiments . Murine primary stroma was isolated from harvested femur marrow without ACK treatment and then cultured as above . Primary stromal samples were analyzed after >2 weeks growth . Stromal cell exosomes were isolated by sucrose step-gradient then washed in 0 . 22 μm filtered PBS . 10 μl was deposited onto glow discharged carbon formvar 400 mesh copper grids ( Ted Pella 01822 F ) for 3 min , rinsed 15 secs in water , wicked on Whatman filter paper 1 , stained for 45 secs in filtered 1 . 33% ( w/v ) uranyl acetate , wicked and air dried . Samples were imaged at 120kV on a FEI Tecnai Spirit TEM system . Images were acquired as 2048 × 2048 pixel , 16-bit gray scale files using the FEI’s TEM Imaging and Analysis ( TIA ) interface on an Eagle 2K CCD multiscan camera . MOLM14 and K562 cells were stained with DiO ( Thermo Fisher ) according to manufacturer’s protocol . HS-5 ECVs were stained with DiI ( Thermo Fisher ) , washed with PBS , and collected by ultracentrifugation . For experiments using mouse bone marrow , cells were isolated from femurs and tibias , RBCs were lysed using ACK buffer ( 0 . 8% NH4Cl and 0 . 1 mMEDTA in KHCO3 buffer; pH 7 . 2–7 . 6 ) , and lineage-negative cells were isolated by MACS cell separation with the human lineage cell depletion kit ( Milteny Biotec ) . Cells were incubated with a cytokine mix ( IL-3 , IL-6 , SCF ) in addition to DiO . DiO-stained cells were combined with DiI-stained ECVs and incubated for 24 hr at 37°C . Cells were washed , placed on poly-D-lysine coated chamber slides , and DAPI-stained . Z-stack imaging was performed on an Olympus IX71 inverted microscope . Images were processes using the Fiji software package ( Schindelin et al . , 2012 ) . ECVs , or exosomes isolated by sucrose step-gradient , were resuspended in proteinase K buffer ( Tris-HCl pH8 , 10 mM CaCl2 ) and then incubated with 200 μg/ml proteinase K at room temp for 30 min . 5 μL 0 . 1 M PMSF and SDS loading buffer was added and samples were incubated at 98°C for 5 min to stop reaction prior to immunoblots . HS-5 and HS-27 cells were grown to 90% confluence in 4-well chamber microscope slides in R10 ±250 nM PD173074 . Cells were stained with lipophilic tracer DiI , washed , and stained with DAPI . Cells were imaged with Zeiss Axio Observer fluorescent microscope at 10X using AxioVision software . Images were uploaded into CellProfiler software and analyzed for cell size . Cell diameter was determined as diameter [μm]=pixels×0 . 394 μm2/pixel . TRIPZ inducible lentiviral FGF2 and FGFR1 shRNA were purchased from Thermo Fisher Scientific Dharmacon RNAi Technologies ( Waltham , MA , USA ) , along with Dharmacon’s trans-lentiviral shRNA packaging kit with calcium phosphate transfection reagent and HEK293T cells . HS-5 and HS-27 cells were transfected with GIPZ control or FGFR1 TRIPZ , per manufacturer’s protocol . TurboRFP/shRNA expression was induced with 1 μg/ml doxycycline ( Fisher ) for 48 hr , cells were washed in PBS , and then media replaced with exosome-depleted R10 +1 μg/ml doxycycline . Cells and CM were collected after 72 hr for analysis . The vector GeCKO lentiCRISPRv2 was obtained from Addgene . This plasmid contains two expression cassettes , hSpCas9 and the chimeric guide RNA . Guide RNA sequences were obtained from GenScript ( Sanjana et al . , 2014 ) , and oligos with 5’ overhang for cloning into lentiCRISPRv2 were manufactured by Fisher Scientific . The vector was digested with BsmBI and dephosphorylated , the plasmid was gel-purified , and oligonucleotides were ligated after annealing and phosphorylation . Plasmid was amplified in Stbl3 bacteria , purified , and lentivirus was generated in HEK293T cells . Transduced HS-5 cells were selected in puromycin for 5 days , and cultured an additional 5 days before assessing knockout . Animal studies were approved by the OHSU Institutional Animal Care and Use Committee . Fgf2tm1Doe/J were purchased from Jackson Laboratory to breed homozygous +/+ and -/- littermates . Bone marrow from 5-FU treated Fgf2 +/+ donors was spinoculated with pMIG containing BCR-ABL and IRES-GFP reporter as previously described ( Traer et al . , 2012; Agarwal et al . , 2008 ) and 2 × 106 cells were retro-orbitally injected into lethally irradiated ( 2 × 450 cGy administered 4 hr apart ) Fgf2 +/+ and -/- recipients . 75 mg/kg/day nilotinib was administered by oral gavage and mice were monitored weekly with cell blood counts and FACS analysis of GFP in peripheral blood . Diseased mice were subjected to detailed histopathologic analysis . For colony assays , ECVs were isolated ( as above ) from equal numbers of Fgf2 +/+ and -/- primary stromal cells cultured on 10 cm plates for 3 days , with and without 500 nM PD173074 ( 3 day pre-treatment and 3 days during ECV collection ) . Bone marrow from FGF2 +/+ mice was spinoculated with pMIG containing BCR-ABL and IRES-GFP reporter as above , incubated with ECVs overnight and washed 3x the next day . 3% of cells were GFP positive by FACS , and 4 × 103 cells were then plated in 1 ml of MethoCult M3234 Methylcellulose Medium for Mouse Cells without cytokines ( Stemcell Technologies ) in triplicate . Mouse bone marrow colonies larger than 50 cells were counted after 8 days . Graphical and statistical data were generated with Microsoft Excel or GraphPad Prism ( GraphPad Software , La Jolla , CA , USA ) . P value < 0 . 05 was considered significant . The authors declare no competing interests . Dr . Druker is currently principal investigator or co-investigator on Novartis clinical trials . His institution , OHSU , has contracts with these companies to pay for patient costs , nurse and data manager salaries , and institutional overhead . He does not derive salary , nor does his lab receive funds from these contracts .

Leukemias are cancers of white blood cells . The cells grow and divide rapidly , often because of mutations in proteins called kinases . Since the kinase mutations do not occur in healthy cells , they provide a good target for anti-leukemia drugs . Several such kinase inhibitors are effective at treating leukemia patients . However , most leukemia cells develop ways to resist the effects of the kinase inhibitors over time , leading to relapses of the disease . One way that leukemia cells resist kinase inhibitors is by taking advantage of signals coming from supportive cells , known as stromal cells , in the bone marrow . When patients are treated with kinase inhibitors , the bone marrow stromal cells produce more of a signaling protein called FGF2 . The leukemia cells then use FGF2 to survive the effects of the kinase inhibitors . It was not clear how the FGF2 signal reaches the leukemia cells from the bone marrow stromal cells . Now , using biochemical techniques , Javidi-Sharifi , Martinez et al . show that bone marrow stromal cells package FGF2 into small compartments called exosomes . The stromal cells release the exosomes into the bone marrow , and the leukemia cells then engulf and internalize the exosomes . Leukemia cells that had taken up FGF2 in this way were better able to survive kinase inhibitor treatment than leukemia cells that had not . Javidi-Sharifi , Martinez et al . also observed that FGF2 also affects the bone marrow stromal cells themselves , causing them to grow faster , produce more FGF2 and release more exosomes . Blocking the effects of FGF2 on the stromal cells slowed their growth and caused fewer exosomes to be released . In addition , mice whose bone marrow stromal cells could not produce FGF2 survived leukemia for longer than mice whose stromal cells provided protective FGF2 in exosomes to leukemia cells . This suggests that taking advantage of drugs that prevent bone marrow stromal cells from releasing FGF2 in exosomes might improve treatments for leukemia . Further research will be needed to confirm whether this strategy would be effective in humans .

2019

FGF2-FGFR1 signaling regulates release of Leukemia-Protective exosomes from bone marrow stromal cells

To build the spindle at mitosis , motors exert spatially regulated forces on microtubules . We know that dynein pulls on mammalian spindle microtubule minus-ends , and this localized activity at ends is predicted to allow dynein to cluster microtubules into poles . How dynein becomes enriched at minus-ends is not known . Here , we use quantitative imaging and laser ablation to show that NuMA targets dynactin to minus-ends , localizing dynein activity there . NuMA is recruited to new minus-ends independently of dynein and more quickly than dynactin; both NuMA and dynactin display specific , steady-state binding at minus-ends . NuMA localization to minus-ends involves a C-terminal region outside NuMA’s canonical microtubule-binding domain and is independent of minus-end binders γ-TuRC , CAMSAP1 , and KANSL1/3 . Both NuMA’s minus-end-binding and dynein-dynactin-binding modules are required to rescue focused , bipolar spindle organization . Thus , NuMA may serve as a mitosis-specific minus-end cargo adaptor , targeting dynein activity to minus-ends to cluster spindle microtubules into poles . Each time a cell divides , molecular motors help remodel the microtubule cytoskeleton into a bipolar assembly of microtubules called the spindle . The spindle’s bipolar architecture is essential to its function – accurate chromosome segregation . In mammalian spindles , microtubule plus-ends mechanically couple to chromosomes , while microtubule minus-ends are focused into two poles that dictate where chromosomes are transported at anaphase . To build the spindle’s architecture , motors must exert spatially regulated forces on microtubules . Microtubule ends offer platforms for such localized regulation , since they are structurally distinct . Indeed , motor recruitment and activity at plus-ends is well-documented ( Wu et al . , 2006 ) , but motor regulation at minus-ends is less well understood . Dynein is a minus-end-directed motor which slides parallel spindle microtubules to focus their minus-ends into spindle poles ( Heald et al . , 1996; Verde et al . , 1991 ) , working in complex with its adaptor dynactin and the microtubule-binding protein NuMA ( Gaglio et al . , 1996; Merdes et al . , 1996 ) . The clustering of parallel microtubules into poles presents a geometric problem when forces are indiscriminately applied all along microtubules: inversely oriented motors between parallel microtubules will oppose each other , resulting in gridlock , unless symmetry is broken by dynein enrichment at microtubule minus-ends ( Hyman and Karsenti , 1996; McIntosh et al . , 1969; Surrey et al . , 2001 ) . In computational models , localizing a minus-end-directed motor at microtubule ends permits microtubule clustering into asters or poles ( Foster et al . , 2015; Goshima et al . , 2005; Nedelec and Surrey , 2001; Surrey et al . , 2001 ) and the emergence of a robust steady-state spindle length ( Burbank et al . , 2007 ) . More recently , experimental work has shown that dynein-dynactin and NuMA do indeed selectively localize to spindle minus-ends , with dynein pulling on them after kinetochore-fiber ( k-fiber ) ablation in mammalian spindles ( Elting et al . , 2014; Sikirzhytski et al . , 2014 ) . This is consistent with suggestions that dynein and NuMA capture and pull on distal k-fiber minus-ends in monopolar spindles ( Khodjakov et al . , 2003 ) . Altogether , these findings demonstrate the importance ( in silico ) and existence ( in vivo ) of localized dynein activity at spindle microtubule minus-ends . How dynein becomes localized at minus-ends remains an open question . Dynein may be enriched near minus-ends because it walks toward them on microtubules and piles up when it runs out of track; indeed , pile-up of dynein has been observed at minus-ends in vitro ( McKenney et al . , 2014; Soundararajan and Bullock , 2014 ) and can drive minus-end clustering ( Tan et al . , 2017 ) . Alternatively , the exposed α-tubulin interface at minus-ends is structurally distinct and could bind an adaptor protein that specifically recruits dynein , analogous to recruitment at canonical dynein cargoes like organelles ( Kardon and Vale , 2009 ) . NuMA can target dynein-dynactin to the cell cortex ( Lechler and Fuchs , 2005; Nguyen-Ngoc et al . , 2007 ) and thus could be one such adaptor . However , in vitro NuMA has shown no direct affinity for minus-ends specifically , binding all along the microtubule lattice ( Du et al . , 2002; Forth et al . , 2014; Haren and Merdes , 2002 ) or at both ends ( Seldin et al . , 2016 ) , unlike three proteins known to interact directly with minus-ends at mitosis: γ-TuRC ( Zheng et al . , 1995 ) , CAMSAP1 ( Akhmanova and Hoogenraad , 2015; Hendershott and Vale , 2014; Jiang et al . , 2014 ) and KANSL1/3 ( Meunier et al . , 2015 ) . In cells , NuMA is thought to require dynein activity to carry it to minus-ends and spindle poles ( Merdes et al . , 2000 ) , where it anchors spindle microtubules ( Dionne et al . , 1999; Gaglio et al . , 1995; Heald et al . , 1997; Silk et al . , 2009 ) . Thus , it remains unclear whether dynein-dynactin and NuMA have specific binding sites at minus-ends , and if so , whether they are recruited by known minus-end binders . Finally , knowing how dynein is targeted to minus-ends would make it possible to test the in vivo role of minus-end-localized – compared to indiscriminately-localized – forces in spindle organization . Here , we use laser ablation in cells to create new , isolated minus-ends in the mammalian spindle , and quantitative imaging to map protein recruitment to these ends and its mechanistic basis . We demonstrate that NuMA binds at minus-ends independently of dynein , and that NuMA targets dynactin – and thereby dynein activity – to spindle minus-ends . This challenges the prevailing model that dynein delivers NuMA to spindle minus-ends and poles ( Merdes et al . , 2000; Radulescu and Cleveland , 2010 ) . NuMA localization to minus-ends is independent of known direct minus-end binders γ-TuRC , CAMSAP1 , and KANSL1/3 , and it involves both NuMA’s canonical microtubule-binding domain and an additional region of its C-terminus . Thus , NuMA – which is sequestered in the nucleus at interphase ( Lydersen and Pettijohn , 1980 ) – may serve as a mitosis-specific minus-end cargo adaptor , recruiting dynein activity to spindle minus-ends . Both NuMA’s minus-end-binding domain and dynactin-binding domain are required for correct spindle architecture , supporting long-standing in silico predictions that localizing dynein to minus-ends enables effective clustering of parallel microtubules into poles . These findings identify a mechanism for mitosis-specific recruitment of dynein to microtubule minus-ends and , more broadly , illustrate how spatial regulation of local forces may give rise to larger scale cytoskeletal architectures . To visualize the spatial targeting of the dynein-dynactin complex to microtubule minus-ends – which are normally buried in dense mammalian spindles – we used nocodazole washout and laser ablation to create resolvable minus-ends in mitotic cells . First , to determine whether dynein-dynactin and NuMA localize to individual microtubule minus-ends , we treated mammalian PtK2 and RPE1 cells with the microtubule-depolymerizing drug nocodazole and fixed cells 6–8 min after drug washout to capture acentrosomal microtubules with clearly visible plus- and minus-ends ( Figure 1A ) . p150Glued ( p150 , a dynactin subunit ) and NuMA strongly co-localized at one end of these individual microtubules ( Figure 1A ) , with a clear binding preference for minus-ends over the microtubule lattice or the plus-end when polarity was marked by EB1 ( Figure 1B ) . Interestingly , in prophase cells before nuclear envelope breakdown , p150 localized predominantly to plus-ends rather than minus-ends ( Figure 1B; Figure 1—figure supplement 1 ) , consistent with dynactin’s interphase localization ( Vaughan et al . , 1999 ) . Thus , nuclear envelope breakdown ( NEB ) confers dynactin’s preference for minus-ends , suggesting regulated , mitosis-specific spatial targeting . Second , we sought to test whether dynactin and NuMA have finite binding sites at minus-ends by measuring the kinetics of their recruitment . To do so , we used laser ablation of k-fibers in PtK2 cells to create new minus-ends within the spindle body ( Figure 1C–E ) . By spatially and temporally synchronizing the creation of a bundle of minus-ends , laser ablation allowed for dynamic measurements of the recruitment to minus-ends of GFP-tagged dynactin ( Arp1A ) and NuMA and comparison to a direct minus-end binding protein , CAMSAP1 . High dynein background in the cytoplasm prevented accurate recruitment kinetics measurements of the motor itself . Dynactin , NuMA , and CAMSAP1 robustly recognized new microtubule minus-ends within the spindle ( Figure 1C–E; Videos 1–3 ) . The kinetics of dynactin and NuMA recruitment to minus-ends were distinct from CAMSAP1 , which reached a max intensity after approximately 5–10 s , but then decreased in intensity as if its binding sites at the minus-end were gradually obstructed ( Figure 1E–G ) . Unlike CAMSAP1 , dynactin and NuMA intensities reached a stable plateau , suggesting that binding saturates , reaching a steady-state . This finite binding is consistent with a truly end-specific localization ( rather than localization along the unbounded lattice near the minus-end ) . In addition , the rate of NuMA or dynactin accumulation at new minus-ends did not correlate with the length of k-fiber stubs created by ablation ( Figure 1H ) , which could indicate that recruitment rate is set by the number of individual microtubule minus-ends ( which is similar across k-fibers [McEwen et al . , 1997] ) rather than k-fiber length . Interestingly , NuMA intensity at new minus-ends increased at a faster rate and saturated sooner than dynactin ( Figure 1F , G ) . This observation hints at a model in which NuMA targets dynein-dynactin to minus-ends , and it is less easily consistent with the idea that dynein delivers NuMA to minus-ends . If true , this hypothesis would explain why dynactin’s minus-end preference arises after NEB , when NuMA is released from the nucleus into the mitotic cytoplasm . The finding that NuMA localizes to minus-ends more quickly than dynactin could suggest that NuMA directly or indirectly recognizes the exposed α-tubulin structure of minus-ends and subsequently recruits dynein-dynactin ( ‘Structural Recognition , ’ Figure 2A ) . Alternatively , dynein-dynactin could walk toward minus-ends , carrying NuMA , and pile up or dwell there ( ‘Walk and Pile-up , ’ Figure 2B ) . To test the ‘Walk and Pile-up’ model , we inhibited dynein-dynactin in PtK2 spindles by over-expressing p50 ( dynamitin ) , resulting in fully unfocused k-fiber arrays ( Figure 2C ) . The ‘Walk and Pile-up’ model predicts that in the absence of dynein-dynactin transport , NuMA – which requires dynactin for its interaction with dynein ( Merdes et al . , 2000 ) – should not reach minus-ends . Instead , GFP-NuMA was robustly recruited to minus-ends created by k-fiber ablation ( Figure 2C; Video 4 ) , indicating that NuMA can localize to minus-ends independently of dynein activity . Similarly , NuMA was recruited to ablation-created minus-ends when dynein was inhibited by p150-CC1 over-expression ( Figure 2—figure supplement 1A , B ) . Thus , NuMA localizes to minus-ends without dynein carrying it there – consistent with early observations in extract asters and spindles ( Gaglio et al . , 1996; Heald et al . , 1997 ) but in contrast to the prevailing view that dynein delivers NuMA to minus-ends ( Merdes et al . , 2000; Radulescu and Cleveland , 2010 ) . Slower NuMA accumulation kinetics after p50 overexpression suggest that dynein-dynactin-NuMA complex formation may aid rapid NuMA recruitment , but dynein ‘Walk and Pile-up’ alone cannot explain NuMA’s minus-end affinity ( Figure 2D , E ) . To confirm that NuMA localizes to minus-ends even after genetic dynein deletion , we used an inducible CRISPR/Cas9 HeLa cell line to knock out dynein’s heavy chain ( DHC ) ( McKinley and Cheeseman , 2017 ) . After dynein knockout ( Figure 2—figure supplement 1C , D ) , both NuMA and dynactin localized robustly to k-fiber minus-ends ( Figure 2F; Figure 2—figure supplement 1E , F ) . Together , these data indicate that NuMA can bind to minus-ends , either directly or indirectly , in the absence of dynein and are consistent with NuMA-mediated recognition of minus-ends ( ‘Structural Recognition , ’ Figure 2A ) . In addition , they suggest that NuMA may interact with and target dynactin to minus-ends even in the absence of dynein , when dynein-dynactin complexes cannot form and dynein motility cannot deliver dynactin close to minus-ends . If NuMA does target dynein-dynactin to minus-ends , localizing force there , we would expect NuMA to be required for dynein to transport minus-ends created by k-fiber ablation ( Elting et al . , 2014; Sikirzhytski et al . , 2014 ) . To test this hypothesis , we made inducible CRISPR/Cas9 RPE1 cell lines to knock out NuMA ( Figure 3—figure supplement 1 ) ( McKinley et al . , 2015; Wang et al . , 2015 ) . Indeed , when NuMA was knocked out – causing elongated , heterogeneously disorganized spindles with detached centrosomes – ablation-created minus-ends were no longer transported toward poles by dynein ( Figure 3A–C; Video 5 ) . These data are consistent with a previous finding that NuMA antibody injection prevents distal k-fiber looping events in monopolar spindles , in which dynein likely pulls on free k-fiber minus-ends ( Khodjakov et al . , 2003 ) . Thus , NuMA is required for dynein activity at spindle microtubule minus-ends . Dynein activity at minus-ends could require NuMA because NuMA modulates dynein-dynactin’s ability to pull on microtubules , or because NuMA localizes dynein to minus-ends . Given the findings in Figures 1 and 2 , we hypothesized that NuMA localizes dynein activity by recruiting dynactin to minus-ends . Indeed , after NuMA knockout , dynactin ( GFP-Arp1A ) was no longer detectable at k-fiber minus-ends created by laser ablation ( Figure 4A; Video 6 ) . Immunofluorescence experiments confirmed that in the absence of NuMA , dynactin remained on the spindle but no longer localized to minus-ends ( Figure 4B ) . Interestingly , dynein’s localization within the spindle ( labeled using an antibody against dynein intermediate chain ) was less minus-end-specific than dynactin’s , both before and after NuMA deletion ( Figure 4C ) . In addition , nocodazole washout experiments revealed that dynactin ( p150 ) localization to individual microtubule minus-ends was lost upon NuMA deletion ( Figure 4D ) . Instead , dynactin frequently co-localized with EB1 at plus-ends , similar to its interphase and pre-NEB localization pattern ( Figure 4D; Figure 1B ) . Thus , the data indicate that NuMA is required for the transport of minus-ends by dynein because NuMA localizes to minus-ends and recruits dynactin there . How is NuMA targeted to minus-ends , independently of dynein ? In vitro , canonical microtubule-binding regions of NuMA have shown no preference for minus-ends relative to the lattice or plus-end of purified microtubules ( Du et al . , 2002; Forth et al . , 2014; Haren and Merdes , 2002; Seldin et al . , 2016 ) . Given this lack of minus-end-specific binding in vitro , we hypothesized that NuMA is indirectly recruited to spindle minus-ends by one of three known direct minus-end binders active at mitosis: γ-TuRC ( Zheng et al . , 1995 ) , CAMSAP1 ( Hendershott and Vale , 2014; Jiang et al . , 2014 ) ( Figure 1; CAMSAP2 and 3 are phosphorylated at mitosis and no longer interact with microtubules [Jiang et al . , 2014] ) , and KANSL1/3 ( Meunier et al . , 2015 ) . To test this hypothesis , we treated RPE1 cells with 30 μM gatastatin to block γ-TuRC binding ( Chinen et al . , 2015 ) ( Figure 5A; Figure 5—figure supplement 1A , B ) . We also made inducible CRISPR/Cas9 RPE1 cell lines to knock out CAMSAP1 or KANSL1 ( Figure 5—figure supplement 1C , D ) , as KANSL1 depletion has been shown to delocalize the entire KANSL complex ( Meunier et al . , 2015 ) . CAMSAP1 knockout caused a small reduction in spindle length ( Figure 5—figure supplement 1E ) , consistent with the spindle minus-end protecting function ascribed to its Drosophila homolog , Patronin ( Goodwin and Vale , 2010 ) . To our surprise , however , none of these perturbations qualitatively altered NuMA localization at spindle poles ( Figure 5A ) . To check for a more subtle contribution of γ-TuRC , CAMSAP1 , or KANSL1 to NuMA localization at minus-ends , we performed k-fiber ablations after 30 μM gatastatin treatment , CAMSAP1 knockout , or KANSL1 knockout and quantified GFP-NuMA recruitment kinetics at new minus-ends . NuMA recruitment to new minus-ends remained robust , and recruitment timescales were statistically indistinguishable from control ( Figure 5B–C ) . Thus , the data indicate that the direct mitotic minus-end binders γ-TuRC , CAMSAP1 , and KANSL1/3 are not responsible for NuMA’s localization to spindle microtubule minus-ends . Given this lack of involvement of direct minus-end binding proteins , we sought to confirm using super-resolution microscopy that NuMA specifically localizes at individual spindle microtubule minus-ends . Three-dimensional stochastic optical reconstruction microscopy ( 3D STORM [Huang et al . , 2008; Rust et al . , 2006] ) of NuMA at k-fiber minus-ends created by ablation revealed organized clusters of NuMA puncta , rather than a lawn of molecules along the lattice near the minus-end ( Figure 5D–G ) . The spacing between puncta was consistent with the distance between individual microtubules within k-fibers ( Figure 5H ) as measured by electron microscopy in the same cell type ( PtK2 ) ( McDonald et al . , 1992 ) , supporting the idea that these NuMA puncta are built on individual microtubule minus-ends . A yet undiscovered minus-end binding protein may recruit NuMA; alternatively , NuMA may have direct minus-end-specific binding ability that has not been recapitulated in vitro . To define the NuMA domain required for spindle minus-end localization , we performed rescue experiments with different NuMA truncations in the NuMA-knockout background ( Figure 6A ) . This avoids the C-terminus-mediated oligomerization ( Harborth et al . , 1999 ) with endogenous protein that can complicate interpretations of localization . In this NuMA-knockout background , full-length NuMA ( ‘FL’ ) localized to spindle minus-ends , as did a bonsai construct ( ‘N-C’ ) with most of the central coiled-coil removed ( Figure 6B ) . To our surprise , ‘N-C’ rescued spindle architecture as effectively as full-length NuMA , indicating that the extraordinary length of NuMA protein is not essential to its function in spindle structure ( Figure 6C ) . Importantly , NuMA’s C-terminus ( ‘C’ ) alone localized to minus-ends even in the absence of endogenous full-length protein ( Figure 6B ) . Its intensity at minus-ends was less striking than that of full-length or N-C protein at poles , likely because minus-ends are not as densely concentrated in disorganized spindles , and perhaps because NuMA’s N-terminus facilitates formation of higher order NuMA assemblies . However , the preference of NuMA ‘C’ for spindle ends was clear . Because dynein-dynactin interacts with NuMA’s N-terminus ( Kotak et al . , 2012 ) , the minus-end localization of NuMA’s C-terminus provides further support for dynein-independent minus-end recognition . We sought to more closely define which sections within NuMA’s C-terminus ( a . a . 1701–2115 ) are involved in microtubule minus-end localization . The second half of the C-terminus ( ‘C-tail2’ , a . a . 1882–2115 ) , which contains residues previously implicated in NuMA-microtubule interactions ( Chang et al . , 2017; Du et al . , 2002; Gallini et al . , 2016; Haren and Merdes , 2002 ) , localized all along spindle microtubules with no minus-end preference ( Figure 6B ) . Similarly , C-tail2A ( a . a . 1882–1981 ) bound along the lattice , while C-tail1 ( a . a . 1701–1881 ) did not bind microtubules ( Figure 6—figure supplement 1 ) . However , in combination ( ‘C-tail1+2A’ , a . a . 1701–1981 ) they localized at minus-ends ( Figure 6B ) . Indeed , the C-tail1+2A region was sufficient for recruitment to new minus-ends created by ablation , even in the absence of endogenous NuMA ( Figure 6D; Video 7 ) . Which pieces of the tail1+2A region are specifically required for minus-end localization , and what they each do , will require further dissection . Interestingly , related NuMA residues were recently shown to bind at both plus- and minus-ends and play a role in spindle orientation ( Seldin et al . , 2016 ) . In sum , our data indicate that NuMA’s localization to spindle microtubule minus-ends is independent of dynein , independent of known direct minus-end binding proteins , and mediated by its C-terminal residues 1701–1981 ( ‘C-tail1+2A’ ) . Importantly , NuMA’s C-terminus ( ‘C’ ) localized to minus-ends but could not rescue proper pole focusing or spindle architecture , unlike full-length protein or N-C ( Figure 6C ) . This suggests that NuMA’s function in spindle organization requires both minus-end binding ( via its C-terminus ) and the ability to recruit dynactin to minus-ends ( via its N-terminus ) ( Kotak et al . , 2012 ) . Consistent with this hypothesis , NuMA’s C-terminus ( ‘C’ ) was unable to recruit dynactin to minus-ends , while its N-terminus and C-terminus fused ( ‘N-C’ ) did ( Figure 6E; Table 1 ) . To test whether recruiting dynein-dynactin to the sides of microtubules was sufficient for proper microtubule clustering into poles , we fused NuMA’s N-terminus to the Tau microtubule-binding domain ( ‘N-Tau’ ) . N-Tau localized along the length of spindle microtubules and recruited dynactin there ( Figure 6E ) . Despite combining dynein-dynactin binding and microtubule lattice binding , N-Tau was not enriched at minus-ends and was unable to rescue spindle architecture ( Figure 6C; Table 1 ) . Like a traditional cargo adaptor , NuMA may target force to spindle minus-ends using a cargo ( minus-end ) binding module ( ‘C’ ) and a dynactin-recruitment module ( ‘N’ ) . Furthermore , the inability of N-Tau to rescue spindle architecture in the absence of endogenous NuMA suggests that specifically targeting dynein-dynactin to minus-ends , not just all along spindle microtubules as N-Tau does , is critical for organizing a focused , bipolar spindle . Our data indicate that NuMA localizes to spindle microtubule minus-ends independently of dynein and minus-end binding proteins γ-TuRC , CAMSAP1 , and KANSL1/3 . NuMA then targets dynactin to minus-ends , localizing dynein motor activity there . In addition , targeting dynein to the end of its track could permit amplification by motor pile-up , as NuMA at minus-ends captures processive dynein complexes . Altogether , our findings are consistent with a model in which NuMA confers minus-end targeting of the dynein-dynactin complex upon nuclear envelope breakdown ( NEB ) , when NuMA is released from the nucleus . Indeed , active microtubule clustering by dynein is first observed coincident with NEB ( Rusan et al . , 2002 ) . Thus , NuMA may provide both spatial ( minus-end-specific ) and temporal ( mitosis-specific ) regulation of dynein-powered force ( Figure 7A ) . The data also indicate that NuMA’s minus-end localization at mitosis is mediated by the tail1+2A region of its C-terminus , while previously identified microtubule-binding domains tail2A or tail2B ( Du et al . , 2002; Gallini et al . , 2016; Haren and Merdes , 2002 ) are not sufficient . We propose three hypotheses for tail1+2A-mediated minus-end recognition . First , a yet-unidentified minus-end binding protein or the minus-end-directed kinesin HSET ( Gaglio et al . , 1996; Mountain et al . , 1999 ) could localize NuMA via an interaction that requires this longer segment of the NuMA tail . During the preparation of this manuscript , ASPM was identified as a novel mammalian spindle minus-end binder , but ASPM deletion does not affect NuMA localization ( Jiang et al . , 2017 ) . Second , NuMA may recognize both plus- and minus-ends via microtubule curvature sensing , as previously proposed ( Seldin et al . , 2016 ) , and yet be actively excluded from plus-ends within the spindle by other proteins . Third , NuMA tail1+2A may bind minus-ends directly in cells , perhaps using post-translational modifications ( Compton and Luo , 1995; Gallini et al . , 2016; Magescas et al . , 2017; Yan et al . , 2015 ) not previously recapitulated in vitro . Models of spindle assembly in silico predict that enriching a minus-end-directed motor at microtubule ends can break gridlock between parallel microtubules and allow robust minus-end clustering into poles ( Burbank et al . , 2007; Goshima et al . , 2005; Hyman and Karsenti , 1996; Surrey et al . , 2001 ) . However , a lack of mechanistic information and tools made it difficult to test such hypotheses in vivo . The present study reveals a mechanism for direct recruitment of the motor to minus-ends , via NuMA , and is consistent with the prediction that targeting dynein-dynactin to the sides of microtubules is not sufficient for robust spindle organization . Both the minus-end-binding module ( ‘C’ ) and dynein-dynactin-binding module ( ‘N’ ) of NuMA are required for bipolar focusing , while the distance between them is not critical for function . Fusing the dynein-dynactin-binding module to the Tau microtubule binding domain is not sufficient , suggesting a requirement for minus-end-specific forces ( Figure 6; Table 1 ) . However , we cannot formally exclude that features of the NuMA C-terminus ( missing from ‘N-Tau’ ) other than minus-end binding enable rescue of pole focusing , or that while N-Tau recruits dynactin ( Figure 6D ) , it cannot effectively couple to dynein . Altogether , the work is consistent with a model for mammalian spindle organization in which targeting poleward force to microtubule minus-ends ( by NuMA-mediated dynactin recruitment ) is critical for organizing microtubules into a focused , bipolar array ( Figure 7B ) . More broadly , the emergence of pole architecture at mitosis illustrates how spatial regulation of molecular-scale activities ( like NuMA localizing dynein activity at individual minus-ends ) can give rise to complex and diverse cellular-scale structures . The data suggest that NuMA in the mammalian spindle body may function as a traditional cargo adaptor , with a cargo ( minus-end ) -binding module and a motor-binding module . In the spindle body context , microtubule minus-ends are the cargo , just as the cortex can be framed as cargo for cortical dynein – analogous to more canonical interphase cargo like membranous vesicles or organelles . This framework raises the question of whether NuMA additionally serves as a dynein activator , inducing highly processive motility like interphase dynein adaptors can ( McKenney et al . , 2014; Schlager et al . , 2014 ) . Dynein’s localization in mammalian mitosis is more ubiquitous than that of dynactin ( Figure 4C; Figure 2—figure supplement 1C ) ; high cytoplasmic levels of dynein prevented us from detecting clear dynein enrichment at ablation-created minus-ends , for example ( data not shown ) , and dynein localization is not as altered by NuMA knockout as dynactin localization ( Figure 4C ) . These observations could be explained if NuMA and dynactin at minus-ends selectively activate dynein there – through a conformational shift that renders the motor more processive , for example – rather than simply selectively localizing it there . In other words , dynein’s localization within the spindle body may be less tightly regulated than its activity . The loss of dynein activity at minus-ends observed after NuMA knockout ( Figure 3C ) may stem from a lack of dynein activation ( without NuMA and dynactin present at minus-ends ) rather than a lack of dynein enrichment . Unlike known dynein activators , NuMA is thought to not only homodimerize but also oligomerize into higher order assemblies ( Harborth et al . , 1999; Saredi et al . , 1997 ) . NuMA’s oligomerization ability suggests that it could assemble teams of motors and invites a comparison to motor-clustering in microdomains on large cellular cargoes , like phagosomes ( Rai et al . , 2016 ) . The increased force and processivity provided by teams of NuMA-dynactin-dynein complexes on mitotic minus-ends could enable transport and clustering of minus-ends despite high loads and friction created by dense microtubule crosslinking and – in the case of k-fiber minus-ends – coupling to chromosomes . PtK2 cells were cultured in MEM ( 11095; Thermo Fisher , Waltham , MA ) supplemented with sodium pyruvate ( 11360; Thermo Fisher ) , nonessential amino acids ( 11140; Thermo Fisher ) , penicillin/streptomycin , and 10% heat-inactivated fetal bovine serum ( FBS ) ( 10438; Thermo Fisher ) . RPE1 and HeLa cells were cultured in DMEM/F12 with GlutaMAX ( 10565018; Thermo Fisher ) supplemented with penicillin/streptomycin and 10% FBS . For Tet-on inducible CRISPR-Cas9 cell lines , tetracycline-screened FBS ( SH30070 . 03T; Hyclone Labs , Logan , UT ) was used . Cell lines were not STR-profiled for authentication . All cell lines tested negative for mycoplasma . Cells were maintained at 37°C and 5% CO2 and were transfected with DNA using ViaFect ( E4981; Promega , Madison , WI ) 48 hr ( RPE1/HeLa ) or 72 hr ( PtK2 ) before imaging . sgRNAs were designed against 5’ exons of NuMA , CAMSAP1 , and KANSL1 using http://crispr . mit . edu . sgRNAs are listed in Key Resource Table . The plasmid used to express sgRNAs under control of the hU6 promoter ( pLenti-sgRNA ) was a gift from T . Wang , D . Sabatini , and E . Lander ( Whitehead/Broad/MIT ) . An RPE1 cell line containing doxycycline-inducible human codon-optimized spCas9 was a gift from I . Cheeseman ( Whitehead/MIT ) and was generated as described in ( McKinley et al . , 2015 ) using a derivative of the transposon described in ( Wang et al . , 2014 ) . We infected this inducible-spCas9 RPE1 cell line with each pLenti-sgRNA as described in ( Wang et al . , 2015 ) using virus expressed in HEK293T cells and 10 μg/mL polybrene and selected with 6 μg/mL puromycin . For each targeted gene , we tested 3 independent sgRNA sequences , each of which generated indistinguishable spindle phenotypes ( data not shown ) , and picked one line for subsequent studies . Four days before each experiment , spCas9 expression was induced with 1 μM doxycycline hyclate . For live imaging , cells were plated on glass-bottom 35 mm dishes coated with poly-D-lysine ( MatTek Corporation , Ashland , MA ) and imaged in a stage-top humidified incubation chamber ( Tokai Hit , Fujinomiya-shi , Japan ) maintained at 30°C and 5% CO2 . To visualize tubulin , 100 nM siR-Tubulin dye ( Cytoskeleton , Inc . , Denver , CO ) was added 2 hr prior to imaging , along with 10 µM verapamil ( Cytoskeleton , Inc . ) . Under these conditions , there was no detected defect in spindle appearance or microtubule dynamics . As described elsewhere ( Elting et al . , 2014 ) , cells were imaged using a spinning disk confocal inverted microscope ( Eclipse Ti-E; Nikon Instruments , Melville , NY ) with a 100 × 1 . 45 Ph3 oil objective through a 1 . 5X lens , operated by MetaMorph ( 7 . 7 . 8 . 0; Molecular Devices , Sunnyvale , CA ) . Laser ablation ( 30 3-ns pulses at 20 Hz ) with 551 nm light was performed using the galvo-controlled MicroPoint Laser System ( Andor , Belfast , UK ) . For laser ablation experiments , images were acquired more slowly prior to ablation and more rapidly after ablation ( typically 7 s prior and 3 . 5 s after ablation ) . For immunofluorescence , cells were plated on #1 . 5 25 mm coverslips coated with 1 mg/mL poly-L-lysine . Cells were fixed with 95% methanol + 5 mM EGTA at −20°C for 3 min , washed with TBS-T ( 0 . 1% Triton-X-100 in TBS ) , and blocked with 2% BSA in TBS-T for 1 hr . Primary and secondary antibodies were diluted in TBS-T+2% BSA and incubated with cells overnight at 4°C ( primary ) or for 20 min at room temperature ( secondary ) . DNA was labeled with Hoescht 33342 ( Sigma , St . Louis , MO ) before cells were mounted in ProLongGold Antifade ( P36934; Thermo Fisher ) . Cells were imaged using the spinning disk confocal microscope described above . Antibodies: mouse anti-α-tubulin DM1α ( T6199; Sigma ) , rabbit anti-α-tubulin ( ab18251; Abcam , Cambridge , UK ) , rabbit anti-NuMA ( NB500-174; Novus Biologicals , Littleton , CO ) , mouse anti-p150-Glued ( 610473; BD Biosciences , San Jose , CA ) , mouse anti-α-tubulin DM1α conjugated to Alexa488 ( 8058S; Cell Signaling , Danvers , MA ) , mouse anti-dynein intermediate chain ( MAB1618MI; Millipore , Billerica , MA ) , rabbit anti-EB1 ( sc-15347; Santa Cruz Biotechnology , Dallas , TX ) , rabbit anti-KANSL1 ( PAB20355; Abnova , Taipei City , Taiwan ) , rabbit anti-CAMSAP1 ( NBP1-26645; Novus Biologicals ) , mouse anti-actin ( MAB1501; Millipore ) , rabbit anti-γ-tubulin ( T3559; Sigma ) , and camel nanobody against GFP coupled to Atto488 ( gba-488; ChromoTek , Hauppauge , NY ) . PtK2 cells expressing GFP-α-tubulin ( gift of A . Khodjakov , Wadsworth Center ) were plated on photo-etched , gridded coverslips ( G490; ProSciTech , Kirwan , Australia ) coated with 1 mg/mL poly-L-lysine ( P-1524; Sigma ) and imaged at 29–30°C in a homemade heated aluminum coverslip holder using the confocal microscope and ablation system described above . 20–30 s after k-fiber ablation , imaging media was replaced with fixative ( as above ) chilled to −80°C , and the coverslip holder was placed on ice for 1 min . Cells were incubated with 3% BSA in PBS for 1 hr at room temperature , and then with primary antibodies overnight at 4°C . Secondary antibodies ( ( anti-mouse Cy3B; Jackson Immunoresearch , West Grove , PA ) ; anti-rabbit AF647 ( Life Tech , Carlsbad , CA ) ) were incubated for 30 min at room temperature . Antibody incubations were followed by four washes with 0 . 2% BSA in PBS . Samples were stored in PBS during confocal imaging , and coverslip grid was used to re-find the individual ablated cell . For 3D STORM imaging , samples were mounted on glass slides with a standard STORM imaging buffer consisting of 5% ( w/v ) glucose , 100 mM cysteamine , 0 . 8 mg/mL glucose oxidase , and 40 µg/mL catalase in 1M Tris-HCI ( pH 7 . 5 ) ( Huang et al . , 2008; Rust et al . , 2006 ) . Coverslips were sealed using Cytoseal 60 . STORM imaging was performed on a homebuilt setup based on a modified Nikon Eclipse Ti-E inverted fluorescence microscope using a Nikon CFI Plan Apo λ 100x oil immersion objective ( NA 1 . 45 ) . Dye molecules were photoswitched to the dark state and imaged using either 647- or 560 nm lasers ( MPB Communications , Montreal , CAN ) ; these lasers were passed through an acousto-optic tunable filter and introduced through an optical fiber into the back focal plane of the microscope and onto the sample at intensities of ~2 kW cm−2 . A translation stage was used to shift the laser beams towards the edge of the objective so that light reached the sample at incident angles slightly smaller than the critical angle of the glass-water interface . A 405 nm laser was used concurrently with either the 647- or 560 nm lasers to reactivate fluorophores into the emitting state . The power of the 405 nm laser ( typical range 0–1 W cm−2 ) was adjusted during image acquisition so that at any given instant , only a small , optically resolvable fraction of the fluorophores in the sample were in the emitting state . Emission was recorded with an Andor iXon Ultra 897 EM-CCD camera at a framerate of 220 Hz , for a total of ~120 , 000 frames per image . For 3D STORM imaging , a cylindrical lens of focal length 1 m was inserted into the imaging path so that images of single molecules were elongated in opposite directions for molecules on the proximal and distal sides of the focal plane ( Huang et al . , 2008 ) . Two-color imaging was performed via sequential imaging of targets labeled by AF647 and Cy3B . The raw STORM data was analyzed according to previously described methods ( Huang et al . , 2008; Rust et al . , 2006 ) . To inhibit γ-tubulin , 30 μM gatastatin ( gift of Takeo Usui and Ichiro Hayakawa , University of Tsukuba and Okayama University , respectively ) ( Chinen et al . , 2015 ) was added 25–60 min before imaging ( Figure 5A–C , Figure 5—figure supplement 1A ) . To test microtubule re-growth in the presence of gatastatin ( Figure 5—figure supplement 1B ) , all cells were treated with 10 μM STLC for 6 hr to create monopolar spindles . For cold treatment , cells were then placed on ice; after 1 hr , 30 μM gatastatin or equivalent ( 0 . 1% ) DMSO was added . After 10 more minutes on ice , cells were moved to room temperature for 1 min before fixation . Control cells ( no ice ) were incubated in media containing 30 μM gatastatin or DMSO for 11 min at room temperature before fixation . For all microtubule re-growth experiments after nocodazole washout , cells were treated with 5 μM nocodazole ( M1404; Sigma ) for 15 min at 37°C . After three washes , cells were incubated at room temperature for 6–8 min before fixation and immunofluorescence . 2xGFP-Arp1A was made by inserting EGFP from pEGFP-N1 ( Clontech , Takara Bio USA , Mountain View , CA ) by Gibson assembly between GFP and Arp1A of GFP-Arp1A ( human Arp1A in a pBABE variant , Addgene 4432; gift from I . Cheeseman , Whitehead Institute ) ( Kiyomitsu and Cheeseman , 2012 ) . 2x-GFP-Arp1A localized correctly to kinetochores and poles , and spindle organization was unperturbed . To make Cas9-resistant GFP-NuMA ( ‘GFP-NuMA_resistant’ ) , full-length human NuMA ( NM_006185 . 3 ) with silent mutations ( 5’-GTGTCAGAGAGACTGGACTTT-3’ mutated to 5’-GTTAGTGAACGCTTGGATTTT-3’ , preserving amino acids 57–62 of NP_006176 . 2 ( ‘VSERLD’ ) ) was synthesized and cloned ( Epoch Life Science , Missouri City , TX ) into pEGFP-N1 at BglII and EcoRI sites . Subsequent truncations of NuMA ( ‘N-C’ , ‘C’ , ‘C-tail1’ , ‘C-tail2’ , ‘C-tail2A’ , ‘C-tail2B’ ) were synthesized and cloned ( Epoch Life Science ) into ‘GFP-NuMA_resistant’ at BglII and HindIII sites . To make GFP-N-Tau , NuMA amino acids 1–1410 from ‘GFP-NuMA_resistant’ followed by a flexible linker and MAPTau ( NM_01684 . 1 ) from pmEmerald-MAPTau-C-10 ( gift from M . Davidson , Florida State University ) were synthesized and cloned ( Epoch Life Science ) into ‘GFP-NuMA_resistant’ at HindIII and SalI sites . Other plasmids used: DsRed-p150217-548 ( CC1; amino acids 217–548 of chicken p150 in pDsRed-N1 , Clontech , gift from T . Schroer , Johns Hopkins University ) ( Quintyne and Schroer , 2002 ) ; mCherry-p50 ( chicken p50 in mCherry-C1 , Clontech , gift from M . Moffert and T . Schroer , Johns Hopkins University ) ( Shrum et al . , 2009 ) ; GFP-NuMA ( human NuMA in pEGFP-N1 , Clontech , gift from D . Compton , Dartmouth Medical School ) ( Kisurina-Evgenieva et al . , 2004 ) ; GFP-CAMSAP1 ( human CAMSAP1 in pEGFP-C1 , Clontech , gift from A . Akhmanova , Utrecht University ) ( Jiang et al . , 2014 ) . To determine the percentage of p150 at plus-ends vs . minus-ends ( Figure 1B , Figure 4D ) , we used single microtubules where both ends were clearly visible . We found that EB1 consistently labeled just one end , the plus-end . We determined p150 localization relative to the EB1-labeled plus-end and calculated the percentage of p150 at each location within each cell . Percentages for multiple cells were averaged for Figures 1B and 4D . Pre-NEB cells were distinguished from post-NEB cells by the exclusion of microtubules from the nucleus , circle-shaped chromosome packing in the nucleus , and , when possible , NuMA localization within the nucleus . Kymographs of GFP-Arp1A , GFP-NuMA , and GFP-CAMSAP1 puncta and pole position over time ( Figure 1C–E , Figure 2C , Figure 6D ) were generated in ImageJ ( Version 2 . 0 . 0/1 . 51 hr ) . To measure GFP intensity at ablation-created minus-ends over time ( Figure 1F , Figure 2D , Figure 2—figure supplement 1E , Figure 5B ) , we used a home-written Matlab ( R2012a Version 7 . 4 ) program to integrate GFP intensity within a 1 . 4 μm-diameter circle centered on the manually-tracked k-fiber minus-end , and to measure local background intensity within a surrounding 2 . 7 μm-diameter ‘donut’ . Code is available at https://github . com/chueschen/IntensityAtMinusEnd ( Hueschen , 2017; copy archived at https://github . com/elifesciences-publications/IntensityAtMinusEnd ) . After background subtraction , the intensity measured at the cut site during the three frames before ablation ( k-fiber intensity ) was averaged and set to zero . For NuMA and Arp1A , we then fit a sigmoid function ( y=a1+e- ( x-b ) c where y = intensity and x = time ) to each trace , normalized to plateau height ( a = 1 ) , and solved for the time b at which y = 0 . 5*a to determine time to half-maximum intensity ( Figure 1G , Figure 2E , Figure 2—figure supplement 1F , Figure 5C ) . For CAMSAP1 , we normalized each trace to peak height ( mean intensity from t = 5 s to t = 20 s ) and found the first point at which intensity passed 0 . 5 to determine time to half-maximum intensity . Finally , to generate mean intensity traces , data from all traces were collected into 5 s wide bins in time and their intensities were averaged . Stub length ( distance between k-fiber plus- and minus-ends , Figure 1H ) was measured in ImageJ at the first frame following ablation . Minus-end position data ( Figure 3C ) were generated by manual tracking of ablation-created k-fiber minus-ends ( marked by GFP-CAMSAP1 ) and spindle poles in time-lapse videos , using a second home-written Matlab program ( Elting et al . , 2014 ) . Nearest neighbor distances between NuMA puncta in STORM imaging ( Figure 5H ) were measured as the center-to-center distance from each NuMA puncta to its nearest neighboring puncta . NuMA truncation rescue capability ( Figure 6C ) reports the percentage of bipolar spindles with two focused poles compared to disorganized spindle architecture characteristic of NuMA knockout ( detached centrosomes , loss of k-fiber focusing into two poles ) . Percentage was calculated for each experiment ( n = 3–5 experiments ) and then averaged . All data are expressed as average ±standard error of the mean ( SEM ) . Calculations of correlation coefficients ( Pearson’s r ) and p values were performed in Matlab . One-way ANOVA and Tukey post-hoc tests ( Figures 1G , 5C and 6C ) were performed in Microsoft Excel and Matlab . All other p values were calculated using two-tailed unpaired t-tests with GraphPad Software . Quoted n’s are described in more detail in Figure Legends , but in general refer to individual biological structures analyzed ( biological replicates , for example , individual spindle lengths , individual k-fiber ablations , etc . ) . Time-lapse images ( Figures 1C–E , 2C , 3A , 4A , 5D and 6D ) show a single spinning disk confocal slice , as do immunofluorescence images of microtubule re-growth after nocodazole or cold treatment ( Figures 1A–B and 4D , Figure 5—figure supplement 1B ) and post-ablation confocal immunofluorescence images ( Figure 5E , Figure 2—figure supplement 1A–B ) . 3D STORM images ( Figure 5F–G ) show a single 600 nm slice in Z . Immunofluorescence images of spindles ( Figures 2F , 4B , C , 5A , 6B and E and Figure 1—figure supplement 1 , Figure 2—figure supplement 1D , Figure 6—figure supplement 1B ) show max intensity projections ( 1–2 μm in Z ) of spinning disk confocal Z-stacks . Videos show a single spinning disk confocal Z-slice imaged over time and were formatted for publication using ImageJ and set to play at 35x relative to real time . Videos were corrected to play at a constant frame rate , even when the acquisition rate was not constant .

Every time a cell divides , it needs to duplicate its DNA and evenly distribute it between the two new ‘daughter’ cells . To move and distribute DNA , the cell builds a large machine called a spindle , which is made of stiff cables called microtubules . Many proteins , including a motor called dynein , help to organize the spindle’s microtubules . One of dynein’s jobs is to cluster all microtubules at the two tips of the spindle , pulling them into shape . Without this clustering , spindles have the wrong shape and structure and can make mistakes when moving DNA . Microtubules have a ‘plus’ end and a ‘minus’ end , and motor proteins usually only travel in one specified direction . Dynein , for example , moves towards the minus end of microtubules , which is where most of the clustering happens . It can form a complex with other proteins that help clustering , one of which is called NuMA . Until now , it was thought that dynein transports NuMA to the minus ends . To test this model , Hueschen et al . studied dividing human cells under a microscope and isolated minus ends with the help of a laser . The experiments showed that , in fact , NuMA gets to minus ends independently of dynein . Once it is on the minus ends , NuMA grabs hold of another protein complex called dynactin , which then gathers dynein . Dynein then pulls the spindles into shape from the minus ends . When NuMA was experimentally removed from the cells , dynein-dynactin complexes were scattered along the entire length of the microtubule instead of pulling specifically on minus-ends , which resulted in disorganized spindles . Thus , where dynein complexes pull determines what spindle shape they build . Hueschen et al . also showed that dynein complexes only pull on minus-ends while the cell divides , which makes sense , because NuMA remains hidden in the cell nucleus for the rest of the time . Together , these results suggest that NuMA makes sure dynein pulls specifically on the minus-ends of the microtubules to tighten the spindle at the right time . A next step will be to test how the mechanical properties of the spindle are changed without dynein pulling on minus-ends . A better knowledge of how different proteins work together to build the spindle and help cells divide can help us understand what goes wrong when cells divide abnormally , as in the case of cancer cells .

2017

NuMA recruits dynein activity to microtubule minus-ends at mitosis

Although molecular recognition is crucial for cellular signaling , mechanistic studies have relied primarily on ensemble measures that average over and thereby obscure underlying steps . Single-molecule observations that resolve these steps are lacking due to diffraction-limited resolution of single fluorophores at relevant concentrations . Here , we combined zero-mode waveguides with fluorescence resonance energy transfer ( FRET ) to directly observe binding at individual cyclic nucleotide-binding domains ( CNBDs ) from human pacemaker ion channels critical for heart and brain function . Our observations resolve the dynamics of multiple distinct steps underlying cyclic nucleotide regulation: a slow initial binding step that must select a 'receptive' conformation followed by a ligand-induced isomerization of the CNBD . X-ray structure of the apo CNBD and atomistic simulations reveal that the isomerization involves both local and global transitions . Our approach reveals fundamental mechanisms underpinning ligand regulation of pacemaker channels , and is generally applicable to weak-binding interactions governing a broad spectrum of signaling processes . Most cellular signaling pathways require or are modulated by the binding of small molecules to integral proteins . However , our understanding of the dynamic events involved in these molecular recognition processes comes primarily from inferences based on downstream activity initiated by binding , or ensemble measures that average over and thereby obscure the underlying mechanistic steps . In contrast , single-molecule observations reveal dynamics and heterogeneity of conformational transitions that are otherwise averaged over in ensemble measurements , and thus are a means to probe specific molecular transitions providing important clues to the physical basis for binding ( Joo et al . , 2008; Csermely et al . , 2010; Greives and Zhou , 2014; Guo and Zhou , 2016; Ruiz and Karpen , 1997; Miller , 1997 ) . Single-molecule approaches have provided mechanistic insight in many areas: for example , patch-clamp recordings from single ion channels reveal the network of states that underlie gating of the central pore ( Lape et al . , 2008; Mukhtasimova et al . , 2009; Colquhoun and Lape , 2012; Purohit et al . , 2014 ) , whereas optical techniques such as single-molecule FRET ( smFRET ) allow tracking of conformations and structural movements of individual domains ( Akyuz et al . , 2015; Cooper et al . , 2015; Wang et al . , 2014; Vafabakhsh et al . , 2015; Landes et al . , 2011; Wang et al . , 2016 ) . However , similar resolution of the fundamental mechanisms underlying individual ligand binding events that initiate or modulate downstream domain movements and pore gating are lacking , primarily due to technical challenges in imaging with both sufficient temporal resolution and at concentrations necessary to drive many physiologically relevant recognition processes . For fluorescence based approaches ( Funatsu et al . , 1995 ) , a major challenge hampering resolution of single binding events with low affinity is the diffraction limit of light microscopy . At the high concentrations necessary to drive these binding reactions , the number of fluorescent ligands within the diffraction-limited excitation volume becomes appreciable , thereby obscuring resolution of individual fluorophores . Unfortunately , many physiologically relevant recognition processes have affinities in the micromolar range , which precludes single-molecule resolution with traditional microscopy techniques including total internal reflection ( TIRF ) or confocal microscopy . To observe micromolar affinity binding events at single molecules , we used zero-mode waveguide ( ZMW ) nanofabricated devices ( Levene et al . , 2003; Zhu and Craighead , 2012 ) . ZMWs limit optical excitation to a sub-diffraction-limited volume such that even at micromolar concentrations there are sufficiently few ligands excited that binding of a single fluorophore in the excitation volume can be resolved . As a notable exception to the overall lack of single-molecule binding observations for physiological processes , ZMWs have been used to great effect to study translation events at individual ribosomes and single-molecule electrochemistry , and have enabled single-molecule genomic sequencing ( Uemura et al . , 2010; Korlach et al . , 2010; Zhao et al . , 2013 ) . Here , we combined ZMWs with smFRET to resolve individual specific binding events of a fluorescent cyclic nucleotide derivative ( fcAMP ) ( Kusch et al . , 2010 ) with micromolar affinity for its receptor CNBD from hyperpolarization-activated cyclic nucleotide-gated ( HCN ) channels critical for oscillatory neuronal activity in the brain and pacemaking in the heart . Although binding of cyclic nucleotides is known to enhance HCN voltage-dependent activation , the mechanisms that underlie this regulation remain unclear . Previous studies of cyclic nucleotide ( e . g . cAMP ) regulation have relied primarily on ensemble channel currents ( Chen et al . , 2007 ) , or more recently on ensemble fluorescence from fcAMP ( Kusch et al . , 2012; Benndorf et al . , 2012; Thon et al . , 2015 ) , to deduce the dynamics of cyclic nucleotide association . Although fcAMP provides a more direct measure of binding than does downstream pore current , both measurements reflect ensemble-averaged data that obscures resolution of the individual steps involved in the binding process . Resolving the dynamics of these steps is important because it provides a rationale for assigning the effect of specific perturbations to distinct mechanistic steps – an invaluable tool for deconstructing the pathway by which binding is transduced to functional changes elsewhere such as at the pore gate . To resolve ambiguity in current ensemble-based models of cyclic nucleotide association , we dissected the intrinsic binding dynamics at single molecules to reveal that cAMP binding involves multiple conformational transitions: an initial binding step that is appreciably slower than expected for a diffusion-limited encounter complex , partly due to selection of the ‘receptive’ conformation , and a subsequent ligand-induced isomerization of the CNBD . Our single-molecule observations in both monomeric and tetrameric CNBD complexes , in conjunction with the first unique X-ray structure of the unliganded CNBD and molecular dynamics ( MD ) simulations , resolve the dynamic mechanisms underlying cyclic nucleotide association at HCN channels to a level of unprecedented detail . For optical tracking of the ligand we used a fluorescent cyclic nucleotide conjugate ( fcAMP ) that modulates HCN2 channel function in a very similar manner to native non-conjugated cAMP ( Kusch et al . , 2010 ) . The purified CNBD from HCN2 channels was engineered to contain a single accessible cysteine residue ( C508A/C584S/E571C ) near the binding pocket that was specifically labeled with a fluorescent FRET acceptor ( Figure 1A–B ) , hereafter referred to simply as CNBD unless stated otherwise . The only other cysteine residue C601 is buried in the hydrophobic core of the CNBD as indicated by both crystal structures ( Zagotta et al . , 2003; Lolicato et al . , 2011 ) and our inability to label this position with a maleimide reactive fluorophore . Ensemble fluorescence anisotropy revealed that both native and mutated CNBDs exhibited similar affinity for fcAMP ( Figure 1—figure supplement 1A ) . Efficient FRET due to specific binding of fcAMP to the acceptor-labeled CNBD was confirmed in bulk solution ( Figure 1—figure supplement 1B ) . 10 . 7554/eLife . 20797 . 003Figure 1 . Imaging ligand binding to single molecules in ZMWs . ( A ) HCN2 channel subunit transmembrane topology is homologous to canonical voltage-gated potassium channel subunits with the exception of a CNBD following the pore lining S6 helix . ( B ) Isolated CNBD colored by secondary structure with fluorescent acceptor DyLight 650 maleimide attached at position E571C and bound donor fcAMP ( cAMP + DyLight 547 ) shown . ( C ) ZMW smFRET imaging setup ( inset shows a rendering of an array of ZMWs ) and ( D ) cartoon of an individual ZMW with a single fcAMP-bound acceptor-labeled CNBD tethered to the optical surface within the aluminum ( Al ) nanopore ( drawing is not to scale ) . Bound fcAMP ( blue sphere ) near the bottom of the ZMW is directly excited and emission from the acceptor ( red sphere ) on the CNBD due to FRET is observed . In contrast , freely diffusing fcAMP molecules are shown near the top of the ZMW where they are outside of the effective near-field observation volume , and thus not observed . The scale bar to the right of the ZMW indicates a typical length constant for the exponentially decaying observation volume S ( z ) , which was estimated by extrapolating reported values for various ZMW diameters ( Levene et al . , 2003 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20797 . 00310 . 7554/eLife . 20797 . 004Figure 1—figure supplement 1 . Bulk solution fluorescence imaging of fcAMP binding . ( A ) Bound probability obtained by measuring fcAMP bulk solution anisotropy plotted as a function of CNBD concentration for wild type and mutant HCN2 CNBDs . ( B ) Solution FRET between bound fcAMP and acceptor-labeled CNBD . Sensitized emission spectra ( excitation at 532 nm ) under various conditions . The emission max of fcAMP is 566 nm and the acceptor on the CNBD is 675 nm . Under FRET conditions , fcAMP serves as a donor and the CNBD as an acceptor . To check specificity of binding , we used excess of non-fluorescent cAMP to displace bound fcAMP . DOI: http://dx . doi . org/10 . 7554/eLife . 20797 . 00410 . 7554/eLife . 20797 . 005Figure 1—figure supplement 2 . Imaging ZMW arrays . Brightfield ( A ) and fluorescence ( B ) images of an array of ZMW nanoholes with diameters of approximately 200 nm . Bright spots in the fluorescence image reflect the subset of ZMWs that contain a fluorescently labeled CNBD molecule ( one of these is indicated by an arrow ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20797 . 005 For single-molecule measurements , isolated acceptor-labeled CNBDs were sparsely deposited in ZMW arrays bathed in concentrations of fcAMP ranging from 0 . 1–10 µM . Donor and acceptor emission was simultaneously recorded from arrays of distinct ZMWs in response to alternating direct excitation of either donor or acceptor ( Santoso et al . , 2008 ) ( Figures 1C–D and 2A , Figure 1—figure supplement 2 ) . Representative fluorescence time traces for fcAMP binding to single CNBD molecules are shown in Figure 2 . Unlike typical FRET experiments where both the donor and acceptor are immobilized , the donor ( fcAMP ) was free to diffuse in and out of the excitation volume surrounding the immobile acceptor on the CNBD . Because the time for free fcAMP to diffuse across the excitation volume ( ~1 ms ) is extremely short compared to the duration of a single image frame ( 100 ms ) , we did not resolve the diffusion of fcAMP to or from the CNBD , but only observed increased fluorescence when fcAMP remained within the excitation volume for a time period comparable to or longer than the frame duration , as when bound to the CNBD . In addition , observation that excitation of a bound donor resulted primarily in an increase in acceptor emission , with little to no observable increase in donor emission , suggests that FRET efficiency between bound fcAMP and the nearby acceptor was close to 100% . This response is not due to blinking of a weakly excited acceptor during donor excitation because direct excitation of the acceptor on alternating frames resulted in stable acceptor emission intensities until the acceptor bleached . Furthermore , whenever fcAMP remained bound at the time of acceptor bleaching , we immediately observed emission from the donor . The FRET signal shown in Figure 2B could also be abolished by competition with excess non-fluorescent cAMP ( Figure 2—figure supplement 1 ) , and thus represents a time-dependent binary readout for specific fcAMP binding at single molecules . 10 . 7554/eLife . 20797 . 006Figure 2 . Singe-molecule FRET imaging of binding dynamics at micromolar concentrations . ( A ) Cartoon depicting smFRET during fcAMP binding . Direct excitation of the donor fcAMP results in stimulated emission from the acceptor on the CNBD due to efficient FRET while the donor is bound up until the acceptor bleaches , after which only emission from the donor is observed . ( B ) Single-molecule fluorescence time series for fcAMP binding to individual acceptor-labeled CNBDs within ZMWs . Simultaneous emission from donor ( blue ) and acceptor ( red ) upon donor excitation at 532 nm was interleaved every other frame with emission from acceptor ( magenta ) upon direct excitation at 640 nm . Acceptor fluorescence for both excitation conditions is overlaid with the idealized time series ( black ) . fcAMP concentration is indicated in the upper left of each donor fluorescence trace . DOI: http://dx . doi . org/10 . 7554/eLife . 20797 . 00610 . 7554/eLife . 20797 . 007Figure 2—figure supplement 1 . Specific fcAMP binding at single molecules . ( A–C ) Single-molecule fluorescence time series for individual acceptor-labeled CNBDs in the absence ( A ) and presence ( B ) of 3 µM fcAMP , and during competition between 3 µM fcAMP and 5 mM non-fluorescent cAMP ( C ) . Simultaneous emission from donor ( blue ) and acceptor ( red ) upon donor excitation at 532 nm was interleaved every other frame with emission from acceptor ( magenta ) upon direct acceptor excitation at 640 nm . The lack of smFRET events during competition with cAMP indicate a lack of non-specific binding . ( D ) Summary of the binding event rate from smFRET traces across molecules for the same conditions described above . DOI: http://dx . doi . org/10 . 7554/eLife . 20797 . 007 Figure 2B provides a visual illustration of both the challenge of resolving single binding events at micromolar concentrations and the advantages of ZMWs . First , at the concentrations required to drive fcAMP binding , fluorescence from individual bound donors cannot be reasonably resolved apart from the freely diffusing donors within a diffraction-limited spot using traditional approaches such as confocal or TIRF microscopy . In contrast , ZMWs reduce the excitation volume beyond that achievable with TIRF , such that even at micromolar concentrations a sufficiently small number of fluorophores are excited that single bound donors can be detected ( Levene et al . , 2003; Zhu and Craighead , 2012 ) ( e . g . donor fluorescence after acceptor bleaching in Figure 2B ) . The combination of ZMWs and smFRET can further reduce background fluorescence from free donor and yet allow resolution of single binding events while limiting artifacts due to nonspecific binding at the surface ( e . g . smFRET in Figure 2B ) . Furthermore , alternating between donor and acceptor excitation provides explicit information on the number of molecules within each ZMW in the form of acceptor bleach steps ( e . g . acceptor fluorescence in Figure 2B ) . The probability of being in a bound state across all time points and all molecules as a function of fcAMP concentration was described by a binding curve with an apparent dissociation constant of 1 . 5 µM ( Figure 3A ) . The similar affinities obtained with both single-molecule and bulk solution anisotropy measurements ( 2 . 3 µM; Figure 1—figure supplement 1A ) suggest that non-specific interactions with the ZMW surface are weak , and that our observations reflect inherent dynamics of monomeric isolated CNBDs . 10 . 7554/eLife . 20797 . 008Figure 3 . Single-molecule cyclic nucleotide binding dynamics at HCN2 CNBDs . ( A ) Bound probability from the total fraction of time spent bound for all single molecules as a function of fcAMP concentration fit with the equation Bmax/ ( 1+Kd/[fcAMP] ) , where Bmax = 0 . 75 is the maximal bound probability and Kd = 1 . 5 µM is the apparent dissociation constant . ( B , C ) Histograms of unbound and bound single-molecule dwell time distributions for events from all molecules combined for ( B ) various concentrations of fcAMP and ( C ) distributions for 1 µM fcAMP overlaid with maximum likelihood estimates for monoexponential ( blue dashed ) and biexponential ( red ) distributions . Exponential fits with estimated parameters and confidence intervals for all tested fcAMP concentrations are shown in Figure 3—figure supplement 1 . ( D ) Contour plots of two dimensional histograms for the average bound time versus average unbound time per molecule at several fcAMP concentrations . Color bar denotes number of molecules . Symbols denote time constants from maximum likelihood biexponential fits ( open circle and triangle ) and their amplitude-weighted average ( asterisk ) . Similar contours for all tested fcAMP concentrations are shown in Figure 3—figure supplement 2 . ( E ) Kinetic models between unbound ( U* ) and bound ( B* ) states . The model ID number is indicated to the left of each model . Optimized rate constants are given in Table 1 . ( F ) Differences in the Akaike Information Criteria ( AIC ) for optimized models shown in ( E ) . ( G ) Comparison of observed dwell time histograms with simulated data from model 4 . Histogram abscissas for ( B ) , ( C ) and ( G ) were truncated to facilitate visualization of the faster components . ( H ) Cartoon illustrating a tetrameric CNBD complex formed by appending tetramerizing GCN4 coiled-coil to the N-terminus . ( I ) Example fluorescence time series ( blue ) for fcAMP binding events at a single tetramer in a ZMW overlaid with the idealized trace ( black ) . At 300 nM fcAMP , the probability that more than one CNBD is bound at any given time is low . ( J ) Dwell time distributions for the first binding step in CNBD tetramers . Bound lifetimes are biexponentially distributed as for monomeric CNBDs . DOI: http://dx . doi . org/10 . 7554/eLife . 20797 . 00810 . 7554/eLife . 20797 . 009Figure 3—figure supplement 1 . Maximum likelihood estimates of biexponential parameters for single-molecule dwell time distributions . ( A ) Histograms of unbound and bound single-molecule dwell time distributions ( gray ) overlaid with maximum likelihood estimates for monoexponential ( blue dashed ) and biexponential ( red ) distributions . Histogram abscissas were truncated to facilitate visualization of the faster components . ( B ) Time constants and their relative amplitudes from biexponential maximum likelihood estimates of unbound and bound dwell time distributions shown in ( A ) as a function of fcAMP concentration . Error bars are 95% confidence intervals . DOI: http://dx . doi . org/10 . 7554/eLife . 20797 . 00910 . 7554/eLife . 20797 . 010Figure 3—figure supplement 2 . Dwell time correlations within single molecules . Contour plots of two dimensional histograms for ( A ) the average bound time versus average unbound time per molecule , ( B ) first order correlation between the dwell times of sequential unbound ( event i ) and bound ( event i + 1 ) events , ( C ) second order correlation between successive unbound events i and i + 2 ( separated by a bound event ) , and ( D ) successive bound events ( separated by an unbound event ) . Color bar denotes number of molecules and fcAMP concentrations are indicated in the upper left of each plot . Symbols in ( A ) denote time constants from maximum likelihood biexponential fits ( open circle and triangle ) and their amplitude-weighted average ( asterisk ) ( see Figure 3—figure supplement 1 ) . The lack of clear correlations between events in ( B-D ) is consistent with a model where each binding event gives rise to the same equilibria between bound states . DOI: http://dx . doi . org/10 . 7554/eLife . 20797 . 01010 . 7554/eLife . 20797 . 011Figure 3—figure supplement 3 . CNBD tetramer . Size exclusion chromatography confirms that the GCN4pLI-HCN2 ( CNBD ) tetramer is stable in solution and exhibits no detectable dissociation into monomers . ( Inset ) SDS-PAGE overloaded with respect to GCN4pLI-HCN2 to access its purity . Under these conditions , GCN4pLI-HCN2 is a monomer . DOI: http://dx . doi . org/10 . 7554/eLife . 20797 . 01110 . 7554/eLife . 20797 . 012Figure 3—figure supplement 4 . Binding with and without smFRET . Comparison of dwell time distributions from monomeric CNBDs either before acceptor bleaching ( i . e . from smFRET trace ) or after acceptor bleaching ( i . e . fcAMP fluorescence alone ) . See Figure 2B for example traces . The nearly identical distributions suggest that analysis of fcAMP fluorescence alone is a valid binding reporter at lower concentrations where such events can be directly resolved . DOI: http://dx . doi . org/10 . 7554/eLife . 20797 . 012 As expected for a ligand binding reaction , distributions of unbound dwell times derived from fitting idealized smFRET traces shifted to shorter durations with increasing fcAMP concentration , whereas the bound time distributions were essentially independent of fcAMP concentration ( Figure 3B ) . Maximum likelihood estimates of unbound and bound dwell time distributions required two exponential components each ( Figure 3C , Figure 3—figure supplement 1 ) , whereas monoexponential distributions resulted in poor descriptions of the data , and triexponential distributions reduced to biexponential distributions in all cases ( i . e . one component had zero amplitude ) . The probability that any of the observed unbound or bound time distributions were monoexponential as opposed to biexponential was less than 0 . 001 as determined from a χ2 distribution with two degrees of freedom corresponding to twice the difference in their log likelihoods . The biexponentially distributed dwell times suggest that either each CNBD can sample two unbound and two bound metastable conformations with two corresponding rate constants for each process , or our data is comprised of a bimodal population of CNBD molecules , each with distinct single-step binding dynamics . In the latter case , the correlation between the average bound versus unbound time for each isolated CNBD should cluster into two groups corresponding to the time constants for binding and unbinding in each population . However , at all observed fcAMP concentrations the per molecule average unbound versus bound time distribution was centered around the weighted average of the two pairs of time constants and broadly distributed between them ( Figure 3D , Figure 3—figure supplement 2A ) . This behavior suggests dynamic heterogeneity , where each individual CNBD molecule was able to repeatedly interconvert between two unbound ( U1 and U2 ) and two bound ( B1 and B2 ) configurations during a typical recording ( ~1 min . ) . Furthermore , individual bound durations did not depend on the preceding unbound duration , consistent with sequential models of binding ( Figure 3—figure supplement 2B–D ) . To explore the functional dynamics underlying our data , we compared likelihoods of several kinetic models using hidden Markov modeling ( HMM ) ( Bronson et al . , 2009; Nicolai and Sachs , 2013 ) ( Figure 3E; Table 1 ) . Models were globally optimized for smFRET time series from all molecules and fcAMP concentrations and ranked according to their Akaike information criterion ( AIC ) ( Akaike , 1974 ) ( Figure 3F ) . Consistent with dwell time distributions discussed above , models with two unbound ( U1 , U2 ) and two bound ( B1 , B2 ) states performed better than models with fewer states , whereas adding additional unbound and bound states did not improve the likelihood . Remarkably , all of the models with two bound states converged to a similar set of rate constants governing an initial binding step followed by a subsequent first order transition ( Table 1 ) . The most direct interpretation of this finding is that the CNBD undergoes a reversible conformational change between distinct cAMP-bound conformations . Importantly , simulated data from such a model reproduced the experimentally observed dwell time distributions ( Figure 3G ) . 10 . 7554/eLife . 20797 . 013Table 1 . Kinetic model rate constants . Optimized rate constants ( s-1 or M-1s-1 ) for models shown in Figure 3E . U* and B* denote unbound and bound states , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 20797 . 013ModelU1→ B1 B1→ U1 B1→B2 B2→B1 U1→U2 U2→U1 U1→B2 B2→U1 11 . 3 × 105 0 . 34------21 . 4 × 105 0 . 910 . 520 . 31----31 . 3 × 105 0 . 980 . 490 . 23--0 . 10 × 105 0 . 0442 . 3 × 105 0 . 950 . 510 . 310 . 040 . 15--52 . 2 × 105 1 . 000 . 490 . 250 . 040 . 150 . 14 × 105 0 . 0362 . 4 × 105 1 . 000 . 480 . 270 . 010 . 04--72 . 8 × 105 1 . 110 . 850 . 560 . 170 . 55--ModelU2→B2B2→U2U2→U3U3→U2B2→B3B3→B260 . 24 × 105 0 . 02----7--0 . 020 . 070 . 030 . 08 To test whether the binding dynamics that we observed in monomeric CNBDs were reflective of their functional dynamics in tetrameric channels , we generated CNBD tetramers using a GCN4 tetramerization motif N-terminal of the CNBD to mimic the S6 helices of the channel pore ( Figure 3H , Figure 3—figure supplement 3 ) . Although neither monomeric nor tetrameric constructs are attached to a true channel pore , the tetramer is a useful construct for testing the role of inter-CNBD interactions that are likely to play a role in channel regulation ( e . g . cAMP induces tetramerization of isolated CNBDs even in the absence of a channel pore [Lolicato et al . , 2011] ) . To resolve the dynamics of the first of four binding steps , we observed fluorescence from bound fcAMP at tetramers deposited in ZMWs bathed in 300 nM fcAMP ( Figure 3I ) . At this concentration , single binding events can be resolved directly from the emission intensity time course of fcAMP ( see Figure 2B , Figure 3—figure supplement 4 ) , and thus acceptor labels on tetramers were bleached prior to recording fluorescence from direct excitation of fcAMP ( in contrast , smFRET was needed to resolve binding at higher fcAMP concentrations in monomers; see Figure 2B ) . Analysis of the first binding step in CNBD tetramers confirms that the biexponential nature of the bound time distributions are an inherent property of CNBDs both in isolation and in tetrameric complexes ( Figure 3J ) . Taken together , dwell time analysis for monomers and tetramers and HMM modeling show that following binding , individual CNBDs interconvert between two bound conformations with distinct lifetimes ( as discussed below ) . To gain structural perspective into the observed single-molecule dynamics , we turned to X-ray crystallography . Several structures of the cAMP-bound ( holo ) CNBD from eukaryotic HCN channels have been solved ( Zagotta et al . , 2003; Xu et al . , 2010; Lolicato et al . , 2011 ) . However , a unique conformation of the unliganded ( apo ) CNBD from HCN channels has been recalcitrant to crystallization , with the only reported crystal of the apo form containing a conformation essentially identical to the ligand-bound form ( Taraska et al . , 2009 ) ( less than 1 Å RMSD difference between the two ) . Distance constraints from nuclear magnetic resonance ( NMR ) spectroscopy has been used to generate a structural model of the apo form of the HCN2 CNBD ( Saponaro et al . , 2014 ) , but this structure has many features that are not consistent with other cyclic nucleotide binding domains found in the Protein Data Bank ( Clayton et al . , 2004; Kim et al . , 2005; Schunke et al . , 2011 ) . These discrepancies include complete unfolding of the conserved P-helix in the phosphate binding cassette ( PBC ) and an unexpectedly high degree of conformational flexibility in the β-roll domain . Hence , we crystallized and solved the structure of the apo HCN2 CNBD as part of a fusion protein with the maltose binding protein ( MBP ) in order to provide a high-resolution counterpart to the existing holo crystal structures . An overview of the structure and representative electron densities are presented in Figure 4—figure supplement 1 , and crystallographic statistics are given in Table 2 . The crystallized construct lacks the first three α-helices of the C-linker , making it unable to tetramerize . As discussed below , this was a key factor because tetramers favor a closed conformation typical of the nucleotide-bound CNBD . The majority of the crystal contacts are formed by MBP , which is significantly more ordered as compared to the CNBD ( mean B-factors for MBP = 22 . 9 Å2 , and CNBD = 43 . 2 Å2 ) , making it unlikely that crystal packing interactions contribute significantly to the observed structure of the HCN2 CNBD ( Figure 4—figure supplement 3 ) . As discussed below , the overall conformation of the CNBD monomer is substantially the same as that found by NMR ( Saponaro et al . , 2014 ) , further suggesting that the structure observed in the crystal is not appreciably altered by crystal contacts or fusion with MBP . 10 . 7554/eLife . 20797 . 014Table 2 . Crystallographic statistics . DOI: http://dx . doi . org/10 . 7554/eLife . 20797 . 014Data collectionSpace groupP21Unit cell dimensions a , b , c ( Å ) 61 . 5 , 42 . 0 , 198 . 4 α , β , γ ( ° ) 90 , 90 . 9 , 90Resolution ( Å ) 24 . 82–2 . 07 ( 2 . 11–2 . 07 ) Unique reflections62519Redundancy*3 . 4 ( 3 . 4 ) Average [I/s] *7 . 6 ( 2 . 2 ) Completeness ( % ) *99 . 9 ( 100 ) Rmerge ( % ) *11 . 5 ( 66 . 7 ) RefinementNumber of atomsprotein7913maltose46solvent381Rwork ( % ) *18 . 0 ( 21 . 6 ) Rfree ( % ) *22 . 1 ( 26 . 0 ) Twin fraction0 . 122Average B-factors , ( Å2 ) protein ( MBP ) , protein ( HCN2 ) 22 . 9 , 43 . 2solvent28 . 9R . M . S . deviations , bond angles ( ° ) 1 . 322R . M . S . deviations , bond lengths , ( Å ) 0 . 009Ramachandran plot ( % ) favored98allowed2* Values in parentheses are for the outer resolution shell . The large-scale structural changes between apo and holo structures involve both N- and C-terminal α-helical fragments that undergo a rigid body rotation around hinges that correspond to the points where they emerge out of the rigid β-roll cage ( Figure 4A–B ) . The result of these rotations is the closed conformation in which the C-helix swings toward the ligand binding site and caps bound cAMP . The core β-roll structure is essentially invariant to the presence of ligand ( Figure 4C ) . Strikingly , when the N- and C-terminal helical termini are treated as a single module , they too can be well superposed between the apo and holo conformations ( Figure 4D ) . This indicates that the hinge-like motions of both termini ( Figure 4E–F ) are coordinated in a way that the α-helical structures combine to form a single rigid body domain . The coordinated movement of N- and C-terminal domains is necessitated by the fact that motion of the C-helix toward cAMP introduces numerous steric clashes with the apo conformation of the N-terminal domain ( Figure 4G ) . These clashes are relieved in the holo conformation by an upward movement of the N-terminus by as much as 7 Å ( compare Figure 4A and B ) , ultimately placing the N-terminal C-linker into a tetramerization-competent state . This conformational coupling between the two α-helical domains enables cAMP-induced structural changes to be transmitted across the entire CNBD toward the gating pore . This idea is illustrated schematically for a pair of CNBDs in Figure 4H . 10 . 7554/eLife . 20797 . 015Figure 4 . Comparison of apo and holo forms of the HCN2 CNBD . ( A ) X-ray crystal structures of apo ( this work ) and ( B ) holo ( PDB 3U10 ) conformations of the HCN2 CNBD colored according to protein secondary structure . Bound cAMP in the holo structure is shown as spheres . Helical domains are labeled D' , E' , A , B , C and P as was previously done for the holo structure ( Zagotta et al . , 2003 ) . ( C–D ) Apo structure ( red ) overlaid with either the β-roll domain ( C ) or the α-helical termini ( treated as a single domain ) ( D ) of the holo structure ( blue ) . The apo and holo structures superimpose with an RMSD of 0 . 49 Å over their β-roll domains , or 2 . 04 Å over their α-helical terminal domains ( N-terminal helix excludes the initial D' segment ) . Hinge-like rotations that account for bulk of conformational changes between apo and holo structures: ( E ) N-terminal helical fragment ( residues 508–534 ) , ( F ) C-terminal helical fragment ( residues 607–634 ) . ( G ) Steric clashes between holo conformation of the C-terminus ( blue ) and apo conformation of the N-terminus ( red ) . Residues that would clash are illustrated as spheres ( M515 , P516 , L517 and L615 , M621 , A624 , F625 ) . ( H ) Cartoon illustrating the cAMP-induced rotation of the α-helical domains ( cyan ) about the rigid β-roll cage ( magenta ) that both caps the bound ligand ( yellow ) and places the N-terminal region in a favorable state for coordinating intersubunit interactions with neighboring CNBDs ( indicated by a faded CNBD ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20797 . 01510 . 7554/eLife . 20797 . 016Figure 4—figure supplement 1 . Overview of the X-ray structure . ( A ) X-ray crystal structure of the HCN2 CNBD in the absence of ligand as a fusion protein with MBP . CNBD segments rainbow colored from N- ( blue ) to C-terminus ( red ) . MBP shown in gray . ( B ) CNBD colored by B-factor . ( C ) Fragments of the composite omit electron density map contoured at 1s level that are representative of the HCN2 CNBD regions with different intrinsic flexibility . Note that the quality of the electron density is uniformly high in the β-roll , whereas the density becomes progressively weaker along the C-helix , indicating its high degree of conformational flexibility in the absence of the ligand . DOI: http://dx . doi . org/10 . 7554/eLife . 20797 . 01610 . 7554/eLife . 20797 . 017Figure 4—figure supplement 2 . Comparison of the X-ray and NMR structures of the apo CNBD . X-ray ( red ) and NMR ( cyan ) structures of the apo CNBD superposed over all Cα atoms . The P-helix in the X-ray structure is highlighted in gold . Although the α-helical termini are similar in both X-ray and NMR structures , significant differences are observed in the β-roll domain in stark contrast to the apo versus holo crystal structures ( Figure 4 ) . These differences include the PBC ( arrowhead ) and all of the loops that comprise the nucleotide binding site ( arrows ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20797 . 01710 . 7554/eLife . 20797 . 018Figure 4—figure supplement 3 . Molecular packing in the crystal of apo MBP-HCN2 . Shown are six asymmetric unit cells viewed down the b axis . MBP is shown in white and HCN2 in rainbow representation . HCN2 moieties are arranged in a loosely packed layer that connects MBP layers that contribute most of the crystal contact areas . DOI: http://dx . doi . org/10 . 7554/eLife . 20797 . 018 The trigger that induces these large-scale conformational changes upon cAMP binding is the localized conformational change in the PBC ( Berman et al . , 2005 ) whose P-helix moves toward the ligand and undergoes a subtle transition from a mostly 310-helix to a mostly α-helix ( Figure 4C , Table 3 ) . In contrast to the reported NMR apo structure where the P-helix residues are fully unfolded , the P-helix remains helical in the X-ray apo structure reported here ( Figure 4—figure supplement 2 ) , consistent with apo versus holo crystal structures of CNBDs from both MlotiK1 ( Clayton et al . , 2004; Schunke et al . , 2011 ) and the regulatory subunit of PKA ( Kim et al . , 2005 ) . 10 . 7554/eLife . 20797 . 019Table 3 . cAMP-dependent change in H-bonding within PBC . DOI: http://dx . doi . org/10 . 7554/eLife . 20797 . 019Apo ( this work ) Holo ( PDB 3U10 ) carbonylamided , Åpatterncarbonylamided , ÅpatternG581C5843 . 42i + 3 G581———E582L5853 . 02i + 3 E582L5863 . 07i + 4 I583L5863 . 10i + 3 I583L5863 . 10i + 3 I583T5872 . 71i + 4 I583T5872 . 71i + 4 C584———C584R5882 . 91i + 4 Although crystal X-ray and solution NMR structures are in agreement on large-scale movements of the α-helical termini , they substantially differ in large stretches of residues that comprise the β-roll ( Figure 4—figure supplement 2 ) . In the apo-structure presented here , the β-roll is essentially superimposable to those observed in the different holo structures solved under a variety of conditions ( Zagotta et al . , 2003; Taraska et al . , 2009; Saponaro et al . , 2014 ) . Thus , the β-roll domain , save for the P-helix and the flexible T566-E571 loop , represents a semi-rigid cage built around the cAMP binding pocket . Considering the slow time scale associated with the U1-B1 and B1-B2 transitions , and the fact that there is little explicit experimental characterization for the B1 state , we do not anticipate that molecular dynamics ( MD ) simulations for these transitions are warranted . However , MD simulations are useful for further understanding how the ligand ( cAMP ) stabilizes the holo structure at both local and global scales . Along this line , we have carried out microsecond MD simulations for the holo structure with cAMP removed from the binding site , and for the apo structure with cAMP manually docked into the binding site; cAMP-bound holo simulation and apo simulation without any ligand were also carried out to serve as controls and references ( Figure 5 ) . These simulations are not designed to probe the conformational transition pathway; indeed , the holo simulation with cAMP removed does not correspond to a process in the proposed kinetic model ( Figure 6 ) . Instead , by monitoring structural responses to a change in the ligation state , we are able to identify structural motifs that are most directly stabilized by cAMP binding; this is not straightforward to accomplish by examining the static crystal structures alone , especially when allosteric effects are implicated . 10 . 7554/eLife . 20797 . 020Figure 5 . Global and local structural features stabilized by cAMP . ( A ) RMSDs from apo ( this work ) and holo ( PDB 3U10 ) crystal structures for the HCN2CNBD ( residues 515–632 ) during 1 µs simulations starting in either the apo or holo structure both with and without cAMP in the binding site . Structures corresponding to the first ( blue ) and last ( yellow ) frames of the simulation for the holo structure with cAMP removed are shown ( N-terminal helices omitted for clarity ) aligned to the β-roll domain . Removal of cAMP from the holo structure resulted in the B , C and P helices swinging outwards towards their apo positions . ( B ) The P-helix adopts a mostly α-helix or mostly 310-helix in the holo and apo crystal structures , respectively . Hydrogen bonds are indicated by dashed lines . ( C ) Carbonyl to amide hydrogen bond distances for select P-helix residues over the course of MD simulations starting in the holo ( upper ) and apo ( lower ) form either with ( left ) or without ( right ) cAMP . The stable α-helical form in the presence of cAMP transitions to a stable mostly 310-helix upon removal of cAMP from the holo structure . Upon addition of cAMP to the apo form , the i/i + 4 interactions associated with the α-helical form become more stable while the i/i + 3 interaction associated with the 310 form is weakened . DOI: http://dx . doi . org/10 . 7554/eLife . 20797 . 02010 . 7554/eLife . 20797 . 021Figure 6 . A structural model of HCN2 CNBD binding dynamics . A structural model of cAMP ( red spheres ) binding dynamics at monomeric CNBDs from HCN2 channels . Rate constants ( s-1 or M-1s-1 ) were optimized using HMM modeling of idealized single-molecule fcAMP binding time series as discussed in the text . Model depicts selective binding of cAMP ( horizontal transition ) to the apo form of the CNBD ( U1 ) and a subsequent isomerization ( vertical transitions ) of the ligand-bound CNBD ( B1 ) to its holo form ( B2 ) . Isomerization of the unliganded CNBD ( U2 ) prevents cAMP binding due to occlusion of the binding site by the C-helix . DOI: http://dx . doi . org/10 . 7554/eLife . 20797 . 021 As shown in Figure 5 , changing the occupancy of the binding site induces structural transitions at both local and global scales; the fact that these changes are observed at the microsecond scale in unbiased MD simulations indicate that they are tightly coupled to cAMP binding . At the local scale , the most visible trend concerns the structural stability of the P-helix , which adopts a 310 helical configuration in the apo crystal structure but an α-helix in the holo crystal structure . In the holo-cAMP simulation , two key i/i + 4 helical contacts ( E582-L586 and C584-R588 ) are seen to readily break , while an i/i + 3 contact ( G581-C584 ) characteristic of a 310 helix is formed; in the holo+cAMP control simulations , the aforementioned i/i + 4 contacts remain stable during the microsecond simulation while the G581-C584 contact is not formed . Similarly , in the apo+cAMP simulation , the two i/i + 4 helical contacts are clearly more stable than in the apo-cAMP control simulation ( although the C584-R588 contact constantly breaks and forms with a duration for the bound state spanning from ~10–100 ns ) , while the G581-C584 contact becomes less stable once cAMP is docked into the apo structure . These consistent trends from independent simulations clearly highlight the direct impact of cAMP binding on the secondary structure of the P-helix . At the global scale , changing the ligation state in the holo and apo structures induces limited structural transitions during the microsecond simulations , as expected considering the diffusive and slow nature of the U1-B1 and B1-B2 transitions . Nevertheless , it is encouraging that removing cAMP from the holo structure is seen to induce a substantial displacement of the C-terminal α-helices ( B , C helices ) and the P-helix , which all swing outwards away from the binding site towards their positions in the apo structure , similar to the cAMP-dependence of relative distances between many residue pairs observed with FRET and electron paramagnetic resonance ( EPR ) ( Taraska et al . , 2009; Puljung and Zagotta , 2013 ) . Although the holo-cAMP simulation does not correspond to the putative encounter complex in the kinetic scheme , the observed structural relaxation helps reveal the motifs whose stabilization relies most critically on cAMP binding . Such information provides further understanding on the impact of cAMP binding , especially regarding the coupling between local and distal structural transitions . Typical optical approaches utilizing fluorescent ligands are incapable of resolving low affinity binding events at the single-molecule level due to the large background fluorescence from the high ligand concentrations necessary to drive association . Here , we show that ZMW’s in conjunction with smFRET can resolve individual specific binding events in the presence of concentrations of at least 10 µM fluorescent species , extending single-molecule approaches to more physiologically relevant lower affinity binding events than previously possible while eliminating the influence of non-specific binding ( e . g . Figure 2—figure supplement 1 ) ( Zhu and Craighead , 2012 ) . Our observations resolve the dynamics of the distinct events that underlie cyclic nucleotide association in HCN channels , and reveal that cAMP selectively binds to the ‘receptive’ state of the CNBD , and thereafter induces an isomerization between two distinct bound conformations . As expected given its modular nature ( e . g . isolated CNBDs retain similar binding affinities to full length channels and form cAMP-induced tetramers in solution [Lolicato et al . , 2011] ) , we show that this isomerization occurs in both monomeric and tetrameric CNBD complexes , and thus is likely to reflect inherent properties of the CNBD in channels . We also successfully crystallized the HCN2 CNBD in a unique unliganded state , which provides a high resolution counterpart to previous ligand-bound structures to aid in structural interpretation of the binding mechanism . The single-molecule binding data combined with the structural information suggest a model where the initial binding step reflects selective cAMP association to the apo form of the CNBD , and the subsequent isomerization involves a coordinated rotation of the N- and C-terminal α-helices about the rigid β-roll whereby the C-helix caps the bound ligand as in the holo structure ( Figure 6 ) . HMM modeling suggests that binding occurs selectively from one of two unbound states , which implies that the interchange between these states involves an occlusion of the binding site in the apo form , likely by the C-helix . The slow dynamics for this process imply a relatively large energy barrier that is reduced upon binding cAMP , consistent with EPR observations that the C-helix adopts a cAMP-dependent equilibrium between apo and holo positions ( Puljung et al . , 2014; Deberg et al . , 2016 ) . Based on both the model predictions that transitions between U2 and B2 states occur either infrequently or not at all , and the structural observation that the C-helix occludes access to the binding site in the holo form , we rule out cyclic models that allow binding to the isomerized state ( e . g . model 6 in Figure 3E ) . Our observations provide the first direct evidence that the binding rate is relatively slow , which suggests that fcAMP binding to the ‘receptive’ state of the CNBD is rate-limited by additional processes other than simple diffusion , such as ligand reorientation , desolvation and structural rearrangements in the binding site including the P-helix 310 to α-helix transition . Thus , the initial encounter complex is either relatively unstable , such that brief encounters shorter than our frame rate were not observed , or it sees a significant barrier to its formation . Regardless , following formation of the initial bound complex the CNBD undergoes a reversible isomerization between two bound forms . Comparison of apo and holo X-ray structures shows that the N- and C-terminal helices are rotated in the isomerized state where the C-helix caps the binding site ( Figure 4 ) . Thus , the observed isomerization may reflect dynamic movement of the C-helix between apo and holo positions ( Figure 6 ) . However , we cannot rule out the possibility that the C-helix also moves during the initial binding step; indeed , by perturbing the ligation state of the CNBD in holo and apo structures in otherwise unbiased molecular dynamics simulations at the microsecond scale , we observe that the cyclic nucleotide has a direct impact on the structure of the P-helix and orientation of the B , C helices in the C-terminal region ( Figure 5 ) . In this case , the subsequent isomerization may involve rearrangements in the N-terminal C-linker , a region that was truncated in order to obtain the apo structure , and therefore for which we have incomplete structural information regarding its unliganded form . Nonetheless , these data establish the dynamics of the distinct events that underlie both selective binding and cAMP-induced conformational changes that regulate the channel pore . We note , however , that although it is highly plausible that fcAMP and cAMP bind with similar dynamics given their similar affinities and functional effects , we cannot rule out the possibility that cAMP association dynamics differ from those observed here for fcAMP . The large-scale conformational change of the HCN2 CNBD induced by cAMP binding can be viewed as a rotation of the N- and C-terminal helices , taken as a single domain , in relation to the rigid β-roll . Crucially , the C-helix that caps the bound ligand moves in conjunction with the N-terminal helices connected to the tetramerization module ( C-linker ) and channel pore . Thus , a reciprocal relationship between ligand binding and tetramerization inevitably follows: ligand binding is expected to shift the equilibrium toward tetramerization and , in turn , anything that promotes tetramerization would be expected to enhance ligand binding . This idea is schematically illustrated in Figure 4h , and provides a facile explanation for cAMP-induced tetramerization of monomeric CNBDs in bulk solution and lower affinity for cAMP of CNBDs lacking a C-linker ( Lolicato et al . , 2011 ) . The notion that ligand activation involves isomerization of a ligand-bound receptor to a ‘flipped’ or ‘primed’ configuration has been predicted primarily on the basis of analysis of downstream current recordings ( Lape et al . , 2008; Mukhtasimova et al . , 2009; Colquhoun and Lape , 2012; Purohit et al . , 2014; Thon et al . , 2015; Goldschen-ohm et al . , 2014; Gielen et al . , 2012 ) . However , these various bound states have previously never been directly observed . Our studies here provide direct evidence for such a binding process , and also shed new light on a long-standing debate about the transition pathways that define the binding mechanism–conformational selection versus induced fit . We find that the initial step involves selective binding to the ‘receptive’ state ( U1 ) , which thereafter undergoes a conformational change that traps the ligand in the bound conformation ( B2 ) . Our observations reveal the dynamics and structural detail of the distinct steps that underlie cyclic nucleotide regulation of HCN pacemaker channels , and lay the necessary foundation to probe the molecular details involved in each separate step of multi-subunit complexes . The combination of dynamic and structural observations provide a general approach for revealing the molecular details governing signaling based on weak-binding , a critical and previously inaccessible class of molecular recognition processes . All cloning and mutagenesis was performed using QuikChange-like protocols as described in detail previously ( Klenchin et al . , 2011 ) . All constructs were sequence-verified over the entire ORF length on both strands . Biotin ligase expression construct was made by amplifying full length BirA gene from E . coli genomic DNA and cloning it into a pET21 backbone as an MBP fusion protein . The plasmid backbone contains an unidentified defect that results in about 5-fold lower plasmid copy number than the typical pET vectors . The protein linker sequence between MalE and BirA was SSSSGTASGGATTSENLYFQGG . HCN2 fragment was originally obtained as a synthetic DNA ( Integrated DNA Technologies ) with the sequence that was codon-optimized for expression in E . coli . All residue numbering refers to a mouse HCN2 protein . CNBD fragments were always expressed as fusions with an N-terminally 8His-tagged MBP , cloned into a derivative of pET28 plasmid . For protein expression intended for single molecule imaging experiments , the final construct included , after TEV protease recognition site , an AviTag sequence ( Beckett et al . , 1999 ) , flexible linker and thrombin recognition site . The resulting protein thus contained , after cleavage and purification , the sequence GGLNDIFEAQKIEWHEGASGGSSGGSSGGLVPRGS at the N-terminal end of the HCN2 residues D443-N640 . HCN2 construct intended for crystallization ( residues E494-N640 ) was fused to MBP through a double alanine linker . The GCN4pLI-HCN2 tetrameric construct was the same as the monomeric one except that it had the following sequence inserted between the end of AviTag ( GAS ) and the start of flexible linker ( GGS ) sequences: RMKQIEDKLEEILSKLYHIENELARIKKLLGER For co-expression of biotin ligase and HCN2 constructs ( monomeric or tetrameric versions ) , BL21 CodonPlus ( DE3 ) -RIL cells were sequentially transformed first with the HCN2 construct , selecting transformants overnight on kanamycin/chloramphenicol plates , then with BirA construct , selecting overnight on ampicillin/kanamycin/chloramphenicol plates . Several clones were picked to inoculate 125 ml of MDG medium ( Studier , 2005 ) and cultured overnight at 37°C . 30 ml of the resulting culture was used to inoculate 1 L of LB medium in 2 L shake flasks that were grown at 37°C until OD600 of about 0 . 5 ( all OD600 values refer to measurements done in Beckman DU-640 spectrophotometer ) , at which point 1 ml of 100 mM solution of biotin in DMSO was added to each flask . After additional 30 min of shaking , the cultures were cooled on ice and induced with 1 mM IPTG . After 20 hr of growth at 16°C , cells from 4 L of culture were pelleted , washed in 1 L of ice-cold 20 mM Tris , 100 mM NaCl and 2 mM EDTA , pH 8 . 0 , the cell paste frozen in liquid nitrogen and stored at −80°C until needed . Expression of the construct for crystallization followed the same outline except that seed culture used was grown at 30°C overnight in MDG with kanamycin/chloramphenicol and no biotin was added before induction at OD600 1 . 0 . The biotinylated HCN2 constructs containing the entire C-linker sequences were purified as follows . Unless otherwise stated , all procedures were performed at 4°C . 10 g of frozen cells were resuspended in 60 ml of buffer A ( 20 mM HEPES , 200 mM NaCl , 25 mM imidazole , 0 . 5 mM TCEP , 10% v/v glycerol , pH 7 . 5 ) with an addition of extra 0 . 5 mM TCEP and protease inhibitors ( house-made cocktail equivalent to Roche’s 'cOmplete EDTA-free' tablets ) . The cells were disrupted with ten cycles of sonication on ice-water bath at ~93 W power output while monitoring suspension temperature , keeping cycles short enough to prevent temperature raising above 8°C and resuming at 2–3°C . The suspension was spun for 30 min at 48 , 000 g and the supernatant was loaded by gravity onto a 6 ml Ni-NTA ( Qiagen ) equilibrated with buffer A . The column was then washed by gravity with 200 ml of buffer A followed by two 50 ml wash steps with modified buffer A containing higher final imidazole concentrations , 32 and 40 mM . Remaining bound protein was eluted with 18 ml of modified buffer A containing 250 mM imidazole . Approximately 1 mg of TEV protease per 50 mg of eluted protein was added and the mixture was dialyzed overnight against 2 L of 20 mM HEPES , 100 mM NaCl , 0 . 5 mM TCEP , pH 7 . 5 in dialysis bags with 15 kDa cut-off ( Spectrum Laboratories cat . # 132122 ) . Following dialysis , a precipitate that formed in the case of mutant HCN2s was spun down for 20 min at 48 , 000 g and the supernatant loaded by gravity onto a fresh 10 ml Ni-NTA column equilibrated with 20 mM HEPES , 200 mM NaCl , 0 . 5 mM TCEP , 10% glycerol , pH 7 . 5 . After 100 ml wash with the same buffer that removes minor contaminants , the untagged HCN2 was eluted isocratically with 10 mM imidazole in the equilibration buffer . The eluted protein was concentrated by ultrafiltration ( Amicon Ultra-15 , 10 kDa cut-off ) at room temperature to approximately 7 mg/ml and frozen in liquid nitrogen as 0 . 3 ml aliquots in screw cap tubes . Purification of the MBP-HCN2 fusion for crystallization followed the same protocol as for biotinylated HCN2 with the following exceptions: purification used 15 g of cells and 100 ml of buffer A; ethylene glycol was used in place of glycerol in buffer A; first Ni-NTA column was 10 ml in volume and the second column was 13 ml; the buffer used during the second Ni-NTA contained 100 mM NaCl and no glycerol . The final protein fraction was concentrated by ultrafiltration ( Amicon Ultra-15 , 50 kDa cut-off ) to approximately 140 mg/ml , frozen in liquid nitrogen as droplets of about 30 µl and stored at –80°C until needed . Purification of the GCN4pLI-HCN2 tetramer followed identical steps to the monomer up to the elution from the first Ni-NTA column . Thereafter , the eluate ( 20 ml ) was dialyzed against 20 mM HEPES , 100 mM NaCl , 1 mM TCEP , 0 . 1 mM EDTA , pH 7 . 5 overnight and dialyzate mixed with an equal volume of 40 mM HEPES , 600 mM NaCl , 20% glycerol , 1 mM TCEP , pH 7 . 5 before adding ~1:20 of TEV protease by protein mass for cleavage overnight at 4°C . Precipitated HCN2 tetramer was separated from MBP by centrifugation at 2500 g for 15 min and the pellet resuspended in 40 mM HEPES , 600 mM NaCl , 20% glycerol , 2 mM TCEP , 0 . 1% LDAO ( Sigma , cat . no . 40236 ) , pH 7 . 5 ( Buffer B ) , followed by homogenization in glass-teflon Potter-Elvehjem homogenizer . Following 30 min centrifugation at 40 , 000 g , the solubilized fraction was loaded onto a fresh Ni-NTA column and the GCN4pLI-GCN2 fusion eluted in the Buffer B containing 20 mM imidazole . The eluate ( ~13 mg/ml protein ) was flash-frozen in liquid nitrogen as 0 . 6 ml aliquots and stored at –80°C until needed . In all cases , protein concentration was determined from A280 using theoretical extinction coefficient values derived by the Protparam tool ( Gasteiger et al . , 2005 ) . The C508A/C584S/E571C mutant of CNBD was used for all the single molecule fluorescence experiments . To remove TCEP that may interfere with labeling , a 0 . 3 ml of 7 mg/ml purified biotinylated monomeric protein was buffer-exchanged at room temperature on a 5 ml spin gel-filtration column packed with Bio-Gel P6 ( Fine grade , Bio-Rad ) equilibrated with degassed buffer C ( 20 mM HEPES , 200 mM NaCl , 0 . 2 mM EDTA , 10% glycerol , pH 7 . 5 ) . To increase the recovery , 0 . 3 ml of buffer B was additionally applied immediately after protein application to the column . Working as quickly as possible , the protein was further diluted to 2 mg/ml and mixed with an equal volume of 0 . 14 mM of maleimide DyLight 650 solution that was freshly prepared by diluting 4 . 7 mM stock ( made with dry DMSO ) into degassed buffer B . After two hours of incubation in the dark at room temperature , the reaction was quenched by adding DTT to 20 mM final , the protein was concentrated by ultrafiltration to ~0 . 5 ml ( Amicon Ultra-4 , 10 kDa cut-off ) and purified from unreacted dye by gel filtration on Superdex 10/300 column using buffer C containing 1 mM DTT . Eluted protein fractions were pooled , mixed with 1/19 th volume of 200 mg/ml BSA solution made in buffer C containing 10 mM DTT and frozen in liquid nitrogen as droplets of about 20 µl that were stored at −80°C until needed . Control experiment with the unreactive C508A/C584S mutant demonstrated that less than 1% of total protein in the sample incorporates the dye following the labeling procedure . This confirms the impression from the crystal structures that the third cysteine , C601 , is buried at all times in the beta-roll hydrophobic environment and is not accessible to the hydrophilic reactive dye . The small degree of incorporation that can be seen can be equally due to a small fraction of surface-denatured protein in the mixture as well as some minor cysteine-containing contaminants that are not even seen on a typical protein gel . The GCN4pLI-HCN2 tetramer was labeled using the identical procedure except that during all dilution , buffer-exchange and purification by gel-filtration steps a degassed and argon-saturated buffer B was used . Bulk solution FRET and anisotropy measurements were performed with a spectrofluorometer ( HORIBA Scientific Fluoromax-4 ) in a buffer consisting of 20 mM HEPES , 100 mM NaCl , 20 mM imidazole , 10% glycerol and 0 . 5 mM TCEP . Emission spectra of acceptor-labeled CNBD and fcAMP were obtained with 532 nm excitation , 5 nm slit widths for both excitation and emission , 2 nm steps and 100 ms integration time per step . Anisotropy for fcAMP during titration of non-fluorescent CNBD was measured in a 150 µl cuvette for excitation at 530 nm and emission at 565 nm with 2 nm slit widths and an integration time of 10 s . The fraction of bound CNBDs was computed from the anisotropy curve by first subtracting the anisotropy of freely diffusing fcAMP in the absence of CNBD and then fitting the resulting normalized curve to the equation 1/ ( 1+Kd/[CNBD] ) , where Kd is the dissociation constant for fcAMP binding . Arrays of ZMWs ( Pacific Biosciences ) with a biotin-doped PEG layer on their bottom surfaces were first incubated for 2 min in 0 . 05 mg/ml streptavidin ( Prospec , cat # PRO-791 ) , then washed by exchanging the solution volume five times with wash buffer , which consisted of phosphate buffered saline ( PBS; pH 7 ) supplemented with 50 mM NaCl , 2 mg/ml bovine serum albumin ( BSA ) , 1 mM Trolox , 2 . 5 mM protocatechuic acid ( PCA ) , and either 1 mM TCEP or 2 mM DDT . Next , biotinylated acceptor-labeled monomeric CNBDs were diluted in wash buffer to a working concentration of ~0 . 1 nM and deposited on ZMW arrays by incubating for 10 min followed by 5–10 solution volume exchanges with wash buffer . This resulted in most ZMWs containing either zero or one CNBD molecule . Finally , the solution volume was exchanged for imaging buffer , which consisted of wash buffer plus 250 nM protocatechuate 3 , 4-dioxygenase from Pseudomonas sp . ( PCD; the oxygen scavenging counterpart to PCA ) and various concentrations of 8- ( 2-[DY-547]-aminoethylthio ) adenosine-3' , 5'-cyclic monophosphate ( fcAMP; BioLog ) . Evaporation of the small buffer volume ( ~75 µL ) bathing the ZMWs within a recording period of up to a couple of hours was essentially abolished by placing a glass coverslip over the top of the array . ZMW arrays were placed on an inverted microscope ( Olympus IX-71 ) and imaged under either 532 or 640 nm laser excitation ( Coherent ) . Laser power at the sample was 60 W/cm2 at 532 nm and 25 W/cm2 at 640 nm . Lasers were fed into a single AOTF ( Laser Launch ) that enabled computer control over the excitation time course at each wavelength . The effective observation volume , which is a product of the excitation intensity and emission observation probability , decays exponentially along an axis perpendicular to the sample surface at the bottom of each ZMW . The decay profile depends on both wavelength and ZMW diameter ( Levene et al . , 2003; Zhu and Craighead , 2012 ) . Fluorescence emission from 150–200 nm diameter ZMWs first passed through a multiband dichroic and filter cube for imaging Cy3/Cy5 and equivalent dyes ( Semrock Brightline Cy3/Cy5-A-OMF ) , after which the donor and emission spectra were split with a 650 nm longpass dichroic ( Semrock Brightline ) and bandpass filtered using pairs of edge filters ( donor: 532–632 . 8 nm; acceptor: 632 . 8–945 nm; Semrock EdgeBasic and Brighline ) and then imaged on two separate 512 × 512 EMCCD cameras ( Andor iXon Ultra X-9899 ) at a frame rate of 10 Hz . Both imaging and alternating wavelength excitation on interleaved frames was controlled with Metamorph software ( Molecular Devices ) . Unless specified otherwise , all analysis was accomplished with custom software programs written in either Matlab ( The MathWorks , Natick , MA; https://github . com/marcel-goldschen-ohm ) , Python or ImageJ . Briefly , for each ZMW array , the simultaneously recorded image time series for donor ( fcAMP ) and acceptor-labeled CNBD emission intensities were each further split into two sets of interleaved frames corresponding to direct excitation of either the donor or acceptor . The resulting four image time series are denoted IDD ( donor emission upon direct excitation ) , IDA ( donor emission upon acceptor excitation ) , IAA ( acceptor emission direct excitation ) and IAD ( acceptor emission upon donor excitation ) . The locations of ZMWs in IAA ( which are the same in IAD ) that contained one or more CNBD molecules was obtained from a thresholded mask of the average intensity in the first ~20 frames of IAA , which corresponds to a time period where most acceptors remained unbleached . ZMW locations were further refined by fitting a two dimensional Gaussian to the local intensity height map . The corresponding ZMW locations in IDD were obtained by applying a two dimensional affine transformation that mapped the acceptor emission image from one camera to the donor emission image in the other camera . The time-dependent fluorescence at each ZMW was obtained by projecting the average image intensity in a 5-pixel diameter circle centered on the ZMW location throughout each image time series . IDA was indistinguishable from a constant background , and thereby ignored in further analyses . Only those ZMWs that exhibited a single acceptor bleach step in IAA corresponding to one CNBD molecule were considered for further analysis . The smFRET response IAD for each ZMW was corrected for cross talk due to donor emission in the acceptor channel by subtracting 10% of IDD , as determined from donor emission in the acceptor camera in the absence of donor . Next , IAD was baseline corrected by subtracting the mean intensity after acceptor bleaching , and in some cases where it was visually apparent also subtracting a very slow exponential decay ( t ~100 s ) . Binding dynamics were determined by idealization of IAD prior to acceptor bleaching using the software vbFRET ( Bronson et al . , 2009 ) and allowing for a maximum of two idealized levels . vbFRET accepts as input both a donor ( ID ) and acceptor ( IA ) time series , and idealizes the smFRET response given by IA/ ( IA+ID ) . However , IDD was not anticorrelated with IAD due to the fact that the donor was on the freely diffusing ligand , and therefore not present in the excitation volume throughout the experiment . Thus , to avoid adding noise during the idealization , IAD was idealized directly by providing vbFRET with IA=IAD and ID= max ( IAD ) -IAD , which results in a smFRET response that is an arbitrarily scaled copy of IAD . The idealized records were overlaid on the raw IAD data series after applying the reverse scaling and baseline shifting , and examined visually . ZMWs with either no idealized events or whose idealization did not pass visual inspection were removed from further analysis . The signal to noise ratio ( S/N ) for IAD was computed as the ratio of the rescaled idealized amplitude to the RMS noise for all unbound time points . The distribution of S/N for all ZMWs was well described by a gamma function , and ZMWs with S/N < 2 were rejected from further analysis . The idealization and HMM analyses described here were able to reproduce known models given simulated data at this S/N cutoff . The idealized smFRET time series were interpreted as a binary unbound versus bound time-dependent signal . HMM analysis of single-molecule binding events was performed with QuB ( Qin et al . , 2000; Nicolai and Sachs , 2013 ) . Models were globally optimized to simultaneously describe idealized binding time series for all molecules across all tested fcAMP concentrations with a dead time of 200 ms . For each molecule , the first and last event was removed from the analysis to avoid interpretation of events truncated by our recording window or bleaching of the acceptor , respectively . The number of molecules ( and binding or unbinding events ) that went into the analysis at each fcAMP concentration was: 0 . 1 µM: 1234 ( 1808 ) , 0 . 3 µM: 988 ( 1950 ) , 1 µM: 2028 ( 7795 ) , 3 µM: 1064 ( 4952 ) and 10 µM: 1085 ( 2025 ) . This was more than enough events for the analysis to reproduce known models with similar rates from simulated data . The observation time window prior to acceptor bleaching , and hence the number of events per molecule varied stochastically . Nonetheless , inclusion of molecules with few events ( i . e . that bleached rapidly ) did not grossly distort our modeling results as very similar rate constants were obtained from the subset of molecules exhibiting 10 or more events prior to bleaching . Analysis of fcAMP binding at tetrameric CNBDs was performed in a similar fashion to that for monomeric CNBDs , except that excitation was constant at 532 nm and IDD as opposed to IAD was idealized as described above . The presence of single tetramers was assumed based on sparse labeling ( conditions where ~5% of ZMWs contained a tetramer were first established by bleach-step analysis of acceptor-labeled tetramers ) . The number of tetramers ( and binding or unbinding events ) that went into the analysis was 147 ( 8554 ) , where the larger average number of events per molecule as compared to the monomer reflects the longer imaging periods due to lack of truncation by acceptor bleaching . Given the 100 ms duration of each image frame , our analysis is limited to the detection of dynamics on a comparatively slower time scale . Furthermore , for example , missed brief unbound events will cause bound durations to appear longer than they actually are , which likely contributes to some distortion of our reported dwell times . Most importantly , if such an artifact were to underlie our observation of multiple bound durations , then we predict that with increasing fcAMP concentration ( and decreasing average unbound duration ) we would observe increasingly lengthy apparent bound durations . However , bound time distributions were not obviously concentration dependent , nor were the relative amplitudes or time constants of their maximum likelihood biexponential fits ( Figure 3—figure supplement 1 ) . Thus , any distortions to the dwell times are minor , and our observations reflect aspects of the cAMP association process . Furthermore , our observed kinetics are roughly consistent with prior predictions of cAMP association rates based on macroscopic observations using patch clamp fluorimetry ( Kusch et al . , 2012 ) . Crystals of the MBP-HCN2 were grown by vapor diffusion at room temperature from a 1:1 mixture of protein at 10 mg/ml ( diluted from frozen stock with 5 mM HEPES , 200 mM NaCl , 6 mM maltose , 2 mM TCEP , pH 7 . 5 ) and precipitant solution that contained 34–36% dimethyl polyethylene glycol 500 ( Sigma cat . # 445886 ) , 240 mM KNO3 , 20 mM MgCl2 and 100 mM Bis-Tris , pH 6 . 0 . Typically , 0 . 2–0 . 4 ml of protein:precipitant solution was prepared and drops of varying sizes ( 5–15 µl ) were hung over 0 . 5 , 0 . 75 or 1 . 0 ml well solutions of precipitant . Microcrystals formed readily in the oil phase overnight but over a few weeks clusters of stacked plates nucleated randomly in some of the drops in the aqueous phase . Individual plate fragments that could by dissected away from the stacks had typical dimensions of 100 × 200 × 20 µm . Crystals were cryoprotected by transferring into a precipitant solution containing 6 mM maltose , 5 . 5% v/v of 1 , 6-hexanediol for one minute followed by loop mounting and dipping into liquid nitrogen . Diffraction data were collected at the GM/CA beamline 23-ID-B ( Advanced Photon Source , Argonne National Laboratory , Argonne , IL ) and processed with HKL2000 ( Otwinowski and Minor , 1997 ) . Free reflections were selected with PHENIX ( Adams et al . , 2010 ) using ‘use-lattice-symmetry’ option . The structure was solved by molecular replacement with two copies of MPB ( PDB ID 1ANF ) using Phaser ( Mccoy et al . , 2007 ) . The initial model was built by ARP/wARP ( Perrakis et al . , 2001 ) . This was followed by iterative cycles of manual model building in Coot ( Emsley and Cowtan , 2004 ) and restrained and TLS refinement in Refmac ( Skubák et al . , 2004 ) . The crystal that gave the highest quality data appears to have been pseudo-merohedrally twinned with a twin fraction of approximately 0 . 15 . To account for this , a twin refinement against amplitudes was included in the Refmac refinement protocol . Large-scale domain motions were analyzed using DynDom ( Girdlestone and Hayward , 2016 ) . The structural alignments over a rigid β-roll cage excluded P-helix and flexible loops and were performed with CCP4 program Superpose ( Krissinel and Henrick , 2004 ) using Cα atoms of residues 533–550 , 553–565 , 571–581 and 589–606 . The alignments over alpha-helical termini used residues ranges 510–531 and 608–632 . Unless otherwise stated , the holo structure used in the alignments was PDB 3U10 . All-atom MD simulations were carried out with NAMD 2 . 10 ( Phillips et al . , 2005 ) using the CHARMM36 force field ( MacKerell et al . , 1998; Best et al . , 2012; Hart et al . , 2012 ) in the NVT ensemble at 310 K using Nosé-Hoover and Langevin piston pressure coupling protocols and a one fs time-step . NAMD’s Particle Mesh Ewald algorithm was used for electrostatics with a grid spacing of 1 Å . All nonbonded interactions were treated with a 12 Å cutoff and a switching function turned on at 10 Å . For both the apo ( this study ) and holo ( PDB 1Q5O ) structures , simulations were performed in the presence or absence of cAMP in the binding site . For the monomeric holo structure , only residues 494 to 634 were included for consistency with simulations of the apo structure which lacked most of the C-linker . The initial apo structure in the presence of cAMP was generated by aligning the backbone atoms of the β-roll for the apo crystal structure to the same atoms in the holo structure by minimizing their RMSD , and then inserting the coordinates of cAMP from the holo structure into the apo structure . The built-in CY35 patch for the CHARMM36 force field was used for the topology of cAMP ( Hart et al . , 2012 ) . The N- and C-termini were neutralized using the ACE and CT3 patches ( Best et al . , 2012 ) , respectively , to mitigate charge effects due to the truncation of the protein sequence . Each system was solvated in TIP3P water with 150 mM KCl using the built-in plugin in VMD ( Humphrey et al . , 1996 ) and energy minimized for 500 fs with all heavy atoms constrained . Next , constraints on all non-hydrogen atoms other than Cα and cAMP ( if present ) were released , and an additional 7 ps of energy minimization was performed . With Cα and cAMP still constrained , systems were heated to 310 K over 4 ps , and then equilibrated at constant pressure in the NPT ensemble for 1 ns . The average system dimensions over the final 10 ps were determined and used during the production runs , which were performed in the NVT ensemble as described above . All analysis was performed using VMD ( Humphrey et al . , 1996 ) . Trajectories shown in Figure 5 were smoothed by applying a running average with 30 ns window . The atomic coordinates and structure factors for the apo HCN2 CNBD have been deposited in the Protein Data Bank ( PDB ) under the accession code 5JON .

Certain cells in the heart and brain show rhythmic bursts of electrical activity . Such electrical activity is a caused by ions moving in or out of the cells though a number of ion channel proteins in the cell surface membrane . The voltage across this cell membrane regulates the activity of these so-called pacemaking channels , and so do small molecules like cAMP . Nevertheless , it remained poorly understood how cAMP binding altered how the channels work . This was because researchers had been unable to unambiguously resolve the early binding events , because the available techniques were too limited . Goldschen-Ohm , Klenchin et al . have now overcome these technical limitations and observed individual molecules of cAMP ( which had been first labeled with a fluorescent tag ) binding to the relevant parts of a pacemaking channel from humans . This approach revealed that the binding process happens via a sequence of discrete steps . First , cAMP selectively binds when the binding site of the ion channel adopts a specific shape , called its “receptive” state . Second , part of the protein rotates which changes the shape of the binding site and traps the bound cAMP in place . The trapped molecule is not released until the binding site reverts to its previous shape . These new findings provide the groundwork for future studies to dissect how different parts of pacemaking channels change shape and interact to control these channels’ activities .

2016

Structure and dynamics underlying elementary ligand binding events in human pacemaking channels

The Ca2+-sensor synaptotagmin-1 that triggers neuronal exocytosis binds to negatively charged membrane lipids ( mainly phosphatidylserine ( PtdSer ) and phosphoinositides ( PtdIns ) ) but the molecular details of this process are not fully understood . Using quantitative thermodynamic , kinetic and structural methods , we show that synaptotagmin-1 ( from Rattus norvegicus and expressed in Escherichia coli ) binds to PtdIns ( 4 , 5 ) P2 via a polybasic lysine patch in the C2B domain , which may promote the priming or docking of synaptic vesicles . Ca2+ neutralizes the negative charges of the Ca2+-binding sites , resulting in the penetration of synaptotagmin-1 into the membrane , via binding of PtdSer , and an increase in the affinity of the polybasic lysine patch to phosphatidylinositol-4 , 5-bisphosphate ( PtdIns ( 4 , 5 ) P2 ) . These Ca2+-induced events decrease the dissociation rate of synaptotagmin-1 membrane binding while the association rate remains unchanged . We conclude that both membrane penetration and the increased residence time of synaptotagmin-1 at the plasma membrane are crucial for triggering exocytotic membrane fusion . Synaptic transmission in the nervous system is mediated by the regulated release of neurotransmitters from presynaptic nerve terminals . Before triggering , neurotransmitters are stored in synaptic vesicles . Upon arrival of an action potential , voltage-gated Ca2+-channels open , leading to a rise in cytoplasmic Ca2+ concentration , which triggers the fusion of synaptic vesicles with the presynaptic plasma membrane . The primary Ca2+ sensor for synchronous neuronal exocytosis is synaptotagmin-1 ( syt-1 ) , a resident protein of synaptic vesicles that consists of a short luminal domain , a transmembrane α-helix and two cytoplasmic C2 domains , referred to as C2A and C2B . C2 domains were first identified in protein kinase C ( PKC ) , and have since been detected in many diverse proteins ( Corbalan-Garcia and Gomez-Fernandez , 2014 ) . The C2 domains of syt-1 are composed of structurally conserved eight-stranded anti-parallel β-sandwiches of ~130 residues . These domains contain Ca2+-binding loops at one end of the β-sandwich motif . Upon Ca2+ binding , the electrostatic surface potential of the C2 domains is altered . As a result , Coulombic attraction develops towards the head groups of anionic phospholipids , leading to binding of the C2 domains to the membrane associated with completion of the Ca2+ coordination sphere . Moreover , a conserved polybasic lysine patch located on the C2B domain also binds to anionic lipid in the absence of Ca2+ , being particularly attracted to multivalent phosphoinositides ( PtdIns ) ( Jahn and Fasshauer , 2012; Südhof , 2013; Chapman , 2008; Brose et al . , 1992; Corbalan-Garcia and Gómez-Fernández , 2014; Südhof , 2012 ) . Membrane binding of syt-1 has been widely studied ( Kuo et al . , 2009; Wang et al . , 2003; van den Bogaart et al . , 2012; Vrljic et al . , 2011; Kuo et al . , 2011; Radhakrishnan et al . , 2009; Li et al . , 2006; Bai et al . , 2004 , 2002; Schiavo et al . , 1996 ) , mostly using a cytoplasmic fragment including both C2 domains ( termed the syt-1 C2AB fragment ) . However , the mechanism of syt-1 binding to the membrane remains a matter of controversy . For example , recent studies have reported that a double Lys-to-Ala mutation in the polybasic lysine patch ( termed the KAKA mutant ) does not alter the binding of the syt-1 C2AB fragment to vesicles containing phosphatidylserine ( PtdSer ) and/or phosphatidylinositol-4 , 5-bisphosphate ( PtdIns ( 4 , 5 ) P2 ) ( Radhakrishnan et al . , 2009 ) . By contrast , the same mutant was reported in other studies to decrease the binding of the C2AB fragment to vesicles ( Vrljic et al . , 2011 ) , even in the absence of PtdIns ( Li et al . , 2006 ) , or to almost completely abolish the binding to soluble PtdIns ( 4 , 5 ) P2 , in either the absence or the presence of Ca2+ ( van den Bogaart et al . , 2012 ) . On the other hand , several studies suggest that not only the tandem C2AB fragment ( Vrljic et al . , 2011 ) but also the individual C2B ( van den Bogaart et al . , 2012 ) and C2A domains ( Guillen et al . , 2013; Zhang et al . , 1998 ) might bind to PtdIns predominantly through the Ca2+-binding loops in the presence of Ca2+ , and as a consequence , might compete with PtdSer to complete the coordination sphere of bound Ca2+ ( Honigmann et al . , 2013 ) . To resolve these discrepancies , and to shed light on the binding mechanism of syt-1 to its main lipid effectors , PtdSer and PtdIns , we examined the kinetics of syt-1 binding to vesicles containing different amounts of PtdSer and PtdIns ( 4 , 5 ) P2 . We also used isothermal titration calorimetry ( ITC ) to measure the affinities of the C2AB fragment towards the head groups of PtdSer and PtdIns . Moreover , we performed nuclear magnetic resonance ( NMR ) experiments to determine the residues that are affected by binding to PtdSer and PtdIns ( 4 , 5 ) P2 , and electron paramagnetic resonance ( EPR ) experiments to examine the orientation of the C2AB fragment with respect to the membrane surface . Our results demonstrate that PtdSer and PtdIns ( 4 , 5 ) P2 act in a synergistic manner to enhance the penetration depth of syt-1 and to reduce the dissociation rate from the membrane . Furthermore , our data provide strong evidence that PtdIns binds to the polybasic lysine patch on the C2B domain and does not compete with PtdSer for sites in the Ca2+-binding loops . Importantly , enhanced binding to the membrane in the presence of Ca2+ is exclusively due to a decrease in the dissociation rate ( koff ) , whereas the association rate ( kon ) remains unchanged . Although synergistic , the binding of the syt-1 C2AB fragment is dramatically different for PtdSer- and PtdIns ( 4 , 5 ) P2-containing membranes , suggesting that the modes of binding to these two lipids are distinct in nature . In the first series of experiments , we carried out stopped-flow experiments in order to investigate the binding kinetics of the soluble domain of syt-1 ( termed the C2AB fragment ) to vesicles containing different molar ratios of PtdSer and PtdIns ( 4 , 5 ) P2 . We monitored Förster resonance energy transfer ( FRET ) between the tryptophan residues of the C2AB fragment and dansyl-labeled phospholipids that were incorporated into the liposomes . The time course of the fluorescence traces was fitted to a mono-exponential function in order to determine the observed rate constant ( kobs ) Figure 1a . The measured kobs values were plotted as a function of vesicle concentration and fitted to a linear equation , in which the slope yields kon , the y intercept yields koff and the Kd was calculated by koff/kon ( Hui et al . , 2005 ) ( Figure 1b ) . 10 . 7554/eLife . 15886 . 003Figure 1 . PtdIns ( 4 , 5 ) P2 and PtdSer act cooperatively in the membrane binding of the C2AB fragment . ( a ) Representative time courses of dansyl emission for rapid mixing of 0 . 25 µM syt-1 C2AB fragment ( final concentration ) with increasing concentrations of large unilamellar vesicles ( final liposome concentrations ranging between 0 . 3 nM and 3 nM ) containing PtdChol:PtdSer ( 80:20 , molar ratio ) , measured at 25°C . The apparent rate constant kobs was determined by fitting the traces to a monoexponential function ( solid black lines ) . ( b ) The dependence of kobs on the vesicle concentration: data from two different sets of liposomes used for the stopped-flow experiment shown in ( a ) . The non-zero y-intercept of the linear regression curve yields koff and the slope yields kon . ( c ) Representative absorbance time course ( n = 2 ) for rapid mixing of C2AB fragment ( 0 . 25 µM , final concentration ) and vesicles ( ~2 . 5 nM ) . Absorbance was monitored at 450 nm for vesicles containing PtdChol/PtdSer/PtdIns ( 4 , 5 ) P2 ( 75:20:5 , molar ratio ) . No aggregation of vesicles was observed in our conditions , even when lipid mixtures with the highest affinity to C2AB fragment were used . ( Inset ) A scale-up of the traces showing that there is no vesicle aggregation in our conditions in concordance to the results of Vennekate et al . ( 2012 ) . ( d ) Kd values , calculated as the ratio of koff/kon , ( e ) kon and ( f ) koff of C2AB fragment binding to liposomes containing different concentrations of PtdSer and PtdIns ( 4 , 5 ) P2 . All values were calculated from stopped-flow experiments carried out by rapid mixing of C2AB fragment with vesicles containing different amounts of PtdSer and/or PtdIns ( 4 , 5 ) P2 , and 100 µM free Ca2+ at 25°C ( n = 5–10 ) . Higher affinities ( lower Kd ) were observed in the presence of both PtdSer and PtdIns ( 4 , 5 ) P2 , solely resulting from a decrease in koff . Columns with the same color indicate values that were not significantly different from each other . Black boxes indicate either lack of binding or low binding that was too noisy for a reliable quantitative analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 15886 . 003 In our approach , we used a low protein concentration to avoid binding of the C2AB fragment to opposing membranes in the presence of Ca2+ ( resulting in vesicle clustering ) ( Vennekate et al . , 2012 ) and other cooperative effects ( Figure 1c ) . Our conditions allowed us to measure reliably the binding kinetics and affinity of the C2AB fragment to liposomes containing various molar ratios of PtdSer ( 0 , 10 and 20% ) and PtdIns ( 4 , 5 ) P2 ( 0 , 2 . 5 and 5% ) . As expected , no binding was observable in the absence of PtdSer and PtdIns ( 4 , 5 ) P2 and , unfortunately , the fluorescence traces for binding to vesicles containing 10% PtdSer or 2 . 5% PtdIns ( 4 , 5 ) P2 could not be reliably fitted because of the low signal-to-noise ratio under our experimental conditions . However , vesicles containing both PtdSer and PtdIns ( 4 , 5 ) P2 yielded a lower apparent dissociation constant ( Kd ) than vesicles containing either PtdSer or PtdIns ( 4 , 5 ) P2 alone , even when the vesicles had similar charge density ( Figure 1d ) . Surprisingly , no significant difference was observed in the bimolecular association rate constant ( kon ) in almost all the tested lipid mixtures ( Figure 1e ) , In other words , the decrease in affinity is solely due to a drop in the unimolecular dissociation rate constant ( koff ) ( Figure 1f ) . Thus , under these experimental conditions , the association rate for binding of the C2AB fragment to the vesicles is diffusion-limited and the binding equilibrium is controlled by the off-rate . The results described above suggest that the C2AB fragment binds to PtdSer in a different manner than to PtdIns ( 4 , 5 ) P2 . To determine whether the orientation of the C2AB fragment differs when bound to membranes containing either PtdSer or PtdIns ( 4 , 5 ) P2 in the presence of Ca2+ , we performed EPR experiments using spin-labeled variants of the protein . Insertion of the spin-labeled side chain R1 ( see Figure 2a ) into the hydrophobic core of the membrane alters the rotamer sampling and motion of the spin label . As consequence , the EPR spectra from membrane-embedded spin labels are significantly broadened when the C2AB fragment is fully bound to vesicles ( Figure 2a ) . In addition , using progressive power saturation ( Freed et al . , 2011; Frazier et al . , 2003 ) , we determined the membrane insertion depth of labels attached to the Ca2+-binding loops of the C2A domain ( residues 173 and 234 ) , the Ca2+-binding loops of the C2B domain ( residues 304 and 368 ) and the polybasic lysine patch ( residue 329 ) . When bound to bilayers containing 5% PtdIns ( 4 , 5 ) P2 , labels in the Ca2+-binding loops exhibit more motional averaging than they do when bound to vesicles composed of either 20% PtdSer ( Figure 2a ) or 10% PtdSer + 2 . 5% PtdIns ( 4 , 5 ) P2 ( data not shown ) . This suggests a shallower membrane penetration for R1 when bound to bilayers composed of only PtdIns ( 4 , 5 ) P2 rather than PtdIns ( 4 , 5 ) P2 and PtdSer , in agreement with previous results of experiments based on continuous wave power saturation . As previously reported ( Herrick et al . , 2006; Kuo et al . , 2009 ) , R1 sidechains at sites 173 , 234 , 368 and 304 penetrate the membrane in the presence of PtdSer regardless of whether PtdIns ( 4 , 5 ) P2 is present , but failed to deeply penetrate the membrane in vesicles containing only PtdIns ( 4 , 5 ) P2 . As shown in Table 1 , the insertion depth of the Ca2+-binding loops was 4–6 Ångstroms ( Å ) shallower in membranes composed of PtdIns ( 4 , 5 ) P2 than in membranes containing PtdSer . By contrast , a label in the polybasic patch , 329R1 , was closer to the membrane in PtdIns ( 4 , 5 ) P2-containing vesicles than in PtdSer-containing vesicles , indicating that there is a difference in orientation . Using the power saturation data in Table 1 , the orientations of C2B domain in the presence of membranes containing either PtdSer or PtdIns ( 4 , 5 ) P2 were determined ( see Materials and methods ) and are shown in Figure 2b . The presence of PtdIns ( 4 , 5 ) P2 is found to promote a tilt of the C2B domain , which is consistent with its interaction with the polybasic face . It should be kept in mind that the depth measurements are averaged values and that these models represent an average of what must be a dynamic system . Together with the stopped-flow measurements described above , these data indicate that PtdSer and PtdIns ( 4 , 5 ) P2 play a synergistic role in the binding of syt-1 to the membrane . When both PtdSer and PtdIns ( 4 , 5 ) P2 are present , the Ca2+-binding loops of syt-1 penetrate deeper into the bilayer than they do when the membrane contains PtdIns ( 4 , 5 ) P2 alone , leading to a tighter binding of the C2AB fragment to the membrane , which is reflected by a decrease in the dissociation rate . 10 . 7554/eLife . 15886 . 004Figure 2 . Orientation of the C2 domains bound to membranes containing either PtdSer or PtdIns ( 4 , 5 ) P2 . ( a ) EPR spectra from sites 173 , 234 , 304 , 329 and 368 in solution or bound to bilayers containing PtdChol/PtdSer ( 80:20 , molar ratio ) or PtdChol/PtdInsP2 ( 95:5 , molar ratio ) . The broader spectra reflect diminished amplitudes and rates of R1 label motion when the C2AB fragment is bound to vesicles containing PtdChol/PtdIns ( 4 , 5 ) P2 ( 95:5 , molar ratio ) or PtdChol/PtdSer ( 80:20 , molar ratio ) ( n = 2–3 ) . R1 is the spin-labeled side chain produced by derivatizing cysteine with the MTSL spin label . ( b ) Docking orientation of the C2B domain ( PDB ID: 1K5W [Fernandez et al . , 2001] ) at the membrane interface , polybasic patch in blue . In this figure , the result from Xplor-NIH ( Materials and methods ) was aligned with a membrane simulation generated by CHARMM-GUI ( Jo et al . , 2008 ) for a bilayer with PtdChol:PtdInsP2 or PtdChol:PtdSer at a molar ratio of 97:3 or 80:20 , respectively . PtdIns ( 4 , 5 ) P2 ( pink sticks ) was manually docked to the C2B domain . Blue sticks correspond to residues 321–326 and red sticks correspond to R1-labeled residues used in EPR experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 15886 . 00410 . 7554/eLife . 15886 . 005Table 1 . Depth parameters and approximate positions of spin labels attached to the C2AB fragment . DOI: http://dx . doi . org/10 . 7554/eLife . 15886 . 005MutantLipid compositionDepth parameter ( Φ ) Approx . distance to phosphate plane ( Å ) *M173R120% PtdSer+1 . 20 ± 0 . 10+8 . 85% PtdIns ( 4 , 5 ) P2−0 . 01 ± 0 . 20+5 . 510% PtdSer + 2 . 5% PtdIns ( 4 , 5 ) P2+0 . 66 ± 0 . 10+7 . 4F234R120% PtdSer−0 . 10 ± 0 . 10+5 . 25% PtdIns ( 4 , 5 ) P2−1 . 50 ± 0 . 03–0 . 610% PtdSer + 2 . 5% PtdIns ( 4 , 5 ) P2 +0 . 00 ± 0 . 10+5 . 5V304R120% PtdSer−0 . 39 ± 0 . 20+4 . 35% PtdIns ( 4 , 5 ) P2−1 . 30 ± 0 . 10+0 . 610% PtdSer + 2 . 5% PtdIns ( 4 , 5 ) P2−0 . 23 ± 0 . 10+4 . 8T329R120% PtdSer−2 . 00 ± 0 . 14–5 . 45% PtdIns ( 4 , 5 ) P2−0 . 80 ± 0 . 14+2 . 810% PtdSer + 2 . 5% PtdIns ( 4 , 5 ) P2−2 . 10 ± 0 . 12–7 . 3G368R120% PtdSer+0 . 60 ± 0 . 10+7 . 25% PtdIns ( 4 , 5 ) P2−0 . 64 ± 0 . 05+3 . 610% PtdSer + 2 . 5% PtdIns ( 4 , 5 ) P2+0 . 18 ± 0 . 10+6 . 0*Positive distances lie on the hydrocarbon side of a plane defined by the lipid phosphates; negative distances reside on the aqueous side . Distances were estimated using a calibration curve empirically determined as described previously ( Herrick et al . , 2006 ) . Depth parameters are typically averages of two to four measurements and the error represents either standard deviations or errors propagated from the error in the measurement of ΔP½ . These measurements were carried out at lipid concentrations high enough so that C2AB fragment is effectively completely membrane bound in the presence of 1 mM Ca2+ . To further characterize the binding of the C2AB fragment to PtdIns ( 4 , 5 ) P2 and PtdSer , we performed isothermal titration calorimetry ( ITC ) experiments in the presence of Ca2+ using the head groups of PtdSer and PtdIns ( 4 , 5 ) P2 , O-phospho-L-serine ( O-PSer ) and inositol-1 , 4 , 5-triphosphate ( InsP3 ) , respectively , as ligands for our titration instead of vesicles . Our rationale was that by using the head group of the phospholipid , we could determine the role of each specific phospholipid in the binding of the C2AB fragment to the membrane , avoiding the interferences of other phospholipids and unspecific electrostatic and hydrophobic interactions between the protein and the membrane . In our experiments , only small changes in the binding enthalpy were observed upon injection of O-PSer into a solution containing the C2AB fragment , precluding a quantitative assessment of the thermodynamics of binding . This did not change even when we used a short-chain analogue of PtdSer ( 1 , 2-hexanoyl-sn-glycero-3-phospho-L-serine , C6-PtdSer ) instead of O-PSer . By contrast , binding of InsP3 to the C2AB fragment was highly exothermic , showing a stoichiometry of 1:1 and an affinity of 14 ± 2 µM ( Figure 3a ) . 10 . 7554/eLife . 15886 . 006Figure 3 . PtdIns ( 4 , 5 ) P2 binds preferentially to the polybasic patch of the C2B domain . ( a ) Binding of the C2AB fragment to phospholipid head groups measured by isothermal titration calorimetry ( ITC ) . Titration of ~50 µM C2AB fragment ( 50 mM HEPES , pH 7 . 4 , 150 mM NaCl and 1 mM CaCl2 ) with InsP3 ( inositol-1 , 4 , 5-triphosphate ) ( n = 3 ) , O-phosphoserine ( O-PSer ) ( n = 2 ) and 1 , 2-hexanoyl-sn-glycero-3-phospho-L-serine ( C6-PtdSer ) ( n = 2 ) , in the presence of saturating Ca2+ at 25°C . Only small heats were observed upon addition of C6-PtdSer and O-PSer , while InsP3 binds specifically ( Kd = 14 ± 2 µM ) with a stoichiometry of 1:1 . ( b ) Averaged-weighted chemical shifts ( △HN , N ) in 15N-1H correlated HSQC NMR spectra of syt-1 in the presence of InsP3 and Ca2+ . Chemical shift changes are widely distributed in the polybasic region of C2B domain . Small chemical shift are also seen in the calcium-binding loops of the C2B and C2A domains . Measurements were made under normal ionic strength ( 150 mM NaCl , 50 mM MES , pH = 6 . 3 , and 3 mM Ca2+ ) at a frequency of 600 MHz for protons . ( c ) Chemical shifts are color coded and mapped onto the structures of C2B and C2A domains according to the color bar ( n = 2–3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15886 . 00610 . 7554/eLife . 15886 . 007Figure 3—figure supplement 1 . Chemical shift of 1H-15N HSQC of C2B domain in the presence of InsP3 or O-PSer . ( a ) Chemical shifts of 1H-15N HSQC of C2AB fragment in the absence ( black ) and in the presence of 2 mM InsP3 ( red ) . The expanded section of the HSQC spectrum shows chemical shift changes upon titration with different concentrations of InsP3 ( 0 mM , blue; 0 . 1 mM , green; 0 . 2 mM , maroon; 0 . 3 mM , purple; 0 . 5 mM , red; 1 mM , cyan and 2 mM , black ) . ( b ) Average weighted chemical shifts ( △HN , N ) in the presence of 7 mM O-phosphoserine . Measurements are made under normal ionic strength conditions ( 150 mM NaCl , 50 mM MES , pH = 6 . 3 , and 3 mM Ca2+ ) at a frequency of 600 MHz for protons . DOI: http://dx . doi . org/10 . 7554/eLife . 15886 . 007 To gain insight into the binding sites , we examined the average weighted chemical shift changes in a 1H-15N HSQC NMR spectrum of 15N-labeled C2AB fragment upon addition of either InsP3 or O-PSer in the presence of Ca2+ . In agreement with the lack of enthalpy changes , addition of O-PSer to the C2AB fragment did not induce significant changes in the chemical shifts ( Figure 3—figure supplement 1a ) . By contrast , the addition of InsP3 induced chemical shift changes primarily in the polybasic lysine patch of the C2B domain , with the highest chemical shifts observable for residues K325 , K326 and K327 , confirming previous reports which established the polybasic lysine patch of the C2B domain as the primary binding site of PtdIns ( 4 , 5 ) P2 ( Vrljic et al . , 2011; Li et al . , 2006; Bai et al . , 2004; van den Bogaart et al . , 2012 ) ( Figure 3b , c and Figure 3—figure supplement 1b ) . In addition to the polybasic lysine patch , residues located in the Ca2+-binding sites of the C2A and C2B domains exhibited significant but smaller chemical shift changes ( Figure 3b , c ) . To determine whether the Ca2+-binding sites contribute to the overall binding affinity of InsP3 , we measured InsP3 binding by ITC to C2AB-fragment mutants in which Ca2+-binding was inactivated either in the C2A domain ( C2aB: D178A , D230A , D232A ) or in the C2B domain ( C2Ab: D309A , D363A , D365A ) or both ( C2ab: D178A , D230A , D232A , D309A , D363A , D365A ) ( Radhakrishnan et al . , 2009 ) . The experiments were carried out in the presence and absence of Ca2+ ( Figure 4a , b ) , with the wild-type C2AB fragment as control . In the absence of Ca2+ , the binding affinity of the wild-type protein was significantly lower than in the presence of Ca2+ . This reduction is mediated by the Ca2+-binding site of the C2B domain: it was abolished in the C2Ab and C2ab mutants but not in the C2aB mutant ( Figure 4d ) . Apparently , the negatively charged side chains in the Ca2+-binding site reduce the affinity of the negatively charged InsP3 through unfavorable Coulombic interactions . Neutralizing these charges by Ca2+-binding or , alternatively , by their deletion by mutagenesis are equally effective in preventing the thermodynamic penalty resulting from repulsive interactions . In line with that proposition , the lower affinities were mainly due to a decrease in binding enthalpy rather than to reduced entropy ( Figure 4—figure supplement 1; Figure 4—figure supplement 2 ) , as one would expect for a primarily electrostatic effect . Nevertheless , no change of the binding stoichiometry was observed in any of the different conditions and mutants tested ( Figure 4c ) . Next , we repeated the ITC titrations using the isolated C2A and C2B domains of syt-1 . As expected , the C2B domain bound InsP3 in a manner similar to the C2AB fragment . By contrast , binding of InsP3 to the C2A domain yielded only small enthalpy changes , confirming that C2B is indeed the domain responsible for InsP3 binding ( Figure 4e ) . 10 . 7554/eLife . 15886 . 008Figure 4 . Ca2+ increases the affinity of PtdIns ( 4 , 5 ) P2 to the polybasic patch by means of shielding the negative charges of the Ca2+-binding site in the C2B domain . ( a , b ) Representative ITC titrations ( n = 3–6 ) of ~50 µM of the wild-type C2AB fragment , mutant proteins in which either one or both Ca2+ binding sites are mutated , and the KAKA mutant in which some of the charges in the polybasic patch are removed . Titrations were carried out with InsP3 as shown in Figure 3a in ( a ) the presence or ( b ) the absence of Ca2+ . Lines represent the fitting of the different titrations . ( c ) Number of binding sites determined from the fits of the ITC experiments of the different C2 domains shown in ( a ) and ( b ) . All C2 domains show one binding site for InsP3 in the presence or absence of Ca2+ . In absence of Ca2+ , binding of InsP3 to the KAKA mutant was completely abolished . ( d ) Dissociation constants ( Kd ) calculated from the experiments shown in ( a ) and ( b ) . In the presence of Ca2+ , only the KAKA mutant shows a significant decrease in the binding to InsP3 . In the absence of Ca2+ , mutations in the C2B Ca2+-binding site rescue the binding to InsP3 , where as the KAKA mutant shows weak unspecific binding . ( e ) Titration of ~50 µM C2A domain ( in the presence of Ca2+ ) and C2B domain ( in the presence or absence of Ca2+ ) with InsP3 ( n = 2 ) compared to that of the C2AB fragment in same conditions ( see legend ) . The C2B domain is responsible for InsP3 binding to syt-1 . ( f ) Titration of ~50 µM C2B domain–SNARE-complex ( 1:1 , molar ratio ) with InsP3 in the absence of Ca2+ ( n = 2 ) . The C2B domain binds preferentially to InsP3 in physiological conditions . ( g ) Titration of ~15 µM C2AB domain/SNARE-complex ( 1:1 , molar ratio ) with InsP3 in the absence of Ca2+ ( n = 2 ) . We set the number sites to one for the fitting due to the uncertainty resulting from the low concentrations used . ( h ) Titration of SNARE-complex ~15 µM with different syt-1 fragments in the absence of Ca2+ ( n = 2 ) . The Kd values from ( g ) and ( h ) should be interpreted with caution because the low concentration used for the titrations resulted in a high uncertainty on the fitting . ( i ) Representative raw data for titration from ( g ) and ( h ) . Interestingly , titration of the C2AB–SNARE complex presented an endothermic profile , whereas titration of InsP3 presented an exothermic profile . DOI: http://dx . doi . org/10 . 7554/eLife . 15886 . 00810 . 7554/eLife . 15886 . 009Figure 4—figure supplement 1 . Thermodinamic binding parameters of the different C2AB fragments . ( a ) Binding enthalpy ( △H ) and ( b ) entropy ( -T*△S ) for ITC titration in Figure 4a , b . DOI: http://dx . doi . org/10 . 7554/eLife . 15886 . 00910 . 7554/eLife . 15886 . 010Figure 4—figure supplement 2 . Representative titrations ( n ≥ 3 ) of ~50 µM of the C2AB fragment used in Figure 4a , b in the presence or absence of Ca2+ . DOI: http://dx . doi . org/10 . 7554/eLife . 15886 . 010 Recently , it has been reported that , in the absence of Ca2+ , syt-1 binds preferentially to the SNARE complex rather than to membranes containing PtdIns ( 4 , 5 ) P2 ( Brewer et al . , 2015 ) . To test this hypothesis , we measured InsP3 binding to the isolated C2B domain in the presence of SNARE complex and in the absence of Ca2+ ( to avoid precipitation of protein [Dai et al . , 2007] ) by ITC . The presence or absence of SNARE-complex made no difference to InsP3 binding to the isolated C2B domain ( Figure 4f ) , in agreement with previous observations from our group ( Park et al . , 2015 ) . This evidence suggests that , under physiological conditions , the C2AB fragment binds preferentially to PtdIns ( 4 , 5 ) P2 . A recent study has reported three distinct syt–SNARE interfaces involving both C2A and C2B domains ( Zhou et al . , 2015 ) . Therefore , we next repeated the experiment described above using the C2AB fragment and lower protein concentration , ~15 µM instead of ~50 µM , to reduce the chances of protein aggregation . Interestingly , the presence of the SNARE complex resulted in a moderate decrease in affinity of InsP3 to the C2AB fragment , in contrast to above results using the C2B domain ( Figure 4g , i ) . To clarify this discrepancy , we decided to examine the syt–SNARE interaction using ITC . Our results show that the SNARE complex binds the C2AB fragment with higher affinity than it does the C2B domain , and that K326 and K327 play an important role in this binding ( Figure 4h , i ) , in agreement with recent findings ( Brewer et al . , 2015; Wang et al . , 2016 ) . Previous studies have shown that the mutation of these two Lys residues in the polybasic lysine patch to Ala ( K326A and K327A , KAKA mutant ) abolishes syt-1 binding to PtdIns ( 4 , 5 ) P2 ( van den Bogaart et al . , 2012 ) . Using ITC , we studied the binding of InsP3 to the KAKA mutant in the presence and absence of Ca2+ . In the absence of Ca2+ , binding to InsP3 was abolished ( Figure 4b ) , as previously reported ( van den Bogaart et al . , 2012 ) . However , in the presence of Ca2+ , the KAKA mutant was able to bind to InsP3 , but with lower affinity than the wild-type C2AB fragment ( Figure 4a , d ) . To further investigate the binding of the KAKA mutant to the membrane in the absence of Ca2+ , we performed stopped-flow experiments in the absence of Ca2+ . No binding was detectable using FRET , probably because of the low concentration of protein used in our pseudo-first-order conditions , which did not provide enough signal for detection ( data not shown ) . To overcome this limitation , we carried out a vesicle sedimentation assay to measure the fraction of protein bound to sucrose-loaded vesicles under true equilibrium conditions . Using sedimentation , the KAKA mutant was found to bind with reduced affinity to either PtdSer- or PtdIns ( 4 , 5 ) P2-containing bilayers in the presence of Ca2+ ( Figure 5a , b , c and Table 2 ) . Remarkably , the KAKA mutation produced only a small effect on the binding to PtdSer-containing bilayers ( Figure 5a ) , but dramatically reduced the membrane affinity of C2AB fragment to 2% PtdIns ( 4 , 5 ) P2 bilayers ( Figure 5b ) . In these membranes , the KAKA mutant altered the membrane-binding free energy of C2AB fragment by 7 . 5 kJ/mole . In the absence of Ca2+ , the KAKA mutant failed to bind to PtdSer- or PtdIns ( 4 , 5 ) P2-containing bilayers ( Figure 5d , e ) . However , this variant did bind to membranes composed of both PtdSer and PtdIns ( 4 , 5 ) P2 ( Figure 5f ) , indicating a synergistic activity of PtdSer and PtdIns ( 4 , 5 ) P2 towards syt-1 C2AB fragment binding . 10 . 7554/eLife . 15886 . 011Figure 5 . Equilibrium binding of the C2AB fragment and polybasic mutants to PtdChol , PtdSer and PtdIns ( 4 , 5 ) P2 bilayers . ( a , b , c ) Ca2+-dependent and ( d , e , f ) Ca2+-independent partitioning of C2AB fragment into ( a , d ) PtdChol/PtdSer , ( b , e ) PtdChol/ PtdIns ( 4 , 5 ) P2 and ( c , f ) PtdChol/PtdSer/ PtdIns ( 4 , 5 ) P2 bilayers . ( d , e ) No binding of the KAKA mutant was observed in the presence of PtdSer or PtdIns ( 4 , 5 ) P2 alone in absence of calcium . At equivalent charge densities , removal of lysine residues within the polybasic face reduced the membrane affinity in vesicles containing PtdSer and PtdIns ( 4 , 5 ) P2 ( c ) more than it did in vesicles containing either ( a ) PtdSer or ( b ) PtdIns ( 4 , 5 ) P2 in the presence of calcium . The lines represent fits to the data using Eq [1] ( Materials and methods ) . Reciprocal molar partition coefficients obtained from data are listed in Tables 2 and 3 ( n = 2–3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15886 . 01110 . 7554/eLife . 15886 . 012Table 2 . Reciprocal molar partition coefficients , K , for the Ca2+-dependent membrane affinity of syt-1 C2 domains . DOI: http://dx . doi . org/10 . 7554/eLife . 15886 . 012Lipid compositionSyt1-C2AB WTSyt1-C2AB K326ASyt1-C2AB K326A/K327A7 mol % PtdSer1 . 0 ( ±0 . 1 ) x 1033 . 4 ( ±0 . 4 ) x 1022 mol % PtdIns ( 4 , 5 ) P22 . 2 ( ±0 . 2 ) x 1045 . 1 ( ±0 . 5 ) x 1031 . 1 ( ±0 . 1 ) x 1035 mol % PtdSer + 0 . 5 mol % PtdIns ( 4 , 5 ) P25 . 0 ( ±0 . 7 ) x 1042 . 2 ( ±0 . 3 ) x 1041 . 0 ( ±0 . 1 ) x 104Values of K are expressed in units of M-1 . The errors represent the uncertainly obtained from the nonlinear regression used in fitting the data to K . The reciprocal of K corresponds to the accessible molar lipid concentration at which 50% of the protein is membrane bound ( n = 2–3 ) . The affinity between the C2AB fragment and membranes containing both PtdSer and PtdIns ( 4 , 5 ) P2 ( Table 3 ) in the absence of Ca2+ indicates that at an accessible lipid concentration of 20 µM , 50% of the protein is bound . However , in full-length syt-1 , the C2 domains are tethered near the membrane interface and thus experience a very high local lipid concentration . Assuming the domains lie within 4 nm of the interface , the effective membrane concentration seen by the C2 domains will be well over four orders of magnitude higher than 20 µM . This represents a shift in the binding free energy to favor membrane association by about 25 kJ/mole , may be even higher when considering that syntaxin and PtdIns ( 4 , 5 ) P2 form clusters that increase the local concentration of PtdIns ( 4 , 5 ) P2 in the plasma membrane ( Honigmann et al . , 2013; van den Bogaart et al . , 2011 ) . Of course , the local concentration of SNARE proteins is also high , so that significant competition with PtdIns ( 4 , 5 ) P2 for syt-1 binding cannot be excluded . We conclude that the affinity of C2AB fragment to membranes containing PtdIns ( 4 , 5 ) P2 and PtdSer is significant even in the absence of Ca2+ ( Table 3 ) . 10 . 7554/eLife . 15886 . 013Table 3 . Reciprocal molar partition coefficients , K , for the Ca2+-independent membrane affinity of syt1 C2 domains . DOI: http://dx . doi . org/10 . 7554/eLife . 15886 . 013Lipid compositionSyt1–C2AB WTSyt1–C2AB K326A/K327A25 mol % PtdSer2 . 7 ( ±0 . 3 ) x 102not detected2 mol % PtdIns ( 4 , 5 ) P21 . 1 ( ±0 . 2 ) x 103not detected10 mol % PtdSer + 2 mol % PtdIns ( 4 , 5 ) P24 . 7 ( ±0 . 9 ) x 1042 . 7 ( ±0 . 3 ) x 102Values of K are expressed in units of M-1 . The errors represent the uncertainly obtained from the nonlinear regression procedure used in fitting the data to K . The reciprocal of K corresponds to the accessible molar lipid concentration at which 50% of the protein is membrane bound ( n = 2–3 ) . Taken together , our results demonstrate that PtdIns ( 4 , 5 ) P2 primarily binds to the polybasic lysine patch of the C2B domain , which includes two Lys residues ( K326 and K327 ) , rather than to the Ca2+-binding sites of C2B domain . As discussed below , the relative affinity of the C2AB fragment for PtdIns ( 4 , 5 ) P2 versus its affinity for the SNAREs suggests that PI ( 4 , 5 ) P2 is the likely target of the polybasic patch of the C2B domain , in agreement with recent reports ( Wang , 2016; Park et al . , 2015 ) . Furthermore , Ca2+ increases C2B's affinity for PtdIns ( 4 , 5 ) P2 , probably by neutralizing the negatively charged side chain at the Ca2+-binding site of the C2B domain , allowing a tighter binding of the polybasic patch to PtdIns ( 4 , 5 ) P2 . Although we focus on PtdIns ( 4 , 5 ) P2 because it is the most abundant phosphoinositide in the plasma membrane ( Ueda , 2014 ) , it has been reported that syt-1 also binds to other phosphorylated phosphatidylinositols ( PtdInsPx ) ( Vrljic et al . , 2011; Wang et al . , 2011 ) . Phosphatidylinositol phosphates are lipids that only differ from each other in the number of phosphate groups and/or the phospho site of the inositol ring , with different species being specifically associated with different intracellular membranes ( Chasserot-Golaz et al . , 2010 ) . To gain insight into the binding of syt-1 to different phosphatidylinositols , we measured the binding kinetics of the C2AB fragment to vesicles containing 55% phosphatidylcholine ( PtdChol ) , 22% phosphatidylethanolamine ( PtdEth ) , 11% phosphatidylserine ( PtdSer ) , 11% cholesterol ( Chol ) and 1% of different phosphatidylinositols ( PtdInsPx ) . Phosphorylated phosphatidylinositides increased the apparent affinities ( Figure 6a ) , with the bimolecular association ( kon ) and the unimolecular dissociation rate ( koff ) constants being enhanced and reduced , respectively ( Figure 6b , c ) . The 5–10-fold decrease in koff and Kd ( koff/kon ) was progressively higher in the presence of bisphosphorylated phosphatidylinositols ( PtdInsP2 ) and triphosphorylated phosphatidylinositols ( PtdInsP3 ) than in the presence of monophosphorylated phosphatidylinositols ( PtdInsP ) ( Figure 6a , c ) . While PtdInsP2 or PtdInsP3 increased the association rate constant moderately by 2–3-fold , PtdInsP or phosphatidylinositols ( PtdIns ) had no measurable effect on the binding kinetics ( Figure 6b ) . Thus , these results suggest that syt-1 does not discriminate between the different tested phosphorylated phosphatidylinositol stereoisomers , and that the increase in affinity is mainly due to the higher charge density of the different PtdInsPx species . Our findings are in contrast to those of a recent study that came to a different conclusion based on a lipid overlay and a binding sensor assay ( Vrljic et al . , 2011 ) . To further examine these conflicting results , we measured the binding of the head groups of different phosphatidylinositol phosphates to the C2B domain of syt-1 using ITC ( Figure 6d ) . Surprisingly , Ins ( 1 , 3 , 5 ) P3 showed lower affinity than the other tested stereoisomers ( Figure 6e ) because of a less favorable binding enthalpy ( Figure 6—figure supplement 1 ) . 10 . 7554/eLife . 15886 . 014Figure 6 . Ca2+ and phosphoinositides increase the membrane affinity of syt-1 by decreasing the dissociation rate of the syt-1–membrane complex . ( a ) Kd , ( b ) kon and ( c ) koff calculated from stopped-flow experiments carried out by rapid mixing , as in Figure 1 , of C2AB fragment with vesicles containing PtdChol/PtdSer/PtdEth/Chol/PtdInsPX ( 55:11:22:11:1 molar ratio ) , where X = 0–3 phosphate groups , in 20 mM HEPES ( pH 7 . 4 ) , 150 mM KCl , 1 mM EGTA and 1 . 1 mM CaCl2 ( 100 µM free Ca2+ ) at 37°C ( n = 5–10 ) . An increased number of phospho groups of phosphoinositides increases the affinity by decreasing the dissociation rate ( koff ) . A minor increase in kon was detected in the case of PtdInsP2–3 . ( d ) Thermodynamic characterization of the binding of C2B domain to head groups of physiological phosphoinositides by ITC . Titration of ~50 µM C2B domain ( 50 mM HEPES ( pH 7 . 4 ) , 150 mM NaCl and 1 mM CaCl2 ) with main stereoisomers of InsP3–4 at 25°C ( n = 2 ) . ( e ) Dissociation constants ( Kd ) calculated for experiments from d . ( f ) Kd , ( g ) kon and ( h ) koff calculated from stopped-flow experiments carried out at different Ca2+concentrations with vesicles containing PtdChol/PtdSer/PtdEth/Chol/PtdIns ( 4 , 5 ) P2 ( 55:11:22:11:1 molar ratio ) at 37°C ( n = 5–10 ) . Ca2+ decreases the rate of dissociation drastically ( h ) and thereby increases the affinity ( f ) . No significant difference was observed for the rate of association ( kon ) ( g ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15886 . 01410 . 7554/eLife . 15886 . 015Figure 6—figure supplement 1 . ITC parameters of the C2B domain binding to different isomers of InsP3 . ( a ) Number of binding sites , ( b ) binding enthalpy ( △H ) and ( c ) entropy ( -T*△S ) for ITC titration in Figure 6d . Titration of ~50 µM C2B domain ( in the presence of Ca2+ ) with different isomers of InsP3 . DOI: http://dx . doi . org/10 . 7554/eLife . 15886 . 015 Finally , we studied the binding kinetics of the C2AB fragment to vesicles containing PtdChol:PtdSer:PtdEth:Chol:PtdIns ( 4 , 5 ) P2 ( 55:22:11:11:1 molar ratio ) at four different Ca2+concentrations ( 50 , 100 , 500 and 1000 µM ) . We were unable to detect binding at Ca2+ concentrations lower than 50 µM . Nevertheless , for all other tested concentrations , the estimated bimolecular association rate constants ( kon ) showed minor differences ( Figure 6g ) , meaning that the increase in the Ca2+concentration did not drive the C2AB fragment to the membrane faster , at least not in this concentration regime . In contrast , the dissociation rate constant ( koff ) showed a substantial decrease at concentrations over 50 µM ( Figure 6h ) with no further decrease at Ca2+ > 100 µM . Because of this , the dissociation constant ( Kd ) , followed a similar trend to the unimolecular dissociation rate constant ( koff ) ( Figure 6f , h ) . In summary , these results suggest that the association rate of the C2AB fragment is largely Ca2+-independent . The increased membrane affinity of the C2AB fragment in the presence of Ca2+ is due to a decreased dissociation rate from the membrane . As a result , Ca2+ lengthens the period during which syt-1 remains bound once membrane binding has occurred . In the present study , we used well-defined kinetic ( stopped-flow ) and structural ( EPR and NMR ) methods to better understand how the C2AB fragment of the vesicular Ca2+ sensor synaptotagmin 1 ( syt-1 ) binds to membranes . Three main conclusions can be derived from our data . First , PtdSer and PtdIns ( 4 , 5 ) P2 act synergistically to promote deeper penetration of the Ca2+-binding site of the C2B domain into the membrane . Second , Ca2+ increases the affinity of syt-1 for PtdIns ( 4 , 5 ) P2 , which binds preferentially to the polybasic lysine patch of the C2B domain , and Ca2+ acts by neutralizing the negative charge of the Ca2+-binding site . Third , Ca2+ and phosphoinositides increase the affinity of syt-1 to membrane exclusively by reducing its dissociation rate . It is well established that syt-1 binds to anionic phospholipids ( mainly phosphatidylserine and phosphoinositides ) via distinct regions , and that these interactions facilitate penetration of the syt-1 C2 domains into the membrane . However , neither the contribution of the individual lipids nor the conformational arrangement of the bound C2B domain was understood in molecular detail ( Kuo et al . , 2009; van den Bogaart et al . , 2012; Kuo et al . , 2011; Radhakrishnan et al . , 2009; Li et al . , 2006; Bai et al . , 2002; Schiavo et al . , 1996; Araç et al . , 2006; Davletov and Sudhof , 1993; Bai and Chapman , 2004; Bai et al . , 2004 ) . Our results now show that PtdSer and PtdIns ( 4 , 5 ) P2 act cooperatively by binding preferentially to different binding regions that influence each other . In the absence of Ca2+ , binding is mediated primarily by the interaction between the polybasic lysine patch and PtdIns ( 4 , 5 ) P2 , in which K326 and K327 play an important role as reported previously ( van den Bogaart et al . , 2012; Bai et al . , 2004; Araç et al . , 2006; Li et al . , 2006 ) . In the absence of Ca2+ , the interaction of syt-1 with the bilayer is limited to an interfacial absorption of the C2B domain via its polybasic region , where membrane insertion of C2B Ca2+-binding loops is prevented by charge repulsion between the negatively charged Ca2+-binding site and the negatively charged membrane interface ( Bai et al . , 2004; Kuo et al . , 2009 ) . Intriguingly , the negatively charged binding site of the C2B domain also reduces the affinity of this domain for PtdIns ( 4 , 5 ) P2 by electrostatic repulsion ( see model in Figure 7 , left ) . In the presence of Ca2+ , several effects operate synergistically to increase membrane affinity . First , Ca2+ binding polarizes the Ca2+-binding loops of the C2B domain and PtdSer completes the coordination sphere of the bound Ca2+ , allowing syt-1 to penetrate into the membrane . Second , the neutralization of charge at the Ca2+ binding site of the C2B domain reduces electrostatic repulsion at the polybasic region and promotes the association of PtdIns ( 4 , 5 ) P2 ( Figure 7 , right ) . This cross-talk between the two sites explains the conflicting conclusions regarding the role of the Ca2+-binding sites and the polybasic lysine patch in binding to anionic membrane lipids . The Ca2+-binding sites preferentially bind to PtdSer , for which they form a defined binding pocket ( Honigmann et al . , 2013 ) but not to phosphoinositides , in contrast to previous suggestions ( Vrljic et al . , 2011; Zhang et al . , 1998; van den Bogaart et al . , 2012 ) . Calcium increases the affinity of syt-1 towards PtdIns ( 4 , 5 ) P2 simply by neutralizing negative charge at the Ca2+-binding site . It should be noted that by using InsP3 , the soluble head group of PtdIns ( 4 , 5 ) P2 , there is no membrane interface in this experiment and no charge repulsion between the C2AB fragment and liposomes containing PtdIns ( 4 , 5 ) P2 . This fact could explain the non-specific interactions with the Ca2+-binding sites ( Figure 3b , c ) and the weak effect of the presence of the SNARE complex on InsP3–C2AB fragment binding ( Figure 4g , i ) . 10 . 7554/eLife . 15886 . 016Figure 7 . Model of the membrane-binding mechanism of syt-1 . In the absence of Ca2+ , syt-1 is attached to the presynaptic membrane interface . Syt-1 binds to PtdIns ( 4 , 5 ) P2 through its C2B polybasic patch ( Bai et al . , 2004; Kuo et al . , 2009 ) in transient encounters , but the negative charges of the Ca2+-binding pockets prevent penetration of the C2 domains into the presynaptic membrane because of the electrostatic repulsion between them , leading to a high rate of dissociation . Upon Ca2+ influx , Ca2+ binding neutralizes the negative charge of the Ca2+-binding sites . As a consequence , phosphatidylserine completes the sphere of coordination of Ca2+ and allows the insertion of the hydrophobic residues at the tips of the C2 domains . Simultaneously , the polybasic patch enhances its affinity to phosphoinositides , leading to deeper penetration of the C2B Ca2+-binding site into the bilayer . Together , these events decrease the rate of dissociation of syt-1 from the membrane and enhance its penetration into the core of the presynaptic membrane , which eventually leads to SNARE-mediated membrane fusion . Nuclear magnetic resonance structures of the C2A domain ( PDB 1BYN [Shao et al . , 1998] ) and C2B domain ( 1K5W [Fernandez et al . , 2001] ) of syt-1 and a molecular dynamic membrane simulation were rendered using PyMOL Molecular Graphics System ( Schrödinger , LLC , http://www . pymol . org ) . The membrane used in this illustration was generated using the Membrane Builder input generator module in CHARMM-GUI ( Jo et al . , 2008 ) for a bilayer with PtdChol:PhdSer:PtdEth:Chol:PtdInsP2 at a molar ratio of 55:11:22:11:1 . DOI: http://dx . doi . org/10 . 7554/eLife . 15886 . 016 The question then arises of how the binding between the C2AB domain and the SNARE complex fits into the picture . In particular , we need to clarify whether there is synergism or competition between syt binding to SNAREs and syt binding to acidic membrane lipids during the triggering of exocytosis . Clearly , the polybasic patch of the C2B domain is involved in SNARE binding with high affinity as shown earlier ( Brewer et al . , 2015; Zhou et al . , 2013 ) and as confirmed by our ITC data . Moreover , mutations that impair syt–SNARE binding also alter syt-membrane binding; in addition , they alter the triggering of exocytosis when the corresponding mutants are introduced into neurons ( Brewer et al . , 2015; Wang et al . , 2016 ) . However , the available evidence suggests that the polybasic region preferentially interacts with PtdIns ( 4 , 5 ) P2 rather than with SNAREs under normal ionic conditions ( Park et al . , 2015; Wang et al . , 2016 ) . For example , pulse EPR and fluorescence cross-correlation spectroscopy indicate that the Syt–PtdIns ( 4 , 5 ) P2 interaction persists under conditions that eliminate the syt–SNARE interaction ( Park et al . , 2015 ) , and the data presented here in Figure 4g , i show that InsP3 binding to the C2AB fragment is only slightly affected by the presence of the SNAREs . Nevertheless , the syt–SNARE interaction is conformationally heterogeneous and quite dynamic , and sites other than the polybasic patch may interact with the SNAREs in the presence of PtdIns ( 4 , 5 ) P2 . This may result in a more complex interplay between PtdIns ( 4 , 5 ) P2 , SNARE proteins and syt-1 under physiological conditions ( Wang et al . , 2016; Zhou et al . , 2015; Park et al . , 2015; Brewer et al . , 2015 ) . Despite the clear evidence for high-affinity syt–SNARE binding in the test tube , we favor the view that syt-phospholipid binding is the main component in the triggering mechanism of syt-1 , with the SNARE binding either being physiologically irrelevant or being synergistic at best . For instance , all models trying to picture synaptotagmin complexes with partially zippered SNARE complexes face serious spatial challenges when trying to fit them into a 3D-model between the vesicle and plasma membrane , particularly when considering that both Munc18 and Munc13 ( Ma et al . , 2013 ) ( or even the larger CAPS proteins [Daily et al . , 2010] ) are thought to be bound to such activated SNARE complexes . Moreover , synaptotagmins or other C2 domain proteins are not required for the basic and evolutionarily conserved SNARE engine . They do not participate in other intracellular fusion reactions and are lacking altogether in more primitive eukaryotic cells such as yeast ( Craxton , 2010 ) . Moreover , the canonical function of Ca2+-binding C2 domains appears to be binding to acidic membrane lipids ( Corbalan-Garcia and Gomez-Fernandez , 2014 ) , see also http://scop . mrc-lmb . cam . ac . uk/scop/ . This is true for soluble proteins whose Ca2+-dependent translocation to plasma membranes is mediated by C2 domains ( classical examples include protein kinase C and phospholipase A ) , and also for transmembrane proteins owning multiple tandem C2 domains , such as ferlins and extended-synaptotagmins ( see e . g Idevall-Hagren et al . ( 2015 ) ; Fernandez-Busnadiego et al . ( 2015 ) ; Pangršič et al . ( 2012 ) ) . In summary , a picture is emerging in which syt-1 continuously probes the membrane with high frequency both in the Ca2+-bound and -unbound state . Binding of Ca2+ increases the dwell time of syt-1 on the membrane by allowing more intimate contact with the bilayer rather than by promoting an enhanced synaptotagmin–membrane association . The model that we propose in Figure 7 is in accordance with data from functional experiments carried out in intact synapses . For instance , mutations in the C2B domain polybasic lysine patch reduce the amplitude of the fast component of EPSC in excitatory neurons ( Li et al . , 2006 ) . Moreover , InsP6 ( which binds to the syt-1 polybasic lysine patch [Joung et al . , 2012] ) suppresses autaptic EPSCs via the C2B domain of syt-1 ( Yang et al . , 2012 ) and weakens the Ca2+-dependent binding of syt-1 to membranes ( Lu et al . , 2002 ) . Thus , the docking/priming function of the polybasic lysine patch facilitates release but is not essential since the Ca2+-binding sites remain intact and still allow for binding to PtdSer and membrane insertion . Indeed , increasing PtdSer concentrations increase the frequency of fusion events in PC12 cells ( Zhang et al . , 2009 ) . Furthermore , mutated proteins in which the hydrophobicity of the Ca2+-binding loops of syt-1 C2 domains were increased enhanced both Ca2+-dependent membrane binding and neurotransmitter release in autaptic hippocampal cultures ( Rhee et al . , 2005 ) . Additionally , a recent report determining the ultrastructure of the mouse hippocampal organotypic culture revealed a reduction of synaptic vesicles within 0–5 nm of the active zone in syt-1 KO synapses . This report supports our model , in which syt-1 continuously probes the plasma membrane in the latest steps of the docking/priming process ( Imig et al . , 2014 ) . Despite all of this detailed knowledge , we still do not know how the synaptotagmin–membrane interaction triggers neurotransmitter release . While , as discussed above , the role of SNARE binding is controversial ( Bacaj et al . , 2015; Zhou et al . , 2015; Brewer et al . , 2015; Park et al . , 2015 ) , it is largely accepted that Ca2+-enhanced binding to acidic membrane lipids is crucial for triggering exocytosis . Our data suggest that two features of the syt-1–membrane interaction play a crucial role: ( i ) the membrane insertion of the Ca2+-binding loops rather than an exclusively electrostatic membrane adsorption , and ( ii ) an increased dwell time of the protein on the membrane surface resulting from an elevated kinetic stability of the syt-1–membrane complex . In particular , the latter finding was unexpected . It is conceivable that the increased dwell time brings the membranes closer together and thus triggers the firing of SNARE complexes . It is also conceivable that the deeper insertion into the membrane destabilizes the bilayer at the prefusion contact site and thus helps to overcome the energy barrier towards fusion . Hopefully , our findings will trigger novel approaches towards elucidating the molecular mechanism of the still enigmatic triggering event in neuronal exocytosis . For stopped-flow measurements , lipid were purchased from Avanti Polar Lipids , except for dansyl-labeled phosphatidylethanolamine ( dansyl-DHPE ) , which was purchased from Invitrogen . Lipids were mixed in appropriate amounts and dried under a stream of nitrogen , and traces of organic solvent were removed under vacuum for at least 3 hr . Lipid mixtures were resuspended in 20 mM HEPES , 150 mM KCl , 1 . 1 mM CaCl2 and 1 mM EGTA ( for 100 µM free Ca2+concentration calculated using http://maxchelator . stanford . edu ) at pH 7 . 4 . Large unilamellar vesicles of around ~100 nm were prepared by extrusion as described earlier ( Arbuzova et al . , 1997 ) . Total phospholipid concentration was determined using the total phosphate determination method ( Böttcher et al . , 1961 ) . For vesicle sedimentation assay and EPR measurements , 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphatidycholine ( POPC ) , 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphatidyserine ( POPS ) , and L-α-phosphatidylinositol-4 , 5-bisphosphate ( ammonium salt ) ( PtdIns ( 4 , 5 ) P2 ) were purchased from Avanti Polar Lipids ( Alabaster , AL ) , to make large unilamellar vesicles ( LUVs ) of 100 mM concentration at the following molar ratios: PtdChol/PtdSer ( 3:1 and 4:1 , molar ratio ) , PtdChol/PtdIns ( 4 , 5 ) P2 ( 98:2 and 95:5 , molar ratio ) , PtdChol/PtdSer/PtdIns ( 4 , 5 ) P2 ( 87 . 5:10:2 . 5 , molar ratio ) . The lipids were prepared as described previously ( Kuo et al . , 2011 ) . The dried lipid film was re-suspended in either sucrose buffer ( 176 mM sucrose , 1 mM MOPS , pH 7 . 0 ) or Ca2+ buffer ( 1 mM CaCl2 , 1 mM MOPS , 100 mM KCl , pH 7 . 0 ) , and vesicles were formed using freeze-thaw cycles and extrusion as previously described ( Kuo et al . , 2011 ) . To produce sucrose-loaded LUVs , another wash step with Ca2+ buffer was used to remove the external sucrose solution followed by ultracentrifugation at 160 , 000 × g for 1 hr . All synaptotagmin-1 proteins were derived from Rattus norvegicus and expressed as pET28a constructs , namely the isolated C2A domain ( aa 97–273 ) , the C2B domain ( aa 262–421 ) , the C2AB fragment of synaptotagmin-1 ( aa 97–421 ) and Ca2+-binding mutants of the soluble domain: C2a*B ( D178A , D230A , and D232A ) , C2Ab* ( D309A , D363A , and D365A ) , and C2a*b* ( D178A , D230A , D232A , D309A , D363A , and D365A ) , have been described before ( Stein et al . , 2007 ) . The KAKA mutant ( K326A , K327A ) has also been described earlier ( Radhakrishnan et al . , 2009 ) . SNARE proteins were composed of syntaxin 1A ( residues 188–259 ) , SNAP-25 ( residues 1–206 ) and synaptobrevin ( residues 1–96 ) . Proteins were expressed in Escherichia coli strain BL21 ( DE3 ) and purified using Ni2+-nitrilotriacetic acid beads ( Qiagen GmbH ) followed by ion exchange and gel filtration chromatography on the ÄKTA system ( GE Healthcare ) as described earlier ( Radhakrishnan et al . , 2009; Fasshauer et al . , 2002 ) . SNARE-complex assembly was performed as previously described ( Fasshauer et al . , 2002 ) by mixing the SNARE proteins and incubating them overnight at 4°C . SNARE-complex was purified by ion exchange and gel filtration chromatography , followed by concentration with a 30 kDa cut-off concentrator ( Sartorius Stedim Biotech GmbH , Göttingen , Germany ) . For EPR , NMR and vesicle sedimentation assay , DNA for rat syt-1 ( P21707 ) was obtained from Dr . Carl Creutz ( Pharmacology Department , University of Virginia ) . In the pGEX-KG , vector encoding amino acid residues 96–421 and 249–421 ( syt-1 C2B ) ( Damer and Creutz , 1994; Kuo et al . , 2009 ) were used to produce constructs 136–260 ( syt-1 C2A ) and 136–421 ( syt1 C2AB ) by ligation into the plasmid vector pGEX-KG following the coding region for GST as described previously ( Bhowmik et al . , 2008 ) . The single native cysteine residue at position 277 was mutated to alanine and single and double alanine mutants ( K326A and K326A/K327A ) were produced in the C2AB construct ( 136–421 ) by QuickChange site-directed mutagenesis ( Agilent , Santa Clara , CA ) . Single cysteine substitutions were also produced to create C2AB mutants of M173C , V304C , L323C , K327C and T329C . All mutagenesis was confirmed by DNA sequencing . The expression and purification of syt-1 C2A , C2B , and C2AB constructs were carried out as previously published ( Kuo et al . , 2009; Herrick et al . , 2009 ) . The wild-type and mutant plasmids were expressed in BL21 ( DE3 ) pLysS cells ( Invitrogen , Grand Island , NY ) , and grown in LB media . The protein was purified using a GST affinity column followed by a second purification step using ion exchange chromatography . An SP column was used for the C2AB fragment and C2B domain , and a Q column was used for the C2A domain to remove any remaining contaminants , such as nucleic acids . Syt-1 C2AB prepared in this manner is correctly folded as indicated by CD ( Herrick et al . , 2006 ) and NMR . This C2AB fragment is also found to bind membranes in a Ca2+-dependent manner . SDS Page indicated that the proteins were pure with appropriate molecular weights of 17 . 4 for C2A domain , 20 . 8 for C2B domain , and 34 . 5 for C2AB fragment . The UV absorbance at 278 nm was used to insure that protein fractions were free of nucleic acid contaminant . The protein concentrations were determined using a Bradford Protein Assay ( Thermo Fisher Scientific , Rockford , IL ) . ITC experiments were carried out as previously described ( Joung et al . , 2012 ) . Titration of ~50 µM protein ( see figure legend ) with different ligands were performed at 25°C in 50 mM HEPES ( pH 7 . 4 ) and 150 mM NaCl , in the presence or absence of 1 mM CaCl2 . Buffer without calcium was prepared as reported ( Radhakrishnan et al . , 2009 ) and checked with rhod-5N ( Invitrogen ) . To obtain the effective heat of binding , results of the titration were corrected using buffer-protein and ligand-buffer controls . Finally , the NITPIC ( Keller et al . , 2012 ) and the Origin9 software ( Origin Labs Inc . ) were used to analyze these data . 1 , 2-hexanoyl-sn-glycero-3-phospho-L-serine ( C6-PtdSer ) was titrated under critical micellar concentration ( CMC ) to avoid the formation of C6-PtdSer micelles ( http://www . avantilipids . com ) that interfere in the titration . Kinetic experiments were carried out on an Applied Photophysics SX . 20 stopped-flow spectrophotometer ( Applied Photophysics , Surrey , UK ) in 20 mM HEPES ( pH 7 . 4 ) , 150 mM KCl , and 1 mM EGTA and different free Ca2+concentrations ( calculated using http://maxchelator . stanford . edu ) , as described previously ( Hui et al . , 2005 ) . FRET was used to monitor the time course of the C2AB fragment vesicle binding . The excitation wavelength was set at 280 nm and a 470 nm cut-off filter was used to collect the dansyl emission for the different vesicle concentrations tested . The resulting time courses were fit to a single-exponential function:F ( t ) =F0+Aobs∗e ( −kobs∗t ) where F ( t ) equals the observed fluorescence at time t , F0 is a fluorescence offset representing the final fluorescence , Aobs equals the amplitude , and kobs is the observed rate constant . Observed rate constants were plotted as a function of lipid vesicle concentration ( v ) , calculated assuming 90 , 000 phospholipid molecules per vesicle ( Arbuzova et al . , 1997 ) , and fitted with the equation:kobs=kon[v]+koff where kon represents the apparent association constant , and koff the apparent dissociation rate constant . The ratio of koff to kon provides the calculated apparent vesicle dissociation constant ( Kd ) . An ultracentrifugation technique ( Buser and Mclaughlin , 1998 ) was used to measure the equilibrium membrane binding of C2AB and C2AB mutants to membranes of varied lipid composition . For experiments where tryptophan emission was used to detect C2AB , final protein concentrations of 0 . 2–0 . 35 µM were used with sucrose-loaded LUVs at lipid concentrations ranging from 0 . 02 mM to 15 mM . The vesicles and C2AB were incubated at room temperature for 10 min , followed by centrifugation at 160 , 000 × g for 1 hr to pellet the LUVs . The concentration of C2AB in the supernatant was determined and used to calculate the fraction of membrane-bound protein ( fb ) . A phosphate assay ( Ames , 1966 ) was used to determine the final lipid concentrations . BODIPY-maleimide labeled C2AB was used in cases were binding affinities were high , in order to maintain sufficiently low protein-to-lipid ratios ( Buser and Mclaughlin , 1998 ) . For PtdChol: PtdSer LUVs , accurate binding affinities were achieved at lipid to protein molar ratios of 140:1 or higher . For each experimental condition , at least two measurements of the fraction of bound protein , fb , were made to determine a reciprocal molar partition coefficient , K . This partition coefficient represents the accessible lipid concentration , [L] , where half the protein is bound and is given by: ( 1 ) fb=K[L]1+K[L] The value of K ( M-1 ) and the standard errors were determined from the data using the fitting function in OriginPro 7 . 5 ( Origin Lab , Northampton , MA ) . C2A and C2B were expressed in BL21 ( DE3 ) pLysS cells , and grown in minimal media where 15NH4Cl ( Cambridge Isotopes , Andover , MA ) was the sole nitrogen source . The purification followed the same protocol mentioned above . The final protein at a concentration of 0 . 4–0 . 8 mM in NMR buffer ( 3 mM CaCl2 , 150 mM NaCl , 50 mM MES , pH 6 . 3 ) was used for two-dimensional 1 hr 15N HSQC experiments ( Ubach et al . , 2001 ) . The ligands ( o-phosphoserine or InsP3 ) were titrated into C2A or C2B at concentrations ranging from 0 . 05 mM to 4 mM . The final NMR protein samples contained 10% D2O and were placed in Shigemi tubes ( Sigma-Aldrich , St . Louis , MO ) . The experiments were conducted in Bruker 600 MHz NMR spectrometer at temperature 27°C , and DDS ( 4 , 4-dimethyl-4-silapentane-1-sulfonic acid ) in NMR buffer was used as a chemical shift standard . The NMR data were processed in NMRPipe ( Delaglio et al . , 1995 ) and the residue assignments were matched in Sparky ( Goddard and Kneller , 2008 ) from the PDB NMR resonance assignments of C2A ( PDB 1BYN ) and C2B ( PDB 1K5W ) ( Fernandez et al . , 2001; Shao et al . , 1997; Shao et al . , 1998 ) . The C2AB mutants , M173C , V304C , L323C , K327C , T329C , were purified and spin labeled using MTSL ( 1-oxy-2 , 2 , 5 , 6-tetramethylpyrroline-3-methyl; Santa Cruz Biotech , Dallas , TX ) . Each sample was spin labeled at a 1:1:10 mole ratio of protein:DTT:MTSL , and the mixture was passed through a HiPrep 26/10 desalting column ( GE Healthcare , Pittsburgh , PA ) to remove any free spin labels . The spin-labeled protein was concentrated to 50–200 uM using an Amicon centrifugal concentrator ( EMD Millipore , Billerica , MA ) . EPR spectra and power saturation measurements were performed on an X-band Bruker EMX spectrometer using a room temperature ER 4123D dielectric resonator . Samples in the Ca2+-free condition contained 5 mM EGTA . The samples with LUVs had a final lipid concentration of 25 mM with a lipid-to-protein ratio of at least 200:1 , ensuring that all protein was completely membrane associated . All spectra were 100 Gauss scans recorded at 2 mM incident microwave power and 1 G modulation . The spectra were baseline corrected and normalized by double integration using a LabVIEW ( Austin , TX ) -based program ( Dr . Altenbach , UCLA ) . Continuous wave power saturation experiments were performed as previously described ( Kuo et al . , 2011 ) , using 12 microwave power steps ranging from 0 . 6 to 36 mW . A LabVIEW-based program ( Dr . Altenbach , UCLA ) was used to plot the data points and determine P1/2 values from which ΔP1/2 values for NiEDDA and O2 were obtained . These data yielded depth parameters , Φ , which were used to estimate the distance ( x ) from the lipid phospholipid phosphates using an empirically derived calibration curve ( Frazier et al . , 2002 ) . Membrane depth data obtained from EPR power saturation were used to generate models for the position of the C2B domain in bilayers containing either PtdSer or PtdIns ( 4 , 5 ) P2 . The approach utilized a version of Xplor-NIH ( Schwieters et al . , 2003 ) that included a plane distance potential , and followed the general procedure described previously ( Herrick et al . , 2009 ) . Briefly , to the high-resolution NMR structure for the C2B domain ( PDB ID: 1K5W [Fernandez et al . , 2001] ) the site-scan feature of the software package MMM ( Polyhach et al . , 2011; Polyhach and Jeschke , 2010 ) was used in conjunction with the warsh rotameric library to determine possible rotamers for the positions of interest . The most probable rotamer was then attached to each site . The depth parameters shown in Table 1 were used as point-to-plane distance restraints and the labeled structure was subjected to simulated annealing and energy minimization to dock the domain to the membrane interface . The spin label atoms and all protein backbone atoms were fixed for the duration of the simulated annealing runs . For these point-to-plane restraints , the distance range was set to the uncertainty in the label position , which was determined from the experimental error in the membrane depth parameter and the empirically derived calibration curve ( Herrick et al . , 2006 ) . Approximately 100 structures were generated from each simulated annealing run . The structures were analyzed and figures were generated using the program PyMOL ( Schrödinger , Cambridge , MA ) . All data are shown as means ± SD ( n = 2–10 ) , and all statistical analyses were performed by one-way ANOVA and Fisher post hoc test ( p-value < 0 . 05 ) .

The human nervous system contains billions of neurons that communicate with each other across junctions called synapses . When a neuron is activated , the levels of calcium ions inside the cell rise . This causes molecules called neurotransmitters to be released from the neuron at a synapse to make contact with the second neuron . The neurotransmitters are stored inside cells within compartments known as synaptic vesicles and are released when these vesicles fuse with the membrane surrounding the cell . Proteins called SNAREs regulate the membrane fusion process . These proteins assemble into bundles that help to drive vesicle and cell membranes together . Another protein called synaptotagmin-1 sticks out from the vesicle membrane and senses the levels of calcium ions in the cell to trigger membrane fusion at the right time . Synaptotagmin-1 has two regions that can bind to calcium ions , known as the C2 domains . When calcium ion levels rise , these domains insert into the cell membrane by binding to two fat molecules in the membrane called phosphatidylserine ( PtdSer ) and phosphatidylinositol 4 , 5-bisphosphate ( PtdInsP2 ) . Synaptotagmin-1 also interacts with the SNARE proteins , but it is not known whether synaptotagmin-1 triggers fusion by binding directly to SNAREs , or by the way it inserts into the cell membrane . Pérez-Lara et al . used several biophysical methods to investigate how synaptotagmin-1 binds to PtdSer and PtdInsP2 . The experiments show that these molecules bind to different regions of synaptotagmin-1 and work together to attach the protein to the cell membrane and insert the C2 domains . Calcium ions increase the affinity of synaptotagmin-1 binding to the cell membrane by making it harder for synaptotagmin-1 to separate from the membrane , rather than by increasing its ability to bind to it . Further experiments show that synaptotagmin-1 prefers to bind to membranes that contain PtdInsP2 over binding to the SNARE proteins . Together , the findings of Pérez-Lara et al . suggest that calcium ions may trigger the release of neurotransmitters by trapping synaptotagmin-1 at the cell membrane rather than by directly affecting how it interacts with SNARE proteins . Further work will be needed to establish exactly how the SNARE proteins , PtdInsP2 and synaptotagmin-1 interact .

2016

PtdInsP2 and PtdSer cooperate to trap synaptotagmin-1 to the plasma membrane in the presence of calcium

Red blood cells ( RBC ) must coordinate their rate of growth and proliferation with the availability of nutrients , such as iron , but the signaling mechanisms that link the nutritional state to RBC growth are incompletely understood . We performed a screen for cell types that have high levels of signaling through mTORC1 , a protein kinase that couples nutrient availability to cell growth . This screen revealed that reticulocytes show high levels of phosphorylated ribosomal protein S6 , a downstream target of mTORC1 . We found that mTORC1 activity in RBCs is regulated by dietary iron and that genetic activation or inhibition of mTORC1 results in macrocytic or microcytic anemia , respectively . Finally , ATP competitive mTOR inhibitors reduced RBC proliferation and were lethal after treatment with phenylhydrazine , an inducer of hemolysis . These results identify the mTORC1 pathway as a critical regulator of RBC growth and proliferation and establish that perturbations in this pathway result in anemia . Cells must coordinate their rate of growth and proliferation with the availability of nutrients . mTOR , a serine–threonine kinase , is one the key proteins responsible for nutrient signaling in eukaryotic cells . mTOR is activated by conditions that signal energy abundance , such as the availability of amino acids , growth factors , and intracellular ATP . Activated mTOR then phosphorylates a set of downstream targets that promote anabolic processes , such as protein translation and lipid biosynthesis , while suppressing catabolic processes such as autophagy ( Zoncu et al . , 2011 ) . mTOR resides in two cellular complexes that have distinct functions and regulation ( Loewith et al . , 2002; Sarbassov et al . , 2004 ) . mTOR complex 1 ( mTORC1 ) is sensitive to inhibition by the natural product rapamycin and contains the protein Raptor . Key targets of mTORC1 include S6 kinase ( S6K ) , which influences cell size , and the eIF4-E binding protein ( 4E-BP1 ) , which regulates cell proliferation through effects on cap-dependent translation ( Shima et al . , 1998; Dowling et al . , 2010 ) . mTOR complex 2 ( mTORC2 ) is resistant to rapamycin and contains the protein Rictor . mTORC2 phosphorylates and activates several kinases in the AGC family , such as Akt and SGK , on a sequence known as the hydrophobic motif ( Sarbassov et al . , 2005 ) . As Akt itself activates mTORC1 by phosphorylation of the Tsc1/2 complex , these two kinases reciprocally regulate each other in response to growth factor signals . TOR was discovered in yeast , where it functions as a nutrient sensor regulating cell growth and proliferation ( Heitman et al . , 1991 ) . This cell-autonomous function is conserved in higher organisms , and there has been rapid progress in delineating the molecular pathways by which mTOR controls basic cellular processes such as protein translation ( Zoncu et al . , 2011 ) . Less is understood about how mTOR signaling links nutrient signals to cellular growth to regulate the physiologic state of multicellular organisms such as mammals . However , the availability of mice bearing conditional alleles of essential mTORC1 and mTORC2 components has made it possible to begin to dissect the physiologic functions of these signaling complexes using cell-type-specific Cre drivers ( Chen et al . , 2008; Gan et al . , 2008; Gan and DePinho , 2009; Kalaitzidis et al . , 2012 ) . In a previous report , we developed a systematic approach to identify the cell types that show high levels of mTORC1 activity in vivo ( Knight et al . , 2012 ) . This approach takes advantage of the fact that activation of mTORC1 leads to rapid phosphorylation of ribosomal protein S6 ( rpS6 ) on five C-terminal serine residues ( Figure 1A; Ser 235 , 236 , 240 , 244 , 247 ) ( Pende et al . , 2004; Roux et al . , 2007 ) . As rpS6 is a structural component of the ribosome , this phosphorylation introduces a ‘tag’ on the ribosomes from cells that have active mTORC1 signaling . We thus used phospho-specific antibodies to S6 to selectively immunoprecipitate these phosphorylated ribosomes from cells with high levels of TORC1 signaling in mouse brain homogenates , thereby enriching for the mRNA expressed in a subpopulation of activated cells ( Figure 1A ) ( Knight et al . , 2012 ) . 10 . 7554/eLife . 01913 . 003Figure 1 . Reticulocytes have elevated mTORC1 signaling . ( A ) Diagram showing the phosphorylation sites on ribosomal protein S6 and strategy for immunoprecipitation of phosphorylated ribosomes . ( B ) Quantification by RNA-Seq of the abundance of each transcript in the pS6 244 immunoprecipitate ( y-axis ) vs the total hypothalamus ( x-axis ) . Hba-a1 and hbb-b1 are enriched in the pS6 immunoprecipitate and labeled . ( C ) Total RNA was prepared from dissected hypothalami of mice that had been perfused with saline ( black bars ) or not perfused ( white bars ) , and the relative abundance of each transcript was quantified by Taqman . Hba-a1 and hbb-b1 but not actin and pomc are depleted from the hypothalamus by perfusion . Values are normalized to rpl23 . ( D ) Lysates from the mouse brain or RBCs were blotted for pS6 at the indicated sites . Neuron-specific enolase ( NSE ) and alpha globin are specific markers for the brain and RBCs , respectively . ( E ) Quantification of the relative phosphorylation of S6 at the indicated sites in the brain vs RBCs . Values are expressed as the ratio of pS6 to total S6 . ( F ) Western blotting for pS6 235/236 in ribosomes purified from a range of mouse tissues . *p < 0 . 05 . ***p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 01913 . 00310 . 7554/eLife . 01913 . 004Figure 1—figure supplement 1 . α-globin is not expressed in a specific population of hypothalamic neurons . ( A ) Microarray analysis reveals that hba-a1 , hbb-b1 , and the neuropeptide vip are among the most highly enriched genes in pS6 immunoprecipitates from the hypothalamus . ( B ) VIP is expressed in a specific population of neurons in the suprachiasmatic nucleus that show high levels of pS6 . No specific staining for α-globin could be detected . ( C ) Enrichment of hba-a1 and hbb-b1 is similar in immunoprecipitates from homogenates of the cortex and hypothalamus . As these regions contain non-overlapping neural populations , this indicates that hba-a1 and hbb-b1 enrichment is unlikely to reflect specific expression in a neural population with high levels of pS6 . DOI: http://dx . doi . org/10 . 7554/eLife . 01913 . 004 Our previous report focused on the use of this approach to identify markers for activated neurons in the mouse brain . However , we also noted that genes encoding the protein subunits of hemoglobin were highly enriched in our pS6 immunoprecipitates . In this study , we report that these transcripts are derived from reticulocytes , immature red blood cells ( RBCs ) , that we find have especially high levels of mTORC1 signaling . We further use a combination of pharmacologic , genetic , and nutritional perturbations to delineate a critical role for mTORC1 signaling in RBC development and the pathogenesis of anemia , suggesting that this pathway links the availability of iron to cell growth and hemoglobin synthesis during erythropoiesis . We recently described a method for molecular profiling of activated neurons in the mouse brain ( Knight et al . , 2012 ) . This approach takes advantage of the fact that ribosomal protein S6 is phosphorylated following neural activity ( Lenz and Avruch , 2005; Villanueva et al . , 2009; Zeng et al . , 2009; Valjent et al . , 2011; Bertran-Gonzalez et al . , 2012 ) . These phosphorylated ribosomes can then be immunoprecipitated from mouse brain homogenates , enriching for the mRNA expressed in a subpopulation of activated cells ( Figure 1A ) . During the course of these studies , we noticed that Hba-a1 and Hbb-b1 were highly enriched transcripts in pS6 immunoprecipitates from the mouse hypothalamus and other brain regions ( Figure 1B ) . Hba-a1 and Hbb-b1 encode α- and β-globin , the protein subunits of hemoglobin . As hemoglobin is not highly expressed in the brain , the enrichment of these transcripts was unexpected and we set out to clarify their cellular origin . We initially considered the possibility that hemoglobin might be expressed in a specific population of neurons that have high levels of pS6 at baseline . For example , VIP neurons of the suprachiasmatic nucleus ( SCN ) have high levels of pS6 , and VIP mRNA is highly enriched in pS6 immunoprecipitates from the hypothalamus ( Figure 1—figure supplement 1 ) . However , consistent with the data from the Allen Brain Atlas , we were unable to detect specific α-globin expression in the SCN or any other hypothalamic region by immunostaining or in situ hybridization . We thus considered the possibility that the globin RNA was not derived from a specific neural population but from another cell type ( figure 1—figure supplement 1 ) . Hba-a1 and Hbb-b1 are most abundantly expressed in reticulocytes , immature RBCs that circulate in the blood . To test whether the Hba-a1 and Hbb-b1 transcripts originated from the circulating cells , we perfused mice with saline to remove blood from the tissue and then quantified the amount of globin mRNA remaining in hypothalamic extracts . Perfusion removed approximately 95% of Hba-a1 and Hbb-b1 mRNA from hypothalamus but had no effect on transcripts expressed in neurons or glia , such as Actb or Pomc ( Figure 1C ) . These data show that the vast majority of Hba-a1 and Hbb-b1 mRNA in the brain originates from the circulating cells . To determine if Hba-a1 and Hbb-b1 were the only enriched erythroid transcripts in the blood , we scanned the RNAseq data for altered expression of other genes expressed in cells of the erythropoietic lineage . In contrast to transcripts for Hbb , we failed to find enrichment for erythroid catalase , carbonic anhydrase II , two cytoplasmic proteins , or sprectrin-a , spectrin-b , and ankyrin , which are membrane proteins . The observation that the globin transcripts in our pS6 immunoprecipitates were derived from circulating cells suggested that reticulocytes were the source of this RNA and that reticulocytes might have unusually high levels of pS6 . Furthermore , since pS6 is widely used as a marker for the activation of the mTORC1 pathway ( Meyuhas , 2008 ) , the data further suggested that reticulocytes might have particularly high levels of mTORC1 signaling . To test these possibilities , we first used western blotting to quantify the level of pS6 in the lysates from the brain and RBCs . Consistent with our ribosome profiling data , reticulocyte lysates had a much higher level of pS6 at both Ser 235/236 and Ser 240/244 compared to extracts from the brain as a whole ( Figure 1D , E ) . We then extended this analysis by purifying the ribosomes from a panel of mouse tissues , including fat , testis , lungs , liver , heart , pancreas , stomach , and kidney , and then analyzed the level of pS6 in these tissues compared to blood . Remarkably , reticulocytes had the highest level of pS6 in any tissue we examined ( Figure 1F ) . Thus , our ribosome profiling data from the brain revealed the unexpected finding that mTORC1 signaling is highly active in the RBC lineage . Reticulocytes are highly specialized for hemoglobin synthesis , and we wondered whether the elevated mTORC1 activity in this cell type might reflect its uniquely high demand for protein translation . The mTORC1 pathway plays a general role in linking nutrient availability to protein translation in diverse tissues ( Sengupta et al . , 2010 ) , but relatively little is known about its function and regulation in RBCs . However , recent work has shown that mTORC1 can be regulated by the availability of iron in cell lines ( Ohyashiki et al . , 2009 ) and the brain ( Ndong et al . , 2009; Fretham et al . , 2013 ) , and , in turn , that mTORC1 activity can modulate downstream enzymes that control intracellular iron metabolism ( Bayeva et al . , 2012; La et al . , 2013 ) . These data raise the possibility that mTORC1 activity in RBCs might help coordinate the rate of translation with the availability of iron , although this had not been directly tested . To test whether iron can regulate mTORC1 signaling in RBCs , wild-type mice were made iron deficient by placing them on a low iron diet combined with daily injections of the iron chelator deferoxamine . Complete blood counts were performed after 1 month and showed that the treated animals had developed microcytic anemia: the low iron cohort showed a reduction in RBC volume ( 58 . 7 vs 52 . 3 fL , p < 0 . 001 for reticulocytes; 48 . 3 vs 46 . 1 fL for mature RBC , p < 0 . 001 ) and RBC hemoglobin content ( 15 . 3 vs 13 . 2 pg for reticulocytes , p < 0 . 001; 14 . 0 vs 13 . 4 pg for mature RBC , p < 0 . 01 ) relative to control animals ( Figure 2A , B ) . To assess the level of mTORC1 signaling in these cells , we prepared lysates from RBCs from animals in each cohort and performed western blots for pS6 . Iron deficiency greatly reduced the level of pS6 in RBCs at both Ser 235/236 as well as Ser 240/244 , the latter of which is a specific marker for mTORC1 activity ( Roux et al . , 2007; Meyuhas , 2008 ) ( Figure 2C , D ) . Consistent with previous data , we observed a similar inhibition of mTORC1 signaling when we treated the erythroleukemia cell line K562 with iron chelators in vitro ( Figure 2E ) , ( Ohyashiki et al . , 2009 ) . Thus iron deficiency results in a marked reduction in mTORC1 signaling in RBCs in vitro and in vivo . 10 . 7554/eLife . 01913 . 005Figure 2 . mTORC1 signaling in RBCs is regulated by iron . ( A ) Mean cell volume ( MCV ) of reticulocytes and mature RBCs in mice that were challenged with a low iron diet ( black bars ) or maintained on a chow diet ( white bars ) for 1 month . ( B ) Cellular hemoglobin of reticulocytes and mature RBCs in mice that were challenged with a low iron diet ( black bars ) or maintained on a chow diet ( white bars ) for 1 month . ( C ) Western blot for pS6 from RBCs in mice that that were challenged with a low iron diet ( black bars ) or maintained on a chow diet ( white bars ) for one month . ( D ) Quantification of the relative phosphorylation of S6 at the indicated sites in the RBCs from mice on a low iron ( black bars ) or chow ( white bars ) diet . Values are expressed as the ratio of pS6 to total S6 . ( E ) K562 cells were treated with the iron chelator deferoxamine ( DFO ) , deferasirox ( DFS ) , or vehicle ( 0 . 05% DMSO ) for 24 hr , and then lysed and levels of pS6 analyzed by western blotting . *p < 0 . 05 . ***p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 01913 . 005 The regulation of mTORC1 signaling by iron suggests a possible mechanism for linking nutrient status , in this case iron availability , to RBC size . To explore the role of mTORC1 in RBC growth and hemoglobin synthesis , we first used a genetic approach to either increase or decrease mTORC1 signaling selectively in hematopoietic cells . mTORC1 is under tonic inhibition by the Tsc1/Tsc2 complex , and Tsc1 deletion results in constitutively increased mTORC1 signaling . Previous work has shown that inducible deletion of Tsc1 in adult animals using either Mx1Cre or Gt ( ROSA ) 26SorCre-ERT2 promotes rapid cycling of the hematopoietic stem cells ( HSC ) and is associated with decreased hematopoiesis due to HSC exhaustion ( Chen et al . , 2008; Gan et al . , 2008; Gan and DePinho , 2009 ) . However , Mx1 and Gt ( ROSA ) 26Sor are expressed in multiple tissues and do not provide information on the effects of cell autonomous Tsc1 deletion in the RBC lineage . We thus assessed the effect of mTORC1 activation in hematopoietic cells by using Vav1Cre mice to delete a floxed allele of Tsc1 ( Kwiatkowski et al . , 2002 ) . Vav-1 is a guanosine nucleotide exchange factor that is exclusively expressed in the hematopoietic lineage ( Zhang et al . , 1994 ) . We selected a VavCre driver because it has been reported to catalyze highly efficient excision of loxP flanked sequences in erythrocytes and in previous studies enabled analyses of several different loss of function phenotypes in the RBC lineage ( Pan et al . , 2007; Gan et al . , 2010; Mortensen et al . , 2010; Schultze et al . , 2012 ) . In contrast , several other Cre lines that use different erythrocyte or hematopoietic-specific promoters have been shown to often produce variegated effects with incomplete recombination in RBCs ( Peterson et al . , 2004; Mortensen et al . , 2010 ) . Vav1CreTsc1fl/fl animals ( hereafter called TSC-KO ) were born at the expected ratio but were smaller than their littermates and had increased mortality in the neonatal period . However , animals that survived to 8 weeks of age had a normal appearance . Western blotting revealed that RBCs from TSC-KO mice had markedly increased levels of pS6 relative to littermate controls , confirming that these cells have constitutive activation of mTORC1 signaling ( Figure 3A ) . 10 . 7554/eLife . 01913 . 006Figure 3 . Tsc1 deletion in hematopoietic cells results in hyperchomic macrocytic anemia . ( A ) Western blots for pS6 at Ser 235/236 and Ser 240/244 in VavCreTsc1fl/fl mice ( TSC-KO ) or littermate controls . ( B ) Mean cell volume ( MCV ) and hemoglobin content per cell of reticulocytes and mature RBCs from control ( white bars ) and TSC-KO ( black bars ) mice . ( C ) RBC number , hematocrit , and total blood hemoglobin from control ( white bars ) and TSC-KO ( black bars ) mice . ( D ) Spleen weight and reticulocyte percentage from control ( white bars ) and TSC-KO ( black bars ) mice . ( E ) Percentage of bone marrow cells corresponding to each erythrocyte progenitor subtype from control ( white bars ) and TSC-KO ( black bars ) mice . Values were determined by flow cytometry for Ter119 and CD71 as described in methods . ( F ) Percentage of splenocytes corresponding to each erythrocyte progenitor subtype from control ( white bars ) and TSC-KO ( black bars ) mice . ( G ) Percentage of TSC-KO mice remaining alive at the indicated times after transition to low iron diet ( red ) or control diet ( black ) . No mortality was observed in littermate controls on either diet . ( H ) Quantification of mean cell volume , hemoglobin content per cell , and percentage of cells with low hemoglobin , in control and TSC-KO mice exposed to a control ( white bars ) or low iron ( black bars ) diet . *p < 0 . 05 , **p < 0 . 01 , ****p < 0 . 0001 . DOI: http://dx . doi . org/10 . 7554/eLife . 01913 . 006 Analysis of peripheral blood from TSC-KO mice revealed a constellation of changes consistent with macrocytic anemia ( Figure 3 ) . Compared to littermate controls , TSC-KO mice had larger reticulocytes ( 66 . 7 vs 58 . 6 fL , p < 0 . 0001 ) and larger mature RBCs ( 54 . 5 vs 48 . 1 fL , p < 0 . 0001 ) and the larger RBCs from TSC-KO mice likewise had significantly more hemoglobin ( 17 . 4 vs 15 . 1 pg in reticulocytes; 16 . 2 vs 13 . 9 pg in mature RBCs , p < 0 . 0001 ) . In addition , the percentage of cells with elevated hemoglobin was dramatically increased in both young and old RBCs ( 29 . 2 vs 5 . 00% in reticulocytes; 13 . 5 vs 0 . 58% in mature RBCs , p < 0 . 0001 ) . Despite this increase in RBC size and hemoglobin content , TSC-KO mice were anemic compared to littermate controls ( 39 . 4 vs 45 . 7% hematocrit , p < 0 . 0001; Figure 3C ) . The reduced hematocrit of TSC-KO mice resulted from a reduction in the RBC number ( 7 . 58 vs 9 . 89 × 10^9 cells/mL , p < 0 . 0001 ) , such that the total amount of hemoglobin in the peripheral blood was reduced ( 1 . 12 vs 1 . 28 g/dL , p < 0 . 01 ) . Thus constitutive activation of mTORC1 leads to a macrocytic anemia with an increase in hemoglobin per cell ( Chen et al . , 2008; Gan et al . , 2008; Gan and DePinho , 2009 ) . We also noticed that TSC-KO mice had a higher percentage of circulating reticulocytes ( 5 . 04 vs 2 . 97% , p < 0 . 01 ) with more than threefold increase in the size of their spleens ( 360 vs 80 . 5 mg , p < 0 . 01 ) relative to littermate controls ( Figure 3D ) . These changes are often characteristic of stress erythropoiesis , suggesting erythropoiesis in the spleen of TSC-KO mice is activated to compensate for reduced RBC production by the bone marrow . To confirm this , we isolated splenocytes and bone marrow cells ( BMCs ) from TSC-KO and control mice and used flow cytometry to quantify the abundance of RBC progenitors in these two compartments . Consistent with this possibility , TSC-KO mice showed a decrease in the number of erythroid progenitors in the bone marrow ( EryB: 5 . 08 vs 16 . 3 , p < 0 . 01 ) , and a reciprocal increase in the number of erythroid progenitors in the spleen ( EryA: 3 . 15 vs 10 . 4% , p < 0 . 01 ) ( Figure 3E ) . Thus , we find that constitutive activation of mTORC1 is sufficient to recapitulate a constellation of features that characterize macrocytic anemia , including an increase in RBC size and hemoglobin content coupled to a decrease in RBC counts and hematocrit . RBCs must balance the rate of globin synthesis with the availability of iron , so that free globin peptides do not aggregate in the absence of heme . An excess of free globin decreases red blood cell survival in a number of conditions including thalassemia . Thus , chronic iron deficiency triggers an adaptive response in which hemoglobin synthesis and RBC size are reduced , resulting in a microcytic hypochromic anemia in which the low heme levels are balanced by lower levels of globin synthesis . During iron deficiency the activity of heme-regulated eIF2alpha kinase ( HRI ) , a protein kinase that is selectively expressed in RBCs is increased and acts to reduce hemoglobin synthesis . This kinase is phosphorylated and inhibited by free heme and is disinhibited with iron depletion ( Chen , 2007 ) . HRI knockout mice do not reduce globin synthesis after iron deficiency indicating that activation of this kinase plays an important role in the adaptive response to dietary iron deficiency ( Han et al . , 2001 ) . However , since mTORC1 is inhibited by iron deficiency ( Figure 2 ) , this raised the question of whether the constitutive mTORC1 activation could also exacerbate the response to iron deficiency . To test this we challenged two cohorts of TSC-KO mice as well as littermate controls with either a low iron or control diet for two months and then measured the effect on RBC growth and proliferation . Similar to the effects of an HRI deficiency on the response to iron deficiency , TSC-KO mice became visibly moribund and showed progressive mortality beginning at 3 weeks on an iron deficient diet , such that only 64% of the TSC-KO mice survived the 2-month experiment ( Figure 3G ) . By contrast , littermate controls were more normal in appearance and had a higher survival . Thus chronic mTORC1 activation in RBCs is maladaptive in the context of iron deficiency . Complete blood counts after 2 months revealed that , similar to controls , TSC-KO mice showed a reduction in the RBC volume and hemoglobin content on the low iron diet . ( Figure 3H ) . The iron deficient TSC-KO mice also displayed splenomegaly and reticulocytosis . We also noted premature mortality in TSC-KO mice whether or not they were on a low iron diet . Autopsies of two TSC-KO mice , one on chow and another on an iron deficient diet at 6 months of age were performed . Both mice had extramedullary hematopoiesis with multifocal myeloid hyperplasia which was described as moderate . Consistent with the greater mortality of animals on low iron diet , the iron deficient TSC-KO mouse showed increased number of phagocytosed red blood cells in multiple organs compared to the chow-fed TSC-KO animal . These data are consistent with the possibility that constitutive activation of the mTOR pathway in the setting of iron deficiency , which was further exacerbated by phlebotomy for blood collection , decreased RBC viability and increased mortality at younger ages . ( Kazemi et al . , 2007 ) All of the TSC-KO mice died prematurely by ∼6 months of age and both animals also developed histiocytic sarcomas which showed diffuse infiltration of neoplastic round cells in thymus , lung , liver , kidneys , spleen , bone marrow , meninges , uterus , and lymph nodes . Note , a previous report showed that animals with increased mTOR signaling as a result of a Pten mutation in HSCs using Mx1Cre mice also developed histiocytic sarcomas ( Lee et al . , 2010 ) . Since mTORC1 activity is increased by iron and constitutive mTORC1 activation results in macrocytic anemia , we next considered the possibility that mTORC1 inhibition would cause microcytic anemia . Germline deletion of either mTOR or the essential mTORC1 component Raptor results in embryos that die shortly after implantation ( Guertin et al . , 2006 ) , which has complicated analysis of the requirement for mTORC1 signaling in erythropoiesis . Experiments using inducible deletion of mTORC1 subunits in the hematopoietic cells have yielded variable results . For example , an inducible deletion of Rptor in either Mx1Cre or ERT2Cre animals was reported to be well tolerated in adult animals ( Hoshii et al . , 2012; Kalaitzidis et al . , 2012 ) . By contrast , inducible deletion of the mTOR kinase in adult mice has recently been reported to result in severe anemia and lethality within 17 days ( Guo et al . , 2013 ) . The more dramatic phenotype following mTOR deletion may reflect the role of other mTOR complexes in erythropoiesis or reflect differences in recombination efficiency and selectivity between these experiments . To directly address the role of mTOR during the development of the hematopoietic system , we mated Vav1Cre to mice with a floxed allele of Rptor ( de Boer et al . , 2003 ) . Initial crosses between Vav1CreRptorfl/+ and Rptorfl/fl animals failed to yield any viable Vav1CreRptorfl/fl ( hereafter called RAP-KO ) offspring ( >100 offspring genotyped from >20 litters ) . The Vav1Cre transgene is expressed beginning at E13 , which corresponds to the onset of definitive erythropoiesis in the fetal liver . We therefore dissected and analyzed embryos from these crosses between E13 and E17 . 5 and found that embryos with the Vav1CreRptorfl/fl genotype were present at the expected ratio and morphologically normal . However , RAP-KO embryos were strikingly pale compared to their littermates ( Figure 4A ) . Pallor during development is characteristic of a diverse set of mouse mutants that have a defect in definitive erythropoiesis . Mutations that cause a complete block of erythropoiesis , such as loss of erythropoietin or its receptor , lead to embryonic lethality at approximately E13 ( Wu et al . , 1995 ) . In contrast , we found that many RAP-KO embryos continued to develop throughout gestation . Monitoring of the delivery from pregnant females from these crosses led to the identification of several live-born RAP-KO pups , but , similar to the embryos , these animals were extremely pale and survived less than 1 day ( Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 01913 . 007Figure 4 . Raptor deletion in hematopoietic cells results in hypochromic microcytic anemia and perinatal lethality . ( A ) Embryos dissected at E16 . 5 illustrating the pallor of VavCreRaptorfl/fl ( RAP-KO ) mice relative to littermate controls . ( B ) Complete blood counts from the fetal liver of E17 RAP-KO embryos and littermate controls . ( C ) Western blotting for the level of α-globin and actin in RAP-KO mice or littermate controls . ( D ) Quantification of the percentage of fetal liver cells corresponding to various erythroid progenitors at E16 . 5 by flow cytometry . *p < 0 . 05 , **p < 0 . 01 , ***p < 0 . 001 , ****p < 0 . 0001 . DOI: http://dx . doi . org/10 . 7554/eLife . 01913 . 00710 . 7554/eLife . 01913 . 008Figure 4—figure supplement 1 . Newborn RAP-KO pups are extremely pale . Picture of RAP-KO pup ( left ) and normal littermate ( right ) within 1 hr of birth . DOI: http://dx . doi . org/10 . 7554/eLife . 01913 . 008 VavCre is expressed in all cells of the hematopoietic lineage . Therefore , the perinatal lethality of RAP-KO embyros could in principle result from Rptor deletion in other cell types such as lymphocytes . To exclude this possibility , we knocked out Rptor specifically in lymphoid cells using an hCD2Cre driver . In contrast to RAP-KO mice , hCD2CreRptorfl/fl mice that have Rptor deleted selectively in the B and T cells ( de Boer et al . , 2003 ) had normal birth-rates and survival , indicating that Raptor loss in the lymphocytes is unlikely to contribute to the perinatal lethality of RAP-KO animals . Thus there were no other major defects in any of the other hematopoietic compartments that might have contributed to their early lethality; lymphocyte percentages were comparable in the RAP-KO with their littermates , and the total white cell count was higher ( p < 0 . 01 ) in the RAP-KO . There were no significant differences in the number of basophils , platelets , and monocytes . These data show that the abnormalities in the hematopoietic system of mice with Vav-cre specific inactivation of mTORC1 are limited to erythroid cells . Consistent with their visual pallor , hematological analysis of fetal livers from RAP-KO embryos revealed a broad impairment in erythropoiesis ( Figure 4B ) . RAP-KO embryos were anemic ( Hct 15 . 4 vs 27 . 5% , p < 0 . 01 ) with reduced RBC numbers ( 1 . 72 vs 2 . 67 × 10^9 cells/mL , p < 0 . 001 ) . Analysis of specific erythroid progenitors by flow cytometry ( Figure 4D ) revealed that RAP-KO embryos have an increased abundance of the earliest hematopoietic progenitors ( ProE , 5 . 95 vs 13 . 7% , p < 0 . 001 ) and a decrease in the abundance of more mature populations ( EryC , 6 . 15 vs 2 . 70% , p < 0 . 01 ) relative to littermates . This shift in the distribution of fetal liver progenitors is similar to what has been observed in previous studies of the bone marrow of adult mice following inducible deletion of the mTOR kinase , though in these studies the hematopoietic phenotype was attributed to increased apoptosis of bone marrow RBCs ( Guo et al . , 2013 ) . We also found that , similar to the effect of iron deficiency , RBCs from RAP-KO embryos were smaller than those from littermate controls ( 92 . 8 vs 106 fL , p < 0 . 05 ) and had less total hemoglobin ( 24 . 0 vs 28 . 5 pg/cell , p < 0 . 0001 ) ( Figure 4B ) . The defect in hemoglobin synthesis was particularly severe , as there was an eightfold increase in the percentage of cells that were hypochromic in RAP-KO embyros ( 39 . 3 vs 4 . 71 , p < 0 . 0001 ) . We confirmed this reduction in hemoglobin expression biochemically by western blotting , which showed that RAP-KO embyros had reduced α-globin expression compared to the littermate controls ( Figure 4C ) . Thus genetic ablation of mTORC1 produces a severe microcytic hypochromic anemia that results in perinatal lethality . These data are consistent with the possibility that a reduction of mTORC1 plays a role in the development of the hypochromic microcytic anemia of iron deficiency . However , because the Rptor knockout was lethal during development , we were unable to use genetics to assess this possibility in adult animals . As an alternative , we employed a pharmacologic approach to inhibit mTORC1 in adult mice . In previous studies , treatment of adult animals with the mTORC1 inhibitor rapamycin was shown to have minimal effects on erythropoiesis . For example , chronic rapamycin treatment of rats results in microcytosis without the development of anemia ( Diekmann et al . , 2012 ) . Likewise , the phenotype of mice following an inducible Cre-mediated deletion of Rptor has led to the conclusion that mTORC1 may be dispensable for the regulation of hematopoiesis in adults ( Hoshii et al . , 2012; Kalaitzidis et al . , 2012 ) ( but see also Guo et al . , 2013 ) . To examine the acute requirement for mTOR signaling in adult erythropoiesis , we treated normal mice for 3 days with either rapamycin or MLN0128 , a selective , ATP competitive mTOR inhibitor that is being clinically tested as a treatment for several types of human cancer ( Hsieh et al . , 2012; Infante et al . , 2012 ) . We assayed the effect of these drugs on RBCs in the peripheral blood and bone marrow . Consistent with the previous reports , we found that rapamycin had only a modest impact on adult erythropoiesis , with a reduction in the percentage of circulating reticulocytes ( 3 . 15 vs 2 . 36% , p < 0 . 01 ) but no change in reticulocyte size or hemoglobin content ( Figure 5A ) . 10 . 7554/eLife . 01913 . 009Figure 5 . ATP competitive mTOR inhibitors but not rapamycin block S6K signaling , growth , and proliferation of RBCs . ( A ) Reticulocyte percentage , mean cell volume , and hemoglobin content per cell in mice treated with MLN0128 , rapamycin , or vehicle for 3 days . ( B ) Quantification by flow cytometry of the percentage of bone marrow cells corresponding to various erythroid progenitors following treatment with MLN0128 , rapamycin , or vehicle for 3 days . ( C ) Western blotting for the level of pS6 240/244 and total S6 in RBCs isolated from mice treated with vehicle , KU-0063794 ( 10 mg/kg ) , MLN0128 ( 2 mg/kg ) , or rapamycin ( 10 mg/kg ) for 1 , 2 , or 6 hr . ( D ) Western blotting for the level of pS6 240/244 and total S6 from the livers of the same animals as in panel ( C ) . *p < 0 . 05 , **p < 0 . 01 , ***p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 01913 . 00910 . 7554/eLife . 01913 . 010Figure 5—figure supplement 1 . RBCs have intrinsically reduced sensitivity to rapamycin . ( A ) Whole mouse blood was washed with HBSS , resuspended in RMPI + 10% FBS with the indicated concentrations of inhibitors and then incubated at 37°C for 45 min . Cells were then washed , hypotonically lysed , and the lysate was blotted for pS6 . ( B ) Stress reticulocytes have reduced rapamycin sensitivity in vivo . Mice were treated with phenylhydrazine on days 0 and 2 . On day 5 mice were injected with either rapamycin ( 10 mg/kg ) , INK-128 ( 5 mg/kg ) , or vehicle . 2 hr later the animals were sacrificed , peripheral blood was collected , washed , and subjected to hypotonic lysis . At the same time livers were dissected and homogenized . Lysates from liver ( top ) and RBCs ( bottom ) were then blotted for pS6 . DOI: http://dx . doi . org/10 . 7554/eLife . 01913 . 010 In contrast , acute MLN0128 treatment had profound effects on RBC growth and proliferation ( Figure 5A ) . The percentage of circulating reticulocytes was reduced by more than 75% ( 3 . 15 vs 0 . 58% , p < 0 . 001 ) and the remaining reticulocytes were smaller ( 61 . 6 vs 54 . 7 fL , p < 0 . 001 ) and had less hemoglobin per cell ( 15 . 3 vs 12 . 7 pg , p < 0 . 001 ) . The reduction in reticulocyte number was paralleled by a decrease in the proliferation of erythroid progenitors in the bone marrow ( Figure 5B ) . MLN0128 dramatically reduced the abundance of the early erythroid progenitors EryA ( 23 . 4% vs 0 . 7% , p < 0 . 001 ) and EryB ( 37 . 3% vs 6 . 0% , p < 0 . 001 ) . By contrast , rapamycin caused more modest reduction in only the EryA population ( 23 . 4% vs 14 . 2% , p < 0 . 05 ) . Thus acute mTOR inhibition by MLN0128 , but not rapamycin , strongly blocked RBC growth , hemoglobin synthesis , and the maturation of erythroid progenitors in the bone marrow of treated mice . MLN0128 fully inhibits both mTORC1 and mTORC2 ( Hsieh et al . , 2012 ) , whereas rapamycin blocks only a subset of signaling pathways that are activated by mTORC1 ( Feldman et al . , 2009 ) . This raised the possibility that the differences between the effects of these two drugs could be a result of inhibition of mTORC2 by MLN0128 but not rapamycin . Alternatively , it is possible that mTORC1 in RBCs was not inhibited by rapamycin to the same extent as it was by MLN0128 . In a prior report , a genetic deletion of the essential mTORC2 component rictor had little effect on hematopoiesis , suggesting that TORC2 does not play a major role ( Magee et al . , 2012 ) . For this reason , we compared first the effects of MLN0128 and rapamycin on mTORC1 signaling . Mice were treated with either rapamycin , MLN0128 , or a third , structurally unrelated ATP competitive mTOR inhibitor ( KU-0063794 ) ( Garcia-Martinez et al . , 2009 ) , and the level of pS6 in RBCs was analyzed by western blotting . MLN0128 and KU-0063794 both potently inhibited S6 phosphorylation in RBCs ( Figure 5C ) . The effect of these drugs was similar at 1 and 2 hr , although MLN0128 showed more durable inhibition at later time points ( Figure 5C ) . By contrast , we unexpectedly found that rapamycin only weakly inhibited S6 phosphorylation in RBCs ( Figure 5C ) and that this rapamycin resistance could not be overcome even by chronic rapamycin treatment ( daily for 4 weeks ) at high doses ( 10 mg/kg ) . This effect was specific to RBCs , as rapamycin potently and durably blocked S6K signaling in other tissues from the same animals ( such as liver , Figure 5D ) . Because S6 is not a direct target of mTORC1 , we also assessed the phosphorylation state of 4E-BP1 in animals treated with MLN0128 ( Figure 6B ) . These data showed that phosphorylation of 4E-BP1 was markedly decreased in treated animals . Furthermore , the extent of this reduction was highly correlated with the reduction in the level of phosphorylation of S6 ( Figure 6B ) . 10 . 7554/eLife . 01913 . 011Figure 6 . mTORC1 signaling is required for induction of stress erythropoiesis . ( A ) Survival of mice treated with MLN0128 ( 1 mg/kg ) , rapamycin ( 10 mg/kg ) or vehicle following challenge with phenylhydrazine injection ( 50 mg/kg ) on days 0 and 2 . No lethality was observed by injection of either drug or vehicle alone . ( B ) Western Blotting for the level of p4E-BP1 and 4E-BP1 and pS6 240/244 and total S6 in phenylhydrazine and MLN0128-treated mice . ( C ) Time course of changes in peripheral blood populations following phenylhydrazine challenge of mice treated daily with vehicle or rapamycin . Analysis performed by flow cytometry using Retic-Count as desribed in methods . Autoflourescent RBCs represent damaged RBCs that presumably contain Heinz bodies composed of damaged hemoglobin . ( D ) Time course of changes in erythroid progenitors in the spleen determined by flow cytometry following phenylhydrazine challenge of mice treated daily with vehicle or rapamycin . ( E ) Time course of spleen weight in mice challenged with phenylhydrazine ( square ) or vehicle ( triangle ) and then treated daily with rapamycin ( red ) or vehicle ( black ) . Images of representative spleens from each group are shown below . ( F ) Concentration of erythropoietin in the peripheral blood from mice challenged with phenylhydrazine and then treated with rapamycin ( red ) or vehicle ( black ) daily . *p < 0 . 05 , **p < 0 . 01 , ***p < 0 . 001 , ****p < 0 . 0001 . DOI: http://dx . doi . org/10 . 7554/eLife . 01913 . 011 The reduced effect of rapamycin was also cell autonomous , as RBCs cultured in vitro were sensitive to ATP competitive mTOR inhibitors but not rapamycin ( Figure 5—figure supplement 1 ) . Thus , mTORC1 signaling through the S6K pathway is selectively resistant to rapamycin in RBCs . While certain mTORC1 substrates such as 4E-BP1 are known to display variable rapamycin sensitivity ( Feldman et al . , 2009 ) , rapamycin resistant S6K signaling has to our knowledge never been described ( Choo and Blenis , 2009 ) . This unexpected reduction in the sensitivity of RBCs to rapamycin is consistent with our observation that ATP competitive mTOR inhibitors cause a more pronounced inhibition of RBC growth and proliferation than rapamycin . The effects of MLN0128 on RBC growth and proliferation were greatly diminished after 4 weeks of treatment , at which point the RBC counts and hemoglobin levels of treated mice were nearly normal . This result suggests that there is tachyphylaxis or some other homeostatic response that can compensate for the chronic but not acute mTOR inhibition in RBCs . As acute erythropoiesis is particularly critical under conditions of erythropoietic stress , we next assayed the effect of mTOR inhibitors following drug-induced hemolysis ( Paulson et al . , 2011 ) . Mice were treated with phenylhydrazine on days 0 and 2 along with daily injections of either an mTOR inhibitor or vehicle . Phenylhydrazine causes irreversible oxidative damage to RBCs ( Figure 6C , right ) and triggers acute hematopoiesis with a shift in erythropoiesis from the bone marrow to the spleen ( Figure 6E ) , the proliferation of splenic progenitors ( Figure 6D ) , and the release of stress reticulocytes into the circulation and that rapidly repopulate the hematopoietic system ( Figure 6C , center ) . Vehicle-treated mice showed a normal response to phenylhydrazine-induced hemolysis and recovered normally within 1 week ( Figure 6A ) . By contrast , mice treated with MLN0128 were visibly moribund by day 2 , with reduced mobility , weight loss , and a reduction in body temperature . By day 3 all animals had died or had to be euthanized for animal welfare reasons ( Figure 6A ) . The dramatic mortality of animals treated with MLN0128 indicates that mTOR signaling is essential for the acute response to hemolytic stress . Western blots were performed using protein from the RBCs of phenylhydrazine-treated mice with and without MLN0128 administration . These data showed high levels of phosphorylation of p4E-BP1 , a direct target of mTORC1 and pS6 , an indirect mTorc1 target , in mice treated with phenylhydrazine alone . However , this phosphorylation was completely absent in phenylhydrazine-treated mice that also received MLN0128 . ( Figure 6B ) . As we previously found that rapamycin only partially inhibited mTORC1 signaling in RBCs , we also tested the effect of this drug in the same assay . Approximately 50% of rapamycin-treated animals survived the 8-day experiment , and the onset of mortality occurred several days later than the animals treated with MLN0128 ( Figure 6A ) . Rapamycin treatment also delayed the release of stress reticulocytes into the blood ( Figure 6C ) and reduced by twofold the increase in spleen size that occurred following phenylhydrazine ( Figure 6E ) . These results indicate that the partial mTORC1 inhibition by rapamycin delays but does not block the induction of stress erythropoiesis . To characterize rapamycin's effects on stress erythropoiesis in more detail , we performed a time course analysis of erythroid progenitors in the spleen by flow cytometry in animals treated with phenylhydrazine with and without rapamycin . Early erythroid progenitors ( EryA and EryB ) were present at very low levels in the spleen on day 0 but were rapidly mobilized by phenylhydrazine treatment in control animals , reaching more than 30% of total splenocytes by day 6 ( Figure 6D , black ) . This proliferative response was significantly reduced at all time points tested by rapamycin treatment ( Figure 6D , red ) . By contrast , rapamycin treatment had no effect on erythropoietin levels , which , similar to control mice , increased approximately 400 fold following phenylhydrazine treatment . This suggests that the defect induced by rapamycin results from mTORC1 inhibition in the red cell lineage ( Figure 6F ) . Taken together , these data reveal a critical role for mTORC1 signaling in mediating the rapid growth and proliferation of erythroid progenitors during stress erythropoiesis . ATP competitive inhibitors block this response much more potently than rapamycin , resulting in a lethal failure to adapt to hemolytic stress . As MLN0128 and other ATP competitive mTOR inhibitors are currently being evaluated in clinical trials , these results suggest that further studies will be necessary to assess the effect of these agents under conditions that trigger stress erythropoiesis , such as hypoxia , blood loss , or drug-induced hemolysis . The mTORC1 pathway coordinates protein translation with nutrient availability in organisms ranging from yeast to humans , and the physiologic functions of mTORC1 signaling in specific mammalian tissues have begun to be elucidated . Within the hematopoietic system , an extensive literature has investigated the role of mTORC1 signaling in leukocyte development and function ( Gan et al . , 2008; Gan and DePinho , 2009; Janes et al . , 2010; Hoshii et al . , 2012; Kalaitzidis et al . , 2012; Magee et al . , 2012; Guo et al . , 2013 ) , including its role in hematopoietic stem cells ( HSCs ) and rapamycin-mediated immunosuppression . By contrast , much less is known about the role of the mTORC1 pathway in erythrocytes . Our attention was drawn to the potential significance of mTORC1 signaling in RBCs by screening for cell types in the adult brain that have elevated levels of pS6 , a ribosomal protein that is a downstream target of mTORC1 activation ( Knight et al . , 2012 ) . In that study , we used immunoprecipitation of polysomes with an anti-pS6 antibody to enrich for mRNAs from cells that had activated mTORC1 signaling . This study revealed an unexpected enrichment of globin transcripts in the precipitated polysomes . In this study , we report that circulating reticulocytes are the source of these enriched globin transcripts and that these cells have an especially high level of S6 phosphorylation . Since mTOR is known to link changes in nutrient availability to changes in translation and cell growth , these data suggested that mTOR could play an important role in fine tuning the physiologic state of RBCs in response to alterations in the availability of critical nutrients such as iron . We found that mTORC1 in RBCs is regulated by iron and that iron deficiency results in decreased mTORC1 activity in RBCs in vivo . We further showed that activation of mTORC1 by deletion of Tsc1 results in severe macrocytic anemia , while genetic inhibition of mTORC1 results in a lethal microcytic anemia . Thus bidirectional modulation of mTORC1 signaling is sufficient to cause either a macrocytic and microcytic anemia , both of which can be caused by nutrient deficiency . We further showed that pharmacologic inhibition of mTOR by ATP competitive inhibitors and to a lesser extent rapamycin resulted in an acute blockade of erythropoiesis in adults and that mTORC1 signaling is critically required for survival , following hemolytic stress . Finally , we identified rapamycin resistant S6K signaling in RBCs as a biochemical correlate of the differential effects of these drugs on RBC function . Thus , these studies delineate a critical role for mTORC1 signaling in RBC growth , development , and proliferation , and define how this pathway is selectively modulated by distinct pharmacologic , genetic , or nutritional perturbations leading to anemia . While few studies have directly examined the role of mTORC1 signaling in RBCs , there is some evidence that dysregulation of protein translation downstream of mTORC1 may play a role in the development of anemia . The hereditary anemias , Diamond-Blackfan and 5q-syndrome , are caused by haploinsufficiency of ribosomal proteins S14 and S19 , and recent data show that these anemias can be ameliorated by pharmacologic doses of L-leucine ( Jaako et al . , 2012; Payne et al . , 2012 ) , which activates mTORC1 signaling through S6K and promotes translation . Thus , our finding that RBCs depend critically on signaling through the mTORC1 pathway is consistent with the broader view that RBCs are uniquely sensitive to perturbations that affect protein translation . A reduced activity of mTORC1 is known to reduce the size of many cell types . Consistent with this , a reduced level of mTORC activity leads to a microcytic anemia either as a result of iron deficiency or a red cell specific mutation in Raptor . As mentioned , HRI , another iron-dependent kinase , is also known to play a critical role in mediating the response to iron deficiency ( Kazemi et al . , 2007 ) . Further studies may reveal whether the roles of this kinase and mTORC1 in the response to iron deficiency are redundant or if there is cross-talk between these two pathways . Our data further indicate that increased mTORC1 activity in animals with a red cell specific TSC mutation can increase RBC size , similar to the effects of B12 and folate deficiency . Further studies may reveal whether a deficiency of these nutrients leads to mTORC1 activation or whether a different mechanism contributes . A surprising finding from our study was that S6K signaling in RBCs is rapamycin resistant yet remains sensitive to ATP competitive mTOR inhibitors . Countless studies have shown that rapamycin potently blocks S6 phosphorylation in diverse cellular contexts , and the recent mTOR crystal structure has elucidated the structural origin of this sensitivity ( Yang et al . , 2013 ) . The biochemical basis for the rapamycin resistance of RBCs is unclear , but it is intriguing that pharmacokinetic studies have reported that rapamycin preferentially accumulates in RBCs ( Yatscoff et al . , 1995; Yatscoff , 1996 ) . This has been proposed to result from the elevated expression of ( unidentified ) rapamycin binding proteins in the erythrocytes . It is thus possible that RBCs express an unusual complement of FKBPs that blunt rapamycin's ability to block mTORC1 signaling . This would provide an explanation for why the widespread clinical use of rapamycin as an immunosuppressant is not associated with a concomitant impairment of erythropoiesis . There is considerable interest in the potential of mTOR inhibitors for the treatment of cancer . While rapamycin analogs have been used clinically for more than a decade , the first ATP competitive mTOR inhibitors are currently in the early-stage clinical trials . A key question for these drugs is how their side-effect profile will compare to widely used rapalogs . Some studies have shown that ATP-competitive mTOR inhibitors actually have weaker effects on lymphocytes than rapamycin , suggesting that these drugs may be safer and less immunosuppressive ( Janes et al . , 2010 ) . Our data by contrast indicate that ATP competitive mTOR inhibitors block erythropoiesis much more potently than rapamycin , particularly under conditions of erythropoietic stress . Consistent with these data , anemia has been identified as a dose-limiting toxicity in some patients enrolled in the phase I trial of MLN0128 ( Infante et al . , 2012 ) . As impairment of erythropoiesis is a common side effect of many chemotherapies , it will be important to monitor how these drugs interact with ATP competitive mTOR inhibitors that potently block RBC function . These data also suggest the need for further studies of patients on these medications in clinical settings where hematopoiesis may be disrupted . The results reported here demonstrate how the transcriptional profiling of activated cells can reveal the cell types that respond to specific stimuli . A basic challenge for biology is to understand how complex physiologic processes emerge from the action of individual cell types , genes , and pathways . Advances in mouse genetics have made it possible to test hypotheses about the function of genes in specific tissues using methods such as Cre-lox recombination . Yet , the existence of thousands of distinct cell types in mammals suggests a need for new approaches to prioritize and guide this effort . In this paper and a companion report ( Knight et al . , 2012 ) , we have described a new way to map biochemical events onto cell types within a complex tissue . Our approach is based on capturing RNA from cells in proportion to a biochemical signal such as mTORC1 activation , and then using RNA sequencing to quantify this signal across numerous cell-type-specific marker genes in parallel . We have previously used this technique to identify neurons in the mouse brain that are activated during a behavior ( Knight et al . , 2012 ) . In this study , we describe how this same approach has revealed a critical role for mTORC1 signaling in a hematopoietic cell type that was only a trace contaminant in our original experiments . These studies illustrate how unbiased genomic approaches such as RNA sequencing can be powerfully repurposed to measure the functional activity of cell types in a complex tissue . We believe the broader application of these approaches has the potential to transform how biologists monitor and analyze complex physiologic systems . Vav-iCre ( #008610 ) , hCD2-iCre ( #008520 ) Tsc1fl/fl ( #005680 ) , and Rptorfl/fl ( #013188 ) mice were obtained from Jackson laboratory . Rabbit anti-pS6 235/236 ( Cell Signaling , Danvers MA #4858 , 1:1000 ) , rabbit anti-pS6 240/244 ( Cell Signaling #5364 , 1:1000 ) , mouse anti-total S6 ( Cell Signaling #2317 , 1:500 ) , rabbit anti-p4E-BP1 Ser65 ( Cell Signaling #9456 , 1:500 ) , rabbit anti-4E-BP1 rabbit anti-alpha globin ( Epitomics , Burlingame , CA #EPR3608 , 1:5000 ) , rabbit anti-neuron-specific enolase ( Immunostar , Hudson , WI #22521 , 1:100 ) , and HRP-conjugated rabbit anti-actin ( Cell Signaling #5125 , 1:2500 ) were used for immunoblotting . KU-0063794 was from Selleckchem ( Houston , TX ) , MLN0128 was from Active Biochem ( Maplewood , NJ ) , phenylhydrazine was from Sigma ( St . Louis , MO ) , and rapamycin was from LC Labs ( Woburn , MA ) . K562 cells were grown in RPMI supplemented with 10% dialyzed FBS . Cells were treated for 2 hr with either deferoxamine ( 30 μM ) , deferasirox ( 50 μM ) , or vehicle ( 0 . 05% DMSO ) . Cells were then collected by centrifugation , washed , and lysed in a 1% NP40 buffer . The hypothalamic ribosome profiling experiment that revealed enrichment of hba-a1 and hbb-b1 has previously been described ( Knight et al . , 2012 ) . For perfusion experiments , control mice were killed by cervical dislocation and the hypothalamus was dissected . Perfused mice were anesthetized with isoflurane , transcardially perfused with PBS for 5 min , and then the hypothalamus was dissected . Hypothalamic homogenates were prepared as previously described ( Knight et al . , 2012 ) , and total RNA was purified using the RNeasy Micro Kit . The indicated transcripts were then quantified by qPCR using internally quenched probes ( Integrated DNA Technologies , Coralville , IA ) with the Taqman Gene Expression Master Mix ( Applied Biosystems , Foster City , CA ) . Values were normalized to the abundance of rpl23 . Mice were maintained on either a standard chow diet containing 220 ppm iron ( Purina 5015 ) or a low iron diet containing 2–6 ppm iron ( Harlan , TD80396 ) . Mice were additionally given subcutaneous injections of deferoxamine ( 150 mg/kg ) or vehicle ( HBSS ) 5 days a week . Reticulocyte lysates for western blotting were generated by collecting blood in EDTA capillaries by cardiac puncture and diluting into HBSS + 20 mM EDTA . This blood was pelleted ( 3 min at 3000 rpm at 4°C ) , washed three times with HBSS/EDTA , and then subjected to hypotonic lysis on ice for 20 min with occasional mixing . This was then centrifuged ( 5 min at 3000 rpm at 4°C ) , and the supernatant was collected and used for western blotting . Freshly isolated spleens were mechanically dissociated and strained through a 70-μm strainer in the presence of cold staining buffer ( 0 . 2% BSA and 5 mM glucose in PBS ) . Bone marrow cells were isolated from the femur using cold buffer . Fetal liver cells were extracted from E14 . 5 embryos , dissociated mechanically by pipetting in cold buffer . All cells were washed twice and resuspended to approximately 1 . 5 × 106 cells/200 μl . Cells were then stained with PE-Cy-conjugated Ter119 and PE-conjugated CD71 at a concentration of 2 . 5 μg/ml for 20 min at 4°C in the dark . Cells were washed and resuspended in staining buffer ( 1 mL ) for flow cytometry . Control samples included unstained cells and single-color controls to calculate compensation . Cells were analyzed on an LSRII ( BD Biosciences , San Jose , CA ) flow cytometer and the data were analyzed with FlowJo software . To determine the percentage of reticulocytes in the blood , peripheral blood ( 5 μl ) was collected in EDTA-coated capillary tubes and diluted into 1 mL of BD Retic Count/Thiazole Orange Reagent ( BD #349204 ) . Cells were stained in the dark at room temperature for 30 min . Control ( unstained ) samples were diluted into 1 ml buffer ( 0 . 1% BSA and 1 mM EDTA ) . Samples were analyzed on a LSRII ( BD Biosciences ) flow cytometer and the data were analyzed with FlowJo software . Complete blood counts were performed on peripheral blood collected using EDTA-coated capillaries . Peripheral blood was diluted 10-fold into cold buffer ( HBSS containing 20 mM EDTA ) and subjected to automated analysis using a Bayer Advia120 hematology analyzer . Animals were given daily intraperitoneal injections of either rapamycin ( 10 mg/kg ) or twice daily injections of INK-128 ( 2 mg/kg ) or vehicle . All drugs were formulated in a solution of 5% PEG400 + 5% Tween80 in PBS and delivered in a total volume of 200 μl .

To multiply and grow , cells need to create more of the molecules—such as proteins—that make up their structure . This only happens if the cell has a good supply of the nutrients used to build the proteins . Red blood cells are particularly sensitive to the supply of nutrients , especially iron , which is a key component of the hemoglobin molecules that enable the cells to transport oxygen around the body . A lack of iron can lead to a shortage of red blood cells and a condition called anemia . People with mild forms of anemia may feel tired or weak , but more severe forms of anemia can cause heart problems and even death . A protein called mTOR forms part of a protein complex that helps alert the cells of many different organisms to the presence of nutrients . mTOR can add phosphate groups to ribosomes—the molecular machines that translate molecules of mRNA to build proteins . In 2012 , researchers developed a technique called Phospho-Trap that can isolate these phosphorylated ribosomes from cells . Cells with an activated mTOR complex express more mTOR protein and in turn have more ribosomes that are modified . Examining the mRNA molecules associated with these ribosomes can reveal which proteins are produced in greater amounts in these cells . Previous experiments using Phospho-Trap found the proteins that make up hemoglobin in unexpectedly high amounts in the mouse brain . Now , Knight et al . —and other researchers involved in the 2012 work—have established that the hemoglobin was not coming from the brain cells but from immature red blood cells circulating within the brain . These immature blood cells were found to have a highly active mTOR complex that promotes the production of hemoglobin and new blood cells . Using genetic techniques in mice , Knight et al . found that the mTOR complex can cause anemia if it is underactive or overactive . Underactive mTOR complexes cause a type of anemia that produces small red blood cells and is usually triggered by a lack of iron . This made sense because mTOR is known to regulate both protein production and cell size . Boosting the activity of the mTOR complex leads to a type of anemia in which the cells are much larger than normal , and which is normally associated with inadequate amounts of folate and B12 vitamins . When Knight et al . gave mice a drug that inhibits the mTOR protein , the mice developed anemia that resolved when the treatment stopped . However , mice that were given the mTOR inhibitor at the same time as a drug that destroys red blood cells , all died within days . Clinical trials are currently testing mTOR inhibitors as a possible cancer treatment; however , a common side effect of chemotherapy is that it stops new red blood cells being produced . Knight et al . suggest that the red blood cells of patients in these clinical trials must be closely monitored before deciding whether to continue the treatment further .

2014

A critical role for mTORC1 in erythropoiesis and anemia

During land plant evolution , determinate spore-bearing axes ( retained in extant bryophytes such as mosses ) were progressively transformed into indeterminate branching shoots with specialized reproductive axes that form flowers . The LEAFY transcription factor , which is required for the first zygotic cell division in mosses and primarily for floral meristem identity in flowering plants , may have facilitated developmental innovations during these transitions . Mapping the LEAFY evolutionary trajectory has been challenging , however , because there is no functional overlap between mosses and flowering plants , and no functional data from intervening lineages . Here , we report a transgenic analysis in the fern Ceratopteris richardii that reveals a role for LEAFY in maintaining cell divisions in the apical stem cells of both haploid and diploid phases of the lifecycle . These results support an evolutionary trajectory in which an ancestral LEAFY module that promotes cell proliferation was progressively co-opted , adapted and specialized as novel shoot developmental contexts emerged . Land plants are characterized by the alternation of haploid ( gametophyte ) and diploid ( sporophyte ) phases within their lifecycle , both of which are multicellular ( Niklas and Kutschera , 2010; Bowman et al . , 2016 ) . In the earliest diverging bryophyte lineages ( liverworts , mosses and hornworts ) the free-living indeterminate gametophyte predominates the lifecycle , producing gametes that fuse to form the sporophyte . The sporophyte embryo develops on the surface of the gametophyte , ultimately forming a simple determinate spore-producing axis ( Kato and Akiyama , 2005; Ligrone et al . , 2012 ) . By contrast , angiosperm ( flowering plant ) sporophytes range from small herbaceous to large arborescent forms , all developing from an indeterminate vegetative shoot apex that ultimately transitions to flowering , and gametophytes are few-celled determinate structures produced within flowers ( Schmidt et al . , 2015 ) . A series of developmental innovations during the course of land plant evolution thus simplified gametophyte form whilst increasing sporophyte complexity , with a prolonged and plastic phase of vegetative development arising in the sporophyte of all vascular plants ( lycophytes , ferns , gymnosperms and angiosperms ) . Studies aimed at understanding how gene function evolved to facilitate developmental innovations during land plant evolution have thus far largely relied on comparative analyses between bryophytes and angiosperms , lineages that diverged over 450 million years ago . Such comparisons have revealed examples of both sub- and neo-functionalization following gene duplication , and of co-option of existing gene regulatory networks into new developmental contexts . For example , a single bHLH transcription factor in the moss Physcomitrella patens regulates stomatal differentiation , whereas gene duplications have resulted in three homologs with sub-divided stomatal patterning roles in the angiosperm Arabidopsis thaliana ( hereafter ‘Arabidopsis’ ) ( MacAlister and Bergmann , 2011 ) ; class III HD-ZIP transcription factors play a conserved role in the regulation of leaf polarity in P . patens and Arabidopsis but gene family members have acquired regulatory activity in meristems of angiosperms ( Yip et al . , 2016 ) ; and the gene regulatory network that produces rhizoids on the gametophytes of both the moss P . patens and the liverwort Marchantia polymorpha has been co-opted to regulate root hair formation in Arabidopsis sporophytes ( Menand et al . , 2007; Pires et al . , 2013; Proust et al . , 2016 ) . In many cases , however , interpreting the evolutionary trajectory of gene function by comparing lineages as disparate as bryophytes and angiosperms has proved challenging , particularly when only a single representative gene remains in most extant taxa – as is the case for the LEAFY ( LFY ) gene family ( Himi et al . , 2001; Maizel et al . , 2005; Sayou et al . , 2014 ) . The LFY transcription factor , which is present across all extant land plant lineages and related streptophyte algae ( Sayou et al . , 2014 ) , has distinct functional roles in bryophytes and angiosperms . In P . patens , LFY regulates cell divisions during sporophyte development ( including the first division of the zygote ) ( Tanahashi et al . , 2005 ) , whereas in angiosperms the major role is to promote the transition from inflorescence to floral meristem identity ( Carpenter and Coen , 1990; Schultz , 1991; Weigel et al . , 1992; Blázquez et al . , 1997; Souer et al . , 1998; Molinero-Rosales et al . , 1999 ) . Given that LFY proteins from liverworts and all vascular plant lineages tested to date ( ferns , gymnosperms and angiosperms ) bind a conserved target DNA motif , whereas hornwort and moss homologs bind to different lineage-specific motifs ( Sayou et al . , 2014 ) , the divergent roles in mosses and angiosperms may have arisen through the activation of distinct networks of downstream targets . This suggestion is supported by the observation that PpLFY cannot complement loss-of-function lfy mutants in Arabidopsis ( Maizel et al . , 2005 ) . Similar complementation studies indicate progressive functional changes as vascular plant lineages diverged in that the lfy mutant is not complemented by lycophyte LFY proteins ( Yang et al . , 2017 ) but is partially and progressively complemented by fern and gymnosperm homologs ( Maizel et al . , 2005 ) . Because LFY proteins from ferns , gymnosperms and angiosperms recognize the same DNA motif , this progression likely reflects co-option of an ancestral LFY gene regulatory network into different developmental contexts . As such , the role in floral meristem identity in angiosperms would have been co-opted from an unknown ancestral context in non-flowering vascular plants , a context that cannot be predicted from existing bryophyte data . The role of LFY in non-flowering vascular plant lineages has thus far been hypothesized on the basis of expression patterns in the lycophyte Isoetes sinensis ( Yang et al . , 2017 ) , several gymnosperm species ( Mellerowicz et al . , 1998; Mouradov et al . , 1998; Shindo et al . , 2001; Carlsbecker et al . , 2004; Vázquez-Lobo et al . , 2007; Carlsbecker et al . , 2013 ) and the fern Ceratopteris richardii ( hereafter ‘Ceratopteris’ ) ( Himi et al . , 2001 ) , which has been used as a model of fern development for a number of years ( Hickok et al . , 1995 ) . These studies reported broad expression in vegetative and reproductive sporophyte tissues of I . sinensis and gymnosperms , and in both gametophytes and sporophytes of Ceratopteris . Although gene expression can be indicative of potential roles in each case , the possible evolutionary trajectories and differing ancestral functions proposed for LFY within the vascular plants ( Theissen and Melzer , 2007; Moyroud et al . , 2010 ) cannot be resolved without functional validation . Here we present a functional analysis in Ceratopteris that reveals a stem cell maintenance role for at least one of the two LFY homologs in both gametophyte and sporophyte shoots and discuss how that role informs our mechanistic understanding of developmental innovations during land plant evolution . The LFY gene family is present as a single gene copy in most land plant genomes ( Sayou et al . , 2014 ) . In this regard , the presence of two LFY genes in Ceratopteris ( Himi et al . , 2001 ) is atypical . To determine whether this gene duplication is more broadly represented within the ferns and related species ( hereafter ‘ferns’ ) , a previous amino acid alignment of LFY orthologs ( Sayou et al . , 2014 ) was pruned and supplemented with newly-available fern homologs ( see Materials and methods ) to create a dataset of 120 sequences , ~50% of which were from the fern lineage ( Supplementary file 1–3 ) . The phylogenetic topology inferred within the vascular plants using the entire dataset ( Figure 1—figure supplement 1 ) was consistent with previous analyses ( Qiu et al . , 2006; Wickett et al . , 2014 ) . Within the ferns ( 64 in total ) , phylogenetic relationships between LFY sequences indicated that the two gene copies identified in Equisetum arvense , Azolla caroliniana and Ceratopteris each resulted from recent independent duplication events ( Figure 1 ) . Gel blot analysis confirmed the presence of no more than two LFY genes in the Ceratopteris genome ( Figure 1—figure supplement 2 ) . Given that the topology of the tree excludes the possibility of a gene duplication prior to diversification of the ferns , CrLFY1 and CrLFY2 are equally orthologous to the single copy LFY representatives in other fern species . The presence of two LFY genes in the Ceratopteris genome raises the possibility that gene activity was neo- or sub-functionalized following duplication . To test this hypothesis , transcript accumulation patterns of CrLFY1 and CrLFY2 were investigated throughout the Ceratopteris lifecycle ( shown as a schematic in Figure 2 for reference ) . The developmental stages sampled spanned from imbibed spores prior to germination of the haploid gametophyte ( Figure 3A ) , to differentiated male and hermaphrodite gametophytes ( Figure 3B–D ) , through fertilization and formation of the diploid sporophyte embryo ( Figure 3E ) , to development of the increasingly complex sporophyte body plan ( Figure 3F–K ) . Quantitative real-time PCR ( qRT-PCR ) analysis detected transcripts of both CrLFY1 and CrLFY2 at all stages after spore germination , but only CrLFY2 transcripts were detected in spores prior to germination ( Figure 3L ) . A two-way ANOVA yielded a highly significant interaction ( F ( 10 , 22 ) = 14 . 21; p<0 . 0001 ) between gene copy and developmental stage that had not been reported in earlier studies ( Himi et al . , 2001 ) , and is indicative of differential gene expression between CrLFY1 and CrLFY2 that is dependent on developmental stage . Of particular note were significant differences between CrLFY1 and CrLFY2 transcript levels during sporophyte development ( Supplementary file 4 ) . Whereas CrLFY2 transcript levels were similar across sporophyte samples , CrLFY1 transcript levels were much higher in samples that contained the shoot apex ( Figure 3F , G ) than in those that contained just fronds ( Figure 3H–K ) . These data suggest that CrLFY1 and CrLFY2 genes may play divergent roles during sporophyte development , with CrLFY1 acting primarily in the shoot apex and CrLFY2 acting more generally . Functional characterization in P . patens previously demonstrated that PpLFY promotes cell divisions during early sporophyte development ( Tanahashi et al . , 2005 ) . To determine whether the spatial domains of CrLFY1 expression are consistent with a similar role in Ceratopteris embryo ( early sporophyte ) development , transgenic lines were generated that expressed the reporter gene B-glucuronidase ( GUS ) driven by a 3 . 9 kb fragment of the CrLFY1 promoter ( CrLFY1pro::GUS ) . This promoter fragment comprised genomic sequence encoding the entire published 5’UTR ( Himi et al . , 2001 ) plus a further 1910 bp upstream of the predicted transcription start site ( Figure 1—figure supplement 2A ) . In the absence of a genome sequence , repeated attempts to isolate an analogous fragment of CrLFY2 sequence were unsuccessful ( see Materials and methods for details ) . Construct maps plus DNA blot and PCR validation of transgenic lines are shown in Figure 4—figure supplements 1–4 . GUS activity was monitored in individuals from three independent transgenic lines , sampling both before and up to six days after fertilization ( Figure 4A–O ) , using wild-type individuals as negative controls ( Figure 4P–T ) and individuals from a transgenic line expressing GUS driven by the constitutive 35S promoter ( 35Spro ) as positive controls ( Figure 4U–Y ) . Notably , no GUS activity was detected in unfertilized archegonia of CrLFY1pro::GUS gametophytes ( Figure 4A , F , K ) but by two days after fertilization ( DAF ) GUS activity was detected in most cells of the early sporophyte embryo ( Figure 4B , G , L ) . At 4 DAF , activity was similarly detected in all visible embryo cells , including the embryonic frond , but not in the surrounding gametophytic tissue ( the calyptra ) ( Figure 4C , H , M ) . This embryo-wide pattern of GUS activity became restricted in the final stages of development such that by the end of embryogenesis ( 6 DAF ) GUS activity was predominantly localized in the newly-initiated shoot apex ( Figure 4D , E , I , J , N , O ) . Collectively , the GUS activity profiles indicate that CrLFY1 expression is induced following formation of the zygote , sustained in cells of the embryo that are actively dividing , and then restricted to the shoot apex at embryo maturity . This profile is consistent with the suggestion that CrLFY1 has retained the LFY role first identified in P . patens ( Tanahashi et al . , 2005 ) , namely to promote the development of a multicellular sporophyte , in part by facilitating the first cell division of the zygote . Both mosses and ferns form embryos , but moss sporophyte development is determinate post-embryogenesis ( Kato and Akiyama , 2005; Kofuji and Hasebe , 2014 ) whereas fern sporophytes are elaborated post-embryonically from indeterminate shoot apices ( Bierhorst , 1977; White and Turner , 1995 ) . The Ceratopteris shoot apex comprises a single apical cell that generates daughter cells through asymmetric divisions , and individual lateral organs ( fronds and root ) arise from their own apical cells specified within the grouped descendants of these daughter cells ( Hou and Hill , 2002; Hou and Hill , 2004 ) . CrLFY1pro::GUS expression in the shoot apex at the end of embryogenesis ( Figure 4E , J , O ) and elevated transcript levels in shoot apex-containing sporophyte tissues ( Figure 3L ) suggested an additional role for CrLFY1 relative to that seen in mosses , namely to promote proliferation in the indeterminate shoot apex . To monitor CrLFY1 expression patterns in post-embryonic sporophytes , GUS activity was assessed in CrLFY1pro::GUS lines at two stages of vegetative development ( Figure 5A–O ) and after the transition to reproductive frond formation ( Figure 5—figure supplement 1A–L ) . Wild-type individuals were used as negative controls ( Figure 5P–T; Figure 5—figure supplement 1M–P ) and 35Spro::GUS individuals as positive controls ( Figure 5U–Y; Figure 5—figure supplement 1Q–T ) . In young sporophytes ( 20 DAF ) , GUS activity was primarily localized in shoot apical tissues and newly-emerging frond primordia ( Figure 5A , F , K ) , with very little activity detected in the expanded simple fronds produced at this age ( Figure 5B , G , L ) . In older vegetative sporophytes ( 60 DAF ) , which develop complex dissected fronds ( Figure 5C , H , M ) , GUS activity was similarly localized in the shoot apex and young frond primordia in two out of the three fully characterized lines ( Figure 5D , I , N ) and in a total of 8 out of 11 lines screened ( from seven independent rounds of plant transformation ) . GUS activity was also detected in developing fronds in regions where the lamina was dividing to generate pinnae and pinnules ( Figure 5E , J , O ) . In some individuals GUS activity could be detected in frond tissues almost until maturity ( Figure 5C ) . Notably , patterns of CrLFY1pro::GUS expression were the same in the apex and complex fronds of shoots before ( 60 DAF ) ( Figure 5C–E , H–J , M–O ) and after ( ~115 DAF ) the reproductive transition ( Figure 5—figure supplement 1A–L ) . Consistent with a general role for CrLFY1 in promoting cell proliferation in the shoot , GUS activity was also detected in shoot apices that initiate de novo at the lamina margin between pinnae ( Figure 5Z–AD ) . Together these data support the hypothesis that LFY function was recruited to regulate cell division processes in the shoot when sporophytes evolved from determinate to indeterminate structures . To test the functional significance of CrLFY expression patterns , transgenic RNAi lines were generated in which one of four RNAi constructs targeted to CrLFY1 , CrLFY2 or both were expressed from the maize ubiquitin promoter ( ZmUbipro ) . Construct maps plus DNA blot and PCR validation of transgenic lines are shown in Figure 6—figure supplements 1–5 . Genotypic screening identified 10 lines which contained the complete transgene cassette and three lines that contained a fragment of the transgene which included the antibiotic resistance marker but not the RNAi hairpin ( Table 1 ) . In the no hairpin control ( NHC ) plants , post-embryonic shoot development initiated with the production of simple , spade-like fronds from the shoot apex ( Figure 6A ) as in wild type . In eight transgenic lines , sub-populations of sporophytes developed in which this early stage of sporophyte development was perturbed , one line ( E8 ) failing to initiate recognizable embryos ( Figure 6B ) and the remainder exhibiting premature shoot apex termination , typically after producing several distorted fronds ( Figure 6C–H ) . Sub-populations of phenotypically normal transgenic sporophytes were also identified in some of these lines ( Figure 6I–L ) . The two remaining lines exhibited less severe shoot phenotypes , one undergoing shoot termination after the production of simple ( B19a ) or lobed ( B19b ) fronds at the stage when control sporophytes produced complex dissected fronds ( Figure 6M–O ) , and the other ( D4b ) completing sporophyte development but reduced in size to approximately 50% of controls ( Figure 6P , Q ) . Despite the predicted sequence specificity of CrLFY1-i3 and CrLFY2-i4 ( Supplementary file 5 ) , qRT-PCR analysis found that all four RNAi constructs led to suppressed transcript levels of both CrLFY genes ( Figure 6R , S ) . The severity of the shoot phenotype was correlated with the level of endogenous CrLFY transcripts detected across all lines ( Figure 6R , S ) , with relative levels of both CrLFY1 and CrLFY2 significantly reduced compared to controls in all early-terminating sporophytes ( E8 , G13 , C3 , D2 , D4a , D13 , F9 ) ( p<0 . 01 or less ) . In phenotypically normal transgenic siblings CrLFY2 transcript levels were not significantly lower than controls ( indeed in line D2 , levels were higher p<0 . 01 ) whereas CrLFY1 levels were significantly reduced ( p<0 . 0001 ) , as in arrested siblings . Together , these data indicate that CrLFY2 can compensate for some loss of CrLFY1 , but at least 22% of CrLFY1 activity is required for normal development ( line D4a pn , Figure 6J , S ) . It can thus be concluded that CrLFY1 and CrLFY2 act partially redundantly to maintain indeterminacy of the shoot apex in Ceratopteris , a role not found in the early divergent bryophyte P . patens , nor known to be retained in the majority of later diverging flowering plants . In six of the RNAi lines that exhibited sporophyte developmental defects , it was notable that 50–99% of gametophytes arrested development prior to the sporophyte phase of the lifecycle ( Table 1 ) . This observation suggested that LFY plays a role in Ceratopteris gametophyte development , a function not previously demonstrated in either bryophytes or angiosperms . During wild-type development , the Ceratopteris gametophyte germinates from a single-celled haploid spore , establishing a single apical cell ( AC ) within the first few cell divisions ( Figure 7A ) . Divisions of the AC go on to form a two-dimensional photosynthetic thallus in both the hermaphrodite , where a notch meristem takes on growth ( Figure 7B ) , and male sexes ( Figure 7C ) ( Banks , 1999 ) . In contrast , the gametophytes from six RNAi lines ( carrying either ZmUbipro::CrLFY1-i3 , ZmUbipro::CrLFY1/2-i1 or ZmUbipro::CrLFY1/2-i2 ) exhibited developmental arrest ( Figure 7D–J ) , which in five lines clearly related to a failure of AC activity . The point at which AC arrest occurred varied , in the most severe line occurring prior to or during AC specification ( Figure 7D ) and in others during AC-driven thallus proliferation ( Figure 7E–I ) . Failure of AC activity was observed in both hermaphrodites ( Figure 7E ) and males ( Figure 7H , I ) . The phenotypically least-severe line exhibited hermaphrodite developmental arrest only after AC activity had been replaced by the notch meristem ( Figure 7J ) . A role for CrLFY in maintenance of gametophyte AC activity was supported by the detection of CrLFY transcripts in the AC and immediate daughter cells of wild-type gametophytes by in situ hybridization ( Figure 7K–N ) . By contrast CrLFY transcripts were not detected in arrested ZmUbipro::CrLFY1/2-i1 lines ( Figure 7O–R ) in which the presence of the transgene was confirmed by genotyping of individual arrested gametophytes ( Figure 7—figure supplement 1 ) . CrLFY1 and CrLFY2 transcripts could not be clearly distinguished in situ due to sequence similarity ( see Supplementary file 6 ) , and hence the observed phenotypes could not be ascribed to a specific gene copy . However , these data support a role for at least one CrLFY homolog in AC maintenance during gametophyte development , and thus invoke a role for LFY in the regulation of apical activity in both the sporophyte and gametophyte phases of vascular plant development . The results reported here reveal a role for LFY in the maintenance of apical cell activity throughout gametophyte and sporophyte shoot development in Ceratopteris . During sporophyte development , qRT-PCR and transgenic reporter lines demonstrated that CrLFY1 is preferentially expressed in the shoot apex ( whether formed during embryogenesis or de novo on fronds , and both before and after the reproductive transition ) ; in emerging lateral organ ( frond ) primordia; and in pinnae and pinnules as they form on dissected fronds ( Figures 3–5 ) . Notably , active cell division is the main feature in all of these contexts . CrLFY2 transcript levels were more uniform throughout sporophyte shoot development , in both dividing tissues and expanded fronds ( Figure 3 ) , and expression has previously been reported in roots ( Himi et al . , 2001 ) . Simultaneous suppression of CrLFY1 and CrLFY2 activity by RNAi resulted in developmental arrest of both gametophyte and sporophyte shoot apices , with any fronds produced before termination of the sporophyte apex exhibiting abnormal morphologies ( Figures 6 and 7 ) . The severity of phenotypic perturbations in sporophytes of transgenic lines correlated with combined CrLFY1 and CrLFY2 transcript levels , with wild-type levels of CrLFY2 able to fully compensate for up to a 70% reduction in CrLFY1 levels ( Figure 6 ) . The duplicate CrLFY genes therefore act at least partially redundantly during shoot development in Ceratopteris . A function for LFY in gametophyte development has not previously been reported in any land plant species . In the moss P . patens , PpLFY1 and PpLFY2 are expressed in both the main and lateral apices of gametophytic leafy shoots but double loss-offunction mutants develop normally , indicating that LFY is not necessary for maintenance of apical cell activity in the gametophyte ( Tanahashi et al . , 2005 ) . By contrast , loss of CrLFY expression from the gametophyte shoot apex results in loss of apical cell activity during thallus formation in Ceratopteris ( Figure 7 ) . The different DNA binding site preferences ( and hence downstream target sequences ) of PpLFY and CrLFY ( Sayou et al . , 2014 ) may be sufficient to explain the functional distinction in moss and fern gametophytes , but the conserved expression pattern is intriguing given that there should be no pressure to retain that pattern in P . patens in the absence of functional necessity . The thalloid gametophytes of the two other extant bryophyte lineages ( liverworts and hornworts ) resemble the fern gametophyte more closely than mosses ( Ligrone et al . , 2012 ) , but LFY function in these contexts is not yet known . Overall the data are consistent with the hypothesis that in the last common ancestor of ferns and angiosperms , LFY functioned to promote cell proliferation in the thalloid gametophyte , a role that has been lost in angiosperms where gametophytes have no apical cell and are instead just few-celled determinate structures . The range of reported roles for LFY in sporophyte development can be rationalized by hypothesizing three sequential changes in gene function during land plant evolution ( Figure 8 ) . First , the ancestral LFY function to promote early cell divisions in the embryo was retained in vascular plants after they diverged from the bryophytes , leading to conserved roles in P . patens ( Tanahashi et al . , 2005 ) and Ceratopteris ( Figure 4 ) . Second , within the vascular plants ( preceding divergence of the ferns ) this proliferative role expanded to maintain post-embryonic apical cell activity , and hence to enable indeterminate shoot growth . This is evidenced by CrLFY activity at the tips of shoots , fronds and pinnae ( Figures 4–6 ) , all of which develop from one or more apical cells ( Hill , 2001; Hou and Hill , 2004 ) . Whether fern fronds are homologous to shoots or to leaves in angiosperms is an area of debate ( Tomescu , 2009; Vasco et al . , 2013; Harrison and Morris , 2018 ) , but there are angiosperm examples of LFY function in the vegetative SAM ( Ahearn et al . , 2001; Zhao et al . , 2018 ) , axillary meristems ( Kanrar et al . , 2008; Rao et al . , 2008; Chahtane et al . , 2013 ) and in actively dividing regions of compound leaves ( Hofer et al . , 1997; Molinero-Rosales et al . , 1999; Champagne et al . , 2007; Wang et al . , 2008; Monniaux et al . , 2017 ) indicating that a proliferative role in vegetative tissues has been retained in at least some angiosperm species . Consistent with the suggestion that the angiosperm floral meristem represents a modified vegetative meristem ( Theiben et al . , 2016 ) , the third stage of LFY evolution could have been co-option and adaptation of this proliferation-promoting network into floral meristems , with subsequent restriction to just the flowering role in many species . This is consistent with multiple observations of LFY expression in both vegetative and reproductive shoots ( developing cones ) in gymnosperms ( Mellerowicz et al . , 1998; Mouradov et al . , 1998; Shindo et al . , 2001; Carlsbecker et al . , 2004; Vázquez-Lobo et al . , 2007; Carlsbecker et al . , 2013; Moyroud et al . , 2017 ) and suggests that pre-existing LFY-dependent vegetative gene networks might have been co-opted during the origin of specialized sporophyte reproductive axes in ancestral seed plants , prior to the divergence of angiosperms . The proposed evolutionary trajectory for LFY function bears some resemblance to that seen for KNOX protein function . Class I KNOX genes are key regulators of indeterminacy in the vegetative shoot apical meristem of angiosperms ( Gaillochet et al . , 2015 ) , and are required for compound leaf formation in both tomato and Cardamine hirsuta ( Bar and Ori , 2015 ) . In ferns , KNOX gene expression is observed both in the shoot apex and developing fronds ( Sano et al . , 2005; Ambrose and Vasco , 2016 ) , and in P . patens the genes regulate cell division patterns in the determinate sporophyte ( Sakakibara et al . , 2008 ) . It can thus be speculated that LFY and KNOX had overlapping functions in the sporophyte of the last common ancestor of land plants , but by the divergence of ancestral angiosperms from gymnosperms , KNOX genes had come to dominate in vegetative meristems whereas LFY became increasingly specialized for floral meristem function . Unlike LFY , however , there is not yet any evidence for KNOX function in the gametophyte of any land plant lineage , and thus if a pathway for regulating stem cell activity was co-opted from the gametophyte into the sporophyte , the LFY pathway is the more likely one . All experimental work was conducted using Ceratopteris richardii strain Hn-n ( Warne and Hickok , 1987 ) . Plant growth conditions for Ceratopteris transformation and DNA gel blot analysis of transgenic lines were as previously described ( Plackett et al . , 2015 ) . A dataset of 99 aligned LFY protein sequences from a broad range of streptophytes was first retrieved from Sayou et al . ( 2014 ) . The dataset was pruned and then supplemented with further sequences ( Supplementary file 1 ) to enable trees to be inferred that would ( i ) provide a more balanced distribution across the major plant groups and ( ii ) infer fern relationships . Only a subset of available angiosperm sequences was retained ( keeping both monocot and dicot representatives ) but protein sequences from other angiosperm species where function has been defined through loss-of-function analyses were added from NCBI – Antirrhinum majus FLO AAA62574 . 1 ( Coen et al . , 1990 ) , Pisum sativum UNI AAC49782 . 1 ( Hofer et al . , 1997 ) , Cucumis sativus CsLFY XP_004138016 . 1 ( Zhao et al . , 2018 ) , Medicago truncatula SGL1 AY928184 ( Wang et al . , 2008 ) , Petunia hybrida ALF AAC49912 . 1 ( Souer et al . , 1998 ) , Nicotiana tabacum NFL1 AAC48985 . 1 and NFL2 AAC48986 . 1 ( Kelly , 1995 ) , Eschscholzia californica EcFLO AAO49794 . 1 ( Busch and Gleissberg , 2003 ) , Gerbera hybrida cv . ‘Terraregina’ GhLFY ANS10152 . 1 ( Zhao et al . , 2016 ) , Lotus japonicus LjLFY AAX13294 . 1 ( Dong et al . , 2005 ) and Populus trichocarpa PTLF AAB51533 . 1 ( Rottmann et al . , 2000 ) . To provide better resolution within and between angiosperm clades , sequences from Spirodela polyrhiza ( 32G0007500 ) , Zostera marina ( 27g00160 . 1 ) , Aquilegia coerulea ( 5G327800 . 1 ) and Solanum tuberosum ( PGSC0003DMT400036749 ) were added from Phytozome v12 . 1 ( https://phytozome . jgi . doe . gov/pz/portal . html ) . Genome sequence from the early-diverging Eudicot Thalictrum thalictroides was searched by TBLASTX ( Altschul et al . , 1990 ) ( https://blast . ncbi . nlm . nih . gov/Blast . cgi ? PROGRAM=tblastx&PAGE_TYPE=BlastSearch&BLAST_SPEC=&LINK_LOC=blasttab ) with nucleotide sequence from the Arabidopsis LFY gene . A gene model was derived from sequence in two contigs ( 108877 and 116935 ) using Genewise ( Birney et al . , 2004 ) ( https://www . ebi . ac . uk/Tools/psa/genewise/ ) . Gymnosperm sequences were retained from Ginkgo biloba and from a subset of conifers included in Sayou et al . ( 2014 ) , whilst sequences from conifers where in situ hybridization patterns have been reported were added from NCBI – Pinus radiata PRFLL AAB51587 . 1 and NLY AAB68601 . 1 ( Mellerowicz et al . , 1998; Mouradov et al . , 1998 ) and Picea abies PaLFY AAV49504 . 1 and PaNLY AAV49503 . 1 ( Carlsbecker et al . , 2004 ) . Fern sequences were retained except Angiopteris spp sequences which consistently disrupted the topology of the tree by grouping with gymnosperms . To better resolve relationships within the ferns , additional sequences were identified in both NCBI and 1KP ( Matasci et al . , 2014 ) databases . The protein sequence from Matteuccia struthiopteris AAF77608 . 1 MatstFLO ( Himi et al . , 2001 ) was retrieved from NCBI . Further sequences from horsetails ( 2 ) , plus eusporangiate ( 1 ) and leptosporangiate ( 53 ) ferns were retrieved from the 1KP database ( https://db . cngb . org/blast/ ) using BLASTP and the MatstFLO sequence as a query . Lycophyte and bryophyte sequences were all retained , but the liverwort Marchantia polymorpha predicted ORF sequence was updated from Phytozome v12 . 1 ( Mpo0113s0034 . 1 . p ) , the hornwort Nothoceros genome scaffold was replaced with a translated full length cDNA sequence ( AHJ90704 . 1 ) from NCBI and two additional lycophyte sequences were added from the 1KP dataset ( Isoetes tegetiformans scaffold 2013584 and Selaginella kraussiana scaffold 2008343 ) . All of the charophyte scaffold sequences were substituted with Coleochaete scutata ( AHJ90705 . 1 ) and Klebsormidium subtile ( AHJ90707 . 1 ) translated full-length cDNAs from NCBI . The new/replacement sequences were trimmed and amino acids aligned to the existing alignment from Sayou et al . ( 2014 ) using CLUSTALW ( Li et al . , 2015 ) ( Supplementary file 2 and 3 ) . The best-fitting model parameters ( JTT + I + G4 ) were estimated and consensus phylogenetic trees were run using Maximum Likelihood from 1000 bootstrap replicates , using IQTREE ( Nguyen et al . , 2015 ) . Two trees were inferred . The first contained only a subset of fern and allied sequences to achieve a more balanced distribution across the major plant groups ( 81 sequences in total ) ( Figure 8 ) , whereas the second used the entire dataset ( 120 sequences ~ 50% of which are fern and allied sequences – Figure 1—figure supplement 1 ) . The data were imported into ITOL ( Letunic and Bork , 2016 ) to generate the pictorial representations . All branches with less than 50% bootstrap support were collapsed . Relationships within the ferns ( Figure 1 ) were represented by pruning the lycophyte and fern sequences ( 68 in total ) from the tree containing all available fern sequences ( Figure 1—figure supplement 1 ) . Because no reference genome has yet been established for Ceratopteris ( or any fern ) , CrLFY copy number was quantified by DNA gel blot analysis . Ceratopteris genomic DNA was hybridized using both the highly conserved LFY DNA-binding domain diagnostic of the LFY gene family ( Maizel et al . , 2005 ) and also gene copy-specific sequences ( Figure 1—figure supplement 2 ) . CrLFY1 and CrLFY2 share 85% amino acid similarity , compared to 65% and 44% similarity of each to AtLFY . DNA gel blotting and hybridization was performed as described previously ( Plackett et al . , 2014 ) . The results supported the presence of only two copies of LFY within the Ceratopteris genome . All primers used in probe preparation are supplied in the Key Resources Table . Genomic sequences for CrLFY1 and CrLFY2 open reading frames ( ORFs ) were amplified by PCR from wild-type genomic DNA using primers designed against published transcript sequences ( Himi et al . , 2001 ) . ORFs of 1551 bp and 2108 bp were obtained , respectively ( Figure 1—figure supplement 2 ) . Exon structure was determined by comparison between genomic and transcript sequences . The native promoter region of CrLFY1 was amplified from genomic template by sequential rounds of inverse PCR with initial primer pairs designed against published CrLFY1 5’UTR sequence and additional primers subsequently designed against additional contiguous sequence that was retrieved . A 3 . 9 kb contiguous promoter fragment was isolated for CrLFY1 containing the entire published 5’UTR and 1 . 9 kb of additional upstream sequence ( Figure 1—figure supplement 2 ) . Repeated attempts were made to obtain a CrLFY2 promoter fragment but this proved impossible in the absence of a reference genome . Some sequence contiguous with the CrLFY2 ORF was obtained by inverse PCR using primers designed against the previously published 5’UTR sequence of the CrLFY2 transcript ( Himi et al . , 2001 ) . This sequence was extended to 1016 bp in length using additional primers against the isolated genomic sequence but this fragment did not contain the entire published 5’UTR . Numerous rounds of inverse PCR generated a second 3619 bp genomic fragment containing sequence identical to the remaining 5’UTR ( see Figure 1—figure supplement 2 , Supplementary file 7 ) but the presumed connecting sequence between these two fragments could not be amplified despite many attempts . It was eventually concluded that either the intervening promoter fragment was too long to amplify or that it was too GC rich for amplification . All primers used in ORF amplification and inverse PCR are supplied in the Key Resources table . The contiguous sequences obtained for the CrLFY1 and CrLFY2 genomic loci have been submitted to Genbank ( accessions MH841970 and MH841971 , respectively ) . RNA was extracted from Ceratopteris tissues using the Spectrum Total Plant RNA kit ( Sigma-Aldrich , St . Louis , MO ) and 480 ng were used as template in iScript cDNA synthesis ( Bio-Rad ) . CrLFY1 and CrLFY2 locus-specific qRT-PCR primers were designed spanning intron 1 . Amplification specificity of primers was validated via PCR followed by sequencing . qRT-PCR of three biological replicates and three technical replicates each was performed in a Bio-Rad CFX Connect with iTaq Universal SYBR Green Supermix ( Bio-Rad , Hercules , CA ) . Primer amplification efficiency was checked with a cDNA serial dilution . Efficiency was determined using the slope of the linear regression line as calculated by Bio-Rad CFX Connect software . Primer specificity was tested via melting curve analysis , resulting in a single peak per primer set . CrLFY expression was calculated using the 2- ΔΔCt method ( Livak and Schmittgen , 2001 ) and normalized against the geometric mean of the expression of two endogenous reference genes ( Hellemans et al . , 2007 ) , CrACTIN1 and CrTATA-BINDING PROTEIN ( TBP ) ( Ganger et al . , 2015 ) . The standard deviation of the Ct values of each reference gene was calculated to ensure minimal variation ( <3% ) in gene expression . Error bars represent ± the standard error of the mean of the 2 ΔΔ Ct values . All primers used in qRT-PCR are supplied in the Key Resources table . Relative expression values of CrLFY from qRT-PCR were compared by one or two-way analysis of variance ( ANOVA ) for developmental stages followed by Tukey’s or Sidak’s multiple comparisons , respectively . To test whether genes were downregulated in transgenic RNAi lines , two-way ANOVA was perfomed with gene ( CrLFY1 or CrLFY2 ) and transgenic line as factors , with ‘gene’ as a repeated factor when all transgenic lines had the same number of replicates . Where appropriate , expression of each gene in each line was compared to the expression of the respective control by Dunnet comparisons . Control plants had been transformed and were hygromycin-resistant , but did not contain the RNAi hairpin that triggers gene silencing ( non-hairpin controls , NHC ) . For all experiments , NHCs were grown alongside transgenic lines . qRT-PCR of transgenic lines was necessarily conducted across several plates , each including a representative NHC , and statistical comparisons were performed within each plate relative to its respective control . The significance threshold ( p ) was set at 0 . 05 . All statistical analyses were performed in Prism v . 6 . 0 ( GraphPad Software , Inc . , La Jolla , CA ) . The CrLFY1pro::GUS reporter construct ( Figure 4—figure supplement 1 ) was created by cloning the CrLFY1 promoter into pART7 as a NotI-XbaI restriction fragment , replacing the existing 35S promoter . A β-Glucuronidase ( GUS ) coding sequence ( Ulmasov , 1997 ) was cloned downstream of pCrLFY1 as an XbaI-XbaI fragment . The same GUS XbaI-XbaI fragment was also cloned into pART7 to create a 35Spro::GUS positive control ( Figure 4—figure supplement 4 ) . The resulting CrLFY1pro::GUS::ocs and 35Spro::GUS::ocs cassettes were each cloned as NotI-NotI fragments into the pART27-based binary transformation vector pBOMBER carrying a hygromycin resistance marker previously optimized for Ceratopteris transformation ( Plackett et al . , 2015 ) . All primers used in GUS reporter component amplification are supplied in the Key Resources table . RNAi constructs were designed and constructed using the pANDA RNAi expression system ( Miki and Shimamoto , 2004 ) . Four RNAi fragments were designed , two targeting a conserved region of the CrLFY1 and CrLFY2 coding sequence ( 77% nucleotide identity ) using sequences from either CrLFY1 ( CrLFY1/2-i1 ) or CrLFY2 ( CrLFY1/2-i2 ) , and two targeting gene-specific sequence within the 3’UTR of CrLFY1 ( CrLFY1-i3 ) or CrLFY2 ( CrLFY2-i4 ) ( Figure 6—figure supplement 1 ) . Target fragments were amplified from cDNA and cloned into Gateway-compatible entry vector pDONR207 ( Invitrogen , Carlsbad , CA ) . Each sequence was then recombined into the pANDA expression vector via Gateway LR cloning ( Invitrogen , Carlsbad , CA ) . All primers used in RNAi target fragment amplification are supplied in the Key Resources table . Transformation of all transgenes into wild-type Hn-n Ceratopteris callus was performed as previously described ( Plackett et al . , 2015 ) . T0 sporophyte shoots were regenerated from transformed callus tissue , with each round of transformation using multiple separate pieces of callus as starting material . Transgenic T1 spores were harvested from these T0 shoots , germinated to form T1 gametophytes and then self-fertilized to produce T1 sporophytes . T1 sporophytes were assessed for T-DNA copy number by DNA gel blot analysis ( Figure 4—figure supplement 2; Figure 6—figure supplement 3 ) and the presence of full-length T-DNA insertions was confirmed through genotyping PCR ( Figure 4—figure supplements 3 and 4 ) . All primers used in genotyping reactions are supplied in the Key Resources table . For characterization of RNAi lines , T2 spores were collected from individuals that either contained the full transgene construct or from segregants in which the RNAi hairpin was absent . GUS activity analysis in CrLFY1pro::GUS transgenic lines was conducted in the T1 generation . GUS staining was conducted as described previously ( Plackett et al . , 2014 ) . Optimum staining conditions ( 1 mg/ml X-GlcA , 5 μM potassium ferricyanide ) were determined empirically . Tissue was cleared with sequential incubations in 70% ethanol until no further decolorization occurred . GUS-stained gametophytes were imaged with a Zeiss Axioplan microscope and GUS-stained sporophytes imaged with a dissecting microscope , both mounted with Q-imaging Micro-published 3 . 3 RTV cameras . Images were minimally processed for brightness and contrast in Photoshop ( CS4 ) . Phenotypic characterization of RNAi transgenic lines was conducted in the T2 or T3 generation . Isogenic lines were obtained by isolating hermaphrodite gametophytes in individual wells at approximately 7 DPS ( or when the notch became visible , whichever came first ) and flooding them once they had developed mature gametangia ( at approximately 9 DPS ) . All transgenic lines were grown alongside both wild-type and no hairpin controls , and phenotypes observed and recorded daily . Gametophytes exhibiting altered phenotypes were imaged at approximately 10 DPS with a Nikon Microphot-FX microscope . Sporophytes with abnormal phenotypes were imaged with a dissecting microscope . Antisense and sense RNA probes for CrLFY1 and CrLFY2 were amplified and cloned into pCR 4-TOPO ( Invitrogen ) and DIG-labelled according to the manufacturer’s instructions ( Roche , Indianapolis , IN ) . Probes were designed to include the 5’UTR and ORF ( CrLFY1 521 bp 5’UTR and 1113 bp ORF; CrLFY2 301 bp 5’UTR and 1185 bp ORF ) ( Supplementary file 6 ) . All primers used in in situ probe amplification are supplied in the Key Resources table . We were unable to identify fragments that distinguished the two genes in whole mount in situ hybridizations . Tissue was fixed in FAA ( 3 . 7% formaldehyde , 5% acetic acid; 50% ethanol ) for 1–4 hr and then stored in 70% ethanol . Whole mount in situ hybridization was carried out based on Hejátko et al . ( 2006 ) , with the following modifications: hybridization and wash steps were carried out in 24-well plates with custom-made transfer baskets ( 0 . 5 mL microcentrifuge tubes and 30 µm nylon mesh , Small Parts Inc . , Logansport , IN ) . Permeabilization and post-fixation steps were omitted depending on tissue type to avoid damaging fragile gametophytes , Acetic Anhydride ( Sigma-Aldrich ) and 0 . 5% Blocking Reagent ( Roche ) washing steps were added to decrease background staining , and tissue was hybridized at 45°C . Photos were taken under bright-field with a Q-imaging Micro-publisher 3 . 3 RTV camera mounted on a Nikon Microphot-FX microscope . Images were minimally processed for brightness and contrast in Photoshop ( CS4 ) .

The first plants colonized land around 500 million years ago . These plants had simple shoots with no branches , similar to the mosses that live today . Later on , some plants evolved more complex structures including branched shoots and flowers ( collectively known as the “flowering plants” ) . Ferns are a group of plants that evolved midway between the mosses and flowering plants and have branched shoots but no flowers . The gradual transition from simple to more complex plant structures required changes to the way in which cells divide and grow within plant shoots . Whereas animals produce new cells throughout their body , most plant cells divide in areas known as meristems . All plants grow from embryos , which contain meristems that will form the roots and shoots of the mature plant . A gene called LEAFY is required for cells in moss embryos to divide . However , in flowering plants LEAFY does not carry out this role , instead it is only required to make the meristems that produce flowers . How did LEAFY transition from a general role in embryos to a more specialized role in making flowers ? To address this question , Plackett , Conway et al . studied the two LEAFY genes in a fern called Ceratopteris richardii . The experiments showed that at least one of these LEAFY genes was active in the meristems of fern shoots throughout the lifespan of the plant . The shoots of ferns with less active LEAFY genes could not form the leaves seen in normal C . richardii plants . This suggests that as land plants evolved , the role of LEAFY changed from forming embryos to forming complex shoot structures . Most of our major crops are flowering plants . By understanding how the role of LEAFY has changed over the evolution of land plants , it might be possible to manipulate LEAFY genes in crop plants to alter shoot structures to better suit specific environments .

2018

LEAFY maintains apical stem cell activity during shoot development in the fern Ceratopteris richardii 

Longevity mechanisms increase lifespan by counteracting the effects of aging . However , whether longevity mechanisms counteract the effects of aging continually throughout life , or whether they act during specific periods of life , preventing changes that precede mortality is unclear . Here , we uncover transcriptional drift , a phenomenon that describes how aging causes genes within functional groups to change expression in opposing directions . These changes cause a transcriptome-wide loss in mRNA stoichiometry and loss of co-expression patterns in aging animals , as compared to young adults . Using Caenorhabditis elegans as a model , we show that extending lifespan by inhibiting serotonergic signals by the antidepressant mianserin attenuates transcriptional drift , allowing the preservation of a younger transcriptome into an older age . Our data are consistent with a model in which inhibition of serotonergic signals slows age-dependent physiological decline and the associated rise in mortality levels exclusively in young adults , thereby postponing the onset of major mortality . The most widely used standard to measure aging of an organism is the quantification of lifespan ( Partridge and Gems , 2007 ) . Lifespan relates to aging , as the latter causes the degeneration of tissues and organs , thereby increasing mortality due to systemic functional tissue failure ( Balch et al . , 2008; Bishop et al . , 2010; David et al . , 2010; Haigis and Sweet-Cordero , 2011; Taylor and Dillin , 2011; Gladyshev , 2013; Burkewitz et al . , 2015; Currais , 2015 ) . Several genetic and pharmacological strategies have been shown to prolong the lifespan of various organisms , including C . elegans ( Kenyon et al . , 1993; Kaeberlein et al . , 1999; Curran and Ruvkun , 2007; Evason et al . , 2008; Onken and Driscoll , 2010; Alavez et al . , 2011; Chin et al . , 2014; Ye et al . , 2014; Tatum et al . , 2015 ) . Mutations in age-1 or daf-2 , for example , slow degenerative processes occurring throughout life , thereby constantly lowering mortality rates ( Johnson , 1990; Kenyon et al . , 1993; Taylor et al . , 2014 ) . Age-associated degenerative processes such as a decline in proteostatic capacity are not necessarily restricted to older organisms but can also be observed in young adults ( Labbadia and Morimoto , 2015a; 2015b ) . This raises the possibility of degenerative processes that occur only in young adults and thus specifically contribute to the rise of mortality during young adulthood . Any longevity mechanisms preventing such a degenerative process would specifically slow mortality rates during the period of young adulthood , effectively prolonging its duration to postpone the onset of major age-associated mortality around midlife ( Bartke , 2015 ) . However , to identify such mechanisms would require mortality-independent metrics of age-associated change , as age-associated mortality rates during young adulthood are difficult to determine by demographic analysis against the back drop of non-aging-related death events ( Partridge and Gems , 2007; Beltran-Sancheza et al . , 2012 ) . In C . elegans , mortality-independent metrics of aging include age-associated decline of various behaviors or physiological parameters such as movement or stress resistance ( Huang et al . , 2004; Bansal et al . , 2015 ) . Molecular markers of aging include sets of genes whose expression change with age , such as micro-RNAs , electron transport chain ( ETC ) components , or genes involved in posttranslational modifications such as methylation ( Budovskaya et al . , 2008; de Magalhaes et al . , 2009; Pincus et al . , 2011; Horvath et al . , 2015 ) . However , aging also increases DNA damage , affects nuclear architecture , chromatin complexes , chromatin modifications , and the transcriptional machinery ( Mostoslavsky et al . , 2006; Scaffidi and Misteli , 2006; Feser et al . , 2010; Greer et al . , 2011; Maures et al . , 2011; Fushan et al . , 2015 ) . Therefore , an emerging alternative approach to measure specific gene expression changes with age is to quantify the progressive imbalance in gene expression patterns as a function of age . Two such approaches , one measuring transcriptional noise , the cell-to-cell variation in gene expression , and the other measuring decreasing correlation in the expression of genetic modules , showed a loss of co-expression patterns with age ( Bahar et al . , 2006; Southworth et al . , 2009 ) . These studies suggest that age-associated changes can be measured independently from mortality by tracking the loss of gene expression patterns that are observed in young animals . In the present study , we set out to investigate the mechanisms by which the atypical antidepressant mianserin extends lifespan by recording the transcriptional dynamics of mianserin-treated and untreated C . elegans across different ages . These studies revealed that aging causes transcriptional drift , an evolutionarily conserved phenomenon in which the expression of genes change in opposing directions within functional groups . These changes cause a transcriptome-wide loss in mRNA stoichiometry and loss of co-expression patterns in aging animals , as compared to young adults . Mianserin treatment reduced age-associated transcriptional drift across ~80% of the transcriptome , preserving many characteristics of transcriptomes of younger animals . We used transcriptional drift along with mortality analysis as metrics to monitor aging and find that mianserin treatment extended lifespan by exclusively slowing age-associated changes in young adults , thereby postponing the onset of mortality . To better understand how aging changes gene expression patterns in a eukaryotic organism , and how these changes are affected by longevity , we measured gene expression changes in mianserin-treated or untreated C . elegans by RNA-sequencing ( RNA-seq; Figure 1a ) . Cohort #1 was a time series to study how gene expression patterns change over time in control ( water ) animals or in animals treated with mianserin on day 1 of adulthood ( 24 hr after L4 stage ) . Cohort #2 was designed to study dosage effects of increasing concentrations of mianserin with aging , and cohort #3 was designed to study the effects of delayed mianserin-treatment of worms treated at day 3 or 5 of adulthood ( Figure 1a ) . Lifespan of a sub-population of each cohort was simultaneously assessed to ensure the effect of mianserin . 10 . 7554/eLife . 08833 . 003Figure 1 . Transcriptional drift-variance increases with age . ( a ) Schematic of RNA-seq experiment . In cohort #1 , water or mianserin was added on day 1 of adulthood and RNA samples were harvested on day 1 ( water only ) , day 3 ( d3 ) , day 5 ( d5 ) and day 10 ( d10 ) . In cohort #2 , animals were treated with water or increasing concentrations of mianserin ( 2 , 10 or 50 µM ) on day 1 ( d1 ) and RNA was harvested on day 5 ( d5 ) for RNA-seq . In cohort #3 , water or 50 µM mianserin was added on day 1 , day 3 , and day 5 , and RNA was harvested on day 10 ( d10 ) for RNA-seq . ( b ) Venn diagrams of the number of GOs enriched for genes that decrease expression with mianserin ( down , dark blue circle ) increase expression with mianserin ( up , light blue circle ) or are enriched for both ( intersection ) . ( c ) Venn diagrams of the number of GOs enriched for genes that decrease expression with age ( down , gray circle ) increase expression with age ( up , white circle ) or are enriched for both ( intersection ) . ( d ) Heat map depicting log2 changes in gene expression for oxidative stress genes elicited by increasing concentrations of mianserin ( yellow , increased expression; blue , decreased expression ) ( e ) Mianserin decreases expression of redox genes that increase with age and increases expression of genes that decrease with age . ( f ) Mianserin reverts age-associated changes on the level of GOs . Venn diagrams of the number of GOs enriched for genes that decrease expression with mianserin ( down , dark blue circle ) and increase with age ( up , white circle ) or vice versa ( down with age , gray circle; up with mianserin , light blue circle ) . ( g ) Mianserin reverts age-associated changes on the level of individual genes . Volcano plot shows the negative log10 of P-values as a function of log2 fold changes of 3 , 367 genes that significantly change expression from day 1 to day 3 in samples of water-treated control animals ( black ) or samples from age-matched mianserin-treated animals ( 50 µM , blue ) . As animals age , gene expression levels change ( “drift” ) away from levels observed in young adults ( yellow line ) . Mianserin treatment attenuates age-associated gene expression changes preserving expression levels as seen in young adults . ( h ) Drift-plot shows log fold change ( old/young ) as a function of age for each gene involved in oxidative phosphorylation ( gray lines . KEGG: cel 04142 ) . Superimposed are Tukey-style box-plots to graph the increases in drift-variance across the entire pathway . Gene expression changes are classified into type I , which describes activation or repression of the entire pathway and into type II , which describes changes among genes relative to each other ( drift-variances ) , see red arrows . ( i ) Drift-plot for lysosomal genes ( KEGG: cel 00190 ) . See Figure 1—source data 1–5 , Figure 1—figure supplement 1 and Table 1 for additional information on data-sets . Also see Methods section for transcriptional drift calculation in each figure panel . DOI: http://dx . doi . org/10 . 7554/eLife . 08833 . 003 10 . 7554/eLife . 08833 . 004Figure 1—source data 1 . RNA-seq gene expression data . DOI: http://dx . doi . org/10 . 7554/eLife . 08833 . 004 10 . 7554/eLife . 08833 . 005Figure 1—source data 2 . Gene ontologies changing in response to mianserin treatment . DOI: http://dx . doi . org/10 . 7554/eLife . 08833 . 005 10 . 7554/eLife . 08833 . 006Figure 1—source data 3 . Gene ontologies changing in response to age . DOI: http://dx . doi . org/10 . 7554/eLife . 08833 . 006 10 . 7554/eLife . 08833 . 007Figure 1—source data 4 . Differentially expressed genes in response to age . DOI: http://dx . doi . org/10 . 7554/eLife . 08833 . 007 10 . 7554/eLife . 08833 . 008Figure 1—source data 5 . Differentially expressed genes in response to mianserin treatment . DOI: http://dx . doi . org/10 . 7554/eLife . 08833 . 008 10 . 7554/eLife . 08833 . 009Figure 1—figure supplement 1 . This figure relates to Figure 1 in main text . Expression patterns of GO annotations are disrupted with age . Representative pie charts show a cross-section of 50 out of 249 GO annotations enriched for genes that change in opposing direction as animals age ( day 3 , 5 , and 10 ) . The fraction of genes whose expression increase with age ( yellow ) , the fraction of genes whose expression decrease with age ( black ) , and the fraction of genes that maintain the expression seen in young day 1 adults ( white ) are shown . GOs are sorted and represented in the figure , starting with GOs that show the least disruption in the upper left , and the GO’s with the most extreme changes in the lower right . As animals’ age progresses from day 3 , 5 to 10 , more and more genes change expression in opposing directions disrupting the transcriptional stoichiometry observed in young day 1 animals . None of these 50 pie charts , as is , allows any statements on how the functional states of the physiological processes they represent change with age . The GO names and number of genes ( n ) belonging to each GO are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 08833 . 009 Comparison of gene expression profiles of age-matched mianserin-treated and untreated controls , showed that approximately 3 , 000–6 , 000 genes changed with age in response to mianserin treatment ( FDR<0 . 1 , Figure 1—source data 1 ) ( Robinson and Oshlack , 2010; Kim et al . , 2013; Lawrence et al . , 2013 ) . We separated genes into sets that showed increased or decreased expression in response to mianserin , to conduct gene-set enrichment analysis . This revealed hundreds of gene ontologies ( GO ) that changed in response to mianserin ( Figure 1—source data 2 ) ( Ashburner et al . , 2000; Mi et al . , 2005 ) . We observed that many GOs were enriched for both , genes that increased as well as decreased as a consequence of aging . This observation complicated any interpretation on whether pathways were activated or inhibited in response to mianserin , and how the associated function ( GO ) relates to mianserin-induced lifespan extension ( Figure 1b ) . We observed a similar scenario by conducting gene-set enrichment analysis for gene expression changes in response to age in untreated animals . As seen with mianserin , many GO annotations were enriched for both up- as well as downregulated genes at any given age ( Figure 1c; Figure 1—source data 3 , 4 ) , making it difficult to interpret whether those pathways are being activated or inhibited with age . We generated 50 representative pie charts out of the 249 GO annotations that contained genes that increased or decreased in expression by day 10 due to aging . These charts suggested that as animals age and become older , genes change expression in opposing directions , disrupting relative mRNA ratios within the GO , when compared to young adults ( Figure 1—figure supplement 1 ) . Thus , aging changed the stoichiometric relationship between mRNAs belonging to the same functional group ( GO ) . In many cases , the fractions of genes that increased , decreased or did not change in expression showed no consistent pattern , nor provided any insight into the pathway activity ( Figure 1—figure supplement 1 ) . Because the expression patterns observed in many GOs were difficult to interpret in terms of functional change , we turned to investigate expression changes in the superoxide detoxification pathway , a well-defined cellular function that declines with age ( Ashburner et al . , 2000; Mi et al . , 2005; Kumsta et al . , 2011; Bansal et al . , 2015; Rangaraju et al . , 2015a ) . As expected from our previous studies ( Rangaraju et al . , 2015a ) , the expression levels of some superoxide detoxification genes were higher in mianserin-treated animals compared to age-matched controls ( Figure 1d ) . Exceptions were the expression levels of sod-4 and sod-5 , which were lowered upon mianserin treatment ( Figure 1d ) . However , plotting expression changes of superoxide detoxification genes as a function of age ( Figure 1e , left panel ) revealed again a scenario in which genes changed in opposing directions as seen in the pie charts for many GOs before ( Figure 1—figure supplement 1 ) . Some mRNAs including those of sod-4 , -5 increased with age , while some decreased ( sod-1 , -2 , prdx-2 , 3 , 6 ) and some did not change ( ctl-1 , 2 , 3 ) , leading to an overall 5-10-fold change in stoichiometric balance among superoxide detoxification-associated mRNAs by day 5 ( Figure 1e , left panel ) . More interestingly , if the expression of an sod increased with age , mianserin treatment prevented the increase and if the expression of an sod decreased with age , mianserin prevented the decrease ( Figure 1e , right panel ) . Thus , when we took the mRNA expression levels of young animals into account , the emerging picture suggested that mianserin treatment attenuated age-associated gene expression changes . We therefore asked whether the complex gene-set enrichment patterns observed comparing mianserin-treated and untreated samples ( Figure 1b , c ) could be explained by mianserin preventing expression changes due to age . Indeed , many GO annotations that increased expression with age were decreased by mianserin treatment and vice versa ( Figure 1f ) . This attenuation of age-associated changes by mianserin treatment was even more pronounced for individual genes ( Figure 1g ) . Analyzing cohort #1 showed a significant change in expression levels of 3 , 367 genes , as the animals aged from day 1 to day 3 , and a change in 5 , 947 genes from day 1 to day 10 ( FDR < 0 . 1 ) ( Figure 1g , significant genes only ) . Mianserin treatment reduced these age-associated expression changes in over 90% of cases . Including all age-associated expression changes for the 19 , 196 different transcripts present in our data-set , we found that mianserin treatment attenuated age-associated changes in transcription in 15 , 095 out of 19 , 169 genes ( 80% , binomial P < 10–100 ) . Thus , most of the changes observed between mianserin-treated and untreated animals are due to mianserin preventing transcriptional changes with age . When we excluded all genes that changed due to age and were attenuated by mianserin , we obtained a much smaller gene-set consisting of mianserin-induced changes that was enriched for GOs related to stress , xenobiotic and immune-responses , as well as genes associated with aging and the determination of lifespan ( Table 1 , Figure 1—source data 5 ) . These GOs have been previously shown to be regulated by serotonin in C . elegans with the exception of the xenobiotic response ( Zhang et al . , 2005; Petrascheck et al . , 2007; Rangaraju et al . , 2015a ) . Thus , accounting for age-associated transcriptional changes dramatically simplified a seemingly very complex gene-expression pattern ( Figure 1b , c ) . It revealed that mianserin affected expression of a small set of physiological functions that are known to be regulated by serotonin and have been shown to be required for mianserin-induced lifespan extension or for aging in general ( Garsin et al . , 2003; Rangaraju , et al . , 2015; Petrascheck , et al . , 2007 ) ( Table 1; Figure 1f; Figure 1—source data 5 ) . 10 . 7554/eLife . 08833 . 010Table 1 . GO annotations enriched for genes upregulated by mianserin during all ages , assessed by RNA-seq ( day 3 , 5 and 10 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08833 . 010GOEnriched P-valueresponse to stimulus4 . 47E-08response to stress5 . 83E-05response to xenobiotic stimulus3 . 25E-07defense response4 . 66E-05innate immune response1 . 56E-02immune response1 . 62E-02immune system process1 . 62E-02aging6 . 63E-05multicellular organismal aging6 . 63E-05determination of adult lifespan6 . 63E-05Note: No process was specifically downregulated for all three ages . Based on these observations , we classified gene expression changes for groups of genes into two types . Type I changes describe whether the overall expression across an entire functional group/pathway increases or decreases i . e . whether the pathway is up or down regulated with age . Type II changes describe the relative changes in gene expression among genes within functional groups with respect to each other . We named the type II change transcriptional drift . As animals age , genes within functional groups change expression levels in opposing directions resulting in the disruption of the co-expression patterns seen in young adults . To analyze the effects of aging on transcriptional drift ( type II ) , we designed graphs that plot the log-fold changes ( log [old/young reference day1] ) in gene expression as a function of age . Such a plot can be constructed for whole transcriptomes as well as for any functional subset of genes , for example , genes involved in oxidative phosphorylation or lysosome biology ( Figure 1h , i ) . In young adults , the log-fold change is 0 and values close to 0 therefore suggest gene expression as seen in young adults ( Figure 1h , i ) . To quantify transcriptional drift changes with age ( type II ) , we calculated the variance of the log-fold change for genes involved in each pathway . For the purpose of this study , we will refer to this variance as drift-variance ( see Materials and methods ) . If gene expression ratios within a pathway stay constant with age , drift-variance will stay small . If a majority of genes within a pathway change expression in opposing directions or if the rates by which they change differ dramatically , drift-variance will increase . Note that “transcriptional drift” is different from “transcriptional noise” in that the former analyzes variance among genes within the same biological replicates , whereas the latter analyzes variance of the same genes among biological replicates . Hence , how far the aging transcriptome deviates away from the transcriptome seen in young adults can be graphed in a Tukey-style box plot , which plots the drift-variance as a function of age ( Figure 1h , i ) . We will refer to these plots as drift-plots ( Figure 1h; Figure 2—figure supplement 1a–d ) . We constructed drift-plots for all 19 , 196 genes in the data of cohort #1 , which revealed a dramatic increase in drift-variance with age , showing a progressive loss of mRNA stoichiometries and co-expression patterns observed in young-adults ( Figure 2a , shaded region encompassing the whiskers of Tukey-plot ) . This effect was also seen in other publicly available data-sets of aging C . elegans transcriptomes and drift-variance continued to increase with age at least until day 20 ( Figure 2—figure supplement 1e ) . Mianserin treatment attenuated the effect of aging across the whole transcriptome and preserved the co-expression patterns observed in young-adults into later age . To test whether transcriptional drift is driven by a small subset of mRNAs or a transcriptome-wide phenomenon , we randomly divided the transcriptome into subsamples of ~1 , 000 genes . Each subsample showed identical increases in drift-variance with age , confirming a transcriptome-wide effect ( Figure 2—figure supplement 1f ) . 10 . 7554/eLife . 08833 . 011Figure 2 . Transcriptional drift-variance is attenuated by two longevity paradigms . ( a ) Drift-plots show that mianserin attenuates increasing drift-variance with age . Note that drift-variance in 10-day-old mianserin-treated animals is the same as in untreated 3-day-old control animals ( dotted red line ) . ( b ) Drift-plots show that increasing concentrations of mianserin cause drift-variance to decrease . Drift-variance was measured on day 5 by RNA-seq . ( c ) Corresponding to b , lifespan curves show that increasing concentrations of mianserin leads to a dose-dependent increase in survival . ( d ) Drift-plots show that initiating mianserin treatment at later ages reduces ( d3 ) or abolishes ( d5 ) its effect on transcriptional drift . Drift-variance was measured on day 10 by RNA-seq . ( e ) Corresponding to d , lifespan curves show initiating mianserin treatment at later ages reduces ( d3 ) or abolishes ( d5 ) its effect on lifespan . ( f ) Log-fold change of xenobiotic gene expression on day 10 when mianserin was added on day 1 or day 5 , compared to control animals treated with water on day 1 . Adding mianserin on day 1 or day 5 leads to comparable changes . ( g ) Drift-plots show daf-2 RNAi attenuates increasing drift-variance with age in a manner dependent on daf-16 . Left: vector control , middle: daf-2 RNAi , right: daf-16/daf-2 RNAi . P-values for transcriptional drift plots are calculated by robust Levene’s test , which compare variances and not mean values . ***P<0 . 001 . All error bars show drift-variance . See Figure 2—figure supplement 1–2 for additional information on calculating drift-variance and Table 2 . Also , see Methods section for transcriptional drift calculation in each figure panel . DOI: http://dx . doi . org/10 . 7554/eLife . 08833 . 01110 . 7554/eLife . 08833 . 012Figure 2—figure supplement 1 . This figure relates to Figure 1c , d and Figure 5a , b in main text . ( a ) Relationship of ( a ) fold-changes in gene expression as measured by qRT-PCR to b ) RNA-seq counts to ( c ) transcriptional drift and ( d ) drift-variance plots . Fold-changes in gene expression in older ( day 5 ) animals by mianserin are mostly caused by mianserin preserving the expression levels seen in young animals , thus leading to small drift-variances for groups of genes . ( e ) Additional transcriptional drift plots for aging C . elegans based on GEO data-sets GSE21784 and GSE46051 . Transcriptional drift increases continuously up until at least day 20 towards the end of the lifespan . ( f ) Transcriptional drift is observed across the entire transcriptome . Random sub-sampling generating ten sets of ~1 , 000 genes and plotting their drift-variance shows that transcriptional drift is a phenomenon present across the entire transcriptome and is not driven by small subsets of genes . DOI: http://dx . doi . org/10 . 7554/eLife . 08833 . 01210 . 7554/eLife . 08833 . 013Figure 2—figure supplement 2 . Egg RNA does not affect drift-variance . ( a ) DIC photomicrograph of eggs obtained from FUDR ( 120 µM final ) treated animals . Eggs are terminally arrested around the ventral closure ( “bean stage” , 400–500 nuclei ) and show a shrunken cell mass . Birefringent gut granules are observed in the middle of the eggs . Images were taken ~48 hr after FUDR treatment . ( Scale bar = 20 µm ) . ( b ) Number of adult worms that produce eggs 24 hr after FUDR treatment . Of the 298 worms evaluated , all of the animals developed germline with eggs inside . ( c ) Treatment with FUDR dramatically reduces the RNA content in eggs . Total RNA was extracted from FUDR-treated whole wt ( N2 ) worms , from eggs isolated from FUDR-treated N2 worms after ~28 hr of FUDR treatment and from eggs from non-FUDR-treated N2 worms ( the same time point as the RNA-seq young reference ) , . ***P<0 . 001 , comparison between whole worms and eggs treated with FUDR , unpaired t-test , n=3 , Error bars S . E . M; ##P<0 . 01 , comparison between eggs treated with FUDR and no FUDR , unpaired t-test , n=3 , Error bars S . E . M . ( d ) Electrophoresis of RNA extracted from whole worms or eggs isolated from FUDR treated animals . Same number of animals used for each sample . Comparison of equal volumes ( 10 µl ) of total RNA loaded from FUDR-treated whole worms and eggs isolated from FUDR-treated animals , resolved in an agarose gel . ( e ) Original drift plot from Figure 2a is shown again for comparison . Note that box in the middle of the drift plot , which is a Tukey-plot , represents the interquartile mean , or 50% of the transcriptome that changes less with age . As drift is also observed in the interquartile mean , drift is not driven by extreme outliers , but by the majority of the genes across the entire transcriptome . ( f ) Drift plot generated from our data-set only including genes that were also detected in the CF512 sterile strain data-set from ( Murphy et al . , 2003 ) . ( g ) Drift plot generated after removing 7 , 292 genes involved in egg-related functions detected from an eggs-only RNA-seq data-set ( Osborne Nishimura et al . , 2015 ) . ( h ) DIC photomicrograph of eggs obtained from untreated and FUDR-treated animals carrying the Pgcy-8::GFP reporter for AFD neurons . ( i ) Fluorescence microscopy images show AFD neurons in eggs derived from untreated adults ( left panel , white arrows ) but not in eggs obtained from FUDR-treated adults ( middle panel ) , confirming that FUDR treated eggs do not progress past the “bean stage” . FUDR does not inhibit Pgcy-8::GFP expression in adults ( right panel ) . ( j ) Overlay of h and i . ( k ) Drift plots using our data-set including only the genes that are highly enriched in AFD , ASE or NSM neurons ( Etchberger et al . , 2007; Spencer et al . , 2014 ) . As FUDR arrests embryonic development before the birth of these neurons , the drift-plots cannot be influenced by RNA derived from eggs . Explanations for Figure 2—figure supplement 2 In the experiments presented in the main manuscript , we used FUDR to sterilize the animals from which we subsequently extracted RNA for RNA-seq . Thus , our samples contained fractions of egg RNA . The following control experiments and analysis show that the fraction of RNA in our samples coming from eggs is small and does not influence the phenomenon of transcriptional drift and its attenuation by mianserin . We first isolated eggs from FUDR-treated and untreated animals . FUDR treatment causes the cell mass inside the eggs to shrink and to terminally arrest at around bean stage ( 400–500 nuclei ) ( Figure 2—figure supplement 2a ) . FUDR-treated animals all contained similar numbers of eggs 24 hr after FUDR treatement ( n=298 ) ( Figure 2—figure supplement 2b ) Note that many of the reported FUDR side-effects such as a lack of germline are not observed in 96-well liquid culture ( Gomez-Amaro et al . , 2015 ) . Extracting RNA from whole worms or eggs isolated from whole worms showed that FUDR-treated eggs contained 5 times less RNA compared to untreated eggs . The fraction of RNA originating from the eggs in FUDR-treated worms was roughly ~5% ( Figure 2—figure supplement 2c , d ) . We next asked whether this fraction could in anyway influence the phenomenon of transcriptional drift . The original plots ( Figure 2a , or Figure 2—figure supplement 2e ) of the entire transcriptome show that drift-variance increases in the interquartile mean ( boxes ) showing that it is not driven by a set of outlier genes , making it unlikely that the 5% fraction would influence drift-variance ( Krzywinski and Altman , 2014 ) . Nevertheless , to test possible interference , we calculated drift plots for various subsets of our data excluding transcrips expressed in eggs . The Murphy data were derived from CF512 ( sterile ) animals and thus any genes detected do not originate from eggs . We therefore excluded all genes not detected by Murphy et al from our data-set and recalculated drift . The resulting drift plot still shows a dramatic increase in drift-variance and attenuation by mianserin ( Figure 2—figure supplement 2f ) . A potential problem with the approach used in Figure 2—figure supplement 2f is that it only removed eggs/germline genes that are specific for eggs but that it did not remove genes that are present in both eggs and soma . We therefore removed all genes that were identified in C . elegans eggs by RNA-seq from our data-set to plot Figure 2—figure supplement 2g ( Osborne Nishimura et al . , 2015 ) . Of the 7 , 700 transcripts identified in eggs , 7 , 200 were present in our data-set . Note that this approach removes all ubiquitously expressed genes like ribosomal , mitochondrial and similar housekeeping genes that are present in both embryos and soma . Even though this operation removes only 7 , 200 out of 19 , 196 individual genes present in the data-set , these 7 , 200 genes account for 73% of total mRNA counts . Despite this dramatic reduction in overall mRNA transcripts , the drift plot combining the remaining 11 , 904 genes ( mostly low expressing genes ) confirms an increase in drift-variance with age that is suppressed by mianserin ( Figure 2—figure supplement 2g ) . To identify gene-sets that cannot possibly originate from the FUDR-treated eggs we exploited the specific arrest in embryonic development caused by FUDR . The DIC images suggested that FUDR arrests embryonic development before the birth of AFD , ASE and NSM neurons . If so , genes in our data-set that are specifically expressed in these neurons have to originate from the adult somatic tissue . To test that FUDR treatment prevents the birth of these neurons , we imaged eggs of C . elegans carrying a Pgcy-8::GFP transgene ( AFD marker ) ( Figure 2—figure supplement 2h , i , j ) . Eggs from untreated animals showed a clear expression of the marker while FUDR-treated eggs did not ( Figure 2—figure supplement 2i , j ( n>100 ) ) . FUDR did not repress the expression of the Pgcy-8::GFP transgene in adults , showing that the lack of a Pgcy-8::GFP signal in FUDR-treated eggs is due to an arrest before the neurons are born and not due to inhibition of the reporter expression by FUDR . As AFD neurons are born before ASE and NSM neurons , these results suggested that none of these three neurons are present in FUDR-treated eggs ( Sulston et al . , 1983 ) . After having established the absence of AFD , ASE and NSM neurons in eggs derived from FUDR treated animals , we then used the published gene-sets that are highly enriched in these three neuron types ( AFD , ASE , NSM ) to construct drift-plots ( Etchberger et al . , 2007; Spencer et al . , 2014 ) . Even for these highly restricted sets of genes , drift-variance dramatically increased with age and was repressed by mianserin . Taken together , these results show that the RNA contamination from FUDR-treated eggs is minimal and that this residual amount does not influence our results . DOI: http://dx . doi . org/10 . 7554/eLife . 08833 . 013 We previously showed , that the effect of mianserin to extend lifespan is dose-dependent ( Petrascheck et al . , 2007 ) . To explore a possible quantitative relationship between longevity and drift-variance , we generated drift-plots for transcriptomes of animals treated with increasing doses of mianserin ( Figure 1a , cohort #2 ) . Increasing doses of mianserin progressively increased longevity and decreased drift-variance as measured in 5-day-old animals ( Figure 2b , c; Table 2 ) . Thus , remarkably , by varying the dose of a single molecule , it was possible to control the degree to which aging drives the loss of transcriptional co-expression away from patterns observed in young adults . These results suggested a quantitative relationship between mianserin-induced longevity and its effect on drift-variance . 10 . 7554/eLife . 08833 . 014Table 2 . Survival data for lifespan of RNA-seq experimental cohorts . DOI: http://dx . doi . org/10 . 7554/eLife . 08833 . 014StrainTreatmentTreatment added on [day]Conc . [µM]Change in lifespan [%] Expt . 1/ Expt . 2/ Expt . 3P-value Expt . 1/ Expt . 2/ Expt . 3Mean Lifespan [days] Expt . 1/ Expt . 2/ Expt . 3Number of animals Expt . 1/ Expt . 2/ Expt . 3N2Waterd1019 . 33/ 17 . 2/ 20 . 45132/ 149/ 130N2Miad12+7/ +12/ -40 . 20/ 0 . 04/ 0 . 2520 . 64/ 19 . 23/ 19 . 67125/ 133/ 151N2Miad110+30/ +16/ +62 . 5E-07/ 3 . 7E-03/ 0 . 5525 . 09/ 19 . 92/ 21 . 7494/ 138/ 136N2Miad150+46/ +39/ +251 . 1E-19/ 1 . 9E-15/ 2 . 8E-0828 . 25/ 23 . 92/ 25 . 6395/ 131/ 125N2Miad350+15/ +14/ +12 . 0E-03/ 9 . 3E-04/ 0 . 2922 . 23/ 19 . 69/ 20 . 75121/ 134/ 152N2Miad550-8/ +8/ -20 . 18/ 0 . 06/ 0 . 8417 . 79/ 18 . 52/ 20 . 13123/ 151/ 139Summary of all lifespan experiments performed in parallel for cohorts 1 and 2 of the RNA-seq studies in Figure 2c , e . The treatments , water or mianserin , at the indicated concentrations ( conc . ) were added on indicated day ( D ) of adulthood and lifespan ( days ) was scored until 95% of animals were dead in all tested conditions . All values ( Change in lifespan [%] , P-values ) were calculated for the pairwise comparison between mianserin-treated and water-treated animals of the same condition , in 3 independent experiments ( expts . ) . Statistical analysis was performed using the Mantel–Haenszel version of the log-rank test . Mean lifespan [days] and number of animals in each experiment are indicated . Our previous studies had also shown that mianserin does not extend lifespan when added to 5-day-old post-reproductive adult animals ( Petrascheck et al . , 2007 ) . Thus , we next tested whether mianserin attenuates transcriptional drift-variance independently of longevity by treating older animals . Mianserin did not attenuate transcriptional drift-variance when added on day 5 ( Figure 2d ) . Adding mianserin on day 3 of adulthood caused a small extension of lifespan and a corresponding small attenuation of drift-variance , further supporting a quantitative relationship between suppression of drift-variance and extension of lifespan ( Figure 1a , cohort #3 , Figure 2d , e; Table 2 ) . However , mianserin fully induced the xenobiotic response by up to 1 , 000-fold irrespective of whether added on day 1 or day 5 ( Figure 2f ) . Therefore , the lack of an effect of mianserin when added to day 5 adults cannot be attributed to reduced drug uptake . Taken together , these results show that mianserin does not attenuate drift-variance when it does not extend lifespan . We next asked whether the attenuation of drift-variance is unique to mianserin or whether it is observed in other lifespan-extension paradigms ( Figure 2g ) . We asked whether reduced insulin signaling also attenuates drift-variance by analyzing the previously published gene expression data-sets of long-lived C . elegans daf-2 RNAi-treated and vector control animals ( Murphy et al . , 2003 ) . Analyses of drift-variance for these data-sets showed that treatment with daf-2 RNAi attenuated drift-variance ( Figure 2g ) . Moreover , mianserin and daf-2 RNAi attenuated age-associated drift of overlapping sets of genes . Of the 6 , 958 genes for which expression levels were detected at all ages in both data-sets , 58% ( 4 , 078 genes , binomial P= 6 . 3e-47 ) were attenuated by both longevity-extending mechanisms . This overlap is consistent with experiments showing that these two longevity mechanisms partially overlap , potentially explaining why mianserin only causes a +11% lifespan extension in daf-2 ( e1370 ) mutant animals instead of 31% seen in the parallel wild-type experiments ( Petrascheck et al . , 2007 ) . Thus , lifespan extension by mianserin or daf-2 RNAi attenuates transcriptional drift in overlapping sets of genes . Conversely , suppressing longevity by daf-16 ( RNAi ) prevented the attenuation of drift-variance by daf-2 ( RNAi ) and increased it beyond what was seen in control animals ( Figure 2g ) . Thus , the activation of DAF-16 target genes leads to the attenuation of transcriptional drift in thousands of genes across the transcriptome . Taken together , these results show that drift-variances increase with age in C . elegans and are attenuated in two different longevity paradigms ( Figure 2a , g ) . From a technical perspective , the comparison between the mianserin data and the Murphy data ( Murphy et al . , 2003 ) also shows that the phenomenon of transcriptional drift is robust enough not to be influenced by the presence of eggs in the animals or the method of sterilization , as our study used FUDR and the Murphy et al . ( 2003 ) study used sterile mutants ( Figure 2a , g; Figure 2—figure supplement 2 ) . The results above suggested that preserving low drift-variance in transcriptomes preserves longevity . We therefore asked whether attenuating drift-variance in specific pathways preserves homeostatic capacity , the ability of pathways to appropriately respond to a stimulus or stress . Throughout life , organisms respond to stimuli by activating or repressing transcriptional programs , an ability that is lost with age . We hypothesized that one way by which regulatory ability may be lost could be due to a failure to return to their precise steady-state transcriptional levels after stimulation . This would give rise to increases in drift-variance ( Figure 3a ) , as seen in the drift plots for oxidative phosphorylation or lysosome biology ( Figure 1h , i ) . In this model , slight initial deviations in gene expression levels would be compounded over time resulting in imbalanced stoichiometries between pathway components resulting in functional decline with age ( Figure 3a ) . 10 . 7554/eLife . 08833 . 015Figure 3 . Preserving low drift-variances in redox pathways preserves redox capacity into old age . ( a ) Model for the occurrence of transcriptional drift with age . Genes belonging to the same pathway appropriately respond to a stimulus but subsequently fail to return to steady-state levels . Repeated stimuli compound this effect leading to increases in transcriptional drift . If multiple genes within a pathway have propensity to drift in one or the other direction drift-variance increases with age . ( b ) Drift-plots show increases in drift-variance in multiple KEGG or GO annotations associated with redox processes . P-values compare variance , not mean , n: No . of genes in each category . *P<0 . 05 , **P<0 . 01 , ***P<0 . 001 , Levene’s test . Error bars; drift-variance ( c ) Fold increase in survival of N2 wild-type ( wt ) mianserin treated vs . untreated animals when challenged with paraquat at different ages . The protective effect of mianserin increases with age . *P<0 . 05 , t-test , Error bars: S . E . M . ( d ) Fold increase in survival of wt ( N2 ) treated vs . untreated animals when challenged with paraquat on day 10 . Delaying mianserin treatment into later life reduces its protective effect . *P<0 . 05 , t-test , Error bars: S . E . M . ( e ) Linear regression of log fold-changes in gene expression with age for genes previously shown to change upon oxidative stress . Genes upregulated in response to oxidative stress ( n=252 ) increase with age , and genes downregulated in response to oxidative stress decrease ( n=88 ) with age . Mianserin attenuates age-associated expression changes in oxidative stress genes in the direction indicated by blue arrows . Shading: 95% confidence interval . ***P<0 . 001 , Wilcoxon rank-sum test . See Tables 3–5 for detailed statistics and Methods section for transcriptional drift calculation in each figure panel . DOI: http://dx . doi . org/10 . 7554/eLife . 08833 . 01510 . 7554/eLife . 08833 . 016Figure 4 . Preserving redox capacity into old age requires the serotonin receptor SER-5 . ( a ) Survival of wt ( dotted lines ) or serotonin receptor mutants and serotonin synthesis mutant ( bold lines ) treated with water ( black ) or mianserin ( blue ) on day 1 , followed by increasing concentrations of paraquat on day 5 . ( b ) Bar graph shows fold protection as a ratio of survival of mianserin-treated vs . water-treated GPCR mutant animals ( ( Mia/water ) -1 ) . *P<0 . 05 , **P<0 . 01 , ***P<0 . 001 , n . s . , not significant , t-test; Error bars: S . E . M . See Figure 4—figure supplement 1 , and Tables 6 and 7 for detailed statistics . DOI: http://dx . doi . org/10 . 7554/eLife . 08833 . 01610 . 7554/eLife . 08833 . 017Figure 4—figure supplement 1 . This figure relates to Figure 4a in main text . ( a ) Survival of wt and two independent alleles of ser-5 mutants , ser-5 ( tm2647 ) or ser-5 ( tm2654 ) , treated with water or mianserin on day 1 , followed by increasing concentrations of paraquat on day 5 of adulthood . ( b ) Hierarchical clustering of fold change [serotonin antagonist/DMSO] in protection of wt ( N2 ) and ser-5 mutant animals , when treated with DMSO or serotonin antagonists on day 1 followed by paraquat on day 5 , shows the degree of similarity in protection between 8 structurally different serotonin antagonists ( left ) and the requirement of ser-5 for these antagonists to protect from oxidative stress . ( c ) Bar graphs quantifying transcriptional drift by qRT-PCR ( log fold-changes in gene expression ) in 5-day-old N2 and ser-3 ( ad1774 ) animals ( left panel ) , and N2 and ser-4 ( ok512 ) animals ( right panel ) treated with mianserin , relative to water-treated N2 , determined by qRT-PCR . Mianserin treatment of ser-3 ( ad1774 ) and ser-4 ( ok512 ) strains result in a drift pattern , similar to those seen in N2 . Thus , these receptors are neither required for drift-attenuation in redox genes , nor for the age-associated increase in oxidative stress resistance ( Figure 4 ) . Error bars: S . E . M . For detailed statistics , see Tables 6 and 7 . DOI: http://dx . doi . org/10 . 7554/eLife . 08833 . 01710 . 7554/eLife . 08833 . 018Figure 5 . Mianserin attenuates drift-variance in peripheral tissues via SER-5 . ( a ) Bar graphs quantifying transcriptional drift ( log fold-changes in gene expression ) as measured by qRT-PCR in 5-day-old N2 and ser-5 ( ok3087 ) animals treated with mianserin , relative to water-treated N2 . Mianserin treatment increases expression of genes drifting down with age and decreases expression of genes drifting up with age in N2 , but not in ser-5 ( ok3087 ) mutants . ( See 5b ) . *P<0 . 05 , **P<0 . 01 , ***P<0 . 001 , t-test; Error bars: S . E . M . ( b ) Log fold-change in gene expression as a function of age for stress response genes shown in a . Blue arrows indicate how mianserin treatment corrects age-associated changes in gene expression toward an expression pattern as seen in young adults . ( c ) Bar graphs quantifying log fold-changes in gene expression in 1-day-old N2 and ser-5 ( ok3087 ) animals treated with paraquat , relative to water-treated N2 animals . N2 and ser-5 ( ok3087 ) show an identical response to paraquat . ( d ) Mianserin treatment on day 1 of adulthood enhances transcription of sod and hsp-16 . x genes in response to an 8h paraquat treatment on day 5 in wt ( N2 ) animals compared to water treated controls . In contrast , mianserin treatment of ser-5 ( ok3087 ) animals did not enhance transcription of sod and hsp-16 . x genes . mRNA levels of genes were evaluated by qRT-PCR and plotted as fold induction ( PQ/water ) ( Y-axis ) for each gene . ( e ) Survival plot of mianserin-treated and untreated N2 and ser-5 ( ok3087 ) animals . ***P<0 . 001 , *P<0 . 05 , Mantel–Haenszel version of the log-rank test . f ) Percent increase in lifespan as a function of mianserin concentration . Mutations in ser-5 or synaptic components rendered the animals partially or completely resistant to mianserin-induced lifespan extension . See Figure 5—figure supplement 1 for additional data , and Table 8 for detailed statistics . DOI: http://dx . doi . org/10 . 7554/eLife . 08833 . 01810 . 7554/eLife . 08833 . 019Figure 5—figure supplement 1 . This figure relates to Figure 5e in main text . ( a ) Kaplan-Meier graphs for lifespan of wt ( dotted line ) and synaptic mutant animals treated with water ( black ) or mianserin ( blue ) . Synaptic transmission is required for mianserin-induced lifespan extension . For detailed statistics , see Table 8 . ( b ) Kaplan-Meier graphs for lifespan of wt ( dotted lines ) , ser-5 ( ok3087 ) , ( solid lines ) treated with DMSO or serotonin antagonists namely: Dihydroergotamine , Metergoline , Amperozide , Methiothepin , Ketanserin , Mirtazapine , LY-165 , 163/PAPP or mianserin , on day 1 of adulthood . All 8 serotonergic antagonists completely or partially depend on ser-5 . DOI: http://dx . doi . org/10 . 7554/eLife . 08833 . 01910 . 7554/eLife . 08833 . 020Figure 6 . Mianserin extends lifespan by specifically slowing age-associated changes in early adulthood . ( a ) PCA plot of RNA-seq data . Each circle represents one RNA-seq sample with the age , in days , indicated . Mianserin-treated day 10 samples show the same transcriptional age as untreated day 3 animals , dotted red line . ( b ) Mortality curves ( moving average ) constructed using Gompertz equation for lifespan experiments from 15 independent experiments of ~100 animals each treated with water or mianserin 50 µM ( n>1500 total for each condition ) . Mianserin treatment causes a 7–8 day parallel shift in log mortality as compared to the water-treated animals . ( c ) Survival of wt animals treated with mianserin for 8 hr , 1 day , 5 days or throughout life was determined and compared to water treated control animals . Removing mianserin after 8 hr or 1 day lessens its effect on lifespan , while removing mianserin on day 5 or maintaining treatment throughout life showed a comparable effect . ( d ) Mean survival of wt animals treated with water or mianserin for 8 hr , 1 day , 5 , 10 , 15 days or throughout life was plotted as a function of mianserin exposure in days . Mianserin treatment for 5 to 10 days was required and sufficient for an optimal lifespan extension . ( e ) Distinct modes of lifespan extension: Proportional lifespan extension leads to a proportional extension across life whereas period-specific lifespan extension leads to a reduced rate of age-associated degeneration during a specific period only . Mianserin reduces the rate of age-associated changes in early adulthood , thereby postponing mortality levels by 7–8 days causing a ‘period-specific lifespan extension’ . ( f ) Model for how mianserin modulates age-associated mortality in early adulthood . Blocking serotonergic signaling via SER-5 decreases transcriptional drift-variance with age in redox genes , leading to preserved homeostatic capacity in redox function , which subsequently delays age-associated mortality . ( g ) Mianserin does not affect reproductive longevity . Wt animals were treated with water or mianserin ( 50 µM ) on day 1 followed by counting the number of viable eggs laid by them on day 1 , day 2 , day 3 and day 4 . h ) Chymotrypsin-like 26S proteasome activity measured from wt animals treated with water or mianserin ( 50 µM ) on day 1 followed by proteasome activity assay on day 2 ( upper panel ) or day 5 ( lower panel ) . Mianserin treatment does not lead to an increase in proteasome activity , unlike long lived germline-less animals . Error bars S . E . M . See Figure 6—figure supplement 1 for additional data and detailed statistics . DOI: http://dx . doi . org/10 . 7554/eLife . 08833 . 02010 . 7554/eLife . 08833 . 021Figure 6—figure supplement 1 . This figure relates to figure 6 in main text . ( a ) Mortality curves constructed using Gompertz equation for lifespan experiments of wt ( N2 ) animals from 15 independent experiments of ~100 animals each treated with water or mianserin 50 µM ( n>1500 total for each condition ) . The shift in log mortality as a function of time with mianserin treatment is parallel to the water-treated animals . See table below for aggregate data showing hazard/mortality for water and mianserin treatment . ( b ) Power of detection for 500 , 1000 and 1500 animals in each cohort as used in Figure 6b ( α=0 . 01 ) . Monte-Carlo simulations based on a parametric model derived from our data were used to determine the power of detection . A lifespan extension of 1 day corresponds to a 5% increase in lifespan . ( c ) Drift-plots show changes in drift-variance in proteasome pathway ( KEGG annotation: 03050 ) associated with 38 genes involved in proteasome activity in animals treated with water or mianserin ( 50 µM ) on day 1 and harvested on day 3 , 5 and 10 . Attenuation patterns of drift-variance with mianserin treatment corresponds functionally to changes in proteasome activity on day 2 and day 5 ( See panel a ) . Mianserin slightly increases transcriptional drift on day 5 and slightly reduces proteasome activity function . P-values compare variance , not mean , **P<0 . 01 , Levene’s test . Error bars; drift-variance . DOI: http://dx . doi . org/10 . 7554/eLife . 08833 . 02110 . 7554/eLife . 08833 . 022Figure 7 . Transcriptional drift-variance increases with age in various species . ( a ) Transcriptional drift-variance in gene expression from different mouse tissues aged 13 to 130 weeks . Drift-plots show an increase in drift-variance with age in mouse brain , kidney , liver , lung and spleen ( b ) Drift-variance plotted as a function of age for different organs . To obtain drift-variance values for young animals , a single transcriptome was set aside and used a reference . ( c ) Drift-plot for gene expression from 32 human brains ( frontal cortex ) plotted as a function of age in years . Data binned in 20-year increments . ( d ) Drift-variance plotted as a function of age in years for individuals . Each dot corresponds to one brain sample ( frontal cortex ) . Shading indicates 95% confidence interval ( ρ=0 . 603 , P=0 . 0014 ) . ( e ) Drift plots show a higher transcriptional drift-variance in BJ fibroblasts ( BJ ) and fibroblasts from Hutchinson Gilford progeria syndrome ( HGPS ) , when compared to H9 embryonic stem cells . Reprogramming the BJ and HGPS cells to induced pluripotent stem cells ( iPSCs ) leads to a partial reversal of the transcriptional drift-variance to a lower variance corresponding to the young phenotype of the iPSCs . See Figure 2—figure supplement 1 for additional information on transcriptional drift calculation , and Methods section for transcriptional drift calculation in each figure panel . DOI: http://dx . doi . org/10 . 7554/eLife . 08833 . 022 Our previous studies showed that mianserin protected C . elegans from oxidative stress by a neuronal mechanism that modulated peripheral stress response genes ( NEUROX ) ( Rangaraju et al . , 2015a ) . We therefore constructed drift plots for redox-associated pathways that showed that mianserin indeed increased the overall expression of oxidative stress response genes ( type I ) relative to age-matched controls but also attenuated transcriptional drift ( type II ) ( Figure 3b; Table 3 ) . 10 . 7554/eLife . 08833 . 023Table 3 . Gene ontology ( GO ) pathways of relevance to this study that are differentially regulated by mianserin . DOI: http://dx . doi . org/10 . 7554/eLife . 08833 . 023KEGG / GO IDKEGG / GO TermNumber of Genes observedLevene’s test for variance ( Difference in transcriptional drift- variance ) Water D1 vs . water DxLevene’s test for variance ( Difference in transcriptional drift- variance ) water Dy vs . mianserin DyTranscriptome19 , 196D3 : P < 1 . 0E-100 D5 : P < 1 . 0E-100 D10: P < 1 . 0E-100D3 : P < 1 . 0E-100 D5 : P < 1 . 0E-100 D10: P < 1 . 0E-100KEGG:Cel00030Pentose phosphate pathway17D3 : P = 0 . 0096 D10: P <1 . 0E-5D3 : P <1 . 0E-4 D10: P = 0 . 01GO: 0006979Response to oxidative stress67D3 : P <1 . 0E-10 D10: P <1 . 0E-16D3 : P <1 . 0E-4 D10: P = 0 . 001GO: 0045454Cell redox homeostasis52D3 : P <1 . 0E-6 D10: P <1 . 0E-10D3 : P <1 . 0E-4 D10: P = 0 . 029GO: 006749Glutathione metabolism13D3 : P <1 . 0E-4 D10: P <1 . 0E-7D3 : P =0 . 041 D10: P <1 . 0E-4GO: 0007186G-protein coupled receptor signaling335D3 : P <1 . 0E-24 D10: P < 1 . 0E-100D3 : P <1 . 0E-4 D10: P <1 . 0E-4GO: 0016209Antioxidant activity34D3 : P <1 . 0E-8 D10: P <1 . 0E-10D3 : P = 0 . 002 D10: P = 0 . 06Summary of gene changes with RNA-seq transcriptome analysis in Figure 3b . GO ID is the Gene Ontology identification number . GO Term is the Gene Ontology term for the biological process . Dx = age in days for the animals indicated , compared with D1 water-treated animals . Dy = age in days for water- and mianserin-treated animals , compared on the same day of age indicated . We therefore asked whether mianserin treatment increased resistance to oxidative stress by either directly activating the oxidative stress response or whether attenuating transcriptional drift would preserve homeostatic capacity into older age ( Rahman et al . , 2013 ) . Animals were treated with water or mianserin on day 1 of adulthood , followed by treatment with the reactive oxygen species ( ROS ) generator paraquat on day 3 , 5 , or 10 ( Figure 3c ) . On day 3 of adulthood , no difference in stress resistance between mianserin-treated and untreated animals was observed . As animals grew older ( day 5 and day 10 ) , mianserin treatment greatly improved stress resistance ( Figure 3c; Table 4 ) . Again , as with lifespan , delaying the start of mianserin treatment to day 3 and day 5 progressively reduced its protective effect on stress resistance , this time measured in animals subjected to paraquat on day 10 of adulthood ( Figure 3d; Table 5 ) . Thus , mianserin treatment specifically improves stress resistance in older ( day 5 and day 10 ) but not in younger ( day 3 ) animals consistent with a model in which it preserves the homeostatic capacity of redox function . 10 . 7554/eLife . 08833 . 024Table 4 . Survival data for paraquat stress resistance assays . DOI: http://dx . doi . org/10 . 7554/eLife . 08833 . 024StrainTreatmentConc . [µM]Treatment added [day]PQ 100 mM , added [day]Survival after PQ [%] ( expt . 1 ) Survival after PQ [%] ( expt . 2 ) Survival after PQ [%] ( expt . 3 ) Mean , Survival after PQ [%]S . D . , Survival after PQ [%]P-valueNo . of wellsTotal no . of animalsN2Water0d1d370 . 043 . 162 . 258 . 413 . 948450N2Mia50d1d387 . 347 . 953 . 963 . 021 . 37 . 72E-0148390N2Water0d1d555 . 856 . 266 . 159 . 35 . 848436N2Mia50d1d595 . 596 . 192 . 094 . 52 . 24 . 24E-0348435N2Water0d1d1063 . 337 . 441 . 747 . 513 . 948400N2Mia50d1d1091 . 982 . 185 . 486 . 45 . 02 . 85E-0248390Summary of all stress resistance assays performed in Figure 3c . The treatments , water or mianserin ( Mia ) , at the indicated concentrations ( conc . ) were added on day 1 of adulthood . Paraquat ( PQ ) was added to a final conc . of 100 mM on day 3 ( d3 ) , day 5 ( d5 ) or day 10 ( d10 ) and survival after PQ [%] was calculated 24 hr after the respective PQ addition . Mean and standard deviation ( S . D . ) of survival after PQ [%] were calculated from 3 independent experiments ( expts . ) . P-values were calculated between water and mianserin-treatments on the same day of PQ addition , using unpaired t-test . The total number of wells and animals from which data were collected are indicated . 10 . 7554/eLife . 08833 . 025Table 5 . Survival data for paraquat stress resistance assays , mianserin added on different days . DOI: http://dx . doi . org/10 . 7554/eLife . 08833 . 025StrainTreatmentConc . [µM]Treatment added dayPQ 100 mM , added daySurvival [%] ( expt . 1 ) Survival [%] ( expt . 2 ) Survival [%] ( expt . 3 ) Mean , Survival [%]S . D . , Survival [%]P-valueNo . of wellsTotal no . of animalsN2Water0d1d1063 . 3037 . 4441 . 7247 . 4813 . 8648400N2Mia50d1d1091 . 8582 . 0585 . 3886 . 434 . 972 . 85E-0248390N2Water0d3d1063 . 9741 . 2538 . 3547 . 8514 . 0248403N2Mia50d3d1078 . 5266 . 2273 . 6272 . 796 . 190 . 07448378N2Water0d5d1057 . 3143 . 8342 . 5747 . 908 . 1648387N2Mia50d5d1068 . 6350 . 5858 . 6259 . 289 . 040 . 1848398Summary of all stress resistance assays performed in Figure 3d . The treatments , water or mianserin ( Mia ) , at the indicated concentrations ( conc . ) were added on day 1 ( D1 ) , day 3 ( D3 ) or day 5 ( D5 ) of adulthood . 100mM Paraquat ( PQ ) was added on day 10 ( D10 ) and survival [%] was calculated after 24 hr . Mean and standard deviation ( S . D ) of survival [%] were calculated from 3 independent experiments ( expts . ) . P-value calculated between water and mianserin-treatments using t-test . The total number of wells and animals from which data were collected are indicated . To further distinguish between a model in which mianserin directly activates an oxidative stress response from one that preserves the homeostatic capacity by attenuating drift-variance , we asked whether mianserin enhanced ( direct activation ) or attenuated ( preserving capacity ) genes that change in response to oxidative stress ( Figure 3e ) . Oliveira et al . identified 252 genes that were upregulated and 88 genes that were downregulated in young C . elegans in response to oxidative stress , and can therefore be considered an experimentally determined oxidative stress signature ( Oliveira et al . , 2009 ) . We hypothesized that a direct activation of the oxidative stress response by mianserin would mimic the increase in expression of the 252 genes and the decrease in the expression of the 88 genes as seen in response to oxidative stress . However , we observed an attenuation rather than an activation of the oxidative stress signatures , consistent with preserving homeostatic capacity rather than a direct activation . Genes that increased in response to oxidative stress ( 252 ) showed a lower expression while genes that decreased ( 88 ) in response to oxidative stress showed a higher expression in age-matched mianserin-treated animals ( Figure 3e ) . Consistent with the functional data , differences in the oxidative stress signature were only observed in older animals ( day 5 , 10 ) , but not in younger day 3 animals . These results are consistent with a model in which mianserin treatment preserves the redox system from age-associated decline , thus improving redox capacity in older age . In mammals , mianserin antagonizes serotonergic signals sent by 5-HT2A/C receptors ( Gillman , 2006 ) . We next asked whether preservation of redox capacity and reducing drift-variance in redox pathways by mianserin depends on serotonergic signaling . To identify the serotonergic receptor , we treated multiple mutants , each deficient in signaling by a single G-protein coupled receptor ( GPCR ) with mianserin on day 1 , followed by increasing concentrations of paraquat on day 5 to induce oxidative stress ( Figure 4a , b; Table 6 ) . Mianserin was unable to protect multiple ser-5 mutant alleles ( ok3087 , tm2647 , tm2654 ) from oxidative stress ( Figure 4a , b; Figure 4—figure supplement 1a; Table 6 ) . In addition , seven structurally distinct serotonergic antagonists/inverse agonists also protect from oxidative stress in a ser-5 dependent manner ( Figure 4—figure supplement 1b; Table 7 ) . Furthermore , mianserin did not protect animals unable to synthesize serotonin ( tph-1 ( mg280 ) ) ( Figure 4a; Table 6 ) ( Sze et al . , 2000 ) . 10 . 7554/eLife . 08833 . 026Table 6 . Survival data for paraquat stress resistance assays . DOI: http://dx . doi . org/10 . 7554/eLife . 08833 . 026StrainTreatmentConc . [µM]PQ conc . [mM]Survival after PQ [%] ( expt . 1 ) Survival after PQ [%] ( expt . 2 ) Survival after PQ [%] ( expt . 3 ) Survival after PQ [%] ( expt . 4 ) Survival after PQ [%] ( expt . 5 ) Survival after PQ [%] ( expt . 6 ) Mean , Survival after PQ [%]S . D . , Survival after PQ [%]P-valueNo . of wellsTotal no . of animalsN2Water0089 . 998 . 995 . 898 . 293 . 993 . 295 . 03 . 448548Water01576 . 4888295 . 395 . 591 . 788 . 27 . 748578Water02574 . 291 . 38092 . 985 . 180 . 484 . 07 . 248531Water05066 . 267 . 863 . 881 . 961 . 667 . 868 . 27 . 148530Water07550 . 161 . 144 . 164 . 642 . 451 . 852 . 48 . 948545Water010036 . 234 . 435 . 553 . 52354 . 739 . 612 . 348503Mia50010010099 . 510010099 . 299 . 80 . 31 . 71E-0248556Mia501510098 . 287 . 610098 . 810097 . 44 . 93 . 52E-0248523Mia502596 . 298 . 89598 . 410098 . 297 . 81 . 84 . 54E-0348529Mia50509595 . 994 . 595 . 39998 . 296 . 31 . 81 . 19E-0448536Mia507598 . 989 . 389 . 492 . 697 . 598 . 194 . 34 . 41 . 29E-0548516Mia5010097 . 690 . 890 . 769 . 893 . 995 . 689 . 710 . 11 . 95E-0548539ser-1 ( ok345 ) Water009271 . 389 . 284 . 211 . 224228Water01573 . 357 . 981 . 871 . 012 . 124187Water02571 . 355 . 967 . 865 . 08 . 124209Water05054 . 846 . 442 . 647 . 96 . 224213Water07539 . 456 . 350 . 748 . 88 . 624213Water010024 . 227 . 346 . 632 . 712 . 124224Mia5001001001001000 . 00 . 1324215Mia501598 . 897 . 797 . 698 . 00 . 70 . 0624211Mia502597 . 994 . 298 . 496 . 82 . 31 . 51E-0224194Mia505094 . 895 . 49795 . 71 . 14 . 52E-0324224Mia507593 . 989 . 992 . 492 . 12 . 09 . 87E-0324232Mia5010087 . 489 . 689 . 588 . 81 . 21 . 45E-0224234ser-2 ( pk1357 ) Water0010010095 . 510098 . 92 . 332278Water0158897 . 573 . 792 . 287 . 910 . 232239Water02590 . 31008383 . 289 . 18 . 032206Water05076 . 787 . 273 . 762 . 775 . 110 . 132254Water07573 . 973 . 265 . 25366 . 39 . 732220Water01007259 . 654 . 447 . 758 . 410 . 332220Mia50098 . 910010010099 . 70 . 60 . 5132231Mia50151001001001001000 . 00 . 1032255Mia502598 . 910098 . 996 . 998 . 71 . 30 . 1032228Mia505010010095 . 596 . 998 . 12 . 31 . 71E-0232243Mia507598 . 99596 . 891 . 895 . 63 . 06 . 35E-0332245Mia501009788 . 792 . 395 . 493 . 43 . 73 . 80E-0332210ser-3 ( ad1774 ) Water0010010092 . 697 . 54 . 324176Water0158988 . 586 . 888 . 11 . 224174Water02590 . 585 . 485 . 287 . 03 . 024216Water05081 . 37472 . 175 . 84 . 924176Water07570 . 848 . 658 . 959 . 411 . 124169Water010043 . 746 . 530 . 140 . 18 . 824140Mia50098 . 210095 . 898 . 02 . 10 . 8824176Mia501598 . 910098 . 499 . 10 . 83 . 25E-0424228Mia502593 . 810090 . 294 . 75 . 00 . 1024173Mia505098 . 110093 . 897 . 33 . 24 . 97E-0324174Mia507592 . 49591 . 893 . 11 . 73 . 20E-0224197Mia5010093 . 465 . 682 . 880 . 614 . 01 . 92E-0224180ser-4 ( ok512 ) water0010087 . 610098 . 696 . 66 . 032249Water01510072 . 691 . 384 . 487 . 111 . 632262Water02598 . 267 . 972 . 585 . 581 . 013 . 732224Water0508867 . 183 . 363 . 575 . 512 . 032229Water0756947 . 275 . 861 . 463 . 412 . 332225Water010056 . 348 . 36043 . 252 . 07 . 632204Mia50010095 . 910010099 . 02 . 10 . 4932212Mia501596 . 997 . 297 . 597 . 797 . 30 . 40 . 2132228Mia502597 . 510091 . 795 . 596 . 23 . 50 . 1132230Mia505093 . 896 . 896 . 495 . 395 . 61 . 34 . 31E-0232261Mia507510091 . 588 . 196 . 594 . 05 . 39 . 66E-0332227Mia5010096 . 986 . 590 . 38990 . 74 . 43 . 75E-0432252ser-5 ( ok3087 ) Water0098 . 892 . 29996 . 73 . 924206Water01591 . 183 . 685 . 586 . 73 . 924230Water02586 . 271 . 688 . 282 . 09 . 124222Water05083 . 267 . 575 . 975 . 57 . 924222Water07568 . 4647769 . 86 . 624216Water010065 . 158 . 262 . 962 . 13 . 524232Mia50098 . 693 . 999 . 297 . 22 . 90 . 8524248Mia501596 . 292 . 997 . 495 . 52 . 33 . 90E-0224221Mia50259578 . 490 . 988 . 18 . 60 . 4524184Mia505089 . 577 . 482 . 983 . 36 . 10 . 2524219Mia507573 . 255 . 772 . 567 . 19 . 90 . 7224213Mia501006454 . 679 . 466 . 012 . 50 . 6524200ser-5 ( tm2647 ) Water0097 . 297 . 396 . 997 . 10 . 224248Water01588 . 891 . 287 . 389 . 12 . 024230Water02594 . 489 . 985 . 890 . 04 . 324227Water05079 . 584 . 681 . 781 . 92 . 624228Water07579 . 973 . 560 . 271 . 210 . 024248Water010051 . 65944 . 151 . 67 . 524224Mia50096 . 799 . 294 . 396 . 72 . 50 . 8024233Mia501596 . 788 . 49593 . 44 . 40 . 2324246Mia502597 . 288 . 592 . 492 . 74 . 40 . 4924187Mia505083 . 787 . 885 . 485 . 62 . 10 . 1324234Mia507569 . 777 . 373 . 573 . 53 . 80 . 7424203Mia5010046 . 475 . 170 . 363 . 915 . 40 . 3024196ser-5 ( tm2654 ) Water0081 . 596 . 383 . 487 . 18 . 124232Water01568 . 886 . 675 . 977 . 19 . 024223Water02577 . 189 . 169 . 178 . 410 . 124226Water05055 . 279 . 878 . 471 . 113 . 824254Water07547 . 542 . 555 . 348 . 46 . 524209Water010041 . 23645 . 841 . 04 . 924215Mia50083 . 796 . 390 . 490 . 16 . 30 . 6324232Mia501573 . 770 . 382 . 675 . 56 . 40 . 8224232Mia502566 . 973 . 788 . 276 . 310 . 90 . 8124184Mia505054 . 568 . 854 . 659 . 38 . 20 . 2924200Mia507534 . 941 . 966 . 547 . 816 . 60 . 9524227Mia5010018 . 230 . 640 . 729 . 811 . 30 . 2224187ser-6 ( tm2146 ) Water0098 . 996 . 998 . 610098 . 61 . 332230Water01595 . 189 . 696 . 593 . 73 . 632260Water02597 . 787 . 590 . 384 . 890 . 15 . 632221Water05095 . 397 . 584 . 878 . 188 . 99 . 132256Water07584 . 887 . 17763 . 578 . 110 . 632265Water010082 . 478 . 177 . 95372 . 913 . 432253Mia50010093 . 310096 . 997 . 63 . 20 . 5732278Mia501598 . 896 . 492 . 495 . 93 . 20 . 4932230Mia502597 . 997 . 596 . 491 . 395 . 83 . 00 . 1432190Mia505010010088 . 692 . 595 . 35 . 70 . 2932252Mia507592 . 210088 . 588 . 992 . 45 . 30 . 0732242Mia5010095 . 691 . 393 . 495 . 794 . 02 . 14 . 92E-0232221ser-7 ( tm1325 ) Water0068 . 173 . 394 . 578 . 614 . 024200Water01548 . 149 . 632 . 443 . 49 . 524142Water02545 . 742 . 930 . 939 . 87 . 924152Water0503837 . 836 . 537 . 40 . 824152Water07516 . 420 . 241 . 826 . 113 . 724160Water010025 . 123 . 231 . 626 . 64 . 424134Mia50098 . 898 . 910099 . 20 . 70 . 1324217Mia501595 . 893 . 897 . 295 . 61 . 79 . 18E-0324212Mia502510093 . 497 . 496 . 93 . 32 . 25E-0324193Mia505088 . 59294 . 691 . 73 . 15 . 30E-0424179Mia507591 . 392 . 489 . 491 . 01 . 51 . 37E-0224189Mia5010096 . 991 . 781 . 690 . 17 . 88 . 94E-0424186tph-1 ( mg280 ) Water0097 . 296 . 198 . 297 . 21 . 124148Water02566 . 967 . 87670 . 25 . 024156Water0505247 . 156 . 952 . 04 . 924164Water07532 . 234 . 64838 . 38 . 524148Water010012 . 26 . 742 . 320 . 419 . 224169Mia50094 . 310096 . 997 . 12 . 90 . 9624161Mia502590 . 458 . 778 . 675 . 916 . 00 . 6124159Mia505064 . 861 . 76965 . 23 . 72 . 33E-0224158Mia507552 . 928 . 957 . 946 . 615 . 50 . 4724143Mia501008 . 71 . 840 . 617 . 020 . 70 . 8524150Summary of all stress resistance assays performed in Figure 4a . The treatments , water or mianserin ( 50 µM ) , with their indicated concentrations ( conc . ) were added on day 1 of adulthood . Paraquat ( PQ ) was added in the concentration range of 0 to 100 mM on day 5 and survival after PQ [%] was calculated 24 hr later . Mean and standard deviation ( S . D . ) of survival after PQ [%] were calculated from 3 to 6 independent experiments ( expts . ) . P-values were calculated between water and mianserin-treatments at the same PQ conc . , using t-test . The total number of wells and animals from which data were collected are indicated . 10 . 7554/eLife . 08833 . 027Table 7 . Summary of oxidative stress protection by serotonin antagonists . DOI: http://dx . doi . org/10 . 7554/eLife . 08833 . 027Strain nameFold change in survival after PQ [ ( Drug/DMSO ) -1] Expt . 1Fold change in survival after PQ [ ( Drug/DMSO ) -1] Expt . 2Fold change in survival after PQ [ ( Drug/DMSO ) -1] Expt . 3Fold change in survival after PQ [ ( Drug/DMSO ) -1] Expt . 4Fold change in survival after PQ [ ( Drug/DMSO ) -1] Expt . 5Fold change in survival after PQ [ ( Drug/DMSO ) -1] Expt . 6Fold change in survival after PQ [ ( Drug/DMSO ) -1] Expt . 7Mean , Fold change in survival after PQS . D . , Fold change in survival after PQP-valueDihydroergotamine 88 µMN20 . 620 . 700 . 790 . 191 . 751 . 430 . 910 . 57ser-5 ( ok3087 ) 0 . 450 . 150 . 100 . 230 . 193 . 49E-02Metergoline 33 µMN20 . 540 . 570 . 680 . 941 . 241 . 670 . 940 . 44ser-5 ( ok3087 ) -0 . 05-0 . 27-0 . 11-0 . 12-0 . 130 . 091 . 50E-03Amperozide 13 µMN20 . 930 . 740 . 990 . 922 . 490 . 891 . 160 . 66ser-5 ( ok3087 ) 0 . 300 . 03-0 . 58-0 . 090 . 451 . 63E-02Methiothepin 10 µMN20 . 801 . 080 . 950 . 360 . 772 . 941 . 391 . 190 . 89ser-5 ( ok3087 ) 0 . 070 . 100 . 16-0 . 010 . 080 . 081 . 24E-02Ketanserin 176 µMN20 . 630 . 591 . 131 . 380 . 421 . 710 . 980 . 51ser-5 ( ok3087 ) -0 . 41-0 . 140 . 01-0 . 07-0 . 150 . 181 . 91E-03Mirtazapine 50 µMN20 . 80 . 71 . 10 . 41 . 00 . 81 . 50 . 890 . 35ser-5 ( ok3087 ) 0 . 0-0 . 1-0 . 1-0 . 2-0 . 110 . 071 . 92E-04LY-165 , 163 33/PAPP µMN20 . 480 . 491 . 000 . 940 . 531 . 400 . 810 . 37ser-5 ( ok3087 ) -0 . 030 . 35-0 . 07-0 . 160 . 020 . 233 . 19E-03Mianserin 50 µMN21 . 101 . 111 . 180 . 533 . 241 . 601 . 460 . 94ser-5 ( ok3087 ) 0 . 14-0 . 18-0 . 26-0 . 100 . 214 . 49E-02Summary of all stress resistance assays performed in Figure 4—figure supplement 1b . The treatments , DMSO or serotonin antagonists , with their indicated concentrations ( conc . ) were added on day 1 of adulthood . Paraquat ( PQ ) ( 100 mM ) was added on day 5 and survival after PQ [%] was calculated 24 hr later . Mean and standard deviation ( S . D . ) of survival after PQ [%] were calculated from 3 to 7 independent experiments ( expts . ) . P-values were calculated between N2 and mutant strains for fold change values with indicated small molecule treatments using t-test . We next asked whether SER-5 was also required for mianserin to preserve low transcriptional drift-variances in redox-related genes . We measured redox gene expression levels by qRT-PCR in wild-type 5-day-old N2 and ser-5 ( ok3087 ) animals that were treated with mianserin or water on day 1 ( Figure 5a , b; Figure 1—figure supplement 1a ) . In N2 samples , mianserin increased the expression of stress response genes that drift down with age ( sod-1 , sod-2 , prdx-2 , -3 , -6 ) and decreased the expression of stress response genes that drift up with age ( sod-4 , sod-5 , all hsp-16s ) , an effect that was not observed in ser-5 ( ok3087 ) mutants . In contrast , SER-3 and SER-4 , two receptors we previously showed to be required for lifespan extension by mianserin , were dispensable for stress protection ( Figure 4a , b ) ( Petrascheck et al . , 2007 ) , as well as for the attenuation of drift-variance in redox-associated genes ( Figure 4—figure supplement 1c ) . Thus , in wild-type animals , mianserin treatment preserved low drift-variances in redox-related genes into older age ( day 5 ) , in a ser-5 dependent manner ( Figure 5a , b ) . Importantly , ser-5 mutants were specifically defective in their response to mianserin , but showed no defect in their response to oxidative stress . Young ( day 1 ) wild-type N2 animals and ser-5 ( ok3087 ) mutants showed a nearly identical response to oxidative stress ( Figure 5c ) . The age-specific effects of ser-5 could not be attributed to expression changes , as ser-5 expression remained constant from day 1 to day 10 in our RNA-seq experiment . To test the hypothesis that mianserin preserved the homeostatic capacity of the redox system , as suggested by Figure 3e , we asked whether the treatment with mianserin on day 1 of adulthood led to an enhanced redox gene expression in response to the stressor paraquat in older animals ( day 5 ) . We therefore challenged older mianserin-treated or control animals ( day 5 ) with paraquat for 8 hr and measured redox-gene expression by qRT-PCR ( Figure 5d ) . Mianserin treatment led to an enhanced transcription of redox genes in response to paraquat as compared to age-matched control animals . The enhanced response was ser-5 dependent ( Figure 5d ) . Thus , SER-5 is required for mianserin to attenuate age-associated increases in drift-variance in redox genes , and to preserve the homeostatic capacity of the redox system into older age . Furthermore , lifespan-extension by mianserin was strongly reduced or abrogated in ser-5 , snt-1 and unc-26 mutant animals ( Figure 5e , f; Figure 5—figure supplement 1a; Table 8 ) . Seven additional serotonergic antagonists/inverse agonists also extended lifespan in a manner that was partially or fully dependent on ser-5 ( Figure 5—figure supplement 1b ) . Thus , these results show that inhibiting serotonergic signals via SER-5 extends lifespan , attenuates age-associated drift-variance in the redox system and preserves the homeostatic capacity of the redox system . 10 . 7554/eLife . 08833 . 028Table 8 . Summary of all lifespan data for mianserin . DOI: http://dx . doi . org/10 . 7554/eLife . 08833 . 028Cumulative statisticsStatistics of individual expts . StrainSmall moleculeNo . of expts . Mean lifespan [days] ( +Mia/+water ) change in lifespan [%]S . E . M . No . of animals ( +Mia/+water ) Mean lifespan ( days ) ( +Mia/+water ) change in lifespan [%]P-valueNo . of animals ( +Mia/+water ) N2Mia1226 . 7/19 . 8+35± 7642/57726 . 4/19 . 8+341 . 67E-0877/5925 . 5/21 . 5+196 . 85E-07113/9428 . 1/20 . 1+403 . 71E-1495/10430 . 6/19 . 0+643 . 17E-1557/5026 . 8/21 . 5+251 . 87E-11149/14522 . 6/16 . 6+271 . 61E-23151/125snt-1 ( md290 ) Mia320 . 9/18 . 2+15± 2236/23123 . 3/19 . 9+171 . 84E-0586/9017 . 3/15 . 4+121 . 18E-0279/8022 . 1/19 . 3+152 . 4E-0371/61unc-26 ( e205 ) Mia325 . 0/26 . 7-7± 7135/16527 . 8/26 . 9+30 . 5354/6822 . 2/26 . 5-164 . 52E-0214/2426 . 5/25 . 3+50 . 5267/73ser-5 ( ok3087 ) Mia323 . 4/22 . 2+5± 5496/45823 . 6/20 . 6+154 . 19E-02152/14426 . 4/26 . 2+10 . 85174/14420 . 1/19 . 8-10 . 25170/170Summary of all lifespan experiments performed in Figure 5e , f and Figure 5—figure supplement 1a . N2 and mutant strains were treated with 50 µM mianserin ( Mia ) on day 1 and lifespan [days] was scored until 95% of animals were dead in all tested conditions . Cumulative statistics and statistics of individual experiments are shown . Mean lifespan [days] , change in lifespan [%] and S . E . M . for mianserin-treated ( +Mia ) and water-treated ( +water ) animals from multiple , independent experiments ( expts . ) are shown . Change in lifespan [%] and P-values for individual experiments were calculated using the Mantel–Haenszel version of the log-rank test . Number of animals in individual experiments and all experiments combined are shown . We next asked whether drift-variance could be used as a metric to monitor age-associated change in young adults . Comparing drift-variances between mianserin-treated and untreated animals , we noticed that by day 10 , mianserin-treated animals exhibited a drift-variance slightly lower than that of 3-day-old control animal ( P=0 . 37 ) . This suggested that mianserin-treated animals showed a ~7–8 day delay in age-associated transcriptional change compared to age-matched controls ( Figure 2a ) . Principle component analysis ( PCA ) , a different statistical method to analyze differences between transcriptomes , confirmed this observation ( Figure 6a ) . PCA showed that control samples aligned on the x-axis ( dimension 1 ) according to age and that 10 day-old mianserin-treated animals aligned closer to 3-day-old than to 10-day-old control animals . These results suggested that the physiological shift that results in the 7–8 day lifespan extension observed in mianserin-treated animals at the end of a lifespan assay was already observable by day 10 . We therefore asked whether mianserin slowed age-associated physiological change specifically in early adulthood causing a 7–8 day delay by day 10 . If so , mianserin would be expected to specifically lower the mortality rate in young but not in old adults . However , the number of age-associated death events in young adults is too low to directly determine changes in age-associated mortality rates before the age of day 10 . As we are comparing mortality in animals either treated with water or mianserin that is added to the same population of worms on day 1 of adulthood , we can confidently state that mortality levels are identical between mianserin-treated and untreated adults at the start of the experiment . Any difference in mortality levels observed from day 1 onwards must therefore be the result of a change in mortality rate by mianserin . Plotting a mortality curve for over 3 , 000 mianserin-treated or untreated animals showed a significantly lower mortality level for mianserin-treated animals by day 12 ( Figure 6b , Figure 6—figure supplement 1a ) . Therefore , mianserin treatment decelerated the rise in mortality levels between day 1 and 12 of adulthood . From then on , the mortality curves were parallel showing a 7–8 day shift in mortality across the remaining lifespan . The parallel nature suggested that mianserin did not affect mortality rates past day 12 and that its effect on lifespan was restricted to the period of early adulthood ( Figure 6b , Figure 6—figure supplement 1a ) ( Mair et al . , 2003; Vaupel , 2010 ) . Power calculations confirmed that these mortality curves were sufficiently powered to detect a one day difference in lifespan in over 90% of the experiments ( α=0 . 01 ) ( Figure 6—figure supplement 1b ) ( Ye et al . , 2014 ) . These results further supported a model in which mianserin treatment specifically lowered age-associated change in early adulthood , causing a shift in physiology and mortality that can be observed in transcriptomes by day 10 . We reasoned that if the effect of mianserin on lifespan precedes the onset of mortality and is completed by day 10 , mianserin treatment beyond day 10 should be dispensable . Alternatively , if mianserin still influenced mortality later in life , shorter exposures would lead to a shorter lifespan extension compared to a lifelong exposure . We therefore limited mianserin exposure to 8 hr , 1 , 5 , 10 and 15 days and compared their lifespan with animals treated for the entire life ( Figure 6c , d ) . Exposing the animals for 5 or 10 days was sufficient to extend lifespan to the same extent as lifelong exposure ( Figure 6c , d ) . Shorter exposures ( 8 hr , 1 day ) also extended lifespan , but not by as much , showing that removing mianserin from the culture is an effective means to restrict its action ( Figure 6c , d ) . Taken together , these results are most consistent with a model in which mianserin specifically lowers the rate of age-associated change during the first few days of adulthood , thereby extending their longevity ( Figure 6e ) and postponing the onset of mortality . While the change in age-associated mortality rate during early adulthood is too small to be accuratly determined , when we measured drift-variance , it allowed us to monitor the age-associated change in the transcriptome during early adulthood ( Figure 6e , f ) . Since the effect of mianserin in early adulthood overlapped with the reproductive period ( first 5 days of adulthood ) , we asked whether mianserin treatment increased reproductive lifespan as has been observed in tph-1 ( mg280 ) mutants ( Sze et al . , 2000 ) . Mianserin treatment blocks serotonin-induced egg-laying ( Petrascheck et al . , 2007 ) , but had a minor effect on amount or timing of spontaneous egg-laying and brood size ( Figure 6g ) . Most importantly , mianserin did not increase reproductive longevity ( Figure 6g ) . We further considered the possibility that mianserin acted by a mechanism similar to lifespan extension by germline ablation ( Figure 6h ) . Two previous findings suggested otherwise: i ) Lifespan extension by germline ablation depends on daf-16 , while mianserin does not ( Arantes-Oliveira et al . , 2002; Petrascheck et al . , 2007 ) ; ii ) germline ablation increases lifespan of eat-2 ( ad1116 ) mutants while mianserin does not ( Crawford et al . , 2007 ) . We measured whether mianserin treatment mimicked the increased proteasome activity observed in glp-1 mutants ( Vilchez et al . , 2012 ) ( Figure 6h ) . A 24 hr mianserin treatment did not increase the proteasome activity , as measured by a fluorescence-based assay for chymotrypsin-like activity . On day 5 , mianserin slightly decreased proteasome activity , consistent with a slight increase in drift-variance in proteasome-related genes ( Figure 6h; Figure 6—figure supplement 1c ) . We concluded that mianserin specifically lowers the rate of age-associated change in somatic tissues and does not involve a mechanism directly related to the germline . Our data demonstrate that changes in drift-variance provide a metric for aging that correlates with mortality in C . elegans . To test whether drift-variance also increases with age in mammals , we re-analyzed published gene expression data-sets obtained from aging mouse tissues , aging human brains , and from fibroblasts derived from Hutchinson-Gilford progeria syndrome patients ( Figure 7 ) ( Lu et al . , 2004; Liu et al . , 2011; Jonker et al . , 2013 ) . We calculated drift-variances from brain , kidney , liver , lung , and spleen based on gene expression data-sets from mice aged 13 , 26 , 52 , 78 , 104 and 130 weeks . We calculated drift-variances using 13-week-old mice as a young reference ( see Methods ) and pooled mice into age-bins of 30 , 60 and 100 weeks to reduce variability . Drift-variance increased in all tissues with age ( Figure 7a ) . Compared to the drift-variance changes observed in C . elegans ( Figure 2a ) , these changes however were small . Because the 13-week-old mice were used as reference for young age ( see methods ) , the drift-variance in the 30-week-old group including the 13-week-old sample is artificially low ( Figure 6a , see material and methods ) . To better reflect the actual variance of the 30-week-old group , we set aside the data of one 13-week-old mouse to use as a young reference and recalculated drift-variances for all samples ( Figure 7b ) . This strategy has the advantage that we can observe the real drift-variance for the 30-week-old group by excluding the reference data-set , but has the disadvantage that the results are less robust as they all depend on a single reference sample . Plotting drift-variance for each organ as a function of age confirmed that as mice age , drift-variance increases in all organs ( Figure 7b ) . It will be interesting to learn if the different rates by which drift-variance increases in different organs will also be observed in other data-sets . We re-analyzed the data from Lu et al . that recorded gene expression profiles from 32 human brains aged 26 to 106 years of age ( frontal cortex ) ( Figure 7c ) ( Lu et al . , 2004 ) . For the first plot , we binned the data into 20-year bins and calculated the overall drift-variance for each 20-year bin . As a young age reference , we used the mean gene expression of adults below 30 ( 26 , 26 , 27 , and 29 ) ( see Materials and methods ) . This analysis shows that over the entire population , drift-variance remains relatively stable until the age of sixty , and then starts to rise ( Figure 6c ) . We also plotted the drift-variance of each individual as a function of age . This revealed a significant correlation ( Spearman , rho=0 . 6 , P=0 . 0014 ) between age and drift-variance in the human brain . Irrespective of the age of the mother , the aging process starts afresh for each new generation . We therefore hypothesized that aging must be reversed with each new generation and asked whether it is possible to reverse increases in drift-variances . To address this question , we re-analyzed the data-set generated by Liu et al . who derived induced pluripotent stem cells ( iPSCs ) from fibroblasts of healthy controls ( BJ ) and patients suffering from Hutchinson-Gilford progeria syndrome ( HPGS ) , an accelerated aging syndrome ( Figure 7e ) ( Liu et al . , 2011 ) . As a young-reference to calculate drift-variance , we used human H9 embryonic stem cells ( ESC ) . As expected for a premature aging syndrome , fibroblasts from HGPS patients showed increased drift-variance relative to BJ control fibroblasts ( Figure 7e ) . Furthermore , nuclear reprogramming reduced drift-variance in iPSCs to levels closer to those seen in H9 embryonic stem cells . Thus , increases in drift-variance are reversed by nuclear reprogramming in vitro Aging has been shown to cause DNA damage , degeneration of the nuclear architecture , loss of histones , loss of histone modification ( Kaeberlein et al . , 1999; Scaffidi and Misteli , 2006; Burgess et al . , 2012 ) . These changes contribute to the degenerative phenotypes observed with aging ( Mostoslavsky et al . , 2006; Feser et al . , 2010; Peleg et al . , 2010 ) . In the present study , we used expression patterns of young adults as a reference to monitor the aging process across the transcriptome . We found that aging causes the expression of genes within functional groups to drift apart , causing a loss of co-expression patterns as observed in young adults . We quantified this phenomenon using drift-variance , defined as the variance in gene expression among genes . It is important to distinguish transcriptional noise , which measures the variance of the same genes among samples ( Bahar et al . , 2006 ) , from transcriptional drift , which measures variance among genes within the same samples . At present it is unclear whether transcriptional drift is the consequence of a regulated program or of degenerative changes in the nucleus that lead to a loss of transcriptional control . Consistent with a regulated program are recent findings that the germline actively represses the activation of heat shock promoters via histone methylation , causing a decline in heat shock capacity ( Labbadia and Morimoto , 2015b ) . Consistent with degenerative changes are recent findings that show the loss of histone methylation to cause aberrant gene expression that increases with age leading to a transcriptional drift-like effect ( Somel et al . , 2006; Mercken et al . , 2013; Pu et al . , 2015; Sen et al . , 2015 ) . Irrespective of whether transcriptional drift is the consequence of a regulated program or a degenerative change , its effect on pathway function is likely to be detrimental . Many physiological processes depend on appropriate stoichiometry of their components . Large and persistent deviations in mRNA balance as measured by drift-variance are likely to result in stoichiometric imbalances in protein complexes , negatively affecting proteostasis as has been recently observed ( Houtkooper et al . , 2013; Walther et al . , 2015 ) . Our results modulating drift-variance for redox genes via mianserin and SER-5 certainly suggest that the age-associated increases in drift-variance are associated with regulatory decline ( Figures 3 , 5 ) . Attenuation of transcriptional drift in the redox system was associated with an improved homoestatic capacity , i . e . an improved ability of the redox system to appropriately respond to outward stimuli . Transcriptional drift also provided a useful concept to analyze aging transcriptomes . Accounting for its effects dramatically simplified what was an initially excessively complex expression pattern ( Figure 1 ) . Excluding gene expression changes due to drift left a set of genes that changed expression in response to mianserin treatment that was enriched for genes related to stress , innate immunity , aging and the xenobiotic response . With the exception of the xenobiotic response , which is expected to be triggered by addition of a foreign substance such as mianserin ( Figure 2f ) , all other functions have been linked to serotonin signaling ( Table 1 ) ( Zahn et al . , 2006; Petrascheck et al . , 2007; Rangaraju et al . , 2015a ) . Further , in accordance with the hypothesis that increases in drift-variance are a signature of aging in the transcriptome , we find that drift-variance is attenuated by two longevity mechanisms ( mianserin and daf-2 RNAi ) across large sections of the transcriptome . Many of the age-associated changes that were reversed by mianserin were also reversed by daf-2 RNAi ( 58% ) . This overlap is consistent with chemical epistasis experiments . Treating daf-2 ( e1370 ) mutants with mianserin causes only a partial extension of lifespan ( 11% instead of 31% ) ( Petrascheck et al . , 2007 ) consistent with the idea that many of the genes attenuated by mianserin treatment are already attenuated in daf-2 ( e1370 ) mutants and thus do not further contribute to a lifespan extension . It should be noted that age-associated increases in drift-variance do not contradict the idea that transcription factors regulate longevity . Activation of DAF-16 target genes by daf-2RNAi prevent age-associated drift of thousands of genes , thus resulting in a net decrease of drift , even though a transcriptional program has been induced ( Figure 2g ) . Our experiment did not address the questions whether increasing drift-variance beyond what occurs naturally with age accelerates aging and whether attenuation of transcriptional drift-variance is universal to all longevity mechanisms . At this point , it is prudent to mention possible pitfalls associated with transcriptional drift analysis . Drift-variance calculations require data-sets that include multiple ages ( 3 or more ) as direct statistical comparisons to the young-reference are not permissible . Furthermore , in the context of GO annotations , it is important to realize that if a given GO annotation contains significant numbers of mis-annotated genes , these genes may change expression in a different direction giving the erroneous impression of transcriptional drift . To account for these effects in our study , we i ) used the experimentally determined oxidative stress signature derived from Olivera et al ( Figure 3e ) , and ii ) used a robust Levene’s test to determine statistical differences . The robust Levene’s test uses a 10% trimmed mean , which removes large outliers such as those that would be expected by mis-annotation . These safeguards , however , are only effective if the number of mis-annotated genes is small relative to the total number of genes . Conceptually , transcriptional drift is not a biomarker for aging . It is a metric for aging similar to lifespan measurements that can be used to monitor age-associated physiological changes on the molecular level within groups of genes . Lifespan measurements record the fraction of organisms alive in different cohorts at any given time to compare rates of aging , while drift-variance allows a similar comparison based on transcriptional drift-variance . What made drift-variance measures essential for the present study was that it allowed us to monitor age-associated physiological changes in young animals , at a time when age-associated mortality levels are too low to be accurately determined ( see below ) . Measuring lifespan of mianserin-treated and untreated C . elegans revealed a mean lifespan extension of 7–8 days ( Figure 2 ) . Lifespan measurements detect differences after the majority of the animals have died and make no statements about the period during which the relevant physiological events that lead to an increase in lifespan occur ( Figure 2c , e ) ( Mair et al . , 2003; Partridge and Gems , 2007 ) . The finding that transcriptional drift values in mianserin-treated animals already showed a 7–8 day delay in physiological change as early as day 10 suggested a model in which the physiological events responsible for the 7–8 days lifespan extension take place ( and conclude ) prior to day 10 ( Figure 2a , 6a , e ) . Determining mortality levels at different ages confirmed this model . Mianserin or water is added on day 1 of adulthood to the same preparation of N2 animals . The mortality levels of both cohorts ( water , mianserin ) are therefore identical at the start of the experiment . Thus , the lower mortality level observed on day 12 in mianserin-treated animals is the result of a lower mortality rate prior to day 12 ( Figure 6b ) . Furthermore , mianserin ceases to affect mortality rates past day 12 as evident by highly parallel mortality curves ( Figure 6b ) . As with the results obtained with drift measurements , the most plausible explanation is that mianserin treatment specifically decelerates the rise in mortality in young adults leading to a lower mortality level sometime between day 10 to day 12 that persists throughout life , ultimately revealing itself in a 7–8 day lifespan extension ( ~30–40% increase in lifespan ) ( Figure 6b ) . Analysis of drift-variance , PCA , mortality and survivorship independently arrive at the same 7–8 days delay in physiology , either measured as a feature of transcriptomes or by recording death times . All methods suggest that the delay is completed before day 10 or 12 and therefore occurs during early adulthood . We further experimentally confirmed this suggestion by showing that treatment for the first five or ten days of life was necessary and sufficient to achieve the same lifespan extension observed with lifelong treatment ( Figure 6c , d ) . Even though this period exactly overlaps with the reproductive period , the effect of mianserin appears to be specific to somatic tissue ( Figure 6g , h ) . In contrast to germline ablation , mianserin extends lifespan of daf-16 mutants but not of eat-2 mutants ( Crawford et al . , 2007; Petrascheck et al . , 2007; Vilchez et al . , 2012 ) and does not increase proteasome activity as observed in glp-1 mutants ( Figure 6h ) . It is still possible that the mianserin-induced lifespan extension interacts or depends on the germline , but if it does , the connection is more indirect potentially similar to what has been observed for dietary restriction ( Crawford et al . , 2007 ) . Lifespan extension mechanisms that decelerate the rate of mortality are generally interpreted as slowing the aging process , while a parallel shift as the one we observe with mianserin is interpreted as a constant risk factor that causes a proportional shift in the overall risk of death ( Mair et al . , 2003; Harrison et al . , 2009; Vaupel , 2010; Kirkwood , 2015 ) . Our data do not challenge any of these prior interpretations , but add a further possibility . Parallel shifts may also be brought about by a period extension in which the rate of age-associated physiological change is specifically lowered in young adults . Age-associated mortality in young adults is very low compared to extrinsic mortality factors and thus changes in age-associated mortality rates are difficult to reliably determine ( Partridge and Gems , 2007; Beltran-Sancheza et al . , 2012 ) . Specific changes in mortality rates during early adulthood therefore can go unnoticed but manifest themselves later as parallel shifts at the time when age-associated mortality levels are sufficiently high to be reliably determined . Whether the attenuation of physiological changes specific to young adults that affects later mortality , as seen for mianserin , is the equivalent of slowing aging in young adults is a debate for the general aging community . In summary , this work describes the phenomenon of transcriptional drift and how it can be used as a metric for aging . Using this metric , we show that blocking serotonergic signals by mianserin delays age-associated physiological changes such as transcriptional drift and mortality exclusively during early adulthood , thus extending the duration of this period and postponing the onset of age-associated mortality . Analyzing the RNA-seq data in aging C . elegans , we observed dramatic changes in the transcriptome with age . We simply termed these changes ‘transcriptional drift’ , to emphasize the ambiguity of these changes . These changes could either be the result of regulated changes as part of a biological program , or caused by a progressive loss of transcriptional control with age . Note that a progressive loss of transcriptional control does not necessarily have to result in random changes . A gene that is continuously activated in young animals may be less activated in older animals due to a progressive functional decline in the transcriptional machinery . Thus , a gradual loss of transcriptional control would cause an age-associated decline in expression of that gene in a non-random fashion . Conversely , repressive chromatin is lost with age leading to increases in transcription that are repressed in young animals . As most physiological processes depend at least to some degree on transcriptional regulation , we propose that expression changes of genes within the same pathway that go into opposing directions ( drift-variance increases ) are detrimental for the functionality of the pathway ( as seen for redox pathways in Figure 3b ) . These changes may also allow us to indirectly track the functional decline by measuring transcriptional drift . Transcriptional drift ( td ) is the change in transcript level of a gene at a given age from its level in young animals ( “young reference” ) . As all the subsequent calculations depend on the age chosen for “young reference” we made sure to indicate the age used as a “young reference” for each plot ( see below ) . For all the C . elegans work , the “young reference” age was day 1 , at the onset of reproductive maturity in adulthood . For any gene x , transcriptional drift ( td ) is defined as ( Equation 1 ) . ( 1 ) tdgene x = ( No . of transcriptsage[t]No . of transcriptsyoung reference ) or , which is the same as ( 2 ) tdgene x= ( cpmage[t]cpmyoung reference ) where , ‘cpm’ stands for counts per million; ‘t’ stands for time in days , weeks or years , dependent on the organism . Equation 1 normalizes the level of transcription for all genes to 0 for a young animal . Note: If several biological replicates are available for the age of the young reference , a variance for the young age can be calculated ( see the section below titled ‘Variance for “the young reference”’ ) . To evaluate changes in co-expression , we calculated the drift-variance ( dv ) ( Equation 3 ) over a group of n genes with transcriptional drift-values ranging from tdi=1 to tdn . ( 3 ) drift variance=1n−1∑i=1n ( tdi−td¯ ) 2 Thus , if genes maintain a youthful co-expression pattern , drift-variance stays relatively small . If large fractions of genes within a GO or an entire transcriptome change expression in opposing directions , the drift-variance increases , suggesting a loss of youthful co-expression patterns as shown in Figure 1h , i . If multiple replicate data-sets for the “young reference” age are available , it is possible to plot drift-variance for the young reference as well . There are two ways to incorporate multiple “young reference” data-sets , each of which has its advantages or disadvantages . Method #1 uses all “young reference” samples to calculate a mean gene expression level for each individual gene to generate the “young reference” values for Equation 1 . Method #1 will result in a drift-variance for the “young reference” age as well , but this drift-variance is too small and should not be used for statistical comparisons due to circular referencing . The advantage of method #1 is that the results for all subsequent ages are more robust as the inclusion of several “young reference” samples thereby reducing the overall noise ( used in Figures 2a , g , 3b , 7a , c , e ) . Method #2 allows calculating a real drift-variance value for young animals by setting aside one or several samples as the “young reference . ” These samples are only used as references and therefore do not contribute to the drift-variance in each plot . For the remaining experimental replicates of the same age , transcriptional drift is then calculated using Equation 1 without including any of the “young reference” samples . ” This will result in a drift-variance greater than 0 for the youngest age and show how much drift varies between young animals . Method #2 has the disadvantage that if there are only few young reference samples are available , and only one is used as a young reference , all values of the graph depend on a single reference sample . We used this method #2 to calculate the variances for Figure 7b , d . The case of 7d was ideal as there were 4 samples less than 30 years of age which were set aside as reference and that allowed us to calculate the “young reference”-mean over all 4 samples . As drift-variances for these 4 samples are artificially low due to self referencing they were excluded from the plot . Ideally , an experiment would have 4–6 gene expression replicates for the “young reference” age , in which case , half of them could be used as references , the others as experimental samples . How transcriptional drift and variance relate to measures like fold-changes in transcription is shown in Supplementary Figure 2a–d . To determine whether the differences in variance were statistically different , we used the Brown-Forsythe version of the Levene’s test , as implemented in STATA software . Figure: 1g: Volcano plot used mean cpm values from all three biological replicates . The 0 line ( young reference , day 1 expression , yellow line ) indicates the expected expression level for young day 1 adult animals . Black: Each dot represents one of for the 3 , 367 genes that significantly change expression with age between day 1 and day 3 . The -log10 ( P-value ) of the P-value comparing day3 water vs day 1 water is shown as a function of the the log2 ( cmp day 3 water / cpm day1 water ) . Blue: Same 3 , 367 genes as above . However the -log10 ( P-value ) comparing day3 mianserin vs day 1water is shown as a function of the the log2 ( cmps day 3 mianserin / cpm day1 water ) . Note: both data-sets ( black and blue ) use identical y- coordinates to demonstrate the reduction in age-associated changes upon mianserin-treatment . ( cpm stands for: counts per million ) . Young Reference: To obtain a ‘young reference’ value for each individual gene the mean expression level across all three biological replicates of young day 1 old water-treated C . elegans animals was calculated . Figure 1h , i: Drift plots for genes involved in oxidative phosphorylation ( KEGG pathway: cel 00190 ) and the lysosome ( KEGG pathway: cel 04142 ) . Only one out of three replicates was used to generate these plots . Transcriptional drift for oxidative phosphorylation and lysosomal genes ( line graphs ) was calculated using Equation 1 and plotted as a function of C . elegans age ( gray lines ) . At each age , the transcriptional drift-variance across all genes within the pathway was calculated using Equation 2 and plotted as Tukey-style box plots omitting outliers . Tukey plots were superimposed over the line graphs . See Equation 1 , 3 . Outliers were only omitted for graphical purposes but not for statistical testing ( robust Levene’s test ) . The lines for each gene were included in these two plots , superimposed on the Tukey-style box plot to illustrate the significance and utility of the box plots in visualizing transcriptional drift . Young reference: As a “young reference” value for each individual gene , the expression level of young day 1 old water-treated C . elegans animals was used . Only replicate #1 of our data-set was used . Figure 2a: Drift plots for all 19 , 196 genes in our data-set of water-treated control and mianserin-treated animals . Tukey plots show drift-variance calculated for the entire transcriptome ( Equation 3 ) . See Equation 1 , 3 . Outliers were only omitted for graphical purposes , but not for statistical testing ( robust Levene’s test ) . Young reference: To obtain a “young reference” value for each individual gene , the mean expression level across all three biological replicates of young day 1 old water-treated C . elegans animals was calculated . Figure 2b: Drift plots show transcriptional drift on day 5 for 19 , 196 genes as a function of mianserin concentration . For each concentration , drift-variances were calculated for 5-day-old animals that were treated with increasing concentrations of mianserin on day 1 , and plotted as Tukey-style box plots as a function of mianserin concentrations , excluding outliers . Outliers were only removed for graphical purposes but not for statistical testing ( robust Levene’s test ) . Young reference: To obtain a “young reference” value for each individual gene , the mean expression level across all three biological replicates of young day 1 old water-treated C . elegans animals was calculated . Figure 2d: Drift plots show transcriptional drift on day 10 of adulthood for 19 , 196 genes as a function of age when mianserin-treatment was started . Tukey plots show drift-variance calculated for the entire transcriptome on day 10 ( Equation 3 ) as a function of age at which mianserin-treatment was initiated . Young reference: To obtain a “young reference” value for each individual gene , the mean expression levels across all three biological replicates of young day 1 old water-treated C . elegans animals was calculated . Figure 2f: Log2 fold changes in expression for each gene shown in the y-axis were calculated by the formula: y = log2 ( cpm treatment day 10/cpm water day 1 ) . Figure 2g: The data from Murphy et al . were dowloaded from the Princeton Puma database . Expression values were calculated using the following variables in the data-set: expression value = ch1netmean/ch2normalizednetmean . Drift plots for control- RNAi , daf-2 ( RNAi ) treated and daf-16 ( RNAi ) ; daf-2 ( RNAi ) treated animals were plotted as transcriptional drift-variance as a function of C . elegans age . To plot drift-variance for the entire transcriptome as function of age in days , we binned the data as follows . Day 0 ( 8 hr ) , day 1 ( 24 hr ) , day 2 ( 28 hr , 40 hr , 52 hr ) , day 4 ( 72 hr , 96 hr ) , day 6 ( 144 hr , 196 hr ) . Young reference: As a “young reference” value for each individual gene we used the expression level at 8 hr of age . The young reference was determined for each RNAi treatement specifically ( control RNAi , daf-16 ( RNAi ) ; daf-2 ( RNAi ) , daf-2 ( RNAi ) . Figure 3e: The log fold gene expression with age was calculated for each of the 252 genes that are known to be upregulated in response to oxidative stress and for each of the 88 genes known to be downregulated in response to oxidative stress . We then performed a linear fit for each set of genes for water-treated ( gray ) and mianserin-treated ( blue ) samples . Shaded region shows the 95% confidence interval . Figure 7a , b: 7a ) Drift plots showing transcriptional drift and drift-variance in different tissues across different mouse ages . For each age , the drift-variance was calculated across the entire transcriptome ( Equation 3 ) and plotted as Tukey-style box plots omitting outliers . As only three mice were available for each age , we pooled two ages for each age bin . 7b ) Drift-variance for each tissue as a function of age . Young Reference: 7a: To obtain a “young reference” value for each individual gene , the mean expression level across all three biological replicates of young 13-week-old mice was calculated for each tissue . Young Reference 7b: To obtain “young reference” values for each individual gene , we used one single 13-week-old replicate as a “young reference” from each tissue . The data from the “young reference” did not contribute to the graph and thus show a real transcriptional drift-variance . Figure 7c , d: 7c ) . Drift plots showing transcriptional drift-variance in human gene expression data from frontal cortices as a function of age . For 7c , the data were pooled into 20 year bins . 7d ) Plots drift-variance calculated based on Equation 3 as a function of age for each sample individually . Young Reference: To obtain “young reference” values for each individual gene , the mean gene expression levels was calculated averaging expression levels from 4 samples aged 25 to 29 years and used as the “young reference” value in Equation 1 . Figure 2—figure supplement 1: e ) The transcriptional drift plots were constructed by using the GEO data-sets GSE21784 and GSE46051 , which are independent publicly available data-sets for aging C . elegans . f ) The transcriptional drift plots were constructed by sub-sampling the data from our RNA-seq . We randomly assigned half of all genes ( out of 19 , 196 ) to one of 10 gene-sets each containing ~1000 genes ( 5% ) and plotted the drift-variance for each set . All 10 sets look nearly indistinguishable to Figure 2a . Figure 2—figure supplement 2: f ) The drift plot was constructed by removing all the genes from our data-set that were not detected in the sterile CF512 strain , thereby removing genes likely resulting from eggs and germline . g ) The drift plot was constructed by removing all genes from our data-set that were detected by RNA-seq in isolated C . elegans eggs . k ) Gene-sets enriched in AFD neurons ( left plot ) , ASE neurons ( middle plot ) and NSM neurons ( right plot ) were used to construct drift plots based on their expression in our data-set . Principal components analysis plot ( Figure 6a ) was generated from the counts table using multidimensional scaling as implemented by the plotMDS function in the edgeR package , which computes inter-sample distances as the root-mean-square of the 500 genes with the largest log2 fold-changes between each pair of sample ( the 'leading log fold-change" ) . Solvents used to prepare stock solutions: Paraquat was dissolved in water; mianserin was dissolved either in water or DMSO as mentioned; Mirtazapine , Dihydroergotamine , LY-165 , 163/PAPP , Mirtazapine , Metergoline , Ketanserin , Methiothepin , and Amperozide were dissolved in DMSO; FUDR was dissolved in S-complete ( Table 9 ) . 10 . 7554/eLife . 08833 . 029Table 9 . List of small molecules and chemicals used in this study with informationDOI: http://dx . doi . org/10 . 7554/eLife . 08833 . 029Molecule nameCAS numberCatalog numberManufacturerMianserin HCl21535-47-70997TocrisMirtazapine85650-52-8M3368LKT LaboratoriesDihydroergotamine mesylate6190-39-20475Tocris/R&D systemsLY-165 , 163/PAPP1814-64-8S009SigmaMirtazapine61337-67-5M3368LKT labsMetergoline17692-51-2M3668SigmaKetanserin tartarate83846-83-7S006SigmaMethiothepin mesylate74611-28-2M149SigmaAmperozide HCl86725-37-32746Tocris/R&D systemsParaquat ( Methyl viologen ) 1910-42-5AC227320010Acros OrganicsFUDR50-91-9F0503Sigma-AldrichDMSO67-68-5472301Sigma-Aldrich Detailed descriptions of all strains used in this study are tabulated below . All strains were backcrossed at least 4 times with the N2 Bristol strain . All strains were maintained as described in ( Brenner , 1974 ) . The strains with name starting with VV were generated by outcrossing to N2 Bristol strain in our lab ( Table 10 ) . 10 . 7554/eLife . 08833 . 030Table 10 . List of mutant and fluorescent strains outcrossed and used in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 08833 . 030Strain nameGenotypeNo . of times outcrossedGene nameTransgeneAlleleParent strain ( s ) VV78unc-26 ( e205 ) IV4unc-26e205CB205VV80snt-1 ( md290 ) II4snt-1md290NM204MT15434tph-1 ( mg280 ) II4tph-1mg280MT15434DA1814ser-1 ( ok345 ) X10ser-1ok345DA1814OH313ser-2 ( pk1357 ) X4ser-2pk1357OH313DA1774ser-3 ( ad1774 ) I3ser-3ad1774DA1774AQ866ser-4 ( ok512 ) III5ser-4ok512AQ866VV130ser-5 ( ok3087 ) I4ser-5ok3087RB2277FX2647ser-5 ( tm2647 ) I0ser-5tm2647FX2647FX2654ser-5 ( tm2654 ) I0ser-5tm2654FX2654FX2146ser-6 ( tm2146 ) IV0ser-6tm2146FX2146DA2100ser-7 ( tm1325 ) X10ser-7tm1325DA2100 Lifespan assays were conducted in 96-well plates as described in ( Solis and Petrascheck , 2011; Rangaraju et al . , 2015b ) . Briefly , age-synchronized animals were cultured in S-complete media containing E . coli OP50 as feeding bacteria ( ~2 × 109 bacteria mL−1 ) in 96-well plates , such that 5–15 worms are in each well . At the L4 stage , FUDR was added to prevent animals from producing offspring . Solvent ( water or DMSO ) or small molecules were added on day 1 of adulthood , exposing the worms to control or compound treatment until the end of the assay . When used , DMSO was kept to a final concentration of 0 . 33% v/v . Live animals were scored visually , based on movement induced by shaking and application of light to each well . Animals were scored three times a week , until 95% of animals were dead in all the tested conditions . Statistical analysis was performed using the Mantel–Haenszel version of the log-rank test . Resistance to oxidative stress was determined by measuring survival of mianserin-treated and untreated worms after a 24 hr exposure to the ROS-generator paraquat ( Methyl viologen ) . Experimental worm cultures were set up as described in Lifespan assays . For dose response assays , paraquat was added to a final concentration of 0 , 25 , 50 , 75 , 100 mM on day 5 of adulthood . For paraquat time-course experiment ( Figure 3c ) , paraquat was added 3 days , 5 days , or 10 days after addition of mianserin on day 1 of adulthood . For mianserin time-course experiment ( Figure 3d ) , 50 µM mianserin was added on day 1 , day 3 or 5 of adulthood , followed by 100 mM paraquat on day 10 . For all experiments , survival of worms was assessed 24 hr after paraquat addition and expressed as the percentage of live versus total animals . Mianserin-induced changes in transcription were determined by RNA-seq . A total of 12 conditions were tested each run in three biological replicates . N2 worms were cultured in 96-well plates as described in ( Solis and Petrascheck , 2011 ) . Animals in cohort #1 were treated on day 1 with water ( solvent ) or 50 µM mianserin , and harvested on day 3 , 5 , and 10 of adulthood . Animals in cohort #2 were treated with water ( solvent control ) or mianserin ( 2 , 10 , or 50 µM ) on day 1 of adulthood and harvested on day 5 . Animals in cohort #3 were treated with water ( solvent ) or 50 µM mianserin on day 1 , day 3 and day 5 and harvested on day 10 ( See Figure 1a ) . RNA was also harvested from untreated day 1 adults , to obtain the “young reference” . Harvested animals were washed three times in ice cold Dulbecco’s phosphate buffer saline and frozen in liquid nitrogen . A parallel lifespan assay was conducted for all cohorts to ensure mianserin action . Three biological replicates were harvested for every cohort . To extract RNA , frozen worms were re-suspended in ice-cold Trizol , zirconium beads , and glass beads ( cat # 03961-1-103 and cat # 03961-1-104 ) in the ratio of 5:1:1 respectively , and disrupted in Precellys lysing system ( 6500 rpm , 3 x 10 s cycles ) followed by chloroform extraction . For RNA-seq , the extracted RNA was precipitated and purified further using Qiagen RNAeasy Mini kit columns ( cat # 74104 ) . RNA was precipitated using isopropanol and washed once with 75% ethanol . Integrity of the RNA was confirmed with a Bioanalyzer ( Agilent Technologies , Santa Clara , CA , USA ) . To prepare the library , 100 ng of total RNA per sample was processed using NuGEN Encore Complete DR RNA-seq Prep Kit ( NuGEN; San Carlos; CA , USA ) , as per manufacturer’s instructions . The libraries were sequenced using v2 sequencing chemistry in a HiSeq2000 platform ( Illumina , San Diego , CA , USA ) . A single-read sequencing approach was used with 100 cycles , resulting in reads with a length of 100 nucleotides each . Libraries containing their own index sequences were sequenced in a multiplex manner by pooling six libraries per lane . Resulting sequences were obtained after 20–30 million reads per sample . Sequence data were extracted in FASTQ format and used for data analysis . RNA-seq data were analyzed by aligning the reads to the C . elegans reference genome and transcriptome from WormBase using Tophat 2 ( Kim et al . , 2013 ) , and unambiguously mapped reads were counted for each annotated gene in each sample ( Lawrence et al . , 2013 ) . Data were normalized for sequencing depths ( counts per million , cpm ) but not for gene length as no comparisons between genes within the same sample were made . The quasi-likelihood F-test from the edgeR package ( Robinson and Oshlack , 2010; Lund et al . , 2012 ) was used to test these counts for statistically significant differential gene expression between water- and mianserin-treated samples , while controlling for expression differences between the 3 biological replicates . We performed multiple testing correction by using the Benjamini-Hochberg procedure to compute a false discovery rate ( FDR ) value for each gene , and we considered an FDR less than 10% to be significant ( Benjamini and Hochberg , 1995; Zhang et al . , 2009 ) . All qRT-PCR experiments were conducted according to the MIQE guidelines ( Bustin et al . , 2009 ) , except that samples were not tested in a bio-analyzer , but photometrically quantified using a Nanodrop . All strains were cultured in 96-well plates as described in ( Solis and Petrascheck , 2011 ) . Water ( solvent ) or mianserin were added on day 1 of adulthood and worms were harvested on day 5 . RNA was extracted as described above , followed by DNAse ( Sigma , cat # AMPD1-1KT ) treatment and reverse transcription using iScript RT-Supermix ( BIO-RAD , cat # 170–8841 ) at 42ºC for 30 min . Quantitative PCR reactions were set up in 384-well plates ( BIO-RAD , cat # HSP3901 ) , which included 2 . 5 µl Bio-Rad SsoAdvanced SYBR Green Supermix ( cat # 172–5264 ) or Kapa SYBR Fast master mix ( cat # KK4602 ) , 1 µl cDNA template ( 2 . 5 ng/µl , to final of 0 . 5 ng/µl in 5 µl PCR reaction ) , 1 µl water , and 0 . 5 µl of forward and reverse primers ( 150 nM final concentration for BIO-RAD SYBR mix and 75 nM final for Kapa SYBR mix ) ( see Table below for oligo pairs used for qRT-PCR of genes tested ) . Quantitative PCR was carried out using a BIO-RAD CFX384 Real-Time thermocycler ( 95ºC , 3 min; 40 cycles of 95ºC 10 s , 60ºC 30 s; Melting curve: 95ºC 5 s , 60ºC- 95ºC at 0 . 5ºC increment , 10 s ) . Gene expression was normalized to three reference genes , rcq-5 , crn-3 and rpl-6 , using the BIO-RAD CFX Manager software . Statistical significance was determined using Student’s t-test ( Table 11 ) . 10 . 7554/eLife . 08833 . 031Table 11 . List of oligos used for qRT-PCRDOI: http://dx . doi . org/10 . 7554/eLife . 08833 . 031Gene nameqRT-PCR forward primer ( 5’-3’ ) qRT-PCR reverse primer ( 5’-3’ ) sod-1CGTAGGCGATCTAGGAAATGTGAACAACCATAGATCGGCCAACGsod-2TTCAACCGATCACAGGAGTCGCTCCAAATCAGCATAGTCGsod-3ATGGACACTATTAAGCGCGAGCCTTGAACCGCAATAGTGsod-4ATGTGGAACTATCGGAATTGTGGGTTGAGATTGTGTAACTGGAsod-5ATGGAGACTCAACCGATGGGACCACGGAATCTCTTCCTctl-1AATGGATACGGAGCGCATACAACCTTGAGCAGGCTTGAAActl-2TGATTACCCACTGATCGAGGGCGGATTGTTCAACCTCAGctl-3CAATCTAACGGTCAACGACACCATTGGATGTGGTGAGCAGprdx-2CATTCCAGTTCTCGCTGACATGATGAAGAGTCCACGGAprdx-3GTTCCGTTCTCTTGGAGCTGCTTGTTGAAATCAGCGAGCAprdx-6GGAGAACAATGGCTGATGCATCTGAACATGGCGTTTGChsp-16 . 1ACCACTATTTCCGTCCAGCTTGACGTTCCATCTGAGCCAThsp-16 . 11ACCACTATTTCCGTCCAGCTTGACGTTCCATCTGAGCCAThsp-16 . 2TCGATTGAAGCGCCAAAGAATCTCTTCGACGATTGCCTGThsp-16 . 41TCTTGGACGAACTCACTGGATCTTGGACGAACTCACTGGAhsp-16 . 48CTCATGCTCCGTTCTCCATTGAGTTGTGATCAGCATTTCTCCAhsp-16 . 49CTCATGCTCCGTTCTCCATTGAGTTGTGATCAGCATTTCTCCAcrn-3GAATGCACTCATGAACAAAGTCTAATGTTCGACTGATGAACCGrcq-5GATGTTAGAGCTGTAATTCACTGGATCTCTTCCAGCTCTTCCGrpl-6TTCACCAAGGACACTAGCGGACAGTCTTGGAATGTCCGA Wild-type N2 worms were cultured as described ( Solis and Petrascheck , 2011 ) . Water or Mianserin 50 µM were added on day 1 and 26S proteasome activity was assayed on day 2 and day 5 using the Millipore Proteasome activity kit ( cat# APT280 ) , following manufacturer’s protocol . Equal number of worms per condition were washed off culture media using ice cold Dulbecco’s phosphate buffer saline and freshly lysed using Precellys system ( 6500 rpm , 3 x 10 s cycles ) in assay buffer ( 25 mM HEPES , pH 7 . 5 , 0 . 5mM EDTA , 0 . 05% NP-40 , and 0 . 001% SDS ( w/v ) ) . Chymotrypsin-like proteasome activity in the lysates were assessed using the Suc-LLVY-AMC substrate and fluorogenic AMC substrate cleavage was measured in 20 min intervals for 120 min . A subset of lysates were pre-incubated with Lactacystin ( 12 . 5 µM final ) to ensure specificity of AMC cleavage by 26S proteasome . The amount of cleaved AMC fragments were quantified using TECAN xfluor safire II system at excitation of 360 nm and emission of 480 nm . The resulting readings were normalized to the total protein content in the samples measured using Bradford assay . Mortality curves were generated based on the life table provided in Figure 6—figure supplement 1 , tabulating death times of 15 independent experiments performed over 5 years . Each experiment consisted of 2 cohorts ( water or 50 µM mianserin ) and each cohort consisted of ~100 worms each amounting to ~1500 worms per condition . Power of detection was determined by Monte-Carlo simulations using a parametric model with parameters derived from our survival data of a cohort of over 5 , 026 N2 animals . The power of detection plot ( Figure 6—figure supplement 1 ) shows the probability to detect a true lifespan extension with a significance level α=0 . 01 as a function of percent increase in lifespan for an experiment consisting of n animals . An accuracy of 1 day is the equivalent of a 5% increase in lifespan .

All organisms age , leading to gradual declines in the body’s systems and eventually death . How certain genetic mutations and drugs delay the effects of aging and promote survival to an older age is a question many researchers are exploring . One way this problem is investigated is by looking at how the activity – or expression – of different genes changes during aging . Scientists interested in understanding aging and longevity often study a simple worm called Caenorhabditis elegans . This worm normally lives for about three weeks , and young C . elegans are able to produce offspring within days of hatching . This accelerated life cycle allows scientists to observe the entire lifespan of the worms . Over time , experiments have shown that DNA damage , changes in behavior and changes to gene expression are all markers of aging in the worms . Now , Rangaraju et al . describe how changes in gene expression patterns that begin early in the lives of C . elegans shorten their lifespan . Specifically , in groups of genes that work together , some genes increase expression , while others decrease expression with age . This phenomenon is called “transcriptional drift” and leads to an age-associated loss of coordination among groups of genes that help orchestrate specific tasks . Rangaraju et al . show that an antidepressant called mianserin prevents transcriptional drift in many of C . elegans’ genes: young worms treated with the drug resist the effects of aging on the transcriptome and maintain coordinated patterns of gene expression for longer . Maintaining coordinated patterns of gene expression postpones the onset of age-related bodily declines and extends the life of treated worms by extending the duration of young adulthood and postponing the onset of age-associated death . The drug also appears to protect against stress-induced changes in gene expression . This suggests that some of the age-related shifts in gene expression occur when cells fail to recover normal gene expression patterns after a stressful event . Questions that remain to be investigated in future studies are whether other longevity mechanisms also extend lifespan by preserving coordinated gene expression patterns , and whether other longevity mechanisms act by extending specific periods of life .

2015

Suppression of transcriptional drift extends C. elegans lifespan by postponing the onset of mortality

Distinct anatomical and spectral channels are thought to play specialized roles in the communication within cortical networks . While activity in the alpha and beta frequency range ( 7 – 40 Hz ) is thought to predominantly originate from infragranular cortical layers conveying feedback-related information , activity in the gamma range ( >40 Hz ) dominates in supragranular layers communicating feedforward signals . We leveraged high precision MEG to test this proposal , directly and non-invasively , in human participants performing visually cued actions . We found that visual alpha mapped onto deep cortical laminae , whereas visual gamma predominantly occurred more superficially . This lamina-specificity was echoed in movement-related sensorimotor beta and gamma activity . These lamina-specific pre- and post- movement changes in sensorimotor beta and gamma activity suggest a more complex functional role than the proposed feedback and feedforward communication in sensory cortex . Distinct frequency channels thus operate in a lamina-specific manner across cortex , but may fulfill distinct functional roles in sensory and motor processes . The cerebral cortex is hierarchically organized via feedback and feedforward connections that originate predominantly from deep and superficial layers , respectively ( Felleman and Van Essen , 1991; Barone et al . , 2000; Markov et al . , 2013; Markov et al . , 2014a; Markov et al . , 2014b ) . Evidence from non-human animal models suggests that information along those pathways is carried via distinct frequency channels: lower frequency ( <40 Hz ) signals predominantly arise from deeper , infragranular layers , whereas higher frequency ( >40 Hz ) signals stem largely from more superficial , supragranular layers ( Roopun et al . , 2006; Roopun et al . , 2010; Bollimunta et al . , 2008; Bollimunta et al . , 2011; Sun and Dan , 2009; Maier et al . , 2010; Buffalo et al . , 2011; Spaak et al . , 2012; Xing et al . , 2012; Smith et al . , 2013; van Kerkoerle et al . , 2014; Bastos et al . , 2015; Haegens et al . , 2015; Sotero et al . , 2015 ) . These data have inspired general theories of the functional organization of cortex which ascribe specific computational roles to these pathways and frequency channels ( Fries , 2005; Fries , 2015; Friston and Kiebel , 2009; Wang , 2010; Jensen and Mazaheri , 2010; Donner and Siegel , 2011; Arnal and Giraud , 2012; Bastos et al . , 2012; Adams et al . , 2013; Jensen et al . , 2015; Stephan et al . , 2017 ) . In these proposals , lower frequency activity subserves feedback , top-down communication conveyed predominantly via infragranular layers , whereas high-frequency activity is predominantly carried via projections from supragranular layers and conveys feedforward , bottom-up information . However , evidence for these proposals in humans is largely indirect , and focused on visual and auditory areas ( Koopmans et al . , 2010; Olman et al . , 2012; Fontolan et al . , 2014; Kok et al . , 2016; Michalareas et al . , 2016; Scheeringa and Fries , 2017 ) . Whether it is indeed possible to attribute low and high frequency activity in humans to lamina-specific sources , throughout the cortical hierarchy , remains unclear . Here we leverage recent advances in high precision magnetoencephalography ( Troebinger et al . , 2014b; Meyer et al . , 2017a ) to address this issue directly and non-invasively across human visual and sensorimotor cortices . MEG is a direct measure of neural activity ( Hämäläinen et al . , 1993; Baillet , 2017 ) , with millisecond temporal precision that allows for delineation of brain activity across distinct frequency bands . Recently developed 3D printed head-cast technology gives us more stability in head positioning as well as highly precise models of the underlying cortical anatomy . Together , this allows recording of higher signal-to-noise ratio ( SNR ) MEG data than previously achievable ( Troebinger et al . , 2014b; Meyer et al . , 2017a ) . Theoretical and simulation work shows that these gains allow , in principle , for distinguishing the MEG signal originating from either deep or superficial laminae ( Troebinger et al . , 2014a ) , in a time-resolved and spatially localized manner ( Bonaiuto et al . , 2018 ) . Demonstrating such lamina-specificity non-invasively in healthy human participants would provide important physiological constraints to the development of theoretical accounts about the functional roles of different frequency channels , in particular with regards to the proposed mechanism of inter-regional communication in hierarchical cortical networks . Here , we employed this approach to acquire high SNR MEG data , and directly test for the proposed lamina-specificity of distinct frequency channels in human cortex . We investigated the laminar and spectral specificity of induced visual and sensorimotor activity during a visually cued action selection task . The task was designed to induce well-studied patterns of low- and high-frequency activity in visual ( Müller et al . , 1996; Hari , 1997; Fries et al . , 2001; Busch et al . , 2004; Sauseng et al . , 2005; Yamagishi et al . , 2005; Hoogenboom et al . , 2006; Thut et al . , 2006; Muthukumaraswamy and Singh , 2013; Mazaheri et al . , 2014 ) and sensorimotor cortices ( Pfurtscheller et al . , 1996; Pfurtscheller and Neuper , 1997; Crone et al . , 1998; Cheyne et al . , 2008; Donner et al . , 2009; Huo et al . , 2010; Gaetz et al . , 2011; Haegens et al . , 2011; de Lange et al . , 2013; Tan et al . , 2016; Tan et al . , 2014; Torrecillos et al . , 2015 ) . Participants first viewed a random dot kinematogram ( RDK ) with coherent motion to the left or the right , which in most trials ( 70% ) was congruent to the direction of the following instruction cue indicating the required motor response ( an arrow pointing left equated to an instruction to press the left button , and vice versa; Figure 1A ) . Participants could therefore accumulate the sensory evidence from the RDK to anticipate the likely required response in advance of the instruction cue . However , in incongruent trials , the instruction cue pointed in the opposite direction from the direction of coherent motion of the RDK , and so the opposite response from the expected one was required . The strength of the motion coherence varied between trials , thereby influencing the predictability of the instructed response ( Figure 1B; Donner et al . , 2009; de Lange et al . , 2013 ) . As expected , particpants responded more accurately and quickly during congruent trials , with additional gains in respond speed when the RDK motion coherence was strongest . By contrast , responses were generally slower and participants made more mistakes during incongruent trials ( Figure 1C , D ) . This was demonstrated by a significant interaction between congruence and coherence for accuracy ( χ2 ( 2 ) = 363 . 21 , p<0 . 001 ) , and RT ( F ( 2 , 16187 ) = 25 . 83 , p<0 . 001 ) . Pairwise comparisons ( Tukey corrected ) showed that accuracy was higher and RTs were faster during congruent trials than incongruent trials at low ( accuracy: Z = 7 . 83 , p<0 . 001; RT: t ( 16181 . 94 ) = −8 . 25 , p<0 . 0001 ) , medium ( accuracy: Z = 23 . 71 , p<0 . 001; RT: t ( 16181 . 94 ) = −13 . 94 , p<0 . 001 ) and high coherence levels ( accuracy: Z = 29 . 96 , p<0 . 001; RT: t ( 16181 . 94 ) = −18 . 39 , p<0 . 001 ) . Participants were thus faster and more accurate when the cued action matched the action they had prepared ( congruent trials ) , and slower and less accurate when these actions were incongruent . Participant-specific head-casts minimize both within-session movement and co-registration error ( Troebinger et al . , 2014b; Meyer et al . , 2017a ) . This ensures that when MEG data are recorded over separate days , the brain remains in the same location with respect to the MEG sensors . In all participants , within-session movement was <0 . 2 mm in the x and y dimensions , and <1 . 5 mm in the z dimension , while co-registration error was <1 . 5 mm in any dimension ( estimated by calculating the within-participant standard deviation of the absolute coil locations across recording blocks; Figure 2—figure supplement 1 ) . To assess the between-session reproducibility of our data , we examined topographic maps , event-related fields ( ERFs ) , and time-frequency ( TF ) decompositions for the different task epochs . These data were analyzed in three ways: aligned to the onset of the RDK ( Figure 2A ) , instruction cue ( Figure 2B ) , or button response ( Figure 2C ) . Topographic maps and event-related fields from individual MEG sensors and time-frequency spectra from sensor clusters were indeed highly reproducible across different days of recording within an individual . For the participant shown in Figure 2 , the intra-class correlation coefficient ( ICC ) , a measure of test-retest reliability , was greater than 0 . 9 for all task epochs , and the three measures used to assess reproducibility ( topographic map , RDK , mean within-session ICC = 0 . 95 , between-session ICC = 0 . 94; topographic map , instruction cue , mean within-session ICC = 0 . 94 , between-session ICC = 0 . 97; topographic map , button response , mean within-session ICC = 0 . 97 , between-session ICC = 0 . 99; ERF , RDK , mean within-session ICC = 0 . 94 , between-session ICC = 0 . 97; ERF , instruction cue , mean within-session ICC = 0 . 96 , between-session ICC = 0 . 96; ERF , button response , mean within-session ICC = 0 . 96 , between-session ICC = 0 . 98; TF , RDK , mean within-session ICC = 0 . 97 , between-session ICC = 0 . 97; TF , instruction cue , mean within-session ICC = 0 . 97 , between-session ICC = 0 . 98; TF , button response , mean within-session ICC = 0 . 99 , between-session ICC = 0 . 99 ) . Across all subjects , the mean ICC for all task epochs and reproducibility measures was greater than 0 . 85 ( topographic map , RDK , within-session ICC , M = 0 . 94 , SD = 0 . 03 , between-session ICC , M = 0 . 96 , SD = 0 . 02; topographic map , instruction cue , within-session ICC , M = 0 . 97 , SD = 0 . 03 , between-session ICC , M = 0 . 98 , SD = 0 . 02; topographic map , button response , within-session ICC , M = 0 . 96 , SD = 0 . 03 , between-session ICC , M = 0 . 95 , SD = 0 . 06; ERF , RDK , within-session ICC , M = 0 . 88 , SD = 0 . 08 , between-session ICC , M = 0 . 94 , SD = 0 . 05; ERF , instruction cue , within-session ICC , M = 0 . 93 , SD = 0 . 03 , between-session ICC , M = 0 . 94 , SD = 0 . 03; ERF , button response , within-session ICC , M = 0 . 94 , SD = 0 . 02 , between-session ICC , M = 0 . 97 , SD = 0 . 02; TF , RDK , within-session ICC , M = 0 . 95 , SD = 0 . 03 , between-session ICC , M = 0 . 97 , SD = 0 . 01; TF , instruction cue , within-session ICC , M = 0 . 96 , SD = 0 . 02 , between-session ICC , M = 0 . 98 , SD = 0 . 01; TF , button response , within-session ICC , M = 0 . 98 , SD = 0 . 004 , between-session ICC , M = 0 . 99 , SD = 0 . 004 ) . To address our main question about the laminar specificity of different frequency channels in human cortex , we first examined task-related low- and high-frequency activity from sensors overlying visual and sensorimotor cortices . Attention to visual stimuli is associated with decreases in alpha ( Hari , 1997; Sauseng et al . , 2005; Yamagishi et al . , 2005; Thut et al . , 2006; Mazaheri et al . , 2014 ) and increases in gamma activity in visual cortex ( Müller et al . , 1996; Fries et al . , 2001; Busch et al . , 2004; Hoogenboom et al . , 2006; Muthukumaraswamy and Singh , 2013 ) . In line with previous research , sensors overlying the visual cortex revealed a significant decrease in alpha ( 7–13 Hz ) and increase in gamma ( 60 – 90 Hz ) power following the onset of the RDK and lasting for its duration ( Siegel et al . , 2007 ) . In addition , we observed a burst of gamma activity following the onset of the instruction cue ( Figure 3A; significant time-frequency windows marked , p<0 . 05 , Bonferroni corrected ) . Motor responses are associated with a characteristic pattern of spectral activity in contralateral sensorimotor cortex , with a stereotypical decrease in average beta power prior to movement , followed by a rebound in average beta activity after the response . Moreover , a burst of gamma activity typically occurs around movement onset ( Pfurtscheller et al . , 1996; Pfurtscheller and Neuper , 1997; Crone et al . , 1998; Cheyne et al . , 2008; Huo et al . , 2010; Gaetz et al . , 2011 ) . At the sensor-level , we indeed observed these classic average power changes , with a significant decrease in beta power ( 15 – 30 Hz ) prior to and during the participant’s response along with a subsequent rebound , and a burst of response-aligned gamma ( 60 – 90 Hz ) activity ( Figure 3B; significant time-frequency windows marked , p<0 . 05 , Bonferroni corrected ) . These signals are relevant for testing the proposed role of low and high frequency activity , respectively , for the following reasons . First , the average beta power decrease prior to movement has been linked to various processes related to the preparation and specification of movement ( Donner et al . , 2009; Engel and Fries , 2010; Aron et al . , 2016; Khanna and Carmena , 2017; Spitzer and Haegens , 2017 ) . Moreover , gamma bursts at movement onset are thought to originate from motor cortex , are effector-specific , and are thought to reflect processes related to the feedback control of movements ( Cheyne et al . , 2008; Muthukumaraswamy , 2010 ) and updating of motor plans ( Mehrkanoon et al . , 2014 ) . However , we note that the proposed roles of pre- and post-movement beta and movement-onset gamma complicate the idea of these frequency channels conveying feedback and feedforward control , as seen in sensory cortices ( Bauer et al . , 2014; Fontolan et al . , 2014; van Kerkoerle et al . , 2014; Bastos et al . , 2015; Jensen et al . , 2015; Michalareas et al . , 2016 ) . This is because ( a ) the dynamics of beta activity occur both prior to and after the event ( i . e . , movement ) , whereas corresponding activity changes in sensory cortices are stimulus-driven; ( b ) the movement-onset gamma bursts have been linked to the initiation of movement and hence with descending corticospinal communication ( Cheyne et al . , 2008; Cheyne and Ferrari , 2013 ) ; and ( c ) motor cortex is agranular , which blurs the proposed laminar dissociation between feedback and feedforward information channels . This opens the possibility that movement-related beta and gamma activity may not be organized in the same lamina-specific manner as in sensory cortices . Alternatively , the same lamina-specific organization may have functional roles that are distinct from the proposed feedback and feedforward communication in sensory cortex . Having identified low- and high-frequency visual and sensorimotor signals at the sensor-level , we next asked whether these frequency channels indeed arise predominantly from deep or superficial cortical laminae . Localization of activity measured by MEG sensors requires accurate generative forward models which map from cortical source activity to measured sensor data ( Hillebrand and Barnes , 2002; Hillebrand and Barnes , 2003; Larson et al . , 2014; Baillet , 2017 ) . We constructed a generative model for each participant based on a surface mesh that included both their white matter and pial surfaces , respectively ( Figure 4 , left column ) . This permits comparison of the estimated source activity for visual and sensorimotor activity on the white matter and pial surface . We infer a deep ( white-matter boundary ) laminar origin if the activity in a given frequency band is strongest on the white matter surface , and a superficial ( pial surface ) origin if this activity is strongest on the pial surface . For the purposes of comparison with invasive neural recordings , the deep laminae approximate infragranular cortical layers , and superficial laminae approximate supragranular layers . The veracity of laminar inferences using this analysis is highly dependent on the accuracy of the white matter and pial surface segmentations . Imprecise surface reconstructions from standard 1 mm isotropic T1-weighted volumes result in coarse-grained meshes , which do not accurately capture the separation between the two surfaces , and thus are suboptimal for distinctions between deep and superficial laminae ( Figure 4—figure supplement 1 ) . We therefore extracted each surface from high-resolution ( 800 μm isotropic ) MRI multi-parameter maps ( Carey et al . , 2017 ) , allowing fine-grained segmentation of the white matter and pial surfaces ( Figure 4—figure supplement 1 ) . For each low- and high-frequency visual and sensorimotor signal , the laminar analysis first calculated the unsigned fractional change in power from a baseline time window ( i . e . power change from baseline divided by baseline power ) on the vertices of each surface , and then compared this fractional power change between surfaces using paired t-tests over trials ( Figure 4C , top ) . The resulting t-statistic was positive when the magnitude of the change in power was greater on the pial surface ( superficial ) , and negative when the change was greater on the white matter surface ( deep; Figure 4C , middle ) . To get a global measure of laminar specificity , we averaged this fractional change in power from baseline over the whole brain ( all vertices ) within each surface . For spatially localized laminar inference , we then identified regions of interest ( ROIs ) in each participant based on the mean frequency-specific change in power from a baseline time window on vertices from either surface ( Bonaiuto et al . , 2018 ) . We compared two metrics for defining the ROIs: functionally defined ( centered on the vertex with the peak mean difference in power ) , and anatomically-constrained ( centered on the vertex with the peak mean power difference within the visual cortex bilaterally , or in the contralateral motor cortex ) . In addition to performing paired t-tests over trials using the unsigned fractional change in power from baseline averaged within ROIs , we also examined the distribution of t-statistics across vertices by performing a paired t-test across trials for each white matter/pial vertex pair ( Figure 4C , bottom ) . Next , we asked whether the observed low and high-frequency lamina-specific activity in visual and sensorimotor cortex dynamically varied with task demands in line with proposals about their role in feedback and feedforward message passing ( von Stein et al . , 2000; Fries , 2005; Fries , 2015; Friston and Kiebel , 2009; Wang , 2010; Jensen and Mazaheri , 2010; Donner and Siegel , 2011; Arnal and Giraud , 2012; Bastos et al . , 2012; Adams et al . , 2013; Jensen et al . , 2015; Stephan et al . , 2017 ) . This would provide additional indirect support for the idea that communication in hierarchical cortical networks is organized through distinct frequency channels along distinct anatomical pathways , to orchestrate top-down and bottom-up control . In our task , the direction of the instruction cue was congruent with the motion coherence direction in the RDK during most trials ( 70% ) . As such , if the direction of motion coherence is to the left , the instruction cue will most likely be a leftward arrow . Gamma activity increases in sensory areas during presentation of unexpected stimuli ( Gurtubay et al . , 2001; Arnal et al . , 2011; Todorovic et al . , 2011 ) , and therefore we expected visual gamma activity in supragranular layers to be greater following incongruent instruction cues than after congruent cues . Indeed , the increase in visual gamma on the pial surface following the onset of the instruction cue was greater in incongruent compared to congruent trials ( W ( 8 ) =0 , p=0 . 008; 8/8 participants; incongruent % change from baseline - congruent % change from baseline M = 1 . 64% , SD = 2 . 34%; Figure 7 ) . Changes in sensorimotor beta power during response preparation predict forthcoming motor responses ( Donner et al . , 2009; Haegens et al . , 2011; de Lange et al . , 2013 ) , whereas the magnitude of sensorimotor beta rebound is attenuated by movement errors ( Tan et al . , 2014; Tan et al . , 2016; Torrecillos et al . , 2015 ) . We therefore predicted that , in infragranular layers , the decrease in sensorimotor beta would scale with the motion coherence of the RDK , and the magnitude of the beta rebound would be decreased during incongruent trials when the prepared movement has to be changed in order to make a correct response . The behavioral results presented thus far suggest that participants accumulated perceptual evidence from the RDK in order to prepare their response prior to the onset of the instruction cue . This preparation was accompanied by a reduction in beta power in the sensorimotor cortex contralateral to the hand used to indicate the response ( Figure 6A ) . This beta decrease began from the onset of the RDK and was more pronounced with increasing coherence , demonstrating a significant effect of coherence on the white matter surface ( Figure 8A; X2 ( 2 ) =9 . 75 , p=0 . 008 ) , with beta during high coherence trials significantly lower than during low coherence trials ( 8/8 participants; t ( 7 ) =-3 . 496 , p=0 . 033; low % change from baseline – high % change from baseline M = 2 . 42% , SD = 1 . 96% ) . Following the response , there was an increase in beta in contralateral sensorimotor cortex ( beta rebound ) which was greater in congruent , compared to incongruent trials on the white matter surface ( Figure 8B; W ( 8 ) =34 , p=0 . 023; 7/8 participants , congruent % change from baseline - incongruent % change from baseline M = 5 . 13% , SD = 5 . 19% ) . In other words , the beta rebound was greatest when the cued response matched the prepared response . In this study , we sought to address recent proposals about the role of distinct frequency channels of activity in hierarchical processing ( Fries , 2005; Fries , 2015; Friston and Kiebel , 2009; Wang , 2010; Jensen and Mazaheri , 2010; Donner and Siegel , 2011; Arnal and Giraud , 2012; Bastos et al . , 2012; Adams et al . , 2013; Jensen et al . , 2015; Stephan et al . , 2017 ) ; though see Haegens et al . , 2015; Halgren et al . , 2017 ) . According to these proposals , ascending ( bottom-up ) and descending ( top-down ) information processing occurs through distinct anatomical and frequency-specific channels . Whereas bottom-up information is conveyed via high frequency activity in supragranular layers , top-down information is associated with low frequency activity in infragranular layers . Currently , few studies in humans have tested these proposals , often on indirect grounds ( Koopmans et al . , 2010; Olman et al . , 2012; Fontolan et al . , 2014; Kok et al . , 2016; Michalareas et al . , 2016; Scheeringa and Fries , 2017 ) . Moreover , these studies have generally focused on sensory systems , whereas here we sought to establish the generalizability of these proposals across cortex , and therefore additionally focused on agranular sensorimotor cortex . When interpreting our results , it is therefore important to consider whether or not it is principally possible to achieve the spatial precision needed to distinguish deep versus superficial laminae activity with MEG . As MEG is a direct measure of neural activity , its spatial precision is , in principle , only limited by the signal-to-noise ratio with which data can be recorded , and the analysis techniques used to perform source localization ( Hillebrand and Barnes , 2002; Hillebrand and Barnes , 2003; Hillebrand and Barnes , 2011; Brookes et al . , 2010; López et al . , 2012; Troebinger et al . , 2014b; Meyer et al . , 2017a; Bonaiuto et al . , 2018 ) . Notably , in addition to theoretical considerations that a distinction of sources as close as 2 – 3 mm with MEG is feasible , recent MEG work on the retinotopic organization of visually induced activity provides empirical support for this precision ( Nasiotis et al . , 2017 ) . These authors quantified the smallest detectable change in source location elicited by a shift in the position of a visual stimulus , which was as low as 1 mm . We found that low frequency activity ( alpha , 7 – 13 Hz; and beta , 15 – 30 Hz ) predominately originated from deep cortical laminae , and high frequency activity ( gamma , 60 – 90 Hz ) from more superficial laminae in both visual and sensorimotor cortex . Our analysis included a built-in control: visually induced gamma after both the RDK and the instruction cue localized superficially , reinforcing the proposal that visual gamma generally predominates from superficial laminae . Moreover , laminar specificity was abolished by shuffling the sensors ( Figure 5—figure supplement 1 ) or introducing co-registration error ( Figure 5—figure supplement 2 ) , underlining the need for spatially precise anatomical data and MEG recordings . Importantly , the laminar bias of both low and high frequency signals increased monotonically as the number of trials included in the analysis increased , but this effect was weaker when the sensors were shuffled ( Figure 5—figure supplement 4 ) , and the superficial bias of all signals increased until saturation with the addition of increasing levels of white noise , but high frequency signals saturated at much lower noise levels and the superficial bias became unstable with increasing noise levels ( Figure 5—figure supplement 5 ) . These results suggest that the more superficial localization of gamma signals was not simply due to a trivial relationship between laminar bias and SNR . Additionally , we established that our results were not simply driven by the relative strength of the pial and white matter surface lead fields . While we found a correlation between relative lead field strength and laminar preference ( Figure 5—figure supplement 6A ) , this relationship was constant across frequency bands ( Figure 5—figure supplement 7 ) , and the laminar dissociation held at the single participant level when considering only vertex pairs matched for lead field strength ( Figure 5—figure supplement 6B ) . Moreover , the deep laminar preference of low frequency signals was preserved even when considering only vertex pairs where the white matter vertex was closer to the scalp than the pial vertex ( Figure 5—figure supplement 9 ) . These results suggest that our main analyses were sensitive to the likely source of low- and high-frequency signals ( rather being simply dependent on the relative magnitude of the influence of source activity from the pial versus white matter surface on the MEG sensors ) . However , while the slope of the relationship between relative lead field strength and laminar preference was constant across frequency bands , for gamma signals , this regression fit had an offset of approximately zero ( Figure 5—figure supplement 7 ) . Moreover , the laminar preference of sensorimotor gamma within the anatomically constrained ROIs reversed when considering only vertex pairs in which the white matter vertex was closest to the scalp . Given these issues , the conservative conclusion would be that visual and sensorimotor gamma localize more superficially than visual alpha and sensorimotor beta . One possible confound in our analysis is the estimate of sensor noise . We assumed this to be diagonal . However subsequent tests , based on independent data recorded during a similar time-period , showed off-diagonal structure ( Figure 5—figure supplement 12 ) . Although this structure was the same across frequency bands it will have affected the free energy optimization stage . However , when using a sensor covariance matrix based on empty room noise measurements , the same pattern of laminar preference was observed ( Figure 5—figure supplement 13C ) . The localization of alpha activity to predominately deep laminae of visual cortex is in line with evidence from depth electrode recordings in visual areas of the non-human primate brain ( Maier et al . , 2010; Buffalo et al . , 2011; Spaak et al . , 2012; Xing et al . , 2012; Smith et al . , 2013; van Kerkoerle et al . , 2014 ) . Several studies have found alpha generators in both infra- and supragranular layers in primary sensory areas ( Bollimunta et al . , 2008; Bollimunta et al . , 2011; Haegens et al . , 2015 ) , and it has been suggested that this discrepancy is due to a contamination of infragranular layer LFP signals by volume conduction from strong alpha generators in supragranular layers ( Haegens et al . , 2015; Halgren et al . , 2017 ) . This is unlikely to apply to the results presented here as this type of laminar MEG analysis is biased toward superficial laminae when SNR is low ( Bonaiuto et al . , 2018 ) . However , this analysis is binary ( deep or superficial ) and will be biased toward the region of highest power change , even if the true source distribution populates multiple depths ( Bonaiuto et al . , 2018 ) . We found that gamma activity was strongest in more superficial sources , confirming invasive recordings showing gamma activity arising predominantly from supragranular layers in visual cortex ( Buffalo et al . , 2011; Spaak et al . , 2012; Xing et al . , 2012; Smith et al . , 2013; van Kerkoerle et al . , 2014; but see Nandy et al . , 2017 ) . The mechanisms underlying the generation of gamma activity are diverse across the cortex ( Buzsáki and Wang , 2012 ) , but commonly involve reciprocal connections between pyramidal cells and interneurons , or between interneurons ( Tiesinga and Sejnowski , 2009; Whittington et al . , 2011 ) . The local recurrent connections necessary for such reciprocal interactions are most numerous in supragranular layers ( Buzsáki and Wang , 2012 ) , as are fast-spiking interneurons which play a critical role in generating gamma activity ( Cardin et al . , 2009; Sohal et al . , 2009; Carlén et al . , 2012 ) . It is hypothesized that the laminar segregation of frequency channels is a common organizing principle across the cortical hierarchy ( Wang , 2010; Arnal and Giraud , 2012; Bastos et al . , 2012; Fries , 2015 ) . However , most evidence for this claim comes from depth electrode recordings in primary sensory areas , with the vast majority in visual cortical regions ( Buffalo et al . , 2011; Spaak et al . , 2012; Xing et al . , 2012; Smith et al . , 2013; van Kerkoerle et al . , 2014 ) . While in vivo laminar data from primate sensorimotor cortex are lacking , in vitro recordings from somatosensory and motor cortices demonstrate that beta activity is generated in neural circuits dominated by infragranular layer V pyramidal cells ( Roopun et al . , 2006; Roopun et al . , 2010; Yamawaki et al . , 2008 ) . By contrast , gamma activity is thought to arise from supragranular layers II/III of mouse somatosensory cortex ( Cardin et al . , 2009; Carlén et al . , 2012 ) . The results presented here support generalized theories of laminar organization across cortex , and are the first to non-invasively provide evidence for the laminar origin of movement-related sensorimotor activity . We found that visual gamma was enhanced following the presentation of the instruction cue in incongruent compared to congruent trials . This was in agreement with our predictions , based on the fact that supragranular layer gamma activity is implicated in feedforward processing ( van Kerkoerle et al . , 2014 ) . In our task , the direction of coherent motion in the RDK was congruent with the direction of the following instruction cue in most trials . Participants could therefore form a sensory expectation of the direction of the forthcoming instruction cue , which was violated in incongruent trials . The enhancement of visual gamma following incongruent cues is therefore consistent with the gamma activity increase observed in sensory areas during perceptual expectation violations ( Gurtubay et al . , 2001; Arnal et al . , 2011; Todorovic et al . , 2011 ) as well as layer-specific synaptic currents in supragranular cortical layers during performance error processing ( Sajad et al . , 2017 ) . There are numerous theories for the computational role of beta activity in motor systems . Decreases in beta power prior to the onset of a movement predict the selected action ( Donner et al . , 2009; Haegens et al . , 2011; de Lange et al . , 2013 ) , whereas the beta rebound following a movement is attenuated by both perturbation-induced movement errors and target errors induced by goal displacement ( Tan et al . , 2014; Tan et al . , 2016; Torrecillos et al . , 2015 ) . Our results unify both of these accounts , showing that the level of beta decrease prior to a movement is modulated by the accumulation of sensory evidence predicting the cued movement , while the beta rebound is diminished when the prepared action must be suppressed in order to correctly perform the cued action ( corresponding to a shift in reach target used by Torrecillos et al . , 2015 ) . While our results cannot directly distinguish between feedback and feedforward processes because we did not assess interactions between brain regions ( Bastos et al . , 2015; Michalareas et al . , 2016 ) , they suggest that in the sensorimotor system , low frequency activity can reflect both bottom-up and top-down processes depending on the task epoch . This may occur via bottom-up , feedforward projections from intraparietal regions to motor regions ( Platt and Glimcher , 1999; Hanks et al . , 2006; Tosoni et al . , 2008; Kayser et al . , 2010 ) or top-down , feedback projections from the dorsolateral prefrontal cortex ( Heekeren et al . , 2004; Heekeren et al . , 2006; Curtis and Lee , 2010; Hussar and Pasternak , 2013; Georgiev et al . , 2016 ) . The dissociation between bottom-up and top-down influences during different task epochs could indicate that the decrease in beta and the following rebound are the result of functionally distinct processes . Our ROI-based comparison of deep and superficial laminae can only determine the origin of the strongest source of activity , which does not imply that activity within a frequency band is exclusively confined to either deep or superficial sources within the same patch of cortex ( Maier et al . , 2010; Bollimunta et al . , 2011; Spaak et al . , 2012; Xing et al . , 2012; Smith et al . , 2013; Haegens et al . , 2015 ) . We should also note that in all of our control studies , in which we discard spatial information , a bias towards the superficial ( pial ) cortical surface was present . However , this bias does not increase with SNR for high frequency activity with poor anatomical models ( Figure 5—figure supplement 4 ) , mirroring the results of simulations showing that this type of laminar analysis is biased superficially at low SNR levels ( Bonaiuto et al . , 2018 ) . Moreover , we used white matter and pial surface meshes to represent deep and superficial cortical laminae , respectively , and therefore made no attempt to explicitly account for activity arising from the granular layers . Recent studies have shown that beta , and perhaps gamma , activity is generated by stereotyped patterns of proximal and distal inputs to infragranular and supragranular pyramidal cells ( Lee and Jones , 2013; Jones , 2016; Sherman et al . , 2016 ) . Finally , a new generation of wearable MEG sensors , optically pumped magnetometers ( Boto et al . , 2016; Boto et al . , 2017; Boto et al . , 2018 ) , promises to extend the reach of laminar MEG . These sensors do not require cryogenic cooling and can therefore be placed directly on the scalp surface , directly increasing SNR . This allows participants to make relatively unconstrained and natural movements; future such systems , which were comfortable to wear , would give the possibility of further augmenting the SNR by recording over much longer periods ( Boto et al . , 2018 ) . Such flexibility in participant behavior opens the door to the possibility of testing theories about the changes in hierarchical communication in the brain , either developmentally or in patient populations such as those with movement disorders , autism spectrum disorders and schizophrenia ( Wang , 2010; Wilson et al . , 2011; Gandal et al . , 2012; Wright et al . , 2012; Chan et al . , 2016; Kessler et al . , 2016; Liddle et al . , 2016 ) . Eight neurologically healthy volunteers participated in the experiment ( six male , aged 28 . 5 ± 8 . 52 years ) . The study protocol was in full accordance with the Declaration of Helsinki , and all participants gave written informed consent after being fully informed about the purpose of the study . The study protocol , participant information , and form of consent , were approved by the UCL Research Ethics Committee ( reference number 5833/001 ) . Participants completed a visually cued action decision making task in which they responded to visual stimuli projected on a screen by pressing one of two buttons on a button box using the index and middle finger of their right hand . On each trial , participants were required to fixate on a small ( 0 . 5°×0 . 5° ) white cross in the center of a screen . After a baseline period randomly varied between 1 s and 2 s , a random dot kinematogram ( RDK ) was displayed for 2 s with coherent motion either to the left or to the right ( Figure 1A ) . Following a 500 ms delay , an instruction cue appeared , consisting of a 3°×1° arrow pointing either to the left or the right , and participants were instructed to press the corresponding button ( left or right ) as quickly and as accurately as possible . Trials ended once a response had been made or after a maximum of 1 s if no response was made . The task had a factorial design with congruence ( whether or not the direction of the instruction cue matched that of the coherent motion in the RDK ) and coherence ( the percentage of coherently moving dots in the RDK ) as factors ( Figure 1B ) . Participants were instructed that in most of the trials ( 70% ) , the direction of coherent motion in the RDK was congruent to the direction of the instruction cue . Participants could therefore reduce their mean response time ( RT ) by preparing to press the button corresponding to the direction of the coherent motion . The RDK consisted of a 10°×10° square aperture centered on the fixation point with 100 , 0 . 3° diameter dots , each moving at 4°/s . On each trial , a certain percentage of the dots ( specified by the motion coherence level ) moved coherently through the aperture in one direction , left or right . The remaining dots moved in random directions through the aperture , with a consistent path per dot . The levels were individually set for each participant by using an adaptive staircase procedure ( QUEST; Watson and Pelli , 1983 ) to determine the motion coherence at which they achieved 82% accuracy in a block of 40 trials at the beginning of each session , in which they had to simply respond with the left or right button to leftwards or rightwards motion coherence . The resulting level of coherence was then used as medium , and 50% and 150% of it as low and high , respectively . Each block contained 126 congruent trials , and 54 incongruent trials , and 60 trials for each coherence level with half containing coherent leftward motion , and half rightward ( 180 trials total ) . All trials were randomly ordered . Participants completed three blocks per session , and 1–5 sessions on different days , resulting in 540–2700 trials per participant ( M = 1822 . 5 , SD = 813 . 21 ) . The behavioral task was implemented in MATLAB ( The MathWorks , Inc . , Natick , MA ) using the Cogent 2000 toolbox ( http://www . vislab . ucl . ac . uk/cogent . php ) . Prior to MEG sessions , participants underwent two MRI scanning protocols during the same visit: one for the scan required to generate the scalp image for the head-cast , and a second for MEG source localization . Structural MRI data were acquired using a 3T Magnetom TIM Trio MRI scanner ( Siemens Healthcare , Erlangen , Germany ) , while participants were laying in a supine position . The first protocol was used to generate an accurate image of the scalp for head-cast construction ( Meyer et al . , 2017a ) . This used a T1-weighted 3D spoiled fast low angle shot ( FLASH ) sequence with the following acquisition parameters: 1 mm isotropic image resolution , field-of view set to 256 , 256 , and 192 mm along the phase ( anterior-posterior , A–P ) , read ( head-foot , H–F ) , and partition ( right-left , R–L ) directions , respectively . The repetition time was 7 . 96 ms and the excitation flip angle was 12° . After each excitation , a single echo was acquired to yield a single anatomical image . A high readout bandwidth ( 425 Hz/pixel ) was used to preserve brain morphology and no significant geometric distortions were observed in the images . Acquisition time was 3 min 42 s , a sufficiently short time to minimize sensitivity to head motion and any resultant distortion . Care was also taken to prevent distortions in the image due to skin displacement on the face , head , or neck , as any such errors could compromise the fit of the head-cast . Accordingly , a more spacious 12 channel head coil was used for signal reception without using either padding or headphones . The second protocol was a quantitative multiple parameter mapping ( MPM ) protocol , consisting of 3 differentially-weighted , RF and gradient spoiled , multi-echo 3D FLASH acquisitions acquired with whole-brain coverage at 800 µm isotropic resolution . Additional calibration data were also acquired as part of this protocol to correct for inhomogeneities in the RF transmit field ( Lutti et al . , 2010; Lutti et al . , 2012; Callaghan et al . , 2015 ) . For this protocol , data were acquired with a 32-channel head coil to increase SNR . The FLASH acquisitions had predominantly proton density ( PD ) , T1 or magnetization transfer ( MT ) weighting . The flip angle was 6° for the PD- and MT-weighted volumes and 21° for the T1 weighted acquisition . MT-weighting was achieved through the application of a Gaussian RF pulse 2 kHz off resonance with 4 ms duration and a nominal flip angle of 220° prior to each excitation . The field of view was set to 224 , 256 , and 179 mm along the phase ( A–P ) , read ( H–F ) , and partition ( R–L ) directions , respectively . Gradient echoes were acquired with alternating readout gradient polarity at eight equidistant echo times ranging from 2 . 34 to 18 . 44 ms in steps of 2 . 30 ms using a readout bandwidth of 488 Hz/pixel . Only six echoes were acquired for the MT-weighted acquisition in order to maintain a repetition time ( TR ) of 25 ms for all FLASH volumes . To accelerate the data acquisition and maintain a feasible scan time , partially parallel imaging using the GRAPPA algorithm ( Griswold et al . , 2002 ) was employed with a speed-up factor of 2 and forty integrated reference lines in each phase-encoded direction ( A-P and R-L ) . To maximize the accuracy of the measurements , inhomogeneity in the transmit field was mapped by acquiring spin echoes and stimulated echoes across a range of nominal flip angles following the approach described in Lutti et al . , 2010 , including correcting for geometric distortions of the EPI data due to B0 field inhomogeneity . Total acquisition time for all MRI scans was less than 30 min . Quantitative maps of proton density ( PD ) , longitudinal relaxation rate ( R1 = 1/T1 ) , magnetization transfer saturation ( MT ) and effective transverse relaxation rate ( R2*=1/T2* ) were subsequently calculated according to the procedure described in Weiskopf et al . ( 2013 ) . Each quantitative map was co-registered to the scan used to design the head-cast , using the T1 weighted map . The resulting maps were used to extract cortical surface meshes using FreeSurfer ( see below ) . From an MRI-extracted image of the skull , a head-cast that fit between the participant’s scalp and the MEG dewar was constructed ( Troebinger et al . , 2014b; Meyer et al . , 2017a ) . Scalp surfaces were first extracted from the T1-weighted MRI scans acquired in the first MRI protocol using standard SPM12 procedures ( RRID:SCR_007037; http://www . fil . ion . ucl . ac . uk/spm/ ) . Next , this tessellated surface was converted into the standard template library ( STL ) format , commonly used for 3D printing . Importantly , this conversion imposed only a rigid body transformation , meaning that it was easily reverse-transformable at any point in space back into native MRI space . Accordingly , when the fiducial locations were optimized and specified in STL space as coil-shaped protrusions on the scalp , their exact locations could be retrieved and employed for co-registration . Next , the head-cast design was optimized by accounting for factors such as head-cast coverage in front of the ears , or angle of the bridge of the nose . To specify the shape of the fiducial coils , a single coil was 3D scanned and three virtual copies of it were placed at the approximate nasion , left peri-auricular ( LPA ) , and right peri-auricular ( RPA ) sites , with the constraint that coil placements had to have the coil-body and wire flush against the scalp , in order to prevent movement of the coil when the head-cast was worn . The virtual 3D model was placed inside a virtual version of the scanner dewar such that the distance to the sensors was minimized ( by placing the head as far up within the dewar as possible ) while ensuring that vision was not obstructed . Next , the head-model ( plus spacing elements and coil protrusions ) was printed using a Zcorp 3D printer ( Zprinter 510 ) with 600 × 540 dots per inch resolution . The 3D printed head model was then placed inside the manufacturer-provided replica of the dewar and liquid resin was poured in between the surfaces to fill the negative space , resulting in the participant-specific head-cast . The fiducial coil protrusions in the 3D model now become indentations in the resulting head-cast , in which the fiducial coils can sit during scanning . The anatomical landmarks used for determining the spatial relationship between the brain and MEG sensors are thus in the same location for repeated scans , allowing data from multiple sessions to be combined ( Meyer et al . , 2017a ) . FreeSurfer ( v5 . 3 . 0; Fischl , 2012 ) was used to extract cortical surfaces from the multi-parameter maps . Use of multi-parameter maps as input to FreeSurfer can lead to localized tissue segmentation failures due to boundaries between the pial surface , dura mater and CSF showing different contrasts compared to that assumed within FreeSurfer algorithms ( Lutti et al . , 2014 ) . Therefore , an in-house FreeSurfer surface reconstruction procedure was used to overcome these issues , using the PD and T1 maps as inputs . Detailed methods for cortical surface reconstruction can be found in Carey et al . , 2017 . This process yields surface extractions for the pial surface ( the most superficial layer of the cortex adjacent to the cerebro-spinal fluid , CSF ) , and the white/grey matter boundary ( the deepest cortical layer ) . Each of these surfaces is downsampled by a factor of 10 , resulting in two meshes comprising about 30 , 000 vertices each ( M = 30 , 0940 . 75 , SD = 2 , 665 . 450 . 45 over participants ) . For the purposes of this study , we used these two surfaces to represent deep ( white/grey interface ) and superficial ( grey-CSF interface ) cortical models . MEG recordings were made using a 275-channel Canadian Thin Films ( CTF ) MEG system with superconducting quantum interference device ( SQUID ) -based axial gradiometers ( VSM MedTech , Vancouver , Canada ) in a magnetically shielded room . The data collected were digitized continuously at a sampling rate of 1200 Hz . A projector displayed the visual stimuli on a screen ( ~50 cm from the participant ) , and participants made responses with a button box . All data are archived at the Open MEG Archive ( OMEGA; Niso et al . , 2016 ) and may be accessed via http://dx . doi . org/10 . 23686/0015896 ( Niso et al . , 2018 ) . Participant responses were classified as correct when the button pressed matched the direction of the instruction cue , and incorrect otherwise . The response time ( RT ) was measured as the time of button press relative to the onset of the instruction cue . We analyzed accuracy using a generalized linear mixed model with a logit link function , using correct ( true or false ) in each trial as the dependent variable , congruence ( congruent or incongruent ) and coherence ( low , medium , high ) and their interaction as fixed effects , and a participant-specific intercept as a random effect . Fixed effects were tested using type III Wald χ2 tests . RT was analyzed using a linear mixed model also using congruence as coherence and their interaction as fixed effects , with a participant-specific intercept as a random effect . Fixed effects for this model were estimated using type III Wald F tests with Kenward-Rogers approximated degrees of freedom ( Kenward and Roger , 1997 ) . For both models , planned pairwise follow-up tests were performed using LSMEANS between congruence levels at each coherence level , Tukey corrected . All MEG data preprocessing and analyses were performed using SPM12 ( RRID:SCR_007037; http://www . fil . ion . ucl . ac . uk/spm/ ) using Matlab R2014a ( RRID:SCR_001622 ) and are available at http://github . com/jbonaiuto/meg-laminar ( Bonaiuto , 2018; copy archived at https://github . com/elifesciences-publications/meg-laminar ) . The data were filtered ( 5th order Butterworth bandpass filter: 2–100 Hz , Notch filter: 50 Hz ) and downsampled to 250 Hz . Eye-blink artifacts were removed using multiple source eye correction ( Berg and Scherg , 1994 ) . Trials were then epoched from 1 s before RDK onset to 1 . 5 s after instruction cue onset , and from 2 s before the participant’s response to 2 s after . Blocks within each session were merged , and trials whose variance exceeded 2 . 5 standard deviations from the mean were excluded from analysis . The reproducibility of the topographic maps , ERFs , and time frequency decompositions was quantified for a representative participant by computing the intra-class correlation coefficient ( ICC ) , a measure of test-retest reliability ( Shrout and Fleiss , 1979 ) . This was done within-session over runs used a type 2 k ICC , with the runs modeled as a random effect and the measure given by an average over trials within a run . Similarly , the between-session ICC was type 2 k with sessions modeled as a random effect and the measure given by an average over runs within a session . At the sensor-level , we analyzed three epochs: one aligned to the RDK stimulus ( 0 – 2000 ms ) , one centered on instruction stimulus ( −500 ms to +500 ms ) , and one centered on the participant’s response ( −1000 ms to +1000 ms ) , with 250 ms padding on either side to avoid edge effects . For each epoch type , seven-cycle Morlet wavelets were used to compute power within 2 – 45 Hz in increments of 1 Hz , and a multi-taper analysis was used to computer power within 55 – 115 Hz in increments of 5 Hz ( sine taper , time resolution = 200 ms , time step = 10 ms ) . Power for each epoch type was baseline-corrected using the 500 ms prior to the onset of the RDK stimulus in a frequency-specific manner using robust averaging . Robust averaging is a form of general linear modeling ( Wager et al . , 2005 ) used to reduce the influence of outliers on the mean by iteratively computing a weighting factor for each sample according to how far it is from the mean . The baseline-corrected time-frequency spectrograms were then averaged over a cluster of 15 sensors overlying occipital cortex for visual signals ( MLO53 , MLO43 , MLO32 , MLO52 , MLO31 , MLO51 , MLO41 , MZO02 , MZO03 , MRO52 , MRO42 , MRO31 , MRO53 , MRO43 , MRO32 ) and 18 sensors overlying contralateral motor cortex for sensorimotor signals ( MLC17 , MLC25 , MLC32 , MLC42 , MLC54 , MLC63 , MRC63 , MLP57 , MLP45 , MLP35 , MLP12 , MLP23 , MLC55 , MZC04 , MLP44 , MLP34 , MLP22 , MLP11 ) , and finally smoothed using a Gaussian kernel ( FWHM 8 × 8 Hz frequency bins and 80 ms ) . We used a linear mixed model with subject-specific offsets as random effects to test for significant changes in power from baseline . We used a significance threshold of p<0 . 05 , Bonferroni corrected for multiple comparisons ( over time and frequency ) . Source inversion was performed using the empirical Bayesian beamformer ( EBB; Belardinelli et al . , 2012; López et al . , 2014 ) . The sensor data were first reduced , using singular value decomposition to 180 virtual channels , each with 16 temporal samples ( weighting the dominant modes of temporal variation across the window ) . For uninformative priors , the maximum-likelihood solution to the inverse problem reduces to:J^=QLT ( Qϵ+LQLT ) −1Ywhere J^ is the estimated current density across the source space , Y is ( reduced ) measured data , L is the lead field or sensitivity matrix that can be computed based on the sensor and volume conductor geometry . Qϵ is the sensor noise , and Q is the prior estimate of source covariance . We assumed the sensor level covariance ( Qϵ ) to be an identity matrix ( see discussion ) . Most popular inversion algorithms can be differentiated by the form of Q ( Friston et al . , 2008; López et al . , 2014 ) . Here we used a beamformer prior to estimate the structure of Q ( Belardinelli et al . , 2012; López et al . , 2014 ) where a direct estimate of prior source co-variance ( Q ) is made based on the sensor-level data:Q ( i ) =1LiTLi ( LiT ( YYT ) −1Li+λI ) −1 Q is a diagonal matrix , and each element of the diagonal Q ( i ) corresponds to a source location i . The ( reduced ) sensor level data is Y , the lead field of each element i is Li , T denotes the transpose operator , I is an identity matrix , and λ is a regularization constant . The highest resolution beamformer estimate will be made with λ=0 and this is the default used throughout the paper . Such low values of regularization can , however , become problematic especially when comparing signals occupying different bandwidths at different SNRs ( Brookes et al . , 2008 ) . In order to verify that the differential effects we were observed were not due to regularization , we therefore also implemented an augmented EBB solution in which the Bayesian scheme optimized from a range of source priors each at different levels of regularization ( 0 , 5 , 10 , 50 , 100 and 1000 ) percent of the mean eigenvalue of YYT ( Figure 5—figure supplement 13B ) . The prior estimates of Qϵ and Q are then re-scaled or optimally mixed using an expectation maximization scheme ( Friston et al . , 2008 ) to give an estimate of J that maximizes model evidence . The source level prior was based on the beamformer power estimate across a two-layer manifold comprised of pial and white cortical surfaces with source orientations defined as normal to the cortical surface and a spatial coherence prior ( Friston et al . , 2008 ) , G ( σ ) :σ=0 . 4 ( corresponding to a FWHM of approximately 4 mm ) . We used the Nolte single shell head model ( Nolte , 2003 ) . All analyses were carried out using the SPM12 ( RRID:SCR_007037; http://www . fil . ion . ucl . ac . uk/spm/ ) software package ( see López et al . , 2014 ) for implementation details ) . The laminar analysis reconstructed the data onto a mesh combining the pial and white matter surfaces , thus providing an estimate of source activity on both surfaces ( Figure 4 ) . We analyzed six different visual and sensorimotor signals at different frequencies and time windows of interest ( WOIs ) , using the same frequency bands across participants: RDK-aligned visual alpha ( 7-13Hz; WOI=[0s , 2s]; baseline WOI=[-1s , - . 5s] ) , RDK-aligned visual gamma ( 60-90Hz; WOI=[250ms , 500ms]; baseline WOI=[-500ms , -250ms] ) , instruction cue-aligned visual gamma ( 60-90Hz; WOI=[100ms , 500ms]; baseline WOI=[-500ms , -100ms] ) , RDK-aligned sensorimotor beta ( 15-30Hz; WOI=[0s , 2s]; baseline WOI=[-500ms , 0ms] ) , response-aligned sensorimotor beta ( 15-30Hz; WOI=[500ms , 1s]; baseline WOI=[-250ms 250ms] ) , and response-aligned sensorimotor gamma ( 60-90Hz; WOI=[-100ms , 200ms]; baseline WOI=[-1 . 5s , -1s] ) . For each signal , we defined an ROI by comparing power in the associated frequency band during the WOI with a prior baseline WOI at each vertex and averaging over trials . Vertices in either surface with a mean unsigned fractional change in power from the baseline in the 80th percentile over all vertices on that surface ( the top 20% ) , as well as the corresponding vertices on the other surface , were included in the ROI . This ensured that the contrast used to define the ROI was orthogonal to the subsequent pial versus white matter surface contrast . For each trial , ROI values for the pial and white matter surfaces were computed by averaging the unsigned fractional change in power compared to baseline in that surface within the ROI ( | WOI−baseline |baseline ) . For within-participant tests , a paired t-test was used to compare the ROI values from the pial surface with those from the white matter surface over trials ( Figure 4 ) . This resulted in positive t-statistics when the unsigned fractional change in power from baseline was greatest on the pial surface , and negative values when the fractional change was greatest on the white matter surface . All t-tests were performed with corrected noise variance estimates in order to attenuate artifactually high significance values ( Ridgway et al . , 2012 ) . Group-level statistics were performed using one-sample Wilcoxon tests of the unsigned fractional change in power from baseline averaged within ROI ( | WOIpial−baselinepial |baselinepial−| WOIwhitematter−baselinewhitematter |baselinewhitematter ) . The control analyses utilized the same procedure , but each introduced some perturbation to the data . The shuffled analysis permuted the lead fields of the forward model prior to source reconstruction in order to destroy any correspondence between the cortical surface geometry and the sensor data . This was repeated 10 times per session , with a different random lead field permutation each time . The mean unsigned magnitude of the change in power from baseline averaged within ROI was then used as the null hypothesis in the follow-up runs of the main laminar analyses . Each permutation was then used in the laminar analysis for every low and high frequency signal . The co-registration error analysis introduced a rotation ( M = 10° , SD = 2 . 5° ) and translation ( M = 10 mm , SD = 2 . 5 mm ) of the fiducial coil locations in a random direction prior to source inversion , simulating between-session co-registration error . This was done 10 times per session , with a different random rotation and translation each time . Again , each perturbation was used in the laminar analysis for every low and high frequency signal . The SNR analysis used a random subset of the available trials from each participant , gradually increasing the number of trials used from 10 to the number of trials available . This was repeated 10 times , using a different random subset of trials each time , and the resulting t-statistics were averaged . The white noise analysis was used to decrease SNR by progressively adding Gaussian white noise of increasing standard deviation to the sensor level data . For analyses of laminar bias , distance to the scalp was computed using the CAT12 toolbox ( http://dbm . neuro . uni-jena . de/cat/ ) to generate a convex hull surface from the pial surface , and then computing the Euclidean distance between each vertex and the nearest vertex on hull surface ( Van Essen , 2005; Im et al . , 2006; Tosun et al . , 2015 ) , and lead field strength was computed as root mean square of the lead field . Relationships between relative lead field strength or laminar depth and the effect size of the laminar bias were evaluated using per-participant Spearman partial correlation coefficients ( controlling for the effect of laminar depth or relative lead field strength , respectively ) . Each participant’s correlation coefficient was Fisher-transformed and the resulting Z scores were compared against zero using a one sample t-test . The analysis using only vertices where the white matter is closer to the scalp used the same ROIs as the main analysis , but only including vertex pairs where the sulcal depth of the white matter vertex was less than that of the pial vertex . The analysis controlling for the effect of the distance to the scalp used robust regression ( Holland and Welsch , 1977 ) to fit a linear model to the difference ( pial – white matter ) of the unsigned fractional change in power from baseline , averaged over trials . The square root of the distance to scalp surface ( averaged over pial and white matter vertices within each vertex pair ) was used as the independent variable . The main laminar analysis was then run on the residuals of this regression . The patch size analysis ran each inversion using a range of reconstruction patch sizes ( FWHM = 2 . 5 , 5 , 10 , and 20 mm ) , and compared the free energy metric of model fit of each to the mean over all patch sizes . For each visual and sensorimotor frequency band/task epoch combination , induced activity was compared between task conditions on the surface and within the anatomically constrained ROI identified from the corresponding laminar analysis . Seven-cycle Morlet wavelets were used to compute power within the frequency band and this was baseline-corrected in a frequency-specific manner using robust averaging . For each participant , the mean percent change in power over the WOI was averaged over all trials within every condition . Wilcoxon tests for comparing two repeated measures were used to compare the change in power for instruction cue-aligned visual gamma and sensorimotor beta rebound between congruent and incongruent trials . A Friedman test for comparing multiple levels of a single factor with repeated measures was used to compare the sensorimotor beta decrease between low , medium , and high RDK coherence trials . This was followed up by Tukey-Kramer corrected pairwise comparisons . Only trials in which a correct response was made were analyzed .

As we interact with the world around us , signals flow from neuron to neuron and from one brain area to the next . When we look at an object , for example , signals pass along a pathway of areas in the outermost part of the brain , called the cortex . Each area along this visual pathway performs more complex processing than the one before it . But information also flows in the opposite direction along such cortical pathways . These feedback signals enable areas further along the pathway to influence the activity of those before them . Studies in animals suggest that much like a highway , information is travelling in opposite directions within the cortex along different lanes . In mammals , these lanes consist of distinct layers of cells . In the visual cortex of monkeys , feedback signals travel via deeper layers of cortex , whereas feedforward signals travel via the upper layers . Brain activity in the upper layers also has a higher frequency than that in the lower layers . But is this also the case in our own brains ? Bonaiuto et al . used a technique called MEG to measure the frequency of brain activity within the upper and lower layers of cortex in healthy volunteers . The volunteers had to look at images on a screen and then respond by pressing a button . Bonaiuto et al . observed that activity in deeper layers of cortex occurred mostly at lower frequencies , while activity in upper layers mostly happened at higher frequencies . This pattern , which matches that seen in monkeys , was found in both visual cortex and in areas of cortex that help plan and execute movements . In visual cortex , the activity in the upper layers appeared to carry feedforward signals . But in movement-related areas , feedback and feedforward signals were less clearly related to cortical layers . These findings lend support to current theories about how the cortex is organized . They also show that MEG can reveal rapidly changing brain activity at a high spatial resolution . The findings may also provide clues to the origins of brain disorders called oscillopathies . These involve changes in specific frequencies of brain activity , and include schizophrenia and epilepsy , among others .

2018

Lamina-specific cortical dynamics in human visual and sensorimotor cortices

Master regulatory genes of tissue specification play key roles in stem/progenitor cells and are often important in cancer . In the prostate , androgen receptor ( AR ) is a master regulator essential for development and tumorigenesis , but its specific functions in prostate stem/progenitor cells have not been elucidated . We have investigated AR function in CARNs ( CAstration-Resistant Nkx3 . 1-expressing cells ) , a luminal stem/progenitor cell that functions in prostate regeneration . Using genetically--engineered mouse models and novel prostate epithelial cell lines , we find that progenitor properties of CARNs are largely unaffected by AR deletion , apart from decreased proliferation in vivo . Furthermore , AR loss suppresses tumor formation after deletion of the Pten tumor suppressor in CARNs; however , combined Pten deletion and activation of oncogenic Kras in AR-deleted CARNs result in tumors with focal neuroendocrine differentiation . Our findings show that AR modulates specific progenitor properties of CARNs , including their ability to serve as a cell of origin for prostate cancer . Elucidating the cell type ( s ) of origin of cancer and the molecular drivers of tumor initiation is of fundamental importance in understanding the basis of distinct tumor subtypes as well as differences in treatment response and patient outcomes ( Blanpain , 2013; Rycaj and Tang , 2015; Shibata and Shen , 2013; Visvader , 2011 ) . Furthermore , since cancer often originates from stem cells and/or lineage-restricted progenitor cells , the identification of stem/progenitor cells is of considerable significance . In the case of the prostate , however , both the specific identity of stem/progenitor cells as well as cell types of origin for cancer have remained unclear ( Lee and Shen , 2015; Wang and Shen , 2011; Xin , 2013 ) . In the normal prostate epithelium , there are three primary cell types , corresponding to secretory luminal cells , an underlying layer of basal cells , and rare neuroendocrine cells ( Shen and Abate-Shen , 2010; Toivanen and Shen , 2017 ) . Lineage-tracing studies have shown that both luminal and basal cells are mostly lineage-restricted ( unipotent ) in the normal adult mouse prostate as well as during androgen-mediated prostate regeneration ( Choi et al . , 2012; Liu et al . , 2011; Lu et al . , 2013; Wang et al . , 2013 ) . In addition , cells within the basal compartment possess stem/progenitor properties in a range of ex vivo assays as well as during inflammation and wound repair ( Goldstein et al . , 2008; Höfner et al . , 2015; Kwon et al . , 2014; Lawson et al . , 2007; Toivanen et al . , 2016; Wang et al . , 2013 ) . However , recent studies have shown that luminal cells can also display stem/progenitor properties in specific in vivo and ex vivo contexts ( Chua et al . , 2014; Karthaus et al . , 2014; Kwon et al . , 2016; Wang et al . , 2009 ) . Furthermore , there is now considerable evidence supporting a luminal origin for prostate cancer , both in mouse models ( Wang et al . , 2009; Wang et al . , 2014 ) as well as in human tissues ( Gurel et al . , 2008; Meeker et al . , 2002 ) . Androgen receptor ( AR ) plays a central role in many aspects of normal prostate development as well as prostate cancer progression ( Cunha et al . , 2004; Toivanen and Shen , 2017; Watson et al . , 2015 ) . In the prostate epithelium of adult hormonally intact mice , AR is primarily expressed by luminal cells , but is also found in a subset of basal cells ( Lee et al . , 2012; Mirosevich et al . , 1999; Xie et al . , 2017 ) . Several studies have shown that conditional deletion of AR in the adult prostate epithelium results in a short-term increase in proliferation of luminal cells ( Wu et al . , 2007; Xie et al . , 2017; Zhang et al . , 2016a ) , indicating a role for AR in normal prostate homeostasis . Importantly , AR can act as a master regulator of prostate epithelial specification in a fibroblast reprogramming assay ( Talos et al . , 2017 ) . In the context of prostate cancer , tumor recurrence after androgen-deprivation therapy is due to the emergence of castration-resistant prostate cancer ( CRPC ) , which is associated with increased AR activity that can be targeted by second-generation anti-androgen therapies ( Watson et al . , 2015 ) . However , treatment failure following such anti-androgen therapies is frequently associated with the appearance of AR-negative tumor cells , which are typically associated with highly aggressive lethal disease ( Beltran et al . , 2014; Vlachostergios et al . , 2017; Watson et al . , 2015 ) . In some cases , this AR-negative CRPC contain large regions displaying a neuroendocrine phenotype ( CRPC-NE ) ( Beltran et al . , 2016 , 2014; Ku et al . , 2017; Mu et al . , 2017; Zou et al . , 2017 ) . Previous work from our laboratory has identified CARNs as a luminal stem/progenitor cell within the androgen-deprived normal mouse prostate epithelium that is also a cell of origin for prostate cancer ( Wang et al . , 2009 ) . Following androgen administration to induce prostate regeneration , CARNs can generate both luminal and basal progeny in vivo , as well as in renal grafting and organoid assays ( Chua et al . , 2014; Wang et al . , 2009 ) . Although CARNs express AR ( Wang et al . , 2009 ) , it has been unclear whether AR is required for any or all the progenitor properties of CARNs , and whether the intrinsic castration-resistance of untransformed CARNs might resemble the castration-resistance of tumor cells in CRPC . Below , we show that the progenitor properties of CARNs are largely unaffected by loss of AR , whereas their ability to serve as cells of origin for prostate cancer are altered by AR deletion in a context-dependent manner . Notably , cell lines derived from AR-deleted CARNs have molecular profiles that resemble those for CRPC , and AR-deleted CARNs can serve as a cell of origin for focal neuroendocrine differentiation in a novel mouse model of AR-negative prostate cancer . To investigate whether the stem/progenitor properties of CARNs are dependent upon AR function , we have used an inducible targeting approach in genetically engineered mice . For this purpose , we used mice carrying a conditional allele of Ar ( De Gendt et al . , 2004 ) together with the inducible Nkx3 . 1CreERT2 driver ( Wang et al . , 2009 ) and the R26R-YFP reporter to visualize cells and their progeny in which Cre-mediated recombination has taken place ( Srinivas et al . , 2001 ) ; as Ar is an X-linked gene , deletion of a single allele in males is sufficient to confer a hemizygous null phenotype . Since CARNs are Nkx3 . 1-expressing cells found under androgen-deprived conditions , we castrated adult male mice carrying the Cre driver and reporter alleles , followed by tamoxifen induction to induce Cre-mediated activity specifically in CARNs ( Figure 1A ) . Using this strategy , we compared the properties of CARNs in Nkx3 . 1CreERT2/+; R26R-YFP/+ mice , which we denote as ‘control’ mice , with those in Nkx3 . 1CreERT2/+; Arflox/Y; R26R-YFP/+ mice , which we denote as ‘AR-deleted’ mice . We found that the percentage of lineage-marked YFP-positive cells , corresponding to CARNs , was not significantly different ( p=0 . 51 ) between the control ( 0 . 36 ± 0 . 17% , n = 5 mice ) and AR-deleted mice ( 0 . 31 ± 0 . 06% , n = 5 mice ) ( Figure 1B , C ) . Notably , we found that 87 . 1% of the YFP-positive cells in Nkx3 . 1CreERT2/+; Arflox/Y; R26R-YFP/+ mice ( n = 344/395 cells in four mice ) were AR-negative , indicating that AR deletion occurred with high efficiency . Furthermore , these YFP-positive cells expressed the luminal markers cytokeratins 8 and 18 ( CK8 and CK18 ) , but not cytokeratin 5 ( CK5 ) and p63 , indicating that AR deletion does not alter the luminal phenotype of CARNs ( Figure 1D ) . These findings indicate that AR deletion does not affect the frequency or luminal properties of CARNs . To investigate the progenitor properties of AR-deleted CARNs , we examined their ability to generate progeny during androgen-mediated regeneration . We implanted subcutaneous mini-osmotic pumps containing testosterone into control Nkx3 . 1CreERT2/+; R26R-YFP/+ mice as well as Nkx3 . 1CreERT2/+; Arflox/Y; R26R-YFP/+ mice , followed by tissue harvest at 4 , 7 , 14 , and 28 days later; the final 28-day time point corresponds to a fully regenerated prostate ( Figure 2A ) . We found that the YFP-marked cells and cell clusters were similar in the control and AR-deleted prostates at 4 and 7 days after testosterone administration ( Figure 2B , C ) . However , at 14 and 28 days , the control prostates contained many YFP-expressing cell clusters with more than 4 cells , whereas the prostates with AR-deleted CARNs mostly contained YFP-expressing single cells or doublets ( Figure 2B , C ) . To compare the proliferative ability of control and AR-deleted CARNs and their progeny , we pursued BrdU pulse-chase experiments during prostate regeneration . We performed castration and tamoxifen administration on control and AR-deleted mice , followed by androgen-mediated regeneration for 28 days , with administration of daily doses of BrdU either from days 1 through 4 of regeneration or from days 11 through 14 ( Figure 3A , B ) . When BrdU was administered from days 1 through 4 of regeneration , we could readily detect BrdU+YFP+ cells in the control prostates ( 50 . 9 ± 11 . 8% , n = 3 mice ) as well as AR-deleted prostates ( 62 . 9 ± 14 . 9% , n = 3 mice ) ( Figure 3C , E ) . In contrast , when BrdU was administered from days 11 through 14 , we could only detect BrdU+YFP+ cells in the control prostates ( 11 . 1 ± 6 . 2% , n = 3 mice ) , but not in the AR-deleted prostates ( 0% , n = 3 mice ) ( Figure 3D , F ) . This difference suggests that AR-deleted CARNs and/or their progeny have a defect in proliferation during later stages of regeneration , consistent with the analysis of YFP+ cluster size ( Figure 2B ) . Notably , although YFP-expressing basal cells could be readily identified in both control and AR-deleted prostates , there was an increase in the percentage of basal cells within the YFP+ population in the AR-deleted mice ( Figure 2D ) . This difference was evident using either the basal marker CK5 ( 2 . 1% CK5+AR+YFP+ versus 19 . 2% CK5+AR–YFP+ ) or p63 ( 3 . 5% p63+AR+YFP+ versus 14 . 6% p63+AR–YFP+ ) ( Figure 2D ) . These findings indicate that AR-deleted CARNs favor generation of basal progeny and/or that there is decreased proliferation or survival of luminal progeny during regeneration . As a further test of the progenitor properties of AR-deleted CARNs , we examined their ability to generate prostate ducts in a tissue recombination/renal grafting assay . Previously , we had shown that single CARNs were capable of generating ducts in this assay ( Wang et al . , 2009 ) . We isolated YFP-positive cells from control and AR-deleted mice that had undergone castration and tamoxifen induction , and recombined 10 YFP-positive cells together with 2 . 5 × 105 rat embryonic urogenital mesenchyme cells , followed by renal grafting ( Figure 3G ) . We found that both control and AR-deleted CARNs could generate prostate ducts ( Figure 3H ) , but that the AR-deleted CARN-s were significantly less efficient ( 12 . 5% graft efficiency , n = 16 ) compared to the control CARNs ( p=0 . 003; 68 . 8% graft efficiency , n = 16 ) , consistent with a proliferation defect in the AR-deleted CARNs . Based on these findings , we further investigated the properties of CARNs and AR-deleted CARNs by establishing adherent cell lines . Using a novel method based on conditions that we previously established for culture of prostate organoids ( Chua et al . , 2014 ) , we successfully generated adherent cell lines from single YFP+ cells isolated from castrated and tamoxifen-treated Nkx3 . 1CreERT2/+; Arflox/Y; R26R-YFP/+ mice . Genotyping of the resulting lines led to identification of Ar-positive ( non-recombined allele ) and Ar-negative ( recombined allele ) lines , which we term APCA and ADCA ( Ar-Positive CArn-derived and Ar-Deleted CArn-derived ) lines . These cell lines could be propagated as adherent cells in the presence of Matrigel and DHT . Under these conditions , we found that the APCA ( n = 2 ) and ADCA ( n = 2 ) lines were morphologically indistinguishable ( Figure 4A ) . These cell lines were comprised of a mixture of cells expressing basal ( CK5 ) or luminal ( CK8 ) markers or both , as well as Foxa1 , an epithelial marker that encodes a transcriptional partner of AR ( Gao et al . , 2003; He et al . , 2010 ) ( Figure 4A ) . Furthermore , both the APCA and ADCA lines showed robust proliferation at similar levels , as demonstrated by Ki67 immunostaining , CellTiter-Glo assays , and colony formation in the presence or absence of DHT ( Figure 4A–C ) . To determine the relative efficiency of forming APCA and ADCA lines from AR+ and AR– CARNs , respectively , we sorted 60 single YFP+ cells from castrated and tamoxifen-treated Nkx3 . 1CreERT2/+; Arflox/Y; R26R-YFP/+ mice into individual wells of a 96-well plate . We found that six YFP+ cells gave rise to adherent lines , with four of these corresponding to AR+ lines that had failed to undergo Cre-mediated recombination of the conditional Ar allele , and two lines corresponding to AR– lines . After correcting for the 87 . 1% efficiency of recombination of the AR-floxed allele in vivo , these data indicate that the relative plating efficiency for the AR– CARNs compared to AR+ CARNs is 7 . 4% , consistent with the decreased grafting efficiency of AR– CARNs . Notably , we were also able to use this methodology to establish 14 primary human prostate epithelial cell lines from benign prostatectomy specimens at high efficiency . Similar to the mouse APCA cell lines , these HPE ( Human Prostate Epithelial ) cell lines are propagated as adherent cells in the presence of Matrigel and DHT . All these lines display similar marker phenotypes , expressing basal and luminal markers as well as AR and PSA , and are highly proliferative ( Figure 4—figure supplement 1 ) . To assess the ability of the APCA and ADCA cell lines to reconstitute prostate ducts , we performed tissue recombination assays by combining 1 × 105 cells with rat urogenital mesenchyme followed by renal grafting . We found that the APCA lines could generate prostate ducts ( n = 10 grafts with two lines; 100% efficiency ) , some with evidence of secretions , whereas the ADCA lines ( n = 6 grafts with one line; 67% efficiency ) generated ducts that lacked prostate secretions ( Figure 4D ) . Next , we tested the role of AR in this tissue reconstitution assay by treating the mice grafted with APCA cells ( n = 12 grafts with two lines ) with tamoxifen at 7 weeks after grafting in order to induce Ar deletion . We found that tamoxifen treatment resulted in grafts containing prostate ducts composed of a mixture of AR-positive and negative cells , but with a decreased efficiency of graft formation relative to the same APCA lines in the absence of tamoxifen ( 42% versus 100% efficiency ) ( Figure 4D ) . Taken together , these results show that AR deletion decreases the efficiency of prostate duct formation by CARN-derived cells , consistent with the results obtained using AR-deleted CARNs ( Figure 3H ) . Notably , since ADCA cells do not display a growth disadvantage relative to APCA cells in culture , this difference in duct formation is likely to be due to a non-cell-autonomous effect mediated by the urogenital mesenchyme in grafts . To examine the molecular basis for differences between the ADCA and APCA lines ( n = 2 lines each ) , we performed RNA-sequencing followed by bioinformatic analyses . Based on the RNA expression profiling data , we constructed a differential expression signature comparing ADCA cells to APCA cells . Using the resulting ADCA signature to examine pathway enrichment by Gene Set Enrichment Analysis ( GSEA ) ( Subramanian et al . , 2005 ) , we found up-regulation of gene sets involved in DNA replication and repair pathways , as well as cell cycle and apoptosis ( Figure 5A ) , suggesting that cellular proliferation and survival are affected by AR deletion . We also compared the ADCA signature with a signature defined between expression profiles of AR-null and AR-positive mouse prostate luminal cells ( Xie et al . , 2017 ) and found enrichment for up-regulated genes ( Figure 5B ) . Next , we performed a cross-species comparison of the ADCA signature with a signature defined between profiles of human prostate luminal and basal epithelial cells ( Zhang et al . , 2016b ) and found that there was no significant enrichment in either tail ( Figure 5C ) , indicating that AR deletion does not drive APCA cells towards a specific lineage . Furthermore , we performed GSEA comparisons of the ADCA signature with several signatures obtained from analyses of human prostate cancer progression . In particular , we observed enrichment for up-regulated genes when compared to a signature of CRPC from Best and colleagues ( Best et al . , 2005 ) , as well as to a signature of metastatic CRPC from Stanbrough and colleagues ( Stanbrough et al . , 2006 ) ( Figure 5D , E ) . Moreover , we observed a strong enrichment when compared to a signature from Beltran and colleagues ( Beltran et al . , 2016 ) defined between CRPC with neuroendocrine differentiation ( CRPC-NE ) and non-neuroendocrine CRPC ( Figure 5F ) , consistent with our observation of pathway enrichment for gene sets corresponding to axon guidance and small-cell lung cancer ( Figure 5A ) . Finally , we tested the ability of AR-deleted CARNs to serve as a cell of origin for prostate cancer , based on the previous finding that prostate cancer can initiate from CARNs after specific deletion of Pten and androgen-mediated regeneration ( Wang et al . , 2009 ) . We used Nkx3 . 1CreERT2/+; Ptenflox/flox; R26R-YFP/+ controls ( which we term NP-CARN ) and Nkx3 . 1CreERT2/+; Ptenflox/flox; Arflox/Y; R26R-YFP/+ mice ( NPA-CARN ) in an experimental paradigm involving castration , tamoxifen-treatment , and androgen-mediated regeneration for one month . We found that AR deletion resulted in a significant difference between the NP-CARN and NPA-CARN phenotypes , as the NP-CARN controls displayed high-grade prostatic intraepithelial neoplasia ( PIN ) , whereas the NPA-CARN prostates showed a weak phenotype corresponding to diffuse hyperplasia with mild inflammation and increased apoptosis , ( Figure 6A ) . The NPA-CARN prostates contained YFP-positive cells that also expressed phospho-Akt ( pAkt ) , indicating successful deletion of Pten , but these cells were only found as solitary or as small clusters , unlike the large clusters of YFP+pAkt+ cells observed in the control NP prostates ( Figure 6A ) . Furthermore , the NPA-CARN prostates displayed a decreased proliferative index relative to NP-CARN ( 2 . 7% , n = 3 vs . 9 . 2% , n = 3 ) , as well as increased apoptosis ( 2 . 6% , n = 3 vs . 0 . 7% , n = 3 ) ( Figure 6B ) . Taken together , findings indicate that AR is required for tumor initiation following Pten deletion in CARNs . In contrast , AR deletion did not affect tumor initiation following combined deletion of Pten and activation of the oncogenic KrasG12D allele . Using a similar protocol for castration , tamoxifen-treatment , and androgen-mediated regeneration , we compared the phenotypes of Nkx3 . 1CreERT2/+; Ptenflox/flox; KrasLSL-G12D/+; R26R-YFP/+ controls ( NPK-CARN ) and Nkx3 . 1CreERT2/+; Ptenflox/flox; KrasLSL-G12D/+; Arflox/Y; R26R-YFP/+ mice ( NPKA-CARN ) . In both genotypes , deletion of Pten and activation of oncogenic Kras resulted in formation of tumors with large clusters of YFP+ cells that express pAkt and Ras ( Figure 6C ) . Furthermore , both NPK-CARN and NPKA-CARN tumors displayed high proliferative indices ( 20% , n = 3 vs . 19% , n = 3 ) and low frequencies of apoptosis ( 0 . 9% , n = 3 vs . 0 . 8% , n = 3 ) ( Figure 6D ) . Notably , we observed an important difference between the NPK-CARN and NPKA-CARN tumors , as all the NPKA-CARN tumors contained a low but variable percentage of synaptophysin-positive neuroendocrine cells among total epithelial cells ( 0 . 7% , n = 3 ) , which were never observed in the NPK-CARN controls ( 0% , n = 3 ) ( Figure 6E , F ) . We also observed rare cells in all three NPKA-CARN tumors that expressed other neuroendocrine markers such as Chromogranin A , Foxa2 , and Aurora kinase A ( Figure 6E ) . Since the synaptophysin-postive cells co-expressed YFP ( Figure 6E ) , we conclude that transformed AR-negative CARNs can give rise to neuroendocrine cells . Taken together , our analyses have defined specific roles for AR in regulating the progenitor properties of CARNs , and indicate that the intrinsic castration-resistance of CARNs is independent of AR function . We find that targeted deletion of AR does not affect the percentage of CARNs , their luminal marker expression , or their ability to generate basal cells during androgen-mediated regeneration . However , there are fewer luminal progeny from AR-deleted CARNs during regeneration in vivo , and there is a decreased efficiency of prostate duct formation by both AR-deleted CARNs and ADCA cells in renal grafts . Thus , AR deletion in CARNs may primarily affect the proliferation and/or survival of their luminal progeny in vivo , although an effect on CARNs themselves cannot be excluded . Interestingly , our results suggest potential roles of the stroma in modulating the proliferation of CARNs and/or their luminal progeny . Notably , BrdU incorporation assays reveal a proliferation defect of AR-deleted CARNs during later stages of regeneration but not during early regeneration . One possible explanation is that AR activity may cell-autonomously regulate the proliferation of luminal progeny of CARNs; alternatively , however , stromal remodeling during later stages of regeneration may alter non-cell autonomous signals that regulate luminal proliferation . Furthermore , since ADCA cells do not display a growth defect in culture , their decreased efficiency of prostate duct formation in grafts is likely due to a non-cell-autonomous inhibitory effect from the stroma . Our study has also yielded interesting insights into differences between CARNs and other luminal epithelial cells . While this manuscript was in preparation , another study also investigated the requirements of AR in CARNs , and reported that AR-deleted CARNs completely failed to generate progeny during regeneration ( Xie et al . , 2017 ) . This apparent discrepancy may be partially explained by our observation that AR-deleted CARNs can still generate basal progeny , and by the failure of progeny from AR-deleted CARNs to proliferate at later stages of androgen-mediated regeneration . However , we concur that CARNs require AR function to generate viable luminal progeny , which is not the case for most luminal cells during homeostasis or regeneration ( Xie et al . , 2017; Zhang et al . , 2016a ) . Furthermore , the decreased proliferation of AR-deleted CARNs during regeneration contrasts with the transient increase in luminal proliferation observed after inducible AR deletion in the adult prostate epithelium , which is also a non-cell-autonomous effect mediated by the stroma ( Zhang et al . , 2016a ) . Together with our previous finding that CARNs display increased organoid formation efficiency relative to other luminal cells ( Chua et al . , 2014 ) , these findings support the identification of CARNs as a distinct luminal population with stem/progenitor properties , and highlight the complexity of AR functions in the epithelial and stromal compartments . In addition , we note that Xie and colleagues reported that Pten deletion in CARNs resulted in tumor formation after regeneration ( Xie et al . , 2017 ) , unlike the absence of tumors that we observe in NPA-CARN mice . At present , the basis for this discrepancy remains unclear . Our finding that AR deletion results in failure of tumor formation following Pten inactivation could be due to differences between CARNs and bulk luminal cells and/or to differences due to Pten loss in the regressed versus hormonally intact prostate epithelium . In principle , these possibilities could potentially be distinguished using inducible Cre-drivers to delete Pten in bulk luminal cells in regressed versus hormonally intact mice . Our findings on CARNs in mouse models may be of potential relevance for human prostate biology and cancer . Although CARNs are defined in the regressed prostate epithelium , and our in vivo studies involve manipulations performed after castration in mice , there is evidence that CARN-like cells exist in the human prostate from tissue-slice culture experiments ( Zhao et al . , 2010 ) , as well as from analyses of prostate tumors after androgen-deprivation ( Germann et al . , 2012 ) . However , it is less clear whether multipotent luminal progenitors can be identified in the context of the hormonally intact human prostate . Previous lineage-reconstruction studies using patterns of mitochondrial DNA mutations have indicated the existence of multipotent epithelial progenitors ( Blackwood et al . , 2011; Gaisa et al . , 2011 ) , and recent work has provided evidence for multipotent basal progenitors localized to the most proximal region of the prostate as well as more distally located unipotent luminal progenitors ( Moad et al . , 2017 ) . Notably , ex vivo studies of human prostate organoids have demonstrated the existence of bipotential luminal progenitors ( Karthaus et al . , 2014 ) . Thus , we believe that current data favor a general similarity of epithelial lineage relationships in the two species , suggesting that findings deduced from analyses of mice may be translatable to the human prostate . The ability of CARNs to retain progenitor properties even in the absence of AR raises the possibility that CARNs represent a cell of origin for prostate cancers that are particularly susceptible to develop castration-resistance . Notably , under conditions of AR down-regulation , such as those that may occur during aging or inflammation , CARNs that lack tumor suppressors such as PTEN may represent a latent target for subsequent oncogenic events that can confer tumor growth , such as those activating the ERK MAP kinase pathway . Interestingly , our bioinformatic analyses of the ADCA cell line signature shows enrichment with castration-resistance signatures based on expression data from human prostate cancer patients ( Best and Stanbrough signatures ) , consistent with increasing evidence supporting AR-independent mechanisms of castration-resistance ( Beltran et al . , 2014; Vlachostergios et al . , 2017; Watson et al . , 2015 ) . In addition , the observed enrichment with the Beltran CRPC-NE signature suggests a similarity in gene expression programs with advanced cancers that lack AR activity , as neuroendocrine differentiation in prostate tumors is associated with loss of AR expression ( Beltran et al . , 2011 ) . Notably , consistent with a role for AR loss in the emergence of neuroendocrine phenotypes , tumors in NPKA-CARN mice can display focal neuroendocrine differentiation , which has also been recently described in other mouse models of advanced prostate cancer ( Ku et al . , 2017; Zou et al . , 2017 ) . In this regard , we note that the NP-CARN and NPK-CARN mice develop tumor phenotypes similar to those in NP and NPK mice , which have the same genotypes but whose tumors are induced by the Nkx3 . 1CreERT2 driver in hormonally intact adult prostate ( Aytes et al . , 2013; Floc'h et al . , 2012 ) . Interestingly , NP tumors are initially castration-sensitive ( Floc'h et al . , 2012 ) , consistent with the inability of NPA-CARN mice to develop tumors , whereas NPK tumors are castration-resistant ( Aytes et al . , 2013 ) , consistent with the phenotype of NPKA-CARN tumors . The molecular basis for this switch is currently unclear , but it is conceivable that it involves ETS family transcription factors , which are known to interact with AR to positively and negatively modulate its activity ( Baena et al . , 2013; Bose et al . , 2017; Chen et al . , 2013 ) ; interestingly , ETV4 is up-regulated in NPK tumors and may be involved in this switch ( Aytes et al . , 2013 ) . However , the focal neuroendocrine differentiation observed in NPKA-CARN tumors suggests that oncogenic transformation of AR-deleted CARNs can also facilitate transdifferentiation of luminal cells to neuroendocrine fates , as we have demonstrated for a Pten and Trp53 mutant mouse model ( NPp53 ) after anti-androgen treatment ( Zou et al . , 2017 ) . Finally , since tumors initiated from CARNs following combined Pten deletion and Kras activation are at least partially independent of AR from their outset , it is conceivable that such tumors are intrinsically more resistant to second-generation anti-androgen therapies . Interestingly , recent studies have also identified distinct castration-resistant progenitors that express Bmi1 ( CARBs ) that are cells of origin for prostate cancer ( Yoo et al . , 2016 ) . The development of targeted therapies directed at molecular features of CARNs and/or other castration-resistant luminal cells may therefore be relevant for successful combination with anti-androgen therapies . The Nkx3 . 1CreERT2 driver ( Nkx3-1tm4 ( cre/ERT2 ) Mms ) has been previously described ( Wang et al . , 2009 ) . Mice carrying the R26R-YFP ( B6 . 129 × 1-Gt ( ROSA ) 26Sortm1 ( EYFP ) Cos/J ) reporter ( Srinivas et al . , 2001 ) were obtained from the Jackson Laboratory Induced Mutant Resource . Mice carrying the conditional Ptenflox ( B6 . 129S4-Ptentm1Hwu/J ) allele ( Lesche et al . , 2002 ) and the inducible Kraslsl-G12D ( B6 . 129-Krastm4Tyj/Nci ) allele ( Jackson et al . , 2001 ) were obtained from the National Cancer Institute Mouse Models of Human Cancer Consortium Repository . Mice with the conditional Arflox ( B6N . 129-Artm1Verh/Cnrm ) allele ( De Gendt et al . , 2004 ) was obtained from the European Mouse Mutant Archive . Animals were maintained on a congenic C57BL/6N background . Genotyping was performed using the primers listed in Supplementary file 1A . Primer sequences used for genotyping of Ar alleles were previously described ( Yeh et al . , 2003 ) . For lineage-marking and simultaneous deletion of AR in CARNs , Nkx3 . 1CreERT2/+; Arflox/Y; R26R-YFP/+ males were castrated at 8 weeks of age and allowed to regress for 4 weeks , followed by administration of tamoxifen ( Sigma; 9 mg/40 g body weight in corn oil ) by daily oral gavage for four consecutive days , and a chase period of 4 weeks . Administration of testosterone for prostate regeneration ( Sigma; 25 mg/ml in 100% ethanol and diluted in PEG-400 to a final concentration of 7 . 5 mg/ml ) was performed by subcutaneous implantation of mini-osmotic pumps ( Alzet ) that release testosterone solution at a rate of 1 . 875 μg/hr , which yields physiological levels of serum testosterone ( Banach-Petrosky et al . , 2007 ) . For BrdU incorporation experiments , BrdU ( Sigma; 100 mg/kg ) was administered by intraperitoneal injection twice daily for 4 consecutive days , either from days 1 through 4 or from days 11 through 14 during androgen-mediated regeneration . For cell of origin experiments , Nkx3 . 1CreERT2/+; Ptenflox/flox; Arflox/Y; R26R-YFP/+ and Nkx3 . 1CreERT2/+; Ptenflox/flox; KrasLSL-G12D/+; Arflox/Y; R26R-YFP/+ mice as well as corresponding controls were castrated at 8 to 12 weeks of age . One month later , mice were administered tamoxifen , with a chase period of 3 months , followed by androgen-mediated regeneration for 1 month; mice were then euthanized for analysis . All animal experiments were performed according to protocols approved by the Institutional Animal Care and Use Committee at Columbia University Medical Center . Radical prostatectomy samples were obtained from consented patients under the auspices of an Institutional Review Board approved protocol at Columbia University Medical Center . Tissue from benign prostate regions was dissected and transported to the laboratory in DMEM/F12 ( Gibco ) supplemented with 5% FBS . Benign pathology was first determined by H and E-staining of snap-frozen sections , and subsequently confirmed by immunostaining of paraffin sections for p63 and AMACR . Tissue dissociation and isolation were performed as previously described ( Chua et al . , 2014 ) . In brief , mouse prostate tissue from all lobes was dissected in cold phosphate buffered saline ( PBS ) and minced with scissors . For human prostate specimens , tissue was cut into small pieces with scalpels , washed with PBS with 4 mg/ml Gentamicin ( Gibco ) , and then minced with scissors . Both mouse and human prostate tissues were then incubated in DMEM/F12 ( Gibco ) supplemented with 5% FBS and 1:10 dilution of collagenase/hyaluronidase ( STEMCELL Technologies ) at 37°C for 3 hr . Dissociated tissues were spun at 350 g for 5 min , and resuspended in ice-cold 0 . 25% trypsin-EDTA ( STEMCELL Technologies ) , followed by incubation at 4°C for 1 hr . Trypsinization was stopped by addition of Modified Hank’s Balanced Salt Solution ( HBSS ) ( STEMCELL Technologies ) supplemented with 2% FBS . After centrifugation at 350 g , pelleted cells were resuspended with pre-warmed 5 mg/ml dispase ( STEMCELL Technologies ) supplemented with 1:10 dilution of 1 mg/ml DNase I ( STEMCELL Technologies ) , triturated vigorously for 1 to 2 min , and diluted by addition of HBSS/2% FBS . Finally , the cell suspension was passed through a 40 μm cell strainer ( Falcon ) . For flow sorting of mouse prostate epithelial cells , cell suspensions were stained on ice for 25 min with fluorescent-tagged EpCAM ( BioLegend #118214 ) antibody . For isolation of human prostate epithelial cells , we used fluorescent-tagged EpCAM ( BioLegend #324208 , specific for human ) and E-cadherin ( eBioscience #46-3249-82 ) antibodies . The stained cells were spun , and cell pellets washed with HBSS/2% FBS , followed by resuspension in HBSS/2% FBS with 10 µM Y-27632 ( ROCK inhibitor; STEMCELL Technologies ) and a 1:1000 dilution of 0 . 5 mg/ml DAPI to exclude dead cells . Both side-scatter pulse width ( SSC-W ) vs . area ( SSC-A ) and forward side-scatter pulse area ( FSC-A ) vs . heights ( FSC-H ) were used to isolate single dissociated cells . To establish cell lines from lineage-marked CARNs as well as benign prostate epithelial cells , we performed adherent culture in our prostate organoid medium ( Chua et al . , 2014 ) , consisting of hepatocyte medium supplemented with 10 ng/ml epidermal growth factor ( EGF ) ( Corning ) , 10 μM Y-27632 ( STEMCELL Technologies ) , 1x glutamax ( Gibco ) , 5% Matrigel ( Corning ) , 5% charcoal-stripped FBS ( Gibco ) heat-inactivated at 55°C for 1 hr , and supplemented with either 100 nM or 1 nM DHT ( Sigma ) for mouse and human cells , respectively . To derive APCA and ADCA lines , single YFP+ cells from castrated and tamoxifen-treated Nkx3 . 1CreERT2/+; Arflox/Y; R26R-YFP/+ mice were flow-sorted directly into 96-well Primaria plates ( Corning ) , and were monitored daily to assess colony formation . Successful colonies were expanded and genotyped to assess the status of the Arflox allele . For derivation of lines from benign human prostate epithelium , cells expressing either EpCAM and/or E-cadherin were plated into six-well Primaria plates at a density of 100 , 000 cells/well . Passaging of adherent cultures was performed by removal of accumulated Matrigel on surface of the cells by gentle washing . The cells were washed with cold PBS , treated with 0 . 25% trypsin for 5 min at 37°C , and mechanically dissociated . Medium was changed every 4 days . Adherent cells were frozen in media consisting of 80% FBS , 10% complete medium , and 10% DMSO . Each APCA and ADCA line has been propagated continuously for at least eight passages . To assess cell viability , APCA and ADCA lines were plated in 96-well Primaria plates at a density of 1000 cells/well in the presence or absence of DHT . Cell viability was assayed at days 1 , 2 , 4 and 6 after plating using CellTiter-Glo 3D ( Promega ) , with five technical replicates for each time point . In brief , CellTiter-Glo 3D reagent was thawed at 4°C and brought to room temperature prior to use . 100 μl of the reagent was added into each well containing 100 μl of medium . After shaking for 5–10 min , the mixture was then transferred to a 96-well CELLSTAR plate ( Greiner ) , followed by incubation at room temperature for 10 min prior to measurement using a luminometer plate reader . To assess colony formation , APCA and ADCA lines were plated in six-well Primaria plates at a density of 500 cells/well and grown for 9 days . three technical replicates were performed for each line in the presence or absence of DHT . At day 10 after plating , wells were washed with PBS and fixed with 100% methanol for 5 min . The wells were then washed with PBS for three times before staining with filtered 0 . 1% crystal violet solution . After drying the plates , colonies were counted , with a colony defined as a cell cluster containing at least 50 cells . For tissue recombination , 10 YFP+ cells from castrated and tamoxifen-treated Nkx3 . 1CreERT2/+; Arflox/Y; R26R-YFP/+ mice or control Nkx3 . 1CreERT2/+; R26R-YFP/+ mice were combined with 250 , 000 dissociated rat urogenital mesenchyme cells from embryonic day 18 . 5 embryos , and resuspended in 15 μl of 9:1 collagen:setting buffer solution ( 10x Earle’s Balanced Salt Solution ( Life Technologies ) , 0 . 2 M NaHCO3 , and 50 mM NaOH ) . The recombinants were cultured overnight in DMEM with 10% FBS and 100 nM DHT , followed by grafting under the kidney capsules of male NOD . Cg-Prkdcscid Il2rgtm1Sug/JicTac ( NOG ) mice ( Taconic ) . Renal grafts were harvested for analysis at 7–12 weeks after grafting . For the experiment involving APCA and ADCA lines , 100 , 000 cells were recombined with 250 , 000 rat urogenital mesenchyme cells , followed by grafting . At 6 weeks after grafting , some mice implanted with APCA cells were treated with tamoxifen to induce Ar deletion . Grafts were harvested for analysis after 12 weeks of growth and analyzed in paraffin sections for the presence of ducts expressing YFP . ( Note that ducts can also be formed by YFP–cells that are derived from contaminating rat urogenital epithelium due to incomplete separation from the urogenital mesenchyme . ) Graft efficiency was calculated on the basis of the presence of YFP+ ducts in the grafts using control CARNs and on the presence of YFP+AR– ducts in the grafts using AR-deleted CARNs . For cryosections , tissues were fixed in 4% paraformaldehyde in PBS at 4°C overnight , placed in 30% sucrose in PBS overnight , and transferred to 1:1 30% sucrose in PBS and OCT ( Tissue-Tek ) solution for at least 4 hr prior to embedding in OCT . For paraffin sections , tissues were fixed in 10% formalin for 1 to 2 days , depending on size of tissue , prior to processing and embedding . Hematoxylin-eosin staining was performed using standard protocols . For immunostaining , sections underwent antigen-retrieval by heating in citrate acid-based or tris-based antigen unmasking solution ( Vector Labs ) for 45 min . Primary antibodies were applied to sections and incubated at 4°C overnight in a humidified chamber . Alexa Fluors ( Life Technologies ) were used as secondary antibodies . In some cases , tyramide amplification ( Life Technologies ) or ABC Elite ( Vector Labs ) kits together with HRP-conjugated or biotinylated secondary antibodies and NovaRed kit were used for signal detection . For immunofluorescent staining of cells , 5000 adherent cells/well were seeded on a eight-well Lab-Tek Chamber Slide ( Nunc ) , grown for 4–8 days , and fixed with 4% paraformaldehyde for 10 min . After washing the slides with 3 changes of PBS , immunostaining was performed as above without antigen retrieval . Details of antibodies used are provided in Supplementary file 1B . Histological grading of mouse prostate phenotypes was performed according to ( Park et al . , 2002 ) . For lineage-tracing experiments , quantitation of marker staining was performed by manual counting of cells from confocal images taken with a 40x objective . For RNA preparation , APCA and ADCA cell lines at passage 5 or 6 were grown to approximately 70–80% confluency in Primaria 6-well plates in the presence of DHT , and lysed in Trizol . Total RNA extraction was performed using the ‘No Spin’ method of the MagMAX-96 for Microarrays kit ( Ambion ) . Library preparation and RNA sequencing was performed by the Columbia Genome Center using their standard pipeline . In brief , mRNA was enriched by poly-A pull-down , and library preparation was performed using an Illumina TruSeq RNA prep kit . Libraries were pooled and sequenced using an Illumina HiSeq2500 instrument , yielding approximately 30 million single-ended 100 bp reads per sample . RTA ( Illumina ) was used for base calling and bcl2fastq2 ( version 2 . 17 ) for conversion of BCL to fastq format , coupled with adaptor trimming . Reads were mapped to the mouse genome ( UCSC/mm10 ) using STAR ( 2 . 5 . 2b ) and FeatureCounts ( v1 . 5 . 0-p3 ) . RNA-seq data raw counts were normalized and the variance was stabilized using DESeq2 package ( Bioconductor ) in R-studio 0 . 99 . 902 , R v3 . 3 . 0 ( The R Foundation for Statistical Computing , ISBN 3-900051-07-0 ) . Differential gene expression signatures were defined as a list of genes ranked by their differential expression between any two phenotypes of interest ( e . g . APCA versus ADCA lines; CRPC-NE versus CRPC , etc . ) , estimated using a two-sample two-tailed Welch t-test ( for n ≥ 3 ) or fold-change ( for n < 3 ) . For comparison of a mouse gene signature with a human gene signature , mouse genes were mapped to their corresponding human orthologs based on the homoloGene database ( NCBI ) . Signatures were compared using Gene Set Enrichment Analysis ( GSEA ) ( Subramanian et al . , 2005 ) , with the significance of enrichment estimated using 1000 gene permutations . Pathway enrichment analysis was performed using the C2 database , which includes pathways from REACTOME ( Fabregat et al . , 2016 ) , KEGG ( Ogata et al . , 1999 ) , and BioCarta ( http://www . biocarta . com/genes/allpathways . asp ) . Expression data are deposited in the Gene Expression Omnibus database under GSE99233 . Statistical analysis was performed using the Statistical Package for the Social Sciences ( SPSS ) . Data distribution was assessed by the Kolmogorov-Smirnov test . Arcsine transformation was performed on data with non-normal distribution . Two-sample two-tail Welch t-test or Fisher’s Exact Test was performed for comparison between two independent groups as appropriate . No statistical methods were used to pre-determine sample size , and experiments were not randomized; investigators were not blinded to allocation during experiments and outcome assessment .

Most prostate tumors rely on male hormones – called androgens – to survive . Aggressive prostate cancer is often treated with drugs that block androgens , which usually cause the prostate tumors to shrink . One class of the drugs works by binding to and inactivating the androgen receptor protein on prostate cancer cells . However , aggressive prostate tumors can often become resistant to these anti-androgen therapies . It is not clear where the resistant cancer cells come from . In 2009 , researchers showed that the normal prostate contains some cells that appear to be independent of androgens . A subset of these cells – also known as CARNs – can act as stem or progenitor cells that can repair the prostate after injury . These normal androgen-independent cells can also be the cells from which prostate tumors arise . Here , Chua et al . – including one of the researchers from the 2009 study – investigated how these CARN cells behave when the androgen receptor is deleted . When the androgen receptor was genetically removed in CARN cells of otherwise healthy mice , the behavior of CARN cells was unaffected . When the androgen receptor was deleted together with a protein that normally suppresses the formation of tumors , it protected the mice from prostate cancer . However , Chua et al . also observed that deleting the androgen receptor could not prevent the tumor from growing when two cancer-causing mutations were present . These tumors were similar to human prostate tumors that are resistant to anti-androgen therapy . Since CARN cells may also exist in humans , this new way of making prostate cancers in mice may be used to study how these resistances arise in patients . A better understanding of how prostate tumors develop might lead to new treatments in which the androgen receptor is blocked in combination with other new protein targets .

2018

Differential requirements of androgen receptor in luminal progenitors during prostate regeneration and tumor initiation

Lysosomes are classically viewed as vesicular structures to which cargos are delivered for degradation . Here , we identify a network of dynamic , tubular lysosomes that extends throughout Drosophila muscle , in vivo . Live imaging reveals that autophagosomes merge with tubular lysosomes and that lysosomal membranes undergo extension , retraction , fusion and fission . The dynamics and integrity of this tubular lysosomal network requires VCP , an AAA-ATPase that , when mutated , causes degenerative diseases of muscle , bone and neurons . We show that human VCP rescues the defects caused by loss of Drosophila VCP and overexpression of disease relevant VCP transgenes dismantles tubular lysosomes , linking tubular lysosome dysfunction to human VCP-related diseases . Finally , disruption of tubular lysosomes correlates with impaired autophagosome-lysosome fusion , increased cytoplasmic poly-ubiquitin aggregates , lipofuscin material , damaged mitochondria and impaired muscle function . We propose that VCP sustains sarcoplasmic proteostasis , in part , by controlling the integrity of a dynamic tubular lysosomal network . Valosin-containing protein ( VCP ) , the homologue of yeast Cdc48 , is the causative gene for a multisystem degenerative disease that was originally termed IBMPFD to encompass the wide range of debilitating clinical outcomes , including inclusion body myopathy ( IBM ) , Paget's disease of the bone ( PDB ) and frontotemporal dementia ( FD ) ( Watts et al . , 2004 ) . Recently , the list of degenerative disorders that are associated with VCP mutations has expanded to include amyotrophic lateral sclerosis ( ALS ) ( Abramzon et al . , 2012 ) , spastic paraplegia ( Clemen et al . , 2010 ) , scapuloperoneal muscular dystrophy ( Liewluck et al . , 2014 ) and Charcot-Marie-Tooth disease ( Gonzalez et al . , 2014 ) . Currently , there are no viable treatments available to slow or halt progression of VCP-related diseases . Muscle weakness is the first presenting symptom in over 50% of VCP disease patients ( Weihl et al . , 2009 ) , yet very little is known about the muscle pathology of VCP-related diseases . Muscle biopsies from patients with VCP-related diseases display an accumulation of cytoplasmic poly-ubiquitin aggregates ( Watts et al . , 2004; Weihl et al . , 2009; Dolan et al . , 2011 ) , suggesting a major defect in protein clearance . VCP is an AAA-ATPase that has essential functions in ubiquitin-dependent proteolysis . But , pathogenic mutations in VCP do not seem to impair the UPS or ERAD protein degradation pathways ( Tresse et al . , 2010a; Chang et al . , 2011 ) . More recently , VCP has been implicated in autophagy . Specifically , over-expression of VCP mutant transgenes with disease causing mutations leads to an accumulation of autophagosomes ( Ju et al . , 2009; Tresse et al . , 2010a ) , suggesting that VCP functions in processes related to the maturation or fusion of autophagosomes with lysosomes . Lysosomes are the major cellular degradation sites for clearing damaged proteins and organelles . Lysosomes are classically thought to be vesicular organelles , where they serve as depots for cargo delivered via endosomes or autophagosomes . However , in certain systems lysosomes have been observed to adopt non-vesicular morphologies . A particularly dramatic example has been observed in a subset of bone-derived cultured macrophages , where lysosomes form abundant , extended tubules that radiate from the cell center and , in some cases , form an interconnected web throughout the cytoplasm ( Swanson et al . , 1987a; Knapp and Swanson , 1990 ) . Additionally , there are examples where cellular stress , particularly the induction of high levels of autophagy , induces lysosomal membranes to tubulate and undergo scission to produce new vesicular lysosomes , a process referred to as autophagic lysosome reformation ( ALR ) ( Yu et al . , 2010 ) . Despite these observations , lysosome tubules have received little attention and it remains unclear to what extent lysosome tubulation occurs in different cell types and what purpose it serves in vivo . Moreover , the molecular repertoire of factors required for lysosome tubule formation is virtually unknown . Here , we employ fluorescently tagged lysosomal and autophagic markers to study the autophagy-lysosome system in Drosophila muscle cells and investigate the muscle pathology of VCP-related diseases . Remarkably , we find that lysosomes adopt a dynamic , tubular morphology that ramifies throughout the entire sarcoplasm of Drosophila muscle , in vivo . We find that VCP is required for the integrity and dynamics of this tubular network . Disruption of lysosome tubules correlates with defects in autophagosome-lysosome fusion , increased poly-ubiquitin aggregates and the accumulation of lipofuscin material in the sarcoplasm . We show that the human VCP homolog can rescue lysosomal tubulation following loss of VCP in Drosophila muscle , indicating that the functions of VCP in lysosome tubulation are conserved . Finally , we demonstrate that homologous mutations that cause VCP diseases in human patients disrupt the lysosome tubular lattice , suggesting that disruption of lysosome tubules contributes to VCP mutant pathogenesis . Taken together , our data establish a functional link between lysosome tubule dysfunction and the pathology of VCP-related degenerative diseases . To visualize muscle lysosomes in vivo , we expressed RFP-tagged Spinster , which has previously been defined as a late endosomal/lysosomal transmembrane protein ( Sweeney and Davis , 2002; Dermaut et al . , 2005 ) . Remarkably , when Spinster-RFP is expressed in Drosophila muscle , Spin-RFP localizes to an expansive tubular network ( Figure 1A–C and Video 1 ) . Tubules were found evenly distributed throughout the sarcoplasm and formed a web of connections with other tubules ( Figure 1B , C ) . Tubules were observed in every muscle and there were no apparent differences in the tubule abundance or architecture between different muscles . We also observed enlarged vesicular compartments at tubule intersections throughout the muscle ( Figure 1C ) . This network is highly sensitive to all chemical fixation conditions that we have attempted . When subjected to fixation , the tubule lattice collapses ( Figure 1D ) , leaving behind distributed round , Spinster-positive compartments that resemble classically defined late-endosomal , lysosomal structures ( Sweeney and Davis , 2002; Dermaut et al . , 2005 ) . Thus , live imaging is essential to study the function and relevance of this Spin-positive , tubular network . 10 . 7554/eLife . 07366 . 003Figure 1 . Lysosomes adopt an extended dynamic tubular array in Drosophila sarcoplasms . ( A ) Muscles of third instar larvae from segment A2 . ( B ) Representative live image of Spin-RFP expressed in muscles at 63× magnification . Muscle 4 ( B ) is shown . DNA was stained with Hoescht . ( C ) Representative live image of Spin-RFP expressed in muscles using the muscle-specific MHC-Gal4 driver . DNA was stained with Hoescht . ( D ) Representative image of Spin-GFP localization in a muscle that was fixed with 4% PFA prior to imaging . ( E ) Time-lapse images of the Spin-RFP network . Time 0 is represented in magenta , and the 5 min time-point is represented in green . The 2 time points were merged to show new and lost tubule formations over the course of 5 min . Arrows indicate examples of de novo tubule formations and the asterisk indicates a retracted tubule . ( F ) Representative time-lapse sequence of a Spin-GFP tubule fission event . ( G ) Representative time-lapse sequence of a Spin-GFP tubule fusion event . In the last frame , a de novo tubule can be seen extruding from the middle of a pre-existing tubule . ( H–J ) Spin-RFP localization in muscles treated with DMSO ( H ) , Nocodazole ( I ) or LatA ( J ) . ( K ) Spin-RFP localization in muscles expressing Clathrin heavy chain ( Chc ) RNAi . DOI: http://dx . doi . org/10 . 7554/eLife . 07366 . 00310 . 7554/eLife . 07366 . 004Video 1 . Spin-RFP tubular network in Drosophila muscle . Spin-RFP was expressed in muscles and imaged live . Average Z-stacks were assembled to produce a 3D volume projection and various angles of the projections are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 07366 . 004 We next explored the dynamics of this tubular network in time-lapse videos . We find evidence of tubule extension , retraction , scission and fusion ( Figure 1E–G ) . The dominant activity in the network is the dynamic extension and retraction of individual tubules throughout the sarcoplasm ( Figure 1E and Video 2 ) . Often the same tubule was observed to extend and retract , while surrounding tubules remained constant . The impression is that the lattice has both stable , interconnected tubules and dynamic elements that create new connections or retract once a connection is broken . In less frequent instances , we observed scission events from the end of a tubule , resulting in a mobile vesicle ( Figure 1F ) . In a corollary phenomenon , we observed tubule extensions that resulted in tubules fusing to larger nodes in the network ( Figure 1G ) . Finally , we observed de novo formation of tubules extruding from the side of existing tubules ( Figure 1G , last panel ) demonstrating that tubulation can originate from existing tubules , not just pre-existing nodes . Importantly , the existence of a tubular network and tubule dynamics are not an artifact of a dissected neuromuscular preparation . We observed an identical , dynamic tubular lysosomal network in intact larvae that were imaged through the cuticle while restrained in a microfluidics chamber ( Video 3 ) . 10 . 7554/eLife . 07366 . 005Video 2 . Spin-RFP tubule dynamics in Drosophila muscle . Representative time-lapse video of Spin-RFP expressed in muscles . Frames were taken at 10 s intervals . DOI: http://dx . doi . org/10 . 7554/eLife . 07366 . 00510 . 7554/eLife . 07366 . 006Video 3 . Spin-RFP tubule dynamics in an intact larva . A whole un-dissected larva was immobilized in a mircrofluidics chamber and Spin-RFP was imaged in the body-wall muscle through the transparent cuticle . Frames were taken at 10 s intervals . DOI: http://dx . doi . org/10 . 7554/eLife . 07366 . 006 The observed lysosomal network structure and dynamics are consistent with the involvement of either the actin or microtubule cytoskeletons . First , we tested whether the tubules require an intact microtubule cytoskeleton . Treatment with nocodazole for 1 hr completely abolished tubules , indicating that the tubules require microtubule support ( Figure 1H , I ) . In contrast , disruption of the actin cytoskeleton with latrunculin A did not have a significant effect on the tubular network , indicating a non-essential role for the actin cytoskeleton in maintaining lysosome tubules ( Figure 1J ) . Finally , we tested whether Clathrin is essential for tubular network integrity . Clathrin has the capacity to shape membranes and was recently implicated in the process of ALR in cultured mammalian cells ( Rong et al . , 2012 ) , a process that involves limited tubulation from auto-lysosomal compartments . Expression of Clathrin heavy chain RNAi ( Chc-RNAi ) completely disrupted network integrity ( Figure 1K ) . Although Spinster was characterized previously as a lysosomal marker , the tubular structures labeled by Spinster are dramatically different from the classical view of vesicular lysosomes . Therefore , we performed additional experiments to validate that the tubular network is lysosomal . First , we co-imaged Spin-GFP with the low pH fluorescent probe Lysotracker and found complete co-localization ( Figure 2A ) , indicating that the Spinster network is acidic . To verify that the observed tubular network is not an artifact of Spin-RFP over-expression , we stained wild type muscles with Lysotracker to examine lysosome morphology under wild type conditions . Lysotracker staining confirmed the existence of this tubular network in wild type muscle ( Figure 2B ) . We also note that the tubule intersections stained more intensely for Lysotracker than the tubules themselves . This could be due to increased tubule volume at the intersections , or these sites might actually have a lower pH than the tubules themselves . Finally , we co-imaged Spin-RFP with several other organelle markers to verify that Spinster specifically labels lysosomes and does not co-stain other organelles . ER and mitochondria also form tubule structures , but when we co-imaged Spin-RFP with ER-tracker and Mito-tracker fluorescent dyes , Spin-RFP tubules did not co-localize with either ER or mitochondria tubules ( Figure 2C , D ) . Instead , Spin-RFP tubules were interwoven between mitochondria and ER tubules . Additionally , Spin-RFP did not co-localize with markers for early endosomes ( YFP-Rab5 ) , recycling endosomes ( Rab11-GFP ) , medial Golgi ( ManII-GFP ) or trans Golgi ( GalT-YFP ) ( Figure 2E–H ) . We note that Golgi organization in muscles forms vesicular structures rather than the classical Golgi stacks that are observed in most cell types and this is consistent with Golgi organization that has been observed in vertebrate skeletal muscles ( Ralston et al . , 2001 ) . Collectively , we have identified and characterized a dynamic tubular lysosomal network that permeates the entire sarcoplasm of Drosophila body-wall muscle in vivo . 10 . 7554/eLife . 07366 . 007Figure 2 . Spin-RFP tubules do not co-localize with mitochondria , ER , golgi or early endosomes . ( A ) Co-imaging of Spin-GFP and Lysotracker Red staining . ( B ) Lysotracker staining of wild type muscles . ( C–H ) Co-imaging of Spin-RFP with ER tracker ( C ) , Mito tracker ( D ) , YFP-Rab5 ( E ) , Rab11-GFP ( F ) , ManII-GFP ( G ) , GalT-YFP ( H ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07366 . 007 To investigate the molecular underpinnings of the observed lysosomal tubule dynamics , we pursued a candidate-based RNAi screen to identify genes required for lysosome tubulation . We focused on genes that have been implicated in the autophagy-lysosome system and identified the AAA-ATPase VCP as being required for the integrity of the entire tubular lysosome network . Specifically , inhibiting VCP expression by RNAi abolished lysosome tubules , leaving behind vesicles throughout the sarcoplasm ( Figure 3A , B ) . The lysosome vesicles were irregular in their size and shape and appeared clustered , rather than uniformly distributed throughout the sarcoplasm . To determine whether the catalytic ATPase function of VCP is required for the tubular network integrity , we employed a VCP-selective inhibitor DBeQ ( Chou et al . , 2011 ) . Acute inhibition of VCP with DBeQ completely disrupted the tubular network after 3 hr ( Figure 3C , D ) , demonstrating the required catalytic function of VCP . Furthermore , time-lapse imaging of Spin-GFP after treatment with DBeQ for 4 hr revealed that the remaining vesicular lysosome structures are completely static ( Video 4 ) . 10 . 7554/eLife . 07366 . 008Figure 3 . VCP inhibition disrupts the lysosome tubule lattice and human VCP rescues this defect . ( A ) Representative live image of Spin-RFP expressed in muscle using the muscle-specific BG57-Gal4 driver . ( B ) Live image of Spin-RFP in muscles expressing VCP-RNAi using the muscle-specific BG57-Gal4 driver . ( C , D ) Live images of Spin-RFP expressed in muscles that were treated with DMSO ( C ) or the VCP-specific inhibitor DBeQ ( D ) for 4 hr . ( E ) Live image of Spin-GFP expressed in muscles using the muscle-specific BG57-Gal4 driver . ( F ) Live image of Spin-GFP in muscles expressing VCP-RNAi using the muscle-specific BG57-Gal4 driver . ( G ) Live image of Spin-GFP in muscles that co-express VCP-RNAi and human VCP ( hVCP ) using the muscle-specific BG57-Gal4 driver . ( H–J ) Lysotracker staining in wild type ( H ) muscles or muscles expressing parkin-RNAi ( I ) or tbph-RNAi ( J ) . ( K , L ) Spin-RFP localization in muscles treated with DMSO ( K ) or tunicamycin ( TM ) ( L ) . ( M ) Western blot analysis of total Hsc70/BiP protein levels . Tubulin serves as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 07366 . 00810 . 7554/eLife . 07366 . 009Video 4 . Spin-GFP dynamics in muscles expressing VCP-RNAi . Representative time-lapse video of Spin-GFP in muscles expressing VCP-RNAi . Frames were taken at 10 s intervals . DOI: http://dx . doi . org/10 . 7554/eLife . 07366 . 009 To determine if the role of VCP in maintaining the tubular lysosomal network in muscle cells is conserved , we overexpressed human VCP in dVCPRNAi muscles . Human VCP should be completely resistant to dVCPRNAi due to lack of extended stretches of nucleotide identity . Over-expressing human VCP in dVCPRNAi muscles rescued the formation of lysosome tubules in every muscle ( Figure 3E–G ) . These rescue data confirm that VCP knockdown is the cause of disrupted lysosomal network integrity and demonstrate that VCP-dependent activity is conserved in the human VCP protein . Since VCP knockdown destroys the sarcoplasmic tubular lysosomal network prior to obvious cellular degeneration , we speculated that dismantling of the network could be a precursor to cellular degeneration . However , it is also possible that loss of this network is a secondary correlate of impaired muscle health . To address this issue , we examined two additional RNAi-mediated conditions that cause muscle degeneration , looking for the presence or absence of tubular lysosomes . Specifically , we expressed RNAi against parkin and tbph , Drosophila orthologues of genes linked to Parkinson's and ALS , respectively . These RNAi have been shown to cause muscle degeneration in Drosophila ( Diaper et al . , 2013; Cornelissen et al . , 2014 ) . Remarkably , we did not observe any significant effect on lysosome tubules ( Figure 3H–J ) when parkin-RNAi and tbph-RNAi are expressed with BG57-Gal4 , the same Gal4 line used to express VCP-RNAi throughout our studies . VCP has well-established roles in the ERAD pathway and loss of VCP causes ER stress ( Wójcik et al . , 2006 ) . Thus , it is possible that loss of tubular lysosomes is caused indirectly by ER stress . To address this possibility , we treated muscles with tunicamycin ( TM ) to induce ER stress by another means and examined the tubular network . Upon treatment with TM for 4 hr , the Spin-RFP tubular network remained intact ( Figure 3K , L ) . We verified that ER stress was induced by examining levels of Hsc70/BiP , a marker of ER stress , in protein lysates derived from the same animals that were used for imaging . Total protein levels of Hsc70/BiP were significantly increased in both TM and DBeQ treated muscles ( Figure 3M ) . Thus , tubular network disruption is not a byproduct of ER stress . Taken together , our data are consistent with three conclusions: ( 1 ) VCP loss of function disrupts the integrity of the tubular lysosomal network , ( 2 ) the role of VCP in maintaining tubular lysosomes is conserved and ( 3 ) disruption of the tubular network is specific for VCP loss of function and not a secondary byproduct of muscle degeneration or ER stress . Muscle biopsies from patients with VCP-related disease display an accumulation of cytoplasmic poly-ubiquitin aggregates ( Watts et al . , 2004; Weihl et al . , 2009; Dolan et al . , 2011 ) , suggesting a defect in protein clearance . This led us to explore the intersection of the observed tubular lysosomal network and autophagy . First , we co-imaged Spin-GFP with the autophagosome membrane marker mCherry-Atg8a/LC3 to determine the relationship between autophagosomes and the tubular lysosome network . Remarkably , we find that mCherry-Atg8a precisely co-localizes with Spin-GFP-labeled tubules ( Figure 4A ) . The mCherry-Atg8a labeling is widely distributed within the network , but does not label the full extent of every tubule ( Figure 4A ) . Time-lapse imaging revealed mCherry-Atg8a labeling is dynamic within Spin-GFP tubules ( Video 5 ) . Some mCherry-Atg8a puncta appear to traffic through the Spin-GFP tubules suggesting that autophagosome membranes and/or cargos are dynamic within the tubular lysosome network ( Figure 4B ) . Atg8 localizes to the autophagophore membrane prior to the formation of an enclosed autophagosome , at which point Atg8 is shed from the outer autophagosome membrane ( Xie and Klionsky , 2007 ) . After autophagosomes fuse with lysosomes , lysosomal enzymes degrade the inner autophagosome membrane containing Atg8 ( Xie and Klionsky , 2007 ) . Given this sequence of events , we propose that the mCherry-Atg8a positive regions of the lysosome tubules represent ongoing degradation of autophagosome membranes following fusion with the Spin-GFP positive lysosomal network . 10 . 7554/eLife . 07366 . 010Figure 4 . Autophagosomes co-localize with the tubular lysosomal network . ( A ) Representative live image of Spin-GFP and mCherry-Atg8a co-expressed in muscles using the muscle-specific BG57-Gal4 driver . ( B ) Representative time-lapse sequence of Spin-GFP and mCherry-Atg8a in muscle . The arrow follows a mCherry-Atg8a positive puncta trafficking along a Spin-GFP tubule . ( C ) Spin-GFP and mCherry-Atg8a no longer co-localize in muscles expressing VCP-RNAi . White box indicates region shown at higher magnification and separate channels at right . D . Live image of GFP-mCherry-Atg8a in muscles treated with DMSO for 3 hr . Note the lack of GFP signal . ( E ) Live image of GFP-mCherry-Atg8a in muscles treated with the V-ATPase specific inhibitor Concanamycin A ( ConA ) for 3 hr . Note the presence of GFP-positive tubules . ( F ) Live image of GFP-mCherry-Atg8a in muscles treated with the VCP-specific inhibitor DBeQ for 3 hr . Note the presence of GFP-positive vesicles . ( G ) Spin-RFP localization in WT muscles or muscles expressing Atg7-RNAi using the muscle specific BG57-Gal4 driver . DOI: http://dx . doi . org/10 . 7554/eLife . 07366 . 01010 . 7554/eLife . 07366 . 011Video 5 . Spin-GFP and mCherry-Atg8a dynamics in Drosophila muscle . Representative time-lapse video of Spin-GFP and mCherry-Atg8 co-expressed in muscles . Frames were taken at 10 s intervals . DOI: http://dx . doi . org/10 . 7554/eLife . 07366 . 011 Next , we explored the consequence of disrupting VCP activity on Atg8 and Spin-GFP co-localization . When VCP was knocked down and the tubular-lysosomal network was eliminated , Atg8 no longer co-localized with Spin-GFP ( Figure 4C ) . Rather , Atg8-positive vesicles were found closely apposed to Spin-positive , vesicular lysosomal compartments . This close apposition suggests that the autophagosomes can identify and potentially dock against the lysosomal membranes , but fusion of the autophagosome with the lysosome is defective . This is consistent with an established role for the yeast VCP homolog , Cdc48 , in membrane fusion ( Latterich et al . , 1995 ) . To test whether tubules that are positive for both mCherry-Atg8a and Spin-GFP are , indeed , a consequence of fused auto-lysosomes , we imaged a dual fluorescent reporter GFP-mCherry-Atg8a ( Nezis et al . , 2010 ) . In neutral pH conditions , both GFP and mCherry fluoresce . However , in acidic environments such as in the lysosomal lumen , GFP fluorescence is quenched and only mCherry is observed . Thus , autophagosomes that are fused with lysosomes will exhibit mCherry but not GFP fluorescence . It is important to note that Spin-GFP fluorescence is not quenched because the C-terminal GFP tag resides on the cytoplasmic face of lysosomes ( Dermaut et al . , 2005 ) . When we expressed the dual fluorescence reporter in muscles we found no detectable GFP fluorescence in the tubules that label strongly for mCherry-Atg8a ( Figure 4D ) , consistent with other data indicating that the tubules are acidic . To further test the acidic nature of the tubules , we treated muscles with concanamycin A , a specific inhibitor of the lysosomal V-ATPase that is required for lysosome acidification ( Huss et al . , 2002 ) . Upon treatment with ConA for 3 hr , we observed GFP and mCherry positive tubules ( Figure 4E ) . Together , these data further verify that the lysosome tubules are acidic and also demonstrate that acidification is not necessary to maintain the structure of the tubules . Again , we explored the consequence of disrupting VCP activity . Inhibiting VCP function with DBeQ created GFP-positive vesicles ( Figure 4F ) that must be non-acidified compartments , a finding that is consistent with the existence of autophagosomes that have not yet fused with Spin-positive lysosomes . Since the appearance of GFP-positive vesicles occurs in a time frame of minutes to hours , it suggests that there is a continual flux of material through the auto-lysosomal system in muscle cells at steady state . Finally , we examined whether autophagy influx was required to induce or maintain lysosome tubules . Expression of Atg7-RNAi , an RNAi line that has been shown previously to inhibit autophagosome assembly ( Ren et al . , 2009 ) , did not affect lysosome tubules ( Figure 4G ) . Thus , autophagy induction is not a prerequisite for lysosome tubules . Taken together , these data confirm previous reports that VCP is required for autophagosome-lysosome fusion ( Tresse et al . , 2010b ) and further suggest that VCP loss abrogates autophagosome-lysosome fusion by affecting the structural properties of lysosomes . Previous studies have reported that , when over-expressed in cultured cells , mammalian VCP localizes diffusely at the nucleus and throughout the cytoplasm ( Vesa et al . , 2009; Tresse et al . , 2010a; Wang et al . , 2011 ) . To examine VCP localization in live muscles , we generated a UAS-VCP-Venus transgenic fly . Similar to previous reports for mammalian VCP , we observed abundant VCP localization in and around the nucleus and diffusely in the cytoplasm ( Figure 5—figure supplement 1 ) . But , VCP-Venus also concentrated at structures that are labeled by either Spin-RFP ( Figure 5A ) or mCherry-Atg8a ( Figure 5B ) , demonstrating that VCP localizes to auto-lysosomes . 10 . 7554/eLife . 07366 . 012Figure 5 . VCP co-localizes with the tubular auto-lysosomes . ( A ) Representative live image of VCP-Venus and Spin-RFP expressed in muscles using the muscle-specific BG57-Gal4 driver . White box indicates region shown at higher magnification and separate channels at right . ( B ) Representative live image of VCP-Venus and mCherry-Atg8a expressed in muscles using the muscle-specific BG57-Gal4 driver . Inset as in A . ( C , D ) VCP-Venus and mCherry-Atg8a localization in muscles treated with the VCP inhibitor DBeQ ( C ) or the proteasome inhibitor MG132 ( D ) for 3 hr . ( E ) Western blot analysis of total VCP protein levels from muscles in the treatments indicated . Tubulin serves as a loading control . ( F ) Representative time-lapse sequence of VCP-Venus and mCherry-Atg8a after MG132 was washed out . The arrow indicates a tubule extending from a mCherry-Atg8a positive vesicle . DOI: http://dx . doi . org/10 . 7554/eLife . 07366 . 01210 . 7554/eLife . 07366 . 013Figure 5—figure supplement 1 . VCP-Venus localization in Drosophila muscle . DOI: http://dx . doi . org/10 . 7554/eLife . 07366 . 013 Because inhibiting the catalytic function of VCP leads to tubule deterioration , we tested whether inhibiting the catalytic function of VCP would affect its localization to auto-lysosomes . Surprisingly , inhibiting VCP activity with DBeQ for 2 hr triggered the formation of rod-shaped VCP aggregates that effectively sequester VCP-Venus from the sarcoplasm ( Figure 5C ) . The formation of these aggregates is striking , as they resemble rod-like structures characteristic of prion aggregates . We considered two possible reasons that application of DBeQ might increase the propensity for VCP to aggregate . First , DBeQ binding to VCP might initiate aggregation directly by altering the solubility of VCP . Alternatively , the aggregation could be caused by the functions of VCP in other contexts , such as proteasome-dependent protein degradation . Remarkably , when we applied the proteasome inhibitor MG132 , VCP-Venus rapidly aggregated in a manner identical to that observed following DBeQ incubation ( Figure 5D ) . VCP aggregation was not due to increased VCP protein levels as a result of proteasome inhibition , because total VCP protein did not increase significantly upon MG132 or DBeQ treatment ( Figure 5E ) . While the significance of VCP-Venus aggregates remains uncertain , this phenotype provided us with a means to rapidly and reversibly sequester VCP protein and control its access to auto-lysosomal membranes . Since VCP exists as a hexamer , MG132 incubation in muscles over-expressing VCP-Venus should sequester both wild type and Venus-tagged VCP . Application of MG132 induced VCP-Venus aggregate formation , which correlated with dissolution of the tubular lysosomal network ( Figure 5D ) . When MG132 was washed out , VCP aggregates dissolved within 20 min and , as they disappeared , cytoplasmic VCP fluorescence intensity increased ( Figure 5F and Video 6 ) . During this time , cytoplasmic VCP-Venus accumulated at autophagosomes/lysosomes ( mCherry-Atg8a ) and tubules began to reform ( Figure 5F and Video 6 ) . These data indicate that VCP localizes to auto-lysosomes , where it could participate in auto-lysosomes tubulation . 10 . 7554/eLife . 07366 . 014Video 6 . VCP-Venus and mCherry-Atg8a dynamics after MG132 wash out . Muscles co-expressing VCP-Venus and mCherry-Atg8 were treated with the proteasome inhibitor MG132 for 3 hr . MG132 was washed out and time-lapse images were taken every 10 s . DOI: http://dx . doi . org/10 . 7554/eLife . 07366 . 014 We next investigated the consequence of disrupting the auto-lysosome tubule network . We first examined overall muscle function . When VCP RNAi was expressed specifically in muscle , muscle wasting was apparent and third instar larvae exhibited a severe impairment in their ability to move ( Figure 6—figure supplement 1A , B ) . When prodded , the animals would move , but their movements were slow and only lasted for short periods of time . The defect in their motility is not due to defects in the nervous system because synaptic transmission at the NMJ remained intact ( Figure 6—figure supplement 1C , D ) . These data are consistent with compromised muscle function that parallels the muscle weakness observed in human VCP-related diseases . Next , we investigated the degradation capacity of the collapsed tubular lysosomes . For lysosomes to degrade their cargo they must be acidified and the proteolytic enzymes must be present . We first examined the acidification of the lysosomes by co-imaging Spin-GFP with Lysotracker in VCP-RNAi animals and found that the enlarged Spin-GFP vesicles also co-stained with Lysotracker ( Figure 6—figure supplement 2A ) , indicating that they are acidic . Then , we examined whether lysosome enzymes were delivered properly to the lysosomes . Normally , the lysosomal enzyme Cathepsin-L is proteolytically processed in the lysosomal lumen to form a mature enzyme . To further examine the functionality of the lysosomes , we examined processing of Cathepsin-L and found no significant difference in Cathepsin-L processing in VCP-RNAi animals or animals treated with DBeQ ( Figure 6—figure supplement 1B ) . Thus , disruption of the tubular lysosomal network does not appear to affect the proteolytic capacity of sarcoplasmic lysosomes . To this point , our data suggest that loss of VCP disrupts the fusion of autophagosomes with functional lysosomes and , in parallel , causes the collapse of the tubular lysosomal network . Based on this , we expected to find evidence of failed autophagy in VCP knockdown muscle . In wild type muscles stained with a poly-ubiquitin antibody , we observed small puncta around the nucleus and a few small puncta in the cytoplasm ( Figure 6A ) . These small puncta likely represent active sites of protein degradation by the proteasome . However , in muscles expressing VCP-RNAi we observed a dramatic accumulation of cytoplasmic poly-ubiquitin aggregates ( Figure 6B ) . Even acute treatment with DBeQ was sufficient to produce cytoplasmic poly-ubiquitin aggregates ( Figure 6C–E ) . We also note that VCP inhibition caused a dramatic decrease in poly-ubiquitin conjugates around the nucleus , which likely reflects failed delivery of poly-ubiquitinated proteins to the proteasome . Importantly , we find that Spin-positive lysosomes are devoid of poly-ubiquitin staining ( Figure 6F ) , an observation that is consistent with our model of failed fusion of autophagosomes with lysosomes . 10 . 7554/eLife . 07366 . 015Figure 6 . Disruption of the tubular auto-lysosomal network correlates with increased poly-Ubiquitin aggregates , impaired mitochondria and increased lipofuscin granules . ( A , B ) Wild type ( A ) and VCP-RNAi ( B ) expressing muscles were fixed and stained with a poly-Ubiquitin antibody . Nuclei with localized poly-Ubiquitin staining are apparent in A . Nuclei are indicated ( dashed circle ) in B . ( C , D ) Wild type animals were treated with DMSO ( C ) or the VCP-specific inhibitor DBeQ ( D ) , fixed and stained with a poly-Ubiquitin antibody . ( E ) Quantitation of the number of poly-Ubiquitin aggregates per 50 µm2 from wild type muscles treated with DMSO for 4 hr or DBeQ for various times ( n = 9 , *p < 0 . 05 , **p < 0 . 01 ) . ( F ) Localization of Spin-GFP and poly-Ubiquitin in muscles expressing VCP-RNAi . ( G , H ) Mitotracker-C2TMRos staining in control ( G ) and VCP-RNAi ( H ) muscles . ( I–K ) Autofluoresence at 488 nm and lysotracker staining in wild type ( I ) , muscles expressing VCP-RNAi ( J ) , and wild type muscles treated with the VCP-specific inhibitor DBeQ for 4 hr ( K ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07366 . 01510 . 7554/eLife . 07366 . 016Figure 6—figure supplement 1 . Loss of tubular lysosomes correlates with impaired muscle function . ( A ) Wild type and VCP-RNAi gross muscle morphology . ( B ) BG57-Gal4 control animals and animals expressing VCP-RNAi in muscles were assessed for their crawling ability on a petri dish . ( C ) The total distance traveled in 1 min was measured for each animal and averaged ( n = 7 , **p < 0 . 01 ) . ( C ) Representative traces for wildtype muscles and muscles expressing VCP-RNAi . ( D ) Quantitation of EPSPs , mEPSPs and resistance input ( n = 10 , *p < 0 . 05 , **p < 0 . 01 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07366 . 01610 . 7554/eLife . 07366 . 017Figure 6—figure supplement 2 . Lysosomal acidity and Cathepsin processing are maintained in VCP-RNAi expressing muscles . ( A ) Spin-GFP co-imaging with lysotracker in muscles expressing VCP-RNAi . ( B ) Western blot analysis of Cathepsin L processing in the genotypes indicated . Tubulin serves as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 07366 . 017 In addition to clearing protein aggregates from the cytoplasm , autophagy is also responsible for clearing damaged organelles , including mitochondria . We examined the functional pool of muscle mitochondria in vivo using mitotracker-Orange CM-H2TMRos , which selectively stains mitochondria with an active membrane potential . Muscles expressing VCP-RNAi displayed less mitotracker-Orange CM-H2TMRos staining and the visualized mitochondria had an altered morphology , appearing round and dispersed rather than being densely packed , tubular structures ( Figure 6G , H ) . The swollen mitochondria observed in the VCP RNAi expressing muscles are likely defective and should be a prime target for mitophagy-dependent degradation . Taken together with the appearance of polyubiquitin aggregates , these data are consistent with an overall defect in autophagic clearance of proteins and defective organelles . However , since VCP has also been shown to be recruited directly to damaged mitochondria ( Kim et al . , 2013 ) we cannot rule out the possibility that the effects on mitochondrial morphology and membrane potential are due to direct VCP functions at the mitochondrial outer membrane . Finally , we noted an accumulation of lipofuscin granules in VCP-RNAi expressing muscles . Lipofuscin granules are a conglomerate of polymerized non-degradable proteins and lipids that build up in the lysosomal lumen ( Szweda et al . , 2003 ) . A distinguishing property of lipofuscin granules is that they exhibit auto-fluorescence at 488 nm . Wild type or VCP-RNAi muscles were stained with LysoTracker-Red and imaged at both 488 nm ( green ) and 555 nm ( magenta ) . The wild type muscles did not display any detectable auto-fluorescence at 488 nm ( Figure 6I ) . But , we observed strong auto-fluorescent puncta in VCP-RNAi that co-localized with LysoTracker ( Figure 6J ) , indicative of lipofuscin granules . We did not observe autofluorescent puncta to the same extent when wild type muscles were treated with DBeQ for 4 hr ( Figure 6K ) , suggesting that lipofuscin accumulation is a progressive phenotype . Alternatively , lipofuscin granule accumulation could require loss of VCP protein rather than just loss of VCP catalytic activity . These data suggest that maintaining the structural integrity of lysosome tubules is critical for lysosome function . Finally , we asked whether overexpression of VCP transgenes that harbor disease-causing mutations impairs the presence or dynamics of tubular lysosomes . To date , a total of 19 missense mutations in 13 different residues are associated with IBMFD that reside in either the Cdc48 homology domain , the L1 linker domain or the D1 ATPase domain ( Figure 7A and [Ju and Weihl , 2010] ) . We selected one mutation from each of these three domains to examine the effect on lysosome tubules: R155H , R191Q and A232E . R155 is the most common hereditary mutation and is located in the CDC48 homology domain , which is a protein interaction module that plays an important role in VCP substrate binding ( Ju and Weihl , 2010 ) . R191 is located in the linker region between the CDC48 domain and the first ATPase catalytic domain ( D1 ) . A232 is located at the beginning the of the D1 domain and is the most severe clinical mutation ( Watts et al . , 2004 ) . All three residues are conserved in the Drosophila protein ( Figure 7A ) . As a control , we show that over-expression of wild type dVCP does not alter tubular lysosomes ( Figure 7B ) . We then demonstrate that over-expression of each of the mutant VCP transgenes profoundly impairs the tubular lysosomal network ( Figure 7C–F ) . In parallel , we find an increase in sarcoplasmic auto-fluorescence when these VCP mutants are over-expressed ( Figure 7B–G ) . Overexpression of the A229E mutation , which is the most severe clinical mutation , caused the largest increase in auto-fluorescence compared to the other clinical mutations ( Figure 7F ) . Because over-expression of the disease relevant VCP transgenes phenocopies VCP-RNAi expression , these data suggest that the disease mutations are dominant interfering mutations . Thus , overexpression of VCP disease-associated mutations disrupts lysosome tubules in vivo , an effect that causes accumulation of cytoplasmic lipofuscin granules . 10 . 7554/eLife . 07366 . 018Figure 7 . Pathogenic VCP alleles disrupt the tubular auto-lysosomal network . ( A ) Schematic diagram of VCP protein . Top: Human pathogenic VCP mutations are labeled on the cartoon . Bottom: sequence alignment of human VCP and Drosophila VCP/Ter94 pathogenic mutant regions . ( B–F ) Autofluoresence at 488 nm and lysotracker staining in wild type muscles expressing VCP-WT ( B ) , VCP-RNAi ( C ) VCP-R152H ( D ) , VCP-R188Q ( E ) or VCPA-229E ( F ) transgenes . ( G ) Quantitation of the number of auto-fluorescent puncta per 50 µm2 in the genotypes indicated ( n = 9 , *p < 0 . 05 , **p < 0 . 01 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07366 . 018 In most tissues , autophagy is induced upon nutrient starvation , but muscles are one of the unique tissues in which autophagy occurs in the absence of starvation ( Mizushima et al . , 2004 ) . In fact , basal autophagy is required to maintain muscle mass ( Masiero et al . , 2009 ) . The constitutive levels of autophagy in muscle are likely related to the large energy requirement of muscles compared to other cell types and their ability to serve as a source of metabolic energy for other organs ( Sandri , 2010 ) . Muscle sarcoplasms are also unique in that they are shared between multiple nuclei and are much larger in volume compared to many other cell types . We propose that the observed extended , lysosomal network that is maintained by VCP is a cellular solution to ensure highly efficient autophagy throughout the entirety of muscle sarcoplasms . It is as if the lysosomes vascularize the sarcoplasm , ensuring that no portion of the sarcoplasm is ever far from a lysosomal depot where autophagic cargo can be degraded . Likewise , this network could facilitate the local recycling of nutrients throughout the sarcoplasm to effectively meet local demands . A similar argument can be made regarding the autophagy-dependent turnover of mitochondria , which are densely distributed throughout the muscle . In support of this , VCP is critical for the rapid degradation of muscle proteins during muscle atrophy and expression of a dominant negative form of VCP reduces protein degradation by both proteasomes and lysosomes ( Piccirillo and Goldberg , 2012 ) . Lysosomal membranes have been observed to extend tubules and fission off to produce de novo lysosomes following fusion of autophagosomes with lysosomes ( Yu et al . , 2010 ) . This process , termed ALR , serves as a mechanism to recycle lysosomes when autophagic demand is high ( Yu et al . , 2010 ) . It is unknown whether ALR is a constitutive process that participates in ongoing , steady-state proteostasis in vivo or whether it is a process that is specifically induced following stress-induced autophagy . It is possible that the extended , dynamic lysosomal tubule network that we observe is related to ALR , perhaps acting as a platform for efficient ALR throughout the sarcoplasm . In support of this , the mammalian Spinster homolog is required for ALR ( Rong et al . , 2011 ) . However , the relatively low frequency with which we observed scission events and the formation of new small lysosome vesicles is not entirely consistent with this idea . The tubular network identified in this study more closely resembles the lysosome tubular networks that have been observed in cultured macrophage cells ( Swanson et al . , 1987a , 1987b; Knapp and Swanson , 1990 ) . Macrophage lysosome tubules can form a web of tubules that appear to be stably connected throughout the cytoplasm ( Swanson et al . , 1987b ) . The purpose of lysosome tubules in this system is still mysterious . We have now identified specific defects associated with loss of lysosome tubules and our data suggest that the tubular network may distribute auto-lysosomal activity throughout the cell . Clues to the function of tubular lysosomes might also be gleaned from studies of early endosomes . Early endosomes form vesicular-tubular structures similar to what we have observed here ( Huotari and Helenius , 2011 ) . In early endosomes it has been established that tubules can be discrete membrane compartments with a different lipid composition and cargo compared to that within the vesicular body of the early endosome ( Huotari and Helenius , 2011 ) . By analogy , the ability of lysosomes to exist in a tubular-vesicular state in cellular contexts where their functional demand is high , might allow lysosomes to execute diverse functions more efficiently through compartmentalization . VCP and the yeast homologue Cdc48 have been ascribed many functions within the cell including cell cycle progression ( Moir et al . , 1982 ) , UPS and ERAD protein degradation ( Meyer et al . , 2012; Wolf and Stolz , 2012 ) , mitophagy ( Taylor and Rutter , 2011 ) and classical autophagy ( Ju et al . , 2009; Tresse et al . , 2010a; Dargemont and Ossareh-Nazari , 2012; Meyer et al . , 2012 ) . VCP achieves these diverse activities , in part , through its generalized function as an ubiquitin-dependent ‘segregase’ that dissociates protein conjugates tagged with ubiquitin from protein complexes and organelle membranes ( Halawani and Latterich , 2006; Meyer et al . , 2012 ) . For each process that relies upon VCP activity , different cofactors control VCP localization and function ( Baek et al . , 2013 ) . To this list of VCP-mediated activities , we now add the action of VCP in controlling the integrity and dynamics of a tubular lysosomal system and fusion of autophagosomes with tubular lysosomes . Additional , future experimentation will be necessary to determine which specific VCP-mediated molecular mechanism ( s ) is most directly relevant to the integrity and dynamics of lysosomal tubules in muscle . The phenotypic consequences following the loss or inhibition of VCP in Drosophila muscle include the collapse of the tubular lysosomal network , failed fusion of autophagosomes with lysosomes , accumulation of sarcoplasmic poly-ubiquitin aggregates , accumulation of lipofuscin granules , impaired mitochondria and impaired muscle health . All of these phenotypes are hallmarks of degenerative diseases that are associated with mutations in VCP ( Watts et al . , 2004 ) . We can now ascribe some of these phenotypes to the action of VCP at lysosomal membranes . Our data indicate that VCP localizes to autolysosomes and loss of VCP causes collapse of the tubular lysosomal network and failed autophagosome-lysosome fusion . Furthermore , when previously sequestered VCP is released back into the cytoplasm , VCP translocates to dormant autolysosomes and tubulation ensues . Although we cannot distinguish the role of VCP in the initiation and/or maintenance of lysosome tubules , these data argue that VCP acts at the autolysosmal membrane to control autolysosomal dynamics and the progression of autophagic protein clearance . If VCP is necessary for normal activity of the autophagy-lysosome system in muscle as our data suggests , then it seems likely that the accumulation of poly-ubiquitin aggregates and lipofuscin granules are a direct consequence of impaired VCP-dependent lysosomal function . This assertion is further supported by our demonstration that over-expression of pathogenic VCP mutant transgenes disrupt the tubular lysosomal network and also cause accumulation of ubiquitin and lipofuscin material . Importantly , these transgenes in Drosophila do not appear to disrupt VCP roles in the UPS or ERAD pathways , emphasizing the role of impaired autophagy in pathogenic VCP phenotypes ( Tresse et al . , 2010b; Chang et al . , 2011 ) . We acknowledge that lipofuscin granule accumulation following overexpression of pathogenic VCP mutant transgenes could represent a byproduct of proteotoxic stress caused by defects in the UPS or autophagy/lysosome pathways that eventually lead to the transformation of proteins into non-degradable products . Ultimately , more detailed biochemical analyses will be required to elucidate precisely how VCP functions at lysosome membranes and how this activity might be coordinated with other aspects of VCP function throughout the cell . Taken together , our identification of tubular lysosomes in Drosophila muscle challenges the traditional view of vesicular lysosomes and suggests that lysosome structures can be more versatile than previously assumed . Understanding how lysosomes regulate their morphological state will be an exciting avenue for future studies . Third instar larvae were dissected in Schneider's insect cell media ( Gibco , Grand Island , NY ) supplemented with 10% FBS ( Gibco ) and Penicillin/Streptomycin ( Gibco ) and all live imaging was performed in insect cell media . For all imaging experiment , at least 3 muscles in 3 animals were imaged to account for variances between muscles and animals and the most representative images are shown in each figure ( n ≥ 9 ) . The following drugs were diluted in insect cell media to the following final concentrations: 1uM LysoTracker Red DND-99 ( Life Technologies , South San Francisco , CA ) , 1uM ER Tracker green ( Life Technologies ) , 10uM Nocodazole ( Sigma , St . Louis , MO ) , 10uM Latrunculin A ( Life Technologies ) , 10uM DBeQ ( Sigma ) , 1uM MG132 ( Sigma ) , 100 nM Concanamycin A ( Santa Cruz Biotechnology , Dallas , TX ) , 2ug/ml TM ( Sigma ) and 500 nM Mitotracker-Orange-CM-H2TMRos ( Life Technologies ) . LysoTracker , ER Tracker and Mitotracker-Orange-CM-H2TMRos were incubated on the dissected larvae for 1 hr prior to imaging . Nocodazole , LatA , DBeQ , TM , Concanamycin A and MG132 were incubated on the dissected larvae for 3–4 hr prior to imaging . For Ubiquitin staining , larvae were dissected and fixed with 4% PFA for 15 min . After fixation , larvae were washed 4× with PBS-T and incubated with anti-poly-Ubiquitin ( Thermo Scientific , Waltham , MA ) at 1:1000 dilution overnight at 4°C . Larvae were washed again 4× with PBS-T , incubated with a fluorescently labeled secondary antibody at 1:5000 dilution ( Life Technologies ) , and washed again 4× with PBS-T before mounting on a slide with vectashield for imaging . Imaging was performed on an inverted Axiovert 200 microscope ( Zeiss ) using a 100× Plan Apochromat objective ( 1 . 4NA ) . Images were captured with a CoolSnap HQ2 CCD camera ( Photometrics ) and de-convolved using Slidebook 5 . 0 software ( Intelligent imaging innovations , Denver , CO ) . Image quantification was performed with imageJ software ( NIH ) . Volume rendering was performed with Slidebook 5 . 0 software . Any adjustment of brightness or contrast was performed using Slidebook 5 . 0 software , and always applied to the entire image . Third instar larvae were dissected and muscle preparations were immediately transferred into 5× SDS sample buffer and denatured by boiling for 10 min . Proteins were resolved by SDS-PAGE on a 4–12% Bis-Tris gel ( Life Technologies ) , transferred to a nitrocellulose membrane , immunoblotted with primary and HRP-conjugated secondary antibodies ( Life Technologies ) and detected using an ECL chemi-luminescent reagent ( Life Technologies ) . The following primary antibodies were used: anti-VCP ( Cell Signaling , Danvers , MA ) at 1:1000 , anti- GRP78/HSPA5 ( Thermo Scientific ) at 1:2500 , insect anti-Cathepsin L ( R&D Systems , Minneapolis , MN ) at 1:1000 , and anti-Tubulin E7-c ( Developmental Studies Hybridoma Bank , University of Iowa , IA ) at 1:10 , 000 . Secondary HRP conjugated antibodies ( GE Lifesciences , Pittsburgh , PA ) were used at 1:5000 . Sharp-electrode recordings were made from muscle 6 in abdominal segments 2 and 3 from third-instar larvae using an Axoclamp 900A amplifier ( Molecular Devices ) , as described previously ( Frank et al . , 2006 ) . Recordings were made in HL3 saline containing the following components: NaCl ( 70 mM ) , KCl ( 5 mM ) , MgCl2 ( 10 mM ) , NaHCO3 ( 10 mM ) , sucrose ( 115 mM ) , trehalose ( 5 mM ) , HEPES ( 5 mM ) , and CaCl2 ( 0 . 3 mM ) . Mean EPSP , mEPSP amplitude , and Rin were obtained by averaging values across all NMJs for a given genotype . EPSP traces were analyzed with custom-written routines in MATLAB ( Mathworks , Natick , MA ) as previously described ( Gaviño et al . , 2015 ) . mEPSP traces were analyzed in IGOR Pro 6 . 3 ( Wave-Metrics; custom script submitted with this manuscript ) .

Mutations in a gene that produces a protein called Valosin-containing protein ( VCP for short ) causes degenerative diseases that affect the brain , muscle and bone . In nearly half of the individuals with these VCP-related diseases—which can also result in dementia , Paget's disease of the bone and amyotrophic lateral sclerosis ( ALS ) —the first symptom is muscle weakness . Currently , very little is known about how VCP affects muscles . Patients with VCP-related diseases often have problems clearing damaged proteins from their cells , and recent research suggests that VCP is important for forming a cellular structure known as a lysosome . Lysosomes contain powerful enzymes that destroy damaged proteins and other cellular structures that would otherwise accumulate in the cells . In most cells , lysosomes look like bubble-like compartments called vesicles . However , in some types of cells lysosomes have been observed to form a network of tubules that extend throughout the cell interior . However , it remains unclear what these tubules do , how they form in cells and whether they are altered in disease . Johnson et al . analyzed lysosomes in the muscle of the fruit fly species Drosophila melanogaster and discovered that lysosomes were in the form of a network of tubules that spread throughout each muscle cell . These tubules constantly changed in living muscles; extending , retracting , breaking and merging to form a large tubular lysosome network . When Johnson et al . reduced the amount of VCP produced by the muscle cells , via a method called RNA interference , the lysosome tubules broke down into vesicles that were no longer constantly changing . Modifying these defective fly muscle cells so that they produced the human VCP protein caused the tubules to form again . These results suggest that the human and fly VCP proteins are very similar and that they play a key role in either the ability of lysosomes to form tubules or the maintenance of existing tubules . Johnson et al . then engineered flies to produce a version of the VCP protein that had mutations commonly seen in individuals with degenerative diseases . Lysosome tubules did not form correctly in the muscle cells of these flies . These flies also had other abnormalities; for example , their cells showed a great build-up of damaged proteins , and their ability to move their muscles was weaker . These findings suggest that a network of lysosomal tubules is necessary for healthy muscle cells , but how and why these tubular networks are formed or maintained is still mysterious . What causes lysosomal membranes to form tubules ? How do they break and fuse ? And why are they necessary ? Genetic experiments in fruit flies will be a great place to discover these mechanisms and understand the links to degenerative diseases in humans .

2015

VCP-dependent muscle degeneration is linked to defects in a dynamic tubular lysosomal network in vivo

Bicoid ( Bcd ) protein distributes in a concentration gradient that organizes the anterior/posterior axis of the Drosophila embryo . It has been understood that bcd RNA is sequestered at the anterior pole during oogenesis , is not translated until fertilization , and produces a protein gradient that functions in the syncytial blastoderm after 9–10 nuclear divisions . However , technical issues limited the sensitivity of analysis of pre-syncytial blastoderm embryos and precluded studies of oocytes after stage 13 . We developed methods to analyze stage 14 oocytes and pre-syncytial blastoderm embryos , and found that stage 14 oocytes make Bcd protein , that bcd RNA and Bcd protein distribute in matching concentration gradients in the interior of nuclear cycle 2–6 embryos , and that Bcd regulation of target gene expression is apparent at nuclear cycle 7 , two cycles prior to syncytial blastoderm . We discuss the implications for the generation and function of the Bcd gradient . The discovery of the concentration gradient of Bicoid ( Bcd ) protein in the early Drosophila embryo established the existence and functional importance of a morphogen gradient for the first time ( Driever and Nusslein-Volhard , 1988a , 1988b; Frigerio et al . , 1986 ) ; it was a watershed moment in developmental biology . These and subsequent studies showed that Bcd protein is present at the cortex of pre-cellular , syncytial blastoderm embryos , with levels that are highest at the anterior end and that decline exponentially toward the posterior ( Driever and Nusslein-Volhard , 1988b; Gregor et al . , 2007; Spirov et al . , 2009 ) . Although bcd mRNA is concentrated in the anterior cytoplasm of stage 13 oocytes and of embryos immediately after egg laying ( Berleth et al . , 1988; Frigerio et al . , 1986; Riechmann and Ephrussi , 2004 ) , its distribution extends more posteriorly in the embryo at syncytial blastoderm stages ( Berleth et al . , 1988; Frigerio et al . , 1986; Spirov et al . , 2009 ) . Whether the protein gradient forms by passive diffusion following synthesis of Bcd protein at more anterior locations ( Gregor et al . , 2007; Little et al . , 2011 ) , or is produced in place by the bcd mRNA concentration gradient is in dispute ( Fahmy et al . , 2014; Spirov et al . , 2009 ) . After fertilization , nuclei divide rapidly and synchronously eight times in the interior of the embryo , moving outward in a choreographed sequence that places them simultaneously at the surface at nuclear cycle 9 ( nc9 ) . The five division cycles that follow delineate the syncytial blastoderm stages nc10-nc14 . Nuclear divisions cease at nc14 , whereupon the nuclei begin to individuate into single cells and gastrulation ensues . Various measures , including in situ hybridization ( Erickson and Cline , 1993; Pritchard and Schubiger , 1996 ) , RT-PCR ( Harrison et al . , 2010 ) , genome array hybridization ( De Renzis et al . , 2007; Little et al . , 2011; Lu et al . , 2009 ) , RNA seq ( Lott et al . , 2011 ) , DNA footprinting ( Harrison et al . , 2010 ) , chromatin profiling ( Harrison et al . , 2011 ) and ChIP-seq ( Blythe and Wieschaus , 2015 ) , show that the zygotic genome is transcriptionally activated during the syncytial blastoderm period . Oogenesis provides the Drosophila egg with a rich dowry of mRNA that is essential to the development of the early , pre-cellular embryo , and for a number of reasons , the period that precedes the maternal-to-zygotic transition has been considered to depend only on maternal stores and to be independent of the zygotic genome . One , the early nuclear divisions are so rapid ( 9 . 6 min ) that productive gene expression has been deemed impossible . Two , molecular analyses of transcriptional activity have almost universally failed to detect RNA synthesis at pre-syncytial blastoderm stages , even as the sensitivity of the detection methods has increased . Three , comprehensive genetic screens for mutants defective in early development identified many genes that are required maternally , but found no evidence for genes that must be active in the zygote prior to cellularization at nc14 ( Luschnig et al . , 2004; Merrill et al . , 1988; Perrimon et al . , 1984; Schupbach and Wieschaus , 1989 , 1991; 1986 ) . Although these observations have substantiated the idea that the gene products supplied by the mother during oogenesis are sufficient for first thirteen cleavage cycles , this conclusion is based on negative findings , and because it depends on the sensitivity of the analysis , it leaves open the possibility that more sensitive methods might detect zygotic transcripts expressed from a small number of active genes or might recognize phenotypes in mutant embryos that were not revealed by then available histological techniques . Drosophila embryos are heavily populated with yolk and glycogen granules that impede histological studies , and have few obvious morphological features that can be evaluated for dependence on genotype . In addition , the idea that rapidly dividing nuclei are incapable of expression has no experimental basis because the capacity for transcription and translation at early nuclear cycles has not been analyzed . It is possible therefore that the normal transcriptional processes are sufficient for transcription units that are small ( approximately 70% of transcripts made by nc10-12 embryos lack introns; De Renzis et al . , 2007 ) , or it may be that yet unexplored mechanisms produce and use transcripts more rapidly at early stages . There are , in fact , several reports of expression by the zygotic genome in pre-syncytial blastoderm Drosophila embryos . The earliest reported zygotic expression obtained by in situ hybridization is at nc8 for the gene sis-A ( Erickson and Cline , 1993 ) . Evidence for earlier gene expression ( β-galactosidase activity in nc4 embryos ) was reported for a transgene construct that carried the lacZ gene regulated by a FTZ-F2 enhancer fragment ( Brown et al . , 1991 ) . The strongest evidence for functional expression in pre-syncytial blastoderm embryos is for the engrailed ( en ) gene ( Ali-Murthy et al . , 2013; Karr et al . , 1985 ) . En protein synthesized in the embryo was detected by antibody staining in nc2 nuclei , a mutant phenotype was identified in en nc2-3 embryos , and PCR and RNA-seq studies provided evidence for expression of a small cohort of genes prior to nc7 ( Ali-Murthy et al . , 2013 ) . These findings show that development of the early embryo is not entirely pre-programmed and that the processes that orchestrate the early stages are actively and directly regulated . The work described here investigates the development of the early embryo further , focusing on the formation and function of the Bcd gradient . To do so , we needed to develop methods that overcome the technical impediments that have limited analysis of the early stages . For instance , because Drosophila fertilization cannot be made synchronous and because females hold their embryos for variable periods prior to laying , an embryo of any particular stage must be identified histologically and individually , and methods must be used that are sensitive enough to detect expression from a small number of active genes in the small number of nuclei . Although molecular detection methods have improved , antibody staining and in situ hybridization have been ineffective due to the large number of yolk and glycogen granules that reduce signal to noise and sensitivity . By using sensitive methods that we modified for studies of the early embryo , we found that the processes that generate the Bcd protein gradient are more complex and operate earlier than had been appreciated , and that the function of the Bcd gradient begins prior to formation of the syncytial blastoderm . Krüppel ( Kr ) is a 'gap'” gene that is expressed by pre-cellular , syncytial blastoderm embryos in a central band that spans approximately 20% embryo length . Although the earliest reported expression detected by in situ hybridization is nc10 ( Pritchard and Schubiger , 1996 ) , RNA-seq and QT-PCR that we carried out previously detected Kr transcripts prior to nc10 in pre-syncytial blastoderm embryos ( Ali-Murthy et al . , 2013 ) . To identify the nuclei that express Kr transcription prior to nc10 , we evaluated various different techniques for in situ hybridization , and determined that a procedure that uses DIG probes to be the most sensitive for our studies of Kr and bcd transcripts in pre-cellular embryos ( see Materials and Methods and Figure 1—figure supplement 1 ) . Figure 1 shows images of nc7-13 embryos in which Kr transcripts were detected . In situ hybridization can detect sites of nascent transcript production as points of staining or fluorescence ( Femino et al . , 1998; Shermoen and O'Farrell , 1991 ) , and the nc10-13 embryos we analyzed had bright dots in most or all nuclei in the 'Kr band' . The fraction of nuclei with dots was lower in nc7-9 embryos , but most of the nuclei with dots were in the central region where the Kr band forms . The position of the Kr band , but not the number of Kr dots , was sensitive to maternal bcd gene dosage: in nc12 embryos , it was at 43–64% embryo length ( from anterior end ) if mothers were wild type ( had two bcd copies ) and 54–75% if mothers had six . This is indicative of the dependence of Kr expression on maternal Bcd ( Figure 1—figure supplement 2 ) , and is consistent with earlier studies ( Driever and Nusslein-Volhard , 1988a; Hoch et al . , 1991 ) . 10 . 7554/eLife . 13222 . 003Figure 1 . In situ detection of Kr transcipts in pre-cellular embryos . In situ hybridization detected Kr RNA ( red ) in nuclear cycle 7–13 embryos that were produced by females with two ( left two columns ) and six ( right two columns ) bcd gene copies . Embryos are prophase stage , dorsal up and anterior left . Embryo images are projection composites from serial optical sections; high magnification images show nuclei with red fluorescent dots . The number of nuclei with red dots and the dot brightness increases with each successive cycle . Two dots are visible in most nuclei; some have only one but none have three . Yellow dots in nuclear cycle 7–9 embryos indicate dots that were too faint to be visible at low magnification . The width of the band of Kr-expressing nuclei was approximately the same in the two genotypes; its position shifted more posteriorly in the embryos with excess Bcd . DOI: http://dx . doi . org/10 . 7554/eLife . 13222 . 00310 . 7554/eLife . 13222 . 004Figure 1—figure supplement 1 . Methods of in situ hybridization compared . In situ hybridization was carried out with bcd and Kr probes to nuclear cycle 4 and 12 embryos as indicated . Method for DIG was as described in Materials and Methods; for FISH using multiple antisense DNA probes marked with fluorescent tags according to the Stellaris protocol ( Little et al . , 2013; 2011 ) , and QuantiGene ViewRNA Probes according to the Affymetrix protocol . Images were obtained under identical conditions except for laser intensities , which were 33% for DIG , 75% for FISH , and 55% for QuantiGene . DOI: http://dx . doi . org/10 . 7554/eLife . 13222 . 00410 . 7554/eLife . 13222 . 005Figure 1—figure supplement 2 . Dependence of gastrula morphology on maternal genotype . Gastrula stage embryos from females with the indicated genotypes were fixed and stained with DAPI and the relative distance from the anterior end to the cephalic furrow ( vertical red line ) was calculated from the images of embryos ( oriented anterior left , dorsal top ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13222 . 005 The images in panels in Figure 2 show nuclear dots in nc9 and nc11 embryos . The dots were absent from early interphase nuclei , but increased in number and brightness as the cycle progressed , and they reached peak brightness at late interphase-early prophase ( Figure 2A , B ) . The level of fluorescence in the cytoplasm also increased during the nuclear cycles . Both nuclear dot and cytoplasmic fluorescence was higher at nc11 relative to nc9 , and was also higher in embryos produced by mothers with six bcd genes . The panel with a high magnification view of a nc9 pro-metaphase nucleus shows that both dots in this nucleus appear to be at visible chromosome arms ( Figure 2C ) . Nuclei in embryos that have one Kr gene had only one dot ( Figure 2D ) . These observations suggest that the in situ hybridization method detected nascent transcripts that were chromosome-associated , as well as mature transcripts in the cytoplasm . The increase in dot brightness during the nuclear cycle suggests that the in situ hybridization signal is proportional to the quantity ( number and length ) of transcripts , and although the longer nuclear cycle at nc11 relative to nc9 may account for the increased signal at the later cycle , the signal increase observed in embryos with additional Bcd dosage was not expected . 10 . 7554/eLife . 13222 . 006Figure 2 . Characterization of Kr transcripts detected in situ . ( A ) In situ hybridization detected Kr transcripts in mid-interphase to late prophase nuclei as discrete red fluorescent dots in nuclear cycle 9 and 11 embryos . ( B ) Bar graphs depicting measures of red fluorescence nuclear dots in nc11 nuclei and in a boxed area of adjacent cytoplasm quantify the dependence on nuclear cycle stage and Bcd level in embryos from mothers with two and six bcd genes . n=10 for nuclear dots , n=4 for cytoplasm . Differences between 2x and 6x are significant for nuclear dots ( two-tailed value <0 . 0001 and t-test <E-08 ) and for early interphase cytoplasm ( two-tailed value <0 . 0001 and t-test 3E-07 ) . ( C ) High magnification image of a cycle 9 pro-metaphase nucleus revealing the apparent association of two red fluorescent dots with chromosome arms . ( D ) Nuclear cycle 11 nuclei in Kr/+ embryos have only one red fluorescent dot . DOI: http://dx . doi . org/10 . 7554/eLife . 13222 . 00610 . 7554/eLife . 13222 . 007Figure 2—source data 1 . Source data for 2B . Fluorescence intensity of dots in nuclei and cytoplasm of projections images were measured in ImageJ by placing a square of constant dimension centered over each dot of interest . Embryos from mothers with either two or six copies were analyzed at the indicated stages . DOI: http://dx . doi . org/10 . 7554/eLife . 13222 . 007 The number of nuclei with fluorescent dots at nc7 and nc8 was small , and although the position of these marked nuclei is in the central region of the embryo where the Kr band will be robust at later cycles , it is important to know whether the dots in these nuclei mark nuclear transcripts . Chromosomes in nuclei at the cortex of syncytial blastoderm embryos are organized and polarized such that the centromeric regions congregate apically and the telomeres congregate basally ( Marshall et al . , 1996 ) . Chromosome organization in nuclei at earlier , pre-syncytial blastoderm stages has not been examined previously . Figure 3 characterizes nuclear and background dots in nc7 and nc8 embryos with respect to proximity to the embryo cortex . Nuclei at nc7 and nc8 are approximately 34 μm and 22 μm , respectively , from the embryo surface ( Figure 3 C , D ) . Comparison of fluorescence levels of these dots in successive confocal optical sections revealed that whereas the dots in nuclei were in the basal sections , dots not in nuclei ( and attributed operationally to background ) were apical . Because the Kr gene is at the tip of 2R ( polytene band 60F5 ) , this result suggests that nuclei are polarized prior to their arrival at the cortex and it is consistent with the conclusion that the nuclear dots mark sites of Kr transcription . 10 . 7554/eLife . 13222 . 008Figure 3 . Intranuclear localization of Kr transcripts in pre-syncytial blastoderm nuclei . ( A , B ) In situ hybridization detected Kr transcripts ( red fluorescence ) in nuclear cycle 7 and 8 embryos; prophase , dorsal up , anterior left . Yellow circles indicate dots whose red fluorescence was measured in successive optical planes and that are depicted in the bar graphs . ( A ) Two fluorescent non-nuclear dots ( B1 , B2 ) that were near nuclei with red dots were analyzed in the nuclear cycle 7 embryo ( left graph ) ; four fluorescent dots ( N1-4 ) were analyzed ( right graph ) . ( B ) Seven red nuclear dots were analyzed in the nuclear cycle 8 embryo . ( C , D ) Nuclei imaged with anti-nuclear lamin ( C ) and DAPI ( D ) show the placement of nuclei at nuclear cycle 7 ( C ) and nuclear cycle 8 ( D ) and the measured distances ( μm ) between nuclei and the cortex . DOI: http://dx . doi . org/10 . 7554/eLife . 13222 . 00810 . 7554/eLife . 13222 . 009Figure 3—source data 1 . Source data for 3A , B . Fluorescence intensity of dots in nuclei and cytoplasm in optical sections , ordered apical to basal , were measured in ImageJ by placing a square of constant dimension centered of each dot of interest . DOI: http://dx . doi . org/10 . 7554/eLife . 13222 . 009 Previous studies described the Bcd concentration gradient at the syncytial blastoderm stages ( Driever and Nusslein-Volhard , 1988b; Fahmy et al . , 2014; Little et al . , 2011; Spirov et al . , 2009 ) , but the earlier , pre-syncytial blastoderm stages have not been characterized . We developed a method for antibody staining of early embryos that has high signal-to-noise sensitivity ( see Materials and methods ) , and used it to monitor Bcd in the pre-cellular stages . Figure 4 shows dorsal views of nc4-14 embryos that were either from normal mothers ( WT ) , or that were from mothers that were mutant for exuperentia ( exu ) or staufen ( stau ) . Function of the exu and stau genes during oogenesis is essential to localize bcd mRNA at the anterior pole and for the Bcd gradient in syncytial blastoderm embryos ( reviewed in Kugler and Lasko , 2009 ) . The Bcd protein gradient detected by this method in WT and mutant embryos at syncytial blastoderm stages was consistent with previous reports . It extended to approximately 60% embryo length in WT nc12 embryos . In the syncytial blastoderm stages of embryos from mutant mothers , the gradient extended more posteriorly and the apparent levels of Bcd protein were higher than WT , especially at late nc14 when Bcd protein was low in WT . Although the Bcd protein levels appeared to be higher in the mutant embryos , Q-PCR analysis showed that the levels of bcd RNA was similar in the WT and mutants ( Figure 4—figure supplement 1 ) . The late persistence of Bcd protein was particularly pronounced in embryos that lacked maternal stau . Kr expression in exu and stau embryos was robust but misplaced , and was broader than the normal in nc13 and nc14 stau embryos ( Figure 4—figure supplement 2 ) . 10 . 7554/eLife . 13222 . 010Figure 4 . Bcd protein distributions in pre-cellular embryos . Anti-Bcd antibody detected Bcd protein ( red fluorescence ) in nuclear cycle 4–14 WT , exu and stau embryos . Embryos are prophase , dorsal up and anterior left . Images are projection composites from serial optical sections ( 46–50 per embryo , spanning from the dorsal to ventral surface ) . Fluorescence is brightest at the anterior end and in the mutant embryos it is brighter and extends more posteriorly than in the WT . Fluorescence is apparent concentrated in nuclei of nuclear cycle 6–14 embryos . DOI: http://dx . doi . org/10 . 7554/eLife . 13222 . 01010 . 7554/eLife . 13222 . 011Figure 4—source data 1 . Source data for Figure 4—figure supplement 1 . PCR amplification cycle number for bcd RNA isolated from embryos that had developed to the indicated nuclear cycle . DOI: http://dx . doi . org/10 . 7554/eLife . 13222 . 01110 . 7554/eLife . 13222 . 012Figure 4—figure supplement 1 . bcd RNA levels in pre-cellular WT and exu and stau mutant embryos . Q-PCR analysis of RNA isolated from stage-selected embryos does not detect differences in bcd RNA levels between WT , exu or stau mutant embryos . bcd RNA levels decline in nuclear cycle 14 embryos . DOI: http://dx . doi . org/10 . 7554/eLife . 13222 . 01210 . 7554/eLife . 13222 . 013Figure 4—figure supplement 2 . Kr expression in exu and stau mutant embryos . In situ hybridization detected Kr transcripts ( red fluorescence ) in exu and stau nuclear cycle 9 , 10 , 12 and 14 embryos . Embryos are late interphase/prophase , dorsal up , and anterior left . Images are projection composites from serial optical sections . Regions with Kr transcripts are expanded and displaced relative to WT . DOI: http://dx . doi . org/10 . 7554/eLife . 13222 . 013 A concentration gradient of Bcd protein was also observed in nc4 and nc6 embryos that were produced by WT females . The apparent level of Bcd was higher in the nc6 embryos and the gradient extended more posteriorly . In contrast , staining by the anti-Bcd antibody did not detect a gradient distribution of Bcd in nc4 and nc6 embryos from exu or stau mothers ( Figure 4 , upper panels ) . Figure 5 shows a higher resolution analysis of Bcd distribution in pre-syncytial blastoderm embryos . The panels show projection images and include sagittal sections calculated from the stacks of optical sections . In the nc2 embryo , Bcd levels were low medially at the anterior end , high in the surrounding region and extended approximately 20% embryo length posteriorly . In the nc3 embryo , both the 'doughnut' distribution at the anterior pole and the posteriorly extending medial plume were more pronounced . The medial plume was increased in the nc4 embryo , and more so in the nc6 embryo . Nuclei in these embryos were in the interior and enveloped by the medial plume and had multiple fluorescent puncta . The nuclei appeared to concentrate Bcd protein , with higher levels of fluorescence in the more anterior nuclei . Fluorescence was apparent only in the most anterior nuclei of nc4 embryos but could be detected in a concentration gradient that encompasses all of the nuclei in nc6 embryos . Analysis of composite projection images and optical sections showed the Bcd protein detected by the antibody was mostly in the interior regions and increased in amount and extended more posteriorly from nc2-nc6 . Although no medial plume was observed in nc4 and nc6 embryos that were from exu mothers , staining by the anti-Bcd antibody was apparent and fluorescent punctae were visible in nuclei of the nc6 mutant embryo . This indicates that Bcd protein was present in the interior of mutant pre-syncytial blastoderm embryos , although its distribution was apparently more uniform than WT . 10 . 7554/eLife . 13222 . 014Figure 5 . Bcd protein distributions in nuclear cycle 2 to 6 embryos . Anti-Bcd antibody detected Bcd protein ( red fluorescence ) in nuclear cycle 2 to 6 WT and exu embryos . Embryos are prophase , dorsal up and anterior left . Images are projection composites from serial optical sections . Yellow lines indicate the position of mid embryo sagittal and transverse optical sections that are shown below and to the right , respectively , of each embryo . The anterior to posterior fluorescence gradient is apparent in the sagittal sections of WT but not the exu embryos; internal embryo fluorescence is apparent in the transverse sections of both WT and mutant embryos . Nuclei in the WT nuclear cycle 4 ( indicated by yellow dots ) , nuclear cycle 6 and exu nuclear cycle 6 embryos concentrate red fluorescence; high magnification images of the indicated nuclei show DAPI fluorescence and resolve spots of red fluorescence in both WT and exu nuclei . DOI: http://dx . doi . org/10 . 7554/eLife . 13222 . 014 In situ hybridization detected bcd RNA concentrated at the anterior end of pre-cellular stages ( Figure 6 ) in patterns that were similar to Bcd protein ( Figures 4 , 5 ) . bcd RNA was concentrated at the anterior end of nc2 embryos , and in nc4 embryos formed a plume that extended posteriorly in the embryo interior . It was predominantly anterior and cortical in embryos at stages nc8 and older . These images of bcd RNA in syncytial blastoderm stage embryos are consistent with previous reports ( Little et al . , 2011; Spirov et al . , 2009 ) . Higher resolution examination of the anatomy of the RNA distribution at nc1 and nc3 stages ( Figure 7 ) revealed that the RNA was tightly restricted at the anterior end at nc1 , but at nc3 was distributed in an internal plume that had distinct contours and dorsal/ventral asymmetries . The posterior extent of the plume was greatest at nc4 , and was not as great at nc5 and nc6 . 10 . 7554/eLife . 13222 . 015Figure 6 . In situ detection of bcd transcripts in pre-cellular embryos . In situ hybridization detected bcd transcripts ( red fluorescence ) in nuclear cycle 2 , 4 , 6 , 8 and 10 embryos . Embryos are dorsal up and anterior left . Images are mid-embryo optical sections ( left ) and projection composites from serial optical sections ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13222 . 01510 . 7554/eLife . 13222 . 016Figure 7 . In situ detection of bcd transcripts in nuclear cycle 1 , 3 and 5 embryos . In situ hybridization detected bcd transcripts ( red fluorescence ) in nuclear cycle 1 , 3 , and 5 embryos . Embryos are dorsal up and anterior left . Left panels are projection composites from serial optical sections and panels to their right are successive dorsal to ventral optical sections . DAPI-stained image of nucleus shown for the nuclear cycle 1 embryo . DOI: http://dx . doi . org/10 . 7554/eLife . 13222 . 016 Published in situ hybridization and antibody studies report the detection of bcd RNA and protein in embryos , but only bcd RNA in oocytes ( Berleth et al . , 1988; Driever and Nusslein-Volhard , 1988b; Frigerio et al . , 1986; Schnorrer et al . , 2000 ) . Although Bcd protein was detected in unfertilized eggs , the assumption was made that stage 14 oocytes have anterior-localized RNA but no Bcd protein , and that the protein in unfertilized eggs was produced in conjunction with egg activation and egg laying . However , because there have not been suitable methods for fixation and preparation , previous studies of stage 14 oocytes have been limited to in situ hybridization analysis of ovarian sections ( Frigerio et al . , 1986 ) . In contrast to younger oocytes , stage 14 oocytes have vitelline membrane and chorion layers that reduce permeability , but they are sensitive to hypochlorite and organic solvent treatments that are resisted by the fully formed vitelline membrane and chorion layers of subsequent stages . We sought methods to examine bcd RNA and protein in stage 14 oocytes because our analysis revealed that significant levels of Bcd protein are present in early nc1 embryos and unfertilized eggs ( Figure 8 ) . The question we addressed was whether the protein had been produced prior to fertilization or in the short interval between fertilization and nc1 metaphase . 10 . 7554/eLife . 13222 . 017Figure 8 . bcd expression in stage 14 oocytes . α-Bcd antibody detects Bcd protein in nuclear cycle 1 embryos ( A ) and in unfertilized eggs ( B ) . Stage 14 oocytes ( stained with DAPI ) are covered with follicle cells initially and the dorsal appendages are juxtaposed ( stage 14a; ( C ) ) ; the nuclei of the follicle cells elongate and the follicle cells migrate anteriorly ( stage 14b; ( D ) ) , form an anterior ring stage 14c; ( E ) that extrudes distally from the dorsal appendages , which separate from each other stage 14d; ( F ) . High magnification images of DAPI-stained nuclei in ( C , D , F ) show the oocytes to be at prophase ( 14a ) and true metaphase ( 14b , 14d ) ( Gilliland et al . , 2009 ) . In situ hybridization detects bcd RNA at all stage 14 oocytes ( G , I , K ) ; antibody staining detects Bcd protein in stage 14d oocytes ( L ) but not in younger stages ( H , J ) . Orientation anterior left . DOI: http://dx . doi . org/10 . 7554/eLife . 13222 . 01710 . 7554/eLife . 13222 . 018Figure 8—figure supplement 1 . In situ detection of bcd transcripts in stage 12 oocyte . In situ hybridization detected bcd transcripts ( red ) in nurse cells and in the anterior cytoplasm of the stage 12 oocyte . DOI: http://dx . doi . org/10 . 7554/eLife . 13222 . 018 At the beginning of stage 14 ( stage 14a ) , follicle cells completely envelope the oocyte ( Figure 8 ) , the chorionic dorsal appendages are not yet separated , and no nurse cells remain . During stage 14 , the oocyte 'undresses': the follicle cells unfasten from the chorion and re-arrange into longitudinal rows and the nuclei elongate ( stage 14b ) , and the cells begin to migrate toward the anterior end and concentrate in an anterior 'collar' ( stage 14c ) . The follicle cells are shed from the tips of the fully formed and separated dorsal appendages ( stage 14d ) . At stage 14c-d , oocytes exit from the distal end of the ovary and enter the lateral branches of the oviduct . Staining with anti-Bcd antibody did not detect Bcd protein in stage 13 ( not shown ) or stage 14a–b oocytes ( Figure 8H , I ) , but detected Bcd protein in stage 14c ( not shown ) and 14d oocytes ( Figure 8L ) . Bcd protein was present in a tightly delineated band that follows the contours of the oblique concavity at the oocyte’s anterior end . This concavity does not fully resolve into a blunt nose cone until mid nc1 . In situ hybridization also detected bcd RNA in a band that follows the anterior concavity of stage 14 oocytes . This RNA distribution was similar to the Bcd protein except that in contrast to the protein , it was detected at stage 12 ( Figure 8—figure supplement 1 ) and 13 ( not shown ) and all periods of stage 14 . These results show that the translation block that inhibits Bcd protein production is relieved prior to fertilization and egg activation , and shows that Bcd protein is produced and is present precisely at the site where bcd RNA is sequestered . The cortical nuclei at the surface of late syncytial blastoderm embryos have apical-basal polarity that is manifested in the organization of their chromosomes . Whereas the centromeric regions of the chromosomes coalesce to form a chromocenter that is at the most apical aspect of the nuclei , the telomeres congregate basally ( Marshall et al . , 1996 ) . Our in situ hybridization experiments located nascent transcripts at the Kr gene , which is situated just proximal to the telomere of chromosome 2R , and analysis of serial optical sections revealed that the nascent transcripts are at the basal regions of nc7 and nc8 nuclei ( Figure 3 ) . These nuclei are approximately 34 μm and 22 μm , respectively , from the embryo surface , indicating that nuclei in the interior of the embryo are organized and polarized with respect to the cortex . The idea that internal chromosome organization is related to gene expression ( Marshall et al . , 1996 ) is consistent with our finding that these nuclei are transcriptionally active . Nuclear divisions in the pre-syncytial blastoderm embryo are precisely synchronized even as the nuclei move to collectively occupy larger volumes of the embryo at later nuclear cycles , and the choreography they follow as they move to the cortex is highly reproducible . Our finding that nuclei are oriented prior to their arrival at the cortex indicates that their apical-basal polarity is not a consequence of short-range interactions with the cortex , and suggests that a global system of structural organization and regulation exists in the early embryo . The Bcd concentration gradient is generated from mRNA that is made by nurse cells and is transported to the oocyte where it is sequestered in an inactive state at the anterior end ( Frigerio et al . , 1986 ) . Previous studies reported that Bcd protein was absent from stage 13 oocytes but present in embryos and unfertilized eggs ( Driever and Nusslein-Volhard , 1988b ) . Although stage 14 oocytes were not examined , the assumption that Bcd protein is also absent from stage 14 oocytes was the basis for the current model that translation of Bcd protein begins at egg activation . Our finding that Bcd protein is made in late stage oocytes prior to activation and fertilization ( Figure 8 ) revises this model , but does not fundamentally change it . The distribution of Bcd protein in stage 14c and 14d oocytes is consistent with the idea that Bcd protein is made by mRNA that was localized to the anterior end . Its production in the oocyte raises the question whether Bcd protein is made only in the oocyte and not in the embryo , but this seems unlikely because the level of staining by α-Bcd antibody increased during the pre-cellular cycles ( Figure 4 ) . Two models have been proposed for the formation of the Bcd protein gradient . One is based on the observation that both bcd mRNA and Bcd protein distribute in concentration gradients at syncytial blastoderm stages ( nc9-14 ) , and because of the spatial relationship between these distributions , it posits that the Bcd protein gradient is a direct consequence of the location of bcd mRNA ( Fahmy et al . , 2014; Spirov et al . , 2009 ) . This model does not require or invoke a role for protein diffusion , but proposes instead that a process of directed and regulated transport produces a concentration gradient of bcd mRNA that precedes and generates the Bcd protein gradient . The second model also involves a gradient of bcd RNA , but because the bcd RNA was not observed to extend as far posteriorly as Bcd protein , it proposes a major role for Bcd protein diffusion ( Little et al . , 2011 ) . Although our studies of pre-syncytial blastoderm stages do not address the RNA and protein distributions at the later syncytial blastoderm stages , they are likely to be relevant to them . We observed that bcd RNA and Bcd protein formed similar distributions in the interior of pre-syncytial blastoderm embryos ( Figures 4–7 ) . Although these distributions were constantly changing , the approximate match between the RNA and protein patterns remained constant . The RNA and protein were tightly concentrated at the anterior end of stage 14 oocytes , broader in nc1 embryos , and formed a graded internal plume that extended the farthest toward the posterior in nc3-4 embryos . The plume had complex geometries that were not radially symmetric , but after nc6-7 , the internal plume was no longer visible and most of the bcd RNA and protein was near or at the cortex . The cortical distributions were radially symmetric . The mechanisms that generate these distributions are not known , but because the patterns of bcd RNA and protein are so similar , we suggest that the mechanisms are probably related . Although these mechanisms cannot be deduced from the geometries of the distributions , the complex and well-defined shapes and the rapidity with which the distributions form and change would seem to be incompatible with passive diffusion . The bcd RNA and Bcd protein distributions we observed in the pre-cellular embryo reveal two distinct concentration gradients – one in the embryo interior that forms between nc1 and nc6 , and a second one that forms later at the embryo cortex . We assume that the bcd RNA and Bcd protein of the early , first gradient contribute to the second , and therefore call this a two-step model . This model contrasts with the models that have been proposed previously in which the dispersion of protein ( and RNA ) from the anterior pole is continuous during both the pre-syncytial blastoderm and syncytial blastoderm stages ( Figure 9 ) . Although the two-step model is presumably generated by motor-driven directed movement , the high yolk content of the embryo cytoplasm has impeded analysis of its cytoskeletal elements . However , several studies have reported microtubule networks in the pre-syncytial embryo ( Fahmy et al . , 2014; Karr and Alberts , 1986 ) and threads of microtubules up to 50 μm long extending from the cortex into the interior at syncytial blastoderm stages ( Fahmy et al . , 2014 ) . These networks offer a possible mechanism that might transport particles of bcd RNA ( Fahmy et al . , 2014; Spirov et al . , 2009 ) and protein posteriorly to form the internal plume in pre-nc6 embryos ( Figures 4–7 ) , and to re-distribute bcd RNA and protein to the cortex at blastoderm stages . 10 . 7554/eLife . 13222 . 019Figure 9 . Models of Bcd gradient formation . Two contrasting models of Bcd gradient formation are depicted as viewed from a sagittal section in the middle of pre-cellular embryos , oriented anterior left . A model that assumes a continuous redistribution of bcd RNA and Bcd protein ( red ) from the anterior pole ( upper row ) contrasts with the two-step model ( lower row ) in which the bcd products generate a plume in the middle of the embryo during the first four nuclear cycles and then generate a second gradient at the cortex in the syncytial blastoderm stages ( nc9-14 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13222 . 019 Our analysis of Kr expression revealed that transcription initiates as early as nc7 ( Figure 1 ) . Although the number of nuclei with nascent Kr transcripts was small , the internal nuclei with transcripts were at the approximate position along the anterior/posterior axis where older , syncytial blastoderm embryos have robust Kr expression , and this position was influenced by the level of Bcd protein in a similar way as the later , cortical expression . We also detected a concentration gradient of nuclear Bcd protein at the nc6 stage , when the embryo has just 32 nuclei that are deep in the interior ( Figure 5 ) , and observed that the both the internal and cortical Bcd gradients as well as both internal and cortical Kr expression were abnormal in exu and stau mutant embryos ( Figures 4 and Figure 4—figure supplement 2 ) . These results show that Bcd regulates Kr in nuclei that are far from the cortex , and importantly , that the internal plume of Bcd protein is a functional distribution . These findings therefore establish that it is the pre-syncytial blastoderm , internal Bcd gradient that regulates gap gene expression and organizes the embryo’s anterior/posterior axis . Although these findings do not imply what the role of the subsequent cortical Bcd gradient in the syncytial blastoderm stages Bcd gradient might be , it is most likely that the cortical gradient derives from the earlier distribution and therefore that it reflects the outputs of the pre-syncytial blastoderm , internal gradient . Morphogens such as Hedgehog , Wingless , Decapentaplegic , and Fibroblast growth factor distribute in concentration gradients across fields of cells in the tissues of developing animals . Their distributions are generated by transport along actin-based cytonemes ( Roy , 2011 #56 and reviewed in Kornberg , 2014 ) and direct exchange between producing and receiving cells at morphogenetic synapses where release and uptake of secreted proteins is regulated ( Roy , 2014 #57 and reviewed in Kornberg and Roy , 2014 ) . The generation of the Bcd concentration gradients in the pre-cellular embryo would appear to have little in common with the gradients that form by cytoneme-mediated dispersion across fields of cells , but we pose the question whether they do . Neither appears to be dependent on passive diffusion and both appear to involve dispersion along cytoskeletal cables . The critical attribute that these mechanisms share is that they provide ways to regulate movement in space and time . Stage 14 oocytes were dissected from 5 pairs of ovaries ( including the complete female reproductive system ) that were obtained from 4–5 day old , well-fed females . Stage 14a and 14b accounted for 96–97% of stage 14 oocytes . Approximately 3% were stage 14c and 14d . To obtain larger numbers of stage 14c and 14d oocytes , five pairs of ovaries were isolated and incubated in PBS for 20 hr at room temperature; the relative proportion of stage 14 oocytes in these preparations was 14a , 0%; 14b , 2–3%; 14c , 75–80%; 14d , 15–20% . Stage 14a and 14b oocytes were not extruded from the distal tip of the ovary; 14c and 14d oocytes were wholly or partially in a lateral branch of the oviduct . The stage 14c and 14d oocytes were sensitive to hypochlorite treatment and had not completed meiosis I , and were therefore designated as not activated ( Sartain and Wolfner , 2013 ) . Oocytes isolated either without or with incubation were subdivided and either stained with DAPI or processed for in situ hybridization or antibody staining . Results of in situ hybridization and antibody staining with oocytes obtained from freshly dissected ovaries and from ovaries incubated ex vivo were indistinguishable . Embryos were mounted with slight pressure under coverslips and orientation was determined by analysis of serial optical sections . Images were obtained with a Leica SPE confocal microscope and processed with ImageJ . Total RNA was prepared from five embryos with Zymo Research RNA MicroPrep kits ( Cat . #R1060 ) and quantified by absorbance with a nanodrop spectrophotometer . cDNA was prepared using Applied Biosystem High Capacity RNA-to-cDNA kits ( Cat . #4387406 ) starting with approximately 200 ng RNA . The Q-PCR reactions were carried out with a BioRad C1000 Touch Thermal Cycler and SsoAdvanced SYBR Green according to manufacturers’ instructions .

As an embryo develops , a single cell transforms into a collection of different types of cells . One protein that is crucial for this process in fruit fly embryos is Bicoid . Thirty years ago , scientists discovered that Bicoid protein is concentrated at the head end of the embryo and gradually decreases in amount towards the rear end . This concentration gradient of Bicoid protein organizes the embryo body and regulates the expression of many genes , thus directing the cells to develop different identities . Several assumptions had been made about how this gradient is established . It was thought that in the unfertilized egg , the mRNA molecules that will be translated to produce Bicoid proteins are stored in an inactive state in the region of the egg that later develops into the embryo’s head . In the embryo , the mRNA molecules were believed to remain in the head region while being translated , with the newly formed proteins then gradually spreading from this site to create the Bicoid gradient . It was also thought that no Bicoid proteins are stored in the unfertilized egg . However , no known methods were sensitive enough to investigate these assumptions . Now , using newer and more sensitive methods , Ali-Murthy and Kornberg show that Bicoid protein is present in the unfertilized fruit fly egg in the same region as the mRNA molecules that make Bicoid . Furthermore , the Bicoid gradient forms when the embryo has fewer than 32 nuclei , much earlier in development than previously thought . The Bicoid protein also does not appear to spread passively towards the rear of the embryo , but is transported in a more orchestrated manner . Overall , Ali-Murthy and Kornberg’s results suggest that the early fruit fly embryo is more organized and actively regulated than had been previously understood . This paves the way for further studies that use sensitive techniques to investigate this early stage of development .

2016

Bicoid gradient formation and function in the Drosophila pre-syncytial blastoderm

Myogenesis is an evolutionarily conserved process . Little known , however , is how the morphology of each muscle is determined , such that movements relying upon contraction of many muscles are both precise and coordinated . Each Drosophila larval muscle is a single multinucleated fibre whose morphology reflects expression of distinctive identity Transcription Factors ( iTFs ) . By deleting transcription cis-regulatory modules of one iTF , Collier , we generated viable muscle identity mutants , allowing live imaging and locomotion assays . We show that both selection of muscle attachment sites and muscle/muscle matching is intrinsic to muscle identity and requires transcriptional reprogramming of syncytial nuclei . Live-imaging shows that the staggered muscle pattern involves attraction to tendon cells and heterotypic muscle-muscle adhesion . Unbalance leads to formation of branched muscles , and this correlates with locomotor behavior deficit . Thus , engineering Drosophila muscle identity mutants allows to investigate , in vivo , physiological and mechanical properties of abnormal muscles . The musculature of each animal is composed of an array of body wall muscles allowing precision and stereotypy of movements . The somatic musculature of the Drosophila larva - about 30 distinct body wall muscles per each hemi-segment which are distributed in three layers , internal , median and external - is a model to study how a muscle pattern is specified and linked to locomotion behaviour . Each muscle is a single multinucleated fibre with a specific identity: size , shape , orientation relative to the dorso-ventral and antero-posterior body axes , motoneuron innervation and attachment sites to the exoskeleton via tendon cells located at specific positions . Intersegmental tendon cells , where a large fraction of muscles is attached , are distributed in three groups , dorsal , lateral and ventral ( Bate , 1990; Volk and VijayRaghavan , 1994; Schweitzer et al . , 2010; Armand et al . , 1994 ) . Internal muscles also attach to muscle ( s ) in the next segment , forming ‘indirect’ muscle attachment sites ( iMAS ) ( Maartens and Brown , 2015 ) . Drosophila muscle development proceeds through fusion of a founder myoblast ( founder cell , FC ) with fusion-competent myoblasts ( FCMs ) ( Deng et al . , 2017 ) . FCs originate from asymmetric division of progenitor cells ( PCs ) themselves selected from equivalence groups of myoblasts , called promuscular clusters ( PMCs ) . Muscle identity ensues from the expression by each PC and FC of a specific combination of identity transcription factors ( iTFs ) ( de Joussineau et al . , 2012 ) , established in three steps: First , different iTFs are activated in different PMCs , in response to positional information , and this expression is only maintained in PCs . Second , refinement of the iTF code occurs via cross-regulations between different iTFs in PCs and/or FCs and Notch signalling ( Carmena et al . , 1998; Carmena et al . , 2002; Enriquez et al . , 2010 ) . Several PCs can be serially selected from a PMC and give rise to different muscle identities according to birth order , adding a temporal dimension to these regulations . A well-documented example is the distinction between DA2 and DA3 identities ( Figure 1A; Dubois et al . , 2016; Boukhatmi et al . , 2012 ) . Third , transcriptional activation of iTFs in syncytial nuclei after fusion correlates with the activation of identity ‘realisation’ genes acting downstream of some iTFs ( Crozatier and Vincent , 1999; Knirr et al . , 1999; Bataillé et al . , 2010; Bataillé et al . , 2017 ) . The consequence of specific muscle identity defects on locomotion , a question of prime importance for progress on studying human myopathies which affect subsets of muscles , remains largely to be assessed . Genetically controlled muscle identity changes should , in principle , provide suitable models for studying locomotion deficits linked to muscle imbalance . However , mutations for known Drosophila iTFs are embryonic lethal and/or show pleiotropic phenotypes reflecting their multiple expression sites . Here , we took advantage of our previous characterisation of col expression in a single larval muscle , the DA3 muscle and of the two involved cis-regulatory modules ( CRMs ) , early ( E-CRM ) and late ( L-CRM ) , to generate muscle-specific mutants and circumvent lethality/pleiotropy of null mutants . CRM deletions show that E-CRM and L-CRM act redundantly at the PC stage , emphasising that PC is a key stage in specification of muscle identity . L-CRM deletion results into loss of col transcription in DA3 FCs and morphological transformation of the DA3 into a DA2-like muscle . Removal of an auto-regulatory cis-module located in L-CRM specifically abolishes col activation in syncytial nuclei fusing with the DA3 FC . This leads to incomplete DA3 transformations and the formation of bifid/branched muscles of mixed DA3/DA2 morphology . In summary , our data show that i ) the FC transcriptional program must be propagated to syncytial nuclei for a muscle to adopt a specific morphology; ii ) the precise matching of muscle/muscle attachments over the intersegmental border , which leads to a staggered rows pattern , involves a process of selective adhesion controlled by iTFs; iii ) branched muscles affect larval locomotion performance . Branched muscles are typical of late , severe phases of human Duchenne Muscular Dystrophy ( Chan and Head , 2011 ) . Drosophila iTF CRM deletion could be an effective setting for creating muscle-specific transformations and branched muscles , as new paradigms to study myopathies specifically affecting subsets of muscles in humans . Having deleted separately each col CRM allowed to assess the respective roles of iTF transcription before or after the PC step . To compare the different CRM deletions , we placed each of them over the deficiency Df chromosome . We introduced the L-CRM-moeGFP reporter to visualise the DA3 morphology at stage 15 ( Enriquez et al . , 2012 ) . Control ( +/Df ) embryos display reporter expression in the DA3/DO5 and DT1/DO3 PCs at stage 11 and the DA3 muscle at stage 15 ( Figure 2A ) . The same pattern is observed in PCs for all col-CRM deletion strains , consistent with transcript analyses ( Figure 1D ) . L-CRM-moeGFP expression at stage 15 shows that the DA3 morphology is normal in about 80% of segments in colΔE/Df embryos ( Figure 2A–B ) , and we did not pursue the analysis of this deletion strain . Low level GFP expression in colΔL1 . 3/Df embryos , consistent with col transcription data ( Figure 1D ) , shows that the DA3 muscle is most often ( 85 . 2% of segments ) transformed into a DA2-like muscle ( designated below as DA3>DA2; Figure 2A–B ) , like in col null mutant embryos ( Enriquez et al . , 2012 ) . In colΔL0 . 5/Df embryos , i . e , when only the autoregulation module has been deleted , a high number ( 29% ) of branched muscles is observed ( Figure 2A–B ) . Branched muscles correspond to incomplete transformations , with two stable anterior attachment sites , overlapping the DA3 and DA2 sites in wt embryos . The high ratios of either complete ( DA3>DA2 ) or incomplete ( branched ) transformations in L-CRM deletion mutants demonstrate that an iTF CRM deletion strategy is effective for creating viable muscle-specific identity mutants and explore branched muscle properties . Complete vs incomplete transformations in colΔL1 . 3 versus colΔL0 . 5 deletions suggest that proper level and/or maintenance of iTF expression is crucial for proper muscle development . This led us to compare the pattern of Col protein in growing DA3 syncytium between wt , colΔL1 . 3 and colΔL0 . 5 embryos . In either deletion strain , Col is detected in PCs at stage 11 but not in muscles at stage 15 ( Figure 3A ) . However , at stage 14 , some Col protein is still detected in muscle precursors in colΔL0 . 5 , not in colΔL1 . 3 embryos ( Figure 3B ) . To trace the origin of this difference , we examined col transcription in the DA3 PC , FC and stage 14 syncytium using Df/hemizygous embryos which display one hybridisation dot per active nucleus ( Figure 3C ) . In control wt/Df embryos , a dot is systematically detected in the DA3/DO5 PC ( 20/20 segments; five embryos analysed ) , the DA3 FC ( 20/20 ) and 80% of DA3 nuclei at stage 14 ( 6 of 7–8 nuclei per fibre on average; 27 segments ) . A dot is detected as well in the DA3/DO5 PC , in either colΔL1 . 3 ( 21/21 ) or colΔL0 . 5 ( 18/18 ) embryos , reflecting E-CRM activity ( Figure 1D ) . In colΔL0 . 5 embryos , a col hybridisation dot is detected in the DA3 FC ( 19/19 ) and in one nucleus , likely the FC nucleus ( 11/21 segments ) , sometimes two nuclei at stage 14 . In colΔL1 . 3 embryos , however , col transcription is only detected in a minor fraction of FCs ( 4/15 ) and is completely lost at stage 14 ( 0/6–7 nuclei per fibre on average; 24 segments ) . Patterns similar to stage 14 are observed at stage 15 , while at stage 16 , col transcription is detected neither in L-CRM deletion strains nor in control ( Figure 3—figure supplement 1 ) . Since col transcription at , and from , the PC stage appears to be nodal to DA3 identity , we measured the level of col transcripts relative to nautilus ( nau ) , the Drosophila MyoD-MRF serving as internal reference ( Figure 3D and Figure 3—figure supplement 2 ) . As expected , similar levels of col transcription are found in the DA3 PC , in control ( Mean ± sem: 1 . 22 ± 0 . 06; n = 18 ) , colΔL1 . 3 ( 1 . 15 ± 0 . 04; n = 26 ) and colΔL0 . 5 embryos ( 1 . 19 ± 0 . 04; n = 18 ) , which confirms handling by the E-CRM , with only a minor contribution from the L-CRM ( Figure 1D ) . On the contrary , high level of col transcription in the DA3 FC in wt embryos ( 2 . 20 ± 0 . 05; n = 28 ) is dependent upon L-CRM activity , since a drop is observed in colΔL1 . 3 FCs ( 1 . 10 ± 0 . 04; n = 24 ) , p<0 . 001 . More precisely , it is dependent upon the presence of Mef2 and Twist binding sites since a basal level of transcription is still observed when only the col autoregulation module is deleted ( colΔL0 . 5: 1 . 41 ± 0 . 04; n = 25 ) . Quantification of col and nau transcripts shows that the col/nau increase between the PC and FC stages in wt embryo is due to increased col transcription while the level of nau is relatively constant and is unaffected in col L-CRM mutants ( Figure 3—figure supplement 2 ) . Together , the data suggest that binding of Mef2 and Twist is required to prime col transcription in the FC nucleus before autoregulation takes off ( Figure 3E ) . Taken with one another , Col immunostaining , FISH data and col transcription quantification indicate that sustained col transcription in the FC nucleus in colΔL0 . 5 embryos ( Figure 3C–D ) , provides enough Col protein for some uptake by other DA3 nuclei at stage 14 , and this leads to their partial reprogramming to DA3 identity . Partial reprogramming could , in turn , explain the formation of branched muscles retaining some DA3 morphological characters . Moreover , the colΔL1 . 3 and colΔL0 . 5 expression data and deletion phenotypes show that both iTF transcription in the FC and reprogramming of ‘naïve’ syncytial nuclei contribute to ensure robust muscle morphological identity ( Figure 3E ) . We next investigated in more detail how ectopic muscle attachment sites at the origin of transformed and branched muscles are selected and stabilised . Double staining of ready-to-hatch ( st17 ) wt embryos for F-actin and Ilk-GFP , a component of muscle attachment sites ( Zervas et al . , 2011; Sarov et al . , 2016 ) shows that the DA3 posterior edge anchors to dorsal , and anterior edge to lateral intersegmental tendon cells ( Figure 4A ) , giving its final acute shape ( Bate , 1990 ) . Moreover , the DA3 and DA2 , as well as the DA2 and DA1 muscles precisely align with each other over each intersegmental border , forming heterotypic muscle-muscle attachment ( iMAS; [Maartens and Brown , 2015] ) at the origin of the staggered rows disposition of DA muscles . No DA3/DA3 ( homotypic ) iMAS surface is observed ( Figure 4A ) . On the contrary , in colΔL1 . 3 embryos , the anterior edge of DA3>DA2 transformed muscles anchors to dorsal tendon cells instead of lateral intersegmental tendon cells , leading to a ‘dual DA3>DA2 morphology’ , DA2-like at the anterior , and DA3 at the posterior edge . This dual identity leads to iMASs between adjacent DA3>DA2 muscles ( Figure 4A ) . As a consequence , DA2 iMASs are shifted dorsally and the general pattern of DA muscles is affected . To explore the dynamics of DA3 muscle attachment , we live-imaged wt and colΔL1 . 3 embryos , starting at stage 12 ( defined as t0 in Videos 1 and 2 ) . The DA3 muscle is visualised by L-CRM-moeGFP and tendon cell along the entire intersegmental border by stripe Gal4;UASmCD8RFP ( Volohonsky et al . , 2007 ) . In wt embryos between stages 12 and 13 ( Video 1; Figure 4B ) , the DA3 muscle precursor extends protrusions towards dorsal tendon cells , posteriorly , and both dorsal and lateral tendon cells , anteriorly , diverging from a bipolar extension scheme ( Schnorrer and Dickson , 2004 ) . At stage 14 , the posterior DA3 edge makes contacts with dorsal tendon cells before the anterior edge ( s ) reaches the intersegmental border , a time gap previously observed during live-imaging of a ventro-lateral muscle ( Gilsohn and Volk , 2010 ) . In addition , numerous filopodia emanate from the DA3 dorsal surface and contact the DA2 ventral surface , contributing to give the DA3 muscle precursor its transient angled-shape . At late stage 14 , the posterior DA3 attachment widens parallel to the intersegmental border . Its anterior attachment to lateral tendon cells in turn becomes stable , whereas dorsal anterior filipodia appear to be repelled from the intersegmental border where the DA3 muscle in the preceding segment is attached . Remaining filopodia are limited to the rims of DA3 anchoring , possibly suggesting a mechanism of homotypic repulsion . In colΔL1 . 3 embryos , the main axis of the DA3>DA2 muscle precursor elongation is more longitudinal than wt , already at stage 12 ( Video 2; Figure 4B ) . At stage 13 and like in wt , many protrusions emanate from the DA3 dorsal surface , but unlike in wt , protrusions contacting dorsal tendon cells do not retract at later stages when they contact the DA3>DA2 muscle from the preceding segment . Rather , stabilisation of contacts made by these protrusions prefigures abnormal DA3>DA2/DA3>DA2 iMAS formation ( Figure 4A ) . In some segments , contacts with lateral tendon cells are also stabilised , resulting into branched muscles . In summary , imaging of live wt embryos illustrates different steps involved in establishment of the acute DA3 orientation: posterior attachment to dorsal tendon cells at the same time as anterior exploration of dorsal and lateral tendon cell . This is followed by DA3 attachment to lateral , and retraction from dorsal tendon cells ( Figure 4B , Video 1 ) , cumulating into heterotypic DA3/DA2 iMAS stabilisation ( Figure 4A and B ) . This retraction does not take place in colΔL1 . 3 embryos ( Figure 4B and Video 2 ) . Further determining how DA3>DA2/DA3>DA2 homotypic attachment could interfere with the staggered rows pattern of DA muscles , required to visualise at the same time the DA3 and DA2 muscles and tendon cells contours . Vestigial ( Vg ) is expressed in DA muscles ( Deng et al . , 2010; Tixier et al . , 2010 ) . We screened the vg regulatory landscape and characterised one vg CRM active in the DA3 and DA2 ( and VL1 ) muscles , VgM1 ( Figure 5—figure supplement 1 ) . Expressing together VgM1-mCD8GFP-H2bRFP , L-CRM-moeGFP and stripeGal4:UAS-mCD8RFP , and co-staining of stage 16 control embryos for α-Spectrin , a protein enriched at muscle attachment sites , shows that the posterior edge of DA3 precisely aligns with the anterior edge of DA2 ( Figure 5A , Figure 5—figure supplement 1 ) , suggesting heterotypic attractive cues . This attraction is already observed at stage 14 , when the dorsal DA3 and ventral DA2 surfaces contact each other via numerous protrusions ( Figure 5A and B; Figure 5—figure supplement 2 ) . Strikingly , at stage 15 , the contact zone gets away from intersegmental borders and the DA3/DA3 and DA2/DA2 homotypic connections disappear , to give way to heterotypic DA3/DA2 stable iMAS formation . In colΔL1 . 3 embryos , the initial steps , stages 14 and 15 , are similar to wt , except for the absence of stable connection of DA3>DA2 to lateral tendon cells , as seen by live imaging ( Video 2 ) . At stage 16 , DA3>DA2/DA3>DA2 and DA3>DA2/DA2 iMASs are privileged , while similar to control , no homotypic DA2/DA2 iMAS forms , suggesting homotypic repulsion ( Figure 4; Figure 5A ) . Together , the data indicate that the precise pattern of iMASs and muscles is contributed by a combination of attractive and , possibly , repulsive cues downstream of muscle iTFs ( Figure 5B ) . To investigate the impact of embryonic muscle patterning defects on Drosophila larval crawling , we first recorded the fraction of DA3 , DA3>DA2 and branched muscles in 3rd instar wt and colΔLCRM larvae expressing GFP under control of the Myosin heavy chain ( Mhc ) promotor region ( Figure 6A ) . It conforms to the statistics in late embryos ( Figure 2B ) , except for an increased proportion of branched muscles , which could reflect under-evaluation of their number in embryos , due to threshold detection limit of L-CRM-moeGFP expression in thin fibres ( Figure 6B ) . We then examined the pattern of MASs , using scanning electron microscopy ( SEM ) of dissected larval filets ( Figure 6C; Figure 6—figure supplement 1 ) . In wt larvae , as seen in stage 17 embryos ( Figure 4A ) , the anterior edge of the DA3 muscle is anchored to a nodal lateral attachment site shared with the LL1 muscle , while its posterior edge aligns with the anterior edge of the DA2 muscle in the next posterior segment . Precise heterotypic DA2/DA1 iMASs are also visible over each intersegmental border . The regular alignment of the DA1 , DA2 and DA3 muscles both draws strength lines spanning three adjacent segments , and regular tension surfaces between segments ( Figure 6C; Figure 6—figure supplement 1 ) . In colΔL1 . 3 larvae , the anterior DA3 muscle attachment has shifted from lateral to dorsal to form a DA3>DA2/DA3>DA2 homotypic iMAS ( Figure 6C ) . This leads in turn to the formation of narrow , ectopic DA2/DA2 contacts and the DA1-DA2-DA3 strength line is distorted ( Figure 6—figure supplement 1 ) . In case of branched DA3 muscles , two strength lines co-exist ( Figure 6C and Figure 6—figure supplement 1 ) . In conclusion , SEM analyses of larval muscles show that loss of DA3 identity leads to ectopic iMASs between several dorsal internal muscles , distortion of the staggered ends architecture of DA muscles and of their alignment between consecutive segments . Drosophila larval crawling relies upon ordered abdominal body wall muscle contractions ( Heckscher et al . , 2012 ) . The distorted muscle patterns observed in colΔL1 . 3 larvae raised the question of whether it impacted on locomotion . To address this question , we compared the locomotion of +/Df , colΔL1 . 3/Df and colΔL0 . 5/Df larvae using FIM ( FTIR-based Imaging Method ) and the FIMTrack software , which allows tracking simultaneously many larvae and quantitatively describing a variety of stereotypic movements ( Risse et al . , 2013; Risse et al . , 2014; Risse et al . , 2017 ) . Here , we focused on three parameters: crawling speed , stride length , stride duration . We first recorded the ‘walking rate’ , also called crawling speed , during the first 20 s , after larvae have been dropped on the agarose gel . At that time , larvae engage an ‘escape response’ corresponding to an active crawling phase ( Figure 7A ) . Box-plot graphs ( left ) show intra-variability for each of the three genotypes . Beyond this variability , we observe , however , that colΔL0 . 5/Df larvae display a significantly reduced crawling speed on average 1 . 05 ± 0 . 034 mm/sec ( n = 108 ) , compared to 1 . 15 ± 0 . 031 mm/sec for +/Df controls ( n = 118 ) , ( p=0 . 03 ) ( Figure 7A ) . To further investigate the origin of this speed reduction , we measured two crawling speed parameters: stride length and stride duration ( Figure 7B–C ) . A significantly shorter stride was measured for colΔL0 . 5/Df larvae ( 1 . 08 ± 0 . 025 mm ) compared to control +/Df larvae ( 1 . 17 ± 0 . 024 mm ) , ( p=0 . 008 ) . Furthermore , stride duration was extended , from ( 1 . 09 ± 0 . 014 s ) for +/Df larvae to 1 . 15 ± 0 . 017 s for colΔL0 . 5/Df larvae . For both crawling speed , stride length and stride duration , colΔL1 . 3/Df larvae ( n = 112 ) display intermediate values . Yet , differences with either control or colΔL1 . 3/Df larvae fall below the significance threshold level , suggesting that mis-orientation of the DA3>DA2 muscle does not , by itself , significantly impair the efficiency of segment contraction . Since larval crawling integrates information provided by neuronal networks , relayed by synaptic connections between motoneurons ( MNs ) and muscles , the mobility phenotype of colΔL0 . 5/Df larvae could indicate defects motor innervation of DA3>DA2 muscles . The DA3 and DA2 muscles are innervated by the intersegmental nerve ( ISN ) which fasciculates motor axons reaching dorsal muscles ( Hoang and Chiba , 2001; Landgraf and Thor , 2006 ) . To examine the DA3>DA2 and branched DA3 innervation , we used anti-HRP and phalloidin staining to view ISN motoneuron projections and muscles , respectively ( Figure 7—figure supplement 1 ) . In wt embryos , the MN projection which innervates DA3 leaves the ISN ventral to the DA3 position and orients left such that the neuromuscular junction locates to the first anterior third of the muscle ( 100%; n = 28 segments ) . In case of a complete DA3>DA2 transformation , a MN projection is still observed ( 100%; n = 15 segments ) , which leaves the ISN at a more dorsal position than wt , reflecting the dorsal shift of the DA3>DA2 , relative to DA3 muscle . In case of branched muscles , the lower branch is always innervated ( 100% , n = 30 segments ) . Only in 20% of the cases , the second , upper branch is also innervated . These innervation data indicate that fully transformed DA3 muscles are innervated , but only one branch of branched muscles is , in most cases . It remains to be established whether this asymmetric innervation contributes to the reduced crawling speed of colΔL0 . 5 larvae . Some iTFs are transiently transcribed during Drosophila muscle development , for example , Kr and nau , others such as col , ladybird and slouch/S59 , at every step of the process ( Dubois et al . , 2016; Knirr et al . , 1999; Bataillé et al . , 2017; Dubois et al . , 2007; Michelson et al . , 1990; Jagla et al . , 2002 ) . Reporter analyses indicated that col transcription is controlled by two sequentially acting CRMs , of overlapping activity at the PC step , suggesting a handover mechanism between CRMs at this stage ( Enriquez et al . , 2012 ) . However , deletion analyses show that L-CRM activity in PCs is not dependent , but redundant with E-CRM activity , and can be separated from col positive autoregulation that is specific to the DA3 lineage . This leads to a new model where iTF code refinement at each step of muscle identity specification , PMC >PC , PC >FC and FC >syncytial nuclei , is driven by a separate CRM . Distribution of the PC-identity information into two CRMs further supports the idea that iTF regulation at the PC stage is nodal to muscle identity specification ( Carmena et al . , 1998; Dubois et al . , 2016; Enriquez et al . , 2012; Jagla et al . , 2002; Nose et al . , 1998; Kumar et al . , 2015 ) . Our former analysis started to decrypt the combinatorial control of each dorso-lateral muscle identity , which involves at least eight different iTFs , in addition to Nau/MRF [Dubois et al . , 2007] . Persistent , low level expression of Col protein in the DT1 and LL1 muscle precursors upon removal of the col autoregulatory module suggests the existence of both positively and negatively acting iTF-responsive elements in this module . Our present analysis concentrated on the DA3 lineage where col expression is maintained , whereas col is expressed in several PCs . Morphological transformations of other dorso-lateral muscles are observed at variable frequency in col protein null mutants ( Enriquez et al . , 2010 ) , suggesting that E-CRM activity provides robustness to the combinatorial control of identity of these muscles and that robustness is likely also contributed by other iTFs ( Dubois et al . , 2016; Schnorrer and Dickson , 2004 ) . PC/FC specific CRMs have only been functionally identified for a handful of muscle iTFs ( Rivera et al . , 2019 ) . Computational predictions identified , however , several thousand putative muscle enhancers and uncovered extensive heterogeneity among the combinations of transcription factor binding sites in validated enhancers , beside sites for the core intrinsic muscle regulators Tin , Mef2 and Twi ( Sandmann et al . , 2007; Gisselbrecht et al . , 2013; Cusanovich et al . , 2018 ) . Dissecting whether step-specific and redundant/distributed CRM configurations ( Cannavò et al . , 2016; Hong et al . , 2008; Frankel et al . , 2010 ) apply to many muscle iTFs and underlie the progressively refined control of final muscle patterns is a future step . The process by which iTFs determine the final morphological features of each muscle , is not fully understood . Fusing FCM nuclei generally adopt the iTF protein code of the FC nucleus , while propagation of iTF transcription and activation of realisation genes is muscle lineage-specific ( Boukhatmi et al . , 2012; Crozatier and Vincent , 1999; Knirr et al . , 1999; Bataillé et al . , 2010; Bataillé et al . , 2017; Bourgouin et al . , 1992 ) . We have previously shown that Col protein import precedes activation of col transcription in fused FCM nuclei , and correlates with the activation of realisation genes , a sequence of events termed syncytial identity reprogramming ( Bataillé et al . , 2017; Dubois et al . , 2007 ) . Upon loss of col transcription in the DA3 FC ( colΔ1 . 3 embryos ) , there is a complete DA3>DA2 transformation . When col transcription is maintained ( colΔL0 . 5 embryos ) , in the DA3 FC but not propagated to other syncytial nuclei , there is an incomplete DA3 transformation into branched muscles . From this , two conclusions can be drawn: 1 ) Final selection of myotendinous connection sites is intrinsic to FC identity . 2 ) Identity reprogramming of syncytial nuclei is required for robustness of this selection and precise muscle patterns . Live imaging of lateral-oblique and ventral transverse muscles development distinguished 3 phases of muscle elongation ( Schnorrer and Dickson , 2004 ) : 1 ) FC migration , stage 12 , ending with the first FC/FCM fusion event and stretching of the muscle precursor along a given axis; 2 ) bipolar myotube elongation , characterised by the presence of extensive filopodia at both axis ends , in search for attachment sites , stages 13 to 15; 3 ) maturation of myotendinous attachment , stage 16 . The formation of DA3>DA2 and branched muscles in CRM mutants recalls a transient exploration by the wt DA3 muscle precursor of both dorsal and lateral tendon cells , a process deviating from the bipolar migration/attachment scheme ( Schweitzer et al . , 2010; Enriquez et al . , 2012; Schnorrer and Dickson , 2004; Bahri et al . , 2009 ) . Interestingly , ectopic Col expression in the DA2 muscle leads to reciprocal DA2>DA3 transformation as well as branched muscles ( Boukhatmi et al . , 2012 ) . This indicates that selection of dorsal versus lateral tendon cells is a highly controlled process . A few molecules involved in targeted attachment of subsets of muscles to specific tendon cells have been identified: Kon tiki/Perdido , a single pass transmembrane protein and the PDZ protein DGrip for proper elongation of ventral longitudinal muscles ( Schnorrer et al . , 2007 ) ; the ArfGAp protein Git for sensing integrin signaling and halting elongation of Lateral Transverse ( LT ) muscles once their attachment site has been reached ( Bahri et al . , 2009; Richier et al . , 2018 ) . Robo/Slit signaling attracts muscles at segmental borders , the Slit ligand being expressed by tendons , and Robo and Robo2 receptors by elongating muscles . slit also acts as a short-range repellent contributing to the collapse of leading-edge filopodia when a muscle reaches the tendon extracellular matrix ( Ordan and Volk , 2015; Ordan et al . , 2015 ) . In vivo imaging showed that the DA3 muscle fails to stop at the segment border in slit mutants and sometimes branches ( Ordan et al . , 2015 ) . However , this is not observed in colΔLCRM; the DA3 slit branching pattern is rather similar to that frequently observed in nau mutants , with two posterior attachment sites in place of one ( Dubois et al . , 2016; Boukhatmi et al . , 2012 ) . Which realisation genes downstream of iTFs are responsible for the precision of DA attachment sites , that is , proper balancing attraction/repulsion cues , should be the focus of future studies . In addition to attaching to tendon cells , it was previously shown that internal muscles attach to each other ( Maartens and Brown , 2015; Bate and Rushton , 1993 ) . Detailed imaging in both embryos and larvae shows that the DA3/DA2 and DA2/DA1 attachment sites precisely match over each segmental border , such that larval DA1 , DA2 , and DA3 align over three consecutive segments . Recording DA3 and DA3>DA2 muscle development by a combination of live imaging and immunostainings shows that DA3/DA2 matching in wt embryos results from both heterotypic DA3/DA2 attraction and , possibly , homotypic repulsion . Prior attachment of the posterior DA3 edge to dorsal tendon cells leads to retraction of filopodia issued from the homologous muscle in the next adjacent segment while stabilisation of heterotypic contacts results into DA3/DA2 iMAS formation over the intersegmental border . A similar process results in alignment of DA2 with DA1 . Interestingly , DA3>DA2 establish both homotypic iMASs , and heterotypic iMASs with DA2 , suggesting preserved attraction to tendon cells and partial loss of preferential heterotypic adhesion . Prior posterior attachment was previously observed during development of abdominal adult muscles ( Currie and Bate , 1991 ) . Whether this temporal sequence is instrumental in the precise matching of muscles over each segmental border and whether competition between muscles for the same tendon cells is also involved remain to be assessed . Cell matching is a widely used process during embryogenesis to construct complex tissue architecture . Selective filopodia adhesion has recently been shown to ensure precise matching between identical cardioblasts and boundaries between different cell identities in the Drosophila heart . In this case , homotypic matching is linked to differential expression by each cell type of the adhesion molecules , Fasciclin III and Ten-m ( Zhang et al . , 2018 ) . Transcriptome analyses of specific muscles at different developmental times should allow to identify attractive and , possibly , repulsive molecules , acting in muscle precise matching . Drosophila larval crawling is a well-suited paradigm to link muscle contraction patterns and locomotor behaviour . Longitudinal , acute and oblique muscles within a larval segment contract together and , as they begin to relax , the contraction is propagated to the next segment , creating a peristaltic wave from tail to head ( forward locomotion ) , or head to tail ( backward locomotion ) ( Heckscher et al . , 2012 ) . The rhythmic movements of locomotion are part of behavorial routines that facilitate the exploration of an environment . Exploratory routines alternate straight line movement also called ‘active crawling phase’ , with change of direction , the ‘reorientation phase’ ( Günther et al . , 2016; Berni et al . , 2012; Lahiri et al . , 2011 ) . The active larval crawling phase requires an intense , prolonged muscular effort . In this study , we focused on crawling parameters during this phase . The crawling speed during the escape response is not significantly reduced in colΔL1 . 3/Df , compared to control +/Df larvae , indicating that muscle contraction is properly controlled and that a mechanical compensation mechanism for the DA3 mis-orientation could occur . However , it is significantly reduced in colΔL0 . 5/Df larvae . This seems paradoxical because the number of DA3>DA2 transformed muscles is higher in colΔL1 . 3/Df larvae . colΔL0 . 5/Df larvae present many branched muscles , however . While DA3>DA2 are always innervated , only one branch of branched muscles is , most of the time , raising the possibility that branched muscles do not contract properly , or with a gap in time ( see Zarin et al . , 2019 ) . From these different observations , we can conclude: i ) single muscle transformations only moderately impact crawling speed , raising the possibility of biomechanical compensation by other muscles; ii ) branched muscles could be less efficient than fully transformed muscles . At this point , the reason why , – either mechanic weakness , improper innervation or impaired Ca2+ wave propagation , antagonistic force lines upon muscle contraction - may only be object of speculation . The generation of branched muscles in Drosophila identity mutants is one important finding as branched muscle fibres accumulate in humans , following muscle regeneration after damage or in Duchenne muscular dystrophy patients ( Chan and Head , 2011 ) . It opens the possibility to investigate , in vivo , how physiological properties of branched fibres differ from morphologically normal fibres and associated mechanical instability in an otherwise normal muscle pattern . All Drosophila melanogaster stocks and genetic crosses were grown using standard medium at 25°C . The strains used were white[1118] , colLCRM 4–0 . 9 ( Enriquez et al . , 2010 ) , col1 ( Crozatier and Vincent , 1999 ) , sr-Gal4 ( obtained from G . Morata , Madrid , Spain ) . The 12 kn and 10 vg Janelia-Gal4 lines ( GMR ) ( Pfeiffer et al . , 2008 ) , UAS-mcd8RFP , Mhc-GFP , Df ( 2L ) BSC429 , knMI15480 y1 w*; Mi{MIC}knMI15480/SM6a ( BDSC_67516 ) ( Nagarkar-Jaiswal et al . , 2015 ) , vasa-cas9VK00027 ( BDSC_51324 ) , Ilk-GFP ( w1118; P{PTT-GB}IlkZCL3111 ) ( BDSC_6831 ) , lines were provided by the Bloomington Drosophila Stock Center . The col1 and Df ( 2L ) BSC429 strains were balanced using CyO , {wgen11-lacZ} or CyO , {dfd-YFP} and homozygous mutant embryos or larvae identified by absence of lacZ or YFP expression , respectively . Genomic col target sites were identified using http://tools . flycrispr . molbio . wisc . edu/targetFinder/ ( Gratz et al . , 2014 ) . Prior to final selection of RNA guides ( gRNA ) for deletions of col CRMs , genomic PCR and sequencing of DNA from knMI15480 and vasa-cas9VK00027 flies was performed to check for polymorphisms in the targeted regions . Guides targeting E-CRM and L-CRM were inserted in the pCFD4: U6:3-gRNA vector ( Addgene no: 49411 ) as described ( Port et al . , 2014 ) ; ( see http://www . crisprflydesign . org/wp-content/uploads/2014/06/Cloning-with-pCFD4 . pdf ) . All guides were verified by sequencing . The sequences of the oligonucleotides used to construct each gRNA expression plasmid are given in Figure 1—figure supplements 1 and 2 . To delete the core region of L-CRM , vasa-cas9 embryos were microinjected with gRNAs in pCFD4 ( 200 ng/μl ) . To delete the E-CRM , knMI15480 embryos were injected with gRNA in pCFD4 ( 150 ng/μl ) and pAct-Cas9U6 ( 400 ng/μl ) . Each adult hatched from an injected embryo was crossed to the balancer stock snaSco/CyO , {wgen11-LacZ} and 100–200 F1 fly were individually tested for either col CRM deletion by PCR on genomic DNA . A pre-screening for E-CRM deletion was based on the loss of yellow carried by Mimic knMI15480 . The yellow intron ( yi ) , FlyBase ID #FBgn0004034 ( position: 356918–359616 ) was inserted in the lacZ coding region between aa ( Tyr 952 ) and aa ( Ser 953 ) by standard PCR-based cloning position . The resulting fragment was cloned downstream of L-CRM inserted in a pAttB vector , and micro-injected in embryos for chromosomal insertion at position 68A4 . VgM1-moeGFP was constructed by PCR amplification of the GMR69G04 and GMR69G05 overlap . The 1 . 4 kb amplicon ( named VgM1 ) was inserted upstream of moeGFP to generate the pAttB VgM1-moeGFP construct . It was inserted at position 68A4 on the third chromosome . Antibody staining , in situ hybridisation with intronic probes and phalloidin staining were as described previously ( Dubois et al . , 2007 ) . Primary antibodies were: mouse anti-Col ( 1/50; Boukhatmi et al . , 2012; Dubois et al . , 2007 ) , anti-LacZ ( 1/1000; Promega ) , mouse anti α-Spectrin ( 1/200; Hybridoma Bank ) , rabbit anti-GFP ( 1/1000; Torrey Pines Biolabs ) , chicken anti-GFP ( 1/500; Abcam ) , Phalloidin-Texas RedX ( 1/500; Thermofisher Scientific ) . Secondary antibodies were: Alexa Fluor 488- , 555- and 647- conjugated antibodies ( 1/300; Molecular Probes ) and biotinylated goat anti-mouse ( 1/2000; Vector Laboratories ) . To both visualise motoneuron axonal pathways and muscles , fillets of control and colΔL1 . 3 third instar larvae were incubated overnight with Alexa 594-conjugated anti-HRP ( 1/300; Jackson Immunological Research ) and Alexa Fluor 488 Phalloidin ( 1/500; Thermofisher Scientific ) , at 4°C . To prepare fillets , third instar larvae placed in myorelaxant buffer ( Yalgin et al . , 2011 ) , were cut longitudinally on the ventral side to expose the dorsal and dorso-lateral musculature . Fillets were then fixed 1 hr in 4% formaldehyde and washed in PBT . In situ hybridisation with Stellaris RNA FISH probes were done as described by the manufacturer for Drosophila embryos ( https://www . biosearchtech . com ) . The FISH probe sets for col and nau were designed using the Stellaris probe designer ( https://www . biosearchtech . com/stellarisdesigner ) and labelled with Quasar 670 Dye ( col ) and Quasar 570 Dye ( nau ) ( Stellaris Biosearch Technologies ) . One set of 48 oligonucleotides was designed against the first col intron to detect primary nuclear transcripts . Another set of 48 oligonucleotides was also designed against the first and third nau introns . When antibody staining and FISH were combined , the standard immuno-histochemistry protocol was performed first , with 1 U/μl of RNase inhibitor from Promega included in all solutions , followed by the FISH protocol . Confocal sections were acquired on Leica SP8 or SPE microscopes at 40x or 63x magnification , 1024/1024 pixel resolution . Images were assembled using ImageJ and Photoshop softwares . To quantify the level of col nuclear transcripts in FCs and PCs , we calculated the ratio between col and nau hybridisation signals using intronic probes . Before using nau as internal reference we verified that nau transcription level is not modified in col L-CRM mutants . The same laser parameters were set for all intronic probes and at least five different embryos at each stage 11 and 12 were recorded . Optimal Z stacks were acquired at × 40 . ImageJ was used to analyse the data . For each stack , a Sum slices projection was generated . Each region of interest ( ROI ) , corresponding to a DA3 nucleus , was manually drawn , based on Mef-2 immunostaining . The same ROI served to determine the intensity of nau and col signals on the green and red channels , respectively ( Figure 3—figure supplement 2A ) . A threshold was applied to each channel to remove background . Data plots and statistical analyses were performed with Prism 5 . 0 using unpaired t-test . To quantify embryonic phenotypes , L-CRM-moeGFP embryos were immunostained with a primary mouse anti-GFP ( 1/500 ) ( Roche ) and secondary biotinylated goat anti-mouse ( 1/2000 ) ( VECTASTAIN ABC Kit ) . Stained embryos were imaged using a Nikon eclipse 80i microscope and a Nikon digital camera DXM 1200C . A minimum 100 A1-A7 abdominal segments of stage 15–16 embryos were analysed for each genotype . ( +/Df: n = 127 segments - 16 embryos; colΔE/Df: n = 170 segments - 23 embryos; colΔL1 . 3/Df: n = 190 segments - 27 embryos; colΔL0 . 5/Df: n = 103 segments - 13 embryos ) . To quantify larval phenotypes , wandering L3 larva displaying Mhc-GFP reporter line were immobilised between slide and coverslip , and left and right larval sides imaged using Nikon AZ100 Macroscope at 5x magnification . Minimum 260 abdominal segments were analysed for each genotype . ( +/Df: n = 375 segments - 27 larvae; colΔL1 . 3/Df: n = 320 segments - 23 larvae; colΔL0 . 5/Df: n = 264 segments - 19 larvae ) . Embryos were bleach dechorionated and stage 12 embryos manually picked , laterally orientated and mounted on a coverslip coated with heptane glue to prevent drift during imaging . A drop of water was placed on the embryos to maintain their survival . Images were collected on a Leica TCS-SP8 confocal using a 25X water immersion lens . Sections were recorded every 130 to 150 s for the wt embryos and every 120 to 160 s for the colΔL1 . 3 embryos , and z-stacks collected with optical sections at maximum 1 µm interval . Image processing was performed with Fiji ( http://fiji . sc/wiki/index . php/Fiji ) and custom programming scripts in Fiji . The z-stacks projections were corrected in x and y dimensions by manual registration using a reference point tracking . To prepare fillets , third instar wild type and homozygous colΔL1 . 3 larvae raised at 25°C were dissected in myorelaxant buffer , according to Gratz et al . , 2014 . Larvae were cut longitudinally on the ventral side to preserve and expose the dorsal and dorso-lateral musculature . Fillets were then fixed 1 hr in a 4% formaldehyde/2 . 5% glutaraldehyde mixture in 1X PBS , washed in water and dehydrated gradually in ethanol . Fillets were dried at the critical point ( Leica EM CPD 300 critical point apparatus ) , covered with a platinum layer ( Leica EM MED 020 metalliser ) and imaged with a Quanta 250 FEG FEI scanning microscope . We conducted locomotion assays by tracking the trajectory of larvae using the FIM method ( Risse et al . , 2013 ) . Wandering third instar larvae were gently picked up with a paintbrush and transferred to an agar plate . The larvae were then videotaped using a digital camera ( Baumer VCXG53M ) ; lentille ( Kowa LM16HC ) ; infrared filter ( IF093SH35 . 5 ) . Each video containing 5 to 10 larvae per run , on a 1% agarose gel , was recorded at five frames/sec for 20 s . Individual larva were tracked using the FIMTrack software ( Risse et al . , 2014 ) , which provided the position across time of five points regularly spaced along the spine of each animal , from head to tail . Analysis was done by using MATLAB software . Peristalsis cycles were obtained using the derivative of the spine length ( i . e . , sum of the distance across successive point along the spine ) through time , which provide a time series smoothly oscillating around zero . For each peristalsis cycle , we measured Stride length ( centroid displacement across each cycle ) and Stride duration ( cycle duration ) . Walking rate was obtained by measuring the distance of the centroid ( 3rd spine point ) across successive frame . These values were averaged for each individual across the 20 s of recording . Statistical comparisons between genotypes were computed using a linear model with GenoT as fixed effect , and individual larva as a statistical unit .

Each muscle in the body has a unique size , shape and set of attachment points . Animals need all of their muscles to have the correct identity to help maintain posture and control movement . A specific set of proteins , called transcription factors , co-ordinate and regulate gene activity in cells so that each muscle develops in the right way . To create a muscle , multiple precursor cells fuse together to form a muscle fibre , which then elongates and attaches to specific sites . Correct attachment is critical so that the fibre is properly oriented . When this process goes wrong , for example in disease , muscle fibres sometimes attach to the wrong site; they become branched and cannot work properly . Collier is a transcription factor protein that controls muscle identity in the fruit fly Drosophila melanogaster . However , like many transcription factors , Collier also has several other roles throughout the body . This made it difficult to evaluate the effect of the protein on the formation of specific muscles . Here , Carayon et al . managed to selectively deactivate Collier in just one muscle per body section in the larvae of fruit flies . This showed that the transcription factor is needed throughout muscle development; in particular , it is required for muscle fibres to select the correct attachment sites , and to be properly oriented . Affected muscles showed an altered orientation , with branched fibres attaching to the wrong site . Even minor changes , which only affect a single muscle from each body segment , greatly impaired the movement of the larvae . The work by Carayon et al . offers a new approach to the study of muscular conditions . Branched muscles are seen in severe human illnesses such as Duchenne muscular dystrophy . Studying the impact of these changes in a living animal could help to understand how this disease progress , and how it can be prevented .

2020

Intrinsic control of muscle attachment sites matching

In complex biological systems , simple individual-level behavioral rules can give rise to emergent group-level behavior . While collective behavior has been well studied in cells and larger organisms , the mesoscopic scale is less understood , as it is unclear which sensory inputs and physical processes matter a priori . Here , we investigate collective feeding in the roundworm C . elegans at this intermediate scale , using quantitative phenotyping and agent-based modeling to identify behavioral rules underlying both aggregation and swarming—a dynamic phenotype only observed at longer timescales . Using fluorescence multi-worm tracking , we quantify aggregation in terms of individual dynamics and population-level statistics . Then we use agent-based simulations and approximate Bayesian inference to identify three key behavioral rules for aggregation: cluster-edge reversals , a density-dependent switch between crawling speeds , and taxis towards neighboring worms . Our simulations suggest that swarming is simply driven by local food depletion but otherwise employs the same behavioral mechanisms as the initial aggregation . Collective behavior has been widely studied in living and non-living systems . While very different in their details , shared principles have begun to emerge , such as the importance of alignment for flocking behavior in both theoretical models and birds ( Bialek et al . , 2012; Pearce et al . , 2014; Reynolds , 1987 ) . Until now , the study of collective behavior has mainly focused on cells and active particles at the microscale , controlled by molecule diffusion and direct contact between cells or particles ( Köhler et al . , 2011; De Palo et al . , 2017; Peruani et al . , 2012; Starruß et al . , 2012 ) , and on animals at the macroscale , aided by long-range visual cues ( Bialek et al . , 2012; Katz et al . , 2011; Pearce et al . , 2014 ) . Collective behavior at the intermediate mesoscale is less well-studied , as it is unclear what processes to include a priori . At the mesoscale , sensory cues and motility may still be limited by the physics of diffusion and low Reynolds numbers , respectively , yet the inclusion of nervous systems allows for increased signal processing and a greater behavioral repertoire . Do the rules governing collective behavior at this intermediate scale resemble those at the micro- or the macro-scale , some mixture of both , or are new principles required ? C . elegans collective behavior can contribute to bridging this scale gap . Some strains of this 1 mm-long roundworm are known to aggregate into groups on food ( de Bono and Bargmann , 1998 ) ; here we also report an additional dynamic swarming phenotype that occurs over longer time periods . C . elegans represents an intermediate scale not only in physical size but also in behavioral complexity—crawling with negligible inertia , limited to touch and chemical sensing , yet possessing a compact nervous system with 302 neurons ( White et al . , 1986 ) that supports a complex behavioral repertoire ( Hart , 2006; Schwarz et al . , 2015 ) . Wild C . elegans form clusters on food at ambient oxygen concentrations , as do loss-of-function neuropeptide receptor 1 ( npr-1 ) mutants . The laboratory reference strain N2 , on the other hand , has a gain-of-function mutation in the npr-1 gene that suppresses aggregation ( de Bono and Bargmann , 1998 ) , rendering N2 animals solitary feeders . Thus , a small genetic difference ( just two base pairs in one gene for the npr-1 ( ad609lf ) mutant ) has a big effect on the population-level behavioral phenotype . Previous research on collective feeding has focused primarily on the genetics and neural circuits that govern aggregation ( Bretscher et al . , 2008; Busch et al . , 2012; Chang et al . , 2006; Cheung et al . , 2005; de Bono et al . , 2002; de Bono and Bargmann , 1998; Gray et al . , 2004; Jang et al . , 2017; Macosko et al . , 2009 ) , rather than on a detailed understanding of the behavior itself . Rogers et al . ( 2006 ) is a notable exception and includes an investigation of the behavioral motifs that might lead to cluster formation including direction reversals at the edge of clusters . However , we do not know whether these candidate motifs are sufficient to produce aggregation . We also do not know whether aggregation at short times and swarming at longer times are distinct behaviors or different emergent properties of the same underlying phenomenon . In this paper , we use fluorescence imaging and multi-worm tracking to examine individual behavior inside aggregates . We present new and systematic quantification of the aggregation behavior in hyper-social npr-1 ( ad609lf ) mutants ( henceforth referred to as npr-1 mutants ) and hypo-social N2 worms . Next , we draw on the concept of motility-induced phase transitions to explain aggregation as an emergent phenomenon by modulating only a few biophysical parameters . Unlike aggregation driven by attractive forces , in motility-induced phase transitions individuals can also aggregate simply due to their active movement and non-attractive interactions , such as volume exclusion ( avoidance of direct overlap ) ( Redner et al . , 2013a ) . For instance , this concept has contributed understanding to the aggregation of rod-shaped Myxococcus xanthus bacteria , which , similar to C . elegans , also exhibit reversals during aggregation ( Mercier and Mignot , 2016; Peruani et al . , 2012; Starruß et al . , 2012 ) . We build an agent-based phenomenological model of simplified worm motility and interactions . By mapping out a phase diagram of behavioral phenotypes , we show that modulating cluster-edge reversals and a density-dependent switch between crawling speeds are sufficient to produce some aggregation , but not the compact clusters observed in experiments . We found that medium-range taxis towards neighboring worms is necessary to tighten clusters and increase persistence . Finally , combining this model with food depletion gives rise to swarming over time , suggesting that the same behavioral rules that lead to the initial formation of aggregates also underlie the dynamic swarming reported here . Aggregation has most often been characterized as the fraction of worms inside clusters , where individual worms can move in and out of clusters . Here we report an additional dynamic swarming phenotype in aggregating C . elegans that occurs on a timescale of hours . Here , swarming refers to the collective movement of a coherent group of worms across a bacterial lawn ( Figure 1A , Video 1 ) . Because of the long timescale , this behavior is not obvious from manual observations of worms on a plate , but becomes clear in time lapse videos ( Figure 1B and C , npr-1 panels ) . Even though N2 worms do not swarm in our experiments ( Figure 1B and C , N2 panels ) , they can swarm under appropriate conditions , such as when a clonal population has depleted almost all food ( Hodgkin and Barnes , 1991 ) or on unpalatable Pseudomonas fluorescens bacterial lawns ( personal communication from J . Hodgkin and G . M . Preston ) . Thus swarming in C . elegans does not require loss of npr-1 function in all environments . Dynamic swarming occurs with just 40 npr-1 mutants ( Figure 1B , top row ) , making it experimentally feasible to study . Usually a single npr-1 aggregate forms on the food patch and then moves around the lawn in a persistent but not necessarily directed manner ( Figure 1C , left; Figure 1—figure supplement 1 ) , at a steady speed ( Figure 1D ) . The onset of this collective movement appears to coincide with local food depletion , and continues until complete food depletion , at which time the cluster disperses . More than one moving cluster may co-exist , and occasionally a cluster may disperse and form elsewhere when it crosses its previous path ( Figure 1—figure supplement 1 ) , presumably due to local food depletion . The observed pattern of npr-1 cluster motion is reminiscent of a self-avoiding , persistent random walk ( i . e . not returning to areas that the worms have previously been where there is no food left ) . By contrast , after initially forming transient clusters on the lawn , N2 worms move radially outwards with no collective movement ( Figure 1C , right ) . Based on our observation that swarming appears to be driven by food depletion , we hypothesize the phenomenon may be a dynamic extension of the initial aggregation that occurs before depletion . To test this idea , we first sought to identify the mechanisms underlying aggregation . The presence of aggregates is clear in bright field images , but it is difficult to track individual animals in these strongly overlapping groups for quantitative behavioral analysis . We therefore labeled the pharynx of worms with green fluorescent protein ( GFP ) and used fluorescence imaging in order to minimize overlap between animals ( Video 2 ) , making it possible to track most individuals even when they are inside a dense cluster ( Figure 2A ) . We also labeled a small number of worms ( 1–3 animals out of 40 per experiment ) with a red fluorescent protein ( RFP ) -tagged body wall muscle marker instead of pharynx-GFP . These RFP-labeled worms were recorded on a separate channel during simultaneous two-color imaging ( Figure 2B ) , thus allowing both longer trajectories and the full posture to be obtained in a subset of animals . We wrote a custom module for Tierpsy Tracker ( Javer et al . , 2018 ) to segment light objects on a dark background and to identify the anterior end of the marked animals automatically , in order to extract trajectories and skeletons of multiple worms from our data ( Figure 2C ) . We first considered long-range chemotaxis driven by food or diffusible ascaroside pheromone signals as a potential behavioral mechanism . Chemotaxis towards food can likely be ignored as our experiments were performed on thin , even bacterial lawns , and worms are mostly on food during the aggregation phase of the experiments ( 99 . 7 ± 0 . 4% for npr-1 and 99 . 8 ± 0 . 3% for N2 , mean ±S . D . ) . Although ascarosides are important for processes such as mating and dauer formation in C . elegans ( Srinivasan et al . , 2008 ) , it is less clear whether long-range signaling via pheromones plays a role in aggregation ( de Bono et al . , 2002; Macosko et al . , 2009 ) . daf-22 ( m130 ) mutants do not produce ascarosides , but daf-22;npr-1 double mutants aggregate similarly to npr-1 single mutants ( Figure 3—figure supplement 1 ) , consistent with the observation that the hermaphrodite-attractive pheromone icas#3 is attractive to both N2 animals and npr-1 mutants ( Srinivasan et al . , 2012 ) and is thus unlikely to explain the difference in their propensity to aggregate . Moreover , attraction between moving objects is known to produce aggregation in active matter systems ( Redner et al . , 2013a ) , but it is not known whether this applies to worms . Short-range attraction between worms may exist in the form of adhesion mediated through a liquid film ( Gart et al . , 2011 ) , but we have no reason to believe this would differ between npr-1 and N2 strains . Having considered long-range food- or ascaroside-mediated attraction and short-range adhesion , we next focused on behavioral responses to nearby neighbors . While postural changes do not seem to be a main driver of aggregation as principal component analysis of lone versus in-cluster npr-1 worms revealed similar amplitudes in the posture modes ( Figure 3—figure supplement 2 ) , we found experimental evidence for density-dependence of both reversal rates and speed and that these differ between the two strains we studied . Reversals have been previously suggested as a behavior that may enable npr-1 worms to stay in aggregates ( Rogers et al . , 2006 ) . To avoid cluster definitions based on thresholding the distance between worms , we quantified individual worm behavior as a function of local density ( Figure 3A ) instead . Calculating the reversal rates relative to that of worms at low densities , we found that npr-1 mutants reverse more at increased neighbor densities , while N2 animals do not ( Figure 3B ) . Next we calculated the speed distributions of individual worms , binned by local neighbor density . We found that both strains slow down when surrounded by many other worms , but the shift is more pronounced for npr-1 animals . npr-1 worms move faster than N2 at low densities , showing a distinct peak at high speeds . As neighbor density increases , this high speed peak gradually becomes replaced by a peak at low speeds , so that the overall speed distribution for npr-1 resembles that of N2 at very high densities . Thus , npr-1 and N2 animals show different density-dependent changes in their respective speed profiles ( Figure 3C ) . Since the observed transition of the speed profiles could occur due to active behavioral changes as well as restricted movement in clusters , we also considered tracks of individual worms . Using body wall muscle-marked worms allowed us to obtain longer trajectories that could be joined for the duration of an entire video , including cluster entry and exit events . We compared the speed of these tracks with visual assessment of when a worm entered or exited a cluster based on the proximity to pharynx-labeled worms . We found that worms are able to move inside of clusters and observed that speed changes can occur prior to cluster entry and exit events ( Figure 3D , Video 3 and Video 4 ) . This change of speed is neither purely mechanical nor a deterministic response to a certain neighbor density , and suggests a mechanism in which worms probabilistically switch between different speeds . The differences in aggregation behavior between npr-1 and N2 are visually striking , but previous quantification has typically been limited to the fraction of animals in clusters . Using the tracked positions of pharynx-labeled worms ( Figure 4A ) , we calculated the pair-correlation function ( Figure 4B ) , commonly used to quantify aggregation in cellular and physical systems ( Gurry et al . , 2009 ) . We also computed a hierarchical clustering of worm positions ( Figure 4C ) , which is calculated from the same pairwise distances but emphasizes larger scale structure . Using both measures , we found that as a population , npr-1 animals show quantifiably higher levels of aggregation than N2 , especially at scales up to 1 mm ( pair-correlation ‘S1’ , Figure 4D ) and 2 mm ( hierarchical clustering ‘S2’ , Figure 4E ) . We also quantified aggregation using scalar spatial statistics , namely the average standard deviation ( ‘S3’ ) and kurtosis ( ‘S4’ ) of the distribution of positions . This confirms that the positions of npr-1 worms are less spread-out and more heavy-tailed than those of N2 ( Figure 4D ) . To test whether the individual behavioral differences measured between npr-1 and N2 worms are sufficient to give rise to the observed differences in aggregation , we constructed a phenomenological model of worm movement and interactions . The model is made up of self-propelled agents ( Figure 5A ) , and includes density-dependent interactions motivated by the experimental data , namely reversals at the edge of a cluster ( Figure 5B ) and a switch between movement at different speeds ( Figure 5C ) . As a model of collective behavior this differs from those commonly considered in the literature , such as the Vicsek model ( Vicsek et al . , 1995 ) and its many related variants ( Vicsek and Zafeiris , 2012; Yates et al . , 2011 ) . Such models typically feature attractive forces or align the direction of motion at ranges much longer than the size of the moving objects , and result in flocking or clustering with global alignment ( Figure 5D ) , which we do not observe in our experimental data . In contrast , our model needs to produce dynamic , disordered aggregates ( Figure 1B , Figure 2A and Video 2 ) , and should primarily rely on short-range interactions that are motivated by behaviors measured in our data . The density-dependence of the reversal rate and speed switching is implemented as follows: The rate of reversals increases linearly with density with slope r’ , which is a free parameter , and is thus given by rrev = r’ ρ . The reversal rate at zero density is zero as we ignored spontaneous reversals outside of clusters as these were only rarely observed under our experimental conditions ( see Appendix 1 for further discussion of the model construction ) . This parameterization of the reversal rate may be unbounded , but we can prevent unrealistically high reversal rates for a given maximum worm number by choosing our prior distribution of the parameter r’ . The rate of slowing down is similarly approximated as a linear function of density , with free parameter ks’ , and is given by kslow = ks0+ks’ ρ , where ks0 is the slowing rate at zero-density . The rate of speeding up is given by kfast = kf0 exp[-kf’ ρ] , where the exponential decay is chosen to ensure positivity of the rate , and kf0 is the rate at zero density . The rates of slowing down and speeding up at zero density ( ks0 , kf0 ) were obtained from published single-worm experimental data ( Javer et al . , 2018; Yemini et al . , 2013 ) . We initially ran a coarse parameter sweep , sampling uniformly in the two-dimensional parameter space associated with the density-dependence of reversals and speed switching . As a simplifying assumption , the density-dependence of the speeding-up and slowing-down rates was set equal ( k’s = k’f = k’ ) . The remaining parameters , r’ and k’ , were varied to explore the global model behavior . This demonstrates that our model can capture different aggregation phenotypes from solitary movement to aggregation ( Figure 5E ) by varying just two free parameters , and provides important general insights . Inspection of the model simulations shows that each behavior alone ( just reversals or slowing ) does not give the same level of aggregation as when both parameters are modulated ( Figure 5E ) , so that using both behavioral components proves important . Quantifying the aggregation and comparing it to the npr-1 experiment , however , highlights incomplete quantitative agreement with both the pair correlation function and hierarchical clustering distribution ( Figure 5F ) . Thus , we reasoned additional interactions may be required to match the experimentally observed behaviors . To explore improvements in clustering , we extended the model by an attractive taxis interaction . Attraction should intuitively improve clustering , but we knew from our model exploration that an attractive potential between bodies produces undesirable cluster shapes ( Figure 5D ) and reasoned that a long-range interaction may be unrealistic ( Figure 3—figure supplement 1 ) . Thus , we include taxis towards neighboring worms and model worm movement as an attractive persistent random walk . The taxis contribution to a worm’s motile force has an overall strength controlled by parameter ft , with multiple nearby neighbors contributing cumulatively , weighted by 1/r , where r is the distance to a neighboring worm . Neighboring worms beyond a cut-off distance equal to the length of a worm have no contribution . Thus , this taxis interaction is acting at a natural intermediate length scale of our system ( see Appendix 1 for details ) . The resulting extended model has four free parameters: density-dependent reversals ( r′ ) , speed-switching rates ( ks′ , kf′ ) and taxis ( ft ) . To find the parameter combinations that best describe each strain , as well as the uncertainty in the parameter values , we used an approximate Bayesian inference approach ( see Appendix 1 ) . To increase the computational efficiency of our inference pipeline , we excluded infeasible regions of parameter space to reduce the prior distribution of parameters that we need to sample from ( Figure 6—figure supplement 1 ) ( see Appendix 1 ) . We then selected the closest matching simulations from about 27 , 000 simulations for npr-1 and about 13 , 000 simulations for N2 , equally weighting all four summary statistics . Results from our extended model ( Figure 6A , Video 5 and Video 6 ) show markedly improved quantitative agreement with the experiments ( Figure 6B ) . The approximated posterior distributions of the parameters ( Figure 6C–D ) show the most likely values of the parameters for each strain , as well as the uncertainty associated with the individual and joint marginal parameter distributions . In particular , to achieve npr-1-like aggregation , the reversal ( r’ ) and taxis ( ft ) parameters need to be higher than for N2 , albeit not too high . The density-dependence of the slowing rate ( k’s ) is only subtly different between the two strains , while the dependence of the speeding up rate ( k’f ) is greater in npr-1 , but with broader uncertainty . To address whether all three behaviors ( reversals , speed changes , and taxis ) were necessary for aggregation we ran additional simulations: starting from the mean of the posterior distribution for npr-1 ( Figure 6C ) as a reference , we removed individual model components by setting the corresponding parameters to zero . These perturbed simulations show that removing speed switching or taxis from the model disrupts aggregation , while removing reversals reduces the overall quantitative agreement with experimental data ( Figure 6—figure supplement 2 , B–D ) . In some cases , removing individual model behaviors also produced correlations of velocity and orientation between neighbors that are different from what we measure in experiments ( Figure 6—figure supplement 3 ) . Thus , we conclude that we have identified sufficient behavioral components for aggregation , and that these are also necessary to quantitatively match aggregation in npr-1 mutants . Searching for evidence of taxis in the experimental tracking data , we calculated the correlation between worm velocity and the vector towards nearby worms , and found this correlation to be nearly zero in both experiments and simulations for all distances up to 2 mm ( Figure 6—figure supplement 3B1–2 ) , which is larger than the size of a typical worm cluster . This may not be intuitive , and we suspect the reason is twofold: ( a ) the taxis effect is only a small influence on the instantaneous direction of the movement of a worm , compared to persistence and noise; and ( b ) we only tracked the pharynx in our experiments , and reproduced this restriction in our analysis of simulations , but the whole body of the worm is likely giving relevant cues to any chemical or mechanical taxis . Our methodology that enables us to track inside worm clusters therefore brings with it the caveat that there is unseen worm density that affects any potential taxis behavior , but which remains undetectable in our tracking . Thus , our analysis shows that a taxis behavior similar to our simulations may be present in experiments , even if it is difficult to detect with correlation analysis . We compared the other inferred parameters with experimental measurements: The reversal rate shows a similar increase with density that is greater for npr-1 than N2 ( Figure 6—figure supplement 4B ) . The speed switching rates could only be compared indirectly by calculating the ratio of fraction of worms in fast vs . slow movement in experiments ( Figure 6—figure supplement 4C1 ) and model simulations ( Figure 6—figure supplement 4C2 ) . The disagreement may indicate that the exponential form of kf ( ρ ) is only a rough approximation . However , aggregation in the model is not sensitive to speed switching rates , as shown by the broad posterior distributions for the inferred parameters ( Figure 6C–D ) . Since we hypothesize that the swarming we observed at longer time scales may be explained as aggregation under food depletion conditions , we further extended the model to allow the local depletion of food . Food is initially distributed uniformly , and becomes depleted locally by worm feeding ( see Appendix 1 for details ) . Absence of food suppresses the switch to slow speeds , thus causing worms to speed up when food is locally depleted . As a result , we hypothesize that worm clusters begin to disperse but reform on nearby food , leading to sweeping . Selecting the parameter combination best matching the npr-1 strain ( Figure 6 ) and an appropriate food depletion rate ( chosen such that all food was depleted no faster than observed in experiments ) , the resulting simulation produced long-time dynamics qualitatively representative of the experimentally observed swarming ( Figure 7A–B , Video 7 ) . Worm clusters undergo a persistent but not necessarily directed random walk , can disperse and re-form elsewhere , and multiple clusters may co-exist , all of which we observe experimentally . Tracking the centroid of worms in our simulations , we find a comparable cluster speed as the median experimental value of 172 μm/min ( Figure 1D ) for a range of feeding rates ( Figure 7C ) ( feeding rate is an unknown parameter as our model only accounts for relative food concentration ) . Thus , the model indicates that dynamic swarming of npr-1 aggregates may be explained as an emergent phenomenon resulting from individual locomotion , and that the same behavioral mechanisms that produce the initial aggregates , when coupled with local food depletion , give rise to the observed swarming behavior . We have investigated the mechanisms of aggregation and swarming in C . elegans collective feeding using quantitative imaging and computational modeling . We show that while a combination of increased reversals upon leaving aggregates and a neighbor density-dependent increase in speed switching rates is sufficient to produce aggregation , the addition of taxis towards neighbors improves the quantitative agreement between simulations and experiments . Removing any one of the core behavioral mechanisms ( reversals , speed changes , taxis ) from our model either disrupts aggregation or otherwise reduces the quantitative agreement with experiments ( Figure 6—figure supplement 2–3 ) . The proposed taxis might be driven by a shallow O2 or CO2 gradient created by a worm cluster ( discussed further below ) , to additional chemical signals unaffected by daf-22 loss of function , or to another unknown mechanism . By extending the aggregation model to include food depletion , we show that the same behavioral mechanisms also underlie dynamic swarming in the hyper-social C . elegans strain , reminiscent of wild fires and other self-avoiding dynamics . We focused on identifying phenomenological behavioral components giving rise to aggregation , while remaining agnostic as to the sensory cues causing the behaviors . The density-dependent interactions could arise from local molecular signaling , or be mediated through contact-sensing , and the 1/r dependence of the taxis interaction is compatible with a diffusible , non-degrading factor ( such as CO2 , or O2 depletion; dependence would likely be different for a pheromone depending on its degradation rate ) . Given that aggregates break up when ambient O2 concentration is reduced to 7% ( Gray et al . , 2004 ) , the preferred concentration of npr-1 mutants , the most obvious candidate for the sensory cue guiding aggregation is O2 ( Rogers et al . , 2006 ) . A simple hypothesis would be that oxygen consumption by worms locally lowers O2 concentration to the 5–12% preferred by npr-1 mutants , promoting their aggregation . To support this , Rogers et al . ( 2006 ) report low O2 concentrations inside worm clusters . However , non-aggregating N2 worms also prefer O2 concentrations lower than atmospheric ( 5–15% ) ( Gray et al . , 2004 ) . Furthermore , a strong reduction of oxygen concentration inside an aggregate to near 7% is unlikely based on reaction-diffusion calculations: the diffusion of oxygen through worm tissue , or their oxygen consumption , would need to be several orders of magnitude different from estimated values to create O2 gradients as steep as reported by Rogers et al . ( Appendix 2—figure 1 ) . However , as worms have been reported to respond even to small changes in oxygen concentration ( McGrath et al . , 2009 ) , aggregation may still be mediated through a shallower local oxygen gradient . In this scenario , high ambient O2 concentration serves as a permissive signal for aggregation and a shallow oxygen gradient induces worms to stay inside aggregates . Our agent-based simulations are entirely compatible with this picture . Further experiments would be required to test the hypothesis that oxygen is playing such a dual role . One possibility would be to introduce mutations leading to aerobic metabolism deficiencies into npr-1 mutants . Such mutants would still be able to sense ambient oxygen , but are expected to produce an even weaker oxygen gradient in an aggregate . The resulting phenotype could then be compared quantitatively to model predictions , for example with reduced taxis and/or modified rates of density-dependent reversal and speed switching . Additionally , one may seek evidence for the ability of worms to sense a shallow oxygen gradient by repeating the gas-phase aerotaxis experiment described in Gray et al . ( 2004 ) , but with a much smaller gradient ( 19–21% ) in the light of our new calculations , to see if worms can sense and move towards environments where oxygen levels are only slightly below ambient concentrations . Further work quantifying the behavior of individual worms at different oxygen concentrations , such as during oxygen-shift experiments inside flow chambers where single animals experience acute switches between 21% and 19% oxygen , may also help to distinguish oxygen as a direct cue or part of the ‘sensory triggers that can initiate social behavior by activating chemotaxis or mechanotaxis’ ( Gray et al . , 2004 ) . The model of worm movement and interactions presented here was chosen for a balance of simplicity and realism , and is not necessarily unique . Our model comprises a persistent random walk of chain-like worms , which were loosely inspired by work on bacterial systems ( Balagam and Igoshin , 2015 ) . We have adopted Bayesian parameter inference to capture the uncertainty in our parameter estimates , and to enable flexible extension to additional experimental data or comparison of different models in future work . An alternative approach is to be entirely data-driven in the construction of the model and compute interactions between worms directly based on their tracked positions at every time step , as has been done in collective behavior of Myxococcus xanthus ( Cotter et al . , 2017; Zhang et al . , 2018 ) . This approach may require higher worm numbers and improved tracking , to ensure comparably large statistical sample sizes with bacterial studies . We have used experimental data to inform our modeling framework where appropriate ( size , shape , speed of agents , and reversal and speed change rates at zero density ) , and verified that the aggregation outcome is robust and quantitatively similar to experimental results regardless of the amount of noise in the persistent random walk ( Figure 6—figure supplement 2E–G ) , or the presence of undulations in agent movement ( Figure 6—figure supplement 2H ) . We have further verified that aggregation still occurs with shorter simulated worms ( and fewer nodes per worm ) , given they are long enough to detect a contact difference between head and tail when exiting a cluster , which is required to initiate reversals ( Figure 6—figure supplement 5A ) . Lastly , in the model presented here , we have allowed for overlap between worms to reflect a degree of overlap in clusters when worms can crawl over each other . With volume exclusion our model still produces aggregation , although the clusters are less dense and more extended ( Figure 6—figure supplement 5B ) . One advantage of using C . elegans to study animal collective behavior is the opportunity to experimentally control and perturb the system . It should be possible to experimentally modify the key behavioral parameters identified in this paper with mutations or acute stimulus delivery in order to test our model . For example , one can introduce a reversal phenotype with unc-4 mutations , or alter the speed switching rates with mutations that affect the roaming-dwelling transition . Controlled stimulus delivery has already been used in previous oxygen-shift experiments . The resultant experimental outcomes may then be compared to theoretical predictions . Thus , there are ample opportunities for future studies to further integrate experimental and theoretical methods in the study of C . elegans collective behavior . Despite its extensive study in the lab , it is still uncertain whether aggregation and swarming have a function in the wild . Aggregation may serve to protect C . elegans from desiccation or UV radiation associated with the surface environment ( Busch and Olofsson , 2012 ) . C . elegans swarming on unpalatable bacteria may also facilitate predation , perhaps through the collective action of secreted molecules that overcome bacterial defenses ( personal communication from J . Hodgkin and G . M . Preston ) in a manner similar to the well-described cooperative predation strategy used by Myxobacteria xanthus ( Muñoz-Dorado et al . , 2016; Pérez et al . , 2016 ) . Moreover , social versus solitary foraging strategies may confer selective advantages in different food abundance , food quality , and population density environments ( de Bono and Bargmann , 1998 ) . The observation that aggregating strains are less fit in laboratory conditions ( Andersen et al . , 2014 ) suggested that social feeding is not an efficient strategy at least in abundant food conditions . However , the observed fitness difference between aggregating and non-aggregating strains is actually dissociable from the feeding strategy in the lab ( Zhao et al . , 2018 ) , leaving the question unresolved . Furthermore , in other systems , social feeding can increase fitness in natural environments via improved food detection and intake ( Cvikel et al . , 2015; Li et al . , 2014; Snijders et al . , 2018 ) . It would be time consuming to experimentally measure the feeding efficiency of different behavioral strategies for a wide range of food patch sizes , distributions , and qualities . The agent-based model used in this study presents an opportunity to use a complementary approach to finding conditions that may favor social feeding . C . elegans bridges the gap between the commonly studied micro- and macro-scales , and finding the behavioral rules underlying this mesoscale system allows us to consider principles governing collective behavior across scales . Indeed , key behavioral rules identified here for C . elegans aggregation have been observed at other scales . Spontaneous reversals have been implicated in bacterial aggregation at the microscale ( Mercier and Mignot , 2016; Starruß et al . , 2012; Thutupalli et al . , 2015 ) . By contrast , aggregating worms reverse mainly in response to leaving a cluster rather than spontaneously , thus requiring more complex sensory processing and behavioral response than seen in bacterial systems . Furthermore , changes in movement speed are a common feature in motility-induced phase transitions ( Großmann et al . , 2016; Redner et al . , 2013b; Abaurrea Velasco et al . , 2018 ) . The emergent phenomena observed in models of interacting particles generally range from diffusion-limited aggregation to jamming at high volume fractions to flocking of self-propelled rods through volume exclusion ( in two-dimensions ) . In contrast , aggregation in C . elegans occurs at much lower numbers of objects ( tens of worms ) and lower densities ( area fraction of 4–6% ) than typically studied in this field ( thousands of objects at area fractions of 20–80% ) , and the density dependence of motility changes again emphasizes the role of more complex sensing and behavioral modulations common in macroscale animal groups such as fish shoals ( Ward et al . , 2011 ) . Thus , collective behavior of C . elegans at the mesoscale indeed draws from both ends of the size scale and complexity spectrum , linking the physical mechanisms familiar from microscopic cellular and active matter systems with the behavioral repertoire of larger multicellular organisms . Our approach of decomposing aggregation into component behaviors through modeling may also have applications in quantitative genetics beyond the scope of our current study . While hyper-social npr-1 mutants and hypo-social N2 worms show phenotypic extremes , wild isolates of C . elegans aggregate to different degrees ( de Bono and Bargmann , 1998 ) . Previous work has shown that even a very small increase in the phenotypic dimensionality ( from one to two ) can reveal independent behavior-modifying loci ( Bendesky et al . , 2012 ) . Thus inferring model parameters for data from multiple wild C . elegans strains would produce behavioral parameterizations that might serve as a powerful set of traits for finding further behavior-modifying loci . C . elegans strains used in this study are listed in Key Resources Table above . All animals were grown on E . coli OP50 at 20°C as mixed-stage cultures and maintained as described ( Brenner , 1974 ) . All animals used in imaging experiments were synchronized young adults obtained by bleaching gravid hermaphrodites grown on E . coli OP50 under uncrowded and unstarved conditions , allowing isolated eggs to hatch and enter L1 diapause on unseeded plates overnight , and re-feeding starved L1’s for 65–72 hr on OP50 . The strain used here ( Figure 1A and Video 1 ) is DA609 . On imaging day , synchronized adults were collected and washed in M9 buffer twice before several hundred animals were transferred to a seeded 90 mm NGM plate using a glass pipette . After M9 is absorbed into the media , ten-hour time-lapse recordings were taken with a Dino-Lite camera ( AM-7013MT ) at room temperature ( 20°C ) using the DinoCapture 2 . 0 software ( v1 . 5 . 3 . c ) for maximal field of view . Two independent replicates were performed . Step-by-step protocol is available at dx . doi . org/10 . 17504/protocols . io . vybe7sn . All recordings from this dataset are listed in Supplementary file 2 . The strains used here ( Figure 1B ) are DA609 and N2 . Prior to collecting the full dataset , a single batch of OP50 was grown overnight , diluted to OD600 = 0 . 75 , aliquoted for use on each imaging day , and stored at 4°C until use . Imaging plates were 35 mm Petri dishes containing 3 . 5 mL low peptone ( 0 . 013% Difco Bacto ) NGM agar ( 2% Bio/Agar , BioGene ) to limit bacteria growth . A separate batch of plates was poured exactly seven days before each imaging day , stored at 4°C , and dried at 37°C overnight with the agar side down before imaging . The center of an imaging plate was seeded with a single 20 μL spot of cold diluted OP50 one to three hours before imaging . The overnight plate drying step allowed the bacteria to quickly dry atop the media in order to achieve a more uniform lawn by minimizing the ‘coffee ring’ effect that would thicken the circular edge of the bacterial lawn . For each imaging day , synchronized young adults were collected and washed in M9 buffer twice before 40 animals were transferred to a seeded imaging plate using a glass pipette . Imaging commenced immediately following animal transfer in a liquid drop , on a custom-built six-camera rig equipped with Dalsa Genie cameras ( G2-GM10-T2041 ) . Seven-hour recordings with red illumination ( 630 nm LED illumination , CCS Inc ) were taken at 25 Hz using Gecko software ( v2 . 0 . 3 . 1 ) , whilst the rig maintained the imaging plates at 20°C throughout the recording durations . Images were segmented in real time by the Gecko software . The recordings were manually truncated post-acquisition to retain aggregation and swarming dynamics only . The start time was defined as the moment when the liquid dried and the all the worms crawled out from the initial location of the drop , and the end time was when the food was depleted and worms dispersed with increased crawling speed . Twelve independent replicates were performed for each strain . Step-by-step protocol is available at dx . doi . org/10 . 17504/protocols . io . vyhe7t6 . All recordings from this dataset are listed in Supplementary file 2 . The experiments here ( Figure 1—figure supplement 1 ) are identical to those in the bright field standard swarming imaging , except for two differences . First , the imaging plates were seeded with a 75 μL spot of diluted OP50 ( OD600 = 0 . 38 ) and allowed to inoculate overnight at room temperature before being used for imaging the next day . Second , recordings were taken over 20 hr instead of seven . Eight independent replicates were performed for each strain . Step-by-step protocol is available at dx . doi . org/10 . 17504/protocols . io . vyie7ue . All recordings from this dataset are listed in Supplementary file 2 . The strains used here ( Figure 3—figure supplement 1 ) are DA609 , N2 , DR476 , and AX994 . Bacteria aliquots and imaging plates were prepared as in the bright field standard swarming imaging assay . For each imaging day , synchronized young adults were collected and washed in M9 buffer twice before 40 animals were transferred to a seeded imaging plate using a glass pipette . After M9 was absorbed into the media following worm transfer in liquid , imaging plates containing the animals were subjected to a gentle vibration at 600 rpm for 10 s on a Vortex Genie two shaker ( Scientific Industries ) to disperse animals and synchronize aggregation start across replicates . Imaging commenced 20 s after the vibration finish , using the same rig set-up as swarming imaging above , except one-hour recordings were taken . Images were segmented in real time by the Gecko software . At least eight independent replicates were performed for each strain . Automated animal tracking was performed post-acquisition using Tierpsy Tracker software ( http://ver228 . github . io/tierpsy-tracker/ , v1 . 3 ) , which we developed in-house ( Javer et al . , 2018 ) . Images with were tracked with customized parameters to create centroid trajectories , 49-point worm skeletons , and a battery of features . Step-by-step protocol is available at dx . doi . org/10 . 17504/protocols . io . vzje74n . All recordings from this dataset are listed in Supplementary file 2 . The strains used here ( Figure 2 , Videos 2–4 ) are OMG2 , OMG10 , OMG19 , and OMG24 . One-color imaging consisted of pharynx-GFP labeled worms only , whereas two-color imaging also included a small number of body wall muscle-RFP labeled worms that were recorded simultaneously on a separate channel ( thus readily segmented from the rest of the worms ) . The latter was necessary to follow individuals over a long period of time , particularly while inside a cluster , as frequent pharynx collisions inside clusters lead to lost individual identities and broken trajectories . For two-color imaging , animals with different fluorescent markers were mixed in desired proportion ( 1–3 red animals out of 40 per experiment ) during the washing stage before being transferred together for imaging . The data collection paradigm was identical to the bright field pheromone imaging assay in terms of bacteria aliquots , imaging plate preparation , and vibration implementation following animal transfer . The difference is that image acquisition was performed on a DMI6000 inverted microscope ( Leica ) equipped with a 1 . 25x PL Fluotar objective ( Leica ) , a TwinCam LS image splitter ( Cairn ) with a dichroic cube ( Cairn ) , and two Zyla 5 . 5 cameras ( Andor ) to enable simultaneous green-red imaging with maximal field of view . One-hour recordings were taken with constant blue ( 470 nm , 0 . 8A ) and green ( cool white , 1 . 4A ) OptoLED illumination ( Cairn ) , and images were acquired with 100 ms exposure at 9 Hz using Andor Solis software ( v4 . 29 . 30005 . 0 ) . The microscopy room was maintained at 21°C throughout the recording durations . Ten or more independent replicates were performed for each strain . We were able to reproduce stereotyped aggregation dynamics across replicates under our experimental paradigm ( Figure 1—figure supplement 2 ) . Image segmentation and automated animal tracking was performed post-acquisition using Tierpsy Tracker software ( v1 . 3 ) with customized parameters , to create centroid trajectories , obtain two-point skeleton from pharynx-labeled individuals and 49-point midline skeletons from body wall muscle-marked ones , and extract various features . For body wall muscle-marked animals , trajectories were manually joined where broken due to tracking errors . The code for tracking data analysis is available at https://github . com/ljschumacher/wormTrackingAnalysis ( Schumacher et al . , 2019; copy archived at https://github . com/elifesciences-publications/wormTrackingAnalysis ) . Tracked blobs were filtered for minimum fluorescence intensity and maximum area , to exclude any larvae and tracking artifacts , respectively , which appeared on the occasional plate . Local worm densities around each individual were calculated using k-nearest neighbor density estimation , where the density is k divided by the area of a circle encompassing the k-th nearest neighbor . We chose k=6≈ √N and verified based on visual assessment that the overall distribution of local densities changes very little with increasing k . Reversals were detected based on a change of sign of speed from positive to negative , which was calculated from the dot-product of the skeleton vector ( of the pharynx ) and the velocity vector , and smoothed with a moving average over half a second . We only counted reversals that were at least 50 µm in length , and that moved at least half a pixel per frame before and after the reversal . Reversal events thus detected where binned by their local density . For each density bin , reversal rate was estimated as the number of events divided by the time spent in forward motion for that bin . The variability was estimated using a subsampling bootstrap: the reversal rate was estimated 100 times , sampling worm-frames with replacement , and estimating mean and standard deviation . Speed profiles were generated by binning the measured speed values for local density , and then creating a histogram of speed values for each density bin . Summary statistics of aggregation , such as pair-correlation and hierarchical clustering , where calculated as described in Appendix 1 .

Anyone who has watched a flock of birds maneuver through the sky has probably wondered how so many animals coordinate their movements . Often , these seemingly complex group behaviors can be explained by a few simple rules that govern the behavior of the individuals in the group . For example , if each bird flies and reacts to its neighbors in a certain way , the whole flock’s flight pattern results from these individual choices . Computer simulations can help researchers to test how individual behaviors contribute to coordinated group movements . Ding , Schumacher et al . have now used a simulation to uncover the rules that control the behavior of small worms called Caenorhabditis elegans , which form large groups while feeding on bacteria . To gather the data needed to form the computer model , Ding , Schumacher et al . genetically engineered C . elegans worms to produce fluorescent proteins . The fluorescence allows the movements of the worms to be monitored automatically in time-lapse movies made from a series of microscope images . The movies show that worm clusters move together over a patch of food , consuming it as they go . As the food disappears , the whole worm cluster moves to a new area in search of more food . The computer simulation that Ding , Schumacher et al . developed to recreate how the clusters move revealed that individual worms in the group interact according to three rules . Firstly , worms slow down when they have more neighbors . Secondly , when a worm leaves its cluster , it will reverse to rejoin the group . And finally , worms will move towards areas with more neighbors . It is still not known why the C . elegans worms form clusters , but understanding how the individuals in the group interact could help future studies to uncover this reason . Many other organisms benefit from forming similar groups , from single celled bacteria to animals such as birds and fish . The results presented by Ding , Schumacher et al . will therefore help researchers to consider whether there are universal rules that control group behavior .

2019

Shared behavioral mechanisms underlie C. elegans aggregation and swarming

The quality of visual information that is available to an animal is limited by the size of its eyes . Differences in eye size can be observed even between closely related individuals , yet we understand little about how this affects vision . Insects are good models for exploring the effects of size on visual systems because many insect species exhibit size polymorphism . Previous work has been limited by difficulties in determining the 3D structure of eyes . We have developed a novel method based on x-ray microtomography to measure the 3D structure of insect eyes and to calculate predictions of their visual capabilities . We used our method to investigate visual allometry in the bumblebee Bombus terrestris and found that size affects specific aspects of vision , including binocular overlap , optical sensitivity , and dorsofrontal visual resolution . This reveals that differential scaling between eye areas provides flexibility that improves the visual capabilities of larger bumblebees . What an animal can see within its environment is restricted by its visual field , or the total angular region of the world from which light can be absorbed by its photoreceptors . To detect specific objects within this visual field , the eyes need spatial resolution , which is achieved through ( and limited by ) the arrangement of individual receptors that sample the spatial distribution of light ( Land and Nilsson , 2012 ) . Having an eye with spatial resolution allows an animal to detect differences in the intensity of the light reaching it from different directions , and this information is crucial for the myriad of visually guided behaviours exhibited by different species ( Cronin et al . , 2014 ) . Over a given finite area , an eye cannot maximise resolution without sacrificing sensitivity ( the amount of light captured ) ( Land , 1997 ) – increased resolution requires light to be sampled from a decreased region of space , necessarily reducing sensitivity ( Snyder , 1979 ) . As a result , the relative density and optical properties of receptors often vary topologically across an eye , creating variations in visual resolution and optical sensitivity that ‘fine-tune’ the capabilities in certain regions of the visual field . We are familiar with this from our own eyes – our fovea provides high-resolution vision over a small region of space , while our peripheral vision is blurred but views a much larger portion of the world . The eyes of other species have also evolved specializations that enable them to acquire critical information from important regions of the world , such as elongated regions of acute vision for detecting the horizon ( Dahmen , 1991 ) and ‘bright zones’ of high optical sensitivity for discriminating passing prey or potential mates against a bright background ( Straw et al . , 2006 ) . Specialized areas represent a local investment in improving a specific visual capability that is related to an animal’s behavioural and ecological requirements . Determining the topology of visual capabilities across an eye’s entire field of view ( FOV ) can provide important insights into the visually guided behaviours and environment of its owner ( Moore et al . , 2017 ) – for example , the topology of facet size ( either with or without a region of enlarged facets ) on the eyes of male bumblebees indicates their species-preferred mating strategy ( perching or patroling , respectively ) ( Streinzer and Spaethe , 2014 ) . Across a wide range of animal groups , bigger individuals generally have absolutely larger eyes , although eyes do not typically grow linearly with body size but , rather , become proportionally smaller in larger animals ( Jander and Jander , 2002; Howland et al . , 2004 ) , even within a species ( Perl and Niven , 2016a ) . Increasing eye size allows improvements in visual quality; in a bigger eye , resolution can be increased by adding receptors and optical sensitivity can be increased by enlarging the receptor size . The relationship between the growth of any trait and an animal’s total body size is conventionally modeled using a power function ( Y = bxα ) to relate a trait ( Y ) to a measure of body size ( x ) . This provides two parameters that describe allometry , the scaling exponent ( α ) and the initial growth index ( b ) Huxley and Teissier ( 1936 ) , where anatomical features that increase in size at a slower rate than the body are represented by an exponent less than 1 . Scaling exponents for eye size within invertebrate groups are usually higher than those of vertebrates ( which average 0 . 6 ) ( Howland et al . , 2004 ) and even include an unusual positive allometry rate in stingless Meliponine bees ( Streinzer et al . , 2016 ) . Allometry studies within hymenopteran species – in which body size can vary substantially between conspecifics – have shown that , although larger individuals of several ant species primarily invest in increasing their total facet number ( Klotz et al . , 1992; Zollikofer et al . , 1995; Schwarz et al . , 2011 ) , other ants ( Baker and Ma , 2006; Perl and Niven , 2016a ) and bumblebees ( Kapustjanskij et al . , 2007 ) increase both the number of facets and their size . Allometry has even been shown to vary across wood ant eyes , in which differential scaling exponents of facet size – leading to differences in visual capabilities – are found between eye areas ( Perl and Niven , 2016b ) . These insect species all possess apposition compound eyes ( in which each lens focuses light onto an individual receptor ) that , to avoid losing sensitivity , must grow in proportion to the square of a resolution improvement , as both the size and number of lenses must be increased ( Land and Nilsson , 2012 ) . Given that homogeneously increasing the size of a compound eye provides a relatively small improvement in its overall visual capability , we hypothesise that compound eyes are likely to scale non-uniformly , such that the majority of a larger eye will be invested in improving vision in a small portion of the FOV . To test whether increasing compound eye size does indeed lead to the development or improvement of specialised visual regions in larger individuals , it is necessary to link the allometry of eye properties to the visual capabilities they provide . Visual resolution in compound eyes is often estimated by dividing an assumed hemispherical FOV ( Land , 1997 ) by the number of facets , leading directly to the conclusion that resolution has the same ( although negative ) scaling exponent as facet number ( Jander and Jander , 2002 ) . This assumption is not supported , however , by direct measurements of inter-ommatidial ( IO ) angle ( a measure of local visual resolution ) . For instance , IO angles from both desert ant and fruit fly eyes have absolute scaling exponents ( −0 . 40 and −0 . 21 , respectively ) that are lower than those for the facet number ( 0 . 75 and 0 . 58 ) ( Zollikofer et al . , 1995; Currea et al . , 2018 ) . In addition , the scaling exponent of IO angle was found to vary between different regions of Orange Sulphur butterfly eyes , whereas the exponent of facet diameter remained relatively constant across the eye ( Merry et al . , 2006 ) . A complication when comparing corneal topology between different compound eyes is that eye shape can vary substantially , both between groups ( for example , butterflies have nearly hemispherical eyes ( Rutowski , 2000 ) , whereas flatter oval eyes are common in hymenopterans ( Jander and Jander , 2002 ) ) and within groups ( for example , male and female honeybees have drastically different eye shapes ( Streinzer et al . , 2013 ) ) . In the absence of a common reference frame , topologies based on general descriptions of eye shapes ( for example , the well-known dorsal rim area ( Labhart and Meyer , 1999 ) ) are not necessarily suitable for comparing visual capabilities between or within species , because corresponding anatomical areas may ultimately have different fields of view . Linking points on the eye to their projection into the visual world – and defining the visual capabilities that they provide – is necessary not only for comparing vision between species but also for gaining an understanding of how an eye samples information from the environment and how this influences the control of visually guided behaviours , such as foraging , predation , and mating . We have begun to explore the effect of size on visual capacity by comparing the topology across the entire eyes of individuals of the size-polymorphic bumblebee Bombus terrestris . The size differences within this species – which can equal an order of magnitude in mass ( Goulson , 2003 ) – influence several of its visually guided behaviours ( Spaethe and Weidenmüller , 2002; Spaethe and Chittka , 2003 ) . Yet , little is known about how optical sensitivity and visual resolution across a bumblebee’s entire visual field are affected by its eye size . To test our hypothesis that the differential scaling of compound eyes will primarily improve the visual capabilities in only a small region of the visual field , we developed a novel method based on constructing 3D models of apposition compound eyes that were imaged with x-ray microtomography ( microCT , Figure 1A ) ( Baird and Taylor , 2017 ) . These models allowed us to approximate the FOV of an eye from its corneal projection ( CP ) ( Figure 1E ) , over which the facet topology was mapped ( Figure 1C ) . Using this method , we directly calculated the inter-facet ( IF ) angle ( as an approximation of IO angle ) across the world , as well as the projected topologies of eye properties affecting optical sensitivity – namely , the facet diameter and the retinal and lens thicknesses ( Figure 1D ) . Crucially , this technique allowed us to place eye properties and visual capabilities from different eyes into common , world-referenced coordinates . We utilized our technique to investigate how size polymorphism influences B . terrestris vision by calculating the topology of eye properties and the visual capabilities of six worker bees varying in linear body size by ×2 . 8 and in eye volume ( EV ) by ×3 . 9 . As a reference , we also analyzed the eyes of European honeybees ( Apis mellifera ) , a species that has highly consistent worker size . Unlike those of B . terrestris , the eyes of honeybees have been the subject of extensive behavioural , anatomical , and physiological analyzes , and provide important data against which the results obtained from our method can be critically compared . Moreover , we identified key features in the allometry of the visual topology of bumblebees that may represent general rules for how compound eyes scale with size . The CP increases with bumblebee eye size ( Figure 2C ) , but it appears that this increase does not result from simply enlarging the CP in all directions ( Figure 4A , B ) . The dorsolateral limit of the CP of each eye is relatively consistent ( between 30° to 60° elevation ( el . ) and −90° to −60° azimuth ( az . - Azimuth ) , Figure 4B ) , while bigger bees appear to enlarge their CP dorsofrontally ( between 0° to 90° el . and −15° to 75° ( az . - Azimuth ) ) and to a lesser extent ventrolaterally ( between 0° to −60° el . and −105° to −60° ( az . - Azimuth ) ) . When the CP of each bee’s right eye is also considered , it is apparent that all bees have regions of binocular CP overlap ( Figure 4C ) that increase in angular area with eye size ( Figure 2C ) . Increasing binocularity is primarily observed dorsofrontally ( between 30° to 90° el . and −75° to 75° ( az . - Azimuth ) , Figure 4C ) , but a wedge of binocularity is also observed facing directly forwards ( between −30° to 30° el . and −15° to 15° ( az . - Azimuth ) ) . Given the shapes of these areas , these appear to be two distinct binocular regions; they are separated on the smallest bee and merge as the binocular field increases in angular size in bigger bees . The CP is a spatial , but binary , representation of each bee’s field of vision across which the eye properties and visual capabilities vary topologically . Optical sensitivity is influenced by facet diameter and rhabdom length ( and also by acceptance angle ) ( Warrant and Nilsson , 1998 ) and these properties increase with eye size ( Figure 3C and 5A ) . This suggests that larger bees have more sensitive ommatidia than smaller bees if the acceptance angles of receptors are assumed to be equivalent to the calculated IF angles and the retinal thickness is used as a proxy for rhabdom length ( Figure 5—figure supplement 1 ) . Taking into consideration the similar eye-wide averages for IF angle ( Figure 3A ) , an increase in retinal thickness and facet diameters suggests that bigger bees invest in improving their sensitivity rather than their visual resolution . There is , however , substantial variation in the histograms of the measured variables for each eye ( Figure 3A , C and 5A ) , so it will be interesting to investigate the topology of how these are projected into the visual field . The smallest IF angles within each bee’s CP are observed in a laterally positioned vertical band running from −45° until 60° el . at approximately −60° ( az . - Azimuth ) , whereas the greatest IF angles are observed at the posterior and rightmost dorsal limits of the visual field ( Figure 3B ) . All Bombus have a similar average IF angle profile across elevation ( average ~1 . 5° between −30° to 30° el . , Figure 3Bi ) . When averaged across azimuth , it is apparent that larger eyes have greater frontal resolution ( between −45° to 15° ( az . - Azimuth ) , Figure 3Bii ) , although the lowest azimuthal average IF angle is not directed frontally , but rather at ~−60° for all bees . A distinctly different topology is present when projecting the facet diameters into each bee’s CP ( Figure 3D ) . Larger Bombus individuals have larger facets ( Figure 3C ) , and each bee’s facet diameters ( averaged across elevation ) are greatest ventrally ( <0° el . , Figure 3Di ) and smallest dorsally ( >45° el . ) . The facet diameters of each bee are generally similar within these elevation ranges , but they are connected by a transitory range as diameters decrease from an elevation of 0° to 45° . When averaging facet diameter across azimuth , each Bombus has its largest facets facing laterally ( <60° ( az . - Azimuth ) , Figure 3Dii ) . Retinal thickness is projected into visual space with a similar , although less consistent , topology as facet diameter; average thickness generally increases towards the ventral and lateral CPs ( Figure 5B ) . Interestingly , lens thickness has a different topology to the other properties that we have described , as it peaks in the frontal visual field ( Figure 3; Figure 3—figure supplement 2Bii ) but is also correlated with facet diameter ( r = 0 . 65 , Figure 3—figure supplement 5 ) . After examining the projection of these variables into visual space , it initially appears that bumblebee eye properties maintain a similar topology that essentially scales with eye size ( facet diameter ( Figure 3D ) , radius of curvature ( Figure 3—figure supplement 1B ) , retinal thickness ( Figure 5B ) , lens thickness ( Figure 3—figure supplement 2B ) , and CC thickness ( Figure 3—figure supplement 3B ) . Although the IF angle is a function of local facet diameter and radius of curvature , local variations in the IF angles ( Figure 3B ) evidently arise from subtle changes in eye properties . Because the projected topologies varied locally in shape ( e . g . IF angle , Figure 3B ) , we calculated maps of the local scaling exponents for eye properties and visual capabilities in world-referenced coordinates to examine the spatial variation in the allometry of these characteristics . As identified from the IF angle profiles ( Figure 3B ) , scaling exponent maps of IF angle also show that larger bees had improved resolution in their frontal and dorsofrontal CPs ( between −15° to 60° el . and −30° to 10° ( az . - Azimuth ) , Figure 6A ) . However , as the facet diameter maintains an almost uniform exponent across the visual field ( Figure 6B ) , local variations in the scaling rate of the radius of curvature ( Figure 6—figure supplement 1C ) must cause differences in the IF angle exponent across the eye . In larger bumblebees , changes in the eye’s local radius , and consequently its shape , are the primary determinant of changes in IF angle and , indeed , local IF angle has a strong negative correlation to the radius of curvature ( r = −0 . 84 ) but is not correlated to facet diameter ( r = −0 . 07 , Figure 3—figure supplement 5 ) . Both retinal thickness and facet diameter contribute to sensitivity and have positive scaling exponents , although the local exponent of retinal thickness shows substantially more variation ( Figure 6C ) : it is highest in the dorsal hemisphere and lowest in the frontal region associated with binocularity ( Figure 4C ) . Arguably , maps of scaling exponents provide a clearer indication than the actual topologies themselves of where and how the topology of bumblebee eye properties changes as a function of eye size . Until now , we have described eye properties and their resultant effects on visual capabilities to investigate how bumblebees invest in vision as their eye size increases . The eye parameter ( P ) is an additional metric ( calculated as the local IF angle multiplied by the facet diameter ) ( Snyder , 1979 ) that provides an indication of whether facets are optimized for resolution ( where P approaches 0 . 29 if the resolution of the eye is limited by the diffraction of green light ) or sensitivity ( where P may be one or higher ) . Although the average eye parameter increases with eye size ( Figure 3—figure supplement 4A; we assume IO equals IF angle when calculating P ) , we found that all bumblebees had a similar average eye parameter profile between −30° and 10° ( az . - Azimuth ) ( Figure 3—figure supplement 4Bii ) . Scaling exponent maps showed that the eye parameter of larger bees was slightly reduced in a region similar to that in which their IF angles decreased , but increased in the periphery of the CP ( Figure 6 , D ) . This indicates that the dorsofrontally improved IF angles correspond to a region of improved optical resolution in larger bees , which strongly suggests that this region of the eye represents an acute zone optimized for high visual resolution ( Land , 1997 ) . Honeybees have slightly larger eye areas and volumes relative to their ITW than bumblebees ( Figure 2A ) . When compared on the basis of their EV , the average values for honeybee eyes are generally similar to those for medium-sized Bombus ( in terms of IF angle ( Figure 3A ) , facet diameter ( Figure 3C ) , radius of curvature ( Figure 3—figure supplement 1A ) , rhabdom length ( Figure 5A ) , and eye parameter ( Figure 3—figure supplement 4A ) ) . However , the distinguishing characteristics of honeybee eyes are that their values for corneal lens thickness are distinctly lower than those for all bumblebees ( Figure 3—figure supplement 2A ) , and that the extent of their binocular overlap is also smaller than that for to medium-sized Bombus ( Figure 2C ) . The CPs of honeybees were also shifted dorsoposteriorly relative to those of bumblebees ( Figure 4A , B ) , while their binocular CP overlap is limited to their dorsal visual field and does not extend frontally ( Figure 4C ) . After accounting for the differences in their visual fields , the projected topologies of the honeybees are similar to those of medium-sized bumblebees but with two differences: first , honeybees have an obvious increase in retinal thickness in their lateral visual field ( Figure 5Bii , −120° to −75° ( az . - Azimuth ) ) and , second , they have relatively small IF angles in their frontal visual field ( Figure 3Bii , −30° to 30° ( az . - Azimuth ) ) , which is the approximate region where larger bumblebees also increase their visual acuity . Anatomical studies of insect eyes have typically been based on measurements of 2D representations – either from imaging histological cross-sections or from flattened replicas of the eye surface ( Ribi et al . , 1989 ) . These techniques do allow optical dimensions to be measured from the eye’s morphology , but they lose the 3D location of these measurements relative to the remainder of the eye and head . Some studies on compound eyes have also measured 2D approximations of 3D shapes , such as the projected surface area of a compound eye from a given viewpoint ( Spaethe and Chittka , 2003; Kapustjanskij et al . , 2007; Perl and Niven , 2016a ) . However , the projected area of a 3D object depends upon on its size , shape , and on the direction of the observer; for example , a hemisphere viewed along its midline appears as a circle with a projected area that is 50% of the true area . Indeed , if the projected area of compound eyes in this study are calculated from orthographic projections along each cornea’s principle axis , the true surface area is consistently underestimated by 22% . Unfortunately measurements of projected area are not always clearly distinguished from true surface area in the literature . The method that we developed addresses the limitations of these techniques by maintaining the 3D structure of the eye relative to the head and allowing areas and volumes to be measured directly . As an alternative to microCT imaging , digital image registration techniques can also be used to reconstruct a volume from serial sections through an insect eye ( Hung and Ibbotson , 2014 ) . Although a volume reconstructed from sections allows measurements to be made with histological resolution , section distortions and alignment errors may limit the accuracy of calculations that rely on the 3D structure . The radius of curvature of a corneal transect can also be measured from a section ( Schwarz et al . , 2011 ) or a micrograph of a whole eye ( Bergman and Rutowski , 2016 ) and can be combined with a measurement of facet diameter to calculate an approximation of IO angle that is equivalent to our IF angle . When used on butterflies , the micrograph technique closely matches the results of the pseudopupil method . However , measuring the local radius from the eye’s profile only provides information along a 2D transect without describing the eye’s entire visual field , and calculating the IO angle in this way has similar limitations to our method in the presence of skewed CC ( discussed further in the following section and the ‘Supplemental methods’ ) . The pseudopupil technique allows IO angle ( and facet diameter ) to be measured directly from live insects and maintains the topology of these variables with reference to the visual world ( Stavenga , 1979 ) . Nevertheless , it is rarely applied across the entire FOV ( or CP ) – the equipment’s geometry typically limits the measurement range to ±70° in azimuth and elevation ( Rutowski et al . , 2009 ) – and cannot measure the dimensions of internal eye structures that influence sensitivity . It is also difficult to measure IO angle directly on compound eyes that lack a pseudopupil and thus have a uniformly black iris pigment ( as is the case for many hymenopteran eyes , including those of honeybees and bumblebees ) . Antidromic illumination is an alternative in such cases that involves the placement of a light source inside an insect’s head to create a pseudopupil with light emerging from the eye’s facets along the reverse optical path of each receptor . This illumination method has been used to measure IO angles into the frontal visual field of several species ( Kirschfeld , 1973; Spaethe and Chittka , 2003; Dyer et al . , 2016 ) . However , desert ants and honeybees are the only species in which this method has been used to determine the full FOV , and in the latter case , the IO angle topology ( Seidl , 1982; Zollikofer et al . , 1995 ) . The method that we have developed to calculate the CP and IF angle can broadly be applied to any apposition compound eye , regardless of its iris pigmentation , and could even be used on preserved insects with intact corneas . This approach could also be used to estimate the visual properties of superposition eyes ( Land and Nilsson , 2012 ) , although modifications would be required to account for the larger optical effect of crystalline cones in this eye type . Unique to our approach is the ability to describe vision in world-referenced coordinates , which allows our findings to be compared quantitatively to the results of studies on the vision of other species . In addition , we observed that the topologies of local variables in our analysis were spatially autocorrelated ( e . g . Figure 3B ) , and dimensional reduction techniques could be used ( with a larger sample size ) to determine a subset of parameters that describe the variation in eye shape and visual capabilities observed between bumblebees ( Goodhill et al . , 2015; Klingenberg , 2016 ) . The eye anatomy and visual capabilities of honeybees have previously been studied in detail . As such , they are an important reference species for assessing the validity of the analysis method presented here . In addition , honeybees have no distinct size polymorphism ( Streinzer et al . , 2013 ) , which improves the robustness of comparisons between different studies by minimising differences caused by variation in eye size . As a case in point , the honeybees in our study and those analyzed by Streinzer et al . , 2013 were almost identical in body size ( our specimens had ITWs of 2 . 9 mm and 3 . 0 mm vs . 2 . 9 ± 0 . 0 mm ) , and had similar eye surface areas ( both 2 . 4 mm2 vs . 2 . 5 ± 0 . 1 mm2 ) , facet numbers ( 5440 and 5484 vs . 5375 ± 143 ) and maximum facet sizes ( 25 . 4 μm and 25 . 6 μm vs . 25 . 2 ± 0 . 3 μm ) . Thus , we conclude that our method is capable of describing corneal eye properties to within 5% of the values provided by replica-based techniques ( with ‘worst case’ errors of −4 . 4% for surface area , +4 . 8% for facet number , and +2 . 8% for maximum facet diameter ) . The thicknesses of eye components have previously been described from sections through a honeybee’s compound eye ( Greiner et al . , 2004 ) ; in the forwards facing areas of our honeybee eyes , we measured similar lens ( 34 μm and 37 μm vs . 28 μm ) and CC thicknesses ( 46 μm and 50 μm vs . 55 μm ) . However , our retinal thickness measurements were substantially thinner ( 219 μm and 223 μm vs . 320 μm ) , and the value from histology lies above the range that we measured for that parameter . Although our thickness measures are consistent between individuals , the relative difference between our measurement of retinal thickness and that of Greiner et al . ( 2004 ) is substantially larger than the error we estimated for the corneal eye properties . A likely explanation for this difference is that our definition of thickness is based on the local surface NV from the lens and the distance until it intersects the upper surfaces of the CC , retina , and lamina interfaces ( Figure 1D ) , whereas Greiner et al . ( 2004 ) directly measured the rhabdom length from serial sections . Rhabdom length , rather than retinal thickness , is the relevant dimension for calculating optical sensitivity and , as a rhabdom may not necessarily lie perpendicular to the cornea , its length can evidentially exceed our structural thickness measurements ( the ratio by which it does so could also vary across the eye ) . In addition , the mean values and distributions that we calculated for all variables , besides CC thickness ( Figure 3—figure supplement 3 ) , are closely matched in the two honeybees examined in this study , indicating that our analysis method produces highly repeatable results . Previous studies on Apis have not calculated IF angles , but several have used an antidromically illuminated pseudopupil to measure IO angles directly . Values for the average IO angle in a honeybee’s frontal visual field have previously been measured as 1 . 8° ( Kirschfeld , 1973 ) and 2 . 0° ( Seidl , 1982 from Giger , 1996 ) , with the latter recording 1 . 2° in the acute region . In the frontal and acute regions , the IF angles calculated for our honeybees ( frontal: 1 . 4° and 1 . 7°; acute: 0 . 9° and 1 . 0° ) are slightly lower than the reported IO angles , but our measurements exceed those reported by Seidl ( 1982 ) in the dorsal and lateral regions ( Supplementary file 1–Table S2 ) . The primary source of difference between these measurements is that the honeybee’s ommatidial viewing axes can be skewed from the corneal NV and this misalignment varies across the eye ( Snyder , 1979 ) . This skewness is visible in sections through compound eyes ( Baumgärtner , 1928 ) , and unfortunately is not correctable without segmenting the individual CC for use in optical modeling . As a result of this , our method to calculate IF angles underestimated the smallest honeybee IO angles by approximately 30% ( absolute value: −0 . 3° ) but may overestimate other angles by 30% to 60% ( +0 . 8° to +1 . 3° ) . Because we calculate the CP from the corneal surface , we likewise underestimate the honeybees’ complete FOV as this also depends on CC skewing . An individual eye’s FOV is reported by Seidl and Kaiser ( 1981 ) to be nearly hemispherical , whereas the CP of the eyes measured in this study span a quarter ( 25% and 27% ) of the world sphere . The greatest differences between these angular extents appear to occur in the ventral and posterior regions: Seidl and Kaiser ( 1981 ) reported that the honeybee FOV extends down to −90° in elevation and back to −156° in azimuth ( at 0° el . ) , whereas we find that the CP extends to −60° in elevation ( at −60° ( az . - Azimuth ) ) and to −107° in azimuth ( at 0° el . ) . In addition , Seidl and Kaiser ( 1981 ) found frontal binocular overlap from dorsal to ventral regions , whereas our CP only indicated binocular overlap dorsofrontally . Given that pseudopupil measurements found a larger FOV and IO angles that were often larger than the IF angles from our projection method , the optical axes of ommatidia must generally diverge in honeybee eyes to create a larger FOV at the expense of resolution . Although this demonstrates a limitation in using our method to approximate visual resolution with IF angles , it also highlights the importance of developing additional techniques to determine CC orientation if detailed analyzes and comparisons of visual fields are to be made within and between species . The visual systems of bumblebees have received less attention than those of honeybees . Nonetheless , several studies have provided data against which we can compare our results . The body sizes of the two medium-sized bumblebees in our study were similar to those of the B . terrestris workers analyzed by Streinzer and Spaethe ( 2014 ) ( we measured an ITW of 4 . 0 mm for both bees , vs . 3 . 9 ± 0 . 6 mm , as well as maximum facet sizes of 25 . 0 μm and 25 . 2 μm vs . 25 . 1 ± 1 . 9 μm ) . However , the surface areas of our bees' eyes were slightly smaller ( 2 . 2 mm2 and 2 . 4 mm2 vs . 2 . 8 ± 0 . 6 mm2 ) and had fewer facets ( 4941 and 5505 vs . 5656 ± 475 ) . As both eye and body size vary substantially between bumblebee individuals , we did not estimate their ‘worst case’ errors as we did for honeybees . Nonetheless , the similarity in the majority of measurements between our study and that of Streinzer and Spaethe ( 2014 ) suggests good agreement between the different methods used to obtain them and provides further support for the validity of our method . The pseudopupil technique has also been used with antidromic illumination to measure the IO angles from the mediofrontal area of bumblebee eyes as a function their body size ( Spaethe and Chittka , 2003 ) for small ( ITW: 2 . 8 mm to 3 . 0 mm , mean IO angle: 1 . 5° for six bees ) to medium-sized individuals ( 4 . 0 mm to 4 . 2 mm , 1 . 2° from four bees ) . By comparison , we found substantially larger IF angles ( facing frontally ) for our equivalently sized small ( 3 . 0 mm , 2 . 7° ) and medium-sized ( both 4 . 0 mm , 2 . 2° and 2 . 4° ) bees . However , our method shows that the IF angle of bumblebees decreases from their frontal to their lateral visual fields ( Figure 3B ) , and that by −45° ( az . - Azimuth ) , the IF angles ( small bumblebee 1 . 5° , medium bumblebee 1 . 3° ) match the IO angle measurements made by Spaethe and Chittka ( 2003 ) . Eye-referenced locations for IO angle measurements do not provide a clear world-referenced viewing direction , and it is conceivable the measurements of Spaethe and Chittka ( 2003 ) were taken from eye areas directed somewhat laterally , in which case our results provide similar values . This highlights the importance of using a world-reference for visual studies , even when comparing between individuals of the same species . Regardless of eye size , we found that the minimum IF angles of both bee species were oriented somewhat laterally ( Figure 3Bii ) , which also challenges the common assumption that the frontal visual field is always the most relevant for acute insect vision . We found a scaling exponent of 0 . 45 for eye surface area vs . ITW for the B . terrestris workers in this study ( Figure 2A ) . This is substantially lower than the between species allometry rate found across a range of 11 Bombus species that had a similar range of body sizes ( 0 . 73 ) ( Streinzer and Spaethe , 2014 ) , indicating that the absolute investment in compound eyes varies more between Bombus species than between B . terrestris individuals . Similar scaling exponents in other Bombus species would provide a clear indication that , independently of any factors influencing body size , visual requirements have influenced the evolution of eye size in different bumblebee species . We also calculated the scaling exponents of facet number and diameter as a function of eye size for the 11 Bombus species described by Streinzer and Spaethe ( 2014 ) . We found that facet number had a substantially higher scaling rate when making comparisons between Bombus species ( 1 . 39 ) rather than within B . terrestris ( 0 . 61 ) . This suggests that the scaling of facet number , and thus resolution , is likely to be a species-specific adaptation . Conversely , the maximum size of facets had a lower scaling rate between species ( 0 . 26 ) than within B . terrestris ( 0 . 70 ) . A recent study on the allometry of wood ant eyes from different colonies also showed that , despite maintaining similar scaling exponents for total eye size between colonies , two colonies invested in more facets as eye size increased whereas another invested in larger facets ( Perl and Niven , 2016a ) . Evidently , varying the parameters of eye allometry allows for substantial fine-tuning of the visual performance of individuals , both between and within species . The allometry of visual performance in bumblebees may influence which individuals and species can most effectively forage for specific floral resources ( Dafni et al . , 1997 ) . Within a species , variation between individuals’ visual capabilities could impact on their relative foraging ability . Behavioural studies investigating the influence of size on visual performance have shown that larger bumblebees ( i ) do indeed have more acute vision when trained to discriminate visual targets ( Spaethe and Chittka , 2003 ) , ( ii ) are able to fly at lower light intensities ( Kapustjanskij et al . , 2007 ) , and ( iii ) are more efficient foragers ( Spaethe and Weidenmüller , 2002 ) . Spaethe and Chittka ( 2003 ) also found that an increase in body size of 34% halved the minimum angular object size that a bee could identify , yet they noted that the allometric improvement in IO angle that they measured could not directly predict the improvement in behavioural visual acuity . We applied the linear regression equation calculated by Spaethe and Chittka ( 2003 ) ( Angle ( ° ) = 17 . 6–3 . 1 × ITW ( mm ) ) to predict the minimum detectable object size for the small and medium bees in our study ( ITWs 3 . 0 mm and 4 . 0 mm ) , giving visual angles of 8 . 4° and 5 . 2° , a 38% decrease . While these angles are substantially larger than the IF angles calculated using our method , the greatest relative improvement in IF angle ( at any matching direction in the common CP ) between our small and medium-sized bees is 38% , surprisingly close to the relative 34% improvement predicted by the behavioural study . This local acuity improvement is directed frontally ( −9° el . , −15° ( az . - Azimuth ) ) , in a region of the visual field that is well positioned for target discrimination and that has been shown to be important for measuring optic flow for flight control ( Baird et al . , 2010; Linander et al . , 2015 ) . Our findings suggest that it may in fact be possible to predict the relative differences in behavioural measures of visual acuity ( on insects of different sizes ) by measuring the allometry of visual resolution at the relevant location in the visual field . Interestingly , bigger eyes do not always provide increased visual performance – bumblebees with larger eyes do not exhibit differences when discriminating between different periodic patterns ( Chakravarthi et al . , 2016 ) , although the range of body sizes ( 3 . 2– mm to 4 . 3 mm ) was narrower than that of the bees analyzed in the present study . Given that the IO angle on the mediofrontal eye area of medium-sized bumblebees is 1 . 2° ( Spaethe and Chittka , 2003 ) , relatively poor resolution limits have been measured for target detection ( 2 . 3° ) ( Dyer et al . , 2008; Wertlen et al . , 2008 ) and pattern discrimination ( 4 . 8°/cycle ) ( Chakravarthi et al . , 2016 ) . The limits obtained from these behavioural experiments are approximately twice those that would be expected on the basis of the sampling frequency ( Snyder , 1979 ) . Where the cut-off frequency of the optics is lower than the sampling frequency of the ommatidial array , the optics of the compound eye lenses may also limit visual acuity through oversampling , an additional limiting factor that we have not considered here ( Snyder , 1979 ) . Our analysis shows that lens thickness does vary across all bees' eyes ( Figure 3—figure supplement 2B ) , which is likely to result in local differences in focal length and may lead to topological variation in acceptance angle and , thus , the optical cut-off frequency . An approximately 25% increase in acceptance angle has indeed been found between frontally and laterally facing ommatidia in honeybees ( Rigosi et al . , 2017 ) . Two additional considerations when determining visual acuity from anatomical measurements are that the acceptance angle of many insects varies between states of light and dark adaptation ( Warrant and McIntyre , 1993 ) , and that both object illumination and contrast also influence an eye’s effective acuity ( Snyder et al . , 1977; Warrant , 1999 ) . In this study , we found that the CP of bumblebee eyes increased with eye size ( Figure 2C ) . This was a consequence not simply of having a larger eye , but also of a change in eye shape such that the surface was projected onto a larger angular area . We also identified an area of binocular overlap not previously reported in bumblebees . The extent of this corneal binocular overlap , directed both frontally and dorsofrontally ( Figure 4C ) , increased rapidly with body size . Bumblebee workers have been found to approach artificial ( Reber et al . , 2016 ) and natural flowers ( Orth and Waddington , 1997 ) from below , which would place the visual target dorsofrontally , in a region where we also found that IF angle decreases with eye size ( Figure 6A ) . Hence , larger bumblebees would view the flowers they approach with a more acute and larger binocular visual field , which would potentially improve their visual discrimination or landing control relative to that of smaller bees . The potential benefits of this binocular overlap would be an interesting topic for further behavioural investigations . Surprisingly , the scaling exponent ( as a function of body size ) found here for bumblebees’ CP is nearly identical to that found from the optically measured FOV of differently sized butterfly species ( Rutowski et al . , 2009 ) . This is the case for both the visual field of a single eye ( we found 0 . 17 vs . 0 . 14 ) and the binocular regions ( 0 . 79 vs . 0 . 82 ) . By contrast , the FOV of desert ants remains similar over a nearly two-fold increase in head size ( Zollikofer et al . , 1995 ) . Given the common scaling rates shared by the B . terrestris worker’s CP and that of butterflies , we hypothesise that increasing the visual field of each eye ( and the binocular overlap between eyes ) at the identified rates is a general strategy for compound eye enlargement among different groups of flying insects . Although visual field extent has rarely been considered during previous studies of insect vision , increasing FOV size has been shown to improve the performance of visually guided behaviours such as navigation ( Wystrach et al . , 2016 ) and visual motion detection ( Borst and Egelhaaf , 1989 ) , and is likely to improve the ability of larger bumblebees to perform these visually guided behaviours . Local variation in the scaling rate of eye properties will cause eye-dependent variation in the topology of visual capabilities . The region with the lowest IF angle scaling exponent ( leading to improved visual resolution ) is directed dorsofrontally ( Figure 6A ) , while a positive but relatively similar scaling exponent for facet size occurs across the visual field ( Figure 6B ) . By contrast , the scaling rate of IO angles vs . body size of Orange Sulphur butterflies was greatest in the ventral , ventrofrontal , and dorsal eye areas , while their facet diameters were also found to increase at a uniform rate across the areas measured ( Merry et al . , 2006 ) . Again , the results from ants are qualitatively different from those for bees and butterflies: the scaling exponent of facet diameter varied between the eye areas of wood ants , being highest in the dorsal and frontal areas ( Perl and Niven , 2016b ) , whereas a study on desert ants found that IO angle scaled similarly between lateral and dorsal eye areas ( Zollikofer et al . , 1995 ) . Pseudopupil measurements along a vertical transect of the eyes of damselfly species found that the maximum diameters and minimum IO angles were influenced by both eye size and habitat ( Scales and Butler , 2016 ) . Although the scaling exponents along the eye transects were not measured , damselflies living in dim , cluttered habitats appeared ( independently of eye size ) to have more prominent eye specializations than those living in open habitats . To our knowledge , this is the first study to investigate the topology of retinal thickness in insects ( Figure 5 ) , and it is evident from our analysis that retinal thickness varies substantially across all bee eyes . If differences in retinal thickness are translated into equivalent differences in rhabdom length , this would influence optical sensitivity across each eye by 20–50% for bumblebees and by 53% for honeybees ( based on the difference between the minimum and maximum retinal thickenss of each eye and the influence of rhabdom length on sensitivity ) . Retinal thickness is typically highest ventrally and posteriorly ( Figure 5B ) , where higher retinal sensitivity may compensate for the reduced effective aperture that results from the skewed CC in these regions ( Stavenga , 1979 ) . Retinal thickness has a positive scaling exponent across the majority of the visual field ( Figure 6C ) , which would improve the optical sensitivity of larger bees . Unexpectedly , we identified that retinal thickness increases at a greater rate in the dorsal hemisphere and would provide larger bumblebees with relatively increased dorsal sensitivity that may , for instance , assist a bee’s ability to visually discriminate downwards facing flowers that are not directly illuminated by sunlight ( Makino and Thomson , 2012; Foster et al . , 2014 ) . Our results demonstrate that , in addition to facet size , retinal dimensions offer substantial scope for insects to fine-tune optical sensitivity across their visual fields , a point that appears to have been overlooked by previous studies . The visual topologies and scaling exponents that we measured for bumblebees partially support our initial hypothesis that the increased area of a larger eye would be invested primarily in improving the capabilities of a small visual region . The improved visual resolution of larger bees is primarily directed dorsofrontally , but the scaling of facet diameter and retinal thickness would lead to increased optical sensitivity across their entire field of view . As a result , we now hypothesise that , for a given insect group , specific regions in the visual field may have certain ‘ideal’ requirements for resolution and/or sensitivity that are based on the visual information available in their specific habitat and their behavioural ecology . Once the size of an eye allows such a threshold to be reached , additional area could then be more broadly invested in improving the visual capabilities across the visual field . This revised hypothesis incorporates our findings that the differential allometry of Bombus eye properties allows their visual capabilities to be improved both locally or globally across their growing visual field . Analysing the 3D structure of insect eyes to determine a holistic description of their visual capabilities provides insight into how the morphology of eyes has evolved to sample visual information from the world . We find that the differential scaling of the morphology between eye areas allows bigger bumblebees to invest the increased resources of a larger eye in improved sensitivity across an enlarged visual field . Yet , studying the allometry of bumblebees' entire visual topology also indicated specific regions that have a high investment rate , such as the dorsofrontal region of both enlarging binocularity and increasing resolution , or the high rate of thickening in the dorsally facing retina . Important visual information is presumably viewed by bumblebees in these regions of their visual fields , indicating a promising avenue for further behavioural experiments , such as the use of virtual reality to manipulate the visual cues at specific regions in an insect’s FOV ( Stowers et al . , 2017 ) and observational studies to identify what bumblebees view in those regions when flying through natural environments ( Stürzl et al . , 2015 ) . The differential allometry between eye areas undoubtedly endows insects that have larger eyes with improved vision because they have a better capacity to match their visual capabilities to the requirements of both their environment and their behaviour . Bumblebees ( Bombus terrestris ) spanning the typical range of body sizes ( categorized here as small , medium , and large ) were collected from a commercial hive ( Koppert , UK ) . Honeybees ( Apis mellifera ) were collected from hives maintained at the Department of Biology , Lund University , Sweden . Several workers of each species and size category were collected and anesthetized with carbon dioxide gas before being dissected . Samples were preserved by dissecting the left compound eye ( to preserve this alone ) , or by removing the front , bottom , and rear of the head capsule ( to preserve the whole head ) . They were then fixated , stained , and embedded in epoxy resin . See the 'Supplemental methods' for further information about the preparation procedure . We also fixated several completely intact heads of each bee species , which we then dehydrated in ethanol and critical point dried . The inter-tegula width ( ITW ) of each sample was measured with digital callipers to provide a measure of body size ( Figure 1D ) ( Cane , 1987 ) . Tomographic imaging of samples was conducted at the Diamond-Manchester Imaging Branchline I13-2 ( Rau et al . , 2011; Pešić et al . , 2013 ) at the Diamond Light Source , UK . See the ‘Supplemental methods’ for further information about the imaging parameters . Dissected and dried heads were imaged using ×2 . 5 total magnification ( 2 . 6 μm effective pixel size , Figure 1B ) , whereas isolated eyes were imaged with ×4 total magnification ( 1 . 6 μm effective pixel size , Figure 1D ) . We examined the imaged volumes to choose , for further analysis , the two best-preserved compound eye samples from each species and size category ( six bumblebees and two honeybees in total ) and the best-preserved head capsule from each species . Amira ( FEI ) was used to analyze these volumes in three ways: i ) by manually labeling the structures of a compound eye ( Figure 1A ) , ii ) by aligning the labeled compound eyes of a given species onto the scan of a full head ( Figure 1B ) , and iii ) by measuring the facet dimensions on a compound eye ( Figure 1C ) . Additional details of the procedure used to process volumes in Amira are provided in the ‘Supplemental methods’ . We developed Matlab scripts to use data from the volumetric analysis performed in Amira ( labeled volumes , facet measurements and transforms ) to compute the eye properties ( eye surface area , eye volume , facet number , facet diameter , radius of curvature , and thicknesses ( for the lenses , CC , and retina ) ) , visual capabilities ( individual CP , binocular CP , complete CP , IF angle , and optical sensitivity ) and a metric ( eye parameter ) . Note that the calculations of optical sensitivity and eye parameter assume that the IO angle equals the computed IF angle ( which is not met across the entire eye ( Figure 1D ) ) , limiting the accuracy with which we can report these parameters . The italicized variables were determined locally , that is , they were calculated at sampling points that were equally spaced at 25 μm intervals across each bee’s corneal surface . The corneal normal vector ( NV ) of each sampling point was used as an indication of the viewing direction of that part of the eye in space ( Figure 1D ) . We determined which viewing directions occurred inside the CP of each bee ( Figure 1E ) , before interpolating each locally calculated variable from the sampling points onto the world . This allowed locally calculated variables to be represented in both eye- and world-centric coordinates , which are reported using plots of: facet-wise mean values and distributions , projections of CP limits ( and topologies ) onto the world , and profiles across both azimuth or elevation representing the average of a projected parameter ( or the integral of CP ) across 10° bands of visual space . Additional details about this computational analysis procedure and a discussion on its limitations are provided in the ‘Supplemental methods’ . The MATLAB scripts for calculating and plotting eye properties are available for download ( Taylor , 2018; copy archived at https://github . com/elifesciences-publications/compound-eye-plotting-elife ) . The allometry of values calculated from bumblebee eyes was described by fitting the parameters b and α in the power function Y = bxα ( Huxley and Teissier , 1936 ) , after a logarithmic transformation of the size indicator ( x ) and the dependent variable ( Y ) . See the ‘Supplemental methods’ for further information about the allometry procedure . Allometry functions were calculated for a given variable ( if the variable was calculated locally then the facet-wise mean value was used ) measured for all bumblebees ( Supplementary file 1–Table S1 ) . As the topology of most parameters varied across the eye , we also calculated local allometric functions for variables on the basis of their projection into the world and represent these as spatial maps of the scaling exponent of each variable ( Figure 6 and Figure 6—figure supplement 1 ) .

Bees fly through complex environments in search of nectar from flowers . They are aided in this quest by excellent eyesight . Scientists have extensively studied the eyesight of honeybees to learn more about how such tiny eyes work and how they process and learn visual information . Less is known about the honeybee’s larger cousins , the bumblebees , which are also important pollinators . Bumblebees come in different sizes and one question scientists have is how eye size affects vision . Bigger bumblebees are known to have bigger eyes , and bigger eyes are usually better . But which aspects of vision are improved in larger eyes is not clear . For example , does the size of a bee’s eyes affect how large their field of view is , or how sensitive they are to light ? Or does it impact their visual acuity , a measurement of the smallest objects the eye can see ? Scaling up an eye would likely improve all these aspects of sight slightly , but changes in a small area of the eye might more drastically improve some parts of vision . Now , Taylor et al . show that larger bumblebees with bigger eyes have better vision than their smaller counterparts . In the experiments , a technique called microtomography was used to measure the 3D structure of bumblebee eyes . The measurements were then applied to build 3D models of the bumblebee eyes , and computational geometry was used to calculate the sensitivity , acuity , and viewing direction across the entire surface of each model eye . Taylor et al . found that larger bees had improved ability to see small objects in front or slightly above them . They had a bigger area of overlap between the sight in both eyes when they looked forward and up . They were also more sensitive to light across the eye . The experiments show that improvements in eyesight with larger size are very specific and likely help larger bees to adapt to their environment . Behavioral studies could help scientists better understand how these changes help bigger bees and how the traits evolved . These findings might also help engineers trying to design miniature cameras to help small , flying autonomous vehicles navigate . Bees fly through complex environments and face challenges similar to those small flying vehicles would face . Emulating the design of bee eyes and how they change with size might lead to the development of better cameras for these vehicles .

2019

Bumblebee visual allometry results in locally improved resolution and globally improved sensitivity

Brain development relies on an interplay between genetic specification and self-organization . Striking examples of this relationship can be found in the somatosensory brainstem , thalamus , and cortex of rats and mice , where the arrangement of the facial whiskers is preserved in the arrangement of cell aggregates to form precise somatotopic maps . We show in simulation how realistic whisker maps can self-organize , by assuming that information is exchanged between adjacent cells only , under the guidance of gene expression gradients . The resulting model provides a simple account of how patterns of gene expression can constrain spontaneous pattern formation to faithfully reproduce functional maps in subsequent brain structures . Karbowski and Ermentrout , 2004 developed a reaction-diffusion style model of how extrinsic signalling gradients can constrain the emergence of distinct fields from intrinsic cortical dynamics . Their model defines how the fraction of occupied synapses ci⁢ ( x , t ) and the density of axon branches ai⁢ ( x , t ) interact at time t , along a 1D anterior-posterior axis x , for N thalamocortical projections indexed by i . The model was derived from the assumption that the rates at which ai and ci grow are reciprocally coupled . Extending the original 1D model to simulate arealization on a 2D cortical sheet , we use ai⁢ ( 𝐱 , t ) and ci⁢ ( 𝐱 , t ) , and model synaptogenesis as ( 1 ) ∂⁡ci∂⁡t=-α⁢ci+β⁢ ( 1-∑j=1Ncj ) ⁢[ai]k . Accordingly , where the total fraction of synaptic connections sums to one , connections decay at rate α . Otherwise , ci⁢ ( 𝐱 , t ) increases non-linearly ( k>1 ) with the density of axon branching . Axon branching is modelled as ( 2 ) ∂ai∂t= ∇ ⋅ ( D∇ai−ai∑j=1Mγi , j∇ρj ( x ) +χi ) −∂ci∂t . The first term on the right describes the divergence ( indicated by ∇⋅ ) of the quantity in parentheses , which is referred to as the ‘flux’ of axonal branching . The flux represents diffusion across the cortical sheet , at rate D , and the influence of M molecular signalling fields , ρ⁢ ( 𝐱 ) . The influence of a given field ( indexed by j ) on a given thalamic projection ( indexed by i ) , is determined by γi , j , which may be positive or negative in order that axons may branch in the direction of either higher or lower concentrations . Note that computing the divergence in simulation requires cells on the cortical sheet to communicate with immediately adjacent cells only ( see Methods ) . Here χi=0 is a placeholder . The second term on the right represents the coupling between axon branching and synaptogenesis , and an assumption that the spatial distribution of synaptic density across the cortical sheet is broadly homogeneous . As such , the quantity ci can be thought of as the connection density . First , we verified that all results established by Karbowski and Ermentrout , 2004 for a 1D axis could be reproduced using our extension to a 2D cortical sheet . Using an elliptical domain , S , with M=3 offset guidance gradients aligned to the longer axis , N=5 thalamocortical projections gave rise to five distinct cortical fields at locations that preserved the topographic ordering defined by the original γ values . However , we found that specifying N ordered areas required M≈ ( N+1 ) /2 signalling fields . This is because localization of axon densities occurs only when projections are influenced by interactions with two or more signalling gradients that encourage migration in opposing directions . As the number of guidance fields is unlikely to approach the number of individual barrels , modifications to the model were required . We reasoned that an arbitrary number of distinct field locations may be determined by a minimum of two guidance gradients , if the concentration of the projection densities is influenced by competition between projections , and if a projection that interacts more strongly with a given guidance gradient migrates further in the direction of that gradient . Accordingly , projections that interact most strongly with a given guidance gradient would come to occupy cortical locations at which that field has extreme values , leaving adjacent locations available to be occupied by projections with the next strongest interactions , and so forth . This would in principle allow the relative locations of the fields to be specified by the relative values of the interaction parameters , γ , and hence for a topological map in the cortex to be specified by a spatial ordering of the γ values at the level of the thalamus . Such dynamics are quite unlike those described by classic chemospecificity models ( Sperry , 1963 ) , which essentially assume centre-points by specifying conditions in the target tissue that instruct pre-identified afferents to stop growing . Consider , for example , that when simulated in isolation from one-another , all projections in the model described would simply migrate to the extrema of the cortical guidance fields . Testing this reasoning required increasing the strength of the competition between simulated thalamocortical projections for cortical territory , by increasing the tendency for each projection to compete for cortical space in which to branch and make connections . The major modification required was thus to introduce into the model an additional source of competition between thalamic projections . The term in parentheses in Equation 1 represents competition between thalamocortical projections for a limited availability of cortical connections . To introduce competition also in terms of axon branching , whilst ensuring that ai is conserved over time , we redefined ( 3 ) χi ( x , t ) =ϵaiN−1∇∑j≠iNaj . This term contributes to the flux of axonal branching as an additional source of diffusion , scaled by ϵ , which reduces the branching density for a given projection where the branches of other projections are dense . Note that this operation is local to individual afferent projections . In addition , the model we have outlined requires that molecular guidance gradients in the cortex are complemented by graded values of the interaction strengths , γ , at the level of the thalamus . While the precise mechanisms by which thalamic and cortical gradients interact during development have not been fully characterised , the presence of complementary thalamic and cortical guidance gradients has been well established experimentally . In particular , the EphA4 receptor and its ligand ephrin-A5 are distributed in complementary gradients in the somatosensory thalamus and cortex ( Vanderhaeghen et al . , 2000; Miller et al . , 2006 ) . Cells originating in VPM express high levels of EphA receptors and project to the lateral part of S1 , which expresses low levels of ephrin-A5 , and cells originating in the VPL express low levels of EphA receptors and project to the medial part of S1 , which expresses high levels of ephrin-A5 ( see Gao et al . , 1998; Dufour et al . , 2003; Vanderhaeghen and Polleux , 2004; Speer and Chapman , 2005; Torii et al . , 2013 ) . We assume that such patterning arises because the relative strengths of interaction with guidance molecules ( e . g . , ephrin-A5 ) in the cortex are correlated with the relative concentrations of complementary molecules ( e . g . , EphA4 ) in the thalamus , and thus with thalamic position along the axis to which their gradients are aligned . For simplicity , the two simulated thalamic interaction gradients , as well as the two cortical guidance gradients , were initially chosen to be linear and orthogonal . Hence a given pair of γ values corresponds to the coordinate of a barreloid centre in the VPM . Coordinates , in a reference plane defined by the anterior-posterior and medial-lateral axes , were estimated from Figure 5d of Haidarliu and Ahissar , 2001 , and scaled such that γ∈±2 . Note that this scaling is arbitrary because according to the model the coordinates provide relative position information only . A cortical boundary enclosing barrels for 41 macrovibrissae was traced from a cytochrome oxidase stain from Zheng et al . , 2001 ( using original data kindly supplied by the authors ) , and Equations 1–3 were solved for N=41 projections on the resulting domain , S , using M=2 linear signalling gradients aligned with the anterior-posterior and medial-lateral axes . These gradients are shown with the barrel field boundary in Figure 1A for clarity , though like ephrin-A5 they may be thought of as extending across the cortical hemisphere ( Miller et al . , 2006 ) . Simulations were stepped through 30000 iterations of Equations 1–3 ( δ⁢t=0 . 0001 ) . Across a wide range of parameter values , random initial conditions ( a uniform random distribution for a⁢ ( 𝐱 , 0 ) ∈ ( 0 . 2 , 0 . 4 ) , c⁢ ( 𝐱 , 0 ) =0 ) eventually yielded a clear Voronoi-like tessellation of topographically organized thalamocortical projections , confirming that barrel maps can self-organize in the absence of pre-specified centre points . The organization is apparent in a plot of the identity of the projection for which the connection density is maximal at each simulated cortical location , as shown in Figure 1C . Parameter values for this example simulation ( see also Figure 1—video 1 ) were obtained by conducting a full parameter sweep and choosing a combination ( α=3 . 6 , β=16 . 67 , k=3 , D=0 . 5 , ϵ=1 . 2 ) that scored well against the following three measures . First , we used an algorithm introduced by Honda to measure the discrepancy of each barrel shape from a Dirichlet domain shape ( Honda , 1983 ) . Low overall values of this Honda-δ metric obtained from simulated barrels indicate a close correspondence of the simulated barrel field with a Voronoi tessellation , and thus with a biological barrel field ( for mice δ≈0 . 054 , Senft and Woolsey , 1991 , and our analysis of data from Zheng et al . , 2001 indicates that the value for rats is similar ) . For the tessellation that is overlaid on the real barrel field in Figure 1A , δ=0 . 025 , and a reduction in δ in the example simulation over time confirmed that an equivalent ‘good’ Voronoi pattern can emerge within ≈ 20000 iterations ( Figure 1D , red circles ) . Second , we devised a pattern difference measure that is sensitive to deviations in the component shapes and overall topographic registration between two tessellations , η , and we used this measure to compare the simulated barrel fields to the real barrel field from which the boundary shape applied to the simulation was obtained ( see Methods for details ) . A similar reduction in η in the development of the example simulation confirmed that the shapes and arrangement of emergent connection fields came to match those of the real barrel field by around 20000 iterations ( Figure 1D , black squares ) . Third , we measured the connection selectivity , ω , at each location on the cortical sheet , as the connection density of the most dense projection divided by the sum over all projection densities . The overall connection selectivity increased as the barrel map self-organized in the example simulation ( Figure 1D , grey hexagons ) , and the selectivity became concentrated in regions overlapping with the emergent barrel centres ( Figure 1E ) . Against these three metrics we are also able to characterise the robustness of self-organization to the model parameters , and to investigate the sensitivity of the model to variation in its inputs . Figure 2A shows values of δ , η and ω obtained after 30000 iterations , from 216 independent simulations , each representing a unique combination of the model parameters D , ϵ , and the ratio α/β . First , we observe that self-organization is highly robust to the ratio α/β , across five orders of magnitude , with respect to all three metrics . Second , the most strongly Voronoi-conforming patterns ( low Honda-δ ) were generated by simulations in which the diffusion constant D and the strength of competition ϵ were high . Third , strongest overall connection selectivities , ω , were obtained for lower values of D . Fourth , variation in the pattern difference metric , η , indicated that the alignment between real and simulated patterns was greatest for intermediate rates of diffusion , D≈0 . 5 . Together these results indicate that when competition is strong , the rate of diffusion determines a trade-off such that fields emerge to be barrel-shaped when diffusion is fast and they emerge to be more selectively innervated when diffusion is slow . The parameters of the example simulation are indicated in Figure 2A using an asterisk . In Figure 2B , we also present examples of alternative patterns that emerge for different choices of D . Decreasing the rate of diffusion may be considered equivalent to increasing the overall size of the domain , S . Hence , insights into barrel development in species with a larger representation of the vibrissae , which do not have barrel fields , may be gained by studying pattern formation when D is small . In this context , it is interesting to note that for small D , the organization is predicted to be topological but highly irregular , with a general expansion in the territory occupied by the central versus peripheral domains that would presumably manifest as an absence of identifiable barrel fields ( Figure 2Biii ) . Next we conducted a sensitivity analysis to determine the extent to which the quality of the pattern ( after t=30000 iterations ) is affected by perturbations to ( i ) the magnitude and offset of the noise applied to ai at t=0; ( ii ) noise applied to the interaction parameters , γi , j; ( iii ) noise ( at various length scales ) applied to the guidance fields; and ( iv ) the magnitude and orientation of one cortical guidance field relative to the other ( Figure 1A ) . Using the parameters of the example simulation ( Figure 1C ) we established baseline mean and standard deviations from ten independent simulations with initial uniform random values for a⁢ ( 𝐱 , 0 ) ∈ ( 0 . 2 , 0 . 4 ) , to be δ=0 . 089±0 . 004 , η=0 . 2108±0 . 002 , and ω=0 . 2165±0 . 0001 . Repeating with the variation in the initial noise doubled ( a⁢ ( 𝐱 , 0 ) ∈ ( 0 . 1 , 0 . 5 ) ) , or removed altogether ( a⁢ ( 𝐱 , 0 ) =0 . 3 ) , generated distributions of δ , η , and ω that were not statistically different , as established using paired two-sample t-tests . Adding noise to the interaction parameters ( γ ) affected neither the Honda-δ or the connection selectivity measures substantially ( see Figure 3A ) , and an increase in the pattern difference reflected an increase in the occurrence of topological defects only when perturbations became so large as to cause the ordering of γ values from neighbouring thalamic sites to be switched ( see example map Figure 3Bi ) . Adding noise to the cortical guidance field values , ρ1⁢ ( 𝐱 ) and ρ2⁢ ( 𝐱 ) , disrupted pattern formation only for high levels of noise applied at short length scales , which manifested as non-straight edges at the domain boundaries ( Figure 3Bii ) . Varying the slope of one linear gradient ρ1⁢ ( 𝐱 ) while keeping that of the other constant caused elongation of the emergent domains along the corresponding axis ( Figure 3Biii ) , while pattern formation was not strongly influenced by relaxing the assumption that the gradients of the two cortical guidance fields are orthogonal ( Figure 3Biv ) . Overall , the sensitivity analysis revealed that self-organization of barrel-like fields in the model is highly robust to a wide range of sources of perturbation . To further investigate the interplay of genes intrinsic to the developing neocortex and extrinsic factors such as thalamocortical input , we simulated two well known experimental manipulations of barrel development . First , we simulated a seminal barrel duplication paradigm ( Shimogori and Grove , 2005; Assimacopoulos et al . , 2012 ) in which the growth factor Fgf8 , which is normally expressed at the anterior end of the cortical subplate from around E9 . 5 ( Crossley and Martin , 1995 ) , is ectopically expressed ( by electroporation ) also at the posterior pole . We assume that this results in a mirror of the primary barrel cortex boundary along the rostrocaudal axis ( Assimacopoulos et al . , 2012 ) and a mirroring of the anterior-posterior guidance gradient ρ1 at the border between them ( Figure 4A ) . The result after 30000 iterations , and otherwise using the parameters of the example simulation , was two mirror-symmetrical barrel fields comprising 2⁢N barrels ( Figure 4B ) , consistent with the outcome of the original experiments . Finally , to investigate the response of the model to environmental manipulation , we simulated a whisker deprivation experiment . In a critical period comprising the first postnatal days , removal of the whiskers by electrocauterization , plucking , or trimming leads to observable changes in brain structures , including the barrel field ( Jeanmonod et al . , 1981 ) . Amongst other changes , deprivation of individual whiskers leads to smaller barrels ( Kossut , 1992 ) . We simulated trimming of the individual whisker C3 during the critical period by reducing the competitiveness of the C3 thalamocortical projection , ϵ . As a result , the corresponding field size was smaller ( Figure 3C ) , and the size of the fields representing the neighbours of C3 increased in size . A reduction in area to 65% , comparable to that induced by Kossut , 1992 , was obtained in simulation when ϵ was reduced to 86% of its default value ( Figure 3D ) , and the C3 barrel disappeared altogether when ϵ was less than half of its original value . Although individual whisker trimming reduces barrel size , if an entire row is trimmed , barrel sizes for the trimmed row are not obviously changed ( Land and Simons , 1985 ) . We investigated with a simulation in which we varied ϵ for all of the row C projections . Although the barrels which formed for row C did show some reduction in area ( Figure 4D ) , this reduction was small when compared with the individually trimmed C3 simulation . If the effect of trimming any whisker is to reduce its ϵ ( the competitiveness of its projection ) to 86% of its original value , then the model predicts that the cortical barrels for the trimmed C row will retain 91% of their area on average . Together with the results of simulated misexpression , the consistency of the simulated whisker trimming results with those of the original studies demonstrates how the model can be used to investigate the contribution of intrinsic and extrinsic factors to the development of cortical fields . The present results suggest that the key requirements for the emergence of realistic barrel patterning are ( i ) at each cortical location thalamocortical projections compete for a limited number of available synaptic connections ( Equations 1–2 ) , ( ii ) at each location the branching rate of a given projection is reduced by the density of other projections ( Equation 3 ) , and ( iii ) the branch density of each projection is conserved over time . The emergence of barrels in simulation required competition between thalamic projections in terms of synaptic connectivity and also competition in terms of cortical space , as represented by χ , with an implicit requirement for a self/other identifier amongst projections . This latter form of competition may account for the absence of barrels in rodents with larger brains , such as capybara , for which competition for space is presumably weaker ( Woolsey et al . , 1975 ) . Hence , irrespective of whether barrels are necessary for adaptive whisker function , the emergence of somatotopically ordered modular structures may be an inevitable consequence of local competition for cortical territory driven by input from an array of discrete sensory organs ( Purves et al . , 1992 ) . In reality , the Voronoi tessellation is extended by scores of smaller barrels alongside the E-row barrels , which represent the microvibrissae , and presumably form via the same competitive processes . Enforcing here the same boundary condition as used to represent the true edges of the barrel field was necessary to ensure the stability of the simulation , though we acknowledge that this region of the boundary was enforced primarily to keep the number of simulated projections , and hence the overall computational complexity of the simulation , manageable ( simulating an extra projection introduces 13030 new dynamical variables ) . It is important to emphasize that the formulation of the model is entirely local , insofar as simulation requires no information to be communicated from a given cortical grid cell to any but those immediately adjacent ( via diffusion ) . Hence the simulations demonstrate how a self-organizing system , constrained by genetically specified guidance cues and by the shape of the cortical field boundary , can faithfully reproduce an arrangement of cell aggregates in one neural structure as a topographic map in another . Moreover , the present results confirm that somatotopic map formation does not require the pre-specification of centre-points by as yet undetermined additional developmental mechanisms . We concentrated on the representation of the forty-one macrovibrissae that constitute a given barrel field , because their thalamic and cortical correlates are easily identifiable and consistently located , excluding the five rhinal whiskers as their cortical representation is isolated from the main barrel field . We excluded the representation of the microvibrissae to limit the overall complexity of the simulations . The cortical sheet was modelled as a two dimensional hexagonal lattice , which simplifies the computation of the 2D Laplacian . Within a boundary traced around the edge of a rat barrel field ( Figure 1A ) we set the hex-to-hex distance d to 0 . 03 mm , which resulted in a lattice containing 6515 hexes for the simulations shown in Figure 1A , C and D and 12739 hexes for the Fgf8 misexpression study shown in Figure 3 . Each hex contained 82 time-dependent variables: 41 branching densities ( ai ) and 41 connection densities ( ci ) . The rate of change of each of the time-dependent variables ( Equations 1 and 2 ) was computed using a fourth-order Runge-Kutta method . The most involved part of this computation is to find the divergence of the flux of axonal branching , 𝐉i⁢ ( 𝐱 , t ) , the term in parentheses in Equation 2: ( 4 ) ∇⋅Ji ( x , t ) =∇⋅ ( D∇ai−ai∑j=1Mγi , j∇ρj ( x ) +ϵaiN−1∇a^i ) , where a^i≡∑j≠iNaj . Note that the sum of the guidance gradients is time-independent and define 𝐠i⁢ ( 𝐱 ) ≡∑j=1Mγi , j⁢∇⁡ρj⁢ ( 𝐱 ) . Because the divergence operator is distributive , Equation 4 can be expanded using vector calculus identities ( dropping references to 𝐱 and t for clarity ) : ( 5 ) ∇⋅Ji=∇⋅ ( D∇ai ) −∇⋅ ( aigi ) +ϵN−1∇⋅ ( ai∇a^i ) . Applying the vector calculus product rule identity yields ( 6 ) ∇⋅Ji=D∇⋅∇ai−ai∇⋅gi−gi⋅∇ai+ϵaiN−1∇⋅∇a^i+ϵN−1∇a^i⋅∇ai , which has five elements to compute: ( i ) D∇⋅∇ai ( the Laplacian of ai ) ; ( ii ) a time-independent modulator of ai ( because ∇⋅gi is a time-independent static field ) ; ( iii ) the scalar product of the static vector field 𝐠i and the gradient of ai; ( iv ) the Laplacian of a^i; and ( v ) a term involving the gradients of ai and a^i . Each of the divergences can be simplified by means of Gauss’s Theorem following Lee et al . , 2014 . By separating the computation of Equation 4 into parts ( i ) – ( v ) , the no-flux boundary condition , ( 9 ) 𝐉i ( 𝐱 , t ) |boundary=0 , can be fulfilled . On the boundary , the contribution to 𝐉 resulting from the first term of Equation 6 can be fixed to 0 by the ‘ghost cell method’ in which , during the evaluation of ( i ) , a hex outside the boundary containing the same value as the hex inside the boundary is imagined to exist such that the flux of 𝐉 across the boundary is 0 . Then , 𝐠i⁢ ( 𝐱 ) can be tailored so that it , and its normal derivative , approach 0 at the boundary , ensuring that the second and third terms of Equation 6 also contribute nothing to 𝐉 . This is achieved by multiplying 𝐠i⁢ ( 𝐱 ) by a sharp logistic function of the distance , db , from 𝐱 to the boundary , of the form 1/[1+exp⁡ ( 100⁢ ( df-db ) ) ] , where df=0 . 1 mm ≈3⁢d is the boundary fall-off distance . The pattern difference metric , η , incorporates information about the differences in the areas of simulated and experimentally determined ( real ) barrels ( 𝒜isim and 𝒜iexp ) , as well as information contained in the ‘adjacency vector’ for each barrel , 𝒱i , the j-th element of which is the length of the border between barrel i and j . For a well-formed barrel pattern , 𝒱i is a sparse vector . A dimensionless quantity can be obtained from the scalar product of the simulated and experimental adjacency vectors: 1N∑i𝒱isimbisim ⋅ 𝒱iexpbiexp , where for example biexp , is the total length of the border around real barrel i . This quantity is small when the fields that form in simulation have dissimilar neighbour relations to those of the real barrels ( e . g . , Figure 1C , t=1000 ) , and maximal for a precise topological map ( t=10000 ) . A second comparison considers the mean magnitude of the difference between the simulated and experimental vectors: 1N⁢∑i∥𝒱isim-𝒱iexp∥ . This tends to 0 for a perfect match and can separate patterns with straight boundaries from those with ‘noisy edges’ ( as in Figure 3Bii ) . We combined these terms into a single metric: ( 10 ) η=1N∑i‖𝒜isim−𝒜iexp‖×1N∑i‖𝒱isim−𝒱iexp‖1N∑i𝒱isimbisim ⋅ 𝒱iexpbiexp , which has units of mm3 . The code required to reproduce these results is available at https://github . com/ABRG-Models/BarrelEmerge/tree/eLife ( James , 2020 , copy archived at https://github . com/elifesciences-publications/BarrelEmerge ) . The computations described in ( i ) – ( v ) may be found in the class method RD_James_comp2::compute_divJ ( ) which calculates term1 , term2 , term3 , term1_1 and term1_2 , respectively . BarrelEmerge depends on the software library morphologica ( RRID:SCR_018813 ) , which must also be compiled .

How does the brain wire itself up ? One possibility is that a precise genetic blueprint tells every brain cell explicitly how it should be connected to other cells . Another option is that complex patterns emerge from relatively simple interactions between growing cells , which are more loosely controlled by genetic instruction . The barrel cortex in the brains of rats and mice features one of the most distinctive wiring patterns . There , cylindrical clusters of cells – or barrels – are arranged in a pattern that closely matches the arrangement of the whiskers on the face . Neurons in a barrel become active when the corresponding whisker is stimulated . This precise mapping between individual whiskers and their brain counterparts makes the whisker-barrel system ideal for studying brain wiring . Guidance fields are a way the brain can create cell networks with wiring patterns like the barrels . In this case , genetic instructions help to create gradients of proteins across the brain . These help the axons that connect neurons together to grow in the right direction , by navigating towards regions of higher or lower concentrations . A large number of guidance fields could map out a set of centre-point locations for axons to grow towards , ensuring the correct barrel arrangement . However , there are too few known guidance fields to explain how the barrel cortex could form by this kind of genetic instruction alone . Here , James et al . tried to find a mechanism that could create the structure of the barrel cortex , relying only on two simple guidance fields . Indeed , two guidance fields should be enough to form a coordinate system on the surface of the cortex . In particular , it was examined whether the cortical barrel map could reliably self-organize without a full genetic blueprint pre-specifying the barrel centre-points in the cortex . To do so , James et al . leveraged a mathematical model to create computer simulations; these showed that only two guidance fields are required to reproduce the map . However , this was only the case if axons related to different whiskers competed strongly for space while making connections , causing them to concentrate into whisker-specific clusters . The simulations also revealed that the target tissue does not need to specify centre-points if , instead , the origin tissue directs how strongly the axons should respond to the guidance fields . So this model describes a simple way that specific structures can be copied across the central nervous system . Understanding the way the barrel cortex is set up could help to grasp how healthy brains develop , how brain development differs in certain neurodevelopmental disorders , and how brain wiring reorganizes itself in different contexts , for example after a stroke . Computational models also have the potential to reduce the amount of animal experimentation required to understand how brains are wired , and to cast light on how brain wiring is shaped by evolution .

2020

Modelling the emergence of whisker barrels

The majority of mitochondrial proteins are targeted to mitochondria by N-terminal presequences and use the TIM23 complex for their translocation across the mitochondrial inner membrane . During import , translocation through the channel in the inner membrane is coupled to the ATP-dependent action of an Hsp70-based import motor at the matrix face . How these two processes are coordinated remained unclear . We show here that the two domain structure of Tim44 plays a central role in this process . The N-terminal domain of Tim44 interacts with the components of the import motor , whereas its C-terminal domain interacts with the translocation channel and is in contact with translocating proteins . Our data suggest that the translocation channel and the import motor of the TIM23 complex communicate through rearrangements of the two domains of Tim44 that are stimulated by translocating proteins . Mitochondria perform a number of essential cellular functions ranging from production of ATP and diverse other metabolic intermediates to initiation of apoptosis . It is thus not very surprising that disturbances in mitochondrial function are associated with a number of human diseases , including neurodegenerative disorders , diabetes , and various forms of cancer ( Nunnari and Suomalainen , 2012; Quirós et al . , 2015; Youle and van der Bliek , 2012 ) . An essential prerequisite for correctly functioning mitochondria is import of about 1000 different proteins synthesized as precursor proteins in the cytosol . Recent studies revealed that mitochondrial protein import machineries are sensitive indicators of functionality of mitochondria ( Harbauer et al . , 2014; Nargund et al . , 2012; Yano et al . , 2014 ) , demonstrating that a deep understanding of mitochondrial protein import pathways and their regulation will be essential for understanding the role mitochondria have under physiological and pathophysiological conditions . Over half of mitochondrial proteins are synthesized with cleavable , N-terminal extensions called presequences . Import of such precursor proteins requires a coordinated action of the TOM complex in the outer membrane and the TIM23 complex in the inner membrane and is driven by membrane potential across the inner membrane and ATP in the matrix ( Dolezal et al . , 2006; Endo et al . , 2011; Koehler , 2004; Mokranjac and Neupert , 2009; Neupert and Herrmann , 2007; Schulz et al . , 2015; Stojanovski et al . , 2012 ) . The TIM23 complex mediates translocation of presequence-containing precursor proteins into the matrix as well as their lateral insertion into the inner membrane . The latter process requires the presence of an additional , lateral insertion signal . After initial recognition on the intermembrane space side of the inner membrane by the receptors of the TIM23 complex , Tim50 and Tim23 , precursor proteins are transferred to the translocation channel in the inner membrane in a membrane-potential dependent step ( Bajaj et al . , 2014; Lytovchenko et al . , 2013; Mokranjac et al . , 2009; Shiota et al . , 2011; Tamura et al . , 2009 ) . The translocation channel is formed by membrane-integrated segments of Tim23 , together with Tim17 and possibly also Mgr2 ( Alder et al . , 2008; Demishtein-Zohary et al . , 2015; leva et al . , 2014; Malhotra et al . , 2013 ) . At the matrix-face of the inner membrane , precursor proteins are captured by the components of the import motor of the TIM23 complex , also referred to as PAM ( presequence translocase-associated motor ) . Its central component is mtHsp70 whose ATP hydrolysis-driven action fuels translocation of precursor proteins into the matrix ( De Los Rios et al . , 2006; Liu et al . , 2003; Neupert and Brunner , 2002; Schulz and Rehling , 2014 ) . Multiple cycles of mtHsp70 binding to and release from translocating proteins are required for complete translocation across the inner membrane . The ATP hydrolysis-driven cycling of mtHsp70 and thereby its binding to proteins is regulated by the J- and J-like proteins Tim14 ( Pam18 ) and Tim16 ( Pam16 ) as well as by the nucleotide-exchange factor Mge1 ( D'Silva et al . , 2003; Kozany et al . , 2004; Mapa et al . , 2010; Mokranjac et al . , 2006; 2003b; Truscott et al . , 2003 ) . Tim21 and Pam17 are two nonessential components that bind to Tim17-Tim23 core of the TIM23 complex and appear to modulate its activity in a mutually antagonistic manner ( Chacinska et al . , 2005; Popov-Celeketic et al . , 2008; van der Laan et al . , 2005 ) . The translocation channel and the import motor of the TIM23 complex are thought to be coupled by Tim44 , a peripheral inner membrane protein exposed to the matrix ( D'Silva et al . , 2004; Kozany et al . , 2004; Schulz and Rehling , 2014 ) . Like other components of the TIM23 complex , Tim44 is a highly evolutionary conserved protein and is encoded by an essential gene . In mammals , Tim44 has been implicated in diabetes-associated metabolic and cellular abnormalities ( Wada and Kanwar , 1998; Wang et al . , 2015 ) . A novel therapeutic approach using gene delivery of Tim44 has recently shown promising results in mouse models of diabetic nephropathy ( Zhang et al . , 2006 ) . In addition , mutations in Tim44 were identified that predispose carriers to oncocytic thyroid carcinoma ( Bonora et al . , 2006 ) . Understanding the function of Tim44 and its interactions within the TIM23 complex will therefore be essential for understanding how the energy of ATP hydrolysis is converted into unidirectional transport of proteins into mitochondria and may provide clues for therapeutic treatment of human diseases . Tim44 binds to the Tim17-Tim23 core of the translocation channel ( Kozany et al . , 2004; Mokranjac et al . , 2003b ) . Tim44 also binds to mtHsp70 , recruiting it to the translocation channel . The interaction between Tim44 and mtHsp70 is regulated both by nucleotides bound to mtHsp70 as well as by translocating proteins ( D'Silva et al . , 2004; Liu et al . , 2003; Slutsky-Leiderman et al . , 2007 ) . Tim44 is likewise the major site of recruitment of the Tim14-Tim16 subcomplex , recruiting them both to the translocation channel as well as to mtHsp70 ( Kozany et al . , 2004; Mokranjac et al . , 2003b ) . In this way , Tim44 likely ensures that binding of mtHsp70 to the translocating polypeptides , regulated by the action of Tim14 and Tim16 , takes place right at the outlet of the translocation channel in the inner membrane . Tim44 is composed of two domains , depicted as N- and C-terminal domains ( Figure 1A ) . Recent studies suggested that the N-terminal domain is responsible for the majority of known functions of Tim44 . Segments of the N-terminal domain were identified that are important for interaction of Tim44 with Tim16 and with mtHsp70 ( Schilke et al . , 2012; Schiller et al . , 2008 ) . Furthermore , using site-specific crosslinking , residues in the N-terminal domain were crosslinked to the matrix-exposed loop of Tim23 ( Ting et al . , 2014 ) . However , the C-terminal domain of Tim44 shows higher evolutionary conservation . Still , the only function that has so far been attributed to the C-terminal domain is its role in recruitment of Tim44 to cardiolipin-containing membranes ( Weiss et al . , 1999 ) . Based on the crystal structure of the C-terminal domain , a surface-exposed hydrophobic cavity was initially suggested to be important for membrane recruitment ( Josyula et al . , 2006 ) . However , subsequent biochemical studies combined with molecular dynamics simulations , demonstrated that the helices A1 and A2 ( residues 235–262 in yeast Tim44 ) , present in the beginning of the C-terminal domain , are important for membrane recruitment ( Marom et al . , 2009 ) . Deletion of helices A1 and A2 abolished membrane association of the C-terminal domain . Interestingly , attachment of helices A1 and A2 to a soluble protein was sufficient to recruit it to a model membrane ( Marom et al . , 2009 ) . 10 . 7554/eLife . 11897 . 003Figure 1 . The function of Tim44 can be rescued by its two domains expressed in trans but not by either of the domains alone . ( A ) Schematic representation of Tim44 domain structure ( numbering according to yeast Tim44 sequence ) . pre . - presequence ( B and C ) A haploid yeast deletion strain of TIM44 carrying the wild-type copy of TIM44 on a URA plasmid was transformed with centromeric plasmids carrying indicated constructs of Tim44 under control of endogenous promoter and 3'UTR . Cells were plated on medium containing 5-fluoroorotic acid and incubated at 30°C . The plasmid carrying wild-type Tim44 and an empty plasmid were used as positive and negative controls , respectively . ( D ) Total cell extracts of wild-type yeast cells transformed with plasmids coding for indicated Tim44 constructs under GPD promoter were analysed by SDS–PAGE and immunoblotting against depicted antibodies . * , ** and *** - protein bands detected with antibodies raised against full-length Tim44 . DOI: http://dx . doi . org/10 . 7554/eLife . 11897 . 00310 . 7554/eLife . 11897 . 004Figure 1—figure supplement 1 . Two domains of Tim44 do not interact stably with each other . ( A ) Purified His6-Tim44 ( 43–263 ) was incubated with purified Tim44 ( 211–431 ) either in low-salt ( 20 mM Tris/HCl , 50 mM NaCl , 10 mM imidazole , pH 8 . 0 ) or high-salt buffer ( 20 mM Tris/HCl , 300 mM NaCl , 10 mM imidazole , pH 8 . 0 ) for 5 min at 25°C . The NiNTA-agarose beads were added and the mixture gently rolled for 30 min at 4°C . After three washing steps with the same buffer , bound proteins were eluted with the buffer containing 300 mM imidazole . Total ( T , 10% ) , flow-through ( FT , 10% ) , and bound ( B , 100% ) fractions were analyzed by SDS–PAGE followed by Coomassie staining . ( B ) Mitochondria were isolated from yeast cells in which the function of the full-length Tim44 was rescued by coexpression of N- and C-terminal domains separately ( N+C ) . In His9N+C mitochondria , the N-terminal domain contained an additional His9 tag . Mitochondria were solubilized with digitonin-containing buffer and incubated with NiNTA-agarose beads at 4°C . After three washing steps , proteins specifically bound to the beads were eluted with Laemmli buffer containing 300 mM imidazole . Total ( T , 10% ) , flow-through ( FT , 10% ) , and bound ( B , 100% ) fractions were analyzed by SDS–PAGE followed by imunoblotting using antibodies to Tim44 . DOI: http://dx . doi . org/10 . 7554/eLife . 11897 . 004 We report here that the function of the full-length Tim44 cannot be rescued by its N-terminal domain extended to include membrane-recruitment helices of the C-terminal domain , demonstrating an unexpected essential function of the core of the C-terminal domain . Surprisingly , we observed that the two domains of Tim44 , when expressed in trans , can support , although poorly , growth of yeast cells , giving us a tool to dissect the role of the C-terminal domain in vivo . We identify the C-terminal domain of Tim44 as the domain of Tim44 that is in contact with translocating proteins and that directly interacts with Tim17 , a component of the translocation channel . Our data suggest that intricate rearrangements of the two domains of Tim44 are required during transfer of translocating precursor proteins from the channel in the inner membrane to the ATP-dependent motor at the matrix face . We reasoned that if all important protein–protein interactions of Tim44 are mediated by its N-terminal domain and the only function of the C-terminal domain is to recruit Tim44 to the membrane , then a construct consisting of the N-terminal domain , extended to include the membrane-recruitment helices A1 and A2 , should suffice to support the function of the full-length protein . To test this hypothesis , we cloned such a construct in a yeast expression plasmid and transformed it into a Tim44 plasmid shuffle yeast strain . Upon incubation of transformed cells on a medium containing 5-fluoroorotic acid to remove the URA plasmid carrying the wild-type , full-length copy of Tim44 , no viable cells were obtained ( Figure 1B ) . A plasmid carrying the full-length copy of Tim44 enabled growth of yeast cells , whereas no viable colonies were obtained when an empty plasmid was used , confirming the specificity of the assay . We conclude that the N-terminal domain of Tim44 , even when extended to include the membrane-recruitment helices of the C-terminal domain , is not sufficient to support the function of the full-length protein . Furthermore , this result suggests that the C-terminal domain of Tim44 has a function beyond membrane recruitment that is apparently essential for viability of yeast cells . We then tested whether the function of Tim44 can be rescued by its two domains expressed in trans . Two plasmids , each encoding one of the two domains of Tim44 and both including A1 and A2 helices , were co-transformed into a Tim44 plasmid shuffle yeast strain and analyzed as above . Surprisingly , we obtained viable colonies when both domains were expressed in the same cell but not when either of the two domains was expressed on its own ( Figure 1C ) . The rescue was dependent on the presence of A1 and A2 helices on both domains ( data not shown ) , as in their absence neither of the domains could even be stably expressed in yeast ( Figure 1D ) . It is possible that the two domains of Tim44 , both carrying A1 and A2 helices , bind to each other with high affinity and therefore are able to re-establish the full-length protein from the individual domains . To test this possibility , we expressed both domains recombinantly , purified them and analyzed , in a pull down experiment , if they interact with each other . The N-terminally His-tagged N-terminal domain efficiently bound to NiNTA-agarose beads under both low- and high-salt conditions ( Figure 1—figure supplement 1A ) . However , we did not observe any copurification of the non-tagged C-terminal domain . We also did not observe any stable interaction of the two domains when digitonin-solubilized mitochondria containing a His-tagged version of the N-terminal domain were used in a NiNTA pull-down experiment ( Figure 1—figure supplement 1B ) . Thus , the two domains of Tim44 appear not to stably interact with each other . We compared growth rate of the yeast strain carrying the wild-type , full-length version of Tim44 ( FL ) with that of the strain having two Tim44 domains , both containing A1 and A2 helices , expressed in trans , for simplicity reasons named from here on N+C . The N+C strain was viable and grew relatively well on a fermentable carbon source at 24°C and 30°C ( Figure 2A ) . Still , its growth was slower than that of the FL strain at both temperatures . At 37°C , the N+C strain was barely viable . On a nonfermentable carbon source , when fully functional mitochondria are required , N+C did not grow at any of the temperatures tested . Thus , the function of Tim44 can be reconstituted from its two domains separately , although only very poorly . 10 . 7554/eLife . 11897 . 005Figure 2 . N+C cells grow poorly , even on fermentable carbon source . ( A ) Ten-fold serial dilutions of △tim44 cells rescued by the wild-type , full-length copy of Tim44 ( FL ) or by its two domains expressed in trans ( N+C ) were spotted on rich medium containing glucose ( YPD ) or lactate ( YPLac ) , as fermentable and non-fermentable carbon sources , respectively . Plates were incubated at indicated temperatures for 2 ( YPD ) or 3 days ( YPLac ) . ( B ) 15 and 35 µg of mitochondria isolated from FL and N+C cells were analyzed by SDS–PAGE , followed by immunoblotting against depicted mitochondrial proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 11897 . 005 We isolated mitochondria from FL and N+C strains grown on fermentable medium and compared their mitochondrial protein profiles . Immunostaining with antibodies raised against full-length Tim44 detected no full-length protein in N+C mitochondria but rather two faster migrating bands ( Figure 2B ) . Based on the running behavior of the individual domains seen in Figure 1D , the slower migrating band corresponds to the N domain and the faster migrating one to the C domain . This confirms that , surprisingly , the full-length Tim44 is indeed not absolutely required for viability of yeast cells . The endogenous levels of other components of the TIM23 complex were either not changed at all ( Tim17 , Tim23 , and Tim50 ) , or were slightly upregulated ( mtHsp70 , Tim14 , and Tim16 ) , likely to compensate for only poorly functional Tim44 . Levels of components of other essential mitochondrial protein translocases of the outer and inner mitochondrial membranes , Tom40 , Tob55 , and Tim22 , were not altered compared to FL mitochondria . Similarly , we observed no obvious differences in endogenous levels of proteins present in the outer membrane , intermembrane space , inner membrane , and the matrix that we analyzed . We conclude that Tim44 can be split into its two domains that are sufficient to support the function of the full-length protein , although only poorly . Considering the essential role of Tim44 during translocation of precursor proteins into mitochondria , we tested whether the severe growth defect of the N+C strain is due to compromised mitochondrial protein import . When import of precursor proteins into mitochondria is impaired , a precursor form of matrix-localized protein Mdj1 accumulates in vivo ( Waegemann et al . , 2015; Wrobel et al . , 2015 ) . We indeed observed a very prominent band of the precursor form of Mdj1 in total cell extracts of N+C cells , grown at 24°C and 30°C , that was absent in cells containing full-length Tim44 ( Figure 3A ) . Thus , the efficiency of protein import into mitochondria is reduced in N+C cells . 10 . 7554/eLife . 11897 . 006Figure 3 . N+C cells have a strongly impaired import via the TIM23 complex . ( A ) Total cell extracts of FL and N+C cells grown at 24°C and 30°C were analyzed by SDS–PAGE and immunoblotting using indicated antibodies . p - precursor , and m - mature form of Mdj1 . ( B–G and I–J ) 35S-labeled mitochondrial precursor proteins were imported into mitochondria isolated from FL and N+C cells . After indicated time periods , aliquots were removed and Proteinase K ( PK ) was added where indicated . Samples were analyzed by SDS–PAGE , autoradiography and quantification of PK-protected mature forms of imported proteins . pF1β - precursor of the β subunit of FoF1 ATPase . pcytb2 ( 1–167 ) △DHFR - precursor consisting of the first 167 residues with the deleted sorting signal of yeast cytochrome b2 fused to mouse dihydrofolate reductase ( DHFR ) ; pSu9 ( 1–69 ) DHFR - matrix targeting signal ( residues 1–69 ) of subunit 9 of FoF1 ATPase from Neurospora crassa fused to DHFR; pOxa1 - precursor of Oxa1; pDLD1 - precursor of D-lactate dehydrogenase; pcytb2 - precursor of cytochrome b2; AAC - precursor of ATP/ADP carrier; p , i , m - precursor , intermediate , and mature forms of imported proteins; * - in vitro translation product starting from an internal methionine . ** - clipped form of Tim23 . ( H ) Membrane potential of isolated mitochondria was measured using DiSC3 ( 5 ) . Valinomycin was added to dissipate membrane potential . DOI: http://dx . doi . org/10 . 7554/eLife . 11897 . 006 To analyze protein import in N+C mitochondria in more detail , we performed in vitro protein import into isolated mitochondria ( Figure 3B–G , I–J ) . To this end , various mitochondrial precursor proteins were synthesized in vitro in the presence of [35S]-methionine and incubated with isolated mitochondria . The import efficiencies of all matrix-targeted precursors analyzed , pF1β , pcytb2 ( 1–167 ) △DHFR , and pSu9 ( 1–69 ) DHFR , were drastically reduced in N+C mitochondria when compared to wild type . Import of presequence-containing precursor of Oxa1 that contains multiple transmembrane segments was similarly impaired . Likewise , precursor proteins that are laterally inserted into the inner membrane by the TIM23 complex , such as pDLD1 and pcytb2 , were imported with reduced efficiency into N+C mitochondria . In agreement with the established role of Tim44 in import of precursors of a number of components of respiratory chain complexes and their assembly factors , we observed a slightly reduced membrane potential in N+C mitochondria as compared to wild type ( Figure 3H ) . However , precursors of ATP/ADP carrier and of Tim23 , whose imports into mitochondria are not dependent on the TIM23 complex , were imported with similar efficiencies in both types of mitochondria , demonstrating that observed effects are not due to general dysfunction of mitochondria . We conclude that splitting of Tim44 into two domains in N+C cells severely impairs transport of proteins by the TIM23 complex , suggesting that full-length Tim44 is required for efficient import of presequence-containing precursor proteins into mitochondria . Tim44 is thought to play an important role in connecting the translocation channel and the import motor of the TIM23 complex . We thus reasoned that disassembly of the TIM23 complex in N+C mitochondria might be a reason for its reduced functionality . When wild-type mitochondria are solubilized with digitonin , affinity-purified antibodies to Tim17 and to Tim23 essentially deplete both Tim17 and Tim23 from the mitochondrial lysate and precipitate part of Tim50 , Tim44 , Tim14 , and Tim16 ( Figure 4 ) . Similarly , affinity-purified antibodies to Tim16 deplete both Tim16 and Tim14 and precipitate Tim50 , Tim17 , Tim23 , and Tim44 from mitochondrial lysate . We observed essentially the same precipitation pattern when we analyzed digitonin-solubilized N+C mitochondria , demonstrating that the TIM23 complex is properly assembled . Importantly , both N and C domains of Tim44 were recruited to the TIM23 complex . 10 . 7554/eLife . 11897 . 007Figure 4 . The TIM23 complex is assembled in N+C mitochondria . Mitochondria from FL and N+C cells were solubilized with digitonin-containing buffer and mitochondrial lysates incubated with affinity-purified antibodies to Tim17 , Tim23 , and Tim16 prebound to Protein A-Sepharose beads . Antibodies from preimmune serum ( PI ) were used as a negative control . After three washing steps , material specifically bound to the beads was eluted with Laemmli buffer . Total ( 20% ) , supernatant ( Sup , 20% ) , and bound ( Pellet , 100% ) fractions were analyzed by SDS–PAGE and immunoblotting with indicated antibodies . DOI: http://dx . doi . org/10 . 7554/eLife . 11897 . 007 Since the assembly of the TIM23 complex is not affected in N+C mitochondria , we reasoned that an altered conformational flexibility may be a reason behind its reduced function in N+C cells . Chemical crosslinking is currently the most sensitive assay available to analyze the conformation of the TIM23 complex in intact mitochondria . We thus compared the crosslinking patterns of TIM23 subunits in N+C mitochondria to those in FL . In wild-type mitochondria , Tim16 can be crosslinked to mtHsp70 , Tim44 , and Tim14 in an ATP-dependent manner ( Figure 5A ) . In N+C mitochondria , the same crosslinks of Tim16 to mtHsp70 and to Tim14 were observed . The crosslink to Tim44 was , as expected , absent in N+C mitochondria and another crosslink to a smaller protein appeared . In addition , a crosslink between two Tim16 molecules became prominent . Interestingly , this crosslink has previously been observed in mutants in which conformation of the TIM23 complex was altered ( Popov-Čeleketić et al . , 2008 ) . Similarly , we observed prominent changes in crosslinking pattern of the channel component Tim23 ( Figure 5B ) . In addition to the crosslink of Tim23 to Pam17 , observed in both FL and N+C mitochondria , a prominent Tim23-dimer crosslink appeared in N+C mitochondria . 10 . 7554/eLife . 11897 . 008Figure 5 . The TIM23 complex adopts an altered conformation in N+C mitochondria . ( A and B ) Mitochondria from FL and N+C cells were incubated with amino group-specific crosslinker disuccinimidyl glutarate ( DSG ) . Where indicated , mitochondrial ATP levels were altered prior to crosslinking . After quenching of excess crosslinker , mitochondria were reisolated and analyzed by SDS–PAGE followed by immunoblotting with antibodies to Tim16 ( A ) and Tim23 ( B ) . * indicates currently uncharacterized crosslinks . ( C ) Mitochondria from FL and N+C cells were solubilized in digitonin-containing buffer and analyzed by BN-PAGE and immunoblotting with indicated antibodies . DOI: http://dx . doi . org/10 . 7554/eLife . 11897 . 008 To obtain an independent evidence that the conformation of the TIM23 complex is affected in N+C mitochondria , we analyzed the complex by blue native gel electrophoresis . When digitonin-solubilized wild-type mitochondria are separated by BN-PAGE , Tim17 , and Tim23 are present in a 90 kDa complex and , to a lesser degree , in higher molecular weight complexes that additionally contain Tim21 and Mgr2 ( Chacinska et al . , 2005; Ieva et al . , 2014 ) . In contrast , with digitonin-solubilized N+C mitochondria , antibodies to Tim17 and Tim23 revealed slightly shifted bands , in particular of the 90 kDa complex ( Figure 5C ) . Since the 90 kDa complex does not contain any other known subunit of the TIM23 complex , this finding further supports the above notion that the conformation of the translocation channel is changed in N+C mitochondria . We observed no obvious difference in the ca . 60 kDa Tim14-Tim16 complex between FL and N+C mitochondria . As expected , full-length Tim44 , present in FL mitochondria , was absent in N+C mitochondria ( Figure 5C ) . Together , these results demonstrate that the conformation of the TIM23 complex is changed in N+C mitochondria . They further show that alterations in the components traditionally assigned to the import motor affect the conformation of the translocation channel in the inner membrane , supporting the notion of an intricate crosstalk within the complex . The data presented so far suggest that full-length Tim44 is required for optimal conformational dynamics of the TIM23 complex . Furthermore , they suggest that the C-terminal domain has an essential function within the TIM23 complex , beyond mere membrane recruitment . So , what is the function of the C-terminal domain of Tim44 ? We first searched for binding partners of the individual domains . To that end , we recombinantly expressed and purified full-length Tim44 as well as its two domains ( Figure 6A ) . To look for interaction partners of the core domains , both domains now lacked the segment containing A1 and A2 helices . Purified proteins were covalently coupled to the Sepharose beads and were subsequently incubated with mitochondrial lysates . Mitochondria were solubilized with Triton X-100 that , unlike digitonin , dissociates the TIM23 complex into its individual subunits ( except for the Tim14-Tim16 subcomplex that remains stable ) . In this way , direct protein-protein interactions can be analyzed . We observed prominent , specific binding of mtHsp70 , Tim16 , Tim14 and Tim17 , and to a far lesser degree of Tim23 and Tim50 , to full-length Tim44 ( Figure 6B ) . None of the proteins bound to empty beads . Also , we observed no binding of two abundant mitochondrial proteins , porin , and F1βß , demonstrating the specificity of observed interactions . mtHsp70 , Tim16 and Tim14 also efficiently bound to the N-terminal domain of Tim44 , in agreement with previous observations ( Schilke et al . , 2012; Schiller et al . , 2008 ) , and far less efficiently to the C-terminal domain . Since the Tim14-Tim16 subcomplex remains stable in Triton X-100 , it is not possible by this method to distinguish which of the two subunits , or maybe even both , directly interacts with the N-terminal domain of Tim44 . Binding of Tim17 to the N-terminal domain of Tim44 was drastically lower compared to its binding to the full-length protein . Instead , a strong binding of Tim17 to the C-terminal domain of Tim44 was observed . 10 . 7554/eLife . 11897 . 009Figure 6 . C-terminal domain of Tim44 interacts with Tim17 and with a precursor in transit . ( A ) Coomassie-stained SDS-PA gel of recombinantly expressed and purified constructs of Tim44 . FL - full-length , mature Tim44 ( residues 43–431 ) ; N - a construct encompassing the N-terminal domain of Tim44 ( residues 43–209 ) ; Cc - a construct encompassing the core of the C-terminal domain of Tim44 ( residues 264–431 ) . ( B ) Wild-type mitochondria were solubilized with Triton X-100 and incubated with indicated purified constructs of Tim44 covalently coupled to CNBr-Sepharose beads . Beads with no coupled protein were used as a negative control . After washing steps , proteins specifically bound to the beads were eluted by Laemmli buffer and analyzed by SDS–PAGE followed by immunoblotting with the indicated antibodies . Input lane contains 4 . 5% of the material used for binding ( upper panel ) . Binding of mtHsp70 , as a representative of the import motor components , and of Tim17 to different beads was quantified from three independent experiments ( lower panel ) . Binding to FL was set to 1 . ( C ) Antibodies specific for N and Cc domains of Tim44 were affinity purified from rabbit serum raised against full-length Tim44 using respective domains of Tim44 covalently coupled to Sepharose beads , as described under ( B ) . To test the specificity of purified antibodies , indicated Tim44 constructs were loaded on an SDS-PA gel , blotted on a nitrocellulose membrane and obtained membranes were immunoblotted using the purified antibodies , as indicated . ( D ) 35S-labelled matrix targeted precursor protein pcytb2 ( 1–167 ) ∆DHFR was imported into isolated mitochondria from FL and N+C cells in the presence of methotrexate , leading to its arrest as a TOM-TIM23 spanning intermediate . Samples were then crosslinked with disuccinimidyl suberate ( DSS ) , where indicated . After quenching of excess crosslinker , aliquots were taken out for 'total' and the rest of samples solubilized in SDS-containing buffer to dissociate all noncovalent protein–protein interactions . Solubilized material was incubated with indicated affinity-purified antibodies prebound to Protein A-Sepharose beads . Antibodies from preimmune serum ( PI ) were used as a negative control . Material specifically bound to the beads was eluted with Laemmli buffer and analyzed by SDS–PAGE and autoradiography . p - precursor and m - mature forms of pcytb2 ( 1–167 ) ∆DHFR . ( E ) Melting curves of recombinant wild type and Pro282Gln mutant of Tim44 obtained by thermal shift assay . DOI: http://dx . doi . org/10 . 7554/eLife . 11897 . 009 We conclude that the N-terminal domain of Tim44 binds to the components of the import motor , whereas the C-terminal domain binds to the translocation channel in the inner membrane , revealing a novel function of the C-terminal domain of Tim44 . We then asked which of the two domains of Tim44 is in contact with translocating proteins . To answer this question , we first affinity-purified antibodies that specifically recognize cores of the individual domains of Tim44 using the above described Sepharose beads . The antibodies , affinity purified using beads with coupled full-length Tim44 , recognized full-length Tim44 as well as both of its domains ( Figure 6C ) . In contrast , antibodies that were affinity purified using beads with coupled individual domains recognized only the respective domain and the full-length protein ( Figure 6C ) . This demonstrates that we indeed purified antibodies specific for individual domains of Tim44 . Next , we accumulated 35S-labelled precursor protein pcytb2 ( 1–167 ) △DHFR as a TOM-TIM23-spanning intermediate . Briefly , this precursor protein consists of the first 167 residues of yeast cytochrome b2 , with a 19 residue deletion in its lateral insertion signal , fused to the passenger protein dihydrofolate reductase . In the presence of methotrexate , that stabilizes folded DHFR , the b2 part reaches the matrix , whereas the DHFR moiety remains on the mitochondrial surface resulting in an intermediate that spans both TOM and TIM23 complexes . The association of Tim44 and its domains with the arrested precursor protein was analyzed by chemical crosslinking followed by immunoprecipitation with antibodies to full-length Tim44 and its individual domains . In wild-type mitochondria , all three antibodies precipitated a crosslinking adduct of Tim44 to the arrested precursor protein , demonstrating that they are all able to immunoprecipitate the respective antigens ( Figure 6D ) . In contrast , with N+C mitochondria , a faster migrating crosslinking adduct of a Tim44 domain to the arrested precursor protein was immunoprecipitated with the antibodies against the C-terminal domain and against the full-length protein but not with the antibodies against the N-terminal domain . This demonstrates that the C-terminal domain of Tim44 is in close vicinity of the translocating protein . Mutations identified in human patients can frequently point to functionally important residues in affected proteins . In this respect , Pro308Gln mutation in human Tim44 has recently been linked to oncocytic thyroid carcinoma ( Bonora et al . , 2006 ) . Since the mutation maps to the C-terminal domain of Tim44 , we wanted to analyze functional implications of this mutation and therefore made the corresponding mutation in yeast Tim44 ( Pro282Gln ) . We compared thermal stabilities of wild type and mutant Tim44 proteins by thermal shift assay . The melting temperature of wild-type Tim44 was 54°C , whereas that of the mutant protein was 4°C lower ( Figure 6E ) . This demonstrates that the mutation significantly destabilizes Tim44 , providing first clues toward molecular understanding of the associated human disease . The major question of protein import into mitochondria that has remained unresolved is how translocation of precursor proteins through the channel in the inner membrane is coupled to the ATP-dependent activity of the Hsp70-based import motor at the matrix face of the inner membrane . Results presented here demonstrate that the two domain structure of Tim44 is essential during this process . We show here that the two domains of Tim44 have different interaction partners within the TIM23 complex . In this way , Tim44 holds the TIM23 complex together . Our data revealed a direct , previously unexpected interaction between the C-terminal domain of Tim44 with the channel component Tim17 . This result not only assigned a novel function to the C-terminal domain of Tim44 but also shed new light on Tim17 , the component of the TIM23 complex that has been notoriously difficult to analyze . Recent mutational analysis of the matrix exposed loop between transmembrane segments 1 and 2 of Tim17 revealed no interaction site for Tim44 ( Ting et al . , 2014 ) , suggesting its presence in another segment of the protein . Our data also confirmed the previously observed interactions of the N-terminal domain of Tim44 with the components of the import motor ( Schilke et al . , 2012; Schiller et al . , 2008 ) . We did , however , not observe any direct interaction between Tim23 and the N-terminal domain of Tim44 that has previously been seen by crosslinking in intact mitochondria ( Ting et al . , 2014 ) . It is possible that this crosslinking requires a specific conformation of Tim23 only adopted when Tim23 is bound to Tim17 in the inner membrane . This notion is supported by our previous observation that the stable binding of Tim44 to the translocation channel requires assembled Tim17-Tim23 core of the TIM23 complex ( Mokranjac et al . , 2003b ) . We observed a direct Tim17-Tim44 interaction here probably because of a high local concentration of the C-terminal domain when bound to the beads . The core of the C-terminal domain is preceded by a segment that contains two amphipathic , membrane-recruitment helices . This central segment connects the two domains of Tim44 . Intriguingly , the two currently available crystal structures of the C-terminal domains of yeast and human Tim44s showed different orientations of the two helices relative to the core domains ( Handa et al . , 2007; Josyula et al . , 2006 ) . The conformational change was likely induced upon PEG binding to this region of human Tim44 during crystallization ( Handa et al . , 2007 ) . It is tempting to speculate that the same conformational change takes place during translocation of proteins in the mitochondria . Such a conformational change would not only reorient the two helices in respect to the core of the C-domain but also change the relative orientation of N- and C-terminal domains . Since the two domains have different interaction partners within the TIM23 complex , such a change could rearrange the entire complex . The importance of this proposed conformational change in Tim44 is supported by the data presented here . The function of the full-length Tim44 could be reconstituted from its individual domains only very poorly . Also , there is obviously a very strong evolutionary pressure to keep the two domains of Tim44 within one polypeptide chain . N+C strain had to be kept at all times on the selective medium - even after only an overnight incubation on a nonselective medium the full-length protein reappeared ( our unpublished observation ) , likely due to a recombination event between two plasmids . Tim44 can be crosslinked to translocating proteins . Our data revealed that it is the C-terminal domain of Tim44 that interacts with proteins entering the matrix from the translocation channel in the inner membrane . A direct interaction of the same domain with Tim17 would optimally position the C-terminal domain to the outlet of the translocation channel . This raises an interesting possibility that translocating precursor proteins may play an important role in the above postulated conformational changes of Tim44 . A missense mutation Pro308Gln in human Tim44 is associated with familial oncocytic thyroid carcinoma . The corresponding mutation in yeast , Pro282Gln , destabilized the protein but produced no obvious growth phenotype or an in vivo import defect ( our unpublished observations ) , suggesting that the yeast system is more robust . This observation is in agreement with the notion that mutations that would severely affect the function of the TIM23 complex would likely be embryonically lethal in humans . Still , the disease caused by a mutation in the C-terminal domain of human Tim44 speaks for an important role of this domain in the function of the entire TIM23 complex . Furthermore , the mutation maps to the short loop between A3 and A4 helices in the C-terminal domain of Tim44 . Based on the crystal structure of Tim44 , it was previously suggested that the mutation could affect the conformational flexibility of the A1 and A2 helices ( Handa et al . , 2007 ) , intriguingly providing further support for the above postulated conformational changes of Tim44 . Based on the previously available data and the results presented here , we put forward the following model to describe how translocation of precursor proteins through the channel in the inner membrane is coupled to their capture by the ATP-dependent import motor at the matrix face of the channel ( Figure 7 ) . Tim44 plays a central role in this model . We envisage that two domains of Tim44 are connected by the central segment that contains membrane-recruitment helices , like two cherries on the stalks ( Figure 7 insert ) . This central segment of Tim44 recruits the protein to the cardiolipin-containing membranes . There , through direct protein–protein interactions , the C-terminal domain of Tim44 binds to Tim17 and the N-terminal domain to mtHsp70 and to Tim14-Tim16 subcomplex ( 1 ) . In this way , Tim44 functions as a central platform that connects the translocation channel in the inner membrane with the import motor at the matrix face . Additional interactions likely stabilize the complex , in particular that between the N-terminal domain of Tim44 and Tim23 ( Ting et al . , 2014 ) as well as the one between Tim17 and the IMS-exposed segment of Tim14 ( Chacinska et al . , 2005 ) . In the resting state , the translocation channel is closed to maintain the permeability barrier of the inner membrane . During translocation of proteins ( 2 ) , the translocation channel in the inner membrane has to open to allow passage of proteins . Opening of the channel will likely change the conformation of Tim17 that could be further conveyed to the C-terminal domain Tim44 . It is tempting to speculate that this conformational change is transduced to the N-terminal domain of Tim44 through the central , membrane-bound segment of Tim44 , leading to relative rearrangements of the two domains of Tim44 . This change would now allow Tim14-Tim16 complex to stimulate the ATPase activity of mtHsp70 leading to stable binding of the translocating protein to mtHsp70 . mtHsp70 , with bound polypeptide , will then move into the matrix , opening a binding site on Tim44 for another molecule of mtHsp70 ( 3 ) . We speculate that the release of mtHsp70 with bound polypeptide from the N-terminal domain of Tim44 will send a signal back to the C-terminal domain of Tim44 and further to the translocation channel . Multiple cycles of mtHsp70 are required to translocate the entire polypeptide chain into the matrix . Once the entire polypeptide has been translocated , the translocation channel will revert to its resting , closed state , bringing also Tim44 back to its resting conformation ( 1 ) . Thus , the translocation channel in the inner membrane and the mtHsp70 system at the matrix face communicate with each other through rearrangements of the two domains of Tim44 that are stimulated by translocating polypeptide chain . 10 . 7554/eLife . 11897 . 010Figure 7 . A proposed model of function of the TIM23 complex . See text for details . For simplicity reasons , only essential subunits of the complex are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 11897 . 010 Wild-type haploid yeast strain YPH499 was used for all genetic manipulations . A Tim44 plasmid shuffling yeast strain was made by transforming YPH499 cells with a pVT-102U plasmid ( URA marker ) containing a full-length TIM44 followed by replacement of the chromosomal copy of TIM44 with a HIS3 cassette by homologous recombination . For complementation analyzes , endogenous promoter , mitochondrial presequence ( residues 1–42 ) and the 3’-untranslated region of TIM44 were cloned into centromeric yeast plasmids pRS315 ( LEU marker ) and pRS314 ( TRP marker ) and obtained plasmids subsequently used for cloning of various Tim44 constructs . The following constructs were used in the analyzes: Tim44 ( 43–209 ) , Tim44 ( 43–262 ) , Tim44 ( 264–431 ) , and Tim44 ( 210–431 ) . The constructs encompassing the N- and the C-terminal domains of Tim44 were cloned into pRS315 and pRS314 plasmids , respectively . Plasmids carrying the full-length copy of TIM44 were used as positive controls and empty plasmids as negative ones . A Tim44 plasmid shuffling yeast strain was transformed with two plasmids simultaneously and selected on selective glucose medium lacking respective markers . Cells that lost the wild-type copy of Tim44 on the URA plasmid were selected on medium containing 5-fluoroorotic acid at 30°C . For expression in the wild-type background , the above-described constructs of Tim44 , containing endogenous Tim44 presequence , were also cloned into centromeric yeast plasmids p414GPD and p415GPD for expression under the control of the strong GPD promoter . Cells were grown on selective lactate medium containing 0 . 1% glucose . FL and N+C cells were grown in selective glucose medium at 30°C , unless otherwise indicated , and mitochondria were isolated from cells in logarithmic growth phase . DNA sequences coding for various segments of Tim44 were cloned into bacterial expression vector pET-Duet1 introducing a TEV cleavage site between the His6-tag and the protein coding region . The following Tim44 constructs were cloned: Tim44 ( 43–431 ) ( full-length protein lacking the mitochondrial presequence ) , Tim44 ( 43–209 ) ( referred to as N in Figure 6A ) , Tim44 ( 43–263 ) , Tim44 ( 211–431 ) , and Tim44 ( 264–431 ) ( referred to as Cc in Figure 6A ) . Pro282Gln mutation was introduced into the full-length construct using site directed mutagenesis . Proteins were expressed in E . coli BL21 ( DE3 ) at 37°C and purified using affinity chromatography on NiNTA-agarose beads ( Qiagen , Germany ) followed by gel filtration on Superdex 75 column ( GE Healthcare , Germany ) . Unless otherwise indicated , the His6-tags were removed by incubation with the TEV protease . The purified proteins were stored at -80oC in 20 mM HEPES/KOH , 200 mM KCl , 5 mM MgCl2 , pH 7 . 5 , until use . Purified proteins were coupled to CNBr-Sepharose beads ( GE Healthcare , Germany ) according to manufacturer's instructions and stored at 4°C . The beads were used for purification of domain-specific antibodies from the serum raised in rabbits against recombinantly expressed full-length Tim44 . For direct binding analysis , mitochondria isolated from wild-type yeast cells were solubilized with 0 . 5% Triton X-100 in 20 mM Tris/HCl , pH 8 . 0 , 80 mM KCl , 10% glycerol at 1 mg/mL and incubated with Tim44 constructs coupled to CNBr-Sepharose beads for 30 min at 4oC . After three washing steps , specifically bound proteins were eluted with Laemmli buffer . Samples were analyzed by SDS–PAGE and immunoblotting . Thermal stabilities of wild type and P282Q mutant form of Tim44 were analyzed by fluorescence thermal shift assay ( Müller et al . , 2015 ) . Recombinant proteins ( 6 . 2 µM ) in 20 mM HEPES/NaOH , 150 mM NaCl , pH 7 . 1 were mixed with 5x SYPRO Orange and melting curves analyzed in a real-time PCR machine using a gradient from 5°C to 99°C . Three technical replicates of two independent protein purifications were analyzed in parallel . Mutant Tim44 showed significantly decreased thermal stability under all conditions analyzed - in buffers containing different salt concentrations ( 50 , 150 , and 450 mM ) as well as in different buffers and pHs ( HEPES buffer at pH 7 . 1 and phosphate buffer at pH 8 . 0 ) . Previously published procedures were used for protein import into isolated mitochondria , crosslinking , coimmunoprecipitations and arrest of mitochondrial precursor proteins as TOM-TIM23 spanning intermediates followed by crosslinking and immunoprecipitation under denaturing conditions ( Mokranjac et al . , 2003a; 2003b; Popov-Čeleketić et al . , 2008 ) .

Human , yeast and other eukaryotic cells contain compartments called mitochondria . These compartments are surrounded by two membranes and are most famous for their essential role in supplying the cell with energy . While mitochondria can make a few of their own proteins , the vast majority of mitochondrial proteins are produced elsewhere in the cell and are subsequently imported into mitochondria . During the import process , most proteins need to cross both mitochondrial membranes . Many mitochondrial proteins are transported across the inner mitochondrial membrane by a molecular machine called the TIM23 complex . The complex forms a channel in the inner membrane and contains an import motor that drives the movement of mitochondrial proteins across the membrane . However , it is not clear how the channel and import motor are coupled together . There is some evidence that a protein within the TIM23 complex called Tim44 – which is made of two sections called the N-terminal domain and the C-terminal domain – is responsible for this coupling . It has been suggested that mainly the N-terminal domain of Tim44 is required for this role . Banerjee et al . used biochemical techniques to study the role of Tim44 in yeast . The experiments show that both the N-terminal and C-terminal domains are essential for its role in transporting mitochondrial proteins . The N-terminal domain interacts with the import motor , whereas the C-terminal domain interacts with the channel and the mitochondrial proteins that are being moved . Banerjee et al . propose a model of how the TIM23 complex works , in which the import of proteins into mitochondria is driven by rearrangements in the two domains of Tim44 . A future challenge is to understand the nature of these rearrangements and how they are influenced by other components of the TIM23 complex .

2015

Protein translocation channel of mitochondrial inner membrane and matrix-exposed import motor communicate via two-domain coupling protein

Automated detection of complex animal behaviors remains a challenging problem in neuroscience , particularly for behaviors that consist of disparate sequential motions . Grooming is a prototypical stereotyped behavior that is often used as an endophenotype in psychiatric genetics . Here , we used mouse grooming behavior as an example and developed a general purpose neural network architecture capable of dynamic action detection at human observer-level performance and operating across dozens of mouse strains with high visual diversity . We provide insights into the amount of human annotated training data that are needed to achieve such performance . We surveyed grooming behavior in the open field in 2457 mice across 62 strains , determined its heritable components , conducted GWAS to outline its genetic architecture , and performed PheWAS to link human psychiatric traits through shared underlying genetics . Our general machine learning solution that automatically classifies complex behaviors in large datasets will facilitate systematic studies of behavioral mechanisms . Behavior , the primary output of the nervous system , is complex , hierarchical , dynamic , and high dimensional ( Gomez-Marin et al . , 2014 ) . Precise approaches to dissect neuronal function require analysis of behavior at high temporal and spatial resolution . Achieving this is a time-consuming task and its automation remains a challenging problem in behavioral neuroscience . In the field of computer vision , modern neural network approaches have presented new solutions to visual tasks that perform just as well as humans ( Ching et al . , 2018; Angermueller et al . , 2016 ) . Application of these tools to biologically relevant problems could alleviate the costs of behavioral experiments and enhance reproducibility . Despite these enticing advantages , few aspects of behavioral biology research leverages neural network approaches . This lack of application is often attributed to the high cost of organizing and annotating the data sets , or to the stringent performance requirements . Thus , behavior recognition within dynamic environments is an open challenge in the machine learning community and translatability of proposed solutions to behavioral neuroscience remains unaddressed . Behavioral action recognition falls under multiple types of computer vision problems , including action classification , event detection , and temporal action localization . Action classification , a task closely related to image captioning , trains a classifier to apply action labels to manually pre-trimmed video clips . This problem has already been largely solved , with the exceptional performance for networks competing in data sets such as Kinetics-400 , Moments in Time , Youtube-8M , and many other available benchmark data sets ( Wu et al . , 2017 ) . However , this classification does not determine when an action occurs within an untrimmed video . To address this shortcoming , two other tasks have been designed: event detection ( ActivityNet 2019 Task 1 ) and temporal action localization ( ActivityNet 2019 Task 2 ) ( Heilbron et al . , 2015 ) . The objective of event detection is to identify when an event occurs , whereas the objective of temporal action detection is to identify where , when , and who is performing an action in untrimmed video input . The dominant approach for solving these issues has been extending region proposal methods from single images to video data . This involves proposing video tubelets ( Kalogeiton et al . , 2017; Feichtenhofer et al . , 2019 ) , a clip of video in both space and time for a single subject performing a single action . In behavioral neuroscience , previous attempts to operate directly on visual data have utilized unsupervised behavioral clustering approaches ( Todd et al . , 2017 ) . These include seminal work to convert visual data into frequency domains followed by clustering in Drosophila ( Berman et al . , 2014 ) and autoregressive Hidden Markov Model-based analysis of depth imaging data for mouse behavior ( Wiltschko et al . , 2015 ) . Both approaches rely upon alignment of data from a top-down view and while they cluster similar video segments , interpretation of generated clusters is still dictated by the user . It is also unclear how these approaches will perform on sequences of disparate behaviors . Supervised approaches in behavioral neuroscience have abstracted the subject into lower dimensions such as ellipse or key points , followed by feature generation , and classification ( Kabra et al . , 2013; van den Boom et al . , 2017 ) . While these approaches were a significant advance when they were introduced , they are inherently limited by the measurements available from the abstraction . For instance , standard measurements such as center of mass tracking , limit the types of behaviors that can be classified reliably . The field quickly recognized this issue and moved to integrate new measurements for the algorithms to classify behavior . These new features are highly specific to the organism and behavior that the researcher wishes to observe . In Drosophila studies , tracking of individual limbs and wings add new tracking modalities ( Robie et al . , 2017 ) . For mice , modern systems integrate floor vibration measurements and depth imaging techniques to enhance behavior detection ( Quinn et al . , 2003; Hong et al . , 2015; Wiltschko et al . , 2015 ) . Vibration measurements set limits to both the environment and the number of animals , while depth imaging restricts the environment . While others have attempted to automate the annotation of mouse grooming using a machine learning classifier , available techniques are not robust for multiple animal coat colors , lighting conditions , and locations of the setup ( van den Boom et al . , 2017 ) . Recent advances in computer vision also provide general purpose solutions for marker-less tracking in lab animals ( Mathis et al . , 2018; Pereira et al . , 2019 ) . These new techniques provide richer features to extend traditional machine learning techniques for behavioral classification . Human action detection leaderboards suggest that while the approach of pose estimation is powerful , it routinely underperforms compared to end-to-end solutions that utilize raw video input for action classification ( Feichtenhofer et al . , 2019; Choutas et al . , 2018 ) . Here , we use neural networks to directly classify mouse grooming behavior from video . Grooming represents a form of stereotyped or patterned behavior of considerable biological importance consisting of a range of small to large actions . Grooming is an innate behavior conserved across animal species , including mammals ( Spruijt et al . , 1992; Kalueff et al . , 2010 ) . In rodents , a significant amount of waking behavior , between 20 and 50% , consists of grooming ( Van de Weerd et al . , 2001; Spruijt et al . , 1992; Bolles , 1960 ) . Grooming serves many adaptive functions such as coat and body care , stress reduction , de-arousal , social functions , thermoregulation , nociception , as well as other functions ( Spruijt et al . , 1992; Kalueff et al . , 2010; Fentress , 1988 ) . The neural circuitry that regulates grooming behavior has been studied , although much remains unknown . Importantly , grooming and other patterned behaviors are endophenotypes for many psychiatric illnesses . For instance , a high level of stereotyped behavior is seen in autism spectrum disorder ( ASD ) , while in contrast , Parkinson’s disease shows an inability to generate patterned behaviors ( Kalueff et al . , 2010 ) . Therefore , the accurate and automated analysis of grooming behavior represents important value in behavioral neuroscience . We also reasoned that successful development of a neural network architecture for grooming behavior classification would be transferable to other behaviors by changing the training data . We applied a general machine learning solution to mouse grooming and developed a classifier that performs at human level . This classifier performs across 62 inbred and F1 hybrid strains of mice consisting of visually diverse coat colors , body shapes , and sizes . We explored reasons why our network has an upper limit on performance that seems to be concordant with human annotations . Human level performance comes at a cost of a large amount of labeled training data . We identified environmental and genetic regulators of grooming behavior in the open field . Finally , we applied our grooming behavior solution to a genetically diverse mouse population and characterize the grooming pattern of the mouse in an open field . We used these data to carry out a genome wide association study ( GWAS ) and to identify the genetic architecture that regulates heritable variation in grooming and open-field behaviors in the laboratory mouse . Combined we propose a generalizable solution to complex action detection and apply it toward grooming behavior . Behavior varies widely on both time and space scales , from fine spatial movements such as whisking , blinking , or tremors to large spatial movements such as turning or walking , and temporally from milliseconds to minutes . We sought to develop a classifier that could observe and predict complex behaviors produced by the mouse . Grooming consists of syntaxes that are small or micro-motions ( paw lick ) to mid-size movements ( unilateral and bilateral face wash ) and large movements ( flank licking ) Figure 1A . There are also rare syntaxes such as genital and tail grooming . Grooming duration can vary from sub-seconds to minutes . Our approach to annotating grooming classified each frame in a video as the mouse being in one of two states: grooming or not grooming . We specified that a frame should be annotated as grooming when the mouse is performing any of the syntaxes of grooming , whether or not the mouse is performing a stereotyped syntactic chain of grooming . This included a wide variety of postures and action durations which contribute to a diverse visual appearance . This also explicitly included individual paw licks as grooming , despite isolated paw licks not constituting a bout of grooming . Scratching was excluded from being classified as grooming . We investigated the variability in manual grooming annotations by humans by tasking five trained annotators with labeling the same six 5 min videos ( 30 min total , Figure 1 ) . To help human scorers , we provided these videos from a top-down and side view of the mouse ( Figure 1B ) . These videos included C57BL/6J , BTBR and CAST/EiJ mouse strains . We gave each annotator the same instructions to label the behavior ( see Methods ) . We observed a strong agreement ( 89 . 1% average ) between annotators , which is in concordance with prior work annotating mouse grooming behavior ( Kyzar et al . , 2011 ) . To examine disagreements between annotators , we classified them into three classes: missed bout , skipped break , and misalignment ( Figure 1—figure supplement 1 ) . Missed bout calls are made when a disagreement occurs in a not-grooming call . Similarly , skipped break calls are made when a disagreement occurs in a grooming call . Finally , misalignment is called when both annotators agree that grooming is either starting or ending but disagree on the exact frame in which this occurs . The most frequent type of error was misalignment , accounting for 50% of total duration of disagreement frames annotated and 75% of the disagreement calls ( Figure 1—figure supplement 1 ) . Next , we constructed a large annotation data set to train a machine learning algorithm . While most machine learning contests seeking to solve tasks similar to ours have widely varied data set sizes , we leveraged network performance in these contests for design of our data set . Networks in these contests perform well when an individual class contains at least 10 , 000 annotated frames ( Girdhar et al . , 2019 ) . As the number of annotations in a class exceeds 100 , 000 , network performance for this task achieves mean average precision ( mAP ) scores above 0 . 7 ( Girdhar et al . , 2019; Zhang et al . , 2019 ) . With deep learning approaches , model performance benefits from additional annotations ( Sun et al . , 2017 ) . To ensure success , we set out to annotate over 2 million frames with either grooming or not grooming . We aimed to balance this data set for grooming behavior by selecting video clips based on tracking heuristics , prioritizing segments with low velocity because a mouse cannot be grooming while walking . We also cropped the video frame to be centered on the mouse to reduce visual clutter using our tracker ( Geuther et al . , 2019 ) . This cropping centered around the mouse follows the video tube approach , as seen in the current state of the art ( Feichtenhofer et al . , 2019 ) . Based on this , we sampled 1253 short video clips from 157 videos . These video clips represent a diverse set of mice including 60 strains and a large range of body weights ( Figure 2—figure supplement 1A–B ) . Using a pool of seven validated annotators , we obtained two annotations for each of the 1253 video clips totaling 2 , 637 , 363 frames with 94 . 3% agreement between annotations ( Figure 2A ) . We trained a neural network classifier using our large annotated data set . Of the 1253 video clips , we held out 153 for validation . Using this split , we achieved similar distributions of frame-level classifications between training and validation sets ( Figure 2A ) . Our machine learning approach takes video input data and produces an ethogram output for grooming behavior ( Figure 2B ) . Functionally , our neural network model takes an input of 16 112 × 112 frames , applies multiple layers of 3D convolutions , 3D pooling , and fully connected layers to produce a prediction for only the last frame ( Figure 2C ) . To predict a completed ethogram for a video , we slide the 16-frame window across the video . We compared our neural network approach to a previously established machine learning approach for annotating lab animal behavior , JAABA ( Kabra et al . , 2013 ) . Our neural network achieved 93 . 7% accuracy and 91 . 9% true positive rate ( TPR ) with a 5% false positive rate ( FPR ) ( Figure 3A , B , pink line ) . In comparison , the JAABA trained classifier achieved a lower performance of 84 . 9% accuracy and 64 . 2% TPR at a 5% FPR ( Figure 3A , B ) . Due to memory limitations of JAABA , we could only train it using 20% of our training set . To test whether the training set size accounted for this poorer performance by JAABA , training our neural network using 20% of our training set still led to out-performance of JAABA ( Figure 3B ) . When training the neural network using different sized training data sets , we observed improved validation performance with increasing data set size ( Figure 3—figure supplement 1A ) . Scaling the training dataset size for JAABA showed that performance saturated when using 10% of our training data ( Figure 3—figure supplement 1B ) . Using an interactive training protocol recommended by the authors of JAABA , we observed decreased performance . This was likely due to the drastic size difference of the annotated data sets used in training ( 17 , 000 frames , or 0 . 7% of our annotated dataset ) . Interestingly , JAABA using 5% of our training dataset outperformed our neural network using 10% of our training dataset . This suggests that although JAABA may perform better using limited small datasets , both a neural network approach and a larger training dataset are necessary for generalizing on larger and more varied data . Our neural network approach was as good as human annotators , given our previous observations in Figure 1B–C of 89% agreement . We inspected the receiver operating characteristic ( ROC ) curve performance on a per-video basis and found that performance was not uniform across all videos ( Figure 3—figure supplement 2 ) . The majority of the 153 validation videos were adequately annotated by both the neural network and JAABA . However , two videos performed poorly with both algorithms and seven videos showed drastic improvement using a neural network over the JAABA-trained classifier . Manual visual inspection of the two videos where both algorithms performed poorly suggests that they did not provide sufficient visual information to annotate grooming . While developing our final neural network solution , we applied two forms of consensus modalities to improve single-model performance ( Figure 3C ) . Each trained model makes slightly different predictions , due to randomness involved in training . This randomness appears in both network parameter initialization and the order of training batches . By training multiple models and merging the predictions , we achieved a slight improvement on validation performance . Additionally , we also modified the input image for different predictions . Rotating and reflecting the input image appears visually different for neural networks . We achieved 32 separate predictions for every frame by training four models and applying eight rotation and reflection transformations on the input . We merged these individual predictions by averaging the probability predictions . This consensus modality improved the ROC area under the curve ( AUC ) from 0 . 975 to 0 . 978 . We attempted other approaches for merging the 32 predictions , including selecting the max value or applying a vote ( median prediction ) . Averaging the prediction probabilities achieved the best performance ( Figure 3—figure supplement 3A ) . Finally , we applied a temporal smoothing filter over 46 frames of prediction . We identified 46 frames to be the optimal window for a rolling average ( Figure 3—figure supplement 3B ) , which results in a final accuracy of 93 . 7% ( ROC AUC of 0 . 984 ) . Our network can only make predictions using half a second worth of information . To ensure our validation performance is indicative of the wide diversity of mouse strains , we investigated the extremes of grooming bout predictions in our large strain survey data set which was not annotated by humans . While most of the long bout ( >2 min ) predictions were real , there were some false positives in which the mouse was resting in a grooming-like posture . To mitigate these rare false positives , we implemented a heuristic to adjust predictions . We experimentally identified that grooming motion typically causes ellipse-fit shape changes ( W/L ) to have a standard deviation greater than 2 . 5×10-4 . When a mouse is resting , the shape changes ( W/L ) standard deviation does not exceed 2×10-5 . Knowing that a mouse’s posture in resting may be visually similar to a grooming posture , we assigned predictions in time segments where the standard deviation of shape change ( W/L ) over a 31 frame window was less than 5×10-5 to a ‘not grooming’ prediction . Of all the frames in this difficult to annotate posture , 12% were classified as grooming . This suggests that this is not a failure case for our network , but rather a limitation of the network when only using half a second worth of information to make a prediction . This approach handled varying mouse postures and physical appearance well , e . g . coat color and body weight . We observed good performance over a wide variety of postures and coat colors ( Figure 3C–D , Figure 3—videos 1–9 ) . Even nude mice , which have a drastically different appearance than other mice , achieved good performance . Visually , we observed instances where a small number of frame orientations and models make incorrect predictions . Despite this , the consensus classifier made the correct prediction . We designed a variety of grooming behavioral metrics that describe both grooming quantity and grooming pattern . Following prior work ( Kalueff et al . , 2010 ) , we defined a single grooming bout as continuous time spent grooming without interruption that exceeds 3 s . We allowed brief pauses ( less than 10 s ) , but did not allow any locomotor activity for this merging of time segments spent grooming . Specifically , a pause occurred when motion of the mouse did not exceed twice its average body length . From this , we obtained a grooming ethogram for each mouse ( Figure 4A ) . Using the ethogram , we summed the total duration of grooming calls in all grooming bouts to calculate the total duration of grooming . Once we had the number of bouts and total duration , we calculated the average bout duration by dividing the two . For measurement purposes , we calculated the 5 min , 20 min , and 55 min summaries of these measurements . We included 5 and 20 min because these are typical open field assay durations . Using 1 min binned data , we calculated a variety of grooming pattern metrics ( Figure 4B ) . We fitted a linear slope to discover temporal patterning of grooming during the 55 min assay ( GrTimeSlope55min ) . Positive slopes for total grooming duration infer that the individual mouse is increasing its time spent grooming the longer it remains in the open field test . Negative slopes for total grooming duration infer that the mouse spends more time grooming at the start of the open field test than at the end . This is typically due to the mouse choosing to spend more time doing another activity over grooming , such as sleeping . Positive slopes for number of bouts inferred that the mouse is initiating more grooming bouts the longer it remains in the open field test . Using 5 min binned data , we designed additional metrics to describe grooming pattern by selecting which 5 min bin a mouse spent the most time grooming ( GrPeakMidBin ) and the time duration spent grooming ( GrPeakVal ) in that minute . We also calculated a ratio between these values ( GrPeakSlope ) . Finally , when we looked at strain-level averages of grooming , we identified how long a strain remains at its peak grooming ( GrPeakLength ) . We compared a variety of open field measurements including both grooming behavior and classical open field measurements ( Figure 4C ) . We separated these 24 phenotypes into four groups . Grooming quantity describes how much an animal grooms , while grooming pattern metrics describe how an animal changes its grooming behavior over time . Open field anxiety measurements are traditional phenotypic measurements that have been validated to measure anxiety . Open field activity describes the general activity level of an animal . With this trained classifier , we sought to determine whether sex and environment affected grooming behavior in an open field , specifically grooming duration . We used data collected over 29 months for two strains , C57BL/6J and C57BL/6NJ to carry out this analysis . These two strains are substrains that were identical in 1951 and are two of the most widely used strains in mouse studies ( Bryant et al . , 2018 ) . C57BL/6J is the mouse reference strain and C57BL/6NJ has been used by the International Mouse Phenotyping Consortium ( IMPC ) to generate a large amount of phenotypic data ( Brown and Moore , 2012 ) . We analyzed 775 C57BL/6J ( 317F , 458M ) and 563 C57BL/6NJ ( 240F , 323M ) mice tested over a wide variety of experimental conditions and ages . Across all these novel exposures in an open field , we quantified their grooming behavior for the first 30 min ( Figure 5 , 669 hr total data ) . We analyzed the data for effect of sex , season , time of day , age , room origin of the mice , light levels , tester , and white noise . To achieve this , we applied a stepwise linear model selection to model these covariates . Both forward and backward model selection results matched . After identifying significant covariates , we applied a second round of model selection that included sex interaction terms . The model selection identified sex , strain , room of origin , time of day , and season as significant . In contrast , age , weight , presence of white noise , and tester were not significant under our testing conditions . Additionally , the interaction between sex and both room of origin and season were identified as significant covariates . Covariatep-valueSex<2 . 2 × 10−16 ***Strain0 . 0267546 *RoomOrigin5 . 357 × 10−13 ***Morning0 . 0001506 ***Season0 . 0039826 **Sex by RoomOrigin0 . 0001568 ***Sex by Season0 . 0235954 * We found an effect of strain ( Figure 5A , p=0 . 0268 C57BL/6J vs C57BL/6NJ ) on grooming duration . Although the effect size was small , C57BL/6NJ groomed more than C57BL/6J . Additionally , we observed a sex difference ( Figure 5A , p<2 . 2×10-16 males vs females ) . Males groomed more than females in both strains . Since sex had a strong effect , we included interaction terms with other covariates in a second pass of our model selection . The model identified season as a significant covariate ( Figure 5B , p=0 . 004 ) . Surprisingly , the model also identified an interaction between sex and season ( p=0 . 024 ) . Female mice for both strains showed an increase in grooming during the summer and a decrease in the winter . We carried out testing between 8AM and 4PM . To determine if the time of test affects grooming behavior , we split the data into two groups: morning ( 8am to noon ) and afternoon ( noon to 4pm ) . We observed a clear effect of time of day ( Figure 5C , p=0 . 00015 ) . Mice tested in the morning groom more overall . We tested mice of different ages , ranging from 6 weeks to 26 weeks old . At the beginning of every test , we weighed the mice and found them to have a range of 16–42 g . We did not observe any significant effect of age ( Figure 5D , r=-0 . 065 , p=0 . 119 ) or body weight ( r=0 . 206 , p=0 . 289 ) on grooming duration , although we did not test ‘old’ mice , generally considered to be more than 18 months old . We compared the grooming levels of mice that were shipped from production rooms in a nearby building at our institution to our testing room with mice bred and raised in a room adjacent to the testing room ( B2B ) . These production rooms contain a variety of possible confounding variables such as microbiome , noise , and technician-related stress . We specifically note that these room differences should not be due to genetic drift because of JAX’s Genetic Stability Program , which periodically re-derives the strain from frozen embryos ( Taft et al . , 2006 ) . Six production rooms supplied exclusively C57BL/6J ( AX4 , AX29 , AX1 , MP23 , MP14 , MP15 ) , three rooms supplied exclusively C57BL/6NJ ( MP13 , MP16 , AX5 ) , and one room supplied both strains ( AX8 ) . All shipped mice were housed in B2B for at least a week prior to testing . We observed a significant effect for room of origin ( Figure 5E , p=5 . 357×10-13 ) . For instance , C57BL/6J males from AX4 and AX29 were low groomers compared to other rooms , including B2B . Shipped C57BL/6NJ from all rooms seemed to have low levels of grooming compared with B2B . We conclude that room of origin and shipping can both have effects on grooming behaviors . We tested two light levels , 350–450 lux and 500–600 lux white light ( 5600K ) . We observed significant effects of light levels on grooming behavior ( Figure 5F , p=0 . 04873 ) . Females from both strains groomed more in lower light , however males didn’t seem to be affected . Despite this , our model did not include a light-sex interaction , suggesting that other covariates better account for the visual interaction with sex here . The open field assays were carried out by one of two male testers , although the majority of tests were carried out by tester 2 . Both testers carefully followed a testing protocol designed to minimize tester variation , which only involves weighing the mouse and placing the mouse into the arena . We observed no significant effect ( Figure 5G , p=0 . 65718 ) between testers . Finally , white noise is often added to open field assays in order to create a uniform background noise levels and to mask noise created by the experimenter ( Gould , 2009 ) . Although the effects of white noise have not been extensively studied in mice , existing data indicate that higher levels of white noise increase ambulation ( Weyers et al . , 1994 ) . We tested the effects of white noise ( 70 db ) on grooming behavior of C57BL/6J and C57BL/6NJ mice and found no significant difference in duration spent grooming . Although there appears to be a stratification present for both C57BL/6J and C57BL/6NJ females , other cofactors better account for this . Combined , these results indicate that environmental factors such as season , time of day , and room origin of the mice affect grooming behavior and may serve as environmental confounds in any grooming study . We also investigated age , body weight , light level , tester , and white noise and found these cofactors to not influence grooming behavior under our experimental conditions . Next , we used the grooming classifier to carry out a survey of grooming behavior in the inbred mouse . We tested 43 classical laboratory and eight wild derived strains and 11 F1 hybrid mice from The Jackson Laboratory ( JAX ) mouse production facility . These were tested over a 31-month period and in most cases consisted of a single mouse shipment of mice from JAX production . Other than C57BL/6J and C57BL6/NJ , on average we tested eight males and eight females of an average age of 11 weeks for each strain . Each mouse was tested for 55 min in the open field as previously described ( Geuther et al . , 2019 ) . This data set consisted of 2457 animals and 2252 hr of video . Video data were classified for grooming behavior as well as open field activity and anxiety metrics . Behavior metrics were extracted as described in Figure 4 . In order to visualize the variance in phenotypes , we plotted each animal across all strains with corresponding strain mean and one standard deviation range and ethograms of select strains Figure 6 . We distinguish between classical laboratory strains and wild derived inbred strains . We observed large continuous variance in total grooming time , average length of grooming bouts , and the number of grooming bouts in the 55 min open field assay ( Figure 6 ) . Total grooming time varied from 2 to 3 min in strains such as 129 × 1/SvJ and BALB/cByJ to 12 min in strains such as SJL/J and PWD/PhJ . Strains such as 129 × 1/SvJ and C57BR/cdJ had less than 10 bouts , whereas MA/MyJ had almost 40 bouts . The bout duration also varied from 5 s to approximately 50 s in BALB/cByJ and PWD/PhJ , respectively . In order to visualize relationships between phenotypes , we created strain mean and 1SD range correlation plots ( Figure 7 ) . There was a positive correlation between the total grooming time and the number of bouts as well as the total grooming time and average bout duration . Overall , strains with high total grooming time had increased number of bouts as well as longer duration of bouts . However , there did not seem to be a relationship between the number of bouts and the average bout duration , implying that the bout lengths stay constant regardless of how many may occur ( Figure 7 ) . In general , C57BL/6J and C57BL6/NJ fall roughly in the middle for classical inbred strains . We investigated the pattern of grooming over time by constructing a rate of change in 5 min bins for each strain ( Figure 8 ) . There appeared to be visual structures in the data , so we used k-means to identify clusters . We identified three clusters of grooming patterns based on total grooming level and relative changes in 5 min binned grooming data ( Figure 1—figure supplement 1 ) . Type 1 consists of 13 strains with an ‘inverted U’ grooming pattern . These strains escalate grooming quickly once in the open field , reach a peak , and then start to decrease the amount of grooming , usually leading to a negative overall grooming slope . Often , we find animals from these strains are sleeping by the end of the 55 min open field assay . These strains include both high groomers such as CZECHII/EiJ , MOLF/EiJ , and low groomers such as 129 × 1/SvJ and I/LnJ . Type 2 consists of 12 strains that are high grooming strains and do not reduce grooming by the end of the assay . They reach peak grooming early ( e . g . PWD/PhJ , SJL/J and BTBR ) or late ( e . g . DBA/2J , CBA/J ) and then remain at or near this peak level for the remainder of the assay . The defining feature of this group is that a high level of grooming is maintained throughout the assay . Type 3 consists of most of the strains ( 30 ) and shows steady increase in grooming until the end of the assay . Overall , the strains in this group are medium-to-low groomers with a constant low positive or flat slope . We conclude that under our experimental conditions there are at least three broad , albeit continuous , classes of observable grooming patterns in the mouse . We compared grooming patterns between classical and wild derived laboratory strains . Classical laboratory strains are derived from limited genetic stock originating from Japanese and European mouse fanciers ( Keeler , 1931; Morse , 1978; Silver , 1995 ) . Classical laboratory inbred mouse lines represent the genome of Mus musculus domesticus ( M . m domesticus ) 95% and Mus musculus musculus 5% ( Yang et al . , 2011 ) . New wild derived inbred strains were established specifically to overcome the limited genetic diversity of the classical inbred lines ( Guénet and Bonhomme , 2003; Koide et al . , 2011 ) . We observed that most wild derived strains groom for significantly higher duration and have longer average bout length than the classical inbred strains . Five of the highest 16 grooming strains are wild derived ( PWD/PhJ , WSB/EiJ , CZECHII/EiJ , MSM/MsJ , MOLF/EiJ in Figure 6A ) . The wild derived strains also had significantly longer bouts of grooming , with 6 of 16 longest average grooming bout strains from this group . Both the total grooming time and average bout length were significantly different between classical and wild-derived strains ( Figure 1—figure supplement 1 ) . These high grooming strains represent M . m . domesticus and M . M . musculus subspecies , which are the precursors to classical laboratory strains ( Yang et al . , 2011 ) . These wild derived strains also represent much more of the natural genetic diversity of the mouse populations than the larger number of classical strains we tested . This leads us to conclude that the high levels of grooming seen in the wild derived strains better represent the normal levels of grooming behavior in mice . This implies that low grooming behavior may have been selected for in classical laboratory strains , at least as observed in our experimental conditions . We also closely examined the grooming patterns of the BTBR strain , which has been proposed as a model with certain features of autism spectrum disorder ( ASD ) . ASD is a complex neurodevelopmental disorder leading to repetitive behaviors and deficits in communication and social interaction ( Association , 2013 ) . Compared to C57BL/6J mice , BTBR have been shown to have high levels of repetitive behavior , low sociability , unusual vocalization , and behavioral inflexibility ( McFarlane et al . , 2008; Silverman et al . , 2010; Moy et al . , 2007; Scattoni et al . , 2008 ) . Repetitive behavior is often assessed by self grooming behavior , and drugs with efficacy in alleviating symptoms of repetitive behavior in ASD also reduce grooming in BTBR mice without affecting overall activity levels , which provides some level of construct validity ( Silverman et al . , 2012; Amodeo et al . , 2017 ) . We found that total grooming time in BTBR is high compared with C57BL/6J but is not exceptionally high compared to all strains ( Figure 6 ) , or even among classical inbred strains . C57BL/6J mice groomed approximately 5 min over a 55 min open field session , whereas BTBR groomed approximately 12 min ( Figure 6A ) . Several classical inbred strains had similar levels of high grooming , including SJL/J , DBA/1J , and CBA/CaJ . The grooming pattern of BTBR belongs to Type 2 which contains 11 other strains ( Figure 8 ) . One distinguishing factor of BTBR mice is that they have longer average bouts of grooming from an early point in the open field ( Figure 6—figure supplement 1 ) . However , again they were not exceptionally high in average bout length measure ( Figure 6—figure supplement 1 ) . Strains such as SJL/J , PWD/PhJ , MOLF/EiJ , NZB/BINJ had similar long bouts from an early point . We conclude that BTBR display high levels of grooming with long grooming bouts , however , this behavior is similar to several wild derived and classical laboratory inbred strains and is not exceptional . Since we did not measure social interaction and other salient features of ASD , we do not argue against BTBR as an ASD model . In addition to BTBR , perhaps other strains from the type two group could should be explored as ASD models . Next we wished to understand the underlying genetic architecture of complex mouse grooming behavior and open field behaviors , and to relate these to human traits . We used the data from the 51 inbred strains and 11 F1 hybrid strains to carry out a genome wide association study ( GWAS ) . We did not include the eight wild derived strains because they are highly divergent and can skew mouse GWAS analysis . We analyzed the 24 phenotypes categorized into four categories – open field activity , anxiety , grooming pattern , and quantity ( Figure 4 ) . We used a linear mixed model ( LMM ) implemented in Genome-wide Efficient Mixed Model Association ( GEMMA ) for this analysis in order to control for spurious association due to population structure ( Zhou and Stephens , 2012 ) . We first calculated heritability of each phenotype by determining the proportion of variance in phenotypes explained by the typed genotypes ( PVE ) Figure 9A . Heritability ranged from 6% to 68% , and 22/24 traits have heritability estimates greater than 20% , a reasonable estimate for behavioral traits in mice and humans ( Valdar et al . , 2006; Bouchard , 2004 ) , and makes them amenable for GWAS analysis ( Figure 9A ) . We analyzed each phenotype using GEMMA , and considered the resulting Wald test p-value . In order to correct for the multiple SNPs we tested ( 222 , 966 ) , and to account for the correlations between SNPs genotypes , we obtained an empirical threshold for the p-values by shuffling the values of one normally distributed phenotype ( OFDistTraveled20m ) and taking the minimal p-value of each permutation . This process resulted in a p-value threshold of 1 . 4×10-5 that reflects a corrected p-value of 0 . 05 ( Belmonte and Yurgelun-Todd , 2001 ) . We defined quantitative trait loci ( QTL ) in the following manner: adjacent SNPs that have correlated genotypes ( r2≥0 . 2 ) were clustered together in a greedy way , starting with the SNP with the lowest p-value in the genome , assigning it a locus , adding all correlating SNPs and then moving forward to the next SNP with the lowest p-value until all the significant SNPs are assigned to QTL . The genetic architecture of inbred mouse strains dictates large linkage disequilibrium ( LD ) blocks ( Figure 9B ) , resulting in QTL that span millions of base-pairs and contain multiple genes ( Supplementary file 2 ) . GWAS analysis of each phenotype resulted in 2 to 22 QTL ( Figure 9—figure supplement 1 , Figure 9—figure supplement 2 , Figure 9—figure supplement 3 , Figure 9—figure supplement 4 , Figure 9—figure supplement 5 , Figure 9—figure supplement 6 ) . Overall , the open field activity had 15 QTL , anxiety 10 , grooming pattern 76 and grooming quantity 51 QTL , leading to 130 QTL combined over all the tested phenotypes ( Figure 9C ) . We observed pleiotropy with the same loci significantly associated with multiple phenotypes . Pleiotropy is expected since many of our phenotypes are correlated and individual traits may be regulated by similar genetic loci . For instance , we expected pleiotropy for grooming time in 55 and 20 min ( GrTime55 and GrTime20 ) since these are correlated traits . We also expected that some loci regulating open field activity phenotypes may regulate grooming . In order to better understand the pleiotropic structure of our GWAS results , we generated a heat map of significant QTL across all phenotypes . We then clustered these , to find sets of QTL that regulate groups of phenotypes ( Figure 9D ) . The phenotypes were clustered into five subgroups consisting of grooming pattern ( I ) , open field activity ( II ) , open field anxiety ( III ) , grooming length ( IV ) , and grooming number and amount ( V ) ( Figure 9D top x-axis ) . We found seven clusters of QTL that regulate combinations of these phenotypes ( Figure 9D y-axis ) . For instance , clusters A and G are composed of pleiotropic QTL that regulate grooming length ( IV ) and grooming time but QTL in cluster G also regulate bout number and amount ( V ) . QTL cluster D regulates open field activity and anxiety phenotypes . Cluster E contains QTL that regulate grooming and open field activity and anxiety phenotypes , but most of the SNPs only have significant p-values for either open field phenotypes or grooming phenotypes but not both , indicating that independent genetic loci are largely responsible for these phenotypes . We colored the associated SNPs in the Manhattan plot ( Figure 9C ) with colors to mark one of these seven QTL clusters ( Figure 9D ) . These clusters ranged from 13 to 35 QTL , with the smallest being cluster F which is mostly pleiotropic for grooming number , and the largest cluster , cluster G , is pleiotropic for most of the grooming related phenotypes . These highly pleiotropic genes included several genes known to regulate mammalian grooming , striatal function , neuronal development , and even language . Mammalian Phenotype Ontology Enrichment showed ‘nervous system development’ as the most significant module with 178 genes ( p=7 . 5×10-4 ) followed by preweaning lethality ( p=3 . 5×10-3 , 189 genes ) and abnormal embryo development ( p=5 . 5×10-3 , 62 genes ) ( Supplementary file 2 ) . We carried out pathway analysis using KEGG and Reactome databases ( Ogris et al . , 2016 ) . This analysis showed 14 disease pathways that are enriched including Parkinson’s ( 9 . 68×10-9 ) , Huntington’s ( 1 . 07×10-6 ) , non-alcoholic fatty liver disease ( 9 . 31×10-6 ) , and Alzheimer’s ( 1 . 15×10-5 ) as the most significantly enriched . Enriched pathways included oxidative phosphorylation ( 6 . 42×10-8 ) , ribosome ( 0 . 00000102 ) , RNA transport ( 0 . 00000315 ) , and ribosome biogenesis ( 0 . 00000465 ) . Reactome enriched pathways included mitochondrial translation termination and elongation ( 2 . 50×10-19 and 5 . 89×10-19 , respectively ) , and ubiquitin-specific processing proteases ( 1 . 86×10-8 ) . The most pleiotropic gene was Sox5 which associated with 11 grooming and open field phenotypes . Sox5 has been extensively linked to neuronal differentiation , patterning , and stem cell maintenance ( Lefebvre , 2010 ) . Its dysregulation in humans has been implicated in Lamb-Shaffer syndrome and ASD , both neurodevelopmental disorder ( Kwan , 2013; Zawerton et al . , 2020 ) . A total of 102 genes were associated with 10 phenotypes , and 105 genes were associated with nine phenotypes . We limited our analysis to genes with at least six significantly associated phenotypes , resulting in 860 genes . Other genes include FoxP1 , which has been linked to striatal function and regulation of language ( Bowers and Konopka , 2012 ) . Ctnnb1 , a regulator of Wnt signaling , and Grin2b , a regulator of glutamate signaling . Combined , this analysis indicated genes known to regulate nervous system function and development , and genes known to regulate neurodegenerative diseases as regulators of grooming and open field behaviors . The GWAS also begins to define the genetic architecture of grooming and open field behaviors in mice . Finally , we wanted to link genes that are associated with open field and grooming phenotypes in the mouse with human phenotypes . We hypothesized that common underlying genetic and neuronal architectures exist between mouse and human , however , they can give rise to disparate phenotypes in each organism . For example , the misregulation of a pathway in the mouse may lead to over-grooming phenotype but in humans the same pathway perturbation may manifest itself as neuroticism or obsessive compulsive disorder . These relationships between phenotypes can be revealed through identification of common underlying genetic architectures . In order to link human and mouse phenotypes , we carried out PheWAS analysis with the 860 mouse grooming and open field genes with at least 6 degrees of pleiotropy identified in the mouse GWAS . We identified 509 human orthologs out of 860 mouse genes and downloaded PheWAS summary statistics from gwasATLAS . The gwasATLAS ( Release 2: v20190117 ) contains 4756 GWAS from 473 unique studies across 3302 unique traits which are classified into 28 general domains and 192 subchapters obtained from either ICD10 or ICF10 ( Watanabe et al . , 2019 ) . For each human ortholog , we focused on the association in the gwasATLAS Psychiatric domain with gene-level p value ≤ 0 . 001 . We then turned the relationships between genes and psychiatric traits into a weighted bipartite network , where the weight of an edge is represented by the association strength ( -log10 ( p value ) ) between a gene and a trait . We identified eight gene-phenotype modules within this weighted bipartite networks ( Figure 10 ) . These modules contained between 15 and 32 individual phenotypes and between 41 and 103 genes . At the subchapter level , modules were enriched for temperament and personality phenotypes , mental and behavioral disorders ( schizophrenia , bipolar , dementia ) , addiction ( alcohol , tobacco , cannabinoid ) obsessive-compulsive disorder , anxiety , and sleep . Surprisingly , we found orthologs that show high levels of pleiotropy in mouse GWAS and the resulting human PheWAS . FOXP1 is the most pleiotropic gene with 35 human phenotypic associations , while SOX5 is second with 33 associations . In order to prioritize candidate modules for further research , first , we produced a ranked list of modules by calculating modularity score , which measures the strength of division of a network into modules . The high-ranking modules represent most promising candidates for further research ( Newman and Girvan , 2004 ) . Modularity scores of the modules ranged from 0 . 028 ( module 8 ) to 0 . 103 ( module 1 ) . Second , we used Simes’ test to combine the p values of genes to obtain an overall p value for the association with each Psychiatric trait . Then the median of association ( -log10 ( Simes p value ) ) was calculated in each detected module for prioritization . Using this method , module one again ranked at the top of eight modules ( median = 5 . 29 ) ( Figure 10—figure supplement 1 ) . Module one is primarily composed of temperament and personality phenotypes , including neuroticism , mood swings , and irritability traits . Genes in this module have a high level of pleiotropy in both human PheWAS and mouse GWAS . Eight of the 10 most pleiotropic genes from the PheWAS analysis are in this module . Genes in this module include SOX5 noted above , RANGAP1 with 31 associations , and EP300 with 23 significant human phenotypic associations . In conclusion , PheWAS analysis links conserved genes that regulate mouse grooming and other open field behaviors to human psychiatric phenotypes . These human phenotypes include personality traits , addiction , and schizophrenia . This analysis links mouse and human traits through shared underlying genetics and allows us to prioritize gene modules for future work , while serving as a framework for future analysis . Grooming is an ethologically conserved , neurobiologically important behavior that is often used as an endophenotype for psychiatric illnesses . It is a prototypical stereotyped , patterned behavior with highly variable posture and temporal length . Tools to automatically quantify behaviors such as grooming are needed by the behavioral research community and can be leveraged to gain insight into other complex behaviors ( Spruijt et al . , 1992; Kalueff et al . , 2010 ) . We present a neural network approach toward automated vertebrate model organism behavioral classification and ethogram generation that achieves human-level performance . This approach is simple , scalable , and can be carried out using standard open field apparatus , and is expected to be of use to the behavioral neuroscience community . Using this grooming behavior classifier , we analyzed a large data set consisting over 2200 hr of video from dozens of mouse strains . We demonstrated environmental effects on grooming , patterns of grooming behavior in the laboratory mouse , carry out a mouse GWAS , and a human PheWAS to understand the underlying human-relevant genetic architecture of grooming and open field behavior in the laboratory mouse . While the machine learning community has implemented a wide variety of solutions for human action detection , few applications have been applied to animal behavior . This may be in part due to the wide availability of human action data sets and the stringent performance requirements for human bio-behavioral research . We observed that the cost of achieving this stringent performance is very high , requiring a large quantity of annotations . More often than not , experimental paradigms are limited by cost to be short or small enough to cost less to allow manual annotation of the data . Machine learning approaches have been previously applied to automated annotation of behavioral data . We observed that our 3D convolutional neural network outperforms a JAABA classifier when trained on the same training data set . Our neural network achieved 91 . 9% true positive rate with a 5% false positive rate while thee JAABA classifier achieved 64 . 2% true positive rate at a 5% false positive rate . This improvement makes the neural network solution usable for application to biological problems . This improvement is not uniform over all samples but is instead localized to certain types of grooming bouts . This suggests that although the JAABA classifier is powerful , it may be most useful for smaller and more uniform data sets . Experimental paradigms and behaviors with diverse expression will require a more powerful machine learning approach . With the grooming classifier , we determined the genetic and environmental factors that regulate this behavior . In a large data set collected over an 18-month period using two reference strains , C57BL/6J and C67BL/6N , we assessed effects on grooming of several fixed and dynamic factors including , sex , strain , age , time of day , season , tester , room origin , white noise , and body weight . All mice were housed in identical conditions for at least a week prior to testing . As expected , we observe strong effects of sex , time of day , and season . In many but not all studies , and not in the present study , tester effects have been observed in the open field in both mice and rats ( Walsh and Cummins , 1976; McCall et al . , 1969; Bohlen et al . , 2014; Lewejohann et al . , 2006 ) . A recent study demonstrated that male experimenters or even clothes of males elicit stress responses from mice leading to increased thigmotaxis ( Sorge et al . , 2014 ) . To our surprise , we found that room of origin had a strong effect on grooming behavior of C57BL/6J and C57BL/6N . We did not see a clear directionality of effect between shipping mice and those bred in an adjacent room , and in some cases the effect size was high ( z > 1 ) . We hypothesize that this may be due to room-specific stress which has previously been demonstrated to alter grooming ( Kalueff et al . , 2016 ) . Presumably , all external mice had similar experience of shipping from the production rooms to the testing area where they were housed identically for at least 1 week prior to testing . Thus , the potential differential stress experience was in the room of origin where the mouse was born and held until shipping . It is important to note that this shipping was only across buildings on the same campus , and shipping that involves air freight may have more drastic effects . This is a point of caution for use of grooming behavior as an endophenotype . Although we acclimated shipped mice to at least 1 week in an adjacent holding room prior to testing , a longer acclimation period may be required prior to testing . We carried out a large strain survey to characterize and account for genetic diversity in grooming behavior in the laboratory mouse . We found three types of grooming patterns under our test conditions . Type 1 consists of mice that escalate and deescalate grooming within the 55 min open field test . Strains in this group are often sleeping by the end of the assay , indicating a low arousal state toward the end of the assay . We hypothesize that these strains use grooming as a form of successful de-arousal , a behavior that has been previously noted in rats , birds , and apes ( Spruijt et al . , 1992; Delius , 1970 ) . Similar to Type 1 , Type 2 groomers escalate grooming quickly to reach peak grooming; however , this group does not deescalate grooming during our assay . We hypothesize that these strains need a longer time or may have a deficiency in de-arousal under our test conditions . Type 3 strains escalate for the duration of the assay indicating they have not reached peak grooming under our assay conditions . BTBR is a member of the Type 2 group with prolonged high levels of grooming from an early point , perhaps indicating a hyperaroused state , or an inability to de-arouse . BTBR mice have previously been shown to have high arousal states and altered dopamine function which may lead to the sustained high levels of grooming ( Squillace et al . , 2014 ) . We postulate that other strains in the Type 2 grooming class may also show phenotypic features of ASD , warranting further study of ASD-related phenotypes in these strains . Wild derived strains have distinct patterns of grooming compared to classical strains . Wild-derived strains groom significantly more and have longer grooming bouts . In our grooming clustering analysis , most of the wild derived strains belong to Type 1 or 2 , where as most classical strains belong in Type 3 . In addition to M . m domesticus , the wild derived inbred lines we tested represent M . m musculus , M . m castaneous , and M . m molossinus subspecies . Even though there are dozens of classical inbred strains , there are approximately 5 million SNPs between any two classical inbred laboratory strains such as C57BL/6J and DBA2J ( Keane et al . , 2011 ) . Indeed , over 97% of the genome of classical strains can be explained by fewer than 10 haplotypes indicating small number of classes within which all strains are identical by descent with respect to a common ancestor ( Yang et al . , 2011 ) . Classical laboratory strains are derived from mouse fanciers in China , Japan and Europe before being co-opted for biomedical research ( Morse , 1978; Silver , 1995 ) . Wild derived inbred strains such as CAST/EiJ and PWK/PhJ have over 17 million SNPs compared to B6J . Thus , the seven wild derived strains we tested represent far more of the genetic diversity present in the natural mouse population than the numerous classical inbred laboratory strains . Behaviors seen in the wild-derived strains are more likely to represent behaviors in the natural mouse population . Mouse fanciers breed mice for visual and behavioral distinctiveness , and many exhibit them in competitive shows . Mouse fanciers judge mice on ‘condition and temperament’ and suggest that ‘it is useless to show a mouse rough in coat or in anything but the mouse perfect condition’ ( Davies , 1912 ) . Much like dogs and horses , the ‘best individuals should be mated together regardless of relationship as long as mice are large , hardy , and free from disease’ ( Davies , 1912 ) . It is plausible that normal levels of grooming behavior seen in wild mice was considered unhygienic or indicative of parasites such as lice , ticks , fleas , or mites . High grooming could be interpreted as poor condition and would lead the mouse fancier to select mouse strains with low grooming behaviors . This selection could account for low grooming seen in the classical laboratory strains . We used the strain survey data to conduct a mouse GWAS which identified 130 QTL that regulates heritable variation in open field and grooming behaviors . We found that the majority of the grooming traits are moderately to highly heritable . A previous study using the BXD recombinant inbred panel identified one significant locus on chromosome four that regulates grooming and open field activity ( Delprato et al . , 2017 ) . We closely analyzed 862 genes in the QTL interval that are highly pleiotropic and find enriched pathways that regulate neuronal development and function . We then associated those intervals to one of seven clusters which regulate combinations of open field and grooming phenotypes . Mouse grooming can be used as a model of human grooming disorders such as tricotillomania; however , grooming is regulated by the basal ganglia and other brain regions and can be used more broadly as an endophenotype for many psychiatric traits , including ASD , schizophrenia , and Parkinson’s ( Kalueff et al . , 2016 ) . We conducted a PheWAS with the highly pleiotropic genes and identified human psychiatric traits that are associated with these genes . This approach allowed us to link mouse and human phenotypes through the underlying genetic architecture . This approach linked human temperament , personality traits , schizophrenia , and bipolar disorder to mouse phenotypes . Future research is needed to definitively link mouse genetic variants to altered behavior . Our GWAS results are a starting point for understanding the genetic architecture of grooming behavior and will require functional studies in the future to assign causation . In conclusion , we describe a neural network based machine learning approach for action detection in mice and apply it towards grooming behavior . Using this tool , we characterized grooming behavior and its underlying genetic architecture in the laboratory mouse . Our approach to grooming is simple and can be carried out using standard open field apparatus and should be of use to the behavioral neuroscience community . All animals were obtained from The Jackson Laboratory production colonies or bred in a room adjacent to the testing room as previously described ( Geuther et al . , 2019; Kumar et al . , 2011 ) . Partial data of the strain survey was published previously and reanalyzed for grooming behavior ( Geuther et al . , 2019 ) . All behavioral tests were performed in accordance with approved protocols from The Jackson Laboratory Institutional Animal Care and Use Committee guidelines . We selected data to annotate by training a preliminary JAABA classifier for grooming , then clipping video chunks based on predictions for a wide variety of videos . The initial JAABA classifier was trained on 13 short clips that were manually enriched for grooming activity . This classifier is intentionally weak , designed simply to prioritize video clips that would be beneficial to select for annotation . We then applied this weak classifier on a larger library of videos . The video clips are a subset of our previously published dataset and include 157 individual mouse videos that represent 60 different mouse strains ( Geuther et al . , 2019 ) . We clipped video time segments with 150 frames surrounding grooming activity prediction to mitigate chances of a highly imbalanced data set . We generated 1253 video clips which total 2 , 637 , 363 frames . Each video had variable duration , depending upon the grooming prediction length . The shortest video clip contains 500 frames , while the longest video clip contains 23 , 922 frames . The median video clip length is 1348 frames . Also see Supplementary file 1 for additional annotated dataset metadata . From here , we trained seven annotators . From this pool of seven trained annotators , we assigned two annotators to annotate each video clip completely . If there was confusion for a specific frame or sequence of frames , we allowed the annotators to request additional opinions . Annotators were required to provide a ‘Grooming’ or ‘Not Grooming’ annotation for each frame , with the intent that difficult frames to annotate would get different annotations from each annotator . We only train and validate using frames in which annotators agree , which reduces the total frames to 2 , 487 , 883 . Our neural network follows a typical feature encoder structure except using 3D convolution and pooling layers instead of 2D . We started with a 16 × 112 × 112 × 1 input video frames , where 16 refers to the time dimension of the input and one refers to the color depth ( monochrome ) . Each convolution layer that we applied is zero-padded to maintain the same height and width dimension . Additionally , each convolution layer is followed by batch normalization and ReLU activation . First , we applied two sequential 3D convolution layers with a kernel size of 3 × 3 × 3 and number of filters of 4 . Second , we applied a max pooling layer of shape 2 × 2 × 2 to result in a new tensor shape of 8 × 64 × 64 × 4 . We repeated this two 3D convolution and max pool , doubling the filter depth each time , an additional three more times which results in a 1 × 8 × 8 × 32 tensor shape . We applied two final 3D convolutions with a 1 × 3 × 3 kernel size and 64 filter depth , resulting in a 1 × 8 × 8 × 64 tensor shape . Here , we flattened the network to produce a 64 × 64 tensor . After flattening we applied two fully connected layers , each with 128 filter depth , batch normalization , and ReLU activations . Finally , we added one more fully connected layer with only two filter depth and a softmax activation . This final layer was used as the output probabilities for not grooming and grooming predictions for the last of the 16 frames . We trained four individual neural networks using the same training set and four independent initializations . During training , we randomly sample video chunks from the data set where the final frame contains an annotation where the annotators agree . Since we sample a 16 frame duration , this refers to the 16th frame’s annotation . If a frame selected does not have 15 frames of video earlier , the tensor is padded with 0-initialized frames . We apply random rotations and reflections of the data , achieving an 8x increase in effective data set size . The loss function we use in our network is a categorical cross entropy loss , comparing the softmax prediction from the network to a one-hot vector with the correct classification . We use the Adam optimizer with an initial learning rate of 10-5 . We apply a decay schedule of learning rate to halve the learning rate if five epochs persist without validation accuracy increase . We also employ a stop criteria if validation accuracy does not improve by 1% after 10 epochs . During training , we assemble a batch size of 128 example video clips . Typical training would be done after 13–15 epochs , running for 23–25 epochs without additional improvement . We trained JAABA classifiers using two different approaches . Our first approach was using the guidelines provided by the software developers ( JAABA Interactive Training ) . This involves interactively and iteratively training classifiers . The data selection approach is to annotate some data , then prioritize new annotations where the algorithm is unsure or incorrectly making predictions . We continued this interactive training until the algorithm no longer made improvements in a k-fold cross validation . Our second approach was to subset our large annotated data set to fit into JAABA and train on the agreeing annotations . Initially , we attempted to utilize the entire training data set , but our machine did not have enough RAM to handle the entire training data set . The workstation we used contained 96 GB of available RAM . We wrote a custom script to convert our annotation format to populate annotations in a JAABA classifier file . To confirm our data was input correctly , we examined the annotations from within the JAABA interface . Once we created this file , we could simply train the JAABA classification using JAABA’s interface . After training , we applied the model to the validation data set to compare with our neural network models . We repeated this with various sizes of training data sets . Here , we describe a variety of grooming behavioral metrics that we use in following analyses . Following the approach that has previously been proposed , we define a single grooming bout as a duration of continuous time spent grooming without interruption that exceeds 3 s ( Kalueff et al . , 2010 ) . We allow brief pauses ( less than 10 s ) , but do not allow any locomotor activity for this merging of time segments spent grooming . Specifically , a pause occurs when motion of the mouse does not exceed twice its average body length . In order to reduce the complexity of the data , we summarize the grooming duration , number of bouts , and average bout duration into 1 min segments . In order to have a whole number of bouts per time duration , we assign grooming bouts to the time segment when a bout begins . In rare instances where multiple-minute bouts occur , this allows for a 1 min time segment to contain more than 1 min worth of grooming duration . From here , we sum the total duration of grooming calls in all grooming bouts to calculate the total duration of grooming . Note that this excludes un-joined grooming segments less than 3 s duration as they are not considered a bout . Additionally , we count the total number of bouts . Once we have the number of bouts and total duration , we calculate the average bout duration by dividing the two . Finally , we bin the data into one minute time segments and fit a linear line to the data . Positive slopes for total grooming duration infer that the individual mouse is increasing its time spent grooming the longer it remains in the open field test . Negative slopes for total grooming duration infer that the mouse spends more time grooming at the start of the open field test than at the end . This is typically due to the mouse choosing to spend more time doing another activity over grooming , such as sleeping . Positive slopes for number of bouts infer that the mouse is initiating more grooming bouts the longer it remains in the open field test . For k-means clustering of the behavior across strains , we visually inspected the data and decided three clusters was optimal . We z-score transformed grooming features as inputs to the k-means algorithm to determine cluster membership . Finally , we projected the clusters discovered by the k-means to a 2D space formed by principal components ( PC ) . Protocols for data collection were previously described in Geuther et al . , 2019 . In brief , each animal was video recorded from a top-down viewpoint for 55 min of novel open field exposure . Imaging parameters were held constant across videos , including camera distance , zoom , frame rate , and background conditions . No power analysis was used sample size for C57BL/6NJ vs C57BL/6J data since this data is longitudinal control data . Power analysis for the strain survey data showed that with 16 animals ( 8M/8F ) , we have 80% power to detect a effect size of 1 ( Cohen’s d ) . Outliers in the strain survey were removed when individuals measured a value of G⁢r⁢T⁢i⁢m⁢e⁢55⁢m<Q1-1 . 5*I⁢Q⁢R or G⁢r⁢T⁢i⁢m⁢e⁢55⁢m>Q3+1 . 5*I⁢Q⁢R , where Q1 is the first quartile , Q3 is the third quartile , and I⁢Q⁢R is the interquartile range . All behavioral data will be available in the Mouse Phenome Database ( MPD ) , and code and models will be available in Kumar Lab Github account ( https://github . com/KumarLabJax and https://www . kumarlab . org/data/ ) . The phenotypes obtained by the machine learning algorithm for several strains were used to study the association between the genome and the strains behavior . A subset of ten individuals from each combination of strain and sex were randomly selected from the tested mice to ensure equal within group sample sizes . The genotypes of the different strains were obtained from the mouse phenome database ( https://phenome . jax . org/genotypes ) . The Mouse Diversity Array ( MDA ) genotypes were used , di-allele genomes were deduced from parent genomes . SNPs with at least 10% MAF and at most 5% missing data were used , resulting with 222 , 967 SNPs out of 470 , 818 SNPs genotyped in the MDA array . LMM method from the GEMMA software package ( Zhou and Stephens , 2012 ) was used for GWAS of each phenotype with the Wald test for computing the p-values . A Leave One Chromosome Out ( LOCO ) approach was used , each chromosome was tested using a kinship matrix computed using the other chromosomes to avoid proximal contamination . Initial results showed a wide peak in chromosome seven around the Tyr gene , a well-known coat-color locus , across most phenotypes . To control for this phenomenon , the genotype at SNP rs32105080 was used as a covariate when running GEMMA . Sex was also used as a covariate . To evaluate SNP heritability , GEMMA was used without the LOCO approach . The kinship matrix was evaluated using all the SNPs in the genome and GEMMA LMM output of the proportion of variance in phenotypes explained – the PVE and the PVESE were used as chip heritability and its standard error . To estimate the LD decay , pairs of SNPs that are at most 2 . 5 Mbp apart had their genotypes correlation computed using Pearson correlation . The pairs of SNPs were divided into bins according to their distance , each bin size being 5000 bp . The average correlation r2 coefficient of pairs of SNPs in each bin were averaged and plotted against the average SNPs distance and smoothed using loess function . Correlation of r2>0 . 2 was set as a threshold for assigning SNPs to the same QTL . QTL were determined by sorting the SNPs according to their p-values , then , for each SNP , determining a locus centered at this SNP by adding other SNPs with high correlation ( r2>0 . 2 ) to the peak SNP . Each locus was limited to 10 Mbp from the initial peak SNP selected upstream and downstream . The peak SNPs were aggregated from all the phenotypes and the p-values were used to cluster the peaks into clusters using the k-means algorithm implemented in R . We repeated k-means clustering for a variety of number of clusters and visually decided that seven clusters was appropriate for this data . The GWAS results of phenotypes in each cluster were combined by taking the minimal p-value of all the phenotypes in each cluster , for each SNP . The entire set of phenotypes was also combined in the same manner . The GWAS execution was wrapped in an R package called mouseGWAS available on github: https://github . com/TheJacksonLaboratory/mousegwas; Geuther , 2021; copy archived at swh:1:rev:5d2caac2637da442f4b9648ac1eb1f35bd1136cf it also includes a singularity container definition file and a nextflow pipeline for regenerating the results . The command used for generating the results is:export GH=https://raw . githubusercontent . com/TheJacksonLaboratory/mousegwas/grooming nextflow run TheJacksonLaboratory/mousegwas -r grooming \ --yaml $GH/example/grooming_nowild . yaml \ --shufyaml $GH/example/grooming_shuffle . yaml \ --input $GH/example/grooming_paper_strain_survey_2019_11_21 . csv \ --outdir grooming_output --addpostp="--loddrop 0" -profile slurm , singularity . In order to carry out such cross-species association and link the mouse genetic circuit of grooming to human phenotypes , we conducted a phenome-wide association study ( PheWAS ) . First , we identified human orthologs of the 860 mouse grooming and open field genes with at least 6 degrees of pleiotropy . For each human ortholog , we downloaded PheWAS summary statistics from gwasATLAS ( https://atlas . ctglab . nl/ , release 2: v20190117 ) ( Watanabe et al . , 2019 ) . We focused on the association in the Psychiatric domain with gene-level p value ≤ 0 . 001 . Second , in order to visualize and cluster these associations , we represented the relationships between genes and psychiatric traits by a weighted bipartite network , in which the width of an edge between a gene node and a Psychiatric trait node is proportional to the association strength ( -log10 ( p value ) ) . The size of a node is proportional to the number of associated genes or traits and the color of a trait node corresponds to the subchapter level in the Psychiatric domain . To identify modules within this network , we applied an improved community detection algorithm for maximizing weighted modularity in weighted bipartite networks ( Dormann and Strauss , 2014 ) , using bipartite R package ( Dormann et al . , 2008 ) . All the networks were visualized using Gephi 0 . 9 . 2 software ( Bastian et al . , 2009 ) .

Behavior is one of the ultimate and most complex outputs of the body’s central nervous system , which controls movement , emotion and mood . It is also influenced by a person’s genetics . Scientists studying the link between behavior and genetics often conduct experiments using animals , whose actions can be more easily characterized than humans . However , this involves recording hours of video footage , typically of mice or flies . Researchers must then add labels to this footage , identifying certain behaviors before further analysis . This task of annotating video clips – similar to image captioning – is very time-consuming for investigators . But it could be automated by applying machine learning algorithms , trained with sufficient data . Some computer programs are already in use to detect patterns of behavior , however , there are some limitations . These programs could detect animal behavior ( of flies and mice ) in trimmed video clips , but not raw footage , and could not always accommodate different lighting conditions or experimental setups . Here , Geuther et al . set out to improve on these previous efforts to automate video annotation . To do so , they used over 1 , 250 video clips annotated by experienced researchers to develop a general-purpose neural network for detecting mouse behaviors . After sufficient training , the computer model could detect mouse grooming behaviors in raw , untrimmed video clips just as well as human observers could . It also worked with mice of different coat colors , body shapes and sizes in open field animal tests . Using the new computer model , Geuther et al . also studied the genetics underpinning behavior – far more thoroughly than previously possible – to explain why mice display different grooming behaviors . The algorithm analyzed 2 , 250 hours of video featuring over 60 kinds of mice and thousands of other animals . It found that mice bred in the laboratory groom less than mice recently collected from the wild do . Further analyses also identified genes linked to grooming traits in mice and found related genes in humans associated with behavioral disorders . Automating video annotation using machine learning models could alleviate the costs of running lengthy behavioral experiments and enhance the reproducibility of study results . The latter is vital for translating behavioral research findings in mice to humans . This study has also provided insights into the amount of human-annotated training data needed to develop high-performing computer models , along with new understandings of how genetics shapes behavior .

2021

Action detection using a neural network elucidates the genetics of mouse grooming behavior

RasGRPs are guanine nucleotide exchange factors that are specific for Ras or Rap , and are important regulators of cellular signaling . Aberrant expression or mutation of RasGRPs results in disease . An analysis of RasGRP1 SNP variants led to the conclusion that the charge of His 212 in RasGRP1 alters signaling activity and plasma membrane recruitment , indicating that His 212 is a pH sensor that alters the balance between the inactive and active forms of RasGRP1 . To understand the structural basis for this effect we compared the structure of autoinhibited RasGRP1 , determined previously , to those of active RasGRP4:H-Ras and RasGRP2:Rap1b complexes . The transition from the autoinhibited to the active form of RasGRP1 involves the rearrangement of an inter-domain linker that displaces inhibitory inter-domain interactions . His 212 is located at the fulcrum of these conformational changes , and structural features in its vicinity are consistent with its function as a pH-dependent switch . The Ras family of small G-proteins , including Ras and Rap , are molecular switches that transmit signals when bound to GTP ( Rojas et al . , 2012 ) . Ras-family members are activated by specific guanine-nucleotide exchange factors ( that we term GEFs for simplicity here ) , such as Son of Sevenless , that activates Ras ( Boriack-Sjodin et al . , 1998 ) , and Epac1 and Epac2 , that activate Rap ( Rehmann et al . , 2006 ) . GEFs activate Ras-family members by triggering the release of GDP and its replacement by GTP ( Wittinghofer and Vetter , 2011 ) . These GEFs , of which there are several distinct families in humans , activate Ras-family members in response to a variety of signals , and thereby initiate downstream kinase signaling cascades ( Bos et al . , 2007; Cherfils and Zeghouf , 2013 ) . One family of these GEFs is comprised of the Ras guanine-nucleotide releasing proteins ( RasGRPs ) that can activate Ras or Rap , of which there are four in humans . RasGRP1 is most extensively studied in T lymphocytes ( Dower et al . , 2000; Priatel et al . , 2002; Roose et al . , 2005; Daley et al . , 2013 ) , RasGRP2 in neutrophils and platelets ( Lozano et al . , 2016; Stone , 2011 ) , RasGRP3 in B lymphocytes ( Teixeira et al . , 2003 ) , and RasGRP4 in mast cells ( Yang et al . , 2003 ) . However , RasGRPs can be found in other cell types as well , including epithelial lineages ( Depeille et al . , 2015 ) . T cell receptor ( TCR ) or B cell receptor ( BCR ) stimulation leads to an increase of diacylglycerol at the membrane , protein kinase C ( PKC ) activation , and increases in intracellular calcium levels ( Myers and Roose , 2016 ) . RasGRP1 is recruited to the plasma membrane , where it binds diacylglycerol to enable interaction with Ras . Upon receptor stimulation , RasGRP1 is phosphorylated by PKC ( Zheng et al . , 2005; Roose et al . , 2007 ) , which enhances RasGRP1’s GEF activity through unknown mechanisms . RasGRP1 contains a calcium-binding EF hand domain and calcium binding to RasGRP1 induces allosteric changes that release autoinhibition ( Iwig et al . , 2013 ) . Dysregulated RasGRPs can cause aberrant signaling and result in disease . Alterations in RasGRP1 expression contribute to diseases such as skin carcinoma ( Diez et al . , 2009; Luke et al . , 2007 ) , colorectal cancer ( Depeille et al . , 2015 ) , and leukemia ( Ksionda et al . , 2016; Hartzell et al . , 2013 ) . Elevated RasGRP3 expression has been reported in breast cancer ( Nagy et al . , 2014 ) , prostate cancer ( Yang et al . , 2010 ) , lymphoma ( Teixeira et al . , 2003 ) , cutaneous melanoma ( Yang et al . , 2011 ) , and uveal melanoma ( Chen et al . , 2017 ) . A polymorphism ( Arg 519 Gly ) in Rasgrp1 in mice results in T cell abnormalities and autoimmunity ( Daley et al . , 2013 ) . Furthermore , several genetics studies have linked single nucleotide polymorphisms ( SNPs ) in RasGRP1 to human autoimmune disease ( Plagnol et al . , 2011; Qu et al . , 2009 ) , and low RasGRP1 levels have been detected in T lymphocytes from patients with Systemic Lupus Erythematosus ( SLE ) ( Yasuda et al . , 2007 ) , and rheumatoid arthritis ( RA ) ( Golinski et al . , 2015 ) . Complete RasGRP1 deficiency in a patient leads to a novel primary immunodeficiency , with impaired activation and proliferation of the patient's T- and B- cells and defective killing by cytotoxic T cells and NK cells ( Roose , 2016; Salzer et al . , 2016 ) . Rasgrp2 deficiency in mice results in excessive bleeding , caused by defective platelets aggregation and degranulation ( Crittenden et al . , 2004 ) . Moreover , polymorphisms in RasGRP2 , either converting Arg 113 into a stop codon , or missense mutations ( Gly 248 Trp or Ser 381 Phe ) , cause a platelet disorder in patients ( Lozano et al . , 2016;Canault et al . , 2014 ) . The N-terminal portion of the RasGRPs contains the catalytic module that is common to other GEFs that operate on Ras-family members ( Figure 1A ) . This module consists of a Ras-exchanger motif ( REM ) domain followed by a Cdc25 domain ( Boguski and McCormick , 1993;Fam et al . , 1997 ) . As first revealed by structural analysis of the Ras-specific GEF Son-of-Sevenless ( SOS ) , the Cdc25 domain interacts with Ras and is responsible for nucleotide release , and the REM domain provides structural support for the Cdc25 domain ( Boriack-Sjodin et al . , 1998 ) . The remaining portion of the RasGRP proteins consists of a Ca2+-binding EF domain and a C1 domain that binds diacylglycerol for membrane localization ( Beaulieu et al . , 2007; Ebinu et al . , 1998; Zahedi et al . , 2011 ) . RasGRP1 has a C-terminal coiled-coil domain that is missing in the other family members . The common set of domains in RasGRP-1 , -2 , -3 , and -4 show a high degree of sequence conservation ( Ksionda et al . , 2013 ) . A nearly complete understanding of the regulation of SOS has been provided through structural and functional studies ( reviewed in Jun et al . , 2013 ) ; The exchange-factor activity of the catalytic module of SOS is inhibited by the action of the N- and C-terminal segments ( Sondermann et al . , 2004 ) , and the activation of SOS requires allosteric feedback from Ras•GTP binding to a site that is distal to the catalytic site where nucleotide is exchanged ( Roose et al . , 2007; Margarit et al . , 2003; Boykevisch et al . , 2006 ) . Once activated , SOS requires multiple plasma membrane-anchoring mechanisms to signal efficiently to Ras ( Findlay et al . , 2013; Christensen et al . , 2016 ) , and SOS signaling is terminated in part via endocytosis ( Christensen et al . , 2016 ) . The regulation of the Rap-specific GEF Epac2 is also well understood ( Rehmann et al . , 2006; Rehmann et al . , 2008 ) . Much less is known about the regulation and activation of any of the four RasGRP proteins . We had previously analyzed the structural basis for the autoinhibition of RasGRP1 , which is very different from that of SOS ( Figure 1B ) . In SOS , the active site is open in the autoinhibited form , and inactivation appears principally to be the result of the blockage of allosteric Ras binding . Allosteric Ras binding causes a conformational change at the active site in SOS . In contrast , in RasGRP1 , the active site is blocked by the linker connecting the Cdc25 domain to the EF domain ( the Cdc25-EF linker ) ( Iwig et al . , 2013 ) . In addition , we proposed that dimerization of RasGRP1 through the C-terminal coiled-coil domain results in blockage of the membrane-interacting surface of the C1 domain . In the structure of autoinhibited RasGRP1 , the EF and C1 domains are docked on the base of the Cdc25 domain , and this interaction steers the Cdc25-EF linker through the Ras-binding site of the Cdc25 domain ( Iwig et al . , 2013 ) . Calcium binding to the EF-hand of RasGRP1 promotes a conformational change that is likely to result in displacement of the Cdc25-EF linker from the catalytic site ( Iwig et al . , 2013 ) . Thus , the EF domain plays both autoinhibitory and activating roles in regulating RasGRP1 , and a point mutation in the EF domain leads to an autoimmune phenotype in mice ( Daley et al . , 2013 ) . Sequence comparisons suggest that this autoinhibitory mechanism is conserved in the three other RasGRP proteins . By scanning and testing human single nucleotide variants ( SNVs ) from genome databases , we uncovered that a conserved histidine residue in the Cdc25 domain of RasGRP1 ( His 212 ) is critical for autoinhibition . We show that the activity of RasGRP1 is sensitive to cellular pH , and that His 212 is critical for this pH sensitivity . Stimulation of lymphocytes results in an increase in intracellular pH ( pHi ) ( Cheung et al . , 1988; Fischer et al . , 1988;Mills et al . , 1985 ) , which can lead to deprotonation and conversion of histidine residues from positively charged to neutral ( Schönichen et al . , 2013 ) . We find that increasing pHi synergizes with receptor stimulation to activate the cellular RasGRP1-Ras-ERK pathway . Conversely , replacing His 212 by a positively charged lysine residue ( H212K ) prevents the recruitment of RasGRP1 to the membrane , presumably by stabilizing the autoinhibited form . In order to understand the role of His 212 in regulation of RasGRP1 we sought to compare the structure of autoinhibited RasGRP1 , determined previously ( Iwig et al . , 2013 ) , to that of the active form in complex with Ras . We have been unable to crystallize a RasGRP1:Ras complex . Instead , we determined the structures of the catalytic modules of two other members of the family , RasGRP4 and RasGRP2 , bound to nucleotide-free HRas and Rap1B , respectively . These structures reveal a key role for the REM-Cdc25 linker in determining the switch from the autoinhibited to the active states . The structure of this linker is fully resolved in the RasGRP2:Rap1B complex , and the linker conformation is incompatible with the inhibitory docking of the EF domain on the Cdc25 domain . The sequence of the REM-Cdc25 linker is conserved between RasGRP1 and RasGRP2 , and we infer that the activation of the RasGRP1 involves correlated movements in the REM-Cdc25 linker and the Cdc25-EF linker . The structures show that the location of His 212 is such that the charge on this residue could alter the balance between the active and inactive states of RasGRP1 . Single nucleotide variants that cause amino acid substitutions ( missense variants; SNVs ) are frequent and generate human genetic variation: most people inherit ∼12 , 000 missense gene variants ( Abecasis et al . , 2010 ) . We identified all SNVs reported for RasGRP1 in publicly available databases , and took a shotgun approach to test a panel of these SNVs for their ability to alter RasGRP1 regulation ( Figure 1C ) . We have not been able to express full length RasGRP1 and study the autoinhibitory mechanisms in cell-free assays ( Iwig et al . , 2013 ) . Therefore , to assess the potential signaling impact of these SNVs , we used RasGRP1−/−RasGRP3−/− DT40 cells ( a chicken B cell line ) genetically deleted for RasGRP1 and RasGRP3 and reconstituted these cells via transfection with wildtype EGFP-RasGRP1 ( WT ) or a catalytically inactive RasGRP1 ( Arg271Glu ) as before ( Iwig et al . , 2013 ) , or with a panel of RasGRP1 SNVs ( indicated in bold and blue in Figure 1C ) . This assay allows for activity assessment of RasGRP1-ERK signaling ( Iwig et al . , 2013 ) , but also of RasGRP1-mTORC1-p70S6 kinase signaling resulting in phosphorylation of ribosomal protein S6 ( P-S6 ) . Precisely how RasGRP1 signals to the S6 pathway is still unresolved and is not the focus of this study here , but the Arg 519 Gly mutation in Rasgrp1Anaef mice results in higher basal S6 signaling , T cell abnormalities , and autoimmunity ( Daley et al . , 2013 ) . To assess the basal activity of RasGRP1 and its SNVs , we used quantitative flow cytometric analysis of phosphorylated ERK ( P-ERK ) and phosphorylated ribosomal protein S6 ( P-S6 ) levels as a function of the expression level of RasGRP1-EGFP ( Figure 1D ) . Our quantitative flow cytometric analyses revealed that RasGRP1 signals strongly to P-S6 in the basal state; basal signals from RasGRP1 to ERK do occur , but are more modest ( Figure 1E ) . Most SNVs were neutral , with signaling features either similar to WT RasGRP1 or with lower activity . There are numerous possible reasons for SNVs signaling at lower strength , including reduction in protein stability ( data not shown ) . However , the His 212 Tyr SNV signaled more strongly to ERK than WT , indicating altered regulation of RasGRP1 . More detailed analysis of the cellular biochemical traits of the SNVs in the His 212 region demonstrated that His 212 Tyr , but not Ser 220 Leu , Phe 221 Cys , and Phe 226 Leu , resulted in increased basal signals to P-ERK and P-S6 in unstimulated cells as compared to WT RasGRP1 ( Figure 1F ) . His 212 in RasGRP1 is conserved among all vertebrate RasGRPs , and is present in most RasGRP proteins from lower organisms ( Figure 2A ) . This residue is located in the first helix of the Cdc25 domain , and is far from the Ras-binding site . We assessed the activity of RasGRP1 bearing mutations at position 212 in transfected cells . Analysis of the human SNV variant His 212 Tyr and His 212 Ala , both alterations to neutral residues , showed increased basal signals to P-ERK in RasGRP1−/−RasGRP3−/− DT40 cells ( Figure 2B , Figure 2—figure supplement 1A ) as well as in JPRM441 ( Figure 2C , Figure 2—figure supplement 1B ) , a RasGRP1-deficient Jurkat T cell leukemia line that we previously exploited to asses RasGRP1 function ( Roose et al . , 2005; Iwig et al . , 2013 ) . Similarly , the His 212 Tyr and His 212 Ala variants of RasGRP1 signaled stronger to P-S6 in the DT40 cell system ( Figure 2D , Figure 2—figure supplement 1C ) ; PTEN- and SHIP1-deficiency in Jurkat results in hyperactive PI3kinase signals ( Abraham and Weiss , 2004 ) , making analysis of PI3K-dependent signals , such as those to S6 , difficult in Jurkat . Thus , the His 212 Tyr and His 212 Ala variants are less autoinhibited . Mutation of His 212 to positively charged residues , i . e . His 212 Arg and His 212 Lys , maintained autoinhibition of RasGRP1 activity in unstimulated cells with a similar efficiency as WT RasGRP1 ( Figures 2E , F and G , Figure 2—figure supplement 1D , E and F ) . Note that our analysis of the effects of these , and other , mutations in RasGRP1 is necessarily restricted to cell-based assays . As shown previously , the RasGRP1 construct used here is released from autoinhibition when studied in vitro , in solution ( Iwig et al . , 2013 ) . The inability of in vitro measurements to capture details of the regulatory mechanism may reflect the role of dimerization in maintaining autoinhibition – the construct used in these studies lacks the C-terminal dimerization domain , and we have not yet succeeded in expressing and purifying full-length RasGRP1 . The same transfection phospho-flow strategy also allows for evaluation of RasGRP1 function following B cell receptor ( BCR ) stimulation , by analyzing kinase pathway signals in cells that express low levels of RasGRP1-EGFP , which we gate on . Upon BCR stimulation with M4 antibody , we typically observe induced levels of P-ERK and P-S6 that are 7–9 fold and 2–3 fold over baseline , respectively ( Figure 3A ) . Note that induction of ERK phosphorylation is more robust than S6 phosphorylation upon BCR stimulation . Thus , S6 phosphorylation appears to be a more robust event in the basal state ( Figure 1 ) and ERK phosphorylation is a more robustly triggered kinase pathway in stimulated cells . To understand the role of His 212 in the activation of RasGRP1 , we used our transfection phospho-flow platform and BCR stimulation . Changing His 212 to neutral residues , as in the human SNV His 212 Tyr or the designed His 212 Ala variant , resulted in increased levels of BCR-induced phosphorylation of ERK and S6 ( Figure 3B and C ) . Conversely , altering His 212 to positively charged residues , either His 212 Arg or His 212 Lys , impaired the BCR-induced ERK and S6 responses ( Figure 3D and E ) . The transfection platform with EGFP-tagged RasGRP1 also allowed us to investigate the recruitment of RasGRP1 to the plasma membrane ( Bivona et al . , 2003; Daley et al . , 2013 ) . Wildtype and His 212 Tyr versions of RasGRP1-EGFP revealed very similar efficiency of plasma membrane recruitment upon BCR-stimulation ( Figure 3F and G , and Figure 3—figure supplement 1 ) . Most striking was the complete lack of RasGRP1 plasma membrane recruitment for the positively charged mutant His 212 Lys ( Figure 3G and Figure 3—figure supplement 1 ) . This is remarkable because BCR-stimulation generates a robust intracellular calcium flux and increased levels of DAG ( Myers and Roose , 2016 ) , which recruits RasGRP1 to the membrane and activates it . Thus , the His 212 Lys mutation blocks the responsiveness of RasGRP1 to these calcium and DAG cues . Together , these data suggest that the positive charge of His 212 contributes to RasGRP1 autoinhibition in the basal state ( Figure 2 ) , and that a neutral residue at position 212 is essential to allow for RasGRP1 plasma membrane recruitment and signal output to the ERK and S6 pathways ( Figure 3 ) . Given that increases in intracellular pH can result in deprotonation of histidines ( Schönichen et al . , 2013 ) ( Figure 4A ) , we next explored whether RasGRP1 is regulated by intracellular pH through His 212 . Cells in an activated state often display increases in intracellular pH ( pHi ) ; For example , the pHi is elevated in tumor cells , which promotes tumor survival and progression ( Webb et al . , 2011; Grillo-Hill et al . , 2015 ) . Stimulation of DT40 cells by PMA ( a diacylglycerol analogue ) and ionomycin ( calcium ionophore ) , mimicking lymphocyte receptor signaling events that connect to RasGRP1 , resulted in a consistent increase of the intracellular pH ( Figure 4B ) . Intracellular pH can be experimentally changed by regulating the activity of plasma membrane H+ and HCO3− ion transporters . Additionally , treating cells with low concentrations ( 10–20 mM ) of ammonium chloride ( NH4Cl ) increases pHi , and pulsing with NH4Cl results in a substantially lower pH with NH4Cl removal . Decreased pHi after pulsing is retained in the presence of EIPA ( a specific inhibitor of the Na+/H+ exchanger NHE1 ) by blocking H+ efflux ( Figure 4C ) ( Nakanishi et al . , 1992 ) . In our transfected RasGRP1−/−RasGRP3−/− DT40 cells , treating with 15 mM NH4Cl increased pHi , whereas a NH4Cl pulse followed by EIPA treatment resulted in a trend towards decreased pHi ( Figure 4D ) . We chose NH4Cl and EIPA to modulate the pHi so that we could subsequently assess how this may impact RasGRP1 activity following BCR stimulation of cells . As such , we treated cells with NH4Cl or NH4Cl + EIPA prior to BCR receptor stimulation . NH4Cl pretreatment was compared to vehicle pretreatment , and resulted in augmentation of the BCR-induced ERK phosphorylation in RasGRP1−/−RasGRP3−/− DT40 cells that were transfected with WT RasGRP1 ( Figure 4E ) . In contrast , NH4Cl + EIPA treatment did not increase BCR-induced ERK signaling , but resulted in a trend towards decreased ERK signals ( Figure 4F ) . It should be noted that lowering of pHi has been reported to activate cell death pathways ( Lagadic-Gossmann et al . , 2004 ) . However , in our experiments the value of pHi never reached levels low enough to induce cell death effects that are observed under some conditions when pHi <7 . 0 ( Matsuyama et al . , 2000 ) . With these approaches , we investigated if the augmentation of BCR-induced ERK signals at increased pHi was the result of deprotonation of His 212 in RasGRP1 expressed in these cells . We used untransfected cells or RasGRP1−/−RasGRP3−/− DT40 cells transfected with WT RasGRP1 , His 212 Tyr , or His 212 Lys . First , we established that all four cell populations demonstrated very similar pHi increases or decreases upon NH4Cl incubation or a NH4Cl pulse with EIPA , respectively ( Figure 4G ) . We found that increased pHi ( Figure 4H ) or decreased pHi ( Figure 4I ) only impacted the level of ERK signaling when His 212 in RasGRP1 was intact . His 212 Tyr and His 212 Lys lost the pHi-dependent regulation . These data indicate that His 212 in RasGRP1 is a pH sensor , and that increases in intracellular pH and His 212 deprotonation promote RasGRP1 activation . In order to understand the role of His 212 in controlling the activation state of the protein , we wished to determine the structure of RasGRP1 bound to Ras , i . e . , in an active form , but we failed to crystallize this complex . In order to obtain structures of active complexes we shifted our focus to other RasGRP family members , since they are closely related in sequence ( Figure 5A and Figure 5—figure supplement 1 ) . The different RasGRPs have been reported to prefer different members of the Ras subfamily , based on gene knockout experiments ( see review [Ksionda et al . , 2013] ) . We first determined the specificity of all four RasGRP family members for Ras and Rap , using an in vitro GEF assay and only the catalytic domains of RasGRPs . We found that RasGRP1 and RasGRP4 are specific for Ras , whereas RasGRP2 has a clear preference for Rap1B . RasGRP3 works equally efficiently as a GEF for Ras , Rap1B and Rap2A ( Figure 5B ) . These findings are in agreement with published work that compared a large panel of GEFs and their preference for specific small GTPases as substrates ( Popovic et al . , 2013 ) . We were successful in crystallizing the catalytic modules of RasGRP4 bound to HRas and RasGRP2 bound to Rap1B ( see Materials and methods for details ) . Comparison of the structures of the RasGRP4:HRas and RasGRP2:Rap1B complexes show that nucleotide-free HRas and Rap1B are bound in a similar fashion to the two RasGRP proteins ( Figure 5C and D ) , and that the binding mode is similar to that seen in SOS:HRas ( Boriack-Sjodin et al . , 1998 ) and Epac2:Rap1B ( Rehmann et al . , 2008 ) complexes ( Figure 5E and F ) . An intriguing aspect of the RasGRP2:Rap1B structure is that the linker connecting the REM domain to the Cdc25 domain ( REM-Cdc25 linker ) is ordered in its entirety ( Figure 6A and B ) . By contrast , this linker is disordered in the RasGRP4 structure ( Figure 6C ) . The sequence of the linker is conserved between RasGRP1 and RasGRP2 , but is drastically different in RasGRP4 ( Figure 6D ) . The structure of the linker as visualized in the RasGRP2 structure is of potential functional importance , because it is inconsistent with adoption of the autoinhibited structure of RasGRP1 that we have defined earlier ( Iwig et al . , 2013 ) ( compare Figure 6A and B with Figure 6E ) . In the structure of autoinhibited RasGRP1 , the EF domain is docked on top of helix J of the Cdc25 domain ( Figure 6E; the helices are labeled as defined for the Cdc25 domain of SOS [Boriack-Sjodin et al . , 1998] ) . The docking of the EF domain maintains RasGRP1 in an inactive state in two ways . First , the Cdc25-EF linker physically blocks the active site . Second , in the autoinhibited dimer , interactions between the EF domain of one RasGRP1 molecule and the C1 domain of the other block the membrane-interacting face of the C1 domain ( see Figure 1B and [Iwig et al . , 2013] ) . In the structure of RasGRP2:Rap1B , the REM-Cdc25 linker runs along the surface of helix J , and wraps around it by forming an α helix ( Figure 6A ) . The high sequence conservation of the REM-Cdc25 linker between RasGRP1 and RasGRP2 suggests that the linker might adopt the same conformation in the active states of both molecules . Support for this idea is provided by experiments in which we disrupted ion-pairing hydrogen bonds formed between Arg 131 in the REM-Cdc25 linker and Asp 177 in helix B of the Cdc25 domain , in the RasGRP2:Rap1B structure ( Figure 6B ) . We used our cellular assay to test whether mutation of the corresponding residues in RasGRP1 , Arg181 and Asp228 , has any effect on GEF activity . Upon mutation of the ion-pairing residues , only Arg 181 mutation to alanine ( Arg 181 Ala ) resulted in somewhat lower levels of basal RasGRP1 signals to P-ERK ( Figure 6F ) . These mutations did not significantly impact the RasGRP1-p70S6 kinase pathway ( Figure 6F ) , possibly because RasGRP1 signals to S6 are relatively active in the basal state ( Figure 1E ) . Both mutants ( Arg 181 Ala and Asp 228 Ala ) revealed decreased induction of both P-ERK and P-S6 under BCR stimulatory conditions , in comparison to levels induced by wild type RasGRP1 ( Figure 6G ) . Taken together , our results indicate that formation of the ion pair between Arg 181 and Asp 228 is necessary for robust activation of RasGRP1 . Another observation that points to the relevance of this linker conformation is that PMA- or antigen receptor-induced phosphorylation of RasGRP1 at Thr 184 in the REM-Cdc25 linker , or RasGRP3 at the corresponding residue ( Thr 133 ) , is correlated with increased RasGRP activity . Mutation of these threonine residues to alanine results in reduced , but not eliminated , RasGEF activity ( Aiba et al . , 2004 ) , whereas introduction of a negative charge to mimic a phosphorylated threonine results in higher activity ( Roose et al . , 2005 ) . The corresponding residue in RasGRP2 is Thr 134 ( Figure 6D ) and is phosphorylated in lymphocytes analyzed by total phospho-proteomics ( Dr . Doreen Cantrell , set of 11 phospho-peptides from RasGRP2 , personal communication ) . The positioning of the REM-Cdc25 linker in the RasGRP2:Rap1B complexes places the side chain of Thr 134 close to the side chains of the positively charged Lys 172 and Arg 195 in the Cdc25 domain , which provides a possible explanation for the activating effect of phosphorylation of the threonine residue , which adds a negative charge ( Figure 6H ) . The cellular signaling results with variants of His 212 , including the human SNV His 212 Tyr , and the findings that His 212 is a pH sensor , prompted us to investigate how His 212 may impact the transition between the autoinhibited and active states of RasGRP1 . Schematic diagrams for these two states are shown in Figure 7A and B , which indicate the position of His 212 with respect to the structural elements that rearrange in the transition . His 212 is located within helix A of the Cdc25 domain , at the interface between helices A and J . Helix J is connected to helix K , which forms part of the Ras binding site . In the autoinhibited conformation of RasGRP1 , Helix J is part of the platform on which the EF domain is docked ( Figure 7C ) . Thus , the location of His 212 is consistent with a role for this residue in the transition between the active and inactive states of RasGRP1 . In addition , the structure suggests that the charge on the histidine residue will be important . An unusual aspect of this histidine is that there are five negatively charged residues in its vicinity ( Figure 7D ) . Four of these are provided by helix A ( Glu 205 , Glu 207 , Glu 208 and Glu 211 ) and the fifth one ( Glu 404 ) is provided by the loop leading into helix J , which packs against helix A ( Figure 7E ) . We estimated the pKa values of the histidine sidechains in RasGRP1 , using continuum electrostatics as implemented in a web-based server for the program DELPHI ( Sharp and Honig , 1990; Wang et al . , 2016 ) . Using the crystal structure of the cdc25 domain of RasGRP1 ( Iwig et al . , 2013 ) , the pKa value of His 212 is calculated to be 6 . 90 , that is , increased by almost 1 pH unit above the pKa value of an isolated histidine ( 6 . 0 ) . His 212 has the highest calculated pKa value among the nine histidines in the cdc25 , indicating that it is primed to convert from positive to neutral when the intracellular pH increases beyond a neutral value ( Table 1 ) . The calculated pKa value of this residue is reduced significantly when one or more of the four glutamate sidechains located nearby , in helix A , are substituted by alanine . We observed a similar cluster of glutamate residues surrounding a His pH sensor in the focal adhesion-associated protein talin , and computational pKa prediction suggested that these glutamate residues have increased pKa values ( Srivastava et al . , 2008 ) . In both RasGRP1 and talin the glutamate network could be part of the pH sensor with histidine . Supporting this idea , experimental work has shown that histidines with coordinating glutamate residues have increased pKa values when determined experimentally by NMR ( Tishmack et al . , 1997; Hiebler et al . , 2017; Baran et al . , 2008 ) . As noted earlier , mutation of His 212 to non-titratable residues abrogates pH sensitive activation , which is also consistent with a role for this residue in pH sensing . We propose that the configuration of residues seen in the autoinhibited RasGRP1 structure is stable when His 212 is protonated , because the concentration of negative charge in this region would favor a positively charged histidine . Neutralization of the histidine may favor disruption of the clustering of negative charge , most likely by movement of Glu 404 , in the loop leading into helix J ( the other four acidic residues are on the same helix as His 212 , and are less likely to move away ) . In the active structure , the loop bearing Glu 404 packs against the REM-Cdc25 linker ( Figure 7 ) . In the autoinhibited RasGRP1 structure , the REM-Cdc25 linker positions two hydrophobic sidechains ( Leu 199 and Leu 200 ) against the hydrophobic interface formed between helix J and the EF domain ( Figure 7C ) . Release of the leucine residues , due to destabilization of the conformation of the REM-Cdc25 linker , would weaken the interface with the EF domain and help initiate a transition towards the active conformation . Consistent with this idea , we found that mutation of Glu 404 resulted in altered RasGRP1 activity , while mutation of Glu 208 , located within helix A , had no effect . Replacement of Glu 404 by either alanine or arginine both resulted in increased signals to S6 but not to the less basally active RasGRP1-ERK pathway ( Figure 7F and G , and Figure 7—figure supplement 1 ) . In unstimulated cells , RasGRP1 is in an inactive conformation , in which the Cdc25-EF linker prevents Ras binding to the active site ( Iwig et al . , 2013 ) . In dimeric and autoinhibited RasGRP1 , the EF domains from each molecule in a RasGRP1 dimer block the DAG-binding sites on the C1 domains of the dimer partner ( Figure 8 ) . Lymphocyte receptor stimulation results in increased pHi , increased DAG levels , and increased intracelluar calcium levels , and we propose that these three signals coordinately induce a conformational change in RasGRP1 . The structure of the RasGRP2:Rap1B complex that we have now determined establishes an important role for the REM-Cdc25 linker in the transition to the active state . A prominent feature in the active RasGRP2 structure is the formation of an ion-pair between Arg 131 , in the REM-Cdc25 linker and Asp 177 in the Cdc25 domain ( Arg 181 and Asp 228 in RasGRP1 ) . This salt bridge and the position of the REM-Cdc25 linker reinforces the active conformation of RasGRP and is incompatible with the autoinhibited conformation ( Figure 7 ) . The sequence of the REM-Cdc25 linker is conserved between RasGRP1 , RasGRP2 and RasGRP3 ( Figure 6D ) . This suggests that these proteins share a common regulatory mechanism . For RasGRP1 , this mechanism provides the first plausible structural explanation for how phosphorylation of Thr 184 in RasGRP1 ( Zheng et al . , 2005; Roose et al . , 2007 ) results in a stable active conformation ( Figure 8 ) . The sequence of the REM-Cdc25 linker is divergent in RasGRP4 , however , and the sequence of the EF domain indicates that RasGRP4 is unlikely to be regulated by calcium ( Iwig et al . , 2013; Reuther et al . , 2002 ) . Thus it appears that the regulation of RasGRP4 is likely to be different from that of the other three members of the family . Our analysis of SNPs in the RasGRP proteins led to the finding that His 212 in RasGRP1 functions as a pH sensor . Receptor signaling-induced increases in pHi is expected to convert His 212 to the deprotonated and neutral form , which destabilizes the autoinhibited conformation and potentiates the activation of RasGRP1 by calcium flux and DAG formation at the membrane ( Figure 8 ) . His 212 is predicted to have an increased pKa value , which is likely influenced by the adjacent cluster of glutamate residues . Supporting this idea , experimental work has shown that histidines with coordinating glutamate residues have upshifted pKa values when determined experimentally by NMR ( Tishmack et al . , 1997; Hiebler et al . , 2017; Baran et al . , 2008 ) . At this point we cannot formally rule out the possibility that other histidines in RasGRP1 also function as pH sensors . However , the fact that mutation of His 212 to non-titratable residues abrogates pH sensitive activation , strongly suggests that His 212 is essential for the pH-sensitive function of RasGRP . Most studies describing pHi changes upon receptor stimulation in lymphocytes have been published roughly 30 years ago ( Cheung et al . , 1988; Fischer et al . , 1988;Mills et al . , 1985 ) . Experiments in those studies were performed under different conditions with variable outcomes . The exact effects of receptor stimulation on pHi changes in lymphocytes , and the time scales and subcellular localization of these pHi changes remain largely unknown . Likewise , many questions remain unanswered on the subcellular localization of changes in intracellular calcium and the roles of various ion channels in both resting and stimulated lymphocytes ( Feske et al . , 2015 ) . At this point , tools to measure local changes in pHi or calcium levels are not available and how pHi and calcium may regulate signaling proteins locally is not known . Global Ca2+ levels in resting cells are in the 20–200 nM range ( Usachev et al . , 1995; Schwaller , 2010 ) , although it should be noted that there is an enrichment of calcium ions near the negatively charged polar headgroups of phospholipids in the plasma membrane ( Hille , 2001 ) . RasGRP1 binds Ca2+ with low micromolar affinity ( Iwig et al . , 2013 ) , which suggest that RasGRP1 will not bind Ca2+ until the cell is stimulated and calcium levels rise substantially . Debye-Hueckel and Gouy-Chapman models from the early 1900's proposed that ions in an aqueous environment are not evenly distributed . Local entry of calcium through channels in the plasma membrane and local increases in pHi leading to deprotonation of His 212 may provide an interesting interplay in the regulation of RasGRP1 . The connection between deprotonation of His 212 and increases in pHi in stimulated lymphocytes we uncovered here motivates us to revisit these studies on pHi in lymphocytes . More clear is the notion that increased pHi is a hallmark of cancer cells and promotes cancer progression ( Webb et al . , 2011; White et al . , 2017 ) . For example , in drosophila models , increased pHi is sufficient to induce dysplasia , and it potentiates growth and invasion ( Grillo-Hill et al . , 2015 ) . In the light of cancer , it is of interest to note that RasGRP1 expression levels are increased in T cell leukemia ( T-ALL ) cells from patients ( Hartzell et al . , 2013 ) . When we biochemically characterized these T-ALL cell lines with an increased level of RasGRP1 expression we noted that RasGRP1 is constitutively recruited to the membrane where it constantly exchanges the nucleotide on Ras ( Ksionda et al . , 2016; Hartzell et al . , 2013 ) . In agreement , RasGRP1 demonstrates constitutive phosphorylation on Thr 184 in these T-ALL cells ( Ksionda et al . , 2016 ) arguing that RasGRP1 spontaneously adopts an active conformation . We speculate that an elevated pHi in these cancerous T-ALL cells facilitates opening up of the overexpressed RasGRP1 leading to the constitutively high level exchange activity that we observed ( Ksionda et al . , 2016; Hartzell et al . , 2013 ) . Jurkat and low level RasGRP1 expressing JPRM441 cells were previously characterized ( Roose et al . , 2005 ) and described ( Roose et al . , 2007 ) . RasGRP1−/−RasGRP3−/− DT40 cells were described in ( Roose et al . , 2007 ) . Jurkat and JPRM441 were cultured in RPMI1640 ( Hyclone ) , containing 10% Fetal calf serum ( FCS ) , 1% glutamine , 10 mM Hepes , penicillin and streptomycin . DT40 culture medium contained additional 1% chicken serum . After electroporation , cells were recovered in culture medium without penicillin and streptomycin . Starvation of cells was performed in culture medium containing low FCS ( 0 . 2% ) for 3 hr , or in plain RPMI1640 for 30 mins . FACS buffer consisted of Phosphate buffered salt ( PBS ) with 2 mM EDTA , 2% FCS , and 0 . 1% NaN3 . RasGRP1−/−RasGRP3−/− DT40 cells , or low level RasGRP1 expressing JPRM441 cells ( Roose et al . , 2005 ) were transfected by electroporation with DNA plasmids encoding EGFPN1-RasGRP1 wildtype or mutant ( 20 × 106 cells with 20 microgram DNA ) ( Biorad Genepulser Xcel ) . Then cells were rested , stimulated and used for flow cytometry , microscopy , and pHi modulation as described below . The protocol was slightly modified from an earlier published version ( Iwig et al . , 2013 ) . DT40 cells , or JPRM441 cells ( Roose et al . , 2005 ) were transfected as described above , and recovered in culture medium without antibiotics for 3–4 hr . Cells were washed , resuspended in plain RPMI , seeded 0 . 4 × 106 per well in a 96 well round bottom plate , and starved in an incubator for 30 min . DT40 cells were stimulated with anti-IgM ( clone M4 ) , cross-linking the B cell receptor , or with vector ( RPMI ) , and JPRM441 cells were treated with RPMI1640 only for 5 min . Cells were fixated in prewarmed fixations buffer ( BD , Cytofix , BD biosciences , San Jose , CA ) , or in 4% paraformaldehyde in Phosphate buffered saline ( PBS ) , for 15 min at 21°C , washed in FACSbuffer , and permeabilized in either 0 . 5% Phosflow buffer IV ( BD biosciences ) for 15 min at 21°C , or MetOH for 30 min on ice . Barcoding protocols were modified from described methods ( Krutzik et al . , 2011 ) . Pacific Blue and Alexa Fluor 750 carboxilic acid succinimidyl-esters ( Life Technologies , Grand Island , NY ) were added in the methanol or the phosflow buffer IV in titrated serial dilutions , and incubated for 30 min on ice ( Methanol ) , or for 15 min at 21°C ( phosflow buffer IV ) . Cells were washed thoroughly in FACS buffer , and barcoded cells were pooled and incubated for 30 min with antibodies towards P-ERK , P-S6 , and for JPRM441 cells we used cleaved caspase or cleaved PARP to exclude ( pre- ) apoptotic cells from analysis . Cells were washed and analyzed using the LSR II flow cytometer , or LSR Fortessa ( BD Biosciences , San Jose , CA , USA ) . Data were analyzed using Cytobank ( Cytobank Inc , Mountain View , CA , USA ) . In some of the pHi modulation experiments cells were measured without barcoding . Anti-phosphorylated-S6 -PE ( clone D57 . 2 . 2E , #5316 s , diluted 1:200 ) , anti-phosphorylated ERK-AF647 ( clone 137F5 , #5376 , diluted 1:50 ) , cleaved caspase 3- pacific blue ( clone D3E9 , #8788 , diluted 1:200 ) , or anti-phosphorylated ERK ( clone 197G2 , #4377 s , diluted 1:50 ) all from Cell Signaling Technologies , Beverly , MA , USA ) . Unconjugated P-ERK was followed by AffiniPure F ( ab' ) 2 fragment Donkey-anti-Rabbit IgG , conjugated to APC ( #711-136-152 , diluted 1:50 ) or PE ( #711-116-152 , diluted 1:50 , Jackson ImmunoResearch , West Grove , PA , USA ) , or cleaved PARP-AF647 ( clone F21-852 , #558710 , diluted 1:100 , BD Biosciences , San Jose , CA , USA ) RasGRP1-RasGRP3 deficient DT40 cells were transfected by electroporation with DNA plasmids encoding EGFPN1-RasGRP1 . Cells were cultured for 4 hr , and starved for 30 min in RPMI . Cells were stimulated by anti-IgM ( M4 ) , cross-linking the B cell receptor , 20 ng/ml PMA , or with vehicle ( RPMI ) for 2 min . Cells were fixated in 4% paraformaldehyde in PBS , for 15 min , at 21°C , and washed . Cells were FACS sorted ( MoFlo XDP , Beckman Coulter , CA , USA ) for RasGRP1-EGFP low or high expression onto poly-L Lysine coated microscopy slides . For each sample 30–50 RasGRP1-EGFP-low cells were captured on a Zeiss confocal microscope , and scored blindly by 2 independent researchers for localization of RasGRP1 ( cytoplasmic , or membrane . Cells with partial or full membrane localization ( showing a clear line of increased GFP signal at the edge ) were counted as 'membrane localization' . Whilst cells only without clear increased membrane fluorescent signal were counted as 'cytoplasma localization' . Profiles were made with Fiji ( ImageJ ) software . Cells were transfected as described above , recovered for 2 hr in culture medium containing 10% FCS without antibiotics , and then transferred to starvation medium containing 0 . 2% FCS for 3 hr prior to assay . Cells were then were treated with 15 mM ammonium chloride ( NH4Cl ) in RPMI for 10 min to increase pHi , or pulsed with 15 mM NH4Cl for 15 min , spun down to remove NH4Cl , and resuspended in RPMI with 10 mM EIPA ( 5'- ( N-ethyl-N-isopropyl ) amiloride , a selective inhibitor of Na+/H+ exchanger NHE1 activity , Enzo Life Science ) for 5 min to decrease pHi ( Choi et al . , 2013 ) . Next , stimulation with anti-B cell receptor antibody M4 ( 1:1000 ) or 20 ng/ml diacylglycerol analogue PMA ( phorbol 12-myristate 13-acetate ) was performed in presence of NH4Cl , EIPA , or RPMI ( control ) , and after 5 minutes cells were fixated in 4% PFA ( paraformaldehyde ) in PBS ( phosphate buffered salt ) . 0 . 5 × 106 cells per well were plated on a 24 well dishes coated with poly-L Lysine , and induced to be quiescent by maintaining for 3 hr in RPMI with 0 . 2% FCS . Intracellular pH ( pHi ) was determined as previously described ( Choi et al . , 2013 ) in a bicarbonate buffer ( 25 mM HCO3− , 115 mM NaCl , 5 mM KCl , 10 mM glucose , 1 mM KPO4 , 1 mM MgSO4 , and 2 mM CaCl2 , pH 7 . 4 ) by using cells loaded with 1 µM 2′ , 7′-bis- ( 2-carboxyethyl ) −5- ( and-6 ) -carboxyfluorescein ( BCECF; Invitrogen ) . To determine steady-state pHi , fluorescence of BCECF at Ex490/Em530 and Ex440/Em530 was acquired every 15 s for 5 min using a plate reader ( SpectraMax M5; Molecular Dynamics ) and the fluorescence ratios were converted to pHi by calibrating the fluorescence in each well with 10 µM nigericin in 105 mM KCl . In vitro nucleotide exchange rates for RasGRP1-4 proteins ( RasGRP1cat , RasGRP2cat , RasGRP3 , and RasGRP4cat ) and small GTPases ( HRas , Rap1B , and Rap2A ) were measured and analyzed as described previously ( Iwig et al . , 2013 ) , except that the final GTPase concentration was 1 µM instead of 500 nM . RasGRP2cat ( residues 1-394 ) , RasGRP2cat ( 1-425 ) , and RasGRP4cat ( 46-460 ) expression vectors were constructed by inserting the genes into 2CT-10 plasmid ( attaches an N-terminal hexa-histidine tag followed by MBP and TEV protease cleavage site ) , a gift from Scott Gradia ( Addgene plasmid # 55209 ) . RasGRP3cat ( 1-418 ) was cloned into pSMT3 vector . Rap1B ( 1-167 ) , Rap1B ( 1-175 ) , and Rap2A ( 1-167 ) was cloned into pProEX HTb vector ( Invitrogen ) . All proteins correspond to human genes . Non-labeled RasGRPcat domains were expressed in T7 Xpress cells ( NEB ) in Terrific Broth media . RasGRP2cat labeled with selenomethionine ( SeMet ) ( residues 1–394 ) was expressed in BL21 ( DE3 ) cells in M9 minimal media with 50 µg/ml of each amino acid except methionine , 5 µg/ml methionine and 50 µg/ml SeMet . For RasGRPcat domains , the cells were grown at 37°C with 100 µg/ml ampicillin , and induced with 0 . 25 mM IPTG at 15°C for 14–18 hr . The cells were suspended in the Ni-A buffer ( 25 mM Tris pH8 . 0 500 mM NaCl , 10% Glycerol , 20 mM imidazole , and 5 mM β-mercaptoethanol ) supplemented with 200 µM AEBSF , 5 µM leupeptin and 500 µM benzamidine , and lysed by a cell disrupter . The cell lysate was cleared by ultracentrifugation with 142 , 032 x g for 1 hr and applied to HisTrap column ( GE Healthcare ) . The protein was eluted by Ni-A buffer supplemented with 500 mM imidazole . The MBP tag was cleaved off by addition of TEV protease while dialyzing against the buffer ( 25 mM Tris pH8 . 0 , 100 mM NaCl , 10% glycerol , 1 mM TCEP ) . The protein was run through HisTrap column and the flowthrough was collected . The protein was concentrated using Amicon Ultra ultrafiltration unit ( EMD Millipore ) and run on an S-75 size-exclusion column equilibrated with SEC buffer ( 25 mM Tris pH 8 . 0 , 100 mM NaCl , 10% glycerol , 1 mM TCEP ) . The protein was concentrated and stored at −80°C after flash-freezing using liquid nitrogen . HRas ( residues 1–166 ) , Rap1B ( 1-167 ) , Rap1B ( 1-175 ) , and Rap2A ( 1-167 ) were expressed and purified as described previously for HRas ( Iwig et al . , 2013 ) . The RasGRP2:Rap1B complex and the RasGRP4:HRas complex were formed by mixing the proteins at 1:2 ratio of the GEF and either Rap1B or HRas . After addition of alkaline phosphatase ( SIGMA , P0114 ) , the sample was incubated at 4°C overnight . The complexes were purified by gel filtration ( Superdex 75 column ) in SEC buffer . Crystals of SeMet-labeled RasGRP2cat bound to Rap1B were obtained by sitting-drop vapor diffusion ( 500 µL reservoir volume ) by mixing 2 µL of protein ( 30 mg/mL ) with 2 µL of the reservoir solution containing 17% PEG3350 and 10 mM sodium citrate ( pH 5 . 6 ) at 20°C . The crystals were streak-seeded from another crystallization drop using a cat whisker 3 min after mixing the drops . Crystals were harvested for data collection after one month , cryoprotected by incubating them in mother liquor supplemented with 10% xylitol , and flash-frozen in liquid nitrogen . Most of the crystals grown in this condition were in space group P21221 , and exhibited anisotropic diffraction , but the best data were obtained from one crystal with space group I212121 . All X-ray diffraction data were collected at Advanced Light Source beamlines 8 . 3 . 1 and 8 . 3 . 2 . The data were integrated and scaled using XDS ( Kabsch , 2010 ) and Aimless ( Evans , 2006 ) . Crystals of the RasGRP4cat:HRas complex were obtained by hanging-drop vapor diffusion ( 500 µL reservoir volume ) by mixing 1 µL of protein ( 10 mg/mL ) with 1 µL of the reservoir solution containing 20% PEG3350 and 300 mM sodium thiocyanate at 20°C . The crystals were cryoprotected by the reservoir solution supplemented with 20% glycerol before flash-frozen in liquid nitrogen . The space group of the RasGRP4cat:HRas crystals was initially assigned as P6322 , and molecular replacement using Phaser ( McCoy et al . , 2007 ) with the RasGRP1 Cdc25 domain ( PDB ID: 4L9M ) as a search model placed three copies of RasGRP4cat in the asymmetric unit . We were able to place HRas bound to RasGRP4cat as observed in the SOS-HRas complex ( Boriack-Sjodin et al . , 1998 ) , but we could not model the REM domain due to weak electron density . We then processed the data in a lower-symmetry space group ( P3212 ) , The final model contains six copies of the RasGRP4cat:HRas complex . The model was rebuilt using Coot ( Emsley et al . , 2010 ) and the structure was refined by Refmac ( Murshudov et al . , 2011 ) and Phenix ( Adams et al . , 2010 ) in an iterative manner ( Figure 5—source data 1 ) . The application of a twinning operator ( h , k , l → k , h , -l ) gave better electron density for the REM domain . Anomalous diffraction data for crystals of SeMet-labeled RasGRP2cat bound to Rap1B were integrated and scaled as for the RasGRP4cat-HRas data . Phase were determined by molecular replacement using Phaser with RasGRP4cat-HRas as a search model , and refinement was performed with Refmac and Phenix using anomalous data to reduce the model bias with rebuilding the model using Coot ( Figure 5—source data 1 ) .

Complex chain reactions between many kinds of molecules regulate every process in the body . For example , the signaling molecule Ras helps the cell to grow and divide . However , abnormally high levels of Ras signals can cause cancer . Ras is activated by proteins called exchange factors . One of the families of Ras exchange factors – RasGRP – plays important roles in immune and bleeding disorders , and certain cancers . In 2013 , researchers studied the structure of one of these exchange factors , called RasGRP1 , while it was inactivated . Inactive RasGRP proteins have a ‘closed’ structure , which must ‘open up’ when they are activated . Vercoulen , Kondo , Iwig et al . – who include several of the researchers involved in the 2013 study – have now investigated the regulation of various RasGRP family members . The protein structures of the active forms of two family members were determined and compared with the structure of inactive RasGRP1 . In parallel , Vercoulen et al . analyzed how genetic mutations that alter some of the amino acids that make up RasGRP1 affect Ras signaling in cells . This revealed that a particular amino acid , histidine 212 , plays a key role in activating RasGRP1 . Histidines can be in a positively charged or neutral form depending on other surrounding amino acids and on the acidity of the cell’s interior . The interiors of cells that receive an external signal often decrease in acidity , and cancer cells tend to have less acidic interiors than normal cells . Vercoulen et al . reveal that a change in the charge on histidine 212 from positive to neutral opens up the RasGRP1 protein . Histidine 212 therefore acts as an acidity sensor that activates RasGRP1 when the inside of the cell becomes less acidic as external signals are received . Since RasGRP proteins play important roles in many diseases , understanding how cell acidity regulates RasGRPs has wide medical relevance . In the future , the protein structures of the RasGRP family members and the method developed in this study could be used to explore how they contribute to disease .

2017

A Histidine pH sensor regulates activation of the Ras-specific guanine nucleotide exchange factor RasGRP1

PrPC , the cellular isoform of the prion protein , serves to transduce the neurotoxic effects of PrPSc , the infectious isoform , but how this occurs is mysterious . Here , using a combination of electrophysiological , cellular , and biophysical techniques , we show that the flexible , N-terminal domain of PrPC functions as a powerful toxicity-transducing effector whose activity is tightly regulated in cis by the globular C-terminal domain . Ligands binding to the N-terminal domain abolish the spontaneous ionic currents associated with neurotoxic mutants of PrP , and the isolated N-terminal domain induces currents when expressed in the absence of the C-terminal domain . Anti-PrP antibodies targeting epitopes in the C-terminal domain induce currents , and cause degeneration of dendrites on murine hippocampal neurons , effects that entirely dependent on the effector function of the N-terminus . NMR experiments demonstrate intramolecular docking between N- and C-terminal domains of PrPC , revealing a novel auto-inhibitory mechanism that regulates the functional activity of PrPC . Prion diseases , or transmissible spongiform encephalopathies , comprise a group of fatal neurodegenerative disorders in humans and animals for which there are no effective treatments or cures . These diseases are caused by refolding of the cellular prion protein ( PrPC ) into an infectious isoform ( PrPSc ) that catalytically templates its abnormal conformation onto additional molecules of PrPC ( Prusiner , 1998 ) . A similar , prion-like process may play a role in other neurodegenerative disorders , such as Alzheimer’s and Parkinson’s diseases and tauopathies , which are due to protein misfolding and aggregation ( Jucker and Walker , 2013 ) . There is evidence that PrPC , in addition to serving as a precursor to PrPSc , acts as a signal transducer that mediates the neurotoxic effects of PrPSc ( Biasini et al . , 2012; Brandner et al . , 1996; Chesebro et al . , 2005; Mallucci et al . , 2003 ) . Clues to possible mechanisms by which PrPC can initiate neurotoxic activity have emerged from studies of transgenic mice expressing PrP molecules that harbor certain internal deletions within the N-terminal domain . The PrPC molecule consists of a partially unstructured N-terminal domain ( residues 23–125 ) , and a globular , C-terminal domain ( residues 126–230 ) comprising three α-helices and two short , β-strands ( Zahn et al . , 2000 ) . Deletions spanning a 21-amino acid region ( amino acids 105–125 ) at the end of the flexible , N-terminal domain induce a spontaneous neurodegenerative phenotype with certain similarities to natural prion diseases , but without accumulation of PrPSc ( Baumann et al . , 2007; Li et al . , 2007; Shmerling et al . , 1998 ) . Importantly , these phenotypes are dose-dependently suppressed by co-expression of wild-type PrP , suggesting that the wild-type and deleted molecules interact with each other , or compete for binding to a common molecular target that mediates both physiological and pathological effects . The shortest deletion , Δ105–125 ( designated ΔCR , for central region ) , produces the most severe neurodegenerative phenotype , and requires the largest amount of wild-type PrP for rescue ( Li et al . , 2007 ) . In our efforts to understand why these deleted forms of PrP are so neurotoxic , we have discovered that they induce large , spontaneous ionic currents , recordable by patch clamping techniques , when expressed in a variety of cell lines ( Solomon et al . , 2010 , 2011 ) and in primary neurons ( Biasini et al . , 2013 ) . Remarkably , these currents are silenced by co-expression of wild-type PrP in the same cells , paralleling the rescuing effects of wild-type PrP in transgenic mice expressing deleted PrP . This observation suggests that the spontaneous ionic currents themselves , or some closely associated phenomenon , play a role in the neurodegenerative phenotype of these mice . In this study , we uncover novel mechanistic features of the toxicity-inducing activities of PrPC . We show that ligands binding N-terminal domain of PrPC abolish ΔCR PrP-induced currents , as do mutations of positively charged residues at the extreme N-terminus of this domain . Remarkably , expression of the isolated N-terminal domain in the absence of the C-terminal domain also induces spontaneous currents , indicating that the N-terminal domain is capable of acting as an autonomous , toxicity-determining effector . We also demonstrate that anti-PrP antibodies targeting epitopes in the structured , C-terminal domain induce ionic currents in cultured cells expressing wild-type PrPC , and cause degeneration of dendrites on hippocampal neurons . These results , taken together with structural evidence from heteronuclear NMR experiments , suggest a molecular model for PrPC in which the N-terminal domain acts as a neurotoxic effector whose activity is regulated by the C-terminal domain . We speculate that this inter-domain regulatory interaction could play a role in the physiological function of PrPC , and that disruption of this interaction could contribute to pathology in neurodegenerative disorders . Our results also have important implications for the safety of anti-PrP antibody therapies for prion and Alzheimer’s diseases . Our previous studies identified a positively charged , nine amino acid segment at the very beginning of the N-terminal domain ( residues 23–31 , KKRPKPGGW ) that is essential for the current activity of ΔCR PrP , and for the neurodegenerative phenotype of mice expressing another deletion mutant , Δ32–134 ( Solomon et al . , 2011; Westergard et al . , 2011b ) . As demonstrated in these earlier studies , deletion of residues 23–31 abolishes the spontaneous inward currents induced by ΔCR PrP ( Figure 1—figure supplement 1A ) , as does addition of pentosan polysulfate ( PPS ) , a negatively-charged glycosaminoglycan which binds to several regions within the N-terminal domain ( Figure 1—figure supplement 1B ) . To probe further the role of the N-terminal domain in ΔCR PrP-induced currents , we tested the effect of additional ligands ( antibodies and Cu2+ ions ) , as well alteration of positively charged residues within 23–31 region . We tested two different anti-prion antibodies: 100B3 ( Thuring et al . , 2005 ) , targeting residues 24–28 , and POM11 ( Polymenidou et al . , 2008 ) , which binds to the octapeptide repeats ( residues 51–90 ) . These antibodies ( at concentrations of 57 nM and 33 nM , respectively ) , dramatically reduced ΔCR PrP-induced currents ( Figure 1A , B ) . As a control for POM11 , we tested the effect of this antibody on ΔCR/Δ51–90 PrP , which retains current activity ( Figure 1—figure supplement 1 ) as shown previously ( Solomon et al . , 2011 ) , but which lacks the octapeptide repeat region and would not be expected to bind the antibody . As predicted , POM11 had relatively little effect on the currents induced by ΔCR/Δ51–90 PrP ( Figure 1B ) , while 100B3 and PPS were still inhibitory ( Figure 1A , Figure 1—figure supplement 1B ) . These results demonstrate that antibody ligands targeting the extreme N-terminus or the octapeptide repeats inhibit ΔCR PrP-induced currents . Interestingly , even though the octapeptide repeats are not required for current activity , binding of an antibody ligand to this region blocks the currents . 10 . 7554/eLife . 23473 . 003Figure 1 . Ligands binding to the N-terminal domain of PrPC block ΔCR-induced currents . ( A ) Left , representative traces of currents recorded from N2a cells expressing ΔCR or ΔCR/Δ51–90 PrP in the absence ( upper traces ) or presence ( lower traces ) of 100B3 ( 57 nM ) . Right , quantitation of the currents , plotted as the percentage of the total time the cells exhibited inward current ≥200 pA ( mean ± S . E . M . , n = 10 ) . ( B ) Left , representative traces of currents recorded from N2a cells expressing ΔCR or ΔCR/Δ51–90 PrP in the absence ( upper traces ) or presence ( lower traces ) of POM11 ( 33 . 3 nM ) . Right , quantitation of the currents ( mean ± S . E . M . , n = 10 ) . ( C ) Left , representative traces of currents recorded from N2a cells expressing ΔCR or ΔCR/Δ51–90 PrP in the absence ( upper traces ) or presence ( lower traces ) of Cu-pentaglycine ( 100 μM ) . Right , quantitation of the currents ( mean ± S . E . M . , n = 10 ) . Scale bars in all panels: 1 nA , 30 s . *p<0 . 05; **p<0 . 01; ***p<0 . 005 . DOI: http://dx . doi . org/10 . 7554/eLife . 23473 . 00310 . 7554/eLife . 23473 . 004Figure 1—source data 1 . Quantification of ΔCR PrP-induced currents w/o treatment of ligands binding to PrPC N-terminus . DOI: http://dx . doi . org/10 . 7554/eLife . 23473 . 00410 . 7554/eLife . 23473 . 005Figure 1—figure supplement 1 . Mutated forms of PrP induce spontaneous currents . ( A ) Left , whole cell recordings were made from N2a cells transfected with empty vector , or vector encoding WT , ΔCR , ΔCR/Δ51–90 , ΔCR/Δ23–31 , or ΔCR/E3D PrP . Right , quantitation of the currents , plotted as the percentage of the total time the cells exhibited inward current ≥200 pA ( mean ± S . E . M . , n = 10 ) . ( B ) Left , representative traces of currents recorded from N2a cells expressing ΔCR or ΔCR/Δ51–90 PrP in the absence ( upper traces ) or presence ( lower traces ) of PPS ( 100 μg/ml ) . Right , quantitation of the currents ( mean ± S . E . M . , n = 10 ) . *p<0 . 05 , **p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 23473 . 00510 . 7554/eLife . 23473 . 006Figure 1—figure supplement 2 . Surface immunofluorescence staining of PrPC on N2a cells expressing ΔCR after treatment with N-terminal ligands . Living N2a cells co-transfected with a ΔCR PrP-encoding plasmid along with an EGFP marker plasmid were treated with the indicated ligands , and then stained with anti-PrP antibody D18 , followed by fixation and incubation with red fluorescent secondary antibody . Representative fluorescence images show surface expression of PrP ( red ) , EGFP fluorescence ( green ) , and DAPI nuclear staining ( blue ) . Scale bar: 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 23473 . 006 To further investigate the role of the octapeptide repeats , we tested the effect of Cu2+ ions , which are known to bind to histidine residues located within four of the five octapeptide repeats and at positions 95 and 110 ( Millhauser , 2007 ) . We found that treatment with Cu2+-pentaglycine ( 100 μM ) blocked spontaneous currents in N2a cells expressing ΔCR PrP , but not in cells expressing ΔCR/Δ51–90 ( Figure 1C ) . Pentaglycine , with a Kd for Cu2+ similar to that of PrPC ( 40 nM ) , was included as a Cu2+ chelator in order to minimize the concentration of free Cu2+ ( 1 μM ) while still allowing PrPC to compete for Cu2+ ions . This result indicates that Cu2+ coordination to the octapeptide repeats blocks ΔCR PrP currents . Both PPS and copper have been shown to induce endocytosis of PrPC from cell surface ( Brown and Harris , 2003; Pauly and Harris , 1998; Shyng et al . , 1995a ) . To determine whether ligand binding to the N-terminus blocked ΔCR PrP-induced currents by reducing the amount of the mutant protein on the cell surface , living cells were treated with PPS for 1 hr or Cu2+-pentaglycine for 5 min in recording buffer at room temperature , and then surface-stained for ΔCR PrP with anti-PrP antibody . Under these conditions , neither ligand altered the amount of ΔCR PrP on the cell surface ( Figure 1—figure supplement 2 ) , presumably because of the short period of treatment and the fact that the experiment was conducted at room temperature rather than 37°C . As reported previously ( Brown and Harris , 2003; Pauly and Harris , 1998; Shyng et al . , 1995a ) , PPS did induce significant endocytosis after treatment at 37°C for 48 hr ( Figure 1—figure supplement 2 ) . Cells treated with 100B3 or POM11 for 48 hr at 37°C did not show any change in surface level of ΔCR PrP ( Figure 1—figure supplement 2 ) . Thus , the inhibitory effects of PPS , POM11 , 100B3 and Cu2+ did not result from acute changes in localization or trafficking of the mutant protein . The extreme N-terminus of PrPC contains four positively charged amino acids ( 23KKRPKPGGW31 ) . To test the role of these residues in ΔCR-induced currents , we generated a variant of ΔCR ( designated ΔCR/E3D ) in which the three lysine residues ( at positions 23 , 24 , and 27 ) were mutated to glutamic acid and the single arginine residue ( at position 25 ) was mutated to aspartic acid . We found that N2a cells expressing this mutant did not show any currents ( Figure 1—figure supplement 1 ) . This result indicates that the positive charges within the 23–31 segment are essential for ΔCR-induced spontaneous currents . Having shown that the N-terminal domain of PrPC is essential for ΔCR PrP-induced currents , we wished to determine whether N-terminus is , by itself , sufficient to induce spontaneous currents . We constructed a series of chimeric proteins ( collectively designated PrP ( N ) -EGFP-GPI ) consisting of various lengths of the N-terminal domain of PrPC ( residues 23–109 ) fused to an EGFP molecule that was equipped with the GPI addition signal from PrPC ( Figure 2A ) . It was necessary to include the EGFP moiety to enable efficient delivery of the protein to the cell surface ( Heske et al . , 2004 ) ; fusing the N-terminal domain directly to the GPI addition signal results in a protein that is largely retained in the ER and is degraded by the proteasome ( Dametto et al . , 2015 ) . We confirmed cell surface localization of the PrP ( N ) -EGFP-GPI constructs in transfected cells by fluorescence microscopy ( Figure 2B ) . 10 . 7554/eLife . 23473 . 007Figure 2 . The N-terminal domain of PrPC induces ionic currents in the absence of the C-terminal domain . ( A ) Schematic of PrP ( N ) -EGFP-GPI constructs containing the N-terminus of PrP fused to EGFP and the PrP GPI attachment sequence . The colored blocks represent the signal sequence ( blue ) , polybasic residues 23–31 ( yellow ) , different portions of the N-terminus ( grey ) , EGFP ( green ) , and the GPI attachment sequence ( magenta ) . ( B ) Fluorescence image of N2a cells expressing PrP ( 1-109 ) -EGFP-GPI , showing localization of the protein on the cell surface . Scale bar = 10 μm . ( C ) Left , representative traces of currents recorded from N2a cells expressing constructs with different lengths of the N-terminus ( 1–31 , 1–58 , 1–90 , 1–109 and 32–109 ) . Scale bars: 500 pA , 30 s . Right , quantitation of the currents , plotted as the percentage of the total time the cells exhibited inward current ≥200 pA ( mean ± S . E . M . , n = 5 cells ) . *p<0 . 05; **p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 23473 . 00710 . 7554/eLife . 23473 . 008Figure 2—source data 1 . Quantification of N1-GFP-GPI-induced currents N2a cells . DOI: http://dx . doi . org/10 . 7554/eLife . 23473 . 00810 . 7554/eLife . 23473 . 009Figure 2—figure supplement 1 . Currents induced by PrP ( 1-109 ) -EGFP-GPI have the same features as currents induced by ΔCR PrP . ( A ) Representative traces of currents recorded from N2a cells expressing PrP ( 1-109 ) -EGFP-GPI or Δ23–89 PrP . ( B ) Left , representative traces of currents recorded from N2a cells expressing PrP ( 1-109 ) -EGFP-GPI in the absence of treatment , or in the presence of PPS ( 100 μg/ml ) , POM11 ( 16 . 7 nM ) or Cu-pentaglycine ( 100 μM ) . Right , quantitation of the currents , plotted as the percentage of the total time the cells exhibited inward current ≥200 pA . ( mean ± S . E . M . , n = 10 ) . ( C ) Left , representative traces of currents recorded from N2a cells transfected with a plasmid encoding ΔCR PrP , or co-transfected with plasmids encoding ΔCR PrP and WT PrP at the indicated ratios . Right , quantitation of the currents ( mean ± S . E . M . , n = 10 ) . ( D ) Left , representative traces of currents recorded from N2a cells transfected with a plasmid encoding PrP ( 1-109 ) -EGFP-GPI , or co-transfected with plasmids encoding PrP ( 1-109 ) -EGFP-GPI and WT PrP at the indicated ratios . Right , quantitation of the currents ( mean ± S . E . M . , n = 10 ) . ( E ) Quantitation of the currents recorded from N2a cells transfected with a plasmid encoding ΔCR PrP , or co-transfected with plasmids encoding ΔCR PrP and Δ23–31 PrP at the indicated ratios . ( mean ± S . E . M . , n = 10 ) . ( F ) Quantitation of the currents recorded from N2a cells transfected with a plasmid encoding PrP ( 1-109 ) -EGFP-GPI , or co-transfected with plasmids encoding PrP ( 1-109 ) -EGFP-GPI and Δ23–31 PrP at the indicated ratios . ( mean ± S . E . M . , n = 10 ) . ( G ) Left , representative traces of currents recorded from N2a cells expressing ΔCR PrP at holding potentials of −30 mV , −50 mV and −70 mV . Right , quantitation of the current at each holding potential ( mean ± S . E . M . , n = 10 ) . ( H ) Left , representative traces of currents recorded from N2a cells expressing PrP ( 1-109 ) -EGFP-GPI at holding potentials of −30 mV , −50 mV and −70 mV . Right , quantitation of the current at each holding potential ( mean ± S . E . M . , n = 10 ) . Except for panels G and H , the holding potential for all experiments was −70 mV . Scale bar in all panels: 1 nA , 30 s . *p<0 . 05 , **p<0 . 01 , ***p<0 . 005 . DOI: http://dx . doi . org/10 . 7554/eLife . 23473 . 009 We found that cells expressing PrP ( N ) -EGFP-GPI constructs displayed spontaneous ionic currents . The most active currents were observed with a construct encompassing PrP residues 1–109 , with successively lower current activities seen as the constructs became shorter ( Figure 2C , D ) . Importantly , the PrP 32–109 construct was much less active , demonstrating the dependence of current activity on the 23–31 region ( Figure 2C , D ) . Expression of the C-terminal domain of PrPC ( Δ23–89 PrP ) did not induce any currents ( Figure 2—figure supplement 1A ) . We note that ΔCR/Δ51–90 PrP ( Figure 1 and Figure 1—figure supplement 1 ) was more effective at inducing currents than 1–31-EGFP or 1–59-EGFP ( Figure 2 ) . There are at least two possible reasons for this observation . First , the ΔCR/Δ51–90 PrP construct contains additional sequences that are not present in the 1–31-EGFP or 1–59-EGFP constructs . In particular , ΔCR/Δ51–90 PrP contains residues 91–104 , which are absent in 1–31-EGFP and 1–59-EGFP . These additional residues may enhance production of spontaneous currents , consistent with the general observation that PrP-EGFP chimeras incorporating longer stretches of the PrP N-terminus produced more currents ( Figure 2C ) . A second possible explanation is that the EGFP portion of the chimeric constructs may position the PrP N-terminus at a different distance from the membrane , or in a different orientation , than the natural PrPC C-terminus , and this may diminish the ability of the N-terminus to interact with the membrane to produce currents . The spontaneous currents associated with PrP ( N ) -EGFP-GPI had characteristics identical to the currents associated with ΔCR PrP . First , PrP ( N ) -EGFP-GPI currents were sporadic in nature , and were silenced by N-terminal ligands , including PPS , 100B3 , POM11 , and Cu2+-pentaglycine ( Figure 2—figure supplement 1B ) . Second , PrP ( N ) -EGFP-GPI currents , like ΔCR PrP currents , were silenced in a dose-dependent fashion by co-expression of WT PrP ( Figure 2—figure supplement 1C and D ) . Moreover , removal of 23–31 region abolished the ability of WT PrPC to suppress both ΔCR and PrP ( N ) -EGFP-GPI currents ( Figure 2—figure supplement 1E and F ) , which is consistent with the observation that expression of Δ23–31 PrP does not rescue the neurodegenerative phenotype of mice expressing Δ32–134 PrP ( Turnbaugh et al . , 2011 ) . Finally , the currents induced by both ΔCR and PrP ( N ) -EGFP-GPI were voltage-dependent , only being observed at holding potentials below −30 mV ( Figure 2—figure supplement 1G and H ) . The results presented thus far suggest that the C-terminal domain of PrPC may directly regulate the N-terminal domain through a cis-interaction between the two domains . This interaction may be disrupted by deletion of residues in the central region ( as in ΔCR ) , or by substitution of an unrelated protein for the C-terminal domain ( as in PrP ( N ) -EGFP-GPI ) . Recently , Evans et al . ( 2016 ) used 1H-15N HSQC NMR to demonstrate that Cu2+ ions , when bound to the N-terminal , octapeptide repeats , promote contact between these repeats and a C-terminal surface site encompassing helices 2 and 3 . We predicted that this cis interaction would be weakened in ΔCR , thereby accounting for the ability of the liberated N-terminal domain to induce spontaneous currents . To test this prediction , we employed paramagnetic relaxation enhancement to probe the interaction between octapeptide-bound Cu2+ ions and residues in the C-terminal domain . Using the methods of Evans et al . ( 2016 ) , we compared the 1H-15N HSQC NMR spectra of recombinant , wild-type mouse PrP ( WT PrP ) and ΔCR PrP in the presence and absence of one equivalent of Cu2+ ( Figure 3—figure supplement 1 ) . Residues that come in close proximity to Cu2+ have their NMR cross-peaks broadened , resulting in a lower observed intensity . For example , in the absence of Cu2+ , a large cross-peak shown in black corresponding to WT PrP residue E199 was observed ( Figure 3—figure supplement 1–A1 ) . However , upon the addition of one equivalent of Cu2+ the corresponding cross-peak depicted in red for residue E199 was greatly diminished in intensity . On the other hand , when Cu2+ was titrated into ΔCR PrP , only a small shift between positions of the cross-peaks corresponding to E199 was observed ( Figure 3—figure supplement 1–A2 ) . The chemical shift changes , mapped onto the structure of WTPrP in Figure 3A , B , identify those residues in the C-terminal domain affected by Cu2+ binding to the octapeptide repeats . In Figure 3A1 and A2 , residues that do not significantly change upon addition of Cu2+ are indicated by blue bars , while residues that underwent a significant reduction in intensity are indicated by red bars . Cross-peaks that were not identified are not shown in the figure . The large interaction patch seen in WT PrP is clearly reduced in the ΔCR mutant ( compare the residues highlighted in magenta in Figure 3B1/C1 and B2/C2 ) , especially in helices 2 and 3 . These data suggest that deletion of residues in the central region disrupts a Cu2+-driven regulatory interaction between the N- and C-terminal domains . 10 . 7554/eLife . 23473 . 010Figure 3 . ΔCR PrP shows diminished interaction between N- and C-terminal domains based on NMR analysis . ( A ) Reduction in peak intensities of 1H-15N HSQC spectra of WT PrP ( A1 ) and ΔCR PrP ( A2 ) in the presence of Cu2+ . Data are shown only for the structured domain ( residues 90–230 ) . I/I0 values represent the ratios of peak heights in the presence and absence of 1 equivalent of Cu 2+ . Residues with I/I0 values less than 1 . 0 SD below the mean ( dotted line ) are shown in red , with unassigned residues omitted . ( B ) Residues ( labeled and colored magenta ) of WT PrP ( B1 ) and ΔCR PrP ( B2 ) with I/I0values < 1 . 0 SD below the mean , mapped onto a ribbon representation of the NMR structure of mouse PrP ( 120-230 ) ( PDB:1XYX ) . ( C ) Affected residues ( magenta ) of WT PrP ( C1 ) and ΔCR PrP ( C2 ) are mapped onto surface plots of mouse PrP ( 120-230 ) ( PDB:1XYX ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23473 . 01010 . 7554/eLife . 23473 . 011Figure 3—figure supplement 1 . NMR signals for C-terminal residues broaden in the presence of Cu2+ bound to the N-terminal octarepeats . Selected regions of the 1H-15N HSQC paramagnetic relaxation enhancement NMR spectra of WT PrP ( A1 ) and ΔCR PrP ( A2 ) in the absence of metal ( black ) and in the presence of 1 equivalent of Cu2+ ( red ) . Spectra were acquired on a Varian INOVA 600-MHz spectrometer at pH 5 . 5 , 300 μM protein , and 37°C . DOI: http://dx . doi . org/10 . 7554/eLife . 23473 . 011 Previous studies have reported that antibodies targeting specific epitopes in the structured domain of PrPC cause neuronal death when administered in vivo or in brain slices ( Reimann et al . , 2016; Solforosi et al . , 2004; Sonati et al . , 2013 ) . We wondered whether the neurotoxicity of anti-PrP antibodies might be due to their ability to induce ionic currents , similar to the way that ΔCR PrP causes ionic currents in cultured cells ( Solomon et al . , 2010 , 2011 ) and neuronal death in transgenic mice ( Li et al . , 2007 ) . We found that two antibodies targeting overlapping epitopes encompassing helix 1 in the C-terminal half of PrPC , POM1 ( Polymenidou et al . , 2008; Sonati et al . , 2013 ) and D18 ( Doolan and Colby , 2015; Williamson et al . , 1998 ) , induced spontaneous currents in N2a cells expressing WT PrPC ( Figure 4A ) . 10 . 7554/eLife . 23473 . 012Figure 4 . Antibodies against the C-terminal domain induce ionic currents in N2a cells expressing wild-type PrPC . ( A ) Left , representative traces of spontaneous currents recorded from cells expressing ΔCR PrP ( top traces ) , and currents induced by anti-prion antibodies ( POM1 and D18 ) in cells expressing WT PrP ( lower traces ) in the absence ( left-hand traces ) or presence ( right-hand traces ) of PPS ( 100 ug/ml ) . Right , quantitative analysis of the currents , plotted as the percentage of the recording time the cells exhibited inward currents ≥200 pA . ( mean ± S . E . M . , n = 10 ) . ( B ) Left , representative traces of currents recorded from cells expressing WT PrP in the presence of POM1 , POM11 , or POM1+POM11 . Right , quantitative analysis of the currents ( mean ± S . E . M . , n = 10 ) . ( C ) Left , representative traces of currents recorded from cells expressing WT PrP in the presence of D18 , POM11 , or D18+POM11 . Right , quantitative analysis of the currents ( mean ± S . E . M . , n = 10 ) . ( D ) Left , representative traces of currents induced by POM1 ( upper traces ) or D18 ( lower traces ) in cells expressing WT PrP ( left-hand traces ) or Δ23–31 PrP ( center traces ) , or in cells expressing WT PrP after pretreatment with PIPLC ( 1 . 0 units/ml for 4 hr at 37°C ) ( right-hand traces ) . Right , quantitative analysis of the currents ( mean ± S . E . M . , n = 10 ) . Scale bars in all panels: 1 nA , 30 s . **p<0 . 01; ***p<0 . 005 . DOI: http://dx . doi . org/10 . 7554/eLife . 23473 . 01210 . 7554/eLife . 23473 . 013Figure 4—source data 1 . Quantification of anti-prion antibody-induced currents on N2a cells . DOI: http://dx . doi . org/10 . 7554/eLife . 23473 . 01310 . 7554/eLife . 23473 . 014Figure 4—figure supplement 1 . ICSM-18 induces currents in N2a cells . ( A ) Left , representative traces of currents induced in N2a cells over-expressing WT PrP by non-specific IgG ( 33 . 3 nM ) ( upper trace ) , ICSM-18 ( 33 . 3 nM ) ( middle trace ) or ICSM-18 ( 33 . 3 nM ) in the presence of PPS ( 100 μg/ml ) ( lower trace ) . Right , quantitative analysis of the currents , plotted as the percentage of the recording time the cells exhibited inward currents ≥200 pA . ( mean ± S . E . M . , n = 5 ) . **p<0 . 01 . ( B ) Left , representative traces of currents induced in N2a cells over-expressing WT PrP by control scFv antibody ( anti-fluorescein , 200 nM ) ( upper trace ) , ICSM-18 scFv ( 200 nM ) ( middle trace ) , or ICSM-18 scFv ( 200 nM ) in the presence of PPS ( 100 μg/ml ) ( lower trace ) . Right , quantitative analysis of the currents , plotted as the percentage of the recording time the cells exhibited inward currents ≥200 pA . ( mean ± S . E . M . , n = 20 ) . **p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 23473 . 01410 . 7554/eLife . 23473 . 015Figure 4—figure supplement 2 . Antibody-induced currents are dependent on PrPC expression level and are produced by Fab fragments . ( A ) Quantitation of currents produced by D18 or POM1 treatment of transfected N2a cells over-expressing WT PrP , untransfected N2a cells , or PrP-null N2a cells . Currents are plotted as the percentage of the recording time the cells exhibited inward currents ≥200 pA . ( mean ± S . E . M . , n = 10 ) . *p<0 . 05 , **p<0 . 01 . ( B ) Left , representative traces of currents recorded from PrP-over-expressing N2a cells in the presence of D18 or an Fab fragment of D18 . Right , quantitative analysis of the currents ( mean ± S . E . M . , n = 10 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23473 . 01510 . 7554/eLife . 23473 . 016Figure 4—figure supplement 3 . 6D11 , but not other central region antibodies , weakly induces currents in N2a cells . Left , representative traces of currents induced by antibodies 6D11 , POM3 , D13 , and ICSM-35 at the indicated concentrations in N2a cells over-expressing WT PrP . Scale bar: 1 nA , 30 s . Right , quantitation of the currents , plotted as the percentage of the recording time the cells exhibited inward currents ≥200 pA . ( mean ± S . E . M . , n = 10 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23473 . 01610 . 7554/eLife . 23473 . 017Figure 4—figure supplement 4 . POM1 , but not antibodies recognizing other regions of the C-terminal domain , induces currents . Left , representative traces of currents induced by antibodies POM1 , POM4 and POM6 at the indicated concentrations in N2a cells over-expressing WT PrP . Scale bar: 1 nA , 30 s . Right , quantitation of the currents , plotted as the percentage of the recording time the cells exhibited inward currents ≥200 pA ( mean ± S . E . M . , n = 10 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23473 . 01710 . 7554/eLife . 23473 . 018Figure 4—figure supplement 5 . D18 induces currents in Tga20 hippocampal neurons over-expressing WT PrP and in wild-type neurons , but not in PrP knock-out ( KO ) neurons . Top , D18 induces currents in Tga20 hippocampal neurons in the absence ( left ) but not the presence ( right ) of PPS ( 100 μg/ml ) . Scale bar: 1 nA , 30 s . Bottom , D18 induces weak currents in wild-type hippocampal neurons ( left ) , but not in PrP KO neurons ( right ) . Scale bar: 1 nA , 30 s . Pie charts show the proportions of three categories of neurons: Light grey represents neurons from which recordings were made for >5 min without any currents; dark grey represents neurons that exhibited spontaneous currents , and in which voltage-clamped could be maintained throughout the 5 min recording period; black represents neurons in which voltage-clamp was lost during the 5-min recording period . DOI: http://dx . doi . org/10 . 7554/eLife . 23473 . 018 A third anti-prion antibody with a similar epitope , ICSM-18 ( Antonyuk et al . , 2009 ) , had a comparable effect , which was blocked by PPS and was absent with non-specific mouse IgG ( Figure 4—figure supplement 1A ) . Because we were not able to obtain sufficient amounts of ICSM-18 for further electrophysiological experiments , we turned to a single chain version of this antibody ( ICSM-18 scFv ) . Like the holo-antibody , ICSM-18 scFv-induced spontaneous currents on on N2a cells overexpressing WT PrPC ( Figure 4—figure supplement 1B ) , although higher concentrations were required ( 200 nM for the scFv version , compared to 33 . 3 nM for the holo-antibody ) , presumably reflecting the lower avidity of the monovalent scFv antibodies ( Mammen , 1998 ) . The spontaneous currents induced by ICSM-18 scFv were abolished by PPS ( 100 μg/ml ) , and were absent with a control scFv ( Figure 4—figure supplement 1B ) . The properties of the D18- , POM1- , and ICSM-18-induced currents were identical to those of the spontaneous currents associated with ΔCR PrP , in terms of their sporadic nature , and suppression by PPS ( Figure 4A , Figure 4—figure supplement 1 ) . In addition , treatment of WT PrP-expressing N2a cells with POM11 blocked POM1- or D18-induced currents ( Figure 4B and C ) , similar to the inhibitory effect of POM11 on ΔCR PrP currents ( Figure 1 ) . Finally , cells expressing PrP with a deletion of residues 23–31 were resistant to current induction by POM1 and D18 antibodies ( Figure 4D ) , parallel to the situation with ΔCR PrP , and emphasizing the importance of the polybasic region in antibody-induced current generation . One would expect a significant reduction in antibody-induced currents in N2a cells expressing Δ23–31 PrP compared to cells over-expressing WT PrP ( as observed in Figure 4D ) , since the endogenous level of WT PrP in N2a cells is low , and we have shown that untransfected cells display reduced currents after antibody treatment ( Figure 4—figure supplement 2A ) . The fact that the Δ23–31-expressing cells do not display even low levels of current due to endogenous PrPC may be due to some suppression of endogenous PrP expression , or perhaps competition by the deleted protein for binding of the antibodies . We performed several control experiments to demonstrate that the antibody-induced currents were dependent on expression of cell-surface PrPC . First , we pretreated transfected N2a cells expressing WT PrP with phosphatidylinositol-specific phospholipase C ( PIPLC ) at 1 U/ml for 4 hr at 37°C , which cleaves the C-terminal GPI anchor and releases PrPC from the cell membrane . Neither D18 nor POM1 induced currents in PIPLC pretreated cells ( Figure 4D ) . Second , we demonstrated that D18 did not induce currents in N2a cells in which PrP gene expression had been abolished by CRISPR-Cas technology ( Mehrabian et al . , 2014 ) ( Figure 4—figure supplement 2A ) . Finally , we observed that D18 induced currents in untransfected N2a cells , although they were smaller than in transfected N2a cells over-expressing WT PrPC , presumably due to the lower expression level of PrPC in the former cells ( Figure 4—figure supplement 2A ) . Taken together , these results indicate that D18 and POM1-induced currents are PrP-dependent and are due to binding of the antibodies to cell-surface PrPC . To determine whether antibody-induced currents were the result of cross-linking of cell-surface PrPC , or Fc-mediated antibody effector functions , we tested the effect of monovalent Fab fragments . We found that Fab fragments prepared from D18-induced currents on N2a cells ( Figure 4—figure supplement 2B ) . This result , along with the effectiveness of ICSM-18 scFv antibody ( Figure 4—figure supplement 1B ) , indicates that the ability of the C-terminally directed antibodies to induce currents in PrP-expressing cells is not due to cross-linking of cell-surface PrP by bivalent binding , or to Fc-mediated effector functions such as complement fixation . The antibody 6D11 ( Pankiewicz et al . , 2006 ) , whose epitope ( residues 93–109 ) encompasses several positively charged residues following the octapeptide repeats , also induced currents in WT PrP-expressing N2a cells ( Figure 4—figure supplement 3 ) . However , this antibody was less potent than POM1 and D18 , since the currents were much smaller , and higher concentrations were required to produce consistent effects ( 66 . 7 nM for 6D11 , compared to 33 . 3 nM for POM1 and 16 . 7 nM for D18 ) ( Figure 4—figure supplement 3 and Figure 4 ) . 6D11 and three other antibodies targeting this region , D13 ( Williamson et al . , 1998 ) ( epitope: a . a . 95–105 ) , POM3 ( Polymenidou et al . , 2008 ) ( epitope: a . a . 95–100 ) and ICSM-35 ( Khalili-Shirazi et al . , 2007 ) ( epitope: a . a . 93–105 ) , did not induce currents at a concentration of 33 . 3 nM ( Figure 4—figure supplement 3 ) . Taken together , these results indicate that antibodies targeting a positively charged segment at the terminus of the flexible domain are less effective at inducing currents than antibodies binding to helix 1 in the C-terminal , structured domain . To explore further the epitope specificity of the antibody-induced currents , we tested two additional POM antibodies recognizing other epitopes in the C-terminal domain of PrPC . We found that POM4 ( whose epitope encompasses helix 3 and β1 ) and POM6 ( which recognizes helices 1 and 2 ) did not produce detectable currents when applied to WT PrP-expressing N2a cells at a concentration of 33 . 3 nM ( Figure 4—figure supplement 4 ) . We tested whether anti-PrP antibodies induce currents in hippocampal neurons , as well as in N2a cells . Treatment with D18 ( 33 . 3 nM ) induced large , spontaneous inward in hippocampal neurons cultured from Tga20 mice over-expressing wild-type PrP , and also increased the fragility of these neurons ( Figure 4—figure supplement 5 ) . Eight of 10 Tga20 neurons analyzed in the presence of D18 were lost to observation shortly after breaking the patch , usually after recording an initial inward current that did not return to baseline . This phenomenon is similar to what we previously observed in cultured neurons expressing ΔCR PrP ( Biasini et al . , 2013 ) , and may reflect a detrimental effect of the induced currents on the integrity of the neuronal membrane . In contrast , recordings from Tga20 neurons in the presence of D18 and PPS remained stable for 5 min without current activity ( Figure 4—figure supplement 5 ) . Seven out of 15 wild-type neurons analyzed in the presence of D18 exhibited spontaneous currents , with no lost cells . D18 did not induce any currents in neurons from Prnp−/− mice , which lack PrP expression . Given the correlation between ΔCR PrP-induced currents and neurotoxicity , we hypothesized that anti-PrP antibodies might have neurotoxic effects on cultured hippocampal neurons as a result of the currents induced by these antibodies . To test this idea , we treated hippocampal neurons cultured from Tga20 mice with D18 ( 16 . 7 nM ) or POM1 ( 33 . 3 nM ) for 48 hr , and then stained them with an antibody to MAP2 to visualize changes in dendritic morphology . We observed that treatment with D18 or POM1 caused dendrites to assume a characteristic ‘beaded’ appearance , which is typical of several kinds of toxic insults , including hypoxia and glutamate-induced excitotoxicity ( Hasbani et al . , 2001 ) ( Figure 5A ) . This effect was less in wild-type neurons and completely absent in Prnp−/− hippocampal neurons , indicating a dependence on the expression level of PrPC ( Figure 5A ) . Treatment of neurons with non-specific mouse IgG had no effect on dendritic morphology ( Figure 5A ) . Antibody 6D11 , which weakly induced currents in WT PrP-over-expressing N2a cells at 66 . 7 nM ( Figure 4—figure supplement 3 ) induced mild dendrite degeneration in Tga20 neurons ( Figure 5—figure supplement 1A ) , while ICSM-35 at 33 . 3 nM , which did not induce currents , did not cause dendritic degeneration ( Figure 5—figure supplement 1B ) . Similarly , POM4 ( 33 . 3 nM ) and POM6 ( 33 . 3 nM ) had no effect on dendritic morphology ( Figure 5—figure supplement 2 ) . 10 . 7554/eLife . 23473 . 019Figure 5 . Antibodies recognizing the C-terminal domain of PrPC induce dendritic degeneration in hippocampal neurons . ( A ) Top , representative images showing dendrite morphology of cultured hippocampal neurons from Tga20 mice ( which over-express WT PrPC ) , WT mice , or Prnp−/− mice after treatment for 48 hr with D18 ( 16 . 7 nM ) , POM1 ( 33 . 3 nM ) or non-specific IgG ( 33 . 3 nM ) . The cells were stained with an antibody to MAP2 to visualize dendrites . Boxed areas are enlarged below each image . Scale bar = 10 µm . Bottom , quantitation of dendritic degeneration , expressed as the length of beaded dendrite segments as a percentage of total dendrite length , from 10 images in three independent cultures for each experimental condition . Data represent mean ± S . E . M . **p<0 . 01; ***p<0 . 005 . ( B–D ) Quantitation of dendritic beading following treatment with IgG , D18 alone , N-terminal ligand ( PPS , 100B3 , or POM11 ) alone , or D18 together with the N-terminal ligand . Data represent mean ± S . E . M . ***p<0 . 005 . DOI: http://dx . doi . org/10 . 7554/eLife . 23473 . 01910 . 7554/eLife . 23473 . 020Figure 5—source data 1 . Quantification of dendritic degeneration , expressed as the length of beaded dendrite segments as a percentage of total dendrite length . DOI: http://dx . doi . org/10 . 7554/eLife . 23473 . 02010 . 7554/eLife . 23473 . 021Figure 5—figure supplement 1 . 6D11 , but not ICSM35 , has a weak effect on the dendritic morphology of Tga20 hippocampal neurons . ( A ) Representative images showing dendrite morphology after treatment with 6D11 or non-specific IgG ( 66 . 7 nM each ) . ( B ) Representative images showing dendrite morphology after treatment with ICSM-35 or non-specific IgG ( 33 . 3 nM each ) . Neurons were stained for an antibody to MAP2 to visualize dendrites . Scale bar: 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 23473 . 02110 . 7554/eLife . 23473 . 022Figure 5—figure supplement 2 . POM4 and POM6 have no effect on the dendritic morphology of Tga20 hippocampal neurons . Representative images showing dendrite morphology of cultured hippocampal neurons from Tga20 mice ( upper panels ) , WT mice ( middle panels ) , or Prnp−/− mice ( lower panels ) after treatment for 48 hr with POM1 ( 33 . 3 nM ) ( left-hand panels ) , POM4 ( 33 . 3 nM ) ( center panels ) , or POM6 ( 33 . 3 nM ) ( right-hand panels ) . Scale bar: 10 μM . DOI: http://dx . doi . org/10 . 7554/eLife . 23473 . 022 We tested whether antibody-induced dendritic changes , like currents , were dependent on the N-terminal domain of PrPC . Supporting such a correlation , D18 had no effect on dendritic morphology of hippocampal neurons cultured from mice expressing Δ23–31 or Δ23–111 PrP ( Figure 6 ) , demonstrating that the N-terminal domain is essential for both the dendrotoxic and current-inducing effects of the antibody . In addition , we found that co-treatment of neurons with the N-terminal ligands PPS , 100B3 , or POM11 abolished D18-induced dendritic degeneration ( Figure 5B–D ) , analogous to the way these ligands inhibit D18-induced currents ( Figure 4A–C ) . 10 . 7554/eLife . 23473 . 023Figure 6 . D18 does not induce dendritic degeneration in hippocampal neurons cultured from Δ23–31 or Δ23–111 transgenic mice on the Prnp−/− background . Representative images showing dendrite morphology of cultured hippocampal neurons from mice expressing Δ23–31 PrP ( upper panels ) or Δ23–111 PrP ( lower panels ) after treatment for 48 hr with D18 ( 16 . 7 nM ) ( left-hand panels ) or non-specific IgG ( 16 . 7 nM ) ( right-hand panels ) . Neurons were stained for an antibody to MAP2 to visualize dendrites . Scale bar: 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 23473 . 023 Given the proposed use of a humanized version of ICSM-18 as an immunotherapeutic for prion and Alzheimer’s diseases in patients ( Klyubin et al . , 2014 ) , we tested whether ICSM-18 induces dendritic toxicity . We observed that treatment of neurons cultured from Tga20 mice with ICSM-18 ( 6 . 67 nM ) for 48 hr caused significant beading of dendrites , similar to the effects of POM1 and D18 ( Figure 7A ) . At higher concentrations ( 33 . 3 nM ) , ICSM-18 caused significant loss of neuronal cell bodies after 48 hr of treatment ( not shown ) . In control experiments , no morphological effects were observed with non-specific IgG ( 6 . 67 nM ) , or after treatment of neurons from Prnp−/− mice with ICSM-18 ( Figure 7A ) . ICSM-18 scFv ( 200 nM ) also induced dendritic degeneration on neurons cultured from Tga20 mice but not Prnp−/− mice , although the effect was milder than for the ICSM-18 holo-antibody ( Figure 7B ) . 10 . 7554/eLife . 23473 . 024Figure 7 . Anti-prion antibody ICSM-18 induces dendritic degeneration in hippocampal neurons . ( A ) Left , representative images showing dendrite morphology of cultured hippocampal neurons from Tga20 mice ( which over-express WT PrPC ) ( left-hand panels ) or Prnp−/− mice ( right-hand panels ) after treatment for 48 hr with non-specific IgG ( 6 . 67 nM ) ( upper panels ) or ICSM-18 ( 6 . 67 nM ) ( lower panels ) . The cells were stained with an antibody to MAP2 to visualize dendrites . Boxed areas are enlarged below each image . Scale bar = 10 µm . Right , quantitation of dendritic degeneration , expressed as the length of beaded dendrite segments as a percentage of total dendrite length , from 10 images in three independent cultures for each experimental condition . Data represent mean ± S . E . M . ***p<0 . 005 . ( B ) Left , representative images showing dendrite morphology of cultured hippocampal neurons from Tga20 mice ( left-hand panels ) or Prnp−/− mice ( right-hand panels ) after treatment for 48 hr with control scFv ( anti-fluorescein , 200 nM ) ( upper panels ) or ICSM-18 scFv ( 200 nM ) ( lower panels ) . The cells were stained with an antibody to MAP2 to visualize dendrites . Boxed areas are enlarged below each image . Scale bar = 10 µm . Right , quantitation of dendritic degeneration , expressed as the length of beaded dendrite segments as a percentage of total dendrite length , from 10 images in three independent cultures for each experimental condition . Data represent mean ± S . E . M . ***p<0 . 005 . DOI: http://dx . doi . org/10 . 7554/eLife . 23473 . 024 Our previous studies identified a nine amino-acid polybasic region at the N-terminus of PrPC ( residues 23–31 ) that is essential for several toxic activities , including the spontaneous current activity associated with the ΔCR PrP deletion mutant , the antibiotic hypersensitivity of cells expressing ΔCR PrP , and the neurodegenerative phenotype of transgenic mice expressing the deletion mutant Δ34–121 ( Solomon et al . , 2011; Westergard et al . , 2011b ) . In the present study , we have shown that reversal of positive charges within this region ( three lysine residues and one arginine residue ) abolishes ΔCR current activity , as does treatment with ligands ( antibodies , Cu2+ ions , pentosan sulfate ) that bind to this region or to other sites within the flexible N-terminal domain ( residues 23–125 ) . Strikingly , the isolated N-terminal domain fused to an unrelated protein ( EGFP ) has the ability to induce spontaneous ionic currents . The magnitude of these currents is quantitatively related to the length of the N-terminus incorporated ( fusions ending at residues 31 , 58 , 90 , and 109 produce progressively more current ) , and the activity of these constructs is entirely dependent on the presence of the 23–31 region . Taken together , these results suggest that the N-terminal domain of PrPC acts as an autonomous effector of ionic current activity , and that basic amino acids at the extreme N-terminus are essential for this activity . We suggest two possible models to explain how the N-terminal domain induces currents . One model is based on the fact that polybasic residues 23–31 resemble a ‘protein transduction domain’ , originally described in the HIV Tat protein ( Wadia et al . , 2008 ) . Such positively charged domains are capable of penetrating lipid bilayers and creating pores , by virtue of binding to and disrupting membrane phospholipids ( Herce and Garcia , 2007 ) . The N-terminal domain of PrPC may thus function as a ‘tethered’ protein transduction domain capable of spontaneously and transiently penetrating the lipid bilayer to produce pores that allow passage of ions . Consistent with penetration of a positively charged protein domain into or across the cell membrane , we find that the ionic current activity associated with both ΔCR and PrP ( N ) -EGFP-GPI is apparent only at hyperpolarized holding potentials ( <−30 mV ) , similar to the resting potential of neurons ( −60 to −70 mV ) . A second possibility is that the N-terminal domain interacts with other membrane proteins , for example endogenous ion channels or channel-modulating proteins , to induce current activity . Supporting this hypothesis is our observation that co-expression of wild-type PrPC suppresses the ionic currents induced by ΔCR and PrP ( N ) -EGFP-GPI , a phenomenon that could be attributable to competition between the wild-type and mutant forms for a common membrane-associated target protein . Three different antibodies ( POM1 , D18 , ICSM-18 ) targeting the C-terminal domain all induce ionic current activity in cells expressing WT PrPC . Importantly , the properties of these currents are identical to those of the spontaneous currents associated with ΔCR PrP , in terms of their sporadic nature , their blockage by N-terminal ligands ( PPS , antibodies ) , and their absolute dependence on the presence of the polybasic 23–31 region . Based on crystal structures , mutagenesis studies , and peptide arrays , the epitopes of these antibodies , while not identical , are largely overlapping and encompass the outer surface of helix 1 , as well as parts of the β1-α1 loop and helix 3 in the case of POM1 ( Antonyuk et al . , 2009; Doolan and Colby , 2015; Sonati et al . , 2013 ) . Three different antibodies ( POM3 , D13 and ICSM-35 ) recognizing a basic region following the octapeptide repeats ( residues 93–110 ) were ineffective at inducing currents , although a fourth one ( 6D11 ) had a weak effect at high concentrations . Two other antibodies targeting additional epitopes in the C-terminal domain had no effect . Table 1 summarizes the effects of the antibodies on current activity . 10 . 7554/eLife . 23473 . 025Table 1 . Anti-prion antibodies used in this study , and their ability to induce ionic currents and dendritic toxicity . DOI: http://dx . doi . org/10 . 7554/eLife . 23473 . 025AntibodyEpitopeCurrentsDendritic toxicityD18 ( holo and Fab ) 132–156 ( α1 ) YesYesPOM1138-147/204 , 8 , 12 ( α1/α3 ) YesYesICSM18 ( holo and scFv ) 143–156 ( α1 ) YesYes6D1193-109/97-100WeakWeakICSM3593–105NoNoD1395–105NoNoPOM4121-134/218-21 ( β1/α3 ) NoNoPOM6140/145/174/177 ( α1/ α2 ) NoNoPOM11*51–90 ( octarepeats ) NoNo*Blocks currents induced by D18 and POM1 . Together , these findings suggest a novel intramolecular regulatory mechanism controlling the activity of PrPC ( Figure 8A ) . Published crystal structures of both POM1 ( Sonati et al . , 2013 ) and ICSM-18 ( Antonyuk et al . , 2009 ) bound to PrPC indicate that these antibodies do not induce major structural alterations in the PrPC globular domain compared to the unliganded state , arguing against antibody-induced allosteric changes as a toxic mechanism . Rather , our results suggest that antibodies bound to helix 1 disrupt a critical regulatory interaction between the N- and C-terminal domains , thereby liberating the N-terminal domain to produce toxic effects ( Figure 8B ) . The fact both Fab and scFv forms of the relevant antibodies display current-inducing activity suggests that antibody ligands as small as 25–50 kDa are able to disrupt N-C interactions . We propose that a similar loss of regulation occurs when residues within the 95–125 region are deleted ( as in ΔCR ) , or when an unrelated protein is substituted for the C-terminal domain ( PrP ( N ) -EGFP-GPI ) ( Figure 8C , D ) . In this scenario , the toxic activity of the liberated N-terminal domain would be blocked by binding of ligands , including PPS , antibodies , or Cu2+ ions ( Figure 8E ) . 10 . 7554/eLife . 23473 . 026Figure 8 . Models for the neurotoxic effects of PrP . ( A ) The C-terminal domain of PrPC negatively regulates the toxic effector function of the N-terminal domain . +++ , basic residues within the 23–31 region at the extreme N-terminus , which are essential for the toxic action of PrP . ( B ) Binding of monoclonal antibodies to the C-terminal domain disrupts this regulatory interaction , releasing the N-terminal domain to produce toxic effects . ( C ) Deletion of the central region , as in ΔCR PrP , produces a similar loss of regulation , with toxic consequences . ( D ) When EGFP is substituted for the C-terminal domain of PrPC , regulation is also lost . ( E ) Binding of ligands ( PPS , antibodies , Cu2+ ) to the N-terminal domain of ΔCR PrP blocks its ability to exert toxic effects . These ligands would have a similar , inhibitory effect on PrP ( N ) -EGFP-GPI ( not shown ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23473 . 026 The model proposed in Figure 8 is supported by biophysical evidence . NMR studies performed here as well as previously ( Evans et al . , 2016; Spevacek et al . , 2013 ) demonstrate that the N-terminal domain docks onto the C-terminal domain , thereby regulating in cis the ability of the N-terminal domain to promote toxic effector functions under normal conditions . Of note , the docking site encompasses the POM1 epitope ( Evans et al . , 2016 ) . Thus , binding of antibodies to this region would be predicted to disrupt the N-C interaction , leading to toxic activity . We also show here that deletion of the central region ( in ΔCR PrP ) weakens Cu2+-induced N-C interaction , consistent with the hypothesis that the toxicity of the ΔCR mutant results from disrupted regulation of the N-terminal domain . Cu2+ ions are physiological ligands of PrPC ( Millhauser , 2007 ) , and changes in endogenous Cu2+ concentration are likely to modulate the strength of N-C interactions . Cu2+ binding to the octarepeats may also directly suppress the toxic activity of the N-terminal domain , as we have observed for the currents induced by ΔCR PrP and PrP ( N ) -EGFP-GPI . In addition to acutely inducing ionic currents , helix 1 antibodies ( POM1 , D18 , ICSM18 ) cause major changes in dendritic morphology of cultured hippocampal neurons , in particular the appearance of blebs or varicosities , when applied for longer periods of time ( 48 hr ) ( Table 1 ) . These antibody-induced dendritic changes are , like the currents induced by the same antibodies , entirely dependent on the N-terminal domain of PrPC , and are blocked by N-terminal ligands and deletions of residues 23–31 . The parallel characteristics of the ionic currents and the dendritic changes induced by C-terminal antibodies suggest a mechanistic connection between the two phenomena . One possibility is that chronic activation of PrPC-mediated currents leads directly or indirectly to dendritic degeneration . Consistent with this possibility , dendritic varicosities similar to those caused by anti-PrP antibodies are a characteristic feature of glutamate excitotoxicity ( Hasbani et al . , 2001 ) , and glutamate excitotoxicity has been implicated in the neuronal degeneration induced by ΔCR PrP ( Biasini et al . , 2013; Christensen et al . , 2010 ) . A second possibility is that current induction and dendritic degeneration are parallel events that represent two distinct outputs of antibody binding to PrPC . For example , the N-terminal domain of PrPC may stimulate currents by interacting with the lipid bilayer , and dendritic changes by interacting with signal-transducing proteins embedded in the membrane . Several other studies have reported toxic effects of anti-PrP antibodies , including several of the ones used here , although some of these results have been contradictory . It has been reported that antibodies D13 ( Reimann et al . , 2016; Solforosi et al . , 2004 ) , POM1 ( Sonati et al . , 2013 ) , and ICSM-18 ( Reimann et al . , 2016 ) , but not D18 ( Solforosi et al . , 2004 ) , caused acute death ( within 24–48 hr ) of neurons when injected stereotaxically into the hippocampus or cerebellum . Using a similar protocol , however , Klöhn et al . ( 2012 ) found that D13 and ICSM-18 were non-toxic , as was ICSM-35 . Finally , Sonati et al . ( 2013 ) observed that chronic treatment ( 10–21 days ) of cerebellar slices with several C-terminally directed antibodies , including POM1 , caused neuronal death , and they concluded that this effect was dependent on the flexible N-terminal domain . The results presented here are consistent with this proposal . It has been reported that anti-PrP antibodies trigger several toxic mechanisms in cerebellar slices , including generation of reactive oxygen species , calpain activation and stimulation of the PERK arm of the unfolded protein response ( Sonati et al . , 2013 ) . Whether these pathways are operative in our system remains to be determined . Since we observe relatively acute changes in dendritic morphology without loss of neuronal viability , it is possible that the downstream toxic pathways engaged by antibody treatment in our system may be different . The results presented here have several important implications . First , they add to concerns that have been raised regarding the use of anti-PrP antibodies as therapeutic tools for treatment of prion diseases ( White et al . , 2003 ) and Alzheimer’s disease ( Klyubin et al . , 2014 ) , given the potential side-effects of these reagents on neuronal viability at nanomolar concentrations ( Reimann et al . , 2016; Solforosi et al . , 2004; Sonati et al . , 2013 ) . Second , our results also suggest a mechanism by which pathologic ligands that bind to PrPC could produce neurotoxic effects by disrupting the normal regulatory cis-interaction between the N- and C-terminal domains , similar to the action of the antibodies described in this study . The Alzheimer’s Aβ peptide , which binds to PrPC and triggers functional and structural alterations in synaptic transmission ( Laurén et al . , 2009 ) , is an example of such a ligand . PrPSc , which also binds to PrPC ( Solforosi et al . , 2007 ) , might act in a similar fashion , although the fact that mice expressing N-terminally truncated PrPC remain susceptible to prion diseases ( Supattapone et al . , 2001; Turnbaugh et al . , 2012 ) argues against this as the primary pathogenic mechanism in these disorders . Finally , it is possible that the antibody-induced effects we have observed here are a reflection of a physiological activity of PrPC . If so , natural ligands , including proteins , small molecules or metal ions , may exist , whose binding to PrPC regulates an effector activity of the N-terminal domain , similar to the way that we suppose anti-PrP antibodies operate . Copper ions are examples of natural ligands , binding of which to the octapeptide repeats promotes docking of the N- and C-terminal domains ( Evans et al . , 2016; Spevacek et al . , 2013 ) . Endogenous ligands for the globular domain may also exist , which either enhance or disrupt N-C interaction . Previous studies have implicated the N-terminal domain of PrPC in several physiological activities of PrPC ( Parkin et al . , 2007; Pauly and Harris , 1998; Sempou et al . , 2016; Shyng et al . , 1995b; Taylor et al . , 2005; Watt et al . , 2012 ) , some of which may be regulated by interaction with the C-terminal domain . POM1 , POM3 , POM4 , POM6 and POM11 antibodies ( Polymenidou et al . , 2008 ) were provided to J . T . by the University of Zürich , Institute of Neuropathology . Hybridomas producing the human-mouse chimeric monoclonal antibodies D13 and D18 ( Safar et al . , 2002; Williamson et al . , 1998 ) were provided by Anthony Williamson , Dennis Burton , and Bruce Chesebro . Antibodies were affinity-purified using protein A spin columns ( Montage Antibody Purification Kit , EMD Millipore ) . Fab fragments of D18 were prepared using the Pierce Fab preparation kit from Thermo Scientific . The purity of all antibody preparations was verified by SDS-PAGE . The following antibodies were purchased from commercial sources: 100B3 ( Wagening UR , Netherlands ) ; 6D11 ( BioLegend , cat #808001 ) ; ICSM-18 and ICSM-35 ( D-Gen Ltd . ) . Pentosan polysulfate ( average MW = 4500–5000 ) was purchased from Biopharm Australia Pty Ltd . , and phosphatidylinositol-specific phospholipase C from Sigma ( cat #P5542 ) . The genes encoding the scFv form of ICSM18 , constructed as described ( Doolan and Colby , 2015 ) , and anti-fluorescein 4-4-20 scFv ( negative control; obtained from Anne S . Robinson ) were subcloned into pHAGE-CMV-dsRed-UBC-GFP-W ( Addgene plasmid #24526 ) in place of dsRed by restriction digest cloning such that the final construct included an N-terminal Igκ light chain secretion signal and a C-terminal FLAG tag ( DYKDDDDK ) fused to the scFv , while maintaining the separate ORF expressing GFP as a reporter . Lentivirus was generated by co-transfection of 5 × 105 HEK cells with 1 . 3 µg pHAGE expression vector , 1 µg psPax2 , and 0 . 65 µg pMD2 . G ( Addgene plasmids #12260 and #12259 ) with TransIT-293 transfection regent ( Mirus Bio LLC ) . After 72 hr , the supernatant was filtered with a 0 . 45-µm syringe-tip filter and used directly for transduction . CHO-S cells adapted to serum-free suspension culture ( kindly provided by Kelvin H . Lee ) were grown in serum-free culture medium ( Thermo Fisher Scientific , SH3054902 ) in vent-cap shake flasks in a humidified incubator at 37°C/5% CO2 . Stable CHO cell lines were created by transducing 5 × 105 CHO cells in 9 ml with 1 ml of HEK supernatant containing lentiviral particles in the presence 6 µg/ml polybrene ( Sigma-Aldrich ) for 2 days . Transduced cells were separated from non-transduced cells based on GFP reporter expression by fluorescence-activated cell sorting ( FACS ) using a BD FACSAria II FACS machine . Production of scFv was carried out by culturing the CHO lines for 7 days; initial cultures contained 5 × 105 cells/ml . Supernatant was harvested by centrifugation and filtration with a 0 . 22-µm syringe-tip filter before purification by anti-FLAG-affinity chromatography ( Sigma-Aldrich ) and low-pH elution according to the manufacturer’s protocol . Eluted scFv was concentrated and buffer-exchanged with a 10 kDa molecular weight cut-off centrifugal filter , assayed by silver stain to ensure purity >90% , and quantified by Bradford assay . Purified scFv was stored in Tris-buffered saline at 4°C for short-term use , or −20°C for long-term storage . pcDNA3 . 1 ( + ) Hygro plasmids ( Invitrogen ) encoding WT , ΔCR , ΔCR/Δ51–90 , or Δ23–31 PrP have been described previously ( Solomon et al . , 2010 , 2011; Turnbaugh et al . , 2011 ) . PrP ( N ) -EGFP-GPI constructs were created by fusing DNA sequences encoding residues 1–31 , 1–58 , 1–90 , 1–109 or 32–109 of mouse PrP to residues 1–239 of EGFP , followed by a poly-glycine/serine linker ( GGGGS ) 4 and then by residues 222–254 of mouse PrP in order to maintain GPI anchoring of the fusion protein . The ΔCR/E3D plasmid encodes ΔCR PrP with the following mutations: K23E , K24E , R25D and K27E . N2a cells ( Cat . #: ATCC CCL-131 , RRID: CVCL_0470 ) were maintained in DMEM supplemented with nonessential amino acids , 10% fetal bovine serum and penicillin/streptomycin . The N2a cell line we used in this study is mycoplasma free . Cells were transiently transfected using Lipofectamine 2000 with pEGFP-N1 ( Clontech ) , along with empty pcDNA3 . 1 ( + ) Hygro vector , or vector encoding WT or mutant PrPs . Cell-surface expression of all PrP constructs was confirmed by immunofluorescence staining . Hippocampi from the newborn pups of the indicated genotypes were dissected and treated with 0 . 25% trypsin at 37°C for 12 min ( Shen et al . , 2006 ) . Cells were plated at a density of 65 , 000 cells/cm2 on poly-D-lysine-coated coverslips in DMEM medium with 10% F12 and 10% FBS . Prnp−/− ( Zurich I ) mice ( Büeler et al . , 1992 ) on the C57BL6 background , and Tga20 mice ( Fischer et al . , 1996 ) have been described previously , and were obtained from EMMA ( European Mouse Mutant Archive ) . Tg ( PrPΔ23–31 ) mice ( Turnbaugh et al . , 2011 ) and Tg ( PrPΔ23–111 ) mice ( Westergard et al . , 2011a ) have been described previously , and were maintained on a Prnp-/- background . Wild-type C57BL6 mice were obtained from Charles River Laboratories . Recordings were made from N2a cells 24–48 hr after transfection . Transfected cells were recognized by green fluorescence resulting from co-transfection with pEGFP-N1 . Hippocampal neurons were analyzed after 13–15 days in culture . Whole-cell patch clamp recordings were collected using standard techniques . Pipettes were pulled from borosilicate glass and polished to an open resistance of 2–5 megaohms . Experiments were conducted at room temperature with the following solutions: internal , 140 mM Cs-glucuronate , 5 mM CsCl , 4 mM MgATP , 1 mM Na2GTP , 10 mM EGTA , and 10 mM HEPES ( pH 7 . 4 with CsOH ) ; external , 150 mM NaCl , 4 mM KCl , 2 mM CaCl2 , 2 mM MgCl2 , 10 mM glucose , and 10 mM HEPES ( pH 7 . 4 with NaOH ) . Current signals were collected from a Multiclamp 700B amplifier ( Molecular Devices , Sunnyvale , CA ) , digitized with a Digidata 1440 interface ( Molecular Devices ) , and saved to disc for analysis with PClamp 10 software . Immunofluorescence staining of cell surface PrPC on N2a cells was performed by incubating livings cells on ice with D18 antibody , fixing in 4% paraformaldehyde in PBS , and then labeling with Alexa Fluor 594 goat anti-human IgG ( Molecular Probes , Eugene , OR ) . Images of N2a cells were acquired with a Zeiss LSM 710 confocal microscope . Neurons cultured for 14 days were fixed with 4% paraformaldehyde in PBS and permeabilized with 0 . 2% Triton100 . Fixed cultures were then incubated with primary antibodies against microtubule-activated protein 2 ( MAP2; polyclonal; 1:1000; Abcam ) , followed by fluorescent secondary antibody . Confocal microscopic analysis was performed on a Zeiss LSM 710 microscope using a 20X objective lens . Identical acquisition settings were applied to all samples of the experiment . Images were analyzed with the NIH Image J program . Recombinant PrP constructs encoding wild-type mouse PrP ( 23-230 ) and mouse ΔCR PrP ( Δ105–125 ) in the pJ414 vector ( DNA 2 . 0 ) were expressed in E . coli ( BL21 ( DE3 ) Invitrogen ) ( Evans et al . , 2016 ) . Mutations were introduced using PCR-based , site-directed mutagenesis with mutagenic primers ( Invitrogen ) and Phusion DNA Polymerase ( Finnzymes ) . All constructs were confirmed by DNA sequencing . Bacteria were grown in M9 minimal media supplemented with 15NH4Cl ( 1 g/L ) for 1H-15N HSQC experiments or 15NH4Cl and 13C6-glucose ( 2 . 5 g/L ) for 1H , 13C , 15N triple-resonance experiments ( Cambridge Isotopes ) . Cells were grown at 37°C until reaching an optical density ( OD ) of 0 . 6 , at which point expression was induced with 1 mM isopropyl-1-thio-D-galactopyranoside ( IPTG ) . PrP constructs were purified as previously described ( Spevacek et al . , 2013 ) . Briefly , proteins were extracted from inclusion bodies with 8 M guanidium chloride ( GdnHCl ) ( pH 8 ) at room temperature and were purified by Ni2+-immobilized metal-ion chromatography ( IMAC ) . Proteins were eluted from the IMAC column in 5 M GdnHCl ( pH 4 . 5 ) and were brought to pH 8 with KOH and left at 4°C for 2 days to oxidize the native disulfide bond . Proteins were then desalted into 10 mM KOAc buffer ( pH 4 . 5 ) and purified by reverse-phase HPLC on a C4 column . The purity and identity of all constructs were verified by analytical HPLC and mass spectrometry ( ESI-MS ) . Disulfide oxidation was confirmed by reaction with N-ethylmaleimide and subsequent ESI-MS analysis . Lyophilized protein samples were dissolved in degassed Milli-Q-purified H2O and allowed to fully solubilize prior to all experiments . Protein concentrations were determined from the absorbance at 280 nm in GdnHCl buffer using an extinction coefficient of 64 , 840 M−1 cm−1 calculated for mouse PrP ( 23-230 ) and 62 , 280 M−1cm-1 calculated for mouse ΔCR PrP ( Δ105–125 ) . For samples containing Cu2+ , the metal ion was added from stock solutions of Cu ( OAc ) 2 or CuCl2 in H2O in which the Cu2+ concentrations were accurately determined by EPR integration ( Walter et al . , 2006 ) . Samples for NMR contained 200–400 µM PrP in 10 mM MES buffer ( pH 6 . 1 ) with 10% D2O . NMR experiments were conducted on a Varian INOVA 600 MHz spectrometer equipped with a 1H , 13C , 15N triple-resonance cryoprobe . Resonance assignments were first obtained using standard triple-resonance experiments with a 400 µM samples of uniformly 13C , 15N-labeled mouse PrP ( 23-230 ) sample at 25°C . Experiments included HNCO , HN ( CA ) CO , HNCACB , CBCA ( CO ) NH , and CC ( CO ) NH . The assignments were then transferred to the 1H-15N HSQC spectrum at 300 µM and 37°C by following the cross-peaks through concentration and temperature titrations . Resonances were confirmed , and additional resonances were assigned by recording three-dimensional HNCACB and 15N-NOESY-HSQC spectra at 300 µM and 37°C . Finally , resonance assignments were transferred to the 1H , 15N HSQC spectrum of MoPrP ( H95Y/H110Y ) at 300 µM and 37°C by visual inspection . All NMR spectra were processed with NMRPipe and NMRDraw ( Delaglio et al . , 1995 ) , and analyzed using Sparky NMR Analysis and CcpNmr Analysis ( Vranken et al . , 2005 ) . 1H , 15N HSQC spectra were recorded for MoPrP ( H95Y/H110Y ) and MoPrP ( Δ105–125 ) H95Y constructs ( 300 µM ) at 37°C both in the absence of metal ions and in the presence of 300 µM CuCl2 , and the HSQC peak intensities were determined using Sparky NMR Analysis and CcpNmr Analysis . Intensity ratios were analyzed using Kaleidagraph ( Synergy Software ) and residues with intensity reductions greater the one standard deviation of the mean were considered significantly perturbed .

Prion diseases are a group of degenerative illnesses of the brain caused when a molecule called the prion protein ( PrP for short ) adopts the wrong shape . These diseases include the human form of mad cow disease , and are often fatal with no effective treatments or cures . Though the normal activity of PrP is not certain , abnormal PrP can affect the healthy PrP on the surface of brain cells and lead to disease . Similar mechanisms may also contribute to other life-threatening brain disorders , including Alzheimer’s disease and Parkinson’s disease . It had been shown that certain altered PrP proteins caused the death of brain cells by allowing excessive electrical charges to cross the membranes of the cell . These changes led to symptoms in animal models of the diseases . Experiments showed that adding a large amount of normal PrP to the cells could prevent these effects . These studies , however , had not yet resolved how PrP behaves inside cells and how this contributes to disease . Using genetically modified mice and cells grown in the laboratory , Wu et al . investigated the role of different parts of PrP in causing brain cells to degenerate . The experiments showed that one end of the protein , called the N-terminus , is involved in the movement of electrical charges across the cell membrane and is able to cause cell degeneration . By contrast , the other end of the protein , the C-terminus , acts as a regulator for the N-terminus and can prevent cell degeneration . Further investigation revealed that the C-terminus regulates the N-terminus through direct contact . A better understanding of the role of PrP in prion diseases may help to reveal new treatments for these and other degenerative brain disorders . In particular , the new findings highlight that treatments should target the toxic N-terminus of altered PrP and not the regulatory C-terminus . Further study will examine how different molecules in the brain control the interaction between the two ends of PrP in healthy brain cells and how this is altered in diseased cells .

2017

The N-terminus of the prion protein is a toxic effector regulated by the C-terminus

Understanding how option values are compared when making a choice is a key objective for decision neuroscience . In natural situations , agents may have a priori on their preferences that create default policies and shape the neural comparison process . We asked participants to make choices between items belonging to different categories ( e . g . , jazz vs . rock music ) . Behavioral data confirmed that the items taken from the preferred category were chosen more often and more rapidly , which qualified them as default options . FMRI data showed that baseline activity in classical brain valuation regions , such as the ventromedial Prefrontal Cortex ( vmPFC ) , reflected the strength of prior preferences . In addition , evoked activity in the same regions scaled with the default option value , irrespective of the eventual choice . We therefore suggest that in the brain valuation system , choices are framed as comparisons between default and alternative options , which might save some resource but induce a decision bias . Standard decision theory assumes that when faced with a choice , individuals first assign subjective values to each option , and then compare these values in order to select the best option ( Samuelson , 1938; Von Neumann and Morgenstern , 1947 ) . Understanding the neural mechanisms governing this valuation/selection process has become a central aim in the field of decision neuroscience . A large set of fMRI evidence points to the ventro-medial Prefrontal Cortex ( vmPFC ) as a key player in the valuation process ( Bartra et al . , 2013; Clithero and Rangel , 2014 ) . Neural activity in the vmPFC reflects subjective values , either measured with likeability ratings or inferred from binary choices ( Kable and Glimcher , 2009; Rangel and Hare , 2010 ) . In accordance with the idea of a common neural currency ( Levy and Glimcher , 2012 ) , the vmPFC was found to encode the subjective value of many kinds of goods , such as food , money , trinkets , faces , paintings , charities , etc . ( Chib et al . , 2009; Hare et al . , 2010; Lebreton et al . , 2009; Plassmann et al . , 2007 ) . Such value coding was observed not only during choice but also in the absence of choice , during passive viewing of items presented in the attentional focus or when performing a distractive task on these items ( Lebreton et al . , 2009; Levy et al . , 2011; Abitbol et al . , 2015 ) . During binary choices , it has been repeatedly shown that vmPFC activity correlates with the relative value of the two options under consideration ( VA–VB ) . However , the framing of such decision value signal , i . e . what A and B actually represent , remains an unresolved issue . This question is of importance because the brain regions downstream in the decision process cannot operate the appropriate selection without knowing which option is favored by the relative value signal . In particular , the post-decisional frame that has often been reported ( Boorman et al . , 2013; Hare et al . , 2011; Hunt et al . , 2012 ) provides a decision value signal between chosen and unchosen options ( Vch-Vunch ) that cannot be used for making the selection . A spatial frame , based on the location of options ( e . g . , Vleft-Vright ) , has been suggested but not supported by much experimental evidence regarding the vmPFC valuation signal ( Palminteri et al . , 2009; Wunderlich et al . , 2009; Skvortsova et al . , 2014 ) . A more promising suggestion is the attentional frame ( Krajbich et al . , 2010 ) , in which the decision value signal encoded in the vmPFC depends on which option is attended to ( Vatt-Vunatt ) . Such framing provided a good account for vmPFC activity in a choice task where fixation patterns were imposed to subjects , and correctly predicted several features of spontaneous choice behavior by imposing a discount weight on the unattended option value ( Krajbich et al . , 2010; Lim et al . , 2011 ) . Notably , the attentional frame predicts that more fixated options should be more frequently chosen , which might explain why vmPFC activity has been found to correlate with Vch-Vunch in other studies . However , the attentional model assumes that visual exploration is random , which might be true in artificial laboratory tasks where subjects have no information about upcoming options , but not in natural situations where prior knowledge might play a role . Here , we hypothesize that the framing of the decision value encoded in the vmPFC is imposed by prior preferences . In other words , vmPFC activity should scale positively with the value of the option that is preferred a priori , which we call the default option , and negatively with that of the alternative ( Vdef-Valt ) . This hypothesis is compatible with the observation that vmPFC activity correlates with Vch-Vunch , since choices usually follows on prior preferences . Yet , the interpretation is fundamentally different , as Vdef-Valt is a pre-decisional value signal susceptible to drive option selection . Our hypothesis builds on the literature about optimal foraging , which argues that stay/switch choice is the natural case of decision-making ( Stephens and Krebs , 1986 ) . In this framework , staying on a same patch is the default option against which all alternatives must be compared . Several studies investigated such stay/switch decisions and implicated the dorsal anterior cingulate cortex in promoting a shift away from the default option ( Hayden et al . , 2011; Kolling et al . , 2012; Kvitsiani et al . , 2013 ) , while others induced default policies by manipulating prior probabilities of being correct ( Boorman et al . , 2013; Fleming et al . , 2010; Mulder et al . , 2012; Scheibe et al . , 2010 ) . Although experimental manipulations vary across these studies , the default option is always defined as the option that would be selected in the absence of further information processing about its value relative to the alternatives . This definition provides objective criteria to identify the default option in a choice set: it should be selected faster and more frequently than the alternatives . Therefore , our hypothesis implies that prior preferences should ( 1 ) induce a bias in favor of the default option , and ( 2 ) determine the frame of the value comparison process . The purpose of the present study was to examine how these two constraints would shape the brain valuation signal . To do so , we exploited the hierarchical structure of preferences: individuals have global preferences between categories of goods that can be locally reversed when comparing particular items . For instance , someone may prefer pop to jazz music in general , but nonetheless pick Keith Jarrett if the only other option is Britney Spears . In a binary choice , the prior preference at the category level thus designates a default option ( i . e . , the item belonging to the preferred category ) , but the option values still need to be compared at the item level in order to reach a final decision . We conducted an fMRI experiment where participants made binary choices between items belonging to different categories . Preferences between categories were inferred from likeability ratings that were collected for every item before the scanning session . In the following analyses , we first establish the presence of a bias toward the default option in both choice and response time , above and beyond the prior preference between categories . Using computational modeling , we provide evidence that the default bias is best accounted for by a shift in the starting point of a drift diffusion process , which is proportional to the prior preference between categories . Then , we show that the default bias is unrelated to gaze fixation pattern , precluding an attentional framing . Finally , we uncover two effects of prior preference in fMRI data: ( 1 ) vmPFC baseline activity reflects the a priori shift in favor of the default option , and ( 2 ) vmPFC evoked response represents the value of the default option , irrespective of the eventual choice . Prior to the scanning session , participants ( n = 24 ) rated the likeability of items belonging to three different domains ( food , music , magazines ) . Each domain included four categories of 36 items ( see Materials and methods ) . At that time , participants were unaware of these categories . This is because the presentation of items for likeability ratings was blocked by domain but not by categories , which were randomly intermixed . During the scanning session , subjects performed series of choices between two items ( Figure 1 ) , knowing that one choice in each domain would be randomly selected at the end of the experiment and that they would stay in the lab for another 15 min to enjoy their reward ( listening to the selected CD , eating the selected food and reading the selected magazine ) . Trials were blocked in a series of nine choices between items belonging to the same two categories within a same domain . The two categories were announced at the beginning of the block , such that subjects could form a prior preference ( although they were not explicitly asked to do so ) . We quantified this prior preference as the difference between mean likeability ratings ( across all items within each of the two categories ) , which is hereafter denoted as DVCAT . In most cases ( 84 ± 3% on average ) , preferences inferred from mean ratings matched the preferences between categories that subjects directly expressed in post-scanning debriefing tasks . Moreover , the confidence in these choices between categories , which subjects provided on an analog rating scale during debriefing , was significantly correlated to DVCAT ( r = 0 . 44 ± 0 . 06 , t ( 23 ) = 7 . 88 , p = 5 . 10−8 ) . These explicit measures taken after the scanning session therefore validate our quantification of implicit preferences between categories . In the following , we analyze choices and response times to assess the presence of a bias in favor of the default option ( i . e . , the item belonging to the preferred category ) . 10 . 7554/eLife . 20317 . 003Figure 1 . Choice task . Participants performed the choice task inside the MRI scanner . The task was composed of four 12-block sessions . During a block , subjects first saw an instruction screen indicating the reward domain ( e . g . , food ) and the two categories from which choice options were drawn . Then , they had to make a series of nine binary choices , each confronting the two categories with two novel items . The choice was self-paced and feedback on chosen option was provided to the subject . DOI: http://dx . doi . org/10 . 7554/eLife . 20317 . 003 We fitted a simple logistic regression model including a constant , the default option value , denoted VIT ( def ) , and the alternative option value , denoted VIT ( alt ) , to choices expressed in the ‘default vs . alternative’ frame . Regression coefficient estimates showed that the two option values were equally contributive to the choice ( VIT ( def ) : β = 0 . 060 ± 0 . 005 , t ( 23 ) = 11 . 90 , p = 3 . 10−11; VIT ( alt ) : β = −0 . 060 ± 0 . 004 , t ( 23 ) = −14 . 21 , p = 7 . 10−13 ) . Crucially , the constant was significantly positive ( β = 0 . 68 ± 0 . 13 , t ( 23 ) = 5 . 40 , p = 2 . 10−5 ) , bringing evidence for a bias toward the default option . This constant was significantly reduced when including DVCAT in the regression model ( β = 0 . 31 ± 0 . 16 , t ( 23 ) = 1 . 94 , p = 0 . 06 ) , with the effect of DVCAT itself being significant ( β = 0 . 021 ± 0 . 006 , t ( 23 ) = 3 . 53 , p = 2 . 10−3 ) , which established a direct link between prior preference and default bias . We also introduced past choices ( coded 1 vs . −1 when default option was chosen vs . unchosen ) in the regression model but they yielded no significant effect on choice rate . Consistently , the constant estimate was not different when restricting the logistic regression to the first choice in a block ( βfirst = 0 . 58 ± 0 . 49 , βall = 0 . 68 ± 0 . 13 , difference: t ( 23 ) = 0 . 97 p = 0 . 44 ) , confirming that the default bias was not resulting from the history of past choices . To illustrate this result ( Figure 2A ) , we plotted the choice rate , P ( def ) , as a function of the decision value , DVIT=VIT ( def ) -VIT ( alt ) . This plot shows that even when the two options have the same value ( DVIT = 0 on the x-axis ) , the choice rate is not at chance level ( 50% on the y-axis ) , which would denote indifference , but shifted toward the default option ( by 15 . 7 ± 1 . 7% on average ) . Thus , these results provide behavioral evidence for a ‘choice bias’ occurring on top of the decision value ( DVIT ) , i . e . above and beyond what could be predicted by the difference in likeability rating . 10 . 7554/eLife . 20317 . 004Figure 2 . Behavioral results ( MRI experiment ) . ( A ) Probability of choosing the default option , P ( def ) , plotted as a function of decision value , DVIT , divided into 20 bins . Values correspond to likeability ratings given by the subject prior to scanning . Both probabilities observed in choice data ( solid line ) and simulated from the fitted Drift Diffusion Model ( dashed line ) are shown . Choice bias was defined as the difference between the observed probability for a null decision value and the expected equiprobability ( 50% ) . ( B ) Choice response time ( RT ) plotted as a function of the absolute decision value , |DVIT| , divided into 10 bins , separately for trials in which the default option was chosen ( black ) and unchosen ( red ) . Both RT observed in behavioral data ( solid line ) and simulated from the fitted Drift Diffusion Model ( dashed line ) are shown . RT bias was defined as the difference between the intercepts observed for the two types of choice . ( C ) Correlation of choice and RT biases across blocks . ( D ) Choice bias plotted as a function of response time , divided into four bins . Inset illustrates the Drift Diffusion model ( adapted from ( Voss et al . , 2013 ) , with S the starting point , DV the mean drift rate and def / alt the thresholds for choosing default / alternative options . Choice bias was larger for shorter RT , suggesting that it could arise from a bias in the starting point . ( E ) Family model comparison between different theoretical accounts of choice and RT biases . Top: the null model ( ‘Fixed’ ) is compared to models in which either the starting point ( ‘Start’ ) or the drift rate ( ‘Drift’ ) is allowed to favor the default option . Bottom: the model with a single free starting point ( ‘1 free S’ ) is compared to models in which the starting point is varied across blocks , either in proportion to the value difference between categories ‘S=a*DVCAT’ or as a set of 12 independent parameters ( ‘12 free S’ ) . Red line corresponds to 95% exceedance probability . ( F ) Correlation across blocks between DVCAT and starting point S ( from fitting the '12-free-S’ model ) . This suggests that the starting point is adjusted in each block to the average value difference between the two confronted categories . Shaded areas and error bars represent ± inter-subject SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 20317 . 004 To account for choice response time ( RT ) , we fitted a general linear model ( GLM ) including the main effects and the interaction of two factors: the unsigned decision value ( |DVIT| ) and the choice type ( default vs . alternative ) . As typically reported , we found a significant effect of unsigned decision value ( t ( 23 ) = −6 . 8 , p = 6 . 10−6 ) , indicating that choices were longer when option values were closer . We also found a significant effect of choice type ( t ( 23 ) = −5 . 47 , p = 1 . 10−5 ) , indicating that subjects were faster to pick the default option than the alternative . There was no significant interaction between the two factors ( t ( 23 ) = 0 . 59 , p = 0 . 56 ) . Thus the ‘RT bias’ corresponds to the difference between intercepts for a null decision value ( Figure 2B ) . This RT bias means that subjects were significantly faster when choosing the default ( by 357 ± 50 ms on average ) , irrespective of the decision value . To assess whether the choice and RT biases could arise from the same underlying computation , we tested their correlation across blocks ( Figure 2C ) . This is possible in our design because each block corresponds to a confrontation between two given categories , some being very close and others far apart in terms of mean likeability ( i . e . , they vary in terms of DVCAT ) . We fitted a regression model to each block in order to extract choice and RT biases for each pair of categories . Correlation across blocks was estimated at the subject level and then tested against the null hypothesis at the group level . We found a significant correlation between the two biases ( r = 0 . 24 ± 0 . 06 , t ( 23 ) = 3 . 78 , p = 1 . 10−3 ) , suggesting a common underlying mechanism , which we further characterized using computational modeling . To account for both choice and RT distributions , we employed an analytical approximation to the Drift Diffusion Model ( DDM ) . The DDM assumes that choices result from a sequential sampling process , through which a decision variable accumulates evidence until it reaches a boundary ( Ratcliff , 1978; Ratcliff and McKoon , 2008 ) . DDMs were originally developed to explain perceptual decisions but they have already been successfully applied to economic ( value-based ) decisions ( Gold and Shadlen , 2007; Basten et al . , 2010; Krajbich et al . , 2010; 2012 ) . In our DDM , the boundaries corresponded to the default and alternative choices , and the mean of the drift rate was the signed decision value , DVIT ( inset in Figure 2D ) . A priori , the choice and RT biases could arise from a change in the drift rate or from a shift in the starting point , S . The latter possibility is more consistent with the negative correlation that was observed between choice bias and RT ( Figure 2D ) and tested at the group level ( r = −0 . 68 ± 0 . 08 , t ( 23 ) = −8 . 06 , p = 4 . 10−8 ) . Indeed , in the DDM framework , a bias in the starting point has less impact on choices when the decision process lasts longer ( Brunton et al . , 2013 ) . To formally disentangle between these possibilities , we compared a DDM where the starting point is fixed at zero and the drift rate equal to DVIT ( null model ) to six alternative DDMs where either the starting point or the drift rate is allowed to change across subjects , and for some of them across blocks . The first three models ( ‘start family’ ) test the hypothesis of a shift in the starting point . The shift was captured with a single free parameter in model 1 ( ‘1 free S’ ) , with one free parameter per block in model 2 ( ‘12 free S’ ) , or as a free parameter scaled by DVCAT in model 3 ( ‘S=a*DVCAT’ ) . Thus , the starting point was respectively considered constant across blocks ( but possibly different from zero ) , freely adjusted to each block , or proportional to the prior preference . The last three models ( drift family ) test the hypothesis of a change in the drift rate , which in any case was proportional to DVIT . The change was captured with a single additional parameter to DVIT in model 4 , with one additional parameter per block in model 5 , and with an additional term scaled by DVCAT in model 6 . We first conducted a family model comparison to examine the possibilities that the choice and RT biases were due to a shift in the starting point ( models 1–3 ) or a change in the drift rate ( models 4–6 ) , relative to the null model ( Figure 2E , top ) . The most plausible mechanism was the shift in the starting point ( start family: exceedance probability , xp = 0 . 997 ) . Then , we compared the three models within this family ( Figure 2E , bottom ) and found evidence in favor of model 3 ( xp = 0 . 920 ) , suggesting that the starting point varied across blocks proportionally to prior preferences . We verified this conclusion by testing the correlation across blocks between the posterior means of the 12-free-S model and the prior preference DVCAT ( Figure 2F ) . The correlation was significant at the group level ( r = 0 . 35 ± 0 . 07 , t ( 23 ) = 5 . 12 , p = 3 . 10−5 ) , strengthening the idea that prior preference was imposing a shift in the starting point that resulted in both choice and RT biases . Thus , the correlation observed between choice and RT biases was driven by variations in DVCAT across blocks , the two biases trending to zero when DVCAT was close to null . The fact that decision bias was best explained by shifting the starting point made less likely an interpretation in terms of attentional dynamics . This is because in previous studies , gaze fixation pattern was found to affect the drift rate and not the starting point ( Krajbich et al . , 2010 ) . We nevertheless investigated the possibility that the effect of prior preferences on choice and RT biases could be mediated by the pattern of gaze fixation . This possibility would imply that subjects pay more attention to the default option than to the alternative , which we examined using eye-tracking measurements . Another group of participants ( n = 23 ) performed the same series of rating and choice tasks , while their gaze position on the screen was recorded using an eye-tracking device . All the behavioral results described in the previous section were replicated ( Figure 3A and B ) , with a significant bias in both choice ( 15 . 5 ± 1 . 7% , t ( 22 ) = 5 . 12 , p = 4 . 10−4 ) and RT ( 341 ± 42 ms , t ( 22 ) = −6 . 69 , p = 1 . 10−6 ) , and a significant correlation between the two ( r = 0 . 22 ± 0 . 06 , t ( 22 ) = 3 . 87 , p = 8 . 10−4 ) . 10 . 7554/eLife . 20317 . 005Figure 3 . Behavioral results ( eye-tracking experiment ) . ( A ) Probability of choosing the default option plotted as a function of decision value DVIT . The three curves correspond to probabilities actually observed in choice data ( lines with circles ) and simulated from either the fitted attentional Drift Diffusion Model ( aDDM , solid lines ) or the same model fitted with a starting point proportional to prior preference DVCAT ( asDDM , dashed line ) . ( B ) Choice response time ( RT ) plotted as a function of the absolute decision value |DVIT| , separately for trials in which the default option was chosen ( left ) and unchosen ( right ) . The different curves correspond to RT observed in behavioral data ( lines with circles ) and simulated from either the fitted aDDM ( solid line with circles ) or asDDM ( dashed line ) . Note that the aDDM alone cannot reproduce choice and RT biases . ( C ) Proportion of fixations ( number of trials over all trials ) to the default and alternative options at each time point when default is chosen ( left ) or unchosen ( right ) . Curves are time-locked to choice ( button press ) . They do not add up to one because at a given time point in a given trial , subjects may fixate none of the two options . Shaded areas are ± inter-subject SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 20317 . 005 We also replicated a number of results predicted by the attentional Drift Diffusion Model ( aDDM ) , in which a parameter θ down-weights the value of the unattended item in the decision value , hence in the drift rate ( Krajbich et al . , 2010; Krajbich and Rangel , 2011; Lim et al . , 2011; Krajbich et al . , 2012 ) . As predicted by the aDDM , we notably observed that the choice probability was higher for the item fixated last ( t ( 22 ) = −11 . 68 , p = 7 . 10−11 ) , and for the most fixated item during the decision process ( t ( 22 ) = −4 . 71 , p = 1 . 10−4 ) , irrespective of decision value . These results confirm that fixation pattern had the expected effects on choice . However , none of these effects could account for the bias toward the default option that was observed in our task . To test the link between prior preference and gaze fixation pattern , we compared the duration of fixation for the default and alternative options , separately for trials in which the default was chosen and unchosen . We found that the default option was fixated longer when it was chosen ( difference: 81 ± 11 ms , t ( 22 ) = 7 . 52 , p = 2 . 10−7 ) . Conversely , it was the alternative option that was fixated longer when the default option was not chosen ( difference: 41 ± 17 ms , t ( 22 ) = 2 . 41 , p = 2 . 10−2 ) . Consistently , the ANOVA conducted on fixation time revealed a significant main effect of the ‘chosen vs . unchosen’ factor ( F ( 1 , 88 ) = 4 . 03 , p = 0 . 048 ) , but no main effect of the ‘default vs . alternative’ factor ( F ( 1 , 88 ) = 0 . 43 , p = 0 . 41 ) . The interaction was significant ( F ( 1 , 88 ) = 17 . 45 , p = 1 . 10−4 ) , reflecting the fact that the default option was more frequently chosen , and with larger decision values . Thus , fixation duration was indeed predictive of choice , but was not influenced by prior preference . To control for the dynamics of the decision process , we computed the proportion of fixations for each option at each time point . The time courses locked to stimulus onset revealed a clear preference for looking at the left option during the first 250 ms , and then at the right option during the next 250 ms , but no significant preference for the default option at the beginning of the trial . The time courses locked on the response confirmed the fixation bias toward the chosen option , with a similar pattern whether default or alternative option was chosen ( Figure 3C ) . This result was further confirmed by a model comparison showing that fixation duration for each option was better explained ( xp = 0 . 999 ) by a GLM including the unsigned decision value and the choice ( chosen vs . unchosen option ) than by GLMs including an additional regressor that indicated the prior preference ( default vs . alternative option ) . Finally , we compared four variants of the DDM to contrast how fixation pattern and prior preference influence the decision process . The first was the null model , with a starting point S fixed at zero and a weighting factor θ fixed at one . The second was the sDDM selected as the best model in the first experiment , with S proportional to DVCAT and θ still fixed at one . The third was the standard aDDM , with S fixed at zero and a freely fitted θ . The fourth was termed asDDM and included both S proportional to DVCAT and fitted θ . The most plausible model was the asDDM ( xp = 0 . 95 ) , with the weight on DVCAT significantly above zero ( 0 . 02 ± 0 . 003 , t ( 22 ) = 7 . 85 , p = 1 . 10−6 ) , and a θ significantly below one ( θ = 0 . 94 ± 0 . 03 , t ( 22 ) = 2 . 14 p = 0 . 04 ) . The fits of choice and RT are illustrated for the aDDM and asDDM ( Figure 3A and B ) . Although using the fixation pattern ( with θ ) improved the fit , only the prior preference ( with S ) could explain the decision bias toward the default option . We reached similar conclusions when the advantage for the attended option was additively included in the drift rate , on top of decision value ( as in Cavanagh et al . , 2014 ) . In fact , gaze fixation pattern failed to produce the default bias simply because the default option was no more looked at than the alternative option . Our behavioral results establish that prior preferences exert a bias on choices , which in the DDM framework was best explained by a proportional shift in the starting point . We analyzed fMRI data first to examine whether the bias toward the default option could be observed in baseline neural activity , second to assess whether prior preference could frame the comparison between option values that might be implemented in the evoked neural response . In this study , we examined how prior preference shapes the neural representation of decision value . We observed two major phenomena in vmPFC activity: ( 1 ) baseline activity was shifted in proportion to the strength of prior preference , as was the starting point in a drift diffusion model accounting for the decision bias in favor of the default option , ( 2 ) evoked activity signaled the value of the option belonging to the preferred category , suggesting that the choice was framed as a comparison between default and alternative . Although they were not instructed to do so , subjects likely formed prior preferences at the beginning of blocks , when the two categories confronted in the upcoming series of choices were announced on the screen . Preference between the two categories was inferred from likeability ratings averaged across items belonging to each category . In a vast majority of cases , this notion of preference matched the preference directly expressed by the subjects in binary choices between categories made during post-scan debriefing . Moreover , the difference between mean likeability ratings ( DVCAT ) was proportional to the confidence expressed in these choices between categories , in keeping with the notion that choice and confidence proceed from the same decision value ( De Martino et al . , 2013 ) . These debriefing observations validate our notion of prior preference , which then served to designate the default option in the pair of items that was presented for choice . Indeed items from the preferred category could be qualified as default options , because they were chosen faster and more frequently than their alternatives . These choices and RT biases are minimal requirements for a default option , i . e . an option that should be chosen in the absence of further information processing . Such criteria have been used in other paradigms where the goal was to maximize an objective reward , with for instance the default being defined as the pre-selected option in a perceptual decision task ( Fleming et al . , 2010 ) , as the current patch in a foraging task ( Hayden et al . , 2011; Kolling et al . , 2012; Kvitsiani et al . , 2013 ) , or as the long-term best option in a probabilistic instrumental learning task ( Boorman et al . , 2013 ) . These studies reported that when the two option values were similar , subjects ( both humans and monkeys ) favored the default option . This phenomenon has been coined ‘default bias’ , or ‘status quo bias’ in cases where the default option was defined as the pre-selected choice . Here , the same phenomenon was observed in the case of subjective preference . Importantly , the default bias was estimated once option values were matched , therefore it goes beyond what could be predicted from the difference in likeability ratings between categories . This bias could lead to preference reversals , meaning that subjects would pick the default option in spite of the alternative option having received a higher rating . Thus , the default bias denotes suboptimal decision-making , which could be compensated by the fact that following a default policy is on average less costly in terms of time or cognitive resource , than a systematic unbiased comparison of option values . This phenomenon is therefore much different from the optimal use of prior information that has been observed in a variety of perceptual decision-making paradigms , subjects being biased only when tricked with invalid cues ( Link and Heath , 1975; Bogacz et al . , 2006; Scheibe et al . , 2010; Mulder et al . , 2012 ) . Within the drift diffusion framework , the default bias observed in choice and RT was best accounted for by a shift in the starting point . This is consistent with perceptual decision-making studies showing that prior information on probability or payoff is also incorporated in the starting point ( Scheibe et al . , 2010; Summerfield and Koechlin , 2010; Mulder et al . , 2012 ) . However , this is not compatible with the idea that the effect of prior preference on choice and RT biases is mediated by the pattern of gaze fixations . This idea implies that subjects pay more attention to the default option , which through the attentional DDM mechanism would favor the default choice , because the attended option has more weight than the alternative in the drift rate . In fact , subjects looked equally often at the default and alternative options in the eye-tracking experiment . Our results nonetheless confirmed that the pattern of gaze fixation does inform the prediction of choices , in a manner that is nicely captured by the attentional DDM . Thus , although the attentional DDM is perfectly compatible with our data , it could not by itself explain the default bias . The best account of choice and RT was in fact obtained with a model that cumulated the down-weighting of unattended options in the drift rate , as suggested by the attentional DDM , and the shift in the starting point that explains the default bias . In our best model , the shift in starting point was proportional to the prior preference ( DVCAT ) . A striking parallel was found at the neural level , with the prior preference being reflected in the baseline activity of valuation regions including the vmPFC , ventral striatum and posterior cingulate cortex . This is in line with a previous study showing that baseline vmPFC activity is sensitive to contextual factors , both in humans and monkeys , and provide a bias in subsequent valuation processes ( Abitbol et al . , 2015 ) . Other contextual manipulations were found to modulate vmPFC activity and subsequent valuation , for instance mood induction has been shown to affect reward-related vmPFC activity ( Young and Nusslock , 2016 ) . In contrast , cueing manipulation that affected perceptual decisions through a shift in starting point had no influence on vmPFC activity ( Scheibe et al . , 2010; Summerfield and Koechlin , 2010; Mulder et al . , 2012 ) . This dissociation suggests that the recruitment of vmPFC was not related to the general process of changing the accumulation starting point , but to biasing value-based decisions ( as opposed to perceptual decisions ) . In fact the shift in baseline vmPFC activity was maintained throughout the decision process , and was hence added to the evoked activity , which followed a canonical hemodynamic response . As both baseline and evoked activity scaled with the value of the default , respectively at the category and item levels , they together contributed to favoring the default option over the alternative . Thus , the mechanics is analog to the DDM process , but the dynamics is somewhat different . In fact , the neural dynamics is not compatible with the vmPFC implementing the DDM , since we observed no ramping signal corresponding to an accumulation-to-bound process; neither is it compatible with the vmPFC output being sent to a distant accumulator , since the shift in starting point should not be integrated over time . Therefore , we do not suggest that the DDM used to capture behavioral patterns is literally implemented as such in the brain , just that the general logics and some key features appeared to match vmPFC activity during choices . We also note that other types of modeling would have been possible to capture behavioral effects , notably a Bayesian account where prior preference would affect the mean and perhaps the variance of a prior distribution on decision value . The analysis of the evoked response showed that the vmPFC and ventral striatum encode the decision value in a frame that opposes the default to the alternative option . This pre-choice framing supports the idea of an anatomical separation between the valuation and selection processes , with the vmPFC being implicated in the former but not the latter . It could be a very general frame for value coding in the vmPFC , because most studies found a correlation between vmPFC activity and the value of chosen options ( e . g . , Hare et al . , 2011; Boorman et al . , 2013 ) , which are partially confounded with default options as we have shown here . We note that an opposite dissociation was found by Boorman et al . ( 2013 ) , with the vmPFC encoding option values in post-choice frame , and not pre-choice frame . As decision-making dynamics was not explored in this study , it is unclear whether participants truly implemented a default strategy as defined here , which implies an anticipation of a default response , associated with shortening of response time . Accordingly , the representation of chosen option value was largely delayed in comparison to our study ( peaking 10 s after option display ) , possibly related to the necessity of storing expected values in a learning context . Another partial confound is with choice easiness or confidence , which was also found to be integrated in vmPFC activity in addition to value ( De Martino et al . , 2013; Lebreton et al . , 2015 ) . The pre-choice framing could also be reconciled with the theory that the vmPFC encodes the value of the attended option , if we assume that when they have no prior information on the choice , subjects set up a default on the fly , which could be the option they just look at . By contrast , we found a post-choice framing of decision value ( unchosen vs . chosen ) in the dACC and anterior insula , which could be related either to choice difficulty or to the value of shifting away from the default policy , which might require cognitive control ( Hare et al . , 2011; Kolling et al . , 2012; Shenhav et al . , 2013 ) . A last potential issue is that the correlation with decision value ( DVIT ) was driven by the default option , although the default and alternative options had the same weight on choices , and despite the two options being reflected in other regions such as dACC . Our interpretation is that both option values are encoded in the vmPFC on top of the decision value . As a result , the correlation with the alternative option value would be cancelled out , and the correlation with the default option value would be doubled , as suggested by the following equation:Signal ( vmPFC ) =[VIT ( def ) −VIT ( alt ) ]+[VIT ( def ) +VIT ( alt ) ]=2∗VIT ( def ) This interpretation is consistent with both the idea that the vmPFC automatically encodes the value of items that fall under the attentional focus ( Lebreton et al . , 2009; Levy et al . , 2011 ) and the idea that the vmPFC computes a decision value whenever a choice process is engaged ( Plassmann et al . , 2007; Grueschow et al . , 2015 ) . It would also explain why many studies report a correlation with the chosen value alone and not the decision value , as the unchosen value would be cancelled out for the same reasons ( Wunderlich et al . , 2010; Kolling et al . , 2012; Hunt et al . , 2012 ) . Model comparison supported this post-hoc interpretation , as including the two option values ( sum ) on top of the decision value ( difference ) provided the best account of vmPFC activity during choice . Other techniques than fMRI , with better spatial resolution , would be needed to investigate whether the different value representations rely on different populations on neurons . In conclusion , our findings show that when decision-makers have a prior preference , the brain valuation system is configured so as to compare default and alternative options , with prior and novel information being encoded in baseline and evoked activity , respectively . Such framing could have been selected to solve natural decision problems , with the advantage of saving time and/or cognitive resource , and the disadvantage of biasing choice toward the default policy . How the valuation system adapts to artificial economic choices , in which two novel options present themselves simultaneously , still needs to be investigated . One may speculate that the brain would start by defining a default option , and then proceed to the comparison as usual . If this is correct , identifying the trial-wise and/or subject-wise default policy might be essential for understanding how the brain makes value-based decisions . However , we only have a proof of concept here , the generality of the ‘default vs . alternative framing’ remains to be established . Further research is also required to specify the contribution of the different brain regions that are involved in the valuation and selection processes , notably the dACC . The present results suggest that the vmPFC provides a decision value , which is also represented in the ventral striatum . How such decision value is used by the brain to make a selection remains to be explained . The study was approved by the Pitié-Salpétrière Hospital ethics committee . All subjects were recruited via e-mail within an academic database and gave informed consent before participation in the study . They were right-handed , between 20 and 32 years old , with normal vision , no history of neurological or psychiatric disease , and no contra-indication to MRI ( pregnancy , claustrophobia , metallic implants ) . They were not informed during recruitment that they would win food items , music CD and magazines to avoid biasing the sample . In total , 24 subjects ( 23 . 8 ± 2 . 8 years old , 12 females ) were included in the fMRI experiment and paid a fixed amount ( 80€ ) for their participation . In the eye-tracking experiment , 24 right-handed subjects ( 24 ± 3 . 4 years old , 11 females ) were recruited following the same procedure with the same inclusion criteria . No statistical method was used to predetermine sample size , but our sample size is similar to those generally employed in the field . One subject was excluded due to a technical issue with the eye-tracking device . All tasks were programmed on a PC in MATLAB language , using the Psychophysics Toolbox extensions ( RRID:SCR_002881 , Brainard , 1997; Pelli , 1997 ) . Subjects performed the rating task outside the scanner and the choice task during fMRI scanning . Prior to each task , they were instructed and trained on short versions ( 24 trials ) to get familiarized with the range of items and the mode of response . During the rating task , subjects were asked to estimate the likeability of all 432 items that they could potentially obtain at the end of the experiment . These items were blocked by reward domain: food , music and magazines . Unbeknown to subjects , each reward domain was divided into 4 categories of 36 items . The 12 categories were: appetizers , biscuits , candies , chocolate ( food domain ) ; news , comics , cultural , generalist ( magazine domain ) ; French , jazz , rock , urban ( music domain ) . The order of presentation was randomized within each reward domain , such that the categories were intermingled . The series of trials consisted of displaying pictures of the items one by one on the computer screen . A pseudo-continuous rating scale ( 101 points ) was presented below the picture , with three reference graduations ( do not like at all , neutral , like a lot ) . Subjects could move a cursor along the scale by pressing a key with the right index finger to go left or another key with the right middle finger to go right . The cursor was initially positioned at the middle of the rating scale . The rating was self-paced and subjects had to press a button with the left hand to validate their response and proceed to the next trial . At the beginning of each block , the reward domain was announced on a black screen . Likeability ratings were used for pairing options in the choice task . For each domain , mean ratings were used to rank categories according to subjective preference . The most preferred categories ( ranked 1 and 2 ) were opposed to the least preferred ones ( ranked 3 and 4 ) , making a total of 4 oppositions ( 1–3 , 1–4 , 2–3 , 2–4 ) . To generate the series of choices for each opposition , items were sorted in the order of likeability rating . Half the choices varied the difference between ratings while keeping the average constant ( item ranked mean+X was paired with item ranked mean-X ) ; the other half varied the average while keeping the difference minimal ( item ranked X was paired with item ranked X-1 ) . Thus , the mean value and relative value of choice options were orthogonalized . A total of 36 choices were generated for each inter-categorical opposition , and presented in a randomized order . The 36 choices were divided into 4 blocks of 9 trials , presented in 4 different fMRI sessions . As there were 12 possible oppositions ( 4 per domain ) , this makes a total of 432 trials , meaning that each item being presented twice . At the beginning of each block , the domain was announced on a black screen for 0 . 5 s , then the two opposed categories were displayed for 2 to 5 s , followed by a 0 . 5 s fixation cross . Each block was composed of a series of 9 choices . Choice trials started with the display of the two options side by side . The side of a given category as well as the best rated option was counter-balanced across trials . Subjects were asked to indicate their preference by pressing one of two buttons , with their left or right index finger , corresponding to the left and right options . The chosen picture was framed with a white square for 0 . 5 s , followed by a black screen with fixation cross lasting for 0 . 5 to 6 s . Importantly , subjects were not asked to generate a prior preference at the beginning of blocks , when the categories are revealed . They were only told that contextual information would be given before each series of choices , and that it would not require any response from their part . They also knew that at the end of the experiment , one trial per domain would be randomly selected and that they would be given the options chosen in these trials . Following the scanning session , subjects had to complete a debriefing task in which they were presented the opposed categories two by two . They were asked to first select the category that they preferred and then to rate their confidence in their choice using an analog scale . Finally , they spent an additional 20 min in the lab to eat the food item , listen to the music album and read the magazine that they just won . All analyses were performed with Matlab Statistical Toolbox ( Matlab R20013b , The MathWorks , Inc . , USA ) . Two dependent variables were recorded: choice ( which option was selected ) and response time ( between option onset and button press ) . The influence of likeability ratings on these variables was assessed with logistic or linear regression models , as explained in the results section . Regression estimates were computed at the individual level and tested for significance at the group level using one-sample two-tailed t-test . Correlations between variables of interest were also computed at the individual level , using Pearson’s coefficient , and similarly tested at the group level . In the eye-tracking experiment , gaze position was recorded with a 60 Hz sampling frequency using The Eye Tribe device ( http://theeyetribe . com ) , during each block of the choice task . There was no constraint on the head , subjects were simply asked to avoid head movement . A screen providing feedback on the eye position was inserted in the trial sequence every time gaze was lost . The number of excluded trials due to loss of gaze position varied between 0 and 6 , depending on subjects . Fixation duration was computed for each trial and option , as the time during which gaze position was inside a square window delineated the corresponding picture on the screen . Last fixation was defined as the picture being looked at when the choice was made . The proportion of fixation was calculated as the number of trials in which gaze position was on the corresponding picture at a given time point . Note that these proportions for the two options do not add up to one because the gaze position can be outside the two windows . We used the EZ2 analytical approximation for the drift diffusion model ( Wagenmakers et al . , 2007; Grasman et al . , 2009 ) to account for the probability of choosing the default option and the response time , on a trial-by-trial basis . As proposed by ( Ratcliff , 1978; Wagenmakers et al . , 2007; Grasman et al . , 2009 ) , we defined the probability of choosing the default option as:P ( def ) =φ ( −A , S−A ) e2μAσ²−1 As proposed by EZ2 ( Grasman et al . , 2009 ) , we defined the corresponding RT as:RT ( def chosen ) =Tnd+ ( A−S ) * ( φ ( S , A ) +φ ( 0 , A−S ) +2Aφ ( A−S , 0 ) ) −μφ ( A−S , A ) φ ( −A , 0 ) With φ ( x , y ) =e−2μyσ2−e−2μxσ2 , A the amplitude between boundaries , S the starting point , μ the mean of the drift rate , σ the standard deviation of the drift rate and Tnd the non-decision time . To compute response time in trials where the alternative option is chosen , we replace ( μ , S ) by ( −μ , A−S ) . The free parameters A , Tnd , σ , μ and S were estimated with the behavioral data . Different versions of the model were compared to account for the changes in choice and RT patterns that were induced across blocks by the variations in prior preference . In all cases , A , Tnd and σ , were estimated for each individual but constant across blocks . In the null model , μ was proportional to the decision value ( difference in likeability rating between options , DVIT , such that μ=αDVIT ) and S was set to zero . The model space ( see details in the results section ) explored the possibilities that μ and S could differ from their initial setting ( μ=αDVIT + β / S=z ) , vary across blocks ( 12 free α for μ/12 free z for S ) , or be informed by the prior preference ( difference in mean likeability rating between categories , DVCAT , such that μ=αDVIT + βDVCAT / S= βDVCAT ) . In the attentional versions of the model , μ was also informed by gaze fixations , as follows:μ= ( VIT ( def ) −θVIT ( alt ) ) ∗Ddef− ( VIT ( alt ) −θVIT ( def ) ) ∗DaltDdef+Dalt With Ddef and Dalt the total durations of fixation for the default and the alternative options during the considered trial , and θ the weight discounting the value of the unfixated item relative to the fixated one ( Krajbich et al . , 2010 ) . All versions of the drift diffusion model were fitted separately for each individual to choices and RTs using Matlab VBA-toolbox ( available at http://mbb-team . github . io/VBA-toolbox/ ) , which implements Variational Bayesian analysis under the Laplace approximation ( Daunizeau et al . , 2014 ) . This iterative algorithm provides a free-energy approximation for the model evidence , which represents a natural trade-off between model accuracy ( goodness of fit ) and complexity ( degrees of freedom ) ( Friston et al . , 2007; Penny , 2012 ) . Additionally the algorithm provides an estimate of the posterior density over the model free parameters , starting with Gaussian priors . Individual log model evidences were then taken to group-level random-effect Bayesian model selection ( BMS ) procedure ( Penny et al . , 2010 ) . BMS provide an exceedance probability ( xp ) that measures how likely it is that a given model ( or family of models ) is more frequently implemented , relative to all the others considered in the model space , in the population from which participants were drawn ( Rigoux et al . , 2014; Stephan et al . , 2009 ) . Functional echo-planar images ( EPIs ) were acquired with a T2*-weighted contrast on a 3 T magnetic resonance scanner ( Siemens Trio ) . Interleaved 2 mm slices separated by a 1 . 5 mm gap and oriented along a 30° tilted plane were acquired to cover the whole brain with a repetition time of 2 . 01 s . The first five scans were discarded to allow for equilibration effects . All analyses were performed using statistical parametric mapping ( SPM8 , RRID:SCR_007037 ) environment ( Wellcome Trust Center for NeuroImaging , London , UK ) . Structural T1-weighted images were coregistered with the mean EPI , segmented , and normalized to the standard Montreal Neurological Institute ( MNI ) T1 template . Normalized T1-images were averaged across subjects to localize group-level functional activations by superimposition . During preprocessing , EPIs were spatially realigned , normalized ( using the same transformation as for structural images ) , and smoothed with an 8 mm full-width at half-maximum Gaussian kernel . We used four general linear models ( GLMs ) to explain pre-processed time-series at the individual level . The first model ( GLM0 ) was built for whole-brain search of voxels encoding prior preference in baseline activity . It was composed of a finite impulse response function ( FIR ) that included seven time points per trial , from one TR ( −2 . 01 s ) before to five TR ( 10 . 05 s ) after choice onset . The different blocks were modeled in separate regressors , each being parametrically modulated by the two option values ( default and alternative ) . For each time point , we computed a contrast that weighted all trials of a given block by the corresponding prior preference ( DVCAT ) . Four subjects were excluded from this analysis because they presented at least one block without a sufficient variance to estimate the parametric regression coefficients . The second model ( GLM1 ) included a stick function capturing option display ( only one event per trial ) , parametrically modulated by the two option values ( chosen and unchosen ) . The three regressors were convolved with a canonical hemodynamic response function . The third model ( GLM2 ) included two categorical regressors: a boxcar function over blocks and the same stick function as in GLM1 . The boxcar function was parametrically modulated by DVCAT , to account for tonic effects of prior preference . The stick function was parametrically modulated by five variables: chosen option ( default or alternative ) , VIT ( def ) when default chosen , VIT ( alt ) when default chosen , VIT ( def ) when default unchosen and VIT ( alt ) when default unchosen . This allowed computing orthogonal contrasts for the decision value in the pre-choice ( default vs . alternative ) and post-choice ( chosen vs . unchosen ) frames . The regressors were convolved with a canonical HRF to localize brain regions where the evoked response reflected the decision value . In a subsequent analysis the same regressors were convolved with the same FIR as used for GLM0 , in order to examine the dynamics of value coding in regions of interest ( ROI ) . The fourth model ( GLM3 ) was equivalent to GLM1 except that the stick function modeling option display was modulated by the sum and difference of option values , in the pre-choice frame ( default vs . alternative ) . Common variance between the two parametric regressors was removed such that they could explain a unique variance in the BOLD signal . Motion artifacts were corrected in all GLMs by adding subject-specific realignment parameters as covariates of no interest . Regression coefficients were estimated at the individual level and then taken to group-level random-effect analysis using one-sample two-tailed t-test . In ROI analyses they were extracted from spheres of 6 mm radius positioned on coordinates defined independently from the present dataset: for the vmPFC we took the peak coordinates [−2 40 –8] from a meta-analysis of value representation ( Bartra et al . , 2013 ) , and for the dACC we took the peak coordinate [−6 24 34] of a negative correlation with chosen option value ( Boorman et al . , 2013 ) . Four variants of GLM2 were also compared to better characterize value coding in the two ROI . The four regressors modeling option values were replaced by a single regressor: ( 1 ) default option value , ( 2 ) pre-choice decision value ( default minus alternative ) , ( 3 ) chosen option value , ( 4 ) post-choice decision value ( chosen minus unchosen ) . This was meant to assess whether value representation concerned only one option or the difference between the two , and whether it was expressed in a pre-choice or post-choice frame . All models were fitted to individual time-series extracted from vmPFC and dACC ROI , so as to compute group-level exceedance probabilities , following a BMS procedure similar to that used for behavioral data analysis .

If you had the choice of listening to a piece of music by either the singer Céline Dion or jazz pianist Keith Jarrett , which would you pick ? When choosing between two mutually exclusive options , the brain first assigns a value to each . An area called the ventromedial prefrontal cortex ( vmPFC ) compares these two values and calculates the difference between them . The vmPFC then relays this difference to other brain regions that trigger the movements required to obtain the selected option . But what exactly is the vmPFC comparing ? A reasonable assumption is that we approach the decision with an existing preference for one of the options based on our previous experience . Lopez-Persem et al . set out to determine whether and how the vmPFC uses this existing preference – for example , for pop music over jazz – to drive the decision-making process . For the experiments , volunteers were asked to rate how much they liked individual musicians spanning a range of different genres . While lying inside a brain scanner , the subjects then had to choose their favorite from pairs of musicians selected from the list . When making such decisions , volunteers must consider both the overall category ( do I prefer jazz or pop ? ) but also the individual examples ( a pop music fan might choose jazz if the pop option is Britney Spears ) . Lopez-Persem et al . found that the volunteer’s decisions were biased towards their prior preference . Pop music fans chose Céline Dion or Britney Spears more often than would be expected based on the likability ratings they had given the individual artists in the study . Brain imaging revealed that the vmPFC represents choices as ‘default minus alternative’ , where the default is any member of the previously preferred category ( e . g . any pop artist for a pop music fan ) and the alternative is from a different category ( e . g . a jazz artist ) . Baseline vmPFC activity is higher for members of the preferred category , giving these options a head start over the alternatives . Asking volunteers to choose between other types of objects , including food and magazines , produced similar results . The brain thus uses a general strategy for decision-making that saves time and effort , but which also introduces bias . The next step is to work out how downstream brain regions use the vmPFC signal to select the preferred option .

2016

How prior preferences determine decision-making frames and biases in the human brain

Human bromodomain and extra-terminal domain ( BET ) family members are promising targets for therapy of cancer and immunoinflammatory diseases , but their mechanisms of action and functional redundancies are poorly understood . Bdf1/2 , yeast homologues of the human BET factors , were previously proposed to target transcription factor TFIID to acetylated histone H4 , analogous to bromodomains that are present within the largest subunit of metazoan TFIID . We investigated the genome-wide roles of Bdf1/2 and found that their important contributions to transcription extend beyond TFIID function as transcription of many genes is more sensitive to Bdf1/2 than to TFIID depletion . Bdf1/2 co-occupy the majority of yeast promoters and affect preinitiation complex formation through recruitment of TFIID , Mediator , and basal transcription factors to chromatin . Surprisingly , we discovered that hypersensitivity of genes to Bdf1/2 depletion results from combined defects in transcription initiation and early elongation , a striking functional similarity to human BET proteins , most notably Brd4 . Our results establish Bdf1/2 as critical for yeast transcription and provide important mechanistic insights into the function of BET proteins in all eukaryotes . Bromodomains ( BDs ) are reader modules that allow protein targeting to chromatin via interactions with acetylated histone tails . BD-containing factors are usually involved in gene transcription , and their deregulation has been implicated in a spectrum of cancers and immunoinflammatory and neurological conditions ( Fujisawa and Filippakopoulos , 2017; Wang et al . , 2021 ) . The bromodomain and extra-terminal domain ( BET ) family is characterized by the presence of a double BD , which has the highest affinity towards hyperacetylated histone H4 , and an ET domain that interacts with non-histone proteins ( Rahman et al . , 2011; Slaughter et al . , 2021 ) . BET BD inhibitors have shown promising results in treating both blood cancers and solid tumors ( Stathis and Bertoni , 2018 ) . Early studies implicated mammalian BET proteins in regulation of selected lineage-specific genes ( Delmore et al . , 2011; Lovén et al . , 2013 ) ; however , it has become clear that their regulatory roles extend to the majority of genes transcribed by RNA polymerase II ( Pol II ) ( Muhar et al . , 2018; Winter et al . , 2017 ) . While most mammalian tissues express three BET factors ( Brd2 , Brd3 , Brd4 ) ( Uhlén et al . , 2015 ) , many studies have implicated Brd4 as most important for widespread changes in transcription after BET inactivation ( Muhar et al . , 2018; Zheng et al . , 2021; Zuber et al . , 2011 ) . These broad genome-wide expression defects are due to the important roles of Brd4 in both transcription initiation and elongation . Brd4 contributes to recruitment of the transcription coactivator Mediator to enhancers and promoters and cooperates with Mediator in forming nuclear condensates at actively transcribed regions ( Bhagwat et al . , 2016; Han et al . , 2020; Sabari et al . , 2018 ) . Brd4 is also important for productive transcription elongation , but it is unclear what mechanisms are involved . Brd4 associates with the positive transcription elongation factor b ( P-TEFb ) , but no global defects in P-TEFb recruitment to chromatin were observed following inactivation of Brd4 or all BET factors ( Muhar et al . , 2018; Winter et al . , 2017 ) . Proposed alternative mechanisms of elongation control include release of P-TEFb inhibition , histone chaperone activity , or direct kinase activity ( Devaiah et al . , 2012; Itzen et al . , 2014; Kanno et al . , 2014 ) . Finally , Brd4 was shown to affect preinitiation complex ( PIC ) formation at promoters , although its roles in later stages of transcription are believed to be dominant ( Kanno et al . , 2014; Winter et al . , 2017 ) . TFIID is a conserved Pol II factor that , in yeast , regulates transcription of almost all Pol II-transcribed genes ( Donczew and Hahn , 2018; Huisinga and Pugh , 2004; Warfield et al . , 2017 ) . TFIID comprises TBP ( TATA binding protein ) and 13–14 Tafs ( TBP-associated factors ) . TFIID acts as a TBP-DNA loading factor , a promoter recognition factor and a molecular scaffold guiding PIC assembly ( Patel et al . , 2020 ) . Recruitment of metazoan TFIID to promoters is thought to be aided by interactions between two Taf subunits and chromatin marks that are enriched at the +1 nucleosome . The human Taf1 double BD and Taf3 PHD domain recognize acetylated histone H4 tails and trimethylated histone H3 lysine 4 , respectively ( Jacobson et al . , 2000; van Ingen et al . , 2008 ) . In contrast , Taf1 and Taf3 in budding yeast are missing both of these chromatin readers as well as recognizable promoter DNA sequence motifs such as the INR and DPE that are known to interact with metazoan Tafs ( Patel et al . , 2020 ) . Yeast BET family member Bdf1 and its paralogue Bdf2 are genetically redundant , with at least one required for viability . Deletion of bdf1 but not bdf2 results in growth retardation , suggesting a dominant role for Bdf1 ( Matangkasombut et al . , 2000 ) . Bdf1 binds Taf7 and recognizes acetylated histone H4 . Both of these functions were found to support cell growth and transcription at several loci . Based on these findings , it was proposed that Bdf1 double BD substitutes for the missing BDs of yeast Taf1 , linking yeast TFIID to promoter-enriched chromatin marks ( Matangkasombut et al . , 2000; Matangkasombut and Buratowski , 2003 ) . Gene deletion or targeted depletion of Bdf1 revealed a small contribution of Bdf1 to Taf1 recruitment genome-wide with a bias towards a group of genes classified as ‘TFIID-dominated’ and led to a modest defect in transcription at a limited gene set ( Durant and Pugh , 2007; Joo et al . , 2017; Ladurner et al . , 2003 ) . By applying degron-mediated protein depletion and monitoring nascent mRNA levels , we recently showed that yeast protein-coding genes can be classified into two broad categories based on transcription changes after rapid depletion of the coactivators TFIID and SAGA ( Donczew et al . , 2020 ) . We found that the majority of yeast genes are strictly TFIID-dependent while a subset , which we termed coactivator-redundant ( CR ) genes , are co-regulated by TFIID and SAGA . This latter category overlaps ~50% with the gene set earlier termed ‘SAGA-dominated’ ( Huisinga and Pugh , 2004 ) . In contrast , long-term ablation of SAGA via gene deletions showed that nearly all yeast genes are dependent on chromatin modifications directed by SAGA . However , the molecular basis for the TFIID and CR gene classes remains unknown . For example , it is not known if there are other factors that preferentially participate in the TFIID or SAGA-directed pathways . Here , we have examined the functions of Bdf1 and Bdf2 in yeast transcription and found broad genome-wide roles for the Bdfs in both transcription initiation and elongation . Rapid Bdf depletion strongly decreases transcription from the TFIID-dependent genes , and , at many genes , the Bdfs are more important than Tafs for normal levels of transcription . The Bdfs contribute to PIC formation and TFIID and Mediator recruitment to gene regulatory regions . In addition , at many genes , the Bdfs also regulate transcription elongation and this role contributes to the strong defects in mRNA synthesis upon Bdf depletion . These striking functional similarities with mammalian BET factors suggest broad conservation of BET function in eukaryotes . A limitation of previous studies on Bdf1/2 function was that both proteins could not be simultaneously eliminated since at least one Bdf is required for viability . In this study , we used the auxin-degron system ( Nishimura et al . , 2009 ) to achieve rapid depletion of Bdf1 , Bdf2 , or simultaneous depletion of both proteins , and used 4-thioU RNA-seq to monitor changes in newly synthesized mRNA ( Donczew et al . , 2020; Rabani et al . , 2011 ) . We first assessed protein degradation after a 30 min treatment with the auxin indole-3-acetic acid ( IAA ) ( Figure 1—figure supplement 1A ) . Bdf1 was efficiently degraded with <10% of protein remaining in IAA-treated samples . Bdf2 degradation was less complete , with ~15% of protein remaining , so we prepared a bdf2 deletion strain combined with a Bdf1 degron ( Bdf1 , Δbdf2 ) . Growth of all strains with Bdf derivatives was similar to wild type ( Figure 1—figure supplement 1B ) , and analysis of 4-thioU mRNA levels in control dimethyl sulfoxide ( DMSO ) -treated cultures showed that changes in transcription due to these genetic alterations were minimal ( Figure 1—figure supplement 1C ) . Experiments were done in two or three biological replicates , and the variation in nascent mRNA levels between replicate samples was <30% for nearly all experiments ( Figure 1—figure supplement 1D and Supplementary file 1 ) . Bdf2 degradation or bdf2 deletion showed minimal changes in transcription while Bdf1 degradation resulted in modest defects , in agreement with the genetic phenotypes of BDF1/2 mutants ( Figure 1A , Figure 1—figure supplement 1E , and Supplementary file 1 ) . Strikingly , depletion of both Bdf1 and Bdf2 , either through a double degron or by combining a bdf2 deletion with a Bdf1 degron , resulted in a global collapse of transcription with median decreases of ~4-fold and ~5 . 4-fold , respectively ( Figure 1A , Figure 1—figure supplement 1F , and Supplementary file 1 ) . Transcriptional changes in both Bdf1/2 depletion experiments are highly correlated ( r = 0 . 95 ) ( Figure 1—figure supplement 1G ) . The results of the Bdf1 and Bdf1/2 depletion experiments also correlate ( r = 0 . 82 ) , showing redundant roles of Bdf1 and Bdf2 in transcription ( Figure 1—figure supplement 1H ) . We compared results of the Bdf depletion experiment with previously published results of rapid Taf1 depletion ( Donczew et al . , 2020 ) . Surprisingly , we observed that transcription changes upon Bdf or Taf1 depletion correlate but the relationship between them is not linear ( Figure 1B ) . Labeling the points based on the TFIID and CR gene categories revealed that genes less sensitive to Bdf depletion ( on the right of the diagonal line ) include almost all CR genes and one-third of TFIID-dependent genes . Interestingly , most of the TFIID-dependent genes ( 68% ) are more sensitive to Bdf than Taf1 depletion and , for 12% of them , the difference is fourfold or higher ( Supplementary file 1 ) . Importantly , Bdf1 and Taf1 are degraded with similar efficiency ( Figure 1—figure supplement 2A ) . Our results show that , while Bdfs likely act in conjunction with TFIID at many promoters , Bdfs also have important TFIID-independent contributions to transcription of many genes . Comparison of the Bdf results with depletion of Taf13 , another TFIID subunit , leads to similar conclusions ( Figure 1—figure supplement 2A , B ) . Ribosomal protein ( RP ) genes are frequently analyzed as a separate gene category due to high expression levels , common regulatory mechanisms , and important roles in cell growth and stress response ( Zencir et al . , 2020 ) . About 95% of RP genes belong to the TFIID-dependent class ( Donczew et al . , 2020; Huisinga and Pugh , 2004 ) . Surprisingly , we found that nearly all RP genes are significantly less dependent on Bdf1/2 compared with TFIID , an uncommon feature for TFIID-dependent genes ( Figure 1—figure supplement 2C ) . With many TFIID-dependent genes losing >85% of detectable transcription after Bdf1/2 depletion , we investigated how these changes compare to defects caused by depletion of basal components of the PIC . We depleted basal factors with degron tags on either TFIIA subunit Toa1 or the TFIIH translocase subunit Ssl2 followed by 4-thioU RNA-seq . The results are summarized in a boxplot divided into CR and TFIID-dependent genes ( Figure 1C and Supplementary file 1 ) . Depletion of Toa1 or Ssl2 decreases all mRNA transcription independent of gene class with median changes of ~15-fold and ~23-fold , respectively . Interestingly , 38 and 19% of TFIID-dependent genes are similarly sensitive to Toa1 or Ssl2 depletion as they are to Bdf depletion . Combined , our results show that Bdf dependence is as good or even a better classifier for the TFIID and CR gene classes than Taf-dependence . Transcription at many TFIID-dependent genes is severely compromised in the absence of Bdfs while transcription of CR genes is only weakly affected ( Figure 1C ) . Bdf1 is also a subunit of the SWR1 complex that incorporates the histone variant H2A . Z into promoter-proximal nucleosomes ( Krogan et al . , 2003; Zhang et al . , 2005 ) . We used the degron system to rapidly deplete H2A . Z and check if the loss of transcription after Bdf depletion is mediated by defects in H2A . Z deposition ( Figure 1—figure supplement 2D ) . RNA-seq analysis did not reveal significant defects in gene transcription following the rapid loss of H2A . Z ( Figure 1—figure supplement 2E and Supplementary file 1 ) , suggesting that the global transcriptional defects upon Bdf depletion are not mediated via SWR1 and H2A . Z . Bdf1 was initially proposed to constitute a missing part of yeast Taf1 ( Matangkasombut et al . , 2000 ) , but later work classified Bdf1/2 as members of the BET family based on a conserved domain organization ( Wu and Chiang , 2007; Figure 1—figure supplement 3A ) . Since Bdfs have important TFIID-independent functions at many genes , we explored similarities between yeast and human BET proteins . Multiple sequence alignment and phylogenetic analysis of individual BDs of Bdf1 , Brd4 , and human Taf1 ( hTaf1 ) revealed that Bdf1 BDs , especially BD2 , are more closely related to Brd4 than hTaf1 BDs ( Figure 1D and Figure 1—figure supplement 3B ) . Other conserved domains of Bdf1 and Brd4 show ≥50% similarity ( Figure 1E ) . We next compared gene-specific responses to Bdf/BET depletion in yeast and human cells . We used a published RNA-seq dataset from an experiment where BET factors were targeted with the chemical degrader dBET6 in the MOLT4 leukemia cell line ( Winter et al . , 2017; Supplementary file 1 ) . Since the CR/TFIID-dependent classes are biased for TATA-containing/TATA-less categories ( Donczew et al . , 2020; Huisinga and Pugh , 2004 ) , we divided yeast and human genes into two categories based on the presence or absence of a consensus TATA box . We found that TATA-less genes in both systems were significantly more affected by Bdf/BET depletion than TATA-containing genes ( Figure 1F ) . Finally , in the 25% most affected genes in each dataset , we observed an extensive overlap in major gene ontology terms ( Figure 1—figure supplement 3C and Supplementary file 2 ) . Altogether , these results suggest a common biological role of BET factors in yeast and human cells , a conclusion supported by other results presented below . We recently mapped Taf1 and other TFIID subunits using an improved ChEC-seq method , where DNA cleavage by protein-MNase fusions is applied to map genome-wide binding ( Donczew et al . , 2021; Donczew et al . , 2020 ) . TFIID is detectable at over 3000 yeast promoters and is found at both TFIID-dependent and CR genes , which agrees with its general role in transcriptional regulation . We used ChEC-seq to map Bdf1 and Bdf2 binding genome-wide . Comparison of Bdf1/2 and Taf1 ChEC-seq showed similar binding patterns with broad peak spanning between −1 and +1 nucleosomes and signals extending into the gene body , especially for Bdf1 ( Figure 2A and Figure 2—figure supplement 1A ) . We identified 3700 , 3466 , and 3103 promoters bound by Bdf1 , Bdf2 , and Taf1 , respectively . Data for Bdf1 and Bdf2 confirm redundancy between both factors . There is an extensive overlap of promoters bound by both Bdfs and the signals at common bound promoters are highly correlated , although Bdf1 signals are consistently stronger , in agreement with its dominant role ( Figure 2B , Figure 2—figure supplement 1B , and Supplementary file 3 ) . There is also extensive overlap between Bdf1 and Taf1 bound promoters ( Figure 2B ) , although Bdf1 is found less frequently at CR genes ( Figure 2C ) , consistent with the lesser role of Bdfs at this gene class . Interestingly , the Bdf1 chromatin occupancy is only weakly predictive of the gene dependence on Bdf1/2 , as was similarly reported for human BET factors ( Figure 2—figure supplement 1C; Muhar et al . , 2018; Winter et al . , 2017 ) . We validated Bdf1 binding using ChIP-seq as an orthogonal approach . The results are similar to ChEC-seq as well as to previously published ChIP-exo data ( Figure 2—figure supplement 1D and Supplementary file 3; Rhee and Pugh , 2012 ) . The number of bound promoters detected by ChIP-seq is lower due to worse signal to background ratio but shows a good overlap with ChEC-seq results ( Figure 2—figure supplement 1E ) . We also performed a Bdf2 ChIP-seq experiment , but the quality of the data was too low to call peaks . Instead , we calculated Bdf2 signal at promoters bound by Bdf1 and found that the signals correlate well , supporting conclusions obtained by ChEC-seq ( Figure 2—figure supplement 1B , F and Supplementary file 3 ) . However , a major difference between ChEC-seq and ChIP-seq was observed at the RP genes . We detected Bdf1 binding at 15% of RP promoters using ChEC-seq , which is consistent with a limited role of Bdfs at RP genes . Conversely , almost all ( 94% ) RP promoters have a significant Bdf1 ChIP signal , but this result is not predictive of Bdf1 function ( Figure 1—figure supplement 2C and Figure 2—figure supplement 1G ) . Next , we investigated whether Bdf1 binding correlates to histone H4 acetylation levels . We measured the genome-wide acetylation of H4 lysine 12 ( H4K12ac ) , which was shown to be a preferred target for Bdf1 ( Slaughter et al . , 2021; Figure 2—figure supplement 1H ) . Importantly , for both classes of genes , the normalized H4K12ac level at promoters significantly differs depending on the presence of a Bdf1 peak ( Figure 2D and Supplementary file 3 ) . To conclude , at many promoters Bdf1 occupancy predicts higher H4K12ac levels and transcription dependence on Bdfs . It was earlier proposed that Bdf2 redistributes to preferred Bdf1 sites in a bdf1 deletion strain ( Durant and Pugh , 2007 ) . We tested this hypothesis by measuring changes in Bdf1 or Bdf2 occupancy after depleting Bdf2 or Bdf1 , respectively . Following Bdf1 depletion , the Bdf2 signal increases at TFIID promoters at the expense of CR promoters ( Figure 2E ) . The changes at individual promoters are relatively weak but they correlate well with gene dependence on Bdfs ( r = −0 . 52 ) ( Figure 2F and Supplementary file 3 ) . Conversely , Bdf1 does not redistribute in response to Bdf2 depletion ( Figure 2—figure supplement 1I and Supplementary file 3 ) . This result provides more evidence of redundancy between Bdf1 and Bdf2 , with Bdf1 being the dominant factor and Bdf2 serving an auxiliary role . Esa1 , a catalytic subunit of the NuA4 complex , is almost exclusively responsible for acetylation of lysines 5 , 8 , and 12 on histone H4 in yeast ( Chang and Pillus , 2009; Suka et al . , 2001 ) . We compared the roles of Esa1 and Bdfs in transcriptional regulation using an Esa1 degron strain . We used a 1 hr treatment with IAA , which allowed for a substantial loss of both H4K12ac and Esa1 ( Figure 3—figure supplement 1A ) . 4-thioU RNA-seq analysis revealed a global loss of transcription after Esa1 depletion , which has a good but nonlinear correlation with the Bdf experiment ( r = 0 . 73 ) ( Figure 3A , B , Figure 3—figure supplement 1B , and Supplementary file 1 ) . Importantly , Bdf depletion results in stronger transcriptional defects , suggesting that a portion of Bdf function is H4Ac independent . Our findings show that Bdf1/2 , Esa1 , and TFIID cooperate in regulation of many genes , but their contributions to transcription can be significantly different at individual genes ( Figure 3C , Figure 3—figure supplement 1C , and Supplementary file 1 ) . Bdf1 interactions with chromatin were previously shown to be modulated by Esa1 ( Durant and Pugh , 2007; Koerber et al . , 2009 ) . We used ChEC-seq to measure changes in Bdf1 occupancy following Esa1 depletion . In parallel , we measured changes in H4K12ac level using ChIP-seq . After 1 hr of IAA treatment , the H4K12ac chromatin signal decreased dramatically , but Bdf1 was still bound at substantial levels , especially at promoters ( Figure 3D , Figure 3—figure supplement 1D , and Supplementary file 3 ) . Nevertheless , the quantified changes in Bdf1 promoter signal and transcription following Esa1 depletion are correlated , suggesting that the effect of Esa1 on transcription is largely mediated by Bdf1 ( Figure 3E ) . Consistent with our results , Brd4 was also detected at human promoters following BD inhibition or mutation , which explains why BET degradation outperforms BET inhibition in preclinical cancer models ( Bauer et al . , 2021; Kanno et al . , 2014; Winter et al . , 2017 ) . It is possible that the residual binding of Bdfs in the absence of H4 acetylation is due to interactions with other histone acetylation marks . For example , Bdf1 was shown to have weak affinity for acetylated histone H3 ( Matangkasombut and Buratowski , 2003 ) . It has also been proposed that BET recruitment can be mediated by interactions with transcription factors ( TFs ) , coactivators , or chromatin remodelers , and it will be interesting to investigate if such mechanisms play a role in Bdf targeting ( Lambert et al . , 2019; Rahman et al . , 2011; Shen et al . , 2015 ) . Human BET factors are involved in recruitment of Mediator ( Bhagwat et al . , 2016 ) , and Bdf1/2 were proposed to assist in recruitment of TFIID to yeast promoters ( Matangkasombut et al . , 2000 ) . We used ChEC-seq to measure changes in chromatin occupancy of selected subunits of TFIID ( Taf1 , Taf11 ) , Mediator ( Med8 , Med17 ) , and SAGA ( Spt3 ) after Bdf depletion . We first defined promoters bound by each factor and restricted the downstream analysis only to these locations . Different factors mapped in this study had characteristic distributions between the two classes of yeast promoters . Bdf1/2 are enriched at TFIID-dependent promoters , TFIID does not exhibit preference against either class , and SAGA and Mediator are enriched at CR promoters ( Figure 4—figure supplement 1A and Supplementary file 3 ) . We observed modest ( ~1 . 7-fold ) decreases in TFIID and Mediator occupancy at all bound promoters following Bdf depletion ( Figure 4A , B , Figure 4—figure supplement 1B , C , and Supplementary file 3 ) . In contrast , SAGA binding was relatively insensitive to Bdf depletion with only small decreases observed , mostly at CR promoters . Importantly , changes measured for different subunits of TFIID or Mediator correlate ( Figure 4—figure supplement 1D , E and Supplementary file 3 ) . Using the ChIP-seq assay , similar results were observed for Bdf-dependent Taf1 binding ( Figure 4—figure supplement 1C , F and Supplementary file 3 ) . We next tested whether the decreases in TFIID and Mediator recruitment were due to cooperative interactions between these two factors ( Grünberg et al . , 2016; Knoll et al . , 2018 ) or via a more direct role of Bdfs in modulating TFIID and Mediator binding . We used ChEC-seq to measure changes in Taf1 and Med8 promoter signals after depleting Med14 and Taf13 , respectively . We confirmed modest binding cooperativity between both factors , but the decreases in TFIID and Mediator binding were significantly stronger after depleting Bdfs ( compare Figure 4A , B with C , D; Supplementary file 3 ) . This indicates a role of Bdfs in both TFIID and Mediator recruitment beyond the cooperative interactions of these two factors . Altogether , our results illustrate that Bdfs have independent contributions to the recruitment of TFIID and Mediator , although , in the absence of Bdfs , both factors are still recruited to promoters at substantial levels . Our results suggest that the strong transcription defects upon Bdf depletion are not solely mediated by defects in TFIID and Mediator binding . Since the strong decrease in transcription of TFIID-dependent genes could not be explained solely by decreases in TFIID or Mediator binding , we probed for defects in PIC assembly using ChIP-seq to quantitate promoter binding of the basal factor TFIIB . Depletion of either Bdfs or Taf13 caused a global loss of TFIIB , most pronounced at TFIID-dependent promoters ( Figure 5A , Figure 5—figure supplement 1 , and Supplementary file 3 ) . Comparing the defects in TFIIB binding due to Bdf depletion versus TFIID depletion , we found that Bdf depletion caused a significantly larger loss of TFIIB at TFIID-dependent promoters than does TFIID depletion ( Figure 5B and Supplementary file 3 ) . Conversely , at CR genes , we found that TFIID depletion caused a greater loss in TFIIB signal compared with Bdf depletion . These results are consistent with the above RNA-seq experiments . However , comparison of transcriptional and TFIIB binding defects caused by Bdf depletion shows that Bdfs have a broader role in transcription than solely regulating PIC assembly . This larger role of Bdfs is especially apparent for the most Bdf-dependent genes where transcription defects significantly exceed the defect in TFIIB binding ( Figure 5C ) . Combined , our results suggest an additional role of Bdfs in transcription that extend beyond organizing a platform for recruitment of PIC components . At many metazoan promoters , Pol II pauses after transcribing ~20–100 nucleotides , and release of paused Pol II , mediated by phosphorylation of NELF , DSIF , and the C-terminal repeat domain ( CTD ) Ser2 , is a critical step in gene regulation ( Core and Adelman , 2019 ) . Interestingly , Brd4 was shown to be involved in Ser2 phosphorylation ( Muhar et al . , 2018; Winter et al . , 2017 ) . While metazoan-like Pol II pausing does not occur at yeast promoters , Pol II stalling and shifts in Pol II distribution under specific conditions have been observed in both budding and fission yeast ( Badjatia et al . , 2021; Shetty et al . , 2017 ) . We tested whether Bdfs , like the mammalian BET factors , play a role in Pol II elongation . Pol II was quantitated using ChIP-seq for Rpb1 before and after depleting Bdfs and , as a comparison , Taf1 . We first calculated the change in Pol II occupancy along the whole transcribed region . We found that the change in transcription following Bdf depletion is not proportional to the loss of Pol II signal , even though the two results show a good correlation ( Figure 6A and Supplementary file 4 ) . As we observed for TFIIB , the decrease in Pol II signal at the most Bdf-dependent genes is less severe than the loss of transcription . Importantly , we did not detect such difference after depleting Taf1 ( Figure 6—figure supplement 1A and Supplementary file 4 ) . Next , we divided genes into quintiles based on Bdf dependence and plotted the change in Pol II occupancy up to 1 kb downstream from the transcription start site ( TSS ) ( Figure 6B ) . Except for the first quintile ( the least Bdf-dependent genes ) , Bdf depletion caused a similar loss of Pol II at TSSs of all genes . Interestingly , the Pol II loss increased further downstream of TSS at the most Bdf-dependent genes until stabilizing at ~400 bp downstream of TSS . Conversely , at the least Bdf-dependent genes , the Pol II signal partially recovered in the first ~400 bp downstream of TSS relative to the loss at TSS . We calculated the Pol II traveling ratio ( TR ) defined as the ratio of Rpb1 signal at 5′ versus the 3′ end of transcribed region ( Rahl et al . , 2010 ) . We observed a decrease of TR at the least Bdf-dependent genes and an increase of TR at the most dependent genes following Bdf depletion ( Figure 6C and Supplementary file 4 ) . Analysis of individual genes revealed shifts in Pol II distribution towards 3′ or 5′ ends of genes , respectively ( Figure 6—figure supplement 1B ) . We did not detect similar changes in Pol II distribution after depleting Taf1 ( Figure 6—figure supplement 2A , B and Supplementary file 4 ) . Importantly , the change in TR after depleting Bdfs correlates with the change in transcription at individual genes ( Figure 6D ) . There is also a modest correlation between the change in TR and the loss of TFIIB at promoters ( Figure 6—figure supplement 2C ) . Combined with our earlier findings , this suggests that at least a subset of the most Bdf-dependent genes experience transcription defects both at the initiation and elongation stages upon Bdf depletion . This explains the disproportionally large decrease in the amount of nascent RNA . On the other hand , more efficient elongation at the least Bdf-dependent genes seems to partially nullify the initiation defects , leading to little total change in transcription upon Bdf depletion ( Figure 1C ) . To further explore elongation defects , we performed ChIP-seq with antibodies specific to phosphorylated CTD residues Ser5P and Ser2P . Sequential deposition of these marks coordinates the assembly of multiple factors during transcription , and they are involved in successful promoter escape and elongation , respectively ( Harlen and Churchman , 2017 ) . We measured total Ser5P and Ser2P signals along transcribed regions , normalized them by corresponding Rpb1 signals , and calculated the change in CTD phosphorylation status following Bdf depletion ( Figure 6E and Supplementary file 4 ) . In agreement with the above findings that the most Bdf-dependent genes have a defect in elongation , Pol II becomes hypophosphorylated on Ser2 and , to a lesser extent , on Ser5 after depleting Bdfs . Conversely , at the least Bdf-dependent genes , Pol II becomes weakly hyperphosphorylated , in agreement with the suggestion that elongation becomes more efficient at this gene set upon Bdf depletion . Importantly , we did not observe similar changes in Pol II phosphorylation after Taf1 depletion , which validates that Bdfs have a TFIID-independent role in regulating transcriptional elongation ( Figure 6—figure supplement 2D and Supplementary file 4 ) . Bur1 and Ctk1 , homologues of metazoan Cdk9 and Cdk12/13 , are responsible for CTD Ser2P marks in yeast . We used ChIP-seq to measure Bur1 and Ctk1 levels along the transcribed regions and normalized them by corresponding Rpb1 signals . As a comparison , we did a similar analysis for the elongation factor Spt5 . Interestingly , the relative gene body occupancies of Bur1 and Ctk1 , but not of Spt5 , correlate with gene dependence on Bdfs ( Figure 6—figure supplement 3 ) . Next , we calculated the change in Bur1 and Ctk1 relative occupancy following Bdf depletion . We observed approximately twofold increases in occupancy of both kinases relative to Rpb1 at the least Bdf-dependent genes , a result that agrees with observed CTD hyperphosphorylation . Surprisingly , at the rest of the genes , the relative occupancy of both kinases further increased , correlating with gene dependence on Bdfs . In contrast , we observed a uniform , small ( ~1 . 4-fold ) increase in Spt5 occupancy , normalized to Rpb1 , at all tested genes ( Figure 6F and Supplementary file 4 ) . We divided genes into quintiles based on Bdf dependence and plotted the change in Bur1 and Ctk1 occupancy up to 1 kb downstream from TSS ( Figure 6G ) . Similarly as for Rpb1 , but showing the opposite trend , Bur1 and Ctk1 accumulate at the 5′ end of the most Bdf-dependent genes until stabilizing at ~400 bp downstream of TSS . These results illustrate that changes in the recruitment of Spt5 and CTD kinases cannot account for the function of Bdfs in transcription elongation . They also provide further evidence that the observed defect in elongation is restricted to the first ~400 bp in the transcribed region . After development of specific inhibitors targeting BET BDs , BET factors , especially Brd4 , emerged as key regulators of transcription and as promising targets in the therapy of cancer and immunoinflammatory diseases ( Fujisawa and Filippakopoulos , 2017; Wang et al . , 2021 ) . In this work , we investigated the roles of yeast BET family members Bdf1 and Bdf2 . We explored transcription dependence on Bdf1/2 , their genome-wide binding patterns , and their interplay with key components of the transcriptional machinery . We found that Bdf functions extend beyond solely regulating TFIID as expression from most genes is more sensitive to Bdf versus TFIID depletion . Our results establish the Bdfs as critical for normal expression of most yeast genes , with functions both during transcription initiation and elongation . Our work reveals that the roles of yeast BET factors are surprisingly similar to their human equivalents and suggests conserved mechanisms of BET function across eukaryotes . Bdf1 was proposed to act as part of yeast TFIID , substituting for the metazoan Taf1 BDs ( Matangkasombut et al . , 2000 ) . This function might be especially important in yeast as H4 acetylation is enriched at most promoters and yeast promoters lack known DNA sequence motifs that are bound by TFIID . However , we found that that the organization and amino acid sequence of Bdf1 is more similar to human BET factors than to the hTaf1 BDs . Our results showed that Bdfs regulate a similar set of genes as TFIID , but that there are important differences . Transcription from most genes defined earlier as CR is largely insensitive to Bdf depletion , while a large subset of TFIID-dependent genes is significantly more affected by Bdf versus TFIID depletion . Genome-wide mapping revealed that Bdf1/2 are found together with TFIID at many promoters and that they contribute to TFIID recruitment . However , a substantial fraction of TFIID is still bound to promoters in the absence of Bdfs , in agreement with the sub-stoichiometric association between Bdf1 and TFIID ( Sanders et al . , 2002 ) . Interestingly , human BET proteins were shown to associate with TFIID in a proteomic screen and both Brd4 and hTaf1 were proposed to have synergistic effects on gene expression and cell growth ( Lambert et al . , 2019; Sdelci et al . , 2016 ) . Finally , hTaf1 BDs have the same substrate specificity as BET BDs and both Brd4 and TFIID are bound to the majority of active promoters in human cells ( Bhagwat et al . , 2016; Lauberth et al . , 2013; Slaughter et al . , 2021 ) . Based on our findings and the presented evidence from human cells , we propose that cooperation between BET factors and TFIID is a conserved feature of eukaryotic gene regulation . The gain of BDs by metazoan TFIID may result from the increased complexity of gene expression programs and allow for BET-independent TFIID interactions with acetylated histone H4 . Brd4 interacts with Mediator and participates in Mediator recruitment to enhancers and promoters ( Bhagwat et al . , 2016; Wu et al . , 2003 ) . Our experiments revealed that Bdfs are important for normal levels of Mediator occupancy . A direct role of Bdfs in Mediator targeting to chromatin was surprising , and we considered the possibility that other factors are involved in that process . Mediator was proposed to be recruited cooperatively with TFIID ( Grünberg et al . , 2016; Johnson et al . , 2002; Knoll et al . , 2018 ) . We confirmed this modest cooperativity , but we found that the contribution of Bdfs to Mediator recruitment surpasses that of TFIID . There is strong evidence that TFIID and Mediator act as molecular scaffolds to organize the recruitment of other components of transcriptional machinery ( Allen and Taatjes , 2015; Patel et al . , 2020 ) . Human Mediator and Brd4 were also found to participate in the formation of dynamic nuclear condensates at the sites of active transcription ( Han et al . , 2020; Sabari et al . , 2018 ) . Finally , a recent study suggested cooperation of yeast TFIID and Mediator in restricting the diffusion of PIC components to shared subnuclear territories in order to facilitate gene transcription ( Nguyen et al . , 2020 ) . We propose that chromatin tethered yeast BET factors serve as a nucleation center for dynamic recruitment of Mediator and TFIID , which in turn create a platform for efficient PIC assembly . This role of Bdf1/2 seems widespread and likely complementary to the action of sequence-specific TFs ( Figure 7 ) . At the least Bdf-dependent genes , we observed a modest decrease in TFIID , Mediator , and TFIIB recruitment after Bdf depletion , which did not translate into appreciable defects in mRNA synthesis . Conversely , at the most Bdf-dependent genes , the loss of transcription exceeded the loss of Mediator , TFIID , TFIIB , and Pol II recruitment . The unexpected role of the yeast BET factors in transcription elongation provided an explanation for these seemingly conflicting results ( Figure 7 ) . At CR genes and at a small subset of TFIID-dependent genes , Bdf depletion results in a 3′ shift in Pol II distribution and increased CTD Ser2 phosphorylation . This suggests more efficient early elongation of these genes in the absence of Bdfs . In contrast , Bdf depletion at many TFIID-dependent genes causes Pol II accumulation at 5′ gene ends and a loss of CTD Ser2 phosphorylation . This change in Pol II profile is similar to the defects in pause release caused by BET depletion in human cells ( Muhar et al . , 2018; Winter et al . , 2017 ) . At these yeast genes , the combined initiation and elongation defects resulting from Bdf depletion lead to a large decrease in mRNA synthesis . Interestingly , it was recently shown that acute stress in Saccharomyces cerevisiae cells causes Pol II stalling at the +2 nucleosome , which overlaps with the region where we observe dominant changes in Pol II distribution ( Badjatia et al . , 2021 ) . CTD Ser2 residues in yeast are phosphorylated by kinases Bur1 and Ctk1 , homologues of metazoan Cdk9 and Cdk12/13 , respectively . Ctk1 is the dominant CTD kinase in yeast while in metazoans Cdk12/13 seems to be less important , although the interplay between Cdk9 and Cdk12/13 is poorly understood ( Harlen and Churchman , 2017 ) . At Bdf-independent genes , depletion of Bdfs results in a modest increase of Bur1 and Ctk1 occupancy normalized to Rpb1 level , and this agrees with a gain of CTD Ser2P marks at these genes . Surprisingly , at Bdf-dependent genes , the occupancy of both kinases relative to Rpb1 increased significantly in the first ~400 bp of transcribed region after Bdf depletion , even though CTD Ser2 phosphorylation decreases . Interestingly , depletion of Brd4 or all BET factors in human cells was shown to result in a small increase in Cdk9 and Cyclin T1 chromatin occupancy , even though it causes a decrease in CTD Ser2P marks and a defect in pause release ( Muhar et al . , 2018; Winter et al . , 2017 ) . It is not known if Brd4 affects Cdk12/13 recruitment . Several models have been proposed to explain the role of Brd4 in transcription elongation . Brd4 was shown to release Hexim1-mediated inhibition of Cdk9 in vitro , thus acting as a positive regulator of P-TEFb ( Itzen et al . , 2014 ) . It was also proposed that Brd2 and Brd4 have kinase activity and that Brd4 is directly connected to CTD Ser2 phosphorylation independently of P-TEFb ( Denis and Green , 1996; Devaiah et al . , 2012 ) . Bdf1 was also proposed to have an intrinsic kinase activity , which raises an intriguing prospect of its direct involvement in CTD phosphorylation that , if correct , would provide an explanation for our results ( Matangkasombut et al . , 2000 ) . Finally , it is possible that CTD hypophosphorylation at BET-dependent genes may result from defects in recruitment and/or reduction in function of other elongation factors . Since the role ( s ) of the BET factors in elongation are unresolved in both systems , it will be of great interest to determine whether the yeast and metazoan factors use similar mechanisms to regulate elongation . Redundancy between three ( Brd2 , Brd3 , Brd4 ) of the four mammalian BET factors was proposed based on their ubiquitous expression in most tissues , similar binding patterns , and because they are all essential for embryonic development ( Gyuris et al . , 2009; Khoueiry et al . , 2019; Uhlén et al . , 2015 ) . The extent of this redundancy is poorly understood because only a few context-specific examples of shared functions of BET factors have been reported ( Gilan et al . , 2020; Rahman et al . , 2011; Stonestrom et al . , 2015 ) . In most studies , the roles of BET factors in transcriptional regulation were attributed to Brd4 alone ( Muhar et al . , 2018; Zheng et al . , 2021; Zuber et al . , 2011 ) . Our data indicate that yeast BET proteins are functionally redundant under normal growth conditions , with Bdf1 being the dominant factor . The changes in transcription following depletion of Bdf1 or Bdf1/2 are highly correlated , while depletion of Bdf2 does not have significant consequences for the cell . In addition , Bdf1 and Bdf2 have nearly identical genome-wide binding patterns and Bdf2 redistributes from CR to TFIID-dependent genes upon Bdf1 depletion . Interestingly , it was reported that Bdf2 overexpression can suppress temperature- and salt-sensitive phenotypes in a bdf1 deletion strain , but , at the same time , Bdf1 was found to be a negative regulator of Bdf2 expression ( Fu et al . , 2013; Matangkasombut et al . , 2000 ) . This latter finding is also supported by our RNA-seq results where we observed approximately two-fold upregulation of BDF2 transcription following Bdf1 depletion . Importantly , a similar interplay between human BET factors in regulating expression of other family members was recently proposed ( Lambert et al . , 2019 ) . It remains to be investigated if and to what extent Brd2 and/or Brd3 can replace Brd4 in supporting transcription in human cells . Recent years have brought remarkable discoveries highlighting the essential nature of mammalian BET factors . Still , many aspects of their biology remain poorly understood , including details of the relationship with other components of the transcriptional machinery at enhancers and promoters and the extent of redundancies . We uncovered striking similarities between yeast and mammalian BET family members . Many of the most BET-sensitive genes in both systems lack a TATA element in their promoters and control highly similar cellular processes . BET proteins are also universally involved in the recruitment of Mediator to chromatin and modulation of transcriptional elongation . Considering the relative simplicity compared to metazoans , the yeast model system offers unique opportunities for a detailed investigation of different aspects of BET biology , and this work provides a foundation for these future studies . All S . cerevisiae and Schizosaccharomyces pombe strains used in this study are listed in Supplementary file 5 and Key resources table ( Donczew et al . , 2020; Baker Brachmann et al . , 1998; Warfield et al . , 2017; Chen et al . , 2008 ) . S . cerevisiae strains were grown in YPD medium ( 1% yeast extract , 2% peptone , 2% glucose , 20 μg/ml adenine sulfate ) at 30°C with shaking . S . pombe strains were grown in YE medium ( 0 . 5% yeast extract , 3% glucose ) at 30°C with shaking . In experiments involving degron depletion of target proteins , S . cerevisiae strains were treated with 500 μM IAA dissolved in DMSO or with DMSO alone for 30 min ( or 60 min for Esa1-degron strains ) to induce protein degradation followed by protocol-specific steps . In RNA-seq experiments , S . cerevisiae and S . pombe strains were grown to A600 ~1 . 0 . In ChEC-seq experiments , S . cerevisiae strains were grown to A600 ~0 . 6 . In ChIP-seq experiments , S . cerevisiae and S . pombe strains were grown to A600 ~0 . 8 . The number of biological replicates collected in each experiment is listed in Supplementary file 6 . S . cerevisiae strains ( Supplementary file 5 ) were constructed using standard methods . Proteins were chromosomally tagged by yeast transformation and homologous recombination of PCR-amplified DNA . Plasmid pFA6a-3V5-IAA7-KanMX6 ( Chan et al . , 2018 ) , or a derivative with the KanMX marker replaced by URA3 ( pSH1855 ) , was used as a template for generating the IAA7 degron tags . This C-terminal tag contains three copies of the V5 epitope tag followed by the IAA7 degron . For ChEC-seq experiments , proteins were tagged with 3xFLAG-MNase::TRP1 using pGZ110 ( Zentner et al . , 2015 ) . Plasmids p2L-3Flag-NAT and pFA6a-13Myc-Hyg ( a gift from Toshio Tsukiyama , Fred Hutch ) were used as templates to generate PCR fragments for tagging proteins with 3xFlag and 13xMyc epitope tags . A strain expressing free MNase under control of the native BDF2 promoter was constructed the following way . First , the MED8 promoter in pSG79 ( Grünberg et al . , 2016 ) was exchanged with the BDF2 promoter ( containing 500 bp upstream DNA from the BDF2 start codon ) . A XhoI/SacI fragment containing the pBDF2-MNase fusion DNA was inserted to the yeast integrating vector pRS303 ( Sikorski and Hieter , 1989 ) . The resulting pRD15 plasmid was linearized with BstEII and integrated into strain BY4705 . BDF2 deletion strain was constructed by replacing the BDF2 gene with the HPH marker amplified from plasmid pAG32 ( Goldstein and McCusker , 1999 ) . ESA1-containing vectors were constructed by inserting a XhoI/SacI fragment containing the ESA1 gene with its native promoter and terminator into vectors pRS316 ( pRD5 ) and pRS315 ( pRD21 ) . A plasmid shuffle strain was constructed containing WT ESA1 on plasmid pRD5 and a chromosomal deletion of ESA1 in strain SHY1036 . pRD21 was modified by replacing nucleotides 301–354 in the ESA1 ORF with the 3xV5-IAA7 fragment to create the Esa1-degron . Plasmid shuffling was used to replace pRD5 with pRD21 by counter selection on 5-FOA plates . 1 ml cell culture was collected and pelleted from strains after treatment with IAA or DMSO , washed with 500 μl water , then resuspended in 100 μl yeast whole cell extract buffer ( 60 mM Tris , 6 . 8 , 10% glycerol , 2% SDS , 5% 2-mercaptoethanol , 0 . 0025% bromophenol blue ) . After heating for 5 min at 95°C , samples were centrifuged for 5 min at 21K × g and analyzed by SDS-PAGE and western blot where protein signals were visualized by using the Odyssey CLx scanner and quantified using Odyssey Image Studio software ( Li-Cor ) by generating a standard curve using a titration from WT extract . Each protein analyzed was normalized to the amount of the TFIIF subunit Tfg2 . S . cerevisiae strains were streaked from glycerol stocks on YPD plates and incubated for 3 days at 30°C . Single colonies from freshly grown plates were used to start overnight cultures for the spot assay . After reaching saturation , cultures were diluted to A600 = 1 . 0 and 10× serial dilutions were prepared . 10 μl of appropriate dilution was spotted on YPD plates , and the plates were incubated for 2–3 days at 30°C to compare growth rates . All steps were done essentially as described with minor modifications ( Donczew et al . , 2020 ) . Two ( Bdf1 , Bdf2 , Δbdf2 , Bdf1Δbdf2 , H2A . Z ) or three ( Bdf1/2 , Toa1 , Ssl2 ) replicate samples were collected for each experiment . 10 ml S . cerevisiae or 20 ml S . pombe ( strain SHY1058 ) cultures were labeled with 5 mM 4-thiouracil ( 4-thioU ) ( Sigma-Aldrich #440736 ) for 4 min , the cells were pelleted at 3000 × g for 3 min , flash-frozen in liquid N2 , and then stored at −80°C for further use . S . cerevisiae and S . pombe cells were mixed in an 8:1 ratio and total RNA was extracted using reagents from RiboPure yeast kit ( Thermo Fisher Scientific #AM1926 ) using the following volumes: 480 μl lysis buffer , 48 μl 10% SDS , 480 μl phenol/CHCl3/isoamyl alcohol ( 25:24:1 ) per S . cerevisiae pellet +50 μl S . pombe cell solution ( from a single S . pombe pellet resuspended in 850 μl lysis buffer ) . Cells were lysed using 1 . 25 ml zirconia/silica beads ( RPI #9834 ) in a Mini Beadbeater-96 ( BioSpec Products ) for 5 min . Lysates were spun for 5 min at 16K × g , then the following volumes were combined in a 5 ml tube: 400 μl supernatant , 1400 μl binding buffer , 940 μl 100% ethanol . Samples were processed through Ambion filter cartridges until all sample was loaded , then washed with 700 μl Wash Solution 1 , and twice with 500 μl Wash Solution 2/3 . After a final spin to remove residual ethanol , RNA was eluted with 25 μl 95°C preheated Elution Solution . The elution step was repeated , and eluates combined . RNA was then treated with DNaseI using 5 μl DNaseI buffer and 4 μl DNaseI for 30 min at 37°C , then treated with Inactivation Reagent for 5 min at RT . Total RNA samples were stored at −20°C for up to 2 months or at −80°C for longer periods . RNA was biotinylated using 40 μl ( ~40 μg ) total RNA and 4 μg MTSEA biotin-XX ( Biotium #90066-1 ) in the following reaction: 40 μl total 4-thioU-labeled RNA , 20 mM HEPES , 1 mM EDTA , 4 μg MTSEA biotin-XX ( 80 μl 50 μg/ml diluted stock ) in a 400 μl final volume . Biotinylation reactions occurred for 30 min at RT with rotation and under foil . Unreacted MTS-biotin was removed by phenol/CHCl3/isoamyl alcohol ( 25:24:1 ) extraction . RNA was precipitated with isopropanol and resuspended in 100 μl nuclease-free H2O . Biotinylated RNA was purified using 80 μl MyOne Streptavidin C1 Dynabeads ( Thermo Fisher Scientific #65002 ) + 100 μl biotinylated RNA for 15 min at RT with rotation and under foil . Prior to use , MyOne Streptavidin beads were washed in a single batch with 3 × 3 ml H2O , 3 × 3 ml High Salt Wash Buffer ( 100 mM Tris , 7 . 4 , 10 mM EDTA , 1 M NaCl , 0 . 05% Tween-20 ) , blocked in 4 ml High Salt Wash Buffer containing 40 ng/μl glycogen ( Millipore Sigma #10901393001 ) for 1 hr at RT , then resuspended to the original volume in High Salt Wash Buffer . After incubation with biotinylated RNA , the beads were washed 3 × 0 . 8 ml High Salt Wash Buffer , then eluted into 25 μl streptavidin elution buffer ( 100 mM DTT , 20 mM HEPES , 7 . 4 , 1 mM EDTA , 100 mM NaCl , 0 . 05% Tween-20 ) at RT with shaking , then the elution step was repeated , and eluates were combined for a total of 50 μl . Samples were stored at −20°C until further processing . 10% input ( not biotinylated ) RNA ( 4 μl ) was diluted into 50 μl streptavidin elution buffer and processed the same as the biotinylated RNA samples to determine the extent of recovery . 50 μl of each input and biotinylated RNA was adjusted to 100 μl with nuclease-free water and purified on RNeasy columns ( Qiagen #74104 ) using the modified protocol . To each 100 μl sample , 350 μl RLT lysis buffer ( supplied by the Qiagen kit and supplemented with 10 μl 1% βME per 1 ml RLT ) and 250 μl 100% ethanol were added , mixed well , and applied to columns . Columns were washed with 500 μl RPE wash buffer ( supplied by the Qiagen kit and supplemented with 35 μl 1% βME per 500 μl RPE ) , followed by a final 5 min spin at 21K × g . RNAs were eluted into 14 μl nuclease-free water , and the RNA concentration was measured using Qubit HS RNA assay ( Thermo Fisher Scientific #Q32852 ) . Samples were stored at −20°C . One sample per batch prepared in a single day was tested for enrichment of labeled RNA by RT-qPCR , probing both unlabeled and labeled RNA from at least three transcribed genes as previously described ( Donczew et al . , 2020 ) . The purified 4-thioU labeled RNA contained 2–11% contamination of unlabeled RNA . Newly synthesized RNA isolated via 4-thioU labeling and purification was prepared for sequencing using the Ovation Universal RNA-seq Library Preparation Kit with S . cerevisiae AnyDeplete reagent ( Tecan #0364-A01 ) according to the manufacturer’s instructions and 50 ng input RNA . Libraries were sequenced on the Illumina HiSeq2500 platform using 25 bp paired-end reads at the Fred Hutchinson Genomics Shared Resources facility . ChEC-seq was performed as previously described ( Donczew et al . , 2021; Donczew et al . , 2020 ) . In experiments involving Bdf1 , Bdf2 , or Bdf1/2 depletion , six replicate samples were collected . In Esa1 depletion experiment , two replicate samples were collected , as well as in experiments mapping SAGA subunits ( Spt3 , Spt7 ) following Med14 or Taf13 depletion . In all other experiments , three replicate samples were collected . S . cerevisiae 50 ml cultures were pelleted at 2000 × g for 3 min . Cells were resuspended in 1 ml of Buffer A ( 15 mM Tris , 7 . 5 , 80 mM KCl , 0 . 1 mM EGTA , 0 . 2 mM spermine [Millipore Sigma #S3256] , 0 . 3 mM spermidine [Millipore Sigma #85558] , protease inhibitors [Millipore Sigma #04693159001] ) , transferred to a 1 . 5 ml tube , and pelleted at 1500 × g for 30 s . Cell were washed twice with 1 ml of Buffer A and finally resuspended in 570 μl of Buffer A . 30 μl 2% digitonin ( Millipore Sigma #300410 ) was added to a final concentration of 0 . 1% , and cells were permeabilized for 5 min at 30°C with shaking ( 900 rpm ) . 0 . 2 mM CaCl2 was added to the samples followed by incubation for another 5 min at 30°C . 100 μl cell suspension was mixed with 100 μl Stop Solution ( 400 mM NaCl , 20 mM EDTA , 4 mM EGTA ) . Stop Solution was supplemented with 5 μg MNase digested Drosophila melanogaster chromatin . Samples were incubated with 0 . 4 mg/ml Proteinase K ( Thermo Fisher Scientific #AM2548 ) for 30 min at 55°C , and the DNA was purified by phenol/CHCl3/isoamyl alcohol ( 25:24:1 ) extraction and ethanol precipitation . Pellets were resuspended in 30 μl 0 . 3 mg/ml RNase A ( Thermo Fisher Scientific #EN0531 ) ( 10 mM Tris , 7 . 5 , 1 mM EDTA , 0 . 3 mg/ml RNase A ) and incubated for 15 min at 37°C . 60 μl of Mag-Bind reagent ( Omega Biotek #M1378-01 ) was added , and the samples were incubated for 10 min at RT . Supernatants were transferred to a new tube , and the volume was adjusted to 200 μl ( 10 mM Tris , 8 . 0 , 100 mM NaCl ) . DNA was purified again by phenol/CHCl3/isoamyl alcohol ( 25:24:1 ) extraction and ethanol precipitation , and resuspended in 25 μl 10 mM Tris , 8 . 0 . ChIP-seq experiments were performed similarly as described ( Donczew et al . , 2020 ) . Two replicate samples were collected for all experiments except for Bur1 ChIP-seq , where three replicates were collected . 100 ml S . cerevisiae or S . pombe cultures were crosslinked with 1% formaldehyde ( Sigma-Aldrich #252549 ) for 20 min in the above growth conditions , followed by another 5 min treatment with 130 mM glycine . Cells were pelleted at 3000 × g for 5 min , washed with cold TBS buffer , pelleted at 2000 × g for 3 min , flash-frozen in liquid N2 , and then stored at −80°C for further use . Cell pellets were resuspended in 300 μl Breaking Buffer ( 100 mM Tris , 8 . 0 , 20% glycerol , protease inhibitors [Millipore Sigma #04693159001] ) . Cells were lysed using 0 . 4 ml zirconia/silica beads ( RPI #9834 ) in a Mini Beadbeater-96 ( BioSpec Products ) for 5 min . Lysates were spun at 21K × g for 2 min . Pellets were resuspended in 1 ml FA buffer ( 50 mM HEPES , 7 . 5 , 150 mM NaCl , 1 mM EDTA , 1% Triton X-100 , 0 . 1% sodium deoxycholate , protease inhibitors [Millipore Sigma #04693159001] ) and transferred to 15 ml polystyrene tubes . In experiments with antibodies specific against phosphorylated Rpb1 CTD , Breaking Buffer and FA buffer were supplemented with phosphatase inhibitors ( Thermo Fisher Scientific #A32957 ) . Samples were sonicated in a cold Bioruptor sonicator bath ( Diagenode #UCD-200 ) at a maximum output , cycling 30 s on , 30 s off , for a total of 45 min . Samples were spun twice in fresh tubes at 21K × g for 15 min . Prepared chromatin was flash-frozen in liquid N2 , and then stored at −80°C for further use . 20 μl of the chromatin sample was used to estimate DNA concentration . First , 20 μl Stop buffer ( 20 mM Tris , 8 . 0 , 100 mM NaCl , 20 mM EDTA , 1% SDS ) was added to samples followed by incubation at 70°C for 16–20 hr . Samples were digested with 0 . 5 mg/ml RNase A ( Thermo Fisher Scientific #EN0531 ) for 30 min at 55°C and 1 mg/ml Proteinase K for 90 min at 55°C . Sample volume was brought to 200 μl , and DNA was purified by two phenol/CHCl3/isoamyl alcohol ( 25:24:1 ) extractions and ethanol precipitation . DNA was resuspended in 20 μl 10 mM Tris , 8 . 0 , and the concentration was measured using Qubit HS DNA assay ( Thermo Fisher Scientific #Q32851 ) . 20 μl Protein G Dynabeads ( Thermo Fisher Scientific #10003D ) was used for a single immunoprecipitation . Beads were first washed three times with 500 μl PBST buffer ( PBS buffer supplemented with 0 . 1% Tween 20 ) for 3 min with gentle rotation . Beads were resuspended in a final volume of 20 μl containing PBST buffer and 5–8 μl of appropriate antibody ( Key resources table ) . The following antibody volumes were used: H4K12ac , FLAG-Tag , Myc-Tag – 5 μl; Rpb1 CTD ( total or phosphorylated ) – 8 . 5 μl . The bead suspension was incubated for 60 min with shaking ( 1400 rpm ) at RT , washed with 500 μl PBST buffer and 500 μl FA buffer . Beads were finally resuspended in 25 μl FA buffer . 1 . 5 μg S . cerevisiae chromatin and 30 ng S . pombe chromatin ( strain SHY1110 ) were combined , and samples were brought to a final volume of 500 μl . 25 μl of each sample was mixed with 25 μl Stop buffer and set aside ( input sample ) . 25 μl of beads was added to remaining 475 μl of samples followed by incubation for 16–20 hr at 4°C . The beads were washed for 3 min with gentle rotation with the following: three times with 500 μl FA buffer , two times with FA-HS buffer ( 50 mM HEPES , 7 . 5 , 500 mM NaCl , 1 mM EDTA , 1% Triton X-100 , 0 . 1% sodium deoxycholate ) , and once with 500 μl RIPA buffer ( 10 mM Tris , 8 . 0 , 0 . 25 M LiCl , 0 . 5% NP-40 , 1 mM EDTA , 0 . 5% sodium deoxycholate ) . DNA was eluted from beads with 25 μl Stop buffer at 75°C for 10 min . Elution was repeated , eluates were combined and incubated at 70°C for 16–20 hr together with input samples collected earlier . Samples were digested with 0 . 5 mg/ml RNase A ( Thermo Fisher Scientific #EN0531 ) for 30 min at 55°C and 1 mg/ml Proteinase K for 2 hr at 55°C . Sample volume was brought to 200 μl , and DNA was purified by two phenol/CHCl3/isoamyl alcohol ( 25:24:1 ) extractions and ethanol precipitation . DNA was resuspended in 15 μl 10 mM Tris , 8 . 0 , and the concentration was measured using Qubit HS DNA assay ( Thermo Fisher Scientific #Q32851 ) . NGS libraries for ChEC-seq and ChIP-seq experiments were prepared similarly as described ( Donczew et al . , 2020; Warfield et al . , 2017 ) . 12 μl of ChEC samples and 5 μl of ChIP samples were used as input for library preparation . Samples were end-repaired , phosphorylated , and adenylated in 50 μl reaction volume using the following final concentrations: 1X T4 DNA ligase buffer ( NEB #B0202S ) , 0 . 5 mM each dNTP ( Roche #KK1017 ) , 0 . 25 mM ATP ( NEB #P0756S ) , 2 . 5% PEG 4000 , 2 . 5 U T4 PNK ( NEB #M0201S ) , 0 . 05 U T4 DNA polymerase ( Invitrogen #18005025 ) , and 0 . 05 U Taq DNA polymerase ( Thermo Fisher Scientific #EP0401 ) . Reactions were incubated at 12°C 15 min , 37°C 15 min , 72°C 20 min , then put on ice and immediately used in adaptor ligation reactions . Adaptor ligation was performed in a 115 μl volume containing 6 . 5 nM adaptor , 1X Rapid DNA ligase buffer ( Enzymatics #B101L ) and 3000 U DNA ligase ( Enzymatics #L6030-HC-L ) , and reactions were incubated at 20° for 15 min . Following ligation , a two-step cleanup was performed for ChEC-seq samples using 0 . 25× vol Mag-Bind reagent ( Omega Biotek # M1378-01 ) in the first step and 1 . 1× vol in the second step . In case of ChIP-seq samples , a single cleanup was performed using 0 . 4× vol Mag-Bind reagent . In both cases , DNA was eluted with 20 μl 10 mM Tris , 8 . 0 . Library Enrichment was performed in a 30 μl reaction volume containing 20 μl DNA from the previous step and the following final concentrations: 1X KAPA buffer ( Roche #KK2502 ) , 0 . 3 mM each dNTP ( Roche #KK1017 ) , 2 . 0 μM each P5 and P7 PCR primer , and 1 U KAPA HS HIFI polymerase ( #KK2502 ) . DNA was amplified with the following program: 98°C 45 s , ( 98°C 15 s , ramp to 60°C @ 3°C/s , 60°C 10 s , ramp to 98°C @ 3°C/s ) 16–18× , 72°C 1 min . 18 cycles were used for library amplification for ChEC-seq samples and 16 cycles for ChIP-samples . A post-PCR cleanup was performed using 1 . 4× vol Mag-Bind reagent , and DNA was eluted into 30 μl 10 mM Tris , 8 . 0 . Libraries were sequenced on the Illumina HiSeq2500 platform using 25 bp paired-end reads at the Fred Hutchinson Cancer Research Center Genomics Shared Resources facility . Data analysis was performed similarly as described ( Donczew et al . , 2020 ) . The majority of the data analysis tasks except sequence alignment , read counting , and peak calling ( described below ) were performed through interactive work in the Jupyter Notebook ( https://jupyter . org ) using Python programming language ( https://www . python . org ) and short Bash scripts . All figures were generated using Matplotlib and Seaborn libraries for Python; ( https://matplotlib . org; https://seaborn . pydata . org ) . All code snippets and whole notebooks are available upon request . Paired-end sequencing reads were aligned to S . cerevisiae reference genome ( sacCer3 ) , S . pombe reference genome ( ASM294v2 . 20 ) , or D . melanogaster reference genome ( release 6 . 06 ) with Bowtie ( Langmead and Salzberg , 2012 ) using optional arguments ‘-I 10 -X 700 --local --very-sensitive-local --no-unal --no-mixed --no-discordant’ . Details of the analysis pipeline depending on the experimental technique used are described below . SAM files for S . cerevisiae data were used as an input for HTseq-count ( Anders et al . , 2015 ) with default settings . The GFF file with S . cerevisiae genomic features was downloaded from the Ensembl website ( assembly R64-1-1 ) . Signal per gene was normalized by the number of all S . pombe reads mapped for the sample and multiplied by 10 , 000 ( arbitrarily chosen number ) . Genes classified as dubious , pseudogenes , or transposable elements were excluded , leaving 5797 genes for the downstream analysis . As a next filtering step , we excluded all the genes that had no measurable signal in at least one out of 42 samples collected in this work . The remaining 5313 genes were used to calculate coefficient of variation ( CV ) to validate reproducibility between replicate experiments ( Figure 1—figure supplement 1D , Figure 3—figure supplement 1B and Supplementary file 1 ) . This gene set was further compared to a list of 4900 genes we found previously to provide the best reproducibility with a minimal loss of information ( Donczew et al . , 2020 ) . The overlapping set of 4883 genes was used in all plots where genes were divided into TFIID-dependent and CR categories . The results of biological replicate experiments for each sample were averaged . Corresponding samples were compared to calculate log2 change in expression per gene ( IAA to DMSO samples for degron experiments , and BDF2 deletion mutant to WT strain [BY4705] ) ( Supplementary file 1 ) . SAM files for S . cerevisiae data were converted to tag directories with the HOMER ( http://homer . ucsd . edu; Heinz et al . , 2010 ) ‘makeTagDirectory’ tool . Data for six DMSO-treated replicate samples from experiments involving Bdf1 , Bdf2 , or Bdf1/2 depletion were used to define promoters bound by Bdf1 , Bdf2 , Taf1 , Taf11 , Med8 , Med17 , and Spt3 as described below ( Supplementary file 3 ) . Peaks were called using HOMER ‘findPeaks’ tool with optional arguments set to ‘-o auto -C 0 L 2 F 2’ , with the free MNase dataset used as a control . These settings use a default false discovery rate ( FDR; 0 . 1% ) and require peaks to be enriched twofold over the control and twofold over the local background . Resulting peak files were converted to BED files using ‘pos2bed . pl’ program . For each peak , the peak summit was calculated as a mid-range between peak borders . For peak assignment to promoters , the list of all annotated ORF sequences ( excluding sequences classified as ‘dubious’ or ‘pseudogene’ ) was downloaded from the SGD website ( https://www . yeastgenome . org ) . Data for 5888 genes were merged with TSS positions ( Park et al . , 2014 ) . If the TSS annotation was missing , TSS was manually assigned at position −100 bp relative to the start codon . Peaks were assigned to promoters if their peak summit was in the range from −300 to +100 bp relative to TSS . In a few cases , where more than one peak was assigned to the particular promoter , the one closer to TSS was used . Promoters bound in at least four out of six replicate experiments were included in a final list of promoters bound by a given factor and were used to calculate promoter occupancy in all relevant experiments . Coverage at each base pair of the S . cerevisiae genome was calculated as the number of reads that mapped at that position divided by the number of all D . melanogaster reads mapped for the sample and multiplied by 10 , 000 ( arbitrarily chosen number ) . To quantify log2 change in factor promoter , occupancy signal per promoter was calculated as a sum of normalized reads per base in a 200 bp window around the promoter peak summit . Peak summits were defined using HOMER as described above . If no peak was assigned to a promoter using HOMER , the position of the strongest signal around TSS was used as a peak summit . Manual inspection of selected cases confirmed the validity of this approach . Corresponding IAA and DMSO-treated samples were compared to calculate log2 change in occupancy , and the results of biological replicate experiments for each sample were averaged ( Supplementary file 3 ) . Data for Bdf1 was used to call peaks as described for ChEC-seq with input sample used as a control . Promoters bound in at least one out of two replicate experiments were included in a final list of Bdf1 bound promoters as defined by ChIP-seq . For all samples , coverage at each base pair of the S . cerevisiae genome was calculated as the number of reads that mapped at that position divided by the number of all S . pombe reads mapped in the sample , multiplied by the ratio of S . pombe to S . cerevisiae reads in the corresponding input sample and multiplied by 10 , 000 ( arbitrarily chosen number ) . The list of all annotated ORF sequences ( excluding sequences classified as ‘dubious’ or ‘pseudogene’ ) was downloaded from the SGD website ( https://www . yeastgenome . org ) . Data for 5888 genes were merged with TSS positions obtained from Park et al . , 2014 . If the TSS annotation was missing , the TSS was manually assigned at position −100 bp relative to the start codon . Signal per gene was calculated as a sum of normalized reads per base in a fixed window relative to TSS ( defined in figure legends ) . The results of biological duplicate experiments for each sample were averaged and the log2 change in signal per gene was calculated by comparing corresponding IAA and DMSO-treated samples ( Supplementary file 3 ) . In experiments mapping Rpb1 CTD , Bur1 , Ctk1 , and Spt5 , the signal per gene was calculated between TSS and PAS ( Park et al . , 2014; Supplementary file 4 ) . FASTQ files for the Bdf1 ChIP-exo ( SRR397550 ) and H3 ChIP-seq ( SRR6495880 , SRR6495888 , SRR6495913 , SRR6495921 ) experiments were obtained from the SRA . Data were processed as described above except for the use of RPM normalization . The four datasets for H3 were averaged and used to normalize H4K12ac data . Amino acid sequences of yeast and human BET proteins and human Taf1 were downloaded from NCBI RefSeq database . Positions of individual BDs were obtained from Wu and Chiang , 2007 . Multiple sequence alignment of individual BDs was performed using Clustal Omega ( Sievers et al . , 2011 ) . Neighbor-joining tree was visualized with Jalview v 2 . 11 . 1 . 3 using PAM250 scoring matrix ( Waterhouse et al . , 2009 ) . Gene ontology analysis for yeast and human BET regulated genes was done using PANTHER overrepresentation test and PANTHER GO-Slim Biological Process annotation dataset with default settings ( http://www . pantherdb . org; v 16 . 0 ) ( Mi et al . , 2021 ) . Only the most specific , enriched child processes with FDR < 0 . 05 were used to compare BET functions between yeast and human model systems .

When a healthy cell creates new proteins , it activates a standard two-step biological manufacturing process . Firstly , DNA is transcribed from a specific gene to generate a strand of messenger RNA , or mRNA . Next , this mRNA molecule is translated to create the final protein product . This process of converting DNA into mRNA is supported by a series of helper proteins , including proteins from the bromodomain and extra-terminal domain ( BET ) family . Cancer cells can become ‘addicted’ to the process of converting DNA into RNA , leading to the overproduction of mRNA molecules , uncontrolled cell growth and tumor formation . Knocking out BET helper proteins could potentially bring cancer cells under control by halting transcription and preventing tumor growth . However , the precise ways in which BET helper proteins regulate transcription are currently poorly understood , and therefore developing rational ways to target them is a challenge . Building on their previous work , Donczew and Hahn have investigated how two BET helper proteins , Bdf1 and Bdf2 , help to regulate transcription in budding yeast . Using a range of genomic techniques , Donczew and Hahn found that Bdf1 and Bdf2 had important roles for initiating transcription and elongating mRNA molecules . Both BET proteins were also involved in recruiting other protein factors to help with the transcription process , including TFIID and Mediator . Based on these findings , it is likely that cooperation between BET proteins , TFIID and Mediator represents a common pathway through which gene expression is regulated across all eukaryotic organisms . Both Bdf1 and Bdf2 were also found to provide the same functions in yeast as similar BET proteins in humans . Using this robust yeast model system to perform further detailed studies of BET factors could therefore provide highly relevant information to expand our understanding of human biology and disease . Ultimately , this research provides important insights into how two members of the BET family of helper proteins contribute to the control of transcription in yeast . This information could be used to guide the design of new drugs for cancer therapy that target not only BET proteins themselves but also other proteins they recruit , including TFIID and Mediator . Such targeted drugs would be expected to be more harmful for cancer cells than for healthy cells , which could reduce unwanted side effects .

2021

BET family members Bdf1/2 modulate global transcription initiation and elongation in Saccharomyces cerevisiae

Here , we describe a novel method based on intronic MiMIC insertions described in Nagarkar-Jaiswal et al . ( 2015 ) to perform conditional gene inactivation in Drosophila . Mosaic analysis in Drosophila cannot be easily performed in post-mitotic cells . We therefore , therefore , developed Flip-Flop , a flippase-dependent in vivo cassette-inversion method that marks wild-type cells with the endogenous EGFP-tagged protein , whereas mutant cells are marked with mCherry upon inversion . We document the ease and usefulness of this strategy in differential tagging of wild-type and mutant cells in mosaics . We use this approach to phenotypically characterize the loss of SNF4Aγ , encoding the γ subunit of the AMP Kinase complex . The Flip-Flop method is efficient and reliable , and permits conditional gene inactivation based on both spatial and temporal cues , in a cell cycle- , and developmental stage-independent fashion , creating a platform for systematic screens of gene function in developing and adult flies with unprecedented detail . Functional gene annotation requires , among others , the knowledge of phenotypes observed in mutants of the gene of interest , as well as the expression pattern and subcellular localization of the protein . For Drosophila genes whose loss causes early developmental lethality , characterization of gene function in later stages of the animal’s life cycle relies on generating mosaics . Currently , Mosaic Analysis with a Repressible Cell Marker ( MARCM ) ( Lee and Luo , 1999 ) and CRISPR/Cas9-mediated somatic mutagenesis ( Port et al . , 2014; Xue et al . , 2014 ) are the two major techniques used to study mutations in mosaics . While MARCM has been very successfully used in mitotically active , developing tissues , the technique has two major limitations . First , the technique relies on mitotic recombination and can therefore not be used in non-dividing cells . Second , along with the mutation under study , other mutations in the chromosome distal to the FRT site also become homozygous in mutant cells , requiring ‘rescue experiments’ of the gene of interest to validate the results ( Roegiers et al . , 2009 ) . Another strategy is based on CRISPR/Cas9-mediated somatic mutagenesis to generate mosaics ( Port et al . , 2014; Xue et al . , 2014 ) . The limitations of CRISPR/Cas9 are that the identity of the generated lesions for the gene of interest may vary in individual cells in the same animal or tissue . Moreover , the mutant cells are not marked and cannot be distinguished from neighboring wild-type cells . Here , we describe a new flippase ( FLP ) -dependent method ‘Flip-Flop’ that offers several advantages over the current techniques for generating mosaics . ( 1 ) The method does not rely on cell division and can , therefore , be broadly used for conditional gene inactivation in post-mitotic cells such as neurons . ( 2 ) It allows endogenous tagging of proteins with EGFP , which permits multiple applications , and ( 3 ) it simultaneously marks mutant cells with mCherry . We engineered the ‘Flip-Flop’ cassette for conditional gene inactivation . This cassette contains two modules that are placed in opposite orientation: a protein-trap ( PT ) module and a gene-trap ( GT ) module ( Figure 1A ) . The PT module carries a splice acceptor ( SA ) , followed by an in-frame EGFP coding sequence , and a splice donor ( SD ) ( Venken et al . , 2011 ) . The GT module similarly contains a SA , but is followed by the T2A peptide sequence , an mCherry coding sequence , stop codons in all three reading frames , and an SV40 polyA signal . The PT and GT modules are placed in opposite orientations and are flanked by inverted pairs of canonical FRT and FRT14 sites , forming a flip-excision switch ( FLEx ) ( Schnütgen et al . , 2003; Xue et al . , 2014 ) . The entire cassette is nested between two inverted attB sites that facilitate Recombination-Mediated Cassette Exchange ( RMCE ) between the Flip-Flop cassette and a target Minos-Mediated Integration Cassette ( MiMIC ) that resides in a coding intron of the gene of interest . When integrated in the MiMIC in the PT orientation , Flip-Flop should result in expression of the endogenous protein with an internal EGFP tag . The internal EGFP tagging does not or subtly disrupt the protein’s function in 77% of the cases ( Nagarkar-Jaiswal et al . , 2015 ) . The cassette can then be converted from a PT to a GT , in vivo , by inverting the cassette’s orientation through the expression of FLP that acts on the FLEx switch ( Figure 1B ) . Following the switch from the PT to the GT orientation , transcription is precociously terminated by the polyA sequence . When this truncated transcript is translated , the T2A site induces a translational skip ( Tang et al . , 2009 ) , truncating the native protein and re-initiating translation at the mCherry sequence ( Figure 1B ) . Hence , inversion of the Flip-Flop cassette results in the generation of a truncated protein , which is typically non-functional , and simultaneously marks the cells that are actively transcribing the gene with mCherry . While mCherry’s expression pattern recapitulates the spatiotemporal expression pattern of the recipient gene , it does not reproduce the endogenous subcellular localization of the protein ( Figure 1B ) . In summary , expression of FLP will induce a Flip-Flop and produce mCherry-labeled mutant cells in which the gene is inactivated , whereas the surrounding cells are wild-type and express the EGFP-tagged protein . 10 . 7554/eLife . 26420 . 003Figure 1 . Mosaic generation using the Flip-Flop cassette . ( A ) The architecture of the Flip-Flop cassette . The cassette consists of two independent modules ( PT and GT ) , that are oriented in opposite orientations . The PT module contains a splice acceptor ( SA ) , followed by an EGFP tag and a splice donor ( SD ) . The GT module contains a SA sequence , followed by the T2A peptide coding sequence ( which will induce a translational skip ) , the mCherry coding region , stop codons in all three coding frames , and an SV40 polyA transcriptional termination signal . Given the opposite orientation of both modules , only one of the SA sequence will be active with respect to the recipient gene . The two modules are nested within a pair of FRT and FRT14 inverted repeats , forming a flippase-responsive FLEx switch . Finally , the entire cassette is flanked by two inverted attB sequences that permit phiC31-mediated RMCE between the Flip-Flop cassette and pre-existing MiMIC elements . A comparison of the FRT and FRT14 sequence is shown below . The FRT14 sequence varies from the canonical FRT sequence at the residues highlighted in red . ( B ) Schematic showing the inversion of a PT-oriented Flip-Flop cassette , inserted into the coding intron of a hypothetical gene . Upon FLP-expression , the FLEx switch undergoes two recombination events: ( 1 ) recombination between the two FRT sites or between the two FRT14 sites leads to cassette inversion that is followed by ( 2 ) excision of either the pair of FRT sites or the pair of FRT14 sites , both of which have obtained the same orientation due to the flip . Since the remaining unpaired FRT and FRT14 sites are not able to recombine , the cassette will be locked in the GT orientation . Thus , the initial PT orientation allows the gene to be tracked by EGFP-tagged protein expression in tissues . FLP activity inverts the Flip-Flop cassette in random cells , generating a mosaic tissue consisting of cells that did not undergo the flip and are still expressing the EGFP-tagged protein and cells that inverted the Flip-Flop cassette into the GT orientation , which are marked by mCherry expression . ( C ) ( a ) Wing imaginal disc from a effMI05507EGFP-T2A-mCherry/+ third instar larva , expressing Eff-EGFP-Eff ( b ) Wing imaginal disc from a hsFLP; effMI05507EGFP-T2A-mCherry/+ third instar larva showing hsFLP-induced clones that loose the GFP signal ( c ) and express mCherry ( d ) Scale bar = 50 µm . ( D ) ( a ) Stage 6 egg chamber obtained from SNF4AyMI09417EGFP-T2A-mCherry/+ females reveals EGFP-SNF4Aγ-EGFP expression in all follicle cells . ( b ) Stage 6 egg chamber obtained from hsFLP; SNF4AyMI09417EGFP-T2A-mCherry/+ females shows a mosaic tissue with groups of cells that have retained EGFP expression , and others that have inverted the Flip-Flop cassette and can be recognized by loss of EGFP expression ( c ) and gain of mCherry expression ( d ) , inversion of Flip-Flop cassette is marked with loss of EGFP and expression of mCherry . Scale bar = 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 26420 . 00310 . 7554/eLife . 26420 . 004Figure 1—source data 1 . List of constructs generated in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 26420 . 00410 . 7554/eLife . 26420 . 005Figure 1—source data 2 . List of fly strains used in the study . DOI: http://dx . doi . org/10 . 7554/eLife . 26420 . 00510 . 7554/eLife . 26420 . 006Figure 1—figure supplement 1 . Flip-Flop PT insertions in eff and SNF4Aγ generate functional EGFP-tagged proteins . ( A ) ( a ) Gene structure of eff , showing the precise location of MiMIC insertion MI05507 ( red triangle ) as well as the orientation of the MiMIC cassette ( red arrow ) ( based on FlyBase annotation release FB2017_01 ) . The coding exons of isoforms that are tagged are shown in green , the 5’ and 3’ UTRs are shown in blue ( b ) Table describing the various alleles of eff used in this study , the nature of these alleles and their ability to complement a null allele eff[mer4] . ( B ) ( a ) Gene structure of SNF4Aγ showing the precise location of MiMIC insertion MI09417 ( red triangle ) and its orientation ( red arrow ) ( based on FlyBase annotation release FB2017_01 ) . The coding regions of isoforms that are tagged are shown in green , isoforms that is not tagged is shown in orange and the 5’ and 3’ UTRs are shown in blue . ( b ) Table displaying results of complementation tests performed between the various SNF4Aγ alleles used in this study . MiMIC gene trap insertion SNF4AγMI09417 and the T2A-GAL4 derivative –SNF4AγMI09417 T2A>GAL4 represent loss-of-function alleles of SNF4Aγ , as they fail to complement a deficiency ( Df ( 3R ) e-R1 ) spanning SNF4Aγ , as well as each other . The PT Flip-Flop insertion ( SNF4AγMI09417-EGFP-T2A--mCherry ) complements all tested loss-of-function alleles and hence produces a functional , EGFP-tagged protein . DOI: http://dx . doi . org/10 . 7554/eLife . 26420 . 00610 . 7554/eLife . 26420 . 007Figure 1—figure supplement 2 . Crossing schemes for FLP-mediated conditional gene inactivation . ( A ) Males carrying a hsFLP transgene and a null allele of the gene of interest are crossed to females carrying a Flip-Flop PT insertion in the same gene . The progeny is heat shocked to induce FLP expression and initiate Flip-Flop cassette inversion , generating mosaics ( as described in Materials and methods ) and the desired genotypes are analysed . ( B ) A similar scheme for inducing conditional gene inactivation in developing eyes , using ey-FLP . ( C ) Crossing scheme for conditional gene inactivation combining the Flip-Flop cassette with FLP expressed using the GAL4/UAS binary system . ( D ) Crossing scheme for Flip-Flop-mediated conditional gene inactivation using a T2A-GAL4 driver , inserted in the gene of interest . The T2A-GAL4 insertion creates a loss-of-function allele and simultaneously induces GAL4 expression according to the spatial and temporal pattern of the gene of interest . Hence , the T2A-GAL4-induced FLP expression will promote cassette inversion in all the cells that express the gene under study . DOI: http://dx . doi . org/10 . 7554/eLife . 26420 . 00710 . 7554/eLife . 26420 . 008Figure 1—figure supplement 3 . Inversion of the Flip-Flop PT insertion in Cdep using ey-FLP . ( A ) Gene structure of Cdep , showing the precise location of MiMIC insertion MI12769 ( red triangle ) as well as the orientation of the MiMIC cassette ( red arrow ) ( based on FlyBase annotation release FB2017_01 ) . The coding regions of isoforms that are tagged by the MiMIC are shown in green and the 5’ and 3’ UTRs are shown in blue . ( b ) A table describing the various Cdep alleles used in this study , nature of the insertions and the ability of alleles to complement a gene trap alleles of Cdep . ( B ) ( a ) Eye-antennal discs from CdepMI12769EGFP-T2A-mCherry/+ larvae expressing Cdep-EGFP-Cdep ( b ) Eye-antennal discs from ey-FLP; CdepMI12769EGFP-T2A-mCherry/+ show patches of mCherry expressing cells , in which cassette inversion occurred . ( c and d ) Images displaying the individual GFP ( c ) and mCherry ( d ) channels of the eye-antennal disc shown in ( b ) . Scale bar = 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 26420 . 00810 . 7554/eLife . 26420 . 009Figure 1—figure supplement 4 . Inversion of GT-oriented Flip-Flop insertions . ( A–B ) Third instar larval brains from heterozygous animals carrying the effMI05507-T2A-mCherry-EGFP GT-oriented insertion stained for mCherry ( red ) and GFP ( green ) . ( A ) ( a ) Image of a third instar larval effMI05507-T2A-mCherry-EGFP /+ brain ( b ) ey-FLP-mediated cassette inversion leads to EGFP expression in nearly all eyeless–expressing cells in the eye disc ( white dotted line ) and the third instar larval brain ( blue dotted line ) . ( c and d ) Images displaying the individual GFP ( c ) and mCherry ( d ) channels of the larval brain shown in ( b ) , scale bar = 50 µm . ( B ) Heat-shock-mediated FLP-expression induces cassette inversion and Eff-EGFP-Eff expression ( green , a and c ) in random subsets of cells in the third instar larval brain of effMI05507-T2A-mCherry-EGFP /+ animals , scale bar = 50 µm . ( C ) Table showing non-functional PT , and GT insertions in Nedd8 and Sik3 generated in this study . These alleles fail to complement the tested deficiencies . This indicates that the EGFP tag in these genes disrupts the proper functioning or the expression of these genes . Nevertheless , these PT insertions could be flipped efficiently using ey-FLP and hsFLP ( data not shown ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26420 . 00910 . 7554/eLife . 26420 . 010Figure 1—figure supplement 5 . The efficiency of Flip-Flop insertions . Eye-antennal discs from effMI5507EGFP-T2A-mCherry/+ ( A ) , SNF4AγMI09417EGFP-T2A-mCherry/+ ( B ) , Trim9MI12525EGFP-T2A-mCherry/+ ( C ) , Sik3MI15336EGFP-T2A-mCherry/+ ( D ) , Nedd8MI13778EGFP-T2A-mCherry/+ ( E ) , and CdepMI12769EGFP-T2A-mCherry/+ ( F ) stained for EGFP ( green ) , marking the tagged protein and for mCherry ( red ) , marking mutant cells . ( a ) Eye-antennal discs from animals containing PT-oriented Flip-Flop insertions in the respective genes , revealing the expression pattern of the gene of interest . ( b ) ey-FLP-mediated cassette inversion for the respective genes , showing mosaic expression of mCherry upon inversion of the Flip-Flop cassette . ( c and d ) Images displaying the individual GFP ( c ) and mCherry ( d ) channels of the eye-antennal discs shown in ( b ) for each gene . The efficiency of cassette inversion for each gene is shown on right side . Scale bar = 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 26420 . 010 We created PT insertions for nine genes using available MiMIC insertions: effete ( eff ) , Nedd8 , SNF4/AMP-activated protein kinase gamma subunit ( SNF4Aγ ) Chondrocyte-derived ezrin-like domain containing protein ( Cdep ) , Tripartite motif containing 9 ( Trim9 ) , Salt-Induced kinase3 ( Sik3 ) , Ankyrin2 ( Ank2 ) , Circadian trip ( Ctrip ) and Ecdysone-induced protein 63E ( Eip63E ) . Seven of the nine generated PT insertions complement the lethality associated with loss of the corresponding gene , suggesting that these internally tagged proteins are biologically functional , consistent with previous observations ( Nagarkar-Jaiswal et al . , 2015; Venken et al . , 2011 ) . The PT insertion in eff ( Figure 1—figure supplement 1A ) introduces the expression of internally tagged Eff ( Eff-EGFP-Eff ) in the wing discs of effMI05507-EGFP-T2A-mCherry larvae ( Figure 1C a ) . Heat-shock-induced expression of FLP ( hsFLP ) ( Figure 1—figure supplement 2A ) causes loss of the EGFP signal and induces mCherry expression in random subsets of cells ( Figure 1C b-d ) . Similarly , the PT insertion in SNF4Aγ ( Figure 1—figure supplement 1B a ) leads to SNF4Aγ-EGFP-SNF4Aγ expression in adult egg chambers ( Figure 1D a ) . The SNF4Aγ Flip-Flop cassette can be flipped to the GT orientation efficiently , generating large clones of cells expressing mCherry upon heat-shock ( Figure 1D b-d ) . Finally , we tagged Cdep ( Figure 1—figure supplement 3A a-b ) , observed EGFP expression in the eye-antennal discs , and efficiently induced inversions using ey-FLP ( Figure 1—figure supplement 3B and Figure 1—figure supplement 2B ) . Hence , the PT orientation insertions permit gene expression analysis and can effectively be inverted to create mCherry marked mutant cells . A comparison of the PTs generated using a previous , shorter , RMCE construct ( GFSTF; Venken et al . , 2011 ) and the PT of the Flip-Flop cassette did not show any obvious difference in expression pattern or genetic properties based on complementation tests . In summary , if the protein traps are functional with the GFSTF cassette they are also functional with the EGFP-tagged proteins derived from the Flip-Flop . In addition to the above-described PT lines , we generated GT-lines expressing mCherry for five genes: eff , Cdep , Trim9 , Ank2 , and Ctrip . Complementation tests with independently derived loss-of-function alleles ( or deficiencies ) indicate that these GT-alleles are indeed loss-of-function alleles . Note that none of these GT insertions induce an obvious dominant phenotype . We observed efficient conversion of an initial GT insertion to the PT orientation for effMI05507 T2A-mCherry-EGFP , using either the ey-FLP or hsFLP drivers in larval brains . As shown , the use of ey-FLP introduces Eff-EGFP-Eff expression in the eye discs ( area within white dotted lines ) , as well as the area of the brain where eyeless is expressed ( demarcated by blue dotted lines ) ( Figure 1—figure supplement 4A ) . In contrast to the ey-FLP , smaller clones of randomly positioned wild-type and mutant cells can be generated by stochastic FLP expression when applying heat shocks . Indeed , heat-shock-induced hsFLP expression brought about small patches of EGFP-positive cells in effMI05507 T2A-mCherry-EGFP larval brains ( Figure 1—figure supplement 4B ) . This type of experiment creates wild-type cells in an otherwise mutant animal and should help to assess whether a mutant phenotype can be attenuated or reverted , or not . In addition , these experiments may help identify in which tissue an essential gene is required . Finally , for Nedd8 and Sik3 we generated PT lines , but these insertions failed to complement corresponding deficiencies , showing that the internal EGFP tag disrupts protein function . Nevertheless , we were able to induce inversions for both genes using ey-FLP , confirming our observation that flipping the Flip-Flop cassette is efficient ( Figure 1—figure supplement 4C and Figure 1—figure supplement 5D and E ) . In summary , both orientations of the Flip-Flop cassette could effectively be flipped . To examine and compare the efficiency of the Flip-Flop inversions , we tested six genes using the same FLP-source , under identical conditions . For SNF4Aγ , Trim9 and eff , more than 95% of the cells undergo cassette inversion . For Nedd8 , cdep , and Sik3 , more than 70% cells displayed Flip-Flop inversions ( Figure 1—figure supplement 5 ) . Hence , inducing Flip-Flop inversions in eye-antennal discs is highly efficient for all genes tested . Next , we tested if Flip-Flop can be used for gene inactivation in post-mitotic cells by using Trim9 as an example . Previously , Song et al . ( 2011 ) created an imprecise excision of a P-element ( P{GawB}Trim9[NP4638] ) inserted in the 5′-regulatory region of Trim9 and reported that animals lacking the first few amino acids of Trim9 are viable and exhibit a droopy wing phenotype ( Song et al . , 2011 ) . We generated both GT- ( Trim9MI12525P T2A-mCherry-EGFP ) and PT-oriented ( Trim9MI12525EGFP- T2A-mCherry ) Flip-Flop insertions in Trim9 ( Figure 2—figure supplement 1A a ) . The GT insertion is homozygous lethal and does not complement a deficiency ( Trim9Df ) , while the PT insertion is viable in trans over Trim9Df ( Figure 2—figure supplement 1A b ) suggesting that the PT encodes a functional protein . When the PT insertion in Trim9MI12525EGFP- T2A-mCherry/Trim9Df animals was inverted in all cells using a ubiquitously expressed ( Actin-GAL ) FLP , the animals died as second instar larvae . This , together with the lethality observed in the GT insertion , suggests that a strong loss of Trim9 leads to early developmental lethality . We then examined the consequences of neuronal loss of Trim9 , by inverting the Trim9MI12525EGFP T2A--mCherry PT insertion in larval and adult brains using neuronally expressed FLP ( nSyb-GAL4>UAS-FLP ) ( Figure 1—figure supplement 2C and Figure 2—figure supplement 1B ) . Prior to the flip , in the PT orientation , Trim9 is expressed in numerous neurons in the larval brain ( Figure 2A a-d ) and the protein is localised to nuclei and cytoplasm ( Figure 2A e-h ) . These cells do not express mCherry . Following inversion to the GT orientation , we detected mCherry expressing cells in the larval brain ( Figure 2A a’-d’ ) . Upon close examination , we observed three different populations of neurons . First , cells that have undergone cassette inversion and are marked with mCherry ( Figure 2A e’-h’ , indicated by number ( 1 ) Second , cells that are EGFP-positive ( Trim9-EGFP-Trim9 ) and did not flip ( Figure 2A e’-h’ , indicated by number ( 2 ) Third , a population of cells that expresses both Trim9-EGFP-Trim9 as well as mCherry ( Figure 2A e’-h’ , indicated by number ( 3 ) Note that Trim9-EGFP-Trim9 levels are lower in these cells . We reason that these cells have undergone cassette inversion and are hence expressing mCherry , but that the Trim9-EGFP-Trim9 that was produced prior to the flip has not been fully degraded . Thus , the population of cells that express only mCherry must have lost most of the tagged Trim9 protein , whereas Trim9-EGFP-Trim9 perdures in a subset of cells that are yellow . Importantly , this allows analysis of protein perdurance . In addition , it permits researchers to only select and analyse those cells that are truly mutant and have lost the majority of the wild-type protein . Neuronal Trim9 inactivation does not affect the eclosion of the animal but leads to the sterility of short-lived animals with droopy wings , corroborating previous observations associated with the loss of Trim9 ( Song et al . , 2011 ) . 10 . 7554/eLife . 26420 . 011Figure 2 . Inactivation of Trim9 in post-mitotic cells using nSyb-GAL4/UAS-FLP . ( A ) A third instar larval brain of nSyb-GAL4/+; Trim9MI12525EGFP-T2A-mCherry/Trim9Df , raised at 29°C stained for EGFP ( green , a , d and f ) , mCherry ( red , b , d and g ) and Elav ( blue , c , d and h ) . In the absence of UAS-FLP , only GFP signal is detected , reflecting the Trim9-EGFP-Trim9 expression pattern . In contrast , mCherry is not expressed . ( b , d and g ) . Magnification of the region indicated by a white square in panel d ( e–h ) . Third instar larval brains of UAS-FLP/nSyb-GAL4; Trim9MI12525EGFP-T2A-mCherry/Trim9Df , expressing FLP under the control of nSyb-GAL4 show cassette inversion that leads to the loss of EGFP and gain of mCherry expression in neurons ( a’–h’ ) . Magnification of the region indicated by a white square in panel ( d’ ) shows the existence of three populations of neurons ( e’–h’ ) : neurons that express mCherry and have completely lost GFP expression ( 1 ) , neurons that have not undergone cassette inversion and thus still express GFP but lack mCherry expression ( 2 ) and neurons that express both tags , suggesting they have recently undergone cassette flip ( as indicated by mCherry expression ) , and still contain some remaining Trim9-EGFP-Trim9 protein that perdures following the cassette flip , scale bar = 50 µm ( a , b , c , d , a’ , b’ , c’ and d’ ) , scale bar = 5 µm ( e , f , g , h , e’ , f’ , g’ and h’ ) . ( B ) Adult brain of tubP-GAL80ts; Trim9MI12525EGFP-T2A-mCherry/Trim9Df; UAS-FLP/nSyb-GAL4 animals stained for EGFP ( green ) , mCherry ( red ) and Elav ( blue ) . The temperature-sensitive GAL80 ( GAL80ts ) , expressed under a tubulin promoter ( tubP ) was used to prevent nSyb-GAL4 from driving expression of UAS-FLP during development . ( a–d ) Adult brain of tubP-GAL80ts; Trim9MI12525EGFP-T2A-mCherry/Trim9Df; UAS-FLP/nSyb-GAL4 control animals , raised at 18°C display Trim9-EGFP-Trim9 ( green , a and d ) and do not show any mCherry expression ( red , b and d ) . ( a’–d’ ) Adult brains of tubP-GAL80ts; Trim9MI12525EGFP-T2A-mCherry/Trim9Df; UAS-FLP/nSyb-GAL4 animals , raised at 18°C until 4 days after eclosion , and then shifted to 29°C to induce FLP-expression , show mCherry-expressing neuronal cells in which the cassette inversion has taken place ( red , b’ and d’ ) , scale bar = 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 26420 . 01110 . 7554/eLife . 26420 . 012Figure 2—figure supplement 1 . Inactivation of Trim9 in post-mitotic cells through induction of FLP using nSyb-GAL4/UAS-FLP . ( A ) ( a ) Gene structure of Trim9 , showing the precise location of MiMIC insertion MI12525 ( red triangle ) as well as the orientation of the MiMIC cassette ( red arrow ) ( based on FlyBase annotation release FB2017_01 ) . The coding regions of isoforms that are tagged are shown in green and the 5’ and 3’ UTRs are shown in blue . ( b ) A table describing the various Trim9 alleles used in this study , nature of the insertions , and the ability of these alleles to complement a deficiency spanning Trim9 . Though the PT-oriented Flip-Flop insertion in Trim9MI12525EGFP-T2A-mCherry is homozygous lethal , Trim9MI12525EGFP-T2A-mCherry/Trim9Df animals are viable , indicating the presence of a second-site , lethal mutation elsewhere on the Trim9MI12525EGFP-T2A--mCherry chromosome . ( B ) A third instar larval brain of nSyb-GAL4 > UAS-2X-EGFP animals , stained for GFP ( green , a and b ) , and neurons marked by the expression of Elav ( blue , c ) revealing the nSyb-GAL4 expression pattern in a developing larval brain . Scale bar = 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 26420 . 012 Next , we explored the use of the Flip-Flop cassette in post-mitotic , adult neurons . We expressed GAL80ts ( a suppressor of GAL4 at low temperature ) to prevent nSyb-GAL4 from inducing flips during development . tubP-GAL80ts; Trim9MI12525EGFP- T2A-mCherry/Trim9Df; UAS-FLP/nSyb-GAL4 animals were raised at 18°C until the fourth day of adulthood and were subsequently shifted to 29°C . Animals that were not shifted to 29°C did not express mCherry in adult brains ( Figure 2B a-d ) . In contrast , those shifted to a higher temperature on the fourth day of adulthood showed loss of Trim9-EGFP-Trim9 expression and gain of mCherry in a large subset of neurons ( Figure 2B a’-d’ ) . Interestingly , animals shifted to 29°C do not show droopy wing phenotype , but have a reduced lifespan . This suggests that the droopy wing phenotype observed by Song et al . ( 2011 ) is due to the loss of neuronal Trim9 during development , and not because of loss of the gene in adults . To further explore the use of Flip-Flop in adult flies , we selected a subunit of the AMP-activated protein kinase ( AMPK ) complex . The AMPK complex is involved in sensing stresses such as a drop in the ratio of ATP/AMP , hypoxia , ischemia , and heat-shock ( Hardie et al . , 2003 ) . The ATP/AMP ratio is sensed through an allosteric mechanism by the non-catalytic γ subunit of the AMPK complex , which in turn promotes the phosphorylation of the catalytic α subunit enhancing α subunit’s kinase activity ( Sanders et al . , 2007; Scott et al . , 2004 ) . Phosphorylation of AMPK’s targets by the α-subunit activates a signalling cascade that ultimately regulates fatty acid oxidation , autophagy , and mitochondrial biogenesis and boosts the ATP/AMP ratio within the cell ( Hardie et al . , 2012 ) . Sensing the ATP/AMP ratio by the γ subunit is thus critical for the proper function of the AMPK complex , and together with the other AMPK subunits , the γ subunit is crucial for proper energy homeostasis ( Hardie et al . , 2003 ) . In Drosophila , the γ subunit is encoded by SNF4Aγ , with 16 transcriptional isoforms . Of these , 15 should be tagged using MI09417 ( Figure 1—figure supplement 1B a ) . We converted this MiMIC insertion into a T2A-GAL4 ( Diao et al . , 2015 ) allele ( SNF4AγMI09417T2A>GAL4 ) , and a Flip-Flop insertion in the PT orientation ( SNF4AγMI09417EGFP- T2A-mCherry ) ( Figure 3A and Figure 1—figure supplement 1B a ) . SNF4AγMI09417T2A>GAL4 truncates the SNF4Aγ protein and simultaneously expresses GAL4 in the spatiotemporal pattern of SNF4Aγ . The SNF4AγMI09417 T2A>GAL4 allele fails to complement the SNF4AγDf and acts as a loss-of-function allele . On the other hand , SNF4AγMI09417EGFP-T2A-mCherry/SNF4AγMI09417 and SNF4AγMI09417EGFP-T2A-mCherry/SNF4AγDf animals are viable , indicating that the Flip-Flop insertion produces a functional SNF4Aγ-EGFP-SNF4Aγ protein ( Figure 1—figure supplement 1B b ) . 10 . 7554/eLife . 26420 . 013Figure 3 . Developmental and neuronal functions of SNF4Aγ revealed through broad or tissue-specific inactivation . ( A ) Gene structure of SNF4Aγ , displaying one of the transcriptional isoforms and the precise location of MiMIC insertion MI09417 ( red triangle ) as well as the orientation of the MiMIC cassette ( red arrow ) . The structure of the T2A-GAL4 insertion ( SNF4AγMI094177T2>GALl4 ) and the PT oriented Flip-Flop insertion ( SNF4AγMI09417EGFP-T2A-mCherry ) are shown below . ( B ) ( a ) Control third instar larva ( SNF4AγMI09417EGFP-T2A-mCherry/SNF4AγMI09417 T2A>-GALl4; PT ) and third instar larva in which SNF4Aγ is inactivated through cassette flip , driven by the T2A-GAL4 insertion in SNF4Aγ on the other chromosome ( UAS-FLP;; SNF4AγMI094177EGFP-T2A-mCherry/SNF4AγMI09417T2A>GAL4; GT ) ( b ) Expression of mCherry in larvae shown in ( a ) . Arrowhead indicates auto-fluorescence of gut tissue in control animals ( SNF4AγMI04147EGFP-T2A-mCherry/SNF4AγMI09417T2A-GAL4; PT ) , which differs from the mCherry signal in animals that underwent cassette flip ( UAS-FLP;;SNF4AγMI09147EGFP-T2A-mCherry/SNF4AγMI09147 T2A>GAL4 , GT ) . ( c ) Pupae of control ( PT ) and experiment genotypes ( GT ) . Loss of SNF4Aγ leads to premature pupariation and death . ( C ) ey-FLP-mediated cassette inversion in developing eyes leads to electroretinogram ( ERG ) defects . ( Left ) ERG trace obtained from control animals ( SNF4AγMI09417EGFP-T2A-mCherry/SNF4AγDf ) on day 1 and day 30 . ( Right ) ERG trace obtained from experimental animals ( ey-FLP; SNF4AγMI09417EGFP-T2A-mCherry/SNF4AγDf ) in which ey-FLP-mediated cassette inversion induced loss of SNF4Aγ in >95% of the developing eye ( see also Figure 1—figure supplement 5B ) . ( D ) Histogram showing relative ERG amplitude ( indicated by the red double-headed arrow in ( C ) ) , measured on days 1 and 30 of control animals ( PT/green ) and experimental animals with ey-FLP-mediated SNF4Aγ inactivation ( GT/red ) . ( At least eight animals of each genotype were analysed . *** indicates a p-value<0 . 001 and ** indicates a p-value<0 . 01 obtained by performing a student T-test ) . ( E ) Image displaying adult eyes of control ( SNF4AγMI09417EGFP-T2A-mCherry/SNF4AγDf , a–c ) and experimental ( ey-FLP; SNF4AγMI09417EGFP-T2A-mCherry/SNF4AγDf , a’–c’ ) animals stained with phalloidin ( blue ) marking the seven visible photoreceptors arranged within each ommatidium ( red dotted circle ) , SNF4Aγ-EGFP-SNF4Aγ expression ( green ) and mCherry expression ( red ) . ( b , b’ , c and c’ ) mCherry channel ( b , b’ ) and phalloidin staining ( c , c’ ) of the image shown in ( a and a’ ) . ( a’ ) eyFLP-driven loss of SNF4Aγ induces mCherry expression ( b’ ) and leads to defects in photoreceptor arrangement and loss of individual rhabdomeres in ommatidia ( red dotted circle ) ( c’ indicated by numbers ) , whereas control ommatidia consistently contain seven visible rhabdomeres ( c ) , scale bar = 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 26420 . 013 In SNF4AγMI09147 T2A>GAL4 /SNF4AγMI09147EGFP- T2A-mCherry animals , SNF4γ-EGFP-SNF4Aγ is broadly expressed in larvae ( data not shown ) . In the presence of UAS-FLP , the GAL4 from SNF4AγMI09417 T2A>GAL4 will drive FLP expression in cells expressing SNF4Aγ , which will convert the PT insertion into a GT insertion . The GT insertion , while expressing mCherry , leads to the loss of functional protein produced by the SNF4AγMI09147EGFP-T2A-mCherry allele . Hence , UAS-FLP;; SNF4AγMI09417EGFP-T2A-mCherry/SNF4AγMI094177 T2A>GAL4 animals lack expression of both alleles of SNF4Aγ ( Figure 1—figure supplement 2D ) . These animals die as poorly developed pupae and are slim and elongated in the larval stage ( Figure 3B b-c ) . These larvae show a broad expression of mCherry that resembles the EGFP expression pattern from SNF4AγMI09417EGFP- T2A-mCherry allele ( Figure 3B b ) . Next , we induced Flip-Flop inversion in eyes . The Drosophila eye is a highly organised structure that is composed of about 800 small functional units called ommatidia . Every ommatidium contains eight photoreceptor cells ( R1-R7 ) each with a rhabdomere , the light-sensing organelle . Within each ommatidium , rhabdomeres are arranged in a highly stereotypic fashion ( Figure 3E c control panel , indicated by numbers ) . In ey-FLP;;SNF4AγMI09417EGFP-T2A-mCherry/SNF4AγDf animals , almost all photoreceptors inverted from the PT- to GT- orientation , as revealed by mCherry expression ( Figure 3E b’ ) . Inverting the cassette in SNF4Aγ to a GT orientation reduces the amplitude of the electroretinogram ( ERG ) in 1-day-old flies . These animals did not show a stronger reduction in ERG amplitude when measured at day 30 and hence do not show further degeneration ( Figure 3C and D ) . We also observed a reduction in the number of rhabdomeres as well as defects in photoreceptor organisation ( Figure 3E–c’ ) . The loss of SNF4Aγ disrupts energy homeostasis , possibly prevents the photoreceptors from meeting their high-energy demand and leading to a reduction in their activity . Similar defects have been described in alicorn mutant eye clones ( SNF4Aα ) ( Spasić et al . , 2008 ) . It has been previously shown that SNF4Aγ is required for autophagy ( Lippai et al . , 2008 ) . We , therefore , analysed the role of SNF4Aγ in autophagy in mutant cells generated by Flip-Flop inversion . To this end , we induced Flip-Flop-mediated loss of SNF4Aγ in the adult gut and starved adults for 24 hr by growing them on agar lacking amino acids ( aa ) ( Figure 4A ) . To analyse the induction of autophagy , we stained adult gut from these animals for p62 , a marker that is degraded by autophagy . If SNF4Aγ is required for autophagy , loss of SNF4Aγ should lead to p62 accumulation in mutant cells , as they should not induce autophagy . However , we observed a loss of p62 in mutant cells ( mCherry positive ) and robust staining of p62 in control cells ( EGFP positive , indicated by red arrows ) ( Figure 4B and Figure 4—figure supplement 1 ) . Interestingly , this parallels the observation by Sahani et al . ( 2014 ) in vertebrate cells . These authors observed that following prolonged amino acid starvation , the autophagy marker p62 is restored because of an increase in transcription and translation mediated by the availability of autophagy derived aa . Hence , we propose that loss of SNF4Aγ blocks autophagy , deprives cells of aa during starvation , leading to a loss of p62 . Similar data have been documented in Drosophila larvae and mammalian cells ( B'chir et al . , 2014; Colosetti et al . , 2009; Duran et al . , 2011; Erdi et al . , 2012 ) . 10 . 7554/eLife . 26420 . 014Figure 4 . Loss of SNF4Aγ in adults flies leads to autophagic and neurodegenerative phenotypes . ( A ) Flow chart describing the experimental outline to induce autophagy in the adult gut ( B ) Image displaying the adult midgut from amino acid-starved hsFLP; SNF4AγMI09417EGFP-T2A-mCherry/ SNF4AγMI09417 animals , stained for GFP ( a , d ) , mCherry ( b , d ) , p62 ( c , d ) and DAPI ( white ) ( d ) . Red arrows indicate control cells expressing SNF4Aγ-EGFP-SNF4Aγ , revealing higher levels of p62 . In contrast , mutant cells are mCherry positive and have reduced p62 levels , indicating a defect in autophagy induction , scale bar = 5 µm ( C ) Haemotoxylin and Eosin staining on brain sections of SNF4AγMI09417EGFP-T2A-mCherry/SNF4Aγ9417 ( control ) and hsFLP; SNF4AγMI09417EGFP-T2A-mCherry/ SNF4AγMI09417 experimental flies . hsFLP-induced cassette inversion in adult flies causes massive neurodegeneration , evident by the severe structural changes observed in brains of 3- and 10-day-old experimental animals ( b and d ) , whereas age-matched control brains show no sign of neurodegeneration ( a and c ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26420 . 01410 . 7554/eLife . 26420 . 015Figure 4—figure supplement 1 . P62 expression in hsFLP; SNF4AγMI09417EGFP-T2A-mCherry/ TM3 , Sb[1] animals . Image displaying the adult mid gut from amino acid starved hsFLP; SNF4AγMI09417EGFP-T2A-mCherry/ TM3 , Sb[1] animals , stained for GFP ( a , d ) , mCherry ( b , d ) , p62 ( c , d ) and DAPI ( white ) ( d ) . Red arrows indicate cells expressing SNF4Aγ-EGFP-SNF4Aγ ( hsFLP; SNF4AγMI09417EGFP-T2A-mCherry/ TM3 , Sb[1] ) , and white arrows indicate mCherry-expressing cells ( hsFLP; SNF4AγMI09417T2A-mCherry-EGFP/ TM3 , Sb[1] ) . Both the populations show similar levels of p62 . Scale bar = 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 26420 . 015 Previous studies have described that developmental loss of SNF4Aγ leads to neurodegeneration ( Cook et al . , 2012; Tschäpe et al . , 2002 ) . To further characterise the role of SNF4Aγ in Drosophila neurons , we used a post-mitotic GAL4 driver C155-GAL4 , to drive UAS-FLP and obtain GT orientated Flip Flops in all neurons . Such pan-neuronal flip of SNF4AγMI09417EGFP-T2A-mCherry into a non-functional GT leads to severe locomotor defects in adult flies ( data not shown ) . We also examined the role of SNF4Aγ in adults using hsFLP-mediated cassette inversion . Flies of the genotype hsFLP; SNF4AγMI09417EGFP- T2A-mCherry / SNF4AγMI09417 were raised at 18°C and were given a 2-hr heat-shock at 37°C , 3 days after eclosion . Two days after the heat-shock , H and E staining of adult brains revealed severe neurodegeneration . The extent of neuronal loss did not worsen by day 10 ( Figure 4C b and d ) . In addition , these flies typically die 3 to 4 weeks later . Following heat-shock , animals that do not carry hsFLP do not show obvious lesions in the brain ( Figure 4C a and c ) . We conclude that loss of SNF4Aγ leads to neuronal demise in adult flies . In conclusion , the Flip-Flop strategy allows conditional gene inactivation and generation of mosaics for a side-by-side comparison of mutant and wild-type cells in the same tissue , while simultaneously marking wild-type and mutant cells with different fluorescent markers . By using different sources of FLP , one can control the spatial and temporal pattern of mosaic generation and gene inactivation . Using hsFLP , we show that Flip-Flop can be used to severely inhibit gene function during development , as well as in adults in both mitotically active and post-mitotic cells . Flip-Flop can therefore reliably be used to explore gene functions both during development and in adult flies . Recently , a similar Flippase-based conditional gene inactivation method ‘FlpStop’ was published ( Fisher et al . , 2017 ) . Both the FlpStop and Flip-Flop strategies employ a similar Flex switch to induce local cassette inversion . However , there are several important differences between the two technologies . ( 1 ) FlpStop does not tag the endogenous protein with a fluorescent tag , while Flip-Flop produces an internally EGFP-tagged protein . While this may seem beneficial for genes where an internal tag is deleterious to the protein , we argue that for the majority of the genes the internal EGFP tag is not detrimental ( this study; Nagarkar-Jaiswal et al . , 2015 ) . In contrast , endogenous tagging is very advantageous as it permits to track the expression and subcellular localization of a protein , allows anti-GFP antibody-mediated immunoprecipitations followed by mass spectroscopy ( David-Morrison et al . , 2016; Yoon et al . , 2017 ) , enables ChIP sequencing ( Nègre et al . , 2011 ) , and permits mRNA or protein removal with iGFPi ( Neumüller et al . , 2012 ) or deGradFP ( Caussinus et al . , 2011; Nagarkar-Jaiswal et al . , 2015; Urban et al . , 2014 ) . Finally , the EGFP tag allows us for the first time to track protein perdurance in mutant cells . ( 2 ) The second major difference between FlpStop and Flip-Flop relates to how mutant cells are marked . In FlpStop , tdTomato is expressed under the control of the UAS/GAL4 system . Hence , cells that have undergone inversion will express tdTomato , irrespective of the expression pattern of the gene . In contrast , Flip-Flop will label cells with mCherry only if they express the gene of interest . In summary , the ability to track the loss of the protein of interest in mutant cells concomitant with loss of the EGFP signal while simultaneously marking the mutant cells with mCherry from the same regulatory elements distinguishes Flip-Flop from all other mosaic analyses methods . We would like to note that Flip-Flop insertions can be generated based on existing MiMIC insertions ( Nagarkar-Jaiswal et al . , 2015 ) as well as MiMIC-like elements introduced by CRISPR/Cas9 ( Zhang et al . , 2014 ) and can be extended to Drosophila cell culture system using RMCE ( Manivannan et al . , 2015 ) . We , therefore , believe that the use of Flip-Flop will permit functional annotation of numerous genes in unprecedented detail . Flip-Flop constructs were generated using the following steps: ‘attB-FRT14-FRT-SA- ( GGS ) 4-PhaseX-EGFP- ( GGS ) 4-SD-{SA-3xSTOP-SV40pA}reverse orientation-FRT14-FRT-attB’ for all three reading frames were synthesised in pUC57 ( Genewiz ) . The T2A-mCherry fragments with appropriate coding frame were amplified using PCR and cloned as AgeI/HindIII fragment in pUC57-“attB-FRT14-FRT-SA- ( GGS ) 4-PhaseX-EGFP- ( GGS ) 4-SD-{SA-3xSTOP-SV40pA} reverse orientation-FRT14-FRT-attB to create pFlip-Flop-P0 , pFlip-Flop-P1 and pFlip-Flop-P2 . RMCE was performed as described in Nagarkar-Jaiswal et al . , 2015 . Mosaics were generated in larval imaginal discs and brain tissue using FLP expressed under the control of eyeless promoter ( ey-FLP ) or using GAL4 drivers ( C155-GAL4 or nSyb-GAL4 ) to drive the expression of UAS-FLP . Corresponding crosses ( Figure 1—figure supplement 3 ) were raised at 25°C . For experiments where hsFLP was used to generate mosaics , embryos from corresponding crosses ( Figure 1—figure supplement 3 ) were collected for 24–30 hr . These embryos were given a heat-shock in a water bath at 37°C for one hr . Animals were raised at 25°C until the third larval instar stage before analysis . For adult mosaics , appropriate fly crosses ( Figure 1—figure supplement 2 ) were set up at 18°C . Three-day-old adult flies were heat shocked at 37°C in an air incubator for 2 hr . For ovaries ( Figure 1—figure supplement 3c ) , females were dissected 4 days after eclosion . For adult brains ( Figure 4B ) , adults were recovered at 25°C and dissected on days 3 or 10 . Larval brain , imaginal discs , and adult brain staining were performed as described in Nagarkar-Jaiswal et al . , 2015 , adult ovary staining was performed as described in Urban et al . ( 2014 ) . Adult eye staining was performed as described in Jaiswal et al . ( 2015 ) . Primary antibodies: chicken anti-GFP 1:500 ( Abcam , ab13970; RRID:AB_300798 ) , rabbit anti-DsRed 1:500 ( Clontech , 632496; RRID:AB_10013483 ) , rabbit anti p62 1:2000 ( David-Morrison et al . , 2016 ) and rat anti-Elav 1:500 ( DSHB , 7E8A10 ) ( O'Neill et al . , 1994 ) . Secondary antibodies: Alexa 488 ( RRID:AB_142924; Invitrogen , Life Technologies , Grand Island , NY ) , Cy5 ( RRID:AB_2338072 and RRID:AB_2338393 ) and Cy3 ( RRID:AB_2338059 ) conjugated secondary antibodies ( Jackson ImmunoResearch , West Grove , PA ) were used at 1:500 . Phalloidin conjugated with Alexa 647 ( Invitrogen ) was used at 1:500 . Recordings were performed as described in Nagarkar-Jaiswal et al . ( 2015 ) . Adult fly heads were fixed in 8% glutaraldehyde ( EM grade ) and embedded in paraffin . Sections ( 10 µm ) were prepared by a microtome ( Leica ) and stained with Hematoxylin and Eosin as described in Chouhan et al . , 2016 . At least three animals were examined for each genotype .

The instructions needed to build and maintain cells in an organism are encoded in their DNA . There are many different cell types , and each type only needs a small portion of the information found in the DNA to do its job . Hence , only some of the instructions , in the form of genes , need to be active or ‘expressed’ in any given cell type . To understand how a gene works , it is necessary to know in which cell the gene is expressed and where in the cell the gene product – normally a protein – is located . Researchers may study a gene by deleting it , which prevents the protein from being made , or by attaching a new instruction into the gene , which generates a fluorescent tag on the protein to determine where and when it is expressed . Until now , it was not possible to selectively inactivate a gene and simultaneously mark both normal cells containing the protein and mutant cells lacking the protein . Based on an existing tagging approach , Nagarkar-Jaiswal et al . have now developed a method in which normal and mutant cells of fruit flies are marked differently . A gene of interest is tagged with a fluorescent marker called green fluorescent protein ( or GFP ) . The same gene is then inactivated in some of the cells , which are tagged with a red marker called mCherry . Nagarkar-Jaiswal et al . compared normal and mutant cells , and were able to determine how long it takes before the mutant cells become abnormal . With this new method , the role of numerous genes in any tissue of adult flies can be reassessed . This will allow to investigate what happens when a protein is removed in specific cells in adult flies . A future goal will be to apply this method to other animals that are more closely related to humans , such as mice , to gain a clearer picture of the role of genes in different cell types and how faulty genes may cause disease .

2017

A cell cycle-independent, conditional gene inactivation strategy for differentially tagging wild-type and mutant cells

Meristems contain groups of indeterminate stem cells , which are maintained by a feedback loop between CLAVATA ( CLV ) and WUSCHEL ( WUS ) signaling . CLV signaling involves the secretion of the CLV3 peptide and its perception by a number of Leucine-Rich-Repeat ( LRR ) receptors , including the receptor-like kinase CLV1 and the receptor-like protein CLV2 coupled with the CORYNE ( CRN ) pseudokinase . CLV2 , and its maize ortholog FASCIATED EAR2 ( FEA2 ) appear to function in signaling by CLV3 and several related CLV3/EMBRYO-SURROUNDING REGION ( CLE ) peptide ligands . Nevertheless , how signaling specificity is achieved remains unknown . Here we show that FEA2 transmits signaling from two distinct CLE peptides , the maize CLV3 ortholog ZmCLE7 and ZmFON2-LIKE CLE PROTEIN1 ( ZmFCP1 ) through two different candidate downstream effectors , the alpha subunit of the maize heterotrimeric G protein COMPACT PLANT2 ( CT2 ) , and ZmCRN . Our data provide a novel framework to understand how diverse signaling peptides can activate different downstream pathways through common receptor proteins . Stem cell proliferation and differentiation throughout plant life is regulated by a feedback loop between the homeodomain transcription factor WUS and CLV ligand-receptor signaling ( Mayer et al . , 1998; Brand et al . , 2000; Schoof et al . , 2000; Yadav et al . , 2011; Daum et al . , 2014 ) . The secretion of the diffusible glycopeptide CLV3 from the central zone ( CZ ) stem cells of the SAM is believed to initiate signaling through LRR receptors ( Fletcher et al . , 1999; Rojo et al . , 2002; Kondo et al . , 2006; Ohyama et al . , 2009; Nimchuk et al . , 2011b ) , which transmits the signal to restrict the expression of WUS in the organizing center ( OC ) cells . To balance this system , WUS non-cell-autonomously promotes stem cell fate by activation of CLV3 expression ( Yadav et al . , 2011; Daum et al . , 2014 ) . CLV3 is thought to be perceived by multiple receptor kinase and receptor like proteins , including the CLV1 LRR receptor-like kinase ( Clark et al . , 1993; Clark et al . , 1997; Brand et al . , 2000; Ogawa et al . , 2008 ) and the related BAM receptors ( DeYoung et al . , 2006; Deyoung and Clark , 2008; Nimchuk et al . , 2015; Shinohara and Matsubayashi , 2015 ) , or by a heterodimer of the receptor like protein CLV2 and the transmembrane pseudokinase CRN ( Kayes and Clark , 1998; Jeong et al . , 1999; Miwa et al . , 2008; Müller et al . , 2008; Bleckmann et al . , 2010; Zhu et al . , 2010; Nimchuk et al . , 2011a ) , or by the receptor-like kinase RPK2 ( Mizuno et al . , 2007; Nodine et al . , 2007; Kinoshita et al . , 2010 ) . The relationship between CLV1 and CLV2 is not clear- CLV1 can form homodimers , or higher order complexes with CLV2/CRN , to signal co-operatively in the SAM ( Guo et al . , 2010; Somssich et al . , 2015 ) , but it seems that CLV2/CRN is not essential for CLV3 perception or for CLV1 signaling ( Müller et al . , 2008; Nimchuk et al . , 2011b; Nimchuk , 2017 ) . In contrast to CLV1 , CLV2 does not bind CLV3 peptide directly ( Shinohara and Matsubayashi , 2015 ) , and its expression is not restricted to the SAM , suggesting that it might function as a co-receptor in additional pathways beyond CLV3 signaling . Indeed , CLV2 appears to be involved in signaling by several CLE peptides ( Fiers et al . , 2005; Meng and Feldman , 2010; Hazak et al . , 2017 ) and in biotic interactions ( Replogle et al . , 2011; Hanemian et al . , 2016 ) , suggesting it plays diverse functions in plant development and immunity ( Pan et al . , 2016 ) . The multiple roles of CLV2 promote the question of how it confers signal specificity . Two candidate downstream effectors of CLV2 have been identified . One is the transmembrane pseudokinase CRN , discovered in Arabidopsis , and the second is COMPACT PLANT2 ( CT2 ) , the heterotrimeric G protein alpha subunit , discovered in maize ( Bommert et al . , 2013a ) . However , since CRN and CT2 were identified in different species , their molecular and genetic interactions remain unknown . The CLV-WUS pathway is widely conserved ( Somssich et al . , 2016; Soyars et al . , 2016 ) . In maize , THICK TASSEL DWARF1 ( TD1 ) and FEA2 are CLV1 and CLV2 orthologs , and function similarly to restrict inflorescence shoot meristem proliferation ( Taguchi-Shiobara et al . , 2001; Bommert et al . , 2005 ) . Two maize WUS orthologs , ZmWUS1 and ZmWUS2 , have been predicted by phylogenetic analysis , and a ZmWUS1 reporter is expressed in the presumptive organizing center of the inflorescence shoot meristem ( Je et al . , 2016 ) , but these genes have not been functionally characterized ( Nardmann and Werr , 2006 ) . In rice , FLORAL ORGAN NUMBER 1 ( FON1 ) , the CLV1 ortholog , and FON2 , the CLV3 ortholog , similarly function in floral development in a common pathway , as expected ( Suzaki et al . , 2004; Chu et al . , 2006; Suzaki et al . , 2006; Suzaki et al . , 2008; Suzaki et al . , 2009 ) , whereas a second rice CLE peptide gene , FON2-LIKE CLE PROTEIN1 ( FCP1 ) controls stem cell proliferation independent of FON1 ( Suzaki et al . , 2008 ) . The rice WUS homolog , TILLERS ABSENT1/MONOCULM3 functions in axillary shoot meristem formation ( Tanaka et al . , 2015; Lu et al . , 2015 ) , and WUS function in the shoot apical meristems appears to have been taken over by the WUSCHEL RELATED HOMEOBOX4 ( WOX4 ) gene ( Ohmori et al . , 2013 ) . How specificity is achieved is a common question in signal transduction pathways . Recently , we identified a distinct CLV receptor , FASCIATED EAR3 ( FEA3 ) in maize and Arabidopsis , and found that FEA3 controls responses to the maize FCP1 ( ZmFCP1 ) CLE peptide ( Je et al . , 2016 ) . Here , we show that the maize CLV2 ortholog FEA2 also participates in ZmFCP1 signaling , in addition to controlling responses to the maize CLV3 ortholog , ZmCLE7 ( Je et al . , 2016 ) . To ask how specificity from these different CLE peptide inputs is achieved , we first isolated mutant alleles of the maize CRN gene . Consistent with results in Arabidopsis ( Miwa et al . , 2008; Müller et al . , 2008; Bleckmann et al . , 2010; Zhu et al . , 2010; Nimchuk et al . , 2011a ) , we found that fea2 was epistatic to Zmcrn in control of meristem size , but Zmcrn;ct2 double mutants showed an additive enhanced phenotype , suggesting they act in parallel pathways , despite the fact that FEA2 binds both ZmCRN and CT2 in co-immunoprecipitation ( co-IP ) experiments . Strikingly , ct2 and Zmcrn mutants were resistant to different CLE peptides , ZmCLE7 and ZmFCP1 , respectively , but fea2 was resistant to both , suggesting that FEA2 controls responses to different CLE peptides by acting through different downstream effectors . We recently described a new CLE signaling pathway in maize , in which ZmFCP1 peptide signals through FEA3 to restrict ZmWUS1 expression from below its organizing center expression domain ( Je et al . , 2016 ) . To test this model , we used a 2-component transactivation system ( Wu et al . , 2013; Je et al . , 2016 ) to drive ZmFCP1 expression in developing primordia , below the ZmWUS1 domain ( Je et al . , 2016; Nardmann and Werr , 2006 ) . As previously described , this expression reduced meristem size of wild type SAMs ( Je et al . , 2016 ) , however we found that meristem size was only partially rescued when ZmFCP1 expression was transactivated in a fea3 mutant background ( Figure 1A and B ) , suggesting that ZmFCP1 signals through additional receptors . We therefore conducted peptide response assays using fea2 mutants , and found that they were also insensitive to ZmFCP1 peptide treatment , as well as to ZmCLE7 , the maize CLV3 ortholog ( Figure 1C ) ( Je et al . , 2016 ) . Interestingly , fea2;fea3 double mutants restored the size of ZmFCP1 treated meristems to control levels , suggesting that ZmFCP1 signaling is transmitted predominantly through both FEA2 and FEA3 ( Figure 1D ) . fea3 mutants are resistant only to ZmFCP1 , and not to ZmCLE7 ( Je et al . , 2016 ) , so we next asked how FEA2 might transmit signals from different CLE peptides . In maize , FEA2 signals through CT2 , the alpha subunit of the heterotrimeric G protein ( Bommert et al . , 2013a ) , but in Arabidopsis the FEA2 ortholog CLV2 is thought to signal through a membrane bound pseudokinase , CRN ( Miwa et al . , 2008; Müller et al . , 2008; Bleckmann et al . , 2010; Zhu et al . , 2010; Nimchuk et al . , 2011a ) . To ask if CRN also functions in CLV signaling in maize , we identified maize CRN ( ZmCRN ) by phylogenic analysis ( Figure 2—figure supplement 1A ) . As is the case for Arabidopsis CRN , ZmCRN was also predicted to encode an inactive pseudokinase ( Figure 2—figure supplement 1B ) ( Boudeau et al . , 2006; Nimchuk et al . , 2011a ) . We identified a predicted null allele as a Mu transposon insertion from the Trait Utility System in Corn ( TUSC ) resource ( McCarty and Meeley , 2009 ) , 52 bp downstream of the predicted translation start site ( Figure 2A ) . We backcrossed this Mu insertion line three times to the standard B73 inbred line , and dissected homozygous mutant or normal sib samples for meristem analysis . The maize crn ( Zmcrn ) mutants had larger vegetative shoot meristems ( 130 . 0 ± 4 . 1 μm , compared to 109 . 2 ± 4 . 6 μm for normal sibs , P value < 0 . 0001 , two-tailed t test , Figure 2B and C ) , and developed fasciated ear primordia with enlarged and split inflorescence meristems ( Figure 2D ) , reminiscent of other fasciated ear mutants ( Taguchi-Shiobara et al . , 2001; Bommert et al . , 2005; Bommert et al . , 2013a; Je et al . , 2016 ) . Concurrently , we identified a second candidate allele by map-based cloning of a fasciated mutant , fea*148 ( Figure 2—figure supplement 2A ) , from an ethyl methyl sulfonate ( EMS ) screen in the B73 background ( hereafter Zmcrn-148 ) . Zmcrn-148 introduced a stop codon within the predicted pseudokinase domain ( Figure 2A ) , and plants homozygous for this mutation developed a similar fasciated ear phenotype ( Figure 2—figure supplement 2C ) . We next crossed heterozygous Zmcrn-148 plants with Zmcrn mutants . The F1 plants developed fasciated ears , while Zmcrn/+ or Zmcrn-148 /+ heterozygotes had normal ear primordia , suggesting that these mutations were allelic ( Figure 2—figure supplement 3 ) , and confirming that CRN functions in shoot meristem size control in maize , similar to its role in Arabidopsis . ZmCRN was expressed throughout the SAM and more strongly in the peripheral domain and leaf primordia ( Figure 2E , confirmed by laser capture microdissection RNAseq , Figure 2—figure supplement 4 ) . Next , since fea2 and other fea mutants are associated with quantitative variation in kernel row number ( KRN ) ( Bommert et al . , 2013b ) , we took advantage of the identification of ZmCRN to ask if it was also associated with this yield trait . We conducted a candidate gene association study using a maize association panel of 368 diverse inbred lines ( Li et al . , 2013; Liu et al . , 2015 ) . We found that three SNPs in the 3’UTR region of CRN showed significant association with KRN in multiple environments , below the threshold p-value<0 . 001 ( Figure 2—figure supplement 5 and Supplementary File 1 ) . These results suggest that natural variation in ZmCRN may underlie subtle variation in inflorescence meristem size sufficient to enhance KRN , with the potential to benefit maize yields . In Arabidopsis , CRN is thought to signal downstream of CLV2 and correspondingly the double mutants show an epistatic interaction ( Müller et al . , 2008 ) . To ask if this relationship was conserved in maize , we measured the SAM size in a segregating double mutant population . As expected , both Zmcrn and fea2 vegetative meristems were larger than normal ( 166 . 3 ± 8 . 3 μm , or 176 . 1 ± 9 . 8 μm respectively , compared to 139 . 7 ± 4 . 8 μm for normal sibs , P value < 0 . 0001 , two-tailed t test , Figure 3A and B ) , and the Zmcrn; fea2 double mutants ( 177 . 2 ± 13 . 3 μm ) were similar to the fea2 single mutants ( 176 . 1 ± 9 . 8 μm , P value = 0 . 68 , two-tailed t test ) ( Figure 3A and B ) . We also characterized ear inflorescence meristems and found that fea2 had stronger fasciated ears than those of Zmcrn , but the double mutants resembled fea2 single mutants ( Figure 3C ) . Together , these results indicate that fea2 is epistatic to Zmcrn , suggesting that FEA2 and ZmCRN function in a common pathway in maize , as in Arabidopsis . We next asked if ZmCRN and CT2 function in the same or in different pathways , again by double mutant analysis . Both Zmcrn and ct2 mutants had larger SAMs compared with their normal sibs ( 161 . 5 ± 10 . 6 μm , or 157 . 1 ± 11 . 8 μm respectively , compared to 139 . 7 ± 8 . 5 μm for normal sibs , P value < 0 . 0001 , two-tailed t test , Figure 3D and E ) , but the SAMs of double mutants were significantly larger than each single mutant ( 191 . 8 ± 18 . 6 μm , P value < 0 . 0001 , two-tailed t test , Figure 3D and E ) , suggesting an additive interaction . Zmcrn; ct2 double mutant ear inflorescences also showed additive enhancement in fasciation , compared to each single mutant ( Figure 3F ) , confirming the additive interaction between ct2 and Zmcrn . In summary , double mutant analyses and quantification of meristem sizes indicated that ZmCRN functions in the same pathway as FEA2 and , as previously reported , CT2 also functions in the same pathway as fea2 ( Bommert et al . , 2013a ) , but CT2 and ZmCRN themselves function in different pathways . This result is most easily explained by the hypothesis that FEA2 functions in two different pathways , one with CT2 and a second with ZmCRN . To test the two-pathway hypothesis , we tested protein-protein interactions using Co-IP assays . We used an internal YFP fusion of CT2 that we previously found to be biologically active ( Bommert et al . , 2013a ) , and C terminal mCherry or Myc fusions of ZmCRN or FEA2 , respectively , which are predicted to be correctly localized and active , based on similar fusions ( Bleckmann et al . , 2010; Nimchuk , 2017 ) . We first confirmed the expected plasma membrane localization of ZmCRN-mCherry by transient expression and plasmolysis ( Figure 4A ) , consistent with FEA2 and CT2 localization ( Bommert et al . , 2013a ) . ZmCRN-mCherry also co-localized with FEA2-YFP and CT2-YFP on the plasma membrane when they were co-expressed ( Figure 4—figure supplement 1 ) . We then tested pairwise interactions using Co-IP experiments following transient expression . ZmCRN-mCherry was able to pull down FEA2-Myc , but not CT2-YFP , even when FEA2-YFP was also co-expressed ( Figure 4B ) . We confirmed that CT2-YFP was properly expressed , because it could pull down FEA2-Myc ( Figure 4C ) , as previously demonstrated by in vivo co-IPs ( Bommert et al . , 2013a ) . To validate these interactions , a reciprocal Co-IP experiment was carried out , in which all three proteins were co-expressed , and we consistently found that FEA2-Myc could IP CT2-YFP or ZmCRN-mCherry ( Figure 4D ) , further confirming that FEA2 formed complexes with both CT2 and ZmCRN . As an independent test , we also used an optimized BiFC system , with monomeric Venus ( mVenus ) split at residue 210 to reduce background due to false positive interactions ( Gookin and Assmann , 2014 ) . We detected YFP signal when FEA2 fused with the N terminal part of mVenus ( NmVen210 ) was co-expressed with ZmCRN fused with the C terminal part ( CmVen210 ) ( Figure 4—figure supplement 2 ) , confirming a direct interaction between FEA2 and ZmCRN . Similar results were reported in Arabidopsis using BiFC to detect CRN-CLV2 interactions ( Zhu et al . , 2010 ) . However , we failed to detect a YFP signal when FEA2-NmVen210 was co-expressed with CT2-CmVen210 ( Figure 4—figure supplement 2 ) . The interaction between FEA2 and CT2 is well documented in maize by in vivo Co-IP experiments ( Bommert et al . , 2013a ) , and a failure to detect the same interaction using BiFC suggests that their interaction might be indirect , such as in a complex where their interaction is bridged by other protein ( s ) . Lastly , as expected , no signal was detected when CT2-NmVen210 was co-expressed with ZmCRN-CmVen210 ( Figure 4—figure supplement 2 ) , confirming out Co-IP results , and supporting the hypothesis that they do not interact . The FEA2-ZmCRN and FEA2-CT2 interactions appeared to be quite stable , and were not affected by co-infiltration of CLE peptides ( Figure 4—figure supplement 3 ) . In summary , the FEA2 receptor-like protein interacted with both candidate signaling molecules , ZmCRN and CT2 , but these interactions appeared to be in different protein complexes , rather than in a common complex , because ZmCRN was not able to immunoprecipitate CT2 . The activity of CLE peptides can be assayed using synthetic peptide treatments , which suppress the growth of the SAM and root apical meristem ( Ito et al . , 2006; Kondo et al . , 2006 ) . We therefore tested the sensitivity of each mutant to different CLE peptides , using embryo culture , as previously described ( Bommert et al . , 2013a; Je et al . , 2016 ) . ct2 or Zmcrn segregating populations were grown in the presence of different peptides , and shoots fixed and cleared for SAM measurements after 12 days . We found that ct2 mutants were partially resistant to ZmCLE7 , but not to ZmFCP1 peptide ( Figure 5A and B ) , suggesting that CT2 functions specifically in signaling by ZmCLE7 , the maize CLV3 ortholog . In contrast , we found that Zmcrn mutants were partially resistant to ZmFCP1 , but not to ZmCLE7 ( Figure 5C and D ) , suggesting that ZmCRN functions specifically in a ZmFCP1 signaling pathway . To confirm these results , we treated each mutant with both ZmCLE7 and ZmFCP1 together . We found that only fea2 , but not ct2 or Zmcrn mutants , showed resistance to the double peptide treatment ( Figure 5E and F ) . Together , these results suggest that FEA2 functions in both ZmCLE7 and ZmFCP1 signaling pathways , but CT2 and ZmCRN function specifically in ZmCLE7 or in ZmFCP1 signaling , respectively . As FEA3 also acts to transmit the ZmFCP1 signal ( Je et al . , 2016 ) , we used genetic analysis to ask if ZmCRN also functions downstream of FEA3 . In a segregating double mutant population , the SAMs of Zmcrn and fea3 mutants were both larger than normal , as expected ( 160 . 2 ± 6 . 7 μm , or 176 . 8 ± 8 . 2 μm respectively , compared to 142 . 6 ± 6 . 0 μm for normal sibs , P value < 0 . 0001 , two-tailed t test ) , and the fea3; Zmcrn double mutants were larger than the single mutants ( 221 . 5 ± 21 . 2 μm , P value < 0 . 0001 , two-tailed t test ) , suggesting that FEA3 and ZmCRN do not function in a common pathway ( Figure 5—figure supplement 1A and B ) . Similar findings were observed for fea3; ct2 double mutants ( Figure 5—figure supplement 1C and D ) , suggesting that FEA3 and CT2 also do not function in a common pathway . Thus , ZmFCP1 signaling appears to be mediated by two different pathways , one acting through FEA2 coupled with ZmCRN , and another acting through FEA3 working through as yet unknown downstream component ( s ) . In summary , through identification of maize crn mutants , we were able to show that signaling through FEA2 by two different CLE peptides is differentiated using different candidate downstream signaling components; with the ZmCLE7 signal passing through CT2 and the ZmFCP1 signal passing through ZmCRN ( Figure 6 ) . A major question in signal transduction is how multiple inputs can be translated into distinct outputs . CLV-WUS feedback signaling is the central regulatory pathway in shoot meristem development , and perception of CLV3 peptide involves the CLV1 receptor-like kinase and the CLV2 receptor-like protein together with the CRN pseudokinase ( Brand et al . , 2000; Schoof et al . , 2000; Miwa et al . , 2008; Müller et al . , 2008; Bleckmann et al . , 2010; Zhu et al . , 2010; Nimchuk et al . , 2011a ) . However , genetic evidence in both maize and Arabidopsis suggests these receptors function independently , and CLV2 , and its maize ortholog FEA2 , respond to multiple CLE peptides ( Bommert et al . , 2005; Fiers et al . , 2005; Müller et al . , 2008; Guo et al . , 2010; Meng and Feldman , 2010; Je et al . , 2016; Hazak et al . , 2017 ) . So how is the information conferred by these different signals kept separate during transmission through a common receptor ? To address this question and further decipher the FEA2 signaling pathway , we isolated mutants in the maize CRN ortholog , ZmCRN , by reverse genetics and by cloning a newly identified fasciated ear mutant fea*148 . ZmCRN was predicted to encode a membrane localize pseudokinase , like CRN in Arabidopsis ( Nimchuk et al . , 2011a ) , and characterization of the mutants indicated that ZmCRN similarly functioned as a negative regulator of stem cell proliferation . We found that fea2 was epistatic to Zmcrn and that FEA2 and ZmCRN interacted directly , using Co-IP and BiFC assays of proteins transiently overexpressed in N . benthamiana , suggesting that ZmCRN is a signaling component in the FEA2 pathway . Natural variation in the CLV-WUS pathway underlies yield improvements in different crop species including tomato , maize and mustard ( Bommert et al . , 2013b; Fan et al . , 2014; Xu et al . , 2015; Je et al . , 2016 ) , and FEA2 is a quantitative trait locus ( QTL ) for kernel row number ( KRN ) ( Bommert et al . , 2013b ) . In this study , we used a maize association panel of 368 diverse inbred lines to show that ZmCRN also had significant association with KRN under multiple environments ( Li et al . , 2013; Liu et al . , 2015 ) , suggesting that ZmCRN contributes to quantitative variation in this trait . Therefore , ZmCRN could be manipulated for maize yield enhancement . Previously , we identified the alpha subunit of the heterotrimeric G protein , CT2 , as an additional interactor of FEA2 . fea2 is epistatic to ct2 in meristem regulation , similar to its genetic interaction with Zmcrn , and FEA2 interacts with CT2 in vivo , revealing that CT2 , like ZmCRN , is a candidate downstream signaling component of FEA2 ( Bommert et al . , 2013a ) . Although fea2 was epistatic both to ct2 and to Zmcrn , we found that ct2; Zmcrn double mutants had an additive interaction , suggesting they function in parallel , and that the FEA2 signaling pathway branches into these two different downstream signaling components . This idea was supported by peptide assays in different mutants , which suggested that ZmCRN and CT2 function specifically in ZmFCP1 or ZmCLE7 signaling , respectively , while FEA2 is involved in both . Although we used high peptide concentrations , the activity of CLE peptides is known to be enhanced by triarabinosylation ( Ohyama et al . , 2009; Matsubayashi , 2011; Xu et al . , 2015; Corcilius et al . , 2017 ) , and indeed we found that similarly modified ZmCLE7 peptide was about 10 fold more potent than the non-modified form ( Figure 5—figure supplement 2 ) . Consistently with our findings , ZmCRN , CT2 and FEA2 were expressed broadly in the SAM in overlapping domains ( Figure 2—figure supplement 5 ) . These data suggest a novel mechanism in plant receptor signaling , where a single receptor , FEA2 , can transmit signals from two different CLE peptides , ZmFCP1 and ZmCLE7 , through two different downstream components , ZmCRN and CT2 . We thereby shed light on how distinct signaling by different peptides can be achieved through a common receptor . As a candidate receptor or co-receptor for different peptides , FEA2 does not have any close homologs in the maize genome ( Figure 6—figure supplement 1 ) , similar to CLV2 in Arabidopsis , and the relatively mild phenotype of fea2 mutants may be due to compensation by partially redundant parallel signaling pathways , such as through FEA3 ( Je et al . , 2016 ) . Our results are largely consistent with findings in Arabidopsis , that CRN is dispensable for CLV3 perception and signaling ( Nimchuk , 2017 ) , and that CLV2/CRN can function with other CLE ligand-receptor complexes ( Hazak et al . , 2017 ) . However , in Arabidopsis CRN is required for CLV2 trafficking to the plasma membrane ( Bleckmann et al . , 2010 ) . Our results suggest that the maize CLV2 ortholog FEA2 still functions ( with CT2 ) in a crn mutant , so is presumably on the plasma membrane even in the absence of ZmCRN . How then can a single receptor recognize different signals and transmit them differentially ? The most obvious answer depends on the hypothesis that FEA2 and CLV2 are co-receptors that function with LRR RLKs , which binds CLE peptides directly ( Figure 6 ) . This hypothesis is supported by the finding that CLV1 binds CLV3 with high affinity , but CLV2 is unable to bind CLE peptides ( Shinohara and Matsubayashi , 2015 ) , and that CLV2/CRN can function with different CLE ligand-receptor complexes ( Hazak et al . , 2017 ) . There are conflicting results surrounding the interaction between CLV2 and CLV1; some experiments detect their physical interaction , but many of them use over-expression and are prone to false positive results , and clv2 and clv1 act additively in double mutant combinations ( Kayes and Clark , 1998; Müller et al . , 2008 ) . This genetic result suggests they act separately , and the same is true for the orthologs FEA2 and TD1 in maize ( Bommert et al . , 2005 ) . A possible explanation for these conflicting findings is that CLV2 may act with multiple CLE receptor RLKs . This model is supported by the observation that CLV1 homologs , the BAMs , function redundantly with CLV1 , so multiple LRR RLKs do indeed function in meristem size control . This also explains why all intermediate and strong clv1 alleles are dominant negative , as they likely interfere with the activity of other receptor kinase ( s ) that have functional overlap with CLV1 ( Diévart et al . , 2003; Nimchuk et al . , 2015 ) . Despite not knowing the details of specific CLE-receptor interactions , our data show that FEA2 can transmit different peptide signals through two distinct downstream signaling components that most likely converge on the regulation of ZmWUS expression to regulate stem cell proliferation in meristem development ( Figure 6 ) . This suggests a new working model for meristem size regulation , in which ligand binding can be transmitted by a common co-receptor working with different RLKs coupled to distinct signaling proteins . Our model differs from most well-studied ligand-receptor signaling pathways , in which the signaling pathways usually converge ( Couto and Zipfel , 2016 ) . For instance , different microbial ligands such as flagellin and Elongation Factor Thermo unstable ( EF-Tu ) are specifically recognized by the FLAGELLIN-SENSITIVE 2 ( FLS2 ) -BRI1 ASSOCIATED RECEPTOR KINASE ( BAK1 ) or EF-Tu RECEPTOR ( EFR ) -BAK1 RLK complexes , respectively , while signal transduction requires a shared set of cytosolic kinases , including BOTRYTIS-INDUCED KINASE 1 ( BIK1 ) ( Aarts et al . , 1998; Lu et al . , 2010 ) . Nevertheless , a similar principle can be drawn from the different signaling pathways mediated by BAK1 , which functions as a co-receptor for the brassinosteroid ( BR ) receptor , BR INSENSITIVE 1 ( BRI1 ) or for FLS2 . After ligand perception , BR signaling through the BAK1-BRI1 complex is transmitted through the receptor-like cytoplasmic kinase ( RLCK ) BRASSINOSTEROID-SIGNALING KINASE 1 ( BSK1 ) , and flagellin signaling through the BAK1-FLS2 complex is transmitted through a different RLCK , BIK1 ( Li et al . , 2002; Nam and Li , 2002; Chinchilla et al . , 2007; Lu et al . , 2010; Wang , 2012; Sun et al . , 2013 ) . Our study also reveals another source of variation in meristem receptor signaling , by highlighting the role of an additional CLE peptide , ZmFCP1 . The role of FCP1 in meristem maintenance has been characterized in both maize and rice ( Suzaki et al . , 2008; Je et al . , 2016 ) , but not yet in Arabidopsis . In summary , multiple receptor signaling pathways appear to be required to for the perception of different CLE peptide signals to fine tune meristem development . This complex system of multiple peptides , receptors and downstream components presumably confers robustness on the meristem structure , as well as providing flexibility to control meristem development according to different physiological or developmental cues . For example , meristem size responds to stress and developmental transitions , such as floral induction , and different signaling pathways may confer such responsiveness . Our results help explain how meristem size regulation is orchestrated by multiple CLE peptides and receptors , as observed in many species including Arabidopsis , rice , maize and tomato ( Ito et al . , 2006; Strabala et al . , 2006; Suzaki et al . , 2009; Nimchuk et al . , 2015; Xu et al . , 2015 ) . They also support the idea that meristem signaling components are highly conserved between diverse plant species , and a major challenge is to understand how differential regulation of these common components leads to diversity in meristem organization and size across diverse plant taxa . Maize plants were grown in the field or in the greenhouse . The Zmcrn Mu insertion allele was isolated from TUSC lines and was backcrossed three generations to the standard B73 inbred line . The fea*148 allele was isolated in an EMS mutagenesis screen using F2 seed stocks prepared by Prof . Gerald Neuffer , derived from a cross of mutagenized B73 pollen onto A619 ears . One fasciated plant from the segregating fea*148 M2 population from the maize GDB stock center was crossed to the A619 inbred , then selfed to make an F2 segregating population . Pooled DNAs from ~50 mutants or the same number of normal ear plants screened from the segregating F2 population were used for bulked segregant analysis ( BSA ) using a maize SNP50 chip ( Illumina , Inc . ) . The BSA analysis revealed a clear linkage of the mutation on Chromosome 3 at 153–158 Mbp . As ZmCRN was an obvious candidate gene within the region , we sequenced the locus of ZmCRN using the mutant pool DNA and found a C to T mutation in the pseudokinase domain , which led to an early stop codon . To measure meristem size , segregating siblings were genotyped and shoot apices of 7-day-old plants ( Figure 2B ) or 21-day-old plants ( Figure 3A and D ) were dissected , cleared and measured as described previously ( Taguchi-Shiobara et al . , 2001 ) . Measurement was made blindly without the knowledge of the genotypes . All measurements included at least 10 samples of each genotype , and two or three independent biological replicates , and mean values ± s . d . were presented , with significance calculated using two-tailed , two-sample t tests , and significant differences reported as P values . Scanning electron microscopy was performed on fresh tissues of maize using a Hitachi S-3500N SEM , as described ( Taguchi-Shiobara et al . , 2001 ) . For confocal microscopy , tobacco infiltrated tissues were dissected and images were taken with a Zeiss LSM 710 microscope , using 561 nm laser excitation and 580–675 nm emission for detection ZmCRN-mCherry , using 512 nm laser excitation and 518–538 nm emission for detection of CT2-YFP and FEA2-YFP and for BiFC imaging . For plasmolysis of ZmCRN-mCherry , leaf tissues were incubated for 30 min with 800 mM mannitol and imaged . Double mutants were constructed by crossing mutants introgressed into B73 , followed by selfing or backcrossing to the F1 . All plants were subsequently genotyped ( primers are listed in Supplementary file 2 ) . In situ hybridization experiments were performed as described ( Jackson et al . , 1994 ) . Antisense and sense RNAs for ZmCRN were transcribed and used as probes . Primers are listed in Supplementary file 2 . CT2-YFP , ZmCRN-mCherry , or FEA2-Myc expression constructs were infiltrated into 4-week-old Nicotiana benthamiana leaves together with a P19 plasmids to suppress posttranscriptional silencing ( Mohammadzadeh et al . , 2016 ) . The protein extraction and membrane fraction enrichment were described in Bommert et al . , 2013a . Briefly , the infiltrated leaves were harvested 3-d post infiltration . The leaf tissues were ground in liquid nitrogen to a fine powder then suspended in twice the volume of protein extraction buffer containing 150 mM NaCl , 50 mM Tris-HCl pH 7 . 6 , 5% glycerol , and EDTA-free Protease inhibitor cocktail ( Roche ) . After filtration through Miracloth , and centrifugation at 4 , 000 g for 10 min at 4°C , the extract was centrifuged at 100 , 000 g for 1 hr at 4°C to enrich the microsomal membrane fraction . The resulting pellet was re-suspended in the extraction buffer supplemented with 1% Triton X-100 . Lysates were cleared by centrifugation at 100 , 000 g for 30 min at 4°C to remove non-solubilized material . ZmCRN-mCherry was immunoprecipitated using RFP-Trap ( Chromotek ) in membrane solubilization buffer for 40 min followed by washing 3 times with 1 ml of the same buffer . The IP’d proteins were eluted with 50 μl 1xSDS loading buffer at 95°C , followed by standard SDS-PAGE electrophoresis and western blotting . FEA2-Myc was immunoprecipitated using agarose beads conjugated with anti-Myc antibody ( Millipore , 16–219 , RRID:AB_390197 ) . ZmCRN-mCherry was detected using an anti-RFP antibody ( Rockland , 600-401-379 , RRID:AB_2209751 ) , FEA2-Myc was detected using an anti-Myc antibody ( Millipore , 05–724 , RRID:AB_309938 ) , and CT2-YFP was detected using an anti-GFP antibody ( Roche , 11814460001 , RRID:AB_390913 ) . Maize embryos segregating for each mutant were dissected at 10 days after pollination , when the SAM was exposed , and cultured on gel media ( Bommert et al . , 2013a ) containing scrambled peptide ( 30 μM; Genscript ) or ZmFCP1 peptide or ZmCLE7 peptide or a mixture of ZmCLE7 and ZmFCP1 peptides ( Je et al . , 2016 ) . After 12 days , the tissues were harvested for genotyping and the embryos were fixed in FAA ( 10% , formalin , 5% acetic acid , 45% ethanol ) and cleared in methyl salicylate , and SAMs measured by microscopy , as described ( Je et al . , 2016 ) . Triarabinosylated peptides were synthesized as described ( Corcilius et al . , 2017 ) . The two-component transactivation assay was performed as described ( Je et al . , 2016 ) , and the lines were backcrossed into the fea3 mutant background . To measure meristem size , segregating siblings were genotyped and shoot apical meristems of 14-day-old plants ( Figure 1A ) were dissected , cleared and measured as described previously ( Taguchi-Shiobara et al . , 2001 ) . The candidate gene association analysis of ZmCRN with the kernel row number ( KRN ) trait was conducted in a maize association panel with 368 diverse inbred lines ( Li et al . , 2013 ) . 22 SNPs in the ZmCRN gene region were observed based on previously released genotypes in the association panel . This was combined with KRN phenotypic data from five environments and BLUP ( Best Linear Unbiased Prediction ) data , including in Ya’an ( 30°N , 103°E ) , Sanya ( 18°N , 109°E ) and Kunming ( 25°N , 102°E ) in 2009 and Wuhan ( 30°N , 114°E ) and Kunming ( 25°N , 102°E ) in 2010 ( Liu et al . , 2015 ) . The association between ZmCRN and KRN was established by a mixed linear model corrected by population structure , with p-value<0 . 001 as threshold ( Zhang et al . , 2010; Li et al . , 2013 ) . CLAVATA2 and FASCIATED EAR2 orthologs from Arabidopsis thaliana , Solanum lycopersicum , Zea mays , Oryza sativa , and Amborella trichopoda were aligned using MUSCLE ( Edgar , 2004; Ouyang et al . , 2007; Lamesch et al . , 2012; Tomato Genome Consortium , 2012; Amborella Genome Project , 2013; Jiao et al . , 2017 ) . This alignment was converted to a Hidden Markov Model ( HMM ) using HMMER3 . 1b2 ( hmmer . org ) , and was used to identify sequences that bore homology within the genomes of these five species ( e-value cutoff <10e-3 ) . These amino acid sequences were grouped using convex clustering in CLANS ( Frickey and Lupas , 2004 ) , and sequences that did not cluster closely with the CLV2/FEA2 cluster were removed manually , followed by subsequent clustering; this was repeated until no sequences were identified as separate from the CLV2/FEA2 cluster . Initial phylogenetic analyses of these sequences revealed a clade of RLPs sister to the CLV2/FEA2 clade . This subset of RLP sequences was used to build two additional HMMs as described above ( hmmer . org ) , one of which included only monocot RLP sequences . These two RLP HMMs were used to search the five focal genomes again . All of the sequences recovered using both RLP HMMs were combined with the refined subset identified with the CLV2/FEA2 HMM , and iteratively clustered using CLANS until no sequences were identified as separate from the CLV2/FEA2 cluster ( Frickey and Lupas , 2004 ) . The final set of sequences , with any kinase domains removed , were aligned via MAFFT L-INS-I ( Katoh et al . , 2005; Katoh and Standley , 2013 ) . Model selection was performed using PartitionFinder2 ( Lanfear et al . , 2017 ) and phylogenetic analysis under the maximum likelihood information criterion was performed using RAxML with the VT + I + G model and 1000 bootstrap replicates ( Stamatakis , 2014 ) . Signal peptide and transmembrane domains were identified using Phobius , and the presence of a kinase domain was determined using HMMER3 . 1b2 and the Pkinase domain , respectively ( Käll et al . , 2004; Finn et al . , 2016; hmmer . org ) .

Like animals , plants are made up of many different types of cells , which descend from undifferentiated cells called stem cells . Thanks to these cells , plants are able to grow and develop throughout their lives . Stem cells live at the tips of the plant’s shoots and roots . They constantly divide to produce new cells to self-renew or replace specific plant cells in need of repair . Over time , they change – or differentiate – to go on to become part of tissues like leaves , roots , stems , shoots , flowers or fruits . To maintain a continuous pool of undifferentiated stem cells and to make sure that stem cells divide at the correct pace , neighbouring cells emit signals that control the activity of stem cells . The new stem cells that remain close to these ‘maintenance signals’ continue to behave like stem cells , but those displaced away begin to differentiate . Stem cells can receive many different types of signals , but how are these signals filtered and passed onto different places within the cell ? To test this , Je , Xu et al . created maize plants that contained mutations in a number of known signalling molecules to see if these molecules used the same communication pathway . The results showed that stem cells could integrate the different signals . Even if the signals pass through the same receiver ( a receptor protein called FASCIATED EAR2 ) , each signal exits the receptor as a different message , and attaches to a different messenger protein to relay specific information about stem cell maintenance to the cell . A next step will be to test if other plants use the same signalling pathways in the same ways to send messages between cells . A better knowledge about stem cell signals in plants could help to develop more productive crops . Previous work has found that precise control of stem cell pathways can help breed crops with more seeds or bigger fruits . These kinds of changes have been selected naturally by humans since the dawn of civilization , but we need to accelerate these advances to help meet the needs of the growing world population and improve agricultural sustainability .

2018

The CLAVATA receptor FASCIATED EAR2 responds to distinct CLE peptides by signaling through two downstream effectors

Although seasonality is widespread and can cause fluctuations in the intensity and direction of natural selection , we have little information about the consequences of seasonal fitness trade-offs for population dynamics . Here we exposed populations of Drosophila melanogaster to repeated seasonal changes in resources across 58 generations and used experimental and mathematical approaches to investigate how viability selection on body size in the non-breeding season could affect demography . We show that opposing seasonal episodes of natural selection on body size interacted with both direct and delayed density dependence to cause populations to undergo predictable multigenerational density cycles . Our results provide evidence that seasonality can set the conditions for life-history trade-offs and density dependence , which can , in turn , interact to cause multigenerational population cycles . In many organisms , reproduction is confined to seasonal fluctuation in periods of high resource , in which both fecundity ( reproduction ) and viability ( survival ) selection can occur , and periods of low resources , when reproduction stops and natural selection occurs only through viability selection . Consequently , sequential episodes of reproduction and survival caused by seasonality could be a major source of fluctuations in the strength and direction of natural selection ( Darwin , 1859; Lack , 1954; Fretwell , 1972; Schluter et al . , 1991; Bell , 2010; Bergland et al . , 2014 ) , giving rise to classic life-history trade-offs ( Lack , 1947; Roff , 1992; Stearns , 1992; Garland , 2014 ) . More specifically , traits that confer a fecundity advantage , but which are associated with a survival cost , will experience natural selection in one season that is opposed by selection in the subsequent season ( Levins , 1968; Michod , 2006; Bell , 2010; Bergland et al . , 2014 ) . The sequential rather than simultaneous nature of trade-offs driven by seasonality could have important consequence for the trait distribution within and across generations ( Levins , 1968; Grafen , 1988; Michod , 2006 ) and population dynamics ( Ozgul et al . , 2010 ) . One way by which life history trade-offs might arise from seasonal variation in resources is via body size ( Ozgul et al . , 2010 , 2014 ) . Large individuals usually have higher fecundity ( Mueller and Joshi , 2000; Schulte-Hostedde and Millar , 2004 ) , but they could also have lower survival during the non-breeding season when resources are scarce ( Stockhoff , 1991; Reznick et al . , 2000; Munch et al . , 2003; Monaghan , 2008 ) . In addition , large individuals might take more time to grow and require more resources for maintenance ( Munch et al . , 2003 ) , which could negatively impact their survival probability ( Kingsolver and Huey , 2008 ) . This association between fitness and body size in seasonal environments could have important consequences for population dynamics , particularly when selection on body size is density-dependent ( Mueller , 1997; Sinervo et al . , 2000; Travis et al . , 2013 ) . Differences in the selective advantage of body size across seasons could also shed light on the long-standing question about why population densities of many species fluctuate periodically over time ( Elton and Nicholson , 1942; Kendall et al . , 1999; McCauley et al . , 2008; Yan et al . , 2013 ) . For example , when body size is positively related to fecundity , but small individuals survive better in the non-breeding season ( Stockhoff , 1991; Munch et al . , 2003; Monaghan , 2008; Betini et al . , 2014 ) , these opposing patterns of selection could cause population cycles if selection is density-dependent . Specifically , if smaller offspring have higher survival when density is high , then the population will be composed of smaller than average individuals with lower average fecundity in the following breeding season . This lower mean fecundity will reduce population growth rates even though larger individuals will be favoured through fecundity selection . As population size declines , the strength of density-dependent viability selection on body size will also decline , which could cause net selection to reverse and favour larger individuals due to the fecundity benefit of being larger . As large individuals increase in frequency , population size should also increase via an improvement of reproductive output , returning populations to high densities . Although changes in the intensity and direction of natural selection caused by environmental variation are widespread ( Schluter et al . , 1991; Bell , 2010; Thompson , 2013; Bergland et al . , 2014 ) , we have little information about whether opposing episodes of natural selection could arise from seasonality and indirectly affect population dynamics through the feedback loop between ecological ( density dependence ) and evolutionary ( selection and evolution ) processes ( Chitty , 1960; Krebs , 1978; Hairston et al . , 2005; Pelletier et al . , 2009; Ozgul et al . , 2010; Schoener , 2011 ) . Here , we investigated how a seasonal fitness trade-off related to body-size could affect changes in population size and body size over time using replicate populations of Drosophila melanogaster exposed to repeated changes in food resources . In addition to standard breeding conditions for Drosophila , we also created a ‘non-breeding season’ by manipulating the food medium to prevent females from laying eggs during this period . Thus , in this system , breeding and survival were restricted to two sequential and distinct seasons ( hereafter 'breeding' and 'non-breeding' ) . The number of days and amount of food in each season was determined so that both fecundity and non-breeding survival were density-dependent ( Betini et al . , 2013a , 2014 , 2015 ) , which is an important feature of many populations . In Drosophila , as in many other species , the positive correlation between body size and fecundity is well known ( Mueller and Joshi , 2000; Appendix 1 ) and we previously demonstrated that small individuals have higher survival during the non-breeding season when abundance is high ( Betini et al . 2013a , 2014 ) . In addition , populations of D . melanogaster do not show evidence of multigenerational cycles ( Mueller and Joshi , 2000 ) , even when kept under the same conditions as our breeding season ( i . e . ‘aseasonal populations’; Appendix 2 ) . We , therefore , hypothesized that density dependence and opposing episodes of fecundity and viability selection on body resulting from seasonality could cause predictable and repeatable fluctuations in both population and body size . Specifically , we predicted that seasonal fitness trade-offs would cause population size and body size to undergo multigenerational cycles between periods of high abundance , when small individuals predominate , and periods of low abundance , when large individuals are more frequent . Furthermore , we predicted that populations not exposed to viability selection in the non-breeding season would lack periodic fluctuations in population size and body size . We tested whether seasonality could result in multigenerational cycles in population size and body size using three experiments ( Figure 1 ) . In the first experiment , we submitted 45 replicate populations to the seasonal treatment described above and tracked the total number of individuals and body size at the end of the breeding and non-breeding season for 58 generations ( the ‘long-term control’ treatment; Figure 1 ) . In the second experiment , we tested the role of viability selection during the non-breeding season by tracking 13 additional populations over 31 generations using a similar protocol to the ‘long-term control’ , but in which we experimentally prevented viability selection in the ‘non-breeding’ season by providing high levels of food during this season ( the ‘stop-selection’ treatment; Figure 1 ) . This protocol also maintained direct density effects on fecundity and survival , similar to the ones observed in the ‘long-term control’ . In order to address potential environmental changes in the lab , we conducted a third experiment using the same protocol as in the 'long-term control' , but under the same initial conditions and at the same time as the 'stop selection' treatment . This 'short-term control' experiment also had 13 replicate populations tracked over 31 generations ( Figure 1 ) . 10 . 7554/eLife . 18770 . 003Figure 1 . A schematic of the three experiments conducted in this study with accompanying duration , number of replicates and brief summary of their purpose . DOI: http://dx . doi . org/10 . 7554/eLife . 18770 . 003 In addition to these experiments , we also developed a mathematical model to investigate the contributions of both viability selection and delayed density dependence to population dynamics . The ‘stop-selection’ experiment was designed to eliminate viability selection , but might have also reduced potential effects of past densities on fecundity and survival . Such delayed density-dependent effects can also cause populations to cycle ( Stenseth et al . , 2003; Yan et al . , 2013 ) , or lead to more complex dynamics , such as chaos ( May 1973 ) . One potential mechanism for delayed density dependence are carry-over effects , which we have previously identified in this seasonal system ( Betini et al . , 2013a ) . Thus , we first investigated if lag effects were present in all three experiments , and then used the mathematical model to understand whether they played a role in the dynamics of their populations . Over 58 generations , the 45 replicated seasonal populations showed a predictable increase in abundance during the breeding season , where the food medium allowed females to lay eggs , and a decline in the subsequent non-breeding season ( Figure 2a ) , as is typical of many natural seasonal systems . However , the autocorrelation functions ( ACF ) also revealed that these short , seasonal cycles were embedded within longer multigenerational cycles where average population size fluctuated 3-fold ( insert in Figure 2a ) . In these populations , the ACF function was characterized by stationary periodic dynamics , which resulted in an oscillatory decay to zero ( Figure 2a inset ) . 10 . 7554/eLife . 18770 . 004Figure 2 . Population size , changes in body size and selection differentials for body size in the 'long-term control' experiment . ( a ) Population size of seasonal flies cycled over 58 generations; ( b ) Female dry weight before ( blue bars ) and after ( black bars ) the non-breeding season . Vertical bars indicate the mean female dry weight before and after the non-breeding season; ( c ) Increased population size in the non-breeding season led to stronger directional selection for smaller flies . ( d ) Time series of female dry weight measured at the end of the non-breeding season over 38 generations . In ( a and ( d ) , the autocorrelograms ( insets ) showed evidence of cycles in both population size and body size . In ( a solid blue line denotes mean population size for each generation from all replicates and dotted lines denote ±1 s . d . In ( d , the horizontal line within each box represents the median value , the edges are 25th and 75th percentiles , the whiskers extend to the most extreme data points , and points are potential outliers . DOI: http://dx . doi . org/10 . 7554/eLife . 18770 . 004 To investigate the presence of viability selection for small body size and whether this selection was density-dependent , we measured female dry weight in 38 generations from 25 different populations . As expected , there is a negative correlation between population size at the end of the non-breeding season and body size at the end of the non-breeding season ( Pearson's product-moment correlation = −0 . 64; t = −4 . 48 , p<0 . 001 ) , suggesting that density negatively impact body size in the non-breeding season . Mean survival during the non-breeding season was 71% ( ±0 . 21 SD ) and survival was density-dependent ( βsurvival = −0 . 001 , t = −27 . 73 , p<0 . 001 ) . Mean female dry weight was significantly lower after the non-breeding season ( 0 . 279 mg , n = 3620 females ) than before the non-breeding season ( 0 . 381 mg; n = 5258 females; standardized values = 0 . 577 before and −0 . 566 after the non-breeding season; Welch t-test: t = −35 . 90 , df = 1 , 589 . 240 . 24 , p<0 . 001; Figure 2b ) and this viability selection was density-dependent ( Figure 2c; Table 1 , Appendix 3 ) . That is , when population size was high at the start of the non-breeding season , there was stronger selection for smaller flies and this was driven by changes in mean dry weight after the non-breeding season rather than changes in the mean dry weight before the non-breeding season ( Appendix 3 ) . Average dry weight measured after the non-breeding season also showed multigenerational cycles ( Figure 2d ) , as indicated by the autocorrelation function ( Figure 2d , inset ) , varying between average peaks of 0 . 32 mg and lows of 0 . 23 mg . 10 . 7554/eLife . 18770 . 005Table 1 . Parameter estimates obtained from linear mixed effect models to investigate viability selection on body size as a function of thenumber of individuals at the beginning of the non-breeding season . In the ‘long-term control’ , R2LMM ( m ) =0 . 18 and R2LMM ( c ) =0 . 20; in the 'stop selection' treatment , R2LMM ( m ) =0 . 006 and R2LMM ( c ) =0 . 006; and in the 'short-term control' , R2LMM ( m ) =0 . 22 and R2LMM ( c ) =0 . 22 . R2LMM ( m ) is the variance on the response variable that is explained only by the fixed effects and R2LMM ( c ) is the variance that is explained by both fixed and random effects . In all models , the selection differential was the response variable , abundance at the beginning of the non-breeding season was the fixed effect and population ( vial ) was the random effect . DOI: http://dx . doi . org/10 . 7554/eLife . 18770 . 005Parameters Fixed effects estimate SE Df T P 1 . Long-term control Intercept −0 . 181 0 . 081434 . 5−2 . 22 0 . 027Non-breeding abundance −0 . 004 0 . 003745 . 3−12 . 67 <0 . 001 2 . Stop selection Intercept −0 . 341 0 . 159159−2 . 15 0 . 033Non-breeding abundance 0 . 0010 . 0011591 . 060 . 2913 . Short-term control Intercept 0 . 0280 . 184100−0 . 15 0 . 880Non-breeding abundance −0 . 005 0 . 001139−6 . 44 <0 . 001 In order to test the role of viability selection in these multigenerational cycles , we experimentally eliminated viability selection during the non-breeding season in 13 additional populations . Unlike the ‘long-term controls’ ( Figure 2a ) , there was no evidence of multigenerational population cycles in these ‘stop selection’ populations ( Figure 3a inset ) . In addition , body size did not significantly decline after the non-breeding season ( Figure 3b; average body size was 0 . 337 mg before and 0 . 331 after the non-breeding season; n = 1290 and n = 689 females , respectively; standardized values: 0 . 028 before and −0 . 052 , respectively; Welch t-test: t = −1 . 733 , df = 1 , 479 . 400 . 40 , p=0 . 081 ) . There was also no evidence of density-dependent selection ( Figure 3c; Table 1 ) and no evidence of cycles in body size ( Figure 3d inset ) . 10 . 7554/eLife . 18770 . 006Figure 3 . Population size , changes in body size and selection differential for body size in the 'stop-selection' experiment . ( a ) Population size of seasonal flies cycled over 31 generations . Unlike the ‘long-term control , ( b ) there was no significant change in body size after the non-breeding season ( female dry weight before - red bars - and after -black bars- the non-breeding season; vertical bars indicates the mean female dry weight before and after the non-breeding season ) and ( c ) no evidence that selection for smaller flies was density dependence . ( d ) Time series of female dry weight measured at the end of the non-breeding season over 27 generations . In ( a ) and ( d ) , the autocorrelograms ( insets ) showed no evidence of cycles in population or body size . In a solid red line denotes the mean population size for each generation from all replicates and dotted lines denote ±1 s . d . In ( d ) , the horizontal line within each box represents the median value , the edges are 25th and 75th percentiles , the whiskers extend to the most extreme data points , and points are potential outliers . DOI: http://dx . doi . org/10 . 7554/eLife . 18770 . 006 Over 31 generations , and similar to the ‘long-term control’ ( Figure 2a ) , the ‘short-term control' exhibited evidence for multigenerational cycles ( Figure 4a ) . These 'short-term control' populations also experienced viability selection that was density-dependent . Overall , female dry weight significantly decreased from an average of 0 . 371 mg before the non-breeding season ( n = 968 females ) to 0 . 276 mg after the non-breeding season ( n = 623 females; standardized values = 0 . 487 before and −0 . 761 after the non-breeding season; Welch t-test: t = −30 . 601 , p<0 . 001; Figure 4b ) , but the magnitude of this viability selection was stronger when densities were higher at the start of the non-breeding season ( Table 1 , Figure 4c ) . 10 . 7554/eLife . 18770 . 007Figure 4 . Population size , changes in body size and selection differentials for body size in the 'short-term control' experiment . ( a ) Population size of seasonal flies cycled over 31 generations , as suggested by the autocorrelogram ( insets ) , ( b ) female dry weight before the non-breeding season ( blue bars ) was higher than after the non-breeding season ( red bars; vertical bars indicate the mean female dry weight before and after the non-breeding season ) . ( c ) increased population size in the non-breeding season led to stronger directional selection for smaller flies . In ( a ) solid black line denotes mean population size for each generation from all replicates and dotted lines denote ±1 s . d . In ( d , the horizontal line within each box represents the median value , the edges are 25th and 75th percentiles , the whiskers extend to the most extreme data points , and points are potential outliers . DOI: http://dx . doi . org/10 . 7554/eLife . 18770 . 007 We statistically investigated whether fecundity and survival were influenced by density in past seasons ( i . e . delayed density dependence ) in all three experiments: ‘long-term control’ , ‘short-term control’ and ‘stop-selection’ . We used linear mixed effect models with vial ( population ) as a random effect and densities going back up to two generations as fixed effects . In the ‘long-term control’ and ‘short-term control’ , fecundity and survival were influenced by density in the current and past season ( Table 2 and Table 3 ) . In contrast , the ‘stop-selection’ experiment showed evidence of the delayed effects on fecundity ( Table 2 ) , but not on survival ( Table 3 ) , meaning that the ‘stop-selection’ treatment eliminated both viability selection on body size as well as delayed effects of density on survival . 10 . 7554/eLife . 18770 . 008Table 2 . Parameter estimates obtained from linear mixed effect models to investigate the effects of current and past density on fecundity in the ‘long-term control’ , ‘stop-selection’ and ‘short-term control’ experiments . B refers to population size at the beginning of the breeding season in the current season and NB refers to population size at the beginning of the previous non-breeding season . In the ‘long-term control’ , R2LMM ( m ) =0 . 34 and R2LMM ( c ) =0 . 37; in the 'stop selection' treatment , R2LMM ( m ) =0 . 07 and R2LMM ( c ) =0 . 07; and in the 'short-term control' , R2LMM ( m ) =0 . 30 and R2LMM ( c ) =0 . 30 . DOI: http://dx . doi . org/10 . 7554/eLife . 18770 . 008Parameters Fixed effects estimate SE T P 1 . Long-term control Intercept 0 . 4500 . 01237 . 445<0 . 001 B −0 . 212 0 . 008−25 . 82 <0 . 001 NB −0 . 008 0 . 008−1 . 05 0 . 296B * NB 0 . 0260 . 0055 . 07<0 . 001 2 . Stop-selection Intercept 0 . 5510 . 02422 . 91<0 . 001 B 0 . 1500 . 0532 . 820 . 005NB −0 . 156 0 . 034−4 . 53 <0 . 001 B * NB 0 . 06130 . 032 . 310 . 0212 . Short-term control Intercept 0 . 5320 . 01927 . 72<0 . 001 B −0 . 226 0 . 024−9 . 34 <0 . 001 NB −0 . 039 0 . 022−1 . 74 0 . 083B * NB 0 . 0460 . 0172 . 650 . 00810 . 7554/eLife . 18770 . 009Table 3 . Parameter estimates obtained from linear mixed effect models to investigate the effects of current and past density on survival in the ‘long-term control’ , ‘stop-selection’ and ‘short-term control’ experiments . NB refers to population size at the beginning of the non-breeding season in the current generation . B1 , NB1 , B2 and NB2 , refers the population size at the beginning of each season going back 1 or two generations , respectively . In the ‘long-term control’ , R2LMM ( m ) =0 . 35 and R2LMM ( c ) =0 . 36; in the 'stop selection' treatment , R2LMM ( m ) =0 . 99 and R2LMM ( c ) =0 . 99; and in the 'short-term control' , R2LMM ( m ) =0 . 43 and R2LMM ( c ) =0 . 43 . DOI: http://dx . doi . org/10 . 7554/eLife . 18770 . 009Parameters Fixed effects estimate SE T P 1 . Long-term control Intercept −0 . 353 0 . 006−54 . 74 <0 . 001 NB −0 . 186 0 . 006−31 . 19 <0 . 001 B1 0 . 1060 . 00813 . 76<0 . 001 NB1 −0 . 056 0 . 007−8 . 34 <0 . 001 B2 0 . 0440 . 0075 . 67<0 . 001 NB2 −0 . 020 0 . 007−2 . 89 0 . 0042 . Stop-selection Intercept −0 . 489 0 . 001−1 , 068 . 62<0 . 001 NB −0 . 213 0 . 001−412 . 71 <0 . 001 B1 0 . 0010 . 0011 . 3250 . 186NB1 −0 . 001 0 . 001−1 . 235 0 . 217B2 0 . 0010 . 0010 . 4790 . 632NB2 0 . 0010 . 0010 . 2250 . 8223 . Short-term control Intercept −0 . 427 0 . 013−33 . 36 <0 . 001 NB −0 . 223 0 . 0142−15 . 64 <0 . 001 B1 0 . 1440 . 0157 . 22<0 . 001 NB1 −0 . 037 0 . 015−2 . 40 0 . 017B2 0 . 0410 . 0202 . 030 . 043NB2 −0 . 015 0 . 016−0 . 933 0 . 351 A mathematical model including delayed density dependence as well as the effects of body size on survival ( viability selection ) and fecundity resulted in multigenerational cycles in population size ( before red line in Figure 5a ) , similar to those observed in our ‘long-term control’ and ‘short-term control’ ( Figure 2a and Figure 4a , respectively ) . The model without viability selection on body size and delayed density dependence ( i . e . including only effects of current abundance on fecundity and survival ) , resulted in the elimination of multigenerational cycles ( after red line in Figure 5a ) , as observed in our 'stop selection' populations ( Figure 3a ) . The exclusion of only viability selection eliminated the fitness trade-off in body size and allowed larger flies to survive to breed . This led to unstable population dynamics ( i . e . the population crashed after 10 generations; Figure 5b ) because larger flies have greater fecundity and there was a negative interaction between body size and abundance . The exclusion of delayed density dependence alone eliminated the cycles ( Figure 5c ) , suggesting that viability selection and delayed density dependence are necessary for these persistent multigenerational cycles to occur . The model with low heritability ( h2 = 10−5 ) also generated multigenerational population cycles ( Figure 5d ) . 10 . 7554/eLife . 18770 . 010Figure 5 . Predicted population size according to the integral projection model . Time series generated with the model where , ( a ) for the first 40 generations ( before vertical red line ) , when both viability selection and delayed density effects on survival during the non-breeding season and fecundity were operating and , for the last 20 generations , only the effects of current population size on fecundity and survival were modelled ( i . e . no viability selection and no delayed density effects ) ; ( b ) population size excluding the effects of viability selection ( only the effects of delayed density dependence on fecundity and survival were operating ) or ( c ) excluding the effects of delayed density dependence ( only the effects of viability selection were operating ) . In ( d ) , times series were generated as in ( a ) but with low heritability ( see Results for details ) . In ( b ) , population crashed ( i . e . population size <1 ) at generation 10 . In both ( b ) and ( c ) , the model incorporated the effects of current abundance on fecundity and survival and heritability for body size . In ( a ) and ( d ) inset figures represent autocorrelograms for population size obtained from the mathematical model including delayed density dependence , viability and fecundity selection ( before red line ) or without both delayed density dependence and viability selection ( after red line ) . See Material and methods for details and model parameters . DOI: http://dx . doi . org/10 . 7554/eLife . 18770 . 010 Our results provide empirical and mathematical evidence that the interplay between density-dependence and evolutionary trade-offs caused by seasonality can have important consequences for population dynamics . In our experimental system , seasonal variation in resources resulted in a fitness trade-off and multigenerational population cycles . These cycles were observed in the absence of predation , long-term fluctuations in resources , or any explicit negative-frequency dependence , all of which are known to cause regular fluctuations in population size ( Lindström et al . , 2001; Yan et al . , 2013 ) . Thus , the emergence of population cycles caused by three common characteristics of natural populations ( seasonality , a life-history trade-off and density-dependence ) , suggests that these ecological and evolutionary processes could contribute to fluctuations in population size over a wider range of taxa and environments than has previously been considered . Our experimental elimination of viability selection and delayed density-dependence and our mathematical model confirmed the importance of both of these evolutionary and ecological processes in the persistence of multigenerational population cycles resulting from seasonal environments . Recently , it has been proposed that eco-evolutionary dynamics are essential for understanding oscillations in population size ( Hairston et al . , 2005; Hiltunen et al . , 2014 ) , but evidence for evolutionary change ( i . e . genetically based change in ecologically relevant traits ) to feedback and influence demography and population dynamics is rare ( Schoener , 2011; Hendry , 2013 ) ; but see ( Cameron et al . , 2013 ) . Our results suggest that the evolutionary response to selection might not be required to observe feedback between ecological and evolutionary processes , as long as selection is strong . In flies , as in many other species , body size is highly heritable ( Prout and Barker , 1989 ) , and it is reasonable to suppose that , in our populations , offspring from smaller flies tend to be smaller . However , even when genetic variation is low , strong selection can still affect trait distributions within generations ( Grafen , 1988 ) , potentially affecting population size . In our experimental system , selection for body size during the non-breeding season was strong enough to drive observed cycles , potentially in the absence of any genetic change across generations . Simulations with our mathematical model with essentially zero heritability ( h2 = 10−5 as opposed to 0 . 3 ) also generated multigenerational population cycles because viability selection changed the distribution of body size , which subsequently affected fecundity . Thus , changes in the trait distribution within a generation caused by seasonal fitness trade-offs could generate feedback loops between ecological and evolutionary process in traits across generations . We showed that variation in non-breeding abundance can cause density-dependent selection , but our mathematical model suggested that long-lasting effects of past density on survival were essential to generate the multigenerational cycles . These effects were strong in both the ‘long-term control’ and ‘short-term control’ , but not in the ‘stop-selection’ treatment . One mechanism that can generate delayed effects is a density-mediated carry-over effect . We have shown how competition during the non-breeding season can negatively impact the physiological conditions of the survivors , reducing their breeding output in the following breeding season ( Ratikainen et al . , 2008; Betini et al . , 2013a ) . This carry-over effect could also negatively impact survival if individuals during the non-breeding season were in lower body condition because of high abundance at the beginning of the season . Delayed density dependence can also arise from maternal effects , whereby offspring attributes are affected by parental abundance , which have previously been shown to generate population cycles ( Plaistow and Benton , 2009 ) . Both carry-over effects and maternal effects have the potential to interact with viability and fecundity selection because they can affect individual fecundity and survival via changes in body size and condition . Our mathematical model also suggested that the delayed density dependence alone would cause short cycles and population crashes after only a few generations . It is clear from the model that the addition of viability selection allowed for sustained multigenerational cycles , as was observed in our empirical results . These results are in general agreement with previous studies . For example , in a predator-prey system , evolutionary changes in the prey population extended density cycles compared to the typical quarter-period phase lags between prey and predator densities ( Yoshida et al . , 2003 ) . This evolutionary effect is now believed to be widespread in lab systems ( Becks et al . , 2012; Hiltunen et al . , 2014 ) . Thus , it is possible that evolutionary processes affecting trait distributions within and between generations can buffer strong negative feedback caused by ecological processes that are typical in many natural systems ( e . g . overcompensation caused by density dependence ) , representing a different form of evolutionary rescue ( Bell , 2013; Carlson et al . , 2014 ) . Opposing viability and fecundity selection on body size was necessary for the persistence of multigenerational population cycles . Such fitness trade-offs between seasons are likely more common than currently appreciated ( Schluter et al . , 1991; Schmidt et al . , 2005; Schmidt and Conde , 2006 , Schmidt and Paaby 2008; Kingsolver and Diamond , 2011; Behrman et al . , 2015 ) , especially given that many organisms use vastly different habitats over the annual cycle . Although body size caused fitness trade-offs between seasons in our experimental system , many other traits might also exhibit seasonal trade-offs . For example , sexually selected traits enhance mating success , but might also reduce survival during the non-breeding season ( Emlen , 2001 ) . Alternatively , increased development rates can accelerate age of first reproduction , but are often associated with reduced survival during the adult stage ( Stearns , 1992; Kingsolver and Huey , 2008 ) . Seasonal fitness trade-offs also represent a form of balancing selection , which can maintain increased levels of genetic variation at evolutionary equilibrium ( Barton and Keightley , 2002 ) . Seasonal fluctuations in selection might allow for the maintenance of increased adaptive potential in populations ( Shaw and Shaw , 2014 ) , facilitating more rapid responses to directional environmental change ( Bradshaw and Holzapfel , 2006; Bell , 2010; Huang et al . , 2016 ) . This temporal variation in selective pressures might also help to explain why some species are more successful at colonizing new areas , as suggested by studies of seasonal changes in genetic variation in wild populations of D . melanogaster ( Schmidt and Conde , 2006; Bergland et al . , 2014; Behrman et al . , 2015 ) . Thus , the consequences of seasonal fluctuations in resources are , therefore , not restricted to purely ecological pathways . Seasonality can also alter the strength and direction of natural selection and might maintain a population’s adaptive potential over longer evolutionary timescales . Examples of delayed density dependence causing cycles have been documented in natural populations subjected to strong seasonality ( Merritt et al . , 2001; Stenseth et al . , 2003; Yan et al . , 2013 ) . We showed how opposing episodes of selection driven by a life-history trade-off arise in seasonal environments , which could interact with delayed effects to influence population size . Recently , it has been proposed that results from lab studies on consumer-resource dynamics were likely confounded by fast evolutionary changes ( Hiltunen et al . , 2014 ) . The same could be true for lab and field studies on population cycles , if changes in trait distributions are linked to population size . To understand whether this is a common phenomenon in natural populations , one needs to tease apart the evolutionary from the ecological consequences of variation in density . As in our lab system , this is a constraint that needs to be overcome with either improved controlled experiments or mathematical approaches . Nevertheless , given that variation in density is common in many natural systems , measurements of selection and population density in both seasons will provide a better understanding of the dynamics of natural populations and how environmental change might alter such dynamics . To simulate seasonality in populations with non-overlapping generations , we changed food composition to generate two distinct ‘seasons’ ( n = 45 populations ) . During the breeding season , we placed adults in vials ( 28 × 95 mm ) with a dead yeast-sugar medium to lay for 24 hr ( day 0 ) , after which adults were discarded and larvae were allowed to mature to adults ( 16 days ) . Individuals were then transferred ( day 17 ) to a non-breeding environment for 4 days . The non-breeding season consisted of an empty vial of the same size as the breeding vials , but food was provided by a pipette tip filled with 0 . 200 ml of 5% water–sugar solution per day from the top of the vial . This solution was sufficient for many flies to survive ( ~95% survival at low population size ) but did not allow females to lay eggs ( Bownes and Blair , 1986; Betini et al . , 2013a ) . In both seasonal treatments , a consistent amount of food was provided regardless of population size , which caused reproduction and survival to be density-dependent ( Betini et al . , 2013a ) . Each of 45 replicate populations was repeatedly transferred between breeding and non-breeding environments for 58 generations and together they are referred to as ‘long-term controls’ . The quality and amount of the medium during the breeding season mimics the scenario well explored in other Drosophila systems ( Mueller and Joshi , 2000 ) , where the food is of lower quality and more limited for adults compared to larvae . Dynamics of populations experiencing only this medium ( i . e . only the breeding season ) do not show evidence of cycles ( Mueller and Joshi , 2000 ) ; ( Appendix 2—figure 1 ) . To investigate whether opposing selective pressures caused multigenerational cycles in seasonal populations , we experimentally stopped viability selection for a smaller body size in the non-breeding season while preserving density-dependent survival . To do this , we exposed 13 new replicate populations to the same seasonal change in food resources ( described above ) over 31 generations , but provided unlimited access to food during the non-breeding season ( 0 . 8 ml/day instead of 0 . 2 ml/day; initial population size was 5 males and 5 females ) . Average mortality was reduced from 28% ( ±0 . 4; mean ± s . e . ) in the 'long-term control' populations to 0 . 5% ( ±1 ) in these new 'stop selection' treatments . After four days in the non-breeding season , we haphazardly selected the survivors that moved to the breeding season . To preserve the density-dependent survival process during the non-breeding period , the number of survivors for each population was calculated based on a logistic non-breeding survival function parameterized from our 'long-term control' populations ( Wilson , 1994; Figure 6 ) Su=11+ ( Nv ) w10 . 7554/eLife . 18770 . 011Figure 6 . The relationship between survival and population size at the beginning of the non-breeding season . The blue line represents a survival function parameterized with our 'long-term control' populations ( v = 375 . 22 , s . e . =7 . 60 , t = 49 . 43 , p<0 . 001; w = 1 . 83 , s . e . =0 . 08 , t = 22 . 50 , p<0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18770 . 011 where Su is survival ( number of survivors at the end of the non-breeding season divided by the total number of individuals at the beginning of the non-breeding season ) , N is the population size at the beginning of the non-breeding season , and v and w are constant to be estimated from the data . The function was parameterized with our 'long-term control' populations ( 45 replicates over 43 generations ) using the non-linear function nls in R ( R Core Team , 2015 ) . Individuals that moved to the following breeding season were haphazardly selected . Our ‘long-term control’ and ‘stop-selection’ populations were initiated at different periods and with different population sizes . Thus , differences in the lab environment and initial conditions could have influenced the dynamics of these populations . For these reasons , we also initiated an additional 13 replicate populations at the same time and same initial population size as the ‘stop-selection’ populations . In these populations , we used the same protocol as in the ‘long-term control’ , but initial population size ( 5 males and 5 females ) and number of generations ( n = 31 ) was the same as in the ‘stop-selection’ experiment . We measured linear selection differentials ( S ) for female body weight ( i . e . population mean dry weight after – population mean dry weight before the non-breeding season ) in all three experiments ( ‘long-term control’ , ‘stop-selection’ and ‘short-term control’ ) . We used female dry body weight as a proxy for body size . We experimentally confirmed that the mean body size declined along with mean dry weight in surviving females after the non-breeding season by placing individuals from the stock population in the non-breeding season in either high ( n = 300 ) or low ( n = 20 ) abundance . We measured the thorax size in 10 individual females sampled before the non-breeding season and 10 females sampled from the population after the non-breeding season . Mean thorax size after the non-breeding season was significantly lower than mean thorax size prior to the non-breeding season ( Welch t-test: t = −2 . 87 , df = 37 . 66 , p=0 . 006 ) , whereas there was no significant differences in average thorax in the low abundance treatment ( Welch t-test: t = −0 . 55 , df = 36 . 96 , p=0 . 586; average size before = 1 . 318 mm; average size after the non-breeding season for the high abundance treatment was 1 . 225 and 1 . 297 mm for the low abundance treatment ) . We obtained female dry weight from 5% of total population size before and after the non-breeding season . We then sexed , dried and weighed individual females to the nearest 0 . 001 mg using a microbalance . To calculate the linear selection differentials ( S ) for female body weight , we standardized dry weight measures to a mean of zero and unit variance prior to calculating selection ( Brodie et al . , 1995 ) . The significance of our estimates of viability selection during the non-breeding season was assessed using a Welch t-test that tested for differences between mean female dry body weight before and after the non-breeding season . We then used a linear mixed effect model ( LMM ) to investigate whether S was density-dependent . In the LMM , the selection differential calculated for each population in each generation was the response variable , number of individuals at the beginning of the non-breeding season was fitted as a fixed effect and population was fitted as a random effect . In the ‘long-term control’ , we measured female dry weight in 38 generations from an arbitrary number of replicate populations ( 16 to 25 population per generation ) in generations 9 , 10 , 15 , 16 , 21 , 22 , 25 , 26 , 28–58 . Total number of individuals measured was 5258 before and 3620 after the non-breeding season . In the ‘stop-selection’ and ‘short-term control’ , we sampled females in seven populations in generations 1 to 25 ( n = 1591 females and n = 1979 females in the 'short-term control' and 'stop selection' treatments , respectively ) , with the exception of generation 13 in the 'short term control' . To investigate whether population size cycled in our experimental populations , we used the autocorrelation function method ( Turchin and Taylor , 1992; Box et al . , 2013 ) . Cycles can be inferred by investigating the shape of the estimated ACF: population cycles are characterized by stationary periodic dynamics , which result in an oscillatory decay to zero of the ACF estimates . We calculated the ACF for each treatment using the mean population size across all replicates at the start of the breeding and non-breeding season in each generation . For all-time series , we excluded the first three generations to avoid any transient effects caused by the initial population size . We also detrended the time series to eliminate any overall temporal trends in abundance by subtracting the fitted values from a linear regression of average population size on generation ( ‘long-term control’: β = 0 . 468 , s . e . =0 . 177 , t = −2 . 65 , p=0 . 011; ‘short-term control’: β = −1 . 855 , s . e . =0 . 686 , t = −2 . 71 , p=0 . 01 ) . To investigate temporal changes in body size in the 'long-term control' and ‘stop-selection’ populations ( measured as female dry weight ) , we used the same autocorrelation function method described above for the population size , including detrending the data . We investigated the effects of past abundance on fecundity and survival with LMM with vial ( population ) as a random effect in all three experiments: ‘long-term control’ and ‘short-term control’ and ‘stop-selection’ treatment . In previous work , we have shown that fecundity in this system is influenced by an interaction between current abundance at the beginning of the breeding season and at the beginning of the previous non-breeding season ( Betini et al . , 2013a , 2013b ) . Thus , in this study , we also investigated if this interaction significantly explained variation in fecundity in all three experiments . Fecundity was calculated as the number of individuals at the end of the breeding season ( i . e . number of offspring ) divided by the number of individuals at the beginning of the same breeding season ( number of parents ) . Because we did not know how environmental effects ( e . g . physiological condition ) could influence survival in our system , we incorporated the breeding and non-breeding abundance going back two generations as explanatory variables in the model to explain survival . Lag effects have been documented in natural populations subjected to strong seasonality ( Merritt et al . , 2001 , 2001; Stenseth et al . , 2003 ) . Survival was calculated as the number of individuals at the end of the non-breeding season divided by the number of individuals at the beginning of the non-breeding season . Explanatory variables were standardized before analysis by subtracting the sample mean from each observation and dividing each value by the sample standard deviation and response variables were log transformed . We developed a mathematical model to investigate the contributions of both viability selection and delayed density dependence to population dynamics . The model linked quantitative trait evolution with demography and was similar to integral projection modelling approaches that are commonly used to study ecological and evolutionary processes concurrently ( Slatkin , 1979 , 1980; Ellner and Rees , 2006 ) . In particular , the number of individuals with body size y at the next time step t + 1 , n ( y , t + 1 ) , was a function of a projection function k ( y ) and n ( y , t ) , such that n ( y , t + 1 ) = k ( y ) * n ( y , t ) . The projection function k ( y ) accounted for the additive genetic and environmental contributions to body size , as well as both density-independent and density-dependent differences in fecundity or survivourship among body sizes , depending on season . Two population sizes were modeled in discrete time: population size at the beginning of the breeding season , and population size at the beginning of the non-breeding season within each generation . For each season , the mean breeding value for body size and the mean phenotypic value for body size were modelled , where the phenotypic value of an individual was its breeding value plus a random environmental effect . We assumed that breeding and phenotypic values were normally distributed and that the variance in breeding values and the variance in environmental effects remained constant across generations . The assumption of a constant variance in breeding values assumed that mutational variance and changes in genetic architecture ( including linkage and epistatic interactions ) could restore variance that was initially depleted by selection . Slatkin’s model provides for changes in the variance of breeding values due to selection and considers the effect of genetic architecture , such as linkage ( Slatkin , 1979 , 1980 ) . In our experiment , we do not have information on the genetic architecture for body size , nor mutational variance . We did not see a decline in the magnitudes of the rates of change in body size in the experiment , which is consistent with the maintenance of heritability . The fecundity of an individual during the breeding season was a function of its body size and current and past effects of density . Past density influenced fecundity via changes in the physiological conditions of individuals that spent the previous non-breeding season at high densities but survived to breed at low densities ( i . e . a carry-over effect that was modelled as the interaction between density at the end of the breeding season in the previous generation and current breeding density ( Betini et al . , 2013a , 2014 ) . Current density could also negatively influence fecundity via density-dependent effects . There was a stronger negative effect of density on fecundity for large versus small individuals , although large individuals have higher intrinsic fecundity ( Figure 3 in the main text ) . During the 'stop selection' phase of the simulations , we decreased carry-over effects ( parameters φ5and φ6 ) by 25% given that flies had unlimited amount of food in the 'stop selection' treatments and were likely to be in better physiological condition than those in the 'short-term control' treatments . Parameters used in the fecundity function were chosen such that fecundities as a function of body size and densities were within the normal range observed in D . melanogaster . The survival of an individual during the non-breeding season was also a function of its body size ( when viability selection occurs ) and current and past effects of density . As indicated by our previous experiment , larger flies had lower survival and this effect was magnified by increase density ( Betini et al . , 2014 ) . We included the effects of past density beyond the ones described above , going back two generations . During the 'stop selection' phase of the simulations , survival was not a function of body size nor past density , but was instead only affected by current density , following the logistic equation Su for survivorship and the experimental design . To accomplish this and to ensure proper scaling of survival , we assigned all individuals the same phenotype in the survivorship function . Parameters for the survival function were based on values obtained in the long-term control ( Figure 2a in the main text ) . In the context of an individual's breeding and phenotypic values , an individual with breeding value x had an environmental effect y added with a mean effect that was a negative linear function of density and was normally distributed with variance V[e] . Consequently , each breeding value expressed a distribution of phenotypes as a function of the current environment . In the context of survivorship , since each breeding value expressed a distribution of phenotypes , each breeding value expressed a distribution of survivorship and the model integrated across environmental effects to get the mean survivorship associated with a breeding value . This integration of survivorship across environmental effects gave the value of the projection function for a given breeding value . The projection function in the context of fecundity was modelled similarly; breeding values were transformed into phenotypes via a distribution of environmental effects , which when integrated gave the mean fecundity of individuals with a particular breeding value for body size . Population size across seasons was the product of the current population size times the average fecundity ( breeding season ) or average survivorship ( non breeding season ) at the phenotypic level . Below is a list of variables , parameters and functions used in the model . Xi - population size at the beginning of the breeding season i generations ago Yi - population size at the beginning of the non-breeding season i generations ago bX¯ - mean breeding value for body size at the beginning of the breeding season bY¯ - mean breeding value for body size at the beginning of the non-breeding season pX¯ - mean phenotype for body size at the beginning of the breeding season pY¯ - mean phenotype for body size at the beginning of the non-breeding season z - body size Note , a generation consists of a breeding season and then a non-breeding season . At the beginning of the breeding season the mean breeding value ( bX¯ ) is a function of densities in previous generations , such that i > 0 . VA=0 . 003 - additive genetic variance for body size VE=0 . 007 - environmental variance for body size such that h2=0 . 30 ( Prout and Barker , 1989 ) Fecundity function:f ( z , X→ , Y→ ) = ( φ1+ ( φ2z ) 4 ) exp ( − ( φ3+φ4z4 ) ( X0+φ5Y1+φ6X0Y1 ) ) Survival function:s ( z , X→ , Y→ ) =1−z ( υ1 ( X1+Y1 ) +υ2 ( X2+Y2 ) ) 1+ ( z ( Y0+ ( υ3Y0z ) exp ( z ) υ4 ) υ6 ) υ5 Integral projection model:b¯x=∫zminzmax∫zmin−xzmax−xxs ( x+y , X→ , Y→ ) N ( x , b¯Y , VB ) +N ( y , eX ( X1 ) , VE ) dydx∫zminzmax∫zmin−xzmax−xs ( x+y , X→ , Y→ ) N ( x , b¯Y , VB ) +N ( y , eX ( X1 ) , VE ) dydxX0=Y1∫zminzmaxs ( y , X→ , Y→ ) N ( y , p¯Y , VB+VE ) dy∫zminzmaxN ( y , p¯Y , VB+VE ) dyp¯x=∫zminzmax∫zmin−xzmax−x ( x+y ) N ( x , b¯Y , VB ) N ( y , eY ( Y1 ) , VE ) dydx∫zminzmax∫zmin−xzmax−xN ( x , b¯X , VB ) N ( y , eY ( Y1 ) , VE ) dydxb¯y=∫zminzmax∫zmin−xzmax−xxf ( x+y , X→ , Y→ ) N ( x , b¯X , VB ) N ( y , eY ( Y1 ) , VE ) dydx∫zminzmax∫zmin−xzmax−xf ( x+y , X→ , Y→ ) N ( x , b¯X , VB ) N ( y , eY ( Y1 ) , VE ) dydxY0=X0∫zminzmaxf ( y , X→ , Y→ ) N ( y , p¯X , VB+VE ) dy∫zminzmaxN ( y , p¯X , VB+VE ) dyp¯y=∫zminzmax∫zmin−xzmax−x ( x+y ) N ( x , b¯Y , VB ) N ( y , eX ( X1 ) , VE ) dydx∫zminzmax∫zmin−xzmax−xN ( x , b¯Y , VB ) N ( y , eX ( X1 ) , VE ) dydx In the equations above N ( z , m , v ) is the probability density of a normally distributed random variable with a mean of m and variance v . Functions eX ( N ) =−λXNand eY ( N ) =−λYNgive the average environmental effect on the phenotype . φ1=1 . 5 ( intrinsic fecundity for small flies ) φ2=2 . 8 ( rate of increase in fecundity with body size ) φ3=0 . 00007 ( baseline rate of decline in fecundity with density for small flies ) φ4=0 . 04588 ( rate of increase in the magnitude of the decline in fecundity as body size increases ) φ5=1 ( constant characterizing the effects of population size at the end of the breeding season one generation ago; i . e . carry-over effects ) φ6=0 . 001 ( constant characterizing the interaction between current population size and population size at the end of the breeding season one generation ago ) υ1=0 . 005 ( rate of decline in survivorship due to population sizes one generation ago ) υ2=0 . 0026 ( rate of decline in survivorship due to population sizes two generations ago ) υ3=0 . 7 ( constant governing negative effect of the interaction between the current population size and body size ) υ4=0 . 7 ( constant governing the shape of the negative effect of the interaction between current population size and body size on survivorship ) υ5=6 ( a second constant governing the shape of the negative effect of the interaction between current population size and body size on survivorship ) υ6=350 ( a third constant governing the shape of the negative effect of the interaction between current population size and body size on survivorship ) λX=0 . 00001 ( rate of decline in body size [with density] , i . e . the environmental effect of density on body size due to density at the beginning of the breeding season ) λY=0 . 00025 ( rate of decline in body size [with density] , i . e . the environmental effect of density on body size due to density at the beginning of the non-breeding season ) All LMM were performed using the lmer function from the lme4 package ( Bates , 2010 ) and p-values were obtained using the lmerTest package ( Kuznetsova et al . , 2014 ) . All analyses were conducted in R ( R Core Team , 2015 ) . Marginal ( R2LMM ( m ) ) and conditional ( R2L<MM ( c ) ) variance for the LMMs were calculated with MuMIn package according to ( Nakagawa and Schielzeth , 2013 ) . R2LMM ( m ) is the variance on the response variable that is explained only by the fixed effects and R2LMM ( c ) is the variance that is explained by both fixed and random effects ( Nakagawa and Schielzeth , 2013 ) .

Many wild populations go through long cycles in abundance that span several generations . The traditional explanation for such “multigenerational” cycles is that they are driven by predator/prey relationships , the classic example being oscillations between the numbers of lynx and snowshoe hares . Population cycles could also be driven by seasonal changes . For example , traits that help animals to produce large numbers of offspring during the breeding season may reduce the ability of the animal to survive the non-breeding season . Body size is one such trait . Large individuals tend to produce more offspring , but their larger body size means that they find it harder to survive when food is scarce . As a consequence , large individuals should have an advantage and be more common when the population size is low and there are enough resources for all individuals . However , small individuals should be more abundant when population size is high . This trade-off caused by seasonality could set the population in motion towards predictable , multigenerational cycles . To test this idea , Betini et al . established populations of fruit flies that went through ‘breeding’ and ‘non-breeding’ seasons . This was achieved by periodically altering the flies’ food to prevent the females from laying eggs ( in the lab , fruit flies do not normally have non-breeding seasons ) . Over 58 generations , the number of flies in each population cycled between peaks of high and low numbers . When the population contained relatively few flies , there was strong selection for large flies because they have high reproductive success . Hence , the population grew . When the population was large , meaning that the flies had to compete for a limited amount of food , there was strong selection for small flies because they are better able to survive on limited resources . However , small flies also produce fewer offspring on average , resulting in a decrease in population size . When the flies all had sufficient food during the non-breeding season , these regular cycles completely disappeared . A major challenge will be to understand how common this phenomenon is in the wild . Virtually all organisms live in seasonal environments but whether they face strong trade-offs in the expression of traits is not well understood . This is primarily because of the difficulty in following individuals throughout the year .

2017

A fitness trade-off between seasons causes multigenerational cycles in phenotype and population size

Animals and humans have a tendency to repeat recent choices , a phenomenon known as choice hysteresis . The mechanism for this choice bias remains unclear . Using an established , biophysically informed model of a competitive attractor network for decision making , we found that decaying tail activity from the previous trial caused choice hysteresis , especially during difficult trials , and accurately predicted human perceptual choices . In the model , choice variability could be directionally altered through amplification or dampening of post-trial activity decay through simulated depolarizing or hyperpolarizing network stimulation . An analogous intervention using transcranial direct current stimulation ( tDCS ) over left dorsolateral prefrontal cortex ( dlPFC ) yielded a close match between model predictions and experimental results: net soma depolarizing currents increased choice hysteresis , while hyperpolarizing currents suppressed it . Residual activity in competitive attractor networks within dlPFC may thus give rise to biases in perceptual choices , which can be directionally controlled through non-invasive brain stimulation . Perceptual and value-based decisions in humans and animals are often characterized by choice biases ( Hunt , 2014; Nicolle et al . , 2011; Fleming et al . , 2010; Padoa-Schioppa , 2013; Noorbaloochi et al . , 2015; De Martino et al . , 2006; Chen et al . , 2006; Rorie and Newsome , 2005; Tom et al . , 2007 ) . For example , human and nonhuman primate value choices are subject to various biases such as framing effects ( De Martino et al . , 2006 ) , choice repetition biases ( Padoa-Schioppa , 2013; Samuelson and Zeckhauser , 1988 ) , sunk cost effects ( Bogdanov et al . , 2015 ) , and previous payoff biases ( Noorbaloochi et al . , 2015; Rorie et al . , 2010 ) . These biases , which become more pronounced with difficult decisions ( Fleming et al . , 2010; Padoa-Schioppa , 2013 ) , are also observed in human perceptual decision making ( Nicolle et al . , 2011; Fleming et al . , 2010; Noorbaloochi et al . , 2015; Mulder et al . , 2012; St John-Saaltink et al . , 2016; Akaishi et al . , 2014 ) , with correlational evidence for a link between neural and choice variability ( St John-Saaltink et al . , 2016; Hesselmann et al . , 2008; Wyart and Tallon-Baudry , 2009 ) . Choice repetition biases are especially intriguing because they provide a window on decision making outside of the laboratory . In real life , decisions do not occur in discrete and isolated trials with long inter-trial intervals , but rather take place within the context of , and are therefore potentially biased by , previous decisions made in the immediate past . This notion has indeed been elegantly recognized in recent economic decision making work in non-human primates , suggesting that decaying trace activity from the previous choice in competitive neural circuits increases the likelihood of repeating that choice when there is a small subjective difference between the current decision options ( Padoa-Schioppa , 2013 ) . These results have been explained by sustained recurrent activity in competitive attractor networks , which gradually returns to baseline levels following a decision , but can bias activity in the following trial in tasks with short inter-trial intervals ( Rustichini and Padoa-Schioppa , 2015 ) . Indeed , a series of studies have linked perceptual and value-based decision-making with activity in such competitive attractor networks ( Rustichini and Padoa-Schioppa , 2015; Hunt et al . , 2012; Bonaiuto and Arbib , 2014; Hämmerer et al . , 2016; Wang , 2008; Wong et al . , 2007; Wang , 2012 , 2002; Martí et al . , 2008; Mazurek et al . , 2003; Moreno-Bote et al . , 2007; Deco and Rolls , 2005; Deco et al . , 2009; Usher and McClelland , 2001; Bogacz et al . , 2007; Furman and Wang , 2008; Deco et al . , 2013; Braun and Mattia , 2010; Jocham et al . , 2012 ) . Here , we show that carry-over activity in these networks produces a conspicuous bias to repeat difficult choices which is mirrored in the behavior of human participants . We further show that a characterization of this phenomenon in silico allows us to make directional predictions of the effects of transcranial stimulation upon choice bias which are further borne out by behavioural experiments . Specifically , we used a combination of human experimentation and computational modeling to investigate the mechanisms underlying choice hysteresis during perceptual decision making . We used an established and biophysically plausible model of a decision making network that employs competition between neural populations to choose between two alternate response options . Rather than simulating discrete trials and reinitializing the network state at the start of each trial , we sought to emulate the serial dependency between real world choices . We therefore ran the network in continuous blocks of trials , with the final state at the end of each trial serving as the initial state of the next trial ( Rustichini and Padoa-Schioppa , 2015 ) . We confirmed that this produced choice hysteresis in the model behavior , through decaying trace activity from the previous trial biasing selection in the current trial , but only for short inter-stimulus intervals . We then conducted an analogous experiment with human participants and found a similar tendency to repeat previous choices . The model contains variables and parameters with well‐defined anatomical and physiological substrates ( Rustichini and Padoa-Schioppa , 2015; Bonaiuto and Arbib , 2014; Wang , 2008 , 2012 , 2002 ) , allowing for explicit simulation and linkage with the known neurophysiological effects of stimulation . We found that perturbation of the model’s trace activity through simulated changes in the network’s membrane potential led to predictable alterations in choice bias . In human participants , we therefore applied transcranial direct current stimulation ( tDCS ) to left dorsolateral prefrontal cortex ( dlPFC ) , a region implicated in perceptual decision making ( Heekeren et al . , 2004; Kim and Shadlen , 1999; Heekeren et al . , 2006; Philiastides et al . , 2011; Rahnev et al . , 2016; Georgiev et al . , 2016 ) . TDCS is thought to alter neuronal excitability and spontaneous firing rates in brain networks by polarizing membrane potentials in a network ( Rahman et al . , 2013; Nitsche and Paulus , 2011; Bikson et al . , 2004 ) , thus providing an analogous network perturbation to our simulations . Because tDCS leads to subthreshold polarization changes , we were able to subtly alter the spontaneous fluctuations in neural activity within the targeted brain region , noninvasively in our human participants ( Nitsche and Paulus , 2011 , 2000; Kuo and Nitsche , 2012 ) . We found that the predictions generated by the model were closely mirrored by the modulation of choice hysteresis in human participants through application of tDCS over dlPFC . We were thus able to directionally control choice biases in perceptual decision making through causal manipulation of the neural dynamics in dlPFC . The comparison with the model suggests that this control of choice hysteresis arises from an amplification or suppression of sustained recurrent activity , which biases the following decision . We used an established spiking neural model of decision making implementing an attractor network ( Bonaiuto and Arbib , 2014; Wang , 2008; Wong et al . , 2007; Wang , 2012 , 2002; Deco et al . , 2009; Bonaiuto and Bestmann , 2015; Rolls et al . , 2010; Wong and Wang , 2006; Lo and Wang , 2006; Machens et al . , 2005 ) . This model was initially developed to explain the neural dynamics of perceptual decision making and working memory ( Wang , 2002 ) and has been used to investigate the behavioral and neural correlates of a wide variety of perceptual and value-based decision making tasks at various levels of explanation ( Rustichini and Padoa-Schioppa , 2015; Hunt et al . , 2012; Bonaiuto and Arbib , 2014; Hämmerer et al . , 2016; Wang , 2012 , 2002; Furman and Wang , 2008; Jocham et al . , 2012; Bonaiuto and Bestmann , 2015; Rolls et al . , 2010; Wong and Wang , 2006 ) . The model is well suited for computational neurostimulation studies because it is complex enough to simulate network dynamics at the neural level , yet is simple enough to generate population-level ( neural and hemodynamic ) signals , and the resulting behavior allows for comparison with human data ( Hunt et al . , 2012; Bonaiuto and Arbib , 2014; Rolls et al . , 2010 ) . The model also incorporates neurons at a level of detail that allows simulation of tDCS by the addition of extra transmembrane currents with parameter values comparable to previous modeling work ( Hämmerer et al . , 2016; Bonaiuto and Bestmann , 2015; Molaee-Ardekani et al . , 2013 ) , and current understanding of the mechanism of action of tDCS ( Rahman et al . , 2013; Nitsche and Paulus , 2011; Bikson et al . , 2004; Funke , 2013; Radman et al . , 2009; Bindman et al . , 1964 ) . The model consists of two populations of pyramidal cells representing the available response options , which are ‘left’ and ‘right’ in this task ( Figure 1A ) . Each population receives task-related inputs signaling the perceived evidence for each response option . The difference between the inputs varies inversely with the difficulty of the task ( Figure 1A , inset ) , and the rate of each input is sampled according to the refresh rate of the monitor used in our experiment ( 60 Hz , Figure 1B , left column ) . The pyramidal populations are reciprocally connected and mutually inhibit each other indirectly via projections to and from a common pool of inhibitory interneurons . This pattern of connectivity gives rise to winner-take-all behavior in which the firing rate of one pyramidal population ( typically the one receiving the strongest inputs ) increases and that of the other is suppressed , indicating the decision . In difficult trials each input fires at approximately the same rate , while in easy trials one input fires at a high rate while the other fires at a very low rate ( Figure 1B , right column ) . 10 . 7554/eLife . 20047 . 003Figure 1 . Model architecture . ( A ) The model contains two populations of pyramidal cells which inhibit each other through a common pool of inhibitory interneurons . The pyramidal populations receive task-related inputs signaling the momentary evidence for each response option . The mean input firing rate to each pyramidal population varies as a function of the stimulus coherence ( inset ) . Difficult trials have low coherence , easy trials high coherence . tDCS is simulated by modulating the membrane potential of the pyramidal and interneuron populations . ( B ) Firing rates of the task-related inputs ( left column ) and two pyramidal populations ( right column ) during representative trials with low ( top row ) , and high ( bottom row ) coherence . The horizontal dotted lines denote the response threshold ( 20 Hz in this example ) and the vertical dotted lines show the decision time - when one of the pyramidal population’s firing rate crosses the response threshold . DOI: http://dx . doi . org/10 . 7554/eLife . 20047 . 003 We simulated behavior in a perceptual decision making task by scaling the magnitude of the task-related inputs to emulate input from a virtual Random Dot Kinetogram ( RDK ) with varying levels of coherent motion . The behavior was produced by virtual subjects , which were created by instantiating the model with parameters sampled from distributions designed to capture between-participant variability in human populations ( see Materials and methods ) . In order to analyze the behavioral output of the network , we consider a response option to be chosen when the corresponding pyramidal population exceeds a set response threshold . We measured the accuracy of the model’s performance as the percentage of trials in which the chosen option corresponded to the stronger task-related input . For comparison between virtual subjects and human participants , we defined the accuracy threshold as the coherence level required to attain 80% accuracy . The time step at which the response threshold is exceeded is taken as the decision time for that trial ( Figure 1B ) . Because we do not simulate perceptual and motor processes involved in encoding visual stimuli and producing a movement to indicate the decision , this is distinct from the response time measured in human participants . As expected , the model generates increasingly accurate responses at higher coherence levels ( Figure 2A ) . This is because the ‘correct’ pyramidal population is receiving much stronger input than the other , allowing it to more easily win the competition by exerting strong inhibitory influence onto the other pyramidal population pool . In line with previous work , the model predicts a decrease in decision time with increasing coherence ( Wang , 2002 ) ( Figure 2B ) . In terms of model dynamics , when motion coherence is low the sensory evidence for the left and right choices is approximately equal , and therefore the inputs that drive both pyramidal populations are more balanced . As a consequence it takes longer for one population to ‘win’ over the other and for the network to reach a stable state ( Figure 1B ) . 10 . 7554/eLife . 20047 . 004Figure 2 . Effects of simulated network stimulation on model behaviour . ( A ) There was no average change in the decision threshold with either depolarizing or hyperpolarizing stimulation , where the decision threshold reflects the coherence required to reach 80% accuracy . ( B ) Decision time decreases with increasing coherence , with depolarizing stimulation speeding decision time and hyperpolarizing stimulation slowing decisions . ( C ) Depolarizing stimulation decreases and hyperpolarizing stimulation increases decision time , but this effect is reduced with increasing coherence . ( D ) Neural dynamics of the model . The model was run continuously , with the decaying activity of each trial influencing the initial activity at the beginning of the following trial . Depolarizing stimulation delayed the return of this decaying activity to baseline levels , while hyperpolarizing stimulation dampened the overall dynamics of the model and therefore suppressed residual activity . ( E ) When sorted by the choice made on the previous trial ( Left* or Right* ) , the indecision point ( or level of coherence resulting in chance selection of the same choice ) , shifts . This reflects a bias towards repeating that decision . ( F ) The positive shift in indecision point is further increased by depolarizing stimulation and decreased by hyperpolarizing stimulation . ( G ) A logistic regression model was fit to choice behavior with coefficients for coherence and the choice on the previous trial . ( H ) This analysis confirms a positive value for the influence of the previous choice on the current choice ( a1 ) , scaled by the influence of coherence ( a2 ) . Depolarizing stimulation increases this ratio , and hyperpolarizing stimulation reduces it . See Figure 2—source data 1 for raw data . DOI: http://dx . doi . org/10 . 7554/eLife . 20047 . 00410 . 7554/eLife . 20047 . 005Figure 2—source data 1 . Competitive attractor model accuracy , decision time , and choice hysteresis with simulated network stimulation . DOI: http://dx . doi . org/10 . 7554/eLife . 20047 . 005 Turning to our main question about choice biases , we simulated performance of the task by running the model in a continuous session ( Figure 2D ) . Thus , rather than resetting the model state at the start of each trial , as in previous work ( Bonaiuto and Arbib , 2014; Hämmerer et al . , 2016; Wang , 2002; Bonaiuto and Bestmann , 2015 ) , we used the network state at the end of the previous trial as the starting state of the next trial ( Rustichini and Padoa-Schioppa , 2015 ) . The network displays sustained recurrent dynamics , due in large part to the slow time constants of the NMDA receptors modeled in the pyramidal cell populations . As a consequence , residual activity from the previous trial may still influence the dynamics of the network when the task-related inputs of the subsequent trial arrive ( Figure 2D ) . We next asked if the model behavior exhibited any choice hysteresis , and whether this systematically related to any neural hysteresis effects . We analyzed possible choice hysteresis effects in the model behavior by separating trials into two groups based on the decision made in the previous trial ( Left*: trials where left was chosen in the previous , and Right*: trials following rightward choices ) . For each group we then fit the percentage of rightward choices to a sigmoid function of the coherence to the left or right ( Padoa-Schioppa , 2013; Rustichini and Padoa-Schioppa , 2015 ) . We found that this choice function was shifted according to the previously selected direction , reflecting a tendency to repeat the previous choice . This effect was particularly pronounced during difficult trials ( Figure 2E ) . We defined the ‘indecision point’ as the level of coherence where rightward choices were made 50% of the time , and compared this value between Left* and Right* trials for each virtual subject across stimulation conditions . The model predicts a significant shift in indecision point depending upon the choice made in the previous trial ( W ( 19 ) = 21 , p=0 . 002; Figure 2F ) . This result was confirmed with a logistic regression analysis which more precisely accounted for the relative influences of current trial coherence and previous choice on decisions ( Padoa-Schioppa , 2013; Rustichini and Padoa-Schioppa , 2015 ) ( Figure 2G ) , and again found a significant influence of the previous choice on the decision ( W ( 19 ) = 10 , p<0 . 001; Figure 2H ) . The model suggests that biases in decaying tail activity from the previous trial can cause choice hysteresis . One would then expect that perturbation of the neural dynamics of the model alters hysteresis biases in a systematic way . We therefore asked how stimulation of our model altered its dynamics , and how these influence the model’s behavior . We injected an additional trans-membrane current into pyramidal cells and inhibitory interneurons , with the polarity and magnitude based on simulations that reproduce tDCS-induced changes in sensory evoked potentials ( Molaee-Ardekani et al . , 2013 ) and behavior ( Bonaiuto and Bestmann , 2015 ) in vivo , and taking into account the cellular effects of tDCS ( Hämmerer et al . , 2016; Rahman et al . , 2013; Nitsche and Paulus , 2011; Bikson et al . , 2004; Bonaiuto and Bestmann , 2015; Funke , 2013; Radman et al . , 2009; Bindman et al . , 1964 ) . One advantage of combining experimental human studies with computational models is that it allows for interrogation of the putative neural dynamics of the model under different experimental manipulations ( Hämmerer et al . , 2016; Bonaiuto and Bestmann , 2015; Fröhlich , 2015; Bikson et al . , 2015; Bestmann , 2015; de Berker et al . , 2013 ) . Relative to no stimulation , there was no effect of depolarizing or hyperpolarizing stimulation on the model’s accuracy threshold ( depolarizing: W ( 19 ) = 65 , p=0 . 135; hyperpolarizing: W ( 19 ) = 74 , p=0 . 247; Figure 2A ) . This is consistent with previous work showing that for low levels of stimulation intensity ( such as that used in these simulations ) , the resulting shifts in membrane potential are insufficient to completely reverse the model dynamics such that it significantly alters choice accuracy ( Bonaiuto and Bestmann , 2015 ) . However , we found that depolarizing stimulation decreased decision time , whilst hyperpolarization increased it ( Figure 2B ) . We then analyzed the difference in decision time between no stimulation and stimulation conditions at each motion coherence level . In both stimulation conditions , this difference is strongest for difficult , low coherence trials , as indicated by the significant slopes in the linear fits between coherence and decision time difference ( depolarizing: B1 = 89 . 251 , p=0 . 017; hyperpolarizing: B1 = −77 . 327 , p=0 . 034; Figure 2C ) . This is because during difficult trials ( low coherence ) , shifts in membrane potential induced by depolarizing stimulation cause the winning population to reach the response threshold earlier , compared to no stimulation , while hyperpolarizing stimulation delays this event . However , during high coherence trials , the strong difference in task-related input strengths overwhelms the subtle effects of membrane potential changes . These simulations therefore predict that response time should be unaffected by subtle changes in network dynamics caused by stimulation on ‘no brainer’ trials in which strong inputs provide unequivocal evidence for one response over the other . This echoes findings from human experiments that tDCS may interact with task difficulty and/or individual differences in performance ( Benwell et al . , 2015; Jones and Berryhill , 2012 ) . The model thus predicts that network stimulation will affect response time , especially in difficult trials , but leave accuracy largely unaffected . It is predicted that depolarizing and hyperpolarizing stimulation will lead to faster and slower responses , respectively . We obtained qualitatively similar results in simulations controlling for the input parameters and effects of stimulation on interneurons , but not those that violate the known neural effects of stimulation ( Tables 1 and 2 , see Materials and methods ) . 10 . 7554/eLife . 20047 . 006Table 1 . Accuracy threshold statistics . DOI: http://dx . doi . org/10 . 7554/eLife . 20047 . 006DepolarizingHyperpolarizingW ( 19 ) p W ( 19 ) P Total task-related input firing rates = 60 Hz 620 . 108670 . 156Refresh rate = 30 Hz 690 . 179570 . 073Refresh rate = 120 Hz 340 . 008780 . 314Inhibitory interneuron stimulation760 . 279920 . 627Pyramidal cell stimulation only910 . 601890 . 55Uniform stimulation840 . 433430 . 021Reinitialization810 . 37910 . 601Accumulator530 . 052540 . 05710 . 7554/eLife . 20047 . 007Table 2 . Decision time difference statistics . DOI: http://dx . doi . org/10 . 7554/eLife . 20047 . 007DepolarizingHyperpolarizingB1 p B1 p Total task-related input firing rates = 60 Hz 50 . 4420 . 043−56 . 366 0 . 037Refresh rate = 30 Hz 90 . 2720 . 024−83 . 599 0 . 036Refresh rate = 120 Hz 93 . 2890 . 015−90 . 707 0 . 03Inhibitory interneuron stimulation106 . 9580 . 02716 . 9130 . 704Pyramidal cell stimulation only87 . 4960 . 028−77 . 929 0 . 045Uniform stimulation60 . 710 . 15756 . 5220 . 168Reinitialization72 . 8050 . 037−83 . 953 0 . 035Accumulator108 . 859<0 . 001 −44 . 87 0 . 008 In addition to decision time , we found significant effects of model stimulation on choice hysteresis . Depolarizing stimulation increased the indecision point shift , relative to no stimulation , ( W ( 19 ) = 44 , p=0 . 023 ) , whereas hyperpolarizing stimulation decreased it ( W ( 19 ) = 41 , p=0 . 017 ) . This result was echoed in a logistic regression analysis , which showed that depolarizing stimulation increased the relative influence of the previous choice to coherence ( W ( 19 ) = 32 , p=0 . 006 ) , while hyperpolarizing stimulation reduced this ratio ( W ( 19 ) = 42 , p=0 . 019 ) . In other words , the model demonstrated that choice hysteresis is caused by residual activity from the previous trial . Moreover , depolarizing stimulation increases this residual activity , while hyperpolarizing stimulation suppresses it . These results were replicated in alternative simulations using similar assumptions about the effects of stimulation , but not in those where the initial state of the network is reset at the start of each trial , or where the effects of stimulation were qualitatively different ( Table 3 , see Materials and methods ) . 10 . 7554/eLife . 20047 . 008Table 3 . Choice hysteresis simulation statistics . DOI: http://dx . doi . org/10 . 7554/eLife . 20047 . 008Indecision point shifts ( Left*-Right* ) Logistic regression ( a2/a1 ) DepolarizingHyperpolarizingDepolarizingHyperpolarizingW ( 19 ) p W ( 19 ) p W ( 19 ) p W ( 19 ) p Total task-related input firing rates = 60 Hz 420 . 019420 . 019500 . 04500 . 04Refresh rate = 30 Hz 490 . 037320 . 006460 . 028520 . 048Refresh rate = 120 Hz 380 . 013420 . 019440 . 023310 . 006Inhibitory interneuron stimulation only650 . 135800 . 351400 . 015620 . 108Pyramidal cell stimulation only480 . 033440 . 023480 . 033340 . 008Uniform stimulation790 . 332430 . 021990 . 823730 . 232Reinitialization740 . 247960 . 737660 . 145950 . 709Accumulator540 . 057950 . 709710 . 204850 . 455 As can be seen in Figure 3 , each pyramidal population fires at approximately 3–15 Hz prior to the onset of the task-related inputs . We sorted the population firing rates of each trial based on which pyramidal population was eventually chosen , and then split trials into those in which the previous choice was repeated and those where a different choice was made . We found that in trials in which the previous choice was repeated , the mean firing rate of the chosen population was slightly higher than that of the unchosen population prior to onset of task-related input ( Figure 3A , C ) . This effect can be attributed to decaying tail activity from the previous trial , which we refer to as hysteresis bias . This bias was amplified by depolarizing ( W ( 19 ) = 4 , p<0 . 001 ) and attenuated by hyperpolarizing stimulation ( W ( 19 ) = 6 , p<0 . 001; Figure 3C ) . The network was only able to overcome the bias and make a different choice from the one it made in the previous trial when the bias was very small and the model activity was dominated by the task-related inputs ( Figure 3B , D ) . 10 . 7554/eLife . 20047 . 009Figure 3 . Decaying tail activity causes neural hysteresis effects . ( A ) Example trial in which the previous choice was repeated , showing a marked difference in the decaying tail activity from the previous trial between the eventually chosen and unchosen pyramidal populations , prior to the onset of the stimulus . ( B ) Example trial in which the pre-stimulus difference between chosen and unchosen firing rates is small . As a consequence , the bias in activity at stimulus onset is not strong enough to influence the choice . ( C ) The mean difference in firing rates in the 500 ms prior to the onset of the task-related inputs ( magnified region in A ) is amplified by depolarizing ( red ) and suppressed by hyperpolarizing ( green ) stimulation , relative to no stimulation ( blue ) . ( D ) When the bias in pre-stimulus activity is relatively small , the model behavior is dominated by the task-related inputs , and therefore the model is able to overcome the hysteresis bias and make a different choice . See Figure 3—source data 1 for raw data . DOI: http://dx . doi . org/10 . 7554/eLife . 20047 . 00910 . 7554/eLife . 20047 . 010Figure 3—source data 1 . Model prestimulus firing rates in repeated and non-repeated choice trials . DOI: http://dx . doi . org/10 . 7554/eLife . 20047 . 010 If decaying tail activity in the chosen pyramidal population from the previous trial causes behavioral choice hysteresis effects , these effects should diminish with longer inter-stimulus intervals ( ISIs ) . Given a long enough ISI , residual pyramidal activity is more likely to fully decay back to baseline firing rates , allowing unbiased competition on the following trial . The simulations described above used an ISI of 2 s ( matching the human experiment ) , which results in trials often beginning before residual activity from the previous trial has completely decayed ( Figure 2D ) . In additional control stimulations using a range of ISIs , choice hysteresis behavior indeed decreased as the time between stimuli increased , as evidenced by shifts toward zero in the mean indecision point ( main effect of ISI: F ( 3 , 76 ) = 14 . 439 , p<0 . 001; Figure 4A ) and the influence of the previous choice on the current decision ( main effect of ISI: F ( 3 , 76 ) = 24 . 196 , p<0 . 001; Figure 4B ) . 10 . 7554/eLife . 20047 . 011Figure 4 . Behavioral choice hysteresis diminishes with longer interstimulus intervals . ( A ) The mean of the indecision point shift decreases with longer interstimulus intervals ( ISIs ) , reflecting a smaller choice hysteresis effect . ( B ) Similarly , the ratio a1/ a2 from the logistic regression , representing the relative influence of the previous choice on the current choice relative to the influence of coherence , decreases with increasing ISIs . Choice hysteresis is strongest for short ISIs of 1 . 5 s , and disappears for the longest ISI of 5 s . See Figure 4—source data 1 for raw data . DOI: http://dx . doi . org/10 . 7554/eLife . 20047 . 01110 . 7554/eLife . 20047 . 012Figure 4—source data 1 . Model choice hysteresis behavior with increasing ISIs . DOI: http://dx . doi . org/10 . 7554/eLife . 20047 . 012 Two mechanisms determine behavior in competitive attractor network models: recurrent excitation within each pyramidal population and mutual inhibition between these populations via a common pool of inhibitory interneurons . Pure accumulator models such as the drift diffusion model are an alternate class of decision making models that do not include mutual inhibition ( Ratcliff and McKoon , 2008 , 1998; Ratcliff , 1978 ) . In these models , separate units integrate their inputs representing evidence for the corresponding option , and a decision is made when one unit reaches a predefined threshold . We tested whether separate integrators would make the same choice hysteresis predictions as the competitive attractor model . We split the interneuron population into two subpopulations , each exclusively connected with the corresponding pyramidal population ( Figure 5A ) . The pyramidal populations could thus integrate their inputs through their recurrent excitatory connections , but could not exert any inhibitory influence on each other . All other parameters were kept the same , except for the background input firing rate and response threshold , as the resulting network became more sensitive to these values ( see Materials and methods ) . The accumulator version of the model made the same qualitative predictions as the competitive attractor version concerning accuracy and decision time ( Figure 5B–D ) : choice accuracy is not affected by depolarizing ( W ( 19 ) = 53 , p=0 . 052 ) or hyperpolarizing stimulation ( W ( 19 ) = 54 , p=0 . 057 ) , but depolarizing and hyperpolarizing stimulation speeds and slows decision time , respectively , and these effects are reduced with increasing coherence ( depolarizing: B1 = 108 . 859 , p<0 . 001; hyperpolarizing: B1 = −44 . 87 , p=0 . 008 ) . However , the model did not exhibit significant choice hysteresis ( indecision point shift: W ( 19 ) = 77 , p=0 . 296; a2/a1: W ( 19 ) = 86 , p=0 . 478; Figure 5E , F ) . This is because the chosen pyramidal population does not inhibit the other population , allowing decaying trace activity from both populations to extend into the next trial ( Figure 5G ) . Therefore , on the next trial both populations can be similarly biased . Neither depolarizing ( indecision point shift: W ( 19 ) = 54 , p=0 . 057; a2/a1: W ( 19 ) = 71 , p=0 . 204 ) nor hyperpolarizing stimulation ( indecision point shift: W ( 19 ) = 95 , p=0 . 709; a2/a1: W ( 19 ) = 85 , p=0 . 455 ) had any effect on choice hysteresis . 10 . 7554/eLife . 20047 . 013Figure 5 . Accumulator model architecture and simulations . ( A ) In the accumulator version of the model , the inhibitory interneuron population is split into two subpopulations , each connected exclusively to the corresponding pyramidal population . The pyramidal populations can thus only integrate the task related inputs through their recurrent connectivity and cannot inhibit each other . ( B ) Neither depolarizing nor hyperpolarizing stimulation significantly changed the decision threshold . ( C ) As with the competitive attractor model , decision time decreases with increasing coherence , and depolarizing stimulation speeds decision time , while hyperpolarizing stimulation slows decisions . ( D ) The effects of stimulation on decision time are reduced with increasing coherence . ( E ) There is no shift in indecision point when sorting trials based on the previous choice , and stimulation does not affect this . ( F ) The relative influence of the previous choice on the current decision is nearly zero , and this is not changed by stimulation . ( G ) . Neural dynamics of the accumulator model . The losing pyramidal population is not inhibited and therefore residual activity from both populations carries over into the next trial . See Figure 5—source data 1 for raw data . DOI: http://dx . doi . org/10 . 7554/eLife . 20047 . 01310 . 7554/eLife . 20047 . 014Figure 5—source data 1 . Accumulator model accuracy , decision time , and choice hysteresis with simulated network stimulation . DOI: http://dx . doi . org/10 . 7554/eLife . 20047 . 014 We next asked whether the predictions from our simulated stimulation were borne out in the behavior of human participants undergoing tDCS over dlPFC , a region strongly implied in controlling human perceptual choice ( Heekeren et al . , 2004 , 2006; Philiastides et al . , 2011; Rahnev et al . , 2016; Georgiev et al . , 2016 ) . 24 human participants performed the perceptual decision making task simulated in the model , in which they viewed a RDK and were required to indicate the direction of coherent motion ( Figure 6A ) . To test the model’s predictions concerning the effects of perturbed dynamics on choice hysteresis , we applied tDCS over the left dlPFC in order to induce depolarizing or hyperpolarizing network stimulation , or sham stimulation ( Figure 6B , C ) . This region is implicated in perceptual decision making , independent of stimulus and response modality ( Heekeren et al . , 2004 , 2006 ) , and it has been suggested that it operates using competitive attractor networks similar to the one we used ( Bonaiuto and Arbib , 2014; Rolls et al . , 2010; Compte et al . , 2000; Wimmer et al . , 2014 ) . Furthermore , transcranial magnetic stimulation ( TMS ) over this region disrupts perceptual decisions , suggesting that it plays a necessary role in this process ( Philiastides et al . , 2011; Rahnev et al . , 2016; Georgiev et al . , 2016 ) . However , here we employed tDCS , which subtly polarizes membrane potentials through externally applied electrical currents ( Rahman et al . , 2013; Nitsche and Paulus , 2011; Bikson et al . , 2004 ) , instead of disrupting ongoing neural activity , as with TMS . This allowed us to alter the dynamics of the human dlPFC in an analogous way to the model simulations . 10 . 7554/eLife . 20047 . 015Figure 6 . Experiment design and behavioral effects of stimulation in human participants . ( A ) Behavioral task . ( B ) Experimental protocol . ( C ) Electrode positions and estimated current distribution during left dorsolateral prefrontal stimulation . ( D–E ) Neither depolarizing nor hyperpolarizing stimulation altered the decision thresholds of human participants ( model predictions shown in shaded colors in the background ) . ( F–H ) In human participants depolarizing and hyperpolarizing stimulation significantly decreased and increased the decision time , respectively , relative to sham stimulation , matching the model predictions shown in matte colors . See Figure 6—source data 1 for raw data . DOI: http://dx . doi . org/10 . 7554/eLife . 20047 . 01510 . 7554/eLife . 20047 . 016Figure 6—source data 1 . Human participant accuracy and reaction time . DOI: http://dx . doi . org/10 . 7554/eLife . 20047 . 016 As expected ( Palmer et al . , 2005 ) , in sham stimulation blocks , the accuracy of human participants increased ( Figure 6D , E ) and response times decreased ( Figure 6F , G ) with increasing motion coherence . In striking accordance with the predictions of our biophysical model , neither depolarizing nor hyperpolarizing stimulation had an effect on the accuracy threshold of human participants ( depolarizing: W ( 23 ) = 145 , p=0 . 886; hyperpolarizing: W ( 23 ) = 118 , p=0 . 361; Figure 6D , E ) . However , tDCS over left dlPFC in our human participants showed the predicted pattern of effects on response time for depolarizing and hyperpolarizing stimulation , compared to sham stimulation ( depolarizing: B1 = 66 . 938 , p=0 . 004; hyperpolarizing: B1 = −49 . 677 , p=0 . 04; Figure 6F–H ) . The behavior of human participants additionally demonstrated choice hysteresis effects . Just as predicted by the model and in line with experimental work with humans and nonhuman primates ( Padoa-Schioppa , 2013; Samuelson and Zeckhauser , 1988 ) , the indecision points of human participants shifted when sorting their trials according to the preceding choice , and a logistic regression revealed a significant influence of the previous choice relative to coherence on decisions . We performed the same analyses used to examine choice hysteresis in the model on behavioral data from the human participants , now comparing each stimulation block to the preceding sham block in the same session . Exactly as predicted by the model , a shift in indecision point , which indicates a choice hysteresis effect , was observed under sham stimulation ( depolarizing sham: W ( 23 ) = 79 , p=0 . 043; hyperpolarizing sham: W ( 23 ) = 47 , p=0 . 003 ) , and this effect was amplified by depolarizing stimulation ( W ( 23 ) = 78 , p=0 . 04; Figure 7A ) and reduced by hyperpolarizing stimulation ( W ( 23 ) = 59 , p=0 . 009; Figure 7B ) , relative to the preceding sham block . The results of the logistic regression confirmed these results , with a nonzero ratio of the influence of the previous choice to that of the current coherence in the sham stimulation blocks ( depolarizing sham: W ( 23 ) = 73 , p=0 . 028; hyperpolarizing sham: W ( 23 ) = 47 , p=0 . 003 ) , which was increased by depolarizing ( W ( 23 ) = 56 , p=0 . 007; Figure 7D ) and decreased by hyperpolarizing stimulation ( W ( 23 ) = 78 , p=0 . 04; Figure 7E ) relative to the preceding sham block . We therefore found that in both the model and in human participants , polarization of the dlPFC led to changes in reaction times and choice hysteresis effects in a perceptual decision making task . The neural dynamics of the model suggest that these changes are explained by alterations of sustained recurrent activity following a trial , which biases the decision process in the next trial . 10 . 7554/eLife . 20047 . 017Figure 7 . Behavioral effects of stimulation on choice hysteresis . ( A–B ) Depolarizing stimulation positively shifted the indecision point in human participants , while hyperpolarization caused the opposite effects , relative to sham stimulation . ( C ) These results are in line with model predictions ( repeated from Figure 2F ) . ( D–E ) The relative influence of the previous choice in the current decision of human participants was increased by depolarizing and decreased by hyperpolarizing stimulation , relative to sham . ( F ) These results are just as predicted by the model ( repeated from Figure 2H ) . See Figure 7—source data 1 for raw data . DOI: http://dx . doi . org/10 . 7554/eLife . 20047 . 01710 . 7554/eLife . 20047 . 018Figure 7—source data 1 . Human participant behavioral choice hysteresis . DOI: http://dx . doi . org/10 . 7554/eLife . 20047 . 018 We next sought to further test the predictions of our model concerning the effect of ISIs on choice hysteresis . A separate group of participants ( N = 24 ) , performed a version of the task without stimulation in which the ISI was either 1 . 5 s or 5 s ( see Materials and methods ) . Exactly as predicted by the model , trials following short ISIs had larger indecision point shifts ( W ( 22 ) = 67 , p=0 . 031; Figure 8A ) and were more influenced by the previous choice ( W ( 22 ) = 51 , p=0 . 008; Figure 8C ) compared to trials following longer ISIs . In fact with an ISI of 5 s , the indecision point was not significantly shifted ( W ( 22 ) = 82 , p=0 . 089 ) , nor was there a detectable influence of the previous choice on the current decision ( W ( 22 ) = 84 , p=0 . 101 ) , meaning that choice hysteresis had disappeared . The model explains this effect by the decay rate of the sustained recurrent activity in the winning pyramidal population , which has sufficient time to fall to baseline levels with long ISIs . 10 . 7554/eLife . 20047 . 019Figure 8 . Choice hysteresis in human participants diminishes with longer interstimulus intervals . ( A ) The mean of the indecision point in human participants approaches zero with a 5 s interstimulus intervals ( ISIs ) , reflecting a smaller choice hysteresis effect . ( B ) These results are similar to those predicted by the model ( repeated from Figure 4A ) . ( C ) The relative influence of the previous choice on the current decision is smaller with a 5 s ISI than a 1 . 5 s ISI . ( D ) These results match the predictions of the model ( repeated from Figure 4B ) . See Figure 8—source data 1 for raw data . DOI: http://dx . doi . org/10 . 7554/eLife . 20047 . 01910 . 7554/eLife . 20047 . 020Figure 8—source data 1 . Human participant behavioral choice hysteresis with increasing ISIs . DOI: http://dx . doi . org/10 . 7554/eLife . 20047 . 020 The indecision point shifts and logistic parameter ratios were not significantly different between the sham conditions of the stimulation experiment and the 1 . 5 s ISI condition ( Mann-Whitney U test , indecision point shift , depolarizing sham: U = 253 , p=0 . 632 , indecision point shift , hyperpolarizing sham: U = 259 , p=0 . 726 , previous choice influence , depolarizing sham: U = 237 , p=0 . 413 , previous choice influence , hyperpolarizing sham: U = 254 , p=0 . 647 ) . In the stimulation experiment , the overall mean RT in sham conditions was 586 ms , resulting in a mean ISI of 1 . 914 s . It is therefore likely not different enough from the 1 . 5 s ISI condition to judge a significant difference between participant groups with our sample size . Our neural network model displayed decaying tail activity from the previous trial , leading to a ‘choice hysteresis’ effect , or a tendency to repeat the last choice , especially during difficult trials . This effect diminished with longer ISIs which allowed the tail activity to fully decay . Simulation of depolarizing noninvasive brain stimulation in this model decreased decision time and amplified this choice hysteresis effect , while hyperpolarizing stimulation increased decision time and suppressed choice hysteresis . These behavioral effects were caused by changes in the level of sustained activity from the previous trial , placing the network closer or further away to one of its attractor states , biasing the response to one or the other option , thus speeding or slowing decisions . Human participants demonstrated the same choice hysteresis bias , which was diminished with longer ISIs ( but see Akaishi et al . , 2014 ) and modulated by noninvasive brain stimulation over the left dlPFC in the same way . This supports the idea that processes related to perceptual decisions in left dlPFC are underpinned by processes that can be approximated by a competitive attractor network as used here . Although not statistically significant , simulated depolarizing stimulation slightly decreased choice accuracy , while hyperpolarizing stimulation improved it . This is not surprising , given the effects of stimulation on decision time and the use of a firing rate threshold as a decision criterion . However , in previous studies with a similar model we found that decision making accuracy is affected at higher levels of stimulation intensity ( Bonaiuto and Bestmann , 2015 ) . The stimulation intensities used in the present simulations were matched to those used in the human experiments , and not sufficiently high to polarize the network enough to completely override differences in task-related inputs ( left/right motion coherence ) . Previous work has shown that repetitive TMS over left dlPFC reduces both accuracy and increases response time in a similar task ( Philiastides et al . , 2011 ) . However , TMS elicits instantaneous synchronized activity within the area of stimulation , and thus likely disrupts ongoing activity , providing an ‘override’ of difference in task-related inputs . By contrast , tDCS subtly alters neural dynamics through small de- or hyper-polarizing currents without directly eliciting spikes ( Radman et al . , 2009; Bikson et al . , 2013 ) . With the number of trials used and the variability observed in the present task , it is likely that stimulation effects were only apparent in response time because response time is a more sensitive performance metric than choice accuracy . More generally , our results show that the possible mechanisms through which non-invasive brain stimulation alters behavior can be interrogated through the use of biologically informed computational modeling approaches . Computational neurostimulation of the kind employed here may thus provide an important development for mechanistically informed rationales for the application of neurostimulation in both health and disease . Here , we simulated the effects of both depolarizing and hyperpolarizing dlPFC stimulation . In these simulations , currents affected both pyramidal cells as well as interneurons . This modeling choice was based on previous work showing that simulated tDCS must affect both pyramidal cells and interneurons in order to explain changes in sensory evoked potentials observed in vitro ( Molaee-Ardekani et al . , 2013 ) . Some accounts of the neurophysiological effects of tDCS suggest that pyramidal neurons are predominantly affected ( Radman et al . , 2009 ) , but in additional simulations in which stimulation was only applied to the pyramidal cells the results were qualitatively similar ( Tables 1–3 ) . The level of stimulation intensity and montages we used for our human experiments are based on current modeling estimates of the mean field strength in dlPFC and in vitro measurements of pyramidal cell and interneuron polarization as a function of field strength ( Bikson et al . , 2004 ) . We also used individual structural MRIs to optimize electrode placement for each participant , using relatively small stimulation electrodes that straddled our target site in left dlPFC . Specifically , electrode positions relative to the MNI template were determined using current modeling to maximize current flow through the superior frontal sulcus portion of the left dlPFC , creating radial inward or outward current flow . Each participant’s MRI was aligned to the template and the inverse of this transformation was used to derive optimal electrode positions to generate current flow through the same sulcus . This approach ensured that stimulation was relatively confined over left prefrontal cortex , and that the direction of current flow through the target site was comparable across subjects . However , in our simulations neurons were not spatially localized and polarization was applied uniformly to all neurons within a population . Future computational neurostimulation studies should investigate the effects of heterogeneous polarization due to variable patterns of current flow through brain tissue . We targeted the left dlPFC , but perceptual decision making involves a network of cortical regions including the middle temporal area MT ( Salzman et al . , 1992; Britten et al . , 1993; Celebrini and Newsome , 1994 ) and the lateral intraparietal area LIP ( Hanks et al . , 2006; Shadlen and Newsome , 1996 ) . However , the left dlPFC is an attractive target for our aims for a number of reasons . The activity of neurons in dlPFC both predicts the upcoming response and reflects information about the sensory stimuli , suggesting that this region makes an integral contribution to transforming sensory information into a decision ( Kim and Shadlen , 1999 ) . Furthermore , in human perceptual decision making , left dlPFC is activated independent of the both the stimulus type and response modality ( Heekeren et al . , 2004 , 2006; Pleger et al . , 2006; Wenzlaff et al . , 2011; Ruff et al . , 2010; Philiastides and Sajda , 2007; Kovács et al . , 2010; Donner et al . , 2007; Ostwald et al . , 2012; Zhang et al . , 2013 ) , and has been shown to play a causal role in the process ( Philiastides et al . , 2011 ) . We optimized the electrode locations to stimulate the left dlPFC , while leaving premotor and motor cortices unaffected , as stimulation of these regions could induce response biases . This may not have been possible with stimulation over parietal cortex . Finally , neural hysteresis in another frontal region , orbitofrontal cortex , has been linked to behavioral choice hysteresis in value-based decisions in nonhuman primates ( Padoa-Schioppa , 2013 ) . Recent work suggests that this region can indeed be targeted with tDCS ( Hämmerer et al . , 2016 ) , and whether choice hysteresis for value-based decision can be similarly molded as in the present study remains to be seen . However , decaying trace activity in left dlPFC could partially reflect lingering activity in afferent regions , and this possibility could be addressed in future research using a multiregional computational neurostimulation approach . Participants made all responses using the hand contralateral to the site of stimulation ( i . e . the right hand ) . It is therefore not clear what effect ipsilateral stimulation would have . However , in nonhuman primates , neurons in dlPFC are active during perceptual decision making whether the choice is indicated with a button press ( Hussar and Pasternak , 2013 ) or a saccade ( Kim and Shadlen , 1999; Kiani et al . , 2014; Opris and Bruce , 2005 ) . In humans , left dlPFC is activated during perceptual decision regardless of the response modality ( e . g . , saccades versus button presses [Heekeren et al . , 2006] ) , right-handed button presses ( Heekeren et al . , 2006; Ruff et al . , 2010; Philiastides and Sajda , 2007; Donner et al . , 2007 ) , left-handed button presses ( Pleger et al . , 2006 ) , and bimanual button presses ( Wenzlaff et al . , 2011 ) . It is thus reasonable to expect that stimulation of the left dlPFC would affect both hands in the same way . Investigating the potential lateralization of stimulation-induced effects on dlPFC as well as other regions in the perceptual decision making network such as MT and LIP , is an interesting avenue for further research for which the computational neurostimulation approach employed here could be leveraged . We found no choice hysteresis bias in the accumulator version of the model , with or without simulated stimulation . However , compared to the competitive attractor model , the accumulator model made very similar predictions concerning choice accuracy and decision time , as well as the effects of stimulation on these measures . In our simulations , the lack of mutual inhibition in the accumulator model allows residual activity from both pyramidal populations to bias the following decision , resulting in zero net bias . In constructing the accumulator model , we sought to alter the competitive attractor model as little as possible , keeping most parameters at the same value . However , we found that without mutual inhibition , the firing rates of the resulting network were much more sensitive to the noisy background inputs , and therefore we had to restrict the range of background firing rates used and scale the response threshold accordingly . It is possible that there are some sets of parameter values that would cause the accumulator model to exhibit choice hysteresis , but a systematic search of the parameter space of this model is beyond the scope of this study . While the model we used to simulate the dlPFC is well established , it is a general model used to study decision making . As such , it does not take into account the specific architecture and detailed connectivity of the dlPFC; however data at this level of detail in general are not currently available . The choice hysteresis behavior in human participants qualitatively matched that of the model , but the model predicted a larger effect than that observed ( Figures 7 and 8 ) . Factors such as the contribution of other brain regions known to be involved in perceptual decision making may explain these quantitative differences . Future studies should involve increasingly realistic biophysical multi-region models that can take this information into account . tDCS is widely used in basic and translational studies for reversible and controlled modulation of neural circuit activity in the human brain . However , there is a distinct lack of mechanistic models that not only explain how stimulation affects neural network dynamics , but also how these changes alter behavior . There are several conceptual models of the effects of noninvasive brain stimulation , but the explanations that they offer typically make leaps across several levels of brain organization and don’t consider how neural circuits generate behavior ( de Berker et al . , 2013; Bestmann et al . , 2015 ) . Efforts have been made in this direction ( Hämmerer et al . , 2016; Rahman et al . , 2013; Bonaiuto and Bestmann , 2015; Molaee-Ardekani et al . , 2013; Fröhlich , 2015; Bestmann , 2015; Neggers et al . , 2015; Hartwigsen et al . , 2015; Miniussi et al . , 2013; Rahman et al . , 2015 ) , but there have been very few computational models that offer an explanation for the behavioral effects of tDCS in terms of neural circuit dynamics ( Hämmerer et al . , 2016; Bonaiuto and Bestmann , 2015; Douglas et al . , 2015 ) . We here present the first biophysically informed modeling study of tDCS effects during perceptual decision making , and provide detailed hypotheses about the changes in neural circuits that translate to observed behavioral changes during stimulation . We have shown that a biophysical attractor model generates perceptual decision making behavior accurately matching that of human participants . Additionally , we used computational neurostimulation of this model to predict the effect of tDCS over left dlPFC on choice hysteresis , which we then confirmed experimentally . Previous work showing that changes to parameters in diffusion models can explain differences in human perceptual decision making after transcranial magnetic stimulation of left dlPFC ( Philiastides et al . , 2011; Rahnev et al . , 2016; Georgiev et al . , 2016 ) . Our results extend these findings by addressing the neural circuitry behind perceptual decision making processes . This allows us to capture the influence of previous neural activity on the current choice in a natural way , and to offer mechanistic explanations for the effects of stimulation on neural dynamics in the left dlPFC . We provide interventional evidence that the left dlPFC integrates and compares perceptual information through competitive interactions between neural populations , and that decaying trace activity from previous trials influences the current choice by biasing the decision toward the repeating the last choice . The model contains 2000 neurons , and consists of one population of 1600 pyramidal cells and one population of 400 inhibitory interneurons . The pyramidal cells contain two subpopulations of 240 neurons each , selective for the ‘left’ , pL , and ‘right’ , pR , choice options , with the remaining neurons non-selective for either option . The neurons in the pyramidal subpopulations form excitatory reciprocal connections with neurons in the same population , and mutually inhibit each other indirectly via projections to and from a common pool of inhibitory interneurons . The neurons in the pyramidal population project to excitatory synapses ( AMPA and NMDA ) on target cells and the interneurons project to inhibitory synapses on their targets ( GABAA ) . All neural populations receive stochastic background input from a common pool of Poisson spike generators , causing each neuron to spontaneously fire at a low rate . The pL and pR pyramidal subpopulations additionally receive task-related inputs as Poisson distributed random spikes signaling the perceived evidence for each response option using the same scheme as Wang ( Wang , 2002 ) . The mean rates of the task-related inputs , μL and μR , vary linearly with the coherence level of the simulated RDK ( Figure 1A , inset ) . Importantly , the sum of the mean task-related input rates always equals 80 Hz , meaning that decision making behavior has to emerge from network dynamics and input structure and cannot be attributed to differences in the overall level of task-related input stimulation . The firing rate of each task-related input at each time point was normally distributed around the mean ( σ = 4 Hz ) and changed according to the refresh rate of the monitor used in our experiment ( Figure 1B , left column ) . In additional simulations we show that the behavior of the model is qualitatively robust to changes in the total task-related input firing rates and refresh rate ( Tables 1–3 ) . For analysis , we compute mean population firing rates by convolving the instantaneous population firing rate with a Gaussian filter 5 ms wide at the tails . The winner-take-all dynamic of the network causes the firing rates of the pyramidal populations to magnify differences in the inputs . This is due to the reciprocal connectivity and structure of the network , which endows it with bistable attractor states , resulting in competitive dynamics ( Camperi and Wang , 1998; Wilson and Cowan , 1972 ) . As the firing rate of one population increases , it increasingly inhibits the other population via the common pool of inhibitory interneurons . This further increases the activity of the winning population as the inhibitory activity caused by the other population decreases . These competitive dynamics result in one pyramidal population ( typically the one receiving the strongest input ) firing at a relatively high rate , while the firing rate of the other population decreases to approximately 0 Hz ( Figure 1B ) . Here we provide a detailed description of the architecture of the biophysical attractor model we used . We modeled synapses as exponential ( AMPA , GABAA ) conductances , or bi-exponential conductances ( NMDA ) . Synaptic conductances are governed by the following equation: ( 1 ) g ( t ) =Ge−t/τ where G is the maximal conductance ( or weight ) of that specific synapse type ( AMPA or GABAA ) , and τ is the decay time constant for that synapse type . Thus when a spike arrives at this synapse at time t , the conductance , g , is set to its maximal value , G ( because e-t/τ is bounded by 0 and 1 ) , after which it decays at a rate determined by τ . Similarly , bi-exponential synaptic conductances are determined by: ( 2 ) g ( t ) =Gτ2τ2−τ1 ( e−t/τ1−e−t/τ2 ) where τ1 and τ2 are rise and decay time constants . Synaptic currents are computed from the product of these conductances and the difference between the membrane potential and the synaptic current reversal potential , E: ( 3 ) I ( t ) =g ( t ) ( Vm−E ) where Vm is the membrane voltage . NMDA synapses have an additional voltage dependence , which is captured by: ( 4 ) INMDA ( t ) =gNMDA ( t ) ( Vm−ENMDA ) 1+[ Mg2+ ]exp ( −0 . 062Vm ) /3 . 57 where [Mg2+] is the extracellular magnesium concentration . The total synaptic current ( summing AMPA , NMDA , and GABAA currents ) is input into the exponential leaky integrate-and-fire ( LIF ) neural model ( Brette and Gerstner , 2005 ) : ( 5 ) Itotal ( t ) =IAMPA ( t ) +INMDA ( t ) +IGABAA ( t ) ( 6 ) CdVmdt=gL ( Vm−EL ) +gLΔTeVm−VTΔT−Itotal where C is the membrane capacitance , gL is the leak conductance , EL is the resting potential , ΔT is the slope factor ( which determines the sharpness of the voltage threshold ) , and VT is the threshold voltage . After spike generation , the membrane potential is reset to Vr and the neuron cannot generate another spike until the refractory period , τr , has passed . Intra- and inter-population connections are initialized probabilistically with axonal conductance delays of 0 . 5 ms . Parameter values are based on experimental data from the literature where possible ( Hestrin et al . , 1990; Jahr and Stevens , 1990; Salin and Prince , 1996; Spruston et al . , 1995; Xiang et al . , 1998 ) and set empirically otherwise ( Table 4 ) . 10 . 7554/eLife . 20047 . 021Table 4 . Parameter values for the competitive attractor model . DOI: http://dx . doi . org/10 . 7554/eLife . 20047 . 021ParameterDescriptionValueGAMPA ( ext ) Maximum conductance of AMPA synapses from task-related inputs1 . 6nSGAMPA ( background ) Maximum conductance of AMPA synapses from background inputs2 . 1nS ( pyramidal cells ) , 1 . 53nS ( interneurons ) GAMPA ( rec ) Maximum conductance of AMPA synapses from recurrent inputs0 . 05nS ( pyramidal cells ) , 0 . 04nS ( interneurons ) GNMDA Maximum conductance of NMDA synapses0 . 145nS ( pyramidal cells ) , 0 . 13nS ( interneurons ) GGABA-A Maximum conductance of GABAA synapses1 . 3nS ( pyramidal cells ) , 1 . 0nS ( interneurons ) τAMPA Decay time constant of AMPA synaptic conductance2 ms τ1-NMDA Rise time constant of NMDA synaptic conductance2 ms τ2-NMDA Decay time constant of NMDA synaptic conductance100 ms τGABA-A Decay time constant of GABAA synaptic conductance5 ms [Mg2+] Extracellular magnesium concentration1 mM EAMPA Reversal potential of AMPA-induced currents0 mV ENMDA Reversal potential of NMDA-induced currents0 mV EGABA-A Reversal potential of GABAA-induced currents−70 mV C Membrane capacitance0 . 5nF ( pyramidal cells ) , 0 . 2nF ( interneurons ) gL Leak conductance25nS ( pyramidal cells ) , 20nS ( interneurons ) EL Resting potential−70 mV ΔT Slope factor3 mV VT Voltage threshold−55 mV Vs Spike threshold−20 mV Vr Voltage reset−53 mV τr Refractory period2 ms ( pyramidal cells ) , 1 ms ( interneurons ) All connection probabilities are determined empirically so that the network generates winner-take-all dynamics ( Bonaiuto and Arbib , 2014; Bonaiuto and Bestmann , 2015 ) . Recurrent pyramidal population connectivity probability ( the probability that any pyramidal cell projected to an AMPA or NMDA synapse on other cells in the same population ) was 0 . 08 , and recurrent inhibitory interneuron population connections used GABAA synapses and had a connectivity probability of 0 . 1 . Projections from the pyramidal populations connected to AMPA or NMDA synapses on the inhibitory interneurons with probability 0 . 1 , connections from the inhibitory interneuron population to each pyramidal population used GABAA synapses with a connectivity probability of 0 . 2 . Thus , the pattern of connectivity between populations was fixed , but the fine-scale connectivity between individual neurons was probabilistically determined by the connectivity parameters . We simulated depolarizing tDCS by injecting a depolarizing transmembrane current into each pyramidal cell and hyperpolarizing current into each interneuron ( Bonaiuto and Bestmann , 2015; Molaee-Ardekani et al . , 2013 ) , and hyperpolarizing tDCS by adding hyperpolarizing current into pyramidal cells and depolarizing current into interneurons , a distinction which arises in cortex due to differences in orientation and cellular morphology . Note however that the results of our model are robust against these assumptions , and remain qualitatively similar when omitting current from interneurons ( see below ) . The simulated current was added to the input to each exponential LIF neuron: ( 7 ) Itotal ( t ) =IAMPA ( t ) +INMDA ( t ) +IGABAA ( t ) +Istim ( t ) where Istim ( t ) is the tDCS current at time t . The simulated tDCS current was applied for the entire duration of each block of trials . Depolarizing tDCS was simulated by injecting 0 . 75 pA into pyramidal cells and −0 . 375 pA into interneurons , while during hyperpolarizing tDCS stimulation pyramidal cells were injected with −0 . 75 pA and interneurons 0 . 375 pA ( Bonaiuto and Bestmann , 2015; Molaee-Ardekani et al . , 2013 ) . In additional simulations when we applied stimulation only to the pyramidal populations the results were qualitatively similar , however when we applied stimulation only to the interneuron population or uniform stimulation to both populations the results were very different ( see Tables 1–3 and Discussion , [Bestmann et al . , 2015; Bonaiuto and Bestmann , 2015] ) . The latter two stimulation protocols are in contrast with known physiology of polarizing currents ( Rahman et al . , 2013; Radman et al . , 2009 ) and thus served as additional tests for the specificity of our simulated membrane polarization effects on the model . The injected current simulating tDCS slightly changed the resting membrane potential of each neuron ( ±0 . 038 mV with ±0 . 75 pA injected current , ±0 . 019 mV with ±0 . 375 pA ) , within the range found by in vitro tDCS studies ( Rahman et al . , 2013; Bikson et al . , 2004; Radman et al . , 2009 ) . One advantage of computational modeling is that the model can be run for many more trials than can feasibly be tested in human participants . However , this can lead to spuriously low-variance model predictions that cannot reliably be compared with human data . In order to fairly compare model and human behavioral performance , we generated 20 virtual subjects and assessed the effects of stimulation in each subject . Virtual subjects were generated using a random seed to generate fine grained neuron-to-neuron connectivity using the connection probabilities described above . Each virtual subject had a background input firing rate sampled from a range ( 880–950 Hz ) previously used to simulate human participants in a similar decision making task ( Bonaiuto and Bestmann , 2015 ) and a response threshold uniformly sampled from a range of 18–22 Hz to capture inter-subject differences in speed-accuracy tradeoffs . Simulations with the accumulator version of the model used a background rate between 855 Hz and 870 Hz and a response threshold that varied with the rate in the range 19–36 Hz . This was necessary because without mutual inhibition , the resulting network was much less stable and more sensitive to these values . Each virtual subject was tested using the same five coherence levels that human participants were tested with ( 3 . 2 , 6 . 4 , 12 . 8 , 25 . 6 , and 51 . 2% ) , with 20 trials at each level ( 10 trials with coherent motion to the left and 10 to the right ) , for a total of 100 trials per block , randomly ordered . The sum of the two task-related inputs always equaled 80 Hz ( at coherence = 0% both inputs were at 40 Hz , and coherence = 51 . 2% one input was at 60 . 48 Hz and the other at 19 . 52 Hz ) , so the total strength of the input received by the network remains equal across all conditions . Each trial lasted for 3s , with task-related input applied from 1–2s , matching the average time course of the experiment with human participants . Three blocks of trials were run: no stimulation , depolarizing , and hyperpolarizing stimulation . In both stimulation blocks , stimulation was applied for the entire duration of the block: thus matching the experimental procedure in humans where we applied tDCS in a blocked manner . All of our model simulations were implemented in the Python programming language using the Brian simulator v1 . 4 . 1 ( Goodman and Brette , 2008 ) . The differential equations defining the model were solved using Euler integration with a time step of 0 . 5 ms . Model simulation and analysis code is available at https://github . com/jbonaiuto/perceptual-choice-hysteresis . 24 neurologically healthy volunteers participated in the stimulation experiment ( seven male , aged 23 . 75 ± 4 . 25 years ) , and a separate group of 24 participated in a control experiment assessing the influence of ISI ( nine male , aged 23 . 54 ± 3 . 32 years ) . One of the participants in the ISI experiment was excluded from analysis because of their high accuracy threshold ( >25% ) . The required number of participants was determined based on a power analysis of with an alpha of 0 . 05 , power of 0 . 8 and effect sizes estimated from previous tDCS studies targeting dlPFC ( d = 0 . 6–0 . 9 , [Boggio et al . , 2010; Fecteau et al . , 2007; Jo et al . , 2009; Fregni et al . , 2005] ) . Participants gave their informed written consent before participating and the local ethics committee approved the experiments ( reference number 5833/001 ) . Participants completed a perceptual decision making task . Participants sat comfortably at a desk in front of a computer and responded to visual stimuli displayed on a screen by pressing two keys on a keyboard using the index and middle finger of their right hand . The screen had an update rate of 60 Hz and was placed 76 cm from the participants . On each trial , participants were required to fixate in the center of a screen . After 500 ms a RDK was displayed and participants were required to press a key as soon as possible to indicate whether the direction of coherent motion was to the left or the right ( Figure 6A ) . Although only one direction of coherent motion was displayed during each trial , the task is a two alternative forced choice task , and therefore evidence must be accumulated and a decision made between the left and right direction . The RDK consisted of a 5° diameter circular aperture centered on the fixation point ( Ruzzoli et al . , 2010 ) with 0 . 1° diameter dots at a density of 16 . 7 dots/deg2/s ( Britten et al . , 1992 ) , each moving at 5°/s ( McGovern et al . , 2012 ) . The percentage of coherently moving dots was set randomly in each trial to 3 . 2 , 6 . 4 , 12 . 8 , 25 . 6 , or 51 . 2% . Trials ended once a response had been made or after a maximum of 1s if no response was made . The inter-trial interval was 1–2s and varied depending on the response time of the previous trial to make all trials the same length . Combined with the 500 ms fixation period , ISIs were therefore between 1 . 5 and 2 . 5s . Matching the model simulations , each block contained 10 trials for each coherence level with half containing coherent leftward motion and half rightward ( 100 trials total ) . All trials were randomly ordered . Participants were shown cumulative feedback at the end of each block displaying % correct , the mean response time in the most difficult trials , and # correct responses / minute . Before each session , participants completed a training block in which trial-by-trial feedback was given during the first ten trials . We used a within-subject design in which each participant completed three sessions ( depolarizing stimulation; hyperpolarizing stimulation; no stimulation; Figure 6B ) . The order of the stimulation conditions was balanced across participants . The human behavioral task was implemented in Python using PsychoPy v1 . 78 . 01 ( Peirce , 2007 ) . Transcranial Direct Current Stimulation ( tDCS ) was applied over the left dorsolateral prefrontal cortex ( dlPFC; Figure 6C ) using a battery-driven multi-channel direct current stimulator ( NeuroConn , GmbH ) . Specifically , activation within the lateral wall of superior frontal sulcus in posterior left dlPFC relates to perceptual decision making regardless of the response modality ( [x , y , z , =-23 , 29 , 37] , [Heekeren et al . , 2004 , 2006; Ostwald et al . , 2012; Zhang et al . , 2013] ) . We optimized electrode positions for targeting of the left dlPFC using MRI-derived head models of electric field ( EF ) distributions to maximize current flow through this voxel ( HD-Explore and HD-Targets software , v4 . 0 , Soterix Medical , New York , NY , USA ) . We determined that electrode positions on the scalp approximately 5 cm medial and lateral to the nearest point to the dlPFC voxel maximized current flow through the lateral sulcal wall of the target location ( Figure 6C ) , whilst sparing premotor/motor cortices . Inward/anodal ( relative to the cortical sulcal surface ) currents have an opposite effect on neural polarization to outward/cathodal currents ( Rahman et al . , 2013; Bonaiuto and Bestmann , 2015; Bestmann et al . , 2015 ) . Placing the cathode electrode in the medial position and the anodal electrode in the lateral position maximized outward ( cathodal ) current , while the opposite configuration maximized inward current ( anodal ) flow through the target site ( Rahman et al . , 2013; Bikson et al . , 2004; Radman et al . , 2009; Basser and Roth , 2000; Reato et al . , 2010 ) . Individualized electrode positions for each participant were derived using their structural MRI scan . Each participant’s MRI was aligned to the MNI template and the dlPFC coordinate was localized in native space using the inverse co-registration transformation . The coordinate was then used in the neuronavigation software ( Visor ) to mark the nearest point in the superior frontal sulcus , and from this an electrode location on the forehead corresponding to a location on the scalp radial from this target site . Participants completed a total of three sessions spaced approximately one week apart ( Figure 6B ) . In the first session , participants completed three blocks of 100 trials each with three short breaks . Each block lasted 20 min . During each session with stimulation ( depolarizing or hyperpolarizing with the session order balanced across participants ) , participants completed two blocks of trials with sham stimulation , and one with depolarizing or hyperpolarizing stimulation . The first block was always sham and the order of the second and third blocks was balanced across participants and stimulation conditions . This within-subjects design was chosen to maximize statistical power and control for learning effects , as we found in a pilot study that performance on the task improved between sessions as well as between blocks within a session . Behavior from each stimulation block was compared with the sham block directly preceding it in the same session , controlling for both within- and between-session learning . During stimulation blocks , tDCS was applied for 20 min at 2 mA . During sham blocks the stimulation was ramped up to 2mA over 10 s , stimulated for 30 s , and then ramped down to 0 over 10 s . The model predicted that choice biases should diminish with longer ISIs , because of the longer time for neural activity to return to baseline . To test this model prediction , we conducted a control experiment using the same task as that in the main experiment , with the exception that the inter-trial intervals were either 1 or 4 . 5 s long . Combined with the 500 ms fixation duration , this resulted in ISIs of either 1 . 5 or 5 s and these were randomized within blocks . Following a training block in which trial-by-trial feedback was given , participants completed three test blocks without feedback . Each test block contained 20 trials for each combination of ISI and coherence level , with half containing coherent leftward motion and half rightward ( 200 trials total per block ) . All trials were randomly ordered . The analysis was performed on all test blocks . We performed exactly the same analyses on the behavior of the virtual subjects and human participants . Trials in which the participant or virtual subject made no response were excluded , as were trials where the response or decision time was classified as an outlier by the median deviation of the medians applied method ( Rousseeuw and Croux , 1993 ) . Stimulation blocks with human participants were conducted in separate sessions , to avoid carry-over effects of repeated stimulation blocks , and compared with the directly preceding sham block in the same session ( Figure 6B ) . We therefore have separate baselines for each stimulation condition , and consequently separate plots for depolarizing and hyperpolarizing conditions in Figures 6 and 7 . In each analysis , we compared the sham blocks preceding the stimulation blocks in order to verify that they did not significantly differ from each other . The accuracy of the virtual subject and human participant performance was measured as the percentage of trials in which the direction of coherent motion was indicated correctly , and each virtual subject and participant’s accuracy threshold was defined as the motion coherence required to reach 80% accuracy . This was determined by fitting the percentage of correct trials at each coherence level to a Weibull function and taking the inverse of that function at 80% . Wilcoxon tests for comparing two repeated measures were used to compare the accuracy thresholds in each stimulation condition to the no stimulation condition in virtual subjects , and in each stimulation block to the preceding sham block in the same session in human participants . The decision time of the virtual subjects was analyzed by fitting the difference in mean decision times between the stimulation conditions and no stimulation condition at each coherence level to a linear function ( DTstim-DTcontrol= β0+ β1c , where DT is the decision time and c is the coherence ) . The differences between the mean response times of human participants in stimulation blocks and the preceding sham block were also analyzed using linear regression . The two sham conditions ( sham blocks directly preceding depolarizing blocks and those directly preceding hyperpolarizing blocks ) did not differ from each other in accuracy threshold ( W ( 23 ) = 139 , p=0 . 753 ) or response time difference ( B1 = −45 . 011 , p=0 . 213 ) . Choice hysteresis was analyzed in two different ways . The first analysis involved splitting trials into two groups based on the decision made in the previous trial ( Left* , and Right* ) , fitting the percentage of rightward choices in each group to a sigmoid function of the coherence to the left or right , and computing the difference in ‘indecision points’ , or the level of coherence where rightward choices were made 50% of the time , between the two groups ( Padoa-Schioppa , 2013; Rustichini and Padoa-Schioppa , 2015 ) . In the second analysis , we modeled the decision as:R=1/ ( 1+e−X ) X=a0+a1c+a2 ( δn−1 , R−δn−1 , L ) where R is equal to one if right is chosen and 0 otherwise , and c is the coherence level ( negative if to the left , positive if to the right , therefore a1 > 0 ) ( Padoa-Schioppa , 2013; Rustichini and Padoa-Schioppa , 2015 ) . The current trial is n , therefore δn-1 , R is one if the choice in the last trial was rightward , otherwise 0 , and δn-1 , L is one if the choice on the last trial was leftward , otherwise 0 . The term δn-1 , R- δn-1 , L is thus −1 if the previous choice was leftward or one if it was rightward . Choice hysteresis is indicated by a value of a2 greater than zero . We normalized the effect of hysteresis on the choice by the effect of coherence by analyzing the distribution of a2/a1 across virtual subjects for each condition . Wilcoxon tests for comparing two repeated measures were used to compare both the indecision point shift and coefficient ratio in each stimulation condition to the no stimulation condition in virtual subjects , and in each stimulation block to the preceding sham block in human participants . The two sham conditions ( those directly preceding depolarizing blocks and those directly preceding hyperpolarizing blocks ) did not differ from each other in terms of indecision point shift ( W ( 23 ) = 129 , p=0 . 549 ) or logistic regression coefficient ratio ( W ( 23 ) = 123 , p=0 . 441 ) . All data are archived on Dryad ( Bonaiuto et al . , 2016 ) and may be accessed via http://dx . doi . org/10 . 5061/dryad . r1072 . Python code to run analyses and generate figures from the manuscript is available on GitHub: https://github . com/jbonaiuto/perceptual-choice-hysteresis .

When making decisions , people and other animals tend to repeat previous choices even if this is no longer the best course of action . This tendency is especially common when the choice is difficult to make . For example , when people are asked to decide whether groups of dots on a television screen are moving mostly to the left or the right , they often repeat their previous choice when the direction of motion is not clear . Recordings of brain activity in animals suggest that once a choice is made , there is brain activity left over that influences the level of activity at the beginning of the following choice . If this leftover activity is stronger in the brain cells that represent the first choice , it might give this option a head start when another decision is made; this would provide one explanation as to why that same choice is repeated . However , this explanation had not been tested directly . Bonaiuto et al . reasoned that if leftover activity is indeed the cause of choice repetition , directly manipulating this activity in the human brain should alter this tendency in a predictable way . First , computer-based simulations of circuits of brain cells were used to predict what the consequences of such manipulation would be . The model predicted that brain activity left over after a choice is made would indeed cause the choice to be repeated . Moreover , stimulating this virtual circuit did increase or decrease the tendency to repeat choices depending on the type of stimulation used . Bonaiuto et al . went on to confirm that human volunteers who had been asked to complete the “moving dots” task did tend to repeat their choices . Next , the volunteers had a region of their brain , which is known to be important for making choices , stimulated using electrodes placed on their scalp ( a non-invasive method of brain stimulation ) . Exactly as the computer simulations predicted , one form of stimulation made the individual more likely to repeat their previous choice , while another form of stimulation had the opposite effect . These findings show that stimulating the brain via a non-invasive technique can shape the choices that people make in ways that can be predicted by a biologically realistic computer simulation of networks in the brain . The findings also support the idea that leftover activity following a choice might be the biological reason why people tend to go against evidence and repeat previous choices . This new knowledge could be exploited in future studies that try to understand and influence decision making in humans .

2016

Response repetition biases in human perceptual decisions are explained by activity decay in competitive attractor models

Potassium channels are opened by ligands and/or membrane potential . In voltage-gated K+ channels and the prokaryotic KcsA channel , conduction is believed to result from opening of an intracellular constriction that prevents ion entry into the pore . On the other hand , numerous ligand-gated K+ channels lack such gate , suggesting that they may be activated by a change within the selectivity filter , a narrow region at the extracellular side of the pore . Using molecular dynamics simulations and electrophysiology measurements , we show that ligand-induced conformational changes in the KcsA channel removes steric restraints at the selectivity filter , thus resulting in structural fluctuations , reduced K+ affinity , and increased ion permeation . Such activation of the selectivity filter may be a universal gating mechanism within K+ channels . The occlusion of the pore at the level of the intracellular gate appears to be secondary . In voltage-gated potassium channels , the opening of an intracellular gate formed by a ‘bundle-crossing’ of the inner pore-lining helices is hypothesized to allow K ions to flow through the pore , down their electrochemical gradient ( Yellen , 1998 ) . A narrow section of the pore located towards the extracellular side of the membrane , called the selectivity filter because it allows the pore to select for K ions , has been proposed to play double duty as an inactivation gate: following a stimulus , the pore opens at the intracellular bundle-crossing and , in the continued presence of the stimulus , the selectivity filter changes its conformation to prevent further flux of K ions . Thus , in voltage-gated K+ channels , the intracellular bundle-crossing of the pore-lining helices is believed to be the activation gate , while the selectivity filter is believed to be the inactivation gate . This gating mechanism is believed to be shared by a subset of K+ channels such as KcsA and inward rectifier channels ( Phillips and Nichols , 2003; Cuello et al . , 2010a; Cuello et al . , 2010b; Uysal et al . , 2011 ) . However , functional experiments examining state-dependent accessibility of pore blockers ( Contreras and Holmgren , 2006; Wilkens and Aldrich , 2006; Rapedius et al . , 2012; Posson et al . , 2013b; Posson et al . , 2015 ) or thiol-modifying agents ( Flynn and Zagotta , 2001; Zhou et al . , 2011 ) , have shown that in other channels the intracellular pore-lining helices no longer form a bundle crossing that obstructs K+ flux in the closed state , thus strongly implicating the selectivity filter as an activation gate ( Flynn and Zagotta , 2001; Proks et al . , 2003; Bruening-Wright et al . , 2007; Klein et al . , 2007; Contreras et al . , 2008 ) . This suggests that different K+ channels with high sequence and structural homology do not share a universal gating mechanism . Here , we challenge this statement based on our findings with KcsA , a structurally characterized model K+ channel , which was believed to activate by simple opening of the intracellular bundle-crossing gate . We find that this outward movement of the transmembrane helices in KcsA leads to a change in the selectivity filter from closed to conductive during channel activation . We thus propose that the selectivity filter is the main gate in KcsA . Ion permeation through the selectivity filter of K+ channels has been investigated and described based on KcsA X-ray crystallography data ( Morais-Cabral et al . , 2001; Zhou et al . , 2001 ) and molecular mechanics simulations ( Aqvist and Luzhkov , 2000; Bernèche and Roux , 2001 ) . The generally accepted model for permeation involves alternating states of the selectivity filter loaded with two or three ions separated by water molecules , low ion desolvation free energy at the filter entryway , and the occurrence of a knock-on transitional state in which two K ions come in close contact ( Bernèche and Roux , 2001 ) . Structures of KcsA are available with the inner helices in different configurations , making KcsA a good model to investigate ion permeation gating ( Cuello et al . , 2010a; Uysal et al . , 2011 ) . In the ‘closed’ KcsA structure , the inner helices are closed at the bundle crossing gate ( Doyle et al . , 1998; Zhou et al . , 2001 ) , while in the ‘activated’ KcsA crystal structures , the inner helices are wide open at the bundle crossing ( Cuello et al . , 2010b ) . Here , we analyze and compare the mobility of K+ ions within the selectivity filter in both the ‘closed’ and ‘activated’ states of KcsA . We performed free energy calculations in order to describe ion permeation through the selectivity filter of the ‘closed’ KcsA channel , which is thought to be stabilized in a conductive state ( pdb entry 1K4C ) ( Zhou et al . , 2001 ) . The plot in Figure 1A shows a projection of the potential of mean force ( PMF ) as a color-coded energy map that varies as a function of the position of the three K ions along the channel axis . The reduced reaction coordinate Z12 corresponds to the center-of-mass of the two ions on the extra-cellular side , and Z3 to the innermost ion , relative to the center of mass of the selectivity filter . Thus , one can read directly from the plot the relative free energy of the filter configuration corresponding to any position of the three ions . As examples , the molecular structures of the key ion occupancy states corresponding to well and saddle points along the permeation pathway are illustrated . The PMF calculation , which was initiated with K ions in sites S0-S2-S4 separated by water molecules in S1 and S3 , shows deep free energy wells of at least 6 kcal/mol , indicating high K+ binding affinity ( Figure 1A ) . In disagreement with the assumption of a conducting filter , these calculations suggest that ion permeation is rather impeded . Our result did not depend on the initial ion configuration , as a PMF calculation started with K ions in sites S1-S3 and Cav ( cavity ) with water molecules in S2 and S4 , reveals similar free energy wells about 7 kcal/mol deep ( Figure 1—figure supplement 1 ) . Unlike the first PMF , this one shows the formation of a knock-on state achieved via the loss of the water molecule between the 2nd and 3rd ions . Similar occupancy states were observed by other groups who proposed mechanisms where ions in the filter are most often not separated by water molecules during permeation , the higher ion density resulting in greater ionic repulsion and current ( Furini and Domene , 2009; Köpfer et al . , 2014 ) . However , comparison of the PMFs in Figure 1A and Figure 1—figure supplement 1 shows that the loss of a water molecule did not significantly reduce the free energy barriers in our calculations . We sought a different explanation for the slow K+ diffusion . We wondered whether the barriers to permeation we calculated were due to a non-permissive conformation of the KcsA selectivity filter , adopted when the channel’s intracellular activation gate is ‘closed’ . To explore this possibility , we next examined the free energy of ion movement through the selectivity filter of a KcsA channel with an ‘open’ intracellular gate . All KcsA ‘open’ states documented in the literature allow K+ access through the intracellular entryway but display a selectivity filter in a collapsed state , in agreement with functional data showing that KcsA inactivates within seconds after activation by protons ( Chakrapani et al . , 2011 ) . Permeation is prohibited through such a collapsed filter conformation as rotation of the amide planes within the collapsed selectivity filter disrupt K ion binding and lead to new , stabilizing interactions of the amide planes with water molecules that intercalate between the subunits ( Figure 1—figure supplement 2A ) . Removal of these water molecules has been shown to facilitate recovery from inactivation and restore the ‘conductive’ form of the selectivity filter ( Ostmeyer et al . , 2013 ) . Two KcsA structures with an open bundle-crossing and collapsed selectivity filter ( pdb entries 3F7V and 3F5W , with intracellular entryway diameters of 23 Å and 32 Å , respectively ) ( Cuello et al . , 2010b ) were chosen for the preparation of simulations of an open KcsA channel . In line with the study of Ostmeyer et al . , 2013 , we prepared a conductive selectivity filter conformation using a simulation in which the tightly bound water molecules behind the selectivity filter were removed and K ions maintained at the canonical binding sites S1-S3-Cav or S0-S2-S4 using harmonic restraints ( see Materials and methods ) . No direct restraints were applied to the selectivity filter itself in order to minimize the possibility of introducing artifactual structural changes in the channel . This procedure led to a selectivity filter nearly identical in structure with that of the closed state KcsA ( Figure 1—figure supplement 2B ) . The PMF calculation describing ion movement within the selectivity filter of the open state KcsA , based on the 3F5W structure and initiated with ions in sites S0-S2-S4 , reveals small free energy barriers of 2 to 3 kcal/mol , conducive to ion permeation ( Figure 1B ) . A PMF describing a complete permeation event ( the entrance of a K ion on one side and the release of a K ion from the other side ) is shown in Figure 1—figure supplement 3 . Compared to the above simulations using the closed KcsA structure ( Figure 1A ) , these results indicate that a dramatic increase in K ion conductance occurs within the selectivity filter after the channel has switched from a conformation with the intracellular gate closed to one where it is open . Ion permeation in the open state appears to be favored by increased fluctuations and slight enlargement of the selectivity filter . The histogram in Figure 2A , extracted from 20-ns long simulations of the open and closed states , shows that following the opening of the intracellular gate the selectivity filter adopts conformations that are up to 1 Å wider in diameter . This slight widening of the selectivity filter allows for a knock-on transition state in which two K ions accompanied by a water molecule ( as opposed to one K+ and one water ) occupy the S3 and S4 binding sites ( state identified by ( ∗ ) in Figure 1B ) . The knock-on state is not strictly required for permeation and an alternative pathway , referred to as vacancy-diffusion , is also accessible ( Figure 1B , dashed arrow ) . The vacancy-diffusion pathway involves that ions occupying the selectivity filter move first , leaving a vacant site that is filled by an incoming ion ( Schumaker and MacKinnon , 1990; Bernèche and Roux , 2001 ) . Interestingly , in a previous simulation work , permeation via both the knock-on and vacancy-diffusion pathways was observed despite the fact that those calculations were performed using a structure of KcsA in which the intracellular gate is closed ( Bernèche and Roux , 2001 ) . In those simulations , the fluctuations of the selectivity filter required for ion permeation were originating from the intrinsic flexibility of the protein main chain , which was overestimated by CHARMM22 , a previous generation of the CHARMM force field ( see section ‘Force field and ion permeation’ in Materials and methods ) . For comparison , the distance distribution in the selectivity filter of the closed state channel is also plotted for a simulation using the CHARMM22 force field ( Figure 2A , dashed line ) . The plot shows fluctuations similar to those observed in the open channel using the current CHARMM36 force field . This new generation of force field notably contains a correction term for backbone dihedral angles developed on the basis of X-ray crystallographic data and quantum mechanical calculations of the protein backbone conformational energy ( Mackerell et al . , 2004; Best et al . , 2012 ) . The improved force field allows us to better understand how the fluctuations of the selectivity filter are actually controlled by allosteric interactions involving the transmembrane and pore helices . Departing from the permeation mechanism described here where three ions , accompanied by waters , permeate the selectivity filter , Köpfer et al . , 2014 suggested a hard knock-on mechanism in which ion permeation arises from the entrance of a fourth ion , excluding at the same time water molecules from the selectivity filter . The discrepancy between the two proposed permeation mechanisms is in part due to differences in the force field parameters defining the K+-carbonyl interaction , which result in much stronger ion binding affinity to the selectivity filter in the study by Köpfer et al . ( see section ‘Force field and ion permeation’ in Materials and methods ) . However , it is important to consider that , as shown here , the ion binding affinity is essentially determined by the functional state of the selectivity filter . Simulations performed by Jensen et al . ( 2013 ) suggest that the activated state of the channel , which has the lowest ion binding affinity , is not susceptible to the effect of ion parameters . Their simulations of the Kv1 . 2/2 . 1 chimeric channel have shown that the ion force field has little influence on the permeation mechanism since parameters that yield different magnitudes of K+-carbonyl interaction still sustained similar mechanisms involving alternating ions and water molecules ( Jensen et al . , 2013 ) . It remains unclear why permeation in the open KcsA channel appears to be susceptible to the ion force field while permeation in the Kv1 . 2/2 . 1 chimera is not ( Jensen et al . , 2013; Köpfer et al . , 2014 ) . As shown in Figure 2B , the widening of the selectivity filter is associated with movement of the pore helix . Interestingly , only the lower part of the pore helix moves outward upon activation , while the upper part of the pore helix remains mostly unchanged . The lower pore helix segment is in direct contact with the outer transmembrane helix ( TM1 ) at the level of residue L40 ( Figure 3 ) . The side chain of this leucine is within van der Waals distance from the side chains of Ser69 and Val70 at the bottom of the pore helix . The simulations reveal that the distances between the L40 alpha carbon ( Cα ) atoms of opposing subunits increases by about 1 Å as the TM1s move outward upon opening of the intracellular gate ( Figure 2C ) . It is this outward movement of TM1 at the level of L40 that gives the pore helix and the selectivity filter the required room to expand at the base and become conductive . This mechanism predicts that reducing the volume of the residue at position 40 by mutation to a smaller amino acid would similarly provide more space for the pore helix and selectivity filter to expand , and would make it easier for the filter to open and become conductive . Such a mutation would thus favor permeation by reducing the coupling between channel activation and selectivity filter opening . To test this hypothesis , we performed electrophysiological analysis of the L40A KcsA channel mutant . KcsA is a pH-dependent channel that activates upon binding of protons to at least two pH-sensor residues , H25 and E118 ( Figure 4A ) ( Thompson et al . , 2008; Posson et al . , 2013a ) . We performed single channel recording in lipid bilayers to evaluate the effect of the L40A mutation on the pH-dependent gating of KcsA . For these experiments , we employed a routinely-used KcsA variant where inactivation was removed via the previously described E71A mutation ( Cordero-Morales et al . , 2006; Thompson et al . , 2008; Posson et al . , 2013a ) . Figure 4B shows that the L40A mutant is fully activated by protons , similar to the E71A control channel ( open probability increases from 0 to 1 upon pH change ) , and with a similar pH at half activation ( pH0 . 5=5 . 2 , see Table 1 ) . However , the mutant displayed a shallower proton dose response compared with the control channel , indicating a markedly decreased sensitivity to protons ( Figure 4B ) , characterized by a lower Hill coefficient ( Hill coefficient is ~2 for L40A compared to ~4 . 5 for the control channel , see Equation 1 , Table 1 ) . Importantly , the L40A mutant channel opened at a much lower proton concentration than the control channel ( pH 6 , Figure 4B–C ) and displayed intermediate activity over a range of proton concentrations , unlike the very steep H+-dependent activation displayed by the control channel ( Figure 4C ) , suggestive of a channel whose opening is less strongly coupled to proton binding . In order to interpret the effects of the L40A mutant on KcsA gating , we analyzed the pH-dependence with a model we previously used to characterize the effects of mutations of channel residues directly involved in proton-sensing ( Posson et al . , 2013a ) . The gating behavior displayed by L40A was identical to that of H25R , a mutant designed to mimic constitutive protonation at histidine 25 , the KcsA proton sensor where proton binding is most strongly coupled to channel opening ( Figure 4B–C ) ( Posson et al . , 2013a ) . First , the mutation of H25 to arginine removed the proton binding sites at H25 that strongly and cooperatively contributed to the opening of the control channel , yielding a shallower dose-response for the mutant channel ( Figure 4B ) . Second , the arginine substitution mimicked histidine protonation , which dramatically increased the intrinsic gating equilibrium ( Lo ) towards the open state ( Equation 2 , Table 1 ) . Surprisingly , the L40A mutation displays a proton dose-response that is as shallow as H25R ( Figure 4B ) , suggesting that the potency of the H25 sensor is diminished in this mutant by as much as removing the sensor altogether . In addition , the L40A channel begins to open at lower proton doses in much the same way as H25R ( Figure 4C ) . We thus propose that the L40A mutation similarly favors channel opening both by increasing the intrinsic gating ( Lo ) compared to the E71A control and by altering the state-dependence of H25 protonation ( see Materials and methods and Table 1 ) . Both changes are necessary to fit the data and constitute the most parsimonious model to fit the gating behavior of the L40A mutant . We propose that these effects result from partial uncoupling of the bottom of the pore helix from the TM1 helix and from the protonation state of H25 . A residue at a position equivalent to KcsA L40 was previously found to control permeation and open probability in another K channel , a Ca2+-activated K channel ( KCa3 . 1 ) that gates at the selectivity filter ( Garneau et al . , 2014 ) , suggesting a common mechanism . To verify the impact of the L40A mutation on K+ permeation , we performed simulations of this mutant using the closed conformation ( pdb entry 1K4C ) , as well as the 3F7V structure in which the main gate is open less widely than in 3F5W ( Cuello et al . , 2010b ) . For consistency with the electrophysiology experiments described above , these simulations used the E71A inactivation-removed mutant as control . The free energy calculations presented in Figure 5 combine data from independent automated umbrella sampling simulations initiated in different ion occupancy states , as detailed in Figure 5—figure supplement 1 . The PMFs of both Figure 5 and Figure 5—figure supplement 1 lead to the following observations . In the closed conformation ( 1K4C ) of the control channel , the state with 3 ions bound to the filter ( S0-S2-S4 ) is overly stabilized in comparison to states with only 2 ions bound to the filter ( S1-S3-Cav and S2-S4-Cav ) , with a free energy difference of about 11 kcal/mol . Free energy barriers of 6 to 10 kcal/mol are observed between the 2-ion and 3-ion states . The E71A/L40A mutant brings about fluctuations that reduce the ion binding affinity and the relative stability of the 3-ion state . The 2-ion states are stabilized by about 6 kcal/mol , and are thus more accessible . The free energy barriers along the permeation pathway are also reduced to 4–5 kcal/mol . In the partially open conformation ( 3F7V ) , the control channel shows reduced free energy barriers of about 4 kcal/mol . The L40A mutant further reduces the barriers by 1 to 2 kcal/mol , to 2–3 kcal/mol . For comparison , a PMF describing ion permeation in the E71A mutant of the fully open channel ( 3F5W ) shows barriers of 3 kcal/mol ( Figure 5—figure supplement 2 ) . A simulation of the E71A/L40A mutant in the partially activated state ( 3F7V ) was performed with an applied transmembrane voltage of 400 mV and an ion concentration of 800 mM KCl . The time-series analysis in Figure 5—figure supplement 3 shows permeation events in which at most three ions are bound to the selectivity filter with accompanying water molecules , in agreement with the above PMFs and with previously proposed permeation models ( Bernèche and Roux , 2001; Morais-Cabral et al . , 2001 ) . These results support the idea that the L40A mutation facilitates the activation of the channel by reducing the extent of the required conformational change and associated energy to reach the maximal open probability . Our work on KcsA revealed a high ion binding affinity state of the selectivity filter that corresponds to a resting ( closed ) state in which the filter is stabilized , non-conducting but primed for ion conduction , while the intracellular gate is closed . Upon activation , in addition to the large motion of the inner TM2 helix that opens the intracellular gate , the reorientation of the outer TM1 helix releases steric restraints at the selectivity filter allowing ion conduction ( Figure 6A ) . This is in agreement with the observation made by EPR experiments that the conducting state of the KcsA selectivity filter displays larger fluctuations than non-conducting states ( Raghuraman et al . , 2014 ) . Such changes in the structural dynamics of the selectivity filter through activation could also explain the sub-conductance levels with altered selectivity detected by electrophysiology measurements in Shaker ( Chapman et al . , 1997; Zheng and Sigworth , 1997; Chapman and VanDongen , 2005 ) . Our work shows how fluctuations at the selectivity filter have direct impact on its conductance and gating properties , and could potentially be affected by other signaling factors , e . g . temperature and membrane tension ( Rodríguez et al . , 1998; Clarke et al . , 2010; Dong et al . , 2015 ) . The central role played by the outer TM1 helix in KcsA activation potentially explains why residue H25 , found at the intra-cellular end of TM1 , was identified as the strongest pH sensor ( Posson et al . , 2013a ) . The corresponding helix in voltage-dependent channels is S5 , which is directly connected to the voltage-sensor through a linker ( S4-S5 linker ) . This linker is believed to interact directly with the S6 helix to open the channel at the intracellular ‘bundle crossing’ . While S5 was generally thought to play little role in activation gating , it is plausible that the voltage-sensor exerts force on S5 and by doing so regulates the conductance of the selectivity filter through interactions with the pore helix , without necessarily engaging the S6 helix via the S4-S5 linker ( Lees-Miller et al . , 2009; Garg et al . , 2013; Garneau et al . , 2014 ) . In support of this hypothesis , previous results by other groups showed that voltage activation can occur without an S4-S5 linker ( Lörinczi et al . , 2015 ) and that the S4-S5 linker in a subset of voltage-dependent channels does not directly engage the S6 helix ( e . g . EAG channels ( Whicher and MacKinnon , 2016 ) ) . This does not exclude that the inner TM helices may form , in KcsA and other channels ( such as voltage-gated channels ) , a second steric barrier at the intracellular end of the pore . This model also fits the behavior of Ca2+-activated K+ channels , found to ultimately open and close at the selectivity filter , although the gating is accompanied by a large movement of the pore-lining inner helices not unlike that found in voltage-gated channels and KcsA , albeit less obstructive at the intracellular entryway ( Posson et al . , 2015 ) . In conclusion , our work contributes a unified gating mechanism for all K+ channels . We propose that the selectivity filter can be found in at least three different states that determine the functional states of the channel: a restrictive closed state , a dynamic and slightly expanded open state , and a pinched inactivated state ( Figure 6B ) . Transitions between these functional states are controlled by the channel transmembrane domains that provide distant allosteric interactions between the selectivity filter and disparate regulatory sites such as voltage sensor and ligand binding domains ( Panyi and Deutsch , 2006; Clarke et al . , 2010; Cuello et al . , 2010a; Wylie et al . , 2014; Posson et al . , 2015 ) . Thus , we propose that the universal gate in K+ channels is the selectivity filter , whereas the pore-lining helices may undergo large motions that can lead to a secondary closure point at the intracellular bundle-crossing for a subset of K+ channels . The molecular membrane systems were built based on the X-ray structure of the KcsA channel in its high-[K+] closed conformation ( pdb entry 1K4C ) ( Zhou et al . , 2001 ) and the open inactivated structures ( pdb entries 3F7V and 3F5W ) ( Cuello et al . , 2010a ) . Helical turns were added to the transmembrane helices of the two open structures so that these helices have the same length in all constructs , each subunit containing residues 22 to 124 . All residues were assigned their standard protonation state at pH 7 , except Glu71 , which was protonated . The E71A and L40A mutations were applied to the X-ray structures and a new system was built for each independent case . The systems were assembled using the CHARMM-GUI web-service ( Jo et al . , 2008 ) following a protocol developed by Woolf and Roux ( Woolf and Roux , 1994; Wu et al . , 2014 ) . The protein channel , with its symmetry axis aligned along the Z-axis , was embedded in a lipid bilayer of about 130 dipalmitoylphosphocholine ( DPPC ) molecules in the case of WT , or dioleoylphosphocholine ( DOPC ) for E71A constructs . The number of ions in the bulk was adjusted to reproduce an ionic concentration of about 150 mM KCl and to obtain neutral systems , which typically yields a total of 19 cations and 31 anions . The resulting molecular systems each contained about 48 , 000 atoms . All calculations were performed using the CHARMM software version c36 ( Brooks et al . , 2009 ) . The all-atom potential energy function CHARMM36 ( Best et al . , 2012 ) was used for protein and phospholipids , and water molecules were modeled using the TIP3P potential ( Jorgensen et al . , 1983 ) . The Lennard-Jones ( NBFIX ) parameters for the K+-carbonyl oxygen pair interactions were refined so that the solvation free energy of potassium in liquid N-methylacetamide ( NMA ) , a model of the main chain of amino acids , is equal to that of potassium in water ( Roux and Bernèche , 2002 ) . The solvation free energy of K+ in liquid amide is not experimentally known , but by comparison with data available for similar organic solvents one would suggest a value of about −2 kcal/mol in reference to solvation in water ( Marcus et al . , 1988 ) . Given the uncertainty on this value and that it is arguably small , it was decided in the context of the free energy calculations published in Bernèche and Roux ( 2001 ) to set the K+ solvation free energy in liquid amide and water equal . For consistency with this previous work , we maintained the same parameters for the current study . Periodic boundary conditions were applied and long-range electrostatic interactions were calculated using the Particle Mesh Ewald algorithm ( Essmann et al . , 1995 ) . The molecular systems were equilibrated for about 400 ps with decreasing harmonic restraints applied to the protein atoms , the ions and the water molecules localized in the P-loop and the filter . All trajectories were generated with a timestep of 2 fs at constant normal pressure ( 1 Atm ) controlled by an extended Lagrangian algorithm ( Feller et al . , 1995 ) and constant temperature ( 323 . 25 K ) using a Nose-Hoover thermostat ( Evans and Holian , 1985 ) . In the simulations of the open E71A constructs , the conformation of the intracellular gate was maintained by a harmonic root-mean-square-deviation ( RMSD ) restraints ( force constant of 10 kcal/mol•Å with an offset of 0 . 1 Å ) applied to residues 22 to 38 and 100 to 124 . Trajectories of 20 ns were produced for each system ( some in two replicas ) , and were used for the distance analyses ( Figure 2 ) . For the simulation-illustrating ion permeation ( Figure 5—figure supplement 3 ) , the E71A/L40A construct of the partially activated channel ( pdb entry 3F7V ) was used with an increase of the ion concentration to 800 mM KCl and the application of a transmembrane voltage of 400 mV across the simulation box along the Z-axis . The computational method used to prepare the open state channel is designed to mimic ion-mediated recovery from the inactivated selectivity filter state , as shown in Ostmeyer et al . ( 2013 ) . Based on the 3F7V and 3F5W open inactivated structures , the collapsed selectivity filter was remodeled in the putative conducting conformation through equilibration runs of 400 ps in which water molecules were removed from the P-loop , and K+ ions were maintained in state S1-S3-Cav or S0-S2-S4 using harmonic restraints of 10 kcal/mol•Å . To accelerate the transition from the collapsed to the conductive conformation , the NBFIX correction to the ion-carbonyl interactions was not applied for this first phase of the equilibration , as in Ostmeyer et al . ( 2013 ) . The systems were afterward further equilibrated by running simulations using the NBFIX correction to the ion-carbonyl interactions and without any restraint applied to the ions . The molecular systems obtained after 2 ns of simulations were used as initial coordinates for the PMF calculations . The 2D potential of mean force ( PMF ) calculations were performed using the self-learning adaptive umbrella sampling method ( Wojtas-Niziurski et al . , 2013 ) . Starting from a single initial configuration , this self-learning approach automatically constructs simulation windows following the valleys of lower free energy . PMF calculations were initiated from the occupancy states S1-S3-Cav and S0-S2-S4 as specified in the text . The upper free energy limit for the creation of new windows was set to 12 kcal/mol . The reaction coordinates were defined as the distance along the pore axis ( which is aligned with the Z axis ) between an ion , or the center-of-mass of two ions , and the center of mass of the selectivity filter backbone ( residues 75 to 79 ) . Independent simulations of 600 ps were performed every 0 . 5 Å along these reaction coordinates using a biasing harmonic potential with a force constant of 20 kcal/mol•Å2 and were unbiased using the weighted histogram analysis method ( WHAM ) ( Kumar et al . , 1992 ) with a grid spacing of 0 . 1 Å in both dimensions . The first 100 ps of sampling is considered as equilibration and is not included in the final PMF calculation . For each PMF , the reference free energy value is set to zero at the free energy minimum . The statistical error on the PMF calculations is calculated by subdividing the sampling in slices of 100 ps ( Figure 5—figure supplement 4 ) . In a subsequent step , a free energy off-set is attributed to each interval PMF according to a least-square fit to the final PMF considering only the grid points for which the free energy is <6 kcal/mol . Using the PMFs of the last five intervals as an ensemble , the standard deviation is calculated at every grid points ( Figure 1—figure supplement 4 and Figure 5—figure supplement 5 ) . The PMFs presented in Figure 5 were obtained from the combined sampling of two independent automated umbrella-sampling simulations initiated in different ion occupancy states , respectively S1-S3-Cav and S0-S2-S4 . The approach is similar to that routinely applied for combining forward- and backward-free energy perturbation simulations using the Bennett Acceptance Ratio ( BAR ) ( Lu et al . , 2003 ) . At odds with the results presented here , PMF calculations presented in Bernèche and Roux ( 2001 ) described diffusion limited ion permeation in the KcsA selectivity filter despite a closed intracellular gate . These free energy simulations were performed based on the 3 . 2 Å structure of the closed KcsA channel ( pdb entry 1BL8 ) ( Doyle et al . , 1998 ) using CHARMM22 , an earlier generation of the CHARMM force field ( MacKerell et al . , 1998 ) . Since then , correction terms were added to the Φ , Ψ backbone potential energy function in the CHARMM force field to better reproduce crystallographic data and quantum mechanic calculations ( Mackerell et al . , 2004 ) . As documented by MacKerell and co-workers , these corrections , which are now included in the CHARMM36 force field ( Best et al . , 2012 ) , led to a decrease of the backbone RMS fluctuations across all tested proteins ( Mackerell et al . , 2004 ) . Thus , the CHARMM22 force field was biased toward higher backbone fluctuations , which favored ion permeation in simulations of the closed KcsA channel as reflected by the PMF calculations found in Bernèche and Roux ( 2001 ) . The CHARMM36 force field , on the other hand , results in more limited fluctuations of the KcsA selectivity filter that are not sufficient to sustain ion permeation in the closed channel . Repeating the PMF calculation on the 1BL8 structure using the latest force field yielded essentially the same results as those obtained with the 1K4C structure , and thus the differences between those structures are not critical ( data not shown ) . The fluctuations required for ion permeation are recovered when the intracellular gate of the channel opens . The CHARMM22 force field allowed for a description of the ion permeation mechanism in the selectivity filter of K+ channels . The CHARMM36 force field allows us to go further and explain how ion permeation is regulated . Our findings differ from those of Köpfer et al . ( 2014 ) who proposed a hard knock-on mechanism in which the binding of 4 ions to the selectivity filter is required for permeation to take place in the open KcsA channel , versus 3 ions intercalated with water molecules in our simulations . Although Köpfer and colleagues also have used the CHARMM36 force field , they have not , unlike us , applied a NBFIX correction to the interaction between K+ and backbone carbonyl oxygen atoms . This lead to a stronger K+-carbonyl interaction in their CHARMM36 simulations compared to ours , with a difference in K+ solvation-free energy in liquid amide of about 8 kcal/mol . For ion permeation to take place , the ion binding affinity to the selectivity filter needs to be compensated by the ion-ion electrostatic repulsion . Thus , in the systems of Koepfer et al . ( 2014 ) Köpfer et al . , 2014 , the intrinsically higher ion-binding affinity requires a higher number of bound ions to promote ion permeation , impeding at the same time the entrance of water molecules into the selectivity filter . The L40A mutation was made in a construct of the non-inactivating ( Cordero-Morales et al . , 2006 ) KcsA-E71A pQE60 by QuikChange mutagenesis ( Agilent Technologies , Santa Clara , CA ) and verified with sequencing . KcsA protein was expressed and purified as described previously ( Heginbotham et al . , 1999; Thompson et al . , 2008; Posson et al . , 2013a ) . Briefly , KcsA protein expression was induced in BL21 ( DE3 , Invitrogen , Carlsbad , CA ) E . coli cells at 37°C in Luria-Bertani media with 500 μM IPTG . Cells were sonicated ( Thermo Fisher Scientific , Waltham , MA ) in a suspension ( 100 mM KCl , 50 mM Tris , pH 7 . 5 ) and KcsA was extracted with 25 mM n-decyl maltoside ( DM , Anatrace , Maumee , OH ) and then purified from the clarified , soluble fraction in 100 mM KCl , 20 mM Tris , and 5 mM DM , pH 7 . 5 buffer using a Ni2+-affinity column ( Novagen , Merck , Germany ) and gel filtration ( Superdex 200 , GE Healthcare , Chicago , IL ) . KcsA was reconstituted into liposomes using protein to lipid ratios of 0 . 1–2 . 5 μg protein per mg of lipid ( 3:1 POPE: POPG , Avanti Polar Lipids , Alabaster , AL ) . Detergent was removed from protein-lipid mixtures using a gel filtration column ( G50 fine , GE Healthcare ) in buffer ( 400 mM KCl , 5 mM NMG , 20 mM Tris , pH 7 . 5 ) . Liposome aliquots were flash-frozen in liquid nitrogen and stored at −80°C . KcsA L40A/E71A channels were recorded and analyzed as described previously ( Thompson et al . , 2008; Posson et al . , 2013a ) . Briefly , channel liposomes were fused to horizontal planar lipid bilayers ( 3:1 POPE: POPG in decane ) and single channel currents were obtained using 70 mM KCl , 30 mM KOH , 10 mM MOPS , 10 mM succinate , and 10 mM Tris , pH adjusted using HCl to pH 7 for the cis ( extracellular side ) and pH 4–6 for the trans ( cytoplasmic side ) . Single channel recordings of the KcsA L40A/E71A mutant established channel activity ( conductance and open probability ) similar to the previously studied E71A control channel at pH 4 . Rare channels where this was not the case were excluded from the analysis . Subsequent perfusion of the cytoplasmic side to higher pH values was used to determine the pH-dependent gating . Bilayers were voltage-clamped using an Axopatch 200B amplifier ( Molecular Devices , Sunnyvale , CA ) and current recordings were filtered at 2 kHz using a 4-pole Bessel filter , digitized at 25 kHz using a Digidata 1440A ( Molecular Devices ) , and recorded in Clampex 10 . Single channel open probabilities ( Po ) were calculated using Clampfit 10 ( Molecular Devices ) and Po vs . pH data were fit in Origin ( OriginLab , Northampton , MA ) with the Hill equation ( Equation 1 ) : ( 1 ) Po=Pomax1+ ( EC50[H+] ) nH=Pomax1+ ( 10 ( pH−pH1/2 ) ) nH Pomax is the maximal open probability , EC50 is the proton concentration of half-activation ( also represented by pH1/2 ) , [H+] is the proton concentration ( also represented by pH ) , and nH is the Hill coefficient . The Po vs . pH data were also fit to a two proton-binding sites MWC model ( Equation 2 ) ( Posson et al . , 2013a ) : ( 2 ) Po=Lo ( 1+[H+]Ka1open ) 4 ( 1+[H+]Ka2open ) 4Lo ( 1+[H+]Ka1open ) 4 ( 1+[H+]Ka2open ) 4+ ( 1+[H+]Ka1closed ) 4 ( 1+[H+]Ka2closed ) 4=Lo ( 1+10 ( pKa1open−pH ) ) 4 ( 1+10 ( pKa2open−pH ) ) 4Lo ( 1+10 ( pKa1open−pH ) ) 4 ( 1+10 ( pKa2open−pH ) ) 4+ ( 1+10 ( pKa1closed−pH ) ) 4 ( 1+10 ( pKa2closed−pH ) ) 4 Lo is the intrinsic gating equilibrium between open and closed in the absence of protons , [H+] is the proton concentration ( also represented by pH ) , KaopenandKaclosed are the acid dissociation constants for a proton sensor in the open and closed states , respectively ( also represented by pKaopenandpKaclosed ) . Equation 2 contains two proton sensors that are labeled with subscripts 1 and 2 . For the analysis of L40A/E71A , independent adjustment of Lo or pH-sensor pKa values were not sufficient to adequately describe the data . Since the mutant dose response was very similar to that observed in the H25R mutant ( Posson et al . , 2013a ) ( Figure 4 ) , we assumed that L40A altered both Lo and pKaclosed . Using these two free parameters , the best fit resulted in an increased Lo value and a loss of the H25 pH-sensor because the pKaclosed value climbed to a constrained maximum value equal to pKaopen . Subsequently constraining pKaclosed resulted in a fit for Lo , the single free parameter . Model parameters for L40A and relevant pH-sensor mutants ( Posson et al . , 2013a ) are summarized in Table 1 .

Potassium channels are proteins found in almost all living organisms and are vital for many different biological processes . These proteins contain a pore that allows potassium ions to flow through cell membranes , but only when the channel is open . Most channels have a narrowing at the inward side of the pore , which was proposed to form a gate that controls the flow of ions . This gate only opens when the channel activates . Nearer the outward side of the pore is another narrow region called the selectivity filter . This region interacts selectively with potassium ions as they pass through the channel . Not all channels form a tight constriction at the inward end of their pore , yet they still only allow potassium ions to flow through when activated . This suggested that these channels might instead use the selectivity filter as a gate . It would also mean that whilst different potassium channels have similar structures , they do not share a common gating mechanism that controls the flow of ions . Heer et al . studied the bacterial channel called KcsA , which was thought to control the flow of potassium ions via a traditional inward gate . Computer simulations based on this protein’s structure , and experiments with purified KcsA in artificial membranes , showed that the selectivity filter was also involved when KcsA was activated . When the activated channel changed shape , the selectivity filter – which had been constrained by regions forming the pore – could now move and allow potassium ions to flow through the pore . Heer et al . confirmed using a mutant KcsA that the motion of the inward side of the pore upon activation affected the movement of the selectivity filter gate as predicted by the simulation . These findings show that the KcsA channel opens and closes at the selectivity filter . The changes at the inward side of the pore , previously believed to be the gate , firstly enable the selectivity filter to serve as a gate while also forming a secondary gate . Heer et al . propose that most if not all potassium channels may also use this mechanism . These findings illustrate that molecular simulations can be powerful for predicting how changes in the structure of a protein will affect its behavior . This and future studies of other potassium channels will help scientists to better understand the subtle differences between the diverse range of channels found across different organisms .

2017

Mechanism of activation at the selectivity filter of the KcsA K+ channel

A central goal of studying host-pathogen interaction is to understand how host and pathogen manipulate each other to promote their own fitness in a pathosystem . Co-transcriptomic approaches can simultaneously analyze dual transcriptomes during infection and provide a systematic map of the cross-kingdom communication between two species . Here we used the Arabidopsis-B . cinerea pathosystem to test how plant host and fungal pathogen interact at the transcriptomic level . We assessed the impact of genetic diversity in pathogen and host by utilization of a collection of 96 isolates infection on Arabidopsis wild-type and two mutants with jasmonate or salicylic acid compromised immunities . We identified ten B . cinereagene co-expression networks ( GCNs ) that encode known or novel virulence mechanisms . Construction of a dual interaction network by combining four host- and ten pathogen-GCNs revealed potential connections between the fungal and plant GCNs . These co-transcriptome data shed lights on the potential mechanisms underlying host-pathogen interaction . How a host and pathogen manipulate each other within a pathosystem to facilitate their own fitness remains a long-standing question . The difference between the pathogen’s ability to infect and the host’s ability to resist generates the resulting disease symptomology . This interaction forces host-pathogen dynamics to shape the genomes of the two species via adaptive responses to each other ( Dangl and Jones , 2001; Bergelson et al . , 2001; Benton , 2009; Kanzaki et al . , 2012; Karasov et al . , 2014 ) . Plants have evolved a sophisticated set of constitutive and inducible immune responses to cope with constant selective pressures from antagonistic microbes ( Jones and Dangl , 2006 ) . Reciprocally , plant pathogens have also evolved a variety of different invasion and virulence strategies to disarm or circumvent plant defense strategies ( Glazebrook , 2005; Toruño et al . , 2016 ) . This has resulted in complex relations between plant hosts and fungal pathogens for survival and fitness . The plant innate immune system includes several functional layers with overlapping functions to detect and defend against phytopathogens . This multi-layer immune system can be categorized as a signal monitor system to detect invasion , local and systemic signal transduction components to elicit and coordinate responses , and defensive response proteins and metabolites focused on combatting the invading pathogen ( Tsuda and Katagiri , 2010; Corwin and Kliebenstein , 2017 ) . These functional layers , as well as the components within them , are highly interconnected and tightly regulated by the host plant to respond appropriately to various phytopathogens ( Couto and Zipfel , 2016; Tang et al . , 2017 ) . For instance , Arabidopsis utilizes a complex signaling network to regulate the production of indole-derived secondary metabolites , such as camalexin and indole glucosinolates , that contribute to resistance against pathogens ( Kliebenstein et al . , 2005; Clay et al . , 2009; Bednarek et al . , 2009; Frerigmann et al . , 2016; Xu et al . , 2016; Mine et al . , 2018 ) . This layered immune system provides pathogens with numerous targets in the plant immune system that the pathogen can utilize , evade or attack . Most biotrophic pathogens , evolved from commensal microbes , attempt to dismantle the plant immune system by injecting effector proteins into host cells or the inter-cellular space ( Dangl and Jones , 2001; Büttner and He , 2009; Stergiopoulos and de Wit , 2009 ) . For example , the biotrophic bacterial pathogen Pseudomonas syringae can utilize the jasmonic acid ( JA ) signaling pathway through the production of a JA-mimic , coronatine , to enhance its fitness ( Mittal and Davis , 1995; Brooks et al . , 2005; Cui et al . , 2018 ) . Alternatively , necrotrophic pathogens , which often evolved from environmental saprophytic microbes , can utilize toxic secondary metabolites , small secreted proteins , and small RNAs to aggressively attack host defenses while also defending against host-derived toxins ( Choquer et al . , 2007; Arbelet et al . , 2010; Mengiste , 2012; Weiberg et al . , 2013; Kubicek et al . , 2014; Macheleidt et al . , 2016 ) . In addition , pathogens can directly resist downstream defenses as is done by B . cinerea , where it has an ATP-binding cassette ( ABC ) transporter BcatrB that provides resistance by exporting camalexin from the pathogen cell ( Stefanato et al . , 2009 ) . This high level of interactivity between the immune system and pathogen virulence mechanisms generates the final level of disease severity . However , a functional description of this combative cross-kingdom communication between a plant host and necrotrophic pathogen remains elusive . Co-transcriptomic approaches whereby the host and pathogen transcriptomes are simultaneously analyzed provide the ability to systematically map the cross-kingdom communication between plant hosts and their pathogens , both for individual genes and gene co-expression network ( GCN ) levels ( Stuart et al . , 2003; Musungu et al . , 2016; Zhang et al . , 2017; Lanver et al . , 2018; McClure et al . , 2018 ) . Recent advances have enabled the measurement of pathogen in planta transcriptome . For example , in planta measurements of the pathogens’ transcriptome within the biotrophic Arabidopsis-Pseudomonas syringae pathosystem has enabled the investigation of early effects on Arabidopsis host immunity and the consequent effects on bacterial growth ( Nobori et al . , 2018 ) . This enabled the identification of a bacterial iron acquisition pathway that is suppressed by multiple plant immune pathways ( Nobori et al . , 2018 ) . This shows the potential for new hypothesis to be generated by a co-transcriptome approach ( Swierzy et al . , 2017; Westermann et al . , 2017; Lee et al . , 2018 ) . The Arabidopsis-B . cinerea pathosystem is well suited for exploring plant-pathogen interaction to understand host defenses and necrotrophic virulence in ecological and agricultural settings . B . cinerea is a necrotrophic generalist pathogen that attacks a broad range of diverse plant hosts , including dicots , gymnosperms , and even bryophytes ( Williamson et al . , 2007 ) . This necrotrophic pathogen is endemic throughout the world and can cause severe pre- and post-harvest losses in many crops . A high level of standing natural genetic variation within B . cinerea population is hypothesized to facilitate the extreme host range of B . cinerea . This genetic variation affects nearly all known B . cinerea virulence strategies , including penetration and establishment , evading detection , and combatting/coping with plant immune responses ( Atwell et al . , 2015; Walker et al . , 2015; Corwin et al . , 2016b ) . For example , a key virulence mechanism is the secretion of phytotoxic secondary metabolites , including the sesquiterpene botrydial ( BOT ) and the polyketide botcinic acid ( BOA ) that trigger plant chlorosis and host cell collapse ( Deighton et al . , 2001; Colmenares et al . , 2002; Wang et al . , 2009; Rossi et al . , 2011; Ascari et al . , 2013; Porquier et al . , 2016 ) . These metabolites are linked to virulence , but some pathogenic field isolates fail to produce either compounds pointing to additional pathogenic strategies . The combination of a high level of genetic diversity and extensive recombination means that a population of B . cinerea is a mixed collection of virulence strategies that can be used to interrogate by the co-transcriptome . In the present study , the Arabidopsis-B . cinerea pathosystem is used to test how the transcriptomes of the two species interact during infection and assess how natural genetic variation in the pathogen impacts disease development . Isolates were inoculated on Arabidopsis Col-0 wild-type ( WT ) in conjunction with immune-deficient hormone mutants coi1-1 ( jasmonate defense signaling ) and npr1-1 ( salicylic acid defense signaling ) . A collection of 96 isolates of B . cinerea was used for infection , which harbor a wide scope of natural genetic variation within the species ( Atwell et al . , 2015; Corwin et al . , 2016a; Zhang et al . , 2016; Corwin et al . , 2016a; Zhang et al . , 2017; Soltis et al . , 2018; Fordyce et al . , 2018 ) . From individual infected leaves , both Arabidopsis and B . cinerea transcripts at 16 hr post-infection ( HPI ) were simultaneously measured . Arabidopsis transcripts was analyzed previously to identify four host-derived GCNs that are sensitive to natural genetic variation in B . cinerea ( Zhang et al . , 2017 ) . In present analysis , ten fungal pathogen-derived GCNs were identiftied , which encode either known or novel virulence mechanisms within the species . Some of these B . cinerea GCNs responsible for BOT production , exocytosis regulation and copper transport are highly linked with the host’s defense phytohormone pathways . By combining the plant host- and pathogen-GCNs into a single network , a dual-transcriptomic network was constructed to identify potential interactions between the components of plant host innate immune system and fungal pathogen virulence . These connections highlight potential targets for fungal pathogen phytotoxins and prevailing counter-responses from plant host . Collectively , co-transcriptomic analysis shed lights on the potential mechanisms underlying how the host and pathogen combat each other during infection and illustrate the continued need for advancements of in planta analysis of dual-species interaction . To investigate how genetic variation within a pathogen differentially interacts with plant host immunity at the transcriptomic level , we profiled the in planta transcriptomes of 96 B . cinerea isolates infection across three host genotypes , the Arabidopsis accession Col-0 WT and two immune-signaling mutants coi1-1 and npr1-1 that are respectively compromised in JA or salicylic acid ( SA ) driven immunity . This previously described collection of 96 isolates represents a broad geographical distribution and contains considerable natural genetic variation that affects a diversity of virulence strategies within B . cinerea ( Denby et al . , 2004; Rowe and Kliebenstein , 2007; Atwell et al . , 2015; Corwin et al . , 2016b; Zhang et al . , 2016 ) . Four independent biological replicates across two separate experiments per isolate/genotype pair were harvested at 16HPI for transcriptome analysis . A total of 1152 independent RNA samples were generated for library preparation and sequenced on Illumina HiSeq platform ( NCBI accession number SRP149815 ) . These libraries were previously used to study Arabidopsis transcriptional responses to natural genetic variation in B . cinerea ( Zhang et al . , 2017 ) . Mapping the dual-transcriptome reads against the B . cinerea reference genome ( B05 . 10 ) , we identified 9284 predicted gene models with a minimum of either 30 gene counts in one isolate or 300 gene counts across 96 isolates . The total of identified genes corresponds to ~ 79% of the 11 , 701 predicted encoding genes in B05 . 10 reference genome ( Van Kan et al . , 2017 ) . The two different thresholds allowed the identification of pathogen transcripts that express only in a specific isolate . Measuring the abundance of individual pathogen transcripts in relation to the host transcripts can be used as a molecular method to estimate fungal biomass ( Blanco-Ulate et al . , 2014 ) . Given this , we hypothesized that the fraction of total reads that map to B . cinerea might be a biologically relevant indicator of pathogen virulence ( Figure 1—source data 1 ) . Comparing B . cinerea transcript abundance at 16HPI to lesion development at 72HPI revealed a significant partial correlation in the WT Col-0 ( R2 = 0 . 1101 , p-value=0 . 0016 , Figure 1 ) . In contrast to WT , the early transcriptomic activities of most B . cinerea isolates were more vigorous in the two Arabidopsis mutants , resulting in a significant curvilinear relationship between total fraction of B . cinerea reads and final lesion area ( p-value=3 . 914e-07 , p-value=0 . 0001 , respectively , Figure 1 ) . Interestingly , the total reads fraction was better correlated with final lesion area in coi1-1 ( R2 = 0 . 2562 ) than either WT ( R2 = 0 . 1101 ) or npr1-1 ( R2 = 0 . 161 ) . This suggests that early transcriptomic activity from the pathogen can be a partial indicator of pathogen virulence , but also depends on the respective resistance from the plant host . Plant defense phytohormone networks , like SA and JA , help shape the immune responses of a plant host while also shape the virulence gene expression within bacterial pathogens , such as Pseudomonas syringae ( Nobori et al . , 2018 ) . To test how variation in host SA/JA-signaling influences the fungal pathogen transcriptome , we applied a generalized linear model linked with negative-binomial function ( nbGLM ) to each B . cinerea transcript across the experiment . This analysis allowed us to estimate the relative broad-sense heritability ( H2 ) of genetic variation from the pathogen , plant host , or their interaction contributing to each transcript ( Figure 2—source data 1–3 ) . Of the 9284 detectable B . cinerea transcripts , 8603 and 5244 transcripts were significantly influenced by genetic variation in pathogen and host , respectively ( 74% and 45% of predicted B . cinerea gene models , respectively ) ( Figure 2A , Figure 2—source data 3 and 4 ) . While this result shows that the plant phytohormone pathways influence B . cinerea gene expression , the variation in host defense responses ( average H2Host = 0 . 010 ) has far less influence on B . cinerea gene expression than that of the pathogens’ own natural genetic variation ( average H2Isolate = 0 . 152 ) . The host defense hormones also affected B . cinerea gene expression in a genotype-by-genotype dependent manner on 4541 genes ( 39% of B . cinerea predicted gene models , average H2Isolate x Host = 0 . 116 ) ( Figure 2B–2I ) . Illustrating this potential for host x pathogen interactions on pathogen gene expression are the two genes encoding the well-studied polygalacturonase 1 ( Bcpg1 ) and oxaloacetate acetyl hydrolase . The two virulence associated genes showed dramatic expression variation across 96 isolates in different host backgrounds ( Figure 3 , Figure 3—figure supplement 1 , and Figure 2—source data 1 ) . Extending this to 500 genes showing the strongest host x pathogen effect showed that there is a wide range of patterns that differs in the host coi1-1 or npr1-1 background with diverse pathogen strain specific patterns ( Figure 4 ) . One potential complication of this analysis is for sequence variation between the reference B05 . 10 genome and the diverse strains to create artificially low expression estimates . However , very few genes showed consistently low expression within a strain and instead when a gene showed no expression in one host genotype , it was expressed in a different host genotype ( Figure 3 and Supplementary file 1 ) . This conditionality argues against a sequencing error as the sequence has not altered . The genes that did show a loss of expression across all host genotypes within a strain ( i . e . BOT and BOA genes ) were frequently linked to whole gene deletions that abolished their expression ( Soltis et al . , 2019 ) . Thus , while there are likely some sequence variation associated expression errors , they are not a dominant signature in the data . Thus , within the Arabidopsis/B . cinerea pathosystem the pathogens transcriptional responses are influenced by a blend of the pathogens’ natural variation and its interaction with the host , while there is less evidence for the host’s defense responses to unilaterally affect B . cinerea . Future work will hopefully assess how this extends to other host-pathogen systems . This data set also allows us to test for specific B . cinerea transcripts whose early expression is associated with later lesion development . These genes can serve as potential biomarkers of overall pathogen virulence and may elucidate the functional mechanisms driving early virulence in the interaction . To find individual pathogen transcripts link with lesion development , we conducted a genome-wide false discovery rate-corrected Spearman’s rank correlation analysis between 72HPI lesion area and individual B . cinerea transcripts accumulation at 16HPI . We identified 2521 genes ( 22% of B . cinerea predicted gene models ) with significant positive correlations and 114 genes ( 1% of B . cinerea predicted gene models ) with significant negative correlations to lesion area across three Arabidopsis genotypes , respectively ( p-value<0 . 01 , Figure 3—source data 1 ) . The top 20 positively correlated B . cinerea genes contained all seven genes involved in BOT biosynthesis ( Deighton et al . , 2001; Colmenares et al . , 2002; Wang et al . , 2009; Rossi et al . , 2011; Ascari et al . , 2013; Porquier et al . , 2016 ) . In addition to phytotoxins , more than 30 genes of the top 100 lesion-correlated genes encode plant cell wall degrading enzymes , that is glucosyl hydrolases , carbohydrate esterases , cellobiose dehydrogenases and Bcpg1 ( Figure 3 and Figure 3—source data 1 ) ( Gerbi et al . , 1996; Zamocky et al . , 2006; Cantarel et al . , 2009; Van Vu et al . , 2012; Igarashi et al . , 2014; Morgenstern et al . , 2014; Blanco-Ulate et al . , 2014; Tan et al . , 2015; Courtade et al . , 2016; Nelson et al . , 2017; Pérez‐Izquierdo et al . , 2017 ) . Additionally , 10 of the top 100 lesion-correlated genes were annotated as putative peptidase activities , which are critical for fungal virulence ( Movahedi et al . , 1991; Poussereau et al . , 2001a; Poussereau et al . , 2001b; ten Have et al . , 2004; ten Have et al . , 2010 ) . A final classical virulence gene in the top 100 gene list is Bcoah ( Bcin12g01020 ) encoding oxaloacetate acetyl hydrolase , which is a key enzyme in oxalic acid biosynthesis that positively contributes to virulence ( Figure 3—figure supplement 1 and Figure 3—source data 1 ) ( Greenberg et al . , 1994; Williamson et al . , 2007; Walz et al . , 2008; Schumacher et al . , 2012; Schumacher et al . , 2015; Tayal et al . , 2017 ) . In addition , this method identified 37 of the top 100 lesion-correlated genes with no gene ontology ( GO ) terms , which likely represent unknown virulence mechanisms ( Figure 3—source data 1 ) . Thus , this approach readily creates new hypothesis about known and novel pathogen virulence functions . To develop a systemic view of fungal pathogen in planta gene expression , we used a co-expression approach to identify B . cinerea networks that associated with growth and virulence in planta . Using solely B . cinerea transcriptome at 16HPI from Arabidopsis Col-0 WT infected leaves , we calculated Spearman’s rank correlations of gene counts across all B . cinerea isolates , filtered gene pairs with correlation greater than 0 . 8 . We then used the filtered gene pairs as input to construct GCNs . We identified ten distinct GCNs containing more than five B . cinerea genes ( Figure 5 , Supplementary file 1 , Figure 5—figure supplement 1 and Figure 5—source data 1 ) . The largest GCN with 242 genes contains members responsible for phospholipid synthesis , eiosome function , and membrane-associated stress signaling pathways ( Figure 5-Vesicle/virulence ) . The biological function of this GCN suggests its role in fungal membrane- and vesicle-localized processes , which are normally involved with general hyphae growth , fungal cell wall deposition , and exudation of fungal toxins to the intercellular space ( Figure 5—source data 1 ) . The second largest network contains 128 genes that were entirely associated with translation and protein synthesis ( Figure 5-TSL/growth and Figure 5—source data 1 ) . Of the smaller GCNs identified ( 5–20 genes ) , five networks were identified with genes distributed across B . cinerea 16 chromosomes , suggesting that these GCNs arise from coordinated trans-regulation ( Figure 5-Trans-networks and Figure 5—figure supplement 1D , F and H–J ) . These networks are associated with diverse array of virulence functions , including the regulation of exocytosis , copper transport , the production of peptidases and isoprenoid precursors ( IPP ) , and polyketide secretion . In contrast to the whole-genome distributed GCNs , three of the smaller GCNs were predominantly comprised of genes tandemly clustered within a single chromosome with no or a few genes on other chromosomes ( Figure 5-BOA , -Cyclic Peptide , -BOT , Figure 5—figure supplement 1C , E and G ) . A functional analysis showed that all of the genes within these networks encoded known or putative biosynthetic enzymes for specialized metabolic pathways . For example , seven genes responsible for BOT biosynthesis cluster on chromosome 12 and form a small GCN with a Zn ( II ) 2Cys6 transcription factor that is specific to the pathway ( Figure 6A and B , Figure 5—figure supplement 1G and Figure 5—source data 1 ) ( Siewers et al . , 2005; Pinedo et al . , 2008; Urlacher and Girhard , 2012; Moraga et al . , 2016 ) . Similarly , all 13 genes involved in BOA biosynthesis cluster in Chromosome one and form a highly connected GCN ( Figure 5-BOA , Figure 6E , Figure 5—figure supplement 1C and Figure 5—source data 1 ) ( Dalmais et al . , 2011; Porquier et al . , 2019 ) . In addition to previously characterized secondary metabolic pathways , we identified an uncharacterized set of ten genes that cluster on Chromosome 1 ( Figure 5-Cyclic Peptide , Figure 6F , Figure 5—figure supplement 1E and Figure 5—source data 1 ) . These genes share considerable homology with enzymes related to cyclic peptide biosynthesis and may represent a novel secondary metabolic pathway in B . cinerea ( Figure 5—source data 1 ) . The expression of these pathways in planta was extremely variable among the isolates and included some apparent natural knockouts in the expression of the entire biosynthetic pathway ( Figure 6G and Figure 2—source data 1 ) . Isolate 94 . 4 was the sole genotype lacking the entire BOT pathway , while 19 isolates and 24 isolates did not transcribe respectively the BOA and the putative cyclic peptide pathways ( Figure 6E–6G and Figure 2—source data 1 ) . We decomposed the expression of these pathways into expression vectors , referred to as eigengenes , using a principle component analysis and used a linear mixed model to test for a relationship between early expression of secondary metabolic pathways and later lesion area . This showed a significant relationship between the expression of BOT and BOA pathways and lesion area measured at 72HPI ( Supplementary file 2 ) . In contrast , the putative cyclic peptide pathway was only associated with lesion development in a BOT-dependent manner , suggesting that it may have a synergism to BOT ( Supplementary file 2 ) . Thus , in planta analysis of the fungal transcriptome can identify known and novel potential virulence mechanisms and associate them with the resulting virulence . The B . cinerea GCNs measured within Arabidopsis WT provide a reference to investigate how phytohormone-signaling in host innate immunity may shape the pathogen’s transcriptional responses during infection . Comparing the B . cinerea GCN membership and structure across the three Arabidopsis genotypes ( WT , coi1-1 , and npr1-1 ) showed that the core membership within networks was largely maintained but the specific linkages within and between GCNs were often variable ( Figure 7 , Supplementary file 1 , Figure 7—figure supplements 1 , 2 and 3 , and Figure 5—source data 1 ) . For example , the two largest B . cinerea GCNs in WT developed multiple co-expression connections during infection in the JA-compromised coi1-1 host ( Figure 7 and Supplementary file 1 ) . In contrast , some GCNs have a highly robust structure across three host genotypes , including three GCNs associated with BOT , BOA and cyclic peptide production , and GCNs associated with exocytosis regulation , copper transport , and peptidase activity ( Supplementary file 1 , Figure 7—figure supplements 1 , 2 and 3 , and Figure 5—source data 1 ) . In addition , we also identified additional small GCNs that demonstrated host specificity in coi1-1 ( Figure 7—figure supplement 2 and Figure 5—source data 1 ) . In particular , there were four small GCNs that are associated with plant cell wall degradation , siderophores , glycolysis , ROS , and S-adenosylmethionine biosynthesis ( Figure 7—figure supplement 2 ) . Thus , the coordinated transcriptional responses of B . cinerea GCNs are at least partially dependent on variation in the host immune response . Host immunities showed different impacts on expression profiles of genes condensed in individual B . cinerea GCNs ( Figure 7—figure supplement 4 ) . Compared with WT , expression profiles of genes within the largest membrane/vesicle virulence GCN were elevated in the SA- and JA-compromised Arabidopsis mutants on average ( Figure 7—figure supplement 4A ) . Fungal genes associated with copper transport and polyketide production were upregulated under SA-compromised host immunity ( Figure 7—figure supplement 4F and J ) . Whereas , members of GCNs responsible for plant cell wall degradation and siderophore biosynthesis were upregulated under JA-compromised host immunity ( Figure 7—figure supplement 4K and L ) . Finally , GCNs associated with BOT and exocytosis regulation showed robust gene expression profiles across all three Arabidopsis genotypes ( Figure 7—figure supplement 4D and G ) . The above observation indicates host immunity influences the B . cinerea transcriptional response of B . cinerea and suggests that B . cinerea isolates have varied abilities to tailor virulence strategy in response to host immunity . To test the interaction between individual genes from two organisms , we generated Arabidopsis-B . cinerea GCNs using co-transcriptome data under each host genotype . We calculated Spearman’s rank correlation coefficients among 23 , 898 Arabidopsis transcripts and 9 , 284 B . cinerea transcripts . This approach identified three cross-kingdom GCNs ( CKGCNs ) under Arabidopsis WT and JA- or SA-compromised two mutants ( Figure 8 , Supplementary file 5 , and Figure 8—source data 1 ) . Under Arabidopsis WT , a total of 54 hub genes were identified , half from B . cinerea and half from Arabidopsis . Furthermore , CKGCNs contain a majority of genes in the BOT GCN and a small proportion of genes in the vesicle/virulence GCN ( Figure 8—figure supplement 1C ) . For plants , CKGCNs contain a majority of genes from Arabidopsis Defense/camalexin GCN ( Figure 8—figure supplement 1B ) . These CKGCNs also contain genes associated with extensive host defense responses , that is genes encoding membrane-localized leucine-rich repeat receptor kinases ( LRR-RKs ) , stress signal sensing and transduction , tryptophan-derived phytoalexin production , regulation of cell death , cell wall integrity , nutrition transporters , etc . ( Figure 8—source data 1 ) . The topological structure and gene content of the CKGCNs shifted across the three Arabidopsis genotypes ( Figure 8 ) . These changes illustrate how the host genotype can influence the intercommunication in the host-pathogen interaction . To begin assessing how two species influence each other’s gene expression during infection , we constructed a co-transcriptome network using both the host- and pathogen-derived GCNs ( Figure 9 and Figure 9—figure supplement 1 ) . We converted the ten B . cinerea GCNs and the four Arabidopsis GCNs into eigengene vectors that capture the variation of the general expression of all genes within a GCN into a single value ( Zhang et al . , 2017 ) . The Arabidopsis GCNs were defined in response to this same transcriptome but by using solely the host transcripts . Of these four Arabidopsis GCNs , one is largely comprised of genes in Defense/camalexin signaling , two are linked to different aspects of photosynthesis and the fourth is largely comprised of host genes in cell division . We calculated Spearman’s rank coefficients among each GCN eigengene pairs without regard for the species . In this dual transcriptome network , the Arabidopsis/B . cinerea GCN eigengenes are displayed as nodes and positive/negative correlations between the GCNs as edges ( Figure 9 and Figure 9—figure supplement 1 ) . Of the host-derived GCNs , the Arabidopsis Defense/camalexin and Photosystem I ( PSI ) GCNs have a higher degree of centrality than do the Cell Division or Plastid GCNs across all three host genotypes , suggesting that they have the most interactions with B . cinerea GCNs . In contrast , the fungal GCNs’ centrality was more dependent on the host genotype . In WT Col-0 , the highest degrees were associated with the exocytosis regulation , BOT , and IPP , whereas they were more peripheral or even not present in the co-transcriptome network in the npr1-1 or coi1-1 host genotypes . Interestingly , in the WT Col-0 host fungal GCNs ( Copper transport , Exocytosis regulation , BOT and IPP biosynthesis ) that were positively correlated with the host Defense/camalexin GCN showed negative correlations with PSI eigengene . However , the host genotype can change these GCN relationships . In the npr1-1 host , the host Defense/camalexin and PSI GCNs shift to a positive correlation . This may reflect a shift in how the B . cinerea BOT GCN has a positive correlation with the Defense/camalexin GCN in the Col-0 host but a negative correlation in the npr1-1 host genotype . This suggests that there are dynamics in the host-pathogen co-transcriptome that can be interrogated to potentially identify causational relationships . To test if these connections were dependent upon the host immunity , we used the eigengene values derived from fungal GCNs to conduct mixed linear modelling of how they were linked to variation in the host genotype and/or host GCNs ( Supplementary file 3 and 4 ) . Some B . cinerea GCNs ( Vesicle/virulence and TSL/growth , etc . ) were more affected by variation in the host genotypes while others had less host dependency on their expression ( BOT , Copper transport , etc . ) . Collectively , pathogen virulence and host immunity GCNs showed complex connections within dual interaction network identified from co-transcriptome data , suggesting functional relationships between host defense and pathogen virulence mechanisms for future experimentation . One potential complicating factor that may influence the co-transcriptome is variation in germination of the spores between B . cinerea strains . The lack of universal genomic patterns for host x pathogen interactions in the co-transcriptome argues that germination is not causing global effects on the co-transcriptome ( Figures 3 and 4 ) . To begin examining how variation in B . cinerea spore germination may influence the co-transcriptome and our identified link to virulence , we investigated the germination of 19 isolates . This showed that there was some variation in germination with all but a few isolates germinating within the 6–7 hr’ time frame at room temperature ( Figure 9—figure supplement 2 ) . To extend this to an in planta analysis , we utilized an existing microarray study on B . cinerea germination to develop an eigengene that estimates the relative germination between the strains using the in planta transcriptomic data ( Leroch et al . , 2013 ) . We then used this in planta estimation of germination to test if our previously identified co-transcriptome to virulence links were altered by controlling for germination . Using linear models , we ran the same test whereby the major B . cinerea GCNs were tested for a link to virulence although this time , we included the germination eigengene as a co-variate . This analysis showed that the in planta estimation of germination significantly associated with virulence . Critically , even with germination taken into account , all the B . cinerea networks remained significantly associated to lesion area . Some GCNs link , that is BOT and BOA , were largely unaffected by the germination estimates ( Supplementary file 6 ) , showing that some aspects of virulence are independent of spore germination . In contrast other GCNs like the vesicle linked GCN had their link to virulence decreased but not abolished by including the germination co-variate . Thus , while spore germination plays a role in our measurement of the plant-pathogen interaction , it is only one of multiple factors influencing the co-transcriptome and is not imparting a dominant global influence on the observed patterns . Necrotrophic pathogen B . cinerea has evolved an arsenal of virulence strategies to establish colonization and enhance infection within the plant host , including production of secondary metabolites . The co-transcriptome approach shows that the expression of fungal specialized pathways early in infection correlates with later lesion development ( Supplementary file 3 ) . Three secondary metabolite GCNs are clustered within the fungal genome and two of them identified with pathway-specific transcription factors ( Figure 6 , Figure 5 , Figure 5—figure supplement 1 , and Figure 5—source data 1 ) . Further , the expression of these pathways displayed a large range of phenotypic variation across the isolates ( Figure 6G and Figure 2—source data 1 ) . However , the topology and memberships of GCNs for the three pathways are largely insensitive to variation in host immunity . Robustness to host immunity suggests that these GCNs are somehow insulated from the host’s immune response , possibly to protect toxin production from a host counter-attack . The co-transcriptome approach showed the ability to identify known and novel secondary metabolic pathways that mediate plant host and fungal pathogen interaction . Importantly , the dual interaction networks provide hypothesis for pathogen-GCNs responsible for fungal secondary metabolites production link to specific plant host-GCNs ( Figure 9 and Figure 9—figure supplement 1 ) . Specifically , the co-transcriptome approach revealed that B . cinerea GCNs responsible for secondary metabolite production are associated with both plant immune responses and primary plant metabolism ( Figure 9 , Figure 9—figure supplement 1 , Supplementary file 3 and 4 ) . For example , in the WT Col-0 host genotype the BOT GCN shows a strong positive correlation with the Arabidopsis Defense/camalexin GCN , suggesting that BOT production may directly induce the host’s defense system . Concurrently , the BOT GCN is negatively linked to the plant’s PSI GCN , suggesting that BOT may repress the plant’s photosynthetic potential . Critically , this relationship changes in the npr1-1 host genotype with the BOT GCN now having a negative correlation to the Arabidopsis Defense/camalexin GCN . Further work is needed to test if these host/pathogen GCN interactions are causal and how the SA pathway in the host may influence these interactions . Collectively , these results strongly implicate the ability of secondary metabolites biosynthesis to mediate the interactions between pathogen virulence and plant host immunity at the transcriptomic level . The co-transcriptome approach showed the potential to enable us to form new hypotheses about how this linkage may occur . In addition to secondary metabolite biosynthesis , the co-transcriptome identified a number of key virulence mechanisms that could be mapped to the two species interaction . One key GCN is enriched for genes involved in exocytosis associated regulation ( Figure 5-Exocytosis regulation and Figure 5—source data 1 ) . The exocytosis complex is responsible for delivery of secondary metabolites and proteins to the extra-cellular space and plasma membrane in fungi ( Colombo et al . , 2014; Rodrigues et al . , 2015 ) . Additionally , we found many B . cinerea genes associated with secretory vesicles within the membrane/vesicle virulence GCN that likely serve a similar function during infection ( Figure 5-Vesicle/virulence and Figure 5—source data 1 ) . These GCNs also provide support for the role of exocytosis-based spatial segregation of different materials during fungal hyphae growth in planta ( Samuel et al . , 2015 ) . The dual interaction network suggests that the exocytosis regulation and membrane/vesicle virulence GCNs are differentially linked to the Arabidopsis Defense/camalexin GCN , indicating varied connections between fungal secretory pathways and plant immune responses ( Figure 9 and Supplementary file 3 and 4 ) . Another conserved GCN in the B . cinerea species is associated with copper uptake and transport ( Figure 5-Copper transport , Figure 7—figure supplements 1 , 2 and 3 , and Figure 5—source data 1 ) . Although copper is essential for B . cinerea penetration and redox status regulation within plant tissues , further work is required to decipher the precise molecular mechanism involved in acquisition and detoxification of copper . Thus , the co-transcriptome approach can identify both known and unknown mechanisms and links within the host-pathogen interaction . It is largely unknown how plant host immunity contributes to the transcriptomic behavior of the fungus during infection . Even less is known about the role of genetic variation in the pathogen in responding to , or coping with , the inputs coming from the host immune system . In the current study , we found that the host immune system’s effect on pathogen transcripts and GCNs was largely via an interaction with the pathogen genotypes ( Figure 2 , Figure 7 , Figure 7—figure supplement 4 , and Figure 2—source data 4 ) . For example , fungal GCNs associated with membrane/vesicle virulence and fungal growth shifted drastically between the WT and coi1-1 or npr1-1 Arabidopsis genotypes ( Figure 7 ) . In addition , some GCNs only appeared in specific backgrounds . For example , those linked to siderophores and a polyketide production were only identified during infection of the JA-compromised Arabidopsis mutant ( Figure 7—figure supplement 4J and L ) . However , other fungal GCNs , like those involved in secondary metabolism , were largely insensitive to variation in the host immunity ( Figure 7—figure supplements 1 , 2 and 3 , Supplementary file 1 , and Figure 5—source data 1 ) . Critically , the gene membership of these GCNs is largely stable across the collection of pathogen isolates , even while their expression level across the B . cinerea isolates is highly polymorphic ( Figure 5-source data 1and Figure 7—figure supplement 4 ) . This suggests that natural variation in the host immunity and pathogen shapes how the co-transcriptome responds to host’s immune system . Further , the natural variation in the pathogen may be focused around these functional GCNs . Plant disease development is an abstract phenomenon that is the result of a wide set of spatiotemporal biological processes encoded by two interplaying species under a specific environment . In current study , we used late stage lesion area as a quantitative indicator of B . cinerea virulence . We have previously shown that early Arabidopsis transcriptomic response could be linked to later lesion development ( Zhang et al . , 2017 ) . Here , our findings suggest that the late-stage disease development of a B . cinerea infection is determined during the first few hours of infection by the interaction of plant immune and fungal virulence responses . It was possible to create a link between early transcripts’ accumulation and late disease development using solely the B . cinerea transcriptome ( Figure 1 and Figure 3—source data 1 ) . This could be done using either individual pathogen genes , GCNs , or more simply the total fraction of transcripts from the pathogen . As the transcriptomic data were from plant leaf tissue only 16HPI , there is not a significant amount of pathogen biomass and this is more likely an indicator of transcriptional activity in the pathogen during infection . It is possible to develop these methods as possible biomarkers for likely fungal pathogen caused disease progression . The co-transcriptome analysis of a B . cinerea population infection on Arabidopsis identified a number of B . cinerea GCNs that contained a variety of virulence-associated gene modules with different biological functions . The characterization of these GCNs simultaneously identified mechanisms known to enhance B . cinerea virulence and implicated several novel mechanisms not previously described in the Arabidopsis-B . cinerea pathosystem . In addition , the plant-fungus co-transcriptome network revealed the potential interaction between fungal pathogen- and plant host-GCNs . Construction of GCNs within single species , CKGCNs and dual networks shed lights on the biological mechanisms driving quantitative pathogen virulence in B . cinerea and their potential targets in the plant innate immune system . A collection of 96 B . cinerea isolates were selected in this study based on their phenotypic and genotypic diversity ( Denby et al . , 2004; Rowe and Kliebenstein , 2007; Corwin et al . , 2016a; Zhang et al . , 2016; Zhang et al . , 2017 ) . This B . cinerea collection was sampled from a large variety of different host origins and contained a set of international isolates obtained from labs across the world , including the well-studied B05 . 10 isolate . A majority of isolates are natural isolates that isolated from California and can infect a wide range of crops . Isolates are maintained in −80°C freezer stocks as spores in 20% glycerol and were grown on fresh potato dextrose agar ( PDA ) 10 days prior to infection . The Arabidopsis accession Columbia-0 ( Col-0 ) was the wildtype background of all Arabidopsis mutants used in this study . The three Arabidopsis genotypes used in this study included the WT and two well-characterized immunodeficient mutants , coi1-1 and npr1-1 , that abolish the major JA- or SA-defense perception pathways , respectively ( Cao et al . , 1997; Xie et al . , 1998; Xu , 2002; Pieterse and Van Loon , 2004 ) . All plants were grown as described previously ( Zhang et al . , 2017 ) . Two independent randomized complete block-designed experiments were conducted and a total of 90 plants per genotype were grown in 30 flats for each experiment . Approximately 5 to 6 fully developed leaves were harvested from the five-week old plants and placed on 1% phytoagar in large plastic flats prior to B . cinerea infection . We infected all 96 isolates onto each of the three Arabidopsis genotypes in a random design with 6-fold replication across the two independent experiments . A total of twelve infected leaves per isolate/genotype pair were generated . For inoculation , all B . cinerea isolates were cultured and inoculated on three Arabidopsis genotypes as described previously ( Denby et al . , 2004; Corwin et al . , 2016b; Zhang et al . , 2017 ) . Briefly , frozen glycerol stocks of isolate spores were first used for inoculation on a few slices of canned peaches in petri plates . Spores were collected from one-week-old sporulating peach slices . The spore solution was filterred and the spore pellet was re-suspended in sterilized 0 . 5x organic grape juice ( Santa Cruz Organics , Pescadero , CA ) . Spore concentrations were determined using a hemacytometer and suspensions were diluted to 10spores/μL . Detached leaf assays were used for a high-throughput analysis of B . cinerea infection , which has been shown to be consistent with whole plant assay ( Govrin and Levine , 2000; Mengiste et al . , 2003; Denby et al . , 2004; Sharma et al . , 2005; Windram et al . , 2012 ) . Five-week old leaves were inoculated with 4 μL of the spore solution . The infected leaf tissues were incubated on 1% phytoagar flats with a humidity dome at room temperature . The inoculation was conducted in a randomized complete block design across the six planting blocks . All inoculations were conducted within one hour of dawn and the light period of the leaves was maintained . Two blocks were harvest at 16HPI for RNA-Seq analysis . The remaining four blocks were incubated at room temperature until 72HPI when they were digitally imaged for lesion size and harvested for chemical analysis as described previously ( Zhang et al . , 2017 ) . Two B . cinerea infected leaf tissues of the six blocks were sampled at 16HPI for transcriptome analysis , which resulted in a total of 1 , 052 mRNA libraries for Illumina HiSeq sequencing . RNA-Seq libraries were prepared according to a previous method ( Kumar et al . , 2012 ) with minor modifications ( Zhang et al . , 2017 ) . Briefly , infected leaves were immediately frozen in liquid nitrogen and stored at −80°C until processing . RNA extraction was conducted by re-freezing samples in liquid nitrogen and homogenizing by rapid agitation in a bead beater followed by direct mRNA isolation using the Dynabeads oligo-dT kit . First and second strand cDNA was produced from the mRNA using an Invitrogen Superscript III kit . The resulting cDNA was fragmented , end-repaired , A-tailed and barcoded as previously described . Adapter-ligated fragments were enriched by PCR and size-selected for a mean of 300 base pair ( bp ) prior to sequencing . Barcoded libraries were pooled in batches of 96 and submitted for a single-end , 50 bp sequencing on a single lane per pool using the Illumina HiSeq 2500 platform at the UC Davis Genome Center ( DNA Technologies Core , Davis , CA ) . All statistical analysis were conducted within R ( R Development Core Team , 2014 ) . Fastq files from individual HiSeq lanes were separated by adapter index into individual RNA-Seq library samples . The quality of individual libraries was estimated for overall read quality and over-represented sequences using FastQC software ( Version 0 . 11 . 3 , www . bioinformatics . babraham . ac . uk/projects/ ) . We conducted downstream bioinformatic analysis , like reads mapping , normalization and nbGLM model analysis , using a custom script from the Octopus R package ( https://github . com/WeiZhang317/octopus; Zhang , 2018; copy archived at https://github . com/elifesciences-publications/octopus ) . The mapping of processed reads against Arabidopsis and B . cinerea reference genomes was conducted by Bowtie 1 ( V . 1 . 1 . 2 , http://sourceforge . net/projects/bowtie-bio/files/bowtie/1 . 1 . 2/ ) using minimum phred33 quality scores ( Langmead et al . , 2009 ) . The first 10 bp of reads was trimmed to remove low quality bases using the fastx toolkit ( http://hannonlab . cshl . edu/fastx_toolkit/commandline . html ) . Total reads for each library were firstly mapped against the Arabidopsis TAIR10 . 25 cDNA reference genome . The remaining un-mapped reads were then aligned against B . cinerea B05 . 10 isolate cDNA reference genome ( Lamesch et al . , 2010; Lamesch et al . , 2012; Krishnakumar et al . , 2015; Van Kan et al . , 2017 ) and the gene counts for both species were pulled from the resulting SAM files ( Li et al . , 2009 ) . For pathogen gene expression analysis , we first filtered genes with either more than 30 gene counts in one isolate or 300 gene counts across 96 isolates . We normalized B . cinerea gene counts data set using the trimmed mean of M-values method ( TMM ) from the EdgeR package ( V3 . 12 ) ( Robinson and Smyth , 2008; Bullard et al . , 2010; Robinson and Oshlack , 2010 ) . We then ran the following generalized linear model ( GLM ) with a negative binomial link function from the MASS package for all transcripts using the following equation ( Venables and Ripley , 2002 ) :Yegai=Ee+Ee ( Gfg ) +Ee ( Gfg ( Afa ) ) +Ii+Hh+Hh∗Iiwhere the main categorical effects E , I , and H are denoted as experiment , isolate genotype , and plant host genotype , respectively . Nested effects of the growing flat ( Gf ) within the experimental replicates and agar flat ( Af ) nested within growing flat are also accounted for within the model . Model corrected means and standard errors for each transcript were determined for each isolate/plant genotype pair using the lsmeans package ( Lenth , 2016 ) . Raw P-values for F- and Chi Square-test were determined using Type II sums of squares using car package ( Fox and Weisberg , 2011 ) . P-values were corrected for multiple testing using a false discovery rate correction ( Yoav and Daniel , 2001 ) . Broad-sense heritability ( H2 ) of individual transcripts was estimated as the proportion of variance attributed to B . cinerea genotype , Arabidopsis genotype , or their interaction effects . GO analysis was conducted for several B . cinerea gene sets that were identified with high heritability , correlated with lesion size , and condensed in network analysis . We first converted sequences of these B . cinerea genes into fasta files using Biostrings and seqRFLP packages in R ( Qiong and Jinlong , 2012; Pages et al . , 2017 ) . The functional annotation of genes was obtained by blasting the sequences against the NCBI database using Blast2GO to obtain putative GO annotations ( Conesa et al . , 2005; Götz et al . , 2008 ) . The GO terms were compared to the official GO annotation from the B . cinerea database ( http://fungi . ensembl . org/Botrytis_cinerea/Info/Index ) and those obtained by Blast2GO analysis . The official gene annotations for host genes was retrieved from TAIR10 . 25 ( https://apps . araport . org/thalemine/bag . do ? subtab=upload ) . To obtain a representative subset of B . cinerea genes co-expressed under in planta conditions , we generated gene co-expression networks ( GCNs ) among genes in the B . cinerea transcriptome . GCNs were generated using the model-corrected means of 9 , 284 B . cinerea transcripts from individual isolate infection across three Arabidopsis genotypes . Only genes with average or medium expression greater than zero across all samples were considered . This preselection process kept 6372 genes and those with negative expression values were adjusted to set expression at zero before network construction . Spearman’s rank correlation coefficients for each gene pair was calculated using the cor function in R . Three gene-for-gene correlation similarity matrixes were generated independently for each of the three Arabidopsis genotypes . Considering the cutoff for gene-pair correlation usually generates biases of GCN structure and the candidate gene hit , we utilized several cutoff threshold values at 0 . 75 , 0 . 8 , 0 . 85 , and 0 . 9 to filter the gene set . Comparing the structure and content of GCNs among those GCN sets using filtered gene set as input , we selected the correlation threshold at 0 . 8 . A total of 600 , 700 and 494 B . cinerea candidate genes passed the criterion under Arabidopsis WT , mutants coi1-1 and npr1-1 , respectively . To obtain a representative subset of B . cinerea gene candidates across three host genotypes , we selected gene candidates that presented across the above three gene subsets . This process generated a gene set with 323 B . cinerea candidate genes that were common to each of the plant genotype backgrounds and had at least 0 . 8 significant correlations . Using this gene set as kernel , we extended gene candidate sets under each Arabidopsis genotype . The expanded B . cinerea gene candidate set under individual Arabidopsis genotypes was further used as input for gene co-expression network construction . GCNs were visualized using Cytoscape V3 . 2 . 1 ( Java version:1 . 8 . 0_60 ) ( Shannon et al . , 2003 ) . The nodes and edges within each network represent the B . cinerea genes and the Spearman’s rank correlations between each gene pair . The importance of a given node within each network was determined by common network analysis indices , such as connectivity ( degree ) and betweenness . Nodes with higher connectivity and betweenness were considered as hub and bottleneck genes , respectively , and the biological functions of each network were determined by the GO terms of hub and bottle neck genes using Blast2GO . We used model-corrected means of transcripts from three Arabidopsis host genotypes and 96 B . cinerea isolates to construct the cross-kingdom Arabidopsis-B . cinerea GCNs . Model-corrected means of 23 , 959 Arabidopsis transcripts and 6 , 372 B . cinerea transcripts derived from two negative binomial linked generalized linear models were served as input data sets ( Zhang et al . , 2017 ) . Spearman’s rank correlation coefficient was calculated between genes from Arabidopsis and B . cinerea data sets . The gene pairs with positive correlations greater than 0 . 74 under each Arabidopsis genotype were considered to construct cross-kingdom GCNs . To construct a cross-kingdom , dual interaction network of plant-pathogen GCNs , we performed principle component analysis on individual GCNs within each species to obtain eigengene vectors describing the expression of the entire gene network as previously described ( Zhang and Horvath , 2005; Langfelder and Horvath , 2008; Okada et al . , 2016 ) . From these eigengene vectors , we calculated the Spearman’s rank correlation coefficient between the first eigengene vectors for each network . The resulting similarity matrices were used as input to construct the interaction network and Cytoscape was used to visualize the resulting network . All the analyses were conducted using R V3 . 2 . 1 statistical environment ( R Core Team , 2014 ) . To investigate how secondary metabolite induction in B . cinerea contributes to disease development , we conducted a multi-factor ANOVA on B . cinerea three secondary metabolic pathways upon impacts on host genotypes . The three secondary metabolic pathways included the biosynthetic pathways of two well-known secondary metabolites , BOT and BOA , and a cyclic peptide biosynthetic pathway predicted in this study . We calculated the z-scores for all transcripts involved in BOT pathway , the BOA , and the putative cyclic peptide pathway for each isolate/plant genotype pair . The multi-factor ANOVA model for lesion area was:yLesion=μ+T∗A∗C∗Gh+ϵwhere T , A , C , and Gh stand for BOT , BOA , Cyclic peptide , and host genotype , respectively . In addition , we used multi-factor ANOVA models to investigate interactions between GCNs within species for impacts upon host genotypes . The ANOVA models contain all GCNs within a species . The first eigengene vector derived from principal component analysis on each network was used in ANOVA models . The ANOVA model for individual B . cinerea GCNs was:yBcNeti=μ+D∗P∗C∗PSI∗Gh+ϵwhere D , P , C , PSI , and Gh stand for Arabidopsis Defense/Camalexin GCN , Arabidopsis Plastid GCN , Arabidopsis Cell/Division GCN , Arabidopsis PSI GCN , and Host genotypes , respectively . Gh stands for HostGenotype , respectively . BcNeti represents one of the ten B . cinerea GCNs identified in this study . The ANOVA model for individual Arabidopsis GCNs was:yAtNeti=μ+ΣBcNeti+Gh+ϵwhere ∑BcNeti represents the summation of each of the ten B . cinerea GCNs identified in this study: BcVesicle/Viru GCN , BcTSL/Growth GCN , BcBOA GCN , BcExocytoRegu GCN , BcCycPep GCN , BcCopperTran GCN , BcBOT GCN , BcPeptidase GCN , BcIPP GCN , BcPolyketide GCN , while Gh stands for Host genotypes . Interactions among the terms were not tested to avoid the potential for overfitting . AtNeti stands for one of the four Arabidopsis GCNs ( e . g . AtDefense/Camalexin GCN , AtPlastid GCN , AtCell/Division GCN , AtPSI GCN ) . All multi-factor ANOVA models were optimized by trimming to just the terms with a significant P-value ( P-value < 0 . 05 ) . To assess the potential for natural variation in germination time in the isolate collection , 19 B . cinerea isolates were investigated by germination assay . The isolates were grown on PDA . Mature spores were collected in water , filtered and resuspended in 50% grape juice , as previously described , and further diluted to 1000spores/μL . To prevent germination before the beginning of the assay , spores were continuously kept on ice or in the fridge at 4°C . During the germination assay , the spores were maintained at 21°C in 1 . 5 mL tubes . Every hour , the tubes were mixed by manual inversion and sampled for 25 μL that were transferred to microscope slides . The spores within the drops were let to set down shortly . Without using slide covers , the spores were observed within the drops at two locations , used as technical replicates . The spores were categorized and counted based on the picture of every microscope observations taken every hour from 2 to 11 hr . Germination was defined as the hyphae emerged out of the spore . To assess the contribution of germination to the observed B . cinerea transcriptomic networks involved in lesion development , we generated germination estimates based on gene expression by extracting the first principal component of a publicly available time series microarray data including 101 gemination-associated genes ( Leroch et al . , 2013 ) . Based on this principal component , we predicted the level of expression of germination-associated genes for the 96 isolates on the three Arabidopsis genotypes at 16HPI . Theses germination predictions for individual isolates were used in a linear ANOVA model to estimate the co-linearity of the germination eigengene vector to virulence . Using linear ANOVA models with and without this germination eigengene vector , we compared how germination influences the 10 B . cinerea transcriptomic networks involved in lesion area in the three host genotypes . The ANOVA models with and without germination eigengene vector were:YLesion=μ+Germination+ΣBcNeti+Gh+εYLesion=μ+ΣBcNeti+Gh+εwhere Germination represents the scores of first principal components on expressions of germination associated genes from B . cinerea transcriptomic data in this study , ∑BcNeti represents the summation of each of the ten B . cinerea GCNs identified in this study: BcVesicle/Viru GCN , BcTSL/Growth GCN , BcBOA GCN , BcExocytoRegu GCN , BcCycPep GCN , BcCopperTran GCN , BcBOT GCN , BcPeptidase GCN , BcIPP GCN , BcPolyketide GCN , while Gh stands for Host genotypes . Interactions among the terms were not tested to avoid the potential for overfitting .

Infections are complex interactions between two organisms . When a disease-causing microbe and a potential host engage , molecules continuously flow in both directions . This creates an inter-connected loop of messages and counter-messages , attacks , counter-attacks and resistance . This communication determines the final winner and the outcome of the disease . Yet it is technically difficult to measure it from both organisms at the same time , mostly because it is often impossible to tell whether a given molecule came from the microbe or the host . As such , little is known about how most infections play out at the molecular level . Now , rather than looking directly at the communication molecules , Zhang et al . have measured the active genes in samples of a plant infected with a fungus . While a molecule released by the plant may be indistinguishable from one from the fungus , the genes needed to make those molecules will be different in each species . The experiments involved two species where databases of gene sequences already exist: Arabidopsis thaliana , a plant often used in laboratory studies , and a fungus known as Botrytis cinerea , which infects many plants . Zhang et al . showed that the interactions between the two organisms are diverse and , rather than single genes , they largely involve sets of genes that are all switched on together as so-called gene co-expression networks ( or GCNs for short ) . Ten of these networks encoded mechanisms that allow the fungus to attack plant hosts . Further analysis identified potential connections between networks of genes in the plant and fungus . These connections may reveal some of the targets of the fungus’s toxins or counter mechanisms that plants can use to attempt to defend themselves . These findings show that it is possible to listen to the molecular communication between two organisms during an infection . In the future , a similar approach may make it possible to ask if a host plant communicates with all of its possible disease-causing microbes with a few distinct pathways , or if instead , hosts have the flexibility to uniquely communicate with each microbe in a different way .

2019

Plant–necrotroph co-transcriptome networks illuminate a metabolic battlefield

Toothed whales use sonar to detect , locate , and track prey . They adjust emitted sound intensity , auditory sensitivity and click rate to target range , and terminate prey pursuits with high-repetition-rate , low-intensity buzzes . However , their narrow acoustic field of view ( FOV ) is considered stable throughout target approach , which could facilitate prey escape at close-range . Here , we show that , like some bats , harbour porpoises can broaden their biosonar beam during the terminal phase of attack but , unlike bats , maintain the ability to change beamwidth within this phase . Based on video , MRI , and acoustic-tag recordings , we propose this flexibility is modulated by the melon and implemented to accommodate dynamic spatial relationships with prey and acoustic complexity of surroundings . Despite independent evolution and different means of sound generation and transmission , whales and bats adaptively change their FOV , suggesting that beamwidth flexibility has been an important driver in the evolution of echolocation for prey tracking . Echolocation has evolved independently multiple times in mammals and birds ( Kellogg et al . , 1953; Griffin , 1958; Konishi and Knudsen , 1979; Griffin and Thompson , 1982 ) , and allows these animals to orient under conditions of poor lighting . However , only toothed whales ( henceforth ‘whales’ ) and laryngeally echolocating bats ( henceforth ‘bats’ ) use echolocation to detect , locate , and track prey . In these groups , echolocation signals are primarily ultrasonic ( >20 kHz ) and are among the most intense biological sounds in water and in air ( Madsen and Surlykke , 2013 ) . These characteristics mean that in uncluttered spaces these predators can detect small prey many body-lengths ahead of them . Key to increasing the effective sensory range of biosonar is a directional sound beam ( i . e . , a narrow volume of forwardly ensonified space ) , which increases intensity along the acoustic axis and reduces ensonification of objects off-axis . It has recently been proposed that the advantages of a narrow sonar beam while in search of prey may have been the primary driver for the evolution of high frequency sonar signals in both whales and bats ( Koblitz et al . , 2012; Jakobsen et al . , 2013; Madsen and Surlykke , 2013 ) . As whales and bats close in on prey , both are thought to concurrently decrease signal intensity and auditory sensitivity to partially compensate for reduced transmission loss ( Hartley , 1992; Au and Benoit-Bird , 2003; Supin et al . , 2004; Linnenschmidt et al . , 2012; Madsen and Surlykke , 2013 ) , while increasing the signal emission rate for faster updates on prey location ( Griffin , 1958; Morozov et al . , 1972; Au , 1993 ) . Both groups terminate prey pursuits with a low intensity , high repetition rate sequence of echolocation signals called a buzz ( see ref . [Madsen and Surlykke , 2013] for review ) . Signal production rates characteristic of buzzes have likely evolved in these echolocators to facilitate close range prey tracking . However , for a given sonar beam , effective beam diameter will decrease as the distance between predator and prey diminishes . Thus , while a directional sound beam enables longer detection range by restricting the acoustic field of view ( FOV ) , it may be disadvantageous to the echolocator at short ranges , as moving prey may easily vanish at the periphery of the FOV . Directionality increases with aperture size ( e . g . , a bat's gape ) and signal frequency ( Au , 1993 ) and both factors appear to be exploited by some bats to modify their beam ( Surlykke et al . , 2009; Jakobsen and Surlykke , 2010 ) . Unlike bats , whales do not generate echolocation signals in the larynx nor do they emit them through the mouth or the nares ( Ridgway et al . , 1980; Cranford et al . , 1996 ) . Instead , whales have evolved specialized sound-producing structures , the phonic lips , located high in the blowhole ( Figure 1 ) . An echolocation click is generated by pneumatic actuation of the phonic lips as pressurized air is forced past them ( Cranford et al . , 1996 ) . The resulting sound pulse propagates into the fatty melon and is transmitted into the water as a directional beam ( Au et al . , 1986 ) . While the exact function of the melon is not known , its size and properties are expected to affect the radiation pattern of sound from the head ( Varanasi et al . , 1975; Aroyan et al . , 1992; Harper et al . , 2008 ) . As the bulbous melon fills a large proportion of the forehead ( Figure 1 ) , the diameter of the head has come to be considered indicative of radiating aperture size ( Au et al . , 1999 ) . 10 . 7554/eLife . 05651 . 003Figure 1 . Transverse MRI scans of a young harbour porpoise . The scale bars indicate 5 cm ( bar in B applies also to C and D ) . The caudal part of the melon ( yellow , A ) abuts against layers of connective tissue , muscles , and tendons ( red , A ) forming a dense theca , which , along with the skull and a collection of nasal air sacs , reflect the vibrations that originate in the phonic lips ( light brown , A ) into the melon ( Aroyan et al . , 1992; Cranford et al . , 1996 ) . The melon is under control of highly developed facial musculature ( Harper et al . , 2008; Huggenberger et al . , 2009 ) . The fibers and tendons of the muscles ( Mu ) associated with the melon ( Me ) lie at oblique angles relative to the frontal , transverse , and sagittal body planes ( Harper et al . , 2008; Huggenberger et al . , 2009 ) . The actions of these richly innervated muscles can change the three-dimensional shape and/or stiffness of the melon ( Harper et al . , 2008; Huggenberger et al . , 2009 ) , and thus likely adjust the properties of the emitted sound . DOI: http://dx . doi . org/10 . 7554/eLife . 05651 . 003 Compared to bats , whales exhibit little flexibility in the frequency content of their echolocation signals ( Madsen and Surlykke , 2013 ) . We might therefore assume whales are capable of only small changes in beam directionality ( Au et al . , 1995 ) and thus be limited to a rather fixed FOV ( Au et al . , 1999 ) . Yet , given that whales echolocate prey over much longer ranges than bats ( Madsen and Surlykke , 2013 ) and are capable of making much narrower beams ( Au et al . , 1999; Surlykke et al . , 2009; Koblitz et al . , 2012; Jakobsen et al . , 2013 ) , such a constraint is hard to reconcile with the disadvantages associated with a fixed beamwidth during close tracking of prey . In bats , pursuit is often over quickly , often with less than 500 ms elapsing between prey detection and interception ( Kalko , 1995 ) . Prey pursuits by whales can last many seconds ( Johnson et al . , 2004 ) ( Figure 2 ) . Accordingly , we hypothesize that long buzzes in which the whale might follow its target from an uncluttered water column to a highly cluttered sea floor ( Figure 2 ) , and back again , may demand greater beam plasticity than is found in bats . 10 . 7554/eLife . 05651 . 004Figure 2 . Long terminal phase of prey pursuit by an echolocating harbour porpoise . ( A ) Echogram ( see ‘Material and methods’: Live prey capture ) displaying sonar clicks and echoes recorded by an acoustic tag attached to the animal just behind its blowhole ( Johnson et al . , 2004 ) . y-axis ( left ) indicates time elapsed from emitted clicks to returning echoes , expressed also as target range ( right ) . Clicks emitted at rates corresponding to inter-click intervals shorter than 3 . 3 ms time-window are displayed repeatedly . The color scale indicates signal energy from blue ( faint ) to red ( intense ) . As pursuit proceeds from water column ( 0 . 7–1 . 1 m from surface , 2–2 . 5 m above sea floor ) to sea bottom the immediate acoustic scene becomes more cluttered with complex bottom echoes shortly following fish echoes ( from −9 s onwards ) . ( B ) Inter-click intervals color-coded for relative apparent output level ( RAOL; [Wisniewska et al . , 2012] ) of signals as recorded by the tag . RAOL variation may stem from rapid head movements , source level adjustments , and/or beam directionality changes ( with less energy reaching the tag from a narrow beam ) . On two occasions , when the fish escaped into the open space of the water column ( at −16 s and −13 s ) , the porpoise increased its ICIs significantly , beyond the values considered as buzz ( Wisniewska et al . , 2012 ) . However , when the fish escaped to similar distances while being at the bottom ( and thus moving in arguably a more predictable way ) the porpoise increased the ICIs only slightly , which might point to anticipatory acoustic tracking on the part of the echolocating animal . DOI: http://dx . doi . org/10 . 7554/eLife . 05651 . 004 Corroborating this hypothesis , most of the reported estimates of toothed whale beam patterns show relatively large variability ( Evans , 1973; Au et al . , 1986 , 1987 , 1995; Koblitz et al . , 2012 ) . Furthermore , a bottlenose dolphin was recently observed to steer , and modify the width of , its sonar beam when stationed on a bite plate and presented with targets displaced by large angles with respect to its body axis ( Moore et al . , 2008 ) . The dolphin was proposed to use two mechanisms as means of beamwidth modulation: ( i ) phase shifting between two pairs of phonic lips dually actuated for generation of a single click and/or ( ii ) manipulation of the volume and geometry of the melon and the associated air-sacs . However , the latest experimental data suggest that dolphins use a single pair of phonic lips to produce echolocation signals ( Au et al . , 2012; Madsen et al . , 2013; Finneran et al . , 2014 ) and that the strong amplitude dependence of dolphin click spectra gives rise to a variable beam pattern , with more directional signals at higher source levels ( Finneran et al . , 2014 ) . Hence , it remains unclear how and under what circumstances a bottlenose dolphin modulates the width of its sound beam , and whether other whale species , with more stereotyped sonar signals , such as porpoises , are capable of similar beamwidth changes . In this study , we set out to test the hypothesis that whales can change their FOV adaptively during prey interception . To that end , we used two complementary experiments to record the beam pattern and signal frequency content of echolocation clicks from harbour porpoises ( Phocoena phocoena ) closing in on targets . Additionally , we obtained magnetic resonance imaging on a dead harbour porpoise to visualize the sound generating structures , the fatty melon and the associated musculature and present video demonstrating melon deformations in an actively echolocating harbour porpoise . In experiment 1 , three captive porpoises were recorded individually using a linear horizontal array of 8 hydrophones spaced 60 cm apart and submerged 75 cm below the water's surface , as the porpoises captured dead fish ( Video 1 ) . This setup ( Figure 3A ) allowed us to quantify the animals' beam patterns over long ranges as they approached and intercepted natural targets free-floating in the water column . A total of 16 trials were recorded for the three animals ( 5–6 trials/porpoise ) . Only trials in which the porpoises swam directly towards the centre of the array were analyzed , amounting to two trials for Freja and Eigil and one trial for Sif ( a total of 75 clicks ) . Data from the two porpoises , Freja and Eigil , recorded during buzzes show that they could up to triple the width of their sonar beam as they approached prey ( median difference of 7 . 85° for four trials: from a mean of 12 . 9° ( 6 . 6°–30 . 5° , with the large beamwidths produced at short target ranges , Figure 3B ) during regular clicking ( ICI >13 ms; [Wisniewska et al . , 2012] ) to 19 . 3° ( 10 . 2°–28 . 9° ) during buzzes ( ICI ≤13 ms; [Wisniewska et al . , 2012] ) , Figure 3B ) . However , limitations in the linear array recordings prevented us from drawing strong conclusions about the exact extent of beam widening: the angular resolution was 4–12° and there was no way to precisely determine the vertical direction of the porpoise beam . Even though the video observations indicated that the vertical direction of the porpoise head was not changing with range to the array , this could not be adequately quantified . 10 . 7554/eLife . 05651 . 005Video 1 . A representative trial from experiment one . Video shows a porpoise capturing fish in front of the linear hydrophone array . Hydrophones were lowered to a depth of 75 cm along the short side of the pool . Prior to the experiment the blindfolded ( i . e . , wearing opaque silicone eyecups ) porpoise was stationed at the opposite end of the pool . As freshly thawed fish were introduced approx . 3 m from the array , the animal was cued to perform the capture task . The porpoise did not roll throughout the approach . Thus , the observed beam changes could not result from the beam not being rotationally symmetric and the animal rolling consistently during the buzz . DOI: http://dx . doi . org/10 . 7554/eLife . 05651 . 00510 . 7554/eLife . 05651 . 006Figure 3 . Porpoise biosonar beam widens at short target ranges . 3 dB beamwidth recorded in two experimental setups: ( A , B ) three harbour porpoises closing on prey and ( C , D ) a harbour porpoise approaching an aluminum sphere target . ( A , C ) show reconstructed porpoise locations for clicks fulfilling inclusion criteria in one trial per array configuration . Targets and their projections in the x–y plane are marked with dark-blue filled and open circles , respectively . For the small array recordings ( light blue in C ) , the target was displaced outward to 0 . 4 m from the array to maintain high spatial resolution at short ranges . ( B ) Data collected using a horizontal array with effective angular resolution ( EAR ) of ∼12° at ranges of target interception ( N = 75 ) . Data points from Freja , the porpoise participating in experiment two , are represented with circles . ( D ) Data gathered with star-shaped arrays in two configurations: large ( red in C ) , for long-range recordings ( >1 . 3 m from array to sound source , squares , N = 34 ) and small ( light blue in C ) , for greater resolution at short ranges ( <2 m from array to sound source , circles , N = 458 ) ( see Figure 4—figure supplement 1 for a detailed view of hydrophone spacing in the two array configurations ) . Hydrophone spacing provided EAR of ∼5° at the shortest ranges from the source examined . Color in ( B ) and ( D ) indicates inter-click intervals ( ICI ) , with buzz starting at 13 ms ( Wisniewska et al . , 2012 ) . Buzz- and regular-click datasets in ( D ) , used at short- and long-ranges , respectively , have different distributions , but similar medians , because during buzzes the animal repeatedly changed its beamwidth ( Figure 6 ) . Beam of the long-range clicks varied less and is better approximated by the median . DOI: http://dx . doi . org/10 . 7554/eLife . 05651 . 006 We therefore designed experiment two to measure the beamwidth at close target ranges using a non-uniform 2D hydrophone array . One of the three porpoises , Freja , was trained to approach a stationary target surrounded by a star-shaped array of 48 hydrophones in two configurations ( Video 2 ) . Initially , the hydrophones were arranged at increasing intervals away from the array center to cover the full extent of the beam at relatively long ranges while providing more resolution around the target at short ranges ( Figures 3C , 4 , Figure 4—figure supplement 1A ) . We then repeated the experiment using an array of more-tightly spaced hydrophones ( Figures 3C , 4 , Figure 4—figure supplement 1B ) and suspended the target further in front of the array , to increase signal intensity resolution at short ranges . 10 . 7554/eLife . 05651 . 007Video 2 . A representative trial from experiment two . Video shows a blindfolded porpoise closing on an aluminum target in front of a 48-element hydrophone array . The sequence was recorded in a short-range trial , that is , with the array extending to 0 . 5 m on either side of the centre hydrophone and the target moved to 0 . 4 m from array centre . Only clicks recorded when the porpoise was <2 m away from the array , at an angle of <15° to its centre and with acoustic axis within 6 cm from the centre hydrophone were selected for the beamwidth analysis . The video's soundtrack was replaced with audio recording from the camera-synchronized DTAG-3 carried by the porpoise . DOI: http://dx . doi . org/10 . 7554/eLife . 05651 . 00710 . 7554/eLife . 05651 . 008Figure 4 . −3 dB beamwidth variation with range . Colored markers indicate beamwidth estimates based on the best fitting piston aperture , while the black vertical lines show the spread around the best fit ( see lower panels in Figure 4—figure supplement 1 ) . Data were gathered with star-shaped arrays in two configurations ( Figure 3C and Figure 4—figure supplement 1 ) : large ( squares , N = 121 ) and small ( circles , N = 745 ) . Color indicates the least-square error associated with the fits . Only fits with error <0 . 2 were considered in the final analysis and presented in Figure 3 . Distance from the sound source to the tip of the animal's rostrum was 17 cm . DOI: http://dx . doi . org/10 . 7554/eLife . 05651 . 00810 . 7554/eLife . 05651 . 009Figure 4—figure supplement 1 . Array configurations and data fitting in experiment two . Upper panels: hydrophone ( black circles ) arrangement in the large ( A ) and small ( B ) star-shaped arrays . The filled circles show hydrophones selected for analysis ( having received peak levels exceeding the rms noise level of the channel by at least 14 dB ) . We calculated click energy on each of the selected channels and fitted a surface to the values using the Matlab ‘gridfit’ function ( grid spacing 0 . 5 cm ) . The surface was fitted over an area corresponding to ±20° at the range of signal emission . The color of the fitted surface represents the received energy from blue indicating faint to red indicating intense . The location of the beam axis was determined as the peak of the fitted surface and the purple star indicates the hydrophone closest to the estimated beam axis . The thick solid line shows the extent of the 3 dB beamwidth of the best-fitting piston source centered on the estimated beam axis . Middle panels: inter-click intervals used by the porpoise during the approach . The red circles mark the clicks shown in the upper panels , which were emitted at 6 m ( A ) and 0 . 2 m ( B ) from the array . Lower panels: results of the piston-fitting procedure for the click shown in the upper panel . The curve shows the least-square error at different piston sizes . The best fit is marked with a red circle . The dashed line limits the lowest 5% of the extent of the error curve , which was used to estimate the data spread around the best fit ( black circles ) shown in Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 05651 . 009 We recorded 123 trials with the long-range configuration and 93 trials with the short-range setup . Only trials in which the porpoise swam directly toward the target ( within ±15° vertically and horizontally from the array's center ) and repeatedly scanned its beam over the center hydrophone were analyzed . This resulted in 11 trials for the long-range- and 4 trials for the short-range array configuration . For each recorded click , the energy levels received at the hydrophones were fitted to the beam pattern of a circular piston ( Au et al . , 1987; Møhl et al . , 2003 ) . Only beamwidth estimates based on fits with R2 > 0 . 8 ( see Figure 4—figure supplement 1 ) were considered in the analysis , rendering a total of 34 and 458 clicks for the long-range- and short-range array , respectively . The results of this experiment show that the porpoise could almost double its beamwidth in degrees when switching to a buzz ( ICI ≤13 ms; [Wisniewska et al . , 2012] ) at short target ranges ( Figure 3D ) . That is , from a mean half-power beamwidth ( i . e . , the off-axis angle at which the sound energy has decreased by 3 dB relative to the on-axis energy ) of 9 . 1° ( 6 . 7–12 . 3 ) to a maximum of 15 . 1° ( mean = 11 . 0 , min = 8 . 2 ) . In this way the animal increased the ensonified area ahead of it by up to three times ( 200% ) in the terminal buzz phase of its approach to a target , as indicated by the ratio of the area measured at a given target range to the area predicted for that range based on the median −3 dB beam angle calculated for long ranges ( i . e . , predicted area had the porpoise not switched to ‘wide-angle view’ , Figure 5 ) . A significant inverse relationship between the beamwidth of the clicks and distance they were emitted from the target was found for data sets from each experiment ( Experiment 1: F = 56 . 9 , R2 = 0 . 44 , p < 0 . 001; Experiment 2: F = 23 . 6 , R2 = 0 . 05 , p < 0 . 001 ) . Data from experiment two were also analyzed using ANOVA ( see ‘Materials and methods’ ) . The three clusters ( sorted using beamwidth values ) differed significantly with respect to distance to target ( F = 5 . 21 , p < 0 . 006 ) . Specifically , clicks in the group with the greatest beamwidths were emitted significantly closer to the target than clicks in the group with the narrowest beamwidths ( Tukey-Karmer HSD , p < 0 . 01 ) . 10 . 7554/eLife . 05651 . 010Figure 5 . Beam adjustments can triple the ensonified area . ( A ) Approximate detection volume for a harbour porpoise tracking fish in a quiet environment , based on the active sonar equation ( 1 ) and source energy levels as measured . Fixed detection threshold ( Kastelein et al . , 1999 ) of 27 dB re 1 µPa2s and target strength of −36 dB for the Atlantic cod of 29–30 cm ( Au et al . , 2007 ) is assumed . Pattern of the outer beam ( solid line cross section ) is based on beamwidth estimates obtained at short target ranges . The inner , narrow beam ( dashed line ) is based on the directionalities measured at long range , representing predicted beam pattern had the porpoise not switched to ‘wide-angle view’ . ( B ) Relative change in the size of ensonified area ahead of the porpoise as it approaches a target . Surface area was computed as base of a cone with height equal to target range and an opening angle corresponding to the measured −3 dB beam angle ( measured area , solid line in A ) or median −3 dB beam angle calculated for long ranges ( >2 m; predicted area , dashed line in A ) . Color indicates inter-click intervals . Squares and circles mark data points obtained with the large ( N = 34 ) and small ( N = 458 ) array , respectively . The bold horizontal line indicates points where the measured and the predicted areas are equal . DOI: http://dx . doi . org/10 . 7554/eLife . 05651 . 010 The timeline of these changes ( Figure 6A , C ) shows that the porpoise varied the width of its beam within the buzz independently of the inter-click intervals ( ICIs ) , with both narrow and wide beam angles used at the lowest ICIs ( Figure 6A , D ) . Consequently , the short- and long-range datasets have different distributions , but similar means , and the long-range clicks are better described by the mean beamwidth value ( Figure 3D ) . The click centroid frequency ( fc ) dropped by about 1% at the start of a buzz ( from a long-range median of 130 . 4 kHz ( 127–136 . 1 ) to a median of 127 . 7 kHz ( 124 . 1–134 . 3 ) , see the color scale in Figure 6 and Figure 6—figure supplement 1 ) . 10 . 7554/eLife . 05651 . 011Figure 6 . Temporal variation in beamwidth within the terminal buzz . Beamwidth changes in terminal phases of two trials ( A , C ) and their respective inter-click intervals ( ICI; B , D ) . Color-coding represents centroid frequencies ( fc ) of signals . Dashed line in ( A , C ) corresponds to median beamwidth at long ranges ( >2 m ) . The porpoise used different beamwidths whilst maintaining ICIs and vice versa . Both trials were recorded with the small star-shaped array ( light blue in Figure 3C ) , but during the trial shown in ( C , D ) the porpoise was not blindfolded . Only data for clicks fulfilling inclusion criterion are presented . DOI: http://dx . doi . org/10 . 7554/eLife . 05651 . 01110 . 7554/eLife . 05651 . 012Figure 6—figure supplement 1 . Frequency variation alone cannot explain the observed beamwidth changes . Beamwidths of clicks recorded with the linear ( red circles ) and star-shaped ( black circles ) arrays plotted as a function of the clicks' centroid frequencies . Light gray lines represent beamwidth variation with frequency as modeled for static circular pistons of sizes considered in the piston-fitting procedures of the present study ( right y-axis ) . Only centroid frequencies within the ranges measured in the two experiments were considered . Piston models fitting best ( based on the lowest root-mean-square errors ( RMSE ) and sum of squared errors ( SSE ) ) to the beamwidths and centroid frequencies measured with the linear and star-shaped arrays are marked with dashed red and black lines , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 05651 . 012 This negligible drop in frequency cannot explain the large changes in beamwidth ( Figure 6—figure supplement 1; Spearman's correlation tests between centroid frequency and beamwidth: Experiment 1: p = 0 . 76 , Experiment 2: p = 0 . 67 ) . Rather , the animal must vary the size of the effective aperture , likely by effecting rapid muscular deformations of the melon ( Video 3 ) , the position of the phonic lips and/or the size and position of the associated air sacs . To visualize the musculature surrounding the melon , we obtained magnetic resonance imaging ( MRI ) on a dead juvenile harbour porpoise . The scanning images reveal nasal structures that include a complex , richly innervated ( Huggenberger et al . , 2009 ) network of facial muscles ( Figure 1 ) . These muscles are homologous to the muscles dedicated to facial expressions in primates , and should enable fast and subtle changes in the shape of nasal components and their associated air sacs ( Huggenberger et al . , 2009 ) . 10 . 7554/eLife . 05651 . 013Video 3 . Harbour porpoises can manipulate their melon while producing clicks . Video shows a harbour porpoise emitting echolocation click trains during a hearing test . It has been slowed down by a factor of two and synchronized with the output of a porpoise click detector . The porpoise depresses the melon as it switches to high repetition rate click trains . Conformation changes in the nasal complex can modulate the degree of sound collimation in the whale's forehead ( [Harper et al . , 2008; Moore et al . , 2008; Huggenberger et al . , 2009] ) to change the field of view . Courtesy of Lee Miller . DOI: http://dx . doi . org/10 . 7554/eLife . 05651 . 013 Here , we show that harbour porpoises can broaden their biosonar beams in the final phase of target approach ( Figure 3 ) , and , unlike echolocating bats , that they are also able to change their beamwidth within the terminal buzz ( Figure 6 ) . At its broadest , the beamwidth ranged from 15° in experiment two to more than 30° in experiment one . A number of factors may have contributed to this variability including: ( i ) differences in effective angular resolution ( EAR ) between the eight-hydrophone linear array and the 48-hydrophone star-shaped array , ( ii ) differences in the animals' behaviour when approaching a slowly sinking fish vs a stationary aluminum sphere , and , finally , ( iii ) potentially higher clutter originating from the large array behind the target and motivating the porpoise to use a narrower beam . Porpoises can adjust their buzz clicking rate to prey range differently when following a fish in open water than when tracking one in the cluttered and more restricted space close to the sea floor ( Figure 2 ) . The porpoise behaviour and experimental context seem the most likely explanation for the observed differences between the two experiments . This is the first demonstration , to our knowledge , of a whale controlling its acoustic FOV while actively approaching a target . This control is achieved independently of spectral adjustments ( Figure 6 and Figure 6—figure supplement 1 ) , probably by changing the conformation of the melon ( Figure 1; [Huggenberger et al . , 2009] ) , the position of the phonic lips ( Cranford et al . , 2014 ) and the size and shape of the associated air sacs ( Aroyan et al . , 1992 ) . Due to its heterogeneous structure ( Norris and Harvey , 1974; Varanasi et al . , 1975 ) , the melon has long been considered an acoustic impedance matcher that minimizes the reflections and energy loss at the tissue–water interface ( Norris and Harvey , 1974 ) and that provides directionality in the emitted click ( Au et al . , 1999 , 2006 ) . Porpoises have a complex facial musculature ( Figure 1 ) with nervous innervation with 4 . 5 times more neurons than human facial muscles ( Jacobs and Jensen , 1964 ) , leading to the recent proposition that muscle induced deformations of the nasal soft structures such as the melon may provide means to change the FOV ( Huggenberger et al . , 2009 ) . Our acoustic measurements and observations support that hypothesis by demonstrating that the melon and accessory structures apparently operate as the functional equivalent of an adjustable collimating lens of a flashlight . In other words , the porpoise's beam can be dynamically changed from spotlight to floodlight ( and everything in between ) to best suit the circumstances , offering unprecedented flexibility in control of the FOV in an echolocating animal that is unmatched in visual mammals ( Land and Nilsson , 2012 ) . The porpoise's ability to change its FOV within a buzz ( Figure 6 ) implies an even greater flexibility than recently documented in vespertilionid bats ( Surlykke et al . , 2009; Jakobsen and Surlykke , 2010; Jakobsen et al . , 2013 ) . While bats adjust their beams to the environment in which they are operating ( Surlykke et al . , 2009 ) , and the task at hand ( Jakobsen and Surlykke , 2010 ) , changes in their FOV during the buzz are accounted for by a concurrent drop in signal frequency content . Thus in bats , changes in FOV , emission rate , and signal frequency content during the buzz appear to be tightly interconnected ( Ratcliffe et al . , 2013 ) . Our results show that in whales these parameters are independently controlled ( Figure 6 and Figure 6—figure supplement 1 ) . Having a FOV that can be modulated independently of signal emission rate or frequency content ( the latter often being dependent on the amplitude of the signal [Au et al . , 1995; Finneran et al . , 2014] ) may be essential for managing flow of sensory information and optimizing long duration close-range prey tracking in acoustic scenes of varying complexity ( Figure 2 ) . The broad beam is advantageous to the porpoises at close range where it would reduce the likelihood of prey escaping perpendicularly to the approaching porpoise by vanishing from its acoustic FOV . These findings support the hypothesis that porpoises dynamically control their acoustic FOV while tracking prey , and do so by altering the effective size of their radiating aperture . The mechanism underlying these adjustments may be muscle-induced phonic lips repositioning ( Cranford et al . , 2014 ) , and melon and air sac deformations ( Moore et al . , 2008; Huggenberger et al . , 2009 ) . All toothed whales studied to date have similar facial musculature surrounding the melon ( Cranford et al . , 1996; Harper et al . , 2008 ) . All toothed whales then presumably have the ability to modify the melon's shape . Given the greater beam plasticity offered by this mechanism , compared to modulating the frequency content of clicks , we suggest that all toothed whales may be able to shape and modulate their beam this way . Despite the independent evolution and very different means of sound generation and transmission , whales and bats have both evolved mechanisms to change their acoustic FOV while tracking prey . This suggests that beam plasticity has been a key driver in the evolution of echolocation , beyond simple orientation , for improved foraging success . Our results from these small toothed whales suggest that the demands of tracking moving prey over variable distances in complex acoustic environments have favored the evolution of a more sophisticated adjustment mechanism , in which pulse rate and beamwidth can be controlled independently . Compared to bats , the greater dynamic beam plasticity we have observed here likely reflects different sensory and ecological constraints . We propose that dynamic control of acoustic FOV in whales is a mechanism for the inclusion and exclusion of potential sensory information that allows these predators to quickly and repeatedly adjust to changes in habitat and prey trajectories . We recorded echolocation clicks from three blindfolded ( i . e . , wearing opaque silicone eyecups ) porpoises as they swam alone in the 3–4 m deep net-pen across the long side of the pool to capture dead , freshly thawed fish and towards a horizontal linear array of 8 calibrated Reson TC4014 hydrophones spaced 60 cm apart ( Figure 3A and Video 1 ) . The array was deployed at a depth of 75 cm and the fish were introduced approx . 3 m away from its centre , giving the array an effective angular resolution of ∼12° at ranges of target interception ( EAR = atan[0 . 6 m/3 m] ) . Signals were amplified and filtered using a custom-made conditioning box and then simultaneously A/D converted with 16-bit resolution at 500 kHz per channel ( National Instruments PXI-6123 , Austin , TX ) . All trials were monitored using a video camera ( Profiline CTV7040 , Abus , Germany ) synchronized with the audio recordings . Analyses were performed using Matlab ( MathWorks , Natick , MA ) . We localized the animal at the time of each emission using hydrophone arrival-time differences ( Madsen and Wahlberg , 2007 ) . Porpoise positions were then verified with the synchronized videos . For each click , the time of maximum sound pressure on each hydrophone was identified , and the energy density of the signal was measured using a window of 30 µs before and 90 µs after the peak of the signal envelope . Such a window corresponds to the duration of a typical porpoise signal . Assuming spherical spreading , the apparent source level ( ASL ) ( Madsen and Wahlberg , 2007; Finneran et al . , 2014 ) was calculated from the recorded level ( computed as energy flux density in a fixed window [peak-30 µs—peak +90 µs] ) and the range . For a click to be included in the analysis its ratio of signal energy to the immediately preceding noise energy had to exceed 6 dB on all channels , and the maximum ASL should not have occurred on one of the outermost hydrophones . Radiation plots were created by plotting the ASLs against their respective angles relative to the estimated on-axis direction . First , the peak amplitude and angle were adjusted by interpolating between the peak ASL and the ASLs from the two neighbouring hydrophones using Lagrange interpolation ( Menne and Hackbarth , 1986 ) . All off-axis levels were then plotted as a function of the off-axis angle . The resulting transmission beam pattern was interpolated to a grid of 0 . 1° . The circular piston model was used to estimate the directivity index and −3 dB beamwidth of the beam pattern as described in ( Møhl et al . , 2003 ) . A single harbour porpoise ( Freja ) was trained to swim across the short side of the pool and close in on a 50 . 8 mm-diameter spherical aluminum target , suspended by a nylon line just in front ( 5–40 cm , depending on array configuration , see below and Figure 3 ) of the center of an array of 48 small ( 25 × 10 mm ) , custom-built hydrophones ( Wisniewska et al . , 2012 ) ( Video 2 ) . The hydrophones were attached to a mesh of 1 mm-diameter Dyneema string stretched over a 3 × 3 . 5 m ( height by width ) metal frame to form a grid of 5 × 5 cm squares . The hydrophones had a flat ( ±2 dB ) frequency response between 100–160 kHz and were connected to a 48 channel conditioning box with 40 dB gain and a fourth order bandpass filter with −3 dB frequencies at 2 kHz and 200 kHz . The hydrophones were sampled continuously during the trials with 16-bit resolution at 500 kHz/channel by three synchronized National Instruments PXIE-6358 boards and streamed to disk , using custom made software developed in LabVIEW , National Instruments ( Source code 1 ) . Differences in array sensitivity due to hydrophone arrangement and attachment were measured after each data collection and corrected during post-processing . The hydrophones were arranged in two star-shaped configurations ( Figure 3C and Figure 4—figure supplement 1 ) . To map the beam extent at long ranges ( 1 . 3–7 m ) , we used hydrophone spacing increasing toward the edge hydrophones 1 . 05–1 . 13 m from the centre . Consequently , along the vertical and horizontal axes the hydrophones were separated by: 5 , 10 , 15 , 20 , 25 , and 30 cm . Along the diagonal axes the hydrophones were separated by 14 . 1 , 14 . 1 , 21 . 2 , 28 . 3 , and 35 . 4 cm . With hydrophones 14 cm apart , the effective angular resolution was ∼6° at 1 . 3 m from the array's centre . To maintain high spatial resolution at short ranges ( ≤2 m ) , we displaced the target outward to 0 . 4 m from the array and rearranged the hydrophones , resulting in an array extending out to 0 . 5 m with hydrophones separated by 5 , 5 , 5 , 15 , and 25 cm along the vertical and horizontal axes , and by 7 . 1 , 7 . 1 , 14 . 1 , and 21 . 2 cm along the diagonal axes . Thus , a sound source at 0 . 55 m from the array ( i . e . , the shortest range examined ) could have had its beamwidth measured to within ∼5° . We pooled the two data sets together with a range overlap between 1 . 3–2 m . Data points acquired with the wider spacing when the animal was closer than 1 . 3 m were discarded , as were data acquired with the fine spacing when the porpoise was more than 2 m away . The porpoise was equipped with a DTAG-3 multi-sensor tag ( Johnson and Tyack , 2003; Johnson et al . , 2004; Wisniewska et al . , 2012 ) attached with suction cups just behind the blowhole , to allow for measuring the range of the sound source to the target and the array from the difference in time-of-arrival of click–echo pairs . The tag sampled sound with 16-bit resolution at 500 kHz/channel and was synchronized with the array hydrophones ( Wisniewska et al . , 2012 ) . In all but one ( Figure 6C , D ) of the analyzed trials , the porpoise was blindfolded with opaque eyecups . All trials were monitored with a set of GoPro Hero 2 cameras ( two on the heads of the trainers and one approximately 1 . 8 m behind the array; Eye of Mine Action Cameras , Carson , CA ) synchronized with the DTAG-3 recordings . The recorded trials were pre-screened for relatively straight approaches to the array using the videos . All subsequent sound analyses were performed using Matlab . Clicks from the study animal were identified in the DTAG-3 acoustic recordings using a supervised click detector . Spectral cues were used to eliminate occasional misdetections of echoes or signals from other porpoises in the neighbouring pen . Echograms were formed from the sounds recorded with the DTAG-3 . Only trials with clear echoes from the target and the array were submitted to further analysis . Clicks from 13–21 key hydrophones of the large- and small star-shaped array were extracted using a supervised click detector with the synchronized timing of the clicks recorded on the animal as an input . Clicks from all the verified channels were then combined into a single template and used for automatic click detection on the remaining channels . For each click , we identified a subset of channels with peak received levels exceeding the rms noise level of the channel by at least 14 dB . We calculated click energy in a fixed window ( peak-30 µs—peak +90 µs ) on each of the selected channels , fitted a surface to the values using the Matlab ‘gridfit’ function with grid spacing of 0 . 5 cm and determined the location of the beam axis as the peak of the fitted surface ( Figure 4—figure supplement 1 , upper panels ) . We ran a series of computer simulations of our methods applied to virtual sources of known piston sizes to evaluate the influence of ( i ) an animal's bearing in azimuth and elevation and ( ii ) beam axis displacement relative to the centre of the array on the beamwidth measurement error . Based on the results of these simulations , we restricted our analysis to only include clicks emitted when the porpoise was swimming directly toward the target ( within ±15° vertically and horizontally from the array's centre ) and with beam axis within ( i ) ±12 cm from the array centre for the large array , ( ii ) ±8 cm for clicks recorded with the small array and produced up to 1 s before the start of buzz , and ( iii ) ±6 cm for clicks made 1 s before buzz and onwards until the end of the target approach . The estimated error was thereby limited to ≤±0 . 5° . The simulations also verified that at these displacements the location of the beam axis could be estimated with an accuracy of ±1 . 5 cm at the ranges considered in this study . We used hydrophone arrival-time differences to compute the animal's bearing ( azimuth and elevation ) to the centre of the array . For each click , we followed the method of Kyhn et al . ( 2010 ) to fit the energy levels received at the hydrophones to the beam pattern of a circular piston that fulfilled the following criteria: ( i ) it was at the same range to the array as the porpoise emitting the click , ( ii ) it was centered on the estimated beam axis and ( iii ) it transmitted the click recorded on the hydrophone closest to that axis . Given that the orientation of the porpoise relative to the array was constantly changing as the animal was approaching the target ( Video 2 ) , the beam pattern was assumed to be rotationally symmetrical around the acoustic axis . Furthermore , this assumption allowed us to utilize information from all hydrophones fulfilling the signal-to-noise ratio criterion to find the best fitting aperture . We carried out a Monte Carlo simulation using theoretical piston transducers with diameters of 1/3 to 3 times 8 . 3 cm ( i . e . , the best fitting vertical equivalent aperture in [Koblitz et al . , 2012] ) in 0 . 1 cm increments , and the circular piston model of Au et al . ( 1987 ) . The diameter of the piston that matched the data best was found by means of a non-linear least-squares method ( see Figure 4—figure supplement 1 ) . Only fits with R2 > 0 . 8 ( Figure 4 ) were kept for final analysis , from which the −3 dB beamwidth and the equivalent piston radius were extracted . To examine the relative change in the size of ensonified area ahead of the porpoise as it approached a target , we computed the surface area as base of a cone with height equal to target range and an opening angle corresponding to ( i ) the measured −3 dB beam angle and ( ii ) median −3 dB beam angle calculated for long ranges . We pooled beamwidth and distance to target data for all clicks from Experiment 1 for the three porpoises and ran a regression analysis on these data . We then pooled beamwidth and distance to target data for both array configurations from Experiment 2 and ran a regression analysis on these data as well . Additionally , we used hierarchical cluster analysis ( centroid , non-standardized ) to assign click beamwidth data from Experiment 2 to one of three clusters and compared these clusters using ANOVA with respect to click distance to target . All statistical analyses were carried out using JMP v . 11 . 2 ( SAS Institute , Cary , NC , USA ) . To explore the acoustic scene experienced by a toothed whale tracking active prey , we deployed DTAG-3 tags ( with the same recording settings as in experiment two ) on porpoises involved in pursuit of small ( ∼15 cm ) , live trout in the pen complex of Fjord&Belt , where the animals have access to a natural sandy bottom at 3–4 m depth . This unique setting approximates what this shallow-water predator might encounter in its natural surroundings . The tags recorded the whale's echolocation clicks as well as echoes from the fish and other objects and surfaces in the whale's surroundings ( e . g . , water surface and sea floor ) . The recorded sounds were used to form stack plots , or echograms , of sound envelopes synchronized to the outgoing click as in echosounder images ( Johnson et al . , 2004 ) . These allowed us to follow movements of the echolocator and its prey in the environment ( Figure 2 ) . The time delay between the echolocation click and the echo , multiplied by one-half of the sound speed ( 1500 m/s was assumed ) gives the distance to the target . Delays to the surface and bottom echoes approximate the animal's depth and altitude above the sea floor , respectively . Time delays between different echo groups represent their relative proximity . A video of melon deformations can be found online ( Video 3 ) . To visualize the anatomy of the head , a dead specimen was scanned in a 1 . 5T Siemens Avanto MRI system ( Siemens Medical Solutions , Germany ) . A Flash 3D T1 weighted pulse-sequence with the following parameters was used: TR 14 . 8 ms , TE 3 . 38 ms , α = 15° , NEX = 3 , spatial resolution = 0 . 64 × 0 . 64 × 0 . 75 mm . Following acquisition , segmentation and modeling were done using Amira 5 . 3 . 3 ( Visualization Science Group , Germany ) .

Bats and toothed whales such as porpoises have independently evolved the same solution for hunting prey when it is hard to see . Bats hunt in the dark with little light to allow them to see the insects they chase . Porpoises hunt in murky water where different ocean environments can quickly obscure fish from view . So , both bats and porpoises evolved to emit a beam of sound and then track their prey based on the echoes of that sound bouncing off the prey and other objects . This process is called echolocation . A narrow beam of sound can help a porpoise or bat track distant prey . But as either animal closes in on its prey such a narrow sound beam can be a disadvantage because prey can easily escape to one side . Scientists recently found that bats can widen their sound beam as they close in on prey by changing the frequency—or pitch—of the signal they emit or by adjusting how they open their mouth . Porpoises , by contrast , create their echolocation clicks by forcing air through a structure in their blowhole called the phonic lips . The sound is transmitted through a fatty structure on the front of their head known as the melon , which gives these animals their characteristic round-headed look , before being transmitted into the sea . Porpoises would also likely benefit from widening their echolocation beam as they approach prey , but it was not clear if and how they could do this . Wisniewska et al . used 48 tightly spaced underwater microphones to record the clicks emitted by three captive porpoises as they approached a target or a fish . This revealed that in the last stage of their approach , the porpoises could triple the area their sound beam covered , giving them a ‘wide angle view’ as they closed in . This widening of the sound beam occurred during a very rapid series of echolocation signals called a buzz , which porpoises and bats perform at the end of a pursuit . Unlike bats , porpoises are able to continue to change the width of their sound beam throughout the buzz . Wisniewska et al . also present a video that shows that the shape of the porpoise's melon changes rapidly during a buzz , which may explain the widening beam . Furthermore , images obtained using a technique called magnetic resonance imaging ( MRI ) revealed that a porpoise has a network of facial muscles that are capable of producing these beam-widening melon distortions . As both bats and porpoises have evolved the capability to adjust the width of their sound beam , this ability is likely to be crucial for hunting effectively using echolocation .

2015

Range-dependent flexibility in the acoustic field of view of echolocating porpoises (Phocoena phocoena)

Genomic instability is a fundamental feature of human cancer often resulting from impaired genome maintenance . In prostate cancer , structural genomic rearrangements are a common mechanism driving tumorigenesis . However , somatic alterations predisposing to chromosomal rearrangements in prostate cancer remain largely undefined . Here , we show that SPOP , the most commonly mutated gene in primary prostate cancer modulates DNA double strand break ( DSB ) repair , and that SPOP mutation is associated with genomic instability . In vivo , SPOP mutation results in a transcriptional response consistent with BRCA1 inactivation resulting in impaired homology-directed repair ( HDR ) of DSB . Furthermore , we found that SPOP mutation sensitizes to DNA damaging therapeutic agents such as PARP inhibitors . These results implicate SPOP as a novel participant in DSB repair , suggest that SPOP mutation drives prostate tumorigenesis in part through genomic instability , and indicate that mutant SPOP may increase response to DNA-damaging therapeutics . Genomic instability is a fundamental feature of human cancer , and DNA repair defects resulting in impaired genome maintenance promote pathogenesis of many human cancers ( Hanahan and Weinberg , 2011; Garraway and Lander , 2013 ) . In prostate cancer , structural genomic rearrangements , including translocations ( e . g . , TMPRSS2-ERG ) and copy number aberrations ( e . g . , 8q gain , 10q23/PTEN loss ) are a key mechanism driving tumorigenesis ( Visakorpi et al . , 1995; Cher et al . , 1996; Tomlins et al . , 2005; Zhao et al . , 2005; Liu et al . , 2006; Demichelis et al . , 2009; Beroukhim et al . , 2010 ) . Whole genome sequencing ( WGS ) has allowed an unprecedented insight into the alterations underlying cancer . Recently , WGS of treatment naive , clinically localized prostate cancer revealed a striking abundance of genomic rearrangements , in some samples comparable to the number of rearrangements in metastatic prostate cancer ( Baca et al . , 2013 ) . Furthermore , the type ( intrachromosomal vs interchromosomal ) and complexity of rearrangements in these tumors shows remarkable heterogeneity , potentially suggesting distinct mechanisms of instability in different molecular classes of prostate cancer . However , somatic alterations underlying these phenomena remain unexplained . Mutations in SPOP ( Speckle-type POZ protein ) occur in around 10% of prostate cancers and represent the most common non-synonymous mutations in primary prostate cancer ( Barbieri et al . , 2012 ) . SPOP mutations define a distinct molecular class of prostate cancer; they are mutually exclusive with ETS rearrangements but display distinct patterns of somatic copy number alterations ( SCNAs ) ( Barbieri et al . , 2012 ) . Here , we investigated somatic alterations associated with genomic rearrangements in prostate cancer . We show that SPOP mutation is an early event specifically associated with increased intrachromosomal genomic rearrangements . Mechanistically , in vitro and in vivo data suggest that SPOP participates in repair of DNA double strand breaks ( DSB ) , and SPOP mutation impairs homology-directed repair ( HDR ) , instead promoting error-prone non-homologous end joining ( NHEJ ) . To nominate somatic events associated with structural genomic rearrangements in clinically localized prostate cancer , we examined WGS data from 55 treatment naive prostate cancers ( Baca et al . , 2013 ) ( Figure 1A ) . This analysis revealed a bimodal distribution , with a more common , ‘low-rearrangement’ population , and a less frequent ‘high-rearrangement’ population primarily driven by intrachromosomal rearrangements ( deletions , inversions , and tandem duplications ) , rather than balanced interchromosomal rearrangements ( Figure 1B ) . We then analyzed the association between recurrent somatic alterations ( point mutations and SCNAs ) and number of rearrangements ( Figure 1C; Figure 1—figure supplements 1 , 2 ) . Several recurrent deletions , primarily on chromosomes 5q and 6q , were significantly associated with intrachromosomal genomic rearrangements ( Figure 1—figure supplement 1 ) , and these were completely distinct compared to alterations associated with interchromosomal rearrangements ( Figure 1—figure supplement 2 ) . Among recurrent point mutations , only a single lesion—mutation in SPOP—was significantly associated with increased rearrangements ( Figure 1C ) . Consistent with increased intrachromosomal rearrangements , SCNA analysis revealed that SPOP mutant prostate cancers showed significantly higher total copy number alteration burden ( Figure 1D ) . 10 . 7554/eLife . 09207 . 003Figure 1 . SPOP mutant prostate cancer displays increased genomic rearrangements . ( A ) Distribution analysis of genomic rearrangements from 55 clinically localized prostate cancers distinguishes two subpopulations . ( B ) Increased total rearrangements are driven by intrachromosomal rather than interchromosomal rearrangements . 55 clinically localized prostate cancers ordered ( right to left ) according to total rearrangements; numbers of intrachromosomal and interchromosomal rearrangements are displayed . ( C ) Association of recurrent point mutations with intrachromosomal rearrangements . X-axis shows ( −log10 ) p-value . ( D ) SPOP mutant prostate cancers harbor increased total somatic copy number aberration ( SCNA ) burden . The fraction of altered genome , partitioned into bins covering a range from <0 . 01 to ≥0 . 5 , is shown as a histogram for SPOP WT and SPOP mutant tumors . Inset: the percentage of altered genome is significantly increased in SPOP mutant prostate cancers ( p = 0 . 0016 , two-sample Wilcoxon-Mann-Whitney test ) . ( E ) Frequency of somatic copy number alterations in 430 clinically localized prostate cancers . SPOP-mutant cancers ( orange ) and SPOP-wild-type tumors ( gray ) . Length of bars indicates the frequency of copy number alterations . ( F ) Clonality of selected alterations associated with genomic rearrangements , in 430 clinically localized prostate cancers . DOI: http://dx . doi . org/10 . 7554/eLife . 09207 . 00310 . 7554/eLife . 09207 . 004Figure 1—figure supplement 1 . Association of recurrent SCNAs with intrachromosomal rearrangements . Cytobands undergoing deletion are labeled blue . X-axis shows ( −log10 ) p-value . DOI: http://dx . doi . org/10 . 7554/eLife . 09207 . 00410 . 7554/eLife . 09207 . 005Figure 1—figure supplement 2 . Association of recurrent SCNAs or point mutations with interchromosomal rearrangements . Cytobands undergoing deletion are labeled blue , PM refers to recurrent point mutations . ‘ERG rearrangement’ was determined by FISH assay . X-axis shows ( −log10 ) p-value . DOI: http://dx . doi . org/10 . 7554/eLife . 09207 . 00510 . 7554/eLife . 09207 . 006Figure 1—figure supplement 3 . Genomic view of CHD1 and MAP3K7 areas in 430 prostate cancers sorted by SPOP mutation status . Blue indicates genomic deletion . DOI: http://dx . doi . org/10 . 7554/eLife . 09207 . 00610 . 7554/eLife . 09207 . 007Figure 1—figure supplement 4 . Genomic burden of three subset of prostate cancers defined by the aberration state of SPOP , CHD1 , and MAP3K7 . No increased signal is detected with respect to the SPOP class . DOI: http://dx . doi . org/10 . 7554/eLife . 09207 . 00710 . 7554/eLife . 09207 . 008Figure 1—figure supplement 5 . Evolution graphs built from prostate cancer sequencing data . These data were generated based on a previously published algorithm to build tumor evolution paths ( Baca et al . , 2013; Prandi et al . , 2014 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09207 . 008 SPOP mutation frequently co-occurs with specific SCNAs , designating a molecular class of prostate cancer ( Barbieri et al . , 2012; Blattner et al . , 2014 ) ( Figure 1—figure supplement 3 ) . Independent analysis of SCNAs from three publicly available data sets comprising 430 tumors ( Baca et al . , 2013; Barbieri et al . , 2012; Consortium TCGA , 2015 ) , including 47 SPOP mutant prostate cancers , confirmed that the rearrangement-associated deletions ( Figure 1—figure supplement 1 ) were those enriched in SPOP mutant prostate cancer ( Figure 1E ) . When individually comparing SPOP mutations and associated deletions ( CHD1 and MAP3K7 ) , we did not observe significant differences in SCNA burden for any one lesion ( Figure 1—figure supplement 4 ) . Analysis of clonality ( Baca et al . , 2013; Prandi et al . , 2014 ) of specific lesions showed that SPOP mutations were highly clonal compared to loci in the associated deletion peaks , supporting that SPOP mutations precede deletions ( Figure 1F ) . In addition , analysis of dependencies of the lesions supports SPOP mutations preceding CHD1 deletions; no lesions were predicted to precede SPOP mutation ( Figure 1—figure supplement 5 ) . Together , these data nominate a distinct prostate cancer class characterized by early SPOP mutations and genomic instability . We posited that the SPOP mutation impacts genome maintenance and prioritized SPOP for functional studies . We explored the functional role of SPOP in vivo , using zebrafish as a rapidly assessable vertebrate model system . SPOP is highly conserved ( 97 . 3% identical at the amino acid level between human and zebrafish , Figure 2—figure supplement 1A ) . Knockdown of Spop by two different splice-blocking morpholinos ( MO5 , MO7 ) dramatically impaired brain and eye development as well as decreased overall body size ( Figure 2A , B; Figure 2—figure supplement 1F ) , resulted in gene expression changes consistent with p53 activation , and apoptosis measured by TUNEL assay ( Figure 2C , Figure 2—figure supplement 1E ) . Microinjection of human SPOP mRNA rescued these phenotypes , confirming specificity of the morpholino effects ( Figure 2A–C , Figure 2—figure supplement 1F ) . To nominate signaling pathways impacted by Spop , we performed transcriptional profiling using RNA-seq on zebrafish with Spop knockdown and ectopic expression of wild-type SPOP ( SPOP-wt ) and SPOP-F133V , the most commonly mutated residue in prostate cancer ( Figure 2D ) . Consistent with recently reported proteomic data ( Theurillat et al . , 2014 ) and heterodimerization between mutant and wild-type SPOP in our models ( Figure 2—figure supplement 2 ) , transcriptional responses to SPOP-F133V compared with SPOP-wt and Spop morpholino showed a pattern consistent with dominant negative , selective loss of function; SPOP-F133V correlated with SPOP-wt for some gene sets ( Cluster B ) and correlated with Spop morpholino for others ( Cluster A ) ( Figure 2D , Figure 2—source data 1 ) . Gene set enrichment analyses ( GSEA ) revealed gene sets involving DNA repair impacted by modulating SPOP function ( Figure 2—source data 2 ) . Notably , the transcriptional response to F133V correlated highly with BRCA1 inactivation ( Figure 2E ) . While SPOP has been previously proposed as involved in DNA damage and repair ( DDR ) signaling based on in vitro experiments ( Zhang et al . , 2014a ) , these results for the first time implicate a functional role for SPOP in DDR signaling in vivo and suggest this function is selectively impaired by prostate cancer-derived SPOP mutations . To test this hypothesis in human prostate cancers , we performed unsupervised hierarchical clustering of transcriptional data from 11 SPOP mutant and 53 SPOP wild-type tumors , based on the transcriptional signature of BRCA1 inactivation ( MSigDB: M2748 ) . We observed significant segregation of SPOP mutant tumors ( p-value = 9 . 5e−07 , Figure 2F ) using this signature , with less robust segregation of tumors harboring CHD1 or MAP3K7 deletions co-occurring with SPOP mutation ( Figure 2—figure supplement 3 ) . These data suggest that human prostate cancers with SPOP mutations show transcriptional effects similar to BRCA1 inactivation , consistent with a role for SPOP in DSB repair . Supporting this hypothesis , we observed a trend of mutual exclusivity between SPOP mutations and BRCA1 alterations in genomic data ( p-value = 0 . 0496 , Figure 2—figure supplement 4 ) . 10 . 7554/eLife . 09207 . 009Figure 2 . SPOP mediates DNA damage repair . ( A ) Evaluation of SPOP function during zebrafish development . Phenotype of morpholino-mediated Spop knockdown ( MO5 ) in zebrafish embryos at 70 hr post fertilization ( hpf ) . Injection of human SPOP mRNA ( 250 pg ) rescued the phenotype . ( B ) Quantification of the rescue of SPOP phenotype after ectopic expression of human SPOP mRNA . Results are represented as s . e . m . ( C ) Whole mount TUNEL assay to determine apoptosis in zebrafish embryos . Arrows point to apoptotic cells ( brown ) . Shown are representative images . ( D ) Heatmap representation of gene expression differences in zebrafish embryos ectopically expressing SPOP-wt or SPOP-F133V compared to SPOP knockdown by morpholino ( MO ) . The list of genes can be found in Figure 2—figure supplement 3 . Number of genes per block: A ( 198 ) , B ( 429 ) , C ( 223 ) . ( E ) Gene set enrichment analysis ( GSEA ) of RNA sequencing data derived from zebrafish embryos expressing SPOP-wt or SPOP-F133V ( 24 hpf ) . Enrichment plot for the BRCA1 gene signature is shown . Molecular Signatures Database ( MSigDB ) systematic name indicated in brackets . ( NES ) Normalized Enrichment Score . ( FDR ) False Discovery Rate . ( F ) Dendrogram of human primary prostate cancer cases based on BRCA1 knockdown genes ( MSigDB: M2748 ) . Unsupervised clustering of RNA-seq data from human primary prostate cancer with wild-type ( n = 53 ) or mutant SPOP ( n = 11 ) , performed on the BRCA1 knockdown gene signature ( M2748 ) identified in zebrafish embryos by GSEA as shown in ( E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09207 . 00910 . 7554/eLife . 09207 . 010Figure 2—source data 1 . List of genes contained in blocks A , B , and C in the heatmap in Figure 2D . Provided as excel file . DOI: http://dx . doi . org/10 . 7554/eLife . 09207 . 01010 . 7554/eLife . 09207 . 011Figure 2—source data 2 . Results from GSEA comparing zebrafish embryos ectopically expressing SPOP-wt or -F133V . Provided as excel file . DOI: http://dx . doi . org/10 . 7554/eLife . 09207 . 01110 . 7554/eLife . 09207 . 012Figure 2—figure supplement 1 . Spop knockdown in zebrafish using morpholinos results in a developmental phenotype . ( A ) Multiple protein sequence alignment of SPOP from Homo sapiens and zebrafish ( Danio rerio ) reveals 97 . 3% identity between humans and zebrafish on the amino acid level . ( B ) Schematic representation of zebrafish SPOP mRNA . Colored boxes indicate exons . MO5 and MO7 indicate the targeted sites of the SPOP-specific morpholinos . F1 and R5 indicate the sites were the primers bind , which were used to amplify exons 1–5 for molecular validation of morpholino efficiency . ( C ) Image of an agarose gel used to analyze zebrafish SPOP splice variants after morpholino treatment . ( D ) Propidium iodide-based cell cycle analyses of cells from zebrafish embryos ( 24 hr ) treated with or without SPOP-specific morpholino . ( E ) Quantification of TUNEL-positive cells in the midbrain area of zebrafish embryos ( 24 hpf ) treated with SPOP morpholino or in combination with WT-SPOP mRNA as shown in Figure 2C . ( F ) Example of zebrafish embryos ( 48 hpf ) injected with a second splice-blocking morpholino ( MO7 ) . Rescue of the effects of MO7 on zebrafish development by coinjecting WT-SPOP mRNA . ( G ) qPCR-based validation of differentially expressed transcripts identified by RNA-seq . DOI: http://dx . doi . org/10 . 7554/eLife . 09207 . 01210 . 7554/eLife . 09207 . 013Figure 2—figure supplement 2 . SPOP wt and F133V form heterodimers . ( A ) 22Rv1 cells transfected with SPOP wt and F133V constructs tagged with HA or myc underwent immunoprecipitation ( IP ) with anti-HA antibody followed by immunoblot for HA and myc . ( B ) Mouse prostate cells with Cre-inducible SPOP-F133V with a FLAG tag underwent immunoprecipitation ( IP ) with anti-FLAG antibody followed by immunoblot for SPOP . Upper band is flag-tagged mutant SPOP; lower band is endogenous SPOP . DOI: http://dx . doi . org/10 . 7554/eLife . 09207 . 01310 . 7554/eLife . 09207 . 014Figure 2—figure supplement 3 . Dendrogram of human primary prostate cancer cases based on BRCA1 knockdown genes ( MSigDB: M2748 ) . Unsupervised clustering of RNA-seq data from human primary prostate cancer annotated for status of SPOP mutation , CHD1 deletion , and MAP3K7 deletion , performed on the BRCA-1 knockdown gene signature ( M2748 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09207 . 01410 . 7554/eLife . 09207 . 015Figure 2—figure supplement 4 . SPOP mutation and alterations in DNA repair genes in prostate cancer . ( A ) SPOP mutation and somatic BRCA1 gene alterations are mutually exclusive in primary prostate cancer . The cBio portal oncoprint tool was used to graphically summarize the mutations in SPOP and genomic deletions and mutations in BRCA1 in 402 human prostate cancer samples . Samples with no alterations are not displayed . SPOP mutations and BRCA1 alterations are mutually exclusive ( one-tail Fisher's test , p-value = 0 . 0496 ) . ( B ) SPOP mutation and DNA repair pathway alterations in castration-resistant prostate cancer . The cBio portal oncoprint tool was used to graphically summarize the mutations in SPOP and mutations in BRCA2 and ATM in 150 metatstaic castration-resistant prostate cancer samples . DOI: http://dx . doi . org/10 . 7554/eLife . 09207 . 015 Consistent with the hypothesis that SPOP is involved in DSB repair in prostate cells , SPOP forms nuclear foci in prostate cells after γ-irradiation ( Figure 3A ) and stable expression of SPOP-wt conferred resistance to DSB-inducing agents cisplatin and camptothecin ( CPT ) ( Figure 3—figure supplement 1A , B ) . To further define the impact of SPOP mutation in response to DNA damage , we utilized primary prostate cells isolated from transgenic mice with Cre-dependent conditional expression of SPOP-F133V . After transduction with Tamoxifen-inducible Cre ( Cre-ERT2 ) , cells were treated with 4-OH tamoxifen or vehicle and exposed to ionizing radiation ( IR ) ( Figure 3—figure supplement 2 ) . Spop localized in nuclear foci similar to human LNCaP cells in mouse prostate cells ( MPCs ) after IR , with no alteration in foci by expression of SPOP-F133V ( Figure 3A , Figure 3—figure supplement 1I ) . Induction of SPOP-F133V resulted in delayed recovery from IR-induced damage as measured by comet assay ( Figure 3B ) . No differences were seen in apoptosis after IR , as measured by PARP cleavage ( Figure 3—figure supplement 2 ) , subG1 events , or caspase-3 cleavage ( data not shown ) . To functionally characterize how SPOP mutations affect response to DSB , we expressed SPOP-wt and SPOP-F133V in benign prostate epithelial cells ( RWPE ) and prostate cancer cells ( 22Rv1 ) , and examined the induction , recognition , and resolution of CPT-induced DSBs . We found that DSB formation ( measured by γH2AX foci and protein levels ) was not affected by modulating SPOP function with siRNA , or expression of wildtype or mutant SPOP ( Figure 3C , Figure 3—figure supplement 1C , D ) . Furthermore , there were no observed differences in early DNA damage signaling events , such as phosphorylation of ATM , ATR , Chk1 , and Chk2 ( Figure 3—figure supplement 1E–H ) , indicating that SPOP did not affect initial induction and recognition of DSBs or initial steps in DDR signaling . Consistent with this , SPOP showed only limited co-localization with early markers of DSB ( γH2AX , phospho-ATM ) in irradiated prostate cells ( Figure 3—figure supplement 1I ) . 10 . 7554/eLife . 09207 . 016Figure 3 . SPOP mutation impairs HDR and promotes NHEJ and SPOP-wt modulates DSB repair activity similar to BRCA1 . ( A ) SPOP forms nuclear foci after induction of DNA damage by γ-irradiation ( 3GY ) in prostate cells derived from transgenic mice ( MPC ) and human LNCaP cells . Red represents nuclear Spop protein foci . Blue represents nuclear DNA stained with DAPI . ( B ) MPC expressing Cre-inducible SPOP-F133V was infected with tamoxifen-inducible Cre ( CreERT2 ) , and DNA damage was assessed after IR with comet assays . Inset: representative cells showing comet tails after IR . ( C , D , E ) Quantification of γH2AX , 53BP1 , or RAD51 foci in RWPE cells overexpressing WT or F133V mutant SPOP after camptothecin ( CPT ) ( 1 μM ) induced DNA damage . Time indicates the observation interval in minutes including double strand break ( DSB ) induction ( 0–60 min ) and recovery ( 60–180 min ) . Shown are the percentages of cells for each genotype with more than 5 foci per nucleus . Results are represented as s . e . m . ( F ) Representative pictures showing γH2AX , RAD51 , or 53BP1 foci ( red or green ) . Blue represents nuclear DNA stained with DAPI . ( G ) Quantification of Rad51 foci in γ-irradiated ( 2GY ) MPC with tamoxifen-inducible SPOP-F133V . Rad51 foci were counted 30 min post irradiation . ( H ) Representative pictures showing Rad51 foci in mouse prostate epithelial cells before and after γ-irradiation ( 2GY ) . ( I ) Quantification of RAD51 foci in RWPE cells treated with siSPOP or control siRNA and subsequently exposed to CPT ( 1 μM , 1 hr ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09207 . 01610 . 7554/eLife . 09207 . 017Figure 3—figure supplement 1 . SPOP and response to DSB . ( A , B ) Clonogenic assay measuring survival of stable SPOP-expressing 22Rv1 cells after CPT or cisplatin treatment . ( C ) Western blot for the DSB marker γH2AX using protein lysates from RWPE cells before and after CPT treatment ( 30 min ) in RWPE cells either ectopically expressing SPOP-wt or -F133V or ( D ) treated with control ( sicon ) or SPOP targeting siRNA ( siSPOP ) . ( E–H ) Western blot for DNA damage signaling proteins using protein lysates from 22Rv1 ( E , F ) or RWPE ( G , H ) cells before and after CPT treatment ( 30 min ) . Cells either ectopically expressed SPOP-wt or SPOP-F133V or were treated with control ( sicon ) or SPOP-targeting siRNA ( siSPOP ) . ( I ) Immunofluorescence-based partial colocalization of SPOP ( red ) and early DNA damage markers gH2AX or pATM ( green ) . MPC-expressing Cre-inducible SPOP-F133V was transduced with tamoxifen-inducible Cre ( CreERT2 ) . MPC were treated with 4-OH tamoxifen or vehicle and exposed to ionizing radiation ( 3GY ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09207 . 01710 . 7554/eLife . 09207 . 018Figure 3—figure supplement 2 . Prostate cells derived from transgenic mice ( MPC ) expressing Cre-inducible SPOP-F133V were infected with tamoxifen-inducible Cre ( CreERT2 ) and treated with 4-OH tamoxifen ( +Tam ) or vehicle ( −Tam ) and exposed to 15 GY IR followed by Immunoblot for SPOP , PARP , and vinculin ( loading control ) blot after 3 hr recovery . DOI: http://dx . doi . org/10 . 7554/eLife . 09207 . 01810 . 7554/eLife . 09207 . 019Figure 3—figure supplement 3 . SPOP mutation impairs HDR-DSB repair and promotes NHEJ . ( A-F ) Quantification of 53BP1 and RAD51 foci in 22RV1 and RWPE cells ectopically expressing SPOP-wt or SPOP-F133V after camptothecin ( CPT ) ( 1 μM ) or gamma irradiation ( 3GY ) -induced DNA damage . Time indicates the observation interval in minutes including DSB induction ( 0–60 min ) and recovery ( 60–1440 min ) . Shown are the percentages of cells for each genotype with more than 5 foci per nucleus . ( G , H ) Representative pictures of 53BP1 or RAD51 foci ( green ) . Blue represents nuclear DNA stained with DAPI . ( I ) Quantification of 53BP1 foci in RWPE ectopically expressing SPOP-wt , SPOP-F102C or SPOP-F133V mutant SPOP after CPT ( 1 μM , 1 hr ) -induced DNA damage . ( J ) Immunofluorescence staining of MYC-tagged SPOP in RWPE cells confirms nuclear localization . Blue indicates DAPI-stained DNA . Green foci indicate SPOP protein . Error bars represent s . e . m . ( K ) Propidium iodide-based cell cycle analyses of RWPE cells . DOI: http://dx . doi . org/10 . 7554/eLife . 09207 . 019 We next examined specific markers ( 53BP1 , RAD51 ) of the two major DSB repair pathways , HDR and NHEJ . RAD51 is a component of the HDR pathway and a marker for engagement of the HDR machinery ( Baumann et al . , 1996 ) . 53BP1 is a positive regulator of NHEJ that blocks 5′-DNA-end resection and therefore functions at the intersection of HDR and NHEJ; if 53BP1 is not cleared from sites of DSB by HDR components , it promotes error prone NHEJ ( Panier and Boulton , 2014 ) . Strikingly , SPOP-F133V-expressing prostate cells showed delayed clearance of 53BP1 from sites of DSB ( Figure 3D , F , Figure 3—figure supplement 3A ) ; similar effects were seen with another SPOP mutation ( F102C ) commonly observed in prostate cancer ( Figure 3—figure supplement 3I , J ) . Furthermore , SPOP-wt increased RAD51 foci formation compared to controls , while SPOP-F133V-expressing cells showed a dramatic decrease in RAD51 foci formation ( Figure 3E , F , Figure 3—figure supplement 3B ) , consistent with impairment of HDR . Induction of SPOP-F133V in primary MPCs similarly decreased Rad51 foci after IR ( Figure 3G , H ) . Knockdown of SPOP also resulted in a decrease of RAD51 foci , suggesting a selective loss of function of SPOP-F133V in HDR ( Figure 3I ) . We also observed decreased clearance of 53BP1 foci and decreased RAD51 foci formation in SPOP-F133V-expressing cells after gamma irradiation , indicating that this effect is not specific to one mechanism of DSB induction ( Figure 3—figure supplement 3C–H ) . The observed changes in DSB repair were not accompanied by changes in the cell cycle distribution of cells expressing SPOP-wt or F133V under these conditions ( Figure 3—figure supplement 3K ) . We next investigated the role of SPOP in DSB repair using the well established DR-GFP and Pem1-Ad2-EGFP reporter assays as functional readouts for HDR and NHEJ , respectively ( Figure 4A , D ) ( Pierce et al . , 1999; Seluanov et al . , 2004 ) . In the DR-GFP assay , knockdown of SPOP by siRNA decreased HDR competence in human epithelial cells to a similar level of BRCA1 knockdown ( Figure 4B ) . Conversely , ectopically expressed SPOP-wt increased the HDR competence in these cells , with partial loss of this function by mutant SPOP ( Figure 4C ) . In contrast , the Pem1-Ad2-EGFP NHEJ reporter assay indicated an increase of NHEJ activity in SPOP-F133V expressing epithelial cells , while both SPOP siRNA and BRCA1 siRNA increased NHEJ ( Figure 4E , F ) . Taken together , these results suggest that SPOP promotes HDR , while somatic mutation in SPOP , as observed in prostate cancer with increased genomic rearrangements , impairs HDR and promotes error-prone NHEJ . 10 . 7554/eLife . 09207 . 020Figure 4 . SPOP modulates DSB repair activity similar to BRCA1 . ( A ) Schematic overview of the DR-GFP assay used to measure homology-directed repair ( HDR ) activity . ( B , C ) Analysis of the HDR-activity in HEK 293 cells with siRNA knockdown of SPOP or BRCA1 and ectopically expressing SPOP-wt or SPOP-F133V . ( D ) Schematic overview of the Pem1-Ad2-EGFP assay used to measure NHEJ-activity . ( E , F ) Analysis of the NHEJ-activity in HEK 293 cells with siRNA knockdown of SPOP or BRCA1 and ectopically expressing SPOP-wt or SPOP-F133V . All results are represented as s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 09207 . 020 Human cancers with underlying defects in HDR ( such as BRCA1 inactivated breast and ovarian cancers ) ( Bryant et al . , 2005; Farmer et al . , 2005; Audeh et al . , 2010; Tutt et al . , 2010 ) show sensitivity to poly ( ADP-ribose ) polymerase 1 ( PARP ) inhibition . To test if SPOP inactivation conferred sensitivity to PARP inhibition after DSB induction , we utilized siRNA targeting SPOP in human prostate cancer cell lines ( PC-3 , LnCap , 22Rv1 ) , followed by irradiation ( 5GY ) and incubation with the PARP inhibitors olaparib or veliparib . Reduction of SPOP expression increased the sensitivity of these prostate cancer cells to both PARP inhibitors ( Figure 5A–C , Figure 5—figure supplement 1B , C ) . To test if SPOP mutation also sensitizes to PARP inhibition , we treated MPCs ( control vs tamoxifen-induced SPOP-F133V ) with olaparib followed by IR—expression of SPOP-F133V increased sensitivity to olaparib similar to loss of SPOP in human prostate cancer cell lines ( Figure 5D ) . This was further confirmed in 22Rv1 cells ectopically expressing SPOP-F133V , which also showed an increased sensitivity to olaparib in viability assays , while cells ectopically expressing SPOP-wt were relatively resistant ( Figure 5—figure supplement 1A ) . We further confirmed these results in HEK293 cells stably overexpressing the two common SPOP mutants ( Y87N , F133V ) in clonogenic survival assays ( Figure 5E , F ) . 10 . 7554/eLife . 09207 . 021Figure 5 . SPOP mutation sensitizes cells to therapeutic PARP inhibition . ( A–D ) Analysis of sensitivity to the PARP inhibitor olaparib in irradiated ( 5GY ) prostate cancer ( PC-3 , LNCaP , 22Rv1 ) and mouse prostate epithelial cells ( MPC ) after SPOP knockdown ( siSPOP ) or tamoxifen-inducible SPOPF133V expression . BRCA1 knockdown ( siBRCA1 ) and non-targeting siRNA ( sictrl ) served as positive or negative control . The IC50 of olaparib for each genotype is indicated in Molar ( M ) . ( E ) Analysis of the sensitivity of most frequently occurring prostate-specific SPOP mutants Y87N and F133V to olaparib in clonogenic assays using HEK293 cells . ( F ) Representative examples from the clonogenic assay used to assess long-term survival in HEK293 cells stably expressing SPOP-wt or SPOP mutants . ( G ) Impact of SPOP mutation on induction of genomic instability in 22Rv1 cells after olaparib ( 1 μM ) treatment as measured by comet assay . Increased genomic instability was measured by an increase in the tail moment . ( H ) Genomic instability in tamoxifen-inducible SPOP-F133V expressing mouse prostate cells after γ-irradiation ( 15GY ) with and without olaparib ( 1 μM ) treatment as measured by comet assay . All results are represented as s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 09207 . 02110 . 7554/eLife . 09207 . 022Figure 5—figure supplement 1 . SPOP mutation and loss sensitize prostate cancer cells to therapeutic PARP inhibition . ( A ) In vitro drug sensitivity assay of 22Rv1 cells expressing SPOP-wt or SPOP-F133V . Cells were treated with the indicated concentrations of olaparib for 96 hr . LacZ was used as empty control expression vector . ( B , C ) In vitro drug sensitivity assay of 22Rv1 and PC-3 cells treated with control siRNA ( sictrl ) , SPOP , or BRCA1 targeting siRNA ( siSPOP , siBRCA1 ) . After irradiation ( 5GY ) cells were treated with the PARP-1 inhibitor veliparib for 96 hr . IC50 values for each genotype are indicated in Molar ( M ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09207 . 022 To determine if the altered sensitivity to PARP inhibitors was associated with increased DNA damage , consistent with impaired double-strand break repair , we performed single-cell gel electrophoresis ( comet assay ) in prostate cells treated with olaparib . The comet assay revealed an increase of genomic instability in SPOP-F133V RWPE cells after PARP inhibition , similar to SPOP ( and BRCA1 ) knockdown with siRNA ( Figure 5G ) . Similarly , primary MPCs expressing SPOP-F133V showed increased damage after olaparib treatment ( Figure 5H ) . Taken together , these data argue that SPOP mutation confers sensitivity to PARP inhibition due to impaired error-free HDR DSB repair and an increase in error-prone NHEJ . In summary , here we report that SPOP , the substrate recognition component of an E3 ubiquitin ligase , is a regulator of the HDR-based DSB repair machinery . Using functional genetic approaches , we find that prostate-specific SPOP mutants deregulate DSB repair by promoting the error-prone NHEJ pathway ( Figure 6 ) . Importantly , after DSB induction , this loss of function leads to an increased sensitivity of mutant SPOP prostate cancer cells to PARP inhibition . These observations provide the first mechanism for the increased genomic instability of SPOP mutant prostate cancer , but also suggest that similar to other cancers with impaired HDR , this distinct class of cancer might benefit from treatment with clinically established DNA damaging therapeutics . This study therefore provides a rationale for hypothesis-based biomarker-driven clinical trials using PARP inhibitors or other DNA damaging agents in patients with prostate cancer . 10 . 7554/eLife . 09207 . 023Figure 6 . Proposed model of the effects of SPOP mutation on genome instability . In prostate epithelial cells , SPOP-wt promotes error-free HR and maintains genome stability . SPOP mutation impairs HR repair and promotes error-prone NHEJ , leading to increased genomic instability . DOI: http://dx . doi . org/10 . 7554/eLife . 09207 . 023 Genomic rearrangements represent critical events deregulating prostate cancer genomes and driving tumorigenesis , although among individual cancer samples , there is marked variability in number and character of structural rearrangements . Here , we identified a genomically unstable subclass of primary prostate cancer characterized by dramatically increased intrachromosomal rearrangements . This subset of clinically localized , treatment-naive prostate cancers display a degree of genomic instability previously thought only to occur in late stage , metastatic tumors . Importantly , increased rearrangements likely result in higher total SCNA burden , which is associated with more aggressive clinical behavior ( Hieronymus et al . , 2014; Lalonde et al . , 2014 ) . Increased rearrangements were associated with mutations in SPOP , encoding the substrate-recognition component of an E3 ubiquitin ligase complex . SPOP mutations are the most common point mutations in prostate cancer , but their role in promoting prostate cancer pathogenesis remains unclear . SPOP mutations occur early in the history of prostate cancer , based on clonality analysis reported here and presence in the prostate cancer precursor high grade prostatic intraepithelial neoplasia ( HG-PIN ) ( Barbieri et al . , 2012 ) , potentially consistent with a ‘gatekeeper’ role in genome maintenance . Consistent with increased intrachromosomal rearrangements in SPOP mutant cancers , in vivo data nominated a functional role for SPOP in DSB repair , similar to BRCA1 . Interestingly , SPOP loss of function induced a developmental phenotype of neuronal degeneration and apoptosis; similar effects have been reported with other genes modulating DNA repair , including BRCA1 ( Pulvers and Huttner , 2009 ) . SPOP encodes the substrate recognition component of an E3 ubiquitin ligase , and here we identified a role for SPOP in regulating the HDR-based DSB repair machinery . Notably , many established components of the DNA damage response are components of enzymes regulating ubiquitylation ( Ceol et al . , 2011 ) . A number of substrates have been reported as deregulated by prostate cancer-derived SPOP mutations , but the relevance of these in vitro findings to human prostate cancers is still unclear ( Geng et al . , 2013 , 2014; An et al . , 2014; Theurillat et al . , 2014; Zhang et al . , 2014b ) . In contrast , we focused on the phenotypes observed in human prostate cancers harboring SPOP mutations . Using functional genetic approaches , we find that prostate-specific SPOP mutants deregulate DSB repair by promoting the error-prone NHEJ pathway ( Figure 4F ) . The clinical importance of our findings is timely as potent PARP inhibitors , inducing synthetic lethality in cancers with alterations in DSB repair , are being evaluated and utilized in human cancers ( Fong et al . , 2009; Tutt et al . , 2010; Schiewer and Knudsen , 2014 ) . Here , we define a novel biomarker for increased genomic instability in clinically localized prostate cancer , provide the first mechanism by which SPOP mutations induce these alterations , and also suggest that similar to other cancers with impaired HDR , this distinct class of cancer may benefit from treatment with clinically established DNA damaging therapeutics , providing a rationale for genotype-based clinical trials . Cells were purchased from ATCC and cultured according to the manufacturer's instructions . All cells were maintained with 1% penicillin/streptomycin ( 15140-122 , Gibco , Grand Island , NY , United States ) . Cells were treated with camptothecin ( C9911 , Sigma-Aldrich , St . Louis , MO , United States ) , olaparib ( AstraZeneca , United Kingdom ) , veliparib ( Selleckchem , United Kingdom ) , or cisplatin ( Sigma-Aldrich ) at the indicated concentrations . For SPOP expression , lentiviral vectors ( pLenti6/V5-Dest Gateway vector , V496-10 , Invitrogen , Grand Island , NY , United States ) coding for mutant or wild-type SPOP were used . Viruses were titrated to equalize infectivity and protein expression . Viral titers were measured by a p24 ELISA ( Lenti-X p24 rapid titer kit , PT5002-2 , Clontech , Mountain View , CA , United States ) . Viral transduction was performed by using Polybrene transfection reagent ( TR-1003-G , Millipore , Billerica , MA , United States ) at a final concentration of 8 ng/μl . Stable cell lines for WT or mutant SPOP were generated by pLenti viral transduction and subsequent selection by cell sorting for RFP using an Aria-II cell sorter ( BD Biosciences , San Jose , CA , United States ) . Stable selection was monitored over time by confirming RFP- and MYC-tagged SPOP expression . Reagents for siRNA knockdown of SPOP and BRCA1 were purchased from Dharmacon ( Lafayette , CO , United States ) . The siGenome smartpools were used to target SPOP ( M-017919-02-0005 ) , BRCA1 ( M-003461-02-0005 ) as well as non-targeting control ( D-001206-13-05 ) . The siRNA oligonucleotides were transfected with Lipofectamine 2000 at a final concentration of 50–100 pmol and incubated for 48 hr . Human prostate and mouse prostate–epithelial cells ( 1000–5000 ) were seeded into 96-well microtiter plates ( CLS3610-48EA , Sigma Aldrich ) . After irradiation ( 5GY ) cells were treated with olaparib or veliparib in a concentration range of 0 . 0001 μM–100 μM for 72–96 hr . DMSO was used as control . Cell viability was measured by using the CyQuant NF cell proliferation assay ( Life Technologies , C35006 ) according to the manufacturer's instructions . Data normalization and nonlinear regression to calculate IC50 values was performed with Graph Pad Prism . Genomic analysis: SCNA profiles for 402 prostate cancers were obtained through processing of high-density oligonucleotide array data upon signal segmentation and then combined for genome-wide analysis for SPOP mutant and SPOP wild-type tumors . Clonality analysis for selected lesions was performed from sequencing data as in Baca et al . ( 2013 ) . Differential Gene Expression ( DGE ) and Gene Set Enrichment Analysis: After sorting and indexing , the aligned files were used with bedtools ( v2 . 16 . 1 ) and Ensembl Zebrafish Transcriptome ( v70 ) to generate read counts for genes for all samples . These read counts were used with DESeq ( 2_1 . 2 . 10 ) for differential expression . Orthology was also added to the differential expression table in this step . Read counts were also counts per million ( cpm ) normalized using EdgeR ( v3 . 4 . 2 ) package and then pre-ranked using Log2 Fold Change . GSEA ( v2-2 . 0 . 13 ) was run in pre-ranked mode to identify enriched signatures . Primers used for validation of the expression of selected genes are shown in Supplementary file 1 . Clinically localized prostate cancers were selected for transcriptome sequencing as described ( Barbieri et al . , 2012 ) . All samples were collected with informed consent of the patients and prior approval of the institutional review boards ( IRB ) of respective institutions . Additionally , the sequencing and data release of all transcriptome-sequenced samples was reviewed and approved by local IRB . All RNA-seq data are deposited in the NCBI sequencing read archive ( SRA ) under the accession number SRP063952 .

Prostate cancer is the most common type of cancer in men in the UK and USA . Cancers develop when cells in the body acquire genetic mutations that allow the cells to grow rapidly and form a mass known as a tumor . Prostate cancer cells from different individuals can carry different genetic mutations , which affects whether the disease progresses and how the tumors respond to medical treatments . This genetic variety arises in cancer cells partly from a phenomenon known as genomic instability , in which DNA mutations accumulate due to defects in DNA repair . Genetic studies of biopsies taken from human prostate cancers have shown that genomic instability causes chromosomes—the structures in which the cell's DNA is organized—to break and then be stuck back together haphazardly . As a result , fragments of chromosomes can end up in the wrong position , be duplicated , or be lost altogether . All of these mutations could spur on the growth of the tumor . However , it is currently not clear why some prostate cancers are more genomically unstable than others , or what exactly causes this instability . Boysen , Barbieri et al . studied prostate cancer cells taken from patients before they started medical treatment . The experiments show that the cancer cells with high levels of genomic instability also often had mutations in a gene that encodes a protein called SPOP . These mutations occur in about 10 percent of men with prostate cancer and appear early in the development of the tumors . Next , they studied the SPOP protein in zebrafish ( which is nearly identical to human SPOP ) , as well as in mouse and human cells . The experiments show that SPOP normally helps the cell to accurately repair DNA that has been damaged . Mutations in SPOP change the DNA repair process , which lead to genomic instability by increasing the likelihood that broken chromosomes will be stuck back together incorrectly . Further experiments tested drugs known as PARP inhibitors on mouse and human prostate cancer cells . The drugs , which have been recently tested successfully in patients with prostate cancer , block a different method of DNA repair that operates separately to the one that involves SPOP . When both of these pathways were inactivated—one by the SPOP mutation , the other by the drug—the cancer cells died more quickly . Therefore , men that are diagnosed with types of prostate cancer in which the gene that encodes SPOP is mutated might benefit from treatment with PARP inhibitors or other therapies that affect DNA repair .

2015

SPOP mutation leads to genomic instability in prostate cancer

Metazoan genes are embedded in a rich milieu of regulatory information that often includes multiple enhancers possessing overlapping activities . In this study , we employ quantitative live imaging methods to assess the function of pairs of primary and shadow enhancers in the regulation of key patterning genes-knirps , hunchback , and snail-in developing Drosophila embryos . The knirps enhancers exhibit additive , sometimes even super-additive activities , consistent with classical gene fusion studies . In contrast , the hunchback enhancers function sub-additively in anterior regions containing saturating levels of the Bicoid activator , but function additively in regions where there are diminishing levels of the Bicoid gradient . Strikingly sub-additive behavior is also observed for snail , whereby removal of the proximal enhancer causes a significant increase in gene expression . Quantitative modeling of enhancer–promoter interactions suggests that weakly active enhancers function additively while strong enhancers behave sub-additively due to competition with the target promoter . There is emerging evidence that metazoan genes occur in a complex regulatory landscape encompassing numerous enhancers ( Jeong et al . , 2006; Hong et al . , 2008; Frankel et al . , 2010; Perry et al . , 2011; Rada-Iglesias et al . , 2011; Buecker and Wysocka , 2012; Lagha et al . , 2012; Levine et al . , 2014 ) . For example , the mouse Sonic Hedgehog gene is regulated by at least 20 different enhancers scattered over a distance of ∼1 Mb ( Jeong et al . , 2006 ) . Individual enhancers mediate expression in a variety of different tissues , including the brain , floorplate , and limb buds . Multiple enhancers with overlapping regulatory activities are also used to control gene expression within individual cell types . For example , the transcriptional activation of the gap gene hunchback ( hb ) in the early Drosophila embryo is mediated by both a proximal enhancer and distal ‘shadow’ enhancer that independently mediate activation in response to high levels of the Bicoid activator gradient ( Buecker and Wysocka , 2012 ) . Despite overwhelming evidence for multiple enhancers regulating the same gene it is unknown whether they simultaneously interact with the same promoter in a given cell . Here , we use a combination of quantitative live imaging and theoretical modeling to investigate the function of multiple enhancers for the regulation of a common target gene within a single cell type . The fate map of the adult fly is established by ∼1000 enhancers that regulate several hundred patterning genes during the 1-hr interval between two and three hrs after fertilization ( Levine , 2010; Nien et al . , 2011 ) . As many as half of these genes contain ‘shadow’ enhancers with overlapping spatiotemporal activities that are thought to improve the precision and reliability of gene expression ( Hong et al . , 2008; Frankel et al . , 2010; Perry et al . , 2010 , 2011; Lagha et al . , 2012; Miller et al . , 2014 ) . For example , the hb shadow enhancer helps produce a sharp boundary of activation by the Bicoid gradient , while its snail ( sna ) counterpart helps ensure reliable activation under stressful conditions such as high temperatures ( Perry et al . , 2010 , 2011 ) . There is emerging evidence that shadow enhancers are used pervasively in a variety of developmental processes , in both invertebrates and vertebrates ( Lagha et al . , 2012; Arnold et al . , 2013; Miller et al . , 2014; Lam et al . , 2015 ) . The underlying mechanisms by which two enhancers with extensively overlapping regulatory activities produce coordinated patterns of gene expression are uncertain . It is possible that they augment the levels of gene expression above the minimal thresholds required to execute appropriate cellular processes ( Gregor et al . , 2014 ) . However , there is currently only limited experimental evidence for enhancers acting in an additive fashion ( Arnold et al . , 2013; Lam et al . , 2015 ) . An alternative view is that shadow enhancers suppress transcriptional noise and help foster uniform expression among the different cells of a population ( Buecker and Wysocka , 2012 ) . To explore these and other potential mechanisms , we examined the timing and levels of gene activity using bacterial artificial chromosomes ( BAC ) transgenes containing individual enhancers and combinations of primary and shadow enhancers in the early Drosophila embryo . BAC transgenes containing three key patterning genes , hb , knirps ( kni ) , and sna , were examined in living precellular embryos . Quantitative analyses suggest that shadow enhancers mediate different mechanisms of transcriptional activity . For kni , we observe additive , even super-additive , activities of the primary and shadow enhancer pairs . In contrast , the hb enhancers function sub-additively in anterior regions containing saturating levels of the Bicoid activator but function additively in regions where there are diminishing levels of the Bicoid gradient . Strikingly , sub-additive behavior is also observed for sna , in that removal of the proximal enhancer causes a significant increase in gene expression . These observations suggest that the levels of enhancer activity determine the switch between additive and non-additive behaviors . Using theoretical modeling , we suggest that these behaviors can be understood in the context of enhancers competing or cooperating for access to the promoter . Weak enhancers work additively due to infrequent interactions with the target promoter , whereas strong enhancers are more likely to impede one another due to frequent associations . Our results highlight the potential of combining quantitative live imaging and modeling in order to dissect the molecular mechanisms responsible for the precision of gene control in development ( Gregor et al . , 2014 ) and provide a preview into the complex function of multiple enhancers interacting with the same promoter . Previous live-imaging studies have relied on simple gene fusions containing a single enhancer attached to a reporter gene with MS2 stem loops inserted in either the 5′ or 3′ UTR ( Garcia et al . , 2013; Lucas et al . , 2013; Bothma et al . , 2014 ) . Detection depends on the binding of a maternal mRNA-binding fusion protein ( MCP::GFP ) expressed throughout the early embryo . In order to examine the interplay between multiple enhancers , we created a series of BAC transgenes containing complete regulatory landscapes ( summarized in Figure 1 ) . The BAC transgenes contain an MS2-yellow reporter gene in place of the endogenous transcription units ( Figure 1A ) . For each locus , hb , kni , and sna , we examined a series of three BAC transgenes: containing both primary and shadow enhancers , as well as derivatives lacking individual enhancers ( Figure 1B , C ) . As expected , the BAC transgenes containing both enhancers produce robust expression of the MS2 reporter gene which recapitulate endogenous patterns previously measured using mRNA FISH and immunostaining ( Perry et al . , 2010 , 2011 ) ( Figure 1D–I , Videos 1–3 ) . 10 . 7554/eLife . 07956 . 003Figure 1 . Live-imaging of transcriptional activity of hb and kni loci lacking different enhancers . ( A ) General structure of the reporter constructs . A reporter construct with 24 repeats of the MS2 stem loops and the yellow gene was recombined into BACs spanning the hb and kni loci . The 5′ UTR and 3′ UTR of the endogenous genes were left intact . The MCP::GFP protein that binds to the MS2 stem loops is present in the unfertilized egg and in the early embryo . Gene models of ( B ) the hb and ( C ) kni loci showing the location of the primary and shadow enhancers ( Perry et al . , 2011 ) . ( D , F , H ) Snapshots of Drosophila embryos expressing different versions of the hb BAC>MS2 reporter containing different combinations of the two enhancers 10 min into nuclear cleavage cycle 13 ( nc13 ) . The colored bar on the bottom right indicates which enhancer was removed . ( E , G , I ) Snapshots of Drosophila embryos expressing different versions of the kni BAC>MS2 reporter containing different combinations of the two enhancers in nc14 . DOI: http://dx . doi . org/10 . 7554/eLife . 07956 . 00310 . 7554/eLife . 07956 . 004Figure 1—figure supplement 1 . kni BAC expression lacking both shadow and primary enhancers . Fluorescent in situ hybridization of endogenous kni and kni BAC>yellow transgenes . ( A ) Shows an embryo with the fully intact kni BAC>yellow transgene in late nc 14 . ( B , C ) Show embryos with the kni BAC>yellow transgene lacking both primary and shadow enhancers , removing both enhancers abolishes all activity in the stripe domain . In ( A ) an embryo is in late nc14 and ( B ) shows and embryo in early nc 14 . DOI: http://dx . doi . org/10 . 7554/eLife . 07956 . 00410 . 7554/eLife . 07956 . 005Video 1 . Dynamics of hunchback expression . Maximum projection of hb BAC>MS2 transgene from nc10 to gastrulation , MCP::GFP in green and histone in red , anterior to the left and ventral view up . Time elapsed since the start of imaging is indicated in top left . The initial pattern is restricted to the anterior where expression is driven by the primary and shadow enhancers . In late nc13 the central domain enhancer starts to be expressed . DOI: http://dx . doi . org/10 . 7554/eLife . 07956 . 00510 . 7554/eLife . 07956 . 006Video 2 . Dynamics of knirps expression . Maximum projection of kni BAC>MS2 transgene from nc10 to gastrulation , MCP::GFP in green and histone in red , anterior to the left and ventral view down . Time elapsed since the start of imaging is indicated in top left . The dynamics of the anterior and central parts of the pattern are evident . DOI: http://dx . doi . org/10 . 7554/eLife . 07956 . 00610 . 7554/eLife . 07956 . 007Video 3 . Dynamics of snail expression . Maximum projection of snail BAC>MS2 transgene from nc10 to gastrulation , MCP::GFP in green and histone in red , anterior to the left and ventral view up . Time elapsed since the start of imaging is indicated in top left . DOI: http://dx . doi . org/10 . 7554/eLife . 07956 . 007 Enhancer ‘deletions’ were created by substituting native sequences with neutral sequences of similar sizes ( see ‘Materials and methods’ ) . These substitutions remove most of the critical sequences identified by ChIP-Seq assays ( Harrison et al . , 2011; Nien et al . , 2011; Perry et al . , 2011 ) . It is nonetheless possible that critical flanking sequences persist within the transgenes . However , removal of both kni enhancers eliminates detectable transcripts in abdominal regions of early embryos ( Figure 1—figure supplement 1 ) , suggesting that any remaining flanking sequences are insufficient to mediate expression . Qualitative inspection of the hb and kni expression videos suggests that removal of either the primary or shadow enhancer does not cause a dramatic alteration in the overall patterns of gene activity . In order to identify more nuanced changes , we quantified the transcriptional activities of the complete series of BAC transgenes ( Figure 2 ) . The fluorescence intensities of active transcription foci were measured during nuclear cleavage cycles ( nc ) 13 and 14 at different positions across the anterior–posterior ( AP ) axis . These intensities were converted into an absolute number of elongating Pol II molecules by calibrating with internal standards ( see Garcia et al . , 2013 ) . Several embryos were analyzed for each time point , and the data were merged to determine the average behavior as a function of AP position and time . 10 . 7554/eLife . 07956 . 008Figure 2 . Combined effect of multiple enhancers as a function of AP position . ( A , B ) Mean number of Pol II molecules transcribing per nucleus ( NPol II ) in the hb BAC reporters containing different combinations of enhancers as a function of AP position for two time points in nc13 . NPol II is calculated by averaging data from at least three embryos at each AP position . The predicted sum of the individual enhancers is also shown . Note the additivity at the boundary vs the sub-additivity at the core , anterior domain of the pattern . ( C , D ) Mean number of Pol II molecules transcribing per nucleus ( NPol II ) in the kni BAC reporters in nc14 as a function of AP position . For kni , we see super-additive behavior in the beginning of nc14 which then becomes additive later in nc14 . The absolute number of transcribing Pol II molecules was estimated following a previous calibration ( Garcia et al . , 2013 ) . Error bars are the standard error of the mean over multiple embryos . DOI: http://dx . doi . org/10 . 7554/eLife . 07956 . 008 Hb expression was examined during the ∼15 min interphase of nc 13 when both the primary and shadow enhancers are active , but before the onset of later-acting ‘stripe’ enhancers during nc 14 ( Perry et al . , 2011 , 2012 ) . We measured the transcriptional activity of all three hb BAC transgenes ( Figure 1D , F , H ) . Contrary to simple expectations suggested by previous studies ( Arnold et al . , 2013 ) , we find that the two hb enhancers do not function in an additive fashion in anterior regions ( 20–40% egg length , EL ) of the embryo ( e . g . , Figure 2A , B ) . Indeed , the levels produced by the wild-type transgene fall far short of the additive levels predicted by simply summing the levels of expression produced by the transgenes containing either the shadow or primary enhancer alone ( Figure 2A , B ) . Moreover , the removal of the shadow enhancer has no effect on the levels of transcription in anterior regions , which is consistent with the original conception of the shadow enhancer as a ‘back-up’ in the event of stress ( Hong et al . , 2008 ) . A very different scenario is observed in central regions of the embryo ( 40–50% EL ) where hb expression switches from ‘on’ to ‘off’ to form a sharp border ( Gregor et al . , 2007 ) . In this region , the wild-type transgene produces significantly higher levels of expression than either of the transgenes driven by a single enhancer . In fact , these levels correspond to the values predicted by simply adding the activities of the single-enhancer transgenes . Thus , the two enhancers transition from sub-additive to additive behavior in the region of the embryo where there are diminishing levels of the Bicoid activator gradient . We therefore suggest that the hb enhancers function additively only when they are operating below peak capacity ( see below ) . To further explore the activities of multiple enhancers , we examined the kni gene , which is regulated by an intronic enhancer and a distal 5′ enhancer ( Perry et al . , 2011 ) . We focus our analysis on central regions of the abdominal expression pattern since previous studies suggest the occurrence of long-range repressive interactions that establish the borders of the ‘stripe’ ( Perry et al . , 2011 ) . During early periods of nc 14 , the two enhancers function super-additively ( Figure 2C ) . That is , the wild-type kni BAC transgene produces higher levels of expression than the predicted sum of the two transgenes containing either enhancer alone . During later stages of development , there is a twofold reduction in the expression levels of the endogenous gene , and at this time the two enhancers work in a simple additive manner ( Figure 2D ) . Note that the maximum number of elongating polymerase ( Pol II ) complexes falls short of that seen for hb ( compare with Figure 2A , B ) . Understanding the stark difference in the behaviors of the hb and kni enhancer pairs necessitates measuring the absolute strengths of the different enhancers . Using absolute counts of mRNA molecules ( Little et al . , 2013 ) , we calibrated our live fluorescence intensity traces to determine the average numbers of actively elongating Pol II transcription complexes ( Garcia et al . , 2013 ) . The kni transgenes containing single enhancers exhibit as little as fourfold lower levels of expression as compared with the corresponding hb transgenes ( Figure 3A , B ) . At peak activity , the proximal hb enhancer induces ∼50 transcribing Pol II complexes across the yellow reporter gene . By contrast , individual kni enhancers produce an average of only ∼15 elongating Pol II complexes . We propose that the additive and super-additive behaviors of the two kni enhancers reflect their inherently ‘weaker’ activities as compared with the ‘stronger’ proximal hb enhancer ( see ‘Discussion’ ) . Note that , despite these differences , the overall output of transcripts and the overall rate of transcript production are essentially identical for all gap genes ( Little et al . , 2013 ) . 10 . 7554/eLife . 07956 . 009Figure 3 . Combined effect of multiple enhancers as a function of time . ( A ) Time course of the mean number of Pol II molecules transcribing per nucleus ( NPol II ) for the different hb BAC transgenes and sum of individual enhancers at 27% EL for the duration of nc13 . ( B ) kni BAC transgenes activities and the sum of individual enhancer activity at 60% EL for the first 50 min of nc14 . ( C ) sna BAC transgenes and the sum of individual enhancer activities averaged over the central mesoderm for the initial 50 min of nc14 . Error bars are the standard error of the mean over multiple embryos . DOI: http://dx . doi . org/10 . 7554/eLife . 07956 . 009 To test the proposed anti-correlation between enhancer strength and additivity , we analyzed the expression of sna , which is essential for delineating the invaginating mesoderm during gastrulation . sna transgenes containing either the proximal or distal enhancer produce peak transcriptional activities of ∼40 actively transcribing Pol II complexes across the yellow reporter gene , similar to the numbers seen for the proximal hb enhancer ( Figure 3C ) . Thus , both sna enhancers are strong and they exhibit strikingly sub-additive behaviors . In particular , the wild-type transgene displays significantly lower levels of expression than the mutant transgene containing only the shadow enhancer . Thus , strong enhancers not only fail to function additively but also interfere with one another , leading to sub-additive expression levels . This observation is also consistent with an earlier study , which suggested that the weaker proximal enhancer attenuates the activities of the stronger distal shadow enhancer ( Dunipace et al . , 2011 ) . In an effort to understand how multiple enhancers might function additively or sub-additively , we developed a mathematical model for dynamic enhancer–promoter interactions . In this model , a single enhancer interacts with its promoter via a forward rate kon and a backward rate koff ( Figure 4A ) . The relative values of the forward and reverse rates determine the strength of the enhancer–promoter interaction by controlling what fraction of time the two are bound . When the enhancer and promoter interact , the promoter is in the ON state and initiates transcription at a rate r . This rate can be interpreted as the efficiency of enhancer-mediated transcriptional initiation upon enhancer–promoter interaction . Hence , the observable rate of mRNA production depends on the interaction strength given by the ratio kon/koff , and the efficiency r with which transcription is initiated upon interaction ( Figure 4B ) . This scheme can be generalized to include two enhancers ( A and B ) interacting with the same promoter ( Figure 4C and see ‘Materials and methods’ for details of the mathematical analysis ) . In this model only one enhancer can interact with the promoter at a given time . 10 . 7554/eLife . 07956 . 010Figure 4 . Model of enhancer–promoter interactions and its predictions for mRNA production . ( A ) Minimal model of one enhancer engaging a promoter . kon and koff are the rates of promoter engagement and disengagement , respectively , and determine the interaction strength . r is the rate of mRNA production when the promoter is engaged and is a measure of the transcriptional efficiency . The mean number of Pol II molecules transcribing per nucleus ( NPol II ) is proportional to the rate of mRNA production . ( B ) As the interaction strength of a single enhancer is increased , the amount of mRNA produced increases up to a maximum value dictated by the transcriptional efficiency . ( C ) The model in ( A ) can be generalized to allow for multiple enhancers interacting with the same promoter . ( D ) In the regime where the interaction strength of both promoters is weak ( kon/koff = 0 . 01 ) , the amount of mRNA produced by having both A and B is simply the sum of the individual contributions of A and B , ( r = 1 ) . ( E ) In the regime where the interaction strength is large , the combined activity of both enhancers can be significantly less than the sum the individual enhancers . A less efficient enhancer A ( rA = 0 . 2 au ) can interfere with the more efficient enhancer B ( rB = 1 au ) such that their combined activity is significantly less than the sum of the activities of individual enhancers . DOI: http://dx . doi . org/10 . 7554/eLife . 07956 . 010 When the individual enhancers interact infrequently with the promoter ( kon/koff << 1 ) they are unlikely to attempt to engage the promoter simultaneously . In this regime , the enhancers will work additively , and the rate of mRNA production of enhancers A and B will simply be the sum of the production rate of A and B alone as shown in Figure 4D . However , as the strength of promoter–enhancer interactions increases , the combined activity of both enhancers is less than the sum of each individual enhancer . When we model enhancers of different strengths ( Figure 4E ) , the amount of mRNA production is reduced as the enhancer with the weaker transcriptional efficiency interacts more frequently with the promoter . This occurs because the two enhancers compete for access to the promoter , effectively inhibiting one another . Thus , weak enhancers might work additively due to infrequent interactions with the target promoter , whereas strong enhancers interfere with one another due to more frequent interactions ( see below ) . Our quantitative analysis of hb and kni expression provides seemingly opposing results . For kni , we observe additive , sometimes even super-additive , action of the two enhancers within the presumptive abdomen . In contrast , the two hb enhancers do not function in an additive fashion in anterior regions but are additive only in central regions where expression abruptly switches from ‘on’ to ‘off’ . We propose that ‘weak’ enhancers function additively or even super-additively , whereas ‘strong’ enhancers can impede one another ( Figure 5 ) . 10 . 7554/eLife . 07956 . 011Figure 5 . Theoretical expectation and experimental results showing different regimes of combined enhancer action . ( Upper left ) Theoretical predictions ( yellow ) illustrating how the rate of mRNA production from both enhancers , NPol IIPrimary+Shadow , varies with the sum of the activity of the individual enhancers , NPol IIPrimary+NPol IIShadow ( yellow ) . Mean number of Pol II molecules transcribing per nucleus ( NPol II ) is proportional to the rate of mRNA production . The green line shows perfect additivity for comparison . The model predicts additive behavior ( NPol IIPrimary+Shadow≈NPol IIPrimary+NPol IIShadow ) when the rate of production is low and sub-additive behavior ( NPol IIPrimary+Shadow<NPol IIPrimary+NPol IIShadow ) as the production rate increases . As the interaction strength of individual enhancers increases so does the rate of mRNA production , but the combined activity of both enhancers becomes sub-additive . ( Upper right , lower left , lower right ) Transcriptional activity of intact loci vs the sum of activities of individual enhancers for hb , kni , and sna at different times . A green line has been drawn in to indicate where NPol IIPrimary+Shadow is equal to NPol IIPrimary+NPol IIShadow . For hb and kni , the plots show data taken at different AP positions at 10 min into nc 13 and 20 min into nc 14 , respectively , while for sna the datapoints were at different times . Ellipses indicate standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 07956 . 011 Additional support for this view is provided by the analysis of sna . We found that the removal of the proximal enhancer significantly augments expression , consistent with the occurrence of enhancer interference within the native locus . It is also conceivable that a single strong enhancer ( e . g . , hb proximal or sna distal ) already mediates maximum binding and release of Pol II at the promoter , and additional enhancers are therefore unable to increase the levels of expression . However , the increase in the levels of sna expression upon removal of the primary enhancer is inconsistent with this explanation . Perhaps , the proximity of the proximal enhancer to the sna promoter gives it a ‘topological advantage’ in blocking access of the distal enhancer ( Dunipace et al . , 2011 ) . The proximal enhancer might mediate less efficient transcription than the distal enhancer and thereby reduce the overall levels of expression ( see Figure 4E ) . We do not believe that this proposed difference is due to differential rates of Pol II elongation since published ( Garcia et al . , 2013 ) and preliminary studies suggest that different enhancers and promoters lead to similar elongation rates ( ∼2 kb/min; T Fuyaka and M Levine , unpublished results ) . A nonexclusive alternative possibility is that deletion of the proximal enhancer removes associated sna repression elements ( MacArthur et al . , 2009 ) , thereby augmenting the efficiency of the distal enhancer . A minimal model of enhancer–promoter associations provides insights into potential mechanisms . In the parameter regime where such interactions are infrequent the two enhancers display additive behavior . However , in the regime of frequent interactions , enhancers compete for access to the promoter resulting in sub-additive behavior . Enhancer–promoter interaction parameters are likely to vary not only between different enhancers but also as the input patterns are modulated in time and space during development ( Rushlow and Shvartsman , 2012; Kok and Arnosti , 2015 ) . This simple model explains the switch from sub-additive to additive enhancer activities for hb and sna . However , in order to explain the super-additive behavior of the kni enhancers , it would be necessary to incorporate an additional state in the model , whereby both enhancers form an active complex with the same target promoter . Such a complex would have a more potent ability to initiate transcription than individual enhancer–promoter interactions . In summary , we propose that enhancers operating at reduced activities ( ‘weak enhancers’ ) can function in an additive manner due to relatively infrequent interactions with their target promoters . In contrast , ‘strong’ enhancers might function sub-additively due to competition for the promoter ( Figure 4E ) . For hb , this switch between competitive and additive behavior occurs as the levels of Bicoid activator diminish in central regions where the posterior border of the anterior Hb domain is formed . Similarly , stress might reduce the performance of the sna enhancers to foster additive behavior under unfavorable conditions such as increases in temperature ( Perry et al . , 2010 ) . Our study highlights the complexity of multiple enhancers in the regulation of gene expression . They need not function in a simple additive manner , and consequently , their value may be revealed only when their activities are compromised . In brief , BAC clones that map to the region of interest were identified from end-sequenced BAC libraries which can be viewed on a browser at http://pacmanfly . org , and ordered from BacPac Resources ( http://bacpac . chori . org/ ) ( Venken et al . , 2009 ) . These BACs arrive already cloned into a vector containing an attB sequence for targeted integration , mini-white cassette , chloramphenicol resistance and are in the inducible copy number strain EPI300 , Epicentre Biotechnologies ( Madison , WI ) . The following CHORI BACs were used as a starting point: sna BAC ( CH322-18I14-1 ) , hb BAC ( BAC CH322-55J23 ) , kni BAC ( CH322-21P08 ) . BACs requiring modification were first transformed into the recombineering strain SW102 , which was obtained from NCI-Frederick Biological Resources Branch . Cultures containing specific BACs were grown overnight and recombination functions were induced as described ( Perry et al . , 2010 ) . The induced bacteria were electroporated with targeting constructs that were prepared previously by PCR amplification . Targeting constructs were made using a pair of 90 base pair long oligonucleotides . These contained 25 base pairs specific to the region being amplified that was to be swapped into the BAC , and an additional 65 base pairs of sequence homologous to the target BAC flanking the region to be replaced . The homologous regions , or ‘homology arms’ , target the amplified sequence to the region of interest for recombination . After electroporation and a 1 hr recovery period in 2XYT broth , bacteria were plated in a dilution series on LB plates with the appropriate antibiotic for overnight incubation at 30°C . Individual resulting colonies were screened by PCR for appropriate recombination at both homology arm locations . Confirmed positive recombinant colonies were transformed back into EPI300 cells ( Epicentre Biotechnologies ) and reconfirmed by antibiotic marker selection and PCR; PCR products were sequenced for final confirmation . Oligonucleotides for amplification to make homology arm constructs ( 90 base pairs in length ) were from Integrated DNA Technologies; shorter primers for colony screening PCR were from ELIM Biopharmaceuticals . Restriction enzymes were from New England Biopharmaceuticals . Qiagen products were used to isolate plasmid DNAs , gel-purify DNA fragments , and purify PCR products . Qiagen taq polymerase was used in colony PCR screening; Invitrogen Platinum pfx was used to amplify targeting constructs . The first step in the modification was to replace the endogenous coding sequence of sna , hb , and kni genes with that of the yellow-kanamycin reporter gene . The yellow-kanamycin fragment was swapped into the place of the endogenous gene at the ATG start codon at the 5′ end , leaving the 5′ UTR intact . The endogenous 3′ UTR was also left fully intact . In most cases , the different enhancers were replaced with an ampicillin resistance cassette which was PCR amplified from pBluescript . In the case of kni , one of the enhancers is in the intron of the transcribed region and so we replaced enhancer with a fragment of lambda phage DNA using galK positive and negative selection . The next step required the insertion the MS2 stem loop sequences . Copies of the MS2 stem loops were extracted from plasmid pCR4-24XMS2SL-stable ( Addgene 31865 ) and were PCR amplified with primers with appropriate homology sequences . BACs were induced to high copy number using Epicentre BAC autoinduction solution , according to supplier's instructions , and grown overnight for 16–18 hr at 37°C . DNA was prepared for micro-injection using the Invitrogen PureLink HiPure miniprep kit by following manufacturer instructions with described modifications for BACs and cosmids . DNA was diluted to a final concentration of ∼300–400 ng/µl and 1× injection buffer . At least 200 embryos were injected per construct by BestGene Inc . ( Chino Hills , CA ) . The transgenes were integrated into the following landing sites: BDSC 9723 , BDSC 9750 , and BDSC 24749 . Hb lacking shadow and kni lacking primary were integrated into 9750 and 24 , 749 , respectively , while all other transgenes were integrated into 9723 . Female virgins of line yw; Histone-RFP;MCP-NoNLS-GFP ( Garcia et al . , 2013 ) were crossed with males of each reporter line . Collected embryos were dechorinated using bleach and mounted between a semipermeable membrane ( Biofolie , In Vitro Systems & Services ) and a coverslip ( 1 . 5 , 18 mm × 18 mm ) and embedded in Halocarbon 27 oil ( Sigma ) . The flattening of the embryos makes it possible to image a larger number of nuclei in the same focal plane without causing significant changes in early development processes ( Di Talia and Wieschaus , 2012 ) . Embryos were either imaged using a custom-built two-photon microscope ( Liu et al . , 2013 ) and a Zeiss LSM 780 confocal microscope . Imaging conditions on the two-photon microscope were as described in Garcia et al . ( 2013 ) . The average laser power at the specimen was 10 mW , the pixel size was set to 220 nm and a single image consisted of 512 × 256 pixels . At each time point , a stack of 10 images separated by 1 µm was acquired resulting in a final time resolution of 37 s . Confocal imaging was performed using a Plan-Apochromat 40×/1 . 4NA oil immersion objective . The MCP-GFP and Histone-RFP were excited with a laser wavelength of 488 nm and 561 nm , respectively . Fluorescence was detected with two separate photomultiplier tubes using the Zeiss QUASAR detection unit ( gallium-arsenide-phosphide photomultiplier was used for the GFP signal while the conventional detector was used for the RFP ) . Pixel size is 198 nm and images were captured at 512 × 512 pixel resolution with the pinhole set to a diameter of 116 µm . At each time point , a stack of 22 images separated by 0 . 5 µm was captured , spanning the nuclear layer . The final time resolution is 32 s . Analysis was performed as described ( Garcia et al . , 2013 ) and full code can be downloaded from https://github . com/PrincetonUniversity/FlyRNAQuant . Histone-RFP slices were maximum projected for each time point . Nuclei were segmented using an object detection approach based on the Laplacian of Gaussian filter kernel . The segmented nuclei were then segmented and tracked over multiple nuclear cycles . Spots are detected in 3D and assigned to their respectively closest nucleus . When multiple spots are detected in the vicinity of a nucleus only the brightest one is kept . Spot intensity determination necessitates an estimate of the local fluorescent background for each particle . A 2D Gaussian fit to the peak plane of each particle column determines an offset , which is used as background estimator . The intensity is calculated by integrating the particle fluorescence over a circle with a radius of 6 pixels and then subtracting the estimated background . The imaging error is dominated by the error made in the fluorescent background estimation ( Garcia et al . , 2013 ) . It is possible to measure the average fluorescence per polymerase molecule for the hunchback enhancer > MS2 transgene with 24 MS2 repeats ( Garcia et al . , 2013 ) . The quantitative imaging for the BAC transgenes was conducted under the exact same imaging conditions on the same microscope . The BAC transgenes also possess 24 MS2 repeats . However , the specific sequence of the stem loops is slightly different as these repeats have been further optimized to facilitate molecular biology work with them ( Hocine et al . , 2013 ) . Assuming that the MS2 sites are similarly saturated with MCP::GFP protein in both cases we can then use the average fluorescence per polymerase molecule calculated for the hunchback>MS2 transgene to calibrate the BAC fluorescent traces in terms of the absolute number of transcribing polymerases per fluorescent spot . We propose a general scheme for enhancer promoter interactions which makes it possible to model the effect of having multiple enhancers activating a single promoter . Here , the enhancer and promoter engage and disengage with one another with characteristic rate constants kon and koff , respectively ( see Figure 4A ) . The ratio between kon and koff determines the strength of the promoter-enhancer interaction . The system can be found in two states . First , the promoter can be unoccupied . We denote the occupancy of this state with [Enhancer] . Second , the promoter and enhancer can be engaged with an occupancy [Enhancer·Promoter] . Following the reaction scheme shown in Figure 4A , the temporal evolution of the occupancy of the state where the enhancer and promoter are engaged is given by ( 1 ) d[Enhancer ·Promoter]dt=kon[Promoter]− koff[Enhancer ·Promoter] . While the enhancer is engaged with the promoter it is capable of producing mRNA at a rate r , which we will call the transcriptional efficiency . Hence the rate of mRNA production is given by ( 2 ) dmRNAdt=r[Enhancer ·Promoter] . Since the promoter can only be in two states , engaged or not engaged by the enhancer , the occupancy of these two states is constrained by ( 3 ) [Enhancer ·Promoter]+[Promoter]=1 . Assuming steady state in the temporal evolution of the state occupancies results in a rate of mRNA production of ( 4 ) dmRNAdt= konkon+koff*r . In Figure 4B we plot this rate as a function of the interaction strength for different values of enhancer efficiency . When considering two enhancers that can interact independently with the promoter ( Figure 4C ) we need to calculate the occupancy of the state accounting for the promoter interacting with enhancer A , [EnhancerA·Promoter] , and with enhancer B , [EnhancerB·Promoter] . The temporal evolution of these occupancies is given by ( 5 ) d[EnhancerA·Promoter]dt=konA[Promoter]−koffA[Enhancer A·Promoter] , and ( 6 ) d[EnhancerB·Promoter]dt=konB[Promoter]− koffB[Enhancer B·Promoter] . In this case the rate of mRNA production is given by ( 7 ) dmRNAdt=rA[Enhancer A·Promoter]+rB[Enhancer B·Promoter] , where the different promoter occupancy states are constrained by ( 8 ) [Enhancer A·Promoter]+[Enhancer B·Promoter]+[Promoter]=1 . In analogy to the single-enhancer case we now assume steady state for the temporal evolution of the occupancies described by Equations 5 , 6 and use the constraint given by Equation 7 to solve for [EnhancerA·Promoter] and [EnhancerB·Promoter] . Finally we replace these results into Equation 8 leading to ( 9 ) dmRNAdt=rAkonAkoffB+rBkonBkoffAkonBkoffA+konAkoffB+koffAkoffB . Using this equation we determine the rate of mRNA production as a function of the strength ( i . e . , r ) and the efficiency ( i . e . , kon/koff ) of enhancers A and B . These calculations are used to generate the plots shown in Figure 4D , E .

Only a subset of the genes in a cell is active at any time . Gene activation or ‘transcription’ is controlled by specific DNA sequences called promoters and enhancers . Promoters are found next to genes and recruit the protein machinery needed to transcribe the gene . Enhancers are located further away from genes and interact with the promoter to increase gene transcription . Some genes have multiple enhancers with overlapping activities , typically including a primary enhancer and secondary or ‘shadow’ enhancer that is even more distant . Multiple enhancers are used to control gene transcription in different types of cells . For example , shortly after fertilization around 1000 enhancers in the embryo of the fruit fly Drosophila regulate several hundred genes that are important for development . These activities establish which cell type any given cell in the embryo will become . Many of these developmental genes have shadow enhancers , which are thought to make gene activity more precise and reliable . It is not known , however , how several enhancers interact with the same promoter at the same time . For example , do the enhancers' individual effects add together ? Or can the combined effects of multiple enhancers be more ( or even less ) than the sum of their parts ? Bothma , Garcia et al . have now examined how multiple enhancers regulate the activity of three developmental genes ( called hunchback , knirps , and snail ) in early Drosophila embryos . The experiments showed that the individual effects of the knirps enhancers add together as one might expect . On the other hand , the snail enhancers interfere with each other , which means that their combined effect on transcription is less than the sum of the two individual effects . Furthermore , other experiments revealed that the combined effect of the hunchback enhancers depends upon whether another component that is needed for gene activation is in short supply . To understand these observations , Bothma , Garcia et al . then developed a mathematical model . The model proposes that behavior of enhancers depends upon how strongly they interact with the target promoter . Since ‘weak’ enhancers do not interact very often , their effects can easily add together . However , the effects of ‘strong’ enhancers do not add together because they often compete to interact with the promoter . These findings show how multiple enhancers can work in a complex manner to control gene transcription .

2015

Enhancer additivity and non-additivity are determined by enhancer strength in the Drosophila embryo

Excitation-inhibition ( E:I ) imbalance is theorized as an important pathophysiological mechanism in autism . Autism affects males more frequently than females and sex-related mechanisms ( e . g . , X-linked genes , androgen hormones ) can influence E:I balance . This suggests that E:I imbalance may affect autism differently in males versus females . With a combination of in-silico modeling and in-vivo chemogenetic manipulations in mice , we first show that a time-series metric estimated from fMRI BOLD signal , the Hurst exponent ( H ) , can be an index for underlying change in the synaptic E:I ratio . In autism we find that H is reduced , indicating increased excitation , in the medial prefrontal cortex ( MPFC ) of autistic males but not females . Increasingly intact MPFC H is also associated with heightened ability to behaviorally camouflage social-communicative difficulties , but only in autistic females . This work suggests that H in BOLD can index synaptic E:I ratio and that E:I imbalance affects autistic males and females differently . Excitation-inhibition ( E:I ) balance in the brain has been hypothesized to be atypical in many neuropsychiatric conditions ( Rubenstein and Merzenich , 2003; Sohal and Rubenstein , 2019 ) , including autism . Rubenstein and Merzenich originally suggested that some types of autism may be explained by an E:I imbalance that may lead to hyper-excitability in cortical circuitry and potentially enhanced levels of neuronal noise ( Rubenstein and Merzenich , 2003 ) . However , coming to a better understanding of how E:I balance is affected across a heterogeneous mixture of autistic individuals has proven to be challenging because of the limited availability of robust E:I biomarkers that are non-invasive and applicable in humans and which can be measured on a large scale . A majority of the literature about E:I balance in autism extends from investigations of prominent single gene mutations associated with autism and the animal model research around these genes ( Sohal and Rubenstein , 2019; Nelson and Valakh , 2015 ) . This leaves a significant gap in evaluating the E:I theory on a larger majority of the autistic population . While no one theory can fully explain all individuals with an autism diagnosis ( Happé et al . , 2006; Lombardo et al . , 2019a ) , the E:I imbalance theory may have utility for understanding subtypes of autistic individuals ( Lombardo et al . , 2015; Lombardo et al . , 2018a; Lombardo et al . , 2019b ) . Sex/gender may be an important stratifier of relevance for highlighting E:I imbalance subtypes ( Lai et al . , 2015; Lai et al . , 2017a ) . Many highly penetrant autism-associated genes are located on the sex chromosomes ( e . g . , FMR1 , MECP2 , NLGN3 , GABRA3 , ARX , SYN1 ) and are known to lead to pathophysiology implicating E:I dysregulation ( Rubenstein and Merzenich , 2003; Lee et al . , 2017; Bozzi et al . , 2018 ) . Other genes playing important roles in the balance between excitation and inhibition in the brain ( e . g . , MEF2C , GRIK2 , GRIA1 , SCN3A , SCN9A , NPTX2 ) are highly sensitive to androgens in human neuronal stem cells and are highly expressed in ‘social brain’ circuitry such as the default mode network , and in particular , the medial prefrontal cortex ( MPFC ) ( Lombardo et al . , 2018b ) . Optogenetic stimulation to enhance excitation in mouse MPFC also results in changes in social behavior ( Yizhar et al . , 2011; Selimbeyoglu et al . , 2017 ) . These results hint that sex-relevant biological mechanisms affect E:I balance and that key social brain regions such as MPFC may be of particular importance for explaining how E:I imbalance affects social behavior . Sex/gender heterogeneity also leads to differing clinical presentations and compensatory mechanisms in autism that may depend on E:I balance in MPFC . It is known that many cognitively able adult autistic women engage in camouflaging behaviors that tend to compensate or mask their social-communicative difficulties moreso than autistic men ( Lai et al . , 2017b; Hull et al . , 2020; Schuck et al . , 2019 ) . Prior work has shown that whereas autistic males show reduced ventral MPFC ( vMPFC ) self-representation response , autistic females show intact vMPFC self-representation . Furthermore , the degree to which vMPFC shows intact self-representation response in autistic females is associated with enhanced ability to camouflage ( Lai et al . , 2019 ) . If E:I imbalance asymmetrically affects vMPFC function in males versus females , this could help explain differential camouflaging in adult autistic females . To better understand sex-specific E:I imbalance in autism we need better neuroimaging biomarkers that index underlying synaptic E:I mechanisms and which can be deployed on a large-scale for in-vivo investigation in deeply phenotyped cohorts . Here we pursue the idea that spectral properties of neural time-series data ( e . g . , local field potentials ( LFP ) or blood oxygen level dependent ( BOLD ) signal ) could be used to isolate such biomarkers . It has been long known that LFP and resting state fMRI ( rsfMRI ) data exhibits rich spectral properties , with power decreasing as function of frequency ( Bullmore et al . , 2001; Maxim et al . , 2005; He , 2011; Bédard et al . , 2006; He , 2014 ) . Models of neural networks have reported that the E:I ratio has profound effects on the spectral shape of electrophysiological activity ( Brunel and Wang , 2003; Mazzoni et al . , 2008; Lombardi et al . , 2017 ) . Recent work with simplified models has proposed that the exponent of the 1/f spectral power law , an index closely related to the Hurst exponent ( H ) , reflects the extent of E:I imbalance ( Gao et al . , 2017 ) . This suggests that neurophysiologically heightened E:I ratio generates flatter 1/f slope in LFP data and this could drive H ( as measured in BOLD ) to be decreased . In past work we have shown that H is atypically decreased in rsfMRI data of adult autistic males , particularly for social brain areas like MPFC ( Lai et al . , 2010 ) . H is statistically relevant for the concept of ‘neural noise’ since lower levels of H can be interpreted as closer to what would be expected of a completely noisy random signal ( e . g . , white noise produces an H = 0 . 5 ) . Related to H and long-memory characteristics of the rsfMRI time-series , prior work has also shown case-control differences in the intrinsic neural timescale in autism ( e . g . , magnitude of temporal autocorrelation ) ( Watanabe et al . , 2019 ) . However , these prior studies examine primarily male-dominated samples and thus cannot shed insight into sex-related heterogeneity in autism . In this work we aim to better understand how E:I imbalance may differentially affect autistic males and females . To achieve this aim , we first took a bottom-up approach by using in-silico ( i . e . computational ) models of local neuronal microcircuitry to make predictions about how H and 1/f slope in LFP and rsfMRI BOLD data may behave when there are underlying changes in E:I balance . Importantly , our approach takes a major step forward from prior work ( Gao et al . , 2017 ) by utilizing a model that includes interactions within and between excitatory and inhibitory neuronal populations . Next , our in-silico predictions are then tested in-vivo with a combination of rsfMRI and experimental manipulations in mice that either increase neurophysiological excitation or that silence the local activity in the network . Chemogenetic ( i . e . designer receptors exclusively activated by designer drugs; DREADD ) or optogenetic manipulations are optimally suited to these purposes , owing to the possibility of enabling remote control of neuronal excitability with cell-type and regional specificity ( Yizhar et al . , 2011; Ferenczi et al . , 2016 ) . Manipulations of neuronal activity like these in animals are key for two reasons . First , they allow for experimental in-vivo confirmation of in-silico predictions . Second , such work is a key translational link across species ( i . e . rodents to humans ) , given the common use of neuroimaging readouts from rsfMRI ( Balsters et al . , 2020 ) . At the genomic level we then examine what cell types could possibly underlie sex-related heterogeneity in E:I imbalance . Finally , we then turn to the human rsfMRI data to show how E:I imbalance may differ amongst autistic males and females and how such mechanisms may explain individual differences in camouflaging behavior . In a bottom-up fashion , we first worked to identify potential biomarkers of E:I imbalance from neural time-series data such as local field potentials ( LFPs ) . Motivating our in-silico modeling of E:I effects on LFP and BOLD data , we note prior work by Gao and colleagues ( Gao et al . , 2017 ) . This prior work simulated LFP time-series from non-interacting excitatory and inhibitory neuronal populations ( Figure 1—figure supplement 1A ) and showed that spectral properties such as the 1/f slope flatten with increasing E:I ratio ( Figure 1—figure supplement 1B ) . Given the relationship between 1/f slope and H ( Stadnitski , 2012 ) , we show within this modeling approach that as E:I ratio increases , H decreases ( Figure 1—figure supplement 1C ) . However , a limitation of this prior work is that it does not include interactions between excitatory and inhibitory populations nor does it allow for recurrent connections within such populations . To address these limitations , we developed a more biologically plausible recurrent network model of interacting excitatory and inhibitory integrate-and-fire neuronal populations that receive external inputs ( both a sensory driven thalamic input and a sensory unrelated intracortical input ) ( Figure 1A; see Materials and methods for more details ) . From this model , we computed the network’s LFP as the sum of absolute values of all synaptic currents . The absolute value is taken because AMPA synapses are usually apical and GABA synapse are peri-somatic and thus their dipoles sum with the same sign along dendrites ( Mazzoni et al . , 2008; Mazzoni et al . , 2010; Deco et al . , 2004 ) . We computed LFP summing presynaptic currents from both external inputs and recurrent interactions , as real LFPs capture both sources of synaptic activity ( Logothetis , 2008 ) . We have extensively validated this method of computing LFPs from integrate-and-fire networks in previous work on both real cortical data and simulations with networks of realistically-shaped 3D neurons and have shown that it works better than when using alternatives such as the sum of simulated membrane potentials , the signed sum of synaptic currents or a time integration of the spike rate ( Mazzoni et al . , 2008; Mazzoni et al . , 2015 ) . In this in-silico network , we manipulated the E:I ratio by independently varying the strengths of the inhibitory ( gI ) and excitatory ( gE ) synaptic conductances . We called g the relative ratio between inhibitory and excitatory conductances ( g=gI/gE ) . We report simulation results for two levels of strength of thalamic input ( υ0 = 1 . 5 spikes/second and υ0 = 2 spikes/second ) , and we verified that our results hold qualitatively for a wider range of input levels ( 1 . 5 to 4 spikes/second ) . Figures 1B-C show examples of LFP time-series and power spectral densities ( PSDs ) for two values of g , one within an excitation-dominated regime ( g = 5 . 6 ) and the other within an inhibition-dominated regime ( g = 14 . 8 ) . The spectral profiles ( Figure 1C ) display two different regions of frequencies with different spectral properties: a region of steeper negative 1/f slopes at higher frequencies ( > 30 Hz ) and a region of shallower ( small negative and sometimes positive ) slopes at low frequencies ( < 30 Hz ) . Thus , we calculated slopes for the low- and high-frequency regions with piecewise regressions of log power predicted by log frequency . Slopes from the low-frequency ( Figure 1D ) and high-frequency region ( Figure 1E ) increase when g is reduced ( i . e . E:I ratio augmented ) . This means that lower values of g correspond to faster spectra with relatively more power at higher frequencies . Changes in slopes are more prominent in the excitation-dominated region where g is smaller ( that is , E:I ratio is shifted in favor of E ) than the reference value ( g = 11 . 3 ) , which has been shown to be a plausible reference value that reproduces cortical power spectra well ( Mazzoni et al . , 2008; Mazzoni et al . , 2010; Mazzoni et al . , 2015; Mazzoni et al . , 2011; Barbieri et al . , 2014; Cavallari et al . , 2014 ) . An increase in g beyond this reference value ( shifting the E:I balance towards stronger inhibition ) had a weaker effect on slopes . Similar results were obtained when quantifying 1/f slope using the FOOOF algorithm ( Haller , 2018; Figure 1—figure supplement 2 ) , indicating that slopes are not biased by the particular piecewise linear fit procedure . Next , we computed H from the same simulated LFPs . As expected , H decreases with decreasing g ( i . e . increasing E:I ratio ) , but only when g is below the baseline reference value ( Figure 1F ) . These results clearly indicate that , in a biologically plausible computational model of local cortical microcircuitry including recurrent connections between excitatory and inhibitory neuronal populations , changes in synaptic E:I ratio are reflected by and thus could be inferred from the overall LFP readout of 1/f slope or H . Given that E:I ratio in LFP data is related to 1/f slope and H , we next asked whether simulated fMRI BOLD signal from the recurrent model would also show similar relationships . To answer this question , we first had to simulate BOLD data from the LFP data generated from the recurrent model . Our approach to simulating BOLD ( see Materials and methods and Figure 2—figure supplement 1 for how BOLD was simulated from LFP ) , captures several key characteristics about the empirical relationship between LFP and BOLD . Studies with simultaneous LFP and BOLD measured in animals have shown that although BOLD signal correlates with both LFPs and spikes , it correlates more strongly with the LFP than with spikes ( Logothetis et al . , 2001; Magri et al . , 2012; Rauch et al . , 2008; Viswanathan and Freeman , 2007; Lauritzen and Gold , 2003 ) . Further studies with simultaneous LFP and BOLD measured in non-human primates ( Logothetis et al . , 2001; Magri et al . , 2012; Schölvinck et al . , 2010 ) have considered the relationship between frequency-resolved LFPs and BOLD and indicate that LFP power shows time-lagged correlations with the time course of BOLD signal and that different frequency bands vary in how they correlate with measured BOLD signal . In particular , gamma band frequencies tend to show the strongest correlation between LFP power and BOLD signal . Frequency dependency of the EEG-BOLD relationship , with prominent predictive power of the gamma band , is also reported in humans ( Scheeringa et al . , 2011 ) . Remarkably , these empirical observations are recapitulated with simulated LFP and BOLD data from the recurrent model . Figure 2A shows time-lagged correlations between time-dependent LFP power and BOLD . Figure 2B shows that all considered LFP frequency bands ( e . g . , alpha , beta , gamma ) correlate with BOLD , but with the gamma band showing the strongest correlations . Thus , our method for simulating BOLD from recurrent model LFP data retains key empirical relationships observed between real LFP power and BOLD . Simulating BOLD with a simple hemodynamic response function ( HRF ) convolution of the LFP would have not respected the patterns of correlations between LFP power and BOLD observed in empirical data ( i . e . the relative increase in correlation between the gamma band and BOLD with respect to other bands; Figure 2—figure supplement 2 ) . With simulated BOLD from the recurrent model , we next computed H on these data to understand if E:I ratio in the recurrent model is associated with changes in H in BOLD . Strikingly , H in BOLD shows the same dependency on g as observed in LFP data ( Figure 2C-D ) - H in BOLD decreases as E:I ratio is shifted toward higher excitation by lowering the value of g with respect to the reference value . Although H in LFP and BOLD showed similar associations with respect to changes in g , it is notable that the range of H in BOLD is shifted towards smaller values ( Figure 2C-D ) than H in LFP ( Figure 1G-H ) . We also verified that the dependency of H in BOLD on g was largely independent of the details of how BOLD is simulated from LFP . While the results shown in Figure 2 are computed with an HRF that reproduces the correlation function measured between the BOLD signal and the gamma band of LFP ( Magri et al . , 2012 ) , it is notable that these results remained similar when using the canonical HRF instead ( Figure 2—figure supplement 2 ) . Removing the high pass filter from simulation of BOLD response did alter the relative values of correlation between LFP power and BOLD across frequency bands , making the BOLD response more in disagreement with experimental data , but did not change the relationship of decreasing H with decreasing g ( Figure 2—figure supplement 2 ) , suggesting that our conclusions are robust to the details of the model of the LFP to BOLD relationship . In sum , the inferences from the recurrent network model suggest that H in LFP and BOLD data can be utilized as a marker to track changes in underlying synaptic E:I mechanisms . We next investigated manipulations of parameters within the recurrent model that approximate the effects of empirical chemogenetic DREADD manipulations in neurons . These simulations are useful to both gain a better understanding of the empirical BOLD measures under DREADD manipulations presented in the next section , and to better characterize the specificity of the origin of changes in 1/f slopes and H with the E:I ratio . Given that as shown above , in our models changes of H in BOLD mirror those in LFPs , here we present changes in model LFP spectra when simulating these DREADD manipulations . We first studied the specific effect of solely increasing excitation within the recurrent network . This can be achieved experimentally by using the drug clozapine-N-oxide ( CNO ) on the DREADD receptor hM3Dq to increase the excitability of excitatory cells only ( Alexander et al . , 2009 ) . We simulated this kind of increase of excitability of pyramidal cells in the recurrent network model by lowering their voltage threshold ( Vth ) for spike initiation . Progressively lowering Vth from -52 to -53 mV resulted in more positive low-frequency and flatter high-frequency 1/f slopes ( Figure 3—figure supplement 1A ) and also caused decreases in H ( Figure 3A ) . For H , increasing Vth ( i . e . decreasing excitability ) from -52 to -51 mV resulted in little change in H . These results predict that specific increases of excitation , as in the application of the hM3Dq DREADD to enhance excitability of pyramidal neurons , should reduce steepness of the high-frequency slopes and lead to a decrease in H . These results also confirm our above findings that in recurrent networks in which excitatory and inhibitory neurons interact , increases in excitability are easier to detect from changes in 1/f slope or H than decreases in excitation . To study whether the changes in 1/f slopes and H are specific to modulations in excitability of only excitatory neurons , we modeled the combined effect of silencing both excitatory and inhibitory neuronal populations . This silencing of both excitatory and inhibitory neurons can be obtained experimentally by application of the hM4Di DREADD ( see next Section ) . In the recurrent network model , we simulated this silencing of both excitatory and inhibitory cells by decreasing the resting potential , EL , in both excitatory and inhibitory neurons . Decreasing EL from the baseline value of -70 to -75 mV produced varied effects in 1/f slopes ( Figure 3—figure supplement 1B ) and resulted in a slight increase of H ( Figure 3B ) . Note that a moderate increase in H with higher input ( Figure 3B ) was also found when comparing two very different levels of input . Given that a possible non-local action of hM4Di might lead to less excitatory input to the considered area coming from the silencing of nearby regions , this suggests that our conclusion should still hold even in the presence of some non-local DREADD effects . In general , the effects of simulating hM4Di DREADD were far less prominent than those reported above when simulating enhanced excitation specifically ( Figure 3A and Figure 3—figure supplement 1A ) . These results predict overall a very small effect of the hM4Di DREADD on H and 1/f slopes . These results also imply that decreases in H are more likely to result from specific increases in excitation rather than from non-specific decreases of excitability across both excitatory and inhibitory neuronal populations . All of the results thus far report results from our in-silico model of recurrent neuronal networks and their readouts as simulated LFP or BOLD data . The in-silico modeling of BOLD data suggests that if E:I ratio is increased via enhanced excitability of excitatory neurons , then H should decrease . To empirically test this prediction in-vivo , we measured rsfMRI BOLD signal in prefrontal cortex ( PFC ) of mice under conditions where a chemogenetic manipulation ( hM3Dq DREADD ) ( Alexander et al . , 2009 ) is used to enhance excitability of pyramidal neurons . Here we used a sliding window analysis to assess dynamic changes in H over the course of 3 different phases of the experiment – 1 ) a ‘Baseline’ phase where the CNO drug or a SHAM injection had not yet occurred , 2 ) a ‘Transition’ phase directly following CNO or SHAM injection , and 3 ) a ‘Treatment’ phase , whereby CNO has its maximal effect . We find that H is modulated over time by the DREADD manipulation ( condition*time*treatment phase interaction F = 349 . 03 , p<0 . 0001 ) . During the Baseline phase of rsfMRI scanning before the DREADD-actuator CNO was injected , H under DREADD or a SHAM control conditions are not affected ( condition main effect F = 0 . 82 , p=0 . 37; condition*time interaction F = 0 . 36 , p=0 . 54 ) . However , during the Transition phase of the experiment where the CNO begins to have its effects , we find a condition*time interaction ( F = 4 . 94 , p=0 . 0262 ) , whereby H drops over time at a steeper rate during the DREADD condition compared to the SHAM condition ( green line in Figure 3C ) . Finally , during the Treatment phase of the experiment , where the drug exerts its maximal effect , there is a significant main effect of condition ( F = 12 . 92 , p=0 . 0011 ) and no condition*time interaction ( F = 0 . 66 , p=0 . 4182 ) ( blue line in Figure 3C ) ( Table 1 ) . This effect is explained by H being reduced in the DREADD vs SHAM condition . These in-vivo results are directly in line with the in-silico prediction that enhancing E:I ratio via enhancing the excitability of excitatory neurons results in a decrease in H ( i . e . Figure 1F–H , Figure 2C–D , and Figure 3A ) . While the above results show that specific enhancement of excitability in excitatory neurons results in a decrease in BOLD H , it is an important negative control contrast to investigate whether non-specifically reducing the excitability of both excitatory and inhibitory neuronal populations might also affect H . This is an important negative control since if H were to change in a similar direction after this manipulation , it would make interpretations about decrease in H being due to increased E:I ratio via excitation problematic . The in-silico simulation of this manipulation ( Figure 3B ) would predict that H would not be changed much , and that if a change in H were to occur , it would be a slight increase rather than a decrease in H . By expressing the inhibitory hM4Di DREADD ( Stachniak et al . , 2014 ) under the control of a pan-neuronal promoter , we chemogenetically silenced both excitatory and inhibitory neurons in PFC of mice and re-ran the same rsfMRI neuroimaging protocol as before . While a significant 3-way interaction between condition , time , and treatment phase was present ( F = 85 . 8 , p<0 . 0001 ) , there were no strong main effects of condition or condition*time interactions in any of the baseline , transition , or treatment phases of the experiment ( see Table 2 and Figure 3D ) . Overall , these results along with the recurrent model simulation of hM4Di DREADD ( Figure 3B ) bolster strength of the interpretation that enhanced excitation drives decreasing H in BOLD and that H in BOLD would not change appreciably in a situation such as pan-neuronal silencing of both excitatory and inhibitory neurons . Consistent with the idea that heightened excitation leads to flattening of the 1/f slope and reductions in H , we also computed a measure of the fractional amplitude of low frequency fluctuations ( fALFF ) ( Zou et al . , 2008 ) . Given the effect of flattening 1/f slope , we expected that fALFF would show reductions due to the DREADD excitation manipulation but would show no effect for the DREADD silencing manipulation . These expectations were confirmed , as DREADD excitation results in a large drop in fALFF , which shows a stark drop off midway through the transition phase and stays markedly lower throughout the treatment phase when the drug has its maximal effects . In contrast , similar effects do not occur for the DREADD silencing manipulation ( see Figure 3—figure supplement 2 ) . The in-silico and in-vivo animal model findings thus far suggest that excitation affects metrics computed on neural time-series data such as 1/f slope and H . Applied to the idea of sex-related heterogeneity in E:I imbalance in autism , these results make the prediction that excitatory neuronal cell types would be the central cell type affecting such neuroimaging phenotypes in a sex-specific manner . To test this hypothesis about sex-specific effects on excitatory neuronal cell types , we examined whether known autism-associated genes that affect excitatory neuronal cell types ( Satterstrom et al . , 2020; Velmeshev et al . , 2019 ) are highly overlapping with differentially expressed genes in human neuronal stem cells when treated with a potent androgen hormone , dihydrotestosterone ( DHT ) ( Lombardo et al . , 2018b; Quartier et al . , 2018 ) . Genes differentially expressed by DHT are highly prominent within the gene set of autism-associated genes that affect excitatory neurons ( OR = 1 . 67 , p=0 . 03 ) , with most of the overlapping genes being those whereby DHT upregulates expression ( Figure 4A ) . By contrast , genes associated with autism that affect inhibitory neuronal cell types or other non-neuronal cells ( e . g . , microglia , astrocytes , oligodendrocytes ) are not enriched for DHT differentially expressed genes ( inhibitory neurons: OR = 1 . 51 , p=0 . 12; microglia: OR = 0 . 78 , p=0 . 78; astrocytes or oligodendrocytes: OR = 1 . 11 , p=0 . 49 ) . This result suggests that autism-associated genes specifically affecting excitatory neuronal cell types are also susceptible to the male-specific influence of androgen hormones in human neuronal stem cells . We next additionally examined how such DHT-sensitive and autism-associated excitatory neuron genes spatially express in the adult human brain . This analysis would help shed insight on which brain areas might be more affected by such sex-specific effects in autism . A one-sample t-test of gene maps from the Allen Institute Human Brain Atlas ( Hawrylycz et al . , 2012 ) shows that this subset of DHT-sensitive and autism-associated excitatory neuron genes are highly expressed in MPFC , PCC , insula , and intraparietal sulcus , amongst other areas ( Figure 4B–C ) . We next move to application of this work to human rsfMRI data in autistic men and women . If E:I ratio is affected by sex-related mechanisms ( Lombardo et al . , 2018b ) , we predict that H would be differentially affected in autistic males versus females and manifest as a sex-by-diagnosis interaction in a 2 × 2 factorial design ( Sex: Male vs Female; Diagnosis: Autism vs Typically-Developing ( TD ) ) . More specifically , the directionality of our predictions from the in-silico and in-vivo results in Figures 1–3 are that if H reflects E:I ratio , there should be decreased H ( due to enhanced E ) specifically in autistic males but not autistic females . Mass-univariate analysis uncovered one region in ventromedial prefrontal cortex ( vMPFC ) , region p32 , with a sex-by-diagnosis interaction passing FDR q < 0 . 05 ( F ( 5 , 104 ) = 15 . 13 , p=0 . 0001 , partial η2 = 0 . 12 ) ( Figure 5A ) . In line with directionality of our predictions , this interaction effect is driven by a large TD >Autism effect in males ( Cohen’s d = 1 . 30 ) and a small Autism >TD effect in females ( Cohen’s d = −0 . 27 ) ( Figure 5B ) . A similar sex-by-diagnosis interaction appeared when using another metric such as the intrinsic neural timescale ( Watanabe et al . , 2019; Figure 5—figure supplement 1 ) and when H was first calculated at each voxel and then averaged across voxels ( Figure 5—figure supplement 2 ) . While the main effects of diagnosis and sex are not the primary contrast for this study , we report that no significant regions survived FDR q < 0 . 05 for the main effects of diagnosis . However , 61% of brain regions showed an on-average male >female sex difference ( Figure 5—figure supplement 1 ) , which is in keeping with results from other work on sex differences in H ( Dhamala et al . , 2020 ) . In contrast to mass-univariate analysis , we also used partial least squares ( PLS ) analysis as a multivariate alternative to uncover distributed neural systems that express the sex-by-diagnosis interaction . PLS analysis identified one neural system expressing the same sex-by-diagnosis interaction ( d = 2 . 04 , p=0 . 036 ) and included default mode network ( DMN ) areas such as MPFC and posterior cingulate cortex/precuneus ( PCC ) ( Figure 5—figure supplement 1 ) , and other non-DMN areas such as insula , lateral prefrontal cortex , somatosensory and motor cortices , intraparietal sulcus , amongst others ( Figure 5C ) . Many of these regions detected by the PLS analysis were subthreshold of FDR q < 0 . 05 in the mass-univariate analysis , but do show heightened effect sizes in keeping with this sex-by-diagnosis interaction pattern ( e . g . , white and light blue areas in the unthresholded map shown in Figure 5A ) . Detection of these regions in a mass-univariate analysis may require a larger sample size to enhance statistical power . Given that many of these PLS-identified regions of a sex-by-diagnosis interaction appear similar to those that appear in the gene expression map in Figure 4B of DHT-sensitive and autism-associated excitatory genes , we assessed how much each HCP-MMP parcellated regions overlap with the map in Figure 4B . PLS-identified regions in vMPFC ( e . g . , areas p32 and 10r ) overlap by about 73–75% . Areas within the insula ( e . g . , Pol1 , Pol2 , MI ) overlap by around 59–69% . Parietal areas in PCC ( e . g . , v23ab , d23ab ) and intraparietal sulcus ( LIPd ) overlap by around 73–85% ( Figure 5D ) . In prior task-fMRI work we found a similar sex-by-diagnosis interaction in vMPFC self-representation response and a female-specific brain-behavioral correlation with camouflaging ability ( Lai et al . , 2019 ) . Given that adult autistic females engage more in camouflaging on-average ( Lai et al . , 2017b; Hull et al . , 2020; Schuck et al . , 2019 ) , we next asked whether vMPFC H would be related to camouflaging in a sex-specific manner . In autistic females , increased camouflaging was strongly associated with increased H in vMPFC ( r = 0 . 60 , p=0 . 001 ) . However , no significant association was apparent in autistic males ( r = −0 . 10 , p=0 . 63 ) . The strength of this brain-behavioral correlation significantly differed between autistic males and females ( z = 2 . 58 , p=0 . 009 ) ( Figure 5E ) . This result suggests that progressively more intact vMPFC H in autistic females , which are likely reflective of more intact E:I balance , is associated with better ability to camouflage social-communicative difficulties . Beyond this hypothesis-driven comparison of the relationship between H and camouflaging in vMPFC , we also ran correlations with ADI-R , ADOS and AQ scores . ADOS social-communication ( SC ) was negatively correlated with vMPFC H in autistic females ( r = −0 . 51 , p=0 . 008 ) indicating higher H with lower SC severity . This relationship was not present in autistic males ( r = −0 . 04 , p=0 . 83 ) . However , the difference between these correlations was not statistically significant ( z = 1 . 70 , p=0 . 08 ) . ADI-R subdomains , ADOS RRB , and AQ correlations were not statistically significant . In this work we set out to better understand how intrinsic E:I imbalance affects the autistic brain in a sex-specific manner . Evidence from animal models of rare genetic variants associated with autism have typically been used as the primary evidence for the E:I imbalance theory ( Rubenstein and Merzenich , 2003; Sohal and Rubenstein , 2019 ) . However , these variants affect only a small percentage of the autism population . Thus , it is unclear how E:I imbalance might affect the majority of heterogeneous individuals within the total autism population . To bridge this gap we need multi-level methods that can be applied to understand the ‘living biology’ behind actual human individuals ( Courchesne et al . , 2019 ) , such as in vivo neuroimaging data and metrics applied to such time-series data that are linked to actual underlying neural E:I mechanisms ( Markicevic et al . , 2020 ) . Bridging this gap will help us identify mechanistic targets that explain neural and behavioral variability across a much larger portion of individuals in the autism population . Based on earlier work ( Gao et al . , 2017 ) , we reasoned that metrics such as 1/f slope and H in neural time-series data would be relevant as an in vivo neuroimaging marker of E:I mechanisms . Prior work suggested this relationship via a model that considers inhibition and excitation as separate entities ( Gao et al . , 2017 ) . However , excitation and inhibition in the brain are inseparably linked . Results about the relationship between spectral shape and E:I balance obtained with our model of recurrent excitation and inhibition are largely compatible with those obtained with an earlier model of uncoupled excitation and inhibition ( Gao et al . , 2017 ) . The uncoupled model predicts a linear increase of the slope value ( i . e . flatter , less negative slopes ) as E:I ratio increases . This is because in the uncoupled model changing the E:I ratio modifies only the ratio of the contribution to the LFP spectra of excitatory ( faster time constant ) and inhibitory ( slower time constant ) synaptic currents , leading to a linear relationship between slopes and E:I . In contrast , we found that the relationship between E:I and the spectral slope flattens out for high values of I . This , in our view , may in part arise from the fact that , as shown in studies of recurrent network models ( Brunel and Wang , 2003 ) , higher recurrent inhibition leads to higher peak frequency of gamma oscillations ( i . e . an increase of power at higher frequencies ) thus partly counteracting the low-pass filtering effect of inhibitory currents in the uncoupled model . We plan to investigate in future studies how these opposing effects interact in a wider range of configurations and to use these results to gain a better understanding of the relationship between E:I ratio and LFP spectral shape . Furthermore , prior work ( Gao et al . , 2017 ) considered only 1/f slopes in simulated LFP data and did not explore the effect of the transformation between LFP neural activity to BOLD . Our simulations address these problems and significantly extends prior work ( Gao et al . , 2017 ) on the relationship between E:I imbalance and changes in spectral properties of neural signals . We showed that when excitation and inhibition interact in a recurrent network model , flatter 1/f slopes and decreases in H are specific markers of increases in E:I ratio . We also showed that in simulated BOLD signal , H and E:I ratio are associated in a manner similar to the relationships observed with LFP data . Taken together , these results predict that changes in H in neural time-series data can be interpreted as a shift in synaptic E:I ratio that permeates through in LFP or BOLD readouts . Our simple model to generate BOLD from frequency-resolved LFPs reflect several features of the empirical LFP-BOLD relationship - namely the presence of a particularly strong gamma-BOLD relationship and the fact that a better prediction of the BOLD is obtained from the frequency-resolved LFP than from the wideband LFP . However , a limitation of our simple model is that , in its present form , it cannot capture the negative relationship between the power of some low-frequency LFP bands and the BOLD amplitude that has been reported in some studies ( Schölvinck et al . , 2010; Scheeringa et al . , 2011; Mukamel et al . , 2005; Niessing et al . , 2005 ) . Modelling the low frequency LFP to BOLD relationship in greater detail would require significant extensions of our neural model , as lower frequency oscillations are thought to arise from more complex cortico-cortical and thalamocortical loops than those that can be captured by our simple model of a local recurrent circuit with only two classes of neurons and no spatial structure ( Scheeringa and Fries , 2019; Zucca et al . , 2019 ) . An important topic for further modelling work will be to understand how biomarkers of more complex neural feedback loops can be extracted from LFP or BOLD spectral signatures . The power of our in-silico modeling approach is that it provides explicit predictions of what to expect in real BOLD data when synaptic E:I imbalance occurs . Remarkably , these in silico predictions are confirmed in vivo with rsfMRI BOLD data in halothane-sedated mice after experimental chemogenetic manipulations that specifically enhance neural excitation . Intriguingly , and consistent with in-silico predictions , manipulations that silence both excitatory and inhibitory neuronal populations do not have a strong effect on H in BOLD . These results are in line with optogenetic studies showing that specifically enhancing excitation in MPFC seems to have the biggest effects on social behavior in mice ( Yizhar et al . , 2011 ) . The present work clearly shows that enhancement of excitation results in measurable changes in BOLD readouts as decreases in H . This insight allows us to leverage H as an in-vivo rsfMRI biomarker that has strong relevance back to synaptic E:I imbalance . Future extensions of our research might involve refined modelling and the use of chemogenetic manipulations in awake conditions , hence minimizing the possible confounding contribution of anesthesia on baseline E:I balance . With regards to how sex-related heterogeneity in E:I imbalance might manifest in autism , we utilized genomics data and found that autism-associated genes that affect excitatory neuronal cell types are enriched for genes that are differentially expressed by DHT in human neuronal stem cells . This inference extends prior work implicating excitatory neuron cell types in autism-relevant biology ( Satterstrom et al . , 2020; Velmeshev et al . , 2019; Willsey et al . , 2013; Parikshak et al . , 2013 ) by linking genomic mechanisms in these cell types to the male-specific influence of androgen hormones . Importantly , other cell types such as inhibitory neurons do not express autism-associated genes that are also influenced by DHT . Additionally , the DHT-sensitive and autism-associated excitatory genes tend to spatially express in the human adult brain in regions such as MPFC , PCC , insula , and intraparietal sulcus , which have been shown to be affected in autism across a range of task-related and rsfMRI studies ( Lai et al . , 2019; Di Martino et al . , 2009; Di Martino et al . , 2014; Uddin and Menon , 2009; Padmanabhan et al . , 2017; Lombardo et al . , 2010 ) , and which overlap with areas discovered by the PLS analysis to express a sex-by-diagnosis interaction ( Figure 5D ) . Moving to human rsfMRI data on adult individuals with autism , we utilized H as a neuroimaging biomarker of E:I imbalance . Specifically , we examined whether H differs between adult males and females with and without autism . Mass-univariate analysis highlighted one region in vMPFC which showed a sex-by-diagnosis interaction - that is , H was specifically reduced in adult autistic males , but not in autistic females . Reduced H in autistic males is compatible with the inference of elevated E:I ratio potentially driven by enhanced excitation . The observed effect in vMPFC may also be consistent with a ‘gender-incoherence’ pattern ( i . e . towards reversal of typical sex differences in autism ) ( Bejerot et al . , 2012 ) . However , sex-specific normative ranges would need to be better established before interpreting effects in autism as being reversals of normative sex differences . More work with much larger general population-based datasets is needed to establish whether there are robust normative sex differences in H and to describe the normative ranges of H may take for each brain region , sex , and across age . Such work would also help with normative modeling ( Bethlehem and Seidlitz , 2018 ) approaches that would enable identification of which autistic individuals highly deviate from sex-specific norms . Multivariate PLS analysis extended the mass-univariate results by showing that a distributed neural system structurally and functionally connected to vMPFC , such as default mode network ( DMN ) areas like PCC ( Buckner and DiNicola , 2019; Yeo et al . , 2011 ) , as well as intraparietal sulcus and insular cortex ( Uddin and Menon , 2009 ) , also expressed a similar but more subtle sex-by-diagnosis interaction . Interestingly , these regions highlighted by the PLS analysis are remarkably similar to the map of brain regions where autism-associated excitatory and DHT-sensitive genes highly express ( Figure 4B–C , Figure 5D ) . Therefore , important social brain circuitry such as the DMN , and other integrative hubs of the salience network ( e . g . , insula ) that connect DMN to other important large-scale networks ( Uddin and Menon , 2009 ) may be asymmetrically affected by heightened E:I ratio in autistic males more than autistic females . These human rsfMRI results are not only compatible with the in silico predictions and the in vivo mouse rsfMRI data presented here , but are also compatible with several prior lines of work . Our prior work highlighted that DMN functional connectivity in typically developing adolescent males , but not females , is affected by heightened levels of fetal testosterone and this network was heavily comprised of MPFC and PCC ( Lombardo et al . , 2018b ) . In the same work , we showed that a cortical midline DMN subsystem comprising MPFC and PCC highly expresses several genes relevant for excitatory postsynaptic potentials ( e . g . , MEF2C , GRIK2 , GRIA1 , SCN3A , SCN9A , NPTX2 ) . The current findings linking autism-associated genes in excitatory neuron cell types ( Figure 4 ) allow for more precise inferences about the importance of excitatory cell types over and above other inhibitory cell types . This is important given that evidence regarding inhibitory neuronal cell types and their role in E:I imbalance in autism is more mixed ( Horder et al . , 2018; Coghlan et al . , 2012 ) . Importantly , the expression of these genes in human neuronal stem cells is elevated after exposure to the potent androgen DHT ( Lombardo et al . , 2018b ) . Thus , one potential explanation for the male-specific reduction of H in vMPFC could have to do with early developmental and androgen-sensitive upregulation of genes that play central roles in excitatory neuron cell types , and thus ultimately affecting downstream E:I imbalance . Such effects may be less critical in human females and may serve an important basis for sex-differential human brain development ( Kaczkurkin et al . , 2019 ) . These effects may also help explain why qualitative sex differences emerge in autism ( Lai et al . , 2017a; Bedford et al . , 2020 ) . rsfMRI H in autistic adults was also relevant in a sex-specific manner to a clinical behavioral phenomenon known as ‘camouflaging’ . Camouflaging relates to a set of compensatory or masking strategies/mechanisms that allow individuals to cope with their social-communicative difficulties in everyday social situations ( Lai et al . , 2017b; Hull et al . , 2020; Livingston et al . , 2019 ) . It is known that cognitively able adult autistic females tend to engage in more camouflaging behavior than males ( Lai et al . , 2017b; Hull et al . , 2020; Schuck et al . , 2019 ) and the extent to which individual females engage in camouflaging is linked to vMPFC function ( Lai et al . , 2019 ) . One of the most important known functions of vMPFC has to do with self-representation ( Lombardo et al . , 2010 ) and simulating others based on information about the self ( Mitchell et al . , 2006 ) . In prior task-related fMRI work we found a similar sex-by-diagnosis interaction effect whereby males are more impaired in vMPFC self-representation response than their female autistic counterparts . Furthermore , increased magnitude of vMPFC self-representation neural response correlates with increased camouflaging ability , but only in adult autistic females ( Lai et al . , 2019 ) . Strikingly , here we find a similar sex-by-diagnosis interaction effect in vMPFC H as well as a female-specific correlation with camouflaging - as vMPFC H increases , indicative of a more normative or intact level of E:I balance , camouflaging also increases . This converging set of results suggests that intrinsic mechanisms such as E:I balance may be atypical only in cognitively able autistic males at vMPFC . More intact E:I balance in the vMPFC of autistic females may enable better vMPFC-related function ( e . g . , self-representation ) and thus potentially better enable these individuals to camouflage social-communicative difficulties and cope in social situations . Future work changing E:I balance in vMPFC may provide a useful avenue for ameliorating daily life social-communication adaptation and coping difficulties in autistic males and enable them to optimally engage in compensatory processes such as camouflaging to the similar extent as autistic females . It may also be fruitful to examine how intact E:I balance in vMPFC of females may be an expression of protective factors that are hypothesized to buffer risk for autism in females ( Robinson et al . , 2013; Werling , 2016 ) . This work may also be of broader relevance for investigating sex-specific E:I imbalance that affects other early-onset neurodevelopmental disorders with a similar male-bias as autism ( Rutter et al . , 2003 ) . For instance , conditions like ADHD affect males more frequently than females and also show some similarities in affecting behavioral regulation and associated neural correlates ( Chantiluke et al . , 2015 ) . Furthermore , gene sets associated with excitatory and inhibitory neurotransmitters are linked to hyperactivity/impulsivity severity in ADHD , suggesting that E:I-relevant mechanisms may be perturbed ( Naaijen et al . , 2017 ) . It will be important for future work to test how specific sex-specific E:I imbalance is to autism versus other related sex-biased neurodevelopmental disorders . Similarly , future work should investigate how H may change over development . Prior work has shown that H and other related measures such as 1/f slope can change with normative and pathological aging in both rsfMRI and EEG data ( Maxim et al . , 2005; Wink et al . , 2006; Voytek et al . , 2015 ) . Imperative to this work will be the establishment of age and sex-specific norms for H in much larger datasets . Age and sex-specific norms will enable more work to better uncover how these biomarkers may be affected in neurodevelopmental disorders or disorders relevant to neurodegeneration . Such work combined with normative modeling approaches ( Bethlehem and Seidlitz , 2018 ) may help uncover how experiential and environmental effects further affect such metrics . In conclusion , we show that spectral properties of neural time-series data , such as H and 1/f slope , can be utilized in neuroimaging readouts like LFP and BOLD as a biomarker for underlying E:I-relevant mechanisms . In silico predictions from simulated LFP and BOLD data were confirmed in vivo with rsfMRI BOLD data where excitation was enhanced through chemogenetic manipulation . Finally , in application to humans , we show that H in rsfMRI data is reduced in vMPFC and other DMN areas of adult autistic males , but not females . Reduced H is indicative of enhanced excitation and thus points to sex-specific dysregulation of E:I balance in social brain networks of autistic males . This male-specific dysregulation of E:I balance may be linked to sex-differential early developmental events such as androgen-upregulation of gene expression for genes that play important roles in excitatory neurons ( Lombardo et al . , 2018b ) . The intact levels of H in females may help facilitate elevated levels of compensation known as camouflaging to cope with daily social-communicative difficulties . This important female-specific brain-behavioral correlation may also be key for future interventions targeting E:I mechanisms and MPFC-related brain networks to enable better coping with daily social-communicative difficulties . More generally , this work extends the relevance of the E:I imbalance theory of autism beyond evidence from autism-associated rare genetic variants and specify a larger portion of the autism population whereby these E:I mechanisms may be of critical importance . All procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975 , as revised in 2008 . All human participants’ informed consent was obtained in accord with procedures approved by the Suffolk Local Research Ethics Committee . Adult native English speakers ( n = 136 , age range = 18–49 years ) with normal/corrected-to-normal vision participated: n = 33 typically developing ( TD ) males , n = 34 autistic males , n = 34 TD females and n = 34 autistic females ( Table 3 ) . They all reported cis-gender identity based on a single item inquiring their birth-assigned sex and another on their identified gender . Groups were not statistically different on age or full-scale IQ ( FIQ ) on the Wechsler Abbreviated Scales of Intelligence ( WASI ) ( Table 3 ) . Exclusion criteria for all participants included a history of or current psychotic disorders , substance-use disorders , severe head injury , genetic disorders associated with autism ( e . g . fragile X syndrome and tuberous sclerosis ) , intellectual disability ( i . e . Full-scale IQ ( FIQ ) < 70 ) , or other medical conditions significantly affecting brain function ( e . g . epilepsy ) . The inclusion criterion for both male and female autistic participants was a formal clinical diagnosis of International Statistical Classification of Diseases and Related Health Problems 10th Revision ( ICD-10 ) childhood autism or Asperger’s syndrome , or Diagnostic and Statistical Manual of Mental Disorders ( 4th ed . , text rev . ; DSM-IV-TR ) autistic disorder or Asperger’s disorder , as assessed by a psychiatrist or clinical psychologist in the National Health Service , UK . Since all participants were adults , we further considered available information of developmental history to include only those with clinically evident childhood autistic symptoms , for example , from information collected using the Autism Diagnostic Interview–Revised ( ADI-R ) ( Lord et al . , 1994 ) where possible , or from the participants’ clinical diagnosis letters shared with the research team to determine eligibility . We used this clinically based criterion for inclusion for the purpose of sampling autistic individuals currently diagnosed by specialists in mental health services in the daily practice and to align with best clinical practice as recommended by the UK National Institute for Health and Clinical Excellence ( NICE ) guideline ( Pilling et al . , 2012 ) . For assessing levels of autism characteristics , we administered the Autism Spectrum Quotient ( AQ ) ( Baron-Cohen et al . , 2001a ) , module 4 of the Autism Diagnostic Observation Schedule ( ADOS ) ( Lord et al . , 2000 ) , and ADI-R ( Lord et al . , 1994 ) where possible , before the fMRI session . Autistic male and female groups were not significantly different on any ADI-R subdomain scores or Reading the Mind in the Eyes Test ( RMET ) ( Baron-Cohen et al . , 2001b ) performance ( Table 3 ) . We further used criteria for inclusion based on characteristics about data quality ( see next paragraphs for data preprocessing ) . In particular , we excluded participants where the number of volumes was not acquired due to scanner hardware issues ( n = 1 ) , the preprocessing pipeline could not adequately preprocess the data ( e . g . , bad registrations; n = 5 ) . Participants were also excluded if their head motion exceed a mean framewise displacement ( meanFD ) ( Power et al . , 2012 ) of >0 . 4 mm ( n = 8 ) . For the remaining subjects we further visually inspected plots of framewise displacement ( FD ) and DVARS ( Power et al . , 2012 ) traces to determine whether the wavelet despiking step sufficiently attenuated artefact-related variability that would leave DVARS spikes . Here we made a qualitative and consensus judgement amongst authors ( S . T . and M . V . L ) to exclude individuals ( n = 9 ) whereby there were numerous FD spikes above 0 . 5 mm or numerous DVARS spikes leftover after wavelet despiking was applied . Other exclusions included any VIQ or PIQ <70 ( n = 1 ) and co-morbid agenesis of the corpus callosum ( n = 1 ) . The final sample sizes included in all further analyses was n = 29 TD males , n = 23 autistic males , n = 33 TD females , and n = 25 autistic females . The final groups used in all analyses did not statistically differ on age ( diagnosis main effect: F ( 3 , 106 ) = 0 . 03 , p=0 . 85; sex main effect: F ( 3 , 106 ) = 0 . 14 , p=0 . 70; sex-by-diagnosis interaction: F ( 3 , 106 ) = 0 . 25 , p=0 . 61 ) or FIQ ( diagnosis main effect: F ( 3 , 106 ) = 3 . 38 , p=0 . 07; sex main effect: F ( 3 , 106 ) = 0 . 48 , p=0 . 48; sex-by-diagnosis interaction: F ( 3 , 106 ) = 2 . 24 , p=0 . 13 ) ( see Table 3 ) . Imaging was performed on a 3T GE Signa Scanner at the Cambridge Magnetic Resonance Imaging and Spectroscopy Unit . Participants were asked to lie quietly in the scanner awake with eyes closed for 13 min and 39 s during sequential acquisition of 625 whole-brain T2*-weighted echo planar image volumes with the following parameters: relaxation time = 1302 ms; echo time = 30 ms; flip angle = 70°; matrix size = 64×64; field of view = 24 cm; 22 anterior commissure-posterior commissure aligned slices per image volume; 4 mm axial slice thickness; 1 mm slice gap . The first five time-points were discarded to allow for T2-stabilization . During analysis of the Hurst exponent ( H ) for BOLD time-series , due to the discrete wavelet transform using volumes in power of 2 , only the first 512 volumes ( 29 ) were utilized . A high-resolution spoiled gradient anatomical image was acquired for each participant for registration purposes . Preprocessing of the resting state data was split into two components; core preprocessing and denoising . Core preprocessing was implemented with AFNI ( Cox , 1996 ) ( http://afni . nimh . nih . gov/ ) using the tool speedypp . py ( http://bit . ly/23u2vZp ) ( Kundu et al . , 2012 ) . This core preprocessing pipeline included the following steps: ( i ) slice acquisition correction using heptic ( 7th order ) Lagrange polynomial interpolation; ( ii ) rigid-body head movement correction to the first frame of data , using quintic ( 5th order ) polynomial interpolation to estimate the realignment parameters ( 3 displacements and three rotations ) ; ( iii ) obliquity transform to the structural image; ( iv ) affine co-registration to the skull-stripped structural image using a gray matter mask; ( v ) nonlinear warping to MNI space ( MNI152 template ) with AFNI 3dQwarp; ( vi ) spatial smoothing ( 6 mm FWHM ) ; and ( vii ) a within-run intensity normalization to a whole-brain median of 1000 . Core preprocessing was followed by denoising steps to further remove motion-related and other artifacts . Denoising steps included: ( viii ) wavelet time series despiking ( ‘wavelet denoising’ ) ; ( ix ) confound signal regression including the six motion parameters estimated in ( ii ) , their first order temporal derivatives , and ventricular cerebrospinal fluid ( CSF ) signal ( referred to as 13-parameter regression ) . The wavelet denoising method has been shown to mitigate substantial spatial and temporal heterogeneity in motion-related artifact that manifests linearly or non-linearly and can do so without the need for data scrubbing ( Patel et al . , 2014 ) . Data scrubbing ( i . e . volume censoring ) cannot be used in our time-series-based analyses here as such a procedure breaks up the temporal structure of the time-series in such a way that invalidates estimation of the Hurst exponent ( H ) that examine long-memory characteristics . Wavelet denoising is implemented with the Brain Wavelet toolbox ( http://www . brainwavelet . org ) . The 13-parameter regression of motion and CSF signals was achieved using AFNI 3dBandpass with the –ort argument . To further characterize motion and its impact on the data , we computed FD and DVARS ( Power et al . , 2012 ) . Between-group comparisons showed that all groups were similar with respect to head motion as measured by meanFD with no diagnosis ( F ( 3 , 106 ) = 1 . 77 , p=0 . 18 ) or sex ( F ( 3 , 106 ) = 0 . 51 , p=0 . 47 ) main effects or sex-by-diagnosis interaction ( F ( 3 , 106 ) = 1 . 10 , p=0 . 29 ) . All groups showed average meanFD of less than 0 . 2 mm ( see Table 3 ) . Mean time-series for each of the 180 parcels within the Human Connectome Project Multimodal Parcellation ( HCP-MMP ) ( Glasser et al . , 2016 ) were extracted from the final preprocessed data , to estimate H . The estimation of H utilizes a discrete wavelet transform and a model of the time-series as fractionally integrated processes ( FIP ) and is estimated using maximum likelihood estimation . This method utilizing the FIP model for estimating H differs from our prior work ( Lai et al . , 2010 ) , which used a model of fractional Gaussian noise ( fGn ) . fGn is one type of process subsumed under the FIP model . However , the fGn model has the limitation of assuming that the BOLD time-series is stationary and also limits the upper bound of H at 1 . In practice , we have seen that the upper bound of H = 1 from the fGn model results in ceiling effects for many brain regions and subjects . Thus , to remove the assumption of stationarity and upper bound of H = 1 , the FIP model offers more flexibility and potentially added sensitivity due to better estimation of between-subject variability when estimates are near or exceed H = 1 . When H > 1 the time-series is considered non-stationary and has long memory characteristics ( e . g . , is fractal ) . H is computed using the nonfractal MATLAB toolbox written by one of the co-authors ( WY ) ( https://github . com/wonsang/nonfractal ) . The specific function utilized is bfn_mfin_ml . m function with the ‘filter’ argument set to ‘haar’ and the ‘ub’ and ‘lb’ arguments set to [1 . 5 , 10] and [−0 . 5 , 0] , respectively . After H was estimated for each of the 180 HCP-MMP parcels , we used a general linear model to test for sex-by-diagnosis interactions as well as main effects of Sex and Diagnosis in H . These models also incorporated meanFD and FIQ as covariates of no interest . Multiple comparison correction was achieved using an FDR q < 0 . 05 threshold . Visualization of effect sizes for figures was achieved using the ggseg library in R ( https://github . com/LCBC-UiO/ggseg ) . In addition to mass-univariate analysis , we also utilized multivariate partial least squares ( PLS ) analysis ( Krishnan et al . , 2011 ) to highlight distributed neural systems that capture the effect of a sex-by-diagnosis interaction . This analysis was implemented with code from the plsgui MATLAB toolbox ( http://www . rotman-baycrest . on . ca/pls/ ) . A matrix with participants along the rows and all 180 HCP-MMP parcels along with columns was input as the primary neuroimaging matrix for PLS . We also inserted a vector describing the sex-by-diagnosis contrast as the matrix to relate to the neuroimaging matrix . This vector describing the sex-by-diagnosis interaction was computed by matrix multiplication of the contrast vector of [1 , -1 , -1 , 1] to a design matrix that was set up with columns defining TD males , autism males , TD females , and autism females , respectively . The PLS analysis was run with 10 , 000 permutations to compute p-values for each latent-variable ( LV ) pair and 10 , 000 bootstrap resamples in order to compute bootstrap ratios ( BSR ) to identify brain regions of importance for each LV pair . To isolate specific brain regions of importance for a statistically significant LV , we selected the top 20th percentile of brain regions ranked by BSR . Relationships between H and camouflaging were conducted within autistic males and females separately . Pearson’s correlations were used to estimate the strength of the relationship and groups were compared on the strength of the relationship using Fisher’s r-to-z transform as implemented with the paired . r function in the psych library in R . Camouflaging ( consciously or unconsciously compensating for and/or masking difficulties in social–interpersonal situations ) was operationalized as prior work ( Lai et al . , 2017b; Lai et al . , 2019 ) : the discrepancy between extrinsic behavioral presentation in social–interpersonal contexts and the person’s intrinsic status . We used both the AQ score and RMET correct score as reflecting intrinsic status ( i . e . self-rated dispositional traits and performance-based socio-cognitive/mentalizing capability ) , and the ADOS Social-Communication total score as reflecting extrinsic behavioral presentation . The three scores were first standardized ( SADOS , SAQ and SRMET ) within our sample of autistic men and women by mean-centering ( to the whole autism sample in this study ) and scaling ( i . e . divided by the maximum possible score of each ) to generate uniformly scaled measures that can be arithmetically manipulated . The first estimate of camouflaging was quantified as the difference between self-rated autistic traits and extrinsic behaviors ( CF1 = SAQ − SADOS ) , and the second estimate between mentalizing ability and extrinsic behaviors ( CF2 = −SRMET − SADOS ) . Then , using principal component analysis , the first principal component score of CF1 and CF2 ( accounting for 86% of the total variance ) was taken as a single , parsimonious measure of camouflaging for all subsequent analyses . This method was utilized in order to be consistent with prior work which computed the camouflaging metric in an identical fashion ( Lai et al . , 2017b; Lai et al . , 2019 ) . This measure should be interpreted by relative values ( i . e . higher scores indicate more camouflaging ) rather than absolute values . This operationalization only allows for estimating camouflaging in autistic individuals in our cohort , as it partly derives from the ADOS score which was not available in TD participants . This approach remains informative , as qualitative studies suggest that camouflaging in autism can be different from similar phenomenon ( e . g . impression management ) in TD individuals ( Bargiela et al . , 2016; Hull et al . , 2017 ) . All in vivo studies in mice were conducted in accordance with the Italian law ( DL 116 , 1992 Ministero della Sanità , Roma ) and the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health . Animal research protocols were also reviewed and consented to by the animal care committee of the Istituto Italiano di Tecnologia . The Italian Ministry of Health specifically approved the protocol of this study , authorization no . 852/17 to A . G . All surgical procedures were performed under anesthesia . Six to eight week-old adult male C57Bl6/J mice ( Jackson Laboratories; Bar Harbor , ME , USA ) were anesthetized with isoflurane ( isoflurane 4% ) and head-fixed in a mouse stereotaxic apparatus ( isoflurane 2% , Stoelting ) . Viral injections were performed with a Hamilton syringe mounted on Nanoliter Syringe Pump with controller ( KD Scientific ) , at a speed of 0 . 05 μl/min , followed by a 5–10 min waiting period , to avoid backflow of viral solution and unspecific labeling . Viral suspensions were injected bilaterally in PFC using the following coordinates , expressed in millimeter from bregma: 1 . 7 from anterior to posterior , 0 . 3 lateral , −1 . 7 deep . The inhibitory DREADD hM4Di was transduced using an AAV8-hSyn-hM4D ( Gi ) -mCherry construct . Control animals were injected with a control AAV8-hSyn-GFP virus ( www . addgene . com ) . These viral suspensions were injected using a 0 . 3 μL injection volume in n = 15 hM4Di DREADD and n = 19 SHAM mice , respectively . The excitatory DREADD hM3Dq was transduced using an AAV8-CamkII-hM3D ( Gq ) -mCherry construct . Control animals for this experiment were injected with a control AAV8-CamkII-GFP construct . This set of injection were carried out using a 1 μL injection volume in n = 17 hM3Dq DREADD and n = 19 SHAM mice , respectively . We waited at least 3 weeks to allow for maximal viral expression . The animal preparation protocol for mouse rsfMRI scanning was previously described in detail ( Bertero et al . , 2018 ) . Briefly , mice were anesthetized with isoflurane ( 5% induction ) , intubated and artificially ventilated ( 2% maintenance ) . Then isoflurane was discontinued and substituted with halothane ( 0 . 75% ) , a sedative that preserves cerebral blood flow auto-regulation and neurovascular coupling ( Gozzi et al . , 2007 ) . Functional data acquisition commenced 30 min after isoflurane cessation . CNO ( 2 mg/kg for hM4Di and 0 . 5 mg/kg for hM3Dq ) was administered i . v . after 15 min from the beginning of the acquisition both in virally transduced animals and in sham mice . Raw mouse rsfMRI data was preprocessed as described in previous work ( Gutierrez-Barragan et al . , 2019; Liska et al . , 2015 ) . Briefly , the initial 120 volumes of the time-series were removed to allow for T1 and gradient equilibration effects . Data were then despiked , motion corrected and spatially registered to a common reference template . Motion traces of head realignment parameters ( three translations + three rotations ) and mean ventricular signal ( corresponding to the averaged BOLD signal within a reference ventricular mask ) were used as nuisance covariates and regressed out from each time course . All rsfMRI time-series also underwent band‐pass filtering to a frequency window of 0 . 01–0 . 1 Hz and spatial smoothing with a full width at half maximum of 0 . 6 mm . The experimental design of the study allowed for computation of H during time-windows in the rsfMRI scan before drug injection ( i . e . ‘Baseline’ ) , a transition phase where the drug begins having its effect ( i . e . ‘Transition’ ) , and a treatment phase when the drug is thought to have its optimal effect ( i . e . ‘Treatment’ ) . Analysis of condition , treatment phase , time , and all interactions between such factors was achieved using a sliding window analysis . Each window was 512 volumes in length and the sliding step was 1 vol . H is computed at each window and results in an H time-series . The H time-series is used as the dependent variable in a linear mixed effect model ( i . e . using the lme function within the nlme library in R ) with fixed effects of condition , time , treatment phase , and all 2-way and 3-way interactions between such factors as well as a factor accounting for scan day . Random effects in the model included time within mouse as well as treatment phase within mouse , all modeled with random intercepts and slopes . This omnibus model was utilized to examine a 3-way interaction between condition , time , and treatment phase . If this interaction was present , we then split the data by the 3 levels of the treatment phase ( e . g . , Baseline , Transition , and Treatment ) , in order to examine the main effect of condition or the condition*time interaction . Plots of the data indicate each mouse ( grey lines in Figure 3 ) as well as group trajectories for each phase , with all trajectories estimated with a generalized additive model smoother applied to individual mice and group trajectories . The recurrent network model we use represents a standard cortical circuit incorporating integrate-and-fire excitatory and inhibitory spiking neurons that interact through recurrent connections and receive external inputs ( both a sensory driven thalamic input and a sensory unrelated intracortical input , see Figure 1A ) . The network structure and parameters of the recurrent network model are the same ones used in Cavallari et al . , 2014 with conductance-based synapses ( for full details see Cavallari et al . , 2014 ) . The network is composed of 5000 neurons , of which 4000 are excitatory ( i . e . they form AMPA-like excitatory synapses with other neurons ) and 1000 inhibitory ( forming GABA-like synapses ) . Neurons are randomly connected with a connection probability between each pair of neurons of 0 . 2 . Both populations receive two different types of external Poisson inputs: a constant-rate thalamic input and an intracortical input generated by an Ornstein-Uhlenbeck ( OU ) process with zero mean . A description of the baseline reference parameters used in simulations is given in Table 4 . The LFP is computed as the sum of absolute values of AMPA and GABA postsynaptic currents on excitatory cells ( Mazzoni et al . , 2008; Mazzoni et al . , 2015 ) . This simple estimation of LFPs was shown to capture more than 90% of variance of both experimental data recorded from cortical field potentials and of simulated data from a complex three-dimensional model of the dipoles generated by cortical neurons ( Mazzoni et al . , 2008; Mazzoni et al . , 2015; Barbieri et al . , 2014 ) . We changed the E:I ratio by independently varying the strengths of the inhibitory ( gI ) and excitatory ( gE ) synaptic conductances . We called g the relative ratio between inhibitory and excitatory conductances ( g=gI/gE ) . We present results of simulations for two levels of strength of thalamic input ( υ0 = 1 . 5 spikes/second and υ0 = 2 spikes/second ) , and we verified that our results hold qualitatively for a wider range of input levels ( 1 . 5 to 4 spikes/second ) . For the simulations used to compute H and 1/f slope of LFPs , we simulated a 10 s stretch of network activity from which we extracted a 10 s LFP time series used to compute H and 1/f slopes for each individual value of g ( Figure 1B ) . To estimate power spectral density ( PSD ) we computed the Fast Fourier Transform with the Welch’s method , dividing the data into ten overlapping segments with 50% overlap . 1/f slopes were computed with least-squares regressions predicting log power with log frequency . A piece-wise regression was applied to fit two line segments to the PSD – one segment to a low frequency region from 1 to 30 Hz and a second segment to a high-frequency region from 30 to 100 Hz . As a basis for our model translating BOLD from LFP data , we note that prior studies with simultaneous electrophysiological and fMRI recordings in non-human primates have established that BOLD signal amplitude is more closely correlated with LFP than with any other type of neuronal events , such as spikes ( Logothetis et al . , 2001; Magri et al . , 2012 ) . Similarly , simultaneous electroencephalogram ( EEG ) /fMRI studies in humans have found that the BOLD correlates with the EEG ( Scheeringa et al . , 2011; Scheeringa et al . , 2009 ) , which in turns correlates strongly with the LFP ( Whittingstall and Logothetis , 2009 ) . Importantly , the BOLD amplitude at any given time has been found to correlate preferentially with the power of high frequency bands . In particular , the BOLD amplitude correlates strongly with the gamma ( 40–100 Hz ) band . However , the power distribution across frequency bands carries complementary information about the BOLD signal , meaning that each band contributes to the prediction of BOLD and predicting the BOLD signal directly from a wide band ( i . e . the whole LFP spectrum ) leads to poorer predictions of BOLD ( Logothetis et al . , 2001; Magri et al . , 2012; Schölvinck et al . , 2010; Scheeringa et al . , 2011; Goense and Logothetis , 2008; Kilner et al . , 2005 ) . To account for these empirical observations , we have developed a model of BOLD signal that integrates contributions from different bands with a preferential contribution from high frequency bands . To compute the simulated BOLD through the convolutions of the simulated LFPs , we needed to generate longer time-series than the initial 10 seconds simulated for LFPs . However , it was unfeasible to simulate very long BOLD time-series due to limitations on computational resources . We thus created , from the LFP data used for evaluations of individual g values , aggregated LFP time-series corresponding to different intervals of g ( rather than individual values of g ) , as follows . The set of 10-second LFP time-series was divided into 3 equi-populated groups of g: g < 7 . 5 , 7 . 5 < g < 11 and g > 11 . A concatenated LFP time-series was created for each group by randomly concatenating 20 LFP traces , which provided 200-second LFP signals . Low frequencies of the concatenated data were log-log linearly extrapolated based on the low-frequency slopes obtained in LFP log-log linear piecewise fitting . To account for statistical variability , the process of concatenating data was repeated 20 times for each group of g , randomly changing the order in which the individual LFP traces were combined within the group . Once concatenated LFP time-series data was simulated , we compute the simulated BOLD time-series as the LFP data convolved not only with a hemodynamic response function ( HRF ) , as in standard network models ( Deco et al . , 2004; Wijeakumar et al . , 2017; Buss et al . , 2014 ) , but also with a high-pass filter ( HPF ) that gives more predictive power to higher LFP frequencies ( Figure 2—figure supplement 1B ) . We have tested different parameters of the HPF , checking that changing the parameters produce qualitatively similar results and a monotonic correspondence between H of simulated LFP and H of simulated BOLD , and we opted for a HPF with a cutoff frequency ( the frequency where the response is lowered by 3 dB ) of 12 . 5 Hz and with a peak response at 20 Hz . The effect of the HPF was to attenuate low frequencies of the BOLD power distribution , partly compensating the low-frequency enhancement of HRFs , and to shift the peak frequency of BOLD power to 0 . 03 Hz , a value much closer to the peak frequency found in our real BOLD data and in most BOLD studies ( Alcauter et al . , 2015; Allen et al . , 2011 ) with respect to the one that would have been obtained without convolution ( see Figure 2—figure supplement 1 ) . To simulate BOLD response from the LFP data generated from the recurrent model , within the frequency domain we multiplied the LFP spectrum with spectra of the high-pass filter ( HPF ) and the hemodynamic response function ( HRF ) , as follows:FFTBOLD=FFTLFPFFT ( HPF ) FFTHRF+ η Where FFT is the fast Fourier transform operation and η is a constant white noise term . This noise term summarizes neurovascular relationship at frequencies not observable because they are faster than the BOLD acquisition frequencies . We assumed that the amplitude of η was very small and we assigned small values to this noise term . We checked that the exact value of amplitude of η , or the spectral profile of this noise term ( for example , simulating 1/f noise instead of white noise ) , did not alter the monotonic relationship between the simulated H of BOLD and LFP . Finally , the simulated BOLD time-series data was produced by applying the inverse Fourier transform of FFTBOLD and then downsampling the resulting signal to a lower sampling rate , similar to that used in BOLD experiments ( e . g . , 0 . 5 Hz ) . The HRF used for simulating BOLD was the HRF from Magri et al . , 2012 , but similar results were obtained when using a canonical HRF ( see Figure 2—figure supplement 2 ) . To test hypotheses regarding cell types that may be affected by androgen influence , we examined genes linked to autism via rare de novo protein truncating variants that are enriched for expression in specific cell types ( Satterstrom et al . , 2020 ) . Of the 102 genes reported by Satterstrom et al . , we split these lists by enrichments in early excitatory neurons ( C3 ) , MGE derived cortical interneurons ( C16 ) , microglia ( C19 ) , and astrocytes or oligodendrocyte precursor cells ( C4 ) . In addition to high risk mutations linked to autism , we additionally used a list of genes differentially expressed ( DE ) in different cell types within post-mortem prefrontal and anterior cingulate cortex tissue of autistic patients ( Velmeshev et al . , 2019 ) . These DE gene lists were split into cell types , and we examined DE genes in any excitatory neuronal cell class ( L2/3 , L4/L5/6 ) , inhibitory cell classes ( IN-PV , IN-SST , IN-VIP , IN-SV2C ) , microglia , astrocytes ( AST-PP , AST-FB ) , and oligodendrocytes . To test the question of whether cell type autism-associated gene lists were enriched for genes known to be differentially expressed by DHT , we used a previous DE gene list from an RNA-seq dataset of DHT administration to human neuronal stem cells ( Lombardo et al . , 2018b ) . Custom code was utilized to compute enrichment odds ratios and hypergeometric p-values for each enrichment test with different cell type autism-associated lists . The background total for these tests was the total number of genes considered in the original DHT-administration dataset ( 13 , 284 ) . To test how the DHT-sensitive and autism-associated genes in excitatory neurons are expressed across the human adult brain , we used whole-brain maps of expression for each gene in MNI space from the Allen Institute Human Brain Atlas ( Hawrylycz et al . , 2012 ) . Maps for each gene were downloaded from the Neurosynth website ( https://neurosynth . org/genes/ ) and then submitted to a one-sample t-test in SPM12 , with a threshold of FDR q < 0 . 01 . Tidy data and analysis code are available at https://github . com/IIT-LAND/ei_hurst; Trakoshis , 2020a; copy archived at https://github . com/elifesciences-publications/ei_hurst . Source code of the recurrent network model is available at https://github . com/pablomc88/EEG_proxy_from_network_point_neurons; Trakoshis , 2020b; copy archived at https://github . com/elifesciences-publications/EEG_proxy_from_network_point_neurons . Raw RNA-seq data used to identify genes differentially expressed by DHT can be found in Gene Expression Omnibus ( GSE86457 ) . Data for the Allen Institute Human Brain Atlas can be found here: https://human . brain-map . org . Mapping of this data to MNI space can be found at the Neurosynth website ( https://neurosynth . org/genes/ ) .

Autism is a condition that is usually diagnosed early in life that affects how a person communicates and socializes , and is often characterized by repetitive behaviors . One key theory of autism is that it reflects an imbalance in levels of excitation and inhibition in the brain . Excitatory signals are those that make other brain cells more likely to become active; inhibitory signals have the opposite effect . In non-autistic individuals , inhibitory activity outweighs excitatory activity . In people with autism , by contrast , an increase in excitatory activity is believed to produce an imbalance in excitation and inhibition . Most of the evidence to support this excitation-inhibition imbalance theory has come from studies of rare mutations that cause autism . Many of these mutations occur on the sex chromosomes or are influenced by androgen hormones ( hormones that usually play a role on typically male traits ) . However , most people with autism do not possess these particular mutations . It was thus unclear whether the theory could apply to everyone with autism or , for example , whether it may better apply to specific groups of individuals based on their sex or gender . This is especially important given that about four times as many men and boys compared to women and girls are diagnosed with autism . Trakoshis , Martínez-Cañada et al . have now found a way to ask whether any imbalance in excitation and inhibition in the brain occurs differently in men and women . Using computer modeling , they identified a signal in brain scans that corresponds to an imbalance of excitation and inhibition . After showing that the technique works to identify real increases in excitation in the brain scans of mice , Trakoshis , Martínez-Cañada et al . looked for this signal , or biomarker , in brain scans of people with and without autism . All the people in the study identified with the gender that matched the sex they were assigned at birth . The results revealed differences between the men and women with autism . Men with autism showed an imbalance in excitation and inhibition in specific ‘social brain' regions including the medial prefrontal cortex , but women with autism did not . Notably , many of these brain regions are strongly affected by androgen hormones . Previous studies have found that women with autism are sometimes better at hiding or ‘camouflaging’ their difficulties when socializing or communicating than men with autism . Trakoshis , Martínez-Cañada et al . showed that the better a woman was at camouflaging her autism , the more her brain activity in this region resembled that of non-autistic women . Excitation-inhibition imbalance may thus affect specific brain regions involved in socializing and communication more in men who have autism than in women with the condition . Balanced excitation and inhibition in these brain areas may enable some women with autism to camouflage their difficulties socializing or communicating . Being able to detect imbalances in activity using standard brain imaging could be useful for clinical trials . Future studies could use this biomarker to monitor responses to drug treatments that aim to adjust the balance between excitation and inhibition .

2020

Intrinsic excitation-inhibition imbalance affects medial prefrontal cortex differently in autistic men versus women

Information is carried in the brain by the joint spiking patterns of large groups of noisy , unreliable neurons . This noise limits the capacity of the neural code and determines how information can be transmitted and read-out . To accurately decode , the brain must overcome this noise and identify which patterns are semantically similar . We use models of network encoding noise to learn a thesaurus for populations of neurons in the vertebrate retina responding to artificial and natural videos , measuring the similarity between population responses to visual stimuli based on the information they carry . This thesaurus reveals that the code is organized in clusters of synonymous activity patterns that are similar in meaning but may differ considerably in their structure . This organization is highly reminiscent of the design of engineered codes . We suggest that the brain may use this structure and show how it allows accurate decoding of novel stimuli from novel spiking patterns . Noise is prevalent in the nervous system , from ion channels through synapses and single neurons , and up to the system level ( Mainen and Sejnowski , 1995; Schneidman et al . , 1998; Rieke et al . , 1999; Osborne et al . , 2005; Faisal et al . , 2008; Ala-Laurila et al . , 2011 ) . Consequently , even when presented with identical stimuli , the brain may receive different spiking patterns from the sensory neurons ( Mainen and Sejnowski , 1995; Berry et al . , 1997; de Ruyter van Steveninck et al . , 1997; Reich et al . , 2001 ) . The nature of neural noise thus determines how information is encoded in the brain ( Borst and Theunissen , 1999; Stocker and Simoncelli , 2006; Cafaro and Rieke , 2010; Rolls and Treves , 2011 ) , and how it might be read out from neural activity ( Warland et al . , 1997; Dan et al . , 1998; Vargas-Irwin et al . , 2010 ) . To correctly decode , the brain must overcome this noise ( Osborne et al . , 2005; Sreenivasan and Fiete , 2011 ) . In engineering , codes are designed to solve this problem by choosing codewords that are far apart in the space of patterns , relative to the typical noise levels . That way , if noise corrupts a message , it would still be easily distinguishable from the noisy variants of other codewords ( Cover and Thomas , 1991; Sreenivasan and Fiete , 2011; Curto et al . , 2013 ) . It is not clear however , how this issue is resolved in the brain , or how it affects the design of the neural code , where information is carried by the joint activity patterns of large groups of noisy neurons ( Nicolelis et al . , 1995; Maynard et al . , 1999; Mazor and Laurent , 2005; Fujisawa et al . , 2008; Pillow et al . , 2008; Truccolo et al . , 2010; Ganmor et al . , 2011a; Harvey et al . , 2012 ) . It is clear that the nature of correlations among neurons is central in shaping the code's capacity and content , in particular in the context of noise ( Zohary et al . , 1994; Zemel et al . , 1998; Abbott and Dayan , 1999; Diesmann et al . , 1999; Sompolinsky et al . , 2001; Schneidman et al . , 2006; Pillow et al . , 2008; Shlens et al . , 2009; Ganmor et al . , 2011b; Schwartz et al . , 2012; Granot-Atedgi et al . , 2013 ) . However , the functional role of these correlations in encoding information by large populations has been heavily debated ( Nirenberg et al . , 2001; Amari et al . , 2003; Pola et al . , 2003; Schneidman et al . , 2003a , 2006; Averbeck et al . , 2006; Tkacik et al . , 2006; de la Rocha et al . , 2007; Pillow et al . , 2008; Ecker et al . , 2010; Ohiorhenuan et al . , 2010; Oizumi et al . , 2010; Ganmor et al . , 2011a ) , partly because of the difficulty to study them directly at the network level and the limitations of generalizing from small groups of cells to large ones . To uncover the structure and content of neural population codebooks , we must be able to quantify the similarity between population activity patterns . In other words , we need to understand network noise well enough to know which patterns are likely to be interchanged in encoding the same stimulus . Several studies suggested intuitive and computationally efficient measures of spike train similarity ( Victor and Purpura , 1997; van Rossum , 2001; Houghton and Sen , 2008 ) . However , they mostly focused on single neurons and often assumed the functional form of similarity between responses . Despite their computational simplicity , it is not clear whether such simple metrics can adequately represent similarity between neural responses . The Hamming distance , for example , is a very intuitive measure of similarity between population activity patterns , simply defined as the number of neurons that switch state from spiking to silence and vice versa . Yet , it measures similarity in form , or syntactic similarity , but not necessarily similarity in meaning , that is , semantic similarity ( Huth et al . , 2012 ) . This would be analogous to suggesting that ‘night’ and ‘light’ are similar in meaning because they differ by just one letter . Extending such metrics to large populations requires additional assumptions on the way the similarity measures between different cells should be combined . We suggest here that the semantic similarity between neural activity patterns can be measured by comparing what each pattern tells the brain about the stimulus . We demonstrate that in the vertebrate retina , responding to natural and artificial stimuli , we can estimate this similarity between any pair of population responses , using models of population encoding noise . Using this similarity structure , or ‘thesaurus’ , we can map the semantic organization and content of the population codebook and show how it enables the accurate decoding of novel population patterns . The encoding noise determines which population patterns may be used to encode the same stimulus and thus implicitly defines similarity between patterns . We submit that two neural activity patterns are similar , in terms of their meaning , to the extent that they convey similar information to the brain . This constitutes semantic similarity between population patterns , which does not necessarily rely on any syntactic similarity ( for example total spike count ) . Intuitively , this is analogous to synonyms in natural language , which may not look similar , but still have the same meaning ( Pereira et al . , 1993 ) . The meaning of a neural activity pattern r is given by what it tells the brain about its corresponding input , or in the case of sensory systems , the potential stimuli that could have resulted in that response . Since the mapping between stimulus and response is probabilistic , the meaning of r is given by the probability distribution over stimuli conditioned on that response , namely P ( s|r ) . We therefore defined the distance between two neural population activity patterns rk and rl as the dissimilarity between P ( s|rk ) and P ( s|rl ) , ( 1 ) d ( rk , rl ) =DJS ( P ( s|rk ) ||P ( s|rl ) ) , where DJS is the Jensen-Shannon divergence between the distributions ( see ‘Materials and methods’ ) —a symmetric measure of dissimilarity that varies between 0 , for identical distributions , and 1 , for non-overlapping distributions . Here , 0 would imply that the two activity patterns are perfect ‘synonyms’ and have identical meaning , whereas 1 would imply completely different meaning . Rather than directly estimating P ( s|r ) , which is challenging both experimentally and statistically , we can instead estimate P ( r|s ) and infer P ( s|r ) through Bayes' rule , namely P ( s|r ) = P ( r|s ) P ( s ) /P ( r ) , where P ( s ) is the distribution over stimuli ( set to be a uniform distribution over all video frames used in the experiment ) , and P ( r ) is the distribution of the neural responses ( calculated by marginalizing over stimuli , P ( r ) =∑sP ( s ) P ( r|s ) ) . Our experimental design , in which we presented many repeats of the same video , allowed us to draw hundreds of samples from P ( r|s ) , for every s . However , due to the exponential number of possible population activity patterns one cannot directly sample this distribution in full even for groups of 20 neurons . Instead , we construct accurate models of population encoding noise , which we then use to estimate d ( ri , rj ) for any pair of population patterns . The properties of encoding noise , and in particular , the magnitude and importance of correlations between neurons in encoding a stimulus , commonly known as ‘noise correlations’ , have been heavily debated ( Nirenberg et al . , 2001; Schneidman et al . , 2003a; Ohiorhenuan et al . , 2010; Oizumi et al . , 2010 ) . For pairs of neurons , average noise correlations are typically weak ( Cafaro and Rieke , 2010; Ecker et al . , 2010 ) , implying that pairs of neurons are not far from being conditionally independent given the stimulus . If groups of cells were encoding information independently , that is , if P ( r|s ) =ΠiP ( ri|s ) for large populations , then P ( r|s ) would be defined by the individual and independent noise of each neuron and would be easily learned from the noise of each neuron P ( ri|s ) . However , weak pairwise correlations do not imply that large groups are conditionally independent in encoding stimuli ( Schneidman et al . , 2006; Pillow et al . , 2008; Vidne et al . , 2012; Granot-Atedgi et al . , 2013 ) . Moreover , noise correlations are often estimated on average , over a range of different stimuli , and it is not clear what low average noise correlations imply for population encoding of specific stimuli . We therefore estimated the noise correlations at the population level , for each stimulus s . Specifically , we quantified the correlation of the population in encoding the stimulus s , by the multi-information , I ( r|s ) =H[Pind ( r|s ) ]−H[P ( r|s ) ] ( Amari , 2001; Schneidman et al . , 2003b ) , where H[Pind ( r|s ) ] is the entropy assuming neurons are independent given the stimulus—Pind ( r|s ) =ΠiP ( ri|s ) , and H[P ( r|s ) ] is the entropy of the joint population response to the stimulus s ( estimated following Strong et al . , 1998 ) . We note that I ( r|s ) measures the total correlations of all orders among the cells . We found that most stimuli did not evoke a substantial response from the retina , which gave rise to low noise correlation in the network for these stimuli . However , when something ‘interesting’ happened in the video and the ganglion cells increased their firing rates , we found a sharp increase in the degree of network noise correlations for both natural and artificial videos ( Figure 2A , B; see Figure 2—figure supplement 1 for analysis of sampling properties ) . Thus , while on average the network noise correlations may be weak , the population was strongly correlated and far from conditionally independent in response to interesting stimuli . In other words , for these stimuli , the variability or noise at the level of the network is significantly reduced compared to what would be expected from the apparent noise level of individual cells . 10 . 7554/eLife . 06134 . 004Figure 2 . Strong noise correlations , at the population level , at interesting times in the video . ( A ) Population noise correlation , measured by the multi-information over the conditional population responses , at each point in time in response to an artificial video . Thin gray lines correspond to individual groups of 20 cells; average over groups is shown in purple . ( B ) Population noise correlation as a function of average population firing rate for one representative group . Interesting events in the video evoke a vigorous response by the retina , characterized by strong network correlations . ( C ) Distribution of spike counts across different repeats of the same stimulus for the time point marked by black dot in A . Purple dots correspond to empirical estimates , gray line is what we would expect if neurons were conditionally independent , given the stimulus; and red line is the prediction of the maximum entropy pairwise model . ( D–F ) Same as A–C but for a natural video clip . DOI: http://dx . doi . org/10 . 7554/eLife . 06134 . 00410 . 7554/eLife . 06134 . 005Figure 2—figure supplement 1 . Accurate sampling of entropy . ( A ) Pairwise maximum entropy upper bound on entropy ( gray ) and the extrapolation corrected noise entropy ( Treves and Panzeri , 1995; Strong et al . , 1998 ) ( purple ) are shown as a function of the naive noise entropy estimate . Each dot corresponds to the distribution at a single time point in the artificial video; black line marks identity . The bias corrections are on the order of a few percent at most . ( B ) Same as A , but for data taken from the natural video data set ( 693 repeats ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06134 . 00510 . 7554/eLife . 06134 . 006Figure 2—figure supplement 2 . Log likelihood ratio of the pairwise model and the conditionally independent model . ( A ) A pairwise ( second order maximum entropy as in the text ) and an independent model ( product of marginals ) was fit to each time point ( equivalently stimulus ) in the artificial video using 90% of video repeats . The log likelihood ratio of the two models ( log[LikelihoodPairwise/LikelihoodCond . − Indep . ] ) is plotted as a function of time ( equivalently stimulus ) , for the held out 10% of repeats . The likelihoods are similar much of the time , corresponding to low firing epochs , but for many points in time the likelihood of the pairwise model can be orders of magnitude higher . ( B ) Same as A , for the natural video . DOI: http://dx . doi . org/10 . 7554/eLife . 06134 . 006 Indeed , conditionally independent models of population encoding ( which assume no noise correlations ) fail exactly at such interesting stimuli ( Pillow et al . , 2008; Granot-Atedgi et al . , 2013 ) , reflecting that in order to accurately characterize the population responses and the noise , we need a model that takes into account correlations between the cells that depend on the stimulus . We extended the framework of maximum entropy models for neural activity patterns ( Schneidman et al . , 2006; Shlens et al . , 2009 ) , to the population responses to each stimulus ( Granot-Atedgi et al . , 2013 ) : for each stimulus , s , we fit the minimal model that has the same stimulus dependent firing rates and pairwise noise correlations observed over the repeats of that stimulus . This stimulus-dependent pairwise maximum entropy model is given by ( 2 ) P ( 2 ) ( r|s ) =1Z ( s ) exp ( ∑iαi ( s ) ri+∑i<jβij ( s ) rirj ) , where αi ( s ) and βij ( s ) were fit to obey the constraints . We found that P ( 2 ) ( r|s ) gave relatively accurate models of the distribution of patterns that the population displayed across repeats , for the different stimuli ( Figure 2—figure supplement 2 ) , in agreement with Granot-Atedgi et al . ( 2013 ) . In particular , they captured accurately the strong network noise correlations ( Figure 2C ) . We then use the distributions P ( 2 ) ( r|s ) that describe the population encoding noise , to estimate the dissimilarity , d ( ri , rj ) , between any pair of patterns over the population in a reliable and precise manner . Since there are over a million possible population activity patterns for 20 neurons , we present here only the similarity matrix of all the population patterns that were observed in the test data ( half of the data , which was randomly selected and not used to learn the similarity between activity patterns ) . If stimuli were represented by overall population rates ( Bohte et al . , 2000; Amari et al . , 2003; Loebel et al . , 2007; Schwartz et al . , 2012 ) , then responses containing the same number of spiking neurons would be similar , and thus would have a small d between them . Figure 3A shows the similarity matrix d ( ri , rj ) for one representative group of 20 cells , where matrix rows ( and columns ) stand for individual population patterns and are ordered according to the total number of spikes in each response . The lack of structure in the matrices for both artificial and natural stimuli shows that similar spike counts do not imply similar meaning . 10 . 7554/eLife . 06134 . 007Figure 3 . The population code of the retina is comprised of clusters of responses with highly similar meaning . ( A ) Top: Similarity matrices of the population responses of representative groups of 20 neurons to an artificial ( left ) and natural ( right ) video . Each entry in the matrix corresponds to the similarity , d ( see text ) , between two population responses observed in the test data ( responses shown at bottom ) . Matrix rows ( and columns ) are ordered by total spike count in the population responses . Bottom: The population responses corresponding to the entries in the matrix; black ticks represent spikes . Each column is a population activity pattern corresponding to the matrix column directly above . Blue lines mark borders between different clusters . The lack of structure in the matrices implies that population responses with similar spike counts do not carry similar meanings . ( B ) Same as A , only here the matrix is clustered into 120 clusters . Matrix rows ( and columns ) are ordered such that responses from the same cluster appear together . A clustered organization of the population code is clearly evident . ( C ) Same as B , but using the Hamming distance between population responses , instead of the similarity measure d . A simple measure of syntactic similarity does not reveal the underlying clustered organization of the code . DOI: http://dx . doi . org/10 . 7554/eLife . 06134 . 00710 . 7554/eLife . 06134 . 008Figure 3—figure supplement 1 . Clustered organization is highly significant . ( A ) The similarity matrix used in Figure 3 in the main text was first randomly shuffled and only then clustered . Clearly the grouping structure has disappeared , suggesting the structure is a property of the code's organization and not the values in the matrix or the power of the clustering algorithm . ( B ) For each of the ten 20 neuron groups we calculated the C Index ( Hubert and Schultz , 2011 ) ( left , labeled ‘Data’ ) , which measures goodness of clustering; a smaller C Index corresponds to better clustering . For each matrix , we then measured the C Index for 100 randomly shuffled versions and took the minimal value of all shuffled matrices ( right , labeled ‘Shuffled’ ) . None of the shuffled values came close to the real values , indicating a p-value smaller than 0 . 01 for each matrix . ( C ) Correlation coefficient between similarity matrices estimated using different fractions of data . Overall we had 641 repeats of the artificial video . Using 70% of the repeats , we are at ∼0 . 9 correlation with our estimates using all available repeats ( Data are for the same matrix shown in Figure 3 of main text ) . ( D–F ) Same as A–C but for data taken from the natural video data set . DOI: http://dx . doi . org/10 . 7554/eLife . 06134 . 00810 . 7554/eLife . 06134 . 009Figure 3—figure supplement 2 . Semantic similarity between population patterns can be explained by a simple local similarity measure , but not by a global similarity measure . ( A ) We sought a simple formula to approximate the similarity measure d , in the spirit of the measures described in Victor and Purpura ( 1997 ) ; van Rossum ( 2001 ) ; Houghton and Sen ( 2008 ) . The Hamming distance gave a very poor approximation of similarity and even a measure which gave a different weight to each neuron , δ1 ( ri , rj ) =∑μwμ ( riμ−rjμ ) 2 , performed very poorly . Shown is a joint histogram of the similarity values d ( ri , rj ) ( x-axis ) and the corresponding values predicted by a global second order similarity model δ2 ( ri , rj ) =∑μ∑νwμν ( riμ−rjν ) 2 ( y-axis . w parameters fit to train data; results for cross-validated test data are shown ) for the responses of a representative group of 20 neurons ( same as in Figure 3 ) to an artificial video . For clarity , values were normalized about the y-axis , such that each vertical slice sums to one . ( B ) Joint histogram of the similarity values of pairs of responses from the same cluster ( x-axis ) and the similarity predicted by a local second order model for similarity , that is , δ2 applied independently to each cluster . Other details as in A . These results suggest that noise is highly stimulus dependent and cannot be accurately described by a global , stimulus independent , model . Yet , for a given cluster , we can accurately describe its similarity neighborhood , using the appropriate set of single neurons and neuron pairs . ( C ) In support of the previous conclusion , we found that close inspection of large response clusters reveals an obvious structure within clusters . Many of the clusters of similar responses can be characterized as having very precise neurons ( almost always spiking or almost always silent ) , alongside more noisy neurons , which appear to be nearly random within a cluster . These precise and noisy neurons differ from one cluster to another . Shown is a detailed view of the responses in a subset of the clusters , which contain between 20 and 50 different patterns and appear most frequently in the data . Top: All population responses belonging to each cluster ( clusters separated by vertical lines ) . Each horizontal line corresponds to one neuron; each vertical slice is the response of the population to a single repeat . Bottom: The average response of each cluster . The spiking probability of each neuron is represented by its gray scale intensity ( color bar: dark—high spike probability , light—low spike probability ) . ( D-F ) same as ( A-C ) , but for a natural video stimulus . DOI: http://dx . doi . org/10 . 7554/eLife . 06134 . 009 In clear contrast , when we used a hierarchical clustering algorithm on d ( ri , rj ) matrices and grouped responses together according to their similarity ( see ‘Materials and methods’ ) , we found that the ordered matrices exhibit an almost perfect block diagonal structure ( Figure 3B; see Figure 3—figure supplement 1 for statistical analysis ) . Thus , the population code that is used by the retina both for artificial videos , and for natural videos , is arranged in sets , or clusters , of highly similar activity patterns , which are very different from all other patterns . We term these groups of highly similar responses neural ‘synonyms’ , and by analogy we refer to the similarity measure d as a neural thesaurus . We examined whether the Hamming distance , which is a simple and intuitive measure of similarity between population responses ( the number of neurons that differ in their spiking or silence state ) , was sufficient to reveal the structure of the population codebook . We thus constructed the matrices of Hamming distances between all pairs of population activity patterns , and clustered the responses , using the same hierarchical agglomerative clustering . Figure 3C shows the Hamming matrix for the same group of 20 cells from a–d , where matrix rows ( and columns ) are ordered according to the clustering results . We did not find evident structure in the codebook used for natural stimuli , and only slightly more structure was apparent for the artificial stimuli . We emphasize that the results of Figure 3 were typical for many independent choices of groups of cells ( as is summarized later in Figure 5 ) . These results reflect the importance of measuring similarity in meaning and not similarity in structure . Using a syntactic ( structure based ) measure , we would not have been able to uncover the clustered organization of the neural population code that the semantic similarity reveals . Patterns that belong to the same cluster do exhibit some shared structure , namely some of the neurons almost always fire in a specific cluster and others are always silent ( Figure 3B , bottom ) . However , importantly , we found that the semantic similarity structure over all the patterns we observed could not be captured by a simple linear or bilinear function of the population patterns ( Figure 3—figure supplement 2 ) . The clustered structure of the neural population codebook suggests that the same stimulus may be represented by different , yet semantically similar population patterns , or synonyms . Such structure is commonly used in engineered codes in computer science and communications ( Cover and Thomas , 1991; Sreenivasan and Fiete , 2011; Curto et al . , 2013 ) . This gives rise to two important predictions , which we confirmed by cross-validation , using novel ( held out ) test stimuli: 1 . Responses to repeated presentations of a stimulus should come from the same cluster . 2 . Responses from the same cluster should be nearly interchangeable . To directly show the advantages of cluster-based encoding of information by the retina , we quantified the reliability of population patterns used to encode the same stimulus . Because of the noise , the responses to repeated presentations of the same stimulus are so variable that even the most frequent population pattern would occur only a handful of times ( Figure 4A , B ) . However , the reliability of the population code is revealed when instead of focusing on individual patterns , we count how many of the population patterns evoked by the same stimulus belong to the same cluster . Notably , even when the population response was highly unreliable ( i . e . , the most frequent response pattern appeared less than 20% of the time ) , often over 70% of the observed responses would fall within a single cluster ( mostly for the natural video ) , for the cross-validated test data ( see Figure 4—figure supplement 1 for statistical analysis ) . 10 . 7554/eLife . 06134 . 015Figure 4 . Responses to the same stimulus tend to come from the same cluster . ( A ) The probability of the most frequent response across video repeats is plotted as a function of stimulus identity in gray ( stimuli are sorted by reliability ) . In purple , we plot the reliability of the clustered response , that is , the probability of observing a response from the most frequent cluster for each stimulus ( clustering matrix presented in Figure 3 ) . Only the 100 stimuli that evoked the strongest response are shown . Clearly , responses to the same stimulus tend to come from the same cluster , even when the most frequent single response occurs less than 20% of the time , thus the cluster code is far less noisy . ( B ) Same as A but for the natural video data set . DOI: http://dx . doi . org/10 . 7554/eLife . 06134 . 01510 . 7554/eLife . 06134 . 016Figure 4—figure supplement 1 . Increased reliability of clustered responses is highly significant . ( A ) Responses recorded at the same time point across video repeats cluster together significantly . Shown is the probability of the most frequent cluster , plotted as a function of stimulus id ( purple; stimuli sorted by reliability ) , for the clustered response; that is , the probability of the most frequent cluster evoked by each stimulus ( same as Figure 4A ) . For comparison , we also shuffled the cluster assignment of each response and repeated the analysis . Shown in gray is mean ± STD of the most frequent cluster in 100 randomly shuffled cluster assignments . Clearly , for many time points the high reliability of the clustered response is not merely a result of response grouping , but the tendency of responses from the same cluster to be associated with the same stimulus . ( B ) Same as A , but for natural video data set . DOI: http://dx . doi . org/10 . 7554/eLife . 06134 . 016 The most direct test of ‘coding by clusters’ is to ask how much information about the stimulus would be lost if instead of knowing the precise population activity pattern ( exactly which neurons spiked and which ones were silent ) we only knew which cluster the pattern belongs to . We therefore compared the information that the full set of population responses r carry about the stimulus I ( s;r ) , to the information that is carried just by knowing which cluster ( out of k possible ones ) the response belongs to , I ( s;Ck ( r ) ) . To that end , we labeled every population pattern in the test data according to its cluster identity , Ck ( r ) , based on the similarity structure learned from training data ( see ‘Materials and methods’ ) . To avoid any arbitrary assumptions about the number of clusters , we assessed how much information is carried by k clusters , for different values of k ( Figure 5 ) on novel test data that was not used for learning the similarity structure . We found that ∼100 clusters were enough to account for over 80% of the information available about the stimulus in the detailed population patterns , for both types of stimuli . Importantly , clustering the responses based on the Hamming distance between them gave significantly worse results . This clearly reflects that responses in the same cluster have a similar meaning and can indeed be viewed as noisy variants of a noise-free codeword , similar to the ground states or attractors in a Hopfield model ( Hopfield , 1982; Tkacik et al . , 2006 ) . 10 . 7554/eLife . 06134 . 010Figure 5 . Cluster identity conveys most of the information about the stimulus . ( A ) The fraction of information retained about the stimulus when population responses are replaced with the label of the cluster they belong to , plotted as a function of the number of clusters used . Lines correspond to the average of 10 groups of 20 neurons , line widths represent SEM . Purple—clustering based on the similarity measure described in the text , gray—clustering based on the Hamming distance . Inset: Fraction of information as a function of the number of clusters on a linear scale . Individual groups are shown in gray , and the orange line marks the curve corresponding to the representative matrix from Figure 3 . Very few clusters are required to account for most of the information , suggesting responses from the same cluster have a similar meaning . ( B ) Same as A , but for a natural video clip . DOI: http://dx . doi . org/10 . 7554/eLife . 06134 . 01010 . 7554/eLife . 06134 . 011Figure 5—figure supplement 1 . The similarity measure generalizes well across stimuli . ( A ) Fraction of information retained about the stimulus plotted against the number of response clusters . The similarity measure learned from the artificial video train data is applied either to cross-validated test data ( same as Figure 5 of main text; purple line ) , or to the train data itself ( purple dots ) . We see nearly identical performance on cross-validated and non cross-validated data . Shown is the same representative group of 20 neurons as in Figure 3 . ( B ) The number of clusters required to recover over 50% ( filled purple circles ) or 80% ( open triangles ) of the information available about the stimulus , plotted as a function of the inverse number of stimuli in the test data ( train data remained fixed ) , on a semi-logarithmic scale . Also shown , for comparison , is the overall number of observed patterns ( black line ) . Depicted are average and SEM ( may be smaller than markers ) over 10 different groups of 20 neurons . As the number of stimuli in the test set increased , we saw a very mild increase in the number of clusters required to account for either 50% or 80% of the information . In fact , there was no significant increase in the number of clusters when the number of stimuli increased from 200 to 250 ( p > 0 . 4 , sign rank test ) , while the number of observed patterns grew significantly by over 500 ( p < 0 . 01 sign rank test ) . This suggests that the clusters ( or ‘code-words’ ) we identified are relevant to a wide range of stimuli within the same stimulus class . ( C , D ) Same as panels A and B , but for data taken from the natural video data set . DOI: http://dx . doi . org/10 . 7554/eLife . 06134 . 01110 . 7554/eLife . 06134 . 012Figure 5—figure supplement 2 . Clustering aimed at maximizing the mutual information yields similar results to clustering based on similarity alone . ( A ) Fraction of information retained about the stimulus plotted against the number of response clusters ( full field flicker stimulus ) . Responses were either clustered using simple agglomerative clustering ( purple . Same as in main text , see ‘Materials and methods’ ) , or using agglomerative information bottleneck clustering ( Slonim and Tishby , 1999 ) , which explicitly aims to cluster responses such that maximal information about the stimulus is retained ( black ) . Although , we would expect clustering aimed at information maximization to do a better job , after cross-validation ( applying the clustering to novel responses to novel stimuli ) , we see that the simple similarity based clustering performs just as well . The data shown are for the same representative group as used throughout the main text . We note that results are shown for cross-validated test data , and that information bottleneck clustering is a greedy approach with no guarantee of optimality , thus it is possible for similarity based clustering to outperform information bottleneck clustering . ( B ) Same as A , but for the responses to a natural video stimulus . DOI: http://dx . doi . org/10 . 7554/eLife . 06134 . 01210 . 7554/eLife . 06134 . 013Figure 5—figure supplement 3 . Comparing response similarity derived from the conditionally independent model and pairwise model . ( A ) Left: The probability of observing the most frequent response across repeats of the artificial video is plotted as a function of stimulus i . d . in gray; for clarity , the stimuli are sorted by reliability of their responses . In black is the reliability of the clustered response , that is , the probability of the most frequent cluster evoked by each stimulus , where responses were clustered by similarity derived from a conditionally independent model . In purple is the clustered reliability as derived from the pairwise model used in the main text ( same as Figure 4 ) . Only stimuli that evoked a strong response ( at least one spike in over 75% of the repeats ) are shown . Here , reliability is much higher when using the more accurate pairwise model to derive similarity between population responses . Right: Same as left , but for the Natural video data set . Here , differences between conditionally independent and pairwise models are less pronounced . ( B ) Left: Similarity derived from the conditionally independent model is directly compared with the similarity derived from the pairwise model , for test responses to the artificial video . Similarities between response pairs calculated using the conditionally independent model were binned ( x-axis ) and the mean and standard deviation of the similarity for the same pairs was calculated using the pairwise model ( y-axis ) . Black line marks identity . Right: Same as left , but for the natural video . We see that similarity measured using the conditionally independent model is closer to that measured using the more accurate pairwise model under natural stimulation . ( C ) Left: same as Figure 5A , but here we compare the fraction of information as a function of number of clusters curve obtained by using the pairwise model ( purple ) and the one obtained using the conditionally independent model , in which noise correlations are ignored ( gray ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06134 . 01310 . 7554/eLife . 06134 . 014Figure 5—figure supplement 4 . The neural ‘thesaurus’ remains stable across different bin sizes . ( A ) Correlation matrix between the similarity values estimated for all possible response pairs . For a given group of 8 neurons from the artificial video data set , we calculated the full similarity matrix over all 256 possible responses . This was done for bin sizes between 5 and 80 ms . We then calculated the correlation between the similarity matrices for each combination of bin sizes . Shown is an average across 8 randomly chosen groups of 8 neurons . The results indicate that except for extreme differences in bin size , we recover highly consistent similarity structures . ( B ) The fraction of information is shown as a function of the number of clusters in the similarity matrix ( compare to Figure 5 in main text ) . Different bin sizes are indicated by colors . Different lines correspond to different individual 8 neuron groups ( same groups as A ) . We see very similar results except for the very short time bin of 5 ms . ( C , D ) Same as A and B but for data taken from the natural video data set . DOI: http://dx . doi . org/10 . 7554/eLife . 06134 . 014 We therefore conclude that grouping responses by their similarity in the training data identified clusters that conserved the information available about the stimuli in the test data , thus indicating that the similarity we measured is a general feature of the code and not limited to a particular set of stimuli . Importantly , all stimuli in the test data and many of the neural responses they evoked were not observed in the training data that we used to learn d ( ri , rj ) . Thus , the limited number of stimuli and responses observed in the training data were sufficient to identify semantic clusters and predict the similarity between over a million possible neural responses evoked by a multitude of different possible stimuli , which were verified using the test data . We further point out that d ( ri , rj ) was so stable across time and for different selections of train and test data , that the information curves derived from clustering the cross-validated test data and the training data itself were nearly identical ( Figure 5—figure supplement 1 ) . In addition , the number of clusters required to convey 80% of the information seems to saturate with the size of the test set ( Figure 5—figure supplement 1 ) . We conclude that the similarity between neural responses does not rely on the specific stimuli we used and would generalize to other stimuli at least within the same stimulus class . We further emphasize that clustering of the responses was based on similarity alone and was not optimized to maximize information in any way ( which could therefore give even better results ) . Yet , even if we clustered patterns by explicitly trying to maximize the information , we achieved very similar results ( Figure 5—figure supplement 2 ) . This suggests that grouping responses simply by their semantic similarity may be nearly optimal from an information transmission standpoint . We also estimated the semantic similarity between population patterns with simpler models for the population responses to the stimuli—using conditionally independent models of encoding , where for each stimulus s the response is described by P ( r|s ) =ΠiP ( ri|s ) . Although these models give a different and less accurate estimate for the probability distribution than the pairwise model P ( 2 ) ( r|s ) ( Figure 2 ) , we find that grouping the population activity patterns using this model results in a similar clustered structure ( see Figure 5—figure supplement 3 ) . This may suggest that the codebook organization into clusters is sufficiently robust , so that even when using a less accurate model of the neural responses one can identify the organization of the response space ( see more in the ‘Discussion’ ) . Given that almost all information about the stimulus is carried by the identity of the cluster that a given activity pattern belongs to , we asked whether we can map the functional organization of the population codebook . We used Isomap ( Tenenbaum et al . , 2000 ) to present a low dimensional embedding of all the population responses associated with clusters that contain 30–300 responses in three-dimensional Euclidean space ( Figure 6A; see ‘Materials and methods’ and Video 1 for the raw data points in 3D ) . Isomap is an embedding algorithm for high dimensional data that preserves the geodesic distance between points . While this embedding is imperfect ( one would need more dimensions to achieve a nearly perfect embedding; see Figure 6—figure supplement 1 ) , it provides an approximate view of the organization of the population code of a sensory system and coarsely reflects the ‘cloud’ of patterns that make up each cluster . 10 . 7554/eLife . 06134 . 017Figure 6 . Response similarity predicts stimulus similarity . ( A ) The responses belonging to clusters that contain 30–300 patterns were embedded using Isomap . Each ellipse represents the 1 STD Gaussian fit to all responses belonging to a single cluster . The Euclidean distance in the plot approximates the similarity measure d ( see text ) . The coordinates also correspond to the RGB value of each ellipse , thus nearby clusters share similar colors . Same representative group as in Figure 3 . ( B ) Embedding of cluster triggered average waveforms in 2D Euclidean space . For each pair of clusters from panel A , we calculated the inter-cluster distance as the average similarity between pairs of responses , one from each cluster . Clusters were then embedded in 2D space using Isomap in a manner that approximates the calculated distances . Each cluster is represented by the mean stimulus that preceded ( 250 ms ) responses belonging to that cluster . Thus , nearby waveforms belong to similar clusters . Clusters are colored as in panel A , therefore the blue channel corresponds to the third dimension of embedding not shown in the plot . DOI: http://dx . doi . org/10 . 7554/eLife . 06134 . 01710 . 7554/eLife . 06134 . 018Figure 6—figure supplement 1 . Cluster similarity implies stimulus similarity . ( A ) Geodesic vs embedded distances for the embedding shown in Figure 6A . The x-axis is the Geodesic distance between clusters ( i . e . , distances in the neighborhood graph; methods ) , vs the Euclidean distances after embedding in 3D . ( B ) Residual variance as a function of dimensionality for embedding of waveforms in Figure 6A . The residual variance is defined as 1 − r2 ( dG , dIso ) , where r is the correlation coefficient , dG is the geodesic distances between points as defined by the weighted neighborhood graph ( Tenenbaum et al . , 2000 ) , and dIso is the Euclidean distances between points after embedding . ( C ) Same as A , but for embedding of responses in Figure 6B . Due to the large number of population responses embedded ( 2391 ) , we show the joint histogram of the geodesic ( x-axis ) and embedded ( y-axis ) distance values . Colorbar represents frequency of occurrence of distance pairs . ( D ) Same as B , but for embedding of responses in Figure 6B . ( E ) For each pair of clusters shown in Figure 6A , we calculated the correlation coefficient between the cluster triggered average stimulus ( the average of all stimuli preceding responses in a cluster ) of each cluster , and plot it as a function of the distance between the clusters . Inter cluster distance was defined as the average similarity between pairs of responses , one from each cluster . Even though the correlation coefficient is not the ideal measure to quantify similarity among stimuli , we still see a clear and significant relationship between cluster similarity and stimulus similarity . Namely , clusters that are more similar ( smaller values on the x-axis ) have a higher correlation between their associated average stimuli . DOI: http://dx . doi . org/10 . 7554/eLife . 06134 . 01810 . 7554/eLife . 06134 . 019Video 1 . Embedding of responses in 3D using Isomap . Each dot represents a single population response to the artificial video; the Euclidean distance between points approximates the similarity d between them . Similar to Figure 6A , only we explicitly plot every population activity pattern in each cluster . Colors represent different clusters and correspond to the colors in Figure 6A , B . DOI: http://dx . doi . org/10 . 7554/eLife . 06134 . 019 To show the functional correlates of these clusters in coding , we present the similarity between clusters in terms of the stimuli that they encode . To that end , we quantified the distance between each pair of clusters , Cm and Cl , by the average distance between all inter-cluster response pairs , 〈d ( ri , rj ) 〉i∈C1 , j∈C2 . We then embed the clusters in a Euclidean space according to the similarity between them ( again using Isomap ) . Thus , in Figure 6B each cluster is represented by the average of its associated stimulus ensemble ( we emphasize that the stimuli were taken exclusively from the test data ) . Hence , waveforms that are closer in space belong to clusters that are more similar . We find that response similarity measured from the training data predicts stimulus similarity in the test data ( Figure 6—figure supplement 1 shows the correlation between cluster similarity and stimulus similarity; for more accurate metrics on stimulus space , see [Tkačik et al . , 2013] ) . Thus , the similarity measure proposed here for the population responses captures similarity in meaning and generalizes across stimuli . We note that this analysis could only be carried out on the simple one dimensional full field video , as there is no evident way to reduce the dimensionality of natural stimuli . The organization of the retina codebook may also explain how the brain can decode novel stimuli from novel neural activity patterns . Namely , if we observe a response r that we have never seen before , we can now ask what similar responses tell us about the stimulus . Importantly , this can be done since P ( 2 ) ( r|s ) allows us to estimate d ( ri , rj ) even for patterns we have not seen in the past . Indeed , we found that P ( s|r ) for a held-out test response , r , could be well estimated simply by taking P ( s|r′ ) for the response , r′ , which is most similar to it . Similarity was assessed using the thesaurus that we learned from the training data , that is , for different stimuli than the ones we tested on ( Figure 7A , B ) . This approach clearly improved our ability to estimate the stimulus compared to our prior knowledge about the stimulus , as measured by the Jensen-Shannon divergence between the ‘true’ P ( s|r ) and the estimate ( Figure 7C ) . Using a thesaurus based on the Hamming distance clearly degraded performance and reduced the accuracy of the stimulus estimate ( Figure 7C ) . 10 . 7554/eLife . 06134 . 020Figure 7 . Accurate decoding of new stimuli from previously unseen population responses , using a neural thesaurus . ( A ) The conditional distribution over stimuli for one population response , P ( s|r ) , to the artificial video is shown ( black dots ) . P ( s|r ) can be well approximated by the conditional distribution over stimuli P ( s|r′ ) where r′ is the response most similar to r according to the thesaurus d ( ‘r′Nearest’ , purple line ) . Actual responses are shown as inset . The same representative group of 20 neurons shown in Figure 3 was used here . Error bars represent standard errors of the probability estimates B . Same as in A , but for a natural video clip . ( C ) Top: The average Jensen-Shannon divergence between the ‘true’ P ( s|r ) and the estimate described in panel A ( Semantic ) , or for an estimate derived using the Hamming distance instead of our similarity measure ( Hamming ) , for the artificial video data . Also shown is the average divergence from the prior over stimuli ( Prior ) . Plotted are mean and standard errors ( barely discernable ) across all patterns that had at least one close neighbor ( <0 . 25 bits away ) . Bottom: Same as above , but for the natural video data . Having a thesaurus markedly improves our ability to gain some knowledge about never before seen responses , compared to a naive prior , or even to using Hamming distance as a similarity measure . DOI: http://dx . doi . org/10 . 7554/eLife . 06134 . 02010 . 7554/eLife . 06134 . 021Figure 7—figure supplement 1 . Comparing decoding performance using the conditionally independent model and pairwise model . Left: The average Jensen-Shannon divergence between the ‘true’ P ( s|r ) and the estimate based on the most similar response as measured using either the conditionally independent model or the pairwise model used in the main text ( see Figure 7 ) , for responses evoked by an artificial video . Plotted are mean and standard errors ( barely discernable ) across all patterns that had at least one close neighbor ( <0 . 25 bits away ) . The pairwise model performs on average slightly yet significantly better ( p < 10−4 , two-sided paired sign test ) . Right: Same as left , but for the natural video . Again , the pairwise model performs on average slightly yet significantly better ( p < 10−4 , two-sided paired sign test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06134 . 021 Thus , using a very limited set of known responses it is possible to decode novel responses to novel stimuli . This may be crucial when encountering new responses to known stimuli , due to sensory noise , or when generalizing prior knowledge to novel stimuli . We presented a thesaurus for a neural population code—a similarity measure between population activity patterns , based on the meaning of the responses and not on their syntactic structure . This is a general framework for mapping the codebook in any sensory system , since it makes no assumptions about what makes two spike trains similar , or that the similarity structure should obey a metric in the strict mathematical sense . Indeed , it revealed that the neural code of the brain is very different than our usual mathematical intuition: what we may regard as intuitive measures , like the Hamming distance , fail to capture the similarity between neural population responses . Instead , we found a highly structured codebook , organized in groups of responses that carry very similar information , and are distinct from all other responses . This organization of the neural population codebook into a relatively small number of semantically different clusters is reminiscent of the design of engineered codes used in communication . There , to overcome noise , different codewords are placed far apart in space of codewords , to allow for error-correction . Since noise may corrupt a message that is sent through a channel , when a pattern is received , it is compared to all possible codewords and is classified according to the nearest one . Importantly , the distance between codewords must be defined by the nature of the noise . Here , we found a neural population code that seems to be built along similar lines . Our thesaurus revealed that most of the information that the population encodes is held by the identity of the cluster that a specific activity pattern belongs to , and that more detailed information may be carried by the fine structure of the pattern . These results generalized across novel stimuli and scaled slowly with the number of stimuli used ( Figure 5—figure supplement 1 ) . We thus suggest that our analysis reveals a design principle of the population codebook of a sensory circuit . What are the advantages of such a population code design ? We note that for a large neural population , most network activity patterns are observed at most once ( Ganmor et al . , 2011a ) . Thus , the brain frequently receives from its sensory neurons activity patterns that were never seen before . This may occur either due to sensory noise , corrupting the response to a known stimulus but also as a result of a truly novel stimulus . In either case , a semantic similarity measure between neural population responses may explain how these novel inputs may be deciphered: when a novel response pattern is encountered it can take on the information content of similar responses , whose meaning is already known . Therefore , the thesaurus may not only explain how sensory noise can be overcome but also how the brain can generalize across stimuli . Indeed , the thesaurus enabled us to decode neural activity patterns we have not seen before , based on their similarity to other responses . Uncovering the structure and organization of the population code relied on our ability to accurately model the noisy nature of the population response to specific stimuli . It is the nature of the noise that determines how likely two responses are to be interchanged , and consequently the semantic similarity between them . Here , we used pairwise maximum entropy models to capture the ‘noise correlations’ at every time point , ( Figure 2 ) , which were weak on average , but significant at specific times , as was also shown in ( Kohn and Smith , 2005; Pillow et al . , 2008; Cafaro and Rieke , 2010; Granot-Atedgi et al . , 2013 ) . Interestingly , learning a thesaurus using the conditionally independent model of the neural responses to the stimuli , recovered a similarity structure that was on par with that revealed by the pairwise model , in terms of the information that the clusters carried about the stimulus . These results are surprisingly given that the cells were clearly not conditionally independent . One possible explanation is that the organization of the codebook may be so robust , namely that the separation between clusters is sufficiently large , that even an inaccurate model can capture the identity clusters because the errors of the model are smaller than the distance between clusters . The results of ( Schwartz et al . , 2012; Granot-Atedgi et al . , 2013 ) suggest that the contribution of noise correlations to shaping the population response to stimuli will increase significantly with network size . Since by definition , the conditional independent model implies larger encoding noise ( noise entropy ) , this suggests that the clusters would be ‘wider’ in this case . It would therefore be most interesting to explore the thesaurus that these models would give for larger populations of cells , beyond groups of 20 cells that we used here for sampling reasons . Finally , we note again that the pairwise models gave better results for decoding stimuli over the conditionally independent model . Previous studies have suggested mathematically elegant and computationally efficient measures of spike train similarity , relying mostly on edit-distance based metrics ( Victor and Purpura , 1997; van Rossum , 2001; Victor , 2005; Houghton and Sen , 2008 ) . Our approach is fundamentally different , as it does not assume a metric structure , or particular features of the code ( Haslinger et al . , 2013 ) and does not require assumptions about the syntactic similarity between activity patterns . But maybe most importantly , we found that the semantic similarity between population responses could not be approximated by simple linear or bilinear functions of the patterns . This is mostly due to the fact that noise showed a high degree of stimulus dependence ( Figure 3—figure supplement 2 ) . This means that the correlated nature of neural noise shapes the population code differently for different stimuli . The approach we presented here could be immediately applied to other neural circuits ( Bathellier et al . , 2012; Parnas et al . , 2013 ) , and it is therefore important to note where our approach has been restrictive and could be extended . First , our clustering was not optimized to maximize the information that the clusters carry about the stimulus , but only to group similar responses together . Interestingly , such an information maximization clustering approach ( Slonim and Tishby , 1999 ) does not result in significantly more information using fewer clusters ( Figure 5—figure supplement 2 ) . Second , hard clustering of the population responses into k clusters is somewhat arbitrary , and it is possible to consider generalizations to fuzzy rather than hard clustering—where responses may be associated with several clusters , or categories , simultaneously in a probabilistic manner ( Zemel et al . , 1998 ) . Lastly , while both agglomerative and spectral clustering yielded similar results in our analysis of the nature of the retinal population code , it is possible that other ways of clustering would reveal further structure of the code . Experiments were performed on the adult tiger salamander ( Ambystoma tigrinum ) . All experiments were approved by the institutional animal care and use committee of Ben-Gurion University of the Negev and were in accordance with government regulations . Prior to the experiment the salamander was adapted to bright light for 30 min . Retinas were isolated from the eye and peeled from the sclera together with the pigment epithelium . Retinas were placed with the ganglion cell layer facing a multi-electrode array with 252 electrodes ( Ayanda Biosystems , Lausanne , Switzerland ) and superfused with oxygenated ( 95% O2/5% CO2 ) Ringer medium which contains: 110 mM NaCl , 22 mM NaHCO3 , 2 . 5 mM KCl , 1 mM CaCl2 , 1 . 6 mM MgCl2 , and 18 mM Glucose , at room temperature ( Meister et al . , 1994 ) . The electrode diameter was 10 μm and electrode spacing varied between 40 and 80 μm . The array was lowered onto the retina from above by means of a standard mechanical manipulator . Extracellularly recorded signals were amplified ( Multi Channel Systems , Germany ) and digitized at 10k Samples/s on four personal computers and stored for off-line spike sorting and analysis . Spike sorting was done by extracting the amplitude and width from each potential waveform , and then by clustering using an in-house written MATLAB program ( Source code 1 ) . 48/31 retinal ganglion cells were recorded and cleanly isolated in the artificial/natural video experiment . Natural video clips were acquired using a digital video camera ( Sony Handycam DCR-HC23 ) at 30 frames per second . The stimulus was projected onto the salamander retina from a CRT video monitor ( ViewSonic G90fB ) at a frame rate of 60 Hz such that each acquired frame was presented twice , using standard optics ( Puchalla et al . , 2005 ) . The original color videos were converted to gray scale , using a gamma correction for the computer monitor . Artificial full-field flicker stimuli were generated by sampling a uniform gray level from a normal distribution . In both cases , the visual stimulus covered the retinal patch that was used for the experiment entirely . A 10 s clip was taken from each video and played to separate retinas repetitively for approximately 2 hr . Each video was therefore replayed to the retina over 600 times . Analysis was carried out in MATLAB . We examined the responses of ten randomly chosen groups of 20 neurons from each retina . Spikes were binned at 20 ms ( different bin sizes did not qualitatively affect our results , see Figure 5—figure supplement 4 ) . Each response r is a 20 bit binary vector with each bit corresponding to the activity of a single neuron ( 0 representing silence and 1 representing spiking ) in a single time bin . To estimate the probability distribution of responses to a given stimulus P ( r|s ) , we considered the ensemble of responses recorded at a single point in time in the video across repeats . Note that the retina is displayed with the exact same stimulus , including several minutes of stimulus history , at each repeat . We then estimated the single neuron and pairwise spiking probability from the data in order to generate a maximum entropy pairwise model as detailed in previous work ( Ganmor et al . , 2011b ) . Briefly , the maximum entropy pairwise distribution is known to take the form ( Jaynes , 1957 ) P ( 2 ) ( r|s ) =1Z ( s ) exp ( ∑i=1Nαi ( s ) ri+∑i<jβij ( s ) rirj ) . The parameters can be found by optimizing the likelihood of the data L ( Data|α , β ) . Since the log likelihood is concave , the global optimum can be found using gradient methods , with local derivatives ∂L∂αi ( s ) =〈ri〉PData ( r|s ) −〈ri〉P ( 2 ) ( r|s ) , where <f ( r ) >P represents the expected value of some function of the response , f , with respect to the probability distribution P . Data were split into train and test sets . The train set was used to learn the conditional probabilities P ( r|s ) , which were then used to construct P ( s|r ) through Bayes' rule P ( s|r ) = P ( r|s ) P ( s ) /P ( r ) . The dissimilarity matrix between all responses observed in the test set was defined as d ( r1 , r2 ) = DJS ( P ( s|r1 ) ||P ( s|r2 ) ) ( where DJS stands for the Jensen-Shannon divergence , and s represents only train stimuli ) . This matrix was clustered using hierarchical agglomerative clustering with the distance between clusters defined as the average distance between inter-cluster pairs . Different methods of spectral clustering yielded similar results . The number of clusters , k , was systematically varied . For a given value of k , the mutual information between stimulus and clusters I ( s;Ck ( r ) ) was estimated as follows: Each response in the test data was replaced with its cluster label Ck ( r ) ( Ck ( r ) can take a value in {1…k} ) and then mutual information was estimated as I ( s;Ck ( r ) ) = H ( Ck ( r ) ) − H ( Ck ( r ) |s ) , where H is the Shannon entropy . H ( Ck ( r ) ) is the total entropy of different clusters in the data . H ( Ck ( r ) ) =−∑i=1kP ( Ck ( r ) =i ) log ( P ( Ck ( r ) =i ) ) , where P ( Ck ( r ) =i ) is the frequency of the ith cluster , out of k , in the data . H ( Ck ( r ) |s ) is the average conditional entropy of clusters given stimulus s . In our case , we treat each time point in the 10 s video as a unique stimulus , and thus H ( Ck ( r ) |s ) is given by the conditional entropy at each time point across video repeats , averaged over time points—H ( Ck ( r ) |s ) =H ( Ck ( r ) |t ) =−∑t=1T1T∑i=1kP ( Ck ( r ) =i|t ) log⁡P ( Ck ( r ) =i|t ) , where P ( Ck ( r ) =i|t ) is the frequency of cluster i , out of k , at time t in the video . The representative matrices in Figure 3 were chosen as the matrices with the fifth ( out of 10 ) highest average ratio of the within cluster similarity and the overall similarity . The total degree of network correlation was measured using the multi-information ( Amari , 2001; Schneidman et al . , 2003b ) , defined as IN = Hind − H , where Hind is the independent entropy ( sum of all individual neuron entropies ) and H is the entropy estimated from the actual data . In all cases where information or entropy was estimated , we used the method of extrapolation ( Treves and Panzeri , 1995; Strong et al . , 1998 ) to correct for finite sampling biases . These corrections were on the order of a few percent ( see Figure 2—figure supplement 1 ) . For a given response r in the test set , we compared the ‘true’ P ( s|r ) , constructed from P ( r|s ) across test stimuli as previously described , to one of three estimates P˜ ( s|r ) : ( 1 ) The true prior over stimuli P ( s ) , which is a uniform distribution over stimuli . ( 2 ) A nearest neighbor estimate , that is , P˜ ( s|r ) = P ( s|r′ ) where d ( r , r′ ) is minimal across all responses in the test set . ( 3 ) Same as 2 , only r′ is a nearest neighbor in Hamming distance ( arbitrarily chosen out of all nearest neighbors ) . Responses were embedded using the Isomap algorithm ( Tenenbaum et al . , 2000 ) . Given a set of pairwise distances , Isomap finds a configuration of points in k dimensions that most closely recreate the geodesic distances between points . The geodesic distance between two points is either the dissimilarity d between them , if they are connected in the neighborhood graph , or the shortest weighted path if they are not . Two nodes were connected if their dissimilarity was smaller than 0 . 5 bits ( epsilon neighborhood ) . Isomap was chosen over multi dimensional scaling approaches as it gave better results , in particular since dissimilarities are bounded from above ( 1 bit maximum ) . We limited the number of points to embed so that a 2–3 dimensional embedding gave a reasonably faithful reconstruction of the geodesic distances as measured by the residual variance .

Our ability to perceive the world is dependent on information from our senses being passed between different parts of the brain . The information is encoded as patterns of electrical pulses or ‘spikes’ , which other brain regions must be able to decipher . Cracking this code would thus enable us to predict the patterns of nerve impulses that would occur in response to specific stimuli , and ‘decode’ which stimuli had produced particular patterns of impulses . This task is challenging in part because of its scale—vast numbers of stimuli are encoded by huge numbers of neurons that can send their spikes in many different combinations . Furthermore , neurons are inherently noisy and their response to identical stimuli may vary considerably in the number of spikes and their timing . This means that the brain cannot simply link a single unchanging pattern of firing with each stimulus , because these firing patterns are often distorted by biophysical noise . Ganmor et al . have now modeled the effects of noise in a network of neurons in the retina ( found at the back of the eye ) , and , in doing so , have provided insights into how the brain solves this problem . This has brought us a step closer to cracking the neural code . First , 10 second video clips of natural scenes and artificial stimuli were played on a loop to a sample of retina taken from a salamander , and the responses of nearly 100 neurons in the sample were recorded for two hours . Dividing the 10 second clip into short segments provided a series of 500 stimuli , which the network had been exposed to more than 600 times . Ganmor et al . analyzed the responses of groups of 20 cells to each stimulus and found that physically similar firing patterns were not particularly likely to encode the same stimulus . This can be likened to the way that words such as ‘light’ and ‘night’ have similar structures but different meanings . Instead , the model reveals that each stimulus was represented by a cluster of firing patterns that bore little physical resemblance to one another , but which nevertheless conveyed the same meaning . To continue on with the previous example , this is similar to way that ‘light’ and ‘illumination’ have the same meaning but different structures . Ganmor et al . use these new data to map the organization of the ‘vocabulary’ of populations of cells the retina , and put together a kind of ‘thesaurus’ that enables new activity patterns of the retina to be decoded and could be used to crack the neural code . Furthermore , the organization of ‘synonyms’ is strikingly similar to codes that are favored in many forms of telecommunication . In these man-made codes , codewords that represent different items are chosen to be so distinct from each other that even if they were corrupted by noise , they could be correctly deciphered . Correspondingly , in the retina , patterns that carry the same meaning occupy a distinct area , and new patterns can be interpreted based on their proximity to these clusters .

2015

A thesaurus for a neural population code

Sensory inputs are remarkably organized along all sensory pathways . While sensory representations are known to undergo plasticity at the higher levels of sensory pathways following peripheral lesions or sensory experience , less is known about the functional plasticity of peripheral inputs induced by learning . We addressed this question in the adult mouse olfactory system by combining odor discrimination studies with functional imaging of sensory input activity in awake mice . Here we show that associative learning , but not passive odor exposure , potentiates the strength of sensory inputs up to several weeks after the end of training . We conclude that experience-dependent plasticity can occur in the periphery of adult mouse olfactory system , which should improve odor detection and contribute towards accurate and fast odor discriminations . Mammalian brains remain plastic throughout their life span , enabling animals to adapt their behavior to novel conditions . Both structural and functional plasticity have been reported in different sensory systems during development and in adulthood following lesions , passive experience or learning ( Recanzone et al . , 1993; Bao et al . , 2001; Polley et al . , 2004; Accolla and Carleton , 2008; Keck et al . , 2008; Hofer et al . , 2009; Carleton et al . , 2010; Rosselet et al . , 2011; Lai et al . , 2012 ) . While attention has been mostly focused on plasticity mechanisms in higher brain areas ( i . e . , mainly in the cortex ) ( Buonomano and Merzenich , 1998; Bao et al . , 2001; Feldman and Brecht , 2005; Accolla and Carleton , 2008; Carleton et al . , 2010; Rosselet et al . , 2011; Lai et al . , 2012 ) , less is known about the functional changes at earlier stages of sensory processing , especially in response to sensory learning . Here we investigated the functional plasticity of sensory inputs in the mouse olfactory system . Olfactory sensory neurons ( OSNs ) expressing one specialized odorant receptor gene out of a large repertoire ( Malnic et al . , 1999 ) converge in a receptor-specific manner onto anatomical structures in the main olfactory bulb ( OB ) called glomeruli ( Ressler et al . , 1994; Mombaerts et al . , 1996; Sakano , 2010 ) . Odorants activate complex spatio-temporal patterns of glomeruli , which can be monitored with various imaging techniques ( Rubin and Katz , 1999; Uchida et al . , 2000; Wachowiak and Cohen , 2001; Spors and Grinvald , 2002; Bozza et al . , 2004; Bathellier et al . , 2007; Vincis et al . , 2012; Patterson et al . , 2013 ) . The sensory information received in the glomeruli by OB output neurons is then transferred to cortical areas . Several plasticity mechanisms have been reported in olfactory cortical regions ( Quinlan et al . , 2004; Franks and Isaacson , 2005; Stripling and Galupo , 2008 ) as well as OB circuitry ( Saghatelyan et al . , 2005; Marks et al . , 2006; Gao and Strowbridge , 2009; Livneh and Mizrahi , 2012 ) . At the input level , both sensory deprivation ( Cummings et al . , 1997 ) and developmental reorganization ( Zou et al . , 2004; Kerr and Belluscio , 2006 ) have been reported to induce structural plasticity in the glomerular layer . Despite these facts , little is known about learning-mediated functional plasticity of sensory inputs in the adult OB of awake mice . Here , we investigate plasticity at the periphery induced by learning or passive exposure by combining olfactory behavior and functional imaging in awake mice . Olfactory training caused an enhanced sensitivity and potentiation of sensory inputs , which helped the animals to achieve fast and accurate odor discrimination . Most strikingly , this functional plasticity was induced specifically by the learning process but not by a passive exposure to the same odorants , and lasted up to several weeks . To investigate the potential functional plasticity at the level of sensory neurons induced by olfactory learning , mice were trained to discriminate two odorants ( rewarded vs unrewarded ) on a go/no-go operant conditioning paradigm ( Abraham et al . , 2004 , 2010 ) . As perception can vary with the odorant dilution , we used a wide spectrum of dilutions covering several orders of magnitude ranging from 100 to 10−10 ( percentile dilution in mineral oil ) for two different odor pairs ( cineol [Cin] vs eugenol [Eu] and isoamyl acetate [IAA] vs ethyl butyrate [EB] ) ( Figure 1A ) . After training , odorant-evoked input patterns were measured in the olfactory bulb by intrinsic optical signal ( IOS ) imaging ( Figure 1B , C ) . To control for the effect of odorant exposure during olfactory discriminative learning , two other groups of mice were also imaged ( naïve and passively exposed groups , Figure 1D ) . Naïve mice never encountered the odorants used for behavior testing before the imaging session whereas the passively exposed group of mice received the same amount of trials and stimulus exposure than the trained group . 10 . 7554/eLife . 02109 . 003Figure 1 . Experimental design used to assess plasticity of sensory inputs in the mouse OB . ( A and B ) Mice are first trained to discriminate pairs of odorants in an automated olfactometer . At the end of the training , odorants evoked input patterns are monitored on the dorsal OB of awake mice using intrinsic optical signal imaging . ( C ) Timetable of the go/no-go olfactory training for different odorants at different dilutions and then followed by imaging ( Trained group ) . ( D ) Two other groups of mice have been imaged for comparison . Several mice have been passively exposed to the different odorants used for the training and for the same amount of time ( Exposed group ) . A third group of mice that never experienced the odorants served as control ( Naïve group ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02109 . 003 After a habituation phase ( ‘Materials and methods’ ) , mice were trained to discriminate Cin vs Eu ( 1200 trials at 100 , Figure 2A ) . Performance levels reached more than 80% of correct responses after 400 trials and remained high during the following sessions . This task allowed the mice to acquire the procedural aspects of the training . It also allowed us to test the discrimination abilities for this odor pair at low dilutions ( i . e . , high concentrations ) . We then tested the discrimination abilities for different dilutions of the same odor pair . The performance levels remained close to chance levels for 10−10 and 10−5 dilutions , which can be interpreted as a lack of perception/discrimination ( Figure 2A , no difference between the two dilutions , Fisher’s Least Significant Difference [LSD] test , p>0 . 05 ) . On average , mice started to discriminate from 10−3 onwards ( note that one mouse was able to discriminate at 10−5 with an accuracy of ∼70% , Figure 2B ) , reaching ∼70% of correct responses for 10−3 dilution and >90% for 10−2 , 10−1 and 100 dilutions ( Figure 2B ) . Following the Cin vs Eu training , mice were trained for different dilutions of IAA vs EB in the range of 10−10-100 . Performance levels remained close to chance levels for 10−10 and 10−8 dilutions ( no difference between the two dilutions , LSD , p>0 . 1 , paired comparison ) , but mice showed a tendency to learn at 10−6 dilution and hence were trained at this dilution for an additional 300 trials during which their performance reached 80% of correct responses ( comparison between first and second task of 300 trials 10−6 IAA/EB , LSD test p<0 . 005 ) . For lower dilutions ( 10−4 , 10−2 and 100 ) , mice performed systematically above 90% ( Figure 2C ) . 10 . 7554/eLife . 02109 . 004Figure 2 . Defining odor discrimination threshold using wide range of odorant dilutions . ( A ) Discrimination accuracy shown as the average percentage of correct choices for different odorants over wide range of dilutions ( n = 7 and 11 mice for Cin/Eu and IAA/EB tasks , respectively ) . The population of mice showed a tendency for learning to discriminate Cin/Eu from 10−3 onwards and IAA/EB from 10−6 onwards [*: Fisher’s Least Significant Difference ( LSD ) test at least p<0 . 005] . Data are presented as mean ± SEM . ( B and C ) Discrimination accuracy measured on the last 300 trials for different dilutions of Cin/Eu and IAA/EB . ( D and E ) Reaction time ( RT ) measured on the same trials as in ( B and C ) . Data are presented as box plots showing the median in gray . Whiskers represent the maximum and minimum values of the dataset . DOI: http://dx . doi . org/10 . 7554/eLife . 02109 . 004 As we observed high and stable performance levels at different dilutions , we calculated the reaction times , a more sensitive parameter to monitor discrimination behavior ( Abraham et al . , 2004 , 2010 , 2012 ) . Across all concentrations , reaction times decreased with the odorant dilution , reflecting a direct linear correlation with performance levels ( R2 = 0 . 74 , ANOVA F = 12 . 3 p=0 . 04 and R2 = 0 . 82 , ANOVA F = 23 . 2 p=0 . 0085 for Cin/Eu and IAA/EB tasks , respectively ) . For dilutions with high performance levels , reaction times were relatively stable . Mice discriminated the dilutions 10−2 , 10−1 and 100 of Cin vs Eu with similar reaction times ( Figure 2D , one-way repeated measures ANOVA , F = 2 . 84 , p=0 . 13 ) . Different dilutions of IAA vs EB ( 10−4 , 10−2 and 100 ) were also discriminated with similar speeds , though significantly different ( Figure 2E , one-way repeated measures ANOVA , F = 3 . 96 , p=0 . 035; Post-hoc LSD test p=0 . 043 , 0 . 016 and 0 . 63 for 100 vs 10−2 , 100 vs10−4 and 10−2 vs 10−4 , respectively ) . In summary , mice were able to discriminate a broad range of dilutions with similar accuracy and speed above certain odorant dilutions , defining discrimination thresholds . For the population of trained animals , we therefore estimate that the discrimination thresholds for Cin/Eu and IAA/EB are between 10−5 and 10−3 and between 10−8 and 10−6 , respectively . We then investigated the effect of olfactory learning on sensory input representation . In order to assess functional plasticity of the inputs , we monitored odorant-evoked glomerular patterns on the dorsal OB of trained mice , with IOS imaging in awake head-restrained animals ( Vincis et al . , 2012 ) . For comparison , we used naïve mice that had never experienced the odorants and a group of mice that had been passively exposed to the same odorants and dilutions used for training ( Figure 1D ) . For odorant dilutions lower than the discrimination threshold ( ≤10−5 and 10−8 for Cin/Eu and EB/IAA , respectively ) , we could not detect any activated glomeruli on the dorsal OB of any group of mice ( Figure 3A–E ) . For odorant dilutions above the discrimination threshold ( 10−3 to 10−2 and 10−6 to 10−4 for Cin/Eu and EB/IAA , respectively; referred as high dilution hereafter ) , we observed more activated glomeruli in trained animals than in naïve or exposed mice , an effect consistent across all odorants used for training ( Figure 3A–E ) . On average we observed a threefold increase in the average number of activated glomeruli among the trained mice ( Figure 3C ) . This enhancement was due to the associative learning and not due to simple exposure to the odorants since the animals passively exposed to the same stimuli and for the same amount of time did not show a significant change in the number of glomeruli in comparison to naïve mice ( Figure 3C ) . 10 . 7554/eLife . 02109 . 005Figure 3 . Functional plasticity of sensory inputs to the olfactory bulb induced by olfactory learning . ( A and B ) Intrinsic optical signal ( IOS ) imaging of activated glomeruli patterns evoked by different dilutions of Cin/Eu in different groups of mice ( trained , naïve and exposed ) . For each odorant and group , all images for the different dilutions are from the same mouse . For the scale in ( A ) and ( B ) , respectively , min ΔR/R ( ‰ ) = −2 and −2 . 5 , max ΔR/R ( ‰ ) = 1 . 5 and 2 . The magenta arrows highlight that more strongly activated glomeruli are visible in the trained mice at lower dilutions . ( C ) The average number of glomeruli activated by all odorants at higher dilutions ( 10−3 and 10−2 for Cin/Eu , 10−6 and 10−4 for IAA/EB ) is significantly higher in the trained group ( T . , n = 5 mice; LSD test between T . and N . or E . : p<0 . 001 ) than in the naïve ( N . , n = 5 mice ) or exposed ( E . , n = 5 mice; LSD test between E . and N . p=0 . 5 ) groups . ( D and E ) IOS imaging of the activated glomeruli patterns evoked by different dilutions of IAA/EB . ( F ) The average number of glomeruli activated by all odorants at lower dilutions ( 10−1 and 100 for Cin/Eu , 10−2 and 100 for IAA/EB ) is similar for all groups ( for all comparisons , LSD test p>0 . 1 ) . For the scale in ( D ) and ( E ) , respectively , min ΔR/R ( ‰ ) = −3 . 5 and −3 , max ΔR/R ( ‰ ) = 3 and 2 . 5 . Values are represented as mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 02109 . 005 The observed increase in the number of glomeruli could be due to an enhanced sensitivity of the glomeruli normally activated at lower dilutions or due to the recruitment of new glomeruli . At lower dilutions ( 10−1 to 100 and 10−2 to 100 for Cin/Eu and EB/IAA , respectively; referred as low dilution hereafter ) the number of activated glomeruli remained similar across the three groups ( Figure 3F ) , suggesting that associative learning indeed improves the sensitivity of sensory inputs . We then asked if there were any changes in the odorant-evoked activity at lower dilutions . In order to address this point we quantified the amplitude of glomerular responses ( change in reflectance ) for all learned dilutions/odorants . The amplitude of the evoked activity did not differ among the passively exposed and naïve group of mice ( Figure 4A–D , LSD test p=0 . 7 , 0 . 23 , 0 . 81 and 0 . 2 for IAA , EB , Cin and Eu ) . In contrast , compared to these control groups , trained mice consistently showed enhanced IOS amplitudes ( Figure 4A–D ) . In summary , a form of functional plasticity is induced in OB sensory inputs by an olfactory learning , but not by a passive exposure to the same odorants . This potentiation was independent of the chemical class and the dilution of the stimuli ( Figure 4E , F ) . 10 . 7554/eLife . 02109 . 006Figure 4 . Functional plasticity induced by olfactory learning is long lasting and independent of the reward value of odorants . ( A–D ) Quantification of the average change in reflectance ( ΔR/R ) in the glomeruli activated by different odorants and dilutions ( n = 5 mice for all groups , * indicates LSD test p<0 . 001 ) . ( E ) Cumulative distributions of amplitudes of the evoked activity in all glomeruli analyzed in all trained ( n = 512 regions of interest ) , naïve ( n = 431 ) and exposed ( n = 427 ) mice . A significant [Kolmogorov–Smirnov ( K . S . ) test] increase in amplitude is observed for the population of glomeruli recorded in the trained group . ( F-G ) In the trained group , no difference in response amplitude was observed between glomeruli activated by different odorants belonging to different chemical classes ( F ) or between glomeruli activated by rewarded and non-rewarded odorants ( G ) . ( H and I ) The training is potentiating the input strength for several weeks . All values have been normalized to the average amplitude calculated in the naïve group . *: LSD test between trained and other groups at least p<0 . 002 . DOI: http://dx . doi . org/10 . 7554/eLife . 02109 . 006 In the go/no-go operant conditioning paradigm used for olfactory training , one stimulus is associated with a reward , whereas the second stimulus is neither punished nor rewarded . It is therefore possible that the sensory representation of the rewarded stimulus is more strongly potentiated compared to the non-rewarded one . However the amplitude of the glomerular responses evoked across all dilutions was not significantly different between the rewarded and non-rewarded stimuli ( Figure 4G ) . During the behavioral experiments , we started the training with different dilutions of Cin/Eu followed by dilutions of IAA/EB ( Figure 1C ) . This resulted in different post-training delays before recording evoked IOS for different odor pairs ( 4–5 weeks for Cin/Eu and 1-2 weeks for IAA/EB ) and allowed us to study whether the functional plasticity is long lasting or not . We compared the amplitude of the evoked glomerular responses across the three groups and found that the learning-induced potentiation is visible for all dilutions and lasts up to 5 weeks ( Figure 4H , I ) . In conclusion , olfactory learning induces a form of functional plasticity at the sensory input level , which is long lasting and independent of the stimulus reward value . Olfactory learning induces the potentiation of sensory inputs to the OB , increasing both the number of activated glomeruli and the strength of activation when compared to control animals . In order to investigate the behavioral relevance of this form of plasticity , we plotted the relationship between discrimination performance levels and measured glomerular response amplitudes for different odorant dilutions . These data were well fitted with a Boltzmann function for both discrimination tasks ( Figure 5A ) . In the rising phase of the curves , corresponding to the dilutions around discrimination threshold , a small change in input strength caused strong improvement in discrimination accuracy . This was further verified by comparing the animals’ accuracy at dilutions close to their discrimination threshold with the strength of OB inputs . At these specific dilutions , an increase in the performance level ( Figure 5B ) correlated with an increase in input strength; either by an increase in the number of activated glomeruli or by an increase in the amplitude of glomerular responses ( Figure 5C , D ) . 10 . 7554/eLife . 02109 . 007Figure 5 . Discrimination accuracy around discrimination threshold is dependent on input strength . ( A ) Plots showing the relationship between the average input strength monitored by IOS imaging and the discrimination accuracy at different concentrations for different odor pairs . The dotted lines represent Boltzmann function fits in the distributions of points ( Boltzmann fit , R2 > 0 . 95 , F = 511 . 34 , ANOVA p<0 . 032 ) . ( B ) Discrimination accuracy for odorant concentrations close to discrimination threshold . *: Paired t test: 10−3 vs 10−2 Cin/Eu , p<0 . 01 and 10−6 vs 10−4 IAA/EB , p<0 . 01 . ( C ) Average number of activated glomeruli for odorant concentrations close to discrimination threshold . *: Paired t test: 10−3 vs 10−2 Cin/Eu , p<0 . 05 and 10−6 vs 10−4 IAA/EB , p>0 . 1 . ( D ) Quantification of the average change in reflectance ( ΔR/R ) in the glomeruli activated by odorants at concentrations close to discrimination threshold . *: Paired t test: 10−3 vs 10−2 Cin/Eu , p>0 . 1 and 10−6 vs 10−4 IAA/EB , p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 02109 . 007 Since we observed that the learning process potentiated the inputs to the olfactory bulb , we propose that the trained animals would be able to detect and eventually discriminate more easily and rapidly odorants at lower concentrations than control animals ( i . e . , naïve and exposed groups ) . The discrimination threshold for odorants could potentially be shifted by few orders of concentration magnitude . The reported form of functional plasticity , induced by sensory experience , in the glomerular layer of the adult mouse olfactory system differs from previously reported plastic changes observed during the postnatal maturation of the olfactory system ( Zou et al . , 2004; Kerr and Belluscio , 2006 ) . During development , OSNs expressing different receptors can project to the same glomerulus . During the first 2 months of postnatal life ( Zou et al . , 2004 ) , a refinement of the projections occurs , leading to the known concept of a glomerulus receiving only afferents from sensory neurons expressing the same receptor . This form of structural plasticity is accelerated by sensory experience ( Kerr and Belluscio , 2006 ) , but it is not known if this causes any change in sensory input strength . In our study , the observed plasticity is induced during adulthood outside this developmental window , when one glomerulus is homogeneously innervated by one type of sensory neurons , and is associated to the increase of input strength . All forms of sensory experience are not equivalent in triggering plasticity of the sensory inputs . Indeed , we report that associative learning causes a form of plasticity whereas a passive repetitive exposure did not , as similarly observed in cortical areas of other sensory systems ( Bao et al . , 2001; Accolla and Carleton , 2008; Rosselet et al . , 2011; Lai et al . , 2012 ) . The passively exposed group of mice did not show any change in the input strength compared to the naïve ones . Differences between associative learning and passive exposure paradigms may be due to differences in the final concentration and presentation duration of the applied odorants . Though we cannot rule out this possibility , the lack of effect during passive exposure is consistent with a recent study ( Kato et al . , 2012 ) . On the other hand , it is contrasting with the previously reported enhancement of input sensitivity following long passive exposure ( Wang et al . , 1993 ) . In our study the total exposure time was less than an hour for one class of chemical stimuli ( including all dilutions ) whereas the exposure time was much longer in the latter study ( 16 hr daily during 2–6 weeks ) . In addition , the change in sensitivity reported previously was only observed for a barely detectable odorant in a selected mouse strain and not for other odorants . Therefore this non-generalized change in sensitivity might have resulted from extremely long exposure time for special odorants . Our associative learning paradigm potentiated the input strength for several odorants independently of their chemical class and reward value . Previous reports showed that training specifically affected the odorant representation of the rewarded stimulus ( Faber et al . , 1999 ) or the stimulus associated with a foot shock ( Kass et al . , 2013 ) . In contrast , we saw a similar potentiation for all odorants regardless of the reward value of the stimuli . What is the possible role of the reward value in causing the potentiation we observed ? In the go/no-go discrimination-learning paradigm , one stimulus is coupled with a positive reward , whereas the second stimulus is neither punished nor rewarded . This task involves the decision-making process for both rewarded and non-rewarded odors , which involves the assessment of reward value . This may explain the learning-induced potentiation we observed for rewarded as well as non-rewarded odors . The enhancement of sensory input strength for both odorants could activate the inhibitory neurons of the olfactory bulb thereby helping the refinement and the discrimination between odorants ( Abraham et al . , 2010 ) . IOS imaging is primarily monitoring neurotransmitter release from OSNs as evidenced by pharmacology experiments ( Gurden et al . , 2006 ) . Although we cannot completely rule out a possible postsynaptic contribution to the signal , studies comparing IOS with presynaptic-specific imaging readouts provide additional evidence supporting the presynaptic origin of IOS signals in the OB ( Wachowiak and Cohen , 2003; Soucy et al . , 2009 ) . In addition , a similar potentiation of the odor-evoked glomeruli activity induced by fear learning has been recently reported in anesthetized animals while monitoring a genetically encoded reporter of activity expressed specifically in OSNs ( Kass et al . , 2013 ) . Taken together , this suggests that the plasticity is associated with change in OSN activity . Which mechanism could explain a long-term change of OSNs activity ? Firstly , OSN sensitivity could be enhanced , leading to an increase of the odor evoked OSN firing rate causing more release of glutamate in the glomerulus . Secondly , olfactory training could modify OSN turnover ( Schwob et al . , 1992 ) leading to higher number of axons innervating the same glomerulus ( Jones et al . , 2008 ) . Thirdly , a local network mechanism could alter OSN release . Since activation of GABAB ( Aroniadou-Anderjaska et al . , 2000; McGann et al . , 2005 ) and dopamine D2 ( Hsia et al . , 1999; Ennis et al . , 2001 ) receptors on OSN terminals can decrease glutamate release , inhibition of GABA and/or dopamine release would lead to the observed increase of input strength . Finally , a direct increase in glutamate release mediated by neuromodulatory fibers may also account for the observed plasticity . Further experiments are needed to identify the exact mechanisms underlying the long lasting change in input strength induced by this olfactory learning . As we performed imaging experiments in awake mice , we took respiration behavior into account while analyzing the IOS readout . Previous studies have shown how an animal can increase its breathing frequency when sampling an odor ( Verhagen et al . , 2007; Carey et al . , 2009; Wesson et al . , 2009; Shusterman et al . , 2011 ) . The learning process can lead to a modulation of the respiration behavior in trained animals as compared to untrained animals . Moreover , it has been shown that prolonged fast respiration ( >4 Hz ) reduces odor-evoked Ca2+ signals in OSNs ( Verhagen et al . , 2007 ) . It is thus plausible that the modulation of the strength of IOS after training ( Figures 3 and 4 ) could arise from changes in breathing strategy . However , we did not observe significant changes in the breathing frequency of mice when monitored at the beginning and at the end of discrimination training ( Figure 6 ) . The frequency was also comparable to the one previously measured in naïve awake head-restrained animals ( 3 . 1 Hz , Patterson et al . , 2013 ) . Likewise , IOS amplitude does not vary with the animal’s breathing frequency ( Figure 7 ) . Therefore , the IOS potentiation we report here mostly reflects neural changes induced by the learning process rather than respiration modulations . Along this line , a similar potentiation of the odor-evoked glomeruli activity induced by fear learning has been recently reported in anesthetized animals where effects are presumably independent of respiration ( Kass et al . , 2013 ) . Altogether these data suggest that the potential role played by breathing behavior is minor as a source of modulation of sensory input strength in olfactory associative learning tasks . 10 . 7554/eLife . 02109 . 008Figure 6 . Respiration behavior is not altered by olfactory discrimination learning . ( A ) Performance levels shown by mice at the beginning ( black , 300 trials averaged ) and at the end of a discrimination training task ( red , 300 trials averaged ) with Cineol and Eugenol ( *paired t test , p<0 . 001 , n = 6 mice ) . ( B ) Average respiration frequency from the respective training blocks in ( A ) . Sniff frequency remained unaltered during different learning epochs ( *paired t test , p=0 . 8 , n = 6 mice ) . Values are represented as mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 02109 . 00810 . 7554/eLife . 02109 . 009Figure 7 . Odor-evoked intrinsic signals are independent of change in breathing frequency . ( A ) Single trial map of the amyl acetate-evoked activity reported by IOS when the mouse is breathing at 1 . 8 ( left ) and 3 Hz ( right ) . LUT: −0 . 003 to 0 . 003 . The respiration pattern recorded during each trial is shown below each image ( I: inspiration , E: expiration ) ; the light gray vertical bar represents odor presentation ( 5 s ) . ( B ) Average values of glomerular response amplitude ( ΔR/R ) at 1 . 8 Hz , plotted against amplitudes at 3 Hz ( Wilcoxon signed-rank test , n = 104 glomeruli from three mice , p=0 . 2945 ) . ( C ) Average values of glomerular response amplitude at different breathing frequencies ( 2 , 2 . 2 , 2 . 4 , 2 . 6 , 2 . 8 and 3 Hz ) . Values are normalized relative to responses recorded at 1 . 8 Hz ( Repeated measure one-way ANOVA , n = 3 mice , F = 0 . 4499 , p=0 . 6505 ) . Values are represented as mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 02109 . 009 What is the behavioral relevance of this plasticity ? The improved representations mediated by the training can account for discrimination accuracy and reaction time stability across a wide range of odorant dilutions . The observed increase in the sensitivity at higher dilutions of odorants tested should lower the odorant detection thresholds in mice . Interestingly , in human olfaction , trained wine tasters show lower detection thresholds and an increase in the perceived odorants intensity compare to healthy non-trained subjects ( Marino-Sanchez et al . , 2010; Tempere et al . , 2011 ) . Although the effect of the plasticity reported in our study lasted for 5 weeks , this enhanced representation may be lasting for longer even with brief repetitive training and may account for the long lasting perception abilities seen in professional wine tasters . Irrespective of the molecular mechanisms , our study provides physiological evidence for the existence of a functional plasticity at the sensory periphery , which helped the animal to achieve fast and accurate odor discriminations . All experiments were performed on adult male C57BL/6J mice ( 11 weeks old at the beginning of the behavioral experiments , Charles River France ) in accordance with the Swiss Federal Act on Animal Protection and Swiss Animal Protection Ordinance , University of Geneva and the state of Geneva ethics committee ( authorization 1007/3758/2 ) . Odorants used were iso-amyl acetate ( IAA , ≥99% purity ) , ethyl butyrate ( EB , ≥99% ) , 1 , 4-cineol ( Cin , ≥85% ) , eugenol ( Eu , ≥99% ) . All chemicals and mineral oil were obtained from Sigma-Aldrich ( Germany ) or Fluka Chemie ( Germany ) . All olfactory discrimination experiments were performed using four modified eight-channel olfactometers ( Knosys , Lutz , FL ) controlled by custom routine ( kindly provided by Dr Andreas Schaefer , National Institute for Medical Research , UK ) written in Igorpro ( Wavemetrics , Portland , OR ) . Odorants were diluted from 100 to 10−10 percent volume in mineral oil and further diluted 1:20 by airflow . Odorants were made freshly for each task . The task habituation training , olfactory training and reaction/discrimination time measurements were conducted as previously published ( Abraham et al . , 2004 , 2010 , 2012 ) . In brief , a trial is initiated by breaking a light beam at the sampling port opening . This opens one of eight odor valves and a diversion valve ( DV ) that allows all airflow to be diverted away from the animal for 500 ms . After the release of DV , the odor is applied to the animal for 2 s . Trials were counted as correct if the animals met the criteria we set for water delivery ( licking at least once in three out of four 500 ms bins ) upon presentation of S+ or if licking did not occur in more than one out of four 500 ms bins for S− . For correct S+ trials mice can receive a 2–4 µl water reward at the end of 2 s stimulus period . Conversely for the incorrect S+ and correct S− trials no reward is supplied . A trial cannot be initiated unless an inter-trial interval of at least 5 s has passed . This interval was sufficiently long so that animals typically retract quickly after the end of the trial . The minimal inter-stimulus interval was thus 5 s , which seemed to be sufficient as no habituation could be observed ( performance was not correlated with the actual inter-trial interval chosen by the animal , which was around 10–20 s ) . No minimal sampling time was required to artificially enforce the animal to take a decision . Odors are presented in a pseudo-randomized scheme ( no more than two successive presentations of the same odor ) . The trained group of mice was evenly distributed between the setups and the valence of odorants in a pair ( S+ and S− ) was switched between animals . All activated glomeruli included in our quantification had therefore an equal chance to be associated with a rewarded odor and a non-rewarded odor ( Figure 4G ) . Upon presentation of a S+ odor , the animal generally continuously breaks the beam , whereas upon presentation of an S− odor an animal familiar with the apparatus usually quickly retracts its head . Reaction times were calculated as follows: for every time point , beam breakings for S+ and S− odors were compared by bootstrapping , yielding significance value as a function of time after odor onset . The last crossing of the p=0 . 05 line determined the reaction time . In very few cases , this did not coincide with the visually identified reaction time ( point of largest curvature in the log[p]-t plot ) and was corrected after visual inspection . Following the training on 100 Cin vs Eu ( 1200 trials , Figure 2A ) , mice were trained on a pair of natural odorants ( cloves vs camphor ) for another experiment , which took 2 weeks . They were then trained on the different dilutions of odorants reported in this study . For the passive exposure experiment we used the same odorant delivery system , the same number of trials and the same pseudo-randomized sequence of odorants compared to the behavior experiments . Each time , the mice were exposed in their home cage to 2 s of odor plumes at an inter-trial interval ( ITI ) of ∼20–40 s , depending on the odorant pair and calculated from the average ITI’s observed during olfactory training . During odor applications , mice were clearly investigating around the odor tube outlet . Each day 300 trials ( 150 rewarded and 150 non-rewarded ) were presented as this was the highest number of trials finished per day by best performers . For data presented in Figure 6 , we performed an odor discrimination task in head-restrained mice as described previously ( Abraham et al . , 2012 ) . Respiration was detected via a directional airflow sensor and breathing frequency was computed during the 2 s odor presentation ( AWM2100V; Honeywell , Germany ) . For IOS imaging performed in awake head-restrained mice , the animal preparations and head post implantations were done as described previously ( Vincis et al . , 2012 ) . For IOS imaging performed in anesthetized mice ( Figure 7 ) , animals were deeply anesthetized by intraperitoneal injection ( i . p . ) of 3 . 1 μl/g body weight of a mixture consisting of 60 μl Medetomidin ( Dormitor , Pfizer AG , Zurich , Switzerland; 1 mg/ml ) , 160 Midazolam ( Dormicum , Roche Pharma AG , Switzerland; 5 mg/ml ) and 40 μl Fentanyl ( Sintenyl , Sintetica SA , Mendrisio , Switzerland; 50 μg/ml ) . A local anesthetic , carbostesin ( AstraZeneca , Zug , Switzerland ) , was subcutaneously injected before any skin incision . Anesthesia was maintained by periodic dosage ( ∼30 μl i . p . every 30 min ) of mixture containing only Midazolam ( 5 mg/ml ) and Medetomidin ( 1 mg/ml ) . Odorants were delivered for 5 s ( 2 s after recording onset ) using a custom made olfactometer ( Bathellier et al . , 2008; Gschwend et al . , 2012 ) and images were acquired at 700 nm wavelength using the Imager 3001F system ( Optical Imaging Ltd . , Israel ) ( Accolla et al . , 2007; Vincis et al . 2012 ) . The number of repetitions for each odorant/dilution was four . The glomerulus detection procedure was done on individual time frames by drawing regions of interest ( ROI ) . The ROI analyzed for the ΔR/R measurement , were selected based on the glomerular map obtained for each mice at the lowest dilution of four odorants used . This map was then used as the reference across different dilutions of the same odorant to calculate the ΔR/R . The same procedure was adopted for each experimental group . We excluded every region that appeared only in a single frame or that looked like blood vessels . For the experiment shown in Figure 7 , frequency of respiration was detected via a directional airflow sensor ( AWM2100V; Honeywell , Germany ) .

The mammalian brain is not static , but instead retains a significant degree of plasticity throughout an animal’s life . It is this plasticity that enables adults to learn new things , adjust to new environments and , to some degree , regain functions they have lost as a result of brain damage . However , information about the environment must first be detected and encoded by the senses . Odors , for example , activate specific receptors in the nose , and these in turn project to structures called glomeruli in a region of the brain known as the olfactory bulb . Each odor activates a unique combination of glomeruli , and the information contained within this ‘odor fingerprint’ is relayed via olfactory bulb neurons to the olfactory cortex . Now , Abraham et al . have revealed that the earliest stages of odor processing also show plasticity in adult animals . Two groups of mice were exposed to the same two odors: however , the first group was trained to discriminate between the odors to obtain a reward , whereas the second group was passively exposed to them . When both groups of mice were subsequently re-exposed to the odors , the trained group activated more glomeruli , more strongly , than a control group that had never encountered the odors before . By contrast , the responses of mice in the passively exposed group did not differ from those of a control group . Given that the response of glomeruli correlates with the ability of mice to discriminate between odors , these results suggest that trained animals would now be able to discriminate between the odors more easily than other mice . In other words , sensory plasticity ensures that stimuli that have been associatively learned with or without reward will be easier to detect should they be encountered again in the future .

2014

Long term functional plasticity of sensory inputs mediated by olfactory learning

Contamination with exogenous DNA is a constant hazard to ancient DNA studies , since their validity greatly depend on the ancient origin of the retrieved sequences . Since contamination occurs sporadically , it is fundamental to show positive evidence for the authenticity of ancient DNA sequences even when preventive measures to avoid contamination are implemented . Recently the presence of wheat in the United Kingdom 8000 years before the present has been reported based on an analysis of sedimentary ancient DNA ( Smith et al . 2015 ) . Smith et al . did not present any positive evidence for the authenticity of their results due to the small number of sequencing reads that were confidently assigned to wheat . We developed a computational method that compares postmortem damage patterns of a test dataset with bona fide ancient and modern DNA . We applied this test to the putative wheat DNA and find that these reads are most likely not of ancient origin . The evolutionary reconstruction of the past has been greatly enriched by direct interrogation of ancient DNA ( aDNA ) from plants and animal remains ( Shapiro and Hofreiter , 2014 ) . Although a vast proportion of flora and fauna do not fossilize , traces of their DNA may be preserved in sediments allowing the characterization of past biodiversity ( Pedersen et al . , 2015 ) . A challenge to exploiting such resources is the ubiquitous threat of contamination with exogenous DNA . Therefore , special sample preparation procedures have been developed to reduce DNA contamination ( Cooper and Poinar , 2000 ) . Nevertheless , it remains difficult to estimate how well preventive measures work . If contamination is a possible explanation for the result , it is crucial to exclude this possibility by giving positive evidence for the authenticity of aDNA ( Prüfer and Meyer , 2015 ) . Fortunately , a large number of full-length DNA sequences can be generated using next generation sequencing , which allows for the authentication of aDNA . In aDNA an excess of C-to-T ( cytosine to thymine ) substitutions occur at the 5′ and 3′ ends of molecules ( or its mirror image G-to-A ( guanine to adenine ) at the 3′ end , depending on the library protocol employed ) . When considering the 5′ end of sequences , the excess of C-to-T substitutions is highest at the first base and decreases exponentially towards the center ( Figure 1A ) . This pattern is the result of cytosine deamination to uracil in single stranded overhangs ( Briggs et al . , 2007 ) . Since it is present in aDNA-derived sequences but absent in much younger samples , it has been used as an authentication criterion in aDNA experiments ( Krause et al . , 2010; Prüfer and Meyer , 2015 ) . 10 . 7554/eLife . 10005 . 003Figure 1 . Patterns of cytosine to thymine ( C-to-T ) substitutions at the 5′end of known modern and ancient DNA . ( A ) C-to-T substitutions at the 5′ end from a whole library of historic Solanum tuberosum ( ancient DNA ) . The line shows the fit with the exponential distribution and the box the goodness-of-fit p-value . ( B ) C-to-T substitutions at the 5′ end from a whole library of present-day Triticum aestivum ( modern DNA ) . Line and box as in ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10005 . 003 Smith et al . analyzed sediments from Bouldnor Cliff , a submerged archeological site in the United Kingdom , and suggested the presence of domesticated wheat 8000 years ago based on sedimentary ancient DNA ( sedaDNA ) . This is 2000 years earlier than expected based on archeological remains in the British Isles and 400 years earlier than in nearby European sites ( reviewed in Smith et al . ) . Since Smith et al . did not find wheat pollen or archeological remains associated with wheat cultivation , they conclude that the wheat presence in Bouldnor Cliff was the result of trading . In total they produced ∼72 million Illumina reads , of which they robustly assigned 152 to wheat ( Triticum ) , with dozens more ( 160 reads ) to higher taxonomic ranks that include wheat . Smith et al . took state-of-the art preventive measures to avoid contamination and exercised great effort to ensure the accuracy and robustness of their phylogenetic assignments . The authors attempted to authenticate the aDNA molecules based on the expected excess of C-to-T substitutions , but because of the very small number of reads assigned to wheat , they failed to do so using standard approaches . As a result of that , the authors did not present any positive evidence for the ancient origin of their reads . Here we present an approach that compares the pattern of C-to-T substitutions in a set of test reads with the distributions of C-to-T substitutions in reads from known ancient- and modern-DNA and apply this approach to sedaDNA from Smith et al . Although the excess of C-to-T substitutions at the 5′ end occurs at different magnitudes in samples of different ages , the exponential increase of substitutions towards the end is a ubiquitous pattern in aDNA studies ( Sawyer et al . , 2012 ) . In order to score the presence of this pattern in various datasets , we fitted an exponential function and evaluated the goodness of fit by using a one-sided t-test to test for significant exponential decay . As expected , true aDNA libraries show significant goodness-of-fit p-values ( Figure 1A ) , whereas non-significant goodness-of-fit p-values , neither decay nor growth , are observed in libraries derived from modern DNA ( Figure 1B ) . A given C-to-T damage pattern plot can thus be summarized by its goodness-of-fit p-value that when it is significant indicates C-to-T exponential decay at the 5′ end ( Figure 1A ) . We resampled ( with replacement ) 10 , 000 sets of 150 sequences from a library of historic Solanum tuberosum collected in 1846 ( Yoshida et al . , 2013a ) . The number was selected to be comparable to the 152 reads that Smith et al . assigned to wheat . An empirical distribution of goodness-of-fit p-values was generated by performing the goodness-of-fit test for each subsample ( Figure 2A ) . When we evaluate the sedaDNA goodness-of-fit p-value , we find that it falls within the upper 3% of subsamples with the least good fit . We can therefore reject the null hypothesis that the sequences assigned to wheat are as ancient as the historic S . tuberosum library . We repeated the whole procedure using this time a modern wheat library to generate the distribution of goodness-of-fit p-values ( Figure 2A ) and find a better match ( p = 0 . 83 ) . Thus , we cannot reject the hypothesis that the sequences assigned to wheat are of modern origin . 10 . 7554/eLife . 10005 . 004Figure 2 . Authenticity test of DNA reads assigned to Triticum by Smith et al . ( A ) The histograms in the center panel show the empirical distributions of goodness-of-fit p-values of subsamples of 150 reads from ancient and modern DNA ( same libraries as in Figure 1 ) . The dotted red line indicates the location of the goodness-of-fit p-value from reads assigned to wheat in sedimentary ancient DNA . The four surrounding panels show cytosine to thymine ( C-to-T ) substitutions at the 5′ end extracted from different point of the goodness-of-fit p-value distributions , and from the reads assigned to wheat in sedimentary ancient DNA . ( B ) Variation of the empirical p-value of the test depending on the goodness-of-fit p-value of the whole library used to generate the empirical distribution . Numbers adjacent to the points indicate the percentage of C-to-T substitutions at first base . Red arrow indicates the aDNA library used as test in Figure 3A . Purple arrow indicates the library used to generate the empirical distribution of goodness-of-fit p-values in Figure 3A–C . DOI: http://dx . doi . org/10 . 7554/eLife . 10005 . 00410 . 7554/eLife . 10005 . 005Figure 2—figure supplement 1 . Patterns of cytosine to thymine ( C-to-T ) substitutions at the 5′end from a 7 . 000-year-old Mesolithic human from La Braña site in Northern Iberia . The line shows the fit with the exponential distribution . The goodness-of-fit p-value is indicated in the upper right corner . DOI: http://dx . doi . org/10 . 7554/eLife . 10005 . 00510 . 7554/eLife . 10005 . 006Figure 2—figure supplement 2 . Authenticity test of DNA reads assigned to Triticum by Smith et al . The histograms shows the empirical distributions of goodness-of-fit p-values of subsamples of 150 reads from a 7 . 000-year-old Mesolithic human from La Braña site in Northern Iberia . The dotted red line indicates the location of the goodness-of-fit p-value from reads assigned to wheat in sedimentary ancient DNA . DOI: http://dx . doi . org/10 . 7554/eLife . 10005 . 006 We sought to investigate how the test behaves when the empirical distribution of goodness-of-fit p-values is generated from different aDNA libraries . For this purpose we used a set of samples from animal ( Sawyer et al . , 2012 ) and plant remains ( Yoshida et al . , 2013 ) with an age of 85–170 years before present , and scored the sedaDNA wheat sequences against distributions generated from these libraries ( subsamples of 150 sequences again ) . We observed that the goodness-of-fit p-value for the libraries is positively correlated with the empirical p-value for the sedaDNA wheat sequences tested against them ( Figure 2B ) . Using a significance level of 0 . 05 , we rejected the hypothesis that the wheat sequences are of ancient origin with 7 out of 13 libraries used in our test ( Figure 2B ) . Thus , the purportedly 8000-year old wheat sequences show a less pronounced deamination pattern than many plant and animal samples with an age of less of 200 years . Finally , we took a less conservative approach and scored the sedaDNA against a distribution of goodness-of-fit p-values ( subsamples of 150 read ) generated from a 7000-years-old human Mesolithic sample from la Braña site in Northern Iberia ( Olalde et al . , 2014 ) . La Braña is a site with cold environment and stable thermal conditions that has yielded exceptionally well conserved human fossils with ∼50% of human endogenous DNA that reach a ∼15% C-to-T substitution rate at the 5′ end ( Olalde et al . , 2014a ) ( Figure 2—figure supplement 1 ) . We could reject the null hypothesis that the sedaDNA reads are as ancient as the sample from la Braña ( p = 0 . 0014 ) , a sample that is closer in time with the allegedly 8000-year-old wheat reads ( Figure 2—figure supplement 2 ) . It is worth pointing out that almost all 10 , 000 subsamples from la Braña had a very low ( close to 0 ) goodness-of-fit p-value , even though we subsample only 150 reads ( Figure 2—figure supplement 2 ) . We assessed the statistical power of the test by testing both an aDNA ( Figure 3A ) and a modern DNA library ( Figure 3B ) against a distribution built from a bona fide aDNA library , while varying the number of sampled sequences . Whereas the hypothesis that a true aDNA library is ancient was never rejected ( Figure 3A ) , the hypothesis that a modern library has ancient origin could be rejected only when sufficient number of sequences were used for the subsample test ( in tests with more than 300 reads the median empirical p-value was always below 0 . 05 ) ( Figure 3B ) . 10 . 7554/eLife . 10005 . 007Figure 3 . Evaluation of test performance . ( A ) Variation of the empirical p-value of the test depending on the number of reads sampled from an ancient DNA library ( indicated with red arrow in Figure 2B ) . ( B ) Variation of the empirical p-value of the test depending on the numbers of reads subsampled from modern DNA Triticum aestivum library ( same library used to generate the distribution of empirical goodness-of-fit p-values in Figure 2A ) . ( C ) Variation of the empirical p-value of the test depending on the size of sample sets from sedimentary ancient DNA reads mapped directly to the T . aestivum genome . Box-and-whisker plots were built based on 1000 tests . Layers as reported in Smith et al . i . e . layer 1 ( most superficial ) , layer 4 ( more deep ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10005 . 007 Finally , we skipped the phylogenetic curation step applied by Smith et al . to reduce the number of false positive wheat alignments , and mapped all reads sequenced by Smith et al . to the wheat genome . After stringent filtering of sedaDNA mappings we repeated our test varying the size of the subsample sets from 100 to 1000 reads . The empirical p-value was dependent on the number of reads tested , and declined with an increasing number of tested reads for all layers of sediments sequenced in Smith et al ( Figure 3C ) . This pattern resembled the one obtained from a modern DNA library ( Figure 3B ) . As for the phylogenetic curated 152 sequences , we were able to reject the hypothesis that the mapped reads are of ancient origin ( mean p-value < 0 . 05 for all tests with more than 400 reads for layers 1–2 and 4 , and 800 reads for layer 3 ) . Our analysis also shows that the 152 sequences after phylogenetic curation are not a biased subsample from the distribution of all wheat-matching sequences . We were able to reject the hypothesis that the sequences assigned to wheat by Smith et al . are of ancient origin . This is true even when we compared the putative 8000 year old sequences with only century old samples that show much lower deamination signatures . This means that a scenario in which wheat was transported to the Bouldnor Cliff site 8000 years ago is unwarranted . Our approach for authentication of aDNA can be used even with a very small number of sequences , and we hope that it will proof useful to test for positive evidence of authenticity for ancient DNA studies whose conclusions rely heavily on the ancient origin of the analyzed sequences . Reads from most of the samples were downloaded from the European Nucleotide Archive ( Table 1 and Supplementary file 1 ) , with the exception of the Gorilla gorilla reads that were provided directly by the authors ( Sawyer et al . , 2012 ) . Adapters were trimmed for both paired- and single-end runs using the program Skewer ( version 0 . 1 . 120 ) using default parameters ( Jiang et al . , 2014 ) . For paired-end runs ( Supplementary file 1 ) forward and reverse reads were merged requiring a minimum overlap of 10 base pairs ( bp ) using the program Flash ( version 1 . 2 . 11 ) ( Magoc and Salzberg , 2011 ) . Merged or single-end reads were mapped as single-end reads against their respective nuclear or organellar genomes: S . tuberosum nuclear genome ( Potato Genome Sequencing Consortium et al . , 2011 ) , Solanum lycopersicum nuclear genome ( The Tomato Genome Consortium , 2012 ) , Triticum aestivum nuclear genome ( International Wheat Genome Sequencing C , 2014 ) , G . gorilla mitochondrial genome ( Xu and Arnason , 1996 ) , Homo sapiens nuclear genome ( Genome Reference Consortium Human Build 37 ) . The mapping was carried out using BWA-MEM ( version 0 . 7 . 10 ) with default parameters , which include a minimum read length of 30 bp ( Li , 2013 ) . PCR duplicates were removed after mapping using bam-rmdup ( available at https://github . com/udo-stenzel/biohazard ) , which computes a consensus sequence for each cluster of duplicated sequences . Alignments were stored in the bam format ( Li et al . , 2009 ) . 10 . 7554/eLife . 10005 . 008Table 1 . Provenance of samplesDOI: http://dx . doi . org/10 . 7554/eLife . 10005 . 008SpeciesType of DNAAgeReferenceStudy IDSample/run IDMetagenomics sampleSedimentary8030-7908*Smith et al . , 2015PRJEB6766‡ERR567364‡Metagenomics sampleSedimentary8030-7908*Smith et al . , 2015PRJEB6766‡ERR567365‡Metagenomics sampleSedimentary8030-7908*Smith et al . , 2015PRJEB6766‡ERR567366‡Metagenomics sampleSedimentary8030-7908*Smith et al . , 2015PRJEB6766‡ERR567367‡Metagenomics sampleSedimentary8030-7908*Smith et al . , 2015PRJEB6766‡ERR732642‡T . aestivumModernNAChapman et al . , 2015PRJNA250383‡SRR1170664‡S . tuberosumAncient135†Yoshida et al . , 2013PRJEB1877‡ERR267886‡S . tuberosumAncient137†Yoshida et al . , 2013PRJEB1877‡ERR267882‡S . tuberosumAncient149†Yoshida et al . , 2013PRJEB1877‡ERR330058‡S . tuberosumAncient165†Yoshida et al . , 2013PRJEB1877‡ERR267872‡S . tuberosumAncient166†Yoshida et al . , 2013PRJEB1877‡ERR267868‡S . tuberosumAncient166†Yoshida et al . , 2013PRJEB1877‡ERR957324‡S . tuberosumAncient167†Yoshida et al . , 2013PRJEB1877‡ERR267868‡S . lycopersicumAncient136†Yoshida et al . , 2013PRJEB1877‡ERR267884‡S . lycopersicumAncient139†Yoshida et al . , 2013PRJEB1877‡ERR267878‡G . gorillaAncient83†Sawyer et al . , 2012NA107¶G . gorillaAncient100†Sawyer et al . , 2012NA109¶G . gorillaAncient100†Sawyer et al . , 2012NA110¶G . gorillaAncient103†Sawyer et al . , 2012NA114¶Homo sapiensAncient7000*Olalde et al . , 2014PRJNA230689‡SRR1045127*B . P . ( before present years ) . †Calculated from collection date ( in years ) . ‡IDs from the European Nucleotide Archive . ¶IDs from Sawyer et al . , 2012 . We used two different approaches to process the reads from sedimentary DNA ( Smith et al . , 2015 ) . Phylogenetic curated reads: we used a set of 152 reads assigned to tribe Triticeae and to genus Triticum by Smith et al . after phylogenetic curation . However , we consider the complete sequence and do not exclude the initial 10 nucleotides as was done in the original processing ( Smith et al . , 2015 ) . Reads were then aligned to the wheat genome as described above . All sedimentary DNA reads: we aligned independently reads from all four layers sequenced by Smith et al . to the T . aestivum nuclear genome ( International Wheat Genome Sequencing C , 2014 ) . Duplicates were removed and only alignments with mapping quality greater or equal than 30 were used for further analysis . Additionally , we include a sequence complexity filter based on entropy , which removed low complexity reads with entropy less or equal to 50 . The entropy filtering was carried out with prinseq-lite ( version 0 . 20 . 4 ) ( Schmieder and Edwards , 2011 ) . For each set of aligned reads ( complete libraries or subsamples ) the C-to-T substitutions patterns along the 5′ end of the read were assessed using the program PMDtools ( Skoglund et al . , 2014 ) . We fitted an exponential function to the frequency of C-to-T substitutions for the first 20 nucleotides at the 5′ end . The fitting was performed in R ( http://www . r-project . org ) using the nls function , which determines the nonlinear least squares estimates of the parameters in a nonlinear model . The fitting was carried out with the model formula: y ∼ N∗exp ( −rate∗x ) . From the nls fitting we obtained the t-value and degrees of freedom for the rate parameter and then calculated a goodness-of-fit p-value by using a one-sided t-test . Subsets of different alignment numbers were randomly sampled ( with replacement ) 10 , 000 times from alignments stored in the bam format ( Li et al . , 2009 ) . The random sampling was performed using samtools view ( Li et al . , 2009 ) . For every subset of alignments we assessed the fraction of C-to-T substitutions , fitted an exponential function and calculated a goodness-of-fit p-value as explained above . Phylogenetic curated reads: we compare the goodness-of-fit p-value of our test set of 152 sedimentary DNA reads with distributions of goodness-of-fit p-values generated from bona fide modern and ancient DNA . For the distribution of goodness-of-fit p-values from aDNA , we count how many of them are equal or greater than the sedimentary DNA goodness-of-fit p-value . To calculate the empirical p-value of the test we subsequently divided this number by the total number of values in the empirical distribution . With this approach we test the null hypothesis that the test set of reads contains a signal of ancient DNA damage that is comparable or even more pronounced than the signal in the aDNA library used to generate the empirical distribution of goodness-of-fit p-values . For the distribution of goodness-of-fit p-values from modern DNA , we count how many of them were smaller or equal than the sedimentary DNA goodness-of-fit p-value . We calculate the empirical p-value of the test by dividing this number by the total number of p-values in the empirical distributions . With this approach we test the null hypothesis that the test set of reads matches the absence of ancient DNA damage patterns seen in reads of modern origin . All sedimentary DNA reads: We tested independently alignments from each of the layers sequenced by Smith et al . using a bona fide aDNA sample for the generation of the distribution of goodness-of-fit p-values . For each layer we tested 10 sets of different numbers of reads ( from 100 to 1000 reads , with increments of 100 reads ) . For each layer and for each number of reads in the test set we repeated the test and calculated the empirical p-value 1000 times as described above . Other ancient DNA and modern DNA libraries were tested using the same procedure .

Ancient DNA , that is to say DNA extracted from fossils and ancient remains , provides a window into the past lives of humans , animals and plants . But working with ancient DNA is challenging; DNA decomposes with time , and so ancient DNA is often fragmented , damaged and present in tiny quantities . Furthermore , ancient DNA is also easily contaminated by modern DNA from those handling it and its surroundings . Researchers have therefore developed special protocols for working with ancient DNA and tests for its contamination . One approach used to check that DNA is of ancient origin identifies a pattern of damage that is specific to ancient DNA . This damage changes the building blocks that make up DNA , causing one ( called cytosine or C ) to be misread as another ( thymine or T ) . This substitution occurs most frequently at the ends of ancient DNA molecules , and occurs less often along its length , forming a detectable and characteristic pattern of damage . A common way to analyse ancient DNA is to sequence it and then compare the resulting sequences to the genomes of modern organisms to identify its origins . In a study published earlier in 2015 , investigators sequenced the DNA present in sediments obtained from a submerged archaeological site off the coast of the Isle of Wight in the United Kingdom . This previous study identified some DNA fragments that matched sequences in the wheat genome . This led the investigators to conclude that wheat was present in the British Isles around 8000 years ago , some 2000 years earlier than previously thought . However , possibly owing to the small number of fragments that were found , the previous study did not check if the damage pattern matched that expected for ancient DNA . Now , Weiß et al . have developed a new computational method that tests whether DNA shows a typically ancient , or typically modern , pattern of C-to-T substitutions . When this test was used to assess the wheat sequences that were previously claimed to have ancient origins , it revealed that their pattern of DNA damage did not fit statistically with those of ancient DNA . Weiß et al . 's findings contest those of the earlier study , and suggest that the new statistical method could be used to authenticate ancient DNA even when the number of available sequences is low .

2015

Contesting the presence of wheat in the British Isles 8,000 years ago by assessing ancient DNA authenticity from low-coverage data

Interactions between membrane protein interfaces in lipid bilayers play an important role in membrane protein folding but quantification of the strength of these interactions has been challenging . Studying dimerization of ClC-type transporters offers a new approach to the problem , as individual subunits adopt a stable and functionally verifiable fold that constrains the system to two states – monomer or dimer . Here , we use single-molecule photobleaching analysis to measure the probability of ClC-ec1 subunit capture into liposomes during extrusion of large , multilamellar membranes . The capture statistics describe a monomer to dimer transition that is dependent on the subunit/lipid mole fraction density and follows an equilibrium dimerization isotherm . This allows for the measurement of the free energy of ClC-ec1 dimerization in lipid bilayers , revealing that it is one of the strongest membrane protein complexes measured so far , and introduces it as new type of dimerization model to investigate the physical forces that drive membrane protein association in membranes . Membrane protein folding involves the favorable association of non-polar protein interfaces amidst an excess of similarly non-polar lipid solvent ( Popot and Engelman , 1990 ) . Surprisingly , the thermodynamic forces driving this assembly remain poorly understood , due to a shortage of experimental systems where reversible equilibrium association can be observed in membranes . Dimerization models of single-pass transmembrane ( TM ) helices have provided a tractable system for free-energy measurements in detergent micelles ( Fleming et al . , 1997; MacKenzie and Fleming , 2008 ) , and recently , in lipid bilayers ( North et al . , 2006; Chen et al . , 2010; Hong et al . , 2010; Yano et al . , 2011 ) . However , the relatively small change in solvent accessible surface area upon dimerization ( MacKenzie et al . , 1997 ) limits their potential to study protein-specific van der Waals interactions and lipid-solvent-dependent effects , the two driving forces hypothesized to be major players within the membrane environment ( Popot and Engelman , 1990; White and Wimley , 1999; Bowie , 2005 ) . Alternatively , studying the dimerization of multi-TM helix membrane proteins offers a new approach , as these interfaces are much larger , and each subunit is expected to adopt a stable , functional fold that could constrain the reaction to a two-state equilibrium . One example that appears particularly well suited is the homodimeric ClC-ec1 Cl-/H+ antiporter native to Escherichia coli ( Maduke et al . , 1999; Dutzler et al . , 2002 ) . This is a 50-kDa membrane protein that dimerizes via a membrane embedded , non-polar interface lined mainly by isoleucines and leucines ( Figure 1—figure supplement 1A ) . Our previous work showed that insertion of bulky tryptophans at the interface destabilized the dimer in detergent , while preserving functional 2:1 Cl-/H+ transport and structural fold as ascertained by X-ray crystallography ( Robertson et al . , 2010 ) . Furthermore , a distant ClC homologue , ClC-F , shows equilibrium exchange in detergent micelles ( Last and Miller , 2015 ) , raising the possibility of free-energy measurements of ClC dimerization in membranes . Here , we measure the equilibrium dimerization free energy of ClC-ec1 in lipid bilayers by diluting the protein into large membranes and measuring the change in the monomer vs . dimer population . If the system is in a state of dynamic equilibrium , then diluting the protein in the lipid bilayer will shift the population to the monomeric state . To measure the proportion of monomers and dimers as a function of density , we incubated Cy5-labeled ClC-ec1 in large 10 μm diameter multilamellar vesicles ( MLVs ) , then measured the probability that 1 , 2 , or more Cy5-labeled subunits are captured into extruded liposomes by single-molecule photobleaching analysis using total internal reflection fluorescence ( TIRF ) microscopy . This approach measures the monomer-dimer equilibrium in the MLV state at the point of extrusion , and as such reports the statistical mechanical dimerization free energy in the lipid bilayer . The sensitivity of the single molecule approach allows for inspection of the protein at sub-biological densities , i . e . less than one subunit per typical cell membrane . With this technical development in hand , we determined that equilibrium ClC-ec1 subunit exchange occurs on a laboratory timescale and that the reaction follows an equilibrium dimerization isotherm as a function of protein density in the membrane . This allows for the measurement of the free energy of ClC-ec1 dimerization in lipid bilayers and the change in free energy due to tryptophan substitutions at the dimerization interface . This work introduces ClC-ec1 as an ideal platform for investigating thermodynamic driving forces underlying membrane protein assembly in membranes . In a previous study , we showed that tryptophan substitutions I201W and I422W at the dimerization interface of ClC-ec1 ( Figure 1—figure supplement 1 ) yield a functionally folded , monomeric form of the transporter in lipid bilayers ( Robertson et al . , 2010 ) . To set up the system for fluorescence studies , we moved a partially buried cysteine to a more accessible position , C85A/H234C that allows for quantitative labeling by Cy5-maleimide without impacting stability ( Figure 1—figure supplement 2 ) . For simplicity , we will refer to this single-exposed cysteine construct as WT , and tryptophan substitutions as W ( I422W ) and WW ( I201W/I422W ) . To measure the dimerization reaction of ClC-ec1 in lipid bilayers ( Figure 1A ) , we reconstituted Cy5-labeled protein in 2:1 POPE/POPG lipids at different mole fractions ( χ subunit/lipid ) and freeze/thawed the proteoliposomes to produce ~10 µm diameter multilamellar vesicles ( MLVs ) ( Pozo Navas et al . , 2005 ) . This creates a model of an infinite bilayer where subunits can exchange with one another , even at low dilutions ( Figure 1B ) . After equilibration , the membranes are fractionated by extrusion ( Figure 1C ) , forming small liposomes that are then imaged on a TIRF microscope ( Figure 1D ) for single-molecule photobleaching analysis ( Figure 1—figure supplement 3D ) . With single-molecule sensitivity , we can count the number of subunits captured into each liposome ( Figure 1E ) and determine the probability distribution of fluorescent protein occupancy in liposomes ( Figure 1F ) . This approach extends another single-vesicle fluorescence method that examines the behavior of membrane proteins in individual liposomes ( Mathiasen et al . , 2014 ) . However , in our case , we are not studying the state of the protein in the final proteoliposome . For example , a spot that bleaches in two steps could represent a dimer or two independent monomers trapped in the same vesicle . Our approach ignores this ambiguity as it measures the probability that two fluorescent subunits were captured in the same vesicle , reporting the proximity of subunits at the point of extrusion of the large membranes . Therefore , the liposome extrusion step captures the monomer-dimer equilibrium in the prior MLV membrane state and ignores any changes in protein density or lipid composition that might arise during the extrusion process . 10 . 7554/eLife . 17438 . 003Figure 1 . Quantifying ClC-ec1-Cy5 monomers vs . dimers in lipid bilayers by subunit capture into liposomes and single-molecule photo-bleaching analysis . ( A ) Cartoon depicting the equilibrium dimerization reaction of ClC-ec1 in lipid bilayers . Kχ is the mole fraction ( χ subunit/lipid ) equilibrium constant . ( B ) Scaled cartoon of a 75 nm × 150 nm area of lipid bilayer , depicting the population of ClC-ec1 distributed as monomers ( grey ) and dimers ( black ) . ClC-ec1 monomers are ~5 nm across . To allow for subunit exchange at low densities , samples are equilibrated in a large membranes obtained by repeated freeze/thaw cycles to form large multilamellar vesicles ( MLVs ) . Red circles represent Cy5 fluorophores conjugated to subunits with PCy5 ~ 70% labeling yield . ( C ) To quantify the monomer vs . dimer populations in the MLV state , membranes are fractionated ( dashed lines in ( B ) ) by extrusion which captures subunits into liposomes . The statistics of subunit capture into liposomes follows a Poisson distribution that depends on the overall density , liposome size distribution and population stoichiometry . ( D ) Subunit occupancy in liposomes is determined by examining protein-occupied liposomes on a single-molecule TIRF microscope and carrying out photobleaching analysis . ( E ) Image of Cy5-labeled ClC-ec1 in 2:1 POPE/POPG liposomes . Numbers indicate the observed photobleaching steps for each fluorescent spot ( 1-white , 2-red , ≧ 3-green ) . ( F ) Photobleaching probability distribution for a ClC-ec1 sample reconstituted at χ = 7 . 5 × 10–7 subunits/lipid ( 0 . 1 μg/mg ClC-ec1/lipid ) . P1 , P2 and P3+ indicate probabilities of observing single , double and ≧ 3 step photobleaching steps , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 17438 . 00310 . 7554/eLife . 17438 . 004Figure 1—source data 1 . Excel file including data and statistical analysis presented in Figure 1 and Figure 1—figure supplements 1–3 including Ellman’s cysteine reactivity data , Cy5 labeling yields , Cl− transport rates , functional F0 , Cl , subunit/lipid mole fraction quantification , Fraction of protein co-localized with liposomes and F0 , the fraction of unoccupied liposomes measured from co-localization microscopy . DOI: http://dx . doi . org/10 . 7554/eLife . 17438 . 00410 . 7554/eLife . 17438 . 005Figure 1—figure supplement 1 . Design of ClC-ec1 constructs for the study of reversible dimerization in membranes by fluorescence methods . ( A ) Dimerization interface of ClC-ec1 showing the four interfacial helices ( red ) lined by non-polar side chains ( yellow ) . Positions of tryptophan substitutions that shift the protein to the monomeric state , I201W and I422W , are indicated ( arrows ) . ( B ) ClC-ec1 has three endogenous cysteines: the partially exposed C85 and buried C302 & C347 . C85A/H234C introduces a single solvent accessible cysteine for rapid maleimide conjugation . ( C ) Size exclusion chromatography of C85A/H234C ( WT ) , C85A/H234C/I201W/I422W ( WW ) and C85A/H234C/I422W ( W ) in 5 mM DM ( black - raw chromatogram , grey - dimer , red - monomer ) . ( D ) W dissociation into monomers in 5 mM DM over 24 hr at room temperature . The ‘re-run’ sample is the collection of both monomer and dimer elution fractions in ( C ) , diluted to 24 μM and re-loaded onto the size exclusion column . ( E ) Cysteine accessibility assay showing the increase in A412 reporting on TNB2- production upon reaction of Ellman’s reagent ( DNTB ) with free thiols on C85A/H234C in DM micelles . Addition of 0 . 2% SDS denatures the subunit to expose previously buried , reactive cysteines . ( F ) Molar ratio of TNB2- to ClC-ec1 subunits showing one exposed and two buried cysteines in C85A/H234C . Data represented as mean ± SE , n = 2–3 . ( G ) PCy5 labeling yields for ClC-ec1 constructs , and Pnon-specific = PCy5 for constructs lacking H234C . Data represented as mean ± SE ( n = 3–4 ) . There is no significant difference in labeling yields for the three constructs ( p > 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17438 . 00510 . 7554/eLife . 17438 . 006Figure 1—figure supplement 2 . Stability and function of Cy5 labeled ClC-ec1 . ( A ) Size exclusion chromatography profiles of WT , WT-Cy5 , W and W-Cy5 in 5 mM DM ( black – raw chromatogram , grey - dimer , red - monomer ) . ( B ) Normalized chloride efflux from 2:1 POPE/POPG liposomes ( 0 . 4 μm extruded vesicles after freeze/thaw ) for empty vesicles and WT , WT-Cy5 , W or W-Cy5 reconstituted at 1 μg/mg density ( χ = 7 . 5 × 10–6 subunit/lipid ) . Black triangle – initiation of efflux by valinomycin/FCCP , white triangle – release of trapped chloride from unoccupied vesicles by β-OG . ( C ) Normalized chloride efflux rates and ( D ) fraction of chloride trapped in unoccupied liposomes ( FCl , 0 ) , mean ± SE , n = 4–8 . All rates are significantly different from each other ( p < 0 . 05 ) , while there is no significant difference in FCl , 0 with and without Cy5 ( p > 0 . 05 ) . ( E ) Measured mole fraction values of WT-Cy5 samples reconstituted at χ = 7 . 5 × 10–5 subunits/lipid after freeze/thaw and extrusion , mean ± SE , n = 3 ( no significant difference , p > 0 . 05 ) . ( F ) Measured χ vs . reconstituted χ for all samples ( WT-Cy5 , W-Cy5 & WW-Cy5 ) mean ± SE ( n = 2–7 ) . Solid line represents the linear fit through ( 0 , 0 ) with slope = 0 . 50 ± 0 . 02 ( best-fit ± SE ) . The dotted line represents complete correlation . DOI: http://dx . doi . org/10 . 7554/eLife . 17438 . 00610 . 7554/eLife . 17438 . 007Figure 1—figure supplement 3 . Co-localization of ClC-ec1-Cy5 and AF488 labeled 2:1 POPE/POPG liposomes measured by single-molecule TIRF microscopy . ( A ) Images of AF488-labeled liposomes ( left ) , Cy5-labeled ClC-ec1 ( middle ) and the merged image using the overlay function in ImageJ ( right , liposome only spots in blue , protein only spots in red , co-localized spots in magenta ) . ( B ) Fraction of protein spots that co-localize with liposomes as a function of protein density . Data represented as fraction ± SD ( n = 1–2 samples , error bars smaller than points ) . ( C ) Fraction of unoccupied liposomes ( F0 ) as a function of the protein density for WT ( black ) and WW ( green ) . Data represents fraction ± SD ( n = 1–3 samples , error bars are smaller than points ) . Dotted lines represent a simulation of liposome occupancy ( Figure 2—figure supplement 1 ) for the monomer accessible liposome population ( green ) and the dimer accessible liposome population ( black ) . ( D ) Representative raw integrated intensity traces of ClC-ec1-Cy5 spots as a function of time ( acquired at ~1 fps ) demonstrating stepwise photo-bleaching behavior . DOI: http://dx . doi . org/10 . 7554/eLife . 17438 . 007 Using two-color TIRF microscopy on ClC-ec1-Cy5 reconstituted into fluorescent liposomes ( Figure 1—figure supplement 3 ) , we find that nearly all Cy5 spots co-localize with liposomes , demonstrating the fidelity of reconstitution . We then counted photobleaching steps in all the imaged Cy5 spots . While the high signal and low background allows for counting of up to nine discrete steps , we found that single- , double- and ≧ three-step photobleaching probabilities , P1 , P2 and P3+ are sufficient to describe the monomer vs . dimer population in the large MLVs: ( 1 ) P1=# of spots photobleaching in 1 steptotal # of spots ( 2 ) P2=# of spots photobleaching in 2 stepstotal # of spots ( 3 ) P3+=# of spots photobleaching in ≥3 stepstotal # of spots Under ideal experimental conditions , i . e . dilute conditions and 100% fluorescent labeling yield , a single or a double photobleaching step corresponds to a monomer or dimer respectively . In reality , labeling is imperfect , so and P2 include additional states such as singly labeled dimers or two monomers , respectively . In addition , to quantify the dimerization reaction , one must examine the liposome occupancy as a function of increasing protein density in the membrane , a condition that increases the chance of randomly trapping independent subunits in the same liposome , whether associated as dimers or not . The probability for fluorescent subunit capture follows an apparent Poisson distribution that depends on: ( i ) the size distribution of the liposomes , ( ii ) the mole fraction subunit density in the lipid bilayer , ( iii ) Cy5 labeling yields and ( iv ) the monomer-dimer equilibrium in the membrane at the time of extrusion . The first three factors must be known and corrected for in order to properly extract information about the monomer-dimer reaction across a wide range of mole fraction densities . To calculate this correction , we used the cryo-EM liposome size distribution reported by Walden et al . ( Figure 2—figure supplement 1E ) ( Walden et al . , 2007 ) and spectrophotometrically determined Cy5 labeling yields ( Figure 2—figure supplement 1A ) to simulate the capture of non-interacting monomer into the extruded liposome population , i . e . the ideal monomer photobleaching probability distribution PMn ( Figure 2A ) , where n refers to the number of observed photobleaching steps . Measurement of the subunit/lipid mole fraction after freeze/thaw and extrusion show that the experimental mole fraction is 50% of the original reconstituted ( Figure 1—figure supplement 2E , F ) . From here on , χ refers to the observed mole fraction , and it is this value that is considered in all of the simulations . At dilute conditions , i . e . densities less than χ = 2 × 10–6 subunit/lipid ( 0 . 2 μg/mg reconstitution density , see Table 1 ) , PM1 is constant and close to one , reflecting single subunit occupancy in the liposomes . The simulation also predicts a non-zero PM2 that arises from the small amount of non-specific labeling at sites other than the cysteine , resulting in the occasional double-labeled subunit . For χ > 2 × 10–6 to 3 . 8 × 10–4 , PM1 decays to 0 as PM3+ increases to 1 , reflecting the increase in random co-encapsulation of monomers into the same liposome . This has been described before as 'artifactual togetherness' ( Tanford and Reynolds , 1976; Fleming et al . , 1997; Stanley and Fleming , 2005 ) and must be corrected for to extract the true dimerization reaction . Next , we simulated a population of non-interacting dimers to obtain the ideal dimer photobleaching probability distribution PDn ( Figure 2B ) . PD1 and PD2 are comparable across the entire protein density range , arising from the fact that our experimental labeling yield is ~70% , leading to an equal probability of dimers labeled with one Cy5 vs . two Cy5 in the single-molecule range ( Figure 2—figure supplement 1D ) . For χ > 7 . 5 × 10–6 , PD1 and PD2 decrease as the protein density increases reflecting the growing probability of liposomes with more than two dimers , described by PD3+ . To assess the dynamic range for this approach , we performed a chi-squared analysis of the monomer vs . dimer probability distributions at each mole fraction value . The distributions were statistically different ( p ≦ 0 . 0001 ) for all mole fraction values except for the highest measured density , χ = 3 . 8 × 10−4 or 50 μg/mg ( p > 0 . 05 ) . Therefore , the photobleaching probability distributions can distinguish between the monomer vs . dimer state across 5 orders of magnitude: χ = 7 . 5 × 10–10 to 7 . 5 × 10–5 . 10 . 7554/eLife . 17438 . 008Figure 2 . Calculation of the ideal monomer and dimer photobleaching probabilities in 0 . 4 μm extruded 2:1 POPE/POPG liposomes . ( A ) PM1 , 2 , 3+ calculated for χ = 7 . 5 × 10–10 to 3 . 8 × 10–4 subunits/lipid for an ideal , non-interacting monomer . The simulation uses the liposome radius probability distribution from Walden et al . ( Walden et al . , 2007 ) , and experimental fluorescent labeling yields PCy5 = 0 . 72 and Pnon-specific = 0 . 14 ( Figure 2—figure supplement 1 ) . The appearance of noise in the simulated curves arises from the stochastic nature of the simulation . ( B ) The ideal , non-interacting dimer photobleaching probabilities PD1 , 2 , 3+ simulated with the additional constraint that dimers ( ~10 nm ) are excluded from liposomes r < 25 nm . The excluded radius was estimated from fitting the calculated fraction of unoccupied liposomes , F0 , to the experimental data from co-localization imaging ( Figure 1—figure supplement 3 ) . Chi-squared analysis shows that the monomer and dimer distributions are statistically significant for all χ values ( p < 0 . 0001 ) except for 3 . 8 × 10–4subunits/lipid ( p = 0 . 24 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17438 . 00810 . 7554/eLife . 17438 . 009Figure 2—source data 1 . Excel file including data and statistical analysis presented in Figure 2 and Figure 2—figure supplement 1 including the ideal monomer and ideal dimer photobleaching distributions ( walden liposomes , PCy5 = 0 . 72 , Pns = 0 . 14 , bias = 4 ) distribution , chi-squared analysis , pooled protein labeling data . DOI: http://dx . doi . org/10 . 7554/eLife . 17438 . 00910 . 7554/eLife . 17438 . 010Figure 2—figure supplement 1 . Probability distributions used to calculate PMn and PDn . ( A ) Pooled overall labeling yield PCy5 = 0 . 72 ± 0 . 08 and non-specific labeling yieldPns = 0 . 14 ± 0 . 04 , mean ± SE ( n = 10–11 ) . ( B ) Probabilities of labeling in the two-site subunit model: H234C vs . a non-specific site . PH234C = PCy5– Pns = 0 . 58 . ( C ) Probability that a single subunit is labeled with NCy5 = 0 , 1 or 2 . ( D ) Probability that a dimer is labeled with NCy5 = 0 , 1 , 2 , 3 or 4 . ( E ) The cryo-EM liposome radius probability distribution from Walden et al . for E . coli polar lipid membranes extruded through a 0 . 4-µm filter after freeze/thaw . ( F ) Surface area probability distribution Psurface-area for the monomer accessible liposome population ( i . e . all liposomes ) . ( G ) Surface area distribution for the dimer accessible liposome population , where liposomes with radius less than 25 nm are excluded based on F0 co-localization data from Figure 1—figure supplement 3C . DOI: http://dx . doi . org/10 . 7554/eLife . 17438 . 01010 . 7554/eLife . 17438 . 011Table 1 . Lookup table for converting between membrane density units . DOI: http://dx . doi . org/10 . 7554/eLife . 17438 . 011ρ ( μg/mg ) χReconstitution ( subunits/ lipid ) χObserved ( subunits/ lipid ) χ* ( subunits/lipid ) ( χ* ) -1 ( lipids/ subunit ) ρ*area ( subunits/ nm2 bilayer ) ( ρ*area ) -1 ( nm2 bilayer/ subunit ) Box ( nm × nm ) 0 . 00011 . 5 × 10-97 . 5 × 10-103 . 8 × 10-102 , 657 , 101 , 1031 . 3 × 10-9797 , 130 , 33128 , 2330 . 00057 . 5 × 10-93 . 8 × 10-91 . 9 × 10-9531 , 420 , 2216 . 3 × 10-9159 , 426 , 06612 , 6260 . 0011 . 5 × 10-87 . 5 × 10-93 . 8 × 10-9265 , 710 , 1101 . 3 × 10-879 , 713 , 03389280 . 0057 . 5 × 10-83 . 8 × 10-81 . 9 × 10-853 , 142 , 0226 . 3 × 10-815 , 942 , 60739930 . 011 . 5 × 10-77 . 5 × 10-83 . 8 × 10-826 , 571 , 0111 . 3 × 10-77 , 971 , 30328230 . 057 . 5 × 10-73 . 8 × 10-71 . 9 × 10-75 , 314 , 2026 . 3 × 10-71 , 594 , 26112630 . 11 . 5 × 10-67 . 5 × 10-73 . 8 × 10-72 , 657 , 1011 . 3 × 10-6797 , 1308930 . 23 . 0 × 10-62 . 0 × 10-68 . 0 × 10-71 , 328 , 5512 . 5 × 10-6398 , 5656310 . 57 . 5 × 10-63 . 8 × 10-61 . 9 × 10-6531 , 4206 . 3 × 10-6159 , 42639911 . 5 × 10-57 . 5 × 10-63 . 8 × 10-6265 , 7101 . 3 × 10-579 , 71328257 . 5 × 10-53 . 8 × 10-51 . 9 × 10-553 , 1426 . 3 × 10-515 , 943126101 . 5 × 10-47 . 5 × 10-53 . 8 × 10-526 , 5711 . 3 × 10-4797189507 . 5 × 10-43 . 8 × 10-41 . 9 × 10-45 , 3146 . 3 × 10-4159440ρ is the reconstituted mass density of μg of ClC-ec1 subunits per mg of 2:1 POPE/POPG lipids . χReconstitution is the reconstituted mole fraction of ClC-ec1 subunits per lipid . χObserved = χReconstitution * 0 . 50 , determined from protein to lipid quantification assays . χ* is the reactive mole fraction calculated as χObserved/2 , assuming that the reaction occurs between oriented subunits in the membrane . ρ*area is the reactive mole density , subunits per bilayer area , using SAlipid = 0 . 6 nm2 . Box – square root of ( ρ*area ) -1Bolded values indicate the observed dynamic range of the photobleaching approach . The single-molecule photobleaching method allows us to explore extremely dilute densities within the membrane with no loss of signal . At lower densities , the number of fluorescent spots in a field of view decreases , which can easily be compensated for by increasing the number of imaged fields to maintain similar counting statistics . We used this approach to investigate whether experimentally measured P1 , and P3+ reflect the equilibrium population of ClC-ec1 monomers and dimers in the membrane . If the system is in dynamic equilibrium , then the fraction of dimers out of all subunits will depend on the mole fraction density , χ ( subunit/lipid ) , but not the path followed to reach this density . We tested this by comparing the photobleaching data obtained using two distinct methods of setting the final mole fraction , one that starts with the protein in monomeric form , and another that starts with the protein as a dimer . In the first method , monomeric W-Cy5 was reconstituted at χ = 7 . 5 × 10–8 , 7 . 5 × 10–7 and 7 . 5 × 10–6 subunit/lipid by dialysis ( Figure 3A ) . Size exclusion chromatography of W in n-Decyl-β-D-Maltopyranoside ( DM ) micelles shows a mixture of monomers and dimers upon purification ( Robertson et al . , 2010 ) ; however , the protein rapidly dissociates to the monomeric form as revealed by an immediate re-run of the eluted protein ( Figure 1—figure supplement 1D ) . Photobleaching analysis on extruded liposomes derived from these membranes ( Figure 3C ) shows that at the lowest density , the probability distribution resembles the ideal monomer distribution ( Figure 2A ) and as χ increases , the distributions approach the ideal dimer distribution ( Figure 2B ) with P1 decreasing as P2 and P3+ increase . 10 . 7554/eLife . 17438 . 012Figure 3 . ClC-ec1-Cy5 photobleaching probabilities depend on χ and is path independent . The mole fraction density can be set by two different methods: ( A ) Reconstitution of ClC-ec1-Cy5 by mixing detergent solubilized subunits with lipids , followed by dialysis to remove detergent resulting in lipid bilayer formation ( black arrow ) . In this case , bilayers are fused together by freeze/thaw ( red arrow ) and incubated at room temperature prior to extrusion and imaging . ( B ) Dilution of high-density proteoliposomes by freeze/thaw fusion with empty vesicles , followed by incubation at room temperature prior to extrusion and imaging . ( C ) Photobleaching probabilities for W-Cy5 reconstituted at χ = 7 . 5 × 10–8 , 7 . 5 × 10–7and 7 . 5 × 10–6 subunit/lipid . All distributions are statistically significant by chi-squared analysis ( p < 0 . 0001 ) . Data are represented as mean ± SE , n = 3 and 2 counters . ( D ) Photobleaching probabilities for samples diluted 100X or 10X from samples reconstituted at χ = 7 . 5 × 10–6 subunit/lipid and imaged one day after freeze/thaw fusion . Data are represented as mean ± SE , n = 3 and 2–3 counters . Chi-squared analysis shows no significant difference between χ = 7 . 5 × 10–7 reconstituted vs . 10X diluted samples , or χ = 7 . 5 × 10–8 reconstituted vs . 100X diluted samples ( p > 0 . 05 ) . ( E ) Post-freeze/thaw time course of PW1-3+ probabilities for the 100X diluted sample as a function of incubation time at room temperature . 'R' at χ = 7 . 5 × 10–6 subunit/lipid represents the original high-density reconstituted sample prior to dilution ( mean ± SE , n = 3 and 2 counters , collected at t = 15 , 28 and 72 days ) . The freeze/thaw process is indicated by the yellow bar . Time course data represent fraction ± SE ( n = 3 samples and 2–3 counters , points without error bars represent calculated fraction from a single counter ) . 'R' at χ = 7 . 5 × 10–8 subunit/lipid shows the probabilities for samples reconstituted directly at the corresponding dilution ( mean ± SE , n = 2 samples and 2 counters , collected at t = 23 and 71 days ) . ( F ) Photobleaching probabilities for WT-Cy5 reconstituted at χ = 7 . 5 × 10–9 , 7 . 5 × 10–8 , 7 . 5 × 10–7 and 7 . 5 × 10–6 subunit/lipid or ( D ) diluted 1000X , 100X or 10X from samples reconstituted at χ = 7 . 5 × 10–6 subunit/lipid , imaged 1 day after freeze/thaw fusion . Data are represented as mean ± SE , n = 3–4 samples and 2 counters . Chi-squared analysis shows no significant difference between χ = 7 . 5 × 10–7 reconstituted vs . 10X diluted samples , or χ = 7 . 5 × 10–8 reconstituted vs . 100X diluted samples ( p > 0 . 05 ) ; however , the 1000X diluted sample is significantly different from the χ = 7 . 5 × 10–9 reconstituted sample ( p < 0 . 001 ) . ( E ) Post-freeze/thaw time course of PWT1-3+ probabilities for 1000X dilution as a function of incubation time at room temperature . 'R' at χ = 7 . 5 × 10–6 subunit/lipid designates the high-density sample prior to dilution ( mean ± SE , n = 3 and 2 counters , collected at t = 10 , 15 and 95 days ) . Time course data represent mean ± SE ( n = 3 samples and 2–3 counters ) . 'R' at χ = 7 . 5 × 10–9 subunit/lipid represent samples reconstituted directly at the corresponding density , mean ± SE , n = 2 samples and 2 counters , collected at t = 1 and 89 days post-freeze/thaw . DOI: http://dx . doi . org/10 . 7554/eLife . 17438 . 01210 . 7554/eLife . 17438 . 013Figure 3—source data 1 . Excel file including data and statistical analysis presented in Figure 3 and Figure 3—figure supplement 1 including WT and W dilution data , chi-squared analysis between reconstituted and diluted samples , and FRET data . DOI: http://dx . doi . org/10 . 7554/eLife . 17438 . 01310 . 7554/eLife . 17438 . 014Figure 3—figure supplement 1 . FRET measurements of WT-Cy3/Cy5 and W-Cy3 + W-Cy5 in lipid bilayers at χ = 7 . 5 × 10–6 subunit/lipid . ( A ) FRET emission spectrum ( λEX = 535 nm ) of co-labeled WT-Cy3/Cy5 ( PCy5/PCy3 = 1 . 2 ) in freeze/thawed 2:1 POPE/POPG MLVs . The raw spectrum ( black ) is decomposed into Cy3 donor emission ( green ) and FRET-specific Cy5 emission ( red ) . ( B ) Addition of 0 . 2% SDS dissociates the dimer resulting in complete loss of the FRET signal . ( C ) Emission spectrum of separately labeled W-Cy3 + W-Cy5 mixed prior to reconstitution then freeze/thawed to form MLVs ( PCy5/PCy3 = 1 . 4 ) . ( D ) Saturation of FRET as a function of increasing acceptor/donor ratios fit to theoretical curves for dimer ( 2 - green ) , trimer ( 3 – orange ) and tetramer ( 4 – blue ) oligomeric states . WT-Cy3/Cy5 ( white circles ) and W-Cy3 + W-Cy5 ( red circles ) data are shown as symbols . DOI: http://dx . doi . org/10 . 7554/eLife . 17438 . 014 In the second method , we started with W-Cy5 already reconstituted in lipid bilayers at high density , χ = 7 . 5 × 10–6 subunit/lipid ( Figure 3B ) , a condition where W exists almost entirely as dimers . This state was confirmed by macroscopic FRET measurements in MLVs showing that W-Cy3 + W-Cy5 and WT-Cy3/Cy5 yield the same dimer FRET signal ( Figure 3—figure supplement 1 ) . The high-density W-Cy5 membranes were then diluted 100X and 10X by freeze/thaw-mediated fusion with empty vesicles . At 1 day after freeze/thaw , P1-3+ ( Figure 3D ) shows no significant difference with the probability distributions obtained with W-Cy5 samples directly reconstituted at the corresponding χ , 7 . 5 × 10–8 and 7 . 5 × 10–7 subunit/lipid , respectively ( Figure 3C ) . Measurement of the 100X diluted photobleaching probabilities as a function of time after freeze/thaw shows no change up to 27 days at room temperature ( Figure 3E ) . The agreement between the two methods demonstrates that the reaction is reversible and that equilibrium subunit exchange of W-Cy5 occurs during the freeze/thaw fusion step . Since the equilibration kinetics may be different for different constructs , we conducted the same experiment for WT-Cy5 . So far , WT has only been observed to exist as dimers in lipid bilayers , but all experiments conducted have examined the protein at relatively high mole fraction densities χ ≧ 7 . 5 × 10–7 subunit/lipid ( i . e . ρ ≧ 0 . 1 μg/mg ) . We reconstituted WT-Cy5 at χ = 7 . 5 × 10–9 to 7 . 5 × 10–6 subunit/lipid , up to 100-fold lower than the lowest density studied so far . At the lowest density , there is a significant P1 probability ( Figure 3F ) indicating that WT exists as monomers in the lipid bilayer at low dilutions . Similar to the W-Cy5 data , as χ is increased , P1 decreases while P2 and P3+ increase indicating a conversion from monomers to dimers . Next , high-density samples of WT-Cy5 reconstituted at χ = 7 . 5 × 10–6 were diluted 1000X , 100X and 10X with empty vesicles by freeze/thaw-mediated fusion ( Figure 3G ) . The diluted probability distributions show no significant difference compared to the reconstituted distributions except for the 1000X diluted sample corresponding to χ = 7 . 5 × 10–9 subunit/lipid . Since these data were collected one day after the freeze/thaw process , we continued to image the 1000X diluted WT-Cy5 samples for up to 25 days after freeze/thaw , incubating the samples at room temperature . The photobleaching probabilities of the diluted sample slowly converged to the reconstituted χ = 7 . 5 × 10–9 subunit/lipid probabilities with a mid-point of 13 days ( Figure 3H ) . This demonstrates that the reconstituted WT samples reflect the protein population in a dynamic equilibrium , albeit with higher kinetic stability compared to W , a finding that has been observed with other membrane proteins such as Diacylglycerol kinase ( Jefferson et al . , 2013 ) . Next , we measured P1-3+ for the three ClC-ec1 constructs reconstituted across a wide range of mole fraction densities: χ = 7 . 5 × 10–10 to 3 . 8 × 10–4 subunit/lipid . The experimental photobleaching distribution for WW-Cy5 , PWW1-3+ ( Figure 4A ) resembles the ideal monomer distribution PM1-3+ ( Figure 2A ) . For W-Cy5 ( Figure 4B ) and WT-Cy5 ( Figure 4C ) , the experimental photobleaching distributions PW1-3+ and PWT1-3+ exhibit three phases: ( i ) at low densities they mimic the monomer distribution , ( ii ) as the density increases , there is a gradual approach to the ideal dimer distribution , and finally ( iii ) at higher densities , P1 and P2 decrease , while P3+ increases indicating the capture of multiple subunits in each liposome . WT and W both demonstrate a monomer to dimer transition; however , the reaction of W is shifted along the χ subunit/lipid axis , indicating the weaker stability of the W dimer . 10 . 7554/eLife . 17438 . 015Figure 4 . Photobleaching probabilities show a monomer to dimer transition that depends on the number of tryptophan residues at the dimerization interface . Left , single subunit of ( A ) WW ClC-ec1 , ( B ) W and ( C ) WT in the lipid bilayer ( beige rectangle , dotted lines ) rotated to show the four helices that form the dimerization interface ( red ) and non-polar residues that line this surface ( yellow , licorice ) . Tryptophan substitutions are shown in yellow VDW representation . Right , experimental Pexpt1-3+ at mole fraction densities χ = 7 . 5 × 10–10 to 3 . 8 × 10–4 subunit/lipid for ( A ) WT-Cy5 , ( B ) W-Cy5 and ( C ) WW-Cy5 . Data are reported as mean ± SE ( n = 2–3 samples and 2–3 counters ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17438 . 01510 . 7554/eLife . 17438 . 016Figure 4—source data 1 . Excel file including data and statistical analysis presented in Figure 4 including raw and averaged data for WT , W and WW photobleaching distributions . DOI: http://dx . doi . org/10 . 7554/eLife . 17438 . 01610 . 7554/eLife . 17438 . 017Figure 4—figure supplement 1 . Robustness of the photobleaching probability distribution . Probabilities are shown for Nstep = 1 ( black ) , 2 ( red ) and >= 3 ( green ) photobleaching steps . ( A ) W-Cy5 sample ( PCy5 = 0 . 66 ) prepared in AF488-labeled lipids . ( B ) Photobleaching distributions from a sample obtained in a different purification of W-Cy5 ( PCy5 = 0 . 73 ) prepared in un-labeled lipids . ( C ) Same data set as in ( B ) counted by a different , blinded individual . All data are represented as fraction ± binomial SD . Chi-squared analysis shows no significant difference ( p > 0 . 05 ) in the photobleaching distributions obtained from samples prepared at the same mole fraction , except for ( C ) χ = 7 . 5 × 10–5 subunits/lipid ( p = 0 . 04 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17438 . 017 Using the data in Figure 4 , we calculated the fraction of dimer ( FDimer ) across the range of experimental χ subunit/lipid densities . To take into account that the protein inserts into membranes in two orientations , but only similarly oriented protein participates in the dimerization reaction , we use the reactive mole fraction scale , χ∗=χ/2 . For each value of χ∗ , we carry out a least-squares fit of the residual sum of squares ( R2 ) of the experimental Pn probabilities to the FDimer weighted linear combination of the ideal monomer PMn and ideal dimer PDnprobability distributions from Figure 2 ( Figure 5—figure supplement 1 ) . The Fdimer vs . χ* data was fit to an equilibrium dimerization isotherm ( Figure 5 ) to determine the mole fraction equilibrium constants and the mole fraction standard state free energies , using a standard state density of χ° = 1 subunit/lipid . The data for the single tryptophan mutant , W , shows a complete reaction from monomer to dimer , yielding KχW = 3 . 7 ± 1 . 6 × 106 ( best-fit ± standard error ( SE ) ) and ΔG°W = −9 . 0 ± 0 . 3 kcal/mole in 2:1 POPE/POPG lipid bilayers at room temperature . Note that as the density increases beyond χ∗ = 1 . 9 × 10–6 , R2 increases indicating that the experimental probability distributions deviate from PDn ( Figure 5—figure supplement 2 ) , due to the observation of more liposomes bleaching in 3+ steps than expected from theory ( Figure 5—figure supplement 1E ) . It is expected that the theoretical calculations at high densities will be sensitive to inaccuracies in the liposome size distribution , especially if larger liposomes or multi-lamellarity have been omitted from the distribution . Another possibility is that there is a small amount of non-specific oligomerization at high mole fractions . To account for these deviations , the fits are weighted by 1/R2 , allowing us to define an observed dynamic range of = 7 . 5 × 10–10 to 3 . 8 × 10–6 subunit/lipid , or 0 . 0001 to 0 . 5 μg/mg ClC-ec1/lipid . While WW only shows the initial part of the reaction , this rise occurs within this dynamic range and is thus likely reflecting the onset of dimerization . Fitting the data to a dimerization isotherm yields KχWW = 2 . 1 ± 0 . 8 × 105 lipids/subunit and ΔG°WW = −7 . 3 ± 0 . 2 kcal/mole . Fitting of WT data yields leads to KχWT = 2 . 1 ± 0 . 5 × 108 and ΔG°WT = −11 . 4 ± 0 . 1 kcal/mole , although the baseline must be constrained ( Y0 = 0 . 07 ) since it is not possible to go to lower dilutions where the all-monomer state is expected to be observed in the membrane . With this , the results demonstrate that substitution at I422W ( W ) destabilizes ClC-ec1 dimerization by ΔΔGW-WT = 2 . 4 ± 0 . 3 kcal/mole , while the additional tryptophan at I201W ( WW ) destabilizes the dimer by an additional ΔΔGWW-W = 1 . 7 ± 0 . 4 kcal/mole , and ΔΔGWW-WT = 4 . 1 ± 0 . 2 kcal/mole overall . 10 . 7554/eLife . 17438 . 018Figure 5 . FDimer vs . the reactive mole fraction χ* . FDimer is estimated by least-squares fitting of the experimental Pexpt1-5+ photobleaching probabilities to ( 1-FDimer ) *PM1-5+ + FDimer*PD1-5+ , where PM1-5+ and PD1-5+ are the calculated ideal monomer and dimer distributions . The reactive mole fraction , χ* , is equal to half of the experimental mole fraction ( χ/2 ) , assuming that the reaction only occurs between similarly oriented protein in the membrane . Data is shown for WT-Cy5 ( black ) , W-Cy5 ( red ) and WW-Cy5 ( green ) , with symbols representing mean ± SE ( n = 2–3 samples and 2–3 counters ) . Dotted lines represent best-fits to the equilibrium dimerization isotherm , weighted by the inverse of the minimum residual sum of squares ( R2 ) calculated in the estimation of FDimer ( Figure 5—figure supplement 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17438 . 01810 . 7554/eLife . 17438 . 019Figure 5—source data 1 . Excel file including data and statistical analysis presented in Figure 5 including raw and averaged FDimer data , minimum R2 values , standard error of the estimate ( SEE ) and 1/R2 weights on the fit . Results from fitting to the dimerization isotherm are presented in Table 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 17438 . 01910 . 7554/eLife . 17438 . 020Figure 5—figure supplement 1 . Least-squares estimation of FDimer . ( A ) Pexpt1-5+ photobleaching distributions for W-Cy5 reconstituted at χ = 7 . 5 × 10–8 , ( C ) 7 . 5 × 10–7 and ( E ) 7 . 5 × 10–6 subunits/lipid . Data are represented as fraction ± binomial SD . The least-squares prediction for FDimer is plotted to the right of the experimental distribution , along with the PM , and PD at the given mole fraction values . ( B , D , F ) The corresponding residual sum of squares function ( R2 ) as a function of FDimer . DOI: http://dx . doi . org/10 . 7554/eLife . 17438 . 02010 . 7554/eLife . 17438 . 021Figure 5—figure supplement 2 . Sum of squared residuals ( R2 ) as a function of the mole fraction density . Left , R2 and right , FDimer estimates as a function of the reactive mole fraction χ* . ( A , B ) WT-Cy5 , ( C , D ) W-Cy5 , ( E , F ) WW-Cy5 . Solid lines represent best-fits to the equilibrium isotherm weighted by 1/R2 . The dotted line in ( B ) is the fit while constraining the baseline ( Y0 ) to 0 . 065 , as is automatically obtained for W-Cy5 and WW-Cy5 data . Numbers in ( B ) represent the post freeze/thaw incubation time for that sample . DOI: http://dx . doi . org/10 . 7554/eLife . 17438 . 021 With these results , we can compare and contrast ClC-ec1 with other dimerization models , and take a step toward a generalized understanding of membrane protein stability in the lipid bilayer solvent . In 2:1 POPE/POPG , we find that ClC-ec1 is a high-affinity dimer with KχWT ClC-ec1 ~108 lipids/subunit , second in stability to GpA in POPC , which has a reported equilibrium constant of KχGpA ~109 ( Hong et al . , 2010 ) . The strength of GpA dimerization is remarkable considering its small 200 Å2 interface , but it has been shown that this stability involves the GxxxG helix-packing motif ( Lemmon et al . , 1992 ) , which allows for backbone flexibility that maximizes VDW packing and hydrogen bonding inside the membrane ( Smith et al . , 2002 ) . ClC-ec1 lacks this specialized motif , so why then is the dimer so stable in lipid bilayers ? To compare between these two very different proteins , we normalize the free energies by the total buried surface area , to obtain a binding efficiency per Å2 of the dimerization interface ( Day et al . , 2012 ) . In this manner , GpA in POPC is highly efficient , contributing -30 cal/mole per Å2 , whereas wild-type ClC-ec1 only exhibits -5 cal/mole per Å2 stability . When GpA dimerization was measured in E . coli polar lipid membranes , similar to our 2:1 POPE/POPG bilayers , the equilibrium constant shifted to ~5 × 105 lipids/subunit due to electrostatic destabilization by negatively charged lipids ( Hong and Bowie , 2011 ) . Even though GpA is slightly weaker than ClC-ec1 in these lipids , its efficiency is still higher , with -20 cal/mole per Å2 . Even though ClC-ec1 and GpA reach similar stabilities in the membrane , the dramatic differences in the dimerization efficiencies suggest they do so by different physical mechanisms . In contrast , dimerization efficiencies of transmembrane helices that do not contain GxxxG motifs have values comparable to ClC-ec1 . For example , the Serine Zipper ( North et al . , 2006 ) is -8 cal/mole per Å2 and poly-LEU-ALA ( Yano et al . , 2002 ) is -5 cal/mole per Å2 , assuming a dimerization interface of 300 Å2 . These are often referred to as models of inert helix dimerization , and it is expected that there is minimal conformational change in the helices upon association . In that respect , ClC-ec1 follows an inert surface model with relatively weak interaction efficiency , but it achieves an overall stability comparable to GpA by virtue of its large dimerization interface . The dimer state buries a remarkable 2400 Å2 of protein surface area that would otherwise be interacting with lipids . Decomposition of the free energy into protein-protein , protein-lipid and lipid-lipid terms yields ( Lemmon and Engelman , 1994; White and Wimley , 1999 ) : ( 4 ) ΔGdimerization∘= ( ΔGprotein−protein+nΔGlipid−lipid ) −2nΔGprotein−lipid which shows that the free energy is strongly dependent on the number of lipids that solvate the dimerization interface , n . Since n will be relatively large for ClC-ec1 then it is predicted that membrane dependent driving forces , such as hydrophobic mismatch ( Lee , 2004; Andersen , 2007 ) and changes in lipid entropy ( Lagüe et al . , 2001; Katira et al . , 2016 ) could play a larger role in ClC dimerization compared to smaller , single TM-helix dimerization models . The tryptophan mutagenesis acts as a starting point for quantifying the physical forces associated with protein assembly in membranes . Since the dimerization interface is highly complementary in shape ( Robertson et al . , 2010 ) , addition of a single tryptophan is expected to act as a steric wedge and disrupt many of the VDW interactions between the protein side-chains . However , this ~2 kcal/mole destabilization likely overestimates the VDW contribution since tryptophans will also stabilize the monomeric state by interacting with lipids at this surface . Therefore , this amounts to a relatively small change in dimer stability , indicating that we either maintain many of the VDW contacts in the dimer complex , or that protein-protein VDW interactions are not the major driving force for dimerization . Further investigation of tryptophans as a function of number and position , as well as other residue substitutions , will surely inform on the underlying relationship . In any case , identifying the driving forces that govern ClC-ec1 dimerization will require further experiments that quantify enthalpic and entropic changes while varying protein and lipid dependent variables . Fortunately , the structural integrity and stability of the individual subunits makes investigation of the protein and membrane at different temperatures possible . The method of single-molecule counting of subunit capture into liposomes addresses many of the challenges previously encountered when studying membrane protein association . Membrane proteins are prone to non-equilibrium aggregation when reconstituted at high densities . Thus , the single-molecule approach enables us to explore low densities where the two-state reaction is expected to dominate . This technique is particularly well suited for studying high-affinity membrane protein complexes that may only show dissociation behavior at sub-biological densities . Our lowest experimental density corresponds to 1 subunit per 50 E . coli inner membranes , assuming a 4 µm2 surface area consisting of ~107 lipids ( Prats and de Pedro , 1989 ) . Note that the lower limit of the biological mole fraction for E . Coli is χ = 2 × 10–7 subunit/lipid , assuming 2 subunits are expressed in the cell . Based on the dimerization isotherm in Figure 5 , at this mole fraction , 90% of subunits will be found in the dimer form . This means that any reasonable level of expression ( 10–100 copies per cell ) will drive the reaction into a density range where the protein will exist as associated dimers , with negligible probability of observing the dissociated monomeric state . An additional benefit of the photobleaching approach is that it provides checks on the fidelity of reconstitution , as aberrant behavior such as aggregation would skew the photobleaching distribution toward exceptionally large occupancies . Furthermore , we expect that this method will be useful in the study of other membrane protein systems , even if the structure is unknown , as long as the protein can be purified , quantitatively labeled with fluorophores , and tested for proper function ( Stockbridge et al . , 2013 ) . By combining single molecule approaches with robust membrane protein systems like ClC-ec1 , we expect that the measurement of membrane protein association reactions in membranes will no longer be considered a technical challenge , but instead lead to discoveries about fundamental physical driving forces within the lipid bilayer . All isoforms were inserted into a pASK vector containing a hexa-histidine tag at the C-terminus . Site-directed mutagenesis was carried out by QuickChange ( Agilent , Santa Clara , CA ) followed by DNA sequencing of the full gene . List of experimental constructs for site-specific labeling: C85A/H234C ( WT ) MW = 51 , 997 g/mole , ε = 46 , 020 M-1 cm-1; C85A/H234C/I422W ( W ) MW = 52 , 070 g/mole , ε = 51 , 700 M-1 cm-1; C85A/H234C/I201W/I422W ( WW ) MW = 52 , 146 g/mole , ε = 57 , 410 M-1 cm-1 . List of constructs used to calculate the non-specific labeling: C85A ( WTnon-specific ) MW = 52 , 031 g/mole , ε = 45 , 900 M-1 cm-1; C85A/I422W ( Wnon-specific ) MW = 52 , 104 g/mole , ε = 51 , 590 M-1 cm-1; and C85A/I201W/I422W ( WWnon-specific ) MW = 52 , 177 g/mole , ε = 57 , 280 M-1 cm-1 . Molecular weight and extinction coefficients calculated using the Peptide Property Calculator at http://biotools/nubic . northwestern . edu/proteincalc . html . Expression and purification of ClC-ec1 was carried out as previously described ( Maduke et al . , 1999; Robertson et al . , 2010 ) . BL21-AI E . coli competent cells ( Thermo Fisher Scientific , Waltham , MA ) were transformed with the plasmid and then 2 L Terrific Broth supplemented with ampicillin was inoculated and grown at 37°C . Protein expression was induced with anhydro-tetracycline at OD600 = 1 . 0 . After 3 hr of induction , cells were harvested , then lysed by sonication in buffer supplemented with 5 mM reducing agent TCEP ( Tris ( 2-carboxyethyl ) phosphine; Soltec Bioscience , Beverly , MA ) and pH adjusted to 7 . 5 . Protein extraction was carried out with 2% n-Decyl-β-D-Maltopyranoside ( DM; Anatrace , Maumee OH ) for 3 hr at room temperature . Cell debris was pelleted down and the supernatant was run on a 2 mL column volume ( CV ) TALON cobalt affinity resin ( Clontech Laboratories , Mountain View , CA ) equilibrated in CoWB/TCEP: 100 mM NaCl , 20 mM Tris , 1 mM TCEP , pH 7 . 5 with NaOH , 5 mM DM . After binding , the column was washed with 15 CVs of CoWB/TCEP followed by a low imidazole wash of CoWB/TCEP containing 20 mM imidazole ( Sigma-Aldrich , St . Louis , MO ) . ClC-ec1 was eluted with CoWB/TCEP containing 400 mM imidazole , then concentrated in a 30 kDa NMWL centrifugal filters ( Amicon , EMD Millipore ) to ~500 μL and injected on a Superdex 200 10/30 GL size exclusion column ( GE Healthcare , Little Chalfont , UK ) equilibrated in size exclusion buffer ( SEB ) : 150 mM NaCl , 20 mM MOPS pH 7 . 5 , 5 mM analytical-grade DM , attached to a medium pressure chromatography system ( NGC , Bio-Rad ) . Wild-type ClC-ec1 contains three endogenous cysteines: C85 , C302 and C347 ( Figure 1 – supplementary 1B ) . While these cysteines can be mutated to yield a ‘cys-less’ form of ClC-ec1 ( C85A/C302A/C347S ) that maintain transport function ( Nguitragool and Miller , 2007 ) , we found that the ‘cys-less’ substitution on I201W/I422W expresses but results in aggregated protein upon purification . Examining the structure , C85 is partially accessible to the aqueous solution while C302 and C347 are buried within the protein core . We tested whether substituting C85 with alanine alone would be sufficient to minimize background labeling for our fluorescent experiments . We made a construct C85A/H234C , which introduces an aqueous solvent exposed cysteine near the dimerization interface , for specific labeling by Cy5-maleimide ( Figure 1—figure supplement 1B ) . A cysteine accessibility assay was used to measure the reactivity of –SH groups present in ClC-ec1 ( Ellman , 1959; Riddles et al . , 1983 ) . A 10 mM master stock of Ellman’s reagent ( DNTB , 5 , 5’-Dithio-bis ( 2-nitrobenzoic acid ) ; Sigma-Aldrich ) was freshly prepared in reaction buffer ( 0 . 1 M sodium phosphate , 1 mM EDTA , pH 8 . 0 ) then diluted to 5 mM with SEB . Reaction of the thiolate anion with DTNB produces 2-nitro-5-thiobenzoate ( TNB- ) that ionizes to TNB2- and absorbs light at 412 nm . A412 was monitored by UV-VIS spectroscopy ( Nanodrop 2000c , Thermo-Fisher Scientific ) for the protein at 10 μM ( 300 μL ) for 5 min to establish a baseline . Following this , 20 μl of the Ellman’s reagent working stock was added ( [protein] = 9 . 4 μM , [DNTB] = 313 μM ) , and the reaction monitored for ten minutes . To estimate the total number of cysteines in the protein , 40 μl of 2% SDS in SEB was added to denature the protein , exposing the buried cysteines C302 and C347 ( [protein] = 8 . 3 μM , [DNTB] = 278 μM , 0 . 2% SDS ) , and the reaction was monitored for 45 min , until a steady saturation of A412 nm was reached . Absorbance at 412 nm was background subtracted by the absorbance at 750 nm to correct for baseline drift during the measurement . An extinction coefficient of 14 , 150 M-1 cm-1 was used to calculate [TNB2-] . Addition of Ellman’s reagent to C85A/H234C in 5 mM DM shows an instantaneous increase in A412nm signal ( Figure 1—figure supplement 1C ) indicating rapid and specific conjugation with H234C . Addition of 0 . 2% SDS shows that the internal cysteines are reactive in the SDS denatured state . We calculated the molar ratio of TNB2- produced per ClC-ec1 subunit in DM , and found that there is little reactivity in C85A . All constructs on the C85A/H234C background shows a single reactive thiol in 5 mM DM and a total of three reactive thiols in 5 mM DM + 0 . 2% SDS , indicating that introduction of tryptophan substitutions do no affect the fold of C85A/H234C in DM micelles ( Figure 1—figure supplement 1D ) . Cy5-maleimide dye was obtained as lyophilized powder as either 1 mg ( GE Healthcare ) or 50 mg ( Lumiprobe , Hannover , Germany ) , stored as 10 mM master stocks ( 50 μl each ) in anhydrous DMSO ( Thermo Fisher Scientific ) at -80°C . Single-use 5 μl working stocks of 10 mM strength were also prepared and stored at -80°C to avoid multiple freeze/thaw cycles of the fluorophores . Both master and working stocks were stored in boxes in the presence of anhydrous CaSO4 ( Drierite , W A Hammond Drierite Co Ltd . , Xenia , OH ) . The fluorophore conjugation reaction was carried out in SEB with 10 μM ClC-ec1 subunits and 50 μM Cy5-maleimide for 12–15 min at room temperature in dark . At the end of the reaction , 100-fold molar excess of cysteine was added to quench the maleimide reaction ( from freshly prepared 100 mM stock in SEB , pH adjusted to ~7 . 5 ) . The ‘free’ dye was separated from the labeled protein by binding the reaction mixture to a 250 μL cobalt affinity resin column equilibrated with 15 CV CoWB ( no TCEP ) in a Micro-Bio spin chromatography column ( Bio-Rad Laboratories , Hercules CA ) , washed 15 CV with CoWB and then eluted with 400 mM imidazole in CoWB , manually collecting only the fluorescently labeled protein . To remove the interfering absorbance of imidazole at 280 nm , the labeled protein was added to a 3 mL Sephadex G50 size exclusion column ( Sigma-Aldrich ) equilibrated in CoWB ( no TCEP ) . The fluorescently labeled protein was eluted after addition 2–2 . 5 mL of CoWB to the column . The concentration and labeling efficiency of protein calculated from the UV-VIS absorbance spectrum of the sample and λmax of ClC-ec1 ( 280 nm ) and Cy5 ( 655 nm ) as follows: ( 5 ) [subunit]=\ A280− ( Afluorophore\ ×CFfluorophore ) εsubunit ( 6 ) Pfluorophore=\ Afluorophore[subunit]\ ×εfluorophore where εsubunit ( M-1 cm-1 ) is the molar extinction coefficient for the ClC-ec1 isoforms , Afluorophore is the absorbance of Cy3 or Cy5 at λmax , CFfluorophore is the correction factor for fluorophore absorbance at 280 nm ( CFCy3 = 0 . 08 CFCy5=0 . 02 ) and ϵfluorophore is the molar extinction coefficients for the fluorophore ( εCy3 = 1 . 5 × 105 M-1 cm-1 at 565 nm , and εCy5 = 2 . 5 × 105 M-1 cm-1 at 665 nm ) . There was no significant difference in labeling yields of the C85A/H234C constructs ( Figure 1—figure supplement 1G ) allowing for the calculation of a pooled average labeling yield of PCy5 = 0 . 72 ± 0 . 08 ( n = 10 ) . For C85A constructs lacking the reactive H234C , we found that there was no significant difference in labeling and measured a pooled average of Pnon-specific = 0 . 14 ± 0 . 04 ( n = 11 ) , representing the non-specific labeling yield . We tested the various C85A/H234C constructs in 5 mM DM for monomer vs . dimer behavior by size exclusion chromatography ( Figure 1—figure supplement 1C ) and found similar behavior compared to the tryptophan substitutions on the wild-type background ( Robertson et al . , 2010 ) : C85A/H234C elutes as a dimer , C85A/H234C/I201W/I422W elutes as a monomer , while C85A/H234C/I422W yields a mixture of monomers and dimers upon initial purification , which quickly dissociates into monomers as shown by re-injecting the eluted protein on the size exclusion column . The profiles of un-labelled protein , and protein labeled with Cy5 do not show any significant difference ( Figure 1—figure supplement 2 ) . Alexa-Fluor 488 ( AF488 ) SDP ester ( Thermo Fisher Scientific ) was used to label the primary amine on POPE ( 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoethanolamine; Avanti Polar Lipids , Alabaster , AL ) . AF488 was selected as it is spectrally removed from the protein labeling Cy5 channel , and it does not partition into membranes ( Hughes et al . , 2014 ) . Dye stocks were prepared in the same manner described for the cyanine-maleimide stocks . Liposomes were prepared from a 2:1 mixture of POPE and POPG ( 1-palmitoyl-2-oleoyl-sn-glycero-3-phospho- ( 1'-rac-glycerol ) ; Avanti Polar Lipids ) as a synthetic mimic of the major phospholipid composition of E . coli polar lipid extract . Briefly , a 2:1 mixture of POPE and POPG in chloroform ( 25 mg/mL ) was dried under a continuous stream of N2 gas , then resuspended in 0 . 5–1 mL pentane and dried again . Labeling buffer ( LB ) : 300 mM KCl , 100 mM NaHCO3 pH 8 . 3 was added to the lipids for a final concentration of 20 mg/mL ( 27 mM total lipids: 18 mM POPE , 9 mM POPG ) . The lipid solution was sonicated in a cylindrical bath sonicator ( Avanti Polar Lipids ) for 15 min until turbid , then 35 mM CHAPS ( Sol-grade; Anatrace ) was added and sonication continued until the solution was transparent ( 30–60 min ) . AF488 SDP ester was added to the lipid-CHAPS suspension at a final 0 . 3% mole fraction of total lipids , an amount that allows for visualization of all liposomes as measured by protein co-localization as a function of dye mole fraction . Roughly , the smallest liposomes in the population ( r = 10 nm ) are estimated to have five dyes assuming a labeling efficiency of 50% . The reaction proceeded at room temperature for 4 hr and then was stopped with 1 . 5 M Tris ( pH 8 . 0 ) . The lipids were stored at room temperature , in the dark , until fluorescently labeled protein was ready for reconstitution ( 1–3 hr ) , and then combined with the Cy5-labeled protein as described in the next section . AF488 fluorescence in the spent dialysis buffer was measured in the fluorometer at λex = 485 nm , showing no detectable free fluorophore at the end of 48 hr . For experiments that did not require fluorescent labeling , lipids were resuspended in either Reconstitution Buffer for functional or microscopy studies ( RB-F ) : 300 mM KCl , 20 mM Citrate pH 4 . 5 with NaOH , or Reconstitution Buffer for FRET ( RB-FRET ) : 150 mM NaCl , 20 mM Citrate , 10 mM MES , 20 mM Hepes , pH 7 . 0 with NaOH , with the change to NaCl allowing for the addition of SDS to samples . CHAPS ( 35 mM ) solubilized lipids were combined with protein from 0 . 0001 to 50 μg ClC-ec1 per 1 mg of lipids , corresponding to χ = 7 . 5 × 10–10 to 3 . 8 × 10–4 protein/lipid mole fraction . The protein-lipid-detergent mixture was dialyzed in cassettes ( NMWL 10 kDa; ThermoFisher Scientific ) at 4°C against 4 L of the appropriate reconstitution buffer for 48 hr with buffer changes every 8–12 hr . After completion of dialysis , the proteoliposomes were harvested from the cassettes , freeze/thawed ( see 'TIRF microscopy of proteoliposomes' ) then stored at room temperature , in the dark until further use . Quantification of Cy5-labeled protein in proteoliposomes was performed by solubilizing 10 μL of vortexed sample into 190 μL of SEB supplemented with 35 mM CHAPS detergent ( Anatrace ) and 10% CTAB ( Sigma-Aldrich ) . A standard curve was created alongside each set of samples by mixing serially diluted purified Cy5-labeled ClC-ec1 in the above buffer volume . Protein fluorescence was quantified in a 96-well plate using a Typhoon FLA 9500 Scanner ( 632nm laser/LPR emission filter permitting light >665nm ) . In general , the protein labeling yields show little variability between preps , allowing for the determination of amount of protein in each sample well compared to the standard curve . Quantification of lipids was performed by ashing proteoliposome samples in 8 . 9 N Sulfuric Acid ( Sigma-Aldrich ) at 200–215°C for 25 min . Samples were then digested further with concentrated hydrogen peroxide ( Sigma-Aldrich ) for 30 min at 200–215°C . To each sample milliQ , 2 . 5% Ammonium Molybdate ( Sigma-Aldrich ) , and 10% Ascorbic Acid ( Macron , Center Valley , PA ) was added . Color was developed by heating the samples at 100°C for 7 min . Samples were placed into a 96-well plate and the absorbance of each sample at 820 nm in a UV-VIS spectrophotometer plate reader . The samples are compared to a standard curve prepared with sodium phosphate dibasic ( RPI , Mount Prospect , IL ) along side each batch of measurements to determine the molar amount of phosphate ( Fiske and Subbarow , 1925 ) . To measure Cl- transport , un-labeled or fluorescent ClC-ec1 isoforms were reconstituted into liposomes in high chloride RB-F at 1 µg/mg , χ = 7 . 5 × 10–6 . These proteo-liposomes were freeze-thawed seven times , then extruded 21 times through a 0 . 4 µm polycarbonate membrane ( Whatman Nuclepore Track-Etched Membranes ) . The external buffer was exchanged by passing 100 µL of the liposome sample through a 2 . 5 mL Sephadex G-50 size exclusion column equilibrated in low-chloride buffer ( ExB ) : 150 mM K2SO4 , 1 mM KCl , 20 mM Citrate at pH 4 . 5 with NaOH . This sets up a Cl− gradient , however , efflux of Cl- ions does not occur because of the opposing potential driving force . A potentiometer to measure Cl- efflux was setup using silver chloride electrodes ( Walden et al . , 2007 ) . Efflux was measured by the potentiometer in reference to 1 M KCl . For the recording , 1 . 8 mL of ExB was added to the measurement cell , followed by 15 µL of 10 mM KCl for calibration of the signal . Liposomes ( ~200 µL ) in ExB were added to the measurement cell , and transport was initiated by addition of 1 µM K+ ionophore valinomycin and 2 µM of protonophore FCCP ( Sigma-Aldrich ) . To normalize and compare Cl- efflux across samples , total Cl- concentration was measured by breaking the liposomes by adding 40 µl of 1 . 5 M ß-OG , releasing the remaining chloride trapped inside unoccupied liposomes . The traces were normalized for total Cl− , then fit to a two-component exponential relaxation function to determine kClC and FCl , 0 ( Walden et al . , 2007 ) . Leak from empty 2:1 POPE/POPG liposomes was measured to determine kleak . Förster Resonance Energy Transfer ( FRET ) was measured for upon W-Cy3 + W-Cy5 mixed during reconstitution or co-labeled WT-Cy3/Cy5 in the freeze/thawed MLV state . Fluorescence was measured using a fluorometer with double monochromators on excitation and emission sides to detect fluorescence in highly scattering MLVs ( Fluorolog 3–22 , Horiba Jobin Yvon , Edison , NJ ) . Fluorescence emission spectra were collected while exciting the donor ( λEx = 530 nm , λEm = 540–850 nm , 2 nm slit width , 0 . 1 s integration time and averaged from 8 to 128 independent sequential scans ) or acceptor Cy5 ( λEx = 640 nm , λEm = 650–850 nm , 2 nm slit width , 0 . 5 s integration time ) . FRET was measured by correcting the donor emission spectrum for background fluorescence arising from the buffer and membrane , and for direct excitation of Cy5 when λEx = 530 nm . A correction factor for direction excitation of Cy5 was determined as 0 . 12 ± 0 . 01 ( n = 3 ) by measuring Cy5-only samples excited at λEx = 530 nm vs . λEx = 640 nm . The reported FRET signal is the area normalization of the FRET-specific Cy5 emission , ICy5-FRET , over the total emission , ICy3 and ICy5-FRET: ( 7 ) FRET = ∑λ=540840IλCy5−FRET∑λ=540840IxCy3+∑λ=540840IλCy5−FRET The FRET signal was plotted as a function of increasing acceptor to donor ratio ( PCy5/PCy3 ) , fit to a single site binding curve , then normalized by the maximum FRET and fit to the oligomeric model function as described in Fung et al . ( Fung et al . , 2009 ) : ( 8 ) Normalized FRET = ( R+1 ) n−Rn−1 ( R+1 ) n−Rn−1+n where R = PCy5/PCy3 and n is the number of subunits in the oligomer . A multi-wavelength single molecule total internal reflection fluorescence microscope was built following the CoSMoS design ( Friedman et al . , 2006; Larson et al . , 2014 ) . The microscope is equipped with 488 and 637 nm excitation lasers ( OBIS , Coherent Inc . , Santa Clara , CA ) each with variable attenuators for control of laser power . The lasers were focused onto the back focal aperture of a high numerical aperture 60X objective lens ( Olympus TIRF 60X 1 . 49 NA , Olympus Corporation , Tokyo , Japan ) via a 3 mm diameter micro-mirror ( MicroMirrorTIRF platform from Mad City Labs Inc . , Madison WI ) . The illuminated area on the coverslip is ~2500 µm2 . Emitted fluorescence was filtered and focused onto the CCD chip of iXon3 888 EM/CCD camera ( Andor Technology Ltd . , Belfast , Nothern Ireland ) . Glass slides ( Gold-seal , 24 × 60 mm no . 1 . 5 thickness ) and coverslips ( Gold-seal , 25 × 25 mm no . 1 . 0 thickness ) were prepared as follows: slides and cover slips were placed in slide mailers , 5 at a time , then sonicated in presence of 0 . 1% Micro-90 detergent ( Cole-Parmer , Vernon Hills , IL ) for 30 min . After sonication , slides and coverslips were rinsed with MilliQ water 10 times to remove all traces of detergent , and then sonicated in presence of absolute ethanol for 30 min followed by rinsing 10 times with MilliQ water . Finally , glass coverslips and slides were sonicated in 0 . 2 M KOH for 5 min to render the surface hydrophilic . After final rinsing with water the coverslips and slides were stored in MilliQ water in a clean air hood for a period of 7–10 days . Proteoliposomes harvested after dialysis were freeze-thawed seven times by incubating the samples in a dry ice/ethanol bath for 15 min , followed by incubation at room temperature for 15 min . At this step , the membranes fuse to form large multilamellar vesicles ( MLVs ) on the order of 10 µm in diameter ( Pozo Navas et al . , 2005 ) , i . e . 1 × 109 lipids per lamella . The samples were stored at room temperature with 0 . 02% NaN3 until imaging , 1–95 days post freeze/thaw ( see Figure 5—figure supplement 2 for specific time points of imaging ) . The formation of large membranes allows for the investigation of the protein state across a wide span of protein density ( = 7 . 5 × 10–10 to 3 . 8 × 10–4 subunits/lipid ) , with the lowest value representing roughly two subunits in a liposome lamella of ~300 µm2 – i . e . , close to infinite dilution . Prior to imaging , the membranes were extruded ( LiposoFast-Basic; Avestin , Ottawa , Canada ) 21 times through a 0 . 4 µm Nuclepore membrane to form liposomes for TIRF microscopy with a defined size distribution ( Figure 2—figure supplement 1E ) . Between samples , the extruder apparatus was dismantled , flushed profusely with hot tap water , then sonicated in presence of dilute detergent ( 0 . 5% Dial dish soap ) followed by sonication in deionized H2O water to maintain extremely low background . After washing , buffer was passed through the extruder and loaded onto the microscope to confirm that there was no contamination between samples . Liposomes were diluted in low-adhesion tubes and allowed to passively bind to glass slide ( Johnson et al . , 2002 ) . The density of fluorescent spots on the glass slide was maintained at 0 . 02 to 0 . 09 spots/µm2 to minimize overloading ( ~50–200 spots per field ) . For experiments at low protein density , the slide was loaded with a high number of liposomes since the probability of protein occupied liposomes was rare . At higher mole fractions , liposomes were serially diluted before loading onto coverslips . Images were acquired at the rate of ~1 frame per second ( fps ) , EM gain set to 300 and laser incident power typically set to 15 µW for AF488 imaging and 240 µW for Cy5 imaging in order to obtain long photo-bleaching traces while maintaining good signal to noise for single molecule spots . In each field , Cy5 imaging was carried out before AF488 imaging as we observed the 488 nm laser convert Cy5 to a dark state . After initial liposome-protein co-localization experiments to measure F0 and reconstitution efficiency , samples for protein photobleaching experiments were typically reconstituted with regular POPE/POPG mixture without any fluorescent label on the lipids . All imaging was carried out in RB-F buffer , filtered three times through a 0 . 22 µm filter ( Millex-GS MCE , Merck Millipore , Billerica , MA ) . Cy5 blinking ( Ha and Tinnefeld , 2012 ) was avoided by imaging the proteoliposomes in the absence of oxygen scavengers . Depletion of oxygen results in longer dwell time in the triplet state which often leads to fluorescent blinking . Triplet-state quenchers , such as Trolox , are often added to the imaging solutions that contain oxygen scavengers to reduce blinking behavior . However , Trolox has been shown to alter lipid bilayer properties by partitioning into membranes ( Alejo et al . , 2013 ) . Instead , we found that avoiding oxygen scavengers altogether allowed us to observe long-lived photobleaching intensity traces of Cy5 without blinking behavior . In addition , the removal of the oxygen scavengers helps to reduce background contamination as we are working with non-passivated clean slides , and we have observed components of the glucose oxidase/catalase oxygen scavenger system demonstrate autofluorescence when bound to the slide . No filtering of images was necessary , as laser power was adjusted for optimal signal to noise while ensuring relatively long photobleaching traces . Photobleaching data was collected in both unlabeled and AF488-labeled liposomes , showing no significant difference in probabilities ( Figure 4—figure supplement 1 ) . Image files were analyzed in a MATLAB-based CoSMoS analysis program ( Friedman and Gelles , 2015 ) . Fluorescent spots were auto-detected based on intensity thresholds selecting 4 × 4 pixel areas of interest ( AOIs ) around the peak fluorescence . The AOIs were integrated over time; typically 300 frames were acquired at 1 frame per second ( fps ) . Integrated intensity trajectories extracted from these AOIs were manually classified by counting the number of steps before complete photo-destruction of Cy5 . To test for subjectivity of counting , 2–3 individuals analyzed the same data set in a blinded approach ( Figure 4—figure supplement 1 ) . For probabilities calculated from an individual population , we report fraction ± standard deviation ( SD ) , calculated as the binary uncertainty p ( 1−p ) N/N , where p is the probability of a Cy5 spot bleaching in n steps and N is the total number of Cy5 spots in the sample , typically 200–500 . In most cases , the probabilities are calculated over 2–6 samples and 2–3 manual counters , with values reported as mean ± standard error ( SE ) across both samples and counters . The capture of ClC-ec1-Cy5 into extruded liposomes is a Poisson process ( Maduke et al . , 1999; Walden et al . , 2007 ) . Therefore , the probability of observing a certain number of Cy5 fluorophores , NCy5 = 1 , 2 , 3 , . . . in a liposome is related to the Poisson distribution . However , under experimentally realistic conditions , the actual distribution will differ due to the following reasons: ( 1 ) extruded liposome sizes are heterogenous , ( 2 ) microscopy visualizes Cy5 and not the protein itself , so un-labeled subunits or subunits with bleached fluorophores are not counted , and ( 3 ) there are multiple labeling possibilities for monomers and dimers due to site-specific and non-specific labeling . As long as the liposome size distribution and fluorescent labeling yields are determined experimentally , the expected P1 and P2 photobleaching probabilities for an ideal monomer or dimer can be calculated . We created a MATLAB script ( Mathworks , Natick , MA ) that simulates the random process of subunit encapsulation given a certain number of subunits , Nsubunits , and liposomes , Nliposomes , and the cryo-EM determined liposome size distribution , Pradius , for 0 . 4 µm extruded liposomes from freeze/thawed E . coli polar lipid membranes ( composition ~2:1 POPE/POPG ) ( Walden et al . , 2007 ) . The experimental mole fraction , χ , is calculated using the following equation: ( 9 ) χ=NsubunitsSAlipidNliposomes8π∑r​Pradius ( r ) r2 using a surface area per lipid , SAlipid , equal to 0 . 6 nm2 ( Murzyn et al . , 2005 ) . The simulation follows by creating a matrix of Nliposomes for each sub-population with radius r , and randomly inserting each protein species into these liposomes . In our case , we consider only the all-monomer and all-dimer condition , where Nprotein , M=Nsubunits and Nprotein , D=Nsubunits/2 . We now discuss how to calculate ( i ) Plabel , M ( NCy5 ) and Plabel , D ( NCy5 ) – the distribution of the fluorescent labeling states for the monomer or dimer , ( ii ) Nliposomes , M ( r ) and Nliposomes , D ( r ) – the number of liposomes that are accessible to monomers or dimers , and ( iii ) Nprotein , M ( r ) and Nprotein , D ( r ) – the total number of monomers or dimers to be inserted into Nliposomes , M ( r ) and . A brief description of the derivation of the dimerization equilibrium isotherm is provided here ( Wyman and Gill , 1990 ) . The equilibrium dimerization reaction scheme can be written as follows for monomers , M , and dimers , D: ( 23 ) M+M⇌D Membrane proteins react in a two-dimensional lipid bilayer and so the scale of the reaction coordinate must be appropriately selected . As discussed in the literature ( White and Wimley , 1994; 1999; Fleming , 2002; Zhang and Lazaridis , 2006 ) , the subunit/lipid mole fraction scale , χ , is a conventional choice for studying membrane protein association , as it can apply to reactions in both detergent micelles and lipids . For reactions in lipid bilayers , the area density scale , ρArea ( subunit/nm2 ) is intuitive and often used ( Hong et al . , 2010 ) . In the end , both scales are imperfect and require further corrections to account for the differences in the sizes of subunit and lipid solvent molecules ( White and Wimley , 1994; Fleming , 2002 ) . For consistency with the current literature , we report the equilibrium constant Kχ , dissociation constant Kd and standard state free energy ΔG° on the subunit/lipid mole fraction scale ( Table 2 ) , but also include the subunit/nm2 the area density scale for reference . The area density scale is calculated from the mole fraction scale using the following equation: ( 24 ) ρArea=2χSAlipid10 . 7554/eLife . 17438 . 022Table 2 . Summary of dissociation constants , equilibrium association constants and standard state free energy based on the best-fit parameters of FDimer vs . the reactive mole fraction , χ* or area density , ρArea* . DOI: http://dx . doi . org/10 . 7554/eLife . 17438 . 022Mole fraction scale ( χ* ) standard state = 1 subunit/lipidArea density scale ( ρArea* ) standard state = 1 subunit/nm2ClC-ec1 constructKd ( subunits/lipid ) Kχ ( lipids/subunit ) ΔG°χ ( kcal/mole ) Kd ( subunits/nm2 ) Kρ ( nm2/subunit ) Eq . Box ( nm × nm ) ΔG°ρ ( kcal/mole ) WTmean ± SE4 . 7 ± 1 . 1 × 10-92 . 1 ± 0 . 5 × 108-11 . 4 ± 0 . 11 . 6 ± 0 . 4 × 10-86 . 4 ± 1 . 4 × 1078002 ± 3794-10 . 7 ± 0 . 1Y0const . = 0 . 0795% CI2 . 6 to 6 . 8 × 10-91 . 2 to 3 . 1 × 108-11 . 6 to -11 . 18 . 5 × 10-9 to 2 . 3 × 10-83 . 5 to 9 . 3 × 1075919 to 9644-10 . 9 to -10 . 4Wmean ± SE2 . 7 ± 1 . 1 × 10-73 . 7 ± 1 . 6 × 106-9 . 0 ± 0 . 38 . 9 ± 3 . 8 × 10-71 . 1 ± 0 . 5 × 1061049 ± 707-8 . 3 ± 0 . 3Y0 = 0 . 07 ± 0 . 0695% CI3 . 5 × 10-8to 5 . 0 × 10-72 . 3 to 5 . 1 × 106-9 . 5 to -8 . 51 . 2 × 10-7to 1 . 7 × 10-61 . 5 × 105to 2 . 1 × 106387 to 1449-8 . 8 to -7 . 7WWmean ± SE4 . 7 ± 1 . 8 × 10-62 . 1 ± 0 . 8 × 105-7 . 3 ± 0 . 21 . 6 ± 0 . 6 × 10-56 . 3 ± 2 . 4 × 104251 ± 155-6 . 6 ± 0 . 2Y0 = 0 . 06 ± 0 . 0295% CI1 . 2 to 8 . 3 × 10-65 . 0 × 104 to 3 . 7 × 105-7 . 7 to -6 . 83 . 9 × 10-6to 2 . 8 × 10-51 . 5 × 104to 1 . 1 × 105122 to 332-7 . 0 to -6 . 1Best-fit parameters are reported as mean ± standard error ( SE ) and 95% confidence intervals ( CI ) . The area density scale is calculated by converting the mole fraction scale using SAlipid = 0 . 6 nm2 per lipid in a single leaflet and is not corrected for differences in the subunit vs . lipid volume . Eq . Box denotes the box size defined by the equilibrium constant Keq . Y0 indicates the baseline offset parameter that is either fitted or constrained . where SAlipid is the surface area per lipid in nm2 . Assuming ideal dilute conditions , the mole fraction equilibrium constant for dimerization is: ( 25 ) Kχ*=χD* ( χM* ) 2 where χ∗ is the total reactive mole fraction , equivalent to χ/2 , that subunits are randomly incorporated into the membrane ( Matulef and Maduke , 2005 ) and that the reaction only occurs between oriented subunits . χ∗D is the dimer/lipid mole fraction , χ∗M is the monomer/lipid mole fraction , and Kχ∗ is the dimerization equilibrium constant in inverse mole fraction units ( i . e . lipid/subunit ) ( Fleming , 2002 ) . The total mole fraction of subunits in the membrane is: ( 26 ) χ∗=χM∗+2χD∗ The fraction of protein in the dimer state ( FDimer ) is derived by substituting ( 26 ) into ( 25 ) to determine χD: ( 27 ) FDimer=2χD∗χ∗=1+4χ∗Kχ∗−1+8χ∗Kχ∗4χ∗Kχ∗ To determine FDimer from the photobleaching analysis , we used the theoretical calculations of the ideal monomer probabilities , PDn=1−5+ , and ideal dimer probabilities , PDn=1−5+ as the respective all-monomer and all-dimer signals . Least-squares analysis was carried out on the sum of squared residuals ( R2 ) between the experimental data , Pexptn , and a linear combination of PMn=1−5+ , and , PDn=1−5+ , weighted by FDimer ( 28 ) R2=∑n=1−5+ ( Pnexpt− ( ( 1−FDimer ) ⋅PnM+FDimer⋅PnD ) ) 2 The minimum R2 value corresponds to the predicted FDimer for a given photobleaching distribution ( Figure 5—figure supplement 1 ) . FDimer vs . χ∗ was fit to the equilibrium dimerization isotherm above with the addition of a baseline offset Y0: ( 29 ) FDimer= ( 1−Y0 ) ∗ ( 1+4χ∗Kχ∗−1+8χ∗Kχ∗4χ∗Kχ∗ ) +Y0 Analysis of the R2 as a function of mole fraction density shows that the quality of the fits of the experimental distributions to the theoretical distributions deviate for χ∗ >1 . 9 × 10–6subunits/lipid ( Figure 5—figure supplement 2 ) , which could arise from inaccuracies in the liposome size distribution by exclusion of larger liposomes or multilamellar liposomes , or a small proportion of non-specific oligomerization . To account for this uncertainty in FDimer estimation at high densities , we weighted the fits by 1/R2 , essentially limiting the data to a dynamic range of χ∗ = 3 . 8 × 10–6 to 1 . 9 × 10–6 . The weighted non-linear fits were carried out using the non-linear regression function in MATLAB . The mole fraction standard state free energy is calculated as: ( 30 ) ΔG∘=−RT ln ( χ∘Kχ∗ ) where R is the gas constant ( 1 . 987 2036 cal/mole K ) , T is the temperature ( ~298 K ) and χ° is the standard state mole fraction density 1 subunit/lipid . Note that this standard state is not physically relevant , but provides a normalization point for other measurements to be compared . On the area density scale , the standard state density ρ° = 1 subunit/nm2 is used . For comparison , the typical standard state of 1 M on the molar scale is equivalent to 1 subunit/1 . 6 nm3 or 1 subunit/54 water molecules assuming a molecular volume of water of 30 Å3 ( White and Wimley , 1999 )

Cells are encapsulated by membranes that form a barrier between the inside of the cell and the outside world . These membranes primarily consist of fatty molecules called lipids , but they are also packed with proteins such as ion channels and transporters that control which molecules pass in and out of the cell . It is proposed that membrane proteins fold spontaneously inside of the cell membrane to adopt the structures that allow them to carry out their function . While it is generally understood why proteins fold spontaneously when they are in water , it is less clear why this occurs for membrane proteins in the oily cell membrane . Proteins fold into specific shapes because of favorable interactions between different surfaces of the molecule and because the folded structure increases the number of states available to the protein and the surrounding environment . Measuring a quantity known as the “free energy” reports the net balance between these thermodynamic factors and is the first step towards understanding why a membrane protein adopts its particular stable structure in the cell membrane . However , few experimental systems are suitable for studying this reaction in membranes . One important group of membrane proteins that offers a new approach to studying this question is the CIC family of channels and transporters . These are large proteins that in Escherichia coli bacteria and other organisms have only been observed as a “dimer” made up of two identical CIC molecules ( or “monomers” ) . It is however possible that within cell membranes , the CIC transporter proteins switch between their dimer and monomer forms . Reducing the number of proteins in the membrane could reveal these monomers , and allow the free energy associated with forming a dimer from two monomers to be measured . Chadda et al . diluted E . coli CIC protein from natural cell membranes into large synthetic membranes to reduce the number of proteins far below the amount normally seen in cells . Examining the membranes using a technique called single molecule fluorescence microscopy revealed that CIC does exist as monomers when present in low amounts in a membrane . Furthermore , measuring the free energy associated with forming a dimer showed that ClC is one of the strongest membrane protein dimers measured so far . Chadda et al . also found that CIC is more likely to be in its monomer form if a bulky amino acid called tryptophan is added to the interface at which two CIC molecules bind to each other . Future studies will investigate the mechanism that underlies this change in stability . Ultimately , CIC could serve as a model system to study the forces associated with protein assembly in membranes and answer fundamental questions about membrane protein folding .

2016

The dimerization equilibrium of a ClC Cl−/H+ antiporter in lipid bilayers

Using a new bioinformatic method to analyze ribosome profiling data , we show that 40% of lncRNAs and pseudogene RNAs expressed in human cells are translated . In addition , ~35% of mRNA coding genes are translated upstream of the primary protein-coding region ( uORFs ) and 4% are translated downstream ( dORFs ) . Translated lncRNAs preferentially localize in the cytoplasm , whereas untranslated lncRNAs preferentially localize in the nucleus . The translation efficiency of cytoplasmic lncRNAs is nearly comparable to that of mRNAs , suggesting that cytoplasmic lncRNAs are engaged by the ribosome and translated . While most peptides generated from lncRNAs may be highly unstable byproducts without function , ~9% of the peptides are conserved in ORFs in mouse transcripts , as are 74% of pseudogene peptides , 24% of uORF peptides and 32% of dORF peptides . Analyses of synonymous and nonsynonymous substitution rates of these conserved peptides show that some are under stabilizing selection , suggesting potential functional importance . In the central dogma , mRNAs are translated into proteins that carry out biological functions . On a genomic scale , translated regions are identified as open reading frames ( ORFs ) that are longer ( typically >100 amino acids ) than expected by chance , given sequence composition . In addition to mRNAs , mammalian cells contain other RNA transcripts generated by RNA polymerase II that are polyadenylated , spliced , and capped , but may not code for protein . One category consists of thousands of long RNAs that lack long open reading frames and have been considered to be non-coding ( Guttman et al . , 2009; 2010; Trapnell et al . , 2010; Cabili et al . , 2011 ) . A few lncRNAs play key regulatory roles in various biological processes via functional RNA domains that regulate chromatin modifications , DNA transcription , mRNA stability , and translation ( Rinn and Chang , 2012; Batista and Chang , 2013; Ulitsky and Bartel , 2013 ) . However , the biological functions of most lncRNAs remain unknown . The human genome also encodes thousands of pseudogenes , which are homologous to protein-coding genes but have lost their coding ability and/or are not expressed ( Vanin , 1985 ) . Pseudogenes can function as competing endogenous RNAs ( ceRNAs ) regulating other RNA transcripts by competing for microRNAs ( Salmena et al . , 2011 ) . Some pseudogenes are differentially expressed in human cancers ( Kalyana-Sundaram et al . , 2012; Han et al . , 2014 ) , but it is unknown if the RNAs expressed from pseudogenes are translated or have biological functions . By definition , noncoding RNAs should not be translated into protein , but this can be difficult to ascertain using informatics alone because they contain short open reading frames that could be potentially translated . Even if a peptide is expressed from a putative non-coding RNA , it is difficult to determine whether the peptide has a biological function or is a mere by-product of an RNA that performs the biological function . However , there are a few examples of lncRNAs that are in fact translated into short peptides with biological roles ( Galindo et al . , 2007; Kondo et al . , 2010; Magny et al . , 2013; Pauli et al . , 2014 ) . In addition , a number of mammalian mRNAs contain so-called 5’ untranslated regions ( 5’UTRs ) with one or more ORFs upstream of their canonical protein-coding regions ( uORFs ) . Due to the scanning mechanism for translational initiation in which ribosomes scan in a 5’ to 3’ direction from the mRNA cap to find an initiation codon ( Sonenberg and Hinnebusch , 2009 ) , uORFs have the potential to regulate translation of the primary protein-coding ORF ( Calvo et al . , 2009; Barbosa et al . , 2013 ) . For example , translation of the uORFs in the yeast GCN4 gene strongly inhibits translation of Gcn4 under normal conditions ( Hinnebusch , 2005 ) . However , during amino acid starvation , ribosomes reinitiate translation at the canonical AUG codon , thereby permitting increased synthesis of Gcn4 ( Hinnebusch , 2005 ) . In human cells , bioinformatic analyses and limited functional testing indicate that uORFs can inhibit protein production , but genome-wide functional analysis has yet to be performed ( Calvo et al . , 2009; Barbosa et al . , 2013 ) . Ribosome profiling , the sequencing of ribosome-associated RNAs , represents a powerful assay for assessing translation in vivo in an unbiased manner on a genome-wide scale ( Ingolia et al . , 2009; Ingolia , 2014 ) . In particular , ribosome profiling in mammalian cells reveals many reads derived from lncRNAs and 5’ UTRs , and lncRNAs and 5’UTRs can be co-purified with 80S ribosome , indicating that these transcripts are translated ( Ingolia et al . , 2011; 2014 ) . However , unlike canonical protein coding-genes translated from mRNAs , many lncRNAs do not have a predominant ORF based on the ribosome release or disengagement scores ( Chew et al . , 2013; Guttman et al . , 2013 ) . However , due to a variety of limitations , previous analyses typically did not explicitly identify in-frame translated ORFs , and they identified only several hundred translated regions that do not correspond to canonical protein-coding regions . Importantly , ribosome profiling reads do not necessarily represent its active translation , due to potential artifacts from non-ribosomal entities and scanning ribosomes ( Guttman et al . , 2013; Ingolia et al . , 2014 ) . Systematic examination of translation requires a computational method to identify bona fide translated ORFs in an unbiased fashion . Here we develop a method , RibORF , to analyze ribosomal profiling data and identify translated ORFs that combines alignment of ribosomal A-sites , 3-nt periodicity , and uniformity across codons . RibORF can effectively distinguish in-frame ORFs from overlapping off-frame ORFs , and it can distinguish reads arising from RNAs that are not associated with ribosomes . Using RibORF , we identify thousands of translated ORFs in lncRNAs , pseudogenes , and mRNA regions upstream ( 5’UTRs ) and downstream ( 3’UTRs ) of protein-coding sequences . Our results suggest that cytoplasmic noncoding RNAs are translated , and that some of these translated products are likely to be biologically meaningful based on their evolutionary conservation . We performed ribosome profiling ( Figure 1A ) in two isogenic human cancer cell models: a Src-inducible mammary epithelial model and a Ras-dependent fibroblast model ( Hirsch et al . , 2010 ) . Cells were treated either with cycloheximide , which inhibits translational elongation of ribosomes throughout the mRNA coding region , or harringtonine , which traps the ribosome at the site of translational initiation . After removing reads aligned to rRNAs and multiple genomic locations , we generated 44 . 0 and 21 . 2 million unique mappable reads upon cycloheximide treatment for breast epithelial and fibroblast cell transformation models , respectively . For harringtonine treatment , we obtained 5 . 9 and 9 . 0 million unique mappable reads for breast epithelial and fibroblast cells , respectively . 10 . 7554/eLife . 08890 . 003Figure 1 . Ribosome profiling reveals in vivo translation with single nucleotide resolution . ( A ) Ribosome profiling experiment . ( B ) Read distribution ( reads/million mappable reads; RPM ) around start and stop codons of canonical protein coding genes . ( C ) Fractions of reads in 1st , 2nd and 3rd nucleotides of codons in the indicated types of ORFs . ( D ) Read distribution in the protein-coding gene CPSF2 . The RPM value was calculated for every 20-nt region along the transcript . ( E ) Distribution of reads across human genome . ( F ) Read distribution of the snoRNA gene SNORA49 in cells treated with cycloheximide ( Chx ) or harringtonine ( Harr ) . ( G ) Distribution of PME values in the indicated types of ORFs . DOI: http://dx . doi . org/10 . 7554/eLife . 08890 . 00310 . 7554/eLife . 08890 . 004Figure 1—figure supplement 1 . Ribosome profiling data . ( A ) RPF length distribution . ( B ) The read distribution of RPFs around start and stop codons of canonical mRNA ORFs . RPFs were grouped based on their length . DOI: http://dx . doi . org/10 . 7554/eLife . 08890 . 004 The length of ribosome-protected fragments ( RPFs ) ranges primarily between 24–31 nts ( Figure 1—figure supplement 1A ) . Notably , RPFs with different length have variable distances between the 5’ end and the ribosome A-site , as defined by canonical ORFs in protein-coding genes ( Figure 1—figure supplement 1B ) . We used these offset distances in known protein-coding genes to account for the read length distribution and thereby align RPFs to specific A-site nucleotides throughout the entire dataset . Most expressed protein-coding ORFs show a clear 3-nt periodicity corresponding to codon triplets ( Figure 1B , C ) . The 1st nucleotides of codons in an ORF contain about 65% of reads , while the 2nd and 3rd have 24% and 11% , respectively ( Figure 1B , C ) . In addition , reads in protein-coding genes are uniformly distributed across codons in an ORF ( Figure 1D ) . 73% of ribosome profiling reads map to canonical ORFs of mRNAs . 2% and 4% map to 5’UTRs and 3’UTRs of mRNAs , respectively , and 9% map to lncRNAs and pseudogenes , suggesting pervasive non-canonical translation ( Figure 1E ) . Consistent with previous reports ( Ingolia et al . , 2011; Guttman et al . , 2013 ) , some ribosome profiling reads map to short noncoding RNAs , including small nucleolar RNA ( snoRNA ) . As snoRNAs are located in nucleus , they should not be accessible to translation machinery located in cytoplasm . Indeed , the sequence reads in the snoRNAs map to a very narrow region and are comparable in the cycloheximide- and harringtonine-treated samples ( Figure 1F ) , indicating they do not represent translated regions of these RNAs . To exclude reads that do not represent active translation , we developed a Percentage of Maximum Entropy ( PME ) approach to measure the uniformity of read distribution across codons in a candidate ORF ( See Experimental Procedures ) . A PME value of 1 represents uniform read distribution , indicative of real translation , while smaller values indicate skewed distribution with a minimum value of 0 indicating reads at a single location , expected for reads not derived from translated RNA . As expected , candidate ORFs from short noncoding RNAs show drastically lower PME values , as compared to canonical protein coding ORFs ( Figure 1G ) . Low PME values indicate RNAs that are not translated , but rather are protected in non-ribosomal protein complexes ( Ji et al . , 2015 ) . Based on the 3-nt periodicity ( Figure 1C ) and uniformity of read distribution across codons ( Figure 1G ) of translated regions , we developed a Support Vector Machine classifier , RibORF , to identify translated ORFs from ribosome profiling data . The model was trained by using canonical protein-coding ORFs as positive examples and off-frame ORFs from protein-coding regions and candidate ORFs from short noncoding RNAs as negative examples . The classifier using both features performed almost perfectly to separate positive and negative examples in a testing set ( Area Under the ROC Curve [AUC] = 0 . 996 ) , with 3-nt periodicity making a greater contribution ( Figure 2A ) . The algorithm performed well for genes expressed at various levels , with AUC values greater than 0 . 993 for ORFs with RPKM > 1 ( Figure 2—figure supplement 1A ) . In addition , the predicted translation probabilities are well correlated in the two cancer models ( R = 0 . 97 ) , indicating the algorithm can be robustly applied to various cell types ( Figure 2—figure supplement 1B ) . 10 . 7554/eLife . 08890 . 005Figure 2 . RibORF identifies translating ORFs . ( A ) Receiver-operating characteristic ( ROC ) curves to measure algorithm performance using different training parameters . ( B ) Types of translated ORFs identified in this study , with ORF number:gene number shown in parenthesis . ( C ) Distribution of reads upon cycloheximide treatment around start codon of predicted positive and negative lncRNA ORFs . Examples of ( D ) a translated lncRNA ( E ) an mRNA with a uORF ( F ) an mRNA with a dORFs; the 3’ most exon is shown . Enlarged figures show 3-nt periodicity can be observed for each codon in Figure 2D–F . DOI: http://dx . doi . org/10 . 7554/eLife . 08890 . 00510 . 7554/eLife . 08890 . 006Figure 2—figure supplement 1 . RibORF algorithm performance . ( A ) ORFs were grouped based on expression levels , and corresponding AUC values were plotted as in Figure 2A . ( B ) Correlation of predicted translating probability of candidate ORFs , using ribosome profiling data from MCF10A-ER-Src cells and fibroblast cells . 1000 randomly selected candidate ORFs were used in the analyses . ( C ) Candidate ORFs were grouped based on predicted translating probability . Fractions of reads in 1st , 2nd and 3rd nucleotides of codons and PME values in different groups were shown . ( D ) Distribution of ribosome profiling reads around start codon of predicted positive and negative uORFs . DOI: http://dx . doi . org/10 . 7554/eLife . 08890 . 00610 . 7554/eLife . 08890 . 007Figure 2—figure supplement 2 . Analysis of ribosome-associated RNA . ( A ) Sucrose gradient fractionation of polyribosomes with fractions indicated . ( B ) Analysis of RNAs associated with 80S monoribosomes ( fraction 1 ) and polyribosomes with 2 ( fraction 2 ) or 3+ ( combining fractions 3–6 ) ribosomes . The RNAs analyzed including seven predicted translated lncRNAs , the IL6 mRNA as a positive control , and non-translating lncRNA ENSG00000256973 . 1 and snoRNA SNORD105 as negative controls . The amounts for the ribosome-associated RNAs are expressed with respect to the amounts of these RNAs in the unfractionated samples prior to sucrose gradient centrifugation . DOI: http://dx . doi . org/10 . 7554/eLife . 08890 . 007 We applied the classifier to predict translated ORFs within lncRNAs , pseudogenes , and mRNAs . Candidate ORFs showed a mixed population of 3-nt periodicity and PME values ( Figure 1C , G ) . Using a stringent cutoff for the probability of prediction ( 0 . 7 with a false positive rate 0 . 67% and a false negative rate 2 . 5%; Figure 2—figure supplement 1C ) , we identified canonical ORFs in 10 , 946 protein-coding genes , and truncated or extended variants in 544 genes ( Figure 2B ) . The canonical ORFs in almost all expressed transcripts were identified . In addition , we identified so-called uORFs in the 5’UTRs of 3842 protein-coding genes , and uORFs overlapping with coding regions ( overlapping uORFs ) in 1054 genes Figure 2B ) . We also identified ORFs located in 3’UTRs of 550 genes , which we term downstream ORFs ( dORFs; Figure 2B ) . In general , translated uORFs and dORFs are expressed from the same transcript as the relevant canonical ORF , although in some cases these may arise from truncated transcripts . Lastly , we identified 1204 ORFs in 510 lncRNAs and 278 ORFs in 161 pseudogenes ( Figure 2B ) . As expected , the predicted translated ORFs show clear 3-nt periodicity and high PME values , while the negative ones do not ( Figure 2C and Figure 2—figure supplement 1C-D ) . Examples of lncRNA ORFs , uORFs and dORFs are shown in Figure 2D-–F , and a full list is presented in Supplementary file 1 . For the well expressed ORFs , we observe 3-nt periodicity for individual codons ( Figure 2D–F ) . Uniform 3-nt periodicity over an extended distance is diagnostic of bona fide translation . In this regard , all 7 tested RNAs encoding non-canonical translated ORFs are associated with 80S monosomes and/or polysomes ( Figure 2—figure supplement 2 ) . Thus , we will refer to the products of translated ORFs as 'peptides' , even though direct biochemical evidence is lacking . In this regard , the peptides represent initial translation products whose stability in vivo is unknown . We suspect that many non-functional peptides will be degraded rapidly and hence difficult to detect biochemically . We did not detect translation for 679 lncRNAs in breast epithelial cells even though RNA-seq analysis indicates that they are expressed at comparable levels to the 510 translated lncRNAs ( p>0 . 05; Figure 3A ) . We hypothesized that the distinction between these two classes is that the untranslated lncRNAs would be preferentially localized in nucleus and not accessible to the translation machinery , whereas the translated lncRNAs would be preferentially localized in the cytoplasm . To test this hypothesis , we examined the cytosolic and nuclear distribution ( C:N ratio ) of lncRNAs , using RNA-seq data from multiple cell lines ( Djebali et al . , 2012; ENCODE , 2012 ) . Indeed , untranslated lncRNAs are less likely to localize to the cytoplasm ( lower C:N ratio ) , than translated ones ( p<10-70; Figure 3B ) . Similar results are observed for lncRNAs in a variety of cell lines ( Figure 3—figure supplement 1A–D ) . Compared to canonical protein coding mRNAs , translated lncRNAs show slightly lower C:N ratios ( p<10-46; Figure 3B ) . Translated pseudogene RNAs are also more likely to be localized in the cytoplasm as compared with untranslated pseudogene RNAs ( Figure 3—figure supplement 1E–G ) . 10 . 7554/eLife . 08890 . 008Figure 3 . RNA subcellular localization is a major determinate of translation efficiency . ( A ) RNA expression levels of lncRNAs with or without translated ORFs and canonical mRNAs in MCF10A-ER-Src cells . ( B ) Relative subcellular location of translated and untranslated lncRNAs and canonical mRNAs . ( C ) Translation efficiency of translated lncRNAs and canonical mRNAs . ( D ) Distribution of translation efficiency of canonical mRNAs , calculated as averaged translation efficiency values in breast epithelial and fibroblast cells . ( E ) Relative subcellular locations of mRNAs grouped based on translation efficiency . DOI: http://dx . doi . org/10 . 7554/eLife . 08890 . 00810 . 7554/eLife . 08890 . 009Figure 3—figure supplement 1 . RNA subcellular localization regulates translation . ( A ) RNA expression levels of expressed lncRNAs with or without translated ORFs and mRNAs in fibroblast cells measure by RNA-seq . ( B ) Translation efficiency of translated ORFs in lncRNAs and canonical ORFs in mRNAs in fibroblast cells . ( C , D ) Relative subcellular location of translated/untranslated lncRNAs and mRNAs . RPKM values were calculated using RNA-seq data for nucleus and cytosol fractions of K562 ( C ) and Hepg2 ( D ) cells . ( E–G ) Relative subcellular localization of translated/untranslated pseudogenes . ( H , I ) mRNAs were grouped based on translation efficiency as in Figure 3D , and relative subcellular locations of mRNAs in K562 ( H ) and Hepg2 ( I ) cells were shown . DOI: http://dx . doi . org/10 . 7554/eLife . 08890 . 009 Translation efficiency of a given RNA is defined as the ratio of translated RNA ( from ribosomal profiling ) : overall RNA ( from RNA-seq ) . In accord with the reduced C:N ratio of translated lncRNAs as compared to mRNAs , lncRNAs also show lower translation efficiency ( p<10-12; Figure 3C ) . However , when corrected for the reduced levels of lncRNAs in the cytoplasm , it appears that the translation efficiency of cytoplasmic lncRNAs and mRNAs are nearly comparable , albeit slightly reduced . Interestingly , the translation efficiencies of mRNAs vary hundreds of fold ( Ingolia et al . , 2009 ) ( Figure 3D ) , and these differences are strongly correlated with localization in the cytosol ( Figure 3E and Figure 3—figure supplement 1H–I ) . The strong relationship between nucleo-cytoplasmic location and translatability of lncRNAs provides strong independent evidence that our classifier effectively identifies translated RNAs . In addition , translation efficiency is strongly correlated with degree of cytoplasmic location , indicating that accessibility of an RNA to the translation machinery is a major determinant of how well it is translated . Over 40% ( 491 out of 1189 ) of expressed lncRNAs encode peptides longer than 10 aa , and 8% ( 98 lncRNAs ) encode peptides longer than 100 aa ( Figure 4A ) . The median length of all peptides translated from lncRNAs ( 43 aa; Figure 4B ) is considerably longer than that of peptides generated from uORFs ( 17 aa ) . Translation of many lncRNAs yields multiple peptides from non-overlapping ORFs , and the median length of the longest peptide translated by a given lncRNA is 62 aa ( Figure 4C ) . Translated lncRNAs use AUG start codons more often than uORFs ( Figure 4—figure supplement 1A , B ) . 10 . 7554/eLife . 08890 . 010Figure 4 . Features and conservation of lncRNA peptides . ( A ) Fraction of expressed lncRNAs that encode peptides longer than a certain length . ( B ) Peptide length encoded by lncRNAs . ( C ) Length of the longest peptide in a given lncRNAs . ( D ) Length of conserved lncRNA peptides . ( E ) LncRNA LOC284023 encodes two peptides , the upstream one being conserved in the mouse lncRNA Chd3os . ( F ) Ka and Ks values of types of conserved lncRNA peptides with Z-Test p-values shown . ( G ) Ka/Ks ratios of types of conserved lncRNA peptides . DOI: http://dx . doi . org/10 . 7554/eLife . 08890 . 01010 . 7554/eLife . 08890 . 011Figure 4—figure supplement 1 . Features of lncRNA translation . ( A ) Start codon of translated ORFs in lncRNAs and mRNAs . ( B ) Start codon of translated ORFs in lncRNA grouped based on length . ( C ) Length of the longest candidate ORFs in a given lncRNAs considering start codon variants ( A/C/G/UUG ) . ( D ) Length of the longest candidate ORFs in a given lncRNAs versus length of the longest peptides translated in a given lncRNAs . DOI: http://dx . doi . org/10 . 7554/eLife . 08890 . 01110 . 7554/eLife . 08890 . 012Figure 4—figure supplement 2 . Conservation of nucleotides encoding lncRNA and pseudogene peptides . ( A ) PhastCon scores of nucleotides encoding lncRNA peptide grouped based on length . The median PhastCon value of translated ORFs in each group was shown . The PhastCon scores of random untranslated sequences of matching sizes and locations are also plotted . ( B ) PhastCon scores of nucleotides encoding pseudogene peptide grouped based on length . The median PhastCon value of translated ORFs in each group was shown . The PhastCon scores of random untranslated sequences of matching sizes and locations are also plotted . ( C ) Fractions of lncRNA and pseudogene peptides with protein domain annotated by Pfam ( including both Pfam-A and Pfam-B ) using default cutoff E-value <1 ) . ( D ) PhastCon scores of nucleotides in ORFs of short lncRNA and pseudogene peptides ( <100 aa ) with or without protein domains . p-values based on the Wilcoxon Rank Sum Test were shown . DOI: http://dx . doi . org/10 . 7554/eLife . 08890 . 01210 . 7554/eLife . 08890 . 013Figure 4—figure supplement 3 . Coding potential of nucleotides encoding lncRNA and pseudogene peptide . ( A ) PhyloCSF scores of nucleotides encoding lncRNA peptide grouped based on length . The PhyloCSF scores of random untranslated sequences of matching sizes and locations are also plotted . Wilcoxon Rank Sum Test p-value comparing ORF sequences and untranslated sequences were shown . And scores of ORFs encoding peptides conserved in mouse and those with Ka/Ks < 0 . 5 were also shown . ( B ) PhyloCSF scores of nucleotides encoding pseudogene peptide grouped based on length . The PhyloCSF scores of random untranslated sequences of matching sizes and locations are also plotted . Wilcoxon Rank Sum Test p-value comparing ORF sequences and untranslated sequences were shown . And scores of ORFs encoding peptides conserved in mouse and those with Ka/Ks < 0 . 5 were also shown . DOI: http://dx . doi . org/10 . 7554/eLife . 08890 . 01310 . 7554/eLife . 08890 . 014Figure 4—figure supplement 4 . BLASTP E-values of peptide sequences encoded by homologous human and mouse ORF . ( A ) LncRNAs ( B ) Pseudogene RNAs BLASTP E-values between human translated ORFs and their randomized sequences were shown as the control . DOI: http://dx . doi . org/10 . 7554/eLife . 08890 . 01410 . 7554/eLife . 08890 . 015Figure 4—figure supplement 5 . BLASTP E-values of peptide sequences encoded by homologous human and mouse peptides . ( A ) uORFs ( B ) Overlapping uORFs ( C ) Internal ORFs ( D ) dORFs BLASTP E-values between human translated ORFs and their randomized sequences were shown as the control . DOI: http://dx . doi . org/10 . 7554/eLife . 08890 . 01510 . 7554/eLife . 08890 . 016Figure 4—figure supplement 6 . The Ka/Ks ratios between human translated ORFs and 50 randomly generated sequences with BLASTP alignment E-value <10-4 . ( A ) ORFs < 50 aa . ( B ) ORFs ≥ 50 aa . DOI: http://dx . doi . org/10 . 7554/eLife . 08890 . 016 For mRNAs , the longest candidate ORFs are virtually always translated into functional proteins , but this is not the case for lncRNAs . The median length of the longest candidate ORF in a given lncRNA is 79 aa , but the longest candidate ORFs is translated only for 56% of the lncRNAs ( Figure 4—figure supplement 1C , D ) . For the remaining 38% of the lncRNAs , the translated ORF was located upstream of the longest ORF . This preferential translation of ORFs located closer to the 5’ ends of the lncRNAs likely reflects the strong preference of translation to initiated at the first AUG codon . The fact that the longest candidate ORF and/or its 5’ proximal location is not necessarily the portion of the lncRNA that is translated indicates the value of the RibORF algorithm . To address the functional significance of peptides translated from lncRNAs , we used four approaches to study their evolutionary conservation . First , we used PhastCon scores based on 44-vertebrate Multiz alignment ( Siepel et al . , 2005 ) to measure conservation of ORF nucleotide sequence among species ( Figure 4—figure supplement 2 ) . Second , we used the PhyloCSF score to study the protein-coding potential of ORF sequences based on 29-mammal genome alignment ( Lin et al . , 2011 ) ( Figure 4—figure supplement 3 ) . Third , we checked the conservation of human peptides in mouse transcripts at the amino acid level and defined them to be conserved if two homologous ORFs encode peptides with a BLASTP alignment E-value <10-4 ( False Discovery Rate < 0 . 0005 for all types and lengths of ORFs; Figure 4—figure supplements 4 and 5 , and Supplementary file 2 ) . Fourth , for lncRNA peptides conserved between human and mouse , we computed the ratio of nonsynonymous ( Ka ) to synonymous ( Ks ) substitution rates of the homologous nucleotide sequences . The Ka/Ks ratio is a commonly used parameter to infer the direction and magnitude of natural selection on peptide sequences ( Hurst , 2002 ) . A ratio smaller than 1 indicates a significant number of nucleotide sequence changes that do not result in protein sequence changes , indicating that the protein is under stabilizing ( negative ) selection and likely to be functional . For these analyses , we excluded the 30 lncRNAs that encode peptides conserved in mouse protein-coding genes and likely to be pseudogenes mis-annotated by GENCODE ( Supplementary file 2 ) . For each translated ORF , we compared its conservation level ( Phastcon and PhyloCSF score ) to untranslated segments that are matched for length and transcript location . Interestingly , at the nucleotide level , translated ORF sequences tend to be more conserved and have higher coding potential than the untranslated sequences ( p<10-4; Figure 4—figure supplements 2A and 3A ) . The pattern is consistent for translated ORFs with different lengths , suggesting that some peptides might be functional . Most lncRNA peptides ( 92% ) do not contain protein domains annotated by Pfam ( Punta et al . , 2012 ) ( Figure 4—figure supplement 2C ) . ORF nucleotide sequences encoding short peptides ( <100 aa ) containing protein domains are more conserved ( p<10-3; Figure 4—figure supplement 2D ) . 93 translated lncRNAs ( 19% of the total ) have homologous lncRNA genes in mouse . From those conserved lncRNA genes , 41 ( 44% ) express conserved peptides , with a median length 69 aa ( Figure 4D , Figure 4—figure supplement 4A , and Supplementary file 2 ) . As expected , these conserved peptides have higher coding potential than non-conserved ones ( Figure 4—figure supplement 3A ) . For example , the human lncRNA LOC284023 expresses a 97 aa peptide encoded by the 5’ end , and a 37 aa peptide encoded downstream ( Figure 4E ) . The 97 aa peptide is conserved in mouse homologous transcript Chd3os , while the 37 aa peptide is not . Interestingly , human lncRNA peptides conserved with mouse peptides encoded by lncRNAs have Ka/Ks ratios significantly lower than 1 ( Figure 4F , G ) . The low Ka/Ks ratios were not due to our BLASTP E-value cutoff ( Figure 4—figure supplement 6 ) . 20 such lncRNAs express peptides with Ka/Ks values smaller than 0 . 5 , and 12 have values < 0 . 3 . Consistently , peptides with lower Ka/Ks values have higher coding potential based on PhyloCSF scores ( Figure 4—figure supplement 3A ) , suggesting that they are evolutionary stabilized and are probably functionally important . The human genome contains 13 , 708 annotated pseudogenes that are derived from ancestral protein-coding genes but generally not expressed as RNAs and believed to have lost their protein-coding capability . However , out of 426 expressed pseudogenes ( ~3% of those annotated ) , 155 ( 36% ) are translated into peptides longer than 10 aa . In addition , 81 expressed pseudogenes ( 19% ) generate peptides longer than 100 aa ( Figure 5A ) , and most ( ~80% ) of these contain at least one protein domain ( Figure 4—figure supplement 2C ) . The median length of pseudogene peptides is 70 aa ( Figure 5B ) , and the median length of the longest peptide translated by a pseudogene is 102 aa ( Figure 5C ) , which is 30 aa longer than lncRNA peptides . 10 . 7554/eLife . 08890 . 017Figure 5 . Features and conservation of pseudogene peptides . ( A ) Fraction of expressed pseudogenes that encode peptides longer than a certain length . ( B ) Peptide length encoded by pseudogenes . ( C ) Length of the longest peptides in a given pseudogenes . ( D ) Length of conserved pseudogene peptides . ( E ) Peptide in a human pseudogene FAM86C2P is conserved in the mouse protein coding gene Fam86 . FAM86C2P also has a homologous human protein coding gene FAM86A . ( F ) Conserved human pseudogene peptides , grouped based on their homologous ORF types in mouse genome . ( G ) Ka and Ks values of types of conserved pseudogene peptides with Z-Test p-values shown . ( H ) Ka/Ks ratios of types of conserved pseudogene peptides . DOI: http://dx . doi . org/10 . 7554/eLife . 08890 . 017 Nucleotide sequences of translated ORFs in pseudogenes are significantly more conserved and have higher coding potential than untranslated sequences of the matching sizes and relative positions , and the pattern is consistent for translated ORFs of various sizes ( p<10-22; Figure 4—figure supplements 2B and 3B ) . 114 pseudogene peptides ( 74% out of those translated ) are conserved in mouse , with a median length 92 aa ( Figure 5D , Figure 4—figure supplements 3B and 4B , and Supplementary file 2 ) that is ~25% the length of the corresponding canonical proteins . For example , the mouse protein-coding gene Fam86 has a homologous protein-coding gene FAM86A in human , and also has a homologous pseudogene FAM86C2P , which is annotated as a long noncoding RNA . We found FAM86C2P is translated into a peptide with 131 aa , while mouse Fam86 protein is 336 aa ( Figure 5E ) . Several internal coding exons in Fam86 are lost in FAM86C2P during evolution . 69% of conserved human pseudogene peptides are homologous to canonical ORFs in mouse mRNAs ( Figure 5F ) . As a class , these conserved peptides show a Ka/Ks ratio significantly lower than 1 ( Figure 5G , H ) , with 50 pseudogenes expressing peptides with Ka/Ks values lower than 0 . 3 . This suggests that , although some human pseudogenes are translated into shorter peptides than their mouse homologs , the peptide sequences are evolutionarily constrained , and hence may play functional roles . In addition , 15% of conserved pseudogene peptides are homologous to mouse pseudogenes , and these peptides also have Ka/Ks ratios even lower than those homologous to mouse canonical ORFs , including 19 with Ka/Ks ratios < 0 . 3 ( Figure 5F–H ) . Thus , pseudogenes with longer evolutionary histories are more likely to encode functional peptides . In contrast , the remaining 16% of conserved pseudogene peptides are homologous to non-canonical ORFs in mouse mRNAs , and these peptides have Ka/Ks ratios close to 1 suggesting they are nonfunctional ( Figure 5F–H ) . The median lengths of uORFs ( 17 aa ) and overlapping uORFs ( 37 aa ) are shorter than those of lncRNAs and pseudogene peptides ( Figure 6A ) . In general , the translation efficiency of uORFs is similar to that of canonical protein-coding sequences ( Figure 6B ) , and this effect is typical for individual genes . However , in accord with previous results linking uORFs to decreased protein levels ( Calvo et al . , 2009; Barbosa et al . , 2013 ) , the translational efficiency of mRNA coding regions is slightly lower for genes containing uORFs ( p<10-34; Figure 6C ) , even though RNA levels of uORF-containing genes somewhat higher than genes lacking uORFs ( p<10-200; Figure 6D ) . However , the relatively high translational efficiency of protein-coding regions in genes containing uORFs suggests that scanning ribosomes often skip the uORF to allow efficient initiation at the protein-coding ORF . 10 . 7554/eLife . 08890 . 018Figure 6 . Features of ORFs encoded by protein coding genes . ( A ) Length distribution of peptides encoded by human protein coding genes . ( B ) Relative translation efficiency comparing non-canonical ORF vs . canonical ORF from the same gene . ( C ) Translation efficiency of canonical ORFs comparing genes with/without uORFs . ( D ) RNA expression level of genes with/without uORFs , measured by RNA-seq . ( E ) ATF4 encoded 3 uORFs and 1 overlapping uORF , whose translation efficiency is much higher than the canonical ORF . ( F ) Start codon types of uORFs showing differential relative expression levels to canonical ORFs . High: >three-fold higher than canonical ORFs . Low: >three-fold lower than canonical ORFs . DOI: http://dx . doi . org/10 . 7554/eLife . 08890 . 01810 . 7554/eLife . 08890 . 019Figure 6—figure supplement 1 . Example genes showing high translation of uORFs . ( A ) RELA ( B ) PTEN ( C ) DICER1 Enlarged figures show supporting read distribution in uORFs . DOI: http://dx . doi . org/10 . 7554/eLife . 08890 . 019 Interestingly , 1 , 1144 genes show >three-fold higher translational efficiency of the uORF than the corresponding protein-coding region , suggestive of translational regulation in a manner similar to Gcn4 ( Supplementary file 3 ) ( Hinnebusch , 2005 ) . These are enriched for 'transcription regulators' ( p<10-8; Fisher’s Exact Test; 237 genes are in the pathway 'regulation of transcription; GO:0045449' ) , particularly zinc finger transcription factors ( p<10-9; Fisher’s Exact Test ) , and protein kinases ( p<10-5; Fisher’s Exact Test; 45 genes are in the pathway 'protein kinase cascade GO:0007243' ) . Interestingly , many AP-1 transcription factors ( ATF4 , ATF5 , ATF2 , and JUN ) have high usage of uORFs , similar to the yeast homolog Gcn4 . For example , ATF4 contains 3 uORFs and 1 overlapping uORF , and the uORF expression is over 300-fold higher than the canonical ORF under normal growth conditions ( Figure 6E ) . However , under stress conditions , ATF4 efficiently re-initiates translation of the canonical ORF , thereby resulting in higher protein expression ( Rutkowski and Kaufman , 2003; Vattem and Wek , 2004 ) . Many other regulatory genes ( e . g . RELA , PTEN and DICER1 ) also show high uORF usage and suppressed translation of the canonical protein regions ( Figure 6—figure supplement 1 ) . The major determinant of uORF translation efficiency is its start codon as 84% of highly translated uORFs ( >three-fold higher than canonical ORFs ) use AUG as start codon , while only 32% of poorly translated uORFs ( >three-fold lower than canonical ORFs ) use AUG ( Figure 6F ) . In contrast to uORFs , the translation efficiency of dORFs is much lower ( 30-fold on average ) than the corresponding protein-coding region , indicating a very low level of translational reinitiation after the canonical stop codon ( Figure 6B ) . However , a small subset of dORFs are translated much more efficiently than the average dORF ( Supplementary file 3 ) . Using the analytical methods described above , we found that nucleotide sequences encoding uORFs and dORFs are more conserved than neighboring untranslated sequences , with 20% human uORF peptides , 46% of overlapping uORF peptides , and 32% of dORF peptides are conserved in mouse ( Figure 7A , Figure 7—figure supplement 1 , and Supplementary file 2 ) . Interestingly , these peptides have Ka/Ks ratios significantly lower than 1 , suggesting they may play functional roles ( Figure 7B , C , and Figure 7—figure supplement 2 ) . While uORFs clearly have an important role in inhibiting downstream expression of the canonical protein ( Morris and Geballe , 2000; Barbosa et al . , 2013 ) ( Figure 6D , E ) our results suggest that some of the encoded peptides are under stabilizing selection . 10 . 7554/eLife . 08890 . 020Figure 7 . Conservation of non-canonical peptides encoded by mRNAs . ( A ) Fraction of human mRNA peptides conserved in mouse . ( B ) Ka and Ks values of conserved mRNA peptides with Z-Test p-values shown . ( C ) Ka/Ks ratios of conserved mRNA peptides . DOI: http://dx . doi . org/10 . 7554/eLife . 08890 . 02010 . 7554/eLife . 08890 . 021Figure 7—figure supplement 1 . Conservation of nucleotides encoding uORF and dORF peptides . ( A , B ) PhastCon scores of nucleotides in uORFs ( A ) and dORFs ( B ) and their neighboring untranslated sequences of matching size and location ( See methods for detail ) were plotted . ( C , D ) PhyloCSF scores of nucleotides in uORFs ( C ) and dORFs ( D ) and their neighboring untranslated sequences of matching size and location were plotted . And scores of ORFs encoding peptides conserved in mouse and those with Ka/Ks < 0 . 5 were also shown . DOI: http://dx . doi . org/10 . 7554/eLife . 08890 . 02110 . 7554/eLife . 08890 . 022Figure 7—figure supplement 2 . Examples of conserved uORF peptides . DOI: http://dx . doi . org/10 . 7554/eLife . 08890 . 022 Although ribosome-profiling experiments indicate that lncRNAs and non-canonical ORFs in mRNAs can be translated ( Ingolia et al . , 2011; 2014; Aspden et al . , 2014; Bazzini et al . , 2014; Ruiz-Orera et al . , 2014 ) , previous methods to identify the translated products have been problematic . First , with one exception ( Bazzini et al . , 2014 ) , they did not use 3-nt periodicity to identify translated ORFs , but rather relied on the longest ORF length and/or maximum read density , which does not provide clear evidence for in-frame translation . Second , with one exception ( Ingolia et al . , 2014 ) , they did not filter the many reads that arise from non-ribosomal complexes and hence are irrelevant to identifying translated proteins . Third , we account for the variable distances between the 5’ end of the sequenced fragment and the ribosome A-site that arises due to imperfect RNase trimming of RPFs , and include RPFs with variable lengths into analyses for maximum sequencing read usage and codon coverage . Alignment of A-sites is critical for observing optimal 3-nt periodicity that characterizes translated regions . With this step , we can observe clear 3-nt periodicity for each codon in well expressed translated ORFs . Fourth , with approaches using harringtonine or lactimidomycin treatment to block translational elongation and hence map the translation initiation site ( Ingolia et al . , 2011; Lee et al . , 2012 ) , additional experiments are required , read peaks are often not precisely located at the start codons , and many genes do not show efficient ribosome pausing at start codons . In addition , the ribosome profiling datasets involving mammalian cells that have been analyzed to date were generated by a polyA tailing procedure that causes inaccuracies in determining the true 5’ end of the RPF . The RibORF algorithm combines ribosome A-site alignment , 3-nt periodicity and uniformity across codons ( PME approach ) to define regions of active translation . Using this approach , we identify a few thousand non-canonical peptides translated from lncRNAs , 5’ UTRs , and 3’UTRs , a dramatic increase over the several hundred previously reported ( Bazzini et al . , 2014 ) . In addition , we show that , although the vast majority of pseudogenes are not transcribed , 36% of expressed pseudogenes are translated into peptides . We believe that the RibORF approach represents a significant improvement over current methods , and it should be generally applicable . Approximately 40% of lncRNAs are translated into peptides >10 aa in length , with a median length of 43 aa . The distinction between translated and untranslated lncRNAs is strongly correlated with whether they are or not in the cytoplasm . Furthermore , the translation efficiency of lncRNAs is comparable to that of mRNAs , indicating that simple access of an RNA to the translation machinery in the cytoplasm is a major determinant of how well it is translated . In this regard , lncRNAs are transcribed by RNA polymerase II , capped , and polyadenylated , and hence are largely indistinguishable from mRNAs with respect to translation . It is unclear why some lncRNAs are predominantly nuclear , whereas others are predominantly cytoplasmic , but it seems unlikely that this is simply a matter of chance . It will be interesting to study whether nucleo-cytoplasmic localization of lncRNAs can be regulated during biological processes . However , the translatability of lncRNAs per se does not indicate whether the peptide is biologically important or even sufficiently stable to be detected . The observation that cytoplasmic lncRNAs are translated suggests that many of the resulting peptides are not themselves biologically functional . In this regard , the majority of lncRNA-encoded peptides are not conserved in mouse or other species . This lack of conservation does not exclude the possibility that these peptides are biologically meaningful , but it seems likely that most lncRNA peptides are not . If so , biological function may be mediated by the lncRNAs themselves or by the act of Pol II transcription , which alters chromatin structure to affect processes such as Pol III transcription ( Moqtaderi et al . , 2010; Oler and Cairns , 2010 ) and V ( D ) J recombination ( Matthews et al . , 2007 ) . However , the possibility remains that some or many lncRNAs represent transcriptional noise ( Struhl , 2007 ) . Although many , and perhaps most , of the various forms of non-canonical peptides are nonfunctional , our results strongly suggest that a minority of them is biologically functional . Some of these non-canonical peptides are conserved in mouse and as a class , these peptides have synonymous and nonsynonymous amino acid substitution rates indicating that are under stabilizing selection and strongly suggesting that they perform biological functions . Our results do not indicate that all of the conserved peptides are biologically functional , nor do they identify specific peptides as being functional . Most likely , functional peptides are those with the lowest Ka/Ks ratios , but these ratios need to be corrected for the number of substitutions analyzed for a given conserved region and the probability that they occur by chance . Nevertheless , a few functional lncRNA peptides have been described in other species ( Galindo et al . , 2007; Kondo et al . , 2010; Magny et al . , 2013; Pauli et al . , 2014 ) , and our results strongly suggest that a significant minority of non-canonical peptides have biological functions . Some previous analyses of lncRNAs tried to eliminate lncRNAs producing potentially functional peptides and removed those with long ORF length , high conservation and protein domains using various cutoffs ( Cabili et al . , 2011; Harrow et al . , 2012 ) . Here we found many lncRNAs after the filtering are translated , and conserved lncRNA and pseudogene peptides have median length 69 aa and 92 aa , respectively , which is shorter than the typical cutoff 100 aa . Our results indicate that ribosome profiling provides significant values to effectively identify translated RNAs in an unbiased manner , and reveal potential functional short peptides . By definition , non-coding RNAs are not translated into protein . However , until the advent of ribosome profiling that directly identifies translated regions of RNAs in an unbiased fashion , non-coding RNAs were defined computationally as lacking ORFs of significant length . Secondary bioinformatic considerations such as codon usage , evolutionary conservation , and protein domain have also been used as part of the definition of non-coding RNAs ( Cabili et al . , 2011; Harrow et al . , 2012 ) . Here we show that ~40% of so-called lncRNAs and pseudogene RNAs are translated in vivo , and hence are not truly non-coding RNAs . Of course , the translation of these RNAs provides no information on whether the resulting peptides are stable/detectable or biologically meaningful . Our results suggest that many , and perhaps nearly all , peptides generated from lncRNAs and pseudogene RNAs arise from the invariable translation of the cytoplasmic RNAs , thereby fortuitously generating peptides of no biological consequence . In such cases , any biological function of these RNAs would depend on the RNA product itself . However , the evolutionary conservation and low Ka/Ks ratios of some peptides generated by lncRNAs and pseudogene RNAs are suggestive that these peptides confer some biological function . An RNAs that generates a functional peptide may also have a biological function as an RNA molecule . Thus , non-coding RNAs can be divided into 3 classes , namely 1 ) true non-coding RNAs that are not translated , 2 ) RNAs that are translated into functionally irrelevant peptides , and 3 ) RNAs that are translated into non-conventional proteins that confer biological function . Furthermore , as the nucleo-cytoplasmic location of RNAs might be regulated by cell-type or environmental conditions , some RNAs that appear to be truly non-coding in our experiments might be translated and give rise to functional peptides in other circumstances . All cultures were performed at 37°C under 5% CO2 . BJ fibroblast cell lines ( EH , EL and ELR ) were cultured on Knockout DMEM ( Thermo Fisher Scientific , Waltham , MA ) with 10% FBS , medium 199 , glutamine and penicillin-streptomycin ( Hahn et al . , 1999 ) . The breast epithelial cell line ( MCF10A-ER-Src ) was grown in DMEM/F12 with 5% charcoal-stripped fetal bovine serum ( Thermo Fisher Scientific , Waltham , MA ) and supplements ( Iliopoulos et al . , 2009 ) . Cells were seeded at 1 × 106 cells per 10-cm culture dish and cultured overnight . MCF10A-ER-Src cells were treated by 1 µM 4-hydroxy-tamoxifen for various time points ( 1 , 4 , and 24 hr ) to induce transformation . Cells were pretreated with cycloheximide ( 100 µg/ml; Sigma-Aldrich , St . Louis , MO ) for 90 s or harringtonine ( 2 µg/ml; Santa Cruz , Santa Cruz , CA ) for 5 min , and detergent lysis was then performed with flash-freezing in liquid nitrogen . For ribosome profiling , DNase I-treated lysates were then treated with RNase I , and ribosome-protected fragments were purified for Illumina TruSeq library construction as previously described ( Ingolia et al . , 2012 ) . For RNA-seq , total RNA was purified from DNase-treated lysates , and ribosomal RNA was depleted with RiboMinus Eukaryote Kit ( Thermo Fisher Scientific , Waltham , MA ) . RNA-seq libraries were prepared with a tagging-based workflow ( Pease and Kinross , 2013 ) . In brief , rRNA-depleted RNA was fragmented at 85°C for 5 min , followed by cDNA synthesis , terminal tagging and PCR amplification with ScriptSeq v2 RNA-Seq Library Preparation Kit ( Epicentre , Madison , WI ) . Ribosome profiling and RNA-seq libraries were sequenced with Illumina HiSeq 2500 . We trimmed 3’ adapters from sequencing reads and then aligned the trimmed reads to human rRNA sequences and removed reads mapping to rRNAs ( 5S , 5 . 8S , 18S , and 28S ) . We then aligned remaining reads to the union of human reference transcript sequences: defined RefSeq; GENCODE lncRNAs; human body Map lncRNAs . The unmapped reads were then aligned to human reference genome sequence ( hg19 ) using Tophat with default parameters ( Trapnell et al . , 2009 ) . RNA-seq reads were mapped using the same steps as ribosome profiling reads . For analysis of subcellular location , RPKM values were calculated from published RNA-seq data from nuclear and cytosolic fractions of MCF7 cells ( Djebali et al . , 2012; ENCODE , 2012 ) . We required a transcript should have over 50 total RNA-seq reads for the calculation . MCF10A-ER-Src cells pretreated with 100 µg/ml cycloheximide for 90 s at 37°C were resuspended in 0 . 7 ml polyribosome lysis buffer [50 mM MOPS-NaOH at pH 7 . 4 , 150 mM NaCl , 15 mM MgCl2 , 0 . 5% Triton X-100 , 100 mg/ml cycloheximide , 7 µl protease inhibitor cocktail ( Cell Signaling Technology , Danvers , MA ) and 3 . 5 µl SUPERase·In ( Ambion , Thermo Fisher Scientific , Waltham , MA ) ] , passed once through a 26-G needle , and incubated at 4°C for 15 min with gentle rotation . Upon centrifugation , the cleared cell lysate was loaded onto a 1050% continuous sucrose gradient and centrifuged at 36 , 000 rpm for 165 min at 2°C with SW41-Ti rotor ( Beckman , Brea , CA ) . Fractions were assayed for RNA ( absorbance at 260 nM ) to determine the locations of the 40S and 60S subunits , 80S monoribosomes , and polyribosomes . RNA purified from these fractions was used to generate cDNA using a 1:1 combination of Oligo ( dT ) 20 and random hexamer and AffinityScript reverse transcriptase ( Agilent , Santa Clara , CA ) . The ribosome-associated amount of indicated RNA from each fraction was calculated by normalizing first to the 18S rRNA amount from that fraction and second to the indicated RNA amount from unfractionated sample loaded onto sucrose gradient . The translation efficiency of an ORF is calculated as the log2 ratio of the ribosome profiling RPKM value: RNA-seq RPKM value . We required the ORFs to have over 10 RNA-seq and ribosome profiling reads to permit a more accurate calculation , and we excluded ORF regions overlapping with other types of ORFS . Protein coding genes were defined by RefSeq database . Short noncoding RNAs were defined by RefSeq database as having length < 200 nt . Pseudogenes were defined by GENCODE and to not overlap with protein-coding genes . lncRNAs were defined by a union set of RefSeq , GENCODE or Human Body Map lncRNAs ( Cabili et al . , 2011; Harrow et al . , 2012 ) . We required a lncRNA to have introns or a length greater than 500 nts and that it does not overlap with any protein-coding gene or pseudogene in the same strand . An expressed lncRNA was defined as transcripts encoding peptides or showing significant RNA expression estimated from RNA-seq ( Cutoffs Benjamini-Hochberg corrected Poisson Test p<10-3 and >10 reads ) . For all types of transcripts , we identified all possible ORFs with a start codon AUG or close variants ( C/U/G ) UG and a stop codon . As the predicted translation probabilities are well correlated in the two cancer models ( Figure 2—figure supplement 1B ) , we combined ribosome-profiling reads in the breast epithelial and fibroblast cells to identify translated ORFs . We required expressed ORFs to have RPKM > 1 in at least one cell line model and over 10 reads . For each ORF , we define the total read number as N , and the encoded peptide length as L . We divide the ORF into smaller regions based on N and L in the following way . If N > L , we define a region length as 1 codon . Otherwise , a region length is defined as floor ( L/N ) . For each region i in an ORF , we calculated the fraction of reads in the region: P ( Xi ) = Ni/N , where Nirepresents number of reads in region i . We then calculate the PME value measuring the H ( X ) = ∑i=1n ( P ( Xi ) * log2P ( Xi ) ) uniformity of read distribution across regions as PME = H ( X ) /max ( H ) , where max ( H ) is the entropy value assuming the reads are perfectly evenly distributed across codons in an ORF . Read genomic locations were adjusted based on offset distance between 5’ end of fragment and A-site , based on parameters shown in Figure 1—figure supplement 1B . The adjusted read locations were used for ORF identification , expression level calculation and visualization . For the model training , we used as a positive set canonical ORFs from coding genes , and as a negative set off-frame ORFs in protein coding regions ( with start codon AUG and stop codons ) and candidate ORFs in short noncoding RNAs . We randomly picked 600 positive examples and 300 negative examples for training , and another 600 positive examples and 300 negative examples for testing . We included two features in the model , including ribosome footprinting 3-nt periodicity calculated as fraction of reads at 1st and 2nd nucleotides of codons in an ORF , and uniformity of read distribution measured by PME values described above . We used Support Vector Machine ( R package 'e1071' ) to build the classifier , with five-fold cross-validation and radial basis kernel . In some cases , we can identify overlapped positive ORFs for one transcript , with the same stop codon but multiple start codons . For these cases , we first picked AUG as start codons if present . We then chose 5’ most start codon as the representative one . But if there is no read between the picked one and the next downstream candidate , we chose the next one as the representative start codon . We used the receiver operating characteristic ( ROC ) curve to evaluate the performance of the RibORF classifier . The ROC curve is created by plotting the true positive rate against the false positive rate at various predicted p-value cutoffs from 0 to 1 . The Area Under the ROC Curve [AUC] value closer to 1 represents better performance of the classifier . As in Figure 2A , we used different training parameters to build the classifier , and the AUC values measuring classifier performances were plotted . We examined whether translated non-canonical ORFs are more conserved and have higher coding potential than untranslated sequences in the same RNAs using PhastCon scores based on multiz alignment of 46 vertebrates ( Siepel et al . , 2005 ) and PhyloCSF scores based on 29-mammal alignment ( Lin et al , 2011 ) , respectively . The PhastCon conservation level and PhyloCSF coding potential of nucleotides in a region were calculated as the average scores across nucleotides . As in Figure 4—figure supplements 2 and 3 , for each translated ORF in lncRNAs and pseudogenes , we randomly picked 50 untranslated segments with the same length . As translated ORFs tend to be located in 5’ end of transcripts , the untranslated segments located in the 5’ end are twice more likely to be picked than the 3’ end ones . However , as we did not observe 5’ end of lncRNAs and pseudogenes are significantly more conserved than 3’ end in untranslated regions , the patterns should be consistent if we do not consider the location bias . As in Figure 7—figure supplement 1 , for translated uORFs and dORFs , we compare their conservation and coding potential levels with their neighboring untranslated regions . If the translated ORF length is L , the neighboring untranslated regions were defined as L/2 region upstream the ORF and L/2 region downstream . We excluded the translated ORFs which are located within L/2 regions of canonical ORFs . We used Liftover ( Karolchik et al . , 2014 ) to identify orthologous genomic locations of human lncRNA ORFs in mouse , and obtained possible ORFs flanking these regions , considering all coding and noncoding transcripts in mouse genome defined by refSeq and GENCODE ( Harrow et al . , 2012 ) . Then we used BLASTP ( Johnson et al . , 2008 ) to compare the similarity between human and mouse ORF peptide sequences . To obtain the expected distribution of BLASTP E-values between non-conserved peptide sequences ( Figure 4—figure supplements 4 and 5 ) , we randomized the nucleotide sequence of each human translated ORF for 50 times and use the BLASTP to compare the human ORF peptide sequence and the randomized the sequence . We consider a human ORF to be conserved in mouse if the two ORFs have a BLASTP alignment E-value <10-4 . Using this cutoff , the False Discovery Rate ( FDR ) is <0 . 0005 for all types and lengths of non-canonical ORFs ( Figure 4—figure supplements 4 and 5 ) . The entire nucleotide and encoded peptide sequences of non-canonical peptides conserved between human and mouse were analyzed by KaKs calculator software to examine nonsynonymous ( Ka ) and synonymous ( Ks ) substitutions and the resulting and Ka/Ks values , using the approximate method 'NG' ( Wang et al . , 2010 ) . As a control to exclude the possibility that low Ka/Ks ratios are an artifact of the our cutoff BLASTP E-value <10-4 , we calculated the Ka/Ks ratios of a given human peptide with 50 randomly generated sequences of the same length as the homologous mouse ORF , and with BLASTP alignment E-value <10-4 . We input the peptide sequences encoded by translated ORFs to the Pfam web server ( http://pfam . xfam . org/search#tabview=tab1 ) . We included both Pfam-A and Pfam-B in the analyses , and used the default cutoff E-value <1 . Gene ontology analyses were done using DAVID database ( Huang et al . , 2009 ) . Unless otherwise stated , p-values were calculated by the Wilcoxon Rank Sum Test . RibORF pipeline is available at http://www . broadinstitute . org/~zheji/software/RibORF . html

Our genes encode the instructions needed to make proteins . When a gene is switched on , it’s DNA is used as a template to make molecules of messenger ribonucleic acid ( RNA ) . These RNAs are then “translated” into proteins by large cell machines called ribosomes . Within the messenger RNA , a long region called an “open reading frame” is the section that encodes the protein . The human genome also contains a vast amount of DNA that is not part of any gene . Cells can produce molecules of RNA from this DNA ( so-called “non-coding RNAs” ) , but these RNAs are not thought to code for proteins because they lack long open reading frames . Non-coding RNAs can also be made from sections of DNA called “pseudogenes” , which have lost their ability to code for proteins over the course of evolution . Furthermore , messenger RNAs also contain short open reading frames in the “untranslated” regions that flank the protein-coding region . The extent to which cells translate non-coding RNAs to produce small proteins ( or peptides ) is not known . “Ribosome profiling” is a powerful method to determine which RNAs are translated , but it is not always possible to distinguish between the RNAs that are genuinely translated and those that just happen to be bound to ribosomes . Ji et al . overcome these limitations by developing a new computational method to analyse data from ribosome profiling . The experiments show that thousands of non-coding RNAs in the human genome are , in fact , translated . This is many more than anticipated and represents approximately 40% of the lncRNAs and pseudogene RNAs , and 35% of untranslated regions in messenger RNAs . Ji et al . also found that a small group of all the lncRNA peptides in the human genome appear to have changed little over the course of evolution , which strongly suggests that they have specific roles in cells . The next challenge is to find out what roles the peptides encoded by these lncRNAs play in cells .

2015

Many lncRNAs, 5’UTRs, and pseudogenes are translated and some are likely to express functional proteins

We identified the neurons comprising the Drosophila mushroom body ( MB ) , an associative center in invertebrate brains , and provide a comprehensive map describing their potential connections . Each of the 21 MB output neuron ( MBON ) types elaborates segregated dendritic arbors along the parallel axons of ∼2000 Kenyon cells , forming 15 compartments that collectively tile the MB lobes . MBON axons project to five discrete neuropils outside of the MB and three MBON types form a feedforward network in the lobes . Each of the 20 dopaminergic neuron ( DAN ) types projects axons to one , or at most two , of the MBON compartments . Convergence of DAN axons on compartmentalized Kenyon cell–MBON synapses creates a highly ordered unit that can support learning to impose valence on sensory representations . The elucidation of the complement of neurons of the MB provides a comprehensive anatomical substrate from which one can infer a functional logic of associative olfactory learning and memory . Neural representations of the sensory world give rise to appropriate innate or learned behavioral responses . Innate behaviors are observed in naïve animals without prior learning or experience , suggesting that they are mediated by genetically determined neural circuits . Responses to most sensory stimuli , however , are not innate but experience-dependent , allowing an organism to respond appropriately in a variable and uncertain world . Thus , most sensory cues acquire behavioral relevance through learning . In Drosophila melanogaster , a number of different forms of learning have been observed in response to sensory stimuli ( Siegel and Hall , 1979; Liu et al . , 1999 , 2006; Masek and Scott , 2010; Schnaitmann et al . , 2010; Ofstad et al . , 2011; Vogt et al . , 2014 ) . In associative olfactory learning , exposure to an odor ( conditioned stimulus , CS ) in association with an unconditioned stimulus ( US ) results in appetitive or aversive memory ( Quinn et al . , 1974; Tempel et al . , 1983; Tully and Quinn , 1985 ) . Olfactory memory formation and retrieval in insects require the mushroom body ( MB ) ( Heisenberg et al . , 1985; de Belle and Heisenberg , 1994 , Dubnau et al . , 2001; McGuire et al . , 2001 ) , an associative center in the protocerebrum ( Figure 1 and Video 1 ) . 10 . 7554/eLife . 04577 . 003Figure 1 . Anatomy of olfactory pathways in the adult fly brain . ( A ) An image of the adult female brain showing the antennal lobes ( AL ) and subregions of the mushroom bodies ( MB; see panel B for more detail ) . The image was generated using a 3D image rendering software ( FluoRender ) ( Wan et al . , 2009; Wan et al . , 2012 ) . The 51 glomeruli of the AL extend projection neurons ( PN ) to the calyx of the MB and the lateral horn ( LH ) . There are a total of ∼200 PN; 6 from the DL3 glomerulus are shown . See Video 1 for an introduction to olfactory circuit . ( B ) Subregions within the MB . The γ lobe , calyx , and pedunculus ( ped ) are displayed separately from other lobes; their normal positions are as shown in panel A and Video 1 . Color-coding is as in panel A . See below for a detailed description of dorsal accessory calyx ( dAC ) and ventral accessory calyx ( vAC ) . ( C ) A schematic representation of the key cellular components and information flow during processing of olfactory inputs to the MB ( see text for references and more details ) . Olfactory receptor neurons expressing the same odorant receptor converge onto a single glomerulus in the AL . A small number ( generally 3–4 ) of PNs from each of the 51 AL glomeruli innervate the MB calyx where they synapse on the dendrites of the ∼2000 Kenyon cells ( KCs ) in a globular structure , the calyx . Each KC exhibits , on average , 6 . 4 dendritic ‘claws’ ( Butcher et al . , 2012 ) , and each claw is innervated by a single PN . There is little order in connection patterns of PNs to KCs . The axons of the KCs project in parallel anteriorly through the pedunculus to the lobes , where KCs terminate onto the dendrites of MB output neurons ( MBONs ) . KCs can be categorized into three major classes α/β , α′/β′ , and γ based on their projection patterns in the lobes ( Crittenden et al . , 1998 ) . The β , β′ , and γ lobes constitute the medial lobes ( also known as horizontal lobes ) , while the α and α′ lobes constitute the vertical lobes . These lobes are separately wrapped by ensheathing glia ( Awasaki et al . , 2008 ) . The α/β and α′/β′ neurons bifurcate at the anterior end of the pedunculus and project to both the medial and vertical lobes ( Lee et al . , 1999 ) . The γ neurons project only to the medial lobe . Dendrites of MBONs and terminals of modulatory dopaminergic neurons ( DANs ) intersect the longitudinal axis of the KC axon bundle , forming 15 subdomains , five each in the α/β , α′/β′ , and γ lobes ( numbered α1 , α2 , and α3 for subdomains in the α lobe from proximal to distal ) ( Tanaka et al . , 2008 ) . Additionally , one MBON and one DAN innervate the core of the distal pedunculus intersecting the α/β KCs ( pedc , see below ) . There are seven types of KCs; five of the seven types have their dendrites in the main calyx , while those of the γd cells form the vAC ( Aso et al . , 2009; Butcher et al . , 2012 ) and those of the α/βp cells the dAC ( Tanaka et al . , 2008 ) . The accessory calyces are thought to receive non-olfactory input since they do not receive input from the PNs from the AL ( Tanaka et al . , 2008 ) . Different KCs occupy distinct layers in the lobes as indicated ( p: posterior; c: core; s: surface; m: medial; a: anterior; and d: dorsal ) . Some MB extrinsic neurons extend processes to only a specific layer within a subdomain , defining elemental subdivisions in the lobes , or ‘synaptic units’ as proposed by Tanaka et al . ( 2008 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04577 . 00310 . 7554/eLife . 04577 . 004Video 1 . Introduction to MB anatomy and the olfactory circuit . DOI: http://dx . doi . org/10 . 7554/eLife . 04577 . 004 Olfactory perception in the fly is initiated by the binding of an odorant to an ensemble of olfactory sensory neurons in the antennae , resulting in the activation of a distinct and topographically fixed combination of glomeruli in the antennal lobe ( Figure 1A , B; reviewed in Vosshall and Stocker ( 2007 ) ; Masse et al . ( 2009 ) ) . Most antennal lobe projection neurons ( PNs ) extend dendrites to a single glomerulus and project axons that bifurcate to innervate two brain regions , the lateral horn and the MB ( Stocker et al . , 1990; Wong et al . , 2002; Jefferis et al . , 2007 ) . The invariant circuitry of the lateral horn is thought to mediate innate behaviors , whereas the MB translates olfactory sensory information into learned behavioral responses ( Heisenberg et al . , 1985 ) . The PN axons synapse onto the dendrites of the Kenyon cells ( KCs ) in the MB calyx; the parallel axons of the KCs form the MB lobes . Odors activate sparse subpopulations of KCs distributed across the MB without spatial preference ( Turner et al . , 2008; Honegger et al . , 2011; Campbell et al . , 2013 ) . Anatomical and physiological studies reveal that each KC receives on average 6 . 4 inputs from a random combination of glomeruli; that is , knowledge of a single input to a KC provides no information about the identity of the additional inputs , and connections differ in different flies ( Murthy et al . , 2008; Caron et al . , 2013; Gruntman and Turner , 2013 ) . Thus , the calyx of the MB discards the highly ordered structure of the antennal lobe . A restoration of order must therefore be imposed downstream to link the KC representation to an appropriate behavioral output . Three classes of KCs extend parallel fibers that form the γ , α′/β′ , and α/β lobes of the MB , where they form synapses with a relatively small number of MB output neurons ( MBONs; Figure 1C ) ( Crittenden et al . , 1998; Ito et al . , 1998; Strausfeld et al . , 2003; Lin et al . , 2007; Tanaka et al . , 2008; Busch et al . , 2009 ) . The MBONs have dendrites in the MB lobes and project axons to neuropils outside of the MB . Modulatory input neurons , including dopaminergic neurons ( DANs ) and octopaminergic neurons ( Nassel and Elekes , 1992; Tanaka et al . , 2008; Busch et al . , 2009; Mao and Davis , 2009 ) , also innervate the MB lobes . The MBONs and DANs send their processes to stereotyped locations , defining spatially restricted ‘subdomains’ in each lobe ( Ito et al . , 1998; Tanaka et al . , 2008; Mao and Davis , 2009; Pech et al . , 2013 ) . However , these studies did not establish the precise anatomical relationships between the subdomains; knowledge of these relationships will be required to understand the structure and logic of MB circuits . The DANs are the most prevalent modulatory neurons in the MB and dopamine is thought to act locally to modify KC–MBON synapses ( Aso et al . , 2010; Waddell , 2013 ) . In accord with this model , DAN activity is required during learning ( Schwaerzel et al . , 2003; Aso et al . , 2010 , 2012; Burke et al . , 2012; Liu et al . , 2012 ) and exogenous activation of DAN subpopulations can serve as an US in associative learning paradigms ( Schroll et al . , 2006; Claridge-Chang et al . , 2009; Aso et al . , 2010 , 2012; Burke et al . , 2012; Liu et al . , 2012 ) . In addition , D1-like dopamine receptors in the KCs are necessary to form olfactory memories ( Kim et al . , 2007 ) . Different populations of DANs are activated by USs of different valence; see Figure 1A of the accompanying paper ( Aso et al . , 2014 ) for summary ( Riemensperger et al . , 2005; Mao and Davis , 2009; Liu et al . , 2012; Das et al . , 2014 ) . Genetic manipulation has also implicated specific subsets of MBONs in the mediation of learned appetitive and aversive behaviors ( Sejourne et al . , 2011; Pai et al . , 2013; Placais et al . , 2013; Aso et al . , 2014 ) . These experiments implicate the DANs as the source of the learning cue and the MBONs as the mediators of behavioral output . The elucidation of the connections between KCs , DANs , and MBONs should provide insight into a problem shared by invertebrate and vertebrate nervous systems: how is meaning imposed on an unstructured ensemble of neurons and how is imposed valence translated into an appropriate behavioral response ? In this study , we developed new genetic reagents and used them to identify the cell types and projections of the neurons comprising the MB lobes . These data provide insight into the potential connections in the MB and suggest how the MB may mediate learned behaviors . We found that the MB lobes are composed of ∼2200 neurons that include 7 KC , 21 MBON , and 20 DAN cell types . The MBONs of a given type exhibit spatially stereotyped dendritic arbors in the MB lobes that form 15 compartments that collectively tile the lobes . Each DAN cell type projects axons to one or at most two of the compartments defined by the MBONs . The alignment of DAN axons with compartmentalized KC–MBON synapses creates an isolated unit for learning that can transform the disordered KC representation into ordered MBON output . The MBON axons project to five discrete neuropils outside of the MB , providing loci for the convergence of all the information necessary for learned associative responses . The elucidation of the full complement of MB neurons and the details of their projections provide an anatomical substrate from which we can infer a functional logic of olfactory learning and memory . Each MB contains ∼2000 KCs that are sequentially generated from four neuroblasts ( Ito et al . , 1997; Lee et al . , 1999; Zhu et al . , 2003; Lin et al . , 2007 ) . The dendrites of the KCs form the MB calyx and their parallel axons form the three MB lobes ( Figure 1 ) ( Crittenden et al . , 1998 ) . The main calyx primarily receives olfactory input from the antennal lobe , whereas the smaller ventral and dorsal accessory calyces are thought to receive non-olfactory input ( see Figure 1C ) ( Tanaka et al . , 2008; Butcher et al . , 2012 ) . The KCs have been divided into three classes , γ , α′/β′ , and α/β , with each class projecting axons to the eponymous lobe ( Crittenden et al . , 1998; Lee et al . , 1999 ) . The split-GAL4 screen and the analysis of the axonal projection patterns of single cells revealed that these three classes of KCs divide into seven cell types ( Figure 3A and Video 2 ) . Each of the four neuroblasts contributes to each of the seven cell types and the dendrites of the KCs generated from the different neuroblasts remain segregated in the main calyx ( Lin et al . , 2007 ) . The parallel axon fibers of each of the seven types of KCs occupy specific layers within the γ , α′/β′ , and α/β lobes ( Figures 6 and 7 ) . Two KC types divide the γ lobe into the main and dorsal ( d ) layers , two types divide the α′/β′ lobe into the middle ( m ) and anterior–posterior ( ap ) layers , and three KC types divide the α/β lobe into the posterior ( p ) , core ( c ) , and surface ( s ) layers ( Figures 6 and 7 , also see Figure 1C ) . Examination of single cell morphologies suggests that each KC may form en passant synapses with target MBONs along the length of its axon , providing each MBON with access to a large number of KC inputs . Five of the seven types of KCs elaborate their dendrites in the main calyx , whereas two types of KCs ( γd and α/βp ) have dendrites exclusively in the ventral and dorsal accessory calyces , respectively ( Figures 6A and 7A ) ( Lin et al . , 2007; Tanaka et al . , 2008; Butcher et al . , 2012 ) . 10 . 7554/eLife . 04577 . 013Figure 6 . KCs of the γ and α′/β′ lobes . Seven KC types identified by split-GAL4 . Representative aligned images of each KC type ( see also Figure 7 ) are shown; the name of the cell type and the approximate number of cells per brain hemisphere are indicated . Three examples of single cell morphologies , segmented from multicolor flip-out experiments , are presented for each cell type illustrating the branching pattern of the KC axons as they project through the lobes . In the single cell images , the cell body , primary neurite , dendrites , and axon in the pedunculus have been false-colored green and axons in the lobes are magenta . Based on co-expression in specific split-GAL4 lines and single cell morphologies , we have divided the KC population into 7 cell types . The number of cells of each type was estimated by counting labeled nuclei in split-GAL4 lines ( see ‘Materials and methods’ ) and is shown in parentheses . ( A ) The γd KCs are thought to be of embryonic origin , because they are not included in a clonal analysis that visualized all post-embryonic KCs ( Lin et al . , 2007; Yu et al . , 2013 ) , and they have morphological similarity to the embryonic born KCs of basal cockroaches ( Farris and Strausfeld , 2003 ) . Their dendritic arbors form a protrusion extending ventral lateral to the main calyx , which we named the ventral accessory calyx ( vAC ) . Their axons occupy the most peripheral layer in the pedunculus , the ventral and anterior layers in the γ1–γ4 compartments , and the dorsal layer in γ5 . Single cell morphologies ( from MB028B and MB355B ) reveal that the γd axons have more branches in γ3–γ5 than in γ1 and γ2 . The vAC may be devoted to non-olfactory inputs; the major types of olfactory projection neurons from the antennal lobe do not innervate this structure ( Butcher et al . , 2012 ) . ( B ) The γ main KCs have their dendrites in the main calyx and their axons occupy about 75% of the volume of the γ lobes . Single cell morphologies ( from MB369B and MB355B ) reveal that each γ main KC branches in all the five compartments of the γ lobe . ( C ) The α′/β′ap KCs have dendrites in the main calyx and project axons to the anterior and posterior layers of the α′/β′ lobes . Single cell morphologies ( from MB461B and MB463B ) reveal that axonal branches from single KCs project to both β′2a and β′2p , where they overlap with distinct sets of output and dopaminergic neurons ( see below ) . ( D ) The α′/β′m KCs have dendrites in the main calyx . In β′2 , their axons are located in the area between the bifurcating axons of α′/β′ap neurons shown in panel C . In the α′ lobe , their axons are medial to those of the α′/β′ap cells . Single cell images are from MB418B and MB369B . DOI: http://dx . doi . org/10 . 7554/eLife . 04577 . 01310 . 7554/eLife . 04577 . 014Figure 7 . KCs of the α/β lobes . ( A ) The dendrites of the α/β posterior ( α/βp ) KCs form a protrusion extending to the dorsal lateral side of the main calyx . This structure has been called the accessory calyx , but we have renamed it as the dorsal accessory calyx ( dAC ) to distinguish it from the ventral AC ( vAC ) ( Figures 1C and 6A ) . The α/βp KC axons project to the posterior layer of the α/β lobe . These are the firstborn α/β KCs and are also known as pioneer α/β KCs ( Lin et al . , 2007 ) . The single cell images were segmented from multicolor flp-out ( MCFO ) brains of MB469B and MB371B . ( B ) The α/β surface ( α/βs ) KCs have dendrites in the main calyx and project axons to the surface layer of the α/β lobes where they form a continuous layer surrounding the α/β core KCs shown in ( C ) . Single cell morphologies of cells ( from MB185B ) reveal that the α/βs KCs have relatively smooth axonal projections in the lobes . ( C ) The α/β core ( α/βc ) KCs have dendrites in the main calyx . They are the last born KCs and their axons occupy the core of the pedunculus and the α/β lobes . They can be morphologically subdivided into inner and outer core cells ( Tanaka et al . , 2008 ) , although the border between the inner and outer core is not well defined and we were unable to make a split-GAL4 driver line that labels only the outer core cells . Single cell morphologies ( from MB594B ) reveal that the axons of the α/βc cells have the fewest branches of the 7 types of KCs . DOI: http://dx . doi . org/10 . 7554/eLife . 04577 . 014 The five KC types ( γmain , α′/β′ap , α′/β′m , α/βc , and α/βs ) that receive olfactory information are each represented by hundreds of neurons per hemisphere and have their dendrites in the main calyx . Each KC cell type sends axonal projections to a spatially segregated layer in the lobes . The dendritic arbors of each KC type also tend to be found in the same regions of the calyx ( Lin et al . , 2007; Leiss et al . , 2009 ) , but those dendritic zones are largely overlapping and individual KCs within a given cell type exhibit variable dendritic projection patterns ( Figures 6 and 7 ) . Moreover , the KCs receive input from an apparently random collection of glomeruli ( Murthy et al . , 2008; Caron et al . , 2013; Gruntman and Turner , 2013 ) . These features are in sharp contrast to most neuronal cell types in the olfactory pathway of the fly that are thought to consist of one to ten neurons that exhibit stereotyped projections ( Yu et al . , 2010 ) , suggesting that their input and output connections are genetically predetermined . These observations suggest a unique function of the KCs in the processing of olfactory information ( see ‘Discussion’ ) . The MBONs extend dendrites that overlap with the KC axons in the MB lobes and project axons outside the MB . By determining the polarity of each cell type using high-resolution confocal imaging along with an analysis of the expression of the presynaptic reporter synaptotagmin-smGFP-HA ( Syt::smGFP-HA; Figure 8 ) , we identified 34 MBONs that comprise 21 different cell types ( Table 1 , Figure 3B , Video 3 ) . We employed immunostaining to identify MBON types as either cholinergic , glutamatergic , or GABAergic ( Figure 9 , Table 1 ) . MBONs that use the same neurotransmitter extend dendrites to adjacent regions of the lobes; cholinergic MBONs in the vertical ( α and α′ ) lobes , glutamatergic MBONs in the medial ( β , β′ , and γ ) lobes , and GABAergic MBONs in an area of the lobes at the intersection between these two regions ( Figure 9 and Video 5 ) . 10 . 7554/eLife . 04577 . 024Figure 8 . Identification of MBONs and visualization of their single cell morphologies . Each MBON is named according to the compartment ( s ) in the MB lobes where its dendrites arborize ( see Table 1 and below ) . For example , MBON-γ2α′1 neurons exhibit dendritic arbors in γ2 and α′1 compartments . ( A ) The projection patterns of MBON-γ2α′1 neurons . Maximum intensity projection confocal images of MB077B driven expression in one brain hemisphere are shown . Visualization with pJFRC225-5XUAS-IVS-myr::smGFP-FLAG in VK00005 labels two MBON-γ2α′1 neurons per hemisphere . ( B ) Labeling of the two MBON-γ2α′1 neurons in different colors using multicolor flp-out ( MCFO; Nern et al . , in preparation; see ‘Materials and methods’ ) . The arbors of the two neurons overlap and are indistinguishable , thus these two cells represent a single cell type . Arrowheads indicate cell bodies . ( C ) MB077B driven expression of a membrane targeted epitope ( green; pJFRC225-5XUAS-IVS-myr::smGFP-FLAG in VK00005 ) and a presynaptically targeted epitope ( magenta; pJFRC51-3XUAS-IVS-Syt::smGFP-HA in su ( Hw ) attP1 ) . The fine processes in the MB lobes are typical of dendrites , whereas the processes of this neuron that are outside the lobes end with varicosities containing the presynaptic marker . ( D ) The morphologies of MBON-γ3 and MBON-γ3β′1 as identified by MB083C driven expression . Using the pJFRC225 membrane-targeted reporter , two neurons innervating the γ3 and β′1 MB compartments in each hemisphere can be seen . Dashed vertical line shows the position of the mid-line . ( E ) Using MCFO , the two cells in one hemisphere were labeled in different colors . Both cells have dendrites in γ3 in both hemispheres but axonal terminals in just the contralateral brain hemisphere . However , one of them ( the green cell ) also has dendrites in the contralateral β′1 compartment , demonstrating that these two cells represent different cell types , which we named MBON-γ3 and MBON-γ3β′1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04577 . 02410 . 7554/eLife . 04577 . 025Figure 9 . Neurotransmitters used by MBONs . The putative neurotransmitters used by MBON cell types were assigned by assessing the immunoreactivity of their axon terminals to antibodies raised against Drosophila vesicular glutamate transporter ( dVGluT ) , Drosophila glutamate decarboxylase 1 ( GAD1 ) , and Drosophila choline acetyl transferase ( ChAT ) ( see ‘Materials and methods’ ) . Single confocal optical sections; axon terminals of different MBONs are shown in green and antibody staining in magenta . ( A ) Axon terminals of MBON-γ5β′2a ( MB210B ) were labeled with anti-dVGluT ( arrowhead ) but not with either anti-GAD1 or anti-ChAT , suggesting that this cell type is glutamatergic . ( B ) Axon terminals of MBON-γ1pedc>α/β ( MB112C ) were labeled with anti-GAD1 ( arrowhead ) , suggesting that this cell type is GABAergic . ( C ) Axon terminals of MBON-γ2a′1 ( MB077B ) were labeled with anti-ChAT ( arrowhead ) , suggesting this cell type is cholinergic . ( D–F ) MBONs of the same neurotransmitter type were given the same color and are displayed together . ( D ) Seven types of glutamatergic MBONs . ( E ) Four types of GABAergic MBONs . ( F ) Eight types of cholinergic MBONs . The neurotransmitter of the two MBON types found in the VT-GAL4 collection was not determined ( but see Figure 5—figure supplement 1 ) . Video 5 illustrates the relative positions of these neurons in the standard brain . DOI: http://dx . doi . org/10 . 7554/eLife . 04577 . 02510 . 7554/eLife . 04577 . 026Video 5 . Arrangement of MBONs by transmitter type . MBONs was color-coded based on putative neurotransmitter as in Figure 9: green , glutamatergic; blue , GABAergic; red , cholinergic . DOI: http://dx . doi . org/10 . 7554/eLife . 04577 . 026 Fourteen MBON cell types consist of only one cell per hemisphere , six types contain two cells , and one type eight cells per hemisphere . In split-GAL4 lines with expression in more than one neuron , single-cell resolution was achieved by using the multicolor flp-out strategy ( MCFO; Nern et al . , in preparation , Figure 8 ) . Single cell analysis revealed that each member of an MBON type exhibits indistinguishable morphology as assessed by light microscopy , and these stereotyped projection patterns are invariant across flies ( see below for all cell types ) . The 21 MBON types elaborate dendritic arbors in insular , segregated domains of the lobes that we call compartments . MBON dendritic arbors within each compartment exhibit little , if any , overlap with arbors in neighboring compartments ( Figure 10 ) . Computational alignment of the dendritic arbors of each of the MBON types within a single reference brain revealed that these compartments collectively tile the MB lobes with minimal overlap ( Figure 10G , I , K ) . The alignment reveals gaps between arbors at four compartment borders; staining of the MB lobes for the presynaptic marker Bruchpilot ( Figure 10—figure supplement 1 ) suggests that these gaps represent areas of reduced synaptic density . Two-color labeling experiments confirmed that the dendritic arbors of different MBONs are segregated in spatially stereotyped compartments ( Figure 11A–C ) . We observed ensheathing glia at the borders between the MB lobes but not between the MBON compartments in each lobe ( Figure 11J–L ) . 10 . 7554/eLife . 04577 . 032Figure 10 . Compartmentalization of the MB lobes . ( A–F ) Tiling of MBON dendrites in the γ lobe . ( A ) A registered image of a brain hemisphere showing 5 MBON cell types that innervate contiguous compartments of the γ lobe . ( B–F ) Confocal images of brains showing expression in the MBONs shown in A; pJFRC225-5xUAS-IVS-myr::smGFP-FLAG in VK00005 and the following split-GAL4 lines were used to generate the images: B , MB210B; C , MB298B; D , MB083C; E , MB077B; and F , MB112C . The cell types are indicated in the panels with the number of neurons of each type in parenthesis; the neurons are shown in white and the nc82 reference stain in orange . ( G–L ) The dendrites of MBONs and the axon terminals of DANs tile the MB lobes , defining 15 compartments . Dendrites of MBONs ( G , I , and K ) and axon terminals of DANs ( H , J , and L ) , aligned to the standard brain , are shown for each lobe . The same false colors were assigned to the DANs and the MBONs of the same compartment . The arrows in ( G , I , and K ) show the four compartment borders where gaps were routinely seen between the MBON dendrites in adjacent compartments; these gaps correspond to areas of reduced synaptic density ( see Figure 10—figure supplement 1 ) . Note that the anterior layer of the β′2 compartment contains dendrites of MBON-γ5β′2a ( I , yellow ) and axon terminals of PAM-β2β′2a ( J , yellow ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04577 . 03210 . 7554/eLife . 04577 . 033Figure 10—figure supplement 1 . Lower density of presynaptic sites at the border between compartments . Comparison of the KC membrane and presynaptic labeling in the γ lobe . ( A ) Membrane-labeled γd and γmain KCs ( MB131B , pJFRC225-5xUAS-IVS-myr::smGFP-FLAG in VK00005 ) ; a substack projection of the γ lobe is shown . ( B ) Presynaptic sites within the γ lobe ( nc82; magenta ) ; nc82 staining outside the γ lobe has been eliminated for clarity . Arrows indicate borders between compartments of the γ lobe where synaptic density is low . ( C–E ) Single confocal slice at the border between the γ5 and γ4 subdomains showing KCs ( C ) , nc82 staining ( D ) , and a merged image ( E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04577 . 03310 . 7554/eLife . 04577 . 034Figure 11 . Two-color labeling experiments demonstrating compartmentalization of the MB lobes . ( A–E ) Two-color labeling of MBONs ( A–D ) or DANs ( E ) from adjacent compartments , or in the case of ( D ) , subdivisions of the same compartment . Neurons were visualized by split-GAL4 and LexA drivers in combination with pJFRC200-10XUAS-IVS-myr::smGFP-HA in attP18 and pJFRC216-13XLexAop2-IVS-myr::smGFP-V5 in su ( Hw ) attP8 , respectively . Substack projections of the compartments are shown . Clear segregation was observed between dendrites of MBONs or axon terminals of DANs from neighboring compartments . ( F ) MCFO labeling of a single brain showing the termini in the MB lobes of three types of PPL1 cluster DANs . The image was generated from line MB060B , which expresses in four types of PPL1 neurons . We were able to confirm all 11 compartment borders tested using either two-color labeling or MCFO experiments; we did not have the required genetic lines to test one of the 12 compartment borders ( between β1 and β2 ) . ( G–I ) Two-color labeling of DANs and MBONs from the same compartment . Single confocal slices are shown . The DANs and MBONs coextend , with each densely arborizing in the entire compartment . ( J–L ) Two-color labeling of ensheathing glia and MBONs . Whereas each of the three lobes ( i . e . , γ , α′/β′ , and α/β ) is separated clearly by ensheathing glia , we did not observe glia between MBON compartments within each lobe . Single confocal slices are shown . Arrows indicate the ensheathing glia separating the axon bundles of the γ , α′/β′ , and α/β neurons in the lobes and pedunculus . The driver lines used are as follows: ( A ) MB310C ( magenta ) , R34B02-LexA ( green ) ; ( B ) MB083C ( magenta ) , R25D01-LexA ( green ) ; ( C ) MB083C ( magenta ) , R25D01-LexA ( green ) ; ( D ) MB062C ( magenta ) , R34B02-LexA ( green ) ; ( E ) MB316C ( magenta ) , R48B03-LexA ( green ) ; ( G ) MB312C ( magenta ) , R53C03-LexA ( green ) ; ( H ) MB083C ( magenta ) , R48B03-LexA ( green ) ; ( I ) MB058B ( magenta ) , R34B02-LexA ( green ) ; ( J ) MB434B ( green ) , R16D08-LexA ( magenta ) ; ( K ) MB112C ( green ) , R16D08-LexA ( magenta ) ; ( L ) MB112C ( green ) , R16D08-LexA ( magenta ) . Scale bar in ( E ) applies to all panels except ( F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04577 . 034 The MB lobes are divided into 15 distinct compartments containing the segregated dendritic arbors of one or a small number of MBONs ( Figure 1C , Figure 12 and Figure 13 ) . These compartments tile the MB lobes , revealing a general organizational principle of the MB output . This organization is in accord with an earlier proposal by Tanaka et al . ( 2008 ) that each of the γ , α′/β′ , and α/β lobes is divided into five domains . 13 of the 21 MBON types extend dendrites to a single compartment , and 8 MBON types project to two compartments ( Figure 12A–C ) . Most of the MBON types innervate KCs from each of the layers within a compartment , but eight types restrict their dendritic arbors to specific layers ( Figure 11D , 12A , C , D ) . 10 . 7554/eLife . 04577 . 035Figure 12 . The arborization patterns of individual cell types within the MB lobes . ( A ) A matrix summarizing the projection patterns in the MB lobes of the KC axons , the MBONs , the DANs , and other modulatory neurons . The 15 lobe compartments and the core of distal pedunculus ( pedc ) are separated with thick vertical lines , and each compartment is further divided into columns representing ‘synaptic units’ ( Tanaka et al . , 2008 ) , elemental subdivisions that are the smallest regions in the MB lobes where a specific set of MBONs , DANs , and KC types overlap . For example , the β1 compartment can be divided into posterior ( β1p ) , surface ( β1s ) , and core ( β1c ) subdivisions containing the axons of the α/βp , α/βs , and α/βc KCs , respectively . The processes of some MBONs and DANs do not arborize in an entire compartment , but only in a subset of its elemental subdivisions . Tanaka et al . ( 2008 ) first described this anatomical feature and called such regions ‘synaptic units’ . An example is shown in Figure 11D , which shows a two-color labeling experiment illustrating the subdivision of the α2 compartment by the dendrites of the two MBONs , one of which arborizes in the domain of the posterior KCs ( α2p ) and the other in the domain of the surface and core KCs ( α2sc ) ; this arrangement is diagrammed in Panel D . The heavy vertical lines divide subdivisions into three groups representing different MBON neurotransmitter types ( also indicated by the colors filling the cells of the matrix ) . The rows of the matrix correspond to the cell types of KCs , MBONs , and DANs . See Table 1 for synonyms for cell type names and references . The colors of the cells of the matrix indicate the putative transmitter for MBONs and cluster of origin for DANs . The color correspondence is given below the matrix . The two MBON types of unknown neurotransmitter type are indicated in light gray . The fully-colored cells of the matrix represent processes of neurons ( dendrites for MBONs and synaptic terminals for DANs ) that are uniformly and densely distributed in that subdivision; fainter colors represent subdivisions that are innervated only sparsely . In three cases , output neurons send axon terminals back into the MB . Such cases are represented by an ‘X’ with the size representing the density of terminals in that subdivision . One DAN cell type ( PAM-γ4<γ1γ2 ) has dendrites within ( as well as outside ) the lobes in the subdivisions indicated by an ‘O’ ( see Figure 14—figure supplement 1E ) . The projection patterns of the GABAergic MB-APL and serotonergic MB-DPM are diagramed; these neurons are MB intrinsic neurons that broadly innervate the MB ( Figure 3—figure supplement 1B , C ) ( Waddell et al . , 2000; Tanaka et al . , 2008 ) . Also diagrammed are the projection patterns of the octopaminergic OA-VPM3 and OA-VPM4 ( Figure 3—figure supplement 1E ) ( Busch et al . , 2009; Busch and Tanimoto , 2010 ) and SIFamide peptidergic neurons ( Figure 3—figure supplement 1F ) ( Verleyen et al . , 2004 ) within the lobes; these neurons have only a small fraction of their terminals within the MB ( Figure 2—figure supplement 6 ) and these sparsely innervate only a subset of the compartments . The following abbreviations are used in the names of the subdivisions: a , anterior; m , middle; p , posterior; d , dorsal; s , surface; c , core; and pedc , pedunculus core . ( B–D ) Diagrams illustrating different circuit motifs found in the MB lobes . ( B ) The terminals of a single DAN type ( PAM-α1 ) and the dendrites of a single MBON type ( MBON-α1 ) occupy a compartment . 16 of the 21 MBON types , like MBON-α1 , arborize dendrites within just one of the three lobes ( i . e . , γ , α′/β′ , and α/β ) , indicating that these MBONs receive inputs from only one of the three KC classes . ( C ) Two DANs ( PAM-γ5 and PAM-β′2a ) innervate the region occupied by one MBON ( MBON-γ5β′2a ) . One DAN fills the γ5 compartment and the other only innervates the anterior elemental subdivision of the β′2 compartment ( β′2a ) ; a single MBON has inputs from the areas defined by both DANs . Four MBON types , like MBON-γ5β′2a , extend dendrites spanning lobe boundaries . ( D ) A single DAN ( PPL1-α′2α2 ) innervate two compartments ( the α′2 and α2 ) , an area covered by three MBON types ( MBON-α′2 , -α2sc , and -α2p3p ) . Eight MBON types , like MBON-α2sc and MBON-α2p3p , arborize dendrites further confined to elemental subdivision ( s ) within a compartment . This suggests that these MBONs receive input exclusively from subtypes of KCs; for example , MBON-α2p3p receives input from α/βp KCs , which presumably carry non-olfactory information ( see Figure 7A ) . Nearly all DANs have their termini confined to a single compartment; however , we identified 3 DAN types that have axon terminals in two compartments . DOI: http://dx . doi . org/10 . 7554/eLife . 04577 . 03510 . 7554/eLife . 04577 . 036Figure 13 . Each KC type transmits information to multiple compartments . ( A ) A simplified version of the matrix shown in Figure 12A that illustrates the contacts between the individual MBON ( top ) and DAN cell types ( bottom ) with the seven types of KCs ( columns ) . Colors of the cells in the matrix indicate neurotransmitter type for MBONs and cluster of origin for DANs . The two MBON types of unknown neurotransmitter type are indicated in light gray . Interestingly , each KC likely synapses with MBONs of all three neurotransmitter types and receives modulatory DAN input from both PAM and PPL1 clusters . ( B ) Diagrams of the MB lobes . The axons of the seven types of KCs are shown as straight vertical lines without branches and the boxes represent each of the 37 elemental subdivisions defined in Figure 12A . The number of KCs of each type , based on cell counting ( see ‘Materials and methods’ ) , is indicated . The size of the box representing each subdivision and calyces indicates its volume , as determined by measurements performed on confocal stacks . The volumes of lobes are not simply proportional to the number of KCs they contain and are not uniform along their lengths , presumably reflecting differences in synaptic density . The convergence ratio from KCs to single MBONs could range from as high as ∼2300:1 ( for MBON-β2β′2a that arborizes in β2 and β′2a of both hemispheres ) to ∼90:1 ( for MBON-α2p3p ) , assuming that each KC forms synapses in each elemental subdivision . In the top diagram , the color of the box represents the neurotransmitter used by the MBONs that have dendrites in that subdivision . In the bottom diagram , the color of the boxes represents the cluster of origin of the DANs innervating that subdivision . The following abbreviations are used in the names of the elemental subdivisions: a , anterior; m , middle; p , posterior; d , dorsal; s , surface; c , core; and pedc , pedunculus core . DOI: http://dx . doi . org/10 . 7554/eLife . 04577 . 036 The identification of the full complement of 21 MBON types highlights the extensive convergence of 2000 KCs onto just 34 MBONs , a number even smaller than the number of glomeruli in the AL . Thus , the high-dimensional KC representation of odor identity is transformed into a low-dimensional MB output . This suggests that the MBONs do not represent odor identity but instead provide a representation that may bias behavioral responses ( see ‘Discussion’ ) . Two clusters of dopaminergic neurons ( PPL1 and PAM ) have previously been shown to project axon terminals to specific regions within the MB lobes and transmit information about reward and punishment to the MB to guide learning ( Schwaerzel et al . , 2003; Claridge-Chang et al . , 2009; Mao and Davis , 2009; Aso et al . , 2010 , 2012; Burke et al . , 2012; Liu et al . , 2012 ) . Our split-GAL4 screen identified over 100 DANs of 20 types ( Figure 3C , Table 1 and Video 4 ) . Each DAN type contains a small number of neurons: DAN types from the PPL1 cluster contain one or two cells per hemisphere and DAN types from the PAM cluster contain up to ∼20 cells per hemisphere ( Table 1 , see Figures 14–16 for each cell type , see ‘Materials and methods’ for classification ) . 10 . 7554/eLife . 04577 . 027Figure 14 . Compartments with dendrites of glutamatergic MBONs . Representative images of neurons that have been aligned to the standard brain are shown . Glutamatergic MBONs and DANs that project to the same compartments in the lobes are displayed together . The names of cell types are given in a standard format that includes information about the subdivisions innervated . Short names based on simple numbering , as well the original name in cases where the cell type has been previously described at the single cell level , are also shown ( gray font ) . The number of cells per cell type found in each brain hemisphere is shown in parentheses . Some cell types were not separated by split intersections and images show mixtures of cell types in these cases . Except for PAM-γ4<γ1γ2 and PPL1-γ1 , DANs have dendritic branches in the ipsilateral hemisphere and axons that bilaterally innervate the same MB compartments in both hemispheres . The distribution of neurites outside the MB lobes is shown in more detail in Figure 18—figure supplement 1 . The split-GAL4 drivers for each cell type are listed in Table 1 and Supplementary File 1 . ( A ) The dendrites of MBON-γ5β′2a arborize in the contralateral γ5 and β′2a . The major axon of MBON-γ5β′2a projects ipsilaterally to the superior medial protocerebrum ( SMP ) , whereas a very thin axon projects to the CRE and SMP in the other hemisphere ( see Figure 14—figure supplement 1A ) . ( B ) PAM-γ5 neurons . Dendrites of these neuron arborize in the same regions of SMP where the MBON-γ5β′2a neurons terminate , suggesting a possible recurrent loop ( see text and Figure 20C ) . ( C ) PAM-β′2a neurons . ( D ) The dendrites of MBON-β2β′2a bilaterally arborize in the β2 compartment and β′2a subdivision and its axon projects ipsilaterally to the superior intermediate protocerebrum ( SIP ) and superior lateral protocerebrum ( SLP ) ( see also Figure 14—figure supplement 1B ) . ( E ) PAM-β2 neurons . ( F ) PAM-β2β′2a neurons have sparse terminals in the anterior layer of β′2a and even sparser terminals in the core layer of β2 . The dendrites of PAM-β2β′2a and PAM-β2 are spatially segregated , suggesting that distinct upstream neurons regulate their activity . ( G ) MBON-β′2mp arborizes in the contralateral β′2mp; its main axon projects to the CRE and SMP on the same side and its minor axon to the ipsilateral side ( Figure 14—figure supplement 1C ) . A second output neuron , MBON-β′2mp-bilateral , sparsely arborizes its dendrites in the β′2mp compartment and projects dense axons bilaterally ( see Figure 14—figure supplement 1D ) . ( H ) PAM-β′2p neurons . ( I ) PAM-β′2m neurons . ( J ) The MBON-γ4>γ1γ2 dendrites arborize in the contralateral γ4 and its axon projects both within the lobes to γ1 and γ2 and to regions outside the lobes . ( K ) PAM-γ4 neurons and PAM-γ4<γ1γ2 neurons . The image shows a mixture of two cell types; PAM-γ4<γ1γ2 is unusual , in which it has some of its dendrites within the MB ( in the γ1 and γ2 compartments; see Figure 14—figure supplement 1E ) . ( L ) The MBON-β1>α dendrites arborize in the contralateral β1 and its axon innervates the α1 , α2 , and α3 compartments within the lobes as well as areas outside the lobes ( see Figure 14—figure supplement 1F ) ; the terminals in α3 are concentrated in the surface and posterior layers . ( M ) PAM-β1 neurons ( bottom ) . The posterior layer of β1 is more densely innervated by a second PAM cluster cell type , PAM-β1ped , that also projects to the posterior end of the pedunculus ( see Figure 14—figure supplement 1G ) . ( N ) The dendrites of MBON-α1 arborize in α1 and its axons project to the posterior SIP and SLP . Two neurons with identical morphology are present in each hemisphere . Although we observed this cell type in MCFO analysis of the dVGlut-GAL4 line OK371 ( Mahr and Aberle , 2006 ) , its terminals showed much weaker immunoreactivity to the anti-dVGluT than the other putative glutamatergic neurons . ( O ) PAM-α1 neurons ( bottom ) have terminals that extend slightly outside the area arborized by the dendrites of MBON-α1 to the distal end of the pedunculus . DOI: http://dx . doi . org/10 . 7554/eLife . 04577 . 02710 . 7554/eLife . 04577 . 028Figure 14—figure supplement 1 . Neurons of the glutamatergic compartments . Morphologies of individual cells were determined by MCFO using the indicated split-GAL4 lines . Brain midlines are indicated by dashed lines . ( A ) MBON-γ5β′2a ( MB210B ) . ( B ) MBON-β2β′2a ( MB014B ) . One of the only three MBON cell types , with MBON-γ3 and MBON-γ3β′1 , to have bilateral dendritic branches in the MB . ( C ) MBON-β′2mp ( MB011B ) . ( D ) MBON-β′2mp bilateral ( MB210B ) . This neuron has sparse dendritic arbors in the β′2mp . Unlike MBON-β′2mp , this cell type sends dense axonal projections to both hemispheres . This cell type was only found in split-GAL4 lines ( for example , MB011B and MB210B ) together with other MBONs . ( E ) PAM-γ4 and PAM-γ4<γ1γ2 ( MB312B ) . Processes of PAM-γ4<γ1γ2 neurons in the γ1 and γ2 are devoid of Syt::smGFP-HA signals and on this basis are considered to be dendritic ( data not shown ) . ( F ) MBON-β1>α ( MB434B ) . ( G ) PAM-β1 and PAM-β1ped neurons ( MB194B ) . The PAM-β1ped neurons project through the pedunculus to the anterior edge of the calyx . Thus , unlike PPL1-γ1pedc that broadly innervates the pedc , this neuron appears to extend through the pedc to the remainder of the pedunculus without extensive arborization . DOI: http://dx . doi . org/10 . 7554/eLife . 04577 . 02810 . 7554/eLife . 04577 . 029Figure 15 . Compartments with dendrites of GABAergic MBONs . ( A ) MBON-γ3 and MBON-γ3β′1; the morphologies of these cell types are described in more detail in Figure 8D , E . ( B ) MBON-β′1 ( top right ) ; this MBON type is unusual in a number of ways . First , there are , on average , 8 cells per hemisphere ( i . e . , the number of cells fluctuate between 7–9 in MB057B ) compared with one or two for the other MBON cell types . Second , they are the only MBONs from the lobes whose dendrites arborize , in addition to the lobes , in neighboring neuropils ( CRE and SMP ) . Since CRE and SMP contain zones where the terminals of other MBONs converge ( see below ) , these neurons may sum the outputs from a number of MB compartments . Finally , this is the only MBON cell type that projects to the lateral accessory lobe , an output region of the central complex . ( C–D ) Three types of PAM cluster DANs innervate the γ3 and β′1 compartments: PAM-γ3 neurons ( C ) ; PAM-β′1ap neurons; and PAM-β′1m neurons ( D ) . ( E ) The dendrites of MBON-γ1pedc>α/β ( top ) arborize in the ipsilateral γ1 and the core of the pedunculus , where the axons of the α/β KCs are found . Its axon projects bilaterally to the α/β lobes and contralaterally to the core of the pedunculus and , to a lesser extent , outside the MB lobes; in α3 , its terminals are enriched in the surface layer ( similar to those of MBON-β1>α , see Figure 17E ) . ( F ) PPL1-γ1pedc; a DAN of the PPL1 cluster with terminals that overlap with the dendrites of MBON-γ1pedc>α/β . One additional PPL1 cluster DAN innervates γ1 sparsely ( PPL1-γ1; not shown , see Table 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04577 . 02910 . 7554/eLife . 04577 . 030Figure 15—figure supplement 1 . The MBON-γ1pedc>α/β and PPL1-γ1pedc innervates the γ1 lobe compartment as well as the core of the distal pedunculus . ( A ) Diagram of innervation by dendrites of MBON-γ1pedc>α/β and axon terminals of PPL1-γ1pedc , which intersect the axon bundle of the two types of γ neurons ( γd and γmain ) in the γ1 compartment and the three types of α/β neurons ( α/βs , α/βp , and α/βs ) at the core of the pedunculus . Registered images of the outline of the γ lobe and the α/β lobes ( B ) , PPL1-γ1pedc ( C ) , MBON-γ1pedc>α/β ( D ) , and both neurons ( E ) are shown from the side together with outline of the γ lobe and the α/β lobes . Ensheathing glia separate the γ1 compartment and the core of the pedunculus; the neurites of MBON-γ1pedc>α/β and PPL1-γ1pedc bifurcate and pass through the glia ensheathing each lobe ( see Figure 11K–L ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04577 . 03010 . 7554/eLife . 04577 . 031Figure 16 . Compartments with dendrites of cholinergic MBONs . ( A ) The dendrites of the two MBON-α′1 cells arborize unilaterally in α′1 . Their thin dendritic branches project through the edge of the α′2 and α′3 compartments . The MBON-α′1 neurons share an axon tract with MBON-α′3ap and MBON-α′3m and terminate in similar region of SIP , SLP , and LH ( see panel D ) . ( B ) MBON-γ2α′1 . This cell type consists of two morphologically identical cells ( see Figure 8B ) that project to the CRE and SMP . ( C ) PPL1-γ2α′1 neuron . ( D ) The dendrites of MBON-α′3ap and MBON-α′3m arborize in the ipsilateral α′3m or α′3ap , respectively , and their axons project bilaterally to the same regions of the LH , SIP , and SLP . The axonal branches of the MBON-α′3m stop at the LH , whereas the MBON-α3′ap axons extend to a region ventral to the LH . ( E ) PPL1-α′3 neuron . ( F ) The dendrites of MBON-α2sc arborize unilaterally in α2sc , and its axon projects bilaterally to the SIP , the SLP , and the dorsal LH . ( G ) MBON-α′2 projects to the CRE and SMP . A few terminals are also found in SIP . ( H ) The two MBON-α2p3p cells arborize their dendrites unilaterally in the posterior layer of the α2 and α3 compartments ( α2p and α3p ) and project to the SMP . The KCs in the posterior layer of the lobes ( the α/βp neurons ) have their dendrites in the dorsal accessory calyx , a region spatially segregated from the dendritic arbors of other KCs ( see Figure 7A ) , and their activity is decreased in response to odors ( Perisse et al . , 2013 ) , suggesting a distinct role for these output neurons . ( I ) PPL1-α′2α2 neuron . ( J ) Two morphologically identical MBON-α3 neurons have dendrites in α3 and terminals in the SMP , SIP , and SLP . ( K ) PPL1-α3 neuron . DOI: http://dx . doi . org/10 . 7554/eLife . 04577 . 031 The axon terminals of the DANs project to specific compartments and , similar to the dendrites of MBONs , tile the entire MB lobes ( Figure 10H , J , L , 14–16 and Video 4 ) . 17 of the 20 DAN types project to only a single compartment ( Figure 12 ) . Two-color labeling and multicolor flp-out of DANs innervating neighboring compartments show clear segregation of their axon termini ( Figure 11E , F and data not shown ) . Two-color labeling experiments revealed overlap of the axon termini of DAN types and the dendritic arbors of cognate MBON types that innervate the same compartment ( Figure 11G–I ) . Computational alignment to a single reference brain extends these observations to all DAN types and further demonstrates that the DAN axon termini tile the MB lobes ( Figure 10H , J , L ) . Thus , different MBON types have access to largely equivalent input from the KCs but are modulated by different DANs ( Figures 12 and 13 ) . In classical learning paradigms , different DAN types respond to different unconditioned stimuli ( US ) ( Riemensperger et al . , 2005; Mao and Davis , 2009; Burke et al . , 2012; Liu et al . , 2012 ) . The compartmental organization we observe suggests that the DANs may convey information about the US to specific MBON types . Dopamine release in specific compartments may modify local KC–MBON synapses to bias behavioral output . We identified three MBONs and one DAN type that appear to interconnect different compartments of the MB lobes ( Figure 12A , Figure 14—figure supplement 1E , and Figure 17A–D ) . Each of these three MBONs projects to the compartments of other MBONs but not back to the compartment occupied by its own dendrites . Thus information flows in one direction , and these MBON connections could create a multi-layered feedforward network ( Figure 17J , see ‘Discussion’ ) . 12 of the MBON types receive input from the KCs but not from other MBONs and therefore read out KC activity as a single output layer . Most of the MBONs having dendrites in the α′/β′ and γ lobes provide such a single-layer readout . In contrast , the glutamatergic MBON whose dendrites arborize in the γ4 compartment projects axons into the γ1 and γ2 compartments as well as to neuropils outside the MB ( Figure 17A , G–I ) . Thus , the γ1 and γ2 MBONs have access to direct KC input as well as to the input provided by the γ4 MBON . The outputs of these two γ lobe compartments thus reflect a two-layer feedforward network . The readout of the α/β lobe is even more elaborate because the γ1 MBON projects to all of the compartments of the α/β lobe ( Figure 17C , E , F ) , and the β1 MBON projects to the α1 , α2 , and α3 compartments ( Figure 17B ) . The α1 , α2 , and α3 MBONs thus represent a four-layer feedforward network with access to the KC activity of the α/β and γ lobes , both directly and indirectly through MBON intermediaries ( Figure 17J ) . 10 . 7554/eLife . 04577 . 037Figure 17 . Three types of MBONs and one type of DAN interconnect multiple compartments . ( A–C ) Registered images of the three MBON cell types whose axons project to other compartments within the MB lobes . ( A ) A glutamatergic output ( MBON-γ4>γ1γ2 ) from γ4 sends axons back into the lobes that terminate in γ1 and γ2 . ( B ) A glutamatergic output from β1 ( MBON-β1>α ) terminates in α1 , α2 , and α3 . ( C ) A GABAergic output from γ1 and the core of distal pedunculus ( MBON-γ1pedc>α/β ) terminates in the α/β lobes in both hemispheres and in the pedunculus , where α/β KCs bifurcate , in the contralateral side of the brain . ( D ) A DAN cell type ( PAM-γ4<γ1γ2 ) that has dendrites in γ1 and γ2 and terminals in γ4 . This cell type provides potential feedback signals to the MBON-γ4>γ1γ2 ( see panels G–I ) . ( E ) Two-color labeling of the GABAergic output neuron shown in panel C ( MB112C , pJFRC225-5XUAS-IVS-myr::smGFP-FLAG in VK00005 ) and an antibody against vesicular glutamate transporter . The dendrites of MBON-γ1pedc>α/β overlap with glutamatergic terminals in γ1 . The likely source of these glutamatergic terminals is MBON-γ4>γ1γ2 shown in panel A . ( F ) The GABAergic terminals of MBON-γ1pedc>α/β and glutamatergic terminals presumably of MBON-β1>α , the neuron shown in panel B , are in close proximity in the surface layer of α3 . The green signal in the lower left corner of the panel is from neurons outside the MB . This overlap of glutamatergic and GABAergic feedforward projections suggests that they may cooperatively or competitively regulate the excitatory output of MBON-α3 . ( G ) A substack projection showing two-color labeling of the dendrites of the only DAN ( MB312C ) that has dendrites in the MB lobes and axon terminals of an MBON in the γ1 compartment ( R53C03-LexA ) . ( H and I ) Single confocal slices in the γ1 and γ2 compartment showing two-color labeling of the cell types shown in ( G ) . ( J ) Schematic of the circuits within the MB lobes . DAN inputs ( bottom ) and MBON outputs ( top ) from 15 MB lobe compartments and the core of the distal pedunculus ( ped ) are shown ( gray rectangles , middle ) . Colors indicate neurotransmitter types for MBONs and cluster of origin for DANs as indicated . Dendrites are represented as squares and presynaptic terminals as triangles . Three MBONs , indicated by the heavier outline and striped cell bodies , send axonal terminals ( triangles ) back into the MB lobes creating a 4-layer feedforward network . See text for details . Eight types of MBONs receive input from more than one of the 15 compartments in the lobes ( or a compartment plus the ped ) and in five cases those compartments reside in different lobes . As different functions of learning and memory , such as acquisition and retrieval , have been attributed to different KC classes ( Isabel et al . , 2004; Krashes et al . , 2007 ) , MBONs integrating across lobes may function in coordinating different phases of learning and memory . MBON-γ1pedc>α/β and PPL1-γ1pedc innervate the core of the distal pedunculus ( ped ) as well as the γ1 compartment . Three DAN cell types innervate multiple lobe compartments . DOI: http://dx . doi . org/10 . 7554/eLife . 04577 . 037 The DAN cell type that projects axons to the γ4 compartment ( Figure 17D ) provides another form of communication between compartments . This cell type extends dendrites to the γ1 and γ2 compartments , where they appear to receive direct inputs from the axonal termini of MBON-γ4>γ1γ2 ( Figure 17G–I ) , creating a recurrent loop involving modulatory dopamine input . The MBON axons that innervate other compartments within the MB lobes have access to dopamine inputs and thus can potentially be modified by learning . Adaptive multi-layer or ‘deep’ feedforward networks are known to be capable of more complex readout functions than single-layer readouts ( Bishop , 2006 ) . Thus , the four-layer α lobe readout system may support more sophisticated neuronal computations than the one- and two-layer γ or α′/β′ lobe systems . The 34 MBONs represent the sole outputs of the MB lobes . Computational alignment of single-cell images allowed us to localize the axon termini of the MBONs outside the MB . The axon terminals of the MBONs ( Figure 18 ) converge onto five neuropils: the crepine ( CRE; a region surrounding the horizontal/medial lobes ) , the superior medial protocerebrum ( SMP ) , the superior intermediate protocerebrum ( SIP ) , the superior lateral protocerebrum ( SLP ) , and the lateral horn ( LH ) ( Figure 18E ) . Most MBON types project axons to several of these neuropils ( Figure 18—figure supplement 1 ) . Within a neuropil , the axon terminals of each MBON type exhibit distinct and confined projection patterns , suggesting that different MBON types may synapse onto different neurons ( Figure 19 , also see Figures 14–16 for individual MBON projection patterns ) . However , we also observed significant overlap between the axon terminals of different MBONs , suggesting that in certain cases MBONs may converge onto the same post-synaptic target neuron ( Figure 20A , D , E ) . These results , obtained by computational alignment , were confirmed for a subset of MBONs by two-color labeling experiments ( Figure 20J–L; Video 6 ) . 10 . 7554/eLife . 04577 . 038Figure 18 . Projection patterns of the MBON axons and DAN dendrites outside the MB lobes . ( A ) To analyze the projection patterns of the MBONs outside the MB lobes , we first segmented their axon terminals based on the localization of a presynaptic marker . A maximum intensity projection of a confocal stack showing MBON-α′2 neurons labeled with a presynaptic reporter ( magenta ) and a general membrane marker ( green ) . The split-GAL4 line MB018B was used to drive the expression of two constructs: a reporter targeted to membranes ( pJFRC225-5XUAS-IVS-myr::smGFP-FLAG in VK00005 ) and a reporter targeted to presynapses ( pJFRC51-3XUAS-IVS-Syt::smGFP-HA in su ( Hw ) attP1 ) . ( B ) An image of the segmented terminals and dendrites of a MBON-α′2 neuron aligned to the standard brain . After registration of the image , synaptic terminals were segmented based on the preferential labeling by Syt::smGFP-HA . Dendrites were segmented based on their morphology and localization within the MB lobe . ( C ) A registered image of the segmented presynaptic terminals of MBONs showing the areas of the brain innervated by MBONs . The segmented terminals were false-colored based on their putative neurotransmitter . ( D ) A registered image of the segmented dendrites of DANs . The segmented dendrites were false-colored based on their cluster of origin . ( E ) Distribution of MBON terminals and DAN dendrites as quantified in each neuropil found in the adult brain ( see Ito et al . , 2014 for the location and abbreviation for each neuropil [Ito et al . , 2014] ) . After normalizing the signal intensity from different cell types , we separately summed signals from the terminals of glutamatergic , GABAergic , or cholinergic MBONs and dendrites of PPL1 and PAM cluster DANs . 99% of the MBON terminals are distributed in the MB lobes and five neuropils ( CRE , SMP , SIP , SLP , and LH ) . Colors represent the percent of the processes found in each of 38 brain regions . See ‘Materials and methods’ for a description of the quantification method . DOI: http://dx . doi . org/10 . 7554/eLife . 04577 . 03810 . 7554/eLife . 04577 . 039Figure 18—figure supplement 1 . Projection patterns of the individual MBON and DAN cell types . Distribution of MBON terminals ( left panel ) and DAN dendrites ( right panel ) of selected cell types is shown separately . Colors represent the percent of the processes found in each brain region . Note that each MBON type typically projects to multiple neuropils and each DAN type typically has dendrites in multiple neuropils . Note that the atypical neuron with dendrites in the calyx ( MB-CP1 , see Table 1 ) is also shown at the bottom row ( labeled ‘calyx’ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04577 . 03910 . 7554/eLife . 04577 . 040Figure 19 . Distribution of MBON terminals and DAN dendrites in four brain areas . ( A ) Schematic of the brain highlighting the areas that exhibit the highest density of MBON terminals and DAN dendrites . ( B ) A magnified view of a portion of the brain shown in panel A with the CRE , SMP , SIP , and SLP neuropils indicated . ( C ) Distribution of MBON terminals and DAN dendrites showing their further clustering within the CRE , SMP , SIP , and SLP . Heat map representations of neurite density were computed separately for PAM and PPL1 cluster DAN dendrites and for cholinergic ( ACh ) , GABAergic , and glutamatergic ( Glu ) MBON terminals . Each horizontal row shows mean intensities in a 10 × 10 × 10 voxel ( 3 . 8 × 3 . 8 × 3 . 8 μm ) cube for each of the indicated neuronal types ( columns ) , represented in an 8-bit scale ( 0–255 ) . The rows have been sorted based on the sum of the intensity in each volume for all neuronal types . ( D–G ) Distributions of MBON terminals and DAN dendrites in the CRE ( D ) , SMP ( E ) , SIP ( F ) , and SLP ( G ) are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 04577 . 04010 . 7554/eLife . 04577 . 041Figure 20 . Convergence of MBON and DAN processes outside the MB lobes . ( A–C ) Matrices showing co-localization of the processes of MB extrinsic neurons outside the MB . Axon terminals and dendrites from each cell type were first segmented and the degree of overlap calculated ( see ‘Materials and methods’ ) and displayed in the form of a linear heat map with arbitrary units . For those cells in the matrices marked with an asterisk , images of the overlapping neurons are shown in ( D–I ) . The heavy black lines in ( A ) and ( B ) show the three major groups as determined by hierarchical clustering ( Ward's method ) . ( A ) Overlap of the MBON terminals . ( B ) Overlap of DAN dendrites . ( C ) Overlap of the MBON terminals and DAN dendrites . The cells of the matrix outlined by the heavy black lines indicate potential feedback loops where MBON terminals from a compartment lie in close proximity to the dendrites of DANs that innervate that same compartment . Note that the atypical neuron with dendrites in the calyx ( MB-CP1 , see Table 1 ) is also shown in panels A and C ( labeled ‘calyx’ ) . ( D and E ) Convergence ( arrowhead ) of the terminals of MBONs from α′2 and β′2mp ( D ) and of the terminals of MBONs from α3 and α1 ( E ) . The overlap of the axon terminals ( also shown in two-color labeling experiments , panels J and K ) suggests that different MBONs may target the same post-synaptic neurons . It is interesting that MBONs of different neurotransmitter types exhibit overlapping axon terminals ( e . g . , cholinergic MBON-α3 and glutamatergic MBON-α1 , E ) , suggesting that the target neuron may be modulated differently by each of the MBONs; for example , receiving excitatory input from cholinergic MBONs and inhibitory inputs from glutamatergic MBONs . ( F ) Overlap ( arrowhead ) of dendrites of PAM cluster DANs that innervate β′1 and the β′2p . ( G ) Overlap ( arrowhead ) of dendrites of PAM cluster DANs that innervate β1 and β2 . ( H ) Overlap ( arrowhead ) of the terminals of the MBON-γ5β′2 and dendrites of PAM-γ5 . ( I ) Overlap ( arrowhead ) of the terminals of the MBON-γ2α′1 and dendrites of PAM-β2β′2a . ( J–L ) Single confocal slices showing two-color labeling of the axonal terminals of the indicated MBONs . The following split-GAL4 and LexA driver lines were used: ( J ) MB081C and R25D01-LexA; ( K ) G0239 ( Chiang et al . , 2011 ) and R71C03-LexA; ( L ) MB026B and R12C11-LexA . DOI: http://dx . doi . org/10 . 7554/eLife . 04577 . 04110 . 7554/eLife . 04577 . 042Video 6 . Convergence of MBON-α3 with MBON-α1 . Two-color labeling of MBON-α3 and MBON-α1 by G0239 and R71C03-LexA as in Figure 20K . MBON-α1 was segmented from the pattern of R71C03-LexA . The video displays the projection patterns of both MBON types , then zooms to the SIP , where terminals of MBON-α3 and MBON-α1 are located , often within submicron distances of one another . DOI: http://dx . doi . org/10 . 7554/eLife . 04577 . 042 Computational alignment of single cell images of each DAN type identified the dendritic arbors of the DANs outside the MB ( Figure 18D ) . Interestingly , 90% of the dendritic arbors of the DANs reside in four of the neuropils targeted by the MBONs: CRE , SMP , SIP , and SLP ( Figures 18E and 19 ) . Because DANs are activated by unconditioned stimuli ( US ) , these neuropils are likely targets of the US input ( see ‘Discussion’ ) . We observed cases where the dendrites of different DAN types in a single neuropil overlap ( Figure 20B , F , G ) , suggesting that they may share upstream input . Furthermore , we observed overlap between MBON terminals and DAN dendrites within each neuropil , implying that some MBONs may synapse on specific DANs ( Figure 20C , H , I ) . In some cases , the axon terminals of an MBON overlap with the dendrites of the DAN type innervating the same compartment ( Figure 20H ) . This may provide feedback that regulates dopamine release and hence learning . In other cases , axon termini of MBONs from one compartment overlap with the dendrites of DAN types projecting to different compartments ( Figure 20I ) . This provides a pathway through which the activity of one MBON could modulate the synapses between KCs and MBONs of other compartments . The MBONs are thought to elicit learned behavioral responses , and it is therefore of interest that four MBON types project to the lateral horn ( Figure 18—figure supplement 1 ) , a brain region responsible for innate odor responses ( Heisenberg et al . , 1985 ) . These cholinergic MBONs may exploit LH neurons capable of eliciting innate behaviors to generate learned olfactory responses ( Sejourne et al . , 2011 ) . This connection may also serve to modulate innate behavioral responses as a consequence of learning . Most of the MBON outputs , however , do not target the LH but converge onto the neuropils surrounding the MB ( Figure 18 ) . These convergent loci also receive projections from the antennal lobe and the lateral horn and are likely to be major sites of integration and processing of information within the fly brain ( Aso et al . , 2014 ) . Neural representations of odor exist at multiple stations of the olfactory circuit . In the representation of odors by the antennal lobe projection neurons , every odor can be thought of as a point in an approximately 50 dimensional space , each dimension corresponding to a particular glomerulus . The KC representation of odor identity has a dimension at least an order of magnitude larger , not only because there are many more KCs than antennal lobe glomeruli but also because the KCs mix projection neuron inputs nonlinearly ( Gruntman and Turner , 2013 ) . The dimension of the KC representation not only allows for a far greater capacity to respond appropriately to a large number of odors , it can also enhance performance in more complex decision-related tasks . Interestingly , the lack of structure in the input to the KCs , the small number of inputs , and the sparseness of their activity all appear to be tuned to maximize this dimension ( Perez-Orive et al . , 2002; Huerta et al . , 2004; Caron et al . , 2013; Gruntman and Turner , 2013 ) ( Ann Kennedy , Columbia University Thesis ) . The fly has therefore evolved an olfactory circuit that exhibits structural and functional features predicted to optimize its ability to contextualize and respond appropriately to a rich array of olfactory experiences . The convergence of a large number of KCs onto a small number of MBONs indicates that the dimension of the MBON representation is significantly smaller than that of the KCs . The dimension of the MBON can be no greater than their number ( 34 ) and is likely to be considerably smaller because there are only 21 different MBON cell types . Thus , rather than providing a general representation of odor identity , the activity of the individual MBONs is more likely to encode a set of ‘state variables’ that collectively bias behavioral responses to sensory stimuli . This bias is likely to reflect the combined effects of external experiences and the internal state of the fly ( see accompanying paper ) ( Aso et al . , 2014 , Krashes et al . , 2009; Bracker et al . , 2013 ) . Consistent with this view , the odor responses of individual MBONs differ between flies and these differences appear to depend on plasticity ( Hige et al . , unpublished ) . It is tempting to associate individual MBONs with specific behaviors , but the ultimate bias in behavioral response may be represented across the full set of MBONs as a population code . Thus , the high-dimensional representation of odor identity in the KCs may be transformed into a low-dimensional representation that dictates behavioral bias . Just as an individual odor is represented by an ensemble of KCs , a given behavioral bias is likely to be represented by an ensemble of MBONs . In accord with this view , activation or inactivation of combinations of MBON cell types results in more robust effects on behavior than those observed with individual MBONs ( see the accompanying paper ) ( Aso et al . , 2014 ) . We have shown that the MB lobes are divided into compartments that receive input from the KCs and specific DANs and transmit information to a small number of MBONs . Each compartment therefore receives specific dopaminergic input capable of modifying the synapses between the KCs and specific MBONs . These compartments may reflect basic computational units of the MB . The specificity of the dopamine input and its ability to direct learning may therefore transform the unstructured KC representation of odor to an ordered MBON representation encoding behavioral bias . Specific subpopulations of DANs may react to different features in the external and internal world ( Mao and Davis , 2009; Liu et al . , 2012; Tomchik , 2013; Das et al . , 2014 ) . DAN activity may modify the activity of MBONs through learning to provide a representation that is predictive of the implications of an olfactory stimulus . Learning , in this view ( Sutton and Barto , 1998 ) , is a transition from a reactive DAN representation to a predictive MBON representation . The dendritic and axonal projection patterns of the MBONs and DANs suggest feedforward and feedback circuit motifs in the MB lobes . The interactions among the different MBON types may form a multi-layer feedforward readout network ( Figure 17 ) . Processing through such a network significantly expands the computational capacity of a readout system , which can be valuable for more complex learning strategies . Consider a learning scenario in which a fly associates an odor with a strongly aversive US . A conditioned aversive response to the initial odor could occur through plastic changes affecting an ensemble of output neurons , and a low response threshold could allow these responses to generalize to related odors . Upon further experience , some odors within this category might be identified as ‘safe’ or even appetitive . Ethologically important exceptions could be learned if one of the neurons that interconnect compartments ( MBON-γ4>γ1γ2 , MBON-γ1pedc>α/β , and MBON-β1>α ) became responsive to a ‘safe’ odor and inhibited the original trained ensemble of aversive MBONs . Thus , the layered MBON network provides an efficient mechanism for modifying and updating previous learning . Interestingly , the dendrites of the DANs overlap with MBON axons in four of the five MBON projection zones ( Figures 19 and 20 ) . Thus , MBONs may modify the activity of the DANs that modulate their own activity and plasticity , resulting in a recurrent loop . This could provide positive or negative feedback to a specific compartment . Positive feedback might enhance learning to particularly salient stimuli , whereas negative feedback might suppress dopamine release once the correct response has been learned . These recurrent connections may also allow output from one compartment to modulate learning in other compartments , as some MBONs appear to target DANs that innervate non-cognate compartments . The axonal terminals of the MBONs are largely confined to five discrete neuropils in the brain , the SIP , SMP , CRE , SLP , and the LH ( Figures 18 and 19 ) . The CRE , SMP , SIP , and SLP may be sites of convergence for all of the signals relevant for classical conditioning . These convergence zones are major sites of arborization for the dendrites of the DANs and axons of the MBONs . Since DANs are activated in response to an aversive or appetitive US , these neuropils are also likely to receive input from neurons transmitting information about the nature of the US . A US , by definition , elicits an innate behavioral response consistent with its valence , suggesting that the outputs from these neuropils may convey motor commands . Interestingly , dendrites of neurons projecting to the fan-shaped body in the central complex , a brain region coordinating motor actions ( Strauss , 2002 ) , arborize in these neuropils ( Hanesch et al . , 1989; Young and Armstrong , 2010 ) . We therefore postulate an evolutionary primitive circuit in the MBON convergence zones , in which US inputs activate motor command neurons to elicit innate responses . Evolution may have built upon this simple reflex circuit , incorporating pathways from the MB that generate learned behaviors in response to a CS ( Figure 21 ) . 10 . 7554/eLife . 04577 . 043Figure 21 . Schematic of the proposed convergence zone . MB lobes consist of three groups of compartments based on the putative transmitter of MBONs ( glutamate , GABA , and acetylcholine , shown color-coded ) . These compartments are interconnected inside the lobes and the MBONs send converging outputs ( color-coded arrows ) to small subregions within five neuropils: the lateral horn ( LH ) , CRE , SMP , SIP , and SLP . The dendrites of DANs are also confined mostly in the convergence zones within the CRE , SMP , SIP , and SLP . These neuropils , therefore , must receive information encoding the unconditioned stimuli ( US ) , such as sugar and shock , which recruit DANs for memory formation . In the simplest model , these same sensory inputs would activate not just the DANs but also other output neurons of these neuropils that elicit appropriate unconditioned behaviors . MBON , conveying the learned valence of stimuli , might then terminate onto these same output neurons . A subset of projection neurons from the antennal lobe and LH output neurons ( LHONs ) also converge with MBONs in these zones ( see accompanying paper Aso et al . , 2014 ) . In this way , learned and innate responses could use a common set of downstream circuits , originating in the LH , CRE , SMP , SIP , and SLP , to drive behavior . DOI: http://dx . doi . org/10 . 7554/eLife . 04577 . 043 The vertebrate brain consists of interconnected structures comprised of large collections of equivalent neurons whose number often increases with evolutionary complexity . This is in sharp contrast to most brain structures in invertebrates that consist of small number of neurons with stereotyped projections that suggest determined connections . The MB represents an exception . The MB lobes are formed by the axons of a large number of equivalent neurons , the KCs , and as with vertebrate cortical neurons , the number of KCs increases in species with more complex behaviors ( Strausfeld , 2012 ) . Moreover , the inputs to the KCs from olfactory projection neurons are not determined or stereotyped but appear random . Thus , the MB diverges from the highly ordered neural architecture typical of the invertebrate nervous system . The MB may therefore represent an evolutionary primitive brain structure homologous in form and function to structures in the vertebrate brain ( Schurmann , 1974; Laurent , 2002; Tomer et al . , 2010; Farris , 2011 ) . The elucidation of the inputs and outputs of the MB may now permit an understanding of how learning links an abstract representation to a specific behavior , and this may provide insight into higher associative functions in both invertebrate and vertebrate brains . Split-GAL4 and LexA transgenes used enhancers , selected based on GAL4-line expression patterns ( Jenett et al . , 2012 ) , and were constructed as previously described ( Pfeiffer et al . , 2010 ) . VT999036 was from the Vienna Tiles collection and a gift of Barry Dickson . Promoter regions corresponding to the following GAL4 constructs were amplified by PCR: TH-GAL4 ( Friggi-Grelin et al . , 2003 ) , Ddc-GAL4 ( Li et al . , 2000 ) , HL9-GAL4 ( Claridge-Chang et al . , 2009 ) , and Tdc2-GAL4 ( Cole et al . , 2005 ) . All fragments were amplified from genomic DNA except for the upstream region of HL9 , which was amplified from the HL9-GAL4 plasmid in order to conserve the mutated exon B start site . 5′-XbaI and 3′-FseI sites were added to the fragments upstream of GAL4 . Downstream fragments were amplified with added 5′-SpeI ( TH ) or NheI ( Tdc2 , Ddc and HL9 ) sites and 3′-NotI sites . These fragments were then cloned into the corresponding sites on pBPp65ADZpUw and pBPZpGAL4DBDUw vectors ( Pfeiffer et al . , 2010 ) that had been modified to add new restriction sites as follows: the p65ADZp and ZpGAL4DBD segments of these vectors were amplified with the addition of a 5′-XbaI and a 3′-AvrII site and then cloned into pBDP ( Pfeiffer et al . , 2008 ) at 5′-EcoRI and 3′-NotI . Downstream fragments were cloned into the modified AD and DBD vectors at AvrII NotI ( TH ) or NheI NotI ( Tdc2 , Ddc and HL9 ) sites . To generate Trh-p65ADZp and Trh-ZpGAL4DBD , the Trh promoter region was amplified from genomic DNA using primers SM ( 1A ) and BI ( 1S ) ( Alekseyenko et al . , 2010 ) and cloned into pBPp65ADZpw and pBPZpGAL4DBDw using the Gateway system as previously described ( Pfeiffer et al . , 2008 , 2010 ) . pJFRC225-5xUAS-IVS-myr::smGFP-FLAG in VK00005 and pJFRC200-10xUAS-IVS-myr::smGFP-HA in attP18 are described by Viswanathan et al . ( unpublished ) ; smGFP is a non-fluorescent , mutated GFP fused with multiple copies of an epitope tag ( either HA , V5 , or FLAG ) for immunolabeling with various fluorescent dyes . pJFRC51-3xUAS-Syt::smGFP-HA in su ( Hw ) attP1 and pJFRC216-13xLexAop-myr::smGFP-V5 in su ( Hw ) attP8 were generated by standard methods using vectors described in Pfeiffer et al . ( 2010 ) . Using a Syt::smGFP-HA construct with only three copies of the UAS sequence was necessary to decrease expression to a level that generated a >5-fold enrichment of signal in presynaptic boutons relative to other cellular subdomains . UAS-nuclearLacZ ( UAS-nlsLacZ ) for cell counting was previously described ( Baker et al . , 1996 ) . Multicolor flp-out ( MCFO ) is a stochastic method that labels individual cells in different colors using a set of three UAS-STOP-epitope constructs that each expresses a different epitope when the STOP cassette is removed . The STOP cassettes in these constructs are each flanked by FRT sites that are removed in a stochastic way by limited expression of flp recombinase ( Struhl and Basler , 1993 ) . Reagents for MCFO are described in Nern et al . , in preparation . NSyb-QF was generated by PCR amplifying the 1 . 9 kb EcoRI fragment of NSyb from pGWB-NSyb ( gift of Julie Simpson ) , adding the Drosophila Synthetic Core Promoter ( DSCP ) ( Pfeiffer et al . , 2008 ) by overlap-extension PCR and cloning the resulting fragment between the MluI and EcoRI sites of pQUAST ( Potter et al . , 2010 ) replacing the QUAS and hsp70-promoter sequences . A QF coding sequence ( QFrco; gift of Christopher Potter ) was then cloned into this vector using EcoRI and BamHI . Transgenic flies were obtained by standard P-element mediated transgenesis ( Genetic Services , Inc . ) . A total of four independent transformants were identified and in all cases most neurons were positive for QF as assessed by multiple lines bearing QUAS-PA-GFP; we used the line with the highest expression levels . A small group of neurons , including a subset of neurons within the PPL1 cluster , however , did not produce detectable QF activity in any of the NSyb-QF inserts ( data not shown ) . To make MB247-QS , a 247 bp sequence of Mef2 encoding the MB247 enhancer was PCR amplified from genomic DNA , the DSCP sequence added by overlap-extension PCR , and the MB247–DSCP fragment cloned between the MluI and EcoRI sites of pQUAST . The coding sequence of QS was then added using EcoRI and NotI . Five independent transformants were recovered , and all exhibited suppression of QF activity in α/β and γ KCs as assessed with NSyb-QFrco . To generate QUAS-C3PA-GFP and QUAS-SPA-GFP , the C3PA-GFP and SPA-GFP coding sequences , flanked with EcoRI-CAAC ( a Drosophila Kozak sequence ) at the 5′-end and NotI at the 3′-end , were cloned into pQUAST ( Potter et al . , 2010 ) . To generate 10xUAS-C3PA-GFP and 10xUAS-SPA-GFP , the C3PA-GFP and SPA-GFP coding sequences were blunt-end cloned into pJFRC-MUH ( Pfeiffer et al . , 2008 ) that had been digested with NotI and XhoI . Transgenic flies were obtained by phiC31-integrase mediated transgenesis ( Genetic Services , Inc . ) with insertion in attP40 , attP2 , VK00027 , VK00005 for C3PA-GFP and in attP40 for SPA-GFP . To identify enhancer fragments that drive expression in the MB cell types , we screened a database of the adult brain expression patterns of 7000 GAL4 driver lines ( Pfeiffer et al . , 2008; Jenett et al . , 2012 ) . We then generated approximately 400 transgenic lines that express either the transcription activation domain ( p65ADZp ) or the DNA binding domain ( ZpGAL4DBD ) of GAL4 under the control of one of the selected enhancers using the vectors described in Pfeiffer et al . , ( 2010 ) . We also generated p65ADZp and ZpGAL4DBD lines using the control regions from the genes encoding the enzymes for synthesizing monoamine neurotransmitters: tyrosine hydroxylase , dopamine decarboxylase , tryptophan hydroxylase , and tyrosine decarboxylase . To assay the expression pattern produced by an intersection of two enhancers , we visualized GAL4 activity in the progeny of a cross between a line expressing p65ADZp under one enhancer and a line expressing ZpGAL4DBD under the other enhancer . We screened the expression patterns observed in female brains of more than 2500 different p65ADZp-ZpGAL4DBD combinations , each chosen based on our anatomical analyses of the original GAL4 lines as likely sharing expression in a particular cell type . For screening expression patterns generated by p65ADZp and ZpGAL4DBD combinations , we crossed males carrying pJFRC200-10XUAS-IVS-myr::smGFP-HA in attP18; the ZpGAL4DBD transgene in attP2 with virgin females carrying the p65ADZp transgene in either su ( Hw ) attP8 , attP40 , or VK00027 and examined expression in 3- to 10-day old female progeny . For screening , we performed immunohistochemistry as described below in Terasaki 60-well microtiter plates ( Thermo Scientific , Waltham , MA ) containing 8 µl of solution . To obtain polarity and higher resolution information on selected lines , split-GAL4 lines were crossed to pJFRC51-3xUAS-Syt::smGFP-HA in su ( Hw ) attP1; pJFRC225-5xUAS-IVS-myr::smGFP-FLAG in VK00005 and four females brains plus two ventral nerve cords ( VNCs ) were dissected per line and immunolabeled in 2 . 0 ml tubes as described below . We further characterized lines we intended to use in behavioral experiments . First , we compared the intensity and specificity of lines that drive expression in the same cell types by immunostaining in parallel , mounting on the same glass slide , and imaging with identical confocal settings . Second , we screened for expression in non-neuronal tissues by crossing lines with pJFRC2-10xUAS-mCD8::GFP in VK00005 and examining the bodies of F1 progeny by stereo fluorescence microscopy; we observed fluorescence in non-neuronal tissues in 17 out of the 176 split-GAL4 lines ( see Figure 2—figure supplement 7 for examples ) . Figure 2—figure supplements 2–6 and Supplementary file 1 document the lines used in this and the accompanying paper ( Aso et al . , 2014 ) . These include KCs ( Figure 2—figure supplement 2 ) , DANs ( Figure 2—figure supplements 3 and 4 ) , and MBONs ( Figure 2—figure supplement 5 ) . We also made split-GAL4 lines for a variety of other modulatory input cell types that are putatively serotonergic , GABAergic , octopaminergic , and peptidergic ( Figure 2—figure supplement 6 ) , but we did not characterize these lines further in this study as their projections are not limited to localized areas of the MB lobes . Confocal image stacks documenting the expression patterns of the 92 selected lines in adult female brains and VNCs are available online ( http://www . janelia . org/split-gal4 ) . Because the observed expression pattern depends to some extent on the reporter used , and in particular on its site of genomic insertion , we assayed the expression of the 92 selected lines ( Supplementary file 1 ) independently with reporter constructs inserted at each of the two chromosomal sites ( attP18 and VK00005 ) that we intended to employ in future behavioral experiments . Consistent expression patterns were observed when using different UAS reporters; but for some lines , the intensity of expression observed in both the targeted MB neurons and off target cells varied significantly with the reporter used . pJFRC2-10xUAS-IVS-mCD8::GFP and pJFRC225-5xUAS-IVS-myr::smGFP-FLAG in VK00005 expressed more strongly in the targeted MB neurons than pJFRC200-10XUAS-IVS-myr::smGFP-HA in attP18 , but the two reporters ( pJFRC2 and pJFRC225 ) in VK00005 occasionally visualized off-targeted cells that were not visible with pJFRC200 in attP18 . The small number of MB extrinsic neurons expressing split-GAL4 in most lines , generally 1 to 14 cells per brain hemisphere , permitted the use of simple visual inspection to judge the completeness and reproducibility of the expression pattern . However , for lines expressing in subsets of the KCs , often several hundred cells , we also employed computer-assisted cell counting . KC nuclei were visualized with UAS-nlsLacZ and their membranes with pJFRC225-5xUAS-IVS-myr::smGFP-FLAG . The cell body cluster of the KCs was imaged at high-resolution ( 0 . 1 µm × 0 . 1 µm × 0 . 1 µm voxels ) . To improve separation of neighboring nuclei , myr::smGFP-FLAG signals were subtracted from the nlsLacZ channel . Nuclei were segmented using the 3D component analyzer plugin of the Fluorender ( Voxelpress; http://www . voxelpress . com/ ) ; first , we counted nuclei-sized ( mean ± 2SD ) nlsLacZ-stained objects at low threshold , deleted the detected volumes , and then counted the remaining nuclei at higher threshold . The counting results were visualized by randomly assigning colors and numbers to each counted object . This procedure gave results consistent with previous manual counting of c739 and c305a drivers ( Aso et al . , 2009 ) . The signal intensity of nlsLacZ varied between nuclei in the same sample; we counted all detectable objects . Brains of 3- to 7-day old females were dissected in saline ( 108 mM NaCl , 5 mM KCl , 5 mM HEPES , 5 mM trehalose , 10 mM sucrose , 4 mM NaHCO3 , 1 mM NaH2PO4 , 2 mM CaCl2 , 1 mM MgCl2 , pH-7 . 3 ) and mounted dorsal-anterior up in a Sylgard-coated Petri dish . Photoactivation was performed using a two-photon laser scanning microscope ( Ultima , Prairie Technologies , Middleton , WI ) with an ultrafast Ti:S laser ( Chameleon Vision , Coherent , Santa Clara , CA ) modulated by Pockels Cells ( Conoptics , Danbury , CT ) . A water immersion objective ( 60X/1 . 0 NA , Olympus , Japan ) was used for both visualization and photoactivation . Emitted photons were detected by a GaAsP detector ( Hamamatsu Photonics , Japan ) for green fluorescence and a PMT for red fluorescence . The laser was tuned to 925 nm for visualizing the samples ( with an intensity of ∼0 . 7–2 mW as measured after the objective ) and 710 nm for photoactivation . The Pockels Cells bias voltage was adjusted to obtain maximum signal-to-noise ratio when tuned at 710 nm , and photoactivation laser intensity at 710 nm was adjusted to be between 2 and 3 mW as measured after the objective . The photoactivation scan was performed with a pixel size of 0 . 24–0 . 39 μm and pixel dwell time of 2 μs . Each pixel was scanned four times successively using the frame-averaging function of the microscope software ( PrairieView , Prairie Technologies ) ; this was repeated with each repetition separated by 10–30 s . A target volume , such as the MB lobes ( Figure 4 ) or a single MB lobe compartment ( Figure 5 ) , was divided into 8–15 z-slices with 2–5 μm steps , depending on the size of the target as well as the orientation of the sample . The number of repetitions of the photoactivation scan depended on the expression levels of PA-GFP as well as the depth of the photoactivation target . For photoactivation of the MB lobes , the photoactivation scan was repeated 30–60 times . For photoactivation of a single compartment , the photoactivation scan was repeated 90–120 times . All photoactivated samples were prepared for confocal microscopy as follows: fixed for 45 min at RT using 2% PFA/PBL ( 2% paraformaldehyde in 75 mM lysine , 37 mM sodium phosphate buffer , pH 7 . 4 ) , washed multiple times in PBS containing 0 . 3% Triton X-100 ( PBST ) , blocked with 10% normal goat serum in PBST , incubated in primary antibodies overnight at 4°C , washed multiple times in PBST , and incubated in secondary antibodies overnight at 4°C ( or for more than 3 hr at RT ) before final washes with PBST . The samples were mounted using either VECTASHIELD ( Vector Labs , Burlingame , CA ) or SlowFade Gold ( Life Technologies , Grand Island , NY ) and confocal imaging was performed using an LSM510 with a Plan-Neofluar 40X/1 . 3 objective ( Zeiss , Germany ) . The following primary and secondary antibodies were used: rabbit-anti-DsRed ( 1:1000 , Clontech ) , nc82 ( 1:10 , Developmental Studies Hybridoma Bank ) , mouse anti-tyrosine hydroxylase ( 1:100 , EMD Millipore , Germany ) , rat anti-Dopa decarboxylase ( 1:400 , a gift from Jay Hirsch ) , rabbit anti-TexasRed ( 1:500 , Invitrogen , Life Technologies ) , Alexa Fluor 568 goat anti-rabbit ( 1:200 , Life Technologies ) , Alexa Fluor 633 goat anti-rat ( 1:200 , Life Technologies ) , and Alexa Fluor 633 goat anti-mouse ( 1:200 , Life Technologies ) . We used flies expressing PA-GFP pan-neuronally with the exception of the KCs , generated with a Synaptobrevin-GAL4 driver ( NSyb-GAL4 2-1; gift of Julie Simpson ) in combination with MB247-GAL80 , to identify the MB extrinsic neurons by photoactivation of the MB lobes ( Figure 4 ) . Generation of the photoactivation mask for the MB lobes was guided by a red fluorescent protein expressed in KCs ( MB247-DsRed , a gift of Andre Fiala ) . In initial experiments , a volume of ∼800 μm × ∼600 μm × ∼220 μm covering most of the central brain was imaged at a pixel size of 0 . 39 μm × 0 . 39 μm , with z-step size of 3 μm before and after photoactivation of the MB lobes in a hemisphere . This identified five clusters of MB extrinsic neuron cell bodies , each of which reproducibly contained more than two cells ( n = 3 brains , data not shown , see Figure 4 ) . Neurons found reproducibly , but not within these clusters , include MB-APL , MB-DPM , two MBONs located by the contralateral spur ( MBON-γ4>γ1γ2 and MBON-β1>α; MV2 ) , and two MBONs located in the contralateral anterior lateral region ( MBON-γ3 and MBON-γ3β′1 ) . We examined the number of neurons in each of the five clusters except the anterior PAM cluster ( see below ) by comparing higher resolution images of areas containing each cluster taken before and after photoactivation . Two or three clusters were examined per sample , and a volume of ∼200 μm × ∼200 μm × ∼100 μm was imaged at 0 . 39 μm × 0 . 39 μm × 1 μm resolution for each cluster . Images taken before and after photoactivation were aligned using a subpixel registration algorithm ( Guizar-Sicairos et al . , 2008 ) and correlation between registered images . Slight shifts in sample orientation precluded the use of a simple image calculation to identify photoactivated neurons . Instead , the photoactivated cell bodies were visually identified from aligned images assisted by a custom MATLAB interface . This analysis identified two neurons with their cell bodies located in regions that do not contain any of the neurons identified in the split-GAL4 lines . One was located ventral to the calyx ( MBON-γ4γ5 ) and the other was located ventral lateral to the antennal lobe ( MBON-γ1γ2 ) ( see below , and Figure 5G and Figure 5—figure supplement 1 ) . They represent the only two cell types identified by PA-GFP tracing that were not found in the split-GAL4 lines . It is important to note that we were not able to reproducibly visualize neurons with elaborate arbors over multiple neuropils and only diffuse and sparse processes in the MB lobes , such as octopaminergic neurons and SIFamide peptidergic neurons ( data not shown ) . This indicates a limitation for the PA-GFP tracing experiments , at least using the parameters described above , in detecting neurons with a large cytoplasmic volume by just photoactivating PA-GFP molecules within a small fraction of the total volume . We performed photoactivation of individual compartments to more specifically visualize the MB extrinsic neurons innervating each compartment ( Figure 5 ) . PA-GFP was expressed pan-neuronally using the Q-system ( Potter et al . , 2010 ) except in the α/β and γ KCs ( NSyb-QF , MB247-QS , QUAS-PA-GFP ) and each compartment was demarcated by a red fluorescent protein ( myr::tdTomato ) expressed using a split-GAL4 driver . The photoactivation mask was generated using this red marker , and in some cases where the split-GAL4 driver labels two compartments , this was further restricted using basal fluorescence of PA-GFP in the α′/β′ KCs ( for example , to demarcate the β2 compartment using MBON-β2β′2a , Figure 5A ) . Upon photoactivation , the brains were fixed and immunostained for tdTomato and tyrosine hydroxylase or dopa-decarboxylase as described above . The confocal images were processed by a custom MATLAB code to identify photoactivated cells . For each image , the fluorescent intensity of the photoactivated GFP signals was measured by drawing a region-of-interest around tdTomato positive neurons ( i . e . , neurons whose processes had been used as target for the photoactivation ) , and the green channel of the image was thresholded by a pixel intensity value representing mean minus 2× standard deviations of all the pixels within the regions-of-interest . A cell was counted as photoactivated if most of the pixels within the cell , as determined visually , were over this threshold value . Finally , we determined whether these cells are dopaminergic by examining co-localization with tyrosine hydroxylase or dopa-decarboxylase . Photoactivation of each of the 15 compartments identified all the MBONs included in the split-GAL4 lines ( Figure 5 , and data not shown ) . We observed the same number of PA-GFP positive cell bodies for each MBON cell type as were labeled by the corresponding MBON split-GAL4 line ( e . g . , Figure 5A , B ) and therefore these experiments confirmed that the split-GAL4 lines label all neurons within each MBON type . For the MBONs with dendrites in α′1 and α′3 compartments and cell bodies near KC cell bodies , it was not possible to assess whether there are more neurons of these types , because the cell bodies of photoactivated α′/β′ KCs were not distinguishable from those of the MBONs . Photoactivation of the MB lobes resulted in labeling of over 100 cells in the anterior medial cluster including the PAM-DANs ( cluster 5 in Figure 4 ) and it was not feasible to accurately count them by the methods described above . We therefore employed two independent approaches to examine the MB neurons in this cluster . We first performed photoactivation of the MB lobes with flies carrying R58E02-GAL80 ( Liu et al . , 2012 ) transgene that suppresses NSyb-GAL4-mediated PA-GFP expression in all PAM-DANs . We observed a dramatic reduction in the number of photoactivated cells in the anterior medial cluster ipsilateral to the photoactivated lobes as compared to flies without R58E02-GAL80 ( data not shown ) . This suggests that many of the MB extrinsic neurons in the anterior medial cluster are PAM-DANs . Moreover , we observed no photoactivated cells in the contralateral anterior medial cluster when using R58E02-GAL80 , indicating that the MB extrinsic neurons in this region are all PAM-DANs . We then characterized the MB extrinsic neurons in the anterior medial cluster by photoactivation of individual compartments as described above . We identified all the MBONs in this cluster and confirmed their numbers . We observed that there are more photoactivated GFP positive neurons in this cluster than those labeled by the split-GAL4 lines expressed in PAM-DANs ( see for example Figure 5C , D ) . These are likely dopaminergic as confirmed by the immunostaining ( Figure 5E and data not shown ) . It is important to note that the identification of neurons using PA-GFP tracing critically depends on the signal-to-noise ratio of photoactivated fluorescence relative to basal fluorescence of PA-GFP molecules . Neurons may not be detected if they express insufficient levels of PA-GFP , as observed in some cases using NSyb-enhancer , or if the photoactivated volume contains only a small fraction of a neuron's processes . For example , we observed little expression of NSyb-QF in a subset of DANs in the PPL1 cluster as well as in PAM-α1 DANs ( data not shown ) . Thus , we could not assess whether the split-GAL4 lines label all the PPL1-DANs and PAM-α1 DANs . We used dye-filling to visualize one of the MB extrinsic neurons identified by PA-GFP that was not included in the split-GAL4 collection ( Figure 5G ) . Flies were generated in which PA-GFP was expressed pan-neuronally except in the KCs ( genotype in Figure 5G legend ) . No red fluorescent protein was expressed . Photoactivation was targeted to the medial lobes using the absence of basal PA-GFP fluorescence in the KCs to demarcate the MB lobes . Upon photoactivation , the brain was treated with collagenase ( 2 mg/ml , Sigma ) for 30 s , the glial sheath removed using fine forceps , and the brain was mounted again in Sylgard-coated Petri dish in saline . A fire-polished , pulled glass pipette ( 0 . 5 mm ID , 1 . 0 mm OD , Sutter ) was backfilled with a TexasRed dye ( lysine-fixable 3000 MW , Invitrogen ) dissolved in saline . A two-photon microscope was used to guide the pipette to the photoactivated cell body , and the dye was injected into the cell body by iontophoresis using over one hundred 3 ms pulses of 20 Vdc applied every 0 . 5 s . The dye was allowed to diffuse for an additional 10 min and then the brain was fixed and immunostained with anti-TexasRed antibody and nc82 . Four brains were examined and in all cases we identified one neuron with the same morphology . Dissection and immunohistochemistry of fly brains were done as previously described with minor modifications ( Jenett et al . , 2012 ) . Brains and VNCs of 3- to 10-day old female flies were dissected in Schneider's insect medium and fixed in 2% paraformaldehyde in Schneider's medium for 55 min at room temperature ( RT ) . After washing in PBT ( 0 . 5% Triton X-100 in PBS ) , tissues were blocked in 5% normal goat serum ( or normal donkey serum , depending on the secondary antibody ) for 90 min . Subsequently , tissues were incubated in primary antibodies diluted in 5% serum in PBT for 2–4 days on a nutator at 4°C , washed three times in PBT for 30 min or longer , then incubated in secondary antibodies diluted in 5% serum in PBT for 2–4 days on a Nutator at 4°C . Tissues were washed thoroughly in PBT four times for 30 min or longer and mounted on glass slides for imaging ( see below for the mounting protocol ) . The following antibodies were used: rabbit anti-GFP ( 1:1000; Invitrogen; A11122 ) , mouse anti-nc82 ( 1:33 . 3; Developmental Studies Hybridoma Bank , Univ . Iowa ) ( Hofbauer et al . , 2009 ) , rabbit anti-HA ( 1:300; Cell Signaling Technology , Danvers , MA ) , rat anti-FLAG ( 1:200; Novus Biologicals , Littleton , CO ) , mouse anti-Drosophila ChAT ( ChAT4B1; 1: 100; Developmental Studies Hybridoma Bank , Univ . Iowa ) ( Takagawa and Salvaterra , 1996 ) , rabbit anti-GABA ( 1:500; A2052 , Sigma-Aldrich , Switzerland ) , rabbit anti-5HT antiserum ( 1:1000; Sigma-Aldrich , catalog no . S-5545 ) , mouse anti-tyrosine hydroxylase ( LNC1 , Millipore ) , rat anti-DDC ( 1:400; a gift from Dr J Hirsh ) ( Beall and Hirsh , 1987 ) , mouse anti-beta galactosidase ( 1:200; Abcam , Cambridge , MA ) , mouse anti-V5-TAG ( 1:1000; AbD Serotec , UK ) , Dylight-549 conjugated mouse anti-V5 ( 1:500; AbD Serotec ) , rabbit anti-Drosophila GAD1 ( 1:1000; a gift from Dr . FR Jackson ) , rabbit anti-DvGluT ( 1:5000; a gift from Dr . A DiAntonio ) as primary antibodies , and cross-adsorbed secondary antibodies to IgG ( H+L ) : AlexaFluor-488 donkey anti-mouse ( 1:400; Jackson Labs ) , AlexaFluor-594 donkey anti-rabbit ( 1:500; Jackson Labs , Sacramento , CA ) , Cy3 donkey anti-rabbit ( 1:500; Jackson Labs ) , AlexaFluor-647 donkey anti-rat ( 1:300; Jackson Labs ) , AlexaFluor-488 goat anti-rabbit ( 1:800; Invitrogen A11034 ) , and AlexaFluor-568 goat anti-mouse ( 1:400; Invitrogen A11031 ) . For determining the likely transmitter used by each MBON cell type , we immunolabeled brains from flies carrying the appropriate split-GAL4 drivers and pJFRC225-5xUAS-IVS-myr::smGFP-FLAG in VK00005 . Tissues were first incubated in primary antibody against GABA , GAD1 , or DvGluT for 2–3 days at 4°C , washed , incubated in secondary antibody for 2–3 days , and washed overnight . To visualize the MBON , we then incubated tissues in either rabbit anti-GFP or rat anti-FLAG ( depending on the host species of other primary antibody ) for 2–3 days , washed , and then incubated in secondary antibody for 2–3 days . Mixtures of 40–60 brains from 17 split-GAL4 drivers were stained in the same tube and mounted and imaged on the same glass slide to enable an unbiased comparison of immunoreactivity across MBONs . The expression pattern of the myr::smGFP-FLAG was used to genotype the brains . For other protocols , tissues were incubated in mixtures of multiple primary or secondary antibodies . After immunohistochemistry , tissues were post-fixed with 4% PFA in PBS for 4 hr at RT followed by four , 15 min washes in PBT . To improve adhesion during mounting , tissue were washed in PBS ( 15 min ) to remove the Triton and then placed on poly-L-lysine-coated cover slips to which they electrostatically adhere . Tissues were then dehydrated through a series of ethanol baths ( 30% , 50% , 75% , 95% , and 3 × 100% ) for 10 min each and then 100% xylene three times for 5 min each in Coplin jars . Samples were embedded in a xylene-based mounting medium ( DPX; Electron Microscopy Sciences , Hatfield , PA ) , and the DPX was allowed to dry for 2 days before imaging . For comparing expression intensities , up to 60 brains and VNCs were mounted on the same cover slip . Because tissues were attached to the flat surface of the cover slip in the same orientation , the same brain structures were located at the same depth during confocal imaging , facilitating a fair comparison of signal intensity across samples . Imaging was done on an LSM710 confocal microscope ( Zeiss ) . Brains and VNCs were imaged first at low-resolution using a Plan-Apochromat 20x/0 . 8 M27 objective ( voxel size = 0 . 56 × 0 . 56 × 1 . 0 µm; 1024 × 1024 pixels per image plane ) . The region including the neurons of interest was then imaged at higher resolution by using a Plan-Apochromat 63x/1 . 40 oil immersion objective ( voxel size = 0 . 19 × 0 . 19 × 0 . 38 µm; 1024 × 1024 pixels ) . For cell types too large to fit in a single image , regions of interests were scanned separately with multiple tiles ( a maximum of five tiles was required to cover the entire brain and optic lobes ) that were then stitched ( Yu and Peng , 2011 ) . Confocal images were analyzed using the Janelia Workstation , a suite of tools for viewing and analyzing image data ( S Murphy; K Rokicki; C Bruns; Y Yu; L Foster; E Trautman; D Olbris; T Wolff; A Nern; Y Aso; N Clack; P Davies; S Kravitz; T Safford , unpublished ) , ImageJ ( http://imagej . nih . gov/ij/ ) , Fiji ( http://fiji . sc/ ) , and Fluorender ( http://www . sci . utah . edu/software/13-software/127-fluorender . html; [Wan et al . , 2009] ) . Figure panels with black backgrounds are single slices or maximum intensity projections of confocal stacks or substacks . Figure panels with white or gray backgrounds show neurons that have been segmented using FluoRender . Making an anatomical atlas from confocal images of many different split-GAL4 lines depends on being able to align data collected from individual specimens onto the framework of a standard brain . In order to minimize the amount of deformation required in the brain alignment process , we prepared a standard brain ( JFRC2013 ) using fixation and dehydration steps identical to those used to prepare our experimental samples . After stitching ( Yu and Peng , 2011 ) five tiles of high-resolution confocal image stacks covering the entire brain and optic lobes , debris on the brain surface were removed from the image using Fluorender . We then generated a downscaled ( 0 . 38 µm isotropic voxels ) version of the JFRC2013 standard brain for use in alignment . Because confocal imaging time was a limiting factor , we sought to develop a method that enabled alignment of a single 63× confocal image stack that covered only a portion of the brain to a standard model of the entire brain . Our alignment strategy is outlined in Figure 22 . We imaged each brain twice: the entire brain at low-resolution ( 20× objective lens; voxel size , 0 . 56 × 0 . 56 × 1 . 0 µm ) and a confocal stack ( or stacks ) covering the region of interest at high-resolution ( 63× objective lens; voxel size , 0 . 19 × 0 . 19 × 0 . 38 µm ) . The two images were scaled to have isotropic voxels; because the low-resolution images were obtained with an air objective , they were optically flattened compared to the high-resolution images and , thus , required scaling in the z-axis . We then aligned the high-resolution image tile to the low-resolution whole brain image , using the reference nc82 channel , by means of image stitching ( Yu and Peng , 2011 ) , which obtains translations through searching the maximum normalized cross correlation using the fast Fourier transform ( FFT ) . Then affine registration of the low-resolution whole brain image to the JFRC2013 standard brain was used to provide a global alignment of the partial brain image to the standard brain . Finally , a non-linear transformation was applied to locally register the high-resolution tile image to the JFRC2013 standard brain using a symmetric diffeomorphic registration algorithm ( Avants et al . , 2008 ) with a combination of mutual information and normalized cross correlation as the similarity metric in the local alignment . 10 . 7554/eLife . 04577 . 044Figure 22 . Alignment of partial brain images to a standard brain model . ( A ) A low-resolution ( 20× ) confocal stack covering the entire brain and optic lobes ( magenta diagram ) and a single high-resolution ( 63× ) confocal stack of the portion of the brain containing the MB ( blue square ) are collected for each specimen . The 63× tile image is first aligned to a whole brain image of the same brain by scaling and rigid translation using labeling of the presynaptic active zone protein Bruchpilot by the mouse monoclonal antibody nc82 ( Laissue et al . , 1999; Hofbauer et al . , 2009 ) as the reference . The whole brain image is then globally aligned to the JFRC2013 standard brain ( black diagram ) by affine registration . In this way the high-resolution tile image is globally aligned to the standard brain and corresponding target volume in the standard brain is identified . ( B ) The nc82 pattern of the high-resolution tile image is non-rigidly aligned to that of the standard brain . See ‘Materials and methods’ for more detail . ( C ) A 63× image tile of a brain ( green , MB018B-driven expression pattern showing MBON-α′2 neurons; magenta , nc82 staining ) . ( D ) The same tile after alignment ( gray; nc82 pattern of the standard brain ) . ( E ) Portions of optical slices from confocal stacks of three brains from the MB018B driver line . ( F ) After alignment , the dendrites of MBON-α′2 neurons in all three samples can be seen to be confined within the α′2 compartment of the standard brain . ( G ) An alignment to the standard brain of three separately imaged brains , each of which visualized a different PPL1 cluster dopaminergic neuron; compare to Figure 11F , where the same three neurons were imaged in a single brain using MCFO . DOI: http://dx . doi . org/10 . 7554/eLife . 04577 . 044 For generating the atlas , we used only ∼25% of the specimens that showed the best alignment to the standard brain based on the reference nc82 marker . In this way we were able to obtain very high-quality alignments as judged by two criteria . First , the alignments to the standard brain obtained from multiple specimens of the same GAL4 line were very similar ( Figure 22E , F ) . Second , we observed the same relative arrangement of cell types when assessed using images from a single brain ( Figure 11F ) and by alignment of images of separate brains ( Figure 22G ) . We routinely obtained alignment data of a given cell type from multiple brains to allow us to assess biological and alignment variability between samples . Neuropil masks were generated in the JFRC2013 standard brain by alignment to the binary masks of the neuropils defined in Ito et al . ( 2014 ) , followed by manual editing of the neuropil borders guided by the nc82 staining of the JFRC2013 brain . To generate masks for each MB compartment , we averaged the registered intensities of MB extrinsic neurons projecting to the same compartment and samples representing the same KC cell type; we manually defined the borders between neighboring compartments after applying a Gaussian blur filter in 3D ( sigma = 2 ) . After normalization of intensities between images , terminals and dendrites were segmented based on morphology and Syt::smGFP-HA distribution using FluoRender ( Wan et al . , 2009 , 2012 ) . To determine the distribution of projections in different brain regions shown in Figure 18E , signals from the cell types making up each of five groups—glutamatergic , GABAergic and cholinergic MBONs and the PPL1 and PAM DANs—were averaged within a group . Then the total signal within the volume of each neuropil mask for each of the five groups was divided by total signal observed in all neuropils . To calculate signal intensities shown in Figure 19C , the average of signals in 10 × 10 × 10 voxel volumes ( 3 . 8 µm in each dimension ) were used . To estimate degree of overlap between processes of MBONs and DANs in Figure 20 , we used the method previously described by Cachero et al . ( 2010 ) . We had multiple images for each cell type and we treated the two brain hemispheres separately , giving us on average 17 . 4 image pairs per cell type combination . We computed the overlap for each image pair separately; each cell in the matrices shown in Figure 20A–C represents the mean value .

One of the key goals of neuroscience is to understand how specific circuits of brain cells enable animals to respond optimally to the constantly changing world around them . Such processes are more easily studied in simpler brains , and the fruit fly—with its small size , short life cycle , and well-developed genetic toolkit—is widely used to study the genes and circuits that underlie learning and behavior . Fruit flies can learn to approach odors that have previously been paired with food , and also to avoid any odors that have been paired with an electric shock , and a part of the brain called the mushroom body has a central role in this process . When odorant molecules bind to receptors on the fly's antennae , they activate neurons in the antennal lobe of the brain , which in turn activate cells called Kenyon cells within the mushroom body . The Kenyon cells then activate output neurons that convey signals to other parts of the brain . It is known that relatively few Kenyon cells are activated by any given odor . Moreover , it seems that a given odor activates different sets of Kenyon cells in different flies . Because the association between an odor and the Kenyon cells it activates is unique to each fly , each fly needs to learn through its own experiences what a particular pattern of Kenyon cell activation means . Aso et al . have now applied sophisticated molecular genetic and anatomical techniques to thousands of different transgenic flies to identify the neurons of the mushroom body . The resulting map reveals that the mushroom body contains roughly 2200 neurons , including seven types of Kenyon cells and 21 types of output cells , as well as 20 types of neurons that use the neurotransmitter dopamine . Moreover , this map provides insights into the circuits that support odor-based learning . It reveals , for example , that the mushroom body can be divided into 15 anatomical compartments that are each defined by the presence of a specific set of output and dopaminergic neuron cell types . Since the dopaminergic neurons help to shape a fly's response to odors on the basis of previous experience , this organization suggests that these compartments may be semi-autonomous information processing units . In contrast to the rest of the insect brain , the mushroom body has a flexible organization that is similar to that of the mammalian brain . Elucidating the circuits that support associative learning in fruit flies should therefore make it easier to identify the equivalent mechanisms in vertebrate animals .

2014

The neuronal architecture of the mushroom body provides a logic for associative learning