Datasets:

chemNLP

/

ncbi-text

like 0

Interferon regulatory factors (IRFs) are DNA-binding proteins that control interferon (IFN) gene expression. IRF1 has been shown to function as an activator of IFN and IFN-inducible genes, whereas IRF2 represses the action of IRF1 (). Since, interferon induction is followed by translational attenuation; it is plausible that the synthesis of IRF2 protein, which is required to repress and regulate the IFN stimulated genes, is allowed to continue under such condition using an alternate mechanism of translation (). In fact, the repressor of IFN-β promoter, NRF (NF-κB repressing factor) protein has been shown to be translationally regulated to provide sufficient level of NRF protein for the complete silencing of the IFN-β genes (). Initiation of translation is the rate-limiting step of protein synthesis and hence it is tightly regulated. Although the general mode of translation of cellular mRNAs involves cap-dependent translation initiation, a sizeable proportion of mRNAs was shown to be associated with polyribosomes in poliovirus-infected cells at a time when cap-dependent initiation is impaired (). The most widespread mechanism of cap-independent mode of translation initiation is mediated by internal ribosome entry sites (IRESs), which directly recruits ribosome bypassing the requirement for 5′ cap structure and the cap-binding protein eIF4E (,). Many mRNAs that contain IRESs encode proteins that play important roles in cell growth, proliferation, differentiation and regulation of apoptosis (). Stress conditions, such as starvation of growth factors, heat shock, hypoxia and endoplasmic reticulum (ER) stress leads to down regulation of protein synthesis through phosphorylation of eIF2α (). However, a number of cellular mRNAs containing IRES elements such as vascular endothelial growth factor (VEGF) (), (), cat-1 mRNA (), NRF () and PITSLRE kinase () continue to be translated under conditions when cap-dependent translation is severely impaired. Similarly, inhibition of protein synthesis during apoptosis is accompanied by a caspase-dependent cleavage of initiation factor eIF4G (). However, there is strong evidence that translation of death associated protein (DAP5) (), X-chromosome linked inhibitor of apoptosis protein (XIAP) (), apoptotic protease activating factor (Apaf1) mRNAs is maintained under these conditions and is driven by their respective IRES elements. This indicates that these mRNAs containing IRES may probably have a reduced requirement for the intact eIF4G, allowing the translation of mRNAs containing them to continue under stress conditions (). This mode of initiation of translation probably protects cells from hostile conditions or at least help them to tide over transient stress conditions. Here, we have investigated the presence of IRES element in the 5′ untranslated region (UTR) of ‘interferon regulatory factor 2′ or IRF2, which belongs to interferon regulatory factor family (). Our results suggest that IRF2 5′ UTR (177 nt) contains an IRES element, which undergoes translation initiation in an eIF4G-independent manner. Also, it seems that IFN-α treatment does not inhibit the IRF2-IRES function to the extent observed in case of HCV or BiP IRES activity (GRP78). Analysis of the cellular protein binding with the IRF2-IRES showed specific binding of certain cellular factors, which might influence its function under stress conditions. In fact PTB protein has been shown to specifically interact with the IRF2 5′UTR and partial knock down of PTB protein resulted in significant decrease in IRF2-IRES activity. Additionally, we have studied the effect of ER stress on IRF2-IRES function. In cells treated with tunicamycin, the IRF2 protein level as well as the IRES function was found to be largely unaltered. These results suggest that the IRES element of the IRF2 mRNA allow translation initiation under stress condition and may play a role in the cellular response. The cDNA corresponding to the 5′ UTR of IRF2, was amplified from the RNA isolated from HeLa cells and cloned in pCDNA 3.1 (+). The primers were used according to the GenBank sequence NM_002199 and confirmed by DNA sequencing (Gene Bank Acc. No. for IRF2 5′UTR, DQ409328). The construct pRΔENullF was a kind gift from Dr Peter Sarnow (Stanford University). All the bicistronic constructs contain respective 5′UTR sequences (pRIRF2F, pRHAVF and pRBipF) cloned between Renilla luciferase (RLuc) and firefly luciferase (FLuc) genes, in pCDNA 3.1 in between HindIII and EcoRI sites. The eukaryotic promoter less bicistronic construct, the pGEMT-R-IRF-F, containing the IRF2 5′UTR and also the pRΔEnullF bicistronic cassette were cloned in pGEMT easy vector (Promega) under T7 promoter. The T7pRCVB3F was cloned in the pBluescript vector (Stratagene) under T7 promoter. The landscape of structure derived from inactive ΔEMCV IRES sequence was cloned upstream of Rluc gene in the upstream hairpin (uphp) bicistronic plasmid (). The Nsp bicistronic construct (pRNspF) contains 264 nt from La ORF (120–204 amino acid encoding region) between Rluc and Fluc (). For constructing IRF2 monocistronic plasmid (pIRF2Fluc), IRF2-Fluc was digested with HindIII and ApaI enzymes (NEB) from the plasmid pRIRF2F and ligated in HindIII, ApaI digested pCDNA 3.1-Fluc. Coxsackievirus 2A protease gene was amplified from CVB3 cDNA (a generous gift from Nora Chapman, Nebraska) using the primers with BamH1 and EcoR1 sites respectively and cloned in pCDN3.1 His C (pCD2A). The primers used are as follows: Cox(F):5′ATTAggATCCggCgCATTTggACAA3′; Cox(R):5′ACgCgAATTCCTgTTCCATTgCATC 3′. Bicistronic plasmids pRHAVF and pRBipF were constructed as described earlier (,). The primers used for the amplification of IRF2 5′UTR are as follows: IRF2(F)-5′CggCAAgCTTTCTCCTTgTTTTgCT3′;IRF2(R)- 5′ATATgAATTCggTgCCCTCTCAgTg3′. Hela S3, Huh7 cells were maintained in DMEM (Invitrogen) with 10% fetal bovine serum (GIBCO, Invitrogen). Cells were transfected with various bicistronic plasmids and pSV40ß-gal using Tfx 20 reagent (Promega) and luciferase assay was performed using Dual luciferase assay reagent (Promega). In experiments using eukaryotic promoter-less bicistronic constructs, cells were infected with vaccinia virus expressing T7 RNA polymerase, VTF7.3 (generous gift from Dr B. Moss, NIH) () prior to transfection with bicistronic plasmids. Luciferase assay was performed by dual luciferase assay reporter reagent (Promega) in a TD 20/20 luminometer (Turner Design, CA, USA). For the interferon experiment, Huh 7 cells was transfected with the bicistronic plasmids pRIRF2F, pRHCVF, pRBiPF followed by treatment of 1000 IU/ml of IFN–alpha 2b (Virchow Ltd). For the 2A protease experiment, co-transfection was performed using pRCVB3F, pRIRF2F and pRHAVF bicistronic plasmids with Coxsackie 2A plasmid (pCD2Apro) constructs. Luciferase assay was performed after 24 h of transfection. For tunicamycin treatment, cells were incubated in presence of 2.5μg/ml of tunicamycin (Calbiochem) for 14 h. Co-transfection of siRNA with bicistronic plasmid was performed in HeLa S3 cells growing in monolayer using lipofectamine-2000 transfection reagent and optiMEM-I prepared without addition of antibiotic (Invitrogen). Cells were seeded onto 35 mm dishes one day prior to transfection in similar manner. For each transfection, 100 nM of pre-characterized siPTB (Dharmacon) and 1 μg of bicistronic DNA were diluted with optiMEM-I to a final volume of 100 μl. In a separate tube, 6 μl of lipofectamine-2000 was diluted with 94 μl of optiMEM-I to a final volume of 100 μl followed by incubation at room temperature for 5 min. The contents of the two tubes were mixed and incubated at room temperature for 20 min. Subsequently, 800 μl of optiMEM-I was added to the transfection mixture, which was then layered onto cells. Six hours later, the medium was replaced with 2 ml of DMEM (with antibiotic) and 10% FBS. Thirty-six hours post–transfection, the cells were washed, lysed with passive lysis buffer and luciferase enzymes assayed in a similar way. For RNA transfections, capped bicistronic RNAs were synthesized from different constructs RIRF2F, RBipF, RΔEnullF) using T7RNA polymerase (Ribomax kit, Promega). Ten microgram of the above synthesized RNAs were used to transfect HeLa cells using Lipofectamine 2000 and optiMEM-I (Invitrogen) as described above. After 6 h, medium was replaced with 2 ml of DMEM (with antibiotic) and luciferase assay was performed by dual luciferase assay reporter reagent (Promega) after 8 h incubation. Enzyme activity was measured in a TD 20/20 luminometer (Turner Design, CA, USA). The transfection efficiency was normalized and the relative luciferase activities were plotted. To make antisense FLuc probe RNA, pCD Luc DNA was linearized with HindIII (NEB) and transcribed by SP6 RNA polymerase (Promega) and 10 μCi/μl of alpha P UTP (NEN) as per manufacturer's guidelines. The HCV 5′UTR RNA probe was made from HCV-GFP DNA () linearized with EcoRI and was transcribed by T7 RNA polymerase. Similarly, the P-labeled RNA probes corresponding to the 5′UTRs of IRF2 and HAV were made from their respective plasmid DNAs after linearizing with either NcoI or EcoRI and transcribed with either Sp6 or T7 RNA polymerase, respectively. The non-specific RNA was made from linearized pGEMT as described elsewhere (). pRIRF2F, pRBipF, pRΔEnullF bicistronic plasmids were linearized with Pme1 (NEB) and the corresponding bicistronic RNAs were synthesized using Ribomax kit (Promega) following manufacturer's protocol. Total RNA from the HeLa cells, transfected with pRIRF2F, pRCVB3F bicistronic plasmids were extracted using TRIZOL (Sigma), followed by DNase I treatment. Firefly luciferase RNA (Promega) and above extracted RNAs were resolved on a 0.8% agarose–formaldehyde gel, blotted on positively charged Nylon membrane (Millipore) and hybridized with a P-labeled riboprobe corresponding to the FLuc gene. Total RNA from HeLa cells transfected with pRIRF2F bicistronic plasmid was extracted using TRIZOL (Sigma). Reverse transcription was performed using AMV RT (Promega) followed by PCR with taq polymerase (Invitrogen). Huh7 and HeLa cells were harvested and the cell pellet was resuspended in 1× RIPA buffer (10 mM sodium phosphate, pH 7.5, 150 mM NaCl, 2 mM EDTA, 1% sodium deoxycholate, 1% NP-40, 0.1% SDS, 0.1% βME, 1 mM PMSF, 50 mM sodium fluoride). Extracts were suspended with 5× SDS gel loading buffer (100 mM Tris-Cl, 200 mM DTT, 4% SDS, 0.2% bromophenol blue, 20% glycerol) and resolved on SDS–10% polyacrylamide gel, followed by electrotransfer of proteins to nitrocellulose membranes. The expression of IRF2 and tubulin was analyzed using antibody specific to IRF2 (a generous gift from Dr Angela Battistini of Istituto Superiore di Sanita, Rome, Italy) and anti rabbit secondary antibody (SIGMA). For detecting tubulin, anti-tubulin antibody (SantaCruz Biotech) was used followed by anti-mouse secondary antibody (SIGMA). TFIID was detected using anti-TFIID antibody (Santa Cruz Biotech) and eIF4G was detected using anti-eIF4G antibody against N-terminal region (Santa Cruz Biotech). The signal was detected by using enhanced chemiluminescence (ECL) detection kit (Amersham-Pharmacia). Similarly for detecting endogenous PTB, anti-PTB antibody was used (Calbiochem). S10 extract was prepared from HeLa and Huh7 cells as described before (). [α-P] 5′UTR RNAs were allowed to form complex with S10 extracts as described earlier, followed by cross-linking with UV light. The unbound RNAs were digested with RNaseA treatment. The protein–RNA complexes were then resolved in a SDS–10% polyacrylamide gel followed by phosphorimaging analysis. The expression of recombinant PTB from PET28a-PTB (a generous gift from Dr J.G. Patton) was induced by 0.6 mM IPTG in (BL21 DE3) cells transformed with the expression vector. His-tagged protein was purified using Ni-nitrilotriacetic acid agarose (Qiagen) under non-denaturing conditions and eluted with 250 mM imidazole. HeLa cells were treated with either interferon α or tunicamycin for the different time periods as mentioned in the text followed by starvation of the cells for 45 min in MEM-medium lacking methionine (SIGMA). Cells were washed and incubated with 100 μCi of S-methionine (trans-label, BARC) for 45 min at 37°C. Cells were harvested and the pellet was resuspended in 2× IP buffer (2% triton X-100 and 0.1% NP40 in TBS) and kept in ice for 1 h. The supernatant was collected and protein estimation was performed using Bradford's reagent (BIORAD). Equal amount of protein was resolved in a SDS–10%PAGE and analyzed by autoradiography. For IRF2 immunoprecipitation, 150 μg of untreated/treated cell extracts were incubated with IRF2 antibody (Santa Cruz Biotech) overnight at 4°C. The immunocomplex was separated by protein A-sepharose beads (SIGMA) for 2 h at 4°C on a rocker. The beads were washed three times with 1× IP buffer and the bound proteins were analyzed by SDS–10% PAGE followed by detection by autoradiogram. To determine whether IRF2 5′UTR can mediate cap-independent internal initiation of translation to provide basal level of protein under stress; we have investigated the presence of an IRES element in the 5′UTR of IRF2 RNA. For this purpose, IRF2 5′ UTR was amplified by RT-PCR from total RNA isolated from HeLa cells. The nucleotide sequence of the IRF2 5′UTR has been shown in A. Zuker's MFOLD algorithm predicted a stable secondary structure with a minimal free energy of −41.7 kcal/mol (B) (). Similar secondary structure was predicted by MFOLD for the BiP IRES (data not shown). It would be interesting to investigate whether IRF2 5′UTR contains the ‘Y’-type stem-loop structure that has been suggested as the characteristic feature of certain cellular IRESs (). To investigate the presence of IRES element in the 5′UTR of IRF2, the amplified IRF2 5′ UTR was subsequently cloned in a bicistronic construct in between two reporter genes. The upstream reporter (Renilla luciferase) in this bicistronic RNA is translated by cap-dependent mode, whereas the downstream reporter (Firefly luciferase) will be translated if the intergenic region contained a functional IRES element. The bicistronic plasmid, pRIRF2F was transiently transfected into HeLa cells followed by luciferase assay. The results showed appreciable amount of firefly luciferase (Fluc) translation mediated by the IRF2 5′UTR. The Fluc activity was found to be around 12.5-fold higher compared to the negative control bicistronic plasmid, pRΔEnullF. Interestingly, the bicistronic plasmid pRBipF, containing Bip IRES as positive control (), showed ∼17-fold increase in the Fluc activity compared to null bicistronic plasmid control (D and E). However, cap-dependent translation of the renilla luciferase (RLuc) was found to be similar in all the three plasmids as expected. These results indicate that the 5′UTR of IRF2 might contain an IRES element, the activity of which is comparable to that of representative cellular IRES. In order to rule out scanning or ribosomal read-through as the possible reason for the IRF2 5′UTR-mediated translation of the firefly luciferase, we have used IRF2 bicistronic construct pΔERIRF2F, containing region of highly stable secondary structure upstream of renilla luciferase to prevent ribosome loading (F) (). When the pΔERIRF2F plasmid was transfected into HeLa cells, RLuc activity was found to be significantly inhibited. However, no significant change in Fluc translation was observed compared to values obtained with the control IRF2 bicistronic construct pRIRF2F (G). Similarly, when the plasmid pRΔEIRF2F containing the internal hairpin structure inserted downstream of Rluc (G) was trasnsfected into HeLa cells, FLuc activity was not inhibited. Interestingly, FLuc activity was found to be marginally higher compared to control, probably due to a change in IRF2-IRES RNA structure in the context of the internal hairpin. However, Rluc activity remained unchanged as expected (G). The results suggest that the translation of the downstream cistron Fluc was not due to ribosomal read-through of the first cistron. It is possible that in the cells transfected with IRF2 bicistronic plasmid, small amount of monocistronic FLuc RNA is generated from the bicistronic construct due to cryptic promoter activity of the IRF2 5′UTR sequence. In order to rule out this possibility, HeLa cells were transfected with IRF2 bicistronic construct that was cloned in a vector, which lacked conventional eukaryotic promoter but contained a T7 phage promoter (A). Similarly, null and coxsackievirus B3 (CVB3) bicistronic plasmids were used respectively as negative and positive controls in the experiment. Results showed luciferase activity almost equal to background level in absence of any eukaryotic promoter in the transfected constructs. But when cells were transfected with recombinant vaccinia virus expressing T7 RNA polymerase gene (VTF7.3) prior to transfection of the construct, significant levels of Fluc and Rluc activity were detected from the same IRF2 and CVB3 bicistronic constructs (A). In fact Fluc activity mediated by IRF2-IRES was found to be around 15 times that of the null construct. As expected, the viral IRES (CVB3) showed much higher efficiency compared to cellular IRES activity under this condition. The result rules out the possibility of cryptic promoter activity in the IRF2 5′UTR. To rule out the possibility, that IRF2 5′UTR might contain splice sites which generate monocistronic Fluc RNA , northern blot hybridization assay was performed. For this purpose, total RNA was extracted from the cells transfected with IRF2 bicistronic plasmid and probed with a P-labeled riboprobe complimentary to Fluc gene. As a positive control, RNA extracted from cells transfected with a similar bicistronic plasmid containing Coxsackievirus B3 (CVB3) IRES was included in the assay. In our northern blot assay, we failed to detect any smaller RNA products derived from either of the transfected bicistronic plasmids (B). We have also performed RT-PCR analysis from the total RNA extracted from cells transfected with the IRF2 bicistronic plasmid using different sets of primers as shown in C, which showed the presence of intact bicistronic RNA . Additionally, to investigate the presence of shorter monocistronic RNA (if any), we have performed RT-PCR analysis of total RNA as above with different dilutions of input RNA and primer sets as indicated above the panel in D. In this experiment primer set P5/P4 would amplify the region of Fluc gene only, whereas the primer set P3/P4 would amplify the full-length IRF2 5′UTR along with the Fluc gene. In the event of cryptic splice sites within IRF2 5′UTR in the bicistronic construct, the ratio of the amplified products generated by using P5/P4 primers would be significantly more than that of P3/P4 product. However, in our assay we did not find significant differences in the amplified products (D). To further validate the IRF2-IRES activity, we have performed RNA transfection experiment. Capped bicistronic RNAs, synthesized from different plasmid constructs (RΔEnullF RIRF2F, RBipF, RCVB3F) were transiently transfected into HeLa cells and the relative luciferase activities were measured 8h post-transfection. The Fluc activity mediated by IRF2-IRES was found to be 6-fold higher than that obtained for the negative control pRΔEnullF (E). Similar fold increase in Fluc activity was observed for Bip IRES, used as positive control for cellular IRES (E). Interestingly, the Fluc activity mediated by the viral IRES (CVB3) was found to be much higher than the cellular IRESs. However, the results are consistent with recent reports, which suggest that cellular IRESs are not as active as viral IRESs when RNA transfections are performed, possibly because cellular IRESs require ‘nuclear history’ for their optimum activity (,). To further characterize the IRES activity of IRF2 RNA, we have studied the requirement of canonical and non-canonical initiation factors. Picornavirus 2A protease has been shown to cleave eIF4G and shut down cap-dependent translation of cellular mRNAs (). Although viral IRES elements are not sensitive to eIF4G cleavage, hepatitis A virus IRES has been shown to require intact eIF4G for efficient translation (). To investigate whether the IRF2-IRES needs intact eIF4G for its efficient function, we have used the Coxsackievirus B3-2A protease encoding plasmid CVB3-2A. This plasmid was co-transfected with IRF2 bicistronic plasmid in HeLa cells. In addition, bicistronic plasmids containing the IRES element of either Coxsackievirus B3 (CVB3) or Hepatitis A virus (HAV) were also included as controls (A). Western blot analysis of the CVB3-2A protease-treated cell extract showed cleavage of eIF4G in the transfected cells (B). The expression of CVB3-2A protease resulted in drastic decrease of cap-dependent translation of RLuc in case of all three bicistronic plasmids but did not significantly affect Fluc translation mediated by IRF2 and CVB3 IRESs (A). On the other hand, HAV IRES activity was reduced by CVB3-2A expression. Thus it appears that the IRF2-IRES-mediated translation might not require the presence of intact eIF4G. To investigate the cellular protein binding with IRF2 5′UTR, UV cross-linking experiment was performed using HeLa S10. The results showed major binding with p70, p58 and p35 and minor binding with p110, p86 and p42 with the IRF2 5′UTR (C). Interestingly, when S10 was used from HeLa cells treated with CVB3-2A protease, a prominent band of 50 kDa polypeptide (p50) was found to cross-link with the IRF2-IRES. However, the same protein did not show interaction with HAV 5′UTR. Also another polypeptide of ∼110 kDa did not show binding with the HAV 5′UTR when 2A protease-treated cell extract was used (B, compare lanes 1 and 2). It has been shown earlier that the ∼58 kDa band interacting strongly with the HAV 5′UTR is actually polypyrimidine-tract-binding protein (PTB) (). To investigate whether PTB protein also interacts with IRF2, UV cross-linking experiment was performed with purified recombinant PTB protein (A, lane 1). As expected, 100-and 500-fold molar excess of the cold self IRF2 5′UTR RNA competed out the PTB binding with the radiolabeled 5′UTR probe (A, lanes 2 and 3), whereas competition with 100- and 500-fold molar excess of non-specific RNA failed to compete the PTB binding with the IRF2 5′UTR suggesting the specificity of the interaction (A, lanes 4 and 5). Although the precise mechanism of the cellular IRESs is not clear, the requirements of some auxiliary factors, called as IRES -acting factors (ITAFs) is well documented (). To elucidate the possible role of PTB in modulating the IRES activity of IRF2, we partially silenced PTB by transient gene silencing method using siRNAs. For this purpose, bicistronic plasmid DNA containing IRF2-IRES was co-transfected with 100 nM of a pre-characterized siRNA specific for silencing PTB gene (). The results showed significant decrease in IRF2-IRES-mediated translation of Fluc whereas cap-dependent translation of RLuc was not affected significantly (B). However, a non-specific siRNA did not inhibit IRF2-IRES activity (data not shown). As expected, the western blot analysis showed almost 50% decrease in the PTB protein level in the siPTB-treated cells compared to control (C), suggesting that the level of PTB could be critical determinant for the efficient function of the IRF2-IRES element. IRF2 belongs to the interferon regulatory factor family and is known to have a transcriptional repressor activity (). We were interested to explore the physiological significance of the IRF2-IRES activity (if any) in regulating cellular response to interferon. Interferon has been shown to inhibit the IRES activity of hepatitis C virus (HCV) (), and is also known to activate PKR which leads to phosphorylation of eIF2α resulting in inhibition of cap-dependent translation (). As a first step, we have investigated the effect of interferon treatment on the activity of IRF2-IRES and used HCV IRES or Bip IRES as representative controls. For this experiment, the hepatocellular carcinoma cells (Huh7) was chosen as the experimental cell line, since earlier studies with this cell line have demonstrated the inhibitory effect of IFN α on HCV IRES activity (). When Huh7 cells were transfected with either HCV or IRF2 or Bip IRES containing bicistronic plasmids and treated with 1000 IU/ml of IFNα for 24 h, it was observed that HCV IRES activity was significantly inhibited as reported earlier (). However, the extent of inhibition was much less in the case of IRF2-IRES activity as compared to HCV and BiP IRESs (A). Additionally, to monitor the synthesis of IRF2 protein during interferon α treatment, pulse metabolic labeling experiment was performed. For this purpose, Huh7 monolayer cells were incubated with interferon α for 12 h, followed by pulse labeling of proteins using S-methionine. The results showed appreciable decrease in overall protein synthesis, although some of the proteins were found to continue synthesis even after interferon α treatment (data not shown). Immunoprecipitation of the pulse-labeled proteins using anti IRF2 antibody showed no significant change in IRF2 protein synthesis (B). Further, to reconfirm the actual position of the IRF2 band, western blot was performed with the same IP extracts using antiIRF2 antibody (C). The result reconfirms that during IFN treatment, the IRF2 protein level was not significantly reduced. Since, the level of phosphorylated form of eIF2α was found to be increased under similar condition (data not shown), it appears that IRF2-IRES function might continue when cap-dependent translation is largely impaired, suggesting possible role in maintaining intricate balance in the expression of interferon response genes. To gain further insight on the differential activity of HCV and IRF2-IRES under such condition, we performed UV cross-linking assay using S10 extract isolated from either mock or interferon α-treated Huh7 cells. The results showed a similar protein-binding profile. However, in UV cross-linking experiment p60 binding was found to be much more prominent with the IRF2 5′UTR as compared to HCV IRES and seems to be unaffected with interferon treatment (D). It is therefore tempting to speculate that this differential protein binding with the 5′UTRs might influence their respective IRES activity under interferon treatment. Since, our results suggest that the IRES activity of IRF2 is largely unaffected in presence of interferon α, we wanted to extend our study to investigate the IRES function under ER stress, which is known to cause translational shut down by eIF2α phosphorylation (). For this purpose, HeLa cells were transfected with either the bicistronic plasmid (pRIRF2F) containing IRF2 5′UTR or the non-specific bicistronic plasmid pRNspF containing a non-specific sequence (as negative control), followed by the treatment with tunicamycin (known to cause ER stress). Interestingly, under such condition, in the context of bicistronic plasmid, the cap-dependent translation of renilla luciferase activity was severely impaired in both pRNspF and pRIRF2F constructs as expected. However, IRF2-IRES-mediated translation of firefly luciferase was not significantly affected (A). To further investigate the effect in a closer natural context, we have used monocistronic constructs. In this experiment, monocistronic plasmid pCDIRF2FLuc was used where firefly luciferase reporter gene was cloned downstream of IRF2 5′UTR. As control of cap-dependent translation monocistronic plasmid pCDRLuc was used (B). When these plasmids were transfected in HeLa cells, both the monocistronic mRNAs generated were expected to be capped and translated by cap-dependent mode. However, the IRF2-FLuc RNA would have an option to switch to IRES-mediated translation when cap-dependent translation mode is affected due to ER stress (,). When pCDRLuc was transiently transfected in the HeLa cells followed by treatment with tunicamycin, the cap-dependent translation of Renilla luciferase activity was found to be inhibited drastically (∼70%) compared to control (B). However, only marginal decrease in the firefly luciferase activity (∼30%) was observed in presence of tunicamycin when pCDIRF2FLuc plasmid was used. Since we observed that the IRES activity of IRF2 was not significantly altered under tunicamycin treatment, we wanted to study the level of IRF2 protein expression under such condition. Western blot analysis also showed that there was no significant change in the level of IRF2 protein (A), although the level of TFIID protein was found to be diminished in presence of tunicamycin (A). Interestingly, the level of PTB protein was also not changed under such condition (A). We have demonstrated the requirement of PTB protein for IRF2-IRES activity in our earlier assay. Thus to investigate whether availability of PTB could affect the level of IRF2 protein under tunicamycin treatment, we opted for transient silencing of PTB followed by tunicamycin treatment. Equal amount of cell lysates were loaded for western blot analysis using antiIRF2 antibody. The result showed a dramatic decrease in the level of IRF2 under such condition (B), suggesting that the availability of PTB is important for the synthesis of IRF2 protein under tunicamycin treatment. The same blot was stripped and probed with anti-beta actin antibody (loading control), which did not show any appreciable difference (data not shown). Additionally, to monitor IRF2 synthesis we have performed pulse metabolic labeling experiment with the HeLa cells after tunicamycin treatment. This assay would not depend on the stability of the pre-existing protein but will reflect the rate of respective protein synthesis during stress condition. As expected, pulse metabolic labeling of cells treated with tunicamycin for different time periods (4 and 14 h) showed appreciable decrease in overall protein synthesis (C). However, immunoprecipitation of IRF2 protein of the above cell extract clearly demonstrated continued synthesis of IRF2 protein (D) reconfirming earlier observation. The position of IRF2 band was further confirmed by western blot analysis with the same IP extracts using antiIRF2 antibody (E). Taken together, these observations strongly suggest that IRF2 RNA has an IRES element, which is less sensitive to conditions that lead to shutdown of cap-dependent translation of majority of cellular mRNAs and allow basal level of IRF2 protein synthesis to regulate interferon and other cellular stress response. Also, a -acting factor, PTB is important for the IRF2-IRES activity under stress condition and its availability could be an important determinant of the efficiency of the respective IRES activity. Interferon (IFN) stimulates transcription of ‘interferon-stimulated response element (ISRE)’ containing genes by the activation of ‘interferon regulatory factors’ (such as IRF1, IRF7, etc.) (). This activation also results in increased transcription of these regulatory factors. Thus the effect of global attenuation of translation might not affect their protein level as much. However, it is important to modulate the activity of these IRFs as well to maintain the intricate balance and the cellular response. IRF2 protein has been shown to negatively regulate the interferon pathway. We hypothesize, that continuous synthesis of IRF2 protein by its IRES element probably ensures unaltered protein level under this situation, which in turn might regulate IFN-induced gene expression. In fact western blot analysis did show similar levels of IRF2 protein in control and interferon-treated cells in support of our hypothesis. In this study, we have shown evidence for a cap-independent translation initiation or IRES-mediated translation of IRF2 that probably allows a response to various stress conditions. Bicistronic assays were employed to show that the IRF2 5′UTR is capable of mediating internal initiation. Furthermore, using stringent assays we have tried to rule out cryptic promoter or splicing activity. Although, our northern blot and RT-PCR analysis clearly showed intact bicistronic RNAs in transfected cells, it is difficult to absolutely rule out the presence of lower abundance of mono-cistronic RNAs generated due to RNA degradation or spurious splicing activity in the context of RLuc/FLuc bicistronic construct as shown in XIAP IRES (). In general, IFN causes activation of PKR, which leads to the phosphorylation of eIF2-alpha culminating in the suppression of protein synthesis (). Our study shows that in the presence of interferon, when majority of protein synthesis is affected, IRF2 is still synthesized by IRES mode of translation. However, this is not true for all cellular IRES, since BiP IRES activity showed significant inhibition. Also HCV IRES was found to be sensitive to interferon treatment as reported earlier (). Interestingly, analysis of the cellular protein binding with IRF2 and HCV IRES did not show much difference in the profile in absence and presence of IFN-α. However, a 58 kDa polypeptide showed prominent binding with the IRF2 5′UTR only. It is not clear at this stage whether this particular protein is involved in IRF2 IRES function in IFN-treated cells. Additionally, the IRF2 protein level was also found to be unaltered in tunicamycin-treated cells, when majority of protein synthesis is expected to be inhibited during ER stress due to phosphorylation of eIF2α. This could be explained by the IRF2-IRES activity, which was found to be unaltered during the treatment. The observation put forward the idea that IRF2 protein might be involved in the cellular responses in other stress conditions as well. Several cellular IRESs have been shown to be active under conditions when eIF4G is cleaved. Incidentally, it has been shown that p50 (a central one-third part of eIF4G) and p100 (C terminal two-third fragment of eIF4G) can partially replace the function of the intact eIF4G in translation initiation mediated by the EMCV IRES (). In fact, the cleavage product of eIF4G as well as the eIF4G-related protein p97 and its cleavage product p86 has been shown to influence the IRES function (,). Interestingly, IRF2 5′UTR showed binding with polypeptides of ∼50 and 100 kDa when 2A protease-treated cell extract was used. Also in our assay, p86 was found to bind with IRF2 but not with HAV IRES. It appears that the binding of these polypeptides could give selective advantage to the IRF2-IRES activity over HAV IRES under conditions of eIF4G cleavage. PTB protein has been implicated to modulate functions of several cellular mRNAs. It has been shown to act as RNA chaperone and facilitate Apaf1-IRES structure and influence the efficiency of its translation initiation (). It has also been reported to positively regulate the IRES-mediated translation of HIF-1α, p27kip1, etc (,), whereas it negatively regulates unr and BiP IRES function (,). Our results suggest that PTB protein might be required for the efficient activity of the IRF2-IRES, since partial knockdown significantly affected the IRES activity. Interestingly, the cytoplasmic pool of PTB protein has been shown to vary under various physiological stress conditions, such as apoptosis and viral infection (,). Since it appears that the IRF2-IRES remains active under stress conditions and may be regulatable, it would be interesting to investigate how the abundance of PTB protein during such conditions could influence the IRES function. Currently, experiments are in progress to verify the MFOLD predicted secondary structure of the IRF2-IRES and how PTB binding can influence IRF2-IRES function during physiological stress conditions. Although, IRF2 protein is known as a negative regulator of interferon-stimulated genes, it has been implicated to stimulate vascular cell adhesion molecule (VCAM) and regulation of histone H4 genes under various conditions (,). The fact that the IRF2 gene is translationally regulated under stress conditions (such as ER stress and interferon treatment), it might have indirect effect on other gene expression as well. Thus, it is likely that the regulated expression of IRF2 protein under various stress conditions would have major implications on the cellular response. Incidentally, this study constitutes the first report on translational control of interferon regulatory factors by internal initiation. The results might have far reaching implications on the possible role of IRF2 in controlling the intricate balance of cellular gene expression under stress conditions in general.

RNAs are subjected to quality control mechanisms, known collectively as RNA surveillance, which remove transcripts arising through errors in DNA replication, transcriptional fidelity, ribonucleoprotein particle assembly or RNA processing. A major component of the cellular surveillance machinery is the exosome 3′->5′ exonuclease complex, which degrades unwanted transcripts both in nuclear RNA surveillance programmes and in translation-coupled events (). In addition, the exosome precisely generates the 3′ termini of mature RNAs such as 5.8S rRNA, small nucleolar RNAs (snoRNAs) that function in pre-rRNA processing and small nuclear RNAs (snRNA) that facilitate pre-mRNA splicing (,). The exosome also degrades the non-coding 5′ external transcribed spacer (5′ ETS) fragment that is released during pre-rRNA processing (), and functions in normal mRNA turnover pathways (,). Recent high-resolution crystal structures of exosome complexes from archaeal, yeast and human origin () have established that the conserved core complex comprises a hexameric ring of subunits homologous to RNase PH, with three further proteins containing S1 and KH RNA-binding domains located at equivalent positions on one face of the ring. This arrangement of the exosome core complex is analogous to that of PNPase (), a component of the degradosome complex that also functions in both mRNA degradation and RNA processing pathways in prokaryotes (). The core appears to play an important structural role, with all nine components of the yeast or human complex being required for structural integrity of the complex (,). The eukaryotic exosome contains the salt-labile 3′->5′ exonuclease, Rrp44p/Dis3p (), and the 3′->5′ exonuclease Rrp6p (,), neither of which are present in the archaeal exosome (). A hallmark of the exosome complex is that it requires additional proteins to facilitate its exonucleolytic activity in either RNA processing or RNA surveillance pathways. Exosome function in mRNA turnover and cytoplasmic mRNA surveillance pathways requires the associated putative GTPase Ski7p and the Ski complex, comprising the putative RNA helicase Ski2p, Ski3p and Ski8p (,). Nuclear RNA surveillance functions require a TRAMP complex, consisting of the putative RNA helicase Mtr4p, the poly(A) polymerase Trf4p or Trf5p, and the homologous RNA-binding proteins Air1p or Air2p (). Similarly, exosome functions mediated by Rrp6p are dependent upon the small basic protein Rrp47p/Lrp1p (). Rrp6p is a member of the DEDD family of 3′->5′ exonucleases () that includes enzymes that act on both RNA and DNA by a hydrolytic mechanism involving two divalent metal ions (). Rrp6p also contains a helicase and RNase D C-terminal (HRDC) domain that has been proposed to be an RNA-binding domain (,). However, Rrp6p itself does not show stable RNA binding (). In addition to the catalytic and HRDC domains, Rrp6p contains a C-terminal region that contains nuclear localization signals () and an uncharacterized N-terminal domain denoted PMC2NT (). was first characterized as a suppressor mutant of a temperature-sensitive (ts) allele of the gene encoding the nuclear canonical poly(A) polymerase (). Haploid yeast cells with an Δ null allele are viable but show slow growth and a ts-lethal phenotype. Strains lacking Rrp6p have defects in stable RNA synthesis, including the accumulation of 3′ extended processing intermediates of 5.8S rRNA, snRNAs and box C/D snoRNAs (,) and the accumulation of polyadenylated RNAs (,). In addition to its role in stable RNA processing, Δ mutants are also defective in nuclear mRNA surveillance pathways and in the retention of transcripts at the site of transcription (36, and references therein). Mutations of conserved residues in the catalytic domain of Rrp6p cause a loss of function and a cold-sensitive growth phenotype, while mutations within the HRDC domain show only mild RNA processing defects (,). The role of the N-terminal PMC2NT domain has not yet been addressed. The nuclear yeast protein Rrp47p/Lrp1p was identified as a component of the exosome that remained associated with Rrp6p-containing complexes through affinity chromatography purification. Northern blot hybridization experiments and microarray analyses on RNA from Δ strains demonstrated that Rrp47p is required at the same step in stable RNA processing pathways as Rrp6p (,). Subsequent studies revealed a similar relationship between Rrp47p and Rrp6p in nuclear mRNA surveillance pathways (,). The absence of Rrp47p has no significant effect on Rrp6p expression levels or the ability of Rrp6p to bind to the exosome (). The similar but weaker effects on stable RNA processing observed with Δ mutants, compared with Δ mutants, the non-additive phenotype of Δ double mutants and the lack of homology between Rrp47p and any characterized exonuclease led to the suggestion that Rrp47p specifically promotes the exonuclease activity of Rrp6p (). In contrast to Δ mutants, cells lacking Rrp47p do not show a ts-lethal growth phenotype. Therefore, Rrp6p has at least one Rrp47p-independent function that is required for optimal growth. While Rrp47p has been shown both biochemically and genetically to function together with Rrp6p, its precise function in RNA processing and surveillance pathways is not clear. Here we report that yeast Rrp47p is a nucleic acid-binding protein that specifically binds structured nucleic acids. Furthermore, we demonstrate that Rrp47p interacts directly with the catalytic exosome component Rrp6p via its N-terminal PMC2NT domain. Yeast strains expressing a truncated version of Rrp6p lacking the N-terminal region that contains the PMC2NT domain broadly exhibit the same growth phenotypes and stable RNA processing defects as Δ mutants, and fail to accumulate normal levels of Rrp47p. These results provide a mechanistic explanation for how Rrp47p promotes the activity of Rrp6p in degrading or processing structured substrates such as the 3′ extended ‘5.8S+30’ rRNA processing intermediate, snoRNA precursors and intergenic pol II transcripts. To express recombinant His-tagged Rrp47p, an NcoI restriction site was introduced at the initiation codon by PCR and the amplified gene was cloned into pRSETB (Invitrogen) as an NcoI–HindIII fragment (p238). The full-length recombinant GST–Rrp6p expression construct (p243) was made by successively subcloning BamHI–BamHI and KpnI–EcoRI fragments from the yeast expression construct pEG-p65 () (kindly provided by J.S. Butler) into pGEX-2T (GE Healthcare). Digestion of p243 with KpnI, treatment with T4 DNA polymerase, digestion with StuI and religation generated the rrp6Δ212-721 construct (p245). Digestion of p243 with BglII and religation generated the rrp6Δ42-450 construct (p246). To express the PMC2NT domain of Rrp6p as a GST fusion protein, the N-terminal region of encoding residues 13–102 was amplified from genomic yeast DNA by PCR and cloned into pGEX-2T as a BamHI–EcoRI fragment, yielding p249. GST fusions of extended variants of the PMC2NT domain (P176X, p256; L197X, p257) were generated by site-directed mutagenesis of p245 using the Quick-change mutagenesis kit (Stratagene). To express Rrp6p lacking the N-terminal region in , p243 was digested with KpnI, treated with T4 polymerase and digested with EcoRI. The released fragment was cloned into pGEX-2T after digestion with BamHI, treatment with klenow and subsequent digestion with EcoRI. The incurred frameshift at the BamHI–KpnI junction was then removed by deleting the 3′ C residue of the BamHI site by site-directed mutagenesis, as above, generating the rrp6Δ1-212 construct (p266). To express an mutant lacking the N-terminal region in yeast, the 1.8 kb KpnI–EcoRI fragment of was cloned in frame behind two tandem copies of the z domain of protein A and expression of the fusion protein was driven from the promoter (). Specifically, the KpnI–EcoRI region was recovered as a blunt end fragment from p243 by treatment with T4 DNA polymerase and the klenow fragment of DNA polymerase, and blunt-cloned into the EcoRI site at the end of the zz epitope tag, generating the construct p260 (zz-rrp6Δ1-213). To express the N-terminal region of Rrp6p in yeast, the BamHI–EcoRI fragment of plasmid p245 encompassing the truncated coding region was cloned into pEG-KT (). The wild-type, () and null mutants, and the epitope-tagged strain () used in this study were derivatives of BMA38 () (). The allele had previously been shown to be fully functional (). The -TAP derivative of BY4741 () was obtained from Open Biosystems. This strain showed vigorous growth at 37°C, demonstrating that the TAP-tagged Rrp6p protein is functional. The -TAP and strains were generated by PCR-mediated mobilization of the or allele and correct integration at the locus was confirmed by PCR. Yeast strains expressing the zz-rrp6Δ1-213 fusion allele, the corresponding, fully functional full-length Rrp6p derivative () or the -regulated GST-rrp6Δ212-712 fusion protein were generated by plasmid transformation of suitable strains. Yeast transformations with plasmids or PCR-amplified DNA were performed using standard techniques. Transformants were selected by growth on suitable selective media and integrants were screened by PCR on genomic DNA. Strains were routinely grown in minimal media containing 2% glucose or galactose, 0.67% yeast nitrogen base and appropriate supplements. For spot growth assays, the relative concentration of viable cells in saturated precultures was first determined by plating 100 μl of a 10 dilution on solid glucose-based medium lacking uracil and counting the number of colonies obtained. Ten-fold serial dilutions of standardized precultures were prepared and 5 μl aliquots were applied to the surface of solid growth medium and incubated at 26 and 34°C for 4 days. The strain BL21(DE3)LysS was transformed with plasmids encoding full-length or truncated GST-Rrp6p polypeptides and grown up at 30°C in LB medium containing ampicillin and chloramphenicol to an OD of 0.5. Transformants expressing His()-Rrp47p were grown up at 37°C. Expression was induced by adding IPTG to 0.5 mM and the cultures were incubated for a further 4 h before harvesting the cells. Cell lysates were prepared by sonication in 20 mM HEPES pH 7.6, 300 mM NaCl, 10 mM imidazole pH 7.6 and clarified by centrifugation at 15 000 for 30 min. Clarified lysates were mixed with pre-washed Ni-NTA superflow (Qiagen) or glutathione-sepharose (GE Healthcare) resin and after extensive washing with lysis buffer, the bound proteins recovered by elution in lysis buffer containing 250 mM imidazole or 20 mM reduced glutathione. His()-Rrp47p was further purified by ion exchange chromatography and gel filtration. The eluate from the Ni-NTA affinity chromatography was diluted 10-fold with 20 mM HEPES pH 7.6 300 mM NaCl to reduce the imidazole concentration and then mixed with SP-sepharose resin. Bound His()-Rrp47p was eluted with cell lysis buffer containing 500 mM NaCl. After concentration under vacuum, the eluate was resolved through a 25 ml superdex 200 column on an ÄKTA purifier system (GE Healthcare) using an elution buffer of 20 mM HEPES pH 7.6, 300 mM NaCl. Concentrations of protein samples were determined by Bradford assay. Recombinant protein-binding assays were performed by mixing lysates from cells expressing one protein with pull-downs of the partner protein on glutathione-sepharose or Ni-NTA superflow beads. After extensively washing the beads in lysis buffer, retained proteins were eluted, resolved by SDS–PAGE and visualized by staining with Coomassie blue G250 or transferred to nylon membrane and decorated with penta-His monoclonal antibodies (Qiagen) or anti-GST antiserum (Sigma). Yeast cell lysates were made in TBS containing 1 mM PMSF by glass bead extraction and clarified by centrifugation at 30 000 for 20 min. Yeast fusion proteins carrying two copies of the z domain of protein A from , alone or within the larger TAP tag, were analysed by western blot using peroxidase/anti-peroxidase (PAP) antibody (Sigma). A custom-ordered rabbit polyclonal antiserum raised against recombinant His()-Rrp47p (Eurogentec) was used at a dilution of 1:5000 for western analyses. Total cellular RNA was recovered from cells harvested at early log growth in glucose-based minimal medium by phenol/GTC extraction using glass beads (). RNA was resolved through 8% polyacrylamide gels, transferred to Hybond N membrane (GE Healthcare) and hybridized with radiolabelled transcript-specific oligonucleotide probes to detect specific RNAs. The probes used in this study were specific for the following RNAs: snR38 (oligo 272), gagaggttacctattattacccattcagacagggataactg; 5.8S (oligo 236), gcgttgttcatcgatgc; ITS2 (oligo 237), tgagaaggaaatgacgct; (oligo 242), aaggacccagaactaccttg. Band-shift assays were performed on His()-Rrp47p using EcoRI-digested pBluscript II KS (+) (Stratagene) or the synthetic homopolymeric RNAs poly(A) and poly(A)–poly(U) (Sigma). Twenty-microlitre serial dilutions of Rrp47p purified from cell lysates by Ni-NTA superflow chromatography were prepared in EMSA buffer (10 mM Tris–HCl pH 8, 20 mM KCl, 2 mM MgCl, 0.1 mM EDTA, 7% glycerol) and mixed with nucleic acids on ice for 20 min. One microlitre of 30% glycerol-containing bromophenol blue was then added and the bound and non-bound nucleic acid fractions were resolved by electrophoresis through 0.5% agarose gels in 0.5× TBE at 5 V/cm. Resolved nucleic acids were then transferred to Hybond N membrane and detected by hybridization with radiolabelled T7 primer (aatacgactcactataggg) to detect plasmid DNA, or with oligo(dT) to visualize poly(A). For the on-bead-binding assays, recombinant proteins or protein complexes were purified on Ni-NTA or glutathione-sepharose beads, as described above. The beads were then equilibrated in EMSA buffer containing 150 mM NaCl and incubated with radiolabelled ligand in 50 μl of the same binding buffer for 30 min on ice. After binding, the supernatant was removed and the beads were washed three times in binding buffer. The amount of ligand retained on the beads was determined by measuring the Cherenkov radiation in a Beckman Coulter LS650 scintillation counter. Ligands were labelled at their 5′ hydroxyl group with γ[P]-ATP using polynucleotide kinase. The ligands used were poly(A), an ∼40-nt-long synthetic oligodeoxynucleotide (o280, cggaacctcttgactttgagctctatgcacaattcagcttatc), a gel-purified 40-nt-long DNA marker (Promega), tRNA (Sigma) and a CIP-treated ∼0.6-kb long restriction fragment. Rrp6p homologous sequences were identified by BLAST searches of the EMBL non-redundant database using WU-BLASTp2 (). Multiple sequence alignments was performed using ClustalW () and viewed using Jalview (). Protein structure predictions were performed using PHYRE (). To purify yeast Rrp47p for binding assays, the protein was expressed in as a 25.5 kDa His()-tagged fusion and purified to near homogeneity from cell lysates by affinity chromatography using Ni-NTA superflow beads (A). The recombinant protein was further purified by ion exchange chromatography over SP-sepharose beads and subjected to gel filtration analysis. A single peak was observed on the gel filtration column trace (B) that corresponded to recombinant Rrp47p by SDS–PAGE analysis. The elution volume of Rrp47p gave an apparent molecular weight of 152 kDa (mean and median averages of 10 independent experiments), indicative of a hexameric complex. We conclude that Rrp47p is expressed as a multimeric protein complex that is most probably hexameric. The murine Rrp47p homologue C1D has been reported to have DNA-binding activity () and previous analyses suggested that yeast Rrp47p promotes the exonuclease activity of Rrp6p by facilitating recruitment of RNA substrates (). We therefore tested recombinant Rrp47p for DNA and RNA binding in a conventional electrophoretic mobility shift assay (EMSA). Incubation of plasmid DNA with Rrp47p caused a clear retardation at protein concentrations of ∼1 μM in the EMSA (A, upper panel). Increased protein concentrations generated lower mobility complexes (lanes 8 and 9), most probably due to multiple Rrp47p complexes associated with the DNA. No retardation was observed in control experiments where GST protein was incubated with plasmid DNA under the same conditions (Figure S1). We then used this DNA EMSA to test RNAs for their ability to act as competitor substrates. Incubation in the presence of poly(A) RNA had little inhibitory effect on the formation of protein–DNA complexes (A, lower panel), even at 200-fold excess over the plasmid DNA (the faster migrating material in lane 4 migrates with protein–DNA complexes seen in the absence of competitor, e.g. upper panel, lane 9). However, incubation in the presence of the double-stranded RNA poly(A)–poly(U) inhibited DNA binding completely. Titration experiments revealed that poly(A)–poly(U) was a very efficient binding competitor, negating DNA binding at approximately equimolar concentrations (B). Competitive inhibition was also observed in the presence of the natural heteropolymeric RNAs tRNA from and total cellular RNA from yeast (Figure S2). In contrast, single-stranded DNA oligonucleotides failed to compete for binding with Rrp47p (Figure S2). Having established that RNA is an effective competitor to DNA binding by Rrp47p, we then addressed the ability of the protein to interact with RNA directly. EMSA experiments were performed with increasing amounts of purified Rrp47p in the presence of poly(A) or poly(A)–poly(U) RNA and the poly(A) component was detected by northern hybridization using an oligo(dT) probe. Consistent with the competitive inhibition EMSA experiments, incubation of poly(A)–poly(U) with Rrp47p caused a reduced electrophoretic mobility at protein concentrations of ∼1 μM (C). As in the DNA-binding experiments, slower migrating complexes were observed at higher protein concentrations. No gel retardation was observed upon incubation of Rrp47p with poly(A) RNA. The EMSA results were corroborated using an on-bead-binding assay. Double-stranded DNA and tRNA were retained on beads coated with His()-Rrp47p, while there was no significant binding of poly(A) or single-stranded DNA oligonucleotide (Figure S3). We conclude that Rrp47p binds to double-stranded DNA or structured RNA with comparable affinity, with an apparent dissociation constant in the 1 μM range, and shows no detectable interaction with single-stranded nucleic acids. Rrp47p has previously been shown to be a component of the Rrp6p-containing exosome complex (,) but previous studies have not addressed whether the proteins physically interact. To demonstrate a putative direct interaction, Rrp6p was expressed in as an N-terminal GST fusion protein and assayed for its ability to bind His()-Rrp47p in pull-down assays (see Materials and Methods section). The His tag antibody detected an ∼25 kDa protein in pull-downs of lysates containing GST-Rrp6p and His()-Rrp47p (A). No signal was observed if lysate from cells expressing GST was used, or if bound GST-Rrp6p was incubated with lysate from cells transformed with the control vector pRSETB. Specific binding between His()-Rrp47p and GST-Rrp6p was also observed in the reciprocal pull-down experiment on Ni-NTA beads (data not shown). We conclude that Rrp47p interacts directly with Rrp6p in the absence of other exosome proteins. To map the Rrp47p-binding site within the 84 kDa Rrp6p protein, deletion constructs of Rrp6p were made (B) and their ability to interact with His()-Rrp47p was compared with that of full-length Rrp6p in pull-down assays. The GST-Rrp6Δ212-721 mutant contains the N-terminal region of Rrp6p without the catalytic domain, the HRDC domain and the C-terminal domain bar 12 residues at the extreme C-terminus. In contrast, the GST-Rrp6Δ42-450 construct contains the complete C-terminal domain and most of the HRDC domain, but not the catalytic domain and the N-terminal region bar the extreme N- terminus. His()-Rrp47p was retained on beads charged with full-length GST-Rrp6p or the GST-Rrp6Δ212-721 mutant (C, lanes 1 and 2) but no binding was observed with the GST-Rrp6Δ42-450 construct (C, lane 3). A GST-Rrp6p fusion lacking only residues 1–212 (GST-Rrp6Δ1-212) also failed to show Rrp47p-binding activity (D). Although the expression levels of the N-terminal deletion constructs were low, compared with fragments containing the N-terminal region, there was absolutely no detectable binding of His()-Rrp47p after prolonged exposure times while binding to the full-length construct was easily detectable. We conclude that the N-terminal 212 residues of Rrp6p are both necessary and sufficient for binding to Rrp47p . To analyse the N-terminal 212 residues of Rrp6p in more detail, homologues were identified by BLAST searches and the sequences aligned using ClustalW (Figure S4). All Rrp6p homologues contained a common N-terminal domain that is not found in other subgroups of the DEDD family of 3′->5′ exonucleases. Previous bioinformatics analyses () suggested that residues 13–102 of Rrp6p constitute an independent domain, denoted PMC2NT (shown schematically in ). Protein structure prediction programmes indicated that the PMC2NT domain is largely α-helical in nature, while residues 103–212 have low structural content. However, a recent structural analysis of residues 129–536 of yeast Rrp6p () revealed four short α-helices at positions 132–135, 162–165, 184–189 and 208–210 (denoted α1–α4 in and Figure S4). The multiple sequence alignment (Figure S4) indicates that the regions of the protein between α1 and α2 (residues 129–162 in Rrp6p), and around α3 (residues 179–197) contain a number of highly conserved residues. To assay whether the PMC2NT domain was sufficient for Rrp47p binding, we expressed residues 13–102 of Rrp6p as a GST fusion protein in and assayed the recombinant protein for Rrp47p-binding activity, as above. As shown in B, the PMC2NT domain of Rrp6p alone is sufficient for Rrp47p binding. However, increased levels of His()-Rrp47p were observed in pull-downs of Rrp6p fusion proteins that extend up to P176 or L197 (compare , lane 2 with lanes 3–5). This effect is not simply due to differences in the expression levels of the different Rrp6p constructs, since the fusion protein extended to L197 is expressed at lower levels than the PMC2NT fusion (B upper panel, lanes 2 and 4). We conclude that the PMC2NT domain is sufficient for binding with Rrp47p, and that residues between 103 and 197 contribute to this interaction. To address whether Rrp47p binding to Rrp6p and nucleic acid is mutually exclusive or can occur simultaneously, we carried out RNA- and DNA-binding assays on pull-downs of Rrp6p/Rrp47p complexes. Glutathione-sepharose beads were charged with either the N-terminal Rrp6p expression construct (GST-Rrp6Δ212-721) or GST and then incubated with lysate containing His()-Rrp47p or a control extract. The beads were then incubated with radiolabelled nucleic acids (tRNA or a DNA restriction fragment), the non-bound fraction removed and the amount of bound ligand quantified (). Binding of both tRNA and DNA was observed for the assembled Rrp6p–Rrp47p complex. No significant binding was observed in the absence of His()-Rrp47p, consistent with the lack of nucleic acid-binding activity of GST or GST-Rrp6Δ212-721 (Figure S1). Moreover, binding was dependent upon the presence of the truncated GST-Rrp6p fusion. We conclude that Rrp6p-associated Rrp47p can form a stable complex with structured nucleic acids. There is no available data regarding Rrp47p expression levels, although the levels of several other exosome components such as Rrp6p have been determined (). We therefore compared the expression levels of Rrp47p and Rrp6p proteins fused to the same epitope (two copies of the z domain of protein A) by western blot analysis. We also addressed whether the expression level of Rrp47p was dependent upon Rrp6p. Western blot analyses revealed that the expression levels of Rrp47p-zz and Rrp6p-TAP in yeast cells were comparable (A, compare lanes 1 and 3). The expression level of Rrp47p–zz was reduced to very low levels in the absence of Rrp6p (A, lanes 1 and 2). In contrast, the expression of Rrp6p-TAP was not significantly sensitive to the level of Rrp47p (A, and 25). We conclude that Rrp6p is required for the normal accumulation of Rrp47p. Since Rrp47p binds directly to the N-terminal region of Rrp6p ( and ), we reasoned that its interaction with Rrp6p was required for the accumulation of Rrp47p in the cell. To test this hypothesis, we determined Rrp47p-zz expression levels in yeast deletion mutants specifically lacking the N-terminal region of the protein () or comprising essentially just the N-terminal region (). The mutant was expressed as a zz tag fusion to allow a direct comparison of the relative expression levels of Rrp6p and Rrp47p proteins, while the mutant was expressed as a GST fusion to test for association with Rrp47p in pull-down assays. The GST-Rrp6Δ212-721 protein was expressed in yeast under the control of the inducible promoter, allowing repression during growth in glucose-based glucose-based medium and induction during growth in galactose-based medium. Rrp6p was expressed at similar levels as a C-terminal TAP tag or an N-terminal zz tag (B), and the N-terminal deletion did not significantly affect the expression level of the protein (C). As observed above, a very low level of expression of Rrp47p-zz was detected in the Δ strain (C, lane 1). Expression of full-length zz-Rrp6p resulted in the accumulation of a band of identical electrophoretic mobility to Rrp47p-zz (C, compare lanes 1 and 2). The observed levels of this band were comparable to that of zz-Rrp6p and a band of this size was not observed in strains expressing zz-Rrp6p as the only epitope-tagged protein (B, lane 2). In contrast, this protein did not accumulate in strains expressing Rrp6p that lacked the N-terminal region (C, lane 3). Rrp47p-zz levels in an Δ strain were significantly increased upon induction of GST-Rrp6Δ212-721 (D, compare lanes 5 and 6) but not in the vector control (D, lanes 3 and 4). We conclude that the N-terminal region of Rrp6p is necessary and sufficient for normal expression levels of Rp47p in yeast. An interaction between Rrp47p and the N-terminal region of Rrp6p in yeast extracts was demonstrated in pull-down assays. Lysate from strains expressing either GST or the GST-Rrp6pΔ211-721 fusion were incubated with glutathione-sepharose beads and the eluates analysed on western blots using the GST-specific antibody or a rabbit polyclonal antiserum raised against recombinant His()-Rrp47p. Rrp47p was detected in the pull-down from the strain expressing the GST-Rrp6Δ212-721 fusion but not the GST control (F). We conclude that the N-terminal region of Rrp6p is sufficient for Rrp47p interaction in yeast cell extracts. Yeast Δ mutants have a ts-lethal growth phenotype and show slow growth at permissive temperatures (), while Δ mutants exhibit a more moderate non-conditional growth defect. To analyse the contribution of the N-terminal region of Rrp6p to the function of the protein , we first assayed the growth of cells expressing the truncated Rrp6p protein. Yeast Δ strains transformed with plasmids expressing the mutant allele, the full-length zz-Rrp6p protein or the empty cloning vector were grown on selective solid growth medium or in liquid culture and their growth characteristics were compared with that of an isogenic Δ mutant and the parental wild-type strain. Spot growth assays revealed that both the full-length zz-Rrp6p protein and the truncated allele complemented the ts-lethal growth phenotype of the Δ mutant (A). To compare the growth phenotypes of the and Δ strains further, we directly measured their growth rates at 30°C, the standard laboratory growth temperature for , and compared them to Δ and wild-type strains (B). The doubling times of the wild-type strain and the Δ mutant complemented by the full-length zz–Rrp6p construct were 175 min and 163 min, respectively. Growth of the Δ and mutants was comparably impaired, with doubling times of 200 min and 223 min, respectively. The Δ mutant grew considerably more slowly, with a doubling time of 319 min. We conclude that the allele encodes a functional Rrp6p protein that supports growth at a level close to that observed upon loss of Rrp47p function. Yeast strains lacking Rrp6p or Rrp47p accumulate 3′ extended forms of 5.8S rRNA and snoRNAs (,,,). To test the role of the N-terminal region of Rrp6p in stable RNA processing pathways, total cellular RNA isolated from the mutant, the isogenic wild-type strain, a strain expressing full-length zz-Rrp6p and Δ or Δ null mutants was analysed by northern blot hybridization using oligonucleotide probes complementary to the mature 5.8S rRNA, 3′ extended 5.8S rRNA species, and the intron-encoded box C/D snoRNA snR38. The mutant exhibited a 5.8S rRNA processing phenotype that was comparable to that of Δ and Δ mutants (A, lanes 3–5), with a slight increase in the level of 7S pre-rRNA, the 3′ extended precursor to 5.8S rRNA, compared to the wild-type strain, and a large accumulation of the ‘5.8S+30’ processing intermediate. A much weaker accumulation of the band corresponding to the 5.8S+30 intermediate was observed in cells expressing the full-length zz-Rrp6p epitope-tagged protein (A, lane 2), demonstrating that the defect observed in the mutant lacking the N-terminal region is not due to the presence of the epitope tag. The mutant also accumulated snR38 species that were ∼3 nt longer than the mature RNA (denoted snR38+3) or extended to the 3′ splice site (denoted snR38-3′). A similar defect was also observed for U14 snoRNA (data not shown). The snoRNA processing phenotype of Δ mutants is more exacerbated than that of Δ mutants, as demonstrated by the complete block to final trimming of snR38 (B, compare lanes 3 and 4, and 25). The snR38-processing defect observed in the mutant matched that observed in the Δ null mutant. We conclude that loss of the N-terminal domain of Rrp6p elicits stable RNA-processing phenotypes that are consistent with a loss of Rrp47p function. However, loss of the N-terminal region of Rrp6p has a more severe effect on the protein's function than simply abrogating its interaction with Rrp47p. To test an earlier model that Rrp47p functions as an exosome cofactor by docking the Rrp6p enzyme onto the 3′ end of RNA substrates (), we assayed Rrp47p for its ability to bind RNA and to interact with Rrp6p. Recombinant Rrp47p was shown to bind concomitantly to both RNA and Rrp6p. The region of Rrp6p involved in Rrp47p binding was mapped to the functionally uncharacterized N-terminal PMC2NT domain () and we therefore addressed the contribution of this domain to Rrp6p activity . Rrp47p bound poly(A)–poly(U) and tRNA but did not bind poly(A) RNA, and bound to structured RNA or double-stranded DNA with comparable affinity in EMSA analyses. Rrp47p is not related in sequence to any characterized double-stranded nucleic acid-binding protein () and therefore represents a novel class of proteins with this activity. Analysis of the protein sequence using the structure prediction algorithms PHYRE () and PSIPRED () suggest that Rrp47p and its homologues consist of an α-helical N-terminal region, followed by an unstructured region of variable length which has a basic stretch of residues at the C-terminus (50% lysine or arginine residues in the final 22 amino acids of Rrp47p). Residues within this lysine-rich region might contribute to nucleic acid binding () but the nature of the specificity for double-stranded nucleic acids is unclear at present. Rrp47p may potentially form a collar around double-stranded DNA or RNA, analogous to some members of the six-membered ring ATPase family (). Consistent with our results, the Rrp47p homologue C1D has recently been reported to bind structured RNAs such as poly(G) and tRNA (). The RNA-binding activity of Rrp47p is most probably of key importance to its function as an exosome cofactor. The 3′ termini of 5.8S rRNA and snoRNA precursors that accumulate in Δ and Δ strains are predicted to be involved in imperfect double-stranded structures (,). In addition, recent genome-wide expression studies have revealed extensive transcription of both intergenic regions and antisense mRNA (), which could hybridize with mRNA to form stretches of double-stranded RNA. Such intergenic transcripts are nuclear exosome substrates () and are degraded in an Rrp47p-dependent manner (). Furthermore, recombinant Rrp6p is highly active on poly(A) or A/U-rich RNAs but degrades structured substrates rather poorly (,). Rrp47p has also been implicated in processes of DNA repair (,) and telomere length control (), which may reflect its ability to bind DNA. Previous estimates suggest Rrp6p is present at ∼2000 copies per cell (). Our western analyses revealed comparable expression levels of Rrp47p and Rrp6p. The organization of Rrp47p into hexameric complexes suggests that the majority of Rrp6p–exosome complexes in the cell do not contain Rrp47p, a notion supported by the stoichiometric levels of Rrp47p and Rrp6p in non-fractionated exosome complexes (). One possibility is that Rrp6p-containing exosome complexes lacking Rrp47p are required for the Rrp47p-independent function of Rrp6p that is important for optimal cell growth. Recombinant protein studies and analyses of yeast extracts showed that Rrp47p interacts with the ∼200-residue long N-terminal region of Rrp6p. While the PMC2NT domain of Rrp6p (residues 13–102) was sufficient for Rrp47p interaction, binding was improved with extended constructs. Sequence alignments and structural studies () reveal that this region includes a semi-conserved loop flanked by short α-helices (P132-L166). We propose that this loop structure promotes binding of Rrp47p to the PMC2NT domain, either by stabilizing the fold of the PMC2NT domain or by providing additional points of interaction. Furthermore, expression of an mutant lacking the N-terminal region correlated with loss of Rrp47p function by three independent criteria: (i) the depletion of Rrp47p expression levels; (ii) a reduction of cell growth rate; and (iii) the accumulation of 5.8S rRNA and snoRNA processing intermediates. C1D has recently been shown to interact with both RNA and PM/Scl-100 (), the human homologue of Rrp6p, demonstrating that this mode of exosome regulation has been conserved from yeast to humans. Deletion of the N-terminal region had a greater effect on Rrp6p function than simply preventing interaction with Rrp47p. Loss of the N-terminal region of Rrp6p may cause subtle structural defects within the protein; it has been proposed that this region may be important to anchor the HRDC domain to the exonuclease domain (). Alternatively, additional factors involved in snoRNA maturation such as Bcd1p () might also interact with this portion of Rrp6p. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

DNA in the cell nucleus is organized in nucleosomes by repetitive coiling ∼1.8 times around core particles consisting of two copies of each of four different core histones (H2A, H2B, H3 and H4). In higher eukaryotes a fifth histone, H1, is bound to the majority of the nucleosomes, bridging the DNA at the entry and exit of its coiling around the core particle. On average, one nucleosome is present for every 160–240 bp of DNA [the nucleosomal repeat length (NRL)]. The structure of the nucleosome is known in detail (), and several models have been proposed for folding of the nucleosomal filaments into fibers with a diameter of ∼30 nm (), which have been observed by electron microscopy. These models are based on experimental evidence obtained by a number of different techniques, using nuclear preparations, isolated chromatin and reconstituted oligonucleosomes. One of the main differences between the models is whether the DNA, which connects the nucleosomes (linker DNA), is straight or bent. Studies of chromatin in isolated nuclei have provided evidence for an organization of the filament into globular assemblies of nucleosomes (supranucleosomes or superbeads) () as well as for interdigitation of 30 nm fibers (,). Supranucleosomes have been proposed to be caused by dislocations in a cross-linker helical structure (), but no models have been developed to explain the geometry of double fibers. The conformational freedom of the nucleosomal filament, i.e. the number of different conformations it can attain, is limited by collisions between the nucleosomes, and is in first instance determined by the basic geometry of the filament, which is therefore of importance for the structure and function of the chromatin in the cell. One function concerns its role in transcription and transcriptional regulation, in which its dynamic nature () and structural plasticity () must play an important role. Another function concerns the putative architectural role of the chromatin in the nucleus, of which not so much is known. A long-standing question is to which extent the chromatin is a self-organizing polymer and to which extent it is being organized by other structures in the nucleus (,). It is not known whether the basic geometry of the filament by itself makes possible the formation of stable fiber associations. Provided that the linker DNA is straight—and there is evidence to suggest that this is so in the cell ()—conformational variations of the filament are mainly determined by two angles: the change in direction of the DNA every time it coils around a core particle and the angle between the flat faces of consecutive nucleosomes around their interconnecting linker DNA. This angle is often referred to as the rotational angle, because the positions of the core particles in the filament follow the right-handed rotation of the DNA double helix and therefore vary with the linker length. The conformational changes caused by variations in these two angles have been studied by computer models (), but existing models do not take into account the full consequences of the size of the directional change of the DNA at the nucleosome, the asymmetry of rotation around the linker DNA and the conformational limitations imposed by the linker length. The present study shows that these parameters are of major importance for the conformational freedom of the filament, and explores the possible conformations by means of physical models. It provides a survey of different conformations of the nucleosomal filament and how they are related to each other, as well as to existing models of the chromatin fiber. It is shown that repeated or periodic variations in β sequence give rise to different types of helical and looped conformations and that helical conformations are able to associate pairwise as a zip, to form double helices, intercalate and contract to a high density. Finally, the irregular conformation of a filament with a random β sequence is described as well as how it can be converted into a regular helix by chromatin remodeling and limited twisting of the linkers. Although physical models are static and do not account for the linker flexibility, they might be useful in studies of chromatin ultrastructure and serve as reference models for the analysis of transmission electron microscopic images. sub xref ext-link #text The conformation of a nucleosomal filament with straight linkers is determined by a number of geometric parameters, which are shortly introduced in the following (-I). The two most important parameters are the angle (α) between the entering and exiting linkers of the nucleosome, caused by the change in the direction of the linker DNA (δ = 180°−α) and the angle (β) between the flat faces of consecutive nucleosomes. According to the X-ray crystal structure of the nucleosome, the ends of the 146 bp of nucleosomal DNA are separated by ∼75° () corresponding to a default value of α = 105°. The size of α is sensitive to binding of one molecule of linker histone (H1) (,), while β varies with the length of the interconnecting linker. When α and β are constant the filament forms a regular fiber within a cylindrical space, with an axial symmetry determined by α (for β = 0°) or the projection of α (α) into a plane perpendicular to the fiber axis (for β ≠ 0°). Because of the slope of the DNA [γ = 4.5° ()] due to its coiling around the core particle, the two linkers diverge () by an angle (η), as defined by their projections into a plane of symmetry formed by the dyad axis and the symmetry axis of the nucleosome. If the linkers are bent so that their projections into this plane cross each other, they are said to be in a . Linkers have also been proposed to be able to approach each other as a result of an allosteric change of the nucleosome (nucleosome gaping) in which the coils of the DNA are separated like a local opening of a spring (). In the following, the significance of α and β for the conformation the nucleosomal filament will be demonstrated, assuming the linker DNA to be straight, entering and exiting the nucleosomes as tangents to their circular surface. The conformation of the filament is determined by the size of α when β = 0° (-II). At α = 0° the direction of the DNA is changing 180° (δ) at each nucleosome, placing every second nucleosome on top of each other as two columns parallel to the fiber axis, like the edges of a ribbon. The filament unfolds when is increased, forming a linear string of nucleosomes at α = 180°, and α can be increased either by directing the linkers toward each other () or away from each other (). A filament with linkers in the closed conformation uncoils clockwise with a concomitant decrease in the distance between the nucleosomes, as more of the linker DNA becomes associated with the core particles. Conversely, a filament with linkers in the open conformation uncoils counter-clockwise, increasing the distance between the nucleosomes as the terminals of the nucleosomal DNA are peeled off the core particles. Thus, at α = 180° the DNA coils twice around the core particles when linkers are in the closed conformation, but only once if linkers are in the open conformation. , i.e. the fraction of the NRL, which is available for keeping distance between the nucleosomes, thus depends on the size of α and the conformation of the linkers (open or closed). In the following, only the geometry of filaments with negatively closed linkers will be considered, if not stated otherwise. During uncoiling of the filament the two stacks of every second nucleosome rotate in opposite direction of the filament, arranging themselves in stacks parallel to the fiber axis at every value of α < 60° that equals the angle of a regular star, and at every value of α ⩾ 60°, which equals the angle of a regular polygon. With reference to this difference in symmetry fibers will be called for α < 60° and for α ⩾ 60° (compare -II, α = 36 and 108°). The helical symmetry of fibers formed by 0° < α < 180° is characterized by the number of 360° at the fiber periphery (in other words: how many nucleosomes it takes to traverse the fiber periphery 360°), as defined by the helical twist of the nucleosomes (coils/nucleosome). As one nucleosome is placed at the periphery of the fiber for every turn (δ) of the linkers, the fibers are also characterized by the number of , i.e. how many times the linkers have to turn 360° to place all the nucleosomes in a coil (turns/coil). Star and polygon fibers of the same axial symmetry have the same number of coils/nucleosome, but differ in the number of turns/coil. Thus, while filaments with α = 36° (δ = 144°) and α = 108° (δ = 72°) both have 1/5 coil/nucleosome, the former has two turns/coil (5 × 144° = 2 × 360°) and the latter has only one (5 × 72° = 360°). The number of sterically allowed values of β depends on whether or not every second nucleosome is able to pass each other side-by-side at β = 180° (-III). When this is possible, the conformation of the filament is changed by clockwise rotation from a helix at β = 0° to a ribbon at β = 180° with a concomitant 90° change in direction, because the pitch () of the dinucleosomes is reduced to zero. The conformation at β = 180° is possible only for combinations of α and linker lengths, which allow for alignment of every second nucleosome side-by-side, requiring a NRL of at least ∼224 bp for α = 60° and 200 bp for α = ∼85°. Clockwise rotation of the nucleosomes beyond 180° makes the filament coil left-handed in the opposite direction, but a full 360° rotation is not possible because of the change from left- to right-handed coiling, except for filaments shorter than a single coil (-IIIA). The size of β at which the change between right- and left-handed coiling takes place depends on α and the linker length. A star fiber with α = 36° and a NRL of 200 bp will only be able of right-handed coiling (-IIIB). The conformational changes of the filament induced by rotation of the nucleosomes were studied by wire models, prefolded into pentagonal star fiber and polygon fibers at β = 0°, followed by rotation of the nucleosomes in steps of 36° (). In the following, rotations are denoted according to the number of steps from β(±1) to β(±4) (+ for 0° < β < 180° and-for 180° < β < 360°) and β and β for 0 and 180°, respectively. The star fiber (α = 36°) is unable to attain conformations between ∼β(+3) and ∼β(−4), as this requires a NRL of at least ∼275 bp (A). Conformations of a star fiber with linker lengths that are sufficiently long to allow the nucleosomes to pass each other at β have been studied by a computer model (). In the polygon fiber rotation beyond 180° is allowed because of the larger value of α, causing a shift from right- to left-handed coiling (B). At β the polygon fiber has the form of a ribbon with α = η because the direction of the ribbon is perpendicular to the coiling of the DNA around the core particles. The ribbon is therefore not planar but coils right-handed. (For η = 2γ = 9°, the ribbon will have 1/20 coil/nucleosome and 9.5 turns/coil). As β gets near to 360° the polygon fiber approaches the so-called interdigitated model of the chromatin fiber, in which every fifth nucleosome stacks upon each other, forming five left-handed coils of stacked nucleosomes () (except for the linkers not being bent in the present model). A NRL > 200 bp is required for the conformation at β(−1), as shown for two coils of the filament in C. Provided that the linkers are long enough, α can be decreased in the polygon fiber at β to generate a ribbon of stacked nucleosomes in which the orientation of the linkers alternates between positive and negative conformation (D). This type of ribbon can be wrapped on the surface of a cylinder to form a fiber in which linker DNA zigzags up and down the helical axis, and such conformations have been described by two models as the twisted ribbon () and the helical ribbon (), which are derived from filaments having an open and closed linker configuration at β, respectively (data not shown). The axial symmetry of the fiber is changed when |β| is increased from 0° due to the decreasing size of α. The axial symmetry of a polygon fiber is thus changed into that of a star fiber at α < 60°. If, furthermore, the size of α is reduced, this will cause the fiber to become more compressed. The pentagonal polygon fiber can thus be compressed at β(+3) into the triple helix model (), named after its three left-hand rotating columns of every third nucleosome (data not shown). At β(+4) α can be reduced to ∼80°, accompanied by intercalation of the every second nucleosome and a slight rotation, until the linkers are aligned side-by-side in the interior of the fiber (E). This is an example of the double-helical crossed linker model (), which has recently been found to correspond to the crystal structure of the tetranucleosome (). Much less compaction is possible at β(−4) (F). The asymmetry of clockwise and counter-clockwise rotations can be seen by considering a dinucleosome with a defined direction of the DNA from nucleosome no. 1 to no. 2, in which nucleosome no. 2 is rotated either clockwise or counter-clockwise around the interconnecting linker from β = 0° to 180° (-IIIC, D). In both cases the pitch () of the dinucleosomes is reduced to zero. However, as the entry and exit sites of the terminal linkers are not diametrically positioned relative to the axis of rotation at β = 0° the changes in are different for clockwise and counter-clockwise rotation. Thus, by clockwise rotation goes through a slightly negative minimum, and by counter-clockwise rotation, increases to a maximum before reaching zero at β = 180° (-IIID). For the same value of |β| the axial symmetry of nucleosomal filaments therefore differ for β(+) and β(−), the former being more compact than the latter, which has a larger diameter with a hole in the middle (G). Another asymmetry of clockwise and counter-clockwise rotation is due to variations in the angle (η) between the terminal linkers of the dinucleosomes, as defined by their projections into a plane perpendicular to the interconnecting linker (-IIID). As the two linkers are not parallel to each other at β = 0°, but diverge by a small angle (η), ∣η∣ will be larger for clockwise rotations (η + β) than for counter-clockwise rotations (η − β) of the same size. The difference in η is greatest for small values of β as seen from the difference in axial symmetry of fibers with η = 10 and 0° (G). The wire models could only be made with straight linkers leaving the nucleosomes as tangents to their curved surface, because bent linkers cannot be folded with any precision. As the linkers in the nucleosomal filament are in fact bent as a result of the binding of histone H1, this represents a major simplification of the geometry. Thus, the size of α is to a large extent determined by binding of H1, which bridges the linkers close to their entry/exit sites and inhibits their fluctuation (). This makes the entry and exit sites approach each other, so that the DNA makes two full coils around the core particle (as for α = 180°, see -IVA), thereby decreasing the effective linker length. However, binding of H1 stabilizes at a value below 180° by bending of the linkers close to their entry/exit sites, causing the nucleosomes to rotate away from each other around the linker entry/exit sites (-IVB). A conformation of the linkers at α = 0° has been reported, in which the two linker DNA segments are juxtaposed ∼8 nm from the nucleosome center and remain apposed for 3–5 nm before diverging, forming a so-called stem motive () (-IVC). As the stem takes up 25–30 bp of the linker, a NRL of more than ∼200 bp is required to accommodate a stem at each nucleosome plus an additional linker DNA segment between the stems to provide a pitch of the dinucleosome ≠ 0° and to make α > 0° at the entry/exit of the stem. Bending of the linkers at their entry/exit sites by binding of H1 may also change the slope of the linker DNA, reducing the pitch of the dinucleosomes (-IVD and E). Moreover, bent linkers may introduce a tilt angle between the flat faces of the core particles (-IVF). Although the wire models do not reflect the path of the linker as affected by histone H1 they can, by minor adjustments of NRL, η and η, imitate changes in the distance between the nucleosomes that may be caused by effects of linker histone binding on the effective linker length and on the pitch of the dinucleosomes. Repeated β sequences along the nucleosomal filament give rise to a large number of different conformations, some of which are described below using the following nomenclature: Values of β are numbered according to the convention in , β(−1) − β(−4) being underlined and in italics. β Angles without specified values are indicated only by (+) and (−) and nucleosomes are indicated by a dot, when convenient. A fiber with alternating positive and negative values of β can thus be written as (+−) or (•+•−)•. Star- and polygon fibers have been folded with α = 36 and 108°, respectively, unless stated otherwise. Star fibers retain the axial symmetry of a star fiber for all sterically allowed combinations of β(+) and β(−) and little variation in β is allowed owing to the small value of α. Thus, even in a filament with as little variation in β as every fifth nucleosome is touching each other (A). Regular helices are formed by polygon fibers with alternating β(+)- and β(−)-linkers, which make the filament shift repeatedly between forward and backward direction. Fibers with β sequence and can be compacted into short tubes by very small changes in α and β (B and C), while the coiled ribbon at cannot be compacted (D). Another type of helix, which can be circumscribed by an ellipsoid cylinder, is formed by the triple repeat (approximate values). This fiber can be slightly condensed by minor adjustments of the size of η (E and F). The shift between forward and backward direction in the polygon fiber, caused by β(+) and β(−), leads to a different type of fiber when the sign of β is changed at every second linker, as shown in G. The occurrence of β(−) linkers in the repeated β sequence thus causes the filament to change direction at every second nucleosome, forming a fiber with a 3-fold axial symmetry and partially overlapping loops of six nucleosomes (•1•1•••1•) (numbered 1-2-3-4-1-2 in H). Only a single nucleosome in each loop (No. 4) protrudes from the central axis, which consists of groups of three nucleosomes, formed by contacts between one nucleosome from each of three consecutive loops. These nucleosome triplets are located very close to each other and may form dimers. Another conformation, caused by the same periodical shift between β(+) and β(−), but with larger values of β,, consists of large loops of ∼20 nucleosomes (I). Linkers with an orientation close to that of the fiber axis have the potential of functioning as a swivel between intermittent coils. In the presence of such ‘swivel-linkers’ coiling of the intermittent nucleosome sequences becomes independent of each other, making possible a shift between right- and left-handed coiling, such as (A) and (data not shown). In some cases, the presence of swivel-linkers makes possible a reduction of α without rotation of the fiber. This is shown by the sequence in which the ()-linker functions as a swivel-linker, connecting two right-handed trinucleosomes (•3•3•), forming a coil of 6 nucleosomes spanning a little more than 360° (because η of the swivel-linker is larger than 108°) (B). Reduction of α at the terminals of the swivel-linker makes the trinucleosomes approach each other, and reduction of α in the trinucleosomes to ∼72° makes them close like a pair of pincers (C and D). As a result of these changes every second linker becomes super positioned (E), allowing for compaction of the fiber (F). The compaction is limited by collision of nucleosomes connected by every second swivel-linker, which stack edge-to-edge in an orientation close to that of the fiber axis (blue arrows in F). The fiber therefore has ∼6 nucleosomes per 11 nm. If the volume of the fiber is calculated as a cylinder with a diameter equal to two times the radius of gyration (32 nm), this corresponds to a density of 0.15 g DNA/ml, which is not different from the density of the solenoid model (). However, the cross-section area is more closely circumscribed by a rectangle of ∼28 × 26 nm, corresponding to a density of 0.18 g DNA/ml. These values are within the range of densities calculated from metaphase chromosomes in a variety of organisms (). The compact fibers can associate by interdigitation between nucleosomes, which are oriented perpendicular to the fiber axis (red arrows in F) forming sheets, which can be packed into a slightly higher density (G and H). Separation of coils in a fiber by swivel-linkers makes space for interactions with other fibers. Two fibers with the sequence can align side-by-side and interlock like a zip, in which nucleosomes in each fiber fit spatially into a pocket of five to six nucleosomes (•3•3••3•3•) in the other fiber (A). The minimum size of required for the formation of this type of double fiber is between 108° and 90°. In a fiber with β sequence, () swivel-linkers separate right-handed coils of four nucleosomes (•3•0•3•), providing sufficient space for the formation of a double helix in which swivel-linkers in one filament is alternately located on the inside and outside of swivel-linkers in the other filament (B). This type of double fiber can be compressed to ∼70% of its original size by reducing α to ∼90°. Double fibers can also be formed by intercalation of every second nucleosome in filaments with irregular β sequences. This type of double fiber can be considered to be derived from the inter fiber contacts between two filaments at β, in which every second linker is exposed on each of the flat faces. If two such ribbons are placed one on top of the other with the nucleosomes being stacked, every linker in one ribbon has two symmetrically positioned crossover points with two linkers in the other ribbon (E). If the two ribbons are staggered in parallel with the axes of the ribbons, the linkers slide along each other, one crossover moving toward the middle of the linker, the other moving toward the terminal region of the nucleosomal DNA. The nucleosomal DNA of every second nucleosome in one filament will thus have a crossover point with the same region at every second nucleosome in the other filament (F–G). Rotation of the nucleosomes counterclockwise from β makes the two filaments coil left-handed around each other, maintaining the crossovers by intercalation. The ability to form double fibers by intercalation (C,D) is not limited to filaments with closed, negative configuration. In H and I is shown an example of a left-handed double helix with α varying between ∼90° and ∼108°, and ∼β(+3). This type of double fiber is formed by filaments with alternating positive and negative linker configurations, which are folded from filaments with an open linker configuration at β (J). The ability of the filament to form longitudinal loops of six nucleosomes (five linkers) with the β sequence (G and H) raised the question of whether loops could be formed by other pentameric series of β. This was addressed by folding of loops with six nucleosomes (α = 90°), using β = 0°, +90°, 180° and −90°, followed by adjusting the β values to integers of 36°. Many models of such loops could only be folded from filaments with β, and β values different from β often had to be introduced in a specific folding sequence to avoid collisions between the nucleosomes. The loops could be classified according to their opening in response to a linear stretch of the filament, into closed loops, which could be opened after a slight displacement of the terminal nucleosomes, and locked loops, which could only be opened after rotation of the terminal nucleosomes (A and B). The loops were relatively independent of the actual β values and could exist in several different conformations after adjusting β to integers of 36°. Thus, the closed loop (+−+−+), which is shown on C with the β sequence, underwent no major changes in compaction when the β values were changed corresponding to a stepwise increase in linker length up to 10 bp (data not shown). Many of the loops could not be serially connected as the terminal linkers were located too close to each other or to neighboring nucleosomes to provide sufficient space, and an extra nucleosome therefore had to be added to each terminal linker (compare D and E). Some of these octanucleosomes had highly compact conformations with crossed linkers and linkers and nucleosomes touching each other, as in the fibers shown in F and G). Thus, in the octamers in G, 6 of the 7 linkers are either crossing another linker or touching a nucleosome. Random variations in β give rise to irregular conformations which depend on the mean β value, the degree of deviation from the mean value as well as the actual β sequence. NRLs for a large number of different organisms and cell types have been shown to be quantized by integral multiples of ∼10 bp, corresponding to integral multiples of helical turns of linker DNA, with experimentally determined SDs of ∼2–4 bp (). It therefore seemed to be most relevant to explore conformations of polygon fibers with β varying up to β(±3) around a mean value of β. Such fibers could not be folded with high precision, and were therefore made in triplicate in order to account for minor inaccuracies, which will inevitably affect the mutual orientation of nearby nucleosomes. shows a model of a polygon fiber consisting of 65 nucleosomes (α = 90°) with the β sequence shown in . Analysis of three models of this fiber showed the presence of clustered loops of 4–13 nucleosomes (). One class consisted of loops formed by nucleosomes, which tended to stay in touch in all three models (black in D-F). In the model in , these nucleosomes were even not separated by a mild stretch of the fiber, thus forming relatively stable domains. A second class consisted of loops formed by nucleosomes, which were more unstably associated (gray in D-F), being separated by even a mild stretch of the fiber and showing a more variable mutual localization in the three models. Both types of clusters were connected by more flexible segments of the filament with nucleosomes, which were generally not in touch with other nucleosomes (white in D-F). Because of a high flexibility mainly in these regions, the fiber easily formed loops by contacts between nucleosomes at the ends of the fiber. The loop shown in C was stabilized by nucleosome No. 65 fitting into a pocket formed by the first five nucleosomes of the filament, analogous to the zip-motive of the double fiber shown in A. Given the irregular conformation of fibers with random variation in β, it was finally asked whether such fibers could be transformed into more regular fibers, especially the type of fiber, which can be compacted to a high density (F). Two properties of the filament made this possible. First, the displacement of core particles along the DNA changes the two flanking β values in opposite direction by the same amount. Secondly, twist constraints of the linkers are negligible for variations of β corresponding to maximally ±2 bp (). In G–I and it is shown how the fiber can be changed by displacements of core particles to generate a β sequence which, by twisting of the linkers maximally ±2 × 36°, can attain the β sequence, which can be compacted. Changing the conformation in this way step-by-step from one end of the filament to the other provides a simple model for the spread of heterochromatin. As is also the sequence required for formation of one type of double fiber (A), combinations of core particle displacements (chromatin remodeling) and limited linker twisting may thus provide the β sequences necessary for stable inter fiber associations. A major result of the present study is the demonstration of the significance of α for the conformational freedom of the nucleosomal filament. When α is smaller than 60° the filament thus remains a helix with little space for variations in the β sequence, while larger values of α provide it with a freedom to attain conformations, which are not dependent on a helical symmetry, by allowing for an increased variation in β sequence. The size of α in chromatin is not known with certainty, but α has been shown by cryo-EM imaging of native chromatin fibers (,) to decrease from 85 to 35° when the salt concentration is increased. These relatively small DNA entry–exit angles have been suggested to be caused by the stem motive, which has not been observed by scanning force microscopy (SFM). Studies by SFM of chromatin fibers fixed at low ionic strength have yielded average values of α of 100 ± 40° in the presence and 130 ± 40° in the absence of linker histones (), and unfixed mononucleosomes at 10 mM MgCl showed two maxima of α at around 80 and 136° in the absence of H1 and one maximum at 85–88° in the presence of H1 () (mean values differed slightly for samples scanned in air and liquid). These large values of α thus correspond to a polygon fiber geometry. An important property of the polygon fiber is the length expansion at 0° < β < 180° (in contrast to the star fiber, which contracts) and the more than 90° change in direction at β > 180°. These properties are essential for creation of swivel-linkers and for transformation of helical coils into loops by reversing the pitch of the filament. Swivel-linkers can make space for close contacts between two fibers by separating two coils in a filament, thereby forming a pocket of 5–6 nucleosomes in which a nucleosome from another filament can bind. Tandem repeats of the β sequences of such pockets allow for the side-by-side association of two fibers like a zip, and the separation of helical coils by swivel-linkers can also provide space for the formation of double helices. It is tempting to speculate that tandem repeated DNA sequences, which often contain regularly positioned nucleosomes, might be stabilized and compacted by such types of fiber interactions. Another type of double fiber, which does not involve regularly spaced nucleosomes, is represented by the left-handed coiling of two polygon fibers around each other by intercalation of every second nucleosome (C). The positions of close contacts between the intercalated nucleosomes in this type of fiber appear to fit well with the positions of alternating asymmetric protection of nucleosomes in isolated nuclei against digestion by DNase I (), which were shown to give rise to a dinucleosomal DNA repeat pattern (). However, this type of double fiber may not provide the high degree of protection of the linkers, as inferred from analysis of the cleavage pattern, which was explained as a result of intercalation of nucleosomes in helical crosslinker model with linkers being buried in the interior of the fiber (). Although the interdigitated model may not be able to fully explain these observations, the polygon fiber geometry has the potential to provide the chromatin with architectural properties by facilitating the formation of double fibers, thus enabling the filaments to form 3D networks. A special property of the helical conformation formed by β sequence is the ability to compact to a high density by a decrease in α, followed by interdigitation. The extent of compaction (∼6 nucleosomes/11 nm), as well as the radius of gyration (32 nm), agrees with results of neutron scattering and scanning electron microscopy of isolated chromatin at elevated ionic strength (), but the volume of the fiber is smaller because the cross-section area is smaller, and the shape of the fiber allows for interdigitation into sheets, which can be packed to a density corresponding to metaphase chromosomes. This high density is so far only exceeded by two other models in which the compaction was caused by bending of the linkers () and by nucleosome gaping (,). Energy analyses have indicated that the balance between repulsion among linker DNA and internucleosome interactions determines the salt-dependent condensation of the nucleosomal filament (), and positively charged core histone N-terminal tails are important in this process by mediating favorable internucleosomal interactions and screening DNA repulsion (). Compaction of the fiber would seem to require a high degree of charge neutralization, as linker DNA and nucleosomal DNA are brought very close to each other, while face-to-face internucleosomal interactions are only involved in interdigitation of the fibers and in apposition of layers of interdigitated fibers. Since a regular spacing of nucleosomes is rare, it is noteworthy that repeated β sequences required for double fiber formation and compaction can also be formed by fibers with a random distribution of β values after displacement of nucleosomes and limited twisting of the linkers. This supports speculations that the nucleosome repositioning activity of chromatin remodeling factors may provide the chromatin fiber with dynamic properties at the supra-nucleosomal level (). It furthermore shows that variations in β may not only be caused by differences in linker length but also by twisting of the linker. There is therefore no unambiguous correlation between β and linker length. In addition, linkers of the same length can differ slightly in their β values owing to sequence-dependent variations in the helical twist angle of base pairs in the linker DNA. An important consequence of the polygon fiber geometry is the ability of the filament to form loops in response to variations in β. Closed and locked loops formed by periodic variations in β, β(+), β and β(−) remained compact at many different combinations of β values, and the flexibility of the filament was caused only by the interconnecting linkers. The fact that some of these types of loops required the addition of one nucleosome at each terminal linker to provide space for being serially connected is reminiscent of the supranucleosomal organization of the chromatin. The size of supranucleosomes isolated by nuclease digestion thus varies from a minimum of 6 or 8 nucleosomes up to 40 nucleosomes (). Filaments with random variations in β sequence also seem to be able to attain a supranucleosomal organization by the formation of relatively stable loop clusters separated by more flexible segments. It is generally recognized that the nucleosome filament is highly dynamic owing to limited twisting of the linkers within the twist constraints and a rapid exchange between free and bound histone H1 (). The major significance of static models may well be to show conformations that have the potential to be stabilized, either by double fiber formation, by the presence of closed and locked loops, or by cross linking of nucleosomes by nuclear proteins (). The stability of loop clusters in filaments with a random distribution of β may thus be increased by cross-linking of nearby nucleosomes. In light of the importance of α, β and the linker length for the conformation of the nucleosomal filament, what is then the role of these parameters for the organization of chromatin in the nucleus? A discussion of this question has to take into account that the geometry of the linker can be affected by bending. Stretches of poly(dAdT) are often bent and have an increased likelihood of being present in linkers () and linkers may also bend as a result of binding of H1. Moreover, theoretical calculations do not exclude that electrostatic interactions with core histone tail domains () can bend the linker DNA, despite the known large persistence length of the DNA (). Mesoscopic computer models have indeed shown that when chromatin fiber models include real DNA flexibility and charge, then the internucleosome interactions overcome the pre-designed linker DNA geometry and result in establishing new nucleosome chain configurations. However, the internucleosomal interactions appeared to be relatively weak and probably insufficient to maintain highly bent linker DNA (). Bending and twisting of linker DNA may be the result of aggregation of nucleosomes (), which have a propensity to aggregate at high concentration and in high salt, and the sequence of aggregation of newly synthesized nucleosomes to their nearest neighbors, might in fact function as a folding path for nucleosomal filaments in the nucleus. The potential importance of a folding path is seen from models of fibers consisting of hexanucleosomal loops, some of which needed an extra nucleosome at the terminals to give space for being serially connected. Such sequences of eight nucleosomes could not be folded from preformed filaments and are therefore unlikely to be generated from pre-existing filaments as a result of a thermodynamic equilibrium. The narrow distribution of NRLs in every cell type and organism around a single mean value suggests that there is a range of preferred β values in the chromatin. NRLs for a large number of different organisms and cell types have been shown to be quantized by integral multiples of ∼10 bp, corresponding to integral multiples of helical turns of linker DNA (), suggesting that preferred β values might be the same for these organisms. However, as the NRL can be changed experimentally by up to 15 bp without fatal effects (,), it seems that chromatin higher order structure can accommodate to a wide range of β values. This is in fact the case for the closed and locked hexanucleosomal looping of the filament, which depends more on the distribution of (+)- and (−)-linkers than on the actual values of β(+) and β(−). While looping appears to be an inherent property of the filament, occurring even as a result of a stochastic distribution of linker lengths, the location, conformation and length of the loops depend on the actual β sequence. This suggests that for a supranucleosomal organization of the chromatin, changes in linker lengths have a modulating effect on the conformation of the filament, leaving some loops unaltered, changing the conformation of others, destroying some loops and creating new ones, thus leaving the principal structure of the fiber roughly unaltered and within the same average dimensions. The models predict that this will be different for conformations, which are more dependent on a specific β sequence, as expected for regular fibers and for some types of double fibers with repeated β sequences. Since these types of fibers have a potential architectural role in the large-scale organization of the chromatin, changes in β within these segments of the filaments might have fatal effects. An invariable placement of nucleosomes in these regions may therefore be more important than in others. Maintenance of the higher order organization of the genome might thus depend on the existence of series of ‘architectural’ nucleosomes located in invariable positions. Such nucleosomes might possibly belong to the most firmly bound nucleosomes whose positions are encoded by the DNA ().

All oligonucleotide probes and targets were obtained from Integrated DNA Technologies, Inc. (Coralville, IA, USA). Antisense Firefly luciferase (pGL3-Luc 235-252, Promega) and antisense c-myc (564-581, GenBank Accession V00568) MBs were labeled at the 5′-end with a CAL Fluor® Red 610 (Biosearch Technologies) fluorophore, Cal610, and at the 3′-end with Iowa Black RQ quencher, IBRQ. Anti-luciferase MBs were used as a control sequence for these studies because this sequence is not complementary to any endogenous RNA target in MCF-7 cells. For studies requiring quantum-dot conjugations, a biotin-dT group was incorporated in the 3′-stem. The anti-luciferase MB employed was Cal610-CTCAGCGTAAGTGATGTC-IBRQ. The c-myc antisense MB employed was Cal610-GTGAAGCTAACGTTGAGG-IBRQ. Luciferase and c-myc target DNA oligonucleotides were synthesized with the sequences, GTCACGACATCACTTACGCTGAGTTT and GTCACCCTCAACGTTAGCTTCACTTT, respectively. An antisense c-myc 2′--methyl RNA oligonucleotide was synthesized with the sequence mGmUmGmAmAmGm CmUmAmAmCmGmUmUmGmAmGmG. QD–MB conjugates (QD–MBs) were prepared using biotinylated MBs and QD streptavidin conjugates (QD800, Invitrogen). Specifically, 10 μM samples of the MB were incubated with 1 μM QD streptavidin conjugates at molar ratios of 5:1, 6:1, 10:1, 15:1, 20:1 and 100:1 in 50 mM sodium borate, 0.05% Tween, pH 8.3 at 4°C overnight. The specified molar ratios were obtained by adjusting the volumes of the MB and QD sample accordingly. QD–MBs were purified from unbound MBs by gel chromatography (Superdex, GE Healthcare). The concentration of the purified QD–MB was determined by measuring the absorbance of QD800 (ε = 8 000 000 cmM) on a Cary100 spectrophotometer (Varian). Throughout the text the concentration of the QD will be used to represent the QD–MB concentration. QD–MB conjugates possess an average diameter of 15–21 nm (). Estrogen-responsive MCF-7 cells (ATCC, Manassas, VA, USA) were cultured in minimum essential medium (Eagle) with 2 mM -glutamine and Earle's BSS adjusted to contain 1.5 g/l sodium bicarbonate, 0.1 mM non-essential amino acids, 1 mM sodium pyruvate and 10% fetal bovine serum (FBS) in 5% CO at 37°C. It should be noted that estrogen was not removed from the FBS. Also, phenol red was present. It has been reported that phenol red can mimic the action of estradiol at a typical concentration of 30 μM (). For all live cell-imaging experiments (described subsequently), the MCF-7 cells were seeded on a 60 mm petri dish at a confluency of 10–30%. In experiments where the effect of tamoxifen on c-myc expression was evaluated, MCF-7 cells were treated with 1 μM tamoxifen for 48 h prior to microinjection and imaging of QD–MBs. All cell experiments were carried out on cells passaged less than 20 times to ensure that they do not acquire resistance to tamoxifen (). All microscopy measurements were performed on an Olympus IX 81 motorized inverted fluorescence microscope equipped with a Sensicam (Cooke) monochrome digital camera, an X-Cite 120 excitation source (EXFO) and Sutter excitation and emission filter wheels. Automated image acquisition was carried out using IPLab (BD Biosciences). Images of Cal610 and QD800 were acquired using the filter sets (HQ560/55, HQ645/75, Q595LP) and (e460spuv, D800/50, 475dcxru) (Chroma), respectively. A LUC PLAN FLN 40× objective (NA 0.6) was used for all cell imaging studies. Fluorescent standards for comparing day-to-day fluctuations in the fluorescent intensity of the microscope set-up were generated by microinjecting samples containing 100 nM MBs and 1 μM complementary oligonucleotide targets in QMB buffer into paraffin oil. The MBs were hybridized to their complementary targets for 24 h prior to the first day of microinjections. The same sample was used for all fluorescent measurements over the course of the study. Fluorescent images of the water-in-oil bubbles were acquired using the Cal610 filter set. These images were subsequently analyzed with NIH Image J. Specifically, a region of interest (ROI) was drawn around each bubble and the total fluorescent intensity was measured. Similarly, the total fluorescence intensity from an equal sized ROI that was drawn around a ‘background’ region was also measured. The background subtracted fluorescence measurement for each bubble was then calculated, . Measuring the diameter of the bubble and assuming a spherical geometry allowed the volume, , of each bubble to be calculated. The value / was used to compare instrumental fluctuations in fluorescence intensity. c r o i n j e c t i o n o f M C F - 7 c e l l s w a s c a r r i e d o u t u s i n g a F e m t o J e t a n d I n j e c t m a n N I 2 ( E p p e n d o r f ) m i c r o i n j e c t i o n s y s t e m f i t t e d w i t h F e m t o t i p s I ( E p p e n d o r f ) . P r i o r t o u s e , F e m t o t i p s w e r e t r e a t e d w i t h H e x a m e t h y l d i s i l a z a n e ( F l u k a ) f o r 1 0 m i n , f o l l o w e d b y r e p e a t e d w a s h e s i n Q M B b u f f e r . sub #text S a m p l e s c o n t a i n i n g 1 0 0 n M Q D s i n Q M B b u f f e r w e r e m i c r o i n j e c t e d i n t o M C F - 7 c e l l s . T h e c e l l s w e r e b r o u g h t i n t o f o c u s , a n d t h e n i m a g e s w e r e a c q u i r e d f r o m − 2 1 μ m b e l o w t h e f o c a l p l a n e t o 1 5 μ m a b o v e t h e f o c a l p l a n e i n i n c r e m e n t s o f 3 μ m . T o t a l c e l l u l a r f l u o r e s c e n c e w a s t h e n d e t e r m i n e d f o r e a c h i m a g e a s d e s c r i b e d a b o v e . #text italic sub #text italic sub #text italic #text C y t o p l a s m i c R N A f r o m M C F - 7 c e l l s w a s i s o l a t e d u s i n g t h e H i g h P u r e R N A i s o l a t i o n k i t ( R o c h e ) a n d s u b s e q u e n t l y r e v e r s e d - t r a n s c r i b e d t o c D N A u s i n g H i g h C a p a c i t y c D N A r e v e r s e t r a n s c r i p t a s e k i t ( A p p l i e d B i o s y s t e m s ) a s p e r m a n u f a c t u r e r ' s p r o t o c o l . S t a n d a r d c u r v e s f o r t h e a b s o l u t e q u a n t i f i c a t i o n o f c - m y c R N A w e r e g e n e r a t e d b y f i r s t a d d i n g k n o w n a m o u n t s o f s e r i a l l y d i l u t e d c - m y c R N A t o t h e c e l l l y s a t e s o f M C F - 7 c e l l s , i m m e d i a t e l y p r i o r t o R N A p u r i f i c a t i o n . S a m p l e s c o n t a i n i n g n o a d d i t i o n a l c - m y c R N A w e r e u s e d t o g e n e r a t e a b a s e l i n e m e a s u r e m e n t f o l l o w i n g R T - P C R . Q u a n t i t a t i v e R T - P C R w a s p e r f o r m e d o n A B I P R I S M 7 3 0 0 S e q u e n c e d e t e c t i o n s y s t e m u s i n g F A M - l a b e l e d T a q m a n ® p r i m e r s e t s f o r c - m y c a n d β - a c t i n , a n d t h e T a q m a n ® u n i v e r s a l P C R M a s t e r M i x ( A p p l i e d B i o s y s t e m s ) a c c o r d i n g t o t h e m a n u f a c t u r e r ' s p r o t o c o l . Immediately following the cytoplasmic microinjection of MBs, there was a clear increase in fluorescence in the nucleus, while there was no detectable change in fluorescence in the cytoplasm (a). The fluorescent signal in the cytoplasm remained extremely faint and was barely distinguishable from the background. The rate of increase in fluorescence observed in the nucleus varied with injected dose, with lower concentrations of MBs taking longer before the fluorescent signal in the nucleus reached a plateau. Specifically, when MBs at a concentration of 10 μM were injected into the cytoplasm, the fluorescent signal within the nucleus typically reached a plateau between 1 and 2 min; however, when cells were injected with 2.5 μM of MBs the fluorescent signal did not plateau for 40–50 min (b). This difference may largely be due to a faster diffusion into the nucleus, although a higher concentration of MBs may also occupy a higher percentage of the protein machinery, and thus expedite the process of active transport as well. In order to determine whether the increase in fluorescence in the nucleus was in response to undesirable MB hybridization or due to non-specific protein interactions (or nuclease degradation), live cell competitive inhibition studies were performed. Specifically, an excess of linear 2′--methyl RNAs with the same targeting-sequence as the MBs were co-injected with MBs into MCF-7 breast cancer cells. It was expected that if the MB signal were due to target hybridization, then the 2′--methyl RNAs would compete for the same binding sites and reduce/eliminate the fluorescent signal. Despite injections of 10-fold excess 2′--methyl RNAs, we continued to observe an increase in MB fluorescence in the nucleus; however, the rate of increase was slightly reduced (b). Presumably, a slower increase in fluorescence is due to the DNA MBs and the 2′--methyl RNAs utilizing some of the same machinery for active transport into the nucleus. Nonetheless, these observations suggest that the MB signal in the nucleus is likely to be non-specific, involving mechanisms other than hybridization. As noted earlier, when MBs with no known intracellular RNA target were injected into MCF-7 breast cancer cells only very faint fluorescent signals were observed in the cytoplasm. To investigate whether this was due to complete sequestration of the MBs into the nucleus or because MBs remained in a quenched hairpin conformation in the cytoplasm, we attached MBs to QDs so they would not be able to pass through the nuclear pores (a). QDs were chosen over large non-karyophilic proteins such as streptavidin as a means of inhibiting entry into the nucleus because QDs also provide a bright and stable fluorescent signal that can be used for tracking MB localization. Further, the QD fluorescence allows for accurate ratiometric measurements since the QD signal remains unquenched regardless of the conformation of the probe (b). Ratiometric measurements are helpful in removing experimental variability in single-cell kinetic studies because while changes in fluorescent intensity due to instrumental fluctuations will directly alter the intensity of the MB and QD signal individually, their ratio will remain unchanged. Of course, one concern with using QDs is that their large size might interfere with MB-target hybridization; however, comparison of the kinetics of hybridization of free MBs and MB-QD conjugates indicated that the QDs had relatively little affect on hybridization (). This result is in agreement with previous studies where DNA attached to submicron latex particles exhibited rate constants similar (0.2× to 1×) to those of unconjugated DNA in homogeneous solutions (). Since the diffusion boundary layer is so small on micro- and nanoparticles, slower hybridization kinetics would have likely meant that the QD–MB probe was reaction-rate limited (). Perhaps, the separation between the MBs on each QD, likely due to the spacing of biotin binding domains, helps limit any type of steric interference during QD–MB-target hybridization. It has recently been reported that conjugation of MBs to QDs could result in fluorescence resonance energy transfer (FRET) (). Indeed some FRET was observed with the QD–MB conjugates reported here. Specifically, the fluorescence of unhybridized MBs was quenched by 70% upon conjugation to QDs; however, upon the addition of excess complementary target the conjugated MBs only exhibited a 34% reduction in peak fluorescence compared with an equimolar amount of unconjugated MBs. These measurements assume a labeling ratio of nine MBs to each QD. The larger extent of quenching of MB-QD conjugates in the hairpin conformation leads to an improvement in signal-to-background (S:B = 93) compared with unconjugated MBs (S:B = 43). In contrast to the MB fluorescence, the QD signal is increased by 2.3% upon conjugation to MBs. When non-targeted QD–MB conjugates were microinjected into living cells, the QD–MBs were localized entirely in the cytoplasmic compartment and no MB fluorescence was observed in the cell immediately after injection (a), although image analysis did indicate an MB signal slightly above the background in all of the images. The average single-cell fluorescent ratio / was calculated to be 0.094 ± 0.061. No further increase in signal was observed for at least the next 20 min (), indicating the absence of false-positive signals. When QD–MB conjugates were pre-hybridized to complementary oligonucleotide targets prior to microinjection into MCF-7 cells, a bright fluorescent signal in the cell cytoplasm was visible in both the MB and QD images. The fluorescent ratio / was found to be 2.15 ± 0.29, which indicates an average intracellular signal-to-background ratio (i.e. [/]/[/]) of 23:1 (b). The difference between intracellular and solution measurements of signal-to-background is likely due to lower signal-to-noise in live cell studies. Nonetheless, these findings provide evidence that the MBs remain in their quenched hairpin conformation when limited to the cytoplasmic compartment. To gain insight into the cause of the false-positive signal observed in the nucleus following the microinjection of MBs into living cells and to determine the effect of conjugating MBs to QDs on these interactions, a fluorometric analysis of MBs and QD–MBs in the presence of various proteins and nucleuses was conducted. It was found that non-specific interactions with BSA did not result in a substantial increase in fluorescent emission for either the MBs or QD–MBs (a); however, SSB protein did cause a significant increase in fluorescence. Specifically, the MB fluorescence increased to 17.9 ± 0.2% of the signal elicited by a completely hybridized MB, while the QD–MB fluorescence increased to 4.9 ± 0.2% of the signal elicited by a completely hybridized QD–MB. The lesser effect of SSBs on QD–MB fluorescence suggests that steric interference may limit SSB-MB binding when the MBs are conjugated to QDs. Previous studies have shown that when MBs (250 nM) are incubated in the presence of saturating concentrations of SSB (529 nM), the MB signal can increase to more than 90% of the signal elicited by a completely hybridized MB (). It is hypothesized that the smaller increase observed in our experiments is due to incomplete binding at the lower concentrations of MB (45 nM) and SSB (360 nM) tested. When MBs and QD–MBs were incubated with phosphodiesterase I (5′-exonuclease) and phosphodiesterase II (3′-endonuclease), only a small increase in MB and QD–MB fluorescence was observed over a period of 30 min (Supplementary Figure 1). Although a slightly faster rate of increase in fluorescence was observed with unconjugated MBs, the difference is negligible. It appears that the reporter fluorophore and quencher at the 5′- and 3′-end respectively limited exonuclease activity for both MBs and QD–MBs. In contrast to exonucleases, MBs and QD–MBs were both highly susceptible to degradation by S1 endonucleases (b). The fluorescent signal of both probes increased to 60% of their maximum with 30 min. Mung Bean endonucleases, which are known to cleave hairpin loops, also degraded both MBs and QD–MBs, but to a much lesser extent than S1 nucleases. The rate of degradation was similar for MBs and QD–MBs in the presence of endonucleases. Based on our findings, it appears that both single-stranded DNA binding proteins and endonucleases could be responsible for generating false-positive signals in the nucleus following the microinjection of MBs into living cells. Although SSB proteins and nucleases are also present in the cytoplasm, previous reports have demonstrated that there are fewer DNA–protein interactions in the cytoplasm compared with the nucleus and that nuclease activity in the cytoplasm is low (,). These differences could help explain why MBs only elicited detectable false-positive signals in the nucleus. Our findings also suggest that the QD–MB signal may remain quenched following microinjection into living cells not only because of cytoplasmic localization but also due to fewer interactions with SSB proteins. The microscope calibration methodology involved injecting known concentrations of pre-hybridized MBs into paraffin oil, which results in the formation of spherical water-in-oil bubbles (Supplementary inset). With this approach, each bubble serves as a well-characterized fluorescent standard that can be imaged directly on the microscope. It was hypothesized that any changes that were observed in MB fluorescence would correspond to instrumental fluctuations. Although MBs were utilized here, they can easily be replaced with any fluorescent molecule. Since the total fluorescence of each bubble increases linearly with volume, which can vary from bubble to bubble, fluorescent intensity measurements were compared only after being normalized by the respective bubble volume (/). These measurements were conducted multiple times a day for 1 week and whenever semi-quantitative measurements of fluorescence were being performed. We found that within any given day there were only very small fluctuations in the fluorescence intensity of the microscope, i.e an SD of less than 3% in fluorescent intensity measurements (Supplementary Figure 2); however, between days fluorescence intensity measurements differed by as much a 28%. This suggests that each time the light source is ignited it may have a slightly different intensity. Daily measurements of instrumental fluctuations in fluorescence intensity were used to adjust measurements of MB and QD fluorescence accordingly. In addition to instrumental fluctuations, it may be speculated that another potential source of error when quantifying and comparing the fluorescence of multiple cells is any difference in the focal plane. When fluorescently labeled cells are observed under a microscope, there are clear changes in the peak fluorescent signal as the cells are brought in and out of focus. Cells that are in focus exhibit a very bright clear signal, while cells that are out of focus exhibit a fainter fluorescent signal that is diffuse over a much larger area. Despite these visual differences, when using wide-field fluorescent microscopy the number of fluorophores being excited and the number of photons being detected were hypothesized to be the same, assuming no major losses in light due to absorption between the cell and the objective. To validate that total cellular fluorescence is indeed independent of focal plane, QD–MB conjugates were microinjected into MCF-7 cells and images were acquired from 15 μm below the focal plan of the cell to 15 μm above the focal plane in increments of 3 μm (Supplementary Figure 3a). The total QD fluorescence in each image was then measured, making sure to include all of the fluorescent signal within the ROIs that were drawn around each cell. This often required drawing regions of interest well outside the cell itself. As shown in Supplementary Figure 3b, there appeared to be only a small loss in signal as the focal plane moved through the cells. In fact, the percent difference between the total fluorescent signal measured at the focal plane and the total fluorescent signal measured 15 μm above or below the focal plane was less than 3.3%. Therefore, it was concluded that quantitative measurements of QD and MB fluorescence are not dependent on the focal plane. In order to evaluate whether QD–MB conjugates could be used to sensitively detect endogenous gene expression across a population of MCF-7 breast cancer cells, microinjection experiments were performed using antisense c-myc QD–MB conjugates. A fluorescent signal was detectable in the cytoplasm of all the microinjected cells. Interestingly, the difference in total integrated fluorescence between the faintest and brightest cell was more than 5-fold (). It is hypothesized that this observed variation at least partly reflects the stochasticity in gene expression, which has been shown to play a pivotal role in governing cellular fates and disease evolution (). However, instrumental noise also is surely a contributing factor. Two negative control experiments were conducted to examine whether the measured fluorescent signal truly reflected MB hybridization or false-positive events. The first negative control consisted of injecting MBs with no known intracellular RNA target into MCF-7 cells. Image analysis of MB fluorescence revealed signals that were statistically lower ( < 0.01) than antisense c-myc signals. In general, the MB signal could not even be detected visually, and thus the QD fluorescence was necessary to identify the location of the cell. The second negative control consisted of competitively inhibiting MB hybridization with excess linear 2′--methyl RNAs targeting the same sequence. Again, a semi-quantitative analysis of MB fluorescence revealed a signal that was statistically lower than the antisense c-myc signals ( < 0.01). These results suggest that QD–MB conjugates can be used for the specific and sensitive detection of endogenous RNA in single living cells. When MCF-7 cells were treated with tamoxifen, the average measure of total cellular fluorescence was 71% lower than in untreated MCF-7 cells. This is in agreement with previous reports, which found that tamoxifen caused a significant reduction in the global expression of c-myc (,,). Interestingly, while the majority of the cells exhibited an extremely faint fluorescent signal similar to the negative controls, there were a small number of cells that exhibited a fluorescent signal similar to untreated cells. It is hypothesized that these outliers may represent a small percentage of cells that are resistant to tamoxifen. When Quantitative RT-PCR was carried out on MCF-7 cells with and without tamoxifen treatment, we observed a 74 ± 2% decrease in c-myc expression following tamoxifen treatment, which agrees well with the MB data (). These results support the use of MBs that are restricted to the cytoplasmic compartment for the semi-quantitative measurement of endogenous gene expression. Quantitative RT-PCR revealed that the average copy number of c-myc mRNA per MCF-7 cell was 2907 ± 289 for untreated cells and 747 ± 101 for tamoxifen-treated cells. While this data provides insight into the sensitivity of QD–MB conjugates, it is important to note that absolute measurements of copy number are highly sensitive to RNA handling and methodology (,). For comparison, another report indicated a copy number of 8000 c-myc transcripts in MCF-7 cells (,). p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Transcription factors bind to a variety of sequences with different affinities (). The amount of sequence variability within a set of binding sites is limited by physical requirements for binding, as well as the ability for the site to be distinguished from non-sites in the genome (). A range of affinities allows for a subtle regulation of transcription. In the case of activators, higher affinity sites will presumably be bound longer than lower affinity sites, and have a greater probability of stabilizing the initiation complex, which in turn has a greater probability of transcribing a gene. Therefore, the affinity of the protein for a site is a direct indicator of the degree that that site will affect the gene expression. Being able to predict binding affinities for different DNA targets is useful in characterizing genetic regulatory pathways. To do this, we use an information theory-based weight matrix to quantify protein binding to individual sequences (). Information theory was developed by Claude Shannon in the late 1940s to describe the movement of information in communications (). When applied to biological systems it has proven to be useful (,). Based on the frequency of each base at each position in a set of aligned binding sites, we can determine the strength of an individual site in bits of information. This strength is called the individual information, (rate of individual information transfer, bits per site) for a site (). Advantages of this approach are discussed in Materials and Methods. It has been shown that the protein–DNA dissociation constant, , varies with DNA sequences, and can be approximated by different weight matrix approaches (). The information in a binding site should be related to the binding energy (). Binding energy, in turn, is proportional to the logarithm of the ratio of the association () and dissociation () rate constants of binding. Since the on-rate depends on diffusion of the protein to the DNA binding site, we expected that the on-rate would be independent of the binding sequence. This suggests that the information of binding sites () should be linearly related to the logarithm of the off-rate. Others have reported differences in binding rate constants as a function of sequence (), but they did not report any relationship between the rate constants and affinity predictions. No one has shown how information theory predictions of individual binding sites are related to binding and dissociation kinetics. To address this issue, we used surface plasmon resonance (SPR) technology () and electrophoretic mobility shift assays (EMSA) to measure the binding kinetics for 13 Fis binding sites ranging in predicted site strength, based on our information theory approach. Fis is a pleiotropic homodimeric DNA binding protein involved in site-specific recombination, chromosomal compaction and transcriptional regulation (,,). Because many genomic sites have been experimentally identified, a reliable Fis model could be constructed and verified (,), making it a good protein for this analysis. The Fis binding site model was built using the standard Delila programs (,) (), and was originally presented in (). Individual information analysis (,) of Fis binding sites was computed using a weight matrix from the equation: e() is a sample correction value where is 120, the number of Fis binding sites and their complements that make up our frequency matrix. To determine the strength of a site (), a DNA sequence is compared to the () weight matrix and the information contribution of each base is summed across the site. There are several advantages to our approach. First, our models are composed of only experimentally verified binding sites, and do not require a training set of unproven ‘non-sites’ like many neural-networks or HMMs (). Second, our method has no arbitrary parameters, and the theory predicts that all sites with greater than zero bits of information have a negative Δ of binding (). Third, the units of measurement, bits, allow direct comparison between different molecular systems. Fourth, the average for all binding sites that define an () is , or the total information content [the area under a sequence logo ()]. The information content is a measure of the sequence conservation and it is determined by the evolution of the sites in the genome (). We used a Fis model ranging from −7 to +7 throughout this article. This assumes that positions outside this region do not affect binding and it is consistent with known footprinting data (). The small amount of information observed in positions −10 to −8 and +8 to +10 () may correspond to overlapping adjacent Fis sites (). Individual information analysis was done using the program and sequence walkers were generated using (,) (). Thirteen oligos of varying information content were synthesized to measure binding kinetics. Ten of these contain naturally occurring Fis binding sites, where binding has been experimentally verified (). These sites are presented in and . We chose these oligos to cover a spectrum of strengths from 4.9 to 12.7 bits, as assessed by our information theory approach. The three remaining oligos do not contain characterized binding sites, but have been engineered by us to test binding at additional site strengths. The first engineered oligo is the Fis consensus of 5′-ATTGGTTAAATTTTAACCAAT-3′ over the range −10 to +10, containing three extra natural bases on each end (), which is presumably the highest strength site (14.9 bits), and it does not occur in the genome (named con in , ). The second oligo is a slight modification of this consensus, where we mutated the T at position +1 to a G to decrease the strength of the site to 12.8 bits (named mut-con in , ). The third oligo is the Fis anti-consensus of 5′-CGGCTGACCCCGGGTCAGCCG-3′, which is made up of the least favorable base at each position (named anti-con in , ). The kinetics of binding to this sequence are presumably those of nonspecific interactions of Fis with DNA. All sequences were inserted into the same hairpin construct: 5′-GCTATCGCG-[Sequence]-ACGATCGCGC-GAA-GCGCGATCGT-[Complement of Sequence]-CGCGA-3′, where there is a 5′ 4 bp overhang of GCTA to allow for future modification, and a 3 bp loop of GAA in the center. This construct has been shown to form tight hairpins (,). All oligos were synthesized carrying a 5′-biotin tag (Synthegen, LLC) to allow immobilization of the oligos onto NeutrAvidin (NA)-coated sensor chips (B1 chips, Biacore Inc.). To test whether the orientation of a sequence in the hairpin affects binding, we inverted the ndhII-188 sequence in the hairpin to create comp-ndhII-188. NeutrAvidin was purchased from Pierce. EDTA, SDS, NaCl and HEPES (pH 7.4) were purchased from Invitrogen. Potassium glutamate was purchased from Sigma-Aldrich. Tris-HCl (pH 7.5) was purchased from Quality Biological, Inc. Binding experiments were performed on Biacore 2000 and Biacore 3000 instruments (Biacore Inc.). NeutrAvidin was diluted to a final concentration of 25 g/ml in 10 mM sodium acetate, pH 4.5. An immobilization wizard within the Biacore control software was used to immobilize no more than 4000 RU of NA. One RU, or resonance unit, corresponds to a change in the angle of the intensity minimum by 0.0001 as detected by the Biacore. The oligos were diluted to a final concentration of 1 mg/ml in immobilization buffer (10 mM Tris-HCl pH 7.5, 300 mM NaCl, 1 mM EDTA). To prepare double-stranded DNA, the oligos were heated to 95°C for 5 min, snap cooled on ice for 5 min, and incubated at room temperature for 15 min. The sample was then diluted 750-fold in immobilization buffer and injected manually over the surface until between 100 and 150 RUs were captured on the B1 sensor chip. Purified Fis protein () was serially diluted in 1×running buffer (10 mM HEPES pH = 7.4, 350 mM potassium glutamate, 3.4 mM EDTA, 0.01% BSA) to concentrations ranging from 100 nM for the high affinity oligos to 1000 nM for the low affinity oligos and injected at 25°C at a flow rate of 100 μl/min for 90 s. All oligos reached a stochastic steady state of Fis binding. Dissociation times were typically 90–360 s depending upon the stability of the complex. Disruption of any complex that remained bound after dissociation was achieved using two 50 μl injections of regeneration solution (0.1% SDS, 3.4 mM EDTA) followed by one EXTRACLEAN command, a running buffer wash to eliminate carry-over into the next experiment. At the beginning of each cycle, the needle was pre-dipped in running buffer before an injection of 100 μl running buffer. Similarly, each cycle was ended by an injection of 100 μl running buffer and an EXTRACLEAN command. Typically, every concentration of protein was injected twice from separate vials. In order to subtract any background noise from each data set, all samples were also run over a sensor chip surface of NA without oligo and injections of running buffer were performed for every experiment (‘double referencing’) (). Data were fit to a single exponential decay model using both of the programs Scrubber 1.10 () and Biaevaluation 3.1 (Biacore, Inc). Using EMSA, we found that nonspecific binding occurred with the long oligos used in the SPR experiments. Therefore we used hairpin oligos containing a Fis site (−7 to +7) with no additional bases, a loop (5′-GCGAAGC-3′) and the complementary sequence of the Fis site for EMSA. (See Supplementary Data for the sequences used.) Competition EMSA between conF37, a 5′ 6-FAM labeled oligo 5′-GGTTAAATTTTAACC-GCGAAGC-GGTTAAAATTTAACC-3′ (Integrated DNA Technologies) containing the consensus Fis binding site, and unlabeled oligos containing naturally occurring and mutated Fis binding sites, was used to determine the of the sites. When a potassium glutamate-containing buffer was used for EMSA, Fis–DNA complexes smear on a gel, therefore we used the following buffer. Binding reactions were carried out in 10 μl of solution, containing 7.7 mM Bis Tris Propane-HCl, 10 mM NaCl, 0.5% glycerol, 10 mM MgCl, 1 mM DTT, 800 nM Fis, 40 nM labeled conF37 oligo and 1.0, 1.5 or 2.0 μM competitors for 5 min at room temperature, followed by 2.2% agarose gel electrophoresis in 5 mM sodium borate pH = 8.5 () for 20 min at 5 V/cm and the gel was scanned by a FMBIO II fluorescent scanner (Hitachi) with 505 nm emission filter (). (See Supplementary Data for how the data were analyzed.) The Fis sequence logo is consistent with models of the Fis/DNA complex () (,,). Sequence conservation at positions ± 7 above 1 bit suggests that Fis binds two major grooves on the same face of the DNA (). However, the distance between the D helices which bind these two major contacts is less than 10.6 bases, one helical twist of B-form DNA, suggesting that the DNA must bend to enable positions ± 7 to contact the D helices (). Indeed, Fis bends DNA (). The relatively low information content of 7.18 ± 0.23 bits over the range ± 7 bases, suggests that Fis is a fairly prolific binder (2, 30). This is consistent with the observed high concentration of Fis in response to nutrient upshifts (as many as 50 000 dimers per cell) (). Finally, DNA methylation and DNase I hypersensitivity results are consistent with positions of significant sequence conservation (). The correspondence between the physical and biochemical characterization of Fis binding with the sequence conservation supports the information-theory based Fis binding model. We chose ten naturally occurring Fis binding sites and three synthetic sites for kinetic analysis. These sites covered a spectrum of strengths and are reported in . The terms anti-consensus (anti-con) and consensus (con) refer to the weakest and strongest possible sites based on our model respectively (). In order to measure the binding kinetics of these oligos, we used SPR technology. Protein can be flowed over a mat of DNA tethered to a thin gold surface. As the protein associates and dissociates, the change in density on the surface can be monitored, and and can be determined (,). The SPR plots appeared to have one-stage binding, suggesting a simple association–dissociation mechanism (). All data obtained for the Fis dimer (22.4 kDa) on the Biacore machine were transport limited (). That is, the kinetics of binding that are inferred from these experiments are not only a measurement of binding, but also a measure of the delivery of Fis to the chip surface. However, we were able to measure an apparent or ‘stability’ which is the rate of dissociation of Fis from the surface. Although this is not the true , because of the transport limitation, it is proportional since the rate of transport ( ) is constant for all measurements. Additionally, surface effects such as nonspecific interactions of Fis with the chip surface could affect the SPR measure so that it does not entirely represent or in-solution conditions, but as with the rate of transport, such effects should also be constant for all measurements. The stability kinetics measurement is strongly correlated to the individual information of the sites, with = 0.84 (). These values are presented in . The complexes of Fis with oligos ndhII and comp-ndhII had similar stabilities (7.4 × 10 and 6.2 × 10 s respectively) suggesting that orientation within the hairpin had little affect on the stability measurement. The dissociation of the protein from the anti-con oligo is faster than the dissociation from the weakest observed natural binder cin-336, 0.22 s versus 0.12 s. This is presumably related to the energy difference between the weakest possible specific binding and nonspecific binding for Fis. The stability of the protein with the consensus and mutated consensus is very high, 9.4 × 10 and 8.7 × 10 s, respectively. The logic of our experiment is based on a series of simple relations: Although our Biacore experiments gave the relationship of Equation () (), they did not give us values. To investigate , we performed competitive EMSA experiments to determine s (). and : The experiment was repeated and similar results were obtained (data not shown). Since s and were measured by different techniques, the relative ratios between the sites should be correct but they may differ from the absolute values by an unknown multiplicative factor. On the log scale, this is in the additive constant. Using the s measured by EMSA and the s measured in the Biacore experiments, we calculated according to Equation () for each DNA. Unexpectedly, we observed that is related to the information. against and against , and we found that 49% of the variance of and 78% of the variance of is explained by the variance of (Supplementary Data). Thus most of the off-rate is explained by the information in the sequence. In addition, a good portion of the on-rate is explained by the sequence, implying that another factor—we suggest sequence bendability—may be involved in the initial binding. Are the evolved binding targets of Fis the result of the physical properties of DNA? It is possible that the bases that are specifically contacted have been adapted through natural selection to facilitate binding through bending. If this is true, then there should be a correlation between and . and 85% of the variance is explained by , suggesting that some of the positions are important for both binding and bending (Supplementary Data ). This proposal is consistent with our previous observations on the sequence logo of Fis (). We found that patterns of bases in the Fis sites can be explained in two distinct ways. In , the outer bases at ± 7, mostly G and C, are consistent with direct binding by Fis into the major groove but these contacts are too close to allow the D helices of Fis to fit into the major groove unless the DNA is also bent. Positions ± 4 and ± 3 contain pyrimidines and purines (respectively, on the 5′↣ 3′ strand) which could be contacted directly through the major groove or which could provide a bendable step. Likewise positions −2 to +2 contain A or T which is also consistent with either direct minor groove contacts or with bending into the minor groove. Since the central positions from −4 to +4 do not appear to be contacted in our 3D model (), binding of Fis may first involve specific contacts followed by bending that perhaps releases those contacts. This implies that the binding rate requires DNA sequence-dependent bending. If so, is controlled by the degree of flexibility of the DNA and that, in turn, is controlled by the DNA sequence. However, if Fis makes direct contacts to the central bases while bound (despite our modeling) then DNA sequence should determine the strength of binding, and this is indeed observed. We are led to suggest that both bending and direct contacts are involved in both of the on- and off-stages of Fis binding. Similar experiments relating the information content of binding sites for other proteins that do not bend DNA as strongly as Fis may reveal further insights into the binding process. The experiments described here suggest that is mostly dependent upon the logarithm of . It has previously been shown that the average for all sites is , the sequence conservation of a set of binding sites (). Therefore, the results imply that the sequence conservation (the amount of variability among a set of binding sites) for a protein is directly related to the binding kinetics of that protein to its targets. A stronger binding protein that covers the same length of DNA will have a less variable site. Another aspect is that evolves to match the information needed to find the sites in the genome, , which is a function of the size of the genome and number of sites (,). As a protein evolves to bind a greater number of targets, the average specific binding energy of that protein to its targets would decrease by increased . Our experiment provides preliminary data supporting a distinction between two approaches to understanding the DNA recognition process. In , no data points were obtained between the anti-consensus at −30.6 bits and 0 bits, however the lowest positive Fis site, at 4.9 bits has a around −3 and the anti-consensus is around −2 so the curve is linear with a negative slope to near zero bits and then presumably is essentially flat from there to −30.6 bits. As shown in , a similar result occurs with a plot of binding energy ( ) versus information. We suggest that this apparent break at zero bits is a manifestation of the Second Law of Thermodynamics and the channel capacity. That is, the Second Law predicts that sites with positive information should have negative Δ values and those with negative information should not bind because they have positive Δ values (). Shannon's channel capacity theorem predicts threshold effects in coded systems where there is a sharp boundary between recognized and unrecognized signals (). The break in the curve therefore provides support for a coding interpretation of the binding interaction between Fis and DNA. This is in contrast with thermodynamic theories of binding, which generate a scale starting at the consensus, and which do not predict a specific boundary (). The individual information appears to be well correlated to the kinetics of binding. This not only gives greater confidence in our previous information theory based models, but also shows that it is a reliable approach to characterize genetic systems . Furthermore, the relationship between information and energy is subtle (), and this correlation helps ground the information theory approach into thermodynamics. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Studies of the structural dynamics of biopolymers, such as nucleic acids and proteins, and their complexes in solution are keys to understanding their functions and mechanisms in living organisms. Nucleic acids, especially RNA, are extremely versatile and flexible molecules that are capable of interacting with many other molecules, and even subtle local variations in structure and dynamics are functionally significant. For detecting the local structural changes, fluorescent probes using base analogs, such as 2-aminopurine (AP) (), pteridines (), cyclic cytosines (,), 5-(fur-2-yl)uridine () and hydrocarbons (,), provide useful information that may not be obtained by a static structural analysis. The fluorescent intensity of AP at defined positions in nucleic acids varies depending on its structural environment, thus making it useful as a probe of local structural changes. For example, stacking interactions with the neighboring bases quench the AP fluorescence. Using this technique, several functional RNA molecules, such as ribozymes and aptamers, were analyzed (). However, the use of fluorescent base analogs is still restricted, because their site-specific incorporation into nucleic acids relies solely on chemical synthesis. Although the combination of chemical synthesis and enzymatic ligation is now a routine and reliable procedure for DNA preparation, it is still difficult and laborious to prepare RNA fragments with long chains. Thus, the further development of the site-specific incorporation of fluorescent base analogs into RNA could expand the fluorescence techniques, which promise to be broadly applicable. One of the attractive methods for the site-specific fluorescent labeling of RNA is the expansion of the genetic alphabet by an unnatural base-pair system (). This system enables the enzymatic incorporation of extra components into RNA at desired positions by transcription mediated by the extra base pairs. Recently, we developed unnatural base pairs, such as 7-(2-thienyl)imidazo[4,5-b]pyridine (denoted by ) and pyrrole-2-carbaldehyde (denoted by ) (), 2-amino-6-(2-thienyl)purine (denoted by ) and 2-oxopyridine (denoted by ) (), and and imidazolin-2-one (denoted by ) (). Each base pair has a specific and characteristic selectivity in replication and transcription. For example, the – pair functions complementarily in replication and transcription, enabling the site-specific incorporations of and into DNA and RNA by polymerases. In addition, the – pair can be used unidirectionally in transcription for incorporating and modified , such as fluorophore-linked bases, into RNA opposite in DNA templates. The base is strongly fluorescent (excitation: 299 and 352 nm; emission: 434 nm; quantum yield: 41%) (). Both the excitation and emission centers of are shifted to longer wavelengths, relative to those of AP. In addition, the quenching of the fluorescence in nucleic acids is sensitive to its stacking environment, but is less than that of AP, and the fluorescence is efficiently detectable even in nucleic acids. Thus, the base would be more useful as a fluorescent probe for analyzing the local structural dynamics of nucleic acids. For the site-specific incorporation of into RNA, the – pair (A) can be used; the substrate of is selectively incorporated into RNA opposite in DNA templates by T7 transcription (). However, the transcription mediated by the – pair is less efficient relative to that by the natural base pairs. Furthermore, DNA templates containing the – pair cannot be amplified by PCR, due to its insufficient selectivity for precise replication. Therefore, the further development of unnatural base-pair systems for incorporation into RNA should provide a powerful tool for the site-specific fluorescent labeling of RNA molecules. Here, we report an unnatural base-pair system combining a novel – pair with the – pair (A). We initially developed as a pairing partner of , and then serendipitously found that in DNA fragments also functions as a template base for the site-specific incorporation of , as well as , in transcription. The – pair enables the efficient, site-specific incorporation of the fluorescent substrate into RNA, by T7 transcription, opposite in templates. Furthermore, the -containing DNA templates can be amplified by PCR via the – pair () (B). Using this system, we employed the site-specific incorporation of the fluorescent probe to analyze the local structures of RNA hairpins with GNRA loops and tRNA molecules. Reagents and solvents were purchased from standard suppliers without further purification. Electrospray ionization mass spectra (ESI-MS) were recorded on a Waters ZMD 4000 LC/MS system. The DNA templates were chemically synthesized with an automated DNA synthesizer (model 392, PerkinElmer Applied Biosystems, Foster City, CA, USA) using the phosphoramidites of () and the natural bases. The oligonucleotides were purified by gel electrophoresis. The substrates of and 6-(2-thienyl)purine (′) were chemically synthesized from its ribonucleoside (see Supplementary Data). Chemically synthesized DNA templates (10 μM of a 35-mer template strand and a 21-mer non-template strand for 17-mer RNA synthesis; 10 μM of a 37-mer template strand and a 23-mer non-template strand for 19-mer hairpin RNA synthesis; 5 μM of 94-mer template and non-template DNAs for tRNA synthesis) were annealed in a buffer containing 10 mM Tris-HCl (pH 7.6) and 10 mM NaCl, by heating at 95°C and slow cooling to 4°C. Transcription was performed in a reaction buffer (20 μl) containing 40 mM Tris-HCl (pH 8.0), 24 mM MgCl, 2 mM spermidine, 5 mM DTT and 0.01% Triton X-100 in the presence of 1 mM natural NTPs, 0 or 1 mM TP, 2 μCi [γ-P]GTP, 2 μM DNA template and 50 U of T7 RNA polymerase (Takara, Kyoto). By the use of [γ-P]GTP, the transcripts were labeled only at the 5′-end, which facilitated the analyses of the yields. After an incubation at 37°C for 3 h, the reaction was quenched by the addition of a dye solution (20 μl) containing 10 M urea and 0.05% BPB. The mixture was heated at 75°C for 3 min, and the products were analyzed on a 20% polyacrylamide–7 M urea gel. Transcription was performed in the reaction buffer (20 μl) with 10 mM GMP, 1 mM natural NTPs, 0, 1 or 3 mM TP, 2 μCi [α-P]UTP, [α-P]ATP or [α-P]GTP (GE Healthcare), 2 μM template, and 50 U of T7 RNA polymerase (Takara) (,). After an incubation for 3 h at 37°C, the transcription was quenched by the addition of the dye solution. This mixture was heated at 75°C for 3 min, and then was loaded onto a 15% polyacrylamide–7 M urea gel. The full-length products were eluted from the gel with water, and were precipitated with ethanol and 0.05 A units of tRNA. The transcripts were digested by 0.075 U/μl RNase T at 37°C for 2 h, in 15 mM sodium acetate buffer (pH 4.5). The digestion products were analyzed by 2D-TLC, using a Merck HPTLC plate (100 × 100 mM) (Merck, Darmstadt, Germany) with the following developing solvents: isobutyric acid/NHOH/HO (66:1:33 v/v/v) for the first dimension, and isopropyl alcohol/HCl/HO (70:15:15 v/v/v) for the second dimension. The products on the gels and the TLC plates were analyzed with a Bio-imaging analyzer, BAS2500 (Fuji Film). The quantification of each spot was averaged from 3 to 9 data sets. The sequences of the templates for the RNA hairpins containing are listed in the Supplementary Data. Transcription was performed in the reaction buffer with 2 mM natural NTPs, 2 mM TP, 2 μM template and 2.5 U/μl T7 RNA polymerase. After an incubation for 3 h at 37°C, the full-length products were purified on a 15% polyacrylamide–7 M urea gel. The products were resuspended in a buffer containing 10 mM sodium phosphate (pH 7.0), 100 mM NaCl and 0.1 mM EDTA, and the amount of each RNA was determined by the absorbance at 260 nm. The solutions were diluted to 2.5 μM (∼0.5 OD260/ml) with the buffer for fluorescent and UV melting measurements. The sequences of the templates for the yeast tRNA molecules containing are listed in the Supplementary Data. In the sequences, the C2–G71 pair was replaced by G2–C71. The last two nucleosides (G and T) at the 5′-termini of the template strands were replaced with their 2′-O-methylribonucleosides, to reduce the addition of one or more non-templated nucleotides at the 3′-terminus of the nascent transcript (). Transcription was performed in the reaction buffer with 10 mM GMP, 1 mM natural NTPs, 1 mM TP, 0.5 μM template and 2.5 U/μl T7 RNA polymerase. After an incubation for 6 h at 37°C, the transcription was quenched by the addition of an equivalent volume of water and 1.75 volumes of the dye solution. This mixture was heated at 75°C for 3 min, and then was loaded onto a 10% polyacrylamide–7 M urea gel. The full-length products were eluted from the gel with water, and were precipitated with ethanol. The products were resuspended in 450 μl of 10 mM EDTA (pH 8) and were incubated at 75°C for 5 min. Then, using a Microcon YM-10 filter (Amicon), the buffer solution was exchanged to Tm buffer containing 50 mM sodium cacodylate (pH 7.2) and 50 mM KCl. The amounts of tRNA were determined by the absorbance at 260 nm, and the solutions were diluted to 1 μM tRNA in Tm buffers containing either 0.1 mM EDTA, 2 mM MgCl or 5 mM MgCl, for fluorescent and UV melting measurements. Fluorescent and UV melting profiles of each RNA hairpin or tRNA containing at a specific position were recorded at a heating rate of 0.5°C/min from 20 to 90°C, using an FP-6500 spectrofluorimeter (JASCO) equipped with a thermoelectric cell holder, and a UV-2450 spectrophotometer (SHIMADZU), respectively. For the emission spectra, the excitation wavelength was 352 nm with a 3-nm spectral bandwidth. For baseline correction, we independently determined the temperature dependence of the fluorescence for the ribonucleoside. Each melting temperature was calculated by using the IGOR Pro software (WaveMetrics, Inc.). First, we examined the incorporation efficiency and selectivity of the substrate of (TP) using short DNA templates (35-mer) containing one or two bases, in which the unnatural bases were located at complementary sites corresponding to positions 13–15 in the 17-mer transcripts (A). The ability of to function as the template base for incorporation was compared with that of another unnatural base, (). After 3 h of transcription with [γ-P] GTP, the 5′-labeled transcripts were analyzed on a gel (B). The mobility of each of the full-length transcripts and truncated products slightly differed on the gel, depending on its base composition and sequence. In the efficient transcription (B, lanes 1, 7 and 8), 18-mer products, which were obtained by the addition of one non-templated nucleotide to the full-length transcripts (17-mer), were observed, as in the common T7 transcription. The truncated products were also observed in transcription involving the unnatural base pairs (B, lanes 1–6). The 13-mer truncated products resulted from pausing after the incorporation of TP (the upper bands of the 13-mer) or the misincorporation of the natural NTPs, mainly ATP () (the lower bands of the 13-mer). Despite the production of the truncated transcripts, the relative yield (92%) of the full-length transcripts containing one base from the template (B, lane 1) was much higher than that from the template (35%) (B, lane 4), and as high as that obtained from the natural template with the natural NTPs (B, lane 8). Even in transcription using templates containing two bases, the full-length products were observed (B, lanes 2 and 3), although the relative yields were lower than that of the native transcription. In contrast, transcription reactions using the templates containing two bases did not yield full-length transcripts (B, lanes 5 and 6). Thus, the transcription efficiency of the – pair was significantly improved, as compared to that of the – pair. To assess the selectivity of the – pairing in transcription, we analyzed the nucleotide composition of the full-length transcripts obtained from the and natural-base templates. For the analysis, the transcripts were internally labeled with either [α-P] UTP, ATP or GTP, chosen depending on the template sequence to label the 3′-side of the incorporated nucleoside, and then the transcripts were fully digested to nucleoside 3′-phosphates with RNase T. The labeled nucleoside 3′-phosphates were analyzed by 2D-TLC (C), and the nucleotide composition was quantified (). The results confirmed the faithful incorporation using the template containing one base. The labeled -nucleoside 3′-phosphate was observed on the 2D-TLC only when the template containing was used (C, NNN = AC). The selectivity (97%) of the incorporation opposite (, entry 1) was as high as that mediated by the – pair (, entry 2) and that of the natural transcription, and no misincorporation of TP opposite natural bases was observed (C, NNN = CAC and , entries 4, 6). In transcription using the templates containing two bases (NNN = A and C), the selectivity of the incorporation at the second positions decreased to 84–87% (, entries 9, 13), although the selectivity of the incorporation at the first position was high (95%) (, entry 7). This selectivity was improved, to some extent (89–93%), by increasing the concentration (3 mM) of TP (, entries 10, 14). To understand the role of the 2-amino group of in the pairing with in transcription, we also chemically synthesized the substrate of 6-(thienyl)purine (′) () (), which lacks the 2-amino group of , and examined the efficiency and selectivity of the ′-incorporation into RNA opposite or . The ′ substrate was also efficiently incorporated into RNA opposite by transcription (data not shown). The selectivity of the ′ incorporation opposite (95%) (, entry 17) was as high as that of the incorporation opposite . However, the selectivity of the ′ pairing was decreased by 85%, suggesting the presence of a hydrogen-bonding interaction between the 2-amino group of and the 2-keto of . In contrast, we have no evidence for a possible interaction between the 2-amino group of and the 2-aldehyde group of in transcription. The DNA fragment containing the – pair exhibited thermal stability similar to that of the DNA fragment with the ′– pair (data not shown). Using the – pair, we examined the site-specific incorporation of into RNA hairpins by T7 transcription to demonstrate the potential of the fluorescent base as a probe. We incorporated into the loop region of RNA hairpins containing GNRA loops (where N can be any nucleotide and R is either G or A). RNA hairpins with GNRA loops, such as GAAA, GAGA and GCAA loops, exhibit high thermal stability and are extremely common in biologically active RNA molecules. NMR studies (,) and fluorescent probing (,) of the hairpins indicated that the GNRA loops contain a sheared G–A pair between the first G and fourth A in the loop. Although the third base (R) is always stacked with the fourth A in the loop, the second base (N) is less ordered than the third base. Thus, it is commonly believed that the second base is apt to be exposed on the outside of the loop, where it would be accessible for interactions with other molecules. We prepared hairpin fragments containing at the second or third position in the GNRA loop (A and B) by T7 transcription using templates with TP and the natural NTPs. Temperature-dependent melting profiles of each transcript (2.5 μM) were measured, using the fluorescence emission at 434 nm of (excited at 352 nm) and the UV absorbance at 260 nm (C). Since the quenching of the fluorescence by collision events with the solvent increases at higher temperatures, the fluorescence-monitored melting curves of each RNA transcript were normalized by that of the nucleoside monomer of (see Supplementary Data). The incorporation into the loop only slightly reduced the thermal stability of the hairpins, by 1.4–1.5°C, as compared to the unmodified hairpin containing the GAAA loop (melting temperature (Tm) = 68.2°C). Thus, the substitution of the second or third base in the loop with does not induce any substantial change in the entire hairpin structure. The GAA and GAA hairpins (10s and 11s) showed characteristic fluorescent profiles (C), which reflect the general GNRA loop structures. At physiological temperatures, the fluorescent intensity of the GAA loop (10s) was much larger than that of the GAA loop (11s), even though the base was next to the first G base in the GAA loop, and in general, G quenches fluorescence most efficiently among the natural bases. The strong fluorescence emitted by the GAA loop gradually decreased when the temperature was increased. On the contrary, the intensity of the GAA loop quickly increased at around 60–76°C. These observations indicate that the second base in the GAA loop is exposed to the solvent at physiological temperatures, and then the non-specific interactions of with the neighboring bases increase gradually by the denaturation of the hairpin structure with an increase in temperature. In contrast, the third base in the GAA loop stacks with the fourth A base at physiological temperatures, and this base stacking unfolds upon the denaturation of the hairpin structure by increasing the temperature. The melting temperatures obtained from the fluorescent profiles (denoted by Tm) also provide valuable information about the thermal stability of each local region. The Tm value (62.5°C) of the GAA loop was much lower than the UV melting temperature (66.8°C) of the hairpin. This shows the high flexibility of the second base in the GAA loop. In contrast, the Tm value (68.5°C) of the GAA loop was higher than the UV melting temperature (66.7°C), indicating that the stacking of the third base in the GAA loop strongly contributes to the stability of the entire hairpin structure. These results highlight the potential of the base as a fluorescent probe. Next, the ability of the fluorescent probe was tested in a more intricate RNA molecule, tRNA. We incorporated one base into yeast tRNA at six specific positions by T7 transcription, and examined the Mg-induced and temperature-dependent tRNA folding by analyzing the fluorescent properties of at each site in the tRNA. For the incorporation into the tRNA, we chose six sites: positions 16 and 17 in the D-loop (tRNA 16s and 17s), 36 in the anticodon (tRNA 36s), 47 in the extra-loop (tRNA 47s) and 57 and 59 in the TΨC-loop (tRNA 57s and 59s), where each original base does not pair with any other base (). Temperature-dependent melting profiles of the fluorescence emission at 434 nm of (excited at 352 nm) and the UV absorbance at 260 nm of each tRNA transcript (1 μM) were measured ( and ). The UV melting temperature of each tRNA transcript containing (64.6–66.4°C in 2 mM MgCl) was as high as that of the natural tRNA transcript (65.5°C in 2 mM MgCl), suggesting that the substitution of at these positions did not significantly destabilize the global tRNA structure. As in the case of the simple GNRA hairpin, each tRNA containing at the specific position also displayed characteristic fluorescent intensity changes that reflected the local structural features (). The fluorescent profiles in the presence of Mg (2 and 5 mM) clearly fall into two groups: one includes tRNA 16s, 17s and 47s (group 1), and the other includes tRNA 36s, 57s and 59s (group 2). In the presence of 2 or 5 mM Mg, the intensities of the fluorescence of group 1 at low temperature were 1.7- to 3.4-fold larger than those of group 2. The fluorescent intensity of group 1 decreased when the temperature was increased, but the fluorescent intensity of group 2 increased at higher temperatures. These results indicate that the base at position 16, 17 or 47 is exposed to the solvent at physiological temperatures, and the non-specific interactions of with the other bases increase upon the denaturation of the folded structure with an increase in temperature. In contrast, the base in position 36, 57 or 59 stacks with the neighboring bases in the folded structure, and the base stacking is gradually denatured by increasing the temperature. These speculations are quite consistent with the conformations of the original bases at each site in the crystal structure of the tRNA (B–E) () and other structural analyses depending on Mg concentrations (,). The fluorescence intensities of tRNA 16s, 17s and 47s drastically increased by the addition of Mg at a physiological temperature, indicating that the bases at these positions are kept outside by Mg binding to the L-shaped tRNA. Furthermore, the Tm values obtained from the fluorescent profiles reflect the stability of each local structure in the tRNA, by comparison with the Tm values obtained from the UV melting profiles. For example, the Tm value from the fluorescent profile (69.5°C at 2 mM MgCl) of tRNA 36s was higher than that from the UV profile (65.2°C at 2 mM MgCl), indicating the increased stability of the anticodon stem-loop relative to the stability of the entire tRNA structure. In contrast, the low stabilities of tRNA 16s (Tm = 58.0°C at 2 mM MgCl) and 17s (Tm = 59.6°C at 2 mM MgCl) suggest that the partial structure involving the D-loop may be a fragile region within the L-shaped tRNA. Further analyses using these tRNAs will provide valuable information about the dynamics of RNA structures and the interactions with other molecules in aminoacylation and translation. In this study, we have described the selective and effective site-specific incorporation of the fluorescent probe into RNA by T7 transcription mediated by the – pair. We initially developed as a pairing partner of another unnatural hydrophobic base, , and TP was efficiently and selectively incorporated into RNA opposite (). We then found that can also function as a template base for the site-specific incorporation of the fluorescent base into RNA by transcription. The rational replacement of the natural bases in RNA molecules with the fluorescent base enables pinpoint structural analysis. The fluorescent intensity of in RNA molecules sensitively decreases with increasing stacking interactions with neighboring bases, reflecting its local structural features and the structural changes of the RNA molecules. Furthermore, the Tm values obtained from the fluorescent profiles provide useful information about the local structural stability at the incorporation sites in the RNA molecules. In addition to the site-specific incorporation into RNA by T7 transcription, the DNA templates containing can be amplified by PCR using the – pair (). Thus, the fluorescent probing by the unnatural base-pair system combining the – and – pairs provides a powerful tool for studying the dynamics of local conformational changes at a defined position within a large RNA molecule. Our findings using this series of unnatural bases, ′, and , also provide a clue about the mechanisms of base pairing in transcription. In the previous report on the – pair (), we showed that the non-hydrogen-bonded base-pair functions in transcription, suggesting the importance of the shape complementarity between pairing bases, as shown in replication by Kool . (). Since ′ is the analog of in shape, both ′ and sterically fit with , resulting in the high selectivity of the ′– in transcription. In addition, the selectivity of the incorporation opposite was as high as that of the ′ incorporation, and thus the 2-amino group of is not essential for the specific pairing with . Even though the amino group of may clash with the aldehyde group of in the Watson–Crick base-pair geometry, T7 transcription might tolerate such a disparity between pairing bases in the selectivity. Although the hydrogen-bonding interactions may not be absolutely required for the selective base pairing in transcription, the comparison between the – and ′– pairings showed that the hydrogen bonds also assist in increasing the base-pair selectivity. As for the transcription efficiency, our results indicate that the hydrophobicity of the template base is also important. Although the shape of is similar to that of , the transcription efficiency using the template is much higher than that using the template. This might be because of the higher hydrophobicity of relative to , suggesting that the hydrophobicity of the template base is favorable to the interaction with T7 RNA polymerase. In fact, the crystal structures of T7 RNA polymerase elongation complexes revealed the stacking interaction between the template bases and hydrophobic amino acid residues, such as the template ( − 1) base with Y639 and the template ( + 1) base with F644 (,). Additional studies using this series of unnatural bases in combination with other base analogs (,,) could provide further understanding of the transcription mechanisms. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

A high level of transposable element expression is usually deleterious for the organism, leading to mutations and chromosomal rearrangements. Therefore, activity of mobile elements is thought to be under keen cellular control. Silencing of selfish elements is realized through the short RNA species, called repeat associated short interfering RNAs (rasiRNAs) () and also Piwi-interacting RNAs (piRNAs) (). piRNAs play evolutionarily conserved roles in the regulation of transposable elements in insects, mammals and zebrafish () and are accumulated specifically in the germline (). In , the rasiRNA pathway requires members of the ‘Piwi subfamily’ of Argonaute proteins Piwi, Aubergine (Aub) and Ago-3 () but not the ‘Argonaute subfamily’ members, Ago1 or Ago2 (), which guide microRNA and siRNA functions, respectively (). rasiRNAs of 24–28 nt in length are longer than 21–22 nt siRNAs derived from dsRNA or 21–23 nt endogenous microRNAs (,). The increased length of rasiRNAs has aroused a suggestion of a peculiar mechanism of their formation (). In flies neither Dicer-1, which makes microRNAs, nor Dicer-2, which produces siRNAs, are implicated in rasiRNA formation (). Recent publications support a model in which discrete heterochromatic loci produce rasiRNAs that are predominantly antisense to transposons. The antisense rasiRNAs are bound by Piwi and Aub proteins and guide formation of sense rasiRNAs by cleavage of sense transposon transcripts (,). It was demonstrated that short interfering RNAs are implicated in chromatin modifications, such as methylation of histone H3 K9, in yeast, plants and animal somatic cells (). However, it remains unknown whether chromatin-based silencing of selfish elements may be realized in the germline by Piwi-interacting RNAs, in particular by rasiRNAs in flies. We show that the significantly reduced abundance of rasiRNAs derived from a wide range of transposable elements in mutant ovaries is accompanied by the increase of H3 K4 dimethylation, decrease of H3 K9 di/trimethylation and depletion of HP1 content in the chromatin of retrotransposons. We demonstrate that rasiRNA-mediated silencing of tested retrotransposons takes place in ovaries, where it is necessary to protect the genome against transposon-induced mutations in progeny, but not in somatic tissues. Strains bearing and mutations were /TM3, (point mutation in helicase domain of Spn-E), /TM3, (P-element insertion into ) (,), / (P- transposon insertion) () and /TM3 (P-element insertion) (), respectively. P-element transformed flies carrying the LTR- construct were kindly provided by E. G. Pasyukova. Discrimination in X-gal staining experiment of homo- and heterozygous larvae carrying and mutations was done using GFP-expressing balancers CyO, P{w = hsp70: GAL4}P{w = UAS: GFP} and TM3, P{w = hsp70: GAL4}P{w = UAS: GFP}. Total RNA was isolated from dissected ovaries or carcasses using Trizol reagent (Gibco BRL). The first strand of cDNA was synthesized using SuperScript II reverse transcriptase (Gibco BRL) and oligo(dT) primer or specific primer according to the manufacturer's instructions. cDNAs were analyzed by real-time quantitative PCR using SYBR Green. For PCR the following primers were used: 5′-CCGTGGTCAACTTCACCAGCTC-3′ (adh d2) and 5′-TCCAACCAGGAGTTGAACTTGTGC-3′ (adh r2), corresponding to GenBank sequence AE003410.1 for gene; 5′-TCCGCCCAGCATACAGGC-3′ (rp49 s2) and 5′-CAATCCTCGTTGGCACTCACC-3′ (rp49 as2), corresponding to GenBank sequence Y13939 for gene; 5′-GCATGAGAGGTTTGGCCATATAAGC-3′ (cop-s) and 5′-GGCCCACAGACATCTGAGTGTACTACA-3′ (cop-as), corresponding to GenBank sequence XO4456 for ; 5′-CGCAAAGACATCTGGAGGACTACC-3′ (Het-s2) and 5′-TGCCGACCTGCTTGGTATTG-3′ (Het-as2), corresponding to GenBank sequence U06920 for ; 5′-TGAAATACGGCATACTGCCCCCA-3′ (I el s2) and 5′-GCTGATAGGGAGTCGGAGCAGATA-3′ (I el as2), corresponding to GenBank sequence M14954 for element. X-gal staining and β-gal activity assays were performed according to protocols described previously (,). Samples containing 5–15 pairs of ovaries dissected from 1 to 3-days-old females or 4–15 carcasses were used for β-gal activity assay. Measurements of β-gal activity were normalized to the total protein evaluated by the Bio-Rad protein assay kit. RNA preparation was performed as previously described (). Total RNA was isolated from adult ovaries and testes. Cloning of miRNAs was performed as described (). Characterization of cloned small RNAs was performed using local NCBI-BLAST 2.2.13 () against the canonical sequences of transposable elements (); repeats (GenBank accession no. X59157|H-, Z11734|H- and Z11735|H-); miRNAs (, Release 8.0), tRNA () and rRNA (GenBank accession no. M21017). Only hits with 95% and higher similarity to transposable elements and () sequences and 100% similarity to other sequences were used. Parsing of results was done using corresponding BioPerl modules (). Ovaries were dissected from 1 to 10-days-old females in 1X PBS and stored in 1.5 ml tube on ice during isolation (up to 2 h). PBS solution was removed after centrifugation (3500 r.p.m. 1–2 min). 10 mg of material (about 150 ovaries or 100 carcasses) was used for one IP reaction. The chromatin IP assay was performed as described previously (), using polyclonal rabbit antibodies (Upstate): Anti-dimethyl-Histone H3 Lys4 (#07-030), Anti-dimethyl-Histone H3 Lys9 (#07-441), Anti-trimethyl-Histone H3 Lys9 (#07-523) and anti-HP1 (PRB-291C Covance innovative). Anti-TAF1 was kindly provided by G. Cavalli. DNA precipitates were amplified by semiquantitative PCR in the presence of αP dATP or real-time quantitative PCR. PCR product quantities were normalized to input and relations to a fragment of intergenic spacer in the 60D region were calculated. No identified or predicted genes are located 2.5 kb upstream and 4.3 kb downstream of the 60D amplified fragment. The TRANSFAC database search found no binding sites for any known chromatin proteins and transcriptional factors in the fragment. Final enrichment values of sample PCR products were calculated using the following expression: (product) *(60D)/(60D) *(product). The following primers were used for PCR analysis in ChIP: 5′-CAACACTACTTTATATTTGATATGAATGGCC-3′ and 5′-CGAAAGGGGGATGTGCTGC-3′ for amplification of the promoter region of LTR- construct; 5′-CAACACTACTTTATATTTGATATGAATGGCC-3′/5′-GCGTACTTCTCGCCATCAAACG-3′ and cop-s/cop-as (see above) for endogenous promoter region and ORF, respectively; 5′-ACCACGCCCAACCCCCAA-3′/5′-GCTGGTGGAGGTACGGAGACAG-3′ and Het-s2/Het-as2 (see above), corresponding to promoter region and ORF, respectively; 5′-CGTGCCTCTCAGTCTAAAGCCTC-3′/5′-CCCGGATTAGCGGTATTGTTGTT-3′ and I el s2/I el as2 (see above), corresponding to element promoter and ORF, respectively; adh d2 and adh r2 (see above), corresponding to gene; rp49 s2 and rp49 as2 (see above), corresponding to gene; 5′-CGGCGAGGGGGGAAAAGGAC-3′ and 5′-CTTGGCAGCAGGTGGAAAATGTT-3′, corresponding to the 60D intergenic spacer. The () gene encodes a putative DExH box RNA helicase, which is required for rasiRNA-mediated silencing of selfish elements (). Previously it was shown that the mutation leads to the loss of testis short RNAs related to the repeats () and ovarian short RNAs of the LINE element () and LTR retrotransposon (). To address the effect of the gene on total rasiRNA abundance, we cloned short RNAs from homo and heterozygous ovaries and testes (Supplementary Table 1). In ovaries the quantity of rasiRNAs was 5-fold higher than that of miRNAs. This is a drastically increased ratio compared to the one calculated previously for embryos and adult flies (about 0.65 and 0.1, respectively) (). In contrast to ovaries, approximately equal amounts of microRNAs and rasiRNAs were observed in testes. The amount of rasiRNAs cloned from homozygous ovaries was 6.7 and 3.3 times lower than in heterozygotes if normalized to microRNA or to the sum of cloned fragments of ribosomal and transfer RNA, respectively (). Both sense and antisense rasiRNAs abundance was decreased in homozygous ovaries. exerted the most pronounced effects on the amount of rasiRNAs related to LINE elements ( and ) and some LTR retrotransposons ( and ). The total amount of LINE-related rasiRNAs normalized to miRNAs was 20-fold lower in homozygous ovaries, whereas only a 4-fold decrease of LTR retrotransposon rasiRNA abundance was revealed (Supplementary Table 1). To investigate the role of chromatin state in rasiRNA-mediated transposable element silencing, we performed ChIP analysis of chromatin in ovarian nuclei using antibodies specific to known histone modifications. We focused on the three extensively investigated retrotransposons of : element, (LINE elements) and LTR-containing element. These three retrotransposons were shown to be up-regulated due to and other mutations, affecting the rasiRNA pathway in flies (,,). In /+ heterozygous ovaries the chromatin of promoter and coding regions of tested retrotransposons compared with that of the ORF of the ribosomal gene contained a significantly lower level of histone H3 dimethylated at lysine 4 (H3 K4me2), the principal mark of transcriptionally active chromatin () (). On the contrary, chromatin of retrotransposons was enriched with H3 K9me2 and particularly with H3 K9me3 mark, which are specific for inactive chromatin (,) (). In homozygous ovaries we observed an increase in H3 K4me2 and a decrease in H3 K9me3 in promoters, as well as in coding regions of retrotransposons, but not in the chromatin of and genes (). Since methylation of H3 K4 was shown to be a cotranscriptional process (), the increase in H3 K4me2 in the chromatin of retrotransposon coding regions may be considered as a consequence of an elevated level of their transcription. Along with endogenous retroelements, we performed ChIP analysis of a transgenic construct containing the reporter gene driven by LTR (LTR-) located on the X chromosome. We also observed an increase of H3 K4me2 occupancy in homozygous ovaries, but no decrease of the repressive H3 K9me2 and H3 K9me3 marks (Supplementary Figure 2). The absence of this latter effect may be attributed to the euchromatic location of the LTR- transgene compared to the mainly heterochromatic locations of endogenous elements. The level of TAF1 protein, which is a known component of RNA polymerase II transcription initiation complex TFIID (), remained unchanged in homozygous ovaries in and promoters. The TAF1 level was increased 3-fold in the element promoter () and increased 2-fold in LTR- transgenic construct (Supplementary Figure 2). These results allow us to propose that chromatin opening is unlikely to occur as a result of enrichment with basal transcription factors in promoter regions. We detected a significant amount of heterochromatic protein HP1 in the chromatin of element, and retrotransposons in ovaries. caused reduction of HP1 content in retrotransposons, especially for (). The 4-fold decrease of HP1 level was observed in promoters and 6-fold decrease in coding regions of . At the same time, mutations in the HP1-encoding gene, which are available only in heterozygous state, lead to a drastic accumulation of transcripts (,). This indicates that even a 2-fold decrease of HP1 level is sufficient for derepression and the observed loss of HP1 occupancy of chromatin, owing to , causes transcriptional activation. The most pronounced changes of the histone marks and HP1 level in chromatin correlates with the most dramatic increase of transcript level, compared to and element in ovaries (). It was demonstrated that SPN-E, PIWI and AUB proteins are required for heterochromatin formation in somatic tissues of (). Some rasiRNA-pathway components have also been shown to be required for nuclear organization of a chromatin insulator (), functioning of Polycomb chromatin complexes () and variegated repression of a reporter carried by the element () in somatic tissues. The origin of short RNAs in these cases remains unknown. We investigated the regulation of element and in somatic tissues of flies, carrying mutations in the and genes, which control the rasiRNA pathway (). The steady-state levels of element and related to and transcripts were comparable in ovaries, heads and carcasses (flies without ovaries) of /+, /+ and /+ heterozygous flies, indicating that tested retrotransposons are not exclusively germ-line transcribed. Nevertheless, we observed up-regulation of retrotransposon transcripts only in ovaries, but not in carcasses or heads of homozygous and flies (;data not shown). Furthermore, we found no effects of the mutation on the histone modifications in carcasses (). To extend the analysis of retrotransposon expression in somatic tissues, we used a transgenic LTR- construct (A). Activity of β-gal increased 10, 9 and 13 times in extracts of homozygous , and ovaries, respectively, as compared to heterozygous ovaries, whereas the expression level remained unchanged in carcasses (B). Expression of the construct was dramatically increased in germinal nurse cells and developing oocytes of homozygous and trans-heterozygous ovaries (C). expression level remained unchanged in brain, imaginal discs and salivary glands of and larvae as compared with heterozygous or wild-type controls (D; data not shown). Thus, the chromatin-based regulation of tested retrotransposons mediated by rasiRNAs is realized in the germline. We demonstrated that short rasiRNA species, known to be associated with Piwi subfamily proteins (), have a germline-specific function in the maintenance of chromatin modifications of retrotransposons. The s and genes are predominantly expressed in germ cells and their mutant states lead to abnormalities in germ-line development and sterility (). Moreover, evidence of germ-line specificity of the rasiRNA-mediated silencing pathway is supported by the observation, that rasiRNAs are significantly more abundant in the germline than in somatic tissues. Germline-specific silencing of mobile elements is considered an important defense mechanism against mutations caused by mobile element transpositions, because selfish transposable elements are thought to be expressed mainly in germinal cells to ensure their amplification and transmission to the progeny. A distinct function of rasiRNA-mediated silencing concerns the maintenance of telomeric state. Extension of telomeres is realized by germ-line specific transpositions of and LINE elements (). The and genes are implicated in the control of and expression, accumulation of corresponding rasiRNAs and frequency of and attachment to broken chromosome ends in ovaries (). Thus, here we demonstrated the involvement of the rasiRNA pathway in the chromatin modification of beneficial telomeric retrotransposons and dangerous transposable elements. We found that elimination of rasiRNAs in ovaries caused by leads to the decompaction of chromatin of retrotransposons. The decrease of HP1 level and the changes in histone modification patterns, manifesting itself in an increase of H3 K4me2 and decrease of H3 K9me3 were observed. A correlation between the most dramatic increase of transcript abundance, owing to the and mutations (), and the most significant changes of chromatin structure caused by compared to the element, and LTR- (, Supplementary Figure 2) suggest that changes of chromatin structure in ovaries of rasiRNA mutants are accompanied by transcriptional activation of retrotransposons. At the same time, we detected no effects of on the chromatin state of retrotransposons in somatic tissues. The observed germline specificity of rasiRNA-mediated retrotransposon silencing is in apparent contradiction with the observations that the and mutations affect heterochromatin formation in somatic tissues (,) and these genes are required for variegated repression of a reporter carried by the element (). It is appropriate to point out that the size of the short RNAs corresponding to the element () and transgenic copies () (∼23 nt) in somatic tissues is consistent with Dicer-produced siRNAs, but not with rasiRNAs, suggesting that silencing of mobile elements in somatic tissues may be realized via RNAi, but not the rasiRNA pathway. Alternatively, rasiRNA-dependent heterochromatin formation might be induced in early stages of embryonic development and then be epigenetically inherited in somatic tissues in a rasiRNA-independent manner. The mechanism of chromatin modification caused by rasiRNAs remains obscure. Although the Piwi protein was shown to be localized in cell nuclei (,,), we failed to detect Piwi in the chromatin of retrotransposons (data not shown). Possibly, Piwi is associated with the nascent RNA but may easily leave chromatin. It has been suggested that rasiRNAs direct cleavage of retrotransposon transcripts (). We propose that slicing of the nascent transcript mediated by the Piwi protein is capable to transform RNA polymerase II to a silencing complex. A similar model has been put forward to explain the spreading of transcriptional silencing in fission yeast . It has been suggested that the sliced nascent transcript might recruit the silencing machinery to perform chromatin modification (,). Further experiments are required to verify this model of chromatin silencing in the genome. Our observations emphasize the proposed role of rasi(pi)RNAs in the formation of heterochromatin enriched by mobile elements and other repeats. Heterochromatin serves as a genome region to recruit and spread regulatory proteins to control chromosomal processes, including transcription as well as chromosome segregation. Actually, the disturbance of silencing of the repeats in the genome is accompanied by chromosome meiotic non-disjunctions () and was shown to be triggered by mutations (,). Interestingly, spn-E mutations also lead to breakages in ovarian chromosomes that might be caused by chromatin opening. The peculiarities of rasiRNA-dependent chromatin modification in male and female germinal cells require further detailed studies taking into account the known role of heterochromatin in chromosome mechanics (). p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Nitric oxide (NO) is a ubiquitous signaling molecule synthesized from -arginine in an NADPH- and O-dependent reaction catalyzed by enzymes termed NO synthases (NOS) (). NO is important for a variety of physiological functions such as vasodilation, fertilization, differentiation, inflammation and apoptosis (). NO typically activates soluble guanylyl cyclase in cells to catalyze the conversion of GTP to cGMP that in turn activates cGMP-dependent protein kinase and other kinases, which mediate many of the normal physiological functions elicited by NO (). NO also influences cell viability either by inducing cell death when in excess or by protecting cells against various apoptotic or necrotic insults when present at more physiological levels (,). A predominant view is that excessive NO exerts cytotoxic effects in diverse cell types by reacting with superoxide and thereby generating the highly reactive free radical peroxynitrite, which causes non-specific oxidative DNA, protein and lipid damage. Such damage also triggers downstream signaling pathways and gene expression, which might elicit either cellular repair or apoptosis (,). A current view is that physiological NO concentrations can block apoptosis induced by various agents by activating anti-apoptotic/protective/repair proteins such as DNA-PK catalytic subunit, Bcl-X, Bcl-2, cAMP-response element-binding protein (CREB), NAD(P)H Quinone oxidoreductase or heme oxygenase-1 (). Conversely, excessive NO can elicit apoptosis or necrosis through a variety of mechanisms involving either p53-dependent or p53-independent pathways (). NO-dependent S-nitrosylation of proteins is increasingly being recognized as a key post-translational mechanism controlling the functions of certain proteins (). There is accumulating evidence of selective S-nitrosylation of cysteines in proteins that can either block cell death or promote cell death; and many of these phenomena were observed in neurons (,). NO can regulate gene expression at least in part through the S-nitrosylation or phosphorylation of transcription factors or other proteins, and by regulating the cellular localization of transcription factors (). Thus, S-nitrosylation contributes to the regulation of several pathways, including those leading to fos/Jun and NF-κB activation (). Direct S-nitrosylation of the hypoxia-inducible transcription factor HIF-1α increases its DNA-binding capacity (). NO induces phosphorylation of p53 that promotes its nuclear retention in neuroblastoma cells (,). The roles of NO in stimulating the functions of other transcription factors such as EGR-1 and Nrf2, but inhibiting the function of VDR/RXR, are documented (,). It is not clear if all these regulatory mechanisms (e.g. oxidative damage, S-nitrosylation, transcription) act in concert or independently, or function in a cell type-dependent manner. The actual outcome (cell death or repair) has been attributed to the particular concentrations of NO, the redox status of the cell and time-dependent differential gene regulation. To reveal the cellular targets of NO under cell death promoting conditions, it is important to understand not only the specific S-nitrosylation events, but also the signaling pathways activated in response to NO. To this end, we used a novel protein/DNA array approach to determine the profiles of -elements/-factors that respond to apoptosis-inducing concentrations of NO in a neuronal cell line. We have identified a number of -elements that bind to various -factors in a NO-regulated fashion, analyzed the patterns of their regulation, and have begun the identification of the target genes they control in order to reveal the cascade of NO-dependent gene expression in the control of cell viability. We show that is a target gene of NO, and the consequent Bcl-2 up-regulation can limit the amount of apoptosis induced by toxic levels of NO. The human neuroblastoma cell line SH-Sy5y was obtained and grown as described previously (). SNP, DETA-NO, SIN-1 and other chemicals were obtained from Sigma Aldrich. Except where stated, the enzymes used in this study were purchased from New England Biolabs. Platinum Taq polymerase, 10 mM dNTP mix and Lipofectamine transfection reagent were purchased from Invitrogen. The pcDNA- vector was a gift from Victor Yu, Institute of Molecular and Cell Biology, Singapore. The SureSilencing™ shBcl-2 Plasmid (targeting sequence in human Bcl-2: 5′-GAGGATTGTGGCCTTCTTTGA-3′) was purchased from SuperArray (KH00079N). The pGL3 Promoter vector containing the firefly luciferase gene, internal control plasmid pRL-TK that encodes luciferase, and the dual luciferase assay kit were from Promega (Madison, WI, USA). The Hybond ECL nitrocellulose membrane and ECL Western blot analysis kit were from Amersham Pharmacia Biotech. The RNeasy MiniKit was from Qiagen. The nuclear extraction kit was from Panomics Inc., CA, USA. The double-stranded oligonucleotides representing the -elements of AP1, AP2, BRN-3, CRE, EGR and SP1 and their mutants (), and the antibody against β-actin, were from Santa Cruz Biotechnology Inc., Santa Cruz, CA, USA. The antibodies against Bcl-2 were from Upstate Biotechnology or Santa Cruz Biotechnology. The antibodies against caspase-3 were from Cell Signaling Technology or Santa Cruz Biotechnology. The Polyclonal antibody against GAPDH was from Abcam. The Transignal DNA/Protein Array I and Transcription Factor cDNA (TF cDNA) array were from Panomics Inc., CA, USA. The consensus oligonucleotide sequences/mutant sequences of AP1, AP2, BRN-3, CRE, EGR and SP1 as shown in were subcloned in the pGL3 promoter vector to generate the reporter plasmids. The sense and antisense oligonucleotides corresponding to the -elements were synthesized with NheI and XhoI sites, respectively, and cloned at the respective sites in the pGL3 promoter vector. The reporter plasmids were named pGL3- Luc, pGL3- Luc, pGL3- Luc, pGL3- Luc, pGL3- Luc and pGL3- Luc. The mutant reporter plasmids were named pGL3-m Luc, pGL3-m Luc, pGL3-m Luc, pGL3-m Luc, pGL3-m Luc and pGL3-m Luc. The promoter constructs used are described (). SH-Sy5y cells were treated with NO donors for the times indicated, and nuclear proteins were prepared as described (). The template of the protein/DNA array I and the details of the -elements present in the array are described previously (). The protein/DNA array was performed following a procedure adapted from that described (). Briefly, the nuclear extracts were incubated with biotin-labeled DNA probe mix in binding buffer for 30 min at 15°C. Protein/DNA complexes were resolved in a 2% agarose gel and excised. DNA probes were recovered from the protein/DNA complexes and hybridized to the transignal protein/DNA array membrane. Hybridization signals were visualized using horseradish peroxidase-mediated chemiluminescence. The experiment was repeated at least two times with two different membranes. The resulting autoradiograph was scanned by densitometry using a previously described method (), and the fold induction/repression was tabulated along with the standard deviation (±SD) (). SH-Sy5y cells were treated with the NO donors DETA-NO, SIN1 or SNP for various times up to 8 h. Nuclear extracts were prepared using the nuclear extraction kit from Panomics. The various -elements were end-labeled with [γ-P] ATP and T4 polynucleotide kinase. Bandshift and the competition assays were performed by previously described procedures (). Ten microgram of the nuclear extract was used in the gel shift assays. The LipofectAMINE transfection reagent kit was used to transfect the various reporter plasmids in human SH-Sy5y neuroblastoma cells as described previously (). The plasmid pRL-TK encoding luciferase was used as an internal control in each transfection. Thirty-six hours after transfection, the cells were washed three times with PBS and lysed in Passive Lysis buffer from The Dual-Luciferase Reporter Assay System Kit. The same kit was used to assay the samples for luciferase activity (), which was measured with a TD-20e luminometer. SH-Sy5y cells were treated with NO donors for the indicated times, and RNA samples were prepared using a published method (). The transcription factor array (TF array) was performed following the procedure described (). Briefly, the probes were synthesized by combining 10 μg of total RNA (isolated from control and NO-treated SH-Sy5y cells) with 5 μl of TF cDNA primer mix provided by the manufacturer. The total mixture was heated for 2 min at 72°C followed by another 2 min at 42°C. The labeling mix contained 2 μl of Biotin-dUTP, 1 μl of reverse transcriptase and 12 μl of labeling mixture in water. The total labeling mix was mixed with the RNA-TF cDNA primer mix and incubated for 2 h at 42°C for the labeling reaction. After 2 h, 3 μl of the 10× denaturing solution was added to each sample and incubated for 20 min at 68°C. Finally, the probe was neutralized by adding 30 μl of 2× Neutralizing Buffer and incubating for 10 min at 72°C. Hybridization of the denatured probe to the transignal transcription factor cDNA array and subsequent washes were done according to the manufacturer's instructions. Hybridization signals were visualized using Streptavidin–HRP conjugate and horseradish peroxidase-mediated chemiluminescence. The experiment was repeated at least two times with two different membranes. Western blot analysis was performed by methods as described (). For total protein extraction, the cells were lysed in a buffer containing complete protease inhibitor cocktail. After centrifugation, 20 µg of total proteins were electrophoresed in 10% polyacrylamide gels and transferred to an ECL membrane. Immunoblotting was carried out with antibodies in phosphate-buffered saline with 0.2% Tween 20 and 5% BSA. After washing, the membrane was probed with horseradish peroxidase-conjugated donkey antiserum to rabbit or mouse (Chemicon) and developed by the enhanced chemiluminescence method (Amersham Pharmacia Biotech). Total RNA was extracted from the cells after the NO donor treatment using the RNA extraction kit from Qiagen. Semi-quantitative RT-PCR analysis was performed using the one-step RT-PCR kit from Qiagen using the manufacturer's protocol. The primers used for the RT-PCR reactions were as follows. -Forward: 5′ atg gat gat gat atc gcc gcg ctc g 3′ and Reverse: 5′ gaa gca ttt gcg gtg gac gat gga ggg 3′; -Forward: 5′ acc atggcgcac gctgggagaa cagg 3′ and Reverse 5′ cttgtggc ccagataggc acccaggg 3′; -Forward: 5′ acc atgtct cagagcaacc gggagctg 3′ and Reverse 5′ ttt ccgactgaag agtgagccca gcag 3′; -Forward: 5′ acc atggactgtg aggtcaacaa cgg 3′ and Reverse 5′ gtcca tcccatttct ggctaagctc c 3′; -Forward: 5′ acc atgttccaga tcccagagtt tgag 3′and Reverse 5′ ctgg gagggggcgg agcttcccct g 3′; -Forward: 5′ acc atgctg ggcatctgga ccctcc 3′ and Reverse 5′ gaccaagct ttggatttca tttctgaag 3′; -Forward: 5′ acc atgcagcagc ccttcaatta cc 3′ and Reverse 5′ gag cttatataag ccgaaaaacg tctg 3′. In order to ensure the PCR reactions are in the exponential versus linear phase, we performed the PCR reactions for each target gene from 18 cycles up to 30 cycles, with 2 cycle intervals, and have carefully monitored the resulting products. The number of cycles up to which the product amplification is in the linear range for a particular target is taken into account for comparison. Accordingly, the cycling conditions used for the various primer sets were as follows. For : 50°C for 30 min, 95°C for 15 min, followed by 26 cycles of 95°C for 1 min, 72°C for 2 min and a final extension at 72°C for 10 min. For , the same reaction conditions as above were used, but the amplification was performed for only 22 cycles. For and : 50°C for 30 min, 95°C or 15 min, followed by 25 cycles of 95°C for 1 min, 62°C for 1 min, 72°C for 1 min and a final extension at 72°C for 10 min. To knock down Bcl-2, SH-Sy5y cells were stably transfected with the shBcl-2-containing plasmid using Lipofectamine™ 2000, and G418-resistant clones or vector control clones were isolated as described previously (). To generate stable cell lines over-expressing Bcl-2, the plasmids pcDNA-Bcl-2-myc and the pcDNA vector were transfected individually in SH-Sy5y cells using LipofectAMINE 2000 following the manufacturer's protocol. The stable cells were selected in medium with 600 μg of neomycin/ml. After selection, the stable clones were maintained in medium containing 200 μg of neomycin/ml. To measure cell death, the Crystal Violet staining method, the caspase-3 activity assay or the LDH release assay was employed as described (). For crystal violet staining, cells were plated in 6-well plates and either left untreated or treated with SNP, DETA-NO or SIN1 for 4, 8, 12, 16 and 24 h. The cells were stained with 30% Crystal Violet in 10% methanol for 10 min. Excess stain was removed completely by washing with water many times. The cells were dried, and the stain was eluted in 50% methanol, 1% acetic acid. Absorbance was measured at 590 nm (). The activity of caspase-3-like proteases was measured using microtiter plates as described (). The LDH release assay was done according to the published method (). The statistical differences were determined using the Prism software by one-way analysis of variance (ANOVA) followed by the Tukey multiple comparison test. Previous work from our laboratory has demonstrated that NO induces apoptosis in a concentration and time-dependent manner in SH-Sy5y neuroblastoma cells (,). SH-Sy5y cells were treated with two different NO donors, SNP (2 mM) and DETA-NO (1.5 mM) for up to 24 h. Cell death was independently measured by Crystal Violet staining (A) and caspase-3 activation (B), confirming that NO donors (but not the vehicle control, left panel) induce substantial apoptotic death, the onset of which is notably delayed by at least 8 h. To identify the -elements whose binding to various -factors is regulated by NO, SH-Sy5y cells were treated with the NO donor SNP (2 mM) for 4 h. A novel protein/DNA array (Materials and Methods section) was used to identify the -elements whose binding to cognate -factors is increased or decreased by NO at 4 h. This time point was chosen, because it is prior to cell death when the cell is expected to be mounting transcriptional responses to the insult, but before the actual decision to survive or undergo apoptosis is taken. shows that NO increased the binding of -factors to a collection of -elements from ∼1- to 6-fold. The maximum increase of 6-fold was observed with the element, followed by elements such as GATA, AP2, NF-E1 and CRE (∼5-fold) (A and ). A moderate increase of ∼2- to 4-fold was observed with elements such as AP1, BRN-3, CBF, CDP, c-myb, NF-E2, PBX-1, EGR, GRE, SP1, TR and USF-1, whereas elements such as PPAR, MEF-2 and TFIID showed a moderate decrease in -factor binding (∼1- to 4-fold) (A and ). Elements such as SMAD and STAT did not show any increase or decrease in -factor binding with NO treatment (A). The increase in binding of the -factors to -elements observed with the protein/DNA hybrid array was subsequently confirmed by EMSA for some of the -elements (B and C). We could not perform a confirmation for all the -elements that showed increases or decreases in binding in the array because of the large number of samples. For the EMSA, we used more than one NO donor (SIN1, SNP, DETA-NO) to confirm that the observed increase in binding of -factors to -elements is consistent irrespective of the donor compounds. The experiments were also done with different concentrations (0.5–2 mM) of NO donors at four different time points. All the NO donors increased the binding of -factors to the -elements AP1, AP2, CRE, EGR and SP1 very substantially (B and C), and the magnitude of the increases was comparable to that observed with the protein/DNA array (A and ), though it was not identical. There was no consistent increase in the -factor binding to BRN-3. The competition assays with the cold (unlabeled) mutant oligonucleotides and the cold wild-type oligonucleotides showed that only the wild-type oligonucleotides could compete for binding (D). This clearly shows the specificity of binding. Thus, the EMSA results confirm the NO-mediated increase in binding of -factors to many of the -elements observed with the protein/DNA array. We next addressed whether the increase in the binding of -factors to -elements observed with the protein/DNA array and the EMSA translated into transcriptional activation or repression. Reporter plasmids carrying the luciferase gene linked to the various -elements or their mutated non-consensus sequences were constructed and used in transient transfection assays. lists the oligonucleotides harboring the -elements and their mutants used in the construction of the reporter plasmids. The reporter plasmids pGL3- Luc, pGL3- Luc and pGL3- Luc mediated a significant basal activity, which was strongly induced upon NO treatment (SNP 2 mM) in a time-dependent manner (). The NO-induced activation observed with the various reporters after 12 h of NO treatment were ∼5-fold for AP1, ∼3-fold for AP2 and ∼12-fold for CRE (, upper panels). The reporter plasmids pGL3- Luc, pGL3- Luc and pGL3- Luc also showed a substantial basal activity. However, the significant and sustained NO-mediated induction of transcriptional activity seen with the AP1, AP2 and CRE -element reporters upon NO treatment was not seen with BRN-3, EGR and SP1 (, lower panels). The reporter plasmid pGL3- Luc showed ∼1.5-fold activation 2 h after NO treatment, which reduced sharply after 2 h and reached the basal level by ∼8–12 h. The reporter plasmid pGL3- Luc showed very marginal ∼0.2-fold activation by 2 h, but showed ∼1-fold repression by ∼8–12 h after NO treatment. The reporter plasmid pGL3- Luc did not show any significant activation upon NO treatment (). The reporter assays were carried out at least three times with all three NO donors. The data presented in show the representative results obtained using SNP (2 mM) as the NO donor, although the results obtained from all three NO donors showed similar patterns. In pilot experiments with the transient transfections, 0.5–2.0 µg of the different reporter constructs were used, but the fold responses remained very similar (data not shown). This indicates a true result and not an artifact of transfecting excessive amounts of the reporter plasmids. Reporter assays with the plasmids carrying the mutant -elements did not exhibit any substantial basal activity, and the induction with the NO donors was completely abolished (data not shown). Hence, these results show that although NO activates the binding of transcription factors to a collection of -elements, not all these elements mediate transcriptional activation or repression. NO might induce or repress transcription of the -factors that bind to the -elements, thereby contributing to the activation or repression observed with the protein/DNA array, EMSA and reporter assays. To address this issue, we employed a transcription factor cDNA array to measure the NO-regulated transcription factor expression levels. Overall, we did not observe any significant changes in the mRNA expression profiles of the clear majority of the transcription factors upon NO treatment using SNP (2 mM) as the NO donor (). A slight increase in the NO-induced mRNA levels was observed with and , whereas a decrease in mRNA levels was seen with and (). Thus, it cannot be ruled out that the NO-mediated reduction in -Luc activity () may be due to lower expression of EGR2 induced by NO (B). Nevertheless, we infer that NO mostly stimulates the binding of pre-formed transcription factors to various -elements, for example, by post-translational modifications (at the very least for AP-1, AP-2, CRE and Brn-3a), which consequently regulates gene transcription. Is the NO-mediated transcriptional activation we observed mediated by the -elements physiologically relevant in terms of specific gene transcription? To identify the NO-regulated pro- and anti-apoptotic genes that harbor these -elements in their promoter regions, a semi-quantitative RT-PCR analysis was performed for the following genes— and (these genes were selected for the analysis based on the fact that they have some of the -elements that we identified in our array in their promoter regions). No changes in the transcriptional profiles of , () or (data not shown) were observed with any of the NO donors. However, the transcript of the anti-apoptotic gene was significantly up-regulated as early as 2 h by all three NO donors (A). Additional experiments done with IMR-32 neuroblastoma cells also showed a significant up-regulation of the gene as early as 2 h after treatment with the three NO donors (data not shown). An ∼2.5- to 4-fold induction of Bcl-2 protein was observed, which began as early as 2 h after SNP or DETA-NO treatment (B). These experiments together suggest that the gene is a transcriptional target of NO. The gene promoter comprises P1 and P2 regions and has characterized -elements viz., one CRE and two SP1 sites located upstream of the P1 promoter that determine the basal and inducible expression () (). The various promoter constructs used in this study have been described () and are shown schematically in A. SH-Sy5y cells were transfected with various full-length and truncated or mutated reporter plasmids. The full-length (P1 + P2) plasmid LB 322 harboring the gene promoter showed ∼1.5-fold inducible reporter activity with DETA-NO (1.5 mM) (B). The construct LB 124 that comprises only the P1 region of the promoter retained the basal and NO-inducible activity of the full-length construct (LB 322: P1 plus P2). However, the construct LB 335, which has only the P2 region of the promoter, did not mediate any induction with NO (B). In other words, the induction seen with the full-length construct (LB 322) upon NO treatment is solely from the P1 region and not from the P2 region of the promoter. To further characterize the -elements in the P1 region that respond to NO, a series of truncated and mutated reporter plasmids were used. The deletion construct LB 334 with two SP1 elements and one element (A) displayed similar basal and NO-inducible luciferase activity as that of the full-length (LB 322) and the P1 construct (LB 124) (C). However, when the CRE element was mutated (LB 595), the basal activity was retained but the induction by DETA-NO (1.5 mM) was completely lost (C). The constructs with a mutation in the proximal SP1 site (LB 1263) or the distal SP1 site (LB 1283) behaved in a similar manner to the full-length (LB 322) and truncated P1 (LB 124) promoter constructs (C). D shows that experiments with SNP (2 mM) as NO donor gave similar results as those in C. These data strongly implicate the CRE element in NO-mediated gene induction. NO is known to induce apoptosis of SH-Sy5y cells in a time and concentration-dependent manner () (,). To study the functional relevance of CRE-mediated up-regulation of the gene in NO-induced apoptosis, stable cells expressing were generated, and two independent clones were selected and characterized (). Two independent stable cell lines expressed higher levels of Bcl-2 protein as compared to the vector control cells (A, left panel). The NO donors SNP (2 mM) and DETA-NO (1.5 mM) induced ∼30–70% cell death in the wild-type and vector control cells (B), and caspase-3 activity was maximal by 12 h after NO treatment with SNP (2 mM) and DETA-NO (1.5 mM) (C). In contrast, both independent stable cell lines expressing higher levels of Bcl-2 protein were substantially (60–70%) protected from NO-induced cell death (B). In accord with this result, the stable cells exhibited a complete loss of caspase-3 activity, even at 16 h after NO stimulation (C), and caspase-3 cleavage was strongly inhibited in these cells (A, right panel). To corroborate these results and to address whether endogenous Bcl-2 can counteract NO-induced apoptosis, we knocked down Bcl-2 using a commercially validated shRNA (Materials and Methods section). Three individual Bcl-2 knockdown stable clones were generated, two of which almost totally abolished Bcl-2 synthesis (A). In time course experiments, all three Bcl-2 knockdown clones exhibited increased NO-induced cell death and caspase-3 activity compared with the parental and vector control cells (B and C). Consistent with C, all three clones produced greater levels of caspase-3 cleaved fragments, indicative of caspase-3 activation (D). Interestingly clone 1, which had the least Bcl-2 knockdown, showed levels of cytotoxicity, caspase-3 activity and caspase-3 cleaved fragments that were intermediate between the vector control and the fully knocked down clones 1 and 2. Thus, the experiments in and together illustrate the functional importance of NO-induced up-regulation and Bcl-2 synthesis () in counteracting NO-mediated apoptosis of neuroblastoma cells. The plethora of responses elicited by toxic levels of NO may be determined by the particular reactive nitrogen intermediate, chemical modifications (e.g. S-nitrosylation) (,), or signal transduction pathways leading to transcriptional activation and gene expression that are increasingly being recognized as important aspects of the pleiotropic actions of NO (,,). As will be discussed below, an emerging view is that excessive NO concentrations can simultaneously induce cell death promoting and counteracting signaling pathways, the balance and timing of which will determine whether the cell survives or dies. The interactions of transcription factors with -acting regulatory elements may be considered the ultimate step in signal transduction pathways. Studies on NO-mediated transcriptional regulation have identified a very wide range of genes induced or repressed by NO in a cell-specific manner, anticipating that correspondingly diverse numbers of transcription factors may be involved in their expression or repression (,,). In general, identification of transcription factors and their target genes has up till now been cumbersome; and not surprisingly, few studies have succeeded in addressing the actual -elements required for the physiological significance and target genes of the transcription factors activated by NO. Nevertheless, several transcription factors are known to be involved in NO-induced neuronal cell death or differentiation. These include c-Jun/AP-1 (bi-potential regulator of cell viability), CREB (which up-regulates protective Bcl-2), E2F (which down-regulates N-Myc in differentiation) and p53 (which up-regulates several pro-apoptotic genes)(,,,,,). We used a novel protein/DNA array to investigate the -elements that display altered binding to cognate -factors in response to NO, identified a collection of -elements whose binding to -factors is regulated by NO, and analyzed their relevance to gene transcription in some cases. NO does not functionally activate -factor binding to all the known -elements; moreover, some -elements mediate gene activation, while others mediate gene repression, indicating that the effects that we observed are not non-specific. Most of the -elements in the array exhibited increased compared with decreased -factor binding, and some were confirmed by EMSA using different NO donors in varying concentrations. Although increased DNA binding was seen with some -elements, only a subset of them might have physiological relevance in terms of gene activation or repression as revealed by reporter assays. Our transcription factor array analysis revealed CRE as one of the -elements whose binding to cognate -factors is strongly activated by NO, which confirmed our previous studies showing c-Jun/AP1 factors regulate NO-mediated cell survival and cell death in SH-Sy5y neuroblastoma cells (,), and also validates the whole array approach. In addition to c-Jun/AP1, the transcription factors AP-2, CREB, Brn-3a, Brn-3b and E2F have also been implicated in cell survival and cell death in various cell types (). Hence, it is not surprising that we find that excessive NO activates the -elements that are bound by these transcription factors. One of the novel and surprising findings in the present study is that NO induces a substantial ∼2- to 3-fold increase in the activity of AP2, a transcription factor mainly associated with the regulation of gene expression during development (). AP2-regulated genes include those encoding p21, c-Myc, TGF-α and c-Kit, which control cell division, differentiation or apoptosis (). The cranial defects in the developing neuroepithelium and neural crest in AP-2α-deficient mice have been attributed, at least in part, to increased cell death (). Moreover, AP-2 may act to attenuate c-Myc-dependent apoptosis under adverse conditions (). The NO-mediated activity profile of BRN-3 and EGR is unique in the sense that there is an initial moderate spike in -factor activity at the early time point (2 h) followed by a steady decrease in activity (from 4 to 12 h) (). If indeed this translates to an effect in gene transcription it would be an interesting phenomenon of gene expression increasing in the early hours and decreasing steadily during the late hours following NO stress. In fact, this kind of gene profile has been reported for genes following various signaling events (). Hence, it would be interesting to explore the downstream target genes harboring BRN-3 and EGR elements that respond to NO. In fact, some EGR- and Brn-3-regulated apoptotic genes are known, which provides a possible starting point (,,). For instance, among the Brn-3 family of transcription factors, Brn-3a and Brn-3b are known to have different, sometimes antagonistic effects on specific promoters (). Moreover, both Brn-3a and Brn-3b bind to p53 in neuroblastoma cells, but Brn-3a is associated with survival, growth arrest and differentiation, whereas Brn-3b enhances p53-mediated apoptosis (). Another interesting finding in our study is the NO-mediated repression mediated through the -elements such as MEF-2, PPAR and TFIID. The cognate -factors for MEF-2 and TFIID were not known to be regulated by NO, and the significance of this repression by NO remains unknown (,). On the other hand, the response of the PPAR element to NO has been documented (). More interestingly, the iNOS promoter has a functional PPAR-binding element (). Hence a negative feedback regulation of NO synthesis by NO is possible through the PPAR element. However, this hypothesis needs to be verified experimentally. The data obtained with the SP1 reporters suggest that there is no obvious NO-regulated transactivation or transrepression. This finding suggests that although there is a strong increase in -element--factor binding following NO treatment in the SP1 array and EMSA, it does not translate into an increase or decrease in transcriptional activity. The significance of the increase in DNA binding in the absence of any substantial change in transcriptional activity remains to be determined. Finally, using the anti-apoptotic Bcl-2 protein as an example, we demonstrated that toxic concentrations of NO induce mRNA and subsequently its protein via a CRE element in the promoter. Interestingly, mutation in the CRE element did not affect the basal expression of the gene, implying that the CRE may not play any significant role in regulating the basal expression of the gene in SH-SY5Y cells (). Cyclic AMP response element binding protein (CREB) is a major protein that binds to the CRE element. There is evidence that non-toxic levels of NO induce protection from apoptosis via CREB phosphorylation, CRE binding and consequent Bcl-2 up-regulation in neuronal cells (,,). Lethal levels of NO donors might also activate gene expression via CREB (); thus, it is entirely possible that CREB mediates NO-induced Bcl-2 up-regulation in the present study. Indeed, using lethal levels of NO donors, we showed Bcl-2 is up-regulated and -factor binding to CRE elements are both increased well before ∼30–60% cell death is evident. Even the amount of induced Bcl-2 is obviously not sufficient to prevent this (incomplete) cell death occurring in the parental cells, since over-expression of Bcl-2 totally prevents apoptosis. Conversely, complete knockdown of Bcl-2 not only sensitizes the cells to apoptosis at later time points, but also causes a much earlier activation of caspases—an indicator of the impending increased apoptosis. Although Bcl-2 is still induced at later time points when many of the cells (30–60%) are dying, we speculate this induction is occurring in the surviving cells. Thus, we conclude Bcl-2 successfully counteracts NO-induced cell death in a significant fraction of neuroblastoma cells. In other words, it is likely that the up-regulation of Bcl-2 by NO both delays the onset of apoptosis and limits the number of cells that succumb to NO. Our findings are in accord with previous studies which hinted at a role for Bcl-2 by demonstrating a correlation between Bcl-2 levels and resistance of cells to NO-induced apoptosis (,,). We have previously demonstrated that toxic NO levels stimulate the early transcriptional activation of the transcription factor Nrf2, which mitigates apoptosis of neuroblastoma cells via the synthesis of protective and anti-oxidant proteins (). Thus, our functional analysis suggests that like Nrf2, NO-mediated Bcl-2 synthesis contributes to the overall cellular defense against excessive NO. Overall our findings implicate a role for gene expression mediated by various -elements in delaying, counteracting or promoting apoptosis. Whether all the identified transcription factors and their -elements act in concert or independently needs to be investigated, though the failure of the majority of the tested transcription factors themselves to be regulated at the mRNA level by NO tends to argue that in the main, transcription is induced or repressed by various preformed transcription factors downstream of various signal transduction pathways. In summary, we show a DNA/protein array approach can readily reveal novel, global and functionally important transcription factor activities stimulated by signaling and toxic molecules or indeed any stimulus external to the cell. In general, the use of specific inhibitors of signaling proteins or knockout cells, combined with the array approach, will now allow us to determine whether particular signaling pathways are essential for regulating gene expression via one or more defined transcription factors.

Far-field fluorescence microscopy is of key relevance to the life sciences because it enables three-dimensional imaging of cellular constituents with unrivaled specificity. However, for many years its resolution was limited by diffraction to about half the wavelength of light, /2 > 200 nm. Matters began to change when it became clear that the spectral properties of the fluorescence markers, in particular their molecular states, may not only be used to generate signal, but also to dramatically increase the spatial resolution (). For example, stimulated emission depletion microscopy () overcomes the diffraction limit by effectively confining the fluorescent state of the marker to a well-defined region ≪/2. To this end, a beam of light is used to effectively keep the marker in the ground state everywhere except at a point or line where this beam is zero (). This concept has been extended to using any marker which can be optically kept or transferred to a dark state, in particular to photoswitchable organic fluorophores and photoactivatable proteins (). In all cases, the subdiffraction image is assembled sequentially in time, by translating the location of the intensity zero through the sample. When assembling an image in this way the (switchable) fluorophores must undergo several -- cycles. Multiple switching is elegantly avoided if the molecules are recorded individually rather than in an ensemble. This is possible by illuminating deactivated but switchable dyes with so little activation light that only a few isolated molecules are transferred to the bright state at a time. Upon excitation, the -state molecules can fire many fluorescent photons in a row, while their neighboring molecules remain entirely dark. Since a single molecule in the -state already represents the smallest possible spot no additional optical confinement is needed to reduce the fluorescent spot any further. However, unlike in the ensemble case, where the position of the bright marker molecules is given by the position of the zero, the exact position of the switched-on or activated molecule is unknown. By imaging the burst of emitted fluorescence on a pixilated detector such as a camera and exploiting the a priori knowledge that the emission stems from the same point, the position of the molecule can be determined with a precision much higher than the full width at half-maximum of the point-spread function (PSF) of the imaging lens (–). In this way each molecule can be precisely mapped so that an image can be assembled. After recording, the molecule has to be switched off again, meaning that the molecule underwent just a single switching cycle. This single molecule recording method was termed photoactivated localization microscopy (PALM) (), fluorescence photoactivation localization microscopy (FPALM) (), or stochastic optical reconstruction microscopy (STORM) (). Its resolution is determined by the localization precision which is given by with being the number of photons detected per activated molecule. In fact, the resolution of the final image can be somewhat tuned by including only those events where is above a certain threshold. Implemented in a wide-field setup, this novel nanoscopy form is very powerful (). Nonetheless it has so far entailed a number of limitations. Working with low densities of fluorescent molecules (), it is so sensitive to diffuse background that a total internal reflection fluorescence microscope has been preferred for recording. This measure limited the method to two-dimensional samples, specifically to sectioned slices thinner than 100 nm (). Furthermore, extremely long image acquisition times of 2–12 h were reported to form a meaningful image of biological specimens (). We now overcome these limitations by changing both the fluorescent protein and the mode with which we record the biological sample. In particular, employing the fast reversibly photoswitching fluorescent protein (RSFP) rsFastLime (), a variant (V157G) of the reversibly switchable fluorescent protein Dronpa (), in combination with fast and asynchronous recording (), accelerates image acquisition ∼100-fold, cutting down the recording time to 2–2.5 min. Furthermore, our strategy reduces background, thus allowing us to image from within the interior of intact (nonsliced) cells. Because we record nontriggered, spontaneous -- cycles without synchronization of the detector, we refer to this method as PALM with independently running acquisition (PALMIRA). The arrangement of our wide-field imaging setup is illustrated in . The beam of an Argon Ion Laser (Innova 300, Coherent, Santa Clara, CA) running at a wavelength of 488 nm is intensity-controlled by an acousto-optical tunable filter (, AA Opto Electronic, New York, NY) and expanded by a telescope (). Background is removed by a cleanup filter (Z488, 10, Chroma Technology, Rockingham, VT) and the beam is converted from linear to circular polarization by a quarter wave plate () and subsequently coupled into a regular commercial wide-field microscope (DMIRE 2, Leica Microsystems, Wetzlar, Germany). We assure uniform epiillumination of a 10-m field-of-view by under-illuminating the back aperture of the objective lens (HCX APO 100×/1.30 Oil U-V-I 0.17/D, Leica Microsystems). The fluorescent light is collected by the same objective lens, separated from the laser light by a dichroic filter (495DCXR, Chroma Technology) and imaged onto an electron multiplying charge-coupled device (CCD) camera (IXON-Plus DU-860, Andor Technology, Belfast, Northern Ireland). Stray laser light and background outside the dye's emission spectrum is removed by a notch filter (, DNPF488-25, LOT-Oriel, Darmstadt, Germany) and a bandpass filter (, HC525/50, Semrock, Rochester, NY). The planar positions of the fluorophores in a sparsely activated sample are determined by the use of Hogbom's classical CLEAN algorithm () in conjunction with a mask-fitting algorithm of the Airy spot. The CLEAN algorithm provides data segmentation whereas the localization serves to find the precise position of the emitting molecule. This interplay between the two algorithms is illustrated in . The subtraction of the PSF entailed by the CLEAN algorithm resulted in a more adequate treatment of slightly overlapping Airy peaks as was the case for simple connection-based segmentation procedures. The localization itself is equivalent to the Gaussian mask-fitting algorithm described in Thompson et al. () and runs in the form of a fixed point iteration: At the beginning, the starting point is set to the center of the pixel with the highest photon count. Then, in each iteration, the center of mass of the data multiplied by a PSF centered at the point is determined. Usually, the algorithm converges after a few iterations. If after 50 iterations the path of is longer than the size of an Airy disk, this is considered as a divergence, and the corresponding molecule is discarded. During our typical image acquisition times of 2–2.5 min, we observed a sample drift of some tens of nanometers. Therefore, we added fluorescent microspheres to the samples. By tracking these bright particles (typically more than 2000 photons/frame), we corrected for the errors in the determined positions of individual fluorophores during the post-processing analysis. In the case of negligible readout noise, the epifluorescent counterpart image to the superresolved PALMIRA image is given by the sum of the individual frames. The conventional images shown in and were therefore determined by summing up all recorded frames and setting the lowest pixel count to zero. The RSFP rsFastLime was expressed in the strain HMS 174 (DE3) and purified by Ni-NTA affinity chromatography and subsequent size-exclusion chromatography according to standard procedures (). The purified proteins were concentrated to ≈33 mg/ml and taken up in 100 mM Tris-HCl, 150 mM NaCl, pH 7.5. To acquire time traces of isolated molecules as shown in , coverslips were rinsed with deionized water for 5 min and cleaned in a low pressure plasma system (Femto-RF, Diener Electronic, Nagold, Germany). A phosphate-buffered saline (PBS)-based solution with 1.27 nM rsFastLime, 0.1% (w/v) Polyvinyl alcohol (88 mol % hydrolyzed, Polysciences Europe, Eppelheim, Germany) and 0.32% (w/v) L-ascorbic acid (A.C.S. reagent, Sigma-Aldrich, St. Louis, MO) was prepared. The pH was adjusted to 7.4. A 40-l aliquot of the solution was pipetted onto the coverslip, which was then spin-coated for 20 s at 3000 rpm (Spin-Coater KW-4A, Chemat Technology, Northridge, CA). The overall image acquisition time was 40 s (500 frames/s) and the light intensity (488 nm) was 5.0 kW/cm. To label the cytoplasmic membrane with the RSFP rsFastLime, we expressed a fusion protein consisting of the M13 bacteriophage procoat protein fused to rsFastLime. M13 is integrated into the cytoplasmic membrane. It has two membrane-spanning domains with both termini reaching into the cytoplasm. The coding sequence for rsFastLime () was PCR amplified and inserted into a modified pET28 expression vector harboring a M13-GFP fusion () to replace the GFP coding sequence. The fusion protein was expressed in SURE cells (Stratagene, La Jolla, CA). Thin cryosections were prepared as described previously (,). cells were fixed with 2% PFA (1 vol growth medium + 1 vol 4% (w/v) PFA) for 30 min at room temperature. After centrifugation, cells were postfixed with 4% and 0.1% glutaraldehyde in PBS for 2 h on ice. After being washed twice with PBS-0.02% glycine, cells were embedded in 10% gelatin, cooled on ice, and cut into small blocks. The blocks were infused with 2.3 M sucrose in PBS at 4°C overnight, mounted on metal pins and frozen in liquid nitrogen. 200-nm sections were cut at ∼−110°C using a diamond knife (Diatome, Biel, Switzerland) in an ultracryomicrotome (Leica Microsystems, Germany) and collected using a 1:1 mixture of 2.3 M sucrose () and 2% methyl-cellulose containing 40 mM Cysteamine (BioChemika, ≥98.0% (RT) (Sigma-Aldrich)). The cryosections were deposited on electric glow discharge (minus) treated coverslips, which were rinsed beforehand with deionized water for 5 min and cleaned in a low pressure plasma system (Femto-RF, Diener Electronic, Nagold, Germany). The mixture of methyl cellulose and sucrose which completely covered the sample was removed mechanically. A PBS (pH 7.4) based solution with 0.045% (v/v) Fluospheres (FluoSpheres carboxylate-modified microspheres, 0.2 m, Nile red fluorescent (535/575) × 2% solids, Invitrogen, Carlsbad, CA) and 0.09% (w/v) Polyvinyl alcohol was prepared. A 40-l aliquot of the solution was pipetted onto the sample spinning at 3000 rpm in a spin-coating apparatus. The final image was generated by analyzing 70,000 frames. The overall image acquisition time was 140 s (500 frames/s) and the light intensity (488 nm) was 5.0 kW/cm. PtK2 cells were grown as described previously (). Cells were seeded on coverslips (Menzel, Braunschweig, Germany) in a six-well plate (Nunc, Wiesbaden, Germany) and grown to a confluency of ∼80%. For immunofluorescence labeling, cells were fixated for 4–6 min in ice-cold methanol and subsequently blocked for 10 min in PBS containing 1% BSA (blocking buffer). Cells were incubated with primary antibodies (2 g/ml, anti--tubulin mouse IgG biotin conjugated, Invitrogen) diluted in blocking buffer at room temperature for 1 h followed by PBS washes and 5 min blocking in blocking buffer. For bridging the biotinylated anti--tubulin antibody with biotinylated rsFastLime, cells were incubated with Neutravidin (10 g/ml, Invitrogen) at room temperature for 1 h followed by a short fixation with 3.7% PFA at room temperature for 5 min and subsequent PBS washes. Finally, cells were incubated with biotinylated rsFastLime (50 g/ml, FluoReporter Mini-biotin-XX Protein Labeling Kit, Invitrogen) at room temperature for 1 h. For sample preparation, a PBS-based solution with 0.045% (v/v) Fluospheres, 0.09% (w/v) Polyvinyl alcohol, and 0.29% (w/v) L-ascorbic acid was prepared. The pH was adjusted to 7.4–7.5. A 15-l aliquot of this solution was pipetted onto the sample, which was then spin-coated for 40 s at 3000 rpm. The final image was generated by analyzing 60,000 frames. The overall image acquisition time was 120 s (500 frames/s) and the light intensity (488 nm) was increased stepwise from 3.5 kW/cm to 5.0 kW/cm. Our setup () consists of a light source for excitation, which may be complemented by another one for activation, and a CCD camera imaging the sample at a certain frame rate . The imaging speed is determined by the number of molecular coordinates retrieved per second and thus by the activation rate, while the resolution depends on the unambiguous assignment of detected photons to single molecules. The density of activated fluorophores has to be low enough to avoid simultaneous bursts from molecules with overlapping images on the camera. The number of emitted photons before deactivation or bleaching and thus the achievable resolution is a constant for moderate excitation intensities , while the average on-time of a molecule, , is proportional to 1/. The imaging speed can therefore be dramatically increased by simultaneously increasing and the camera frame rate such that ≈ 1/, thus maintaining the highest activation density allowing reliable assignment of photons to individual molecules. PALMIRA also reduces background noise to a minimum: While frame rates faster than 1/ spread bursts over several frames, increasing readout noise, slower frame rates would result in additional accumulation of background. We used a fast CCD camera that allowed us to compress the readout process to several milliseconds. shows three typical recordings of a sparse sample of the reversible photoswitchable protein rsFastLime () at a frame rate of 500 Hz. At this rather high frequency, our camera still operated at 97% duty cycle. The excellent signal/noise ratio of the data is evident in the time trace (). The root-mean-square of the combined background and readout noise typically averaged to <1 photon/pixel. The data shown in were recorded under pure 488 nm irradiation, which not only switches off dyes in the process of fluorescence reading but also activates dark molecules. Therefore, a dynamic equilibrium is formed during image acquisition with most of the molecules in the dark state and a small fraction of the molecules in the bright state. This fraction can be estimated from our single molecule experiments as the number of frames in which the molecule is visible, divided by the frame number of the last occurrence of such a burst from the molecule. The analysis of >600 time traces showed that the dynamic equilibrium keeps <0.2% of the rsFastLime molecules in their bright state. This activation crosstalk keeps PALMIRA simple because, unlike in PALM or STORM, only a single laser line is used for activation and readout. Activation now occurs at arbitrary times during image acquisition, making the synchronization of illumination and image acquisition redundant. However, an activation density of 0.2% might be too high for some dense samples and would have to be reduced either by prebleaching some of the molecules or if possible by using a different excitation wavelength at which the action crosstalk between excitation and activation is different. If the activation density is lower than optimal, image acquisition can be accelerated by applying auxiliary activation light of a different wavelength. Alternatively, one can increase the excitation intensity beyond the optimum ( < 1/) at the expense of noise. To explore the capabilities of PALMIRA, we first imaged 200-nm-thick cryosections of whose cytoplasmic membrane was labeled with rsFastLime (). We obtained the image by dividing the field of view into 20 nm pixels. Each time a molecule was located within the corresponding image region, that was brighter than the threshold, the pixel value was incremented by one. Note that this image assembly differs from that in Betzig et al. (). In the latter the image is formed by plotting a Gaussian around each fitted position, with the standard deviation equaling the fit uncertainty. We have not followed this procedure because it tends to suggest more dynamic range than actually conveyed by the number of molecules detected. In , the dynamic range directly reflects the real number of events. In all experiments the number of photons per activation event followed geometrical distributions. In the threshold was adjusted to 22 photons tuning the resolution to an estimated 50 nm. Nonetheless, image acquisition times of ≈140 s yielded enough events to form clear images with sufficient dynamic range. shows that several bacteria can be identified. Their membranes are much more clearly resolved than in the conventional image from the same sample (). There is no fundamental reason for this method to be limited to (ultra)thin samples. As long as the number of addressable molecules per unit area is small enough and these molecules are located inside a section where the shape of the PSF does not significantly change, it is also possible to record data inside whole cells or even thicker samples. As a matter of fact, operating at the optimal frame rate and hence being less susceptible to background, PALMIRA allowed us to image stained -tubulin inside whole PtK2 cells. compares wide-field () and PALMIRA () recordings of the same area inside the cell. While individual microtubules are not resolved in the conventional image they can be clearly distinguished in the PALMIRA recording. The resolution is even high enough to pinpoint the somewhat inhomogeneous labeling along the filaments. The distribution of detected photons per event for the data presented in is shown in . Due to the threshold of 1.5 photons in the CLEAN algorithm, detection of events with a small number of photon counts was significantly less probable. For large photon counts per event (>30), the histogram approximately follows a geometrical distribution with an expectation value of ≈21 photons. To estimate the resolution, we scrutinized the object indicated in showing an agglomeration of several rsFastLime proteins. The averaged - and -profiles () reveal that the resolution can be tuned to ≈40 nm by only choosing events with 35 photons or more for the data representation. Note that the “gray values” in our images represent the number of single molecule events per pixel. Hence the resolution directly results from the statistics of recording many events. Higher thresholds would improve the resolution further but reduce the dynamic range. In summary, we demonstrated nanoscale far-field fluorescence imaging based on reversible photoswitching and detecting the photon bursts of individual molecules with recording times of 2–2.5 min. These advancements are due to an asynchronous laissez-faire protocol of data acquisition that just matches the camera acquisition time to the duration of the photon bursts. The 100-fold cut-down in recording time as compared to previous experiments lessens the requirements on long-term stability of the setup. Dedicated background suppression techniques such as total internal reflection fluorescence imaging are no longer generally required for single-molecule, localization-based applications. The use of a standard epifluorescence setup consequently allowed us to image features inside whole cells. It is possible that a molecule is counted more than once in our image. This is because the utilized protein is reversibly switchable and because a single burst of emission may be spread over two or more camera frames. However, the overcounting probability is the same for every molecule, ensuring that the image is an unbiased representation of the fluorophore concentration in the sample. Controlling or knowing the average number of total counts of a molecule is not required. An advantage of reversible switching is that molecules that have emitted too few photons to be registered have another chance to finally contribute to the image. Time-lapse microscopy using PALMIRA will probably benefit from reversibility, because the labeling density is much less altered after the recording. Nonetheless, PALMIRA would work equally well with irreversible switches. In the future, individual planes inside a thicker sample will selectively be activated and imaged. This can be accomplished, for example, by multiphoton-induced activation processes. By utilizing three-dimensional position determination techniques inside these individual planes, and by combining the information from several planes, it will be possible to obtain subdiffraction three-dimensional images of, e.g., whole cells, organelles, or transparent nonbiological samples. Higher threshold values improve the resolution at the expense of dynamic range or imaging speed. This possibility will be partially limited by the geometrical distribution of the emitted photons in a burst. Since the main factor determining the image sharpness is the label itself, the development and use of other (reversibly) switchable fluorophores with stronger photon bursts will directly result in sharper images. Our frame acquisition time of 2 ms is sufficient to collect even 100-times more photons per burst, which will lead to an overall resolution of ≈4 nm. Still, the use of faster detectors will accelerate PALMIRA imaging even further. Limiting the amount of crosstalk in prospective markers is of major importance, because it will allow the imaging of dense samples without prior bleaching. The speed of the switching process itself remains of minor importance because in our asynchronous recording protocol it has no influence on the total acquisition time. Finally, our results once more underscore the huge potential of molecular photoswitching for nanoscale imaging with visible light and regular lenses.

Blood–tissue barriers, which comprise endothelial cells connected by tight junctions (; ), protect the brain, peripheral nerve, myocardium, retina, olfactory epithelium, and the inner ear from the external environment. Metabolism in these tissues is fueled by glucose that is transported across the endothelial barrier via a transcellular mechanism mediated by the type I facilitative glucose transport protein, GLUT1 (; ; ). GLUT1 is a prototypic member of the facilitative glucose transporter family () and of the wider major facilitator superfamily (MFS) of structurally and functionally related transport proteins (; ). GLUT1 is expressed at very high levels in endothelial cells and erythrocytes () where it displays substrate affinities, kinetic properties, and inhibitor pharmacodynamics that distinguish it from other GLUT facilitative sugar transporter members (). Glucose metabolism in erythrocytes and endothelial cells is not rate limited by transport because the glucose transport capacities of these cells greatly exceeds their glycolytic capacities (; ). In spite of this, GLUT1-mediated sugar transport displays acute and adaptive regulation in endothelial cells (; ; ) and acute regulation in erythrocytes (; ; ; ; ) where cellular ATP depletion enhances GLUT1-mediated sugar import capacity (; ; ; ; ). Acute responses occur within seconds to minutes and involve stimulation of existing cell surface glucose transporters (; ; ). Adaptive responses occur over several hours in response to hypoxia and hypoglycemia and involve changes in glucose transporter expression (). GLUT1 is a nucleotide binding protein that, when complexed with ATP, displays reduced glucose import capacity but increased affinity for sugar (; ; ). ATP modulation of GLUT1-mediated transport is competitively inhibited by AMP and ADP, but does not require ATP hydrolysis (). Peptide mapping studies of azidoATP-labeled GLUT1 demonstrate that ATP interacts with GLUT1 residues 301–364 (). This sequence spans transmembrane helices 8 and 9 (TM8 and TM9) and cytoplasmic loop 8–9. Residues 332–343 of this region display 50% sequence identity with a component of the adenylate kinase ATP binding pocket () and mutagenesis of key residues within this subdomain abolishes ATP modulation of transport (). These observations suggest that nucleotide binding pocket minimally consists of L8–9 and a portion of TM9. Competitive antagonism of ATP regulation of transport by AMP eliminates the possibility that transport regulation is a simple consequence of nucleotide binding to GLUT1. Rather, regulation must involve nucleotide- induced GLUT1 conformational changes but the nature of these changes is unknown. Neither is it known whether the details of GLUT1 regulation are isoform specific or reflect a mechanism fundamental to all structurally related GLUT family members that extends to transport catalyzed by other structurally and functionally related MFS proteins. GLUT1 regulation appears to involve rapid changes in GLUT1 intrinsic activity while regulation of the insulin-sensitive transporter GLUT4 involves rapid redistributions of GLUT4 proteins between intracellular and cell surface membranes in addition to GLUT4 activation (; ). The results of the present study suggest that ATP binding to GLUT1 causes the GLUT1 carboxyl terminus to interact with GLUT1 cytoplasmic loop 6–7 in a sequence-specific fashion to inhibit transport. Fresh, de-identified human blood was purchased from Biological Specialties Corporation. Protein assays, Pro Blue coomassie stain, and Supersignal chemiluminescence kits were from Pierce Chemical Co. Nitrocellulose and Immobilon-P were purchased from Fisher Scientific. Purified rabbit IgGs raised against synthetic peptides corresponding to GLUT1 subdomains were obtained from Animal Pharm Services, Inc. These are N-Ab (GLUT1 residues 1–13); L2–3-Ab (GLUT1 residues 85–95); L6–7-Ab (GLUT1 residues 217–231); L7–8-Ab (GLUT1 residues 299–311); C-Ab (GLUT1 residues 480–492). All other reagents were purchased from Sigma-Aldrich. Saline comprises 150 mM NaCl, 10 mM Tris-HCl, and 0.5 mM EDTA, pH 7.4. Lysis medium contained 10 mM Tris-HCl and 0.2 mM EDTA, pH 7.2. Stripping solution contained 2 mM EDTA, 15.2 mM NaOH, pH 12. Tris medium contained 50 mM Tris-HCl, pH 7.4. Kaline consisted of 150 mM KCl, 5 mM HEPES, 4 mM EGTA, and 5 mM MgCl. Ammonium bicarbonate was 0.5% (63 mM), pH 9.0. PBS containing Tween (PBS-T) comprised 140 mM NaCl, 10 mM NaHPO, 3.4 mM KCl, 1.84 mM KHPO, 0.1% Tween, pH 7.3. Stop solution comprises ice-cold Kaline plus cytochalasin B (CB) (10 μM) and phloretin (100 μM). Red cells were isolated by as described previously (). Red cell ghosts were formed by reversible hypotonic lysis of washed red cells (). Glucose transporter (plus endogenous lipids) was purified from human erythrocyte membranes in the absence of reductant as described previously (). The resulting GLUT1 proteoliposomes contain (by protein mass) 90% GLUT1, 8% RhD protein, 2% nucleoside transporter (ENT1) and have a lipid:total protein mass ratio of 1:1 (). Experiments were restricted to the use of GLUT1 preparations in which the stoichiometry of proteoliposomal cytochalasin B binding is 0.48 ± 0.07 mol CB per mol nonreduced GLUT1 and 0.9 ± 0.1 mol CB per mol reduced GLUT1. Transport in red cell ghosts was measured as described previously (). Purified GLUT1 (10 μg) was digested with a 20:1 (protein: enzyme) ratio of purified porcine trypsin (Princeton Separations) in 50 mM Tris-HCl (pH 7.5), 5 mM MgCl, 4 mM ATP (pH 7.5). Digestions were performed at 4°C for 60 min, or the indicated time period. Reactions were immediately loaded onto 15% SDS-PAGE. After SDS-PAGE, samples were transferred to nitrocellulose and blocked overnight in 25% nonfat dry milk/PBS-T. Primary antibody (N-Ab [1:200], L2–3-Ab [1:200], L6–7-Ab [1:500], L7–8-Ab [1:200], L8–9 [1:200], C-Ab [1:15,000], δ-Ab [1:1,000]) was incubated in 3% nonfat dry milk/PBS-T for 1 h at room temperature. Blots were washed three times in PBS-T and incubated with horseradish peroxidase (HRP)–conjugated goat anti-rabbit secondary antibody (1:5,000 dilution) at room temperature for 1 h. Blots were washed three times in PBS-T, developed using Pierce SuperSignal West Pico Chemiluminescent substrate and visualized by autoradiography. Purified GLUT1 (200 ng) in PBS was adsorbed to each well of the ELISA plate for 2 h at 37°C. Plates were blocked with PBS + 3% BSA for 2 h at 37°C. Primary antibody (N-Ab [1:200], L2–3-Ab [1:200], L6–7-Ab [1:500], L7–8-Ab [1:200], L8–9 [1:100 to 1:500], C-Ab [1:15,000], δ-Ab [1:1,000]) was added to each well in PBS + 0.1% BSA ± ATP and binding was allowed to proceed for 2 h. The plate was washed five times with PBS, and then each well was incubated with HRP-conjugated goat anti-rabbit secondary antibody in PBS + 0.1% BSA (1:5,000 dilution) for 1 h at 37°C. The plate was washed five times with PBS and wells were developed using 100 μl of 1-Step ABTS solution (Pierce Chemical Co.). Primary IgG binding was quantitated as absorbance at 415 nm. This chimera substitutes the middle loop (L6–7) with that of its rat GLUT4 counterpart and was constructed using a six-step PCR protocol. In PCR 1A, a HindIII primer complimentary to the 5′ end of human GLUT1 and a reverse primer complimentary to nucleotides 600–619 of GLUT1 and 667–685 of GLUT4 was used to generate a fragment containing TM 1–6 (nucleotides 1–620) of GLUT1. In a separate reaction, PCR1B, primer (TGCATCGTGCTGCCCTTCTGTCCTGAGAGCCCCCGA) containing sequence complimentary to nucleotides 600–619 of GLUT1 and 667–685 of GLUT4 (5′end of L6–7) and a reverse primer (containing a NotI restriction site) complimentary to the 3′ end of GLUT4 was used to generate a fragment containing the middle loop and TM 7–12 (nucleotides 667–1531) of GLUT4. In PCR2, the HindIII primer complimentary to the 5′ end of human GLUT1 and the reverse primer (containing a NotI restriction site) complimentary to the 3′ end of GLUT4 were used along with the products from PCR1A and B to generate an intermediate chimera containing sequence from nucleotides 1–620 of GLUT1 and 667–1531 of GLUT4. This intermediate chimera contained TM 1–6 of GLUT1 in frame with L6–7 and TM 7–12 of GLUT4. In PCR 3A, the HindIII primer complimentary to the 5′ end of human GLUT1 and a primer (CGCACCCAGGGGCAGCCTATCCTCATCGCTGTGGTC) complimentary to nucleotides 844–862 of GLUT4 and 815–829 of GLUT1 were used along with the product from PCR2 to generate a fragment containing TM 1–6 of GLUT1 (nucleotides 1–619) in frame with L6–7 (nucleotides 667–862) of GLUT4. In PCR 3B, a primer (CGCACCCAGGGGCAGCCTATCCTCATCGCTGTGGTC) complimentary to nucleotides 844–862 of GLUT4 and nucleotides 815–829 of GLUT1 were used with wild-type GLUT1 plasmid as a template to generate a fragment containing TM 7–12 (nucleotides 815–1480) of GLUT1. In the final reaction, PCR 4, HindIII primer complimentary to the 5′ end of human GLUT1 and the NotI primer complimentary to the 3′ end of GLUT1 were used with the products of PCR 3A and B as a template generating the final product, which contained TM 1–6 (nucleotides 1–619) of GLUT1, followed by L6–7 of GLUT4 (nucleotides 667–862), followed by TM 7–12 (nucleotides 815–1480) of GLUT1. This PCR product was digested with HindIII and NotI and ligated into the mammalian expression vector pcDNA 3.1+, cut with the same enzymes. Sequences were confirmed by sequencing and the chimera was transiently expressed in HEK 293 cells as described previously (). GLUT1 was covalently modified at accessible lysine residues (±ATP) using sulfosuccinimidyl-6-(biotinamido) hexanoate (sulfo-NHS- LC-biotin) as described previously (unpublished data). Detection of modified residues was achieved by limited proteolysis of labeled GLUT1 followed by RP-HPLC separation of fragments and ESI-MS/MS identification of peptides as described previously (unpublished data). #text T h i s s t u d y e x a m i n e s t h e s t r u c t u r a l b a s i s o f G L U T 1 m o d u l a t i o n b y c y t o p l a s m i c A T P a n d A M P . O u r f i n d i n g s s u g g e s t t h a t t h e G L U T 1 c y t o p l a s m i c C t e r m i n u s a n d t h e c y t o p l a s m i c l o o p l i n k i n g T M s 6 a n d 7 i n t e r a c t i n a n A T P - d e p e n d e n t f a s h i o n t o m o d u l a t e t r a n s p o r t .

During embryonic development, columnar epithelial cells polarize their plasma membrane into distinct apical and basolateral domains that are separated by tight junctions. Once established, this polarity has to be maintained, and it has to be ensured that apical and basolateral transmembrane proteins are correctly sorted to the respective target domain (; ). Frequently, a first step in basolateral sorting is the recognition of a targeting determinant encoded in the cytoplasmic tail of transmembrane proteins. Perhaps the best-characterized sorting determinants are similar to endocytosis signals and consist of either LL- or Y-based sorting signals. These basolateral signals are in general cis-dominant over apical sorting information such as - or -glycosylations in a protein's ectodomain (). Although it is still unclear how proteins with LL-based sorting signals are selected for basolateral delivery, a better understanding has been achieved for the Y-based sorting signals (). Typically, Y-based sorting signals are recognized by the medium subunits of heterotetrameric clathrin adaptor protein (AP) complexes (). There are four major types, AP-1 through AP-4, all sharing the same general subunit organization with two large subunits (γ, α, δ, ɛ, and β1–β4), one medium subunit (μ1–μ4), and one small subunit (σ1–σ4; ). Although AP-1, AP-3, and AP-4 are all thought to mediate cargo selection and vesicle formation at the TGN or endosomes, AP-2 is involved in clathrin-mediated endocytosis (; ). Most polarized epithelial cells actually contain two highly similar AP-1 complexes, the ubiquitously expressed AP-1A, thought to mediate endosomal sorting, and the epithelial-specific AP-1B complex involved in basolateral targeting (). AP-1A and AP-1B differ only in the incorporation of their medium subunits μ1A or μ1B, respectively. Despite this close homology, AP-1A and AP-1B have largely nonoverlapping functions and form distinct vesicle populations (, ). The different roles of AP-1A and AP-1B in sorting can partially be explained by their differential localization to the TGN and recycling endosomes, respectively (; ). Besides AP-1B, the only other adaptor complex implicated in basolateral sorting is AP-4 (). Although AP-4 is thought to mediate basolateral sorting at the TGN, AP-1B is involved in recycling internalized cargo back from recycling endosomes to the basolateral membrane. In addition, AP-1B may also regulate basolateral delivery of newly synthesized proteins, which may travel through recycling endosomes on their way to the plasma membrane (, ; ). Despite the clear advances in our understanding of adaptor complexes involved, little is known about the relationship of sorting basolateral cargo at the TGN or the recycling endosomes (e.g., to what extent does biosynthetic cargo move through recycling endosomes?), and the interaction of specific basolateral sorting determinants with the individual μ-chains has not been established; however, the latter information would add considerably to our mechanistic understanding of basolateral sorting pathways. Regardless, once the basolateral transport vesicles are formed, tethering to and fusion with the correct target site has to be ensured. The mammalian exocyst, an eight-subunit complex, is thought to facilitate tethering of basolateral vesicles with the target site, and at least two of its subunits associate with AP-1B vesicles (; ). After tethering, the complex may rearrange to bring vesicle and plasma membrane into close contact to allow for SNARE pairing and subsequent fusion (). SNAREs are soluble NSF attachment protein receptors. Plasma membrane SNAREs of the syntaxin family (e.g., syntaxins 1, 2, 3, and 4) and the synaptosomal-associated protein of 23 kD (SNAP-23) form the so-called target SNAREs (t-SNAREs) that, upon interaction with the vesicle-associated membrane proteins (VAMPs/v-SNAREs), mediate the exocytic fusion reaction (; ; ). Epithelial cells express three different t-SNAREs at the plasma membrane. Syntaxin 3 localizes to the apical domain, syntaxin 4 is basolateral, and syntaxin 2 in renal epithelial cells localizes either to the apical or basolateral membrane, whereas exogenously expressed syntaxin 2 in MDCK cells is nonpolarized (; ). Recently, it has been shown that the correct localization of syntaxin 3 is necessary for epithelial polarity and correct targeting of proteins to the apical membrane (; ; ). The role of syntaxin 4 in basolateral sorting is suspected but has not yet been demonstrated. Furthermore, the nature of the corresponding v-SNARE is still elusive. It has been shown that the addition of tetanus neurotoxin (TeNT) to permeabilized MDCK cells slowed the transport rate of vesicular stomatitis virus glycoprotein (VSVG) to the basolateral domain but not the transport of influenza HA to the apical domain (). Apical targeting was then shown to involve the TeNT-resistant v-SNARE TI-VAMP (VAMP7) and syntaxin 3–dependent fusion (; ; ). However, the exact nature of the SNARE proteins involved in basolateral targeting was not determined, and it remains to be shown whether treatment with TeNT results in any sorting defects to the basolateral domain. TeNT is a highly specific clostridial neurotoxin that has three known targets: synaptobrevin 1 (VAMP 1), synaptobrevin 2 (VAMP 2), and cellubrevin (VAMP 3; ). Although synaptobrevin 1 is mainly brain specific, synaptobrevin 2 has a broader expression profile, including adipocytes and exocrine tissues (; ). In contrast, cellubrevin is ubiquitously expressed and has been shown to play a role in epithelial cell migration (; ). Furthermore, cellubrevin has been implicated in recycling of transferrin receptors (TfnRs) in fibroblasts and apical recycling pathways in MDCK cells (; ; ). However, because apical targeting was shown to be resistant to treatments with TeNT, it seems unlikely that cellubrevin is involved in this pathway (; ). Thus, questions concerning the exact role of cellubrevin in polarized membrane trafficking and its potential connection to adaptor complexes still remain. To better understand the TeNT sensitivity of basolateral sorting and the role of cellubrevin in polarized epithelial cells, we examined MDCK cell lines stably transfected with GFP-cellubrevin (GFP-cb) and TeNT or an enzymatic inactive mutant of TeNT. established MDCK cells stably transfected with cDNAs encoding cellubrevin tagged with GFP at its N terminus (GFP-cb) and TeNT (+TeNT) or the inactive E234Q mutant of TeNT (+mutant TeNT). Here, we tested the polarized localization of GFP-cb in filter-grown, fully polarized MDCK cells. As shown in (top), cellubrevin at least partially colocalized with the marker protein gp58 at the basolateral membrane and in intracellular puncta most likely corresponding to recycling endosomes (, top, arrows). The basolateral localization of GFP-cb may be a result of membrane fusion reactions; therefore, cellubrevin may localize within the basolateral membrane or underneath this domain as part of docked vesicles. It may also be present in early endosomes underlying the basolateral membrane (basolateral early endosomes) after removal from the plasma membrane. To test whether cellubrevin localizes at least in part within the basolateral plasma membrane, we tagged cellubrevin at its luminal, C-terminal end with the myc-epitope (cb-myc). Cb-myc was transiently expressed in polarized MDCK cells. Transfected cells were incubated with anti-myc antibodies before fixation to stain cellubrevin at the membrane. As expected, we detected a fraction of cb-myc at the basolateral surface (Fig. S1, available at ), confirming the localization observed for GFP-cb. Importantly, we did not detect any surface staining at the apical membrane (Fig. S1). Coexpression of GFP-cb with TeNT resulted in cleavage of GFP-cb and cytosolic accumulation of GFP (, bottom; ). Interestingly, stably expressing TeNT in MDCK cells does not lead to a mislocalization of the basolateral marker protein gp58, indicating that overall polarity was not disturbed in this assay (, bottom). Taken together, these data show that cellubrevin localizes in part to the basolateral membrane, where cellubrevin may be involved in fusion events between exocytic vesicles and the basolateral plasma membrane. Moreover, we conclude that the established cell lines coexpressing GFP-cb and TeNT are suitable for analyzing a potential role for cellubrevin in this process. To test whether cellubrevin plays a role in fusion events at the basolateral membrane, we sought to coprecipitate cellubrevin with the basolateral t-SNARE syntaxin 4 or the apical t-SNARE syntaxin 3. To this end, we used defective adenoviruses to express myc-tagged versions of syntaxin 4 or 3 in GFP-cb–expressing MDCK cells. It has previously been shown that myc-tagged syntaxin 3 and 4 localize and function correctly (; ). As shown in , anti-GFP antibodies efficiently precipitated GFP-cb (lanes 4 and 6). Importantly, although syntaxin 4 was abundantly brought down with cellubrevin (lane 6), only small amounts of syntaxin 3 coprecipitated (lane 4). This shows that cellubrevin preferentially forms SNARE pairs with syntaxin 4. Therefore, we hypothesize that cellubrevin plays a role in some vesicle fusion events at the basolateral plasma membrane in MDCK cells. To answer the question of whether cellubrevin is involved in membrane trafficking to the basolateral membrane, we analyzed the sorting phenotypes of various cargos in the presence of TeNT. As shown in , endogenous as well as virally expressed TfnR, which have Y-based sorting motifs () and are basolateral in our control cells, showed a nonpolarized distribution in the presence of TeNT. In contrast, FcII-B2 receptors (FcRs), which have an LL-based sorting motif (; ), remained localized to the basolateral domain in the presence of TeNT (). Next, we tested the surface expression of basolaterally localized VSVG and apical A-VSVG (). Surprisingly, even though VSVG has a Y-based sorting motif like TfnR (), this protein remained at the basolateral membrane in the presence of TeNT (, left). Importantly, A-VSVG sorting to the apical domain was also not disturbed (, right). Finally, we tested the sorting of LDLR and two of its mutants. LDLR contains two basolateral sorting determinants (). The proximal signal more closely situated toward the transmembrane domain, and surrounding a tyrosine residue at position 18 is an NPXY motif (). By contrast, the distal targeting determinant is considered a noncanonical Y-based motif (). The first mutant receptor analyzed was LDLR(Y18A), in which the critical tyrosine 18 was mutated to an alanine. The second mutant construct was truncated at position 27, right after the proximal targeting determinant (LDLR-CT27). We found that sorting of LDLR and LDLR(Y18A) is insensitive to TeNT (, left). In contrast, LDLR-CT27 was missorted in the presence of TeNT (, right). Therefore, only basolateral targeting mediated by LDLR's proximal sorting determinant is sensitive to TeNT expression. In summary, we found that two basolateral receptors, TfnR and LDLR-CT27, are missorted to the apical domain when TeNT is present. Next, we performed rescue experiments to confirm that cleavage of cellubrevin is the reason for the observed TeNT sensitivity of basolateral sorting. There are only three known v-SNAREs that are substrates for TeNT. However, although cellubrevin and synaptobrevin 2 are cleaved very efficiently, cleavage of synaptobrevin 1 is less efficient (; ). Moreover, although cellubrevin is ubiquitously expressed, synaptobrevin 1 and 2 could not be detected in MDCK cells by Western blotting (). Therefore, to analyze rescue, we added an N-terminal RFP tag to TeNT-resistant mutants of human cellubrevin (RFP-cb-VW) and, for control purposes, human synaptobrevin 2 (RFP-syn2-VW) by mutating Q63/F64 or Q76/F77 to VW (), respectively. These constructs, together with LDLR-CT27, were transiently expressed in MDCK cells stably expressing TeNT and GFP-cb. Cells were processed for surface staining of LDLR-CT27 with or without inhibiting protein synthesis with cycloheximide for 2 h before fixation. As shown in , RFP-cb-VW rescued the basolateral sorting of LDLR-CT27 independently of expression levels in ∼70% of all analyzed cells (cells showing lower levels of RFP-cb-VW expression are shown in the left panels). Furthermore, this rescue was independent of ongoing protein synthesis (). In contrast, RFP-syn2-VW rescued only if highly overexpressed (, RFP-syn2-VW, compare right and left panels). Again, the same result was obtained after adding cycloheximide. Overall, RFP-syn2-VW rescued LDLR-27 sorting in ∼30% of all cells analyzed, counting cells independent of their expression levels (). Next, we transiently expressed RFP-cb-VW or RPF-syn2-VW in the cell line stably expressing GFP-cb and found that RFP-cb-VW colocalized with GFP-cb in endosomes and at the plasma membrane, whereas RFP-syn2-VW only partially co localized with GFP-cb in endosomes (Fig. S2, available at ). In addition, we could barely detect RFP-syn2-VW at the plasma membrane (Fig. S2), indicating that even if synaptobrevin 2 were expressed in MDCK cells, it is not likely to be involved in fusion events at the basolateral plasma membrane. In summary, from these rescue experiments, we conclude that cellubrevin is indeed the v-SNARE needed for basolateral sorting. LDLR-CT27 is a receptor that is internalized rapidly from the plasma membrane (). Therefore, by analyzing LDLR-CT27 missorting at steady state, we can so far make the conclusion that LDLR-CT27 is dependent on cellubrevin function during endocytic recycling. To address the question of whether there might already be a TeNT-sensitive step in LDLR-CT27 sorting during biosynthetic delivery, we performed radioactive pulse-chase experiments coupled to vectorial surface biotinylation with MDCK cells stably expressing GFP-cb and TeNT or its enzymatic mutant and virally expressing LDLR-CT27. Surprisingly, after a 1-h chase, newly synthesized LDLR-CT27 arrived directly at the apical surface (). The same was true for earlier time points (30 min), when newly synthesized LDLR-CT27 starts to appear at the plasma membrane (unpublished data). The overall apical missorting measured by this pulse-chase experiment is only ∼50% and seems weaker than the sorting phenotypes observed by immunofluorescence. However, our cell lines were not clonal, as judged by the fact that not all cells in the population expressed GFP-cb and TeNT (compare , , and for appearance of cells without GFP staining). As a result, the coinfection rate of GFP-cb– and TeNT-expressing cells with LDLR-CT27 virus was only ∼50%. As expected, LDLR-CT27 was directly sorted to the basolateral membrane in our control cell lines expressing mutant TeNT (). Therefore, it seems that the sorting of LDLR-CT27 is already sensitive to TeNT during biosynthetic delivery in addition to missorting during recycling. To further analyze which basolateral sorting pathways might depend on cellubrevin, we analyzed cellubrevin's subcellular localization by indirect immunofluorescence. We found that in addition to the plasma membrane, GFP-cb localized to a perinuclear region distinct from the Golgi complex, as demonstrated by costaining with the cis-Golgi marker GM130 (unpublished data) and on endosomal populations throughout the cells. Furthermore, cellubrevin showed in ∼60% of all cells analyzed staining patterns distinct from TGN38, a marker for AP-1A–positive TGN subdomains (; ). In contrast, in virtually all cells analyzed, cellubrevin showed colocalization with TfnR accumulated in recycling endosomes by incubating the cells for 2 h at 20°C before fixation (). Moreover, cellubrevin partially colocalized with γ-adaptin, one of the large subunits of AP-1A and AP-1B (unpublished data). To directly compare cellubrevin's localization relative to AP-1A or AP-1B, we used defective adenoviruses to express HA-tagged μ1A or myc-tagged μ1B (). Again, there was only limited colocalization between GFP-cb and AP-1A–HA (only in ∼15% of the cells analyzed), but we observed colocalization between GFP-cb and AP-1B–myc in 90% of the cells expressing both markers (). Finally, as additional controls, we analyzed cellubrevin tagged at the N or C terminus with the myc epitope transiently expressed in MDCK cells (myc-cb and cb-myc). The myc-tagged cellubrevin proteins confirmed our analysis with GFP-cb. Myc-cb and cb-myc both colocalized with TfnR and showed essentially the same endosomal and plasma membrane staining pattern as GFP-cb (unpublished data). In addition to immunofluorescence analysis, we investigated the colocalization of cellubrevin and AP-1B ultrastructurally by immuno-EM on specimens stably expressing GFP-cb and transiently expressing μ1B-myc. As shown in , we detected colocalization of cellubrevin (15 nm gold) and AP-1B–myc (10 nm gold) in endosomes (, arrows) and in clathrin-coated vesicles (, C and D, arrows). As expected from the immunofluorescence data, we also found cellubrevin in endosomal populations negative for AP-1B labeling, which perhaps represent early endosomes. In summary, we conclude that cellubrevin colocalizes with AP-1B in recycling endosomes and clathrin-coated vesicles. Given that TfnR, cellubrevin, and AP-1B all localize to recycling endosomes, we asked whether cleavage of cellubrevin by TeNT interferes with the colocalization of TfnR and AP-1B in this compartment. Thus, we virally expressed μ1A-HA or μ1B-myc in MDCK cells stably expressing GFP-cb (false color blue) and TeNT or mutant TeNT. As shown in , TeNT expression had no effect on the perinuclear localization of AP-1A–HA in ∼90% of the cells analyzed (data counts not depicted). However, AP-1B–myc staining was scattered throughout the cell in >95% of the cells analyzed (; data counts not depicted). Moreover, costaining for TfnR (false color green) revealed that TfnR staining was also more dispersed in cells stably expressing TeNT as compared with the control cells without TeNT. Most important, although in control cells AP-1B–myc and TfnR colocalized at recycling endosomes together with GFP-cb (; note extensive yellow staining in magnified inset), AP-1B–myc and TfnR no longer colocalized after cleavage of cellubrevin (; note distinct red and green staining in magnified inset). Therefore, in addition to cellubrevin's role in membrane fusion at the basolateral plasma membrane, as evidenced by observed SNARE pairing between cellubrevin and syntaxin 4 (), cellubrevin may be required for the homeostasis of recycling endosomes. Furthermore, the selective disruption of AP-1B staining suggests that cellubrevin plays a role in AP-1B–mediated basolateral targeting. Because AP-1B's localization at recycling endosomes depends on functional cellubrevin, we asked which cargos analyzed in this study are AP-1B dependent and whether there is a correlation between AP-1B dependency and TeNT sensitivity in basolateral sorting. In the past, we designated basolateral transmembrane proteins as AP-1B dependent for sorting based on their behavior in the μ1B-negative porcine kidney epithelial cell line LLC-PK1 (). In this cell line, receptors with Y-based sorting motifs, such as TfnR, LDLR, and VSVG, were missorted to the apical domain, and this sorting phenotype was reversed by exogenously expressing μ1B in LLC-PK1 cells (LLC-PK1∷μ1B; , ; ). In contrast, receptors with LL-based signals, such as FcR, remained located to the basolateral domain independent of μ1B expression (). However, the sorting data available in LLC-PK1 cells were not complete. Therefore, we investigated sorting behaviors of additional cargos in LLC-PK1∷μ1A, LLC-PK1∷μ1B, or LLC-PK1 cells expressing a mutated μ1B (LLC-PK1∷μ1B). μ1B has four mutations (F172A, D174A, W408A, and R410A), which abolish μ1B's binding to sorting peptides (). First, we analyzed LDLR and its mutants LDLR-CT27 (proximal signal intact) and LDLR(Y18A) (distal signal intact). As shown in and Fig. S3 (available at ), we found that LDLR and LDLR-CT27, which are basolateral in MDCK cells, are apical in LLC-PK1∷μ1A cells. In both cases, basolateral sorting was restored in LLC-PK1∷μ1B cells (; ). However, only LDLR-CT27 showed a missorting phenotype in LLC-PK1∷μ1B cells. The different sorting behaviors of LDLR and LDLR-CT27 might be explained if the distal signal is AP-1B independent. Indeed, LDLR(Y18A) was localized to the basolateral membrane independent of μ1B expression (; ). Next, we analyzed TfnR and VSVG. Both cargos are apical in LLC-PK1∷μ1A cells and basolateral in LLC-PK1 cells expressing μ1B. Surprisingly, basolateral sorting of TfnR and VSVG was also restored in LLC-PK1 cells expressing μ1B (; , ; ). The reason for this lack of missorting could be either that μ1B still binds to the respective sorting signals or that TfnR and VSVG interact with putative (AP-1B) coadaptors instead. As expected, FcR and A-VSVG were sorted to the basolateral or apical domain, respectively, in all cell lines tested (; ; ). To gain a better understanding of which alternative adaptor complexes might be involved in basolateral sorting of LDLR(Y18A), TfnR, and VSVG, we tested by yeast two-hybrid assay the interactions of individual sorting signals with various μ-chains (μ1A, μ1B, μ1B, μ2, μ3A, or μ4). To this end, we analyzed the proximal and distal sorting determinants of LDLR and TfnR, as well as the sorting signal of VSVG. As a positive control, we also analyzed the sorting signal of TGN38 ( and Fig. S4, available at ). For LDLR, we found that the distal targeting signal GYSY (present in LDLR[Y18A]) interacted with μ2, μ3A, and μ4, suggesting that LDLR(Y18A) may use AP-3 or AP-4 to exit the TGN. In contrast, the proximal targeting signal of LDLR, NPXY (present in LDLR-CT27), does not interact with any of the μ-chains. However, it should be noted that although the NPXY signal is a well-established endocytic signal, AP-2 does not bind directly to NPXY. In the case of LDLR, this interaction is bridged by the coadaptors ARH/Dab2/numb (). Perhaps similar coadaptors are involved in LDLR sorting from recycling endosomes. TfnR also has two sorting determinants. One is the proximal, endocytic signal YTRF, and the other is the distal signal GNDS, thought to be involved in correct recycling of TfnR (). We found no interactions between the GNDS signal and any μ-chains. In contrast, we found interactions for the YTRF motif with all μ-chains tested. The strongest interactions were observed between YTRF and μ2/μ4, suggesting that TfnR may use AP-4 to exit the TGN during biosynthetic delivery. As expected, we detected no interactions with the mutant μ1B (Fig. S4 B). The situation was similar for TGN38, whose sorting signal () also interacted with all tested μ-chains (strongest with μ3A and μ4). Finally, the YTDI signal of VSVG () interacted with μ1B and, to a weaker extent, with μ4. Interestingly, the interaction between YTDI and μ1B was not disrupted when tested against μ1B (Fig. S4 B), which may explain why VSVG is not missorted to the apical domain in LLC-PK1∷μ1B cells. SNARE proteins are fundamentally important for all known intracellular fusion events between transport vesicles and target membranes. For example, apical targeting depends on the correct sorting of syntaxin 3 to the apical domain, and its missorting to the basolateral domain results in loss of cellular polarity (; ; ). Furthermore, at the apical membrane syntaxin 3 forms SNARE pairs with the v-SNARE, TI-VAMP, and SNAP-23 (). The SNARE pairs involved in basolateral sorting, however, remained elusive. Using TeNT-expressing MDCK cells, we have now provided evidence that at least a subset of basolateral vesicles uses cellubrevin for basolateral sorting. Moreover, we were able to coprecipitate cellubrevin with the basolaterally localized syntaxin 4, indicating that syntaxin 4 and cellubrevin form SNARE pairs during exocytosis at the basolateral membrane. Besides the basolateral membrane, we found cellubrevin localizing to different endosomal populations, including TfnR and AP-1B–positive recycling endosomes. These data fit very well with previous studies demonstrating a role for cellubrevin in TfnR recycling from recycling endosomes in fibroblasts (; ). In addition, by immuno-EM, we found cellubrevin colocalizing with AP-1B in endosomes and clathrin-coated vesicles. It has previously been shown that some v-SNAREs directly interact with APs. For example, the TGN-localized VAMP 4 has an LL-based motif needed for interaction with AP-1 (). Similarly, TI-VAMP interacts with δ-adaptin/AP-3 through its aminoterminal longin domain (). However, we found no evidence that AP-1B directly recognizes cellubrevin and, at least by yeast two-hybrid assay, cellubrevin failed to interact with μ1B (unpublished data). Therefore, we propose that cellubrevin is incorporated into AP-1B vesicles through interactions with putative coadaptors. For instance, in mammalian cells, EpsinR helps incorporating Vti1b into AP-1A vesicles at the TGN (; ). Likewise, in yeast cells, the EpsinR homologue Ent3p interacts specifically with Vti1p (). Future experiments will be aimed at identifying putative coadaptors for AP-1B coats that might help incorporating cellubrevin into AP-1B vesicles. Interestingly, the cleavage of cellubrevin by TeNT results not only in a more scattered endosomal staining of TfnR but also in a loss of perinuclear AP-1B staining. Because the colocalization of TfnR and AP-1B was lost also, these data suggest a role for cellubrevin in maintaining functional recycling endosomes. It seems that without cellubrevin, cargo such as TfnR can no longer enter this compartment. Alternatively, fusion-incompetent AP-1B vesicles may titer out components needed to generate new vesicles, also leading to a disruption of recycling endosomes (indicated in as isolated entities as opposed to the normal tubular network []). Without functional (AP-1B–positive) recycling endosomes, however, cargo that is normally sorted into AP-1B vesicles at recycling endosomes for basolateral delivery will be missorted to the apical membrane if no alternative pathways can be used instead. It has been noted that the rate of basolateral delivery of VSVG, but not the apical delivery of influenza HA protein, is sensitive to treatment with TeNT (). In this study, we show that apical cargos (A-VSVG) and AP-1B–independent basolateral cargos (LDLR[Y18A] and FcR) are sorted correctly to the apical or basolateral domain, respectively, independent of functional cellubrevin. Although we still do not know which adaptor complex sorts FcR to the basolateral membrane, based on yeast two-hybrid interactions, we propose that LDLR(Y18A) is sorted by AP-4, as has been suggested previously for wild-type LDLR (). In contrast, we observed apical missorting of the AP-1B–dependent cargos TfnR and LDLR-CT27. Surprisingly, however, VSVG, the protein perhaps most often used in the literature as “AP-1B–dependent” cargo, was not missorted. These data are summarized in . Although shows the situation without TeNT, shows the scenario with TeNT. For simplicity, we are omitting sorting into lysosomes and retrieval pathways. We found that LDLR-CT27 is missorted during recycling at steady state ( and ) and directly during biosynthetic delivery (). The latter finding suggests that LDLR-CT27 normally traffics to the basolateral surface via recycling endosomes and that no alternative pathways exist. Indeed, we found no interaction between LDLR-CT27 and any adaptor complex μ-chains, and this receptor was entirely dependent on AP-1B for basolateral sorting ( and ). In agreement with this, we previously demonstrated that LDLR specifically cross-linked to AP-1B (). In addition, LDLR-CT27's AP-1B dependence was recently underlined by its apical missorting in MDCK cells, in which μ1B was knocked down by siRNA (Maday, S., and I. Mellman, personal communication). Like LDLR-CT27, TfnR is missorted during recycling. However, unlike LDLR-CT27, TfnR interacted well with alternative adaptor complex μ-chains (). We propose that during biosynthetic delivery, TfnR is sorted via AP-4 directly from the TGN to the basolateral membrane. In agreement with this, a small fraction of TfnR was missorted in MDCK cells incubated with μ4 anti-sense RNA (). Furthermore, showed that knock down of μ1B in MDCK cells by siRNA resulted in missorting of TfnR during recycling, but not during biosynthetic delivery, also indicating that newly synthesized TfnR may be sorted directly from the TGN to the basolateral membrane. We previously found that VSVG sorting to the basolateral membrane was AP-1B dependent (), and this cargo was used in studies to link small GTPases to the AP-1B pathway (). Furthermore, we showed that VSVG moves through recycling endosomes during biosynthetic delivery in semipolarized MDCK cells, indicating an involvement of AP-1B during biosynthetic delivery (). Therefore, we expected to see a missorting phenotype for VSVG in TeNT-expressing MDCK cells. It should be noted however, that tailless VSVG or VSVG with a mutated Y-based motif showed only weak apical missorting (33 and 37%, respectively; ). Furthermore, in MDCK cells incubated with μ1B siRNA, biosynthetic delivery of VSVG to the basolateral membrane still occurred at ∼60% (). Therefore, VSVG may use alternative sorting pathways in the absence of AP-1B or functional sorting signals. The use of alternative adaptor complexes by VSVG may be more pronounced in cells with dysfunctional recycling endosomes because of the expression of TeNT. Indeed, when recycling endosomes were inactivated enzymatically, the majority of VSVG was retained within the cells, but a small fraction was delivered to the basolateral membrane (). VSVG was shown to interact with δ-adaptin/AP-3 (); however, AP-3's main function is to facilitate lysosomal sorting and not plasma membrane delivery. Here, we show by yeast two-hybrid analysis that VSVG interacts with μ4 (). Thus, we propose that VSVG may use AP-4 as an alternative adaptor complex to reach the basolateral membrane. The interplay and possible competition of the different adaptors binding VSVG (AP-1B, AP-3, and AP-4) should now be sorted. In conclusion, we found that, in general, cellubrevin is needed for basolateral sorting of AP-1B–dependent cargo. In addition, this v-SNARE is required for maintaining functional (AP-1B–positive) recycling endosomes. Collectively, these data strongly suggest a functional connection between cellubrevin and AP-1B in membrane trafficking to the basolateral domain. RFP-tagged versions of TeNT-resistant cellubrevin and synaptobrevin 2 were cloned by PCR-based site-directed mutagenesis using human cellubrevin cDNA () and human synaptobrevin 2 cDNA (OriGene Technologies) as templates using the following C-terminal 5′-CTTGGCTGCGCTCGTTTCCCATACAGAAGCGCCTGCCTGCAG-3′ and N-terminal primers 5′-CTGCAGGCAGGCGCTTCTGTATGGGAAACGAGCGCAGCCAAG-3′ to generate cellubrevin(Q63V/F64W) and C-terminal 5′-GCTTGGCTGCGCTTGTTTCCCACACGGAGGCCCCCGCCTGGAG-3′ and N-terminal primers 5′-CTCCAGGCGGGGGCCTCCGTGTGGGAAACAAGCGCAGCCAAGC-3′ to generate synaptobrevin 2(Q76V/F77W). Subsequently, mutated cellubrevin or synaptobrevin 2 was amplified by PCR and cloned into pRKV-RFP as BamHI–HindIII fragments. PRKV-RFP was generated by amplifying RFP as EcoRI–BamHI PCR fragments using FcR-RFP as template. PCR products were verified by sequencing, and no errors were found. Sorting peptides as indicated in were translated into DNA sequences and amplified by PCR as overhangs introducing an EcoRI site at the N terminus for cloning into the multiple cloning site of pAS2-1. As C-terminal primer, we used a sequence priming ∼1 kbp downstream of the multiple cloning site: 5′-CCTGTTACTAGTGGCTGCTGCCAG–3′. The PCR products were cloned as EcoRI–SpeI fragments into pAS2-1. As a result, the sorting peptides were fused in frame with the Gal4 binding domain. The constructs were verified by sequencing, and no errors were found. The different μ-chains cloned into pACT-2 as fusions with the Gal4 activating domain, were a gift from J. Bonifacino (National Institutes of Health, Bethesda, MD). μ1B was amplified using PCR from cell extracts of LLC-PK1∷μ1B cells and cloned into pACT2. Defective adenoviruses encoding LDLR, TfnR, VSVG-ts045-GFP, or A-VSVG-ts045-YFP were as described previously (, ). Defective adenoviruses encoding LDLR(Y18A), LDLR-CT27, or FcR-RFP were prepared by homologous recombination as described previously (). FcR-RFP cDNA in pShuttle was a gift from I. Mellman (Yale University, New Haven, CT), and myc-tagged syntaxin 3 or 4 in pShuttle was a gift from T. Weimbs (University of California, Santa Barbara, Santa Barbara, CA). Stably transfected MDCK cells were maintained in MEM containing 7% (vol/vol) fetal bovine serum, 2 mM -glutamine, 200 μg/ml geneticin, 4 μg/ml puromycin, and 100 μg/ml penicillin/streptomycin as previously described (). LLC-PK1 cells stably transfected with μ1A, μ1B, or μ1B were maintained as described previously (; ). To allow for polarization, cells were seeded on polycarbonate membrane filters at a density of 4 × 10 cells per 12-mm filter (for immunofluorescence) or 8 × 10 cells per 24-mm filter (for biochemical experiments; 0.4-μm pore size; Corning-Costar transwell units) and cultured for 4–6 d with changes of medium in the basolateral chamber every day. For intracellular localization experiments, cells were seeded on Alcian blue–coated coverslips and cultured for 2–4 d. For anti–γ-adaptin staining, cells were fixed in −20°C methanol for 5 min followed by a 5-min incubation in PBS (PBS [0.2 g/liter KCl, 0.2 g/liter KHPO, 8 g/liter NaCl, and 2.17 g/liter NaHPO × 7 HO] plus 0.1 g/liter CaCl and 0.1 g/liter MgCl × 6 HO). Otherwise, cells were fixed in 3% (wt/vol) PFA for 15 min at RT. After fixation, cells were processed for immunofluorescence essentially as described previously (). For cell surface staining, the cultures were washed once with PBS and incubated with antibodies applied to apical and basolateral sides for 7.5 min at RT. Cultures were washed three times with ice-cold PBS and fixed in 3% PFA for 15 min at RT. Filters were then cut out and stained for immunofluorescence microscopy essentially as described. 2 d after seeding, filter-grown MDCK cells were infected with defective adenoviruses encoding double myc-tagged syntaxin 3 or double myc-tagged syntaxin 4 () as described. 1 d after infection, filters were washed three times in ice-cold PBS and cut out, and cells were lysed with 1.25 ml solubilization buffer (50 mM Tris-HCl, pH 8.0, 150 mM NaCl, 1% Triton X-100, and 1× protease inhibitor cocktail [Roche Pharmaceuticals]), passed four times through a 22.5-gauge needle and a 1-ml syringe, and incubated for 30 min on ice followed by a clarifying spin at 4°C in an centrifuge (13,200 rpm for 15 min; Eppendorf). The resulting supernatant was incubated with anti-GFP antibodies bound to protein A beads and rotated end-over-end overnight at 4°C. Immunoprecipitates were washed three times in lysis buffer and eluted with 100 mM glycine, pH 2.7, and 0.5% Triton X-100. The eluate was neutralized with Tris base and supplemented with SDS sample buffer, boiled for 5 min, and run on SDS-PAGE gels followed by blotting onto nitrocellulose membranes and Western blot analysis. MDCK cells expressing GFP-cb were infected with defective adenoviruses encoding μ1A-HA or μ1B-myc as described. Infected cells were incubated at 37°C for 36 h and subsequently fixed for 1 h with 4% PFA (Electron Microscopy Sciences) and 0.25 M Hepes, pH 7.4, at RT. Cells were then left overnight in 8% PFA and 0.25 M Hepes, pH 7.4. Preparation for immunochemistry was as described previously () using rabbit anti-myc (A-14) and anti-GFP (Invitrogen) antibodies, respectively, followed by 10 and 15 nm protein A–gold, respectively (Cell Microscopy Center, Utrecht University, Netherlands). Sections were observed in an electron microscope (Tecnai 12; FEI Company), and images were captured using a charge-coupled device camera (Morada; Olympus) and saved as TIF files. Images were assembled using Photoshop without any alteration other than contrast and brightness. 3 d after seeding, MDCK cells were infected at 14 pfu/cell as described. 16 h after infection, cells were pulse-labeled, chased, subjected to either apical or basolateral biotinylation, and harvested essentially as previously described (). Postlysis supernatant was spun at 100,000 for 30 min (TLA 55 rotor [Beckman Coulter]; MaxE tabletop ultracentrifuge). The supernatant was then incubated with C7 antibodies coupled to protein G beads, and resulting immunoprecipitates were used for subsequent precipitation of biotinylated material essentially as previously described (). Samples were subjected to SDS-PAGE and autoradiography. Radioactive counts were detected using a Phosphorimager (model FLA-5100; Fujifilm). The resultant signals were analyzed using Multi Gauge Version 3 software (Fujifilm). Surface proteins were normalized against the total protein. The data are expressed as a percentage of LDLR-CT27 protein reaching each surface relative to the total amount of surface LDLR-CT27 at each time point. Fig. S1 shows the expression of myc-tagged cellubrevin at the basolateral membrane of MDCK cells. Fig. S2 shows the intracellular localization of RFP-tagged TeNT-resistant cellubrevin or synaptobrevin 2. Fig. S3 shows the sorting of selective cargos to the basolateral or apical plasma membrane in LLC-PK1 cell lines. Fig. S4 shows the yeast two-hybrid interactions of basolateral sorting signals with adaptor complex μ-chains. Online supplemental material is available at .

Adhesion and migration of epithelial sheets are critical for wound healing, organ integrity, and morphogenetic movements during development. Cellular circuits that orchestrate these processes require coordination of integrin function with multiple signaling pathways. Integrins are transmembrane heterodimeric receptors for ECM that convey information bi-directionally between the extracellular environment and intracellular signaling machinery (). Engagement of integrins leads to the concentration of tyrosine kinases and their substrates at focal adhesions, a type of adherens junction that acts as a signaling nexus, a tethering site for actin filaments, and a region for generation of traction force during cell migration. During embryogenesis, lateral epidermal sheets migrate to close a hole in the dorsal epidermis in the process of dorsal closure (DC). DC is executed through cytoskeletal rearrangements and cell shape changes with no accompanying cell division (). Because many proteins involved in DC also function in epithelial migration in other organisms, DC has emerged as an ideal model system to dissect the mechanisms driving migration and fusion of epithelial sheets. During DC, structures related to focal adhesions are assembled at the leading edge (LE) of advancing lateral epithelial cells and integrins are concentrated at these sites (; ). Moreover, genetic analysis has revealed that integrins are essential for normal DC (; ). Based on the established roles of integrins in mammalian systems, these adhesion receptors could influence DC by supporting cell–substratum interactions, modulating signaling pathways, or both. One signaling cascade that is essential for successful execution of DC results in activation of c-Jun amino-terminal kinase (JNK). Fine tuning of JNK output is critical, as both attenuation and hyper-activation of JNK signaling result in a failure of DC. The formation of focal adhesion complexes at the apical borders of the LE cells during DC depends on proper modulation of the JNK cascade (; ), highlighting the potential importance of crosstalk between integrin and JNK signaling. Several cytoplasmic proteins that colocalize with integrins are known to be essential mediators of integrin function in mammalian systems (). One of these, the LIM protein PINCH, interacts with the integrin-linked kinase (ILK) and is critical for adhesion and spreading of mammalian cells (; ). To elucidate the in vivo role and mechanism of action of PINCH, we undertook a genetic analysis of PINCH function. PINCH is encoded by the () locus. Mosaic analysis has revealed a critical role for PINCH in integrin-dependent epithelial cell adhesion in the adult wing (). Homozygous zygotic mutants die late in embryogenesis, exhibiting deficits in both muscle cell adhesion and actin–membrane anchorage (). Involvement of PINCH in both integrin-mediated adhesion and actin–membrane linkages makes it an attractive candidate for coordination of integrin and JNK functions during DC. To determine if PINCH could contribute to DC, we examined its localization in stage 14 embryos. PINCH and β-PS integrin are colocalized in both the LE and the amnioserosa ( A), consistent with PINCH's established role as an integrin effector. The amnioserosa is an extraembryonic tissue present on the dorsal surface of the embryo. As it has been established that coordinated signaling between the amnioserosa and migrating epithelium is key to formation of LE focal complexes (), PINCH could exert an effect in the LE epithelium, the amnioserosa, or both tissues. homozygous mutant embryos rescued with a transgene under the control of the endogenous PINCH promoter display PINCH-GFP at the LE of the advancing epithelial sheets. Within the LE, PINCH is precisely localized to areas of active phosphotyrosine signaling at triangular nodes corresponding to apical adherens junctions ( B, inset). Zygotic mutants proceed normally through DC with complete lethality arising at the embryo-to-larval transition. When maternal PINCH contribution is eliminated, only 12% of cuticles have wild-type morphology. Dorsal puckers and dorsal holes ( C) characteristic of aberrant DC are observed at a 36 and 23% frequency, respectively ( = 180), indicating that maternally inherited PINCH is a key contributor to the process of DC. Moreover, in the absence of maternal PINCH, we also observe epithelial defects in ventral patterning and head involution (Fig. S1, available at ), indicating that PINCH may have additional functions in the developing embryo. Cuticles from embryos lacking both maternal and zygotic PINCH have the same array of phenotypes. PINCH is composed of five LIM domains, each of which could serve as a protein-binding interface. The SH2-SH3 adaptor protein, Nck2, has been reported to interact with mammalian PINCH (). This association is intriguing because the homologue of Nck2, Dreadlocks, interacts directly with Misshapen (Msn), a MAP4K in the JNK signaling cascade (). As with other components of the JNK pathway, null mutations in result in embryonic lethality due to failure of DC. Although we were unable to detect direct binding of PINCH to Dreadlocks in , we uncovered a link between PINCH's role in DC and the JNK cascade by testing for genetic interaction between and . , into the homozygous null background allows partial restoration of DC (P = 0.003; A), suggesting that PINCH functions as a negative regulator of JNK signaling. Puckered (Puc) is a JNK phosphatase whose expression is up-regulated in response to JNK activation, setting up a negative feedback loop (). During DC, JNK-regulated expression of a Puc-LacZ fusion reporter is restricted to the LE cells ( B). In embryos lacking maternal PINCH, expression of the Puc-LacZ fusion protein is disorganized and present in an expanded number of cells, including those beyond the LE border ( B). This phenotype is similar to Puc-LacZ expression observed in loss-of-function mutants and further supports a role for PINCH in the negative regulation of the JNK cascade. Thorax closure is a post-embryonic developmental process with features common to DC, including migration of epithelial sheets and a dependence on JNK signaling. Within the wing disc, cells of the stalk region are functionally similar to LE cells during DC (; ). These cells comprise the eventual fusion site for adjacent imaginal discs and are active in JNK signaling (). Spatially restricted JNK signaling in the stalk of wing disc can be visualized via a Puc-LacZ reporter, and PINCH expression overlaps with Puc-LacZ in this area of active JNK signaling ( C). Therefore, as in DC, PINCH is properly positioned to act as a regulator of the JNK cascade. are semi-viable and a large proportion of the eclosing adults have thorax closure defects ( D). These observations underscore the similarities between thorax closure and DC. heterozygous background, a greater percentage of homozygotes are able to eclose (P < 0.0001; D), supporting the hypothesis that PINCH is a negative regulator of the JNK pathway in both dorsal and thorax closure. We purified PINCH in complex with its binding partners using tandem affinity purification (TAP)–tagged PINCH (TAP-PINCH; ). homozygous mutant embryos rescued with a transgene driven by the endogenous promoter to wild-type levels ( A) afford material for purification of soluble, cytoplasmic TAP-PINCH complexes in the absence of endogenous PINCH protein. Three partners that copurified stoichiometrically with TAP-PINCH from embryos, as well as in complex with TAP-PINCH from cultured S2R cells ( B), were identified by mass spectrometric analysis. Consistent with what is observed in mammalian cells () and our previous findings in (), ILK copurified with PINCH. The homologue of the parvin/actopaxin family of proteins, CG32528, is also present in PINCH protein complexes. Parvin is known to bind ILK and actin in mammalian systems (), but the isolated Parvin/ILK/PINCH complexes are the first to be described in . Additionally, a novel 31-kD protein was identified as CG9031. The CG9031 protein is 55% identical and 74% similar to human RSU-1, a leucine-rich repeat containing protein first identified as a suppressor of cell transformation by -Ras () and subsequently implicated in regulation of MAP kinase signaling, specifically the JNK and ERK cascades, when overexpressed in cultured cells (). Despite its potent ability to act as a tumor suppressor, little is known about the mechanism of action of RSU-1. Its partnership with the PINCH protein allows placement of RSU-1 in a molecular pathway that is linked to integrins. To assess the specificity and nature of the interaction between PINCH and RSU-1, domain-mapping studies were performed in cell culture and in yeast two-hybrid assays. RSU-1 copurifies with full-length His-tagged PINCH, but not with a truncated His-tagged PINCH containing only LIM1–3 ( C), confirming the specificity of the interaction and suggesting LIM4 and/or 5 is the site of binding. ILK, which binds LIM1 of PINCH, copurifies with both full-length and the truncated LIM1–3 version of His-tagged PINCH ( C), serving as a positive control. Both PINCH and ILK are copurified with His-tagged RSU-1 ( C). Moreover, endogenous PINCH and RSU-1 can be coimmunoprecipitated (unpublished data). The site of RSU-1 binding to PINCH was further mapped using yeast two-hybrid analysis. Only cells expressing LIM5 bait/RSU-1 prey activated all three reporters ( D), indicating LIM5 is the site of RSU-1 binding. Consistent with the view that they interact in vivo, PINCH:GFP and RSU-1 are prominently colocalized at integrin-rich muscle attachment sites in embryos ( E). RSU-1, which displays seven leucine-rich repeats with high sequence similarity to small GTPase regulators, is encoded by the locus. We have characterized a P-element insertion allele that disrupts the RSU-1 coding sequence ( A). Flies homozygous for this mutation within are viable and fertile (unpublished data), and lack RSU-1 protein as indicated by Western analysis with multiple anti-RSU-1 antibodies ( B). PINCH and RSU-1 are both expressed in larval wing discs ( C) and similar to wing clones (), the mutation within produces flies with wing blisters ( D) at 60% penetrance. These data are consistent with PINCH and RSU-1 acting in concert to support integrin-dependent adhesion. We have named the CG9031 gene () after the son of Daedalus who had unstable wings. Although elimination of RSU-1 function does not result in dorsal or thorax closure defects (unpublished data), we evaluated the role of RSU-1 in these processes by testing for genetic interactions between and . mutant embryos (P < 0.001; A). Absence of RSU-1 also increases eclosure rates (P < 0.0001) of hypomorphs ( B) and completely suppresses the thorax defects present in animals (unpublished data), suggesting that like PINCH, RSU-1 can function as a negative regulator of JNK signaling. To confirm that the suppression of DC defects by mutation is mediated by the JNK signaling cascade, we eliminated RSU-1 in () embryos that lack zygotic JNK, the terminal kinase in this cascade. mutants (P < 0.001; C), confirming that loss-of-function mutations affect DC by influencing the JNK cascade. Moreover, wing discs isolated from mutants display a 30% increase in active phospho-JNK relative to wild type ( D), providing direct biochemical confirmation that RSU-1 influences JNK activation state in vivo. Although we have not detected any localized accumulation of RSU-1 during DC (unpublished data), RSU-1 is readily detected by Western analysis in stage 13 embryos that are undergoing DC ( E, lane 1). Thus, the temporal pattern of RSU-1 expression is consistent with genetic results that highlight its role in regulation of JNK-dependent morphogenesis. Analysis of PINCH and RSU-1 levels in wild-type versus or mutant embryos provided insight into the physiological significance of their association. In embryos mutant for both maternal and zygotic RSU-1 is dramatically reduced relative to wild-type levels ( E, lane 2). Likewise, in embryos, PINCH levels are also decreased ( E, lane 3). These observations suggest that PINCH and RSU-1 are reciprocally dependent on each other for maximal expression and/or stability. The mechanism for coordinate regulation of RSU-1 and PINCH remains to be determined. Because the phenotypes associated with loss of RSU-1 represent a subset of phenotypes, the processes disturbed in mutants may be exquisitely sensitive to PINCH levels. Alternately, RSU-1 may have functions that are independent of its role in PINCH stabilization. Our data are consistent with a model ( F) in which PINCH could modulate JNK signaling in two distinct ways. First, PINCH is present at areas where JNK is active and antagonizes JNK signaling. This behavior is reminiscent of Puc, a phosphatase regulator of the JNK cascade that establishes a negative feedback loop (). PINCH has no intrinsic catalytic activity, but might recruit proteins that could alter the availability or activity of JNK signaling components. Like Puc, PINCH expression is up-regulated in response to constitutive JNK signaling (). Availability of RSU-1 at these sites of active JNK signaling could independently regulate JNK signaling or modulate the effects of PINCH on JNK through regulation of PINCH stability. Second, PINCH and RSU-1 are required for integrin-dependent adhesion. PINCH has previously been shown to link integrins to the actin cytoskeleton via ILK and Parvin (; ; ), and these connections could influence both integrin-dependent adhesion and signaling. Integrin signaling, through a variety of tyrosine kinases and Rac, stimulates the JNK cascade (); therefore, PINCH may also exert an influence on JNK signaling via integrin. Our findings illustrate that the cellular concentration of PINCH affects the level of RSU-1 and vice versa. Thus, modulation of the ratio of RSU-1 to PINCH could provide a mechanism to regulate JNK signaling during DC and thorax closure in . We hypothesize that PINCH/RSU-1 complexes fine-tune and integrate the JNK and integrin signaling cascades required during morphogenesis, highlighting the potential role of integrin-associated apical junctional complexes as signal coordination points for epithelial morphogenesis. PINCH:GFP and PINCH:TAP transgenics used standard methods. dorsal open rescue, /TM3, flies were crossed to same and to stck/TM3, or /TM3, flies were crossed to same. dorsal open rescue, /CyO or ; /CyO flies were crossed to same. , /TM3, flies were crossed to same and to stck/TM3, or ; /TM3, flies were crossed to same. Embryos lacking maternal PINCH were generated using the FLP-FRT system (). Rabbit polyclonal antisera were generated (Harlan Bioproducts) using antigens of the first and last 15 amino acids of RSU-1. Antibodies used were rabbit anti-ACTIVE-JNK (Promega), anti-JNK (), and anti-PINCH; mouse anti-ILK (BD Scientific), anti-phosphotyrosine 4G10 (Upstate Biotechnology), anti-penta His (QIAGEN), anti-β-PS integrin (CF6G11), and anti-LacZ (40-1a) (DSHB, University of Iowa, Iowa City, IA). embryos and third instar larval wing discs were prepared as described previously (). Confocal images were acquired at RT on an confocal microscope (model FV300; Olympus), using 20× 0.7 NA dry and 60× 1.4 NA oil immersion objectives, and were assembled using ImageJ and Adobe Photoshop 7.0. Third instar larval wing discs ( = 10–25) were homogenized and quantitatively immunoblotted for JNK and activated JNK as described previously (). Plasmids for expression of tagged PINCH or RSU-1 were constructed by standard molecular biology techniques. See Fig. S2 for details (available at ). 10 g pCaspin-TAP–rescued embryos or 5 × 10 S2R cells stably transfected with pMT/TAP-PINCH were washed and homogenized in lysis buffer (TBS, pH 7.9, plus 0.1% Triton X-100 and protease inhibitors) and 125,000 soluble portion was used as described below. S2R cells were grown as recommended (Invitrogen) and lysed, and 30,000 supernatant was batch-bound to 100 μl IgG Sepharose (Amersham Biosciences) prepared per manufacturer's recommendations and equilibrated in lysis buffer. After washing extensively with lysis buffer, proteins were eluted with a step gradient of 100 mM glycine from pH 5.0–2.75. Ni-NTA agarose (QIAGEN) purifications of His-tagged proteins used standard techniques. TAP-PINCH complexes were TCA precipitated and resuspended in Tris buffer, 8M urea, pH 8.6, reduced, and alkylated. Complexes were endoproteinase Lys-C digested (4 h), diluted to 2M urea, and digested with trypsin overnight (). Peptide mixtures were loaded onto a triphasic LC/LC column and analyzed as described previously (). Tandem mass spectra were analyzed using SEQUEST and the sequence database with threshold values of 1.8 (+1), 2.8 (+2), and 3.5 (+3) (). Identities of specific bands were confirmed by sequence analysis. PINCH baits depicted in C were constructed in pGBD-C1 (). The full-length RSU-1 prey is cloned in pACT2. The yeast host strain, PJ69-2a, was transformed with bait and prey, and then reporter activities were assayed as described previously (). Fig. S1 shows pleitropic phenotypes of maternally deficient cuticles. Details of plasmid construction are provided in Fig. S2. Online supplemental material available at .

The rapid increase in the number of sequenced genomes has augmented the need for efficient methods to define the regulatory logic of genes. Meeting this need requires the identification of each regulatory element within a promoter, including repressor-binding sites and upstream activation sequences (UASs), and the identification of the proteins that bind to these elements. These regulatory elements are typically short DNA sequences (4–10 bp) that are abundant in the genome but that only serve regulatory roles in a small subset of these occurrences (). The most common method researchers have used to identify new UASs in yeast places the promoter of interest upstream of a reporter gene, such as (). Through a series of 5′-deletions, the promoter is truncated and the effects of each truncation on reporter gene expression is measured in the presence of a known inducer of that promoter. When induction is lost, it is assumed that a UAS has been deleted. Subsequently, higher resolution truncations of the putative UAS-containing region are used to define a minimal UAS. While this method has been successfully used to dissect dozens of promoters in yeast, it can be labor intensive and can miss UASs that are part of more complex promoter structures containing multiple UASs or repressors. Recently, two new bioinformatic approaches have been developed to identify UASs: phylogenetic footprinting (,) and clustering of co-regulated genes (,). These methods are used to identify motifs that are conserved between related species or highly represented in the promoters of genes that are co-expressed under a given set of conditions. A limitation of these computational approaches is that regulatory regions in yeast are quite small compared with the size of a promoter (typically 600–1000 bp), leading to a low signal-to-noise ratio. As a result, these methods must apply strict selection criteria to yield only the most highly conserved sequences, resulting in a significant number of false negatives due to the degeneracy that occurs in many transcription-factor binding sites. Moreover, the uncertainty surrounding putative UASs identified using current computational methods requires that each predicted UAS be experimentally validated. This validation can be prone to a significant failure rate, since promoter regions can be highly conserved for reasons other than transcription-factor binding, including maintenance of chromatin structure and determination of long-range structure in the nucleus by specifying spatial organization of chromosomes (). The creation of new, efficient experimental methods that not only generate but also simultaneously test hypotheses using a functional transcriptional readout would significantly enhance our ability to define the regulatory logic of promoters of interest. Here we present a highly efficient method for the functional dissection of yeast promoters. This method involves the diversification of promoters using nonhomologous random recombination (NRR) (), followed by the identification of functional promoter variants using an selection or high-throughput screen. We validated this method using the unfolded protein response (UPR), a transcriptional program activated in response to unfolded protein in the endoplasmic reticulum (ER). Using this approach, we rapidly identified known UASs in promoters that are upregulated during the UPR. In addition, this method also identified novel regions in these promoters that are sufficient for a transcriptional response in the presence of a UPR inducer. Our approach may be applicable to any yeast promoter whose activation or repression can be subjected to screening or selection. Media consisted of nitrogen base (Sigma), 2% dextrose and synthetic dropout supplements lacking uracil or uracil and tryptophan (Open Biosystems). All synthetic media was supplemented with 50 μg/ml -inositol (Sigma). X-Gal indicator plates were prepared as previously described (), except that X-Gal was used at a final concentration of 80 μg/ml and adenine was added to a final concentration of 100 μg/ml. yeast genomic DNA was purchased from Promega. Tunicamycin (Sigma) was stored as a 1000 × stock in dimethylformamide and used at a final concentration of 1 μg/ml. strain DH10B was purchased from Invitrogen. Yeast strains W303a () and CP263 (Δ otherwise W303a) were kindly provided by Professor Peter Walter (University of California at San Francisco). Full open reading frame deletion of Crz1, replaced by the gene, was generated using a PCR-based deletion strategy in strain W303a (,). Gene disruption was confirmed by PCR and automated DNA sequencing. pJC104 is a two micron plasmid with a marker that contains a crippled promoter upstream of and four copies of UPRE1 serving as a UAS (). This plasmid was a generous gift from Professor Walter (). A multiple cloning site was inserted in the large XhoI/BglII fragment of pJC104 using the phosphorylated primers 5′-GATCTGGTCACCTAGGTACCGCGGCCGGTAGCCCGGGTCGAC and 5′-TCGAGTCGACCCGGGCTACCGGCCGCGGTACCTAGGTGACCA to yield pProm. The following phosphorylated primers were used to amplify the promoters of and , respectively: K1 (5′-GTGGGAGTCAATCAAATCCC) and K2 (5′-GGTATGTTTGATACGCTTTTTCC); S1 (5′-CATCCAGGATCAAGTATATACC) and S2 (5′-TCTAAGTTTGCGTTCTTGGAAG); H1 (5′-AGAGCCACTATCATCGGC-GAC) and H2 (5′-AGTGGCGGTTGTTGTCGTAGG). Primers HP1 (5′-GCATAGCAGCTCGGCGCCGAGCTGCTATGC), HP2 (5′-GCATAGCAGCTCGGCGCCGAGCTGCTATGC) and HP3 (5′-GCATAGCAGCTCGGCGCCGAGCTGCTATGC) contain BglII, SalI and BstEII restriction sites (underlined), respectively, for cloning into pProm. These primers also contain AscI and NcoI restriction sites (italicized) for removal of hairpins (see subsequently). Primers F1 (5′-CGCCAGCTGCTATGCAGATCT), F2 (5′-CGCCAGCTGCTATGCGTCGAC) and F3 (5′-CGCCAGCTGCTATGCGGTCACC) are used for the final PCR amplifications. Genomic DNA fragments from the and promoters, −1 to −1000 relative to the start of translation, were amplified by PCR using the primers K1/K2, S1/S2 and H1/H2, respectively. NRR was performed on the resulting PCR product as previously described () using primers HP1 and HP2 for the and libraries. HP2 and HP3 were used for the library. Fragments with a desired size range (see text) were purified by agarose gel electrophoresis after the hairpin ligation step. Recombined promoter fragments for the and libraries were digested with NcoI and AscI, then amplified with primers F1 and F2 and cloned into the large BglII/SalI fragment of pProm. Recombined fragments from the library were also digested with NcoI and AscI, then amplified using F2 and F3, and cloned into the large BstEII/SalI fragment of pProm. All libraries were amplified in the strain DH10B. A promoter library (∼1 μg) was transformed into W303a using a standard lithium acetate protocol and selected on plates lacking uracil at 30°C. After 3 days of growth, the yeast colonies were replica plated onto X-Gal plates supplemented with tunicamycin and incubated at 30°C for 24–72 h. The bluest colonies were selected and grown individually in liquid media lacking uracil. Liquid cultures were then spotted onto X-Gal plates lacking uracil with and without tunicamycin to identify those cultures that specifically turned blue only in the presence of tunicamycin. The cells from an aliquot of each liquid culture were lysed using glass beads and a region of DNA encoding the promoter fragment was amplified using primers Seq1 (5′-GGAGACGCATTGGGTCAACAG) and Seq2 (5′-GTGTTTGCGTGTCTATAGAAG). These PCR products were then sequenced to identify promoter fragments. DNA sequences of the promoters from the blue colonies were then aligned to identify consensus regions. Phosphorylated, complementary oligonucleotides containing one, two or three copies of each consensus regions from the X-Gal screening were annealed and then individually cloned into the large BglII/SalI fragment of pProm for further analysis. For the promoter, the consensus regions I–III are as follows: 5′-AATGTACACGTATC, 5′-TTTGAACACGTCAACAAC and 5′-TAGCCAAACGGACA, respectively. Extended consensus region III was 5′-TAGCCAAACGGACAGCTGTCCTCA. consensus region IV was contained in pJC104. The consensus region is 5′-GAAAAGGCCACGTAG. The consensus regions I and II are as follows: 5′-GGCAAAGTGGCTCAGCAT and 5′-TGGTTTTGAACACCTTGTTCTCTTTTGT, respectively. Plasmids containing the consensus regions were then transformed in W303a, CP263 or the Crz1-deleteion yeast strains and assayed for β-galactosidase activity with or without tunicamycin in triplicate using -nitrophenyl-β-galactopyranoside as previously described (). NRR () is a method to diversify nucleic acids that randomly rearranges any sequence using a fragment size defined by the researcher. NRR has been used to study protein topological requirements (), sRNA mechanisms () and mRNA transport signal sequences [(); Liu,J.M. and Liu,D.R., unpublished data). The first step in NRR involves the random digestion of a starting nucleic acid pool using DNase I (). By modulating the digestion time, the average size of the digested fragments can be controlled. After blunting the pieces with T4 DNA polymerase, the DNA fragments are reassembled in the presence of DNA hairpins, allowing for the deletion, insertion and rearrangement of fragments. The average number of crossovers per recombinant is controlled by varying the relative molar ratio of hairpins to fragments. To identify UASs within a promoter, we envisioned that an NRR-diversified promoter of interest could be cloned upstream of a crippled core promoter that drives the expression of the gene () (). The resulting library can be rapidly screened for transcriptional activation only in the presence of a signal known to induce transcription from the promoter of interest (). Because the products of NRR are highly diversified in the size, order, spacing, redundancy and orientation of subsequences, the comparison of DNA sequences among promoter variants that maintain their ability to be induced should rapidly identify essential UASs. This approach requires virtually no knowledge of the promoter of interest other than a means of induction, and therefore could represent a general approach to the discovery of UASs. To validate the proposed method to identify UASs in promoters of interest, NRR coupled with high-throughput screening was applied to three promoters of genes involved in the UPR: and . Three UASs that specifically respond to unfolded protein in the ER have been previously identified as UPR elements 1, 2 and 3 (UPRE1, UPRE2 and UPRE3) (,,). The 1000-base promoter contains one known UAS, UPRE1 (7 bp), that is responsive to the presence of unfolded protein in the ER (). We subjected the promoter of to NRR and screening to determine if UPRE1 could be quickly and accurately identified. The NRR-diversified promoter library was cloned upstream of the crippled promoter driving the expression of for blue/white screening in the presence of tunicamycin. The promoter library contained DNA fragments that were an average of 31 bp in length. The library was transformed into yeast and replica plated onto X-Gal plates supplemented with tunicamycin to induce the UPR. The 60 bluest colonies were isolated, then subjected to a secondary screen on X-Gal plates both with and without tunicamycin. The clones that showed robust transcriptional activation only in the presence of tunicamycin (15 out of 60 colonies) were classified as strong hits. All 60 clones were sequenced, and the promoter fragments were aligned to reveal any consensus regions among the selected promoters (). Promoter fragments were grouped according to blue/white phenotype, with the strong hits at the top of (gray area), and then by sequence to show the consensus regions. This entire process, from initial amplification of the promoter to the sequence alignment, required less than three weeks. Four consensus regions emerged from this analysis, including one consensus region present in most of the strong clones. Gratifyingly, this consensus region (region IV in ) contains the previously described UPRE1. Sequence alignment of the strongest clones yields a consensus region of 15 bp, which contains the 7 bp core of UPRE1 previously shown to be essential for ER stress-mediated activation of the promoter (). Since UPRE1 was the only UPRE previously described for the promoter, we were surprised that three additional consensus regions emerged from the above analysis (). Consensus region I contains UPRE2, a recently discovered UPRE, CACGTA, that is a putative binding site for Gcn4. Region II contains a very similar sequence that differs from UPRE2 by a single nucleotide, CACGT. This sequence has not been specifically characterized as a Gcn4-binding site. Region III does not contain any previously reported UPREs though shares some similarity with UPRE1 (). Of the four consensus regions, region III was the least abundant, occurring in only seven out of 60 clones (). All seven clones contained a common sequence shown in red in , but six of these clones also had a longer consensus region shown in blue. Since comparatively few clones defined the boundaries of this consensus region, both the minimal and extended consensus regions were used for further analysis. To test the functional significance of these consensus regions, fragments of the promoter containing each consensus sequence, including both region III and the extended region III (), were cloned upstream of the construct and analyzed using quantitative β-galactosidase assays (). Regions I, II and IV each showed significant activity only in the presence of tunicamycin. The minimal consensus region III did not show any activity in the presence or absence of tunicamycin, but the extended region III exhibited robust transcriptional induction in the presence of tunicamycin. This result reveals the necessity of carefully choosing the endpoints of a consensus region when only a small number of clones define the region. Each of these consensus regions was also tested in a strain that lacks the gene for Ire1, the protein that initially senses unfolded protein in the ER (). All four regions that showed tunicamycin-dependent activity in a wild-type strain had no activity in a Δ strain (). Ire1 is essential for the interaction of the transcription factors Hac1 and Gcn4 with UPRE1 and UPRE2 (). The Ire1 dependence of transcriptional activation mediated by consensus regions I, II, III (extended) and IV suggests a similar transcriptional pathway. Taken together, the above results demonstrate the ability of a coupled NRR/screening approach to identify both known and unknown upstream-activating regions within yeast promoters. Sil1, a nucleotide exchange factor for Kar2 (), is highly upregulated during the UPR (). The promoter of was analyzed because it has not been previously experimentally dissected, although it contains a sequencing matching UPRE2 (). It is currently not known if addition transcriptional elements are also present in the promoter. The promoter of was subjected to NRR and cloned into the construct. The resulting library had an average DNA fragment length of 27 bp. This NRR-diversified library was screened in the presence of tunicamycin and the 60 bluest colonies were isolated. Using the secondary screen, 12 of the 60 colonies showed robust transcription only in the presence of tunicamycin and these colonies were classified as strong hits (, top). When all 60 colonies were subjected to sequence analysis, only one region was repeatedly selected (). Alignment of clones showing a strong phenotype upon plating on media containing X-Gal in the presence and absence of tunicamycin revealed a 14-base consensus region containing UPRE2 (). No other consensus regions emerged among the sequences. When this consensus region was cloned into the construct, it was responsive to tunicamycin in a wild-type strain (). As expected for UPRE2, this consensus region did not function as a UAS in a Δ strain (). The regulatory logic of the promoter appears to be quite simple; the lack of any other consensus regions suggests that UPRE2 is the sole determinant of transcription from during the UPR. The 14-bp region containing UPRE2 was retained in 100% of our selected clones (), indicating that some part of this sequence which includes eight bases upstream of UPRE2 may be important for a strong transcriptional response. The promoter of , a key transcription factor involved in the UPR, was similarly analyzed by NRR and screening. A UPRE1-like sequence was recently identified in this promoter (), and Hac1 has been shown to bind to this region in an autoregulatory mechanism that is -dependent. A second -independent regulatory mechanism for the promoter has been observed during extensive ER stress, but the UAS responsible is unknown (). To elucidate the various mechanisms of transcriptional regulation, the promoter of was subjected to NRR and cloned into the construct. The resulting library had an average DNA fragment length of 33 bp. When this library was screened in a wild-type strain in the presence of tunicamycin, clones containing one of two consensus regions emerged (). The DNA sequences of clones in groups I and II were aligned to reveal the identities of the minimal consensus regions (). Consensus region II contains the previously described UPRE1-like sequence that has been shown to be sufficient for tunicamycin-induced transcription from the promoter (). The consensus sequence from region I is derived from only six clones, thus making it difficult to precisely identify the endpoints of the minimal region that contains a UAS (). These six clones fall into two categories, one containing a minimal consensus region (red in ) and one containing an extended consensus region (red + blue in ). The consensus region I does not contain a DNA sequence that matches any of the known UPREs, suggesting that a novel transcription factor is likely binding to this region. A search for transcription factor-binding sites in this consensus region revealed only one known motif (), a binding site for the transcription factor Crz1 () (). Crz1 is a calcineurin-dependent transcription factor responsible for the transcriptional response to changes of Ca concentration in the yeast calcium cell survival (CCS) pathway () and has not been previously implicated in the UPR. To test if the consensus regions from the wild-type analysis can serve as UPREs, each region was cloned into the construct and analyzed with quantitative β-galactosidase assays (). These two regions were also tested in the Δ strain to determine if the Hac1/Gcn4 pathway is necessary for transcription. Region II showed robust transcriptional activation in the presence of tunicamycin in the wild-type strain but not in the Δ strain, consistent with the previously described analysis of this region in the promoter (). In contrast, both region I and the extended region I showed similar tunicamycin-induced transcription in both the wild-type and Δ strains, indicating that a transcription factor-binding site likely resides in the minimal consensus region, which contains the putative Crz1-binding site ( and data not shown). Regions I and II were then tested in a Δ strain. While transcription from region II was unaffected by this deletion, all transcriptional induction was lost from region I, further implicating Crz1 as a factor responsible for transcriptional regulation from this region during the UPR. Our results collectively suggest that NRR is well suited to the rapid and efficient identification of -regulatory elements from yeast promoters. All three sets of experiments described here including promoter library construction, screening and sequence analysis collectively required only 3 weeks in total and can easily be performed in parallel for several promoters. The ability to use fragment sizes that approach the sizes of the typical UAS elements allows for the rapid identification of consensus regions that are responsible for transcriptional activity under the chosen experimental conditions. In the three examples shown here, the fragment sizes were 10–50 bp (note that manipulation of double-stranded DNA fragments smaller than ∼ 10 bp can be difficult due to spontaneous fragment denaturation). The ability of the NRR method to introduce crossovers between two or more functional regions in the libraries was particularly helpful in the analysis of the promoter. Although most of the clones containing consensus regions I and II had a weak phenotype upon secondary screening, these weak consensus regions can still yield functional UASs since they can act additively to yield a strong phenotype as observed in clones 25 and 29 (). For an optimum library, the average number of crossovers should match the number of regulatory elements that might act combinatorially in the promoter. While this number cannot be known , ChIP-chip experiments suggest that more than one-third of yeast promoters are bound by more than one transcriptional regulator, indicating that ideal libraries would have an average number of crossovers > 1 (). Our findings further suggest that, unlike previously described bioinformatic approaches (), the NRR method as applied to UPRE identification does not appear to suffer from a significant false-positive or false-negative rate. The method was able to find all four of the sequences that had previously been described as UASs: UPREs 1 and 2 from the promoter, UPRE2 from the promoter and the UPRE1-like sequence from the promoter. Each consensus region revealed by the NRR analysis was identified every time it occurred in these three promoters, suggesting that the NRR method can cover candidate promoter fragment space exhaustively, at least above a certain threshold of activity determined by the screening method used. In addition, every consensus region that was tested as a UAS, when the boundaries were carefully chosen, was responsive to tunicamycin, indicating an absence of false positives from this analysis. Current methods to experimentally dissect promoters usually take advantage of 5′-promoter truncation to identify regions that lead to a loss of inducible signal under the assay conditions. This method, however, can miss UASs that can function independently of and redundantly with downstream UASs, because truncation of a region containing an independent and redundant upstream UAS simply leads to a reduction in signal but not a loss of inducibility. For example, the deletion of a DNA sequence containing the consensus region I from the NRR analysis () was previously shown to cause a 2-fold loss in signal, but no loss of inducibility because the UPRE1-like sequence was still present (). As a result, this region was previously not implicated by researchers who were dissecting the promoter (). Since NRR can analyze DNA fragments independently and combinatorially, without any directional bias for UASs closer to the start of translation, this approach may be better suited than previous methods for the comprehensive characterization of yeast promoters. The identification of a consensus sequence as a UAS does not automatically imply that it is physiologically relevant in the context of the original promoter from which it was discovered. For example, an identified consensus sequence may normally be sequestered in heterochromatin, inaccessible to the transcription factors that bind to it in the artificial context of the assay. Also, a UAS could be controlled by negative regulatory elements that are lost in the analyses described above. However, while the UASs identified by the NRR method may not be relevant in the promoters from which they are derived, it is likely that they have a physiological role in other promoters. The evolution of DNA-binding specificity is under strong positive and negative selective pressures during evolution due to the importance of tight transcriptional regulation to cell survival (,). Therefore, interactions between a transcription factor and a DNA target that lead to a transcriptional readout are likely to be physiologically relevant in some cellular context. Identification of a new UAS allows researchers to identify a list of relevant promoters for which that UAS may be relevant. Each of these promoters can then be analyzed individually further to test for biological significance of the UAS. A new UAS was also discovered in region I of the promoter that responds to tunicamycin in an -independent manner. This UPRE contains a putative binding site for the transcriptional factor Crz1, a calcineurin-dependent transcription factor implicated in Ca signaling (), and its activity is dependent on Crz1. High salt, alkaline pH and cell wall damage lead to increased cytosolic levels Ca, activating calcineurin's phosphatase activity (). Calcineurin dephosphorylates Crz1, leading to its nuclear localization and transcription of its target genes. Previous studies have shown that unfolded protein in the ER leads to an influx of Ca at the plasma membrane in an - and -independent manner, leading to activation of Crz1 (). The CCS pathway is not required for the initial UPR response, but may be required for a prolonged response to tunicamycin treatment (). In addition to the canonical UPR, Walter and coworkers () have described a ‘super-UPR’ that requires at least two ER stress signals, such as inositol deprivation and tunicamycin treatment. Under these conditions, transcription is upregulated in an -independent manner, suggesting the possible existence of a new transcription factor in targeting the promoter. Our identification of a Crz1-dependent putative Crz1-binding site in the promoter and the previous description of Crz1 activation in response to ER stress suggest a possible connection between the CCS pathway and the super-UPR. Cross-talk between these two pathways has not been previously reported, and additional experimental work is needed to explore this hypothesis in depth. In conclusion, NRR coupled with screening is well suited to the dissection of the regulatory logic of yeast promoters, resulting in the identification of UASs with no significant false-positive or false-negative rate in the case of the UPR. From a single set of experiments that requires less than 3 weeks, fairly precise consensus regions can be rapidly defined that are close to the minimal requirements for a UAS that responds to the assay conditions. The speed and efficiency of this method to find UASs represent a significant advance over other current methods used to find UASs. Our initial efforts to validate the method described here have already yielded unexpected UASs that may play a role in the UPR or the super-UPR. Further work is needed to understand the biological role of these new UASs and to test the generality of this method in other promoters and organisms.

Secretory proteins exported via the general secretory (sec) pathway are synthesized as precursors with an N-terminal signal peptide. This signal peptide is removed by leader peptidase I upon export to the periplasm [for a review of export, see ()]. The signal peptide can be divided into three regions: a hydrophilic N-terminus often containing 1–3 positively charged residues, a hydrophobic core and a cleavage site for processing by the respective signal peptidase (). The role ascribed to the positively charged N-terminus of the signal peptide is to provide stable interactions with the negatively charged inner membrane phospholipids. This interaction is thought to be important for targeting the secretory protein translocase (). The positive charge could also aid interaction with the export machinery, signal recognition particle (SRP) () and SecA (). Studies by von Heijne (,) on the charge distribution of 39 and 32 prokaryotic signal sequences reported a net positive charge at the N-terminus of 1.7. Despite the abundance of positively charged residues at the N-terminus of signal peptides, studies have shown that removing the positive charges, such that the net charge at the N-terminus is 0, has differing effects on export. These include a reduced rate of export (,), lower rate of protein synthesis () and no discernable effect (). If there is a net-negative charge, then export is impaired, with increased levels of unprocessed precursor (). These results suggest that a net positive charge is not essential for export to occur, and raises the question of why the majority of secretory proteins have a positive charge at the N-terminus. With the availability of the complete genome sequence of (), and the ability to sort secreted proteins from non-secreted proteins (), distribution of charged residues was analysed in secretory proteins. The codon usage of the charged residues was further analysed and revealed preference for AAA at P2, implying that codon usage was a stronger selective pressure than a requirement for a positive charge. All cloning was carried out in DH5α (). Kanamycin was used at 50 μg/ml. Primers were made by Pro-Oligo (Sigma, Lismore). All PCR reactions were carried out using Phusion Taq (Finnzyme). Ligations using T4 DNA ligase and digests were performed according to manufacturer's instructions (New England Biolabs). DNA sequencing was done using Big Dye Terminator method (Griffith University DNA Sequencing Facility). The wild-type gene and signal sequence mutants were generated by splice-overlap PCR and cloned into the multicloning site of pGBS19 [pUC19 with kanr replacing , ()]. Briefly the pMalE::bla was made with splice-overlap PCR products generated from the 5′bla-MBP/3′MBPss and the 5′MBP-bla/bla_rev primer pairs, amplified with 5′bla-MBP and bla-rev primer pair. The pPhoA::bla construct was generated from the 5′phoA-bla/bla_rev and 5′bla-phoA/3′phoAss primer pairs, amplified with 5′bla-phoA and bla_rev primer pairs. Primers are listed in . The MBP signal sequence was amplified from the plasmid pMALp2e vector while the PhoA signal sequence was amplified from K-12 MG1655. For both constructs, the upstream primer (5′bla-MBP, 5′bla-phoA) incorporated 28 bp upstream of the start codon from pMALp2e, which contains a Shine Dalgarno sequence and an I site. The common downstream primer, bla_rev, incorporates a I site. The splice-overlap PCR product was digested with I and I and ligated into pGBS19. Prospective colonies were selected on LB-kanamycin plates and confirmed by sequencing using terminal primers. Amino acid changes were incorporated on oligonucleotides using pMalE::bla and pPhoA::bla as template for the PCR reaction. The PCR product was amplified using the mutagenic and bla-rev primer pair, digested with I and I and cloned into pGBS19. Prospective colonies were sequenced to check for the correct base changes. The MIC was determined using cultures grown overnight in LB-kanamycin and subcultured 1:100 the following morning. When cells were in mid-log phase, 10 cells were added to a microtitre plate for MIC determination, containing either freshly prepared ampicillin or kanamycin two-fold serially diluted. From the same culture 1 ml was spun down and resuspended in sample buffer for western analysis using a 1:10 000 dilution of a polyclonal β-lactamase antibody (Chemicon International, AB3738). xref ext-link #text ext-link #text To study the distribution of positively charged residues in the signal peptide, first all protein-encoding genes from the K-12 (strain MG1655) were categorized as either secretory or non-secretory based on signalP () analysis of the first 70 amino acids (see Materials and Methods). This analysis generated 466 secreted and 3023 non-secreted genes with 664 not able to be classified in either group. To ensure that the entire N-region was included in the analysis, the distribution of positively charged amino acids was analysed for the first 10 amino acids from the secretory group (). The net charge across this region was 1.84, compared to 0.4 for the non-secreted group. Approximately one-third of the net positive charge in the secretory group is due to lysine at P2 and P3. The extent of the bias for all charged amino acids was analysed using a χ2 test: ∑(observed−expected)2/(expected), with 19 degrees of freedom. The expected values for an amino acid were calculated using its frequency of occurrence at the respective position in all genes. This revealed that at P2, there was a massive bias for lysine ( = 1.72 × 10). At P3 both lysine and arginine were preferred ( = 2.71 × 10 × 10−9, = 0.001, respectively), although arginine was clearly less than lysine (). This appears to be consistent with the reported key role for positively charged residues at the N-terminus of the signal peptide, but raises the question of why there is significant bias only for lysine, rather than any positively charged residues, at P2. Other than a requirement for a positive charge, another possible factor constraining amino acid usage at P2 in secreted proteins could be the removal of the N-terminal methionine by methionyl amino-peptidase (MAP). All bacterial proteins start with a f-Met. The process of f-Met removal involves two steps, with the formyl group first removed by a deformylase (), followed by excision of the N-terminal methionine by MAP. In general, amino acids at P2 that are small promote N-terminal methionine removal by MAP, whereas large amino acids prevent N-terminal methionine removal (,). If the N-terminus of the signal peptide is required to help initiate export by associating with the cytoplasmic membrane, then presumably removal of the N-terminal methionine by MAP, whose binding pocket recognizes the first two residues, would inhibit this process, albeit temporarily. This could form a selection pressure to use amino acids at P2 in secreted proteins that prevent N-terminal methionine removal. Using a program recently described in Frottin (), all secretory and non-secretory proteins were analysed for probability of N-terminal methionine removal (see Materials and Methods). In the secretory group, 78.33% (365/466) are predicted to retain the N-terminal methionine, compared to 57.62% (1742/3023) of non-secretory proteins (). Assuming that there should be no difference between the two groups, this difference is significant ( < 10−5). However, the question still remains as to why only lysine, not arginine and histidine, is preferentially used in secreted proteins at P2. Another factor that could affect amino acid composition at the second position is selection for codons that promote or hinder translation initiation efficiencies. Two independent studies have shown that codon usage at the second amino acid position can affect translation initiation efficiencies (,), which is thought to be the rate limiting step in translation (,). The strength of translation initiation correlates to the nucleotide composition, with adenosine content promoting high translation initiation efficiencies. As this factor is independent of amino acid properties but rather codon dependent, analysis on the codon usage at the second amino acid position in all groups was done using a χ2 test with 60 degrees of freedom (not including stop codons as possibilities). The expected values were calculated using the total frequency of codons in all genes at P2. The data were limited to amino acid families that occurred more than 30 times, as numbers below this are too small to do a χ2 test on. This analysis showed that only the lysine codon AAA, which is used 29.68% of the time in the secreted group at P2, occurred at levels significantly greater than expected ( = 1.7871 × 10−7) (). All other codons, including the other lysine codon AAG, occurred at frequencies that were expected ( = 0.99 for AAG). Extending the same analysis to P3, where there was a bias for the amino acids lysine and arginine, revealed no codon bias for any codon at that position. The bias for AAA at P2, and not for AAG, indicates that the selective pressure is for codon usage and not the amino acid at this position in secretory proteins. In this study, the charge distribution of all sec dependent signal peptides was analysed. This revealed a massive bias for lysine at P2, which could be attributed to a bias for the lysine codon AAA. The preference for lysine codon AAA at P2 was experimentally shown not to be a requirement for a positive charge. Two studies have shown that AAA at P2 is the best initiator of translation (,). We propose that the selective pressure at P2 is for codons that promote faster translation initiation efficiencies. There does not appear to be any selective pressure at P2 for residues that do not promote N-terminal methionine removal. Sequences rich in adenine nucleotides immediately downstream of the start codon have been shown to enhance gene expression, presumably by enhancing translation initiation (). As lysine is encoded by AAA and AAG, it could be these factors that enhance choice of lysine at P2 and P3, not a requirement for a positive charge. This is supported by the ratio of lysine to arginine, which is 4.05:1 at P2 and drops at every position down to 0.41:1 by P9. If the requirement were simply for a positive charge, then one would not expect preferential usage of one basic amino acid over another. This enhances the idea that secretory proteins require higher translation initiation rates, but raises the question why that is necessary? Signal peptides also contain the highest levels of non-optimal codons seen anywhere in the genome (). Studies have shown that the insertion of consecutive non-optimal codons downstream of the start codon significantly lowers protein production compared to insertion of the same codons further downstream (), due to ribosomes dissociating from the transcript prematurely (). Hence for secretory proteins it is likely that ribosomes would disassociate prematurely due to the high levels of non-optimal codons in the signal sequence. Preferential use of AAA at P2, and the high use of adenine rich nucleotides at P3, which promotes rapid translation initiation, would help to counteract that effect, as subsequent ribosomes would quickly replace the previously dissociated ones. This would likely result in more ribosomes per transcript. Conversely factors that promote slow translation initiation would likely result in the spacing between ribosomes being greater, as one ribosome would be able to translate more codons before the next one commences translation. Biasing codons to ensure rapid translation initiation could help recycle chaperones required for export. For example the molecular chaperone SecB delivers the presecretory protein to SecA, while SRP delivers the presecretory to FtsY (). Both SecA and FtsY are inner membrane proteins. Once directed to these proteins, the chaperone is free to associate with a new nascent peptide. If ribosomes are close together on an mRNA transcript, due to increased translation initiation efficiencies, this could allow efficient binding to a new nascent peptide emerging from an upstream ribosome. This time factor may be important, as proteins must be in a loosely folded state to allow protein export (,). If it takes longer for the chaperone to find the next nascent peptide, this could mean the nascent peptide folds into a conformation incapable of export. This study, as well as others (), has found that a positive charge is not required for protein export. Studies have found that a net negative charge is deleterious for export, resulting in increased amounts of unprocessed precursor (). Other than the extreme bias for lysine at P2 and P3, which could be to promote high translation initiation frequencies, there was no bias for a positively charged residue at any other position. This raises the possibility that the overall selective force in the entire N-region of signal peptides is to avoid a net-negative charge. Supporting this is the fact that the negatively charged residues glutamic acid and aspartic acid occur on average 0.01 times from P2-P10 in secretory proteins, compared to 0.81 times for all other genes (data not shown). Given that most secretory proteins are exported to the membrane by SRP or SecB (), there is no requirement for the positive charge to interact with the membrane to initiate export. Once at the membrane, a net-negative charge would interfere with insertion into the membrane, due to the negatively charged phospholipids. Hence the observation that sec dependent signal peptides contain a positive charge at the N-region could be due to selection for lysine at P2 and P3 to promote high translation initiation efficiencies, and an overall selection against a net-negative charge. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

SUMOylation is a protein conjugation process which leads to covalent modification of many proteins involved in transcriptional regulation, protein transport, chromosome segregation and signal transduction in a constitutive or even stimulation-dependent manner (). The three mammalian SUMO isoforms (SUMO1–3) are expressed as precursor proteins and subsequently cleaved at their C-terminus by one of five SUMO-specific proteases (SENP1–3, SENP5 and SENP6) (). A complex consisting of SAE1 and SAE2 (E1) () activates the maturated SUMO and transfers it to the conjugating enzyme Ubc9 (E2) (,). Its interaction with a substrate protein then induces the transfer of the SUMO moiety onto the substrate protein and the formation of the isopeptide bond (). Most SUMOylation processes also seem to be assisted by SUMO ligases, which beside their different functions act as adapters that bind the conjugating enzyme Ubc9 and the substrate protein. Such SUMO ligases are the members of the PIAS (protein inhibitor of activated STAT; PIAS1, PIAS3, PIASxα, PIASxβ and PIASy) protein family (), the Polycomb group protein Pc2 () and the RanBP2 (,), a protein of the nuclear pore. The amount of SUMOylated proteins in the cell is further regulated by the five SUMO-specific proteases, which display SUMO deconjugating activity as well (). The number of identified SUMOylation substrates is steadily increasing especially through proteomic studies () but only for about 60 SUMOylation substrates the function of the SUMOylation is characterized in detail (). Recently, a Ubc9 fusion-directed SUMOylation (UFDS) system was developed that strongly increases the degree of SUMOylation of a specific substrate protein fused to Ubc9 . UFDS is efficient, selective for the SUMOylation sites and independent of SUMO ligases (). This method should be well suited to screen for new SUMOylation substrates and, therefore, we applied UFDS for the identification of new SUMOylation substrates leading to 14 potential new substrate candidates. Verification of eight of these new substrates by UFDS independent methods revealed that the UFDS system is capable of identifying constitutive and induced SUMOylations. The destination vector (pCU-B) for the fusion of open reading frames to the N-terminus of Ubc9 was made by inserting the Gateway RfB recombination cassette (Invitrogen) into the filled-in EcoRI site of the pcDNA3-MCS-Ubc9. The destination vector for the fusion of open reading frames to the C-terminus of Ubc9 was made by amplifying the Ubc9 cDNA using the primers se-5′-CCTCGGATCCGTTATGTCGGGGATCGCCCTCAG-3′ and ase-5′-CCTCGAATTCTGAGGGGGCAAACTTCTTCG-3′ and cloning the PCR product into the BamHI/EcoRI sites of pcDNA3. Then, the Gateway-RfC.1 recombination cassette (Invitrogen) was cloned in the EcoRV site of the pcDNA3-Ubc9-MCS to obtain the destination vector pNU-C.1. The Ubc9-ORF/ORF-Ubc9 fusion protein expression vectors were obtained by recombination of the above described destination vectors with the ORF harboring entry plasmids (from the RZPD Deutsches Ressourcenzentrum für Genomforschung) with the Gateway recombination system (Invitrogen). HEK293 cells were cultured in Dulbecco's modified Eagle's medium with high glucose, complemented with 10% fetal calf serum, 2 mM -glutamine, 100 U/ml penicillin and 100 µg/ml streptomycin. Antibodies against the following proteins or peptides were used: Ubc9 (H81, Santa Cruz) and GFP (B-2, Santa Cruz). Horseradish peroxidase-coupled secondary antibodies were from Santa Cruz. Transfection of 50–80% confluent HEK293 cells was performed in 12-well plates using the Polyfect transfection reagent (Qiagen) according to the instructions of the supplier. Transfectants were grown for 24–48 h, then lysed in 150 µl gel loading buffer (80 mM Tris, pH 6.8, 2% SDS, 5% ß-ME, 0.01% bromphenol blue) and incubated for 10 min at 95°C. For western blot analysis the proteins were separated by SDS-PAGE, blotted on a PVDF membrane and developed with a specific primary antibody, an HRP conjugated secondary antibody, the ECL+ (Amersham) and the LAS-3000 imaging system (Fuji). The analysis of protein functions necessitates the identification and characterization of post-translational modifications that are involved in their regulation. Since SUMOylation of most proteins is hardly detectable , we used UFDS to analyze 46 potential nuclear proteins () for their SUMOylation. For an efficient generation of the expression plasmids for Ubc9 fusion proteins, we constructed Ubc9 fusion destination vectors that allowed the fusion of protein coding sequences to Ubc9 using the ‘Gateway’ cloning-by-recombination system (Invitrogen). We fused the coding sequences of the proteins () to the N-terminus or the C-Terminus of Ubc9 (A and B), to determine if Ubc9 can SUMOylate the fused protein in both structural arrangements. All Ubc9 fusion protein expression vectors were transfected alone or together with an EGFP-SUMO1 expression vector into HEK293 cells. Proteins of the transfectants were analyzed by immunoblotting using a Ubc9 antibody. Of the 46 fusion proteins analyzed, 37 were found to be expressed to different degrees, while 9 fusion proteins could not be detected. Of the expressed fusion proteins, 14 were strongly and 23 were not or only weakly SUMOylated (c.f. C and ). Nine of the strongly SUMOylated proteins were fusions to the C-terminus, while the five others were fusions to the N-terminus of Ubc9. This demonstrates that Ubc9 is able to SUMOylate the fused protein in both structural arrangements. Most of the strongly SUMOylated proteins we identified are involved in transcription (CIAO1, CRSP9, EDF1, FOS, HMGN2, HSF2BP, PC4, TAF10 and ZNRD1). The others are proteins involved in signal transduction by phosphorylation, such as the Casein kinase 2, beta polypeptide CSNK2B, the Ribosomal protein S6 kinase polypeptide 6 (RPS6KA6) and the p50CDC37, an Hsp90 chaperone protein kinase-targeting subunit. Furthermore, the proteasome 26S subunit 6A (PSMC3) and the developmentally regulated GTP binding protein 1 (DRG1) were identified. To verify the newly identified SUMOylation substrates, we analyzed their SUMOylation without the Ubc9 fusion. Because of the unavailability of suitable antibodies for most of the proteins, the SUMOylation was analyzed using GST-tagged proteins which can be pulled down with glutathion-S Sepharose and then be detected with a GST-antibody. Since SUMOylation of many proteins can hardly be detected without the presence of a specific SUMO ligase, we analyzed the potential SUMOylation target proteins also in the presence of the known SUMO ligases PC2, PIAS1/3/xß/γ and RanBP▵ which were ectopically co-expressed. As a further control, SENP2, one of the SUMO deconjugating enzymes, was coexpressed to demonstrate SUMOylation by its disappearance under these conditions. -SUMOylation was first tested for the known SUMOylation substrate p53 (,) and could be detected when coexpressed with EGFP-SUMO1 (A). When SENP2 is coexpressed p53 SUMOylation is clearly diminished (A). SUMOylation of p53 is enhanced by the SUMO ligase PIAS1 () and most strongly by PIASγ () (B). Unexpectedly, in our hands PIASx/ß () did not enhance p53 SUMOylation (B). But we found out that PC2 can enhance the p53 SUMOylation. Of the potential SUMOylation targets we could verify that FOS (A) is SUMOylated when coexpressed with EGFP-SUMO1 and de-SUMOylated in the presence of SENP2. Furthermore its SUMOylation is enhanced by a coexpression of the SUMO ligases PIAS1 and PIASγ. CRSP9 (B) and CDC37 (C) are also SUMOylated when coexpressed with EGFP-SUMO1 and de-SUMOylated in the presence of SENP2, but there was no influence on their SUMOylation by coexpression of SUMO ligases (data not shown). For the remaining 11 SUMOylation substrates, we could not detect SUMOylation when coexpressed with EGFP-SUMO1 and Ubc9 (data not shown). Furthermore, none of the SUMO ligases PC2, PIAS1/3/xß/γ and RanBP▵ induced the SUMOylation of these proteins (data not shown), indicating the involvement of other specific SUMO ligases for these targets. In view of the fact that proteomic studies have identified many stimulation-dependent SUMOylations of substrate proteins (,) and that phosphorylation-dependent SUMOylation sites have been recently described (), we wondered whether UFDS had identified stimulation-dependent SUMOylations. To analyze a stimulation-dependent SUMOylation independent of UFDS, we coexpressed the candidate proteins with a constitutively active MEKK1 (MEKK1ca) (). The constitutive SUMOylations of FOS (data not shown) and CRSP9 (B) were not enhanced by the MEKK1ca coexpression, whereas the SUMOylation of CDC37 by EGFP-SUMO1 was enhanced by coexpressed MEKK1ca and was strongest when both MEKK1ca and Ubc9 were present (C). For five other SUMOylation substrates, CSNK2B (A), TAF10 (B), HSF2BP (C), PSMC3 (D) and DRG1 (E) we detected SUMOylation only when the various MAP-kinase cascades were activated by coexpression of MEKK1ca. Thus, UFDS had identified both target proteins for constitutive and induced SUMOylation. For the six remaining new SUMOylation substrate candidates, we could so far not unequivocally demonstrate SUMOylation in the absence of a fusion to Ubc9. This could be explained by the lack of expression of the specific SUMO ligases or the adequate stimulation in HEK 293 cells or by artificial SUMOylation in the Ubc9 fusion proteins. In spite of the large number of new SUMOylation substrates that were recently identified in proteomic studies, only one of the proteins identified in this study (FOS) was in parallel recognized as SUMOylation substrate (). This clearly makes UFDS a powerful complementary approach for identifying in particular stimulation-dependent protein SUMOylation. Although hundreds of potential SUMOylation substrates have been described recently (), many proteins were probably not accessible to this kind of analysis due to low expression rates or low amount of SUMOylation, which may result from a lack of expression of specific SUMO ligases or of an adequate stimulus. By using UFDS, out of 37 expressed proteins 14 were identified as SUMOylation substrates. Eight of these 14 potential SUMOylation substrates were also SUMOylated without fusion of Ubc9. While three of these are SUMOylated when coexpressed with EGFP-SUMO1 and one exhibited enhanced SUMOylation when coexpressed with known SUMO ligases, five proteins were only SUMOylated when constitutively activated MEKK1 was coexpressed. Therefore, constitutively active MEKK1 mimics various extracellular signal-dependent stimulations of the cell in parallel and serves as an ideal tool to analyze signal-regulated SUMOylation for candidate substrates. However, due to the broad effects of MEKK1ca, no information about the specific pathway involved can be obtained. Further work using specific stimuli and inhibitors of these pathways is needed to understand this signaling in detail. The SUMOylation of many proteins is enhanced by different stress stimuli-like heat shock, ethanol, MG132, serum starvation () or serum stimulation (). Recently, phosphorylation-dependent SUMOylation has been described where phosphorylation(s) C-terminal to the SUMOylation site are necessary () permitting signal-regulated SUMOylation. On the other hand, for the protein STAT1 we have recently shown that SUMOylation inhibits phosphorylation of a site in the vicinity of the SUMOylation (). These findings demonstrate a full regulatory cross talk between phosphorylation and SUMOylation. Here, UFDS identified SUMOylation of substrate proteins in non-stimulated cells, whereas these substrates normally need a special stimulation to become SUMOylated. Therefore, UFDS can not only be used to identify constitutive SUMOylation, but also to identify the induced SUMOylation of proteins which are modified only in the presence of a specific, often unidentified SUMO ligase after a specific stimulation which may also alter subcellular localization or activity of this ligase. Furthermore, these results indicate that UFDS has the potency to study the function of stimulation-dependent SUMOylation without any stimulation of the cell, allowing the characterization of the function of one specific SUMOylated protein avoiding interference by parallel stimulation of other proteins. The characteristics outlined above make UFDS eligible for the identification and analysis of unknown SUMOylation-dependent processes. This is demonstrated by the identification of 14 potential SUMOylation substrates of which only one (FOS) was described as substrate in parallel (). The finding that UFDS is independent of both SUMO ligases () and specific stimuli favors that in case of the remaining six non-verified SUMOylation substrates expression of the specific SUMO ligase or a specific stimulation was not reached in HEK 293 cells. Of course, we can also not exclude that those were artificially SUMOylated by UFDS. Interestingly, the majority of the 14 candidate proteins also represents interaction partners of other described SUMOylation substrates. CDC37 binds to the androgen receptor (), Ciao1 is a binding partner of the Wilms Tumor suppressor protein WT1 (), the Endothelial differentiation-related factor 1 (EDF1) binds to c-Jun () and Fos () and FOS binds to c-Jun as well(). HSF2BP binds to the Heat shock transcription factor 2 (HSF2) () and PC4 interacts with p53 (). TAF10 which we found to be strongly SUMOylated, is part of the TFIID complex that comprises the TATA box binding protein (TBP) and 13 TBP-associated factors (TAFs). Two proteins of this complex, TAF5 and TAF12, have also been reported to become SUMOylated (). These data let us assume that the organization of several multiprotein complexes is assisted by the SUMOylation of more than one SUMOylatable proteins. Nine of the fusion proteins were not expressed and others were strongly reduced in their expression by the coexpression of EGFP-SUMO1. Until now, most reports on SUMO conclude that SUMOylation is not involved in proteasomal protein degradation. We have not yet proved if some of the proteins we investigated are destabilized upon their Ubc9 fusion dependent SUMOylation, but UFDS should be an excellent tool for such investigations at least for proteins which are stated to be destabilized by SUMOylation. Summarizing our data, it is evident that UFDS could be a valuable method for the identification of constitutive and stimulated SUMOylation and for the functional analysis of these kinds of protein SUMOylation leading to new insights on how SUMOylation regulates protein function.

There is a major need to develop fast, cheap and precise detection methodologies that detect DNA samples at extremely low concentration. This ability is critical to basic life sciences, medical diagnosis and treatment, pharmaceutical applications, identification of biological weapons, as well as forensic analysis (). To fulfill this goal, the scientific community is striving to develop new methods and assays that are highly selective and sensitive. Optical/colormetric (), fluorescent (,) and electrochemical () based methods have been reported for detection of DNA samples. Among these new methodologies, optical detection methods, which rely on the hybridization between target DNA and substrate modified with radioactive, fluorescent, chemiluminescent or nanoparticle tags, are of particular interest (). The use of gold nanoparticles (nAu) as labeling tags receives most attention in recent years, due to their unique chemical and physical properties () that can be exploited in the development of highly sensitive detection assays (,). Although still in its infancy, the application of surface-functionalized nAu in sequence recognition has shown great promise in achieving high sensitivity that is difficult to achieve by conventional methods. Mirkin and co-workers have developed a series of nAu–based DNA detection methods, such as scanometric and bio-barcode assays, that reach attomolar and high zeptomolar sensitivity (,,). Such sensitivity may allow the direct detection of genomic DNA and bypass the need of amplification that is usually done using polymerase chain reaction (PCR). Besides sensitivity, quantification and selectivity are the other two important aspects for the evaluation of DNA biosensor devices. DNA quantification is critical for gene expression analysis, detection of DNA mutations or genetic defects, early stage diagnosis of critical illness such as HIV and cancers, and forensic applications (). Furthermore, diagnosis of pathogenic and genetic diseases requires the device to have high selectivity that can discriminate single nucleotide mismatches (,). Single nucleotide polymorphisms (SNPs) are the most abundant form of genetic variation that occur once every 100–300 bases and there are greater than 3 million SNPs in the human genome (). Identification of these SNPs and associate individual SNPs with specific diseases and pharmacological responses are clinically important for medical diagnostics, disease prevention and prognostics (,). These needs have driven intense efforts toward the development of new methodologies that enable quantitative, selective and cost-effective detection of SNP in DNA samples (,). Currently, real-time polymerase chain reaction (RT-PCR) is one of the most frequently used methods for DNA quantification and SNP discrimination in life science and clinical research. However RT-PCR is a time-consuming and labor-intense process, and its selectivity is not always satisfactory even with sophisticated optimizations (,). For commonly used DNA detection systems such as DNA chips, the selectivity and quantification are dependent on the dissociation properties of target DNA hybridized with capture strands immobilized on the chip (). To achieve SNP discrimination, a stringent wash step has to be performed to remove mismatched DNA binding on the capture strands. However, the difference in binding affinity between a perfectly matched target DNA and one with a mismatched base is usually too small to achieve complete discrimination (). Previously, we have shown that gold nanoparticle–DNA (nAu–DNA) conjugates bearing definite number of short DNA (<20 bases) can be prepared by gel electrophoresis isolation followed by restriction endonuclease manipulation of the nAu–bound DNA (). Simply loading short DNA onto the nAu directly followed by gel electrophoresis separation only yields a smear and not individual bands, which correspond to conjugates bearing definite number of DNA. This is because the mobility difference between conjugates bearing different number of short DNA is insignificant. Thus, we reported to first use gel electrophoresis to separate nAu bearing definite number of >50-base DNA strands. Subsequently restriction endonuclease can be used to cleave the long DNA to obtain the short DNA on nAu. In this study, we described a novel gold-nanoparticle (nAu)-based assay methodology that has reliable quantification ability and SNP discrimination selectivity. In this assay, two sets of specially designed nAu–DNA conjugates are fabricated via the gel electrophoresis and restriction endonuclease manipulation methods. These two sets of conjugates with definite number (1, 2, 3…) of short single-stranded DNA (ssDNA) probes are not directly complementary to each other. After mixing, these conjugates do not recognize and group each other until a target DNA that is complementary to both sets of conjugates is introduced. Only conjugate groupings with defined structure (dimer or trimer) can form due to definite number of DNA strands on each nAu. The resulting conjugate groupings are characterized and quantified by agarose gel electrophoresis. The size differentiation ability of gel electrophoresis allows strict discrimination between different conjugate structures (monomer, dimer and trimer) and enables precise quantification of target DNA samples. Furthermore, a strong discrimination between perfectly matched and single base mismatched DNA is achieved since only the perfectly matched target DNA allows the formation of conjugate groupings. Hydrogen tetrachloroaurate (III) trihydrate, trisodium citrate dihydrate, tannic acid, dithiothreitol (DTT), NaCl, MgCl and tris-borate-EDTA (TBE) buffer were purchased from Sigma-Aldrich. The synthetic DNA (modified with thiol linker at the 5′ end) was purchased from Integrated DNA Technologies (IDT). Restriction endonuclease EcoRV (500 000 U/ml) was obtained from New England Biolabs. Thirty percent acrylamide/Bis solution (29:1) and ethidium bromide were obtained from Bio-Rad Laboratories. Agarose was purchased from Cambrex. Dialysis tubing (Float-A-lyzer, MWCO: 3500) was obtained from Spectra/Por. Milli-Q water with resistance >18 MΩ/cm was used throughout the experiments. nAu were synthesized by the reduction of hydrogen tetrachloroaurate (III) trihydrate by trisodium citrate dihydrate and tannic acid (). In order to determine the size and size distribution of the resulting nAu, TEM characterization was performed on a Philips CM300 FEG system operating at 200 kV. At least 200 particles were sized from TEM micrographs via graphics software ‘Image-Pro Express’ (Media Cybernetics). The mean particle diameter was 10 nm and the size distribution was within 15% of the mean diameter. The sequences of DNA used in this study are shown in (complete sequences are shown in Figure S1, Supplementary Data). Two complementary single-stranded DNA (ssDNA), one modified with a thiol linker, was allowed to hybridize to form a double-stranded DNA (dsDNA). Two dsDNA, Strand A′ with a 5′ thiol group and Strand revA′ with a 3′ thiol group, were used to prepare two sets of conjugates (nAu–A′ and nAu–revA′) for conjugate grouping studies. The detailed procedure of these conjugates preparation can be found in our previous work (). Briefly, 2 μM dsDNA (Strand A′ or revA′) was mixed with nAu for 2 h followed by 5-T ssDNA for surface passivation. Excess reagents were then removed by repeated washing and centrifuging the samples with 0.5X TBE. Agarose gel electrophoresis (3% agarose at 5 V/cm, 0.5X TBE as running buffer) was carried out for 4 h to separate the conjugates bearing from one to three dsDNA molecules. As shown previously, discrete bands can easily be identified because of the wine-red color of nAu (). Desired bands were extracted from the gel and the conjugates were then recovered by electrophoretic dialysis (). After recovering from agarose gel, the conjugates were purified by repeated washing (with 50 mM tris buffer, pH 8.0) and centrifuging to ensure complete removal of EDTA which may deactivate the restriction endonuclease. The restriction endonuclease digestion of nAu–bound DNA (Strands A′ and revA′ to Strands A and revA respectively) was performed by incubating the conjugates with 100 units of EcoRV at 37°C in a 200 μl reaction buffer for 20 h. After incubation, the enzyme was deactivated by adding 50 mM EDTA. The digested conjugates were analyzed by polyacrylamide gel electrophoresis to confirm the high digestion efficiency and high yield of nAu conjugated with definite number of DNA strands (Figure S2, Supplementary Data). Two sets of nAu, each carrying a single DNA probe (nAu–A or nAu–revA, 1.5 pmol of each) were mixed in a buffer containing 50 mM tris pH 8.0, 100 mM NaCl and 2 mM MgCl at a 1:1 molar ratio. Target/linker DNA molecules (matched and mismatched, ) were then added for hybridization with Strand A and rev A on the nAu surface in a tail-to-tail configuration (). The ratio between the two sets of nAu–DNA conjugates and the target DNA (nAu–A: nAu–revA: target) varied from 1:1:0.1 to 1:1:1. The hybridization of target DNA was carried out by first heating the sample to 70°C for 2 min to ensure complete melting of the DNA strands and then the samples were slowly cooled down to 25°C at a rate of 0.2°C/min to form stable duplexes. After formation of conjugate groupings, agarose gel electrophoresis (3% agarose at 5 V/cm, 0.5X TBE as running buffer) was carried out at 4°C to characterize and quantify the conjugate groupings. Desired bands of conjugate groupings were visualized by a Syngene GeneGenius UV/white light gel documentation system or a Bio-Rad GS-800 calibrated densitometer. The grouping percentage of nAu–DNA conjugates was calculated by quantifying the relative optical intensity of the gel bands in the same lane using gel analysis software from manufacture, more specifically, by dividing the amount of conjugate groupings by the total amount of conjugates in the same lane (unbound conjugates + conjugate groupings). Each value reported was the average of three individual tests. After that the conjugate groupings were extracted from the gel and recovered by electrophoretic dialysis for TEM characterization. As reported previously, agarose gel electrophoresis was first used to isolate conjugates (nAu–A′ and nAu–revA′) bearing different number of DNA molecules (). Subsequently the conjugates with definite number of DNA strands were exposed to endonuclease EcoRV digestion to cleave Strands A′ and revA′ (80 bp) into Strands A and revA (18 bases). The enzyme digestion efficiency of 10 nm nAu–bound DNA obtained from 12% PAGE (Figure S2, Supplementary Data) shows more than 90% digestion efficiency for both Strands A′ and revA′. The use of endonuclease to cleave nAu–bound DNA has been reported by several groups () including ourselves (,). Our high digestion efficiency here leads to short and homogeneous DNA on nAu which is critical for the subsequent quantitative DNA analysis. Though nanoparticle assemblies has been studied extensively, most studies use nanoparticles with a high loading of DNA and lead to a coordinate effect from multiple DNA linkages. To avoid this scenario, nAu–A and nAu–revA conjugates with single DNA strands are used here. nAu–DNA conjugate dimer formed with various DNA targets/linkers (20 bases to 26 bases) was chosen as a model system. A tail-to-tail structure was selected to reduce the steric hindrance and allow maximum grouping efficiency (). The conjugate dimers formed by nAu–A and nAu–revA grouping are shown in . The stoichiometric ratio of nAu–A, nAu–revA, and target DNA used in hybridization is 1:1:1. Since Strand A and Strand revA are not complementary to each other, a linker is needed for the formation of conjugate dimers. Lane A in corresponds to nAu without Strand A or revA modification (control). Lanes B, C, D, E and F correspond to nAu–A + nAu–revA conjugates with no target DNA, with 26, 24, 22 and 20-base target DNA respectively. Using Lane B as a reference (contains only nAu–A and nAu–revA conjugate monomers), the bands on the bottom of Lanes C/D correspond to the unbound conjugate monomers (share the same mobility with the band in Lane B) and the bands on the top (with slower mobility) correspond to the conjugate dimers. The presence of only a single band in Lane B indicates that there is no nonspecific interaction between the nAu–A and nAu–revA. Therefore, the hybridization of complementary target DNA with these two conjugates, not nonspecific interaction, drives this conjugate dimer formation. With DNA targets of different lengths, the dimer grouping percentages of 26- and 24-base DNA targets are 55.6 and 55.3% respectively, even 1:1:1 ratio is used to allow complete grouping formation. Similar result was previously reported by Alivisatos and co-workers in which complete grouping of nAu–DNA conjugates could not be achieved even when a target with complementary sequence of 100-base was used (,). In our experiment, when the 20- and 22-base DNA targets are used, only a single band corresponding to conjugate monomers is unexpectedly found in the gel (Lanes E and F). With half of the DNA target hybridizes to each conjugate, reducing the target from 24 to 22 bases cuts only a single base in conjugate hybridization. This single base cut, however, results in total absence of dimer groupings. The modification of nAu with 5-T ssDNA as surface passivation in our experiment leads to a highly negatively charged surface. The strong electrostatic repulsion between both conjugates and perhaps target DNA may destabilize the hybridization of short sequences, and thus no dimer is formed. The critical length of DNA target for this particular system is found to be 24 bases. Our nAu–DNA conjugates carry only a single DNA probe and this allows quantitative analysis of target DNA that acts as a linker. This is because each conjugate dimer formation represents a hybridization event between a single complementary target DNA and the two nAu–DNA conjugates. By measuring the dimer formation, in principle, the amount of the target or linker DNA can then be quantified. In our experiments, equal molar of the nAu–A and nAu–revA conjugates was mixed with various ratios of 24-base target DNA (nAu–A: nAu–revA: target DNA ranges from 1:1:0.1 to 1:1:1). The target DNA amount was obtained by quantifying the percentage of conjugate dimers after agarose gel electrophoresis. In , Lane J corresponds to the control where nAu was conjugated without Strand A/revA, and Lanes K is the mixture of nAu–A and nAu–revA without target DNA. Only one band corresponding to conjugate monomers is found in the gel, indicating no dimer formation. Lanes G, H, I, L, M, N and O are samples with various ratios of nAu–A and nAu–revA to target DNA, from 1:1:1 to 1:1:0.1 respectively. Two bands, which correspond to conjugate monomers and dimers, are found in the gel. The amount of the conjugate dimers increases with the increasing molar amount of target DNA. This trend is more clearly shown in after quantifying the dimer band intensity. When the molar ratio of the target DNA is between 0 and 0.5, the dimer grouping percentage is directly proportional to the target DNA ratio. Such linear relationship between the amount of target DNA and dimer formation has good potential for the development of new target quantification assay. When the molar ratio goes beyond 1:1:0.5, both and show that only minor increase in the grouping percentage (less than 20%) is found. For example, 1:1:0.8 and 1:1:1 can only result in 59 and 64% dimer formation respectively. Similar grouping percentage is also found when conjugates with two Strand revA (nAu–2x revA) were used. The top bands in Lanes P, Q and S in correspond to the conjugate trimers that have the lowest mobility among all the conjugates. The second bands from the top are conjugate dimers, and bands at the bottom are conjugate monomers. The maximum grouping percentage (dimers + trimers) of 66% is achieved when molar ratio of 2:1:2 is used. Excess target DNA, however, does not further improve the grouping percentage as the grouping percentage falls slightly to 60% when the ratio of target DNA increases to 2:1:5. Further drop in grouping is observed upon the introduction of large excess of target DNA, as the Lanes S and T (ratio 2:1:20 and 2:1:100) show fade bands of conjugate groupings with only 51 and 36% grouping percentage respectively. This decrease in grouping percentage can be attributed to the conjugate grouping inhibition by the excess target DNA. Since each nAu–DNA conjugate may hybridize with individual target DNA molecule separately, less conjugates are available for grouping and thus the dimer/trimer formation is impeded. To give a direct connection between conjugates with different electrophoretic mobility and their actual structure, dimer groupings were recovered from the gel and visualized by TEM. shows the structure of the conjugate groupings from the dimer band extracted from Lane P. The large majority of the conjugates are participated in the same dimer grouping structure, indicating that each discrete gel band does contain a single type of grouping and not a mixture of several conjugate structures. A small number of monomers and high-order groupings were observed in the TEM image, and this may be due to the disruption during sample extraction from gels as well as during TEM preparation where solvent evaporation may result in particle aggregation. Conventional techniques for the detection of SNP using mass spectrometry or gel electrophoresis to discriminate DNA fragments (,) are time consuming and relatively costly. Chip-based detection methods using fluorescent dyes or nanoparticles as labels have become popular in recent years (,). These newly emerging methods are based on the melting temperature difference between perfectly matched and mismatched DNA duplex, and involve a stringent wash process with precise temperature control as well as skillful personnel and long analysis time. Furthermore, it may be difficult to discriminate DNA targets that exhibit insignificant melting temperature difference. Compared with the existing techniques mentioned above, our newly proposed method offers the advantage of straightforward single-base mismatch discrimination without the need of stringent wash (). As shown in , Lanes U-AA correspond to nAu–A and nAu–revA plus 24-base perfectly matched DNA (PM), single base matched DNA (SM1/SM2/SM3), double base matched DNA (DM), non-complementary DNA (NC) and no target DNA respectively. There is an obvious difference in Lane U where the PM case shows a top dimer band and a bottom monomer band. For the DM and NC cases (Lanes Y and Z), only a single band can be found in the gel, indicating that no dimer structure is formed. For SM1, 2 and 3 cases (Lanes V, W and X), only a single band corresponding to conjugate monomers is found, and no false positive signal is observed. The monomeric structure in these three lanes can be further confirmed by the one without target DNA (Lane AA), which shows exactly the same electrophoretic mobility. Even with end-mismatched SNP (Lane X), which introduces least interruption to the duplex DNA structure and is often difficult to discriminate, this method shows very good discrimination and no sign of conjugate dimers is observed in Lane X. The key strategy to achieve such high selectivity in SNP discrimination is to use unfavorable conditions for hybridization, so that only target DNA with a perfect match has a chance to hybridize with nAu–bound DNA. This unfavorable condition is most likely resulted from the highly negatively charged nAu surfaces (nAu–A) which repel the target DNA from hybridizing with the other conjugate (nAu–revA). As a result, even single base mismatched at the end position (SM3) can lead to sufficient destabilization to the dimers, and only perfectly matched duplex is energetically stable to form dimer grouping. To further confirm the critical role of nAu in SNP discrimination, the melting behavior of free dsDNA (PM, SM1, SM2 and SM3) in the absence of nAu was conducted using real-time PCR (Figure S4, Supplementary Data). The results show that the center-mismatched SM1 and SM2 show wider melting transition and lower peak value compared with PM. However, the end-mismatched SM3 shows almost the same melting transition and is hardly distinguishable compared with PM. This is not unexpected since end mismatching introduces least disturbance to the DNA duplex, the difference in melting behavior between PM and SM3 is too small to support an accurate discrimination. Due to the difficulty of detecting end-mismatched sequences, the location of SNP should be placed at the middle of the probe in the design of probe sequences to enhance discrimination ability. To check the SNP discrimination ability of our system with longer DNA, we used 26-base target DNA to test the same nAu–DNA conjugates. We found the discrimination is as good as that of the 24-base target. As shown in , we mixed, in Lane BB, perfectly matched (PM) target with three other single mismatched sequences (SM4, SM5 and SM6) for conjugate dimer formation. The ratio between nAu–A, nAu–revA and PM/SM4/SM5/SM6 is 1:1:0.25/0.25/0.25/0.25. The result shows that ∼25% of dimer is formed, which is the amount of PM. Therefore the quantitative analysis of target DNA is not interfered by the presence of mismatched DNA. Using SM4, SM5 and SM6 alone, no dimer formation can be found in Lanes CC, DD and EE. SM6 has again an end-mismatched sequence and our nAu conjugate system is able to differentiate it. Therefore, we expect our system to be applied to even longer target DNA without jeopardizing the discrimination ability. The data reported here show that the current conjugate grouping method reaches only 100 fmol range sensitivity, which is not as sensitive as other methods reported in the literature (,,). Thus, the method is likely applicable for analyzing amplified DNA and not genomic DNA directly. Further work to improve the sensitivity is necessary for possible genomic DNA detection. Finally, a possible merit of the method reported here can be its ease of use. Most existing and highly sensitive methods require specialized readout platforms, such as surface plasmon resonance (SPR), fluorescent microarrays, and scanometry. Storhoff and co-workers reported a straightforward DNA detection method by allowing nAu–DNA conjugates to cross-link in the presence of target DNA in solution (). The cross-linked nAu shows a red shift in the scattering spectrum and the result can be read even with naked eyes. Our method also requires fairly simple readout, which is a gel electrophoresis setup and is commonly found in biology and clinical labs. t h i s s t u d y , w e h a v e d e m o n s t r a t e d a n o v e l n A u – b a s e d q u a n t i t a t i v e D N A a s s a y m e t h o d w i t h S N P d i s c r i m i n a t i o n s e n s i t i v i t y . T h i s m e t h o d c o m b i n e s g e l e l e c t r o p h o r e s i s i s o l a t i o n a n d r e s t r i c t i o n e n d o n u c l e a s e m a n i p u l a t i o n t o p r o d u c e p r e c i s e l y c o n t r o l l e d n A u – D N A c o n j u g a t e s w h i c h a l l o w s q u a n t i t a t i v e a n a l y s i s o f D N A m o l e c u l e s b a s e d o n t h e f o r m a t i o n o f c o n j u g a t e g r o u p i n g s b y t a r g e t D N A l i n k a g e . A l i n e a r c o r r e l a t i o n b e t w e e n t h e a m o u n t o f t a r g e t D N A a n d c o n j u g a t e g r o u p i n g s w a s o b t a i n e d a t l o w e r t a r g e t D N A c o n c e n t r a t i o n a n d c a n f u r t h e r b e e x p l o i t e d f o r t a r g e t q u a n t i f i c a t i o n . F o r S N P s t u d y , s i n g l e b a s e m i s m a t c h d i s c r i m i n a t i o n i s a c h i e v e d f o r b o t h t h e e n d - a n d c e n t e r - m i s m a t c h e d c a s e s . p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Many gene families in mammals are expressed in a developmental stage and/or tissue-specific pattern under the influence of a locus control region (LCR) (). LCRs are complex transcriptional enhancer elements located at a distance from target genes. Typically, LCRs comprise clustered DNase I hypersensitive sites (HSs) in a region of chromatin enriched for acetylation of H3 and H4 and di-methylation of H3 K4, histone modifications associated with active chromatin regions (). The sequences within DNase I HSs contain binding motifs for transcriptional activators which are important for establishment of the distinct chromatin structure of the LCR and for LCR enhancer function (). Specific LCR-binding activators may mediate chromatin modifying activity directly (,) or they may recruit or increase chromatin modifying complexes at an LCR, including SWI/SNF type complexes and histone acetyltransferases (HATs) (). It has long been observed that DNase I HSs denote regulatory elements in chromatin such as enhancers and promoters but the nature of these sites has been difficult to ascertain. The current view is that these sites may be nucleosome-free regions of DNA or they may contain ‘altered’ nucleosomes that do not strongly protect DNA from nuclease attack (). One approach to distinguish these possibilities is to determine nucleosome occupancy by chromatin immunoprecipitation (ChIP) with antibodies to total histone H3. H3 occupancy data can be used to correct histone modification profiles and was used on a genome-wide scale to show that active gene promoters are depleted of nucleosomes (). It is likely that nucleosome-free regions are also associated with enhancers/LCRs (). Although it is a prediction that transcription activators will occupy these DNase I HSs in lieu of nucleosomes, this relationship has not been probed at high resolution. The human β-globin locus contains an LCR that confers erythroid-specific enhancer activity to the globin genes which are expressed sequentially during development (). The LCR consists of four HSs (HS1–HS4) that are located far upstream of the globin genes. The sites contain similar binding motifs for erythroid-specific and ubiquitous transcription activators and are generally considered to contribute to LCR enhancer activity collectively and to come into close physical contact in a ‘chromatin hub’ in erythroid cells (). HS5, which flanks the 5′ end of the locus, is also part of the hub and displays certain characteristics of a chromatin insulator including binding of the insulator protein CTCF (). NF-E2 and GATA-1 are among the erythroid activators interacting at the β-globin LCR HSs that play critical roles in transcription activation of the globin genes (). These factors interact differentially at the LCR HSs of the murine globin locus (). Chromatin hub interactions bringing the LCR and actively transcribed globin genes into physical proximity in murine erythroid cells require GATA-1 (). GATA-1 also participates in recruitment of CBP to murine LCR HS3 () and increases LCR recruitment of BRG1, a component of the SWI/SNF nucleosome remodeling complex (). In human erythroid K562 cells that express the embryonic ε and fetal γ-globin genes but not the adult β-globin gene NF-E2 and GATA-1 play distinct roles in the enhancer activity of HS2 (,) and influence recruitment of CBP, p300 and SWI/SNF components to HS2 (,). We previously observed varying levels of DNase I hypersensitivity among the β-globin LCR HSs in human K562 cells using a quantitative approach (). Here we probed the relationship between nucleosome occupancy, activator and co-activator binding and histone modifications at the LCR HSs in a representative population of native mono and di-nucleosomes prepared from K562 cell nuclei. Sequences corresponding to HS1, HS2 and HS4 were highly depleted of nucleosomes. HS5 was less depleted and at HS3 nucleosomes were not lost. Histones remaining at HSs were hyperacetylated on H3 and H4 to varying degrees that did not always correlate with the extent of nucleosome loss. GATA-1 occupied HS1, HS2 and HS4, while NF-E2 was strongly detected only at HS2. The co-activators CBP/p300 and Ash2L occupied HS2 and NF-E2 interaction there was required for Ash2L recruitment. Interestingly, even though HS5 contains motifs for GATA-1 and NF-E2, these factors were not present at HS5 which was otherwise occupied by factors associated with insulator activity. These studies show that different HSs of the β-globin LCR in K562 cells have distinct chromatin structural attributes and factor occupancy which may reflect the organization of the complete LCR. K562 cells were grown in RPMI 1640 medium containing 10% FBS and harvested at confluence of 4–6 × 10 per ml for all experiments. HeLa cells were cultured in DMEM with 10% FBS. K562 cell clones carrying stably maintained episomes that contain HS2 linked to a complete ε-globin gene have been described (). The tandem NF-E2 or GATA-1 binding motifs of HS2 were mutated by site-directed mutagenesis in episomes. Nuclei were prepared from 5 × 10 K562 or HeLa cells, digested with different MNase concentrations (0.0025 units/ul, 0.01 units/ul and 0.04 units/ul) and combined (). Soluble chromatin was fractionated on a sucrose gradient (5–30%). DNA purified from fractions containing mainly mono- and di-nucleosomes was used for further study. To determine MNase sensitivity, 1 ng of purified DNA was compared with 1 ng of total genomic DNA (briefly digested with EcoRI) by real-time quantitative PCR using the comparative Ct method (). In this analysis, nuclease sensitive sequences will be depleted compared to the total genomic sample. Histone modifications were analyzed using purified nucleosomes without formaldehyde cross-linkage (). Nuclei from 5 × 10 K562 cells were digested with different concentrations of MNase as for the sensitivity assay. Soluble chromatin was fractionated on a sucrose gradient (5–30%). Mono- and di-nucleosome were pooled and reacted with antibodies after pre-clearing with protein A agarose. Immunoprecipitated nucleosome-protein A agarose complexes were washed and DNA was eluted as described (). ChIP assays for transcription factors CBP, p300 and Ash2L were carried out using cross-linked chromatin (). Briefly, 2 × 10 K562 cells were incubated in growth medium containing 1% formaldehyde for 10 min at 25°C, and then the cross-linking reaction was quenched by adding glycine to 0.125 M. Chromatin of primarily mono-nucleosome size was prepared by MNase digestion and sonication, and reacted with antibodies after pre-clearing with protein A or A/G agarose. Immunoprecipitated chromatin was bound to protein A or A/G agarose bead, washed and eluted as described (). Antibodies for NF-E2, GATA-1, CBP and p300 and normal rabbit IgG were purchased from Santa Cruz Biotechnology, Santa Cruz, CA, USA. The antibody for Ash2L was prepared by Marjorie Brand. Anti-di-acetylated (K9, K14) histone H3, anti-tetra-acetylated (K4, K8, K12, K16) histone H4, and anti-di-methylated H3 (K4) were obtained from Upstate Biotechnology, Lake Placid, NY, USA. Purified DNA was analyzed by quantitative real-time PCR (ABI Prism 7900) using TaqMan probes and primers (Primer Express 1.0, PE Applied Biosystems). Real-time PCR was carried out with 200 nmol of TaqMan probes and 900 nmol of primers in a 25 µl reaction volume. Data were collected at the threshold where amplification was linear. In cross-linked ChIP assays, the relative enrichment for each primer pair was determined by comparing the amount of target sequence in 1.25% of immunoprecipitated DNA to the amount of target sequence in 0.1% of input DNA. In ChIP assays for histone modifications using nucleosomes, the fold enrichment was determined by comparing equal amounts (1 ng) of immunoprecipitated DNA and input DNA. Sequences of primers and TaqMan probes have been described (). To investigate the presence or absence of nucleosomes in the LCR directly, we digested non-cross-linked nuclei with graded concentrations of MNase and pooled and purified the digests to produce a complete, representative pool of native nucleosomes (). At the lowest concentrations, MNase will attack linker regions between nucleosomes and sequences not protected by a nucleosome such as sites of extensive factor occupancy. At successively higher concentrations, DNA associated with nucleosomes of varying stability will eventually be digested. Nuclei from K562 cells expressing the embryonic ε and fetal γ-globin genes and from non-erythroid HeLa cells that do not express these genes, were digested with different concentrations of MNase, the digests combined, and mono and di-nucleosomes purified by sucrose gradient fractionation (B) (see Methods section). Sensitivity to MNase was determined by comparing nucleosome-associated DNA with genomic DNA at and between LCR HSs using quantitative real-time PCR with TaqMan probes (A). HSs in the LCR cores showed much higher MNase sensitivity than intervening regions between the HSs sites in K562 cells indicating nucleosome loss at these sites as expected (C). However, the magnitude of nucleosome loss was variable at different HSs. Notably, the core regions of HS1, HS2 and HS4 were highly depleted of nucleosomes, whereas DNA corresponding to HS5 was less depleted although all these sites were equally DNase I sensitive (). MNase sensitivity was not notable at HS3. The lack of nucleosome remodeling at HS3 in K562 cells is consistent with a very weak DNase I sensitive site there (,). We note that the relative intensity of nucleosome loss indicated by MNase sensitivity at the LCR core HSs is in good agreement with results obtained from ChIP assays of formaldehyde cross-linked chromatin using an antibody against total histone H3 () (see Supplementary Figure 1) as would be predicted if a fully representative pool of nucleosomes had been obtained by graded MNase digestion. None of the HSs were depleted of nucleosomes in HeLa cells (C). To investigate directly the histone modifications present on nucleosomes that remained associated with the LCR, we analyzed by ChIP H3 and H4 hyperacetylation and di-methylation of H3 K4 using non-cross-linked mono- and di-nucleosomes. Purified nucleosomes were reacted with antibodies to acetylated H3 and H4 and di-methyl H3 K4 and immunoprecipitated DNA was analyzed by quantitative real-time PCR across the β-globin LCR using the TaqMan probes shown in A. Across the LCR sites tested, histones H3 and H4 were hyperacetylated and di-methylated at K4 compared to the repressed necdin gene (A–C). The patterns of the modifications were quite similar and variable across the locus, and we found that peak modification sometimes occurred outside the HSs cores (i.e. HS3 5′, A) consistent with data on histone modification in the murine LCR (). Compared to other sites, HS1, HS3 and HS4 were highly modified and HS2 and HS5 were less so. Thus, there was an imperfect correlation between histone modifications and histone depletion. For example, HS1, HS2 and HS4 were the most highly depleted sites but only HS1 and HS4 appeared as peaks of histone modifications. In addition, HS3 was as highly modified as HS1 and HS4 but was not depleted of nucleosomes. HS5 significantly retained nucleosomes and was only weakly marked by histone acetylation and H3 K4 di-methylation. Weak acetylation of HS5 nucleosomes compared to the other LCR HSs contrasts with findings that the homologous chicken 5′HS4 insulator site and an ectopic chicken 5′HS4 in human cells are very strongly marked by acetylated H3 (,). However, recent studies showed that murine HS5 was not hyperacetylated (). The nucleosome preparation used to obtain the data in and without cross-linking of nuclei results in MNase digestion and loss of sequences not bound to histones. Nucleosome loss in LCR HSs might be a prerequisite for or occur concomitantly with the binding there of transcriptional activators or activator binding might preclude re-deposition of nucleosomes following cell division. To investigate whether the nucleosome-depleted sequences were occupied instead by transcription activators, ChIP was carried out with nuclei cross-linked by formaldehyde to stabilize protein–DNA interactions. The HSs of the human β-globin LCR contain multiple -elements for binding of erythroid and ubiquitous transcriptional activators (A). We focused on binding of the erythroid factors NF-E2 and GATA-1 in K562 cells using ChIP assays and antibodies against these proteins. Cells were cross-linked by formaldehyde treatment and chromatin was fragmented by sonication and MNase digestion to 100–200 bp size, comparable to the nucleosomal chromatin used for the data in and (Supplementary Figure 2). DNA obtained after ChIP was analyzed across the LCR by quantitative real-time PCR using the TaqMan probes shown in A. The analysis showed that NF-E2 and/or GATA-1 were specifically bound at HS1-HS4 (B). NF-E2 association was essentially limited to HS2. Association of GATA-1 was observed at HS1, HS2 and HS4: the GATA-1 signal at HS3 was very low, which might be related to the retention of nucleosomes at that site. At HS5, there was no significant signal for these activators even though GATA-1 and NF-E2 motifs are present in HS5 (A). In earlier work, we showed that the insulator factors CTCF and USF occupy this site (,). The results show that the HSs in the human β-globin LCR have distinct associations with transcriptional activators as they do in the murine β-globin locus and that not all motifs for an activator are occupied in chromatin (,). Histone acetylation and methylation are carried out by acetyltransferase (HAT) and methyltransferase (HMT) complexes, respectively. CBP and p300 HATs are important for histone acetylation in the globin locus (,,). Human H3K4 methyltransferase complexes, of which Ash2L is a shared component (), have not been detected in human globin locus sequences although the mono-, di- and tri-methylated forms of H3 K4 are variably detected across the LCR and human β-globin locus (). We examined the distribution of co-activators CBP, p300 and Ash2L protein across the human β-globin LCR using chromatin prepared by the same procedure used for ChIP assays of transcriptional activators and immunoprecipitation using antibodies against these proteins. The distributions of the three co-activators in the LCR were highly consistent with one another (A and B). A strong signal with all antibodies was detected in HS2, where NF-E2 interaction was strong, while the other LCR HSs showed weak signals. Interestingly, the association of CBP and p300 at HS1 and HS4 was very weak in comparison to HS2 even though strong association of GATA-1 was observed at those sites and GATA-1 is implicated in CBP and p300 recruitment to the LCR (,,). To investigate whether GATA-1 and/or NF-E2 were important for recruitment of Ash2L to HS2, we performed a ChIP assay using K562 cell clones with stable, chromatinized episomes that contain wild-type or GATA-1 or NF-E2-mutated HS2 linked to a complete ε-globin gene. C shows that mutation of the HS2 GATA-1 site did not affect Ash2L recruitment to HS2. However, loss of NF-E2 binding resulted in failure to recruit Ash2L to HS2. Thus, NF-E2 is implicated in recruitment of H3 K4 histone methyltransferase activity to the β-globin LCR. #text S u p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

RNA interference (RNAi) is an evolutionarily conserved mechanism in which double stranded RNA (dsRNA) induces sequence-specific gene silencing (). RNAi is initiated by the RNase III-like enzyme Dicer that processes dsRNAs into 21–25 bp dsRNAs called small interfering RNAs (siRNAs) (). Subsequently, the siRNA is incorporated into the RNA-induced silencing complex (RISC), which uses one strand of the siRNA as a guide to target the complementary mRNA for cleavage (). In mammalian cells, RNAi can be induced via transfection of synthetic siRNAs () or DNA vectors for intracellular expression of short hairpin RNAs (shRNAs) (). RNAi induction via transfection of siRNAs or via stable expression of shRNAs has been shown to be highly effective in inhibiting HIV-1 (). However, HIV-1 can escape from RNAi by introduction of nucleotide substitutions or deletions in the siRNA target sequence (). In addition, HIV-1 can also escape from RNAi-mediated inhibition through mutations that alter the local RNA secondary structure (). To reduce the chance of escape, the virus should be targeted simultaneously with multiple highly efficient siRNAs. We recently identified a large set of potent shRNA inhibitors against different highly conserved HIV-1 sequences (), which can be used to target multiple sites. There are several ways to express multiple effective siRNAs. One possibility is the insertion of multiple shRNA-expression cassettes in a viral vector. However, repeats of the same regulatory sequences, e.g. the H1 polymerase III promoter, may cause genetic instability and reduced titer of the vector system (Ter Brake and Berkhout, submitted for publication). Ideally, the expression of multiple antiviral shRNAs should be coordinated by putting them in a single transcript. Another possibility is to use long hairpin RNAs (lhRNAs), from which multiple siRNAs can be produced. Several reports described efficient RNAi induction by lhRNAs against human deficiency virus 1 (,), hepatitis C virus (,) and hepatitis B virus (). Importantly, intracellular lhRNA expression does not seem to induce non-specific type 1 interferon (IFN) responses in cells, which may occur when dsRNA longer than 30 bp is introduced in mammalian cells (,). Induction of a type I IFN response will lead to non-sequence specific degradation of mRNAs. However, recent studies showed that smaller duplexes can also activate the IFN response (,), which is dose-dependent. To learn how to design effective extended hairpin RNAs (e-shRNAs), from which several effective siRNAs can be produced, we constructed a series of antiviral e-shRNA constructs and examined their silencing activities on luciferase reporters and HIV-1. We determined the minimal hairpin stem length to produce two active siRNAs. Furthermore, we demonstrated that siRNA activity correlates with proper processing of the e-shRNAs. Importantly, we showed that extended hairpin RNA transcripts are highly efficient in inhibiting HIV-1 production, without induction of the IFN response in cells. These results provide building principles for the design of multi-siRNA hairpin constructs for durable inhibition of escape-prone RNA viruses. Hairpin RNA constructs were made by annealing of complementary oligonucleotides and inserting them into the BglII and HindIII site of the pSUPER vector (). The sequences of the oligonucleotides used are listed in Supplementary Table 1. siRNA sequences for constructing the hairpin construct targeting the luciferase gene (GL3) were described previously (,). Luciferase reporters containing 50 nt of the HIV-1 gag, pol and nef gene in the 3′ UTR were made using annealed oligonucleotides, which were ligated between the EcoRI and PstI sites of the firefly luciferase vector pGL3 (Promega). The same procedure was used for construction of the luciferase reporter with a 19 nt gag target. The plasmid encoding the HIV-1 isolate LAI () was used to produce virus in transfected 293T cells. For amplification of plasmids, GT116 cells were transformed with the constructs by electroporation with Gene Pulser II, 25 μF, 200 Ω and 2.5 kV. The bacteria were grown in Luria Broth medium containing 100 μg/ml ampicillin by shaking at 37°C or on LB agar plates containing 100 μg/ml ampicillin at 37°C. Plasmids were isolated using the Qiagen Maxiprep or Midiprep kit (Qiagen) and the sequences were verified using the BigDye Terminator Cycle Sequencing kit (ABI, Foster City, CA, USA). For sequencing of hairpin RNA constructs, a sample denaturation temperature of 98°C was used and 1 M of Betaine was added in the reaction (). The Mfold program () was used to determine the secondary structure of RNA transcripts. The human embryonic kidney (HEK) cell line 293T was maintained in Dulbecco's modified Eagle's medium (DMEM; Invitrogen, Carlsbad, CA, USA) containing 10% fetal calf serum (FCS), 100 U/ml penicillin, 100 U/ml streptomycin and minimal essential medium non-essential amino acids at 37°C and 5% CO. For luciferase assays, HIV-1 inhibition assays and the interferon assay, 293T cells were seeded in 24-well plates at a density of 1.5 × 10 cells per well in 1 ml of DMEM with 10% FCS without antibiotics one day prior transfection. Transfections were performed using Lipofectamine 2000 reagent (Invitrogen) according to the manufacturer's instructions. For small RNA analysis, 1.6 × 10 293T cells were seeded in T25 flasks. The next day, the cells were transfected with 5 μg of the hairpin RNA constructs using Lipofectamine 2000 reagent (Invitrogen) as suggested by the manufacturer. For luciferase assays, cells were co-transfected with 100 ng of firefly luciferase expression plasmid (pGL3; Promega, Madison, WI, USA), 1 ng of renilla luciferase expression plasmid (pRL-CMV) and different amounts of hairpin RNA expression constructs. pBluescript (pBS) was added to the transfection mixture to ensure equal DNA amounts. Transfected cells were lysed at 48 h post-transfection in 150 μl 1× passive lysis buffer (Promega) by gentle rocking for 15 min at room temperature. The cell lysates were centrifuged for 5 min at 4000 rpm at 4°C and 10 μl of the supernatant was used to measure firefly and renilla luciferase activities with the Dual-Luciferase Reporter Assay System (Promega). The relative luciferase activity was calculated as the ratio between the firefly and renilla luciferase activities and corrected for between-session variation (). To determine inhibition of virus production, 293T cells were co-transfected with 250 ng of the HIV-1 pLAI, 1 ng pRL-CMV and 2.5, 10 or 25 ng of the hairpin RNA constructs. pBS was added to the transfection mixture to ensure equal DNA amounts. The cells were grown at 37°C and harvested 48 h after transfection. Virus production was monitored by determining the CA-p24 level in the culture supernatant by ELISA as described previously (). Subsequently, the cells were lysed as described above and 10 μl of the supernatant was used to measure the renilla luciferase activities using the Renilla Luciferase Assay System (Promega). The relative CA-p24 production was calculated as the ratio between the CA-p24 level and the renilla luciferase activity and corrected for between-session variation (). Small RNAs were purified from the transfected 293T cells at 72 h post-transfection using the mirVana miRNA isolation kit (Ambion) according to the manufacturer's instructions. For northern blot analyses, 3 μg of small RNAs was resolved on urea 15% denaturing polyacrylamide gels (Invitrogen). For a size reference, we used the decade RNA molecular weight marker (Ambion, Austin, TX), which was radioactively labeled and pre-treated as suggested by the manufacturer. Prior to loading, the samples were heated for 5 min at 95°C and immediately placed on ice. After electrophoresis, the gels were stained with 2 μg/ml ethidium bromide for 20 min and destained with milliQ water for 10 min. Then the tRNA bands were visualized under UV light to check for equal sample loading. Subsequently, the gels were electrotransferred to positively charged nylon membranes (Boehringer Mannheim, GmbH, Mannheim, Germany). The RNA was crosslinked to the membrane under ultraviolet light at 254 nm wavelength (1200 μJ × 100). Labeled LNA oligonucleotides that are complementary to the antisense strand of the siRNAs pol and nef were used as probes. Probes were 5′ end labeled with the kinaseMax kit (Ambion) in the presence of 1 μl of [γ-]PATP (0.37 MBq/μl, Amersham Biosciences) and purified using the Sephadex G-25 spin columns (Amersham Biosciences) according to manufacturer's protocol. We used LNA oligonucleotides 5′ AAGAGAGTGTAAG 3′ (pol probe) and 5′ GGATGAAGGTATT 3′ (nef probe). Underlined are the positions that contain locked nucleic acids. Hybridizations were performed at 42°C in 10 ml ULTRAhyb hybridization buffer (Ambion) in the presence of the labeled probe, as suggested by the manufacturer. After hybridization, the membranes were washed twice for 5 min at 42°C in 2 × SSC/0.1% SDS and signals were detected by autoradiography using a phosphorimager (Amersham Biosciences). To determine the induction of the IFN system by the e-shRNA constructs, 293T cells were transfected with 25 ng of e-shRNAs using Lipofectamine 2000 reagent (Invitrogen). Transfection of 2 μg transcribed dsRNA was used as a positive control for IFN-β induction as described previously (). Total RNA was isolated from cells 24 h post-transfection using RNeasy mini kit (Invitrogen) according to the manufacturer's protocol. Genomic DNA was removed by DNase treatment using the TURBO DNA-free™ kit (Ambion). First strand cDNA was reverse transcribed using 1 µg of total RNA, Thermoscript™ reverse transcriptase (Invitrogen) and random hexamer primers (Invitrogen). PCR amplification was performed on 2 µl RT product with IFN-β, OAS, MxA, ISG56 and beta actin specific primers. The following primer combinations were used: IFN-β, f: 5′-GATTCATCTAGCACTGGCTGG-3′; r: 5′-CTTCAGGTAATGCAGAATCC-3′ (186 bp product) (), OAS, f: 5′-TCAGAAGAGAAGCCAACGTGA-3′; r: 5′-CGGAGACAGCGAGGGTAAAT-3′ (399 bp product) (), MxA, f: 5′-AGTATGGTGTCGACATACCGGA-3′; r: 5′-GAGTCTGGTAAACAGCCGAATG-3′ (145 bp product) (), ISG56, f: 5′-CTTGAGCCTCCTTGGGTTCG-3′; r: 5′-GCTGATATCTGGGTGCCTAAGG-3′ (137 bp product) () and β-actin f: 5′-GACTACCTCATGAAGATCCTCAC-3′; r: 5′-ATTGCCAATGGTGATGACCTG-3′ (197 bp product). For PCR amplification, the Reddymix™ Master Mix (Abgene) was used in a 50 μl reaction. The PCR program was as follows: 95°C for 3 min, 30 cycles of 30 s at 95°C, 30 s at 57°C, 45 s at 72°C and a final extension for 8 min at 72°C. In an initial attempt to learn the rules for constructing multiple siRNAs in a single, extended hairpin, we tested the importance of the siRNA position within the hairpin stem. We made a first set of three 39 bp hairpin RNA constructs that are expressed from the H1 polymerase III promoter. These RNA hairpins have identical 5 nt loop sequences and top 2 bp of the pSUPER system () and end with a 2 nt UU overhang (A). We positioned a highly effective anti-gag siRNA in the base (39B), center (39C) or top (39T) of the hairpin stem and compared these with the original 21 bp shGag inhibitor (A). The guide/antisense strand of the gag siRNA is boxed in A. This 19 nt sequence is flanked in the 39 bp hairpins by contiguous antisense HIV-1 gag sequences. To determine the inhibitory activities of these hairpin RNAs, two luciferase reporter constructs were generated. One reporter contains a broad 50 nt gag target sequence that should reveal the activity of any siRNA processed from the extended hairpins (B, left). The other reporter contains the minimal 19 nt gag target sequence that only scores the activity of the precise siRNA against gag (B, right). We co-transfected 293T cells with the luciferase reporter constructs and different amounts of the hairpin RNA constructs. Plasmids encoding renilla luciferase were included to correct for transfection variation and to monitor for cell viability that may be affected by off-target effects of the encoded siRNAs. Firefly and renilla luciferase expression was measured 48 h post-transfection and the ratio was used as an indicator for target inhibition. Luciferase expression in the absence of inhibitor was set at 100%. Hairpins 39B and 39T were equally active on the 50 nt gag reporter, but hairpin 39C showed significantly less inhibition (C, left). All extended hairpin RNAs clearly showed less inhibition than shGag. When these constructs were tested on the luciferase reporter with the exact 19 nt gag target, we observed the same ranking order of activity, but the differences are more pronounced (C, right). Construct 39B is most efficient, 39T is marginally active and 39C did not show any inhibitory effect. These findings suggest that processing of the 39 bp hairpins starts at the base of the hairpin RNA transcript, as suggested by the marginal inhibitory activity of 39T on the 19 nt gag reporter. 39T does have some activity on the broad 50 nt gag reporter because it also detects the siRNAs produced from the base of 39T. In case of 39C, the centrally located 19 nt siRNA is probably destroyed because Dicer cleavage will occur near the center of the extended hairpin. Again, some 39C activity is recovered on the 50 nt gag reporter, which likely represents siRNAs processed from the base of the hairpin. To study whether efficient inhibition of two genes can be obtained by a single e-shRNA, we designed extended hairpin constructs encoding two highly effective siRNAs: one against HIV-1 pol gene () and the other against the nef gene (). Two versions were made: pol-nef and nef-pol e-shRNAs (A). We extended the original short hairpin constructs and placed a second siRNA on top of the first siRNA, resulting in a hairpin of 40 bp (A). Our previous findings suggest that processing of 40 bp hairpins by Dicer will start at the base of the hairpin. However, the exact cleavage site of Dicer is unknown, therefore we made four additional constructs with increasing length (41–44 bp) by insertion of 1–4 bp linkers between the two siRNAs (A). The linker sequences were obtained by extending the sense strand of the first siRNA with HIV-1 derived sequences. This design allowed us to study the silencing activities of hairpins with increasing stem length. To evaluate the suppressive activities of these hairpin RNA constructs, we co-transfected 293T cells with the inhibitors and luciferase reporter constructs containing either the 50 nt pol target (Luc-pol) or the 50 nt nef target (Luc-nef) (B). The renilla luciferase expression plasmid was included to control for transfection variation. Firefly and renilla luciferase expression levels were measured 48 h post-transfection and the ratio in the presence of pBS was set at 100%. The irrelevant hairpin RNA construct 39C (against HIV-1 gag) was used as a negative control. The highest inhibitory activity was observed for the siRNAs present at the base of the hairpin. All pol-nef hairpins inhibit the Luc-pol reporter (C, upper left) and all nef-pol hairpins inhibit the Luc-nef reporter efficiently (C, lower right). The activity of the e-shRNAs is comparable to that of the original shRNAs, although the pol-nef constructs are slightly less active than shPol (C, upper left). Poor activity was scored for the top siRNAs in the original 40 bp context, that is nef in the pol-nef construct (C, upper right) and pol in the nef-pol construct (C, lower left). However, addition of spacer residues improved the activity of these top siRNAs. The nef siRNA from the pol-nef 43 construct showed optimal activity that approaches the activity of shNef. Further size increases (44 bp) do not significantly enhance the inhibitory potential. The pol siRNA from the nef-pol constructs also showed optimal activity when the hairpin stem length is 43 bp. Importantly, increased activity of the top siRNA is not at the expense of the siRNA at the base of the hairpin. Thus, a hairpin stem of at least 43 bp is required to produce two functional siRNAs from a single hairpin RNA. Of note, renilla luciferase expression of the transfection control was not affected by the extended shRNAs, indicating that the extended hairpins do not cause cytotoxicity in 293T cells. To address the hairpin length requirement from another perspective, we made two additional mutants in the nef-pol 44 context. We removed either one or two nucleotides (A, mutant 44-1 and 44-2, respectively) from the passenger/sense pol strand, thus effectively reducing the number of basepairs by one or two in the hairpin stem without affecting the guide/antisense strand. Anti-gag construct 39C was used as a negative control and we set the luciferase expression in the presence of pBS at 100%. All three constructs showed efficient silencing of the nef reporter by the siRNAs at the base of the hairpins (data not shown). However, Luc-pol silencing is sensitive to the hairpin length. Removal of one nucleotide marginally reduced Luc-pol silencing activity and removal of two nucleotides showed a significant reduction (B). These findings confirm the results obtained with the original constructs in that the hairpin stem length is critical for e-shRNA activity. To rule out that the silencing activity observed for the extended shRNAs is due to aspecific effects, e.g. IFN induction by the extended hairpins, we made additional nef-pol 44 variants. If the observed inhibition is due to RNAi, it should be sequence-specific. We made three mutants containing one or two nucleotides deletions in the guide/antisense pol strand of nef-pol 44 (C, mutant 44a, 44b and 44c). Anti-gag construct 39C was again used as a negative control and the luciferase expression with pBS was set at 100%. We observed a gradual reduction in the ability of the deletion mutants to silence Luc-pol expression (D), but no effect on Luc-nef silencing was observed (data not shown). These combined results indicate that the e-shRNA produces two functional siRNAs that induce RNAi in a sequence-specific manner. We next asked whether the differences in silencing activity of the hairpin RNA constructs are due to differences in RNA expression, stability or processing into functional siRNAs. The latter possibility seems most likely as the siRNA at the base of the hairpin is always active, unlike the top siRNA. To study this, we transfected 293T cells with the 40–44 variants of the pol-nef and nef-pol hairpin constructs and examined siRNA production by northern blotting. To ensure equal loading of the purified small RNA samples on the polyacrylamide gel, we stained the gel with ethidium bromide to visualize tRNA species and other small RNA molecules. For detection of siRNAs, we used 5′ end-labeled 19 nt LNA probes complementary to the antisense strand of the siRNAs nef and pol (A). The two individual shRNA constructs were used as positive control, showing the mature, fully processed siRNAs. In pol-nef e-shRNA transfected cells, we readily detected pol siRNAs of ∼22 nt that are derived from the base of the hairpin (40–44), but nef siRNA processing from the top of the hairpin increases with spacer length (B). A similar picture was observed for the nef-pol constructs (C). Here, the nef probe detects an approximately equal amount of the siRNA at the base of the hairpin, but significantly more of the top siRNA is processed from constructs 42, and especially 43 and 44. These results correlate with the luciferase data in which we observed stable activity of the siRNA at the base of the hairpin, but increased activity of the top siRNA when a 43 bp stem length is reached (C). Thus, the increased activity is due to more efficient processing of the top siRNA. Precursor hairpin RNAs were not detected in these experiments, suggesting that the expressed hairpins are efficiently processed by the cellular RNAi machinery. Interestingly, we did not detect the top part of the hairpin of constructs 40 and 41 that are not efficiently processed into siRNAs. These results indicate that this part of the e-shRNA is rapidly degraded because it is not processed and taken up by the RNAi machinery. Next, we tested whether these e-shRNAs are indeed efficient inhibitors of HIV-1. The target sites for the encoded pol and nef siRNAs are indicated in the HIV-1 genome (A). To quantify the antiviral effects of these hairpins, we co-transfected the HIV-1 molecular clone pLAI with an increasing amount of the pol-nef and nef-pol construct series (40–44) in 293T cells. HIV-1 production was measured by determining CA-p24 levels in the culture supernatant 2 days post-transfection. CA-p24 levels were normalized to the renilla luciferase activities of the co-transfected control plasmid. Virus production in the presence of the empty vector was set at 100%. We observed optimal inhibition of HIV-1 production by the pol-nef 42 and 43 and nef-pol 43 constructs (B). These results correlate with the luciferase inhibition data in C and the processing results in B and C. As an additional control for the sequence-specificity of the antiviral e-shRNAs, we constructed two control hairpin constructs of 43 bp. One targets two sites within the luciferase reporter gene (hLuc) and the other contains the scrambled sequence of the 43 e-shRNA variant (hSCR). Neither hLuc nor hSCR inhibited HIV-1 production, indicating that virus suppression by the e-shRNAs is fully sequence-specific (C). There are several technical problems associated with the construction of hairpin RNA expression constructs (). First, it is difficult to sequence constructs containing lengthy inverted repeats. Second, these constructs are unstable in , resulting in a high mutation rate in the hairpin region. We experienced similar problems with the e-shRNA constructs. We were able to sequence the constructs using an optimized protocol (98°C sample denaturation and addition of 1M Betaine) (). We attempted to modify the hairpins to allow easy cloning and sequencing, obviously without loss of their silencing activity. Therefore, we designed variants of the pol-nef 43 hairpin that are less stable, without affecting the guide/antisense strand of the hairpin RNA. We mutated or deleted the third nucleotide of the linker in the passenger/sense strand, which results in hairpins with a destabilizing mismatch in the central stem regions (A). Indeed, these modifications allow easier sequencing and improve the genetic stability of the DNA construct (data not shown). A regular sequencing protocol can now be used to sequence these constructs. Next, we tested the silencing activities of the destabilized hairpins compared to the original pol-nef 43 construct on Luc-pol and Luc-nef reporters. In general, we observed no significant loss of RNAi inhibition on the pol and nef reporter, although a slight reduction was observed for the deletion mutant, which creates a 1 nt internal loop (B). To test for HIV-1 inhibition, we co-transfected 293T cells with the pLAI molecular clone, the renilla expression plasmid and an increasing amount of the modified hairpin constructs. All mutant hairpins showed similar suppression compared to the parental pol-nef 43 construct, except for the mutant with the 1 nt deletion (C). These results indicate that a single nucleotide substitution, causing a mismatch in the center of the e-shRNA, is well-tolerated. Because the IFN response can be induced by long dsRNAs, we performed extensive RT–PCR analyses for markers of the IFN response on e-shRNA transfected 293T cells. The expression of the pol-nef e-shRNAs and the 43 bp A mutant (A) does not induce the expression of IFN-β, OAS, MxA and ISG56 mRNAs (). As a positive control, we used transcribed long dsRNAs of 300 bp. The expression of long dsRNA resulted in an IFN response in the transfected cells, showing that 293T cells are able to activate an innate IFN response. HIV-1 replication can be efficiently inhibited via induction of RNAi using shRNA expressing constructs. However, HIV-1 can easily escape from RNAi by introducing mutations or deletions in the target sequence. For effective inhibition of HIV-1 replication, multiple shRNAs should be used simultaneously. In addition, highly conserved sequences within the HIV-1 RNA genome should be targeted to further reduce the chance of escape. The use of e-shRNA may be a promising approach to target HIV-1 at multiple sites, but little is known about the design of effective lhRNAs. In this study, we attempted to explore some principles for the design of e-shRNAs that encode two active siRNAs. We first constructed extended hairpin RNAs of 39 bp and varied the position of a potent siRNA in the stem region. We observed that positioning of the siRNA at the base of the hairpin stem yields optimal RNAi activity. This observation is consistent with the proposed model for Dicer action, in which it counts approximately 22 nt from the 3′ end of the stem region to determine the position of cleavage (,). Consistent with this mechanism, Siolas () showed that synthetic 29-mer shRNAs with 3′ overhangs were processed and from the 3′ end into discrete products of 21 and 22 nt. Next, we combined two effective siRNAs to build extended hairpin RNAs of 40–44 bp. We demonstrate that a minimal length of 43 bp is needed to generate two effective siRNAs. In HIV-1 inhibition assays, we observed an optimal activity when the pol-nef hairpin is 42 or 43 bp. The pol-nef e-shRNA transcripts end with a U and the pol III promoter terminates after the second U, yielding a transcript with a 1 nt instead of a 2 nt 3′ overhang. This may be the cause for the 1 nt difference in length requirement in the HIV-1 inhibition assays. Indeed, hairpins with 1 nt or 2 nt overhangs have different Dicer cleavage sites , generating siRNA products of different sizes (). We showed that RNAi activity observed in luciferase assays correlates with the processing data. Importantly, by removal of nucleotides in the guide/antisense strand of the hairpin RNA, the RNAi activity of the e-shRNAs is significantly impaired, indicating that the inhibition is sequence-specific. We demonstrated that these e-shRNAs profoundly inhibit HIV-1 replication. Furthermore, we designed user-friendly e-shRNA variants by introducing destabilizing mutations to avoid cloning/sequencing problems due to the lengthy inverted repeat sequences, obviously without loss of inhibitory capacity. We found similar knockdown efficiencies for the modified hairpins compared to the original e-shRNA. Several studies have demonstrated that multiple siRNA species can be detected when a lhRNA is expressed in cells (,). However, these studies did not carefully examine the minimal length requirement of the hairpin to produce multiple siRNAs. Here, we show that the siRNA at the base of the hairpin is always active in an e-shRNA, but activation of the top siRNA requires a minimal stem length of 43 bp. Importantly, the e-shRNA constructs effectively inhibit two viral genes simultaneously, which should prevent or delay viral escape. Previous reports show that inclusion of multiple mutations within the sense strand, thus creating G-U basepairs in the hairpin stem, can improve the stability of the plasmid in (,). Interestingly, it has also been shown that introduction of G-U bp can avoid the interferon response (,) and improve the suppressive activity of the hairpin (). We also observed improved stability of a plasmid expressing a modified hairpin with only a single point mutation in the sense strand that destabilized the hairpin. Importantly, inclusion of this point mutation does not reduce the RNAi activity of these hairpins. We showed that expression of the e-shRNAs does not trigger the innate antiviral IFN response. Notably, cytotoxicity can also be induced by overexpression of shRNAs, which leads to competition for and oversaturation of endogenous cellular RNAi components (). This emphasizes the need to use minimal amounts of shRNA expressing vectors and carefully designed shRNAs. Therefore, we think it is more effective to precisely stack potent siRNAs (43 bp) instead of using an extended consecutive sequence. The latter design may produce moderately active siRNAs that may saturate the RNAi pathway. Based on our results we suggest that the use of e-shRNAs with a specific hairpin length to ensure proper processing may be beneficial in silencing RNA viruses in a multi-targeting approach. Furthermore, based on our data and that of others (,), we would advise to introduce multiple point mutations in the sense strand of the hairpin to avoid the interferon response, enhance the stability of the plasmid in and to facilitate sequencing. In conclusion, our results provide building principles for the design of multi-siRNA producing hairpins or e-shRNAs. We demonstrate that the hairpin stem length is critical for proper processing and optimal activity. In addition, expression of the e-shRNAs does not result in activation of the IFN response. Our e-shRNA constructs may be an initial step towards the design of further extended multi-siRNA transcripts. This study stresses the importance of a careful design of such hairpin RNA molecules. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Fragile X syndrome (FXS), the most common cause of inherited mental retardation with a prevalence of 1:4000 in males and 1:6000 in females, is caused by an expansion of a (CGG) sequence in the 5′ UTR of the gene to >200–2000 repeats (). Expansion of the (CGG) tract beyond ∼200 repeats entails hypermethylation of the repeat sequence and of a neighboring CpG island that results in the transcriptional silencing of and absence of the product FMRP protein (). Although carriers of premutation alleles that have >55–200 (CGG) repeats do not develop FXS, some do present various forms of clinical involvement such as minor alterations of physical features and emotional problems (). Notably, ∼20% of the female premutation carriers develop premature ovarian failure (POF) () and one-third of the carrier males are affected by fragile X-associated tremor-ataxia syndrome (FXTAS) (,). In contrast to the absence of transcripts in FXS cells, the levels of mRNA in peripheral blood leukocytes at the upper range of FXS (CGG) premutation were found to increase by 5- to 10-fold relative to normal subjects (,). However, the amounts of FMRP in these cells remain within the normal range or are somewhat below normal at higher premutation repeat sizes (,,). A progressively diminishing association of mRNA with polysomes that have increasing (CGG) repeat sizes was observed in lymphoblastoid cell lines of premutation carriers that over-express mRNA (). It appears, therefore, that expanded premutation (CGG) tracts in mRNA lower the efficiency of its translation . A clue to the underlying mechanism of the declining efficacy of translation was obtained in a study of the impeded translation of mRNA that contained longer (CGG) tracts. Although cells of a mildly affected FXS patient with alleles of 57–285 CGG repeats exhibited normal steady-state levels of mRNA, synthesis of FMRP from transcript with more than 200 repeats was markedly decreased and these transcripts were associated with stalled 40S ribosomal subunits (). A plausible interpretation is that the longer (CGG) sequences form secondary structures that block the migration of the 40S subunit. Thus, (CGG) premutation repeats may conceivably also fold into secondary structure(s) that slow down or obstruct ribosome progression, therefore lowering the efficiency of translation. DNA and RNA (CGG) repeat sequences readily fold into hairpin structures () that may pair to generate stable intermolecular tetraplexes (). Quadruplex (CGG) structures in DNA templates were reported to hinder the progression of DNA polymerases (,). In a likely analogy, (CGG) hairpin and quadruplex structures in mRNA may block polysome formation and impede ribosome progression and productive protein synthesis. If mRNA translation is obstructed by secondary structures of the moderately expanded premutation (CGG) tracts, it should be enhanced by their destabilization. Indeed, several proteins and a low molecular size porphyrin were shown to efficiently disrupt tetraplex (CGG) structures. The human Werner syndrome helicase was reported to catalyze the unwinding of intermolecular quadruplex structures of (CGG) tracts in DNA (,). Member proteins of the hnRNP family; CBF-A (,,), hnRNP A2 and modified hnRNP A1 () and UP-1 (,) were reported to mediate non-enzymatic destabilization of quadruplex forms of (CGG) in both DNA and RNA. Additionally, the quadruplex DNA interacting cationic porphyrin TMPyP4 was shown to disrupt tetrahelical (CGG) in DNA (). In this work, we report that upstream positioned mid-range premutation (CGG) tracts preferentially suppress the and translation of a reporter firefly luciferase gene. Demonstrating that (CGG) tract in RNA forms an intramolecular secondary structure which is likely to impede translation, we show that the tetraplex (CGG) destabilizing proteins hnRNP A2 and CBF-A enhance the translation of mRNA molecules that contain 62 or 99 (CGG) repeats. —pSP6-5′--UTR (CGG)-FL ( = 0, 30, 62, 99) plasmids were prepared by modifications of previously described constructs (). The no-repeat plasmid (high-copy origin) was digested with NaeI and BlpI, while the 30, 62 and 99 CGG-repeat plasmids (low-copy origin) were digested with BglII. The BglII overhang was blunted with mung bean nuclease and digested by BlpI. Thus regions containing the CMV promoter and the 5′-end of the 5′-UTR, up to the BlpI site were removed while creating blunt-end to BlpI-overhang linear vectors. Both high- and low-copy plasmids were ligated to a double-stranded oligomer, 5′-(GGCATTTAGGTGACACTATAGATCAGGCGC)-3′·3′-(CCGTAAATCCACTGTGATATCTAGTCCGCGAGT)-5′ that contained the SP6 promoter and restored the 5′-end of . The formed 5′-UTR sequence starting from transcription site 1 was as previously described () with the following modifications. The first two 5′ AG bases of the 5′-UTR were changed to GA for optimal SP6 transcription, the 3′-end base was C instead of G, and the no CGG-repeat 5′UTR was devoid of both the CGG repeat and the following GC dinucleotide. The (CGG) repeat sequence had two intervening (AGG) triplets at positions 11 and 21, the (CGG) repeat contained an (AGG) interruption at the 11th position and the (CGG) repeat had at least one AGG intervening trinucleotide of undetermined location. Next, bacteriophage T7 promoter was incorporated into the pSP6-5′--UTR(CGG)-FL plasmids. Double-stranded DNA adaptor including the T7 promoter sequence (underlined) 5′-d(TCACGCGTACTTGCTAGCCGC)-3′3′-d(CGCATGAACCGATCGGCGAGT)-5′ was ligated into the BlpI site upstream to the plasmid UTR-(CGG) sequence. Plasmids without a (CGG) tract or that included a (CGG) insert were propagated in 500 ml of LB medium at 37°C for 12 h in (gift of Dr D. Gidoni, Volcani Center, Israel) and purified (Maxi-prep kit, Qiagen). Due to the instability of the (CGG) and (CGG) repeats, plasmids harboring these inserts were grown in multiple 3-ml cultures. Following plasmid purification (Mini-prep kit, Qiagen), lengths of the BlpI and NheI excised inserts were determined by agarose gel electrophoresis and only plasmids that contained validated (CGG) or (CGG) tracts were used. —pCMV--5′-UTR(CGG)-FL plasmid () was contributed by Dr P. Hagerman, UC Davies. To construct plasmids with 0, 30 or 62 repeats, the -5′-UTR(CGG) sequence was excised by BlpI and NheI digestion and the linearized plasmid was ligated with (CGG), (CGG) or (CGG)-containg -5′-UTR sequences that were excised from the respective pT7--5′-UTR(CGG)-FL plasmids by BlpI and NheI digestion. Plasmid propagation and verification of correct repeat sizes were performed as described for the pT7--5′-UTR(CGG)-FL plasmids. —The 5′ untranslated region was amplified from genomic DNA of EBV transformed B lymphoblastoid cell line of a normal subject (gift of Dr P. Chiurazzi, Catholic University, Italy). The UTR tract which included 33 (CGG) repeats interrupted by two (AGG) triplet units at trinucleotides 11 and 21 was PCR amplified using a 5′- exon forward primer 5′-(CGCGGATCCCGCTCAGCTCCGTTTCGGTTTC)-3′ and 3′-end exon primer 5′-(CCGGAATTCTAGAAAGCGCCATTGGAGCCCC)-3′. The amplified fragment was restricted by BamHI and EcoRI digestion and cloned into SP6 promoter containing pCS107 plasmids that were propagated in and purified (Maxi-prep, Qiagen). —Plasmid was constructed to harbor and express cDNA encoding the quadruplex (CGG)-disrupting protein hnRNP A2 or its mutant hnRNP A2 FS that lost the destabilizing activity (). The plasmids pGEX-2T- (gift of Dr Ralph C. Nichols, Dartmouth School of Medicine, NH) or pGEX-2T- () were cut by EcoRI and SalI and the electrophoretically isolated cDNA were ligated into EcoRI and SalI-linearized pCMV2-Flag plasmid (Stratagene). To construct pCMV2-Flag plasmids that express the quadruplex (CGG) destabilizing CBF-A protein or its mutant CBF-A ΔRNP1 that lost the tetraplex disrupting activity (), pCMV2-Flag was cut by EcoRI and dephosphorylated with calf intestinal phosphatase (New England Biolabs). CBF-A or CBF-A ΔRNP1 encoding sequences that were cleaved by the same enzyme out of plasmids () were ligated to the linearized pCMV2-Flag plasmids. The restriction digestion at the border of the Flag and the CBF-A or CBF-A ΔRNP1 encoding sequences caused loss of the reading frame. A proper reading frame was restored by PFU (Promega) catalyzed PCR amplification of the altered DNA region using a 5′ primer 5′-d(CTTGCGGCCGCGAATCCATGGCCGACC)-3′ and a 3′ primer 5′-d(GGTCGGCCATGGATTCGCGGCCGCAAG)-3′. The modified plasmids were cloned and restoration of the reading frame was verified by DNA sequencing. Coupled transcription and translation reactions were carried out using the TNT® transcriptiontranslation system (Promega). Reaction conditions were essentially according to the manufacturer's specifications. Briefly, reaction mixtures contained in a final volume of 25 μl: 1 μl of 25× TNT reaction buffer; 0.5 μg XbaI-linearized pT7--5′-UTR(CGG)-FL plasmid; 12.5 μl 2× TNT rabbit reticulocyte lysate; 0.5 μl TNT T7 RNA polymerase; 5.0 μCi [α-P] UTP (3000 Ci/mmol, Amersham); 20 μM of each amino acid; 40 U RNasin ribonuclease inhibitor (Promega). Following incubation at 30°C for 30 min, the samples were rapidly cooled to 4°C and 5 μl aliquots of each reaction mixture were mixed with 100 μl of luciferase assay Reagent II (Promega). Following a 3 s delay, firefly luciferase activity was determined for 30 s (20/20 luminometer, Turner Biosystems). To measure levels of the radiolabeled RNA, gel-loading dye containing SDS to a final concentration of 0.5% was added to 10 μl aliquots of the reaction mixtures and the samples were electrophoresed through 0.8% agarose gel. The identity of the radioactive RNA bands as full-length luciferase gene transcripts was verified by northern analysis using the luciferase 3′ gene probe 5′-(CTTTCCGCCCTTGGCCTTTATGAGGATC)-3′. The relative levels of the labeled luciferase RNA molecules that contained (CGG) tracts of different lengths were determined by phosphor imager analysis. To investigate the effect of different lengths of (CGG) tracts on transcription, pT7--5′-UTR(CGG)-FL plasmids ( = 0, 30, 62 or 99) were linearized by XbaI restriction digestion and transcribed by T7 RNA polymerase. Reaction mixtures contained in a final volume of 20 μl: 0.5 μg linear plasmid DNA; 1 mM each of the four rNTPs; 3.0 μCi [α-P] UTP (3000 Ci/mmol, Amersham); 40 U of ribonuclease inhibitor (Takara); 80 mM potassium acetate; 40 U of T7 RNA polymerase and T7 polymerase reaction buffer (Promega). The reaction mixtures were incubated at 30°C for 30 min and the RNA polymerization reaction was terminated by the addition of 5 μl of 2.5% SDS in gel loading dye. Following resolution of the radiolabeled RNA by electrophoresis in 0.8% agarose, the gels were dried and radioactivity of the product RNA molecules was quantified by phosphor imaging analysis. That the radioactive RNA bands represented full-length transcripts was confirmed by northern analysis as described for the coupled transcription–translation system. Equal amounts of radiolabeled RNA transcripts of the pT7--5′-UTR(CGG)-FL plasmids were translated . To prepare the RNA, 1 μg of each XbaI-linearized plasmid was transcribed at 37°C for 1.5 h in a final volume of 20 μl of Ampliscribe T7 transcription system (Epicenter Technologies) that contained 3.0 μCi [α-P] UTP (3000 Ci/mmol, Amersham). Plasmid DNA was removed by digestion at 37°C for 15 min with 1 U of RNase-free DNase (Promega) and the RNA transcripts were extracted with ultra pure phenol:chloroform:isoamyl alcohol (25:24:1, Invitrogen) followed by chloroform extraction. The RNA was precipitated and washed with 70% ethanol and dissolved in RNase-free water. Approximately equal amounts of RNA as estimated by absorption at 260 mμ were resolved by agarose gel electrophoresis and the precise amounts of RNA were normalized by the radioactivity of the full-length RNA transcript bands. Equal amounts (0.5 μg each) of the -5′-UTR(CGG)-FL transcripts were translated in rabbit reticulocyte lysate (Flexi® translation system, Promega) as follows. The RNA samples were incubated at 30°C for 30 min in a final volume of 25 μl of reticulocyte lysate translation system that contained 2.0 mM DTT; 1 mM magnesium acetate; 70 mM KCl and 20 U ribonuclease inhibitor (Takara). The translation reaction was terminated by rapid cooling of the samples to 4°C and the luciferase activity was determined in 10 μl aliquots of each reaction mixture as described for the coupled transcription–translation system. One microgram of XhoI-linearized pCS107--5′-UTR(CGG) plasmid DNA was transcribed in AmpliScribe SP6 system (Epicenter Biotechnologies) according to the manufacturer's instructions. The RNA transcript was precipitated and washed with 70% ethanol and resolved by denaturing gel electrophoresis through 10% polyacrylamide, 8 M urea. The gel band that contained the RNA transcript was visualized by UV adsorption, excised and agitated overnight at 4°C in a solution of 0.3 M sodium acetate pH 5.2; 0.5 mM EDTA; 0.1% SDS. Following ethanol precipitation and wash, the eluted RNA transcripts were dephosphorylated at 37°C for 1 h by shrimp alkaline phosphatase (USB). The phosphatase was heat-inactivated at 65°C for 15 min and the RNA was 5′-end labeled using T4 polynucleotide kinase and [γ-P] ATP (). The labeled RNA was resolved by denaturing gel electrophoresis through 10% polyacrylamide, 8 M urea and recovered and isolated by ethanol precipitation as described above. Duplex regions of RNA were distinguished from unpaired tracts by the resistance of the former to digestion by the single-stranded RNA-specific nuclease RNase T1. Labeled RNA transcripts were dissolved in RNase-free water and supplemented with unlabeled yeast RNA carrier to a final concentration of 8 μM. Digestion of native RNA with a specified amount of RNase T1 (Ambion) was conducted at 20°C for 20 min in a reaction mixture that contained in 10 μl final volume 10 mM MgCl and 80 mM potassium acetate in 10 mM Tris-HCl buffer, pH 7.4. Digestion of heat-denatured RNA by RNase T1 was conducted at 55°C for 15 min in the same reaction mixture that also contained 3.5 M urea. The reactions were terminated by adding to each mixture 20 μl of inactivation-precipitation buffer as specified by the manufacturer. Products of the ribonuclease digestion reactions were resolved by denaturing gel electrophoresis in 12% polyacrylamide, 8 M urea. -5′-UTR(CGG) RNA was transcribed and internally labeled in reaction mixtures that contained in a final volume of 20 μl: 1.0 μg XhoI-linearized pCS107--5′-UTR(CGG) plasmid DNA, 10 mM DTT, 40 U RNasin (Promega), 0.5 mM ATP, 3.0 mM CTP, 6.0 mM GTP or 0.5 mM 7-deazaGTP (TriLink Biotechnologies), 8.3 μM UTP, 50 μCi [α-P] UTP (800 mCi/mmol, Amersham), 20 U SP6 polymerase and 4 μl 5× SP6 transcription buffer (Promega). Transcription was conducted at 37°C for 60 min, the DNA was digested by 1 U DNase I at the same temperature for 15 min and the product RNA was phenol extracted and precipitated and washed with ethanol. An intramolecular secondary structure of the RNA was generated by incubating at 4°C for 15 min, mixtures that contained in a final volume of 10 μl: <2.5 nM 5′-P -5′-UTR(CGG) RNA, 120 mM KCl, 1 mM EDTA, 0.5 mM DTT, 20% glycerol in 20 mM Tris-HCl buffer, pH 7.8. Control reactions were performed under identical conditions except that KCl was omitted from the reaction mixture. The RNA was resolved by electrophoresis in non-denaturing 4% polyacrylamide gel in 0.5× TBE buffer, pH 8.2 that contained 10 mM KCl. Human Embryonic Kidney 293 cells (HEK293) were seeded in 0.1% gelatin-coated 10 cm plates and grown to 80–90% confluence at 37°C in 5% CO atmosphere in Dulbecco Modified Eagle's Medium (DMEM) supplemented with 4.5 g/l -glucose, 5 mM -glutamine, 10% fetal calf serum, 83.3 U/ml each Penicillin and Streptomycin and 0.2 μg/ml Amphotericin B (Biological Industries, Israel). Cells that were reseeded at 3.3 × 10 cells per 6 cm gelatin-coated plate, were grown overnight and transiently co-transfected with three plasmids: a vector harboring reporter firefly luciferase (FL) without or with an upstream (CGG) repeat tract; plasmid that encoded normalizing Renilla luciferase (RL) and plasmids expressing quadruplex destabilizing proteins or their inactive mutants. The growth medium was replaced with 4 ml of fresh DMEM to which 100 μl of DMEM were added that contained 6 μl Fugene 6.0 (Roche), 0.2 μg pCMV--5′-UTR(CGG)-FL DNA and 0.02 μg pCMV-RL plasmid (Promega) and 1.8 μg of DNA of one of the following expression plasmids: pCMV2-Flag; pCMV2-Flag- pCMV2-Flag-; pCMV2-Flag or pCMV2-Flag-. The cells were harvested 24 h after transfection, washed once with 1.5 ml cold phosphate buffered saline (PBS) and resuspended in 3 ml PBS. Aliquots of each sample were used to determine FL and RL activities and to conduct semi-quantitative RT-PCR measurements of the levels of their mRNA transcripts. Correction for variations in cell viability and transfection efficiency was performed for each experiment as described () except that the RL-corrected FL activity was normalized to FL activity of cells transfected with reporter pCMV--5′-UTR-FL plasmid with no upstream (CGG) repeat tract. FL and RL activities were determined in lysates of transfected HEK293 cells according to the manufacturer's instructions using the dual luciferase reporter assay system, (Promega). Briefly, the cells were lysed in passive lysis buffer (Promega) and 20 μl of the cell lysate were added to 100 μl luciferase reagent II. Following a 3 s delay, FL activity was measured for 30 s using the 20/20 luminometer. The reaction was terminated by adding 100 μl Stop and Glo reagent to quench the FL activity and after a 3 s delay, the activity of RL was determined for 30 s. In each sample, triply measured values of FL activity were normalized to RL activity. RNA was isolated from HEK293 cells using the total RNA isolation kit (Cartagen, USA) and genomic and plasmid DNA were removed by use of the Turbo DNA-free kit (Ambion). Reverse transcription and amplification reactions were conducted with the Reverse-it one-step RT-PCR kit (ABgene, UK). Each reaction mixture contained in a final volume of 25 μl: 0.05 μg total RNA; 0.3 or 0.9 μCi [α-P] dCTP (3000 Ci/mmol) for the reverse transcription of FL or RL mRNA, respectively, and 5.0 pmol each of FL forward primer 5′-(CTATGAAGAGATACGCCCTGGTTCCTGG)-3′ and FL reverse primer 5′-(GGCAGTTCTATGAGGCAGAGCGAC)-3′ or RL forward primer 5′-(GGGATGAATGGCCTGATATTGAAGAAG)-3′ and RL reverse primer 5′-(CAATTTGTACAACGTCAGGTTTACCACC)-3′. FL or RL RNA were amplified by RT-PCR for 26 or 29 cycles, respectively, of 94°C for 20 s, 53°C for 25 s and 72°C for 1 min. Based on measurements of the levels of radiolabeled FL and RL reverse transcripts for each cycle between the 20th and 35th cycles, it was determined that 26th or 29th cycles were at the middle of the respective linear range of amplification of FL and RL RNA. Every set of reactions included a negative control of a mixture without RT to verify that all the amplification products were copies of RNA template and not of contaminating DNA. Equal aliquot volumes of the FL or RL RT-PCR P-labeled DNA products were electrophoresed at 12 V/cm through 4 or 6% non-denaturing polyacrylamide gels in 0.5× TBE buffer (2.4 mM EDTA in 1.08 mM Tris-borate buffer, pH 8.3) until a bromophenol blue marker reached the end of the gel. DNA size markers (T4 PNK end-labeled 5′-P peqGOLD 100 bp DNA ladder, PeqLab Biotechnologies) were used to identify full-length FL and RL RT-PCR DNA product bands. The identity of these DNA products was verified by parallel electrophoresis of radiolabeled PCR products of the original plasmids DNA with the FL or RL primers. Amounts of the electrophoretically resolved DNA samples were quantified in the dried gels by phosphor imaging analysis. Equal amounts of HEK293 cell lysates protein were resolved by 10% SDS-PAGE. Following transfer to nitrocellulose membrane, the expression of functional or mutant Flag-hnRNP A2 or FLAG CBF-A proteins was detected by use of murine anti-Flag monoclonal antibody (1:1000, Sigma) as the primary antibody and horseradish peroxidase-conjugated goat anti mouse IgG (H + L, 1:10 000, Pierce) as secondary antibody. Horseradish peroxidase activity was detected by use of the Super Signal Wes Pico chemiluminescence substrate (Pierce). We first evaluated in model T7 promoter driven coupled and separate transcription and translation systems, the effects of increasing sizes of normal and premutation-range (CGG) tracts on FL gene expression. DNA and RNA (CGG) repeat sequences were shown to readily form hairpin (,,,,,), tetraplex () and slipped-strand () secondary structures. A likely cause for the observed declining efficiency of translation as a function of the size of the upstream (CGG) mRNA sequence could be the formation of secondary structures by the RNA repeat sequence. To probe the structure of a (CGG) tract in RNA, denatured or native 5′-P labeled RNA transcripts of pCS107--5′-UTR(CGG) DNA were digested by RNase T1 and products of the nucleolytic digestion were resolved by denaturing polyacrylamide gel electrophoresis (see Materials and Methods section). RNase T1 cleaves phosphodiester bonds in single-stranded RNA 3′ to unpaired guanine residues whereas paired guanines resist digestion. As seen in , the -5′-UTR(CGG) denatured RNA transcript in lane 1 was cleaved by RNase T1 at each and every guanine residue. By contrast, as indicated in lanes 2 and 3, the enzyme did not digest portions of the native RNA GC-rich 5′-UTR sequence and particularly of the (CGG) repeat. In fact, nearly all the (CGG) stretch in native RNA-resisted digestion by RNase T1 except for cleavage at the tenth and twentieth (CGG) trinucleotides that, respectively, precede the (AGG) and (AGG) intervening triplets and a faint digestion band at the fifteenth (CGG) trinucleotide. Thus, except for selected unpaired (CGG) trinucleotides, the majority of the guanine residues were engaged in base pairing. These data indicated, therefore, that most of the repeat sequence folded into a secondary structure. Notably, a large portion of the 5′ UTR sequence also resisted digestion by RNase T1 indicating that it too was largely in secondary structure. We next sought to directly demonstrate the ability of the RNA (CGG) tract to form an intramolecular secondary structure. Aliquots of <2.5 nM 5′-P labeled -5′-UTR(CGG) RNA were incubated at 4°C for 15 min without or in the presence of 120 mM KCl and resolved by electrophoresis in non-denaturing polyacrylamide gels that were, respectively, devoid of salt or contained 10 mM KCl. Representative electrophoregrams shown in indicated that the 365-bases-long single-stranded (CGG) RNA migrated in the absence of salt between DNA size markers of 200 and 250 bp. By contrast, in the presence of KCl, the RNA migrated between DNA markers of 100 and 200 bp. Thus, the (CGG) RNA sequence formed in the presence of K ions a rapidly migrating compact structure. The zero-order kinetics of the formation of this species was consonant with it being an intramolecular secondary structure that may represent either a hairpin or quadruplex formation (see Discussion section). Being positioned at the 5′ end of mRNA, such structure could well retard its translation. Whereas transcription is completely silenced in fragile X full mutation cells (), premutation carriers produce 5- to 10-fold higher amounts of mRNA than normal cells (,). Yet, despite the elevated levels of mRNA in carrier cells, they produce FMRP at amounts that are within or below the norm (,,). It thus appears that translation is repressed by the moderately expanded premutation (CGG) repeat sequence. Indeed, evidence was presented to show that premutation (CGG) repeat sizes are linked to lowered association of mRNA with lymphoblastoid cell polysomes () and that the 40S ribosomal subunits stall along mRNA molecules that contain more than 200 repeats (). Data indicated that (CGG) tracts in RNA fold into hairpin (,,,) and tetraplex () structures. A reasonable conjecture, therefore, is that the diminished efficacy of the translation of premutation mRNA is due to retarded ribosome association and stalled progression at secondary structures of the (CGG) tract in premutation mRNA. In this work, we showed that premutation-range (CGG) tracts selectively reduced the efficiency of mRNA translation both and . Second, following a demonstration that (CGG) sequence in RNA was capable of forming quadruplex structures, we showed that the expression of proteins that untangle tetraplex structures of (CGG) increased the efficacy of mRNA utilization. Relative to FL reporter gene with no (CGG) upstream tract, the introduction of 30 (CGG) repeats upstream to the gene stimulated its transcription and translation in coupled T7 promoter-driven transcription–translation system () and in promoter-driven transcription and translation (). The most frequent normal number of (CGG) repeats in the gene and its transcript in normal subjects is 29–33 (). expression of the FL gene () suggested that secondary structures of the (CGG) tracts may promote maximum efficacy of translation mRNA in cells. Such requirement may exert positive selective pressure that results in the preservation of the (CGG) repeat sequence in the human population. As our results showed, 62 or 99 (CGG) repeats that represent the lower range of premutation size slightly repressed () or did not affect () the transcription of the FL gene. Most significantly, both 62 and 99 (CGG) repeats did sharply suppress FL translation both in coupled or in separate systems ( and ). The observed declining ratio of FL protein to mRNA indicated that the premutation repeat tracts reduced the efficiency of utilization of the FL mRNA. Results presented in and demonstrated that the (CGG) RNA repeat maintained a secondary structure and that a large portion of the guanine-rich 5′ UTR sequence itself also folded into secondary configuration (). It is plausible that the nature and geometry of secondary structures that (CGG) RNA tracts formed independently or together with the 5′ UTR tract differed for different sizes of the repeat sequence. Thus, the spatial arrangement of secondary structures of the normal 30 repeats may have promoted the association of mRNA with polysomes and enhanced ribosome progression whereas the different geometries of secondary structures of premutation sized 62 or 99 repeats impeded the translation machinery and limited the efficacy of mRNA utilization. Formation and stabilization of quadruplex nucleic acids require the presence of alkali ion. Thus, the potassium ion dependence of the formation of intramolecular complex by the (CGG) RNA sequence () was consistent with tetrahelical nature of this secondary structure. In contrast to tetraplex formation, hairpins can be generated in the absence of ions (). However, the possibility remained that under our experimental conditions K ions were critical for the stability of hairpins. Hence, although the data presented in were consonant with a tetraplex structure of the (CGG) RNA sequence, the possibility that the observed secondary structure was a hairpin could not be ruled out. Transcription of promoter-driven FL reporter gene and translation of the transcripts in HEK293 cells respond differently to the presence of upstream premutation (CGG) tracts than do T7 promoter-driven transcription and translation. Yet in both the and systems, premutation size repeat tracts ultimately lowered the efficacy of mRNA utilization. Unlike the essentially unaltered extent of transcription ( and ), the amount of FL (CGG) mRNA relative to FL (CGG) mRNA increased 2-fold, FL (CGG) mRNA remained unchanged and FL (CGG) mRNA was elevated by nearly 6-fold (). It is plausible that transcription efficiency was differentially affected by distinct (CGG) configurations in DNA such that different lengths of the repeat sequence formed secondary structures geometries that were better targets for the transcription factors than others. The modulation of transcription by different lengths of the (CGG) repeat in DNA could also reflect different degree of the chromatinization of the different lengths of the repeat tracts and dissimilar nucleosome positioning along these sequences. The observed higher amounts of transcripts that were produced in the presence of 99 (CGG) triplets was in line with the reported increased accumulation of mRNA in leukocytes of premutation carriers (,). The contrasting unchanged or reduced transcript accumulation was likely to be due to the use of T7 in place of the promoter and to the dissimilar reaction conditions of the and transcription systems. The (CGG) tracts had also opposing effect on translation and . Whereas the relative extent of translation of FL mRNA molecules that contained 62 or 99 (CGG) repeats was sharply diminished relative to 30 repeats ( and ), it decreased in HEK293 cells for 62 repeats but was somewhat increased at 99 repeats ( and ). The observed elevated level of protein in the transfected cultured cells was inconsistent with the reported unchanged or slightly reduced amounts of FMRP in cells of premutation carriers (,,). Although the reason for this discrepancy is not clear, cells of premutation carriers (), transfected HEK293 cells ( and ) and translation systems ( and ) all displayed reduced translation efficiency of premutation mRNA. A likely common denominator for all the systems may, therefore, be the impediment to the translation machinery by tetraplex structures of the intermediately expanded (CGG) tract. Results summarized in and and indicated that the RNA binding and tetraplex DNA and RNA-disrupting proteins hnRNP A2 and CBF-A increased the efficacy of (CGG) mRNA translation . Based on our observation that (CGG) RNA formed intramolecular secondary structure ( and ), we assumed that a stable (CGG) unimolecular folded domain in FL mRNA was responsible for its retarded translation (B, C and 2B) and (). In such a case, untangling of the secondary structure should increase the efficiency of translation. We demonstrated in the past that intermolecular tetraplex structures of (CGG) in DNA were disrupted by members of the hnRNP family CBF-A and hnRNP A2 (,,). We also showed that mutations in each of the two conserved domains of both proteins, the RNP1 box and the ATP/GTP-binding fold, specifically abolished their quadruplex disruption activity (). Results described here showed that expression of functional hnRNP A2 in HEK293 cells, but not of its inactive hnRNP A2 FS mutant, alleviated the repression of FL mRNA translation by a (CGG) tract (C and ). Also, active CBF-A, but not its ΔRNP1 mutant, increased the efficiency of the translation of (CGG) FL mRNA (C and ). Although the hnRNP A2 FS and CBF-A ΔRNP1 mutants lost their capacity to disrupt tetraplex structures of the (CGG) sequence, they maintained their nucleic acids binding activity (Cohen, E. and Weisman-Shomer, P., unpublished data). It is likely, therefore, that the enhancement of (CGG) mRNA translation by hnRNP A2 and CBF-A was due to their quadruplex destabilizing activity rather than to their RNA-binding capacity. Whereas both hnRNP A2 and CBF-A were shown to disrupt quadruplex structures of (CGG) in DNA, only hnRNP A2 but not CBF-A destabilized a bimolecular quadruplex structure of synthetic (CGG) RNA (). Because the and conditions for CBF-A action were different, since the likely RNA structure that formed was intra- rather than intermolecular (), and as the CBF-A ΔRNP1 mutant failed to stimulate translation, we assume that similar to hnRNP A2, CBF-A also enhanced the translation of FL (CGG) mRNA by untangling its monomolecular secondary structure. The observed correlation between the ability of hnRNP A2 or CBF-A to unfold quadruplex structures of RNA or DNA and their capacity to alleviate the blocking of translation by the (CGG) tract, was consistent with the folded repeat sequence being a tetrahelix. However, as discussed above, the observed intramolecular secondary structure of the (CGG) repeat may represent a hairpin rather than quadruplex. Such scenario raises the untested prospect that hnRNP A2 and CBF-A can also destabilize hairpins and that similar to their tetraplex unfolding activity, the unwinding of hairpins also requires that the proteins possess intact RNP1 and ATP/GTP motifs. Taken together, our results lent credence to the notion that secondary structures of the expanded (CGG) repeat tract in fragile X premutation mRNA hindered translation and that their disruption alleviated this impediment. This conclusion gained support recently by the report that a quadruplex motif in the 5′ UTR of the human NRAS oncogene mRNA modulated its translation . Furthermore, bioinformatics analysis predicted that 5′ UTR elements in 2922 other human mRNA species can assume tetraplex structure that may potentially regulate their translation (). These results, as well as the present report on the capacity of quadruplex destabilizing proteins to alleviate the impediment to translation that a folded (CGG) tract posed, are in line with the notion that tetrahelical motifs in mRNA may modulate its translation. Although functional hnRNP A2 and CBF-A augment the translation of FL (CGG) mRNA, their mechanisms of action appeared to be dissimilar. Whereas the level of FL (CGG) mRNA was 5.9-fold higher than that of FL mRNA with no repeat tract, hnRNP A2 lowered the amount of FL (CGG) mRNA to that of the FL (CGG) mRNA. At the same time, hnRNP A2 did not affect the amounts of accumulated FL protein that FL (CGG) mRNA encoded ( and C). It is possible that hnRNP A2 suppressed transcription by affecting the conformation and chromatinization of the promoter through its direct interaction with the adjacent (CGG) repeat tract. Yet, although hnRNP A2 lowered the amount of FL (CGG) mRNA, it increased the relative efficacy of its utilization by 5.2-fold, as reflected by the rising ratio of FL protein to mRNA from 0.6 to 3.1 ( and C). By contrast, functional CBF-A increased the relative amount of FL (CGG) transcripts and amount of FL protein by 7.2- and 13.7-fold, respectively ( and C). As a result of the accumulation of FL protein in excess over FL mRNA, CBF-A augmented the efficacy of mRNA usage and increased the ratio of FL protein to mRNA by 3.2-fold from 0.6 to 1.9 ( and C). The different patterns of translation enhancement by hnRNP A2 and CBF-A is not surprising. In addition to its quadruplex (CGG) untangling activity, CBF-A also binds single-stranded DNA and tetraplex forms of some sequences whereas hnRNP A2 also binds RNA and interacts with some DNA structures. It is plausible, therefore, that because of their binding to different targets in DNA or RNA in the cell in addition to their direct interaction with tetraplex (CGG) in RNA, each protein differently affects transcription and translation.

Genome-wide association studies with large sets of single nucleotide polymorphisms (SNP) () are a new option for mapping the genetic variants underlying complex human diseases. However, the power and cost-effectiveness of such studies depends critically upon the properties of the SNP sets used. Consequently, the choice between one of the commercially available marker panels and the construction of a new set is of strong practical significance. No objective criteria other than descriptive measures (e.g. marker number) have so far been used to compare the utility of genome-wide marker sets. More importantly, any sensible assessment of a marker panel requires that recent discoveries about the biology of meiotic recombination are appropriately taken into account (). For example, it has been shown () that the ‘geodesy’ of the human genetic map is fairly homogenous above the centi-Morgan level, but that the correlation between physical and genetic distance is weak at a finer scale, due to rapidly evolving recombination hotspots. Consequently, SNP selection strategies that are based upon the assumption of static linkage disequilibrium (LD) blocks, or that merely employ pairwise LD, may result in sub-optimal marker sets. The utility of a marker set for disease association analysis is determined by a number of factors, including marker number, informativity and spacing, in addition to the local level of LD. In practice, genotyping technologies may pose serious restrictions upon the usability of an individual SNP, irrespective of whether its inclusion might be desirable or not. If such limitations can be ignored, however, then the utility of a marker set should ideally be evaluated by a criterion that: Shannon entropy () is a well-established mathematical concept for assessing the utility of genetic markers. We have recently devised an entropy-based SNP selection approach () that can in principle be adapted to a genome-wide setting. Furthermore, the methodology facilitates estimation of the relative, region-specific efficacy of a given marker set by τ, a quantity that approximates to the relative sample size required to map a causative variant at a given map position, compared to including a maximally polymorphic SNP at the same position (see Methods section). We calculated τ across the genome using publicly available genotype data for HapMap (Phase 2, built 35) () and for the five commercial marker sets of Affymetrix () (100K and 500K) and Illumina () (100K, 300K and 550K). The results were compared to an ‘ideal’ SNP set constructed from HapMap via entropy-based marker selection. Parameter ɛ, which denotes the inverse of the swept radius, was used as a local measure of LD strength (,) and was estimated from HapMap genotype data on the basis of all markers with a minor allele frequency ≥10%. To this end, pairwise haplotype frequencies were estimated from the genotype data using an EM algorithm. Marker-specific ɛ values were estimated by a log-linear regression analysis of ρ and the physical distance to all other markers in a 500 kb window surrounding the marker of interest (), i.e. by fitting model log(ρ) = −ɛ·| − | to marker locations and . Here and in the following, we assumed that the population of interest was characterized by monophyletic inheritance and by a lack of association between unlinked loci, a simplification of the original model of LD decay that was justified by empirical observations made for autosomal markers in Europe and the US (). At inter-marker positions < < , ɛ() was estimated by linear interpolation, i.e. We have previously devised a method for assessing the utility of marker sets for disease association studies (), based upon Shannon entropy (). For the purposes of disease association analysis, a genomic region is assumed to be covered by markers , … , at map positions < … < . Then, the problem of SNP selection reduces to deciding, on the basis of existing genotype or haplotype data, which single marker out of some additional markers , … , to include in order to maximize the mapping utility of the extended panel. Without loss of generality, it can be assumed that this choice is confined to maximizing the utility of the marker set in a given interval, centred at map position . A utility score (:, ) is then constructed that reflects the benefit, with respect to mapping a disease gene at position , of adding to a single marker , Here, (|) = (,) − () denotes the conditional entropy of given . The quantity in formula () can be calculated directly from pairwise haplotype frequencies, known swept radii and known marker locations. Application of the above-mentioned framework to large-scale genome-wide data sets poses additional computational problems since the comprehensive evaluation of all pairwise haplotype frequencies, as required by formulas () and (), is no longer feasible. In this way, the number of pairwise haplotype frequency estimations was limited and the computing time scaled linearly (instead of quadratically) with marker number. Formula () was also used for selecting the first few markers on a given chromosome, successively breaking the chromosome down into shorter intervals by applying formula () to the corresponding interval centers. Marker selection according to formula () commenced for an interval when it was shorter than three times the internal median swept radius. Since equals the predicted allelic association () between and a maximally informative biallelic marker at map position , it follows that On the other hand, the number of individuals required to detect association ρ between and at significance level α and with power 1 − β is approximately equal to For any two marker sets A and B, let τ() and τ() be the τ values obtained with respect to the same location . The methodology described above has been implemented into a suite of JAVA programs interacting with a MySQL relational database for the storage of genotypes and intermediate results. Since the HapMap data set was the most exhaustive one, calculation of swept radii was based upon these markers and genotypes. The software is available as a web service under . Caucasian genotype data for HapMap (Phase II, built 35), Affymetrix 100K and 500K were retrieved from the respective web sites (, ). The marker identities of the Illumina 100K, 300K and 500K sets were retrieved from the Illumina website (); the corresponding genotypes were taken from HapMap or from the Illumina website. Quantity τ measures the relative efficacy of a given marker set to map a causal variant at a specified map position, compared to including a maximally polymorphic SNP at the very same position (see Methods section). Therefore, τ = 1 corresponds to full local efficacy of a marker panel whereas τ = 0 indicates that no information can be extracted locally. For the purpose of comparing different marker sets, τ was calculated here at 10 kb intervals along the human genome (NCBI build 34), except for annotated gaps, heterochromatic, telomeric and centromeric regions. Y chromosomal SNPs were also excluded. Variation of the interval size between 5 and 10 kb for chromosomes 3 and 19 did not yield notably different results (data not shown). It may be argued that, in many instances, only markers located in gene-coding regions are of practical interest for genome-wide disease association studies. In order to take this issue into account, ‘coding’ regions were defined here as all sequences containing one of the ‘RefSeq’ genes of the Golden Path (), including exons, introns and 10 kb of flanking sequence. Marker sets were evaluated on the basis of publicly available genotype data (). Our analyses included CEPH samples from Northern and Western Europe (CEU), from Yoruba in Nigeria (YRI) and from Japanese and Han Chinese people (JPT + CHB). Swept radii 1/ɛ were estimated for different genomic regions on the basis of the available HapMap genotype data. As is exemplified by chromosomes 12 and 19 in the CEU population (A and A), the distribution of 1/ɛ was found to vary considerably along chromosomes and therefore resembled recently published recombination plots in this respect (). The median 1/ɛ of ∼500 kb corresponds closely to previous estimates (). A graphical representation of all swept radii and values obtained in the present study is available at . In the following, our results will be exemplified by a more detailed consideration of chromosomes 12 and 19, which are typical in terms of their size and gene density. When all 180 613 HapMap SNPs on chromosome 12 were included in the analysis, τ values larger than 0.5 were obtained for most of the chromosome (C). By contrast, the 5253 chromosome 12 markers of the Affymetrix 100K set left many intervals with τ close to 0, indicating low efficacy (B). Similar results were obtained for chromosome 19 (). and provide an overview of the distribution of τ along the coding’ regions and the full genomic sequences of the two chromosomes. When all HapMap SNPs were included, the median τ values obtained were 0.70 (interquartile range: 0.56–0.82) for chromosome 12 and 0.66 (interquartile range: 0.52–0.78) for chromosome 19. By contrast, the best commercial marker sets yielded a median τ of 0.59 (interquartile range: 0.45–0.73) for chromosome 12, and of 0.56 (interquartile range: 0.41–0.70) for chromosome 19 in the case of Illumina 550K, and of 0.52 (interquartile range: 0.36–0.67) for chromosome 12 and of 0.41 (interquartile range: 0.26–0.58) for chromosome 19 with the Affymetrix 500K set. A comparison of the two commercially available 100K sets revealed the impact of both, the genotyping technology and the selection strategy upon the mapping efficacy. If only the coding sequence was considered on chromosome 12, the median for Affymetrix 100K was 0.21 (interquartile range: 0.08–0.41), as compared to 0.44 (interquartile range: 0.27–0.61) for Illumina 100K (). The Illumina 100K set, designed primarily for a good coverage of sequences containing annotated transcripts, provides essentially the same efficacy for the coding sequence on this gene-rich chromosome as the Affymetrix 500K set (median τ: 0.41, interquartile range: 0.26–0.58). Similar, albeit less pronounced results were obtained for chromosome 12 (). A genome-wide overview of the efficacy of all SNP sets is given in and, on a chromosome-wise basis, in . Let denote the local coverage of a chromosome or chromosomal region at relative efficacy , achieved by a particular marker set (i.e. equals the proportion of a given genomic region for which τ ≥ ). For the coding regions of chromosome 12, for example, = 0.16 for the Affymetrix 100K set and = 0.48 for Affymetrix 500K (). This means that the two sets cover 16 and 48% of the gene containing sequence, respectively, at 50% or higher relative efficacy. At 80% relative efficacy, the respective figures decrease to 2 and 8%, respectively. A genome-wide overview of the coverage of the different marker sets at 50 and 80% efficacy is given in and, on a chromosome-wise basis, in and . The HapMap markers provide the ‘gold standard’ for the currently achievable coverage of the human genome with informative SNPs. If a fully flexible genotyping technology were available, optimal SNP sets could thus be constructed from HapMap using, for example, entropy-based marker selection. As exemplified for chromosomes 12 () and 19 (), such customized panels would significantly improve the coverage provided by a given number of markers. With 5253 SNPs on chromosome 12, which corresponds to the size of the respective Affymetrix 100K set, HapMap would yield = 0.81, i.e. a more than four times higher coverage than the commercial product. Replacing the Affymetrix 500K set by a similarly sized HapMap set would increase from 0.48 to 0.81 whereas would increase from 0.07 to 0.21. More detailed information about the present study can be found on our web server at . The same site also provides routines for the customized selection of optimal SNP sets from HapMap build 19, using the available Caucasian, Asian and Yoruba genotype data. Under a model of spatially homogenous LD, with constant recombination and mutation rates and a common evolutionary history shared by all chromosomal regions, disease association markers would ideally be spread evenly along the genome. However, the systematic evaluation of both LD and local recombination rates has revealed an inherent non-uniformity of these characteristics (,,). Thus, recombination rates differ between chromosomal segments and between populations, which implies that even closely linked genomic regions may be of substantially different ancestry in individuals from one and the same population (). Consequently, the relationship between LD and physical distance is complex, and combinations of unevenly spaced SNPs may prove more informative than equally spaced markers, depending upon the genomic region of interest.() Previous studies have suggested the existence of ‘haplotype blocks’, i.e. clearly identifiable chromosomal segments that are characterized by a reduced rate of recombination, low haplotype diversity and a high level of internal LD (). In addition, haplotype-tagging SNPs (htSNPs) have been proposed to be capable of identifying haplotypes for substantially larger marker sets from within these blocks (). The practical relevance of this block concept arises from the expectation that htSNPs extract sufficient information from an LD block with respect to co-ancestry while, at the same time, reducing genotyping costs (). A number of computational methods for the construction of htSNP sets have been developed (,,) but for these techniques to be efficient, detailed knowledge of the extended haplotype frequency distribution in the population of interest is required. Moreover, the size and location of haplotype blocks depend critically upon the SNP density and the method of marker selection (,,). Therefore, haplotype tagging appears feasible only when large samples and appropriate family structures are available for the necessary (deterministic or probabilistic) haplotype assignments, the reliability of which decreases with the number and complexity of the haplotypes present (). The idealized picture of static LD blocks, separated by hot spots of recombination, () has recently been challenged by new insights into the biology of meiotic recombination (). The correlation between physical and genetic distance is weak below the centi-Morgan level so that the inference of marker genotypes from htSNP haplotypes is far from being reliable (). Moreover, block-like structures may even occur merely because of genetic drift (). It thus appears as if the tacit assumption underlying the use of the haplotype block concept for disease association mapping, namely that all genetic variation in a block follows the same hierarchical pattern, is often not fulfilled. As a consequence, the usefulness of htSNPs for such studies has generally been questioned (). SNP selection based upon pairwise LD alone has been suggested to avoid the conceptual and computational problems of extended haplotype (or ‘block’) approaches. The use of some SNPs as proxies for other SNPs that are in high LD with the former (), measured by , reduces the redundancy of a SNP set. Thresholds for of at least 0.8 are generally regarded as sufficient to provide good marker coverage for association studies (,). The rationale underlying the pairwise approach is the expectation that high inter-marker LD translates into high LD between some of the markers and potentially causative variants, an assumption that is however unlikely to hold true in general (). Selection of SNPs based upon pairwise LD alone is therefore likely to perform well only with a particularly high and uniform SNP density (). Irrespective of the approach taken, the inherently unknown LD between markers and unknown causal variants has to be extrapolated in one way or another from both physical distance and the local strength of LD. However, marker selection based upon pairwise LD alone does not take distance or individual marker informativity into account. As a consequence, simple pairwise ‘haplotype tagging’ potentially leads to inhomogeneous marker spacing with less than maximum efficacy. Here, we have adapted a recently proposed method for selecting maximally informative marker sets for association studies () to a genome-wide comparison of marker sets. The original approach combines the information content, physical spacing and pairwise LD of individual markers with information on the local LD structure, extracted from available data in the form of swept radii (,). All of these determinants are included in a single, position-specific utility measure that corresponds to the distance-weighted haplotype entropy of the marker set, approximated however by a pairwise score of the same form (see Methods section). The approach is therefore not affected by the computational and conceptual problems of block-based methods and, at the same time, takes physical distance and local LD structure into account when extrapolating LD between markers and causal variants from pairwise inter-marker LD. An extension of the approach has led to the development of a quantitative criterion (τ) that approximates the efficacy of a given marker set to map a disease-causing variant at a position of interest. It should be emphasized that the interpretation of τ as a measure of efficacy is only valid in relative terms, i.e. by comparison to the inclusion of a maximally polymorphic SNP at the site of the causal variant. In general, since the properties of the underlying disease model are unknown, no marker-based quantity can on its own provide information about the absolute power of a marker set to map genetic variants underlying a given phenotype. Owing to recent successes () and its theoretical appeal (), significant funds have been allocated to the concept of genome-wide association analysis in the context of various phenotypes. Researchers are however facing the practical problem of choosing the ‘right’ genotyping technology. In many countries, universal control genotype pools are in the process of being established, and these pools will pre-determine the choice of technology for future studies. Of the currently available marker sets, the Affymetrix 500K ( = 0.68, = 0.19) and Illumina 550K ( = 0.79, = 0.29) products provide the best genomic coverage in Caucasians. The Illumina 550K marker set provides a higher coverage than the 500K Affymetrix set, probably because of the higher flexibility of the Illumina genotyping technology. Pronounced differences between full genomic and ‘coding’ region coverage were only observed for the 100K sets, probably because of the relatively small marker numbers. The good ‘coding’ region coverage provided by the Illumina 100K set highlights the fact that this panel was primarily designed for gene-based association mapping. It should be emphasized, however, that all of the above conclusions were based upon the assumption that all markers were callable, and that practical factors such as genotyping quality, departure from Hardy–Weinberg equilibrium and DNA requirements could be neglected. Furthermore, interesting differences became apparent in terms of in different ethnic groups. Whilst their relative efficacy was approximately the same in the Caucasian and African populations, SNP coverage was notably poorer for all products for the East Asian populations.

Pathologic double-strand DNA breaks (DSBs) arise when ionizing radiation passes near DNA or when a replication fork encounters a nick. In mitotic cells, physiologic double-strand breaks are generated in lymphocytes during V(D)J recombination and class switch recombination (,). Many pathologic DSBs and nearly all physiologic DSBs in mitotic cells are repaired by the non-homologous DNA end joining (NHEJ) pathway. Given this, it is not surprising that individuals born with defects in non-homologous DNA end joining (NHEJ) are sensitive to ionizing radiation and have severe combined immunodeficiency (SCID) (). Many factors involved in NHEJ have been identified based on analysis of human and mouse SCID (), and ∼15–20% of human SCID is due to NHEJ defects. The proteins involved in NHEJ include Ku70, Ku80, DNA-PKcs, Artemis, XRCC4, DNA ligase IV, pol mu, pol lambda and XLF (or Cernunnos) (). NHEJ is thought to begin with Ku binding to the two newly created DNA ends at the double-strand break. Ku improves the affinity of a nuclease complex (Artemis:DNA-PKcs), a ligase complex (XRCC4:DNA ligase IV) and POL X polymerases (pol μ and pol λ). XLF (or Cernunnos) () stimulates the XRCC4-DNA ligase IV complex in ligation assays (). We and others have recently described the ability of XRCC4-DNA ligase IV to ligate incompatible DNA ends (,) and the ability of the XLF (Cernunnos) protein to stimulate the ligation of DNA ends (,). However, two studies observed that XLF (Cernunnos) stimulated ligation by XRCC4-DNA ligase IV at compatible DNA ends (,), whereas one study only observed stimulation at incompatible DNA ends (). Here, we find a marked influence of terminal DNA end sequence on end joining by XRCC4-DNA ligase IV. Nearly all aspects of the sequence effect conform to a model in which the two DNA ends initially anneal at any shared sites of complementarity. However, some aspect of the joining process places the base at position N on the top strand in a sterically constrained position relative to the N+1 base on the anti-parallel strand. In addition, we find that at physiologic Mg concentrations, XLF (Cernunnos) stimulates all DNA end joining, regardless of the extent of terminal microhomology between the two DNA ends, though inefficiently joined ends benefit most by the presence of XLF (Cernunnos, but hereafter called XLF, for simplicity). Finally, we find that XRCC4-DNA ligase IV is capable of ligating single-stranded DNA and long single-stranded overhangs. We discuss how these properties of the NHEJ ligase complex explain aspects of the flexibility of mammalian NHEJ. The purification of Ku has been described. XRCC4:DNA ligase IV complex was purified from baculovirus-insect cell system as described. Native DNA polymerase mu was expressed and purified from , as described previously. Soluble human XLF-myc-his protein was expressed in 293T cells and purified by Ni-NTA agarose beads and Mono Q column as described. T4 DNA ligase was purchased from New England Biolabs (Beverly, MA, USA). The DNA ligation assay was performed in a 10 μl reaction. DNA substrates were first incubated with or without Ku and/or XLF in 1× ligation reaction buffer (25 mM Tris–hydrochloride, pH 7.5, 75 mM NaCl, 72.5 mM KCl, 2 mM DTT, 0.025% Triton X-100 and 100 μM EDTA) supplemented with 10% PEG (MW > 8000 kD), 50 μg/ml BSA and 5% glycerol at room temperature for 15 min. The EDTA was used to eliminate effects of any possible trace divalent cations, but is not necessary. Mg2+ is added where specified above this low level of EDTA. PEG improves the ligation efficiency substantially. The low level of Triton X-100 is irrelevant to the ligation efficiency. Ligation was initiated by adding 10 mM MgCl with the combinations of proteins designated on the gels (XRCC4-DNA ligase IV, and pol mu, where indicated). Reactions were then incubated at 37°C for 30 min or 2.5 min as indicated. After incubation, reactions were stopped, deproteinized with organic extraction and analyzed by 8% or 10% denaturing PAGE gel. Gels were dried, exposed in a PhosphorImager cassette and scanned. We have found that there is a wide variation in the efficiency of DNA end ligation by XRCC4-DNA ligase IV due to minor variations in the DNA end sequences. For example, the end sequences illustrated in B, left panel (–ACC3′ joined to 3′ GGA–) are joined well (on the top) strand by Ku plus XRCC4-ligase IV (lane 1), and this is only slightly increased by the addition of pol mu (lane 2). However, the bottom strand of those same DNA ends is not ligated well (lane 3), unless pol mu is present to fill in the 1 nt gap (lane 4). This illustrates that the two strands of the same pair of DNA ends are not ligated with equal efficiency by XRCC4:DNA ligase IV and some aspect of the sequence at the two ligation points determines this difference. (Note that when we test the top strand for ligation, the bottom strand has an unligatable 5′OH at the junction. Likewise, when we test the bottom strand for ligation, the top strand has an unligatable 5′OH at the junction.) There are hundreds of variations of the two partially complementary DNA ends of A, and we have not analyzed all of them. However, a 1 nt change in the complementary portion, such that the ends are –AGC3′ and 3′ CGA– now results in a substantial reduction (60-fold) in the ability of Ku plus XRCC4-ligase IV to join the top strand of these ends (B, right panel, lanes 3 versus 8). DNA sequence analysis confirmed the junctional sequence. These results illustrate that 1 nt changes in otherwise identical overhangs can markedly affect the joining. Importantly, XLF is able to permit joining to a much greater level for a pair of inefficiently joined ends (B, right panel, lanes 3 versus 4). In contrast, ends joined efficiently are not stimulated further by XLF (lanes 8 versus 9). [Assays done using shorter incubation times and less ligase rule out the possibility that the lack of stimulation is due to the reaction reaching a plateau (unpublished data).] This raises the possibility that base pairings (2 bp here) that are accommodated well by the ligase complex are sufficient to stabilize those ends in a manner that XLF cannot improve upon, whereas more weakly accommodated end sequences benefit from XLF. The following sections describe studies to further explore both the sequence effects on ligation and the XLF effects on ligation efficiency. For ligation with XRCC4-DNA ligase IV and Ku (with or without XLF), we observed a certain pattern of ligation efficiencies for pairs of ends with partial complementarity (2 bp of terminal microhomology with a 1 nt gap on each strand). After annealing at any chance terminal microhomologies, the order of ligation efficiencies for the various pairs of DNA ends fits a pattern most consistent with some steric limitations due to purines at positions 2 and 3 nt from one 3′ end being in conflict with purines on the anti-parallel strand, but at positions that are 1 nt shifted from the initial base pairing. The ligation efficiency is optimal when the purine:purine conflict (R:R) is minimal within a slanted 2 nt region covering each of the two overhangs (Supplementary A; see Supplementary for a diagram of the model). In order to test this model, we created many of the end pair configurations for 3 bp 3′ overhangs that share 2 bp of terminal microhomology. Among the 11 substrate pairs created, nearly the entire set conformed to the model (Supplementary B and Figure 1; see model in Supplementary Figure 2). Specifically, the substrates with the least R:R conflict within the 2 nt region are most efficiently ligated (B, right panel, lane 8 and Supplementary B, lane 13). The substrates with the most R:R conflict within the 2 nt region were least efficiently joined (B, left panel, lane 3; right panel, lane 3 and Supplementary Figure 1, lane 3). Hence, the two DNA ends behave according to a model where they initially anneal and then are subject to steric constraints on what bases can occupy the 3′ overhangs, and these limitations affect ligation across any gaps in the top or bottom strands (Supplementary Figure 2). As mentioned above, we find that ligation of pairs of DNA ends by XRCC4-DNA ligase IV are highly stimulated for ligation by XLF. However, compatible DNA ends (4 bp 3′ overhangs) and pairs of blunt DNA ends are not stimulated for ligation by XLF (). This suggests that substantial terminal homology achieves the same effect as XLF for incompatible DNA ends, namely, stabilization. In previous work, we and others had shown that XLF can stimulate ligation of compatible DNA ends (,). However, our studies were done at concentrations of Mg (2.5 mM) that are closer to physiologic [∼0.3 mM; ()]. We reasoned that our failure to detect XLF stimulation for compatible DNA ends here (B) might be due to use of a higher Mg concentration (10 mM) in the ligation buffer in the current study. To test this, we studied the ligation as a function of the Mg concentration. At concentrations of Mg of 0.2, 0.5, 1 or 2 mM, XLF stimulated ligation of the compatible DNA ends. XLF failed to stimulate at concentrations ranging from 5 to 20 mM Mg (A). For incompatible DNA end joining, XLF stimulated ligation by XRCC4-DNA ligase IV at all Mg concentrations between 0.2 and 20 mM (B and Supplementary Figure 3). These observations help to unify the disparate findings reported by various laboratories in which some found XLF stimulation of compatible DNA end ligation (,) and others did not (). Like terminal microhomology, XLF and Mg may help to stabilize the pair of DNA ends within the ligase complex. For ends that are already stabilized by terminal microhomology and high Mg (5 mM or above), XLF may have no additional stabilizing effect. In our analysis of incompatible DNA end joining, we examined progressively longer overhangs (up to 15 nt poly-dT overhangs). For comparison, we tested for the ligation of poly-dT as single-stranded DNA. We were surprised to detect a ladder of ligation products exceeding 6 unit molecules in length for the poly-dT substrate (). Notably, poly-dA,-dG,-dC and -rU were not ligatable (data not shown). Moreover, a single terminal dT at the 5′ or 3′ end of the single-strand was not sufficient to permit ligation. Five dT nucleotides at both the 5′ and 3′ end were readily ligated, regardless of the internal sequence (data not shown). Neither Ku nor XLF could stimulate this single-strand ligation, probably because Ku does not bind single-stranded DNA and XLF binds it very inefficiently (). In related studies, we have also observed ligation of a DNA duplex with a long poly-dT overhang to a single-stranded poly dT molecule (Supplementary , lanes 2 and 3) and ligation of two duplex molecules, one with a long 3′ dT overhang and the other with a long 5′ dT overhang (lanes 4–8). Hence, the XRCC4-ligase IV complex has remarkable flexibility to ligate single-stranded DNA and long 5′ or 3′ overhangs. Yet the interactions of the single-stranded substrate near the active site of the ligase complex may require more than one dT nucleotide near the 5′ and the 3′ end. #text S u p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

The DNA microarray is one of the most powerful biotechnological tools used to conduct high-throughput analysis of DNA sequences, genetic variations and gene expressions (). To develop efficient DNA microarray systems, one of the most essential and important subjects is how to immobilize probe DNA oligomers on the solid surface so that the resulting hybridization between the targets and the probe is detected clearly and the array can be stored for long periods without cleavage of the probes (). For this purpose, fabrication systems generating covalent bondings between the probe and the solid surface is preferred since DNA probes, which are tightly immobilized on the surface by the covalent bond except in a few cases such as thiol-gold conjugating (), provide high stability of the arrays and reproducibility of the data obtained (,,). For the fabrication of DNA microarray systems, both probe DNA oligomers and solid surfaces are usually modified with reactive organic functional groups (,,), and then by utilizing the chemical activation employing appropriate reagents, covalent bonding is formed between the probe and surface (,,). Several functional groups such as carboxyl, phosphate, aldehyde and amino groups are commonly introduced, and therefore, the relevant chemical activation steps have also been developed according to the combination of the introduced functional groups (,). Amino groups, for instance, have been chiefly employed for both the probe and the surface because of its easy preparation, stable functionality and wide applicability. To attach the probe DNA oligomers covalently on the NH-functionalized surface, the solid surface modified with amino groups are subsequently subjected to chemical activation by use of homobifunctional linkers such as disuccinimidyl glutarate (DSG), phenylene diisothiocyanate (PDC) (,), etc. Although such surface-activation strategies are frequently adopted for covalent bonding formation in DNA microarray fabrication, they have some drawbacks that the activated surface has no long life and the surface should therefore be activated just prior to use, and during the cross-linking reaction, undesirable by-products remain on the surface (,). Also, there is a high possibility for the activated groups to react with free amino groups on the same surface or with the amino groups of DNA nucleobases, which inactivate the immobilized probes (,). As another approach, probe DNA oligomer is subjected to chemical activation step and then the activated probe DNA employed for the covalent bonding formation on the surface (,,). For instance, when the probe DNA oligomers with carboxyl or phosphate groups at the ends are immobilized on the NH-functionalized surface, dehydration reagents such as dicyclohexylcarbodiimide (DCC), 1-ethyl-3-(3-dimethylaminopropyl)carbodiimide (EDC), etc. are employed usefully for their activation (). Such probe-activation strategies, however, have several fundamental problems in DNA microarray fabrication. Since most of the DNA microarray fabrications are based on robotic-spotting process, the probe should be prepared as a reproducible or controllable form for obtaining stable and reliable immobilization performance. However, activated carboxyl or phosphate groups do not have a sufficiently long half-life in aqueous conditions, because the attached activator group is easily hydrolyzed () so that the activated DNA probes are inactivated. In addition, much excess amount of the dehydration reagent is required compared to the probe DNA oligomer, and therefore, by-products are formed. Also, the sample solution containing only the activated probe DNA oligomers, which would be ideal, cannot be obtained easily by general purification methods (). As mentioned above, in the processes of DNA microarray fabrication, activation strategies, which have been used widely for covalent bonding formation between the probe and surface, still possess technical problems. Consequently, simple and direct covalent bonding fabrication method, which does not require additional activation steps and leaves no by-products, is preferred. In the present study, we employed oxanine (Oxa) as a new linker for mediating direct covalent bonding reaction for the immobilization of probe DNA oligomer on the NH-functionalized surface in one-pot mode, as shown in . Oxa was identified in 1996 as a unique lesion generated as one of the main deamination products of guanine (Gua) by NO- or HNO-induced nitrosative oxidation () and the formation mechanism has been identified in detail (). Since Oxa has an -acylisourea structure, an activated-carboxyl group, as illustrated in A, Oxa is expected to react with amino or thiol group of biomolecules (,). However, the practical use of such a unique function has not yet been focused on until date. Recently, we have developed a solid-phase chemical synthesis procedure for incorporating Oxa into DNA oligomers (), so that it is possible to employ Oxa-containing DNA oligomers in the field of biotechnological applications such as DNA microarray fabrication. Probe DNA molecules were prepared by incorporation of phosphoramidite monomer of deoxyoxanosine (dOxo, deoxynucleoside of Oxa) into the 5′-end of DNA oligomers with or without fluorescence labeled at 3′-end by the chemical synthesis. They were then spotted on the NH-functionalized glass slide, and subsequently subjected to the post-spotting treatments such as baking process (conventional method) or mild treatment (substitution method newly adopted in this study). Then, the performance of Oxa as a linker was evaluated by investigating the immobilization efficiency and hybridization efficiency. This study shows that use of Oxa as a linker is efficient for covalent attachment of probe DNA in one-pot mode even under the mild conditions and therefore practically appropriate for DNA microarray fabrication. Reagents for DNA oligomer synthesis including CPG column and appropriately protected normal nucleosides were obtained from Glen Research (Sterling, VA, USA) and solvents for the synthesis from Applied Biosystems (Foster, CA, USA). Other organic solvents were purchased from Nacalai Tesques (Osaka, Japan) and all other chemicals from Wako Pure Chemicals (Osaka, Japan). The glass slides and several NH-functionalized glass slides [25.4 × 76.2 mm (±0.05 mm)] were obtained from Matsunami (Osaka, Japan), Corning (Corning, NY) and Nunc (Wiesbaden, Germany). For purification of synthesized probes and target DNA oligomers, an RP-HPLC system consisting of a Tosoh PX-8020 (controller), DP-8020 (pump), CO-8020 (temperature controller) and PD-8020 (diode detector) and Ultron VX-ODS column [150 × 4.6 mm (for analysis) or 150 × 6.0 mm (for purification), 5 μm; Shinwa Co. (Kyoto, Japan)] were used. Chemical synthesis of DNA probes and target DNA oligomers was carried out on an Applied Biosystems 3400 DNA synthesizer [Applied Biosystems (Foster, CA)]. UV spectra of DNA oligomers were measured on a Shimadzu UV-260 UV-Vis spectrophotometer equipped with an SPR-5 temperature controller. The DNA oligomers were spotted on the glass slides with an inkjet spotter [custom built by NGK Insulators (Nagoya, Japan)]. The spotted glass slides were scanned with DNA MicroArray Scanner Model G2505A of Agilent Technologies (Palo Alto, CA, USA) or Array WoRx of Applied Precision (Issaquah, WA, USA). Probe DNA oligomers with Oxa-nucleotide unit (Oxa-N) at the 5′-end were prepared by the solid-phase chemical synthesis procedure, as reported previously (). The probe sequence, 5′-d(TGTTGTCGAAAATGTCAACG)-3′ (XDHp), was designed from the antisense part of the xylitol dehydrogenase gene (XDH) from . After the synthesis of XDHp, Oxa-N was incorporated at the 5′-end in the final coupling step. Three kinds of probe DNA oligomers with Oxa-N were synthesized; one is the wild type (Oxa-Probe) containing Oxa-N at the 5′-end and the others are those in which carbon 3 (–(CH)–) and carbon 12 (–(CH)–) spacers are inserted between XDHp and Oxa-N (namely, Oxa-c3- and Oxa-c12-Probes, respectively) as shown in . For the analysis of the immobilization efficiency, probe DNA oligomers labeled with fluorescein [ = 495 nm, = 519 nm, ε = 75 000 (M cm)] at the 3′-end were prepared by using fluorescein-bound CPG column (Glen Research). For the analysis of the hybridization efficiency of the probe DNA oligomers immobilized on the glass slide, two types of XDH target sequences, 5′-d(CGTTGACATTTTCGACAACA)-3′ (short XDH target; XDHt-S) and 5′d(CTGCTGCTGTCGCCAAGACCTTCGGTGCTAAGGGTGTCATCGT AGTTGAAGATGGCCAAGGACATTGGTGCTGCTACTCACACCTT)-3′ (long XDH target; XDHt-L, underlined bases are the same sequence as XDHt-S) were prepared and their 5′-ends were labeled with Cy3 [ = 547 nm, = 563 nm, ε = 136 000 (M cm)]. The probe and target DNA oligomers used in this study are listed in . The NH-functionalized surface was prepared on glass slides as follows. After 10 min ultrasonication in acetone and 10 min vacuum drying, to prepare a hydroxyl group-enriched surface, treatment in Piranha solution (70% HSO, 30% HO) was carried out at 55°C for 30 min. After cleaning by sonication in methanol, methanol/toluene [1:1 (v/v)] and toluene (for 10 min each), silanization was performed by immersing the glass slides in 3-aminopropyl-triethoxysilane (APTS) solution (2% in toluene). During the silanization, containers were placed in an orbital shaker with gentle shaking (70 r.p.m.) at room temperature. The glass slides were then cleaned in an ultrasound bath in toluene, toluene/methanol [1:1(v/v)] and methanol (for 10 min each). The glass slides were baked at 110°C for 1 h. The prepared NH-functionalized glass slides were stored at room temperature immediately prior to use. To see the effect of the modifiers attached to the 5′-end on how efficiently probe DNA oligomers were formed on the NH-functionalized surface (the performance of the modifier as a linker), two different evaluations were employed. One is the immobilization efficiency evaluated by immobilizing probes labeled with fluorescein at the 3′-end (Oxa-Probe-F, Oxa-c3-Probe-F and Oxa-c12-Probe-F) and by measuring the fluorescence intensity of each spot. The other is the hybridization efficiency of the target DNA oligomers, labeled with Cy3 at the 5′-end, with the probes without the fluorescein-label (Oxa-Probe, Oxa-c3-Probe and Oxa-c12-Probe). The resulting Cy3-fluorescence intensity of each spot was measured, assuming that the Cy3-fluorescence intensity is correlated with the quantity of the active probe DNA oligomers immobilized on the glass slides. First, on the NH-functionalized glass slides, probe DNA oligomers were spotted as follows; aqueous solutions of probe DNA oligomers (100 pmol/µl) were mixed in the ratio of 1:1 (v/v) with an inkjet-spotting solution consisting of glycerin, glycerol and disaccharides (), the pH of which was controlled by using sodium-phosphate buffers (10 mM). About 100 pl of these spotting probe solutions (50 pmol/µl) was spotted at 3 mm spacing on the NH-functionalized glass slides with an inkjet spotter. As shown in A, the spotted DNA probes were round in shape (∼80 µm in diameter) and clearly separated from each other. No satellite spot was observed. After the post-spotting treatments, in order to remove the unreacted DNA probes from the slides, the incubated slides were rinsed with 2× SSC (saline sodium citrate) containing 0.1% SDS (sodium dodecyl sulfate, 15 min), soaked in 2× SSC containing 0.2% SDS (5 min), washed with water, soaked in ethanol (1 min) and air-dried at room temperature (30 min). To inactivate unspotted areas, the rinsed slides were incubated in a blocking solution containing 1% bovine serum albumin (BSA), 4× SSC and 0.5% SDS at 42°C for 45 min. The prepared slides with the immobilized probe DNA oligomers were stored in a desiccator, and used to see the effects of the modifiers on the efficiency of the immobilization (see IMB of A). Then, the hybridization efficiency was analyzed in another method for evaluating the performance of the modifier as a linker. The prepared slides were treated with hybridization buffer (5× SSC containing 0.5% SDS) containing the target DNA oligomers (10 pM). After covering with a cover slip, the glass slides were kept at 42°C for 16 h. Then, to remove non-hybridized target DNA oligomers, the incubated slides were immersed in 2× SSC containing 0.1% SDS (5 min), and finally agitated in a gently shaking bath in 1× SSC (5 min) and 0.1× SSC (5 min). After spin-drying, the slides were stored in a desiccator, and used to see the hybridization efficiency of the immobilized probe DNA oligomers (see HYB of A). The glass slides were scanned with a microarray scanner such as DNA MicroArray Scanner Model G2505A or Array WoRx, and the fluorescent image intensities and the location of each analyte spot on the slides were measured using the mapping software, GenePixPro Ver 5.0 or 5.2 of Molecular Devices (Sunnyvale, CA, USA). The fluorescence intensity data obtained for 90 spots for each type of probe DNA oligomer were used for the calculation of the statistical data, which are shown in B. The same procedures of spotting, washing and hybridization steps were carried out on the same probe and target DNA oligomers as adopted in the previous section, except for the employment of the new conditions or combinations of post-spotting treatments. As shown in , the temperature, humidity and time of the post-spotting treatments, pH of the spotting solution and the combination of conventional or new post-spotting treatments were investigated. For the evaluation of the performance of the modifier as a linker, the efficiencies of immobilization and hybridization were analyzed, and their analyses procedures were conducted in the same way as described in the previous section, unless otherwise specified. As a preliminary trial, Oxa was explored for its property as a linker, for instance, its reactivity with primary amines and the stability of the resulting product (Supplementary Data, Table S1 and Figures S1 and S2). Since Oxa contains an -acylisourea structure in the 6-membered ring in which the carboxyl group moiety is bound to the carbodiimide group, as shown in A, Oxa is expected to react with amino groups readily to result in the amide-bond formation without any further activation step of carboxyl group. When 2.5 mM dOxo was incubated in the presence of 25 mM hexylamine, a peak due to the product was observed in RP–HPLC chromatogram (Supplementary Data, Figure S1). Spectroscopic analysis including NMR (Supplementary Data, Figure S2) and mass spectroscopy (data not shown) indicated that -acylisourea structure in Oxa of dOxo reacted with the amino group of hexylamine, resulting in the ring-opened product (dOxo-hexylamine, 1-(2-deoxy-β--ribofuranosyl)-5-ureido-1-imidazole-4-carboxylic acid hexylamide), as shown in A. This amide-bond formation was observed in the confined pH range of ∼6.5 − 11 as shown in B. The rate constant for this reaction was measured at pH 9.5 and 25°C as 2.30 × 10 mM min (second-order reaction). Further, the stability of the -glycosidic bond was analyzed and for instance, the sufficiently long half-life at pH 9.5 and 42°C was obtained at 841 h (Supplementary Data, Table S1). It should be noted that the -acylisourea structure of Oxa can be maintained as ring-closed structure up to relatively high pH conditions [p = 9.4 ()]. Considering both that the optimum condition for the active amino groups is above pH 9 and that Oxa is stable in alkali conditions, the use of Oxa as a linker is suitable for the covalent attachment of the probe DNA oligomers on the NH-functionalized surface. Based on these data and the functional merits, Oxa was employed as a new linker, in the present study. As shown in the previous section and , Oxa reacted with amino groups without any activation step and the amide bond was formed in relatively high pH conditions (pH 9–10). It was expected, therefore, that Oxa could provide efficient covalent bonding formation with amino groups, that is, it could be utilized as a new linker for simple and direct covalent attachment of the probe DNA oligomers on the NH-functionalized surface. To see if this is the case, fluorescein-labeled probe DNA oligomers (Oxa-Probe-F, Oxa-c3-Probe-F and Oxa-c12-Probe-F) were spotted on the NH-functionalized glass slides, and the glass slides were subjected to the baking process (1 h incubation at 80°C), conventional post-spotting treatment. The slides were washed to remove the unreacted probe DNA oligomers and then the fluorescence intensity of each spot was measured for estimating the amount of DNA probes immobilized. The typical fluorescein fluorescence of the spot is shown in the line of IMB in A and the intensities of the spots were measured by the scanner. It was found, as represented in B, that the fluorescence intensities of the spots of probe DNA oligomers with Oxa-N were 1761 (Oxa-Probe), 1805 (Oxa-c3-Probe) and 2032 (Oxa-c12-Probe) (black bars). Although the longer spacer such as –(CH)– was found to increase the immobilization efficiency, it was not remarkable. The spotted probe DNA oligomers with Oxa-N were further washed by the use of high-salt 5× SSC for over 16 h. In the cases of the probe DNA oligomers with Oxa-N, the changes in fluorescence intensities were negligible while in the case of those without Oxa-N, significant decrease in the intensities observed (data not shown), indicating that Oxa-N covalently immobilized the probes. The efficiencies for the hybridization of the target DNA oligomers with the probes immobilized on the glass slide were also investigated. Oxa-Probe, Oxa-c3-Probe and Oxa-c12-Probe, which have no fluorescein-label, were spotted analogously on the NH-functionalized glass slide and the spotted slides were also subjected to baking process (1 h incubation at 80°C). After such a conventional post-spotting treatment and washing steps were performed, the target DNA oligomer, a 106-mer (Cy3-XDHt-L) that has a 20-base sequence in the middle complementary to the probe and whose 5'-end was labeled with Cy3 was hybridized. Cy3-fluorescence intensities of each spot were measured as shown in the line of HYB in A. As depicted by the bar graphs in B, the three types of probe DNA oligomers with Oxa-N showed high Cy3-fluorescence intensities. The intensities were 3227 (Oxa-Probe), 3678 (Oxa-c3-Probe) and 4992 (Oxa-c12-Probe) (gray bars). It was suggested that the longer alkyl spacer would produce the larger efficiency for the target recognition. Analogous results were obtained when Cy3-XDHt-S was used as a target, and when non-complementary was utilized as a target, non-specific adsorption of the target was not detected. Several kinds of commercially available aminosilane-modified glass slides were also tested, and analogous results in immobilization and hybridization efficiencies were obtained (data not shown). If the probe DNA oligomer and the functionalized surface are prepared in the same ways, the post-spotting treatment becomes the determinant step, which influences the performance of the modifier as linker on the solid surface. As described in the previous section, baking process of the spotted slide carried out at 80°C for 1 h was employed as conventional post-spotting treatment. However, such a harsh condition does not seem to be proper for the natural property of probe DNA oligomers, so that mild conditions would be ideal for post-spotting treatment, if covalent attachment is possible. In this study, some other substitution methods, which are more compatible to probe DNA oligomer, were investigated as new post-spotting treatments because Oxa is expected to show sufficient reactivity with amino group in the mild conditions even without any activation step. Two kinds of mild treatments, in which the spotted glass slide was incubated at 25 and 42°C, were employed as the post-spotting treatment and the performance of the modifier as a linker was evaluated by investigating its immobilization and hybridization efficiencies. The three probe DNA oligomers with Oxa-N at the 5′-end and fluorescein-label at the 3′-end were spotted on the NH-functionalized glass slides and then, each mild treatment was employed on the spotted slides as the post-spotting treatment by controlling relative humidity (RH). First, in the case of mild treatment at 25°C, it was found that at this temperature condition, the fluorescence intensities increased with increase in RH and reached a plateau when RH was 75% (data not shown). The data obtained at 25°C and RH 75% for 48 h are depicted by the black bars in A. The average fluorescein intensities of the immobilized probe DNA oligomers with Oxa-N were 3991 (Oxa-Probe-F), 4025 (Oxa-c3-Probe-F) and 4203 (Oxa-c12-Probe-F). In the experiments at 42°C, the immobilization efficiencies of the probes were also investigated by controlling RH. In the experiments at 42°C for 24 h, the highest immobilization data was obtained at RH 50%, as represented by the gray bars in A. The average fluorescein intensities obtained were 5096 (Oxa-Probe-F), 5139 (Oxa-c3-Probe-F) and 5298 (Oxa-c12-Probe-F). Then, the hybridization efficiency was also analyzed for the probe DNA oligomers obtained by the employment of the same mild treatments (post-spotting treatment) as performed in the experiment for the immobilization efficiency. The probes without fluorescein-label were used and Cy3-XDHt-L was the target. At 25°C and RH 75% for 48 h, the Cy3-fluorescence intensities of the targets were measured to be 3005, 3582 and 4915 for Oxa-Probe, Oxa-c3-Probe and Oxa-c12-Probe, respectively, as shown by the black bars in B. The average values obtained at 42°C and RH 50% for 24 h were 5590, 6268 and 9551 for Oxa-Probe, Oxa-c3-Probe and Oxa-c12-Probe, respectively, as represented by the gray bars in B. The results obtained at 42°C and RH 50% for 24 h were found to be higher than the results obtained by the baking process as post-spotting treatment (A and B). These results indicate that the performance of Oxa to make a covalent bonding formation with amino groups on the surface is effective even in such mild conditions, resulting in the high-hybridization efficiency. Analogous results were obtained when Cy3-XDHt-S was used as a target, and when non-complementary was utilized as a target, non-specific adsorption of the target was not detected. Another parameter responsible for the performance of Oxa as a linker may be the time of post-spotting treatments. Several spotted glass slides were identically prepared by spotting the probes (Oxa-Probe, Oxa-c3-Probe and Oxa-c12-Probe) on the NH-functionalized surface. Then, under the employment of post-spotting treatments at 42°C and RH 50% (mild treatment), the treated slides were taken according the treatment time. For each slide sample, the resultant hybridization efficiency measured by Cy3-fluorescence intensities (the target: Cy3-XDHt-L) was investigated for the evaluation of the time-dependent performance of Oxa as a linker. As shown in the bar graphs in C, the hybridization efficiency increased depending on the time of the post-spotting treatment. Analogous results were obtained when Cy3-XDHt-S was used as a target, and when non-complementary was utilized as a target, non-specific adsorption of the target was not obtained. Next, the influences of the pH values on the covalent bonding formation between Oxa and amino group on the surface were investigated by controlling the pH of the spotting solutions. The resultant hybridization efficiency was used for the evaluation of the performance of Oxa as a linker. The probe DNA oligomers employed were Oxa-Probe and the target Cy3-XDHt-L. The pH value of the original spotting solutions was ∼7.5. The pH of the spotting solution was adjusted by adding the same portion of 10 mM sodium phosphate buffer with different pHs to the original spotting solutions. After the spotting of the probes, the glass slides were subjected to the post-spotting treatment at 42°C and RH 50%. As the pH was raised from ∼6 to 10, the efficiency was increased at the higher pH, that is, the reaction of Oxa and amino group on the surface was increased as the pH of the spotting solution increased (data not shown). However, as shown in D, as the incubation time increased, the efficiency decreased after 24 h at high pH, such as pH 9.5, while in the original spotting solution, which does not contain additional salts of sodium phosphate, the efficiency increased with time. These results indicated that although high pH and salts enhance the formation of amide bonding between Oxa and amino group on the surface, they may also cause undesirable influence when the spots are formed and the concentrations of the components raised accordingly as the time of post-spotting treatments elapsed. #text To fabricate DNA microarray systems that recognize targets with high reproducibility and without error and are stable during storage, the development of efficient methods by which probe DNA oligomers are immobilized on the solid surface through the covalent bonding formation between the probe and the solid surface is required. Also, it is recommended that such a covalent bonding reaction is performed under mild conditions so that undesirable by-products are not produced on the surface. In the present study, Oxa, which is one of the unique lesions generated from Gua by NO- or HNO- induced nitrosative oxidation and has an -acylisourea structure reactive to amines, was investigated as a functional linker to see if it is useful in DNA microarray fabrication carried out under mild conditions. By solid-state chemical synthesis, a nucleotide unit of Oxa (Oxa-N) was incorporated into the 5′-end of probe DNA oligomers with or without the spacers such as –(CH)– and –(CH)– between Oxa-N and the probe sequence, and utilized to immobilize the probes onto the NH-functionalized glass slide surface. For evaluating the performance of Oxa as a linker, two analyses such as immobilization efficiency and hybridization efficiency were performed. The former was estimated by measuring the fluorescence intensity of the spots for the probe DNA oligomers whose 3′-end was labeled with fluorescein and the latter by immobilizing the probe DNA oligomers without fluorescein-labeling and measuring the fluorescence intensity after the hybridization with the Cy3-labeled target DNA oligomers. Mild conditions were also explored as a new post-spotting treatment for the efficient attachment of probe DNA oligomer on the surface. The temperature and humidity of the post-spotting treatment were controlled and their effects on the performance of Oxa as a linker were investigated. Under the mild conditions of post-spotting treatment such as a temperature of 25°C and a RH of 75%, and 42°C and RH 50%, it was found that probe DNA oligomers with Oxa-N produce high efficiencies of immobilization and hybridization, and that longer alkyl chain spacers lead to better recognition of target DNA oligomers. In addition, the effects of some other parameters such as the time of post-spotting treatment and pH of spotting solutions were investigated to enhance the covalent bonding formation of Oxa with amino group on the surface. It was found that under the mild conditions of post-spotting treatment, the performance of Oxa as a linker increased linearly with the time of post-spotting treatment, but that pH of spotting solution did not give rise to the desirable improvement of the results although at higher pH of the spotting solution, the covalent bonding formation of Oxa with amino group on the surface was accelerated. The newly adopted mild treatment and other conventional post-spotting treatments were sequentially employed on the spots of probe DNA oligomer with Oxa-N, and the synergistic effect was found in the combination of the mild treatment (42°C and RH 50%) and UV irradiation with a total energy of 600 mJ. Consequently, Oxa has been found to be useful to covalently immobilize probe DNA oligomers onto the NH-functionalized glass slide by one-pot reaction under the mild conditions. For covalently immobilizing probe DNA oligomers by the conventional methods on the solid surface, additional activation steps for probe and surface are commonly employed, which require some additional steps such as stringent washing, etc. and produce undesirable by-products which reduce the immobilization efficiency of the spotted DNA probe and results in the low-hybridization efficiency. On the other hand, the currently developed system in which Oxa-N is employed as a linker and immobilization of probe DNA oligomers can be performed under mild conditions that requires no activation step and produces no by-products. Also the resulting high target recognition efficiency is notable. The incorporation of Oxa-N in the probe DNA oligomers as a linker would be employed as one of the advanced methods for the immobilization of the widely used NH-functionalized glass slides. Preparation of various types of DNA molecules with Oxa-N would also be a useful tool for the investigation of physiologically active DNA conjugates with other biological high molecules such as proteins, which is one of the hottest but the most difficult subjects in the current biochemistry and molecular biology. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

spp. are Gram-negative colonic bacteria that harbor a number of mobile genetic elements (,). Mobile elements called conjugative transposons (CTns) play a significant role in the transfer of antibiotic resistance genes (,). Integration of the conjugative transposon CTnDOT into the chromosome requires the joined ends of the conjugative transposon (), an sequence in the chromosome, a host factor and the CTnDOT integrase protein (IntDOT) (,). Recently, we showed that IntDOT belongs to the tyrosine family of recombinases (), but the mechanism of integration and excision of CTnDOT appears to be different from that proposed for phage λ and other tyrosine recombinases (). Tyrosine recombinases utilize a topoisomerase I-like mechanism to mediate DNA rearrangements between two pairs of DNA strands (). The recombination reaction can be divided into two steps and involves two staggered single-strand exchanges between DNA partners on either side of a short DNA sequence called the core or crossover region. In the first step, the active site tyrosine cleaves one strand of each DNA partner at one end of the crossover region. The products of this step are a 3′-phosphotyrosyl linkage of the enzyme to the DNA and a free 5′-hydroxyl group (,). The 5′-hydroxyl group from one DNA strand mediates a nucleophilic attack on the phosphotyrosyl bond of its partner. The two strands are exchanged between the DNA partners forming a Holliday Junction (HJ) (,,). This intermediate is resolved in the second step of the reaction into recombinants by cleavage and strand exchanges between DNA partners at the other end of the crossover region (). Homology between DNA partners within the crossover region is an obligate condition for the recombination reaction. The generally accepted view has been that the homology requirement in λ Int-catalyzed recombination is linked to the strand-swapping and ligation steps of the reaction (,). The HJ is a central intermediate in recombination reactions mediated by tyrosine recombinases. However, this intermediate is extremely short-lived and usually does not accumulate during the reaction. Therefore, the kinetics of its appearance and disappearance have been difficult to study. In any recombination reaction that proceeds via a HJ, there are two possible reaction pathways: the HJ can be formed by cleavage and exchange of the top DNA strands first and then resolved by the second cleavage and exchange of the bottom DNA strand. Alternatively, the bottom strands can undergo recombination first, to generate a HJ which is later resolved by the top DNA strand exchange. Previous work has established that some tyrosine recombinase-mediated reactions, like FLP recombination (), do not have a defined order of strand exchange. On the other hand, the λ Int and Xer systems show strong preferences for cleavage and exchange of one strand first during recombination (,). DNA sequence analyses of and several sites yielded a consensus sequence TTTGCNNNNN (). The TTTGC motif is completely conserved suggesting it is important for recombination. However, the 5 bp non-homologous sequence varies from site to site and recombination can occur between partner sites where all 5 bp are different. Initially the borders of the 5 bp non-homologous sequence were assumed to define the putative staggered sites of cleavage. However, using cleavage assays, we recently showed that one of the cleavage sites is actually 2 bp to the left, between the T and G, in the region of homology (). The fact that one pair of cleavage sites lies within a region conserved in all sites could have implications for the mechanism of cleavage and ligation of the top strands. In this article, we show that the IntDOT recombination reaction proceeds through a HJ intermediate. We also show that there is a strong bias in the order of strand exchanges during the integrative recombination reaction. The top DNA strands are exchanged first. Using cleavage and ligation assays we show that IntDOT is able to cleave and ligate activated substrates in the presence of mismatches within the crossover region. We also report that the 2 bp region of homology between and an sequence on the left side of the crossover region is required for efficient recombination. Finally, we show that recombination occurs by sequential homology-dependent and homology-independent steps. The plasmids and oligonucleotides used in this work are described in Supplementary Data, Tables 2 and 3. The mutations were made in the 2 bp region of homology that is conserved in all known CTnDOT sites. Mutations were made using the Stratagene Quickchange Mutagenesis Kit as described by the supplier. Primers carrying the specific mutations are shown in Table 3, Supplementary Data. The regions of the plasmid were sequenced to confirm the presence of the desired mutation and to insure that there were no additional mutations. One of the DNA strands was 5′ end-labeled with P and purified using G-25 Amersham Biosciences columns. The DNA substrate was prepared by mixing the labeled strand and the unlabeled complementary strand, at a 1:5 molar ratio and annealing them in an annealing buffer (0.1 M KCl, 10 mM Tris–HCl pH 8) by heating to 90°C and cooling to 25°C at a rate of 2°C/min. The DNA substrates containing a nick in either the bottom or top DNA strand were prepared by mixing a labeled strand and two unlabeled oligonucleotides that were complementary to the labeled one, at ratios of 1:5:5, respectively and annealing them as described above. All suicide substrates were phosphorylated at the 5′ end at the cleavage position to prevent rejoining after strand exchange. The and substrates were then incubated in a 20 μl volume containing 30 mM Tris–HCl (pH 7.4), 2 mM DTT, BSA (70 μg/ml), 2.6% glycerol and 50 mM KCl. Proteins and DNA were added to the final concentrations: IntDOT, 70 nM; IHF, 40 nM; DNA, 2 nM; and plasmid DNA, 3 nM. Samples were incubated for 2 h at 37°C. All samples were loaded onto 1% agarose gels and subjected to electrophoresis. The gels were exposed to phosphorimager screens and the reaction efficiencies were quantified using a Fuji FLA-3000 phosphorimager and Fujifilm Image Gauge software (Macintosh v.3.4). Results were calculated where indicated by dividing a total amount of label present in the product band by the total amount of label present in the substrate and product bands, multiplied by 100 to give percent recombination. The intact substrates and substrates that contained a single-stranded DNA nick at the IntDOT cleavage site in the bottom strand were radiolabeled at the 5′ end. The recombination reactions were performed with DNA substrates using IntDOT as described above. Samples were electrophoresed through a 1.5% agarose gel in 50 mM NaOH and 1 mM EDTA, at 35 V for 16 h. The gel was then neutralized by soaking in 500 ml of 1 mM HCl for 1 h and dried on a vacuum slab drier and exposed to phosphorimager screens. The DNA substrates, labeled as indicated in the text, were recombined with in a double volume of the standard recombination reaction. The reaction was terminated by addition of 200 μl of phenol/chloroform/isoamyl alcohol mixture (25:24:1). The mixture was vortexed and spun in a microcentrifuge and the aqueous phase was transferred to a fresh tube. The DNA was ethanol-precipitated and re-suspended in a 50 μl 1× reaction buffer and digested with SspI endonuclease. Samples were electrophoresed through a 1% agarose gel at 90 V for 100 min. The gel was dried on a vacuum slab drier and exposed to phosphorimager screens. A double size standard recombination reaction was terminated by the addition of 0.1 vol of 10% SDS, followed by addition of 0.5 ml of A1 buffer (10 mM Tris–HCl pH 7.5, BSA 20 μg/ml, calf thymus DNA 20 μg/ml, 1% SDS). The sample was vortexed and incubated at 37°C for 4 min. Fifty microliters of 2.5 M KCl was added to the mixture which was placed on ice for 10 min. The precipitate was collected by centrifugation for 2 min at 4°C and re-suspended in 1 ml of cold B buffer (10 mM Tris–HCl pH 7.5, 1 mM EDTA, 100 mM KCl). The supernatant fraction was saved and the pellet was washed twice. The final precipitate was re-suspended in 1 ml of buffer containing 10 mM Tris–HCl pH 7.5, 1 mM EDTA, 100 mM NaCl, 10 mM MgCl and tRNA (20 μg/ml). DNA was precipitated by addition of 2.5 vol of 100% ethanol. The pellet was washed in 70% ethanol, air-dried, and re-suspended in 20 μl of water. The pellet and the supernatant were electrophoresed through a 1% agarose gel. The cleavage and ligation assays have been described previously () and are included in Supplementary Data. We developed an recombination assay based on the one previously described for λ Int (). The IntDOT recombination reaction occurs between a circular site on a supercoiled plasmid and a linear DNA. The product of the reaction is a linear DNA fragment containing both and sites (A). IntDOT was able to catalyze an efficient recombination reaction . Up to 45% of the labeled substrate underwent recombination. As expected, the linear product of the reaction migrated as a 3.6 kb linear fragment on an agarose gel. The recombinant should contain a unique SspI site. As shown in B, digestion of the recombinant with SspI resulted in the formation of two linear DNA fragments of the expected lengths (1.1 and 2.5 kb). Recombinant product should be formed by two sets of strand exchanges between and sites. The top DNA strands can be cleaved and exchanged first to form a HJ, followed by a second cleavage, exchange and ligation of the bottom DNA strands. Alternatively the bottom strands can undergo strand exchange first, to generate a HJ which is resolved by the top DNA strand exchanges. To determine whether CTnDOT recombination proceeds by ordered strand exchanges, we used suicide substrates that contained a nick in one DNA strand at one of the sites of IntDOT cleavage (). Introduction of the nick at a cleavage site should block recombination because the enzyme cannot form the phosphotyrosyl–protein intermediate. On the other hand, an intact strand can be cleaved and exchanged with its DNA partner. Inhibition of the second strand exchange by a nick should lead to the formation of a HJ with a single-strand crossover. Since one DNA partner is circular and the other is linear, a HJ should appear as an ‘α structure’ and would be expected to migrate in an agarose gel like a relaxed plasmid (A) (,). If the nick is introduced in the top strand, an α intermediate should be observed if the bottom strand is exchanged first. Alternatively, if a nick is introduced in a bottom strand, an α intermediate should be observed only if the top strand is exchanged first. If either top or bottom strand can be exchanged first, an α intermediate should be observed with both nicked substrates (A). We were unable to observe the product of recombination in an recombination reaction between supercoiled and labeled containing a nick in the top DNA strand (B, lane 2). To show that lack of a product in the reaction is due to the presence of a nick and not because the substrates were not fully hybridized, we performed reactions with the same suicide substrates that were previously incubated with T4 DNA ligase and ATP to seal the nick. The ligated substrate was able to form the product (B, lanes 5 and 6). In the recombination reaction with an substrate having a nick in the bottom DNA strand, we observed the formation of a product (B, lane 3) with the same electrophoretic mobility as the linear recombinant (B, lanes 3 and 4). The results of this experiment indicate that there is a bias in strand exchange during the recombination reaction and suggest that the top strand is exchanged first. In the recombination reaction with a suicide substrate, we expected to trap a HJ as an α structure intermediate. However, it was surprising to find that actual intermediate had the same electrophoretic mobility as the linear recombinant while an α structure would have migrated slower in an agarose gel. Digestion of an α structure with a restriction enzyme containing a single restriction site should form a nicked HJ (,), with expected migration mobility close to the linear DNA of the same size in agarose gel. Surprisingly, digestion of the product generated two DNA fragments 1.1 and 2.5 kb long which were the same size as SspI generated fragments from a complete recombination reaction (, lanes 1 and 3). This result was unexpected and suggested that after formation of a HJ by a first round of strand exchanges a second DNA cleavage occurs on the bottom strand of DNA that is not followed by DNA ligation. The structure of this product is shown in A. If IntDOT is able to perform a second cleavage on the bottom strand during the reaction with suicide substrate, the protein should remain covalently attached to the 3′-DNA end (A). Treatment of the reaction mixture with SDS/KCl should precipitate only proteins or DNA that is covalently bound to a protein (,). Recombination reactions with suicide and intact substrates were performed and the reactions were divided in half. In one sample, proteins were precipitated from the reaction mixtures and pelleted by centrifugation. The supernatant fractions were saved. The untreated sample and supernatant and pellet from the SDS/KCl-treated sample were then analyzed on an agarose gel. (B) The results of the experiment revealed that stable covalent DNA/protein complexes are formed during recombination reaction with suicide substrates. When intact substrates were used DNA precipitation was negligible (B, lanes 1, 2 and 3). We were able to precipitate DNA from the reaction mixture only when suicide substrates were used (B, lanes 4, 5 and 6). Proteinase K treatment was used to show that this precipitation was due to the covalent linkage of the protein to DNA (B, lanes 7, 8 and 9). Our results indicate that after formation of a HJ in the reaction with a suicide substrate, IntDOT is able to perform a second DNA cleavage in the bottom DNA strand that is not followed by DNA ligation. In order to determine if the product of the reaction with containing a nick in the bottom strand is a DNA molecule with a single-stranded crossover on the top strand, we used denaturing agarose gel analysis. We used intact substrates and substrates with a preexisting nick in the bottom DNA strand, where only one strand was radiolabeled. The recombination reactions were performed and each DNA strand was then separated on an agarose gel under denaturing conditions (C). As predicted, denaturation of the product of the reaction with intact yields a long single-stranded DNA, when either the bottom or the top strand was labeled (C, lanes 5 and 6). Denaturation of the product of the recombination reaction with an suicide substrate, in which the label was placed on the top strand, yielded long single-stranded DNA fragments of the same size as recombinant DNA (C, lane 8). When the bottom strands were labeled, we were unable to detect any product (C, lanes 7 and 9) indicating that the bottom DNA strand did not undergo strand exchange. To confirm that the product of IntDOT-mediated recombination reaction with a suicide is a DNA molecule with the top strand exchanged, we used 2D gel analysis. The recombination reactions were performed with bottom strand nicked substrates containing a label either on the top or on one of the bottom strands. After separation on a native agarose gel, the single strands of each fragment were separated by electrophoresis in the second dimension under denaturing conditions (A–C). If the top strand was labeled and if only the top DNA strand underwent recombination we would expect a long, ssDNA that migrated slowly in the second dimension on a denaturing agarose gel. By contrast, if either bottom strand was labeled we would expect short DNA fragments that migrated faster in the agarose gel, because the bottom DNA strand did not undergo recombination. Our analysis showed that denaturation of the product of these reactions released a long top strand (A) and short bottom strands (B and C). These results indicated that only the top strand was exchanged, whereas the bottom strand did not undergo strand exchange. Previously we demonstrated that the IntDOT top strand cleavage site lies in a region of homology between the T and G of the conserved TTTGCNNNNN sequence. Thus, both the cleavage and ligation reactions involving the top strand take place in an environment where base pairing can occur. In order to determine whether the homology at the left side of the CTnDOT crossover region is required for the recombination reaction to occur or, alternatively, if the reaction can occur if the sites have heterology on the left side, we performed recombination reactions using intact and substrates that contained a GC to CG or AT mutation in the region of homology. The reactions were performed with equimolar concentrations of labeled substrates. The efficiency of recombination reaction was calculated as described in Materials and Methods section. As shown in , when mutated sites ( or ) recombined with wild-type , integration was 1000 times less efficient than for the reactions with wild-type . A similar effect was observed for the reactions where mutated recombined with wild-type (, B). IntDOT was able to catalyze the recombination reaction between wild-type site and the site containing the AT mutation, but the efficiency of this reaction was severely depressed (). To exclude the possibility that the decrease in the recombination activity between wild-type and mutated site could be due to the lack of Int binding to the mutated sequence, not to the disruption of the homology between and , we tested recombination efficiency between and containing the same mutations in the crossover region [A (lane 4) and B (lane 4)]. As shown in , the recombination frequency between and containing the same mutation was similar to the wild-type recombination activity. We also performed competition assays in which an equimolar mixture of duplex wild-type labeled on the top strand and duplex mutated ( or ) labeled on the bottom strand were incubated with wild-type in the reaction mixture. Reaction products were digested with SspI and analyzed on an agarose gel. Digestion of the recombinant with the top DNA strand labeled should give only a 1.1 kb fragment that could be detected on the gel, while digestion of the recombinant containing the label on the bottom DNA strand should generate 2.5 kb detectable fragment only. The only labeled band was the 1.1 kb long DNA fragment (C, lanes 3 and 7). This indicated that only wild-type site was able to recombine with a wild-type site. Alternatively, when we incubated the wild-type labeled on the top strand and mutated labeled on the bottom strand with mutated , we were able to detect only the 2.5 kb long DNA fragment (C, lane 5). Similar results were obtained when GC sequence was changed to AT (C, lane 9). These results indicated that substrates could only recombine with carrying the same 2 bp sequence on the left side of the crossover region. From our data we conclude that 2 bp homology on the left side of the crossover region is critical for the IntDOT recombination reaction while the remaining 5 bp are not conserved. The cleavage and ligation assays allowed us to determine whether lack of base paring within the IntDOT crossover region affected those steps in the recombination reaction. To determine the cleavage rates of the wild-type DNA substrate and substrates containing mismatches introduced in the IntDOT crossover region, a phosphorothioate cleavage assay was used (). In this assay, the oxygen which forms a phosphotyrosyl bond with the catalytic tyrosine is replaced by sulfur. The cleavage reaction leaves behind a 5′ SH group. This reaction is essentially irreversible and the protein remains covalently attached to the 3′ end of the DNA (A). This product can be detected as a shift in migration distance of DNA due to attachment of the protein. We used cleavage substrates containing 1, 5 and 7 mismatches introduced in the CTnDOT crossover region (B). The yield of all those reactions was comparable with wild-type reaction efficiency (C). Results of the cleavage assay showed that the CTnDOT cleavage reactions can occur in the presence of mismatches. Studies of FLP recombination target sites (FRT) demonstrated that homology in the crossover region is required to successfully complete the ligation step of recombination reaction (). It was also shown that the efficiency of the ligation reaction of λ Int was sensitive to mismatches during DNA hairpin formation and very sensitive to the homology at the site of ligation during resolution of HJ (). Since it was previously shown that IntDOT can perform the recombination reaction when all five bases on the right side of the crossover region are different between two sites (), we expected that IntDOT could perform the ligation reaction without homology between DNA partners. We tested if IntDOT is sensitive to mismatches in an -nitrophenol (pNP) assay (,,). The ligation substrate, a pNP oligonucleotide, mimics the activated 3′-phosphotyrosine intermediate that is formed after substrate cleavage. The enzyme binds the DNA, forms a phosphodiester bond, and releases the pNP (A). We measured the ligation rates of the wild-type DNA substrate and the substrates containing 2, 5 and 7 mismatches introduced in the IntDOT crossover region (B). The ligation product can be detected by formation of a 44-base DNA product on a denaturing gel (,). In the experiments reported here, the amount of product was measured as a function of time. The results of these experiments are shown in C. IntDOT was able to perform ligation on the pNP substrates containing up to 7 mismatches. This reaction was less efficient than the wild-type reaction, although half of the substrate was ligated, which was ∼70% of the ligation efficiency of strands that were complementary to the partner site (C). We repeated the same experiment for λ Int and we found that this enzyme was not able to perform the ligation reaction on a similar activated substrate containing a single mismatch within the crossover DNA sequence (C). The ability of λ Int to ligate pre-activated substrate was reestablished after the homology was restored (data not shown). Our observation that the IntDOT-mediated ligation reaction can occur in the absence of base pairing, while other tyrosine recombinases require complete base pairing for ligation, indicates that the architecture of IntDOT catalytic site has significant differences from the other well-studied tyrosine recombinases. A molecular analysis of the mechanism of the IntDOT-mediated recombination reaction depends on the ability to block the reaction at intermediate steps. In tyrosine recombinase reactions, HJs are intermediates but do not usually accumulate because they are rapidly resolved to recombinants or substrates (,). We also did not observe a detectable amount of HJ formed during recombination (B). To determine if the CTnDOT reaction proceeds via a HJ intermediate, we needed to find a way to block the reaction at the intermediate step. Several modifications of reaction conditions that stop tyrosine recombinase reaction at the HJ step have been described. These include incorporation of a non-bridging phosphorothioate in the DNA at the site of cleavage () or use of small synthetic peptides that are known to trap HJs (). We were unable to detect HJs with these methods. We found that the introduction of a nick in either the top or bottom strand influenced IntDOT recombination (). A nick at the top strand inhibited the reaction without accumulation of an intermediate, while a nick at the bottom strand promoted the accumulation of an intermediate (). However, we demonstrated that the product is not a free HJ. After the first cleavage and strand exchange in the reaction with the nicked top strand, IntDOT performs a second cleavage on the bottom DNA strand which is not followed by ligation. The product of this reaction was a linear DNA molecule that contained two nicks in the bottom strand and the protein covalently attached to one of the 3′ DNA ends (). Other well-studied tyrosine recombinases do not form a similar product (), indicating that IntDOT recombination differs from other well-studied tyrosine recombinases reactions. It is known that some tyrosine recombinase-mediated reactions, like FLP recombination (), do not have a defined order of strand exchanges. On the other hand, λ Int and Xer systems show strong preferences for the initial cleavage and exchange of one particular pair of strands (,). We interpret our results to be strong evidence that CTnDOT integrative recombination proceeds by sequential DNA exchanges in which the top DNA strands in each site are cleaved and exchanged first. Most tyrosine recombinases display a strong requirement for sequence identity in the crossover region between DNA partners for completion of the recombination reaction (,,). For example Bauer () showed that the single base-pair mismatch adjacent to the cleavage site in the crossover region in the top strand of a λ site strongly inhibited the reaction. When both sites contained the homologous mutated base pair, recombination was restored. Kitts and Nash () demonstrated that heterology introduced on the right side of the crossover region provided a partial block of the reaction and lead to the accumulation of HJs. They explained this need for homology by a branch-migration model, which involves migration of the junction branch point between the sites of Int cleavage. After the strand exchange, the lack of base pairing between DNA partners provides an energetic barrier to the branch migration process and blocks the second exchange. Landy and coworkers (,,) proposed an alternative strand-swapping-isomerization model, in which homology is sensed during the formation of the new bases between DNA partners, after cleavage and prior to ligation. They showed that the first-strand exchange in λ recombination depends on complementary base pairing between DNA partners. They also demonstrated that λ Int ligation is affected by the presence of mismatches at the site of ligation. Studies with FLP and Cre recombinases showed similar requirements for homology (,) In some cases, cleavage and ligation reactions have been shown to occur in regions of heterology. The topological analysis of the recombination products from assays with supercoiled plasmids containing two heterologous FRT sites on the same molecule revealed that the Flp-mediated reaction can occur in the presence of heterology but the products are rapidly resolved back to substrates (). Sherratt and coworkers () used an Xer-mediated recombination system, which used two plasmids that contained mutated recombination target sites (). They found that formation of the HJ product occurred with some substrates that contained mismatched bases. Experiments monitoring recombination of synthetic HJs that contained varying junction positions showed that ligation occurs between sites with mispaired bases. Ligation was most efficient when the mispaired base was away from the cleavage point and least efficient when it was adjacent to the cleavage site (,). However, the conditions used for these experiments did not employ substrates or intermediates that are normal participants in the respective recombination reactions. Our previous analyses showed that CTnDOT sites contain a 2 bp region of homology on the left side and a 5 bp non-homologous region on the right side of the crossover sequence. This finding suggests that recombination mediated by IntDOT is different from reactions catalyzed by other well-studied tyrosine recombinases and proceeds by homology-dependent and homology-independent strand exchanges. We propose that the IntDOT recombination reaction proceeds by the following general steps. First an intasome composed of IntDOT, IHF (or the host factor) and DNA is formed. Four monomers of IntDOT bind to the four sites in and that flank the crossover sites. The complex captures and undergoes synapsis with a partner site. This step could be similar to synapsis by other tyrosine recombinases, and would not require homology in the crossover regions of and (,). During the first cleavage, strand exchange and ligation reaction, only two out of four monomers are active. Results described in this article strongly indicate that the first cleavage and ligation reactions occur in the conserved region of homology in the top DNA strands ( and ). This event appears to require homology at the left side of the crossover region because reactions with and substrates that contained GC to CG or AT mutations in the 2 bp region of homology are severely depressed (). It is possible that the homology dependence at this step occurs by the strand-swapping-isomerization model according to which homology is sensed during the formation of the new bases between DNA partners before the first ligation and formation of a HJ (). Alternatively, the homology could be required only for ligation of the exchanged strands after cleavage. The next step of the reaction would involve an isomerization of the HJ, where the complex changes conformation that activates the other two partner IntDOT monomers at the sites of the second strand exchange. Since the remaining five bases on the right side of the crossover region are not complementary, the isomerization step would be homology independent because movement of the DNA occurs in the absence of complementary base pairs. It is possible that formation of the two complementary base pairs on the left side is important for the proper isomerization of the junction. The second pairs of cleavage, strand exchange and ligation reactions would form the recombinant products. This step is homology independent as shown in this work and requires ligation of exchanged strands under conditions where they cannot form Watson–Crick base pairs with the new partner. The final step of the IntDOT-mediated recombination would be dissociation of the complex with the recombinant sites containing a five base region of heteroduplex DNA between the sites of strand exchanges. We demonstrated here that IntDOT cleavage and ligation reactions can occur in the presence of mismatches ( and ). Although ligation and cleavage assays were not performed in the context of recombination complexes we confirmed that IntDOT efficiently cleaves mispaired phosphorothioate substrates (). In the ligation assay, IntDOT was able to ligate DNA substrates in the presence of seven mismatches in the crossover region. In contrast, introduction of a single mismatch in a pNP substrate completely abolished λ ligation in a similar experiment (C). These findings emphasize drastically different properties of the two enzymes. IntDOT was classified as a member of the tyrosine recombinase family, although our previous results indicated that the catalytic core of the protein seems to have somewhat different organization than other well-studied tyrosine recombinases (). In this article, we demonstrate that the coordination by the strand exchange catalyzed by IntDOT appears to be distinct from ones used by other tyrosine recombinases such as λ Int. Our observations indicate that the first strand exchange in IntDOT-mediated recombination is a homology-dependent step, while second round of recombination occurs in the region of heterology where neither cleavage, exchange or ligation of the bottom DNA strand require homology. To our knowledge, IntDOT is the only tyrosine recombinase to use this type of two-step mechanism. It will be interesting to determine whether this feature is found in other recombinases encoded by other conjugative transposons. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

RNA editing is a process reported in a wide range of organisms from viruses to mammals and plants where it has different functions such as regulating gene expression, increasing protein diversity or reversing the effect of mutations in the genome (). RNA editing is defined as a site-specific modification of RNA molecules, occurring by nucleotide insertion/deletion, nucleotide substitution or nucleotide modification, usually by deamination of A to I or C to U. Occurring via several molecular mechanisms, different types of editing have been described generally involving a specificity factor (RNA or protein) that recognizes the editing site, an enzyme catalyzing the reaction and sometimes other accessory factors. RNA editing alters the sequence of different types of mRNAs but also tRNAs, rRNAs and small regulatory RNAs () (microRNAs). The number of site-specific editing sites varies considerably between organisms. While about a thousand have been reported in some early diverging land plants (), only a few are known in humans. In many cases, RNA editing is essential for correct production of the protein encoded by the RNA, such as in trypanosomatid and plant organelles or in humans, where this process is essential for the absorption of dietary fats in small intestine by producing the lipid-carrying protein apolipoprotein B48 (). In other cases, RNA editing modulates the functional properties of the encoded protein as in the case of the glutamate and serotonin receptors in the central nervous system (,). RNA editing events result in a single nucleotide polymorphism between genomic DNA and the corresponding RNA sequence. Partial editing is common and results in a mixed edited and unedited RNA population. The unpredictability of RNA editing and the possibility of editing frequency varying with tissue, development and environmental conditions have made it extremely difficult to effectively screen systematically for editing events or for mutants that are affected in the RNA editing processes. Computational approaches have suggested that editing is much more prevalent than previously realized, particularly in primates (), but few of the predicted sites have been experimentally verified. Previously used methods such as cDNA sequencing, primer extension or pyrosequencing are either too expensive, not sensitive enough or too labor intensive for high-throughput screens (). A one-step, high-throughput method that allows both the scanning of transcripts for new editing sites (without any prior knowledge of their nature or location) and the quantification of editing would greatly accelerate RNA editing research. We reasoned that techniques used in clinical and genetics research to detect mutations and determine allele frequency should be suitable for detecting new editing sites and quantifying editing. High resolution melting (HRM) of PCR amplicons is used as a closed-tube method for mutation scanning and genotyping that does not require probes or labeled oligonucleotides, and no purification step is needed (). A PCR is performed using a fluorescent double-stranded DNA dye that can be used in fully saturating conditions. The amplicon is analyzed by melting—the change in fluorescence caused by the release of the intercalating dye from a DNA duplex as it is denatured by increasing temperature is precisely monitored. The presence of heteroduplexes (containing one or more mismatches), that melt at lower temperature, alter the shape of the melting curve. Here, we report the successful adaptation of HRM analysis to scan transcripts for new editing sites and to quantify editing variability in different individuals under various conditions. Our model is the plant which displays a moderate frequency of C-to-U editing in mitochondrial and chloroplast transcripts, but the approach would be easily applicable to RNA samples from any organism and to any type of nucleotide substitution or modification editing that results in a different base being incorporated by reverse transcriptase. DNA was isolated from leaves of Arabidopsis Col-0 as described in Edwards (). Total RNA was extracted with an RNeasy minikit (Qiagen Pty Ltd, Clifton Hill, VIC, Australia) and treated with a DNA-free kit (Ambion, Austin, TX, USA). DNA-free RNA (3 µg) was reverse transcribed for 1 h at 50°C with the SuperScript III Reverse Transcriptase (Invitrogen Australia Pty, Mount Waverley, VIC, Australia) using random hexamers. PCR and RT-PCR products were cloned in the pGEM-T easy vector (Promega, Madison, WI, USA). The primers used to scan plastid transcripts are given in Supplementary Table S1. They allow the amplification of fragments of an average size of 500 bp (ranging from 350 to 1330 bp). The intercalating dye used was LCGreen Plus (Idaho Technology Inc., Salt Lake City, UT, USA). Amplicons were analyzed with the LightCycler 480 software package. First, an absolute quantification analysis was performed to check the amplification curve and the Crossing Point (CP) value. Then, a m calling analysis was done to generate melting curves representing the fluorescence signal (at 450–500 nm) with increasing temperature, melting peaks corresponding to the negative derivative (−d/d) of the fluorescence signal, and to calculate m values. PPE of RT-PCR products and determination of editing efficiency were performed essentially as described in Peeters and Hanson () except that the oligonucleotides used were fluorescently labeled at the 5′ end using 6-carboxyfluorescein (FAM) (Sigma Genosys, Sigma-Aldrich Pty Ltd, Castle Hill, NSW, Australia). PCR templates were generated using the primers listed in Supplementary Table S1. Unincorporated primers and nucleotides were removed using ExoSAP-IT (USB Corporation, Cleveland, OH, USA) treatment following the manufacturer's instructions. To scan transcripts for editing sites, we compared the melting curves of amplification products from genomic DNA (homoduplexes) to amplification products from a mix of genomic (gDNA) and cDNA (containing heteroduplexes if the cDNA has been reverse transcribed from an edited mRNA). The procedure is described in a. Our interest was to gain a complete description of the editing sites in chloroplast transcripts of the model plant as an important step towards characterizing editing factors in plants. To generate amplicons to be analyzed by HRM, primers originally designed to screen chloroplast transcripts for transcription, processing and splicing defects were used (Charles Andrés, Andéol Falcon de Longevialle, ALCB, Claire Lurin and IS, in preparation). This primer set comprises 320 pairs spanning the entire Arabidopsis plastid genome with an average amplicon size of ∼500 bp. Among these primer pairs, 21 couples flank all 28 of the known editing sites (). Four of these pairs flank multiple editing sites. All these pairs were tested and all 21 amplicons were flagged as containing heteroduplexes by the HRM assay, demonstrating the sensitivity of the technique (). Out of the full set of 320 amplicons, besides the 21 known to contain editing sites, 18 other amplicons were found to give different melting curve shapes from gDNA alone versus a mix of gDNA and cDNA. The amplification products were checked by gel electrophoresis and those presenting more than one band (5 out of the 18) were not considered further as multiple PCR products obviously interfere with the melting analysis. To confirm the remaining 13 candidates, we sequenced cloned RT-PCR products. For 6 candidates, no difference in the sequence was found indicating that they are false positives. The remaining 7 candidate amplicons contained single C to T changes consistent with RNA editing, covering six new sites not previously identified in (, ). One of these new sites was covered by two amplicons in the screen and detected in both. Five of the six new sites could be further confirmed by analysis of public EST databases that revealed sequences containing the same C/T polymorphisms. No publicly available ESTs exist for the sixth transcript, . The factors involved in RNA editing in plants are still elusive. Two pentatricopeptide repeat (PPR) proteins have been reported to be essential for the editing of specific sites in the chloroplast transcript of Arabidopsis (,), raising the possibility that this large family of RNA-binding proteins could constitute the specificity factors recognizing the sequence around the target C and recruiting the enzyme to catalyze editing (). We are undertaking extensive HRM screening of mutant lines lacking different proteins of the PPR family, to check for defects in RNA editing. One mutant, , with an insertion in the gene At1g05750 was found to be impaired in the editing of two sites in the transcripts and (). This mutant will be described more fully elsewhere (Charles Andrés, ALCB, Maricela Ramos Vega, Arturo Guevara-García, María de la Luz Gutiérrez-Nava, Araceli Cantero, Luis Felipe Jiménez, Claire Lurin, IS and Patricia León, in preparation). The site in is one of the six new sites discovered in the screen developed during this study. HRM has been used to detect mutations in heterogeneous DNA populations and proved reasonably sensitive (). We decided to adapt this type of sensitivity test to quantify editing (the procedure is shown in b and differs primarily from the screens described previously in that the cDNA is not mixed with genomic DNA before amplification and that much shorter amplicons are required for the best results). We first calibrated the test with mixes of plasmid DNA containing a range of ratios of C to T at known editing sites. In short amplicons of 70 bp, as little as 2.5% T was easily detected and melting peak profiles for amplicons differing by only a few percent in C/T could be distinguished (). In plant organelles, the extent of RNA editing of some sites can vary according to the genotype, the tissue or in different growth conditions (,). No systematic survey has been published so far concerning such changes, which may well have physiological relevance. To gauge the efficacy of the HRM assay for such a survey, we prepared cDNA samples from different genotypes, different organs, or from plants grown in different conditions. The cDNAs were amplified, subjected to HRM analysis and their melting peak profiles were compared to the ones from plasmid mixes with increasing ratios of T as compared to C. Five primer pairs flanking four different sites in the transcripts , , and were tested. The results obtained with the HRM assay were compared to those obtained by a widely used, but much more labor-intensive poisoned primer extension (PPE) assay (). In all, 24 comparisons were made (6 samples × 4 editing sites). In 18 of the 24 comparisons, the editing efficiency measured by HRM matched that measured by PPE (). In the cases where the two approaches gave different values, the values were reasonably close, generally within 10%. Following optimization of the procedures, we employed HRM to analyze chloroplast transcripts from , already extensively analyzed by previous groups (,). This screen covered a total of 126 kb of chloroplast transcripts in four amplification plates and took only 4 days to complete. The screen detected 34 editing sites in total, including all 28 known sites, whilst discovering 6 new editing sites ( and ). Four of these sites in are coding sequences (, , , ) and change the protein sequence translated from these mRNAs. The other two sites are the first to be identified in chloroplast non-coding sequences (in the 3′ UTR of and the intron of ). Despite the sensitivity of the screen, only 11 false candidates needed to be eliminated (a false positive rate of 11/292 or 3.8%), and 5 of those could be ruled out simply by electrophoresis of the amplicons. This high sensitivity coupled with a low false positive rate, the simplicity of use and the affordability of the approach (6 times cheaper per data point than bulk sequencing of RT-PCR products) make HRM screening an extremely attractive new tool for studying RNA editing. We demonstrated that this method is highly suitable for systematically screening mutants for a defect in editing by discovering that the mutant fails to edit two sites, in and . Furthermore, using small amplicons, HRM is sensitive and accurate enough to detect < 2.5% editing and to easily quantify partially edited sites, the editing extent of which can vary in different genotypes, different organs and different conditions. Previous quantification techniques such as poisoned primer extension or pyrosequencing were much more labor intensive, low throughput and prohibitively expensive for large-scale surveys. One of the few drawbacks of the approach is that HRM analysis alone cannot easily detect how many editing sites are present, or where within the amplified region the editing site is positioned. For this, sequencing of the amplified product is required. In conclusion, we have demonstrated that a method originally designed to detect DNA mutations and genotype individuals in clinical research and diagnostics can be simply adapted to research on RNA editing. Currently research in this area is limited by the lack of cheap, effective approaches for screening for new editing sites or for mutants affected in the editing process. The approach described here can be simply and directly applied to samples from any organism, so this breakthrough should stimulate research in many laboratories. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Werner syndrome (WS) is an autosomal recessive disorder characterized by an early onset of age-related pathologies including graying hair, alopecia, arteriosclerosis, osteoporosis, diabetes mellitus and cancer (). The gene mutated in WS, , encodes a RecQ-type DNA helicase (,). WRN also possesses a 3′ - 5′ exonuclease activity residing in a separate domain located at the N-terminus of the protein (,). At the cellular level, WRN deficiency is associated with defects in DNA replication, homologous recombination (HR) and telomere maintenance (). As a result, cells derived from WS patients display a high degree of genomic instability including elevated levels of chromosomal translocations and deletions (). WS cells are hypersensitive to DNA-damaging agents such as 4-nitroquinoline 1-oxide, topoisomerase inhibitors and DNA cross-linkers, suggesting that WRN is actively involved in DNA repair (). Several lines of evidence implicate WRN in the cellular response to DNA double-strand breaks (DSBs). WRN is rapidly recruited to the sites of ionizing radiation (IR)-induced damage (). Moreover, it interacts physically and functionally with a number of proteins that are involved in the DSB-repair process including the MRE11-RAD50-NBS1 (MRN) complex (), the Ku complex (), RAD52 () and DNA-dependent protein kinase (). However, the precise role for WRN in DNA repair remains to be elucidated. The DNA mismatch-repair (MMR) system maintains genomic integrity by correcting DNA replication errors and preventing recombination between divergent sequences (,). Defects in a subset of MMR genes including , , , and are associated with hereditary non-polyposis colon cancer, highlighting the crucial role for MMR in genome maintenance (). In the initiation step of the eukaryotic MMR process, at least three heterodimers, namely MSH2/MSH6 (MutSα), MSH2/MSH3 (MutSβ) and MLH1/PMS2 (MutLα), are involved (). MutSα binds to base-base mismatches and short insertion/deletion loops, while MutSβ can recognize only insertion/deletion loops containing up to 16 extra nucleotides in one strand (). MutLα possesses an intrinsic endonuclease activity, which is activated upon mismatch recognition and introduces incisions in the discontinuous strand of the heteroduplex DNA, generating entry sites for the 5′-3′ exonuclease EXO1 (). Sgs1, the yeast ortholog of WRN, also contributes to the suppression of recombination between divergent DNA sequences (). Heteroduplex rejection during repair of DSBs by the single-strand annealing pathway of HR in yeast requires the mismatch binding and ATPase functions of the Msh2p/Msh6p heterodimer and the helicase activity of Sgs1 (,). These findings led to the proposal that MMR proteins act in conjunction with Sgs1 to unwind DNA recombination intermediates containing mismatches (,). Here we demonstrate that WRN directly interacts with MutSα, MutSβ and MutLα via distinct domains. MutSα and MutSβ are found to stimulate WRN-mediated unwinding of forked DNA duplexes with a 3′-single-stranded (ss) arm. The stimulatory effect of MutSα on WRN-mediated unwinding is enhanced by a single G/T mismatch located in the duplex ahead of the fork in a manner independent of MutLα. These data provide biochemical evidence suggesting that the rejection of homeologous recombination by MMR proteins occurs helicase-mediated unwinding of recombination intermediates. The bacterial expression vectors for the WRN fragments encompassing the amino acid residues 51–449, 949–1432, 500–946, 500–1149, 500–1236, respectively, fused to the C-terminus of glutathione -transferase (GST) were constructed by PCR amplification of corresponding regions of the WRN cDNA and their insertion in pGEX-2TK (Amersham Biosciences) between the EcoRI and BamHI sites. The complete coding region of WRN was amplified by PCR and cloned in pACT2 (Clontech Palo Alto, CA) via SmaI site to construct a yeast two-hybrid (YTH) vector expressing WRN as a fusion with a Gal4 activation domain. MLH1 cDNA comprised of the codons 500–756 was cloned in a YTH vector pBTM116 (Clontech Palo Alto, CA) between the EcoRI and SalI sites, resulting in a fusion with a LexA DNA binding domain. The pBTM16 derivatives expressing other MLH1 variants as well as the full-length yMlh1 were previously described (). The Gal4-hMSH2 (pLJR105), LexA-MSH3 and LexA-MSH6 bait plasmids were also described previously (,). Recombinant human WRN (,), MutSα (), MutSβ () and MutLα () and MutS () were produced and purified as previously described. An antibody against the N-terminal region of WRN encompassing amino acids 1–391 (ISEV-391) was raised in rabbit and purified on an antigen-coupled Sepharose 4A column (Amersham Biosciences). Control IgGs were purified from a rabbit preimmune serum on a 5 ml HiTrap protein G-Sepharose column (Amersham Biosciences). The following human cell lines were used in this study: HEK 293 embryonic kidney cells and AG11395 SV40-transformed WS fibroblasts (Coriell Institute for Medical Research). The HEK 293 cells were maintained in DMEM (Gibco) supplemented with 10% fetal calf serum (Biochrome AG). The WS cells were maintained in MEM containing 15% fetal calf serum and 2 mM -glutamine. Cells were suspended in lysis buffer containing 20 mM Tris–HCl (pH 7.5), 150 mM NaCl, 2 mM EDTA, 0.1% (v/v) Triton X-100, 10% (v/v) glycerol and complete, EDTA-free protease inhibitor cocktail (Roche). After sonication, the suspension was centrifuged at 20 000 for 30 min at 4°C. Aliquots containing 1.6 mg of protein were incubated overnight at 4°C with purified rabbit polyclonal anti-WRN IgGs (2 μg), which was followed by a 2-h incubation with protein A/G-agarose beads (Santa Cruz) at 4°C. Where required, extracts were treated with 50 U of DNaseI (Roche) for 30 min at 25°C prior to addition of antibody. After extensive washing with the lysis buffer, the immunoprecipitates were subjected to electrophoresis in a 7.5% polyacrylamide–SDS gel followed by western blotting. The blots were probed with mouse monoclonal antibodies against WRN (BD Biosciences, 611169), MSH6 (Pharmingen, clone 44), PMS2 (Pharmingen, clone 16-4), MLH1 (BD Biosciences, 554073), MSH2 (Calbiochem, clone NA 26) and MSH3 (Transduction Laboratories, clone 52). Immune complexes were detected using ECL-plus reagent (Amersham Biosciences), with horse anti-mouse IgG-horseradish peroxidase conjugate (Vector) used as a secondary antibody. In the control experiment, IgGs purified from a preimmune rabbit serum were used instead of the anti-WRN antibody. GST–WRN fusion proteins were produced in the BL21-CodonPlus(DE3)-RIL strain (Stratagene) using the expression vectors described above. The fusion proteins were bound to glutathione–sepharose beads (Amersham Biosciences) as previously described (). The beads were incubated with 1 μg of purified MutSα, MutSβ or MutLα in 400 μl of NET-N 100 buffer [10 mM Tris HCl (pH 8.0), 1 mM EDTA, 100 mM NaCl, 0.5% (v/v) NP-40] for 2 h at 4°C. After extensive washing with NET-N 100 buffer, proteins bound to the beads were analyzed by western blotting. Membranes were probed with the monoclonal antibodies described above. MSH3 was detected with a rabbit polyclonal antibody (NTH3) raised against its first 200 amino acids (Eurogentec). In a control experiment, beads were coated with GST protein only. Purified recombinant WRN was diluted to a concentration of 20 nM in carbonate buffer [16 mM NaCO, 34 mM NaHCO (pH 9.6)] and added to wells of a 96-well microtiter plate (50 µl/well). Plates were incubated overnight at 4°C. For control reactions, wells were pre-coated with an equivalent amount of bovine serum albumin (BSA). After aspiration of the samples, the wells were blocked with blocking buffer [phosphate-buffered saline, 0.5% (v/v) Tween-20 and 3% (w/v) BSA] for 2 h at 37°C (200 µl/well). Following blockage, the wells were incubated with increasing concentrations of purified recombinant MutSα, MutSβ and MutLα proteins for 1 h at 37°C. All samples were supplemented with ethidium bromide (EtBr) at a concentration of 50 µg/ml to prevent DNA-mediated interactions. Wells were washed four times with blocking buffer to eliminate unbound proteins, and incubated with the appropriate primary antibody diluted in blocking buffer (mouse monoclonal anti-MSH2 antibody for MutSα and MutSβ, and mouse monoclonal anti-MLH1 antibody for MutLα). Plates were incubated for 1 h at 37°C. After four washings with blocking buffer, horseradish peroxidase-conjugated anti-mouse secondary antibody (1:10 000 in blocking buffer) was added and the plates were incubated at 37°C for 30 min. After extensive washing with blocking buffer, the protein complexes were detected using o-Phenylenediamine dichloride (Sigma) dissolved in 0.1 M citrate-phosphate buffer (pH 5.0) containing 0.03% hydrogen peroxide (1 mg/ml). The reactions were terminated after 5 min by adding 50 μl of 2 M HSO. The plates were scanned in a microplate reader (Molecular Devices) for absorbance at 492 nm. The A values, corrected for background signal in the presence of BSA, were plotted as a function of the concentration of appropriate MMR protein using the GraphPad Prism software. To determine an apparent dissociation constant of each complex (K), the data points were fitted by the hyperbolic function = B*/(K + ) where B is the maximal binding and K is the concentration of ligand required to reach half-maximal binding. YTH analysis was carried out using strains L40 (a t) and Y190 (α, , , , --). The former strain was used for LexA-bait vectors while the latter strain was used for Gal4-bait vectors. Clones carrying the bait and prey plasmids were tested for β-galactosidase activity using a pellet X-gal (PXG) assay as previously described (). Schemes of DNA substrates as well as the sequences of the constituent oligonucleotides are summarized in Supplementary Table 1. The f11-20 oligonucleotide (50-mer) was labeled at the 5′-end using T4 polynucleotide kinase (NEB) and [γ-P]ATP (Amersham Biosciences), and annealed with appropriate oligonucleotides under previously described conditions (). The helicase reaction mixtures (10 μl) contained 50 mM Tris–HCl (pH 7.5), 50 mM NaCl, 2 mM MgCl, 50 μg/ml of BSA, 2 mM ATP, 1 mM DTT, 1 nM DNA substrate and indicated concentrations of MMR proteins (MutSα, MutSβ, MutLα and MutS) and WRN. The MMR proteins were pre-incubated with the DNA substrate on ice for 1 min prior to the addition of WRN. The reactions were incubated at 37°C for 30 min and terminated by the addition of 0.5 reaction volume of buffer S [150 mM EDTA, 2% (w/v) SDS, 30% (v/v) glycerol, 0.1% (w/v) bromophenol blue] followed by treatment with proteinase K (0.1 mg/ml) at 37°C for 10 min. The reaction products were resolved on a non-denaturing 10% polyacrylamide gel [acrylamide to bis-acrylamide, 19:1 (w/w)] run in 1xTBE buffer at 140 V. Radiolabeled DNA species were visualized by a phosphorimager and quantified using ImageQuant software (Molecular Dynamics). Based on genetic studies in yeast, it has been suggested that the proteins involved in the initiation of MMR act in conjunction with RecQ DNA helicases to eliminate DNA recombination intermediates containing mismatches, which can give rise to chromosomal rearrangements (,). We sought to test the validity of this model biochemically using the WRN helicase, one of the five RecQ homologs identified in human cells, whose dysfunction results in chromosomal translocations and deletions. First, we performed an ELISA-based protein-binding assay to investigate whether WRN and the MMR proteins interact physically. Increasing concentrations of purified MutSα (MSH2/MSH6), MutSβ (MSH2/MSH3) and MutLα (MLH1/PMS2) proteins (A) ranging from 0 to 80 nM were incubated in wells that had been pre-coated with WRN at a concentration of 20 nM and subsequently blocked with BSA to prevent non-specific interactions. After extensive washing, the bound MMR proteins were incubated with specific antibodies followed by a colorimetric assay to quantify the binding. In control experiments, MMR proteins were incubated in wells pre-coated only with BSA. We found that all the three MMR heterodimers were specifically bound to WRN-coated wells in a dose dependent manner, indicating a direct interaction (B–D). Interestingly, the apparent dissociation constant of the MutSβ–WRN complex (K = 8.8 nM) was much lower than that estimated for the MutSα–WRN complex (K = 38.5 nM). The dissociation constant of the MutLα–WRN complex (K = 34.9 nM) was similar to that of the MutSα–WRN complex. To test whether WRN and MMR proteins form a stable complex , we immunoprecipitated WRN from extracts of exponentially growing human embryonic kidney cells (HEK 293) and subjected the resulting immunoprecipitate to western blot analysis. This immunoprecipitate was found to contain the MLH1 and PMS2 proteins, components of the MutLα complex, but not the MSH2 and MSH6 proteins, which form the MutSα heterodimer (A, lanes 3 and 4). None of these MMR proteins were detected in the immunoprecipitate obtained with control IgGs (A, lane 2). To exclude the possibility that the observed association of WRN with MLH1 and PMS2 results from independent binding of these proteins to DNA, we pre-treated the cell extracts with DNaseI. We found that this treatment did not alter the level of the MMR proteins in the WRN immunoprecipitate, suggesting that the WRN–MLH1–PMS2 complex is mediated by protein–protein interactions (A, compare lanes 3 and 4). Furthermore, we did not detect PMS2 in an immunoprecipitate obtained with anti-WRN antibody from extracts of the WS cell line AG11395, excluding the possibility that the observed co-immunoprecipitation of MMR proteins with WRN is due to cross-reactivity of the antibody (B). Collectively, these data indicate that MutLα but not MutSα, forms a stable complex with WRN . To identify the MutSα, MutSβ and MutLα-interaction sites on WRN, we performed affinity pull-down assays using a series of WRN fragments fused to GST. These fragments covered the entire WRN polypeptide except for the first 50 amino acids and the region spanning the amino acids 450–499 (A and B). The GST pull-down experiments revealed that the MutSα interaction site on WRN was localized to the region between amino acids 500 and 946, which constitutes the helicase core of WRN composed of the DExH helicase and Zn-binding domains (A and C). MutSβ was found to make contacts not only with the helicase core of WRN, but also with a region spanning amino acids 947–1149 that contains the winged-helix (WH) motif, a common interaction site for most of the WRN partners identified thus far (A and C) (). Notably, the binding affinity of MutSβ to the WH domain of WRN appeared to be much higher that its binding affinity to the helicase core of WRN (C, compare lanes 4–7). The data also indicated that MutSβ binds to WRN more efficiently than MutSα (C, top and middle panels; compare lane 1 with lanes 4–7), which is in agreement with the results of the ELISA assay (). MutLα was found to interact with the helicase core of WRN and with the N-terminal portion of WRN including the exonuclease domain, showing a higher binding affinity to the former domain (A and C, bottom panel; compare lanes 3–7). To identify the subunits of MutSα, MutSβ and MutLα that mediate the interaction with WRN, we performed a quantitative YTHassay with the full-length WRN as prey. The following interactions were examined: MSH2–WRN, MSH6–WRN, MSH3–WRN and MLH1–WRN. We found WRN to interact with MLH1 and MSH2, but not with MSH3 and MSH6 (A and B). This indicates that the MutSα–WRN and MutSβ–WRN interactions are mediated by MSH2, and the MutLα–WRN interaction is mediated by MLH1. However, the inability of MSH3 and MSH6 to interact with WRN in the YTH assay could be a consequence of the fact that these proteins are not soluble when expressed alone (). This is also true for PMS2 (). Therefore, the possibility still exists that these proteins could make additional contacts with WRN. This is particularly likely in the case of MSH3, since our GST pull-down experiments revealed that MutSβ interacts with both the helicase core and the WH domain of WRN, whereas MutSα interacts only with the helicase core of WRN (). In order to identify the WRN interaction domain on MLH1, we tested a series of MLH1 deletion variants for the ability to interact with the full-length WRN in YTH assay. We found that this domain is located at the C-terminus of the MLH1 polypeptide between amino acids 500 and 756 (A and B). This is different from the location of the BLM-interaction site that was mapped to the region spanning amino acids 396–500 (). Next, we tested MutSα for the ability to affect the helicase activity of WRN on DNA substrates containing mismatches. In these experiments, we used a synthetic DNA duplex (49 bp) with a 3′-ss flap (19 nt) resembling a part of the structure that results from annealing of the resected arms of a broken chromosome at regions of homology. On such forked DNA structures, WRN preferentially translocates along the 3′-flap oligonucleotide to unwind the duplex ahead of the fork junction, generating a 3′-tailed duplex. This primary product can be further unwound by WRN into the component strands, as a consequence of loading of a second helicase molecule on the 3′-ssDNA tail (). We prepared a fully matched substrate and a substrate containing a single G/T mismatch located 11 nt ahead of the ss/ds junction (A and B, top panels). WRN alone displayed a very low helicase activity on both structures when present at the same concentration as the DNA substrate (1 nM). However, the helicase activity of WRN on these structures dramatically increased upon inclusion of an 8-fold molar excess of MutSα in the reaction (A–D). Notably, the initial rate of the MutSα-stimulated unwinding reaction with G/T substrate was about 1.7 times higher than that measured with the G/C substrate (Supplementary Table 2). Since MutLα is known to bind to MutSα–heteroduplex complexes (), we investigated whether it can affect the WRN-mediated unwinding of G/T and G/C substrates induced by MutSα. We found that MutLα did not significantly alter the MutSα-dependent helicase activity of WRN on these DNA structures (). Also, it had no effect on WRN-mediated unwinding in the absence of MutSα (, lane 5). To further assess the effect of MutSα on WRN-mediated unwinding of the 3′-flap duplex, we performed a protein titration experiment, in which we varied the concentration of MutSα while keeping WRN and DNA substrate at a fixed concentration of 1 nM. We found that MutSα stimulated the helicase activity of WRN in a concentration-dependent manner, exhibiting a significantly higher activity on the G/T substrate than on the homoduplex substrate (). To explore the specificity of the observed stimulatory effect, we tested human MutSβ as well as MutS for the ability to stimulate DNA unwinding by WRN. We found that MutSβ enhanced the WRN-mediated unwinding of the 3′-flap DNA duplex to a similar extent as seen with MutSα (, compare lanes 2 and 3). In contrast, the MutS protein did not enhance the WRN-mediated DNA unwinding (, lane 8), indicating that the observed stimulatory effect is specific to human MutS homologs. As in the case of MutSα, MutSβ-stimulated helicase activity of WRN was not influenced upon addition of MutLα (, compare lanes 4 and 7) and it was dependent on MutSβ concentration (Supplementary Figure S1). We also found that in the presence of MutSβ, WRN unwound the G/T substrate with the same efficiency as the homoduplex substrate (Supplementary Figure S2). This is consistent with the fact that MutSβ does not bind to base–base mismatches () and supports the conclusion that the observed stimulatory effect of the G/T mismatch on MutSα-dependent unwinding of the 3′-flap duplex by WRN results from the specific binding of MutSα to the mismatch. To gain further insights into the mechanism underlying the stimulation of the helicase activity of WRN by MutSα and MutSβ, we investigated the dependence of this reaction on the configuration of the arms of the fork. Using the same set of oligonucleotides, we prepared the following substrates: a forked duplex with both arms single stranded (splayed arm); a forked duplex with the 3′-arm single stranded and the 5′-arm double stranded (3′-flap duplex); a forked duplex with the 3′-arm double stranded and the 5′-arm single stranded (5′-flap duplex) and a forked duplex with both arms double stranded. Earlier studies revealed that WRN could unwind efficiently all these structures, indicating that it does not require the 3′-arm to be single stranded for loading at the fork (). We found that MutSα and MutSβ strongly stimulated the WRN-mediated unwinding of the splayed arm and the 3′-flap duplex, but had no significant effect on the unwinding of the 5′-flap duplex and the fully double stranded fork (). We also tested these proteins for the ability to stimulate the helicase activity of WRN on 3′-ssDNA-tailed duplex, which is normally a poor substrate for WRN (). We found that neither MutSα nor MutSβ could activate WRN for unwinding of this partial DNA duplex (data not shown). Interestingly, the 3′-tail duplex resulting from unwinding of the 3′-flap structure was unwound by WRN efficiently. This discrepancy can result from the fact that WRN exists as an oligomeric structure, which would facilitate loading of a second molecule of WRN on the 3′-ssDNA generated by unwinding of the duplex ahead of the fork. Although WRN has been implicated in a number of DNA repair processes, the exact DNA transactions mediated by this helicase/exonuclease in the cell remain elusive. Here we show that WRN interacts physically with proteins that are involved in the initiation of MMR and the rejection of recombination between divergent sequences. Most importantly, our experiments revealed that MutSα and MutSβ can stimulate the helicase activity of WRN on forked DNA structures with a 3′-ss arm that resemble intermediates of single-strand annealing pathways of HR. In addition, we found that a single G/T mismatch located ahead of the fork junction increased the efficiency of the MutSα-dependent unwinding by WRN. These data are consistent with a model in which the MMR initiation factors prevent homeologous recombination by activating a DNA helicase for unwinding of recombination intermediates containing mismatches. This model was proposed earlier on the basis of results of heteroduplex rejection assays with yeast and mutants (,). It is possible that MutSα and MutSβ bind to mismatches formed after pairing of sequences of imperfect homology and, following ATP binding, are converted into a DNA sliding clamp as proposed in the case of the MMR pathway (). When the clamp encounters the junction between the heteroduplex and the non-homologous 3′- tail, it binds stably to it and recruits a DNA helicase to disrupt the joined DNA molecule. In agreement with this hypothesis, it has been demonstrated that yeast MutSβ specifically binds to forked DNA structures containing 3′- ssDNA making contacts with the sequences at the ds–ss junction (). These studies also revealed that MutSβ holds the junction in an altered, perhaps more rigid, conformation (). Such structural changes could facilitate the loading of the WRN helicase on the 3′- ssDNA at the junction, which is a prerequisite for duplex unwinding to occur. However, it should be noted that the MutSα-activated unwinding of a 3′-flap duplex by WRN displayed only a moderate dependence on mismatches. It is therefore possible that hereroduplex rejection involves some additional factors that ensure mismatch specificity of this transaction. In our studies, we did not observe any significant modulation of WRN-mediated unwinding by MutLα, even in the presence of MutSα or MutSβ. In agreement with this finding, the yeast Mlh1 and Pms1 proteins have been shown to have only minor roles in the rejection of homeologous recombination relative to the contributions of Msh2 and Msh6 (). Thus, it appears that the physical interaction between WRN and MutLα identified in this study has some other functional implication. Interestingly, MLH1 was shown to interact with various DNA repair factors including MRE11, BACH1, MBD4 and BLM (,). It is, therefore, possible that MLH1 plays a more general role in DNA repair processes. A number of other functional implications for the observed interactions between WRN and the MMR factors can be discussed. Several lines of evidence suggest that WRN promotes replication of telomeric DNA by unwinding G-quadruplex structures that can readily form in G-rich telomeric DNA and impose a barrier for progression of DNA replication forks (,). Strikingly, human MutSα has been shown to bind efficiently to G-quadruplex DNA (). Moreover, Msh2 deficiency in mice is associated with loss of telomeres and an elevated level of telomere end-to-end fusion, a phenotype similar to that manifested by WRN-deficient cells (,). Thus, one can speculate that the MMR proteins can mediate recruitment of the WRN helicase to G-quadruplex structures formed at telomeres and hence facilitate their removal. It has been shown that the human MutSβ and WRN are required along with PCNA, RPA and ERCC1-XPF for uncoupling of psoralen-induced inter-strand DNA crosslinks (ICLs) in cell-free extracts, suggesting a novel ICL-repair pathway in which MutSβ is essential for the recognition of ICLs, while the WRN helicase mediates unwinding of the DNA duplex adjacent to the lesion, which enables strand incision by ERCC1-XPF (,). Our finding that MutSβ physically interacts with WRN and stimulates its helicase activity brings further support for this model and suggests that MutSβ might recruit WRN to the ICL sites. Earlier studies demonstrated that nuclear extracts from several fibroblastoid cell lines derived from WS patients were deficient in repair of base–base mismatches and insertion/deletion loops, suggesting that WRN could have a role in MMR (). However, it is not certain that the MMR-deficiency of these extracts was caused solely by WRN deficiency because complementation experiments with recombinant WRN protein were not performed in this study. Moreover, in some cases, pair-wise mixing of these extracts restored MMR proficiency, making the involvement of WRN in MMR rather questionable. Recently, two other human RecQ homologs, namely RECQ1 and BLM, have been shown to interact physically and functionally with the MMR-initiation factors (,,,,). As in the case of WRN, MutSα was found to stimulate RECQ1-mediated unwinding of a forked structure with a 3′-ss arm (). In contrast, MutSα did not affect unwinding of forked DNA duplexes by BLM (). Instead, MutSα was found to stimulate the ability of BLM to process Holliday junctions (). Further studies will be needed to fully understand the molecular mechanisms by which the RecQ helicases and MMR factors work together to maintain genomic stability. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

RNA interference (RNAi) was initially discovered in by Fire and colleagues, who showed that introduction of long double-stranded RNA (dsRNA) caused a nearly complete inhibition of genes harboring the same sequence (). It was subsequently demonstrated that short ∼21 bp dsRNAs, termed small interfering RNA (siRNA), were functional triggers of RNAi without inducing the innate immune responses associated with longer dsRNA in mammalian cells (). Natural siRNAs are processed from longer dsRNA species derived from, e.g. virus, mobile elements or transgenic RNA by the cytoplasmic RNAse III enzyme Dicer (). Similarly, exogenous 19–27 bp siRNAs are functional if introduced into the cytoplasm (). Here, the siRNA will be incorporated into the RNA-induced silencing complex (RISC) by a RISC loading complex (RLC), which is best described in (), but likely also exists in humans (). By sensing the thermodynamic asymmetry of siRNA duplex ends, RLC distinguishes the siRNA guiding antisense strand from the sense strand, thereby dictating the so-called pre-RISC to assemble asymmetrically on the siRNA duplex (,). Although both strands of the siRNA duplex are initially incorporated into pre-RISC, the RLC-tagged sense strand is subsequently cleaved and released thereby establishing activated RISC which contains only the single stranded antisense strand. Recent data suggest that the catalytic core protein of RISC, the Ago2 endonuclease, initiates sense strand elimination by cleaving it 9 nt from its 5′ end during RISC activation (). Although the helicase activity for unwinding the duplex remains unidentified, these events expose the antisense strand in RISC to the mRNA target, which is subsequently cleaved probably by a similar mechanism. The use of synthetic siRNAs is currently hampered by lack of efficient means of siRNA delivery, low biostability in biological fluids and low specificity of action due to inherent gene off-target effects caused by the microRNA-like behavior of all investigated siRNAs (). Several attempts to reduce off-target effects through chemical modification of synthetic siRNA have been made (,). Since both strands of a siRNA duplex can contribute to off-target effects (), minimizing sense strand incorporation into activated RISC should significantly increase targeting specificity. It is well established that the siRNA strand with the thermodynamically least stable 5′ end is preferentially utilized as antisense strand in activated RISC (,). Accordingly, selective thermodynamic stabilization of sense strand 5′ ends by incorporation of locked nucleic acids (LNA) has been shown to reduce unwarranted gene silencing by the sense strand (,). Here, we apply a radically different design characterized by an intact antisense strand complemented with two shorter 9–13 nt sense strands, together named mall nternally egmented nterfering RNA (sisiRNA, ). We show that only the antisense strand of this construct is capable of gene silencing thereby significantly increasing targeting specificity. Moreover, incorporation of LNA nucleotides into the disrupted sense strand significantly increases serum stability which may be important for applications. Interestingly, the sisiRNA design can functionally accommodate heavily modified antisense strands that are non-functional as standard siRNAs. This potentially allows the application of more highly functionalized siRNA designs. The two reporter constructs pISO and pISO were constructed by annealing equimolar amounts of the following DNA oligoes 5′-GCGACGTAAACGGCCACAAGTTC-3′ and 3′-TCGACGCTGCATTTGCCGGTGTTCAAGGATC-5 (antisense target) or 5′-CTAGGCGACGTAAACGGCCACAAGTTCAGCT-3′ and 3′-CGCTGCATTTGCCGGTGTTCAAG-5′ (sense target) into SacI/NheI digested pISO (kindly provided by David Bartel) () downstream of the firefly luciferase coding sequence. Non-modified and LNA-modified RNA oligoes were prepared on an automated DNA synthesizer as described earlier (). The synthesis of adamantyl and pyrenyl containing RNA oligoes is described elsewhere (J.W., S.W.L., B.R.B., manuscript in preparation). Cells used for EGFP northern, western and flow-cytometry analysis were seeded at ∼20% confluency and transfected using Bio-Rad Silentfect transfection reagent (50 nM final RNA concentration) according to manufacturer's instructions. Twenty-four hours later cells were replenished with fresh medium and incubated for another 24 h before either re-transfection using Lipofectamine2000 (50 nM final RNA concentration) according to manufactures directions or harvested for western blot analysis, northern blot analysis or flow-cytometry analysis (counting approximately 5 × 10 cells and averaged). Western blotting was performed as follows: cells were washed twice in PBS and an equal amount of cells were lysed in 2× SDS sample buffer [4% Sodium Dodecyl-Sulphate (SDS), 20% glycerol, 125 mM Tris/HCl pH 6.8, 0.01 mg/ml Bromphenol Blue, 10% b-mercaptoethanol] at 90°C for 2 × 10 min separated by gentle pippeting. Proteins were separated in 12% SDS acrylamide gels and electroblotted overnight onto a PVDF membrane (Immobilon). The filter was blocked for 1 h with PBS containing 10% w/v milk. EGFP protein was detected using a 1:1000 dilution of a rabbit polyclonal EGFP antibody. The mouse hnRNP C1 antibody was a gift from Seraphin Pinol-Roma. A horseradish peroxidase (hrp) conjugated secondary antibody (DAKO) was used with ECL reagent (Amersham Biosciences) for visualization. EGFP mRNA was analyzed by northern blotting according to standard procedures. siRNA-variants (80 nM) or poly(I:C) (0.8 μg/ml) were transfected into T98G cells using the TransIT-TKO® transfection reagent (Mirus) according to the manufactures protocol. Total RNA was purified using Trizol® reagent (Invitrogen), DNase treated and subjected to oligo-dT-primed reverse transcription. qPCR was performed using the platinum SYBR®Green qPCR supermix (Invitrogen) on a Stratagene Mx3005p qPCR system. Primers used for amplification of ISG56: 5′-AAGGCAGGCTGTCCGCTTA-3′ and 5′-TCCTGTCCTTCATCCTGAAGCT-3′. Primers for amplifying GAPDH: 5′-GAAGGTGAAGGTCGGAGT-3′ and 5′-GAAGATGGTGATGGGATTTC-3′. The PCR conditions are: 1 cycle: 95°C 10 min, 40 cycles: 95°C 30 s, 55°C 1min, 72°C 30 s, 1 cycle: 95°C 30 s, 55°C 1 min, 95°C 1 min. Relative quantification of mRNAs levels were done by using the ▵▵CT-method. The experiments were done in triplicates and ISG56 levels for siRNA-treated cells were normalized to TransIT-TKO treated controls. Annealed LNA-modified sisiRNAs, LNA-modified siRNAs or siRNAs were incubated at 37°C in either 10% or 80% fetal calf serum in DMEM (Gibco). Aliquots of 5 μl (each containing 20 pmol of siRNA) were diluted in 25 μl 1.2× TBE loading buffer (1.2× TBE, 10% glycerol, bromphenol blue) and snap-frozen on dry ice immediately upon sample taking. Samples were run on a 15% native polyacrylamide gel and stained using SYBR Gold® (Invitrogen). H1299 cells were plated in 6-well plates in RPMI supplemented with 10% fetal bovine serum and grown ON to 40–60% confluency. pISO and pISO (1 μg) were co-transfected with 0.002 μg pRluc-N2 (Perkin–Elmer) and the siRNA duplexes (10 nM final concentration) by simultaneous use of 6 μl TransIT-LT1 (Mirus) and 6 μl TransIT-TKO (Mirus) according to the manufactures protocol. The Dual-luciferase assay was done 48 h posttransfection using the ‘Dual-luciferase reporter assay system’ (Promega) according to the manufactures protocol. The luciferase activities were measured on a Lumat LB 950 luminometer (Berthold) and normalized to the renilla luciferase signal. To eliminate sense strand incorporation into activated RISC, we applied a novel siRNAs design characterized by an intact antisense strand complemented with two shorter sense strands. We anticipated that by incorporating LNA nucleotides into such tri-molecule construct, sufficient stability and dsRNA structural mimicry would be achieved to allow RNA interference activity. We initially designed a sisiRNA composed of a 10 and 12 nt sense strand directed towards a previously established functional target in the mRNA encoding enhanced green fluorescent protein (EGFP)(). To stabilize the sisiRNA construct we incorporated LNA at two and four positions in the sense 5′ and 3′ half-strands, respectively, and near the 3′ end of the antisense strand and assembled the construct from these three strands [AS1 + 5′SS1 + 3′SS1 (sisiRNA), ]. Together with a standard siRNA and unrelated control siRNA, the constructs were tested by transfection into an H1299 lung carcinoma cell line that stably expresses destabilized EGFP. Subsequently, the level of EGFP mRNA and protein expression was monitored on the basis of fluorescence microscopy (A), northern blotting (B), western blotting (C) and flow cytometry (D). Treating cells with 50 nM LNA-modified sisiRNA or siRNA yielded a comparable 10-fold knock down after 48 h (B–D). The duration of the knock down effect by the sisiRNA was similar or slightly superior to unmodified siRNA at 120 (5 days) or 180 h (7.5 days) (B and C). Hence, sisiRNA exhibits similar silencing activity in cell culture as compared to standard siRNAs. The activity of the LNA-modified sisiRNA was strictly depending on the presence of all three strands as omitting one or both of the short sense strands (5′SS1 or 3′SS1) eliminated the activity of the sisiRNA (A–C). To investigate if the sisiRNA design is applicable to other target sequences, we additionally targeted the endogenous gene GAPDH in H1299 cells using two sisiRNA designs differing only in the position of the LNA residues in passenger fragments [AS8 + 5′SS6 + 3′SS6 (siGAPDH 1) and AS8 + 5′SS6*+3′SS6* (siGAPDH 2), ] and compared them to unmodified siRNA targeting GAPDH [AS7+SS7 (siGAPDH), ] and LNA-modified siRNA [AS8+SS6 (siGAPDH, LNA), ]. All four constructs resulted in a similar ∼60% reduction in GAPDH mRNA levels (A and B, columns 1–4) as compared to cells transfected with EGFP-specific siRNA or non-transfected controls (A and B, columns 5 and 6). Competitive knock down levels have also been observed with sisiRNA directed towards H-Ras in HeLa cells (M.B.L. unpublished data). Hence, the sisiRNA design has proven highly functional for all tested targets with efficiencies similar to unmodified or LNA-modified siRNA. We speculated that the LNA-modified siLNA and sisiRNA designs may have greater stability compared to unmodified siRNAs, both in terms of premature strand separation and resistance to RNases. In accordance, we found that upon incubation in 80% FCS, unmodified siRNAs were rapidly degraded within 1½ h whereas a large proportion of both LNA-modified sisiRNA and an identical construct, but with continuous sense strand (AS1+SS1, ), remained intact for 780 min (13 h) (A). Notably, we have observed no significant knock down when using a sisiRNA containing only unmodified residues (data not shown) suggesting that LNA modifications are not only beneficial but also essential for the integrity of the sisiRNA design in biological fluids. This is compatible with the observation that LNA-modifications in the complementary part of either the antisense or sense strand are essential for sisiRNA serum stability (B). Hence, we conclude that the sisiRNA design is highly stable in serum despite of the introduction of an internal nick in the sense strand. Chemical modifications of nucleic acids can have a dramatic influence on the cellular immune response in cultured cells and in animals (,). We did, however, not observe any cytotoxic side-effects or growth inhibition in sisiRNA-treated cells as compared to standard siRNAs (data not shown). To analyze that the sisiRNA design does not trigger cellular interferon responses, we transfected the human glioblastoma T98G cell line using 80 nM of the different siRNA constructs and measured the induction of ISG56, which is strongly induced by both types of IFNs () and dsRNA (). ISG56 induction has previously been reported in T98G cells upon siRNA transfection (), yet no significant differences in ISG56 induction were observed between LNA-modified sisiRNA, LNA-modified siRNA and unmodified siRNA (C). In contrast, poly(I:C) induced the ISG56 several hundred fold. Hence, at least in cell culture, sisiRNA appears to be immunologically similar to standard siRNAs. To further optimize the sisiRNA design, we tested a series of different sense and antisense strands. In one experiment, the position of the gap in the sense strand was either shifted one position towards the 3′ end [AS1 + 5′SS3 + 3′SS3 (sisiRNA), ] or one position towards the 5′ end [AS1 + 5′SS4 + 3′SS4 (sisiRNA), ] which since the initiation of this work has been reported to be the natural cleavage site for Ago2 in RISC (). For the sisiRNA design, only a minor decline in silencing was observed whereas the sisiRNA design was slightly less efficient in gene silencing (A, compare columns 1–3). A similar minor decline in knock down efficiency was seen for a sisiGAPDH design as compared to siGAPDH 1 and siGAPDH 2 (data not shown). These data underscore the notion that the position of LNA modifications need not be fixed within the sisiRNA (A and B) and the position of the nick in the passenger strand need not to mimic the natural cleavage site (A). Increasing the gap size of the sense strand of the sisiRNA duplex to 1–2 nt resulted in a dramatic decline in sisiRNA activity irrespectively of gap position (data not shown). In conclusion we find flexibility in positioning of both the LNA modifications and the sense strand central nick, yet the sisRNA design has proven most efficient among the sisiRNAs combinations tested (A, data not shown). The 3′SS1 strand was initially designed to contain an additional U-residue at the 3′ end in order to ease the chemical synthesis. To test whether this residue affects sisiRNA activity, 3′SS1 was synthesized without this terminal U-residue (3′SS5; ). This alteration did not alter the activity of the construct significantly (compare columns 2 and 4, B). To test whether the discontinuity of the intended sense strand eliminates its contribution to gene silencing we inserted the EGFP target sequence for either the siRNA antisense or sense strand within the 3′ UTR of a firefly luciferase reporter construct (). This strategy allowed us to differentially assess the knock down effect derived from the antisense and the sense strand incorporation into activated RISC. As predicted, the LNA-modified sisiRNA construct was significantly more specific than the equivalent siRNA and LNA-modified siRNA duplexes since ∼50% knockdown was constantly seen from the sense strand with standard siRNA design (columns 2 and 4, ). In contrast, the sisiRNA design completely abrogated the silencing of the ‘sense target’ as compared to mismatch controls without compromising the potent knockdown mediated by the antisense strand (columns 5 and 6, ). To test whether it is possible to abrogate the silencing function of an otherwise optimal antisense strand with a nick, another LNA-modified sisiRNA with an intact sense strand and a discontinuous antisense strand (5′AS2, 3′AS2, SS3) was tested for the ability to knock down the antisense and sense targets (). This design completely eliminated silencing of the antisense target yet retained silencing of the sense target to a level comparable to the standard siRNA (columns 7 and 8, ). Collectively these data clearly demonstrate that the sisiRNA construct exhibits a much higher level of specificity for the intended target compared to the usual siRNA design. We and others have previously found that extensive LNA-modification of antisense strands strongly interfere with RNAi activity in standard siRNA designs (B, column 2 and data not shown) (,). Therefore, we initially designed AS1 to contain only two LNA residues near the 3′ end. To investigate whether the discontinuity of the sense strand influences the requirement for unmodified residues in the body of the antisense strand, we tested an LNA-modified sisiRNA with a highly modified antisense strand containing six LNA residues (AS2). This antisense strand is essentially inactive when paired with an all-RNA sense strand (data not shown) or an LNA-modified sense strand (LNA-modified siRNA duplex AS2+SS1, ; B, column 2). Interestingly, the requirement for unmodified residues in the antisense strand was less stringent when using the sisiRNA design (compare columns 2 and 3, B). A similar improvement in knock down efficiency was observed using the 3′ end shortened sense construct and the LNA-modified sisiRNA design (data not shown). To test if a similar effect applies to other types of chemical modifications that do not increase siRNA thermodynamic stability, we tested three designs containing either additional single N2′-adamantyl (AS4 and AS5, ) (synthesis will be described elsewhere) or N2′-pyren-1-yl 2′-amino-LNA-T (AS6, ) modifications in the antisense strand (, aT and pT, respectively). These types of modifications render the siRNA almost non-functional when paired to SS1 in a standard (LNA-modified) siRNA design (B, columns 4, 6 and 8). However, in the context of the sisiRNA design, both the adamentyl and pyrenyl antisense strands resulted in a 40–60% knock down of EGFP expression (, columns 5, 7 and 9). Hence, the sisiRNA design can accommodate a wide variety of bulky chemical modifications that otherwise are incompatible with the activity of standard duplex siRNA. To further characterize the mechanism for the relaxed stringency of antisense strand modification, we investigated the incorporation of the sense strand (SS1) into RISC in lightly modified antisense siRNA (AS1+SS1) as compared to the siRNAs with heavily modified antisense (AS2+SS1, AS4+SS1, AS5+SS1, AS6+SS1). The LNA-modified siRNA (AS1+SS1) caused an ∼60% reduction of reporter levels from the sense target (C, column 2), confirming that the sense strand is indeed incorporated into activated RISC and active. In contrast, the function of the sense strand was virtually lost when it was paired to highly modified antisense strands (AS2, SS1; AS4, SS1; AS5, SS1; AS6, SS1) (C, columns 6, 10, 14 and 18). Hence, the poor silencing by the heavily modified antisense strands (AS2, AS4, AS5 and AS6) is not simply be due to a shifted strand selection (towards SS1) during RISC activation. Instead, heavy modification of siRNA duplexes in the antisense strand seems to abrogate its function at steps prior to RISC activation. The sisiRNA design, however, seems to partly rescue such defects thereby allowing heavily modified antisense strand to be efficiently loaded into activated RISC. To ensure that the knock down effects obtained using siRNA with lightly and heavy modified sisiRNA or siRNA are specific, we synthesized the equivalent set of LNA-modified siRNA and sisiRNA containing five mismatched positions (SS8+AS9, SS8+AS10 and 5′SS8 + 3′SS8+AS9, 5′SS8 + 3′SS8+AS10, respectively; ) and tested them in the luciferase reporter assay (Supplementary Figure 1). The results confirmed the increased potency of heavily modified antisense strands when situated in a sisiRNA design and showed that the knock down was specific to the wild type EGFP target sequence. In this study, we have developed a radical new siRNA design composed of an intact antisense strand complemented with two shorter 10–12 nt sense strands. We show that this three-stranded construct is fully functional and that it has several advantages over the standard 21 nt duplex siRNA designs. (i) The LNA-modified sisiRNAs have similar high potency as compared to standard siRNAs in cell culture, yet has greatly enhanced stability in serum which is expectably important for applications. (ii) The segmented nature of the passenger strand completely alleviates its contribution to unwarranted gene knock down thereby greatly increasing targeting specificity and expectably reducing off-target effects. (iii) The sisiRNA design has the ability to rescue the function of chemically modified antisense strands that are non-functional within the context of a standard siRNA duplex thereby allowing more chemical modification to be incorporated into the antisense strand. (iv) The sisiRNA design has six terminal ends compared to four in normal siRNA which can conveniently be used for tethering functional chemical groups to enhance, e.g. cellular delivery. For instance, it is possible to tether bulky groups like cholesterol to the 5′ end of the downstream sense strand without loosing activity (M.B.L., J.K., J.W., unpublished data). (v) As the yield of synthesis is usually higher for shorter RNA strands, the cost of large-scale synthesis in connection with therapeutic application may be reduced using a sisiRNA design. An important feature of the sisiRNA design is the ability to completely eliminate the contribution of the segmented strand to gene silencing while leaving the RNAi activity of the opposing strand intact (). The resultant increase in gene silencing specificity can be expected to reduce the genome-wide off-targets effects from the sense strand that has been observed for other investigated siRNAs (). Furthermore, as strand selection is primarily determined by the thermodynamic asymmetry of siRNA duplex ends, highly efficient siRNA may be difficult to design if the target sequence is restricted to a thermodynamically unfavorable region, e.g. when the intension is to target single nucleotide mutation or junctions between fused genes. In these instances, the sisiRNA design will ensure that only the unsegmented strand can contribute to gene silencing irrespectively of the thermodynamic profile of the sisiRNA duplex and will thereby eliminate the significant unwarranted silencing conferred by the thermodynamically favored opposing strand. Leuchner () have previously demonstrated that pre-cleaved siRNA, similar to our unmodified sisiRNA, is capable of RISC loading and target cleavage in a cell extract. However, we find that sisiRNAs without LNA residues are non-functional in a cellular context, even if 2′ OMe modified residues are introduced in the short sense strands (data not shown). Based on our stability assays (B), the most likely explanation is that the unmodified strands in sisiRNA are dissociating and degraded and that only the significant increase in Tm, provided by the LNA residues, renders the duplex sufficiently stable under these conditions. An interesting observation is that sisiRNA function does not rely strictly on exact structural mimicry of an intermediate Ago2-cleavage product as the strand nick can be moved 1–2 nt without major loss of silencing efficiency (A, data not shown). In particular, the sisiRNA design mimicking the ‘natural’ Ago2-cleavage product (sisiRNA) seems less efficient than when moving the nick 1 and 2 nt towards the 3′ end of the sense strand (sisiRNA and sisiRNA). Based on data from Leuchner () these constructs are most likely cleaved by Ago2, liberating one or two nucleotides, respectively. It is therefore possible that allowing a ‘natural’ Ago2-cleavage event in the sisiRNA and sisiRNA designs may further help RISC activation by facilitating subsequent steps such as, e.g. sense strand elimination. Hence, we believe that the sisiRNA design introduces novel improvements in siRNAs function beyond those offered by the structural mimicry of natural intermediates in the RNAi pathway. We and others have observed that extensive chemical modifications in the antisense strand of siRNAs generally are incompatible with their function in gene silencing (, data not shown)(). Yet, the specific steps in the RNAi pathway, which are incompatible with extensive siRNA modification, are only poorly defined. Interestingly, the sisiRNA design can significantly enhance the efficiency of heavily modified antisense strands. We demonstrate here that the inability of extensively LNA-, LNA/adamantyl- and LNA/pyrenyl-modified antisense strands to support RISC activity can be partially rescued by the sisiRNA-design, whereas both strands of similarly modified ordinary siRNAs are non-functional (). This shows that heavy modification of an antisense strand abrogates the function of both sense and antisense strands that are individually functional within the context of activated RISC (B and C). This suggests that extensive modification of antisense strands may lead to impairments prior to RISC activation, e.g. siRNA recruitment by the RLC or structural rearrangements within pre-RISC. It may be speculated that the segmented sense strand in the sisiRNA design facilitates the preferential loading of the intact opposing antisense strand into activated RISC and thereby enhance their silencing potential. However, no increase in silencing by the unsegmented strand in neither AS1 + 5′SS1 + 3′SS1 nor 5′AS2 + 3′AS2+SS3 designs was observed as compared to AS1+SS1 (). In agreement, no enhancement in silencing by the sisiRNA design as compared to siRNA was seen in titration assays (0.01–100 nM) suggesting that strand selection is not affected (data not shown). Furthermore, the rescue of silencing by the sisiRNA design seems not to rely on alteration of the siRNA thermodynamic profile as the adamantyl and pyrenyl modifications, if anything, slightly destabilize siRNA duplexes in contrast to the stabilizing effect of the LNA-residues (J.W., unpublished data). Instead, heavily modified siRNAs may be too inflexible for structural rearrangements within pre-RISC during RISC loading or activation. The central strand nick in the sisiRNA design may provide more structural flexibility to the sisiRNA duplex allowing it to better position itself during RISC activation. The possibility to incorporate more extensive chemical modifications into the sisiRNA design as compared to standard siRNAs may have beneficial properties for steps both upstream and downstream of RISC activation in the RNAi pathway. Introducing lipophilic groups like adamantyl and pyrenyl may increase cellular uptake of siRNA duplexes and unnatural modifications in general will increase siRNA stability in intra- and extracellular compartments. Furthermore, modifications in the seed region (nucleotides 2–8 of the antisense strand) may prove essential to minimize inherent gene off-target effects by siRNAs as it has been previously been demonstrated for position 2 in the antisense strand (). Finally, it is possible that increased numbers of LNA residues in the antisense strand may improve the target specificity and affinity. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

A total of 4584 mature miRNAs have been registered in the miRBase (as of May, 2007) (). These miRNAs were identified from various organisms including primates, rodents, birds, fish, worms, flies and viruses. Sequences of the majority of the miRNAs are highly conserved across species, suggesting that miRNAs are important regulators of molecular and cellular processes. miRNAs are believed to function as post-transcriptional suppressors through binding their target mRNAs through base pairing and subsequently inducing either translational repression or mRNA destabilization (). Several studies have shown that miRNAs are involved in the regulation of various cellular processes including cell differentiation (,), proliferation (,) and apoptosis (). Recent studies also provide evidence that miRNAs are directly linked to viral diseases (), neuronal development () and tumorigenesis (). These findings suggest that miRNAs are as important as transcriptional factors in the control of gene expression in higher eukaryotes. miRNAs are transcribed as 100–1000 nucleotide (nt) primary miRNAs (pri-RNAs) by RNA polymerase II, and are modified just like mRNAs, including 5′ capping and 3′ poly(A) tailing (). The miRNA-encoding portion in the pri-miRNA forms a hairpin, which is cleaved by the double-stranded RNA (dsRNA)-specific ribonuclease Drosha and its cofactor DiGeorge syndrome critical region 8 (DGCR8) (). The cleaved hairpin is 60–70 nt long and is called a precursor miRNA (pre-miRNA) (). The pre-miRNA is then transported by Ran-GTP and Exportin-5 (EXP5) to the cytoplasm (). The pre-miRNA is further processed by a second RNase III complex, consisting of Dicer and the trans-activator RNA-binding protein (TRBP), which generates a miRNA duplex containing two mature miRNAs (5′- and 3′-strand miRNAs) each ∼22 nt in length (). It is currently believed that a miRNA-induced silencing complex (miRISC) selects either the 5′- or 3′-strand miRNA, depending on the relative stability of the termini in the pre-miRNA duplex. The strand with lower stability of base pairing in the 2nd–4th nucleotides at the 5′ end of the duplex (called miRNA or guide strand) preferentially binds the miRISC and thus becomes accumulated and functional, whereas the other one (called miRNA* or passenger strand) is degraded (,,). This ‘strand bias’ theory was based upon analyses of the thermodynamic stability profiles of pre-miRNAs and mature miRNAs as well as small interfering RNAs (siRNA) (,,). According to this model, only a small portion of miRNAs displaying similar thermodynamic stability in both the guide and passenger strands is expected to show co-accumulation (), whereas the majority of miRNAs are highly asymmetric and thus are subjected to strand selection (,). However, during our analyses of miRNAs in the miRBase, we found that 234 out of 969 known human and mouse miRNAs were actually miRNA pairs (117 pairs). In addition, many 5′-strand miRNAs were found in one species, whereas its 3′-strand counterparts were present in another species. Since majority of the miRNAs are highly conserved among different species, these observations suggest that both strands are expressed and accumulated as miRNA pairs, but their selective or simultaneous accumulation may be tissue dependent. To test this hypothesis, we first predicted the sister miRNAs for all ‘unpaired’ miRNAs using a miRNA sequence determination rule that we deduced by analyzing the stem-loop structures of the pre-miRNAs for 117 known human and mouse miRNA pairs. We further validated the expression of these predicted sister miRNAs using semi-quantitative PCR and ribonuclease protection assays. We also predicted target genes for paired miRNAs and functionally validated six target genes for a sister pair mir-30e-5p and mir-30e-3p. Our data reveal that selective or co-accumulation of both strands of miRNAs is highly tissue dependent and both strands of a miRNA pair can functionally suppress the expression of their target genes. All human and mouse miRNAs were collected from the miRBase (). miRNA genes were located on each genome using the UCSC genomic browser () (,). A sequence containing a 100-bp upstream segment and a 100-bp downstream segment around the miRNA was used as a pre-miRNA to generate the secondary stem-loop structure using the MFOLD program version 3.2 () (,). Most pre-miRNA sequences could fold into expected stem-loop structure as shown in the miRBase. For the ones that failed to fold into stem-loop structures initially, the flanking sequences were trimmed to get favorable folding. Both strands were then located in the stem-loop structures. Sizes of miRNAs, core sequences and overhangs were analyzed. The size distributions were calculated using Microsoft Excel (Office 2003, Microsoft). Data were plotted and graphed using GraphPad Prism (GraphPad Software). By analyzing the commonality of the 5′- and the 3′-strand miRNA sequences in the duplex structures, we deduced a ‘3-Prime-Counting-22-nt’ rule for the prediction of sister miRNAs based upon the known strands. Using this rule, we predicted sister strands for all currently known unpaired human and mouse miRNAs in the miRBase. To see whether the predicted sister miRNAs matched any of the known miRNAs or Piwi-interacting RNAs (piRNAs), all predicted miRNAs were searched for in the miRBase and in the GenBank at the NCBI website (). Small RNA samples from 12 different mouse tissues (brain, heart, liver, spleen, lung, kidney, stomach, small intestine, colon, ovary, uterus and testis) were isolated using the mirVana™ miRNA isolation kit (Ambion) according to the manufacturer's instructions. Preparation of the small RNA complementary DNA (srcDNA) library and semi-quantitative PCR analyses of miRNAs were performed as described (). All oligos used in this study are shown in the Supplementary Table 1. Semi-quantitative PCR analyses using srcDNAs were performed such that the PCR cycle numbers were empirically determined to ensure that each of the amplification reactions was in the exponential range (20–30 cycles). The house-keeping miRNA let-7d-5p () was used as a loading control. For miRNA detection, RPA was performed using mirVana miRNA Probe Construction Kit (Ambion). For transcription of antisense RNA probes, DNA oligonucleotides (IDT) complementary to 5′-strand or 3′-strand of mir-24, mir-194 and mir-130a with the 12-nt linker sequence were synthesized (Supplementary Table 1). The 12-nt sequence is complementary to the T7 promoter primer provided in the kit. Each synthesized oligonucleotide (200 μM) was annealed with 2 μl of the T7 promoter primer in 6 μl of DNA hybridization buffer at 70°C for 5 min, followed by incubation at RT for 5 min. To get dsDNA template, end-filling reaction was performed by adding 2 μl of Klenow Reaction Buffer, 2 μl of 10 × dNTP Mix, 4 μl of Nuclease-free Water and 2 μl of Exo-Klenow at 37°C for 30 min. transcription was carried out to make RNA probes in a 20 μl reaction containing 1 μl of dsDNA template, 2 μl 10 × Transcription buffer, 1 μl of 10 mM ATP, 1 μl of 10 mM UTP, 1 μl of 10 mM CTP, 5 μl [α-P]GTP at 800 Ci/mmol and 10 mCi/ml (3.125 μM), 2 μl of T7 RNA Polymerase and 7 μl of Nuclease-free water at 37°C for 15 min, as described in the mirVana miRNA Probe Construction Kit Protocol (Ambion). After reaction, 1 μl of DNase I provided in the kit was added and incubated at 37°C for 10 min. The labeled RNA probes were purified using a purification filter cartridge provided in the mirVana Probe & Marker Kit (Ambion). Briefly, Binding/Washing Buffer (12 volumes of the sample) was added to the sample and mixed thoroughly. The mixture was applied to a purification filter cartridge and centrifuged. The cartridge was washed with 300 μl of Binding/Washing Buffer. The RNA probes were recovered with two sequential elutions using 20 μl of pre-heated (95°C) Elution Buffer each time. About 2 μg of small RNAs isolated from brain, heart, liver, spleen, lung and testis were used to hybridize with 1 μl each of the RNA probes, 10 μl of 2 × Hybridization Buffer, 2 μg of Yeast RNA at 42°C for 2 h using mirVana miRNA Detection Kit (Ambion). After hybridization, an aliquot of 150 μl of diluted RNase was added to digest free RNA probes and the samples were incubated at 37°C for 45 min. The digestion reaction was stopped by adding 225 μl of RNase Inactivation/PPT solution. Equal volumes (225 μl) of 100% ethanol were added to purify the probe-miRNA duplexes and the samples were stored at −20°C for 30 min. The samples were centrifuged at 12 000 r.p.m. for 10 min. The pellets were air-dried and resuspended in 15 μl of Gel Loading Buffer II. The samples were incubated at 95°C for 3 min and separated on a 15% denaturing polyacrylamide gel. The gel was photographed on the ChemiDoc XRS imaging system (Bio-Rad). Target genes for each of the sister strands were predicted and retrieved from the miRBase () using the miRanda algorithm (,) with a -value cutoff at 0.05 and also from the PicTar () (,). The predicted targets for each sister strand were downloaded in the .txt format (the option is at the top of the page) and imported into Excel using the ‘Data->Import External Data’ option. Total target numbers for each sister strand were calculated and then average numbers for each sister strand were calculated using the Excel (Microsoft) and graphed using Prism (GraphPad Prism) (). Target genes for the 5′-strand and the 3′-strand of mir-30e were predicted and analyzed using the miRBase Targets Version 4 (). Based on scores and -values, six potential target genes (, , , , and ) were selected for experimental validation (D). A miRNA-target validation vector was constructed from a luciferase reporter vector pGL4.19 [luc2CP/Neo] (Promega) and was named pGL-miTar (Supplementary Figure 7). SV40 promoter region (474 bp) was amplified with a primer pair containing the KpnI (forward) and HindIII (reverse) sites by PCR from the pGL4.19 (for primers, see Supplementary Table 1). The SV40 promoter was then inserted at the multiple cloning sites KpnI–HindIII before the luciferase gene (). An eGFP gene cassette (996 bp, eGFP- SV40 poly A signal) was amplified from the pAd Sh/H1 vector () and subcloned into pcDNA 3.1 TOPO vector. A chimeric intron (78 bp) with XmaI site in the middle used for adding the pre-miRNA sequences was inserted at the middle (335 bp from ATG) of the eGFP-coding region by mutagenesis PCR (eGFP-pre-miRNA/pDNA3.1) (for the chimeric intron sequence, see Supplementary Table 1). The eGFP-pre-miRNA cassette (2056 bp, CMV promoter-eGFPa-chimeric intron-eGFPb-SV40 poly A signal) was digested with BglII and XhoI and used to replace the neomycin gene cassette (1627 bp, BamHI–SalI) in the vector. The pGL-miTar was used for cloning a pre-miRNA and a target sequence into the chimeric intron and the 3′UTR of , respectively (B). A mir-30e precursor (pre-mir-30e, 98 bp) with XmaI site at each end was amplified from the mouse genomic DNA and subcloned into pcDNA3.1 TOPO vector (Invitrogen) (for pre-mir-30e primers, see Supplementary Table 1). The pre-mir-30e was digested with XmaI from the plasmid isolated and subcloned into the pGL-miTar linearized with the same restriction enzyme. Target regions (120–186 bp) for each gene with XbaI site at each end were amplified from mouse testis cDNA and subcloned into the pcDNA3.1 TOPO vector (Invitrogen) (for each target gene primers, see Supplementary Table 1). Each target region was digested with XbaI and subcloned into the pGL-mir-30e linearized with the same restriction enzyme. The sequence accuracy and correct orientation of the pre-mir-30e and target region in the vector were confirmed by sequencing (for sequencing primers, see Supplementary Table 1). Bacteria transformed with each of the six final constructs were cultured in 500 ml LB medium and the plasmids were isolated using an endotoxin-free, EndoFree Plasmid Maxi Kit (QIAGEN). A Dual-Luciferase Reporter Assay System (Promega) was used to examine the effects of miRNAs on their target genes. Each mir-30e-target plasmid (2 μg) was co-transfected with pRL-CMV (10 ng) (Promega) into HEK-293 cells grown on a 6-well plate using PolyFect transfection reagent (20 μl) (QIAGEN). The transfected HEK cells were cultured for 24 h and used for the luciferase assay. The cells were washed with PBS buffer and homogenized with 200 μl of Passive Lysis Buffer. Luciferase Assay Buffer (100 μl) was transferred into a 96-well plate. The lysate (20 μl) was added into two wells (duplicate) of the 96-well plate and mixed. Firefly luciferase activity was measured on GloRunner Microplate Luminometer (Turner BioSystems). The plate was removed from the luminometer, Stop & Glo Reagent (100 μl) was added and mixed. The Renilla luciferase activity was measured on the luminometer. Data recorded on the luminometer were analyzed using Excel (Microsoft) and graphed using Prism (GraphPad Prism). Among the 969 mouse and human miRNAs registered in the miRBase (as of May, 2007) (), 377 (39%) are 5′-strand miRNAs and 358 (37%) are 3′-strand miRNAs. The remaining 234 are paired miRNAs (117 pairs) containing both the 5′-and 3′-strands () (Supplementary Table 2). Among these 234 paired miRNAs, 82 miRNAs (41 pairs) are from the mouse and 152 (76 pairs) are from the human. Interestingly, 114 (57 pairs) out of the 152 paired human miRNAs (76 pairs) registered in the miRBase () (highlighted in Supplementary Table 3) were previously shown to display a thermodynamic preference for the selection of either the 5′- or the 3′-strand (). Moreover, we found an additional 32 pairs of miRNAs expressed as sister pairs in other species including the mouse (), cow (), rat (), chicken (), zebra fish (), frog () or fruit fly () (highlighted in Supplementary Table 4). These 32 paired miRNAs were previously shown to be thermodynamically asymmetric (). These findings suggest that the current ‘thermodynamic stability’ theory on strand selection is not universally followed during natural miRNA biogenesis. We hypothesized that in some tissues miRNAs were expressed and accumulated as sister pairs (5′- and 3′-strand miRNAs) while in other tissues the same miRNAs may be subjected to strand selection by an unknown mechanism. To test our hypothesis, we need to develop a method to accurately predict the sister miRNAs for all unpaired miRNAs and test whether both strands of a miRNA pair are indeed co-accumulated in some tissues. To deduce a sequence determination rule for predicting sister miRNA sequences based upon known unpaired miRNAs, we analyzed the stem-loop structures of pre-miRNAs for all 234 paired miRNAs previously identified from humans (Supplementary Table 3) and mice (Supplementary Table 5). The miRNA sequences in the stem-loop duplexes are almost identical between the mouse and human. A paired miRNA in a stem-loop shows a duplex structure, consisting of a core sequence, where the 5′-strand base pairs with the 3′-strand with an overhang at the 3′ end. The size of the majority of the paired miRNAs (225 out of 234 miRNAs analyzed, 96%) ranges from 20 to 24 nt in length (Supplementary Figure 1A). The overall size distribution between the two species is similar, although human miRNAs appear to be a bit longer than mouse miRNAs. Most of the paired miRNAs (201 miRNAs, 86%) range from 21 to 23 nt in length, among which the 22 nt size is the most frequently found (101 miRNAs, 43%). Overhang sizes are more variable, ranging from 0 to 5 nt (Supplementary Figure 1B) with the 2 nt overhang being the most common (85 miRNAs, 36%). Variation in total size and overhang size may be due to the current cloning strategy, which uses a primer containing two degenerate nucleotides at the 3′ end during cDNA synthesis (). Analyses of the sizes of the core sequences and overhangs in the stem-loop structures of 234 paired mouse and human miRNAs registered in the miRBase allowed us to deduce a ‘3-Prime-Counting-22-nt’ rule for the prediction of sister miRNA sequences (Supplementary Figure 2). The deduced duplex structure is consistent with the observation that the RNase III digests dsRNAs and pre-miRNAs (e.g. pre-mir-30) in a staggered manner, leaving a 2 nt overhang at the 3′ end () which serves as a substrate for Dicer to generate 22–24 nt products in a ruler-like fashion (). To predict the sister strand for a known miRNA, one needs to first locate the first nucleotide (nt) of the known strand and then find its base-pairing nt (BN) on the to-be-predicted strand. A reference nt (RN) is further determined by locating the 1st–4th nt 3′ next to the BN, depending upon whether a gap or gaps are present in the known or the to-be-predicted strand (see Supplementary for step-by-step procedures). The RN will be the last nt of the 3′ end of the predicted sister miRNA strand. Thus, once the RN is determined, one can count 22 nt from the RN toward the 5′ direction and the 22 nt sequence is determined as the predicted sister miRNA. We next randomly selected 47 paired mouse miRNAs, including 22 novel sister miRNAs predicted in this study, to validate their expression in mouse tissues. We had previously developed a PCR-based method for detection and quantification of miRNAs (). This method is accurate, less time-consuming and much more sensitive than the traditional PAGE-based Northern blot analyses. We examined the expression levels of 47 paired miRNAs, including eight let-7 isoforms (let-7a to let-7i) in multiple mouse tissues using the semi-quantitative PCR in conjunction with ribonuclease protection assays (). All of the 47 miRNAs were readily detected ( and , Supplementary Figures 3–6) and each miRNA PCR product was sequenced to confirm their identity. As a representation, duplex structures and expression profiles of both strands of three miRNAs, mir-24, mir-194 and mir-130a are shown in . Both strands of mir-24 have been previously identified from the mouse and human (A). Only the 5′-strand of mir-194 and the 3′-strand of mir-130a were identified in both species, respectively. The 3′-strand of mir-194 (D) and the 5′-strand of mir-130a (G) were predicted using the ‘3-Prime-Counting-22-nt’ rule. Both strands of mir-24, the 5′-strand of mir-194 and the 3′-strand of mir-130a displayed a ubiquitous expression pattern in all the tissues examined (B, E and H). The 3′-strand of mir-194 and the 5′-strands of mir-130a, however, were preferentially expressed in some, but not all of the tissues tested (E and H). For example, the 5′-strand of mir-130a is preferentially expressed in lung and uterus, but almost no expression was detected in liver, spleen, ovary and testis (H and I). The ribonuclease protection assays (C, F and I) showed that the expression patterns of the three miRNAs, mir-24, mir-194 and mir-130a were consistent with those of the PCR assays, further indicating that our semi-quantitative PCR method is reliable. It is unlikely that the detection of predicted sister strands resulted from the degenerating/degenerated miRNA* or passenger strands. The miRNA* or passenger strands not incorporated into the miRISC should be destroyed rapidly and thus not be accumulated in the tissues. In addition, levels of the miRNAs detected by the PCR represented their steady-state levels. Surprisingly, co-expression and co-accumulation of both strands of miRNAs are regulated in a tissue-dependent manner. Certain tissues express both strands as miRNA pairs [e.g. co-expression of both mir-194-5p and -3p in heart, lung, small intestine, ovary and uterus (E and F)], whereas only single strands are accumulated in the other tissues [e.g. selective accumulation of mir-194-5p in brain, liver, spleen, stomach and testis (E and F)]. This spatial regulation of miRNA expression and accumulation implicates that a novel tissue-dependent mechanism rather than the thermodynamic asymmetry is involved in the decision of selective or simultaneous accumulation of sister strands. Consistent with the expression profiles of the three paired miRNAs (), the other 36 paired miRNAs also displayed differential expression patterns in the multiple tissues. The eight let-7 isoforms are ubiquitously expressed in all tissues tested, but expression levels of each miRNA are different in different tissues (A and Supplementary Figures 3 and 6). Among tissue-specific miRNAs observed, many miRNAs (mir-201-5p, mir-202-5p and 3p, mir-465-5p and 3p, mir-470-5p, mir-471-5p and 3p, mir-181b-1-3p) are preferentially expressed in the testis (Supplementary Figures 4C, 4D, 5A, 5C, 5D and B). We further validated the expression of the other 60 paired miRNAs in the miRBase using the same semi-quantitative PCR method (Supplementary Tables 5 and 7, gel pictures available upon request) and found that early identified miRNAs (e.g. let-7s and mir-1 to 200) are mostly ubiquitously expressed in multiple mouse tissues, whereas the ones recently identified (e.g. mir-740s and mir-460s-470s) (,) or predicted sister miRNAs tend to be preferentially or tissue-specifically expressed. Since most of the early identified miRNAs are ubiquitously expressed, it is likely that their ubiquitous expression accounts for their early identification. Also, the finding that not all miRNAs are ubiquitously expressed may explain why many of our predictions, which may be preferentially expressed in only certain tissues, have not yet been cloned. This finding also suggests that there may be many more tissue-specific miRNAs remaining to be identified. Some miRNA genes have multiple copies on the same chromosome or different chromosomes. It is unknown whether multiple copies are all transcribed, or whether only one is transcribed while the others act as pseudogenes and remain silent. We therefore attempted to examine the expression of the same miRNAs transcribed from two loci on different chromosomes. We analyzed three let-7 isoforms (let-7a, c and f), mir-181b and 194, which are all transcribed from two loci on different chromosomes ( and Supplementary Figure 6). Sequences and chromosome locations are shown in Supplementary Table 8. let-7c-5p and two sister miRNAs were ubiquitously expressed in multiple mouse tissues (A). The gene for mir-181b-5p has two copies, with one on chromosome 1 and the other on chromosome 2. Genomic sequences of the mir-181b-3p gene on chromosome 1 (named mir-181b-1-3p) and on chromosome 2 (named mir-181b-2-3p) are slightly different and thus can be distinguished using sequence-specific primers in the semi-quantitative PCR analyses (Supplementary Table 8). mir-181b-5p, mir-181b-2-3p and mir-194-1-5p are ubiquitously expressed in multiple mouse tissues, whereas mir-181b-1-3p, mir-194-1-3p and mir-194-2-3p are expressed in a tissue-preferential manner (B for mir-181b, Supplementary for mir-194). These data suggest that all miRNAs analyzed in this study are indeed transcribed from two loci on different chromosomes. miRNAs encoded by genes with multiple copies have been assumed to be the same and thus grouped under the same miRNA identifiers. In the mouse and human, 161 miRNAs are derived from genes with at least two copies on the same or different chromosome(s). However, the sequence analysis of the sister miRNAs from different loci and the expression data that we show here demonstrate that sequences of sister miRNAs are different and that the expression patterns of each of them are unique. We therefore suggest that all the miRNAs derived from multiple copies of the same genes should be regarded as individual miRNAs. The co-accumulation of both sister strands of a miRNA pair does not necessarily mean that both are functional, since one of the pair may target less, or even may not target any genes. To assess whether both sister miRNAs can target similar numbers of mRNAs, we analyzed the predicted target genes for 103 paired miRNAs (mouse 43 and human 60 pairs). The target genes predicted using the miRanda algorithm showed that both strands of paired miRNAs can target comparable average numbers of potential target genes in mice and humans (). In mice, the average target numbers were 908.9 for the 5′-strand miRNAs and 899.0 for the 3′-strand miRNAs (Supplementary Table 9). Interestingly, human miRNAs appear to have more predicted targets (1083.4 predicted targets for the 5′-strand miRNAs and 1166.5 for the 3′-strand miRNAs, Supplementary Table 10) than the mouse ones using the same algorithm although mice and humans are predicted to have the same number of protein-encoding genes (∼30 000) in each genome (,). Forty-six out of the 59 human miRNA pairs analyzed here were previously thought to show a thermodynamic preference for the selection of either the 5′- (28 miRNAs) or the 3′-strand (18 miRNAs) (Supplementary Table 10) (). However, similar average numbers of targets for the 5′- and 3′-strands were predicted, respectively. The average number of target genes for the 28 5′-strand miRNAs were 1119.0 and that for their 3′-strand sister miRNAs was 1217.5. Likewise, the average number for the 18 3′-strand miRNAs were 1014.6 and that for their 3′-strand sister miRNAs was 1154.9. Using another popular target prediction program the PicTar () (,), we predicted and analyzed the potential targets for each strand of the 40 paired miRNAs (mouse 27 pairs and human 13 pairs) available in this website. The average target numbers predicted by the PicTar algorithm (e.g. 188 for the 5′-strand and 174 for the 3′-strand of mouse paired miRNAs) were less than those predicted by the miRBase (909 for the 5′ strand and 809 for the 3′ strand in mouse paired miRNAs). However, average numbers of targets predicted for each strand of the 40 paired miRNAs appeared to be similar (Supplementary Tables 9 and 10), which is consistent with the target prediction results using the miRanda algorithm. These data suggest that both strands be equally functional. To test whether both sister strands of a miRNA pair can functionally suppress their target gene expression, we developed a miRNA-target validation system using a dual-Luciferase reporter assay. Using this system, we examined functionality of both strands of mir-30e (mir-30e-5p and mir-30e-3p). The paired miRNAs mir-30e-5p and -3p were chosen because they were previously shown to display differential thermodynamic stability, implicating that only the guide strand, corresponding to mir-30e-3p here, could be selectively accumulated and be functional (). mir-30e-5p was cloned from the mouse embryonic stem cell (), whereas mir-30e-3p was cloned from the human HL-60 cells (). Interestingly, our semi-quantitative RT-PCR analyses revealed that both mir-30e-5p and mir-30e-3p were differentially expressed in most of the mouse tissues (A). Consistent with the expression patterns of other miRNA pairs analyzed in this study, each strand showed a tissue-dependent expression: mir-30e-3p appeared to be preferentially accumulated in stomach, whereas mir-30e-5p was selectively accumulated in spleen (A). Among numerous predicted target genes, three for mir-30e-5p (, and ) and three for mir-30e-5p (, and ) were chosen because they showed higher scores with lower -values (Supplementary Table 11). For target validation, we constructed a miRNA-target validation vector pGL-miTar (B). The target region for each of six genes (120–186 bp) was amplified from the mouse testis cDNA and then subcloned into the 3′UTR of the Luciferase gene () (B). The mir-30e precursor (98 bp) amplified from the mouse genomic DNA was inserted into the chimeric intron (36 bp 5′ donor and 36 bp 3′ acceptor sequences) of an gene so that the pre-mir-30e could be spliced from the eGFP transcript to produce the mir-30e sister pair. The remaining eGFP transcript would be spliced, ligated and translated into eGFP, which serves as a marker for monitoring not only transfection efficiency but also correct splicing. Two representative fluorescent images (C) show that HEK-293 cells transfected with either the mir-30e-5p- vector or the mir-30e-3p- construct expressed abundant eGFP. HEK-293 cells transfected with each of the other four miRNA-target validation constructs also expressed abundant eGFP (data not shown), indicating that the transfection was effective and the pre-mir-30e were correctly spliced. The pre-mir-30e processed by the Dicer of the cells should generate mature mir-30e-5p and mir-30e-3p, which would then be incorporated into the miRISC. The mir-30e-RISCs would bind their respective target sequences inserted at the 3′UTR of the gene and thus suppress levels of Luciferase in the assays. Luciferase activity was reduced by 18–48% in the cells transfected by each of the six miRNA-target sequence validation constructs (D). Among the six target genes, showed the greatest reduction (48%) by mir-30e-5p, whereas displayed the least suppression (18%) by mir-30e-3p. As control experiments, we used pGL-mir-30eTar (contains no target sequences in the 3′UTR of the ), or pGL-mir-30e-, which contains the target sequence from the 3′UTR of . was predicted to be targeted by let-7d-5p, but not by either strand of mir-30e. As expected, no reduction in luciferase activity was observed in the cells transfected with pGL-30eTar (no target control) or pGL-mir-30e-. These results suggest that both sister strands of the mir-30e are functional despite their varying efficiency in suppressing their target gene expression. The tissue-dependent co-accumulation of sister miRNAs is physiologically significant because it reveals, for the first time, that one pre-miRNA can produce two functional sister miRNAs which can target two sets of totally different genes. Based on our finding, we propose a ‘Target-Two-Sets-of-Genes-With-One-Pre-miRNA’ model, in which cells from certain tissues can utilize both strands of a sister miRNA pair rather than adopting one strand and discarding the other (). Recent computational identification of target genes of miRNAs show that 148 human miRNAs can potentially target up to 5300 human genes, representing up to 30% of the gene set in the analysis (17 850 orthologous mammalian genes) (). Human tissues express at least 968 miRNAs including 408 miRNAs predicted in this study, suggesting that miRNAs theoretically could target the entire mRNA transcriptome and that all the human mRNA-coding genes are under the regulation of miRNAs. To be functional, each strand of a miRNA pair needs to be incorporated into the miRISC. It is very difficult to show experimentally the incorporation of all sister miRNAs into miRISC, and only a handful of miRNAs have been experimentally verified to be incorporated into RISC (,,), in which artificial target constructs and siRNAs instead of miRNAs were used. In siRNAs, asymmetric siRNAs have been shown to result in asymmetric assembly of siRNA RISC (siRISC) (,). Although it was generally accepted that siRISC and miRISC are functionally interchangeable, recent data show that siRISC and miRISC are distinct complexes that regulate mRNA stability and translation in different interacting modes (). The terminology of miRNA and miRNA* or guide and passenger strands that is used is based upon differences in the thermodynamic stability of the 5′ and the 3′ strands. However, our data suggest that thermodynamic stability apparently is not an exclusively determining factor for the selective accumulation or co-accumulation of both strands of miRNAs. Naming each strand of a paired miRNA according to its strand origin (e.g. mir-30e-5p and mir-30e-3p) will be more accurate and less confusing. p p l e m e n t a r y d a t a a r e a v a i l a b l e a t N A R O n l i n e .

To understand cellular responses to endogenous genotoxic stress, it is important to have insight into the DNA repair process known as base excision repair (BER). BER is involved in repairing base lesions and single-strand breaks that occur thousands of times during the average mammalian cell cycle () and is generally considered to involve two subpathways defined by the size of the repair patch and the enzymes involved (). These subpathways are termed single-nucleotide (SN) BER and long-patch (LP) BER. As these names imply, in SN BER one nucleotide in the damaged strand is excised and replaced (), whereas LP BER excises and replaces several nucleotides in the damaged strand (,). LP BER appears to be a backup subpathway in cases where the SN BER system stalls or where the DNA lesion is refractory to the enzymatic steps of the SN BER subpathway (,,). Components of the BER system are constitutively expressed in mammalian cells, but also exhibit widely divergent tissue-specific expression levels plus up- or down-regulation after genotoxic stress and cytokine exposure (). This dynamic regulatory picture suggests the potential for cell type specific differences in cellular capacity for BER and also for differences in relative use of the two BER subpathways, since there are distinct enzymes and accessory factors involved in each (,,). Methods for characterization of the overall amount and efficiency of BER are limited, and improved methods for quantitative measurement of repair capacity and subpathway choice are needed for studies of extract-mediated BER. In recent years, research in the BER area has been greatly facilitated through measurements of BER activity using oligonucleotide and plasmid substrates containing a lesion in a defined site and sequence context (,,,). There has been debate, however, on advantages and disadvantages regarding the form of DNA used as substrate, i.e. plasmid versus oligonucleotide. For example, oligonucleotide substrates offer the advantage of relative ease of preparation of large amounts of pure material, and hence the option of preparing reaction mixtures with an excess of DNA substrate. Under this condition, steady-state measurements of repair activity of extracts can be obtained using short incubation periods (e.g. 1 min) and small amounts of protein extract (e.g. 1 μg). On the other hand, use of a plasmid substrate with the same lesion has involved reaction mixtures with only minimal concentrations of substrate DNA, in view of the difficulty of preparing large quantities of plasmid material. This limitation imposes a need for longer incubation periods with higher amounts of protein extract, in order to observe conversion of a significant amount of substrate into product. Such substrate-limiting, or efficiency-based, conditions can complicate interpretation of reaction kinetics. In addition, there are concerns regarding competing enzymatic activities such as nucleases, as well as plasmid substrate impurities that reduce the effective concentration of substrate. Nevertheless, there can be advantages in using plasmid substrate molecules. For example, artifacts due to protein binding to DNA ends can be eliminated. Also plasmids allow assembly of multi-protein complexes that are too large to be accommodated on oligonucleotide molecules. In addition, plasmids enable use of clamp proteins requiring long and/or circular substrate molecules, such as PCNA (). Finally, there has been a suggestion that plasmid DNA substrates are more relevant to events than oligonucleotide substrate molecules, however, this latter point has been controversial. Resolution of this debate and uncertainties about choice of substrate DNA can be facilitated by a direct comparison of BER kinetics obtained with each type of DNA bearing the same lesion and sequence context. This comparison has been enabled recently by the introduction of a method for plasmid DNA preparation that yields large amounts of purified plasmid in a relatively short period (). In the present study, we examined cell extract-mediated uracil-initiated BER using plasmid and oligonucleotide substrates with the U:G mispair in the same sequence context. Repair was measured under steady-state conditions for each type of substrate. Values for and were obtained using laboratory reference extracts from bovine testis and mouse embryonic fibroblasts (MEF). The results of these analyses indicated that plasmid and oligonucleotide substrates yielded similar values for uracil-DNA BER activity, although the catalytic efficiency (/) of repair with the oligonucleotide substrate was slightly higher than with the plasmid substrate. Methods for measuring the ratio of the two BER subpathways, SN BER and LP BER, and for evaluation of the LP BER repair patch size are also discussed. Synthetic oligodeoxyribonucleotides were from Integrated DNA Technologies, Inc. (Coralville, IA, USA) and Oligos Etc. Inc. (Wilsonville, OR, USA). The [α-P]dCTP and dTTP (3000 Ci/mmol) were from GE HealthCare (Piscataway, NJ, USA). DNA polymerase was from Stratagene (La Jolla, CA, USA). Endonuclease NB1, T4 DNA ligase and all other restriction enzymes were from New England Biolabs (Beverly, MA, USA). Plasmids were isolated using Qiagen Plasmid Maxi kits from Qiagen (Valencia, CA, USA). Streptavidin-coated magnetic beads (Dynabeads M-280) were from Invitrogen (Carlsbad, CA, USA). Recombinant human DNA polymerase β (Pol β) was overexpressed and purified as described previously (). Human uracil-DNA glycosylase (UDG), apurinic/apyrimidinic endonuclease (APE) and DNA ligase I were purified as described previously (). Lesion-specific BER plasmid substrates were prepared essentially as described previously () with slight modifications. Plasmid pUC19N was derived from pUC19 by inserting a 43-bp phosphorylated oligonucleotide (5′-GATCGAGTCGAATGCATGCTCGAGTCTA GAGGTACCAGATCT-3′) at the HI site position 417. During this process the HI site was lost. The resulting pUC19N contained two NB1 sites flanking 48 bases that could be substituted with a lesion-specific oligonucleotide. The sequence of the pUC19N plasmid was confirmed by sequencing. A 48-mer oligonucleotide in the sense strand that contained a defined lesion base [uracil or tetrahydrofuran (THF) at position 20, underscored above] was replaced between two NB1 sites as follows: pUC19N (100 μg, 57 pmol) was digested with 500 U of NB1 at 55°C for 2 to 3 h, followed by subsequent incubation with 200 pmol 3′-biotin-tagged complementary 48-mer oligonucleotide on a rotary shaker for 30 min at 37°C. Then, streptavidin-coated Dynabeads M-280 (400 pmol), pre-equilibrated with 10 mM Tris–HCl, pH 7.5, 0.5 mM EDTA and 1 M NaCl, were added to the reaction mixture, and the mixture was incubated for 2 h at 37°C. The 48-bp biotin-tagged DNA, adsorbed onto magnetic beads, was separated from the gapped-pUC19N DNA by placing the reaction mixture tube on a magnet for 2 min to collect the beads. The supernatant was carefully removed with a pipette while the tube remained on the magnet. Gapped-plasmid DNA in the supernatant was extracted twice with phenol/chloroform and precipitated with 95% ethanol. A 20-fold excess of uracil or THF lesion-containing 48-mer oligonucleotide (1.0 nmol) was added to the gapped-pUC19N DNA in 10 mM Tris–HCl, pH 8.0 and 50 mM NaCl. The DNA mixture was annealed at 45°C for 4 h, and then incubated at 25°C for 15 h with 5000 U of T4 DNA ligase. In order to purify the closed-circular DNA (ccDNA) plasmid from the un-ligated nicked DNA, two rounds of isopycnic centrifugation in CsCl with ethidium bromide were performed. The yield of lesion-specific plasmid DNA was ∼30–50% of the starting plasmid (i.e. ∼25 pmol). The purity of each plasmid preparation was evaluated by digestion with I and subsequent electrophoretic separation of ccDNA and linearized DNA. In this assay, the starting plasmid, pUC19N (plasmid without a lesion), was linearized by I digestion, whereas the lesion-specific substrates were not. This test for contaminating starting plasmid was considered an important precaution as contamination with the starting plasmid was found to be significant for some preparations, owing to inefficient removal of 48-mer NB1 DNA fragment and religation in the last step described above. The resulting plasmids containing either uracil or THF were designated as pUN1 or pUN2, respectively. Routinely, the ratio of lesion-specific plasmid to starting plasmid was greater than 3 : 1 in the experiments shown here. Unphosphorylated 55-mer oligonucleotide (100 mM) with 3′ inverted dT (5′-TCGGTACCCGGGGATCGAGT CGAATGCATGCTCGAGTCTAGAGGTACCAGAT CT-3′) containing either uracil or THF at position 32 (underscored ) was annealed to its complementary unphosphorylated 55-mer in 10 mM Tris–HCl, pH 8.0 and 100 mM NaCl, and then the DNA solution was diluted to 1 mM in the same buffer. The solution was divided into aliquots and stored at −80°C. The plasmid pGL4.10TKΔKpn (5.0 kb) was constructed from pGL4.10 (Promega, Madison, WI, USA) by inserting the TK promoter between I and III sites. During this process the I site was lost. For the BER plasmid assay, pPAL1 was prepared by replacing the III-I fragment of pGL4.10TKΔKpn with an oligonucleotide that contained a uracil at a new I site and unique I and I restriction sites. To prepare the III-I fragment, pGL4.10TK ΔKpn was digested with I and subsequently treated with calf intestine alkaline phosphatase. The resultant plasmid was digested by III and purified by 1% agarose gel electrophoresis. Then a 73/81-bp hybrid oligonucleotide was prepared by annealing unphosphorylated 73-mer 5′CTTCTTAATGTTTTTGGCATCTTCC ATGGTGGCTTTACCAAGGTAUCAGTAAGTATTA ATTAAGGAGAGCTCA-3′ and 5′-phosphorylated 81-mer 5′-AGCTTGAGCTCTCCTTAATTAATACTTACTGGTACCTTGGTAAAGCCACCATGGAAGAT GCCAAAAACATTAAGAAGGGCC-3′ DNA. Ligation of this oligonucleotide into the III-I digestion product of pGL4.10TKΔKpn was carried out as described (). The closed circular DNA was separated by 1% low-melting temperature agarose gel electrophoresis containing SYBR Gold nucleic acid gel stain (Invitrogen), and the plasmid was recovered using a gel extraction kit (Qiagen). The resulting plasmid, pPAL1, was obtained in microgram quantities. Nuclear extract was prepared from bovine testis essentially as described previously (). Briefly, 100 g of bovine testis was minced in 300 ml buffer A [10 mM Hepes, pH 8.0, 1.5 mM MgCl, 10 mM NaCl, 0.5 mM dithiothreitol (DTT), 10 mM sodium metabisulfite, 0.1 mM 4-(2-aminoethyl)benzenesulfonylfluoride (AEBSF), 1 mM benzamidine, 1 μg/ml leupeptin and 1 μg/ml pepstatin A] and homogenized using four 10 s bursts with a blender at 4°C. The homogenate was pelleted by centrifugation at 10 000 × for 10 min at 4°C. Pellet fraction was resuspended in 150 ml of buffer B (same as buffer A, except it contained 1 M NaCl) and blended using three 5 s bursts. The homogenate was centrifuged at 100 000 × for 1 h at 4°C. The clear supernatant fraction was brought to 40% saturation by adding solid ammonium sulfate slowly with stirring; stirring was continued for 1 h at 4°C. The precipitate was recovered by centrifugation at 14 000 × for 20 min at 4°C. The pellet fraction was resuspended in 50 ml of buffer C (25 mM Hepes, pH 7.5, 100 mM NaCl, 1 mM DTT, 10 mM sodium metabisulfite, 0.5 mM EDTA, 0.1 mM AEBSF, 1 mM benzamidine, 1 μg/ml leupeptin, 1 μg/ml pepstatin A and 20% glycerol) and dialyzed against the same buffer. The dialyzed nuclear extract was clarified by centrifugation at 10 000 × for 20 min. The clear supernatant fraction was referred to as the bovine testis nuclear extract (BTNE) and stored at −80°C. Protein concentration was determined with a Bio-Rad protein assay kit using bovine serum albumin (BSA) as a standard. Whole cell extract was prepared as previously described (). Briefly, cells were washed twice with phosphate-buffered saline at room temperature, detached by scraping, pelleted by centrifugation and resuspended in Buffer I (10 mM Tris–HCl, pH 7.8, 200 mM KCl and protease inhibitor cocktail). An equal volume of Buffer II (10 mM Tris–HCl, pH 7.8, 200 mM KCl, 2 mM EDTA, 40% glycerol, 0.2% Nonidet P-40, 2 mM DTT and protease inhibitor cocktail) was added. The suspension was rotated for 1 h at 4°C, and the resulting extract was clarified by centrifugation at 14 000 rpm at 4°C. The protein concentration of the extract was determined as above, and aliquots were stored at −80°C. Uracil- and THF-containing plasmids were designed as described above to quantify total BER, SN BER and LP BER in the same reaction. A 48-nt NB1 fragment containing uracil or THF () was inserted in pUC19, which upon digestion with strategic restriction enzymes, as indicated in , would generate fragments representing total BER product (41-bp I fragment), SN BER plus LP BER (25-bp I-I fragment) and LP BER (16-bp I-I fragment). BER assays were performed using reaction mixtures of 10 μl final volume that contained 20–500 nM plasmid or oligonucleotide DNA substrate, 10 μg of extract, 50 mM Tris–HCl, pH 7.5, 5 mM MgCl, 20 mM NaCl, 1 mM DTT, 4 mM ATP, 20 μM each of dATP, dGTP and dTTP or dCTP and 2.3 μM [α-P]dCTP or [α-P]dTTP, as indicated. The incubation was at 37°C for 1 to 30 min, as indicated. Reactions were terminated by adding 1 μl of 0.5 M EDTA and 90 μl TE (10 mM Tris–HCl, pH 8.0 and 1 mM EDTA), and this was followed by phenol/chloroform extraction and ethanol precipitation. To facilitate ethanol precipitation, 100 ng of carrier tRNA was added to each reaction mixture. The resulting DNA pellet fraction was dissolved in HO and divided into three equal portions. These DNA-containing portions were subjected to restriction enzyme digestion as follows: (i) no enzyme, (ii) I and (iii) I plus I. The reaction mixtures were incubated for 90 min at 37°C. The resulting reaction products were separated by 20% denaturing polyacrylamide gel electrophoresis (PAGE). The gels were dried, scanned with a Typhoon PhosphorImager and the data were analyzed with ImageQuant software. The BER assay, with uracil-containing plasmid, pPAL1, also was performed under similar reaction conditions as above, and the repaired DNA was restricted with I/I and/or I, as indicated in . The experiments were repeated several times, and representative images and average values of SN BER and LP BER are shown. BER assays using the 55-bp oligonucleotide duplex DNA and cellular extracts were performed under similar reaction conditions as with the plasmid DNA. A plasmid DNA substrate containing either uracil or THF opposite guanine was prepared as outlined in . Uracil opposite guanine was chosen because this mismatch is a widely used initiating lesion for studies of BER. The THF abasic site lesion was chosen as a reference because all BER with this lesion occurs through the LP BER subpathway, since DNA ligation after the gap-filling step in BER is blocked by the persistent THF group (,). A key feature in plasmid preparation was the use of the single-strand cutting restriction enzyme NB1, such that a single-strand gap of 48 nt was introduced into the plasmid and the released 48-mer was removed (, see penultimate step). Eventually, a synthetic 48-mer uracil-containing or THF-containing oligonucleotide was annealed into the gap, and the nicks were ligated with good yield (, final step). The plasmids (pUN1 or pUN2) thus prepared could be purified in milligram quantities, enabling the use of as much as micromolar final concentrations of plasmid substrate in BER reaction mixtures. In addition, these plasmids were designed such that the lesion was flanked with two I sites (A). Digestion of BER reaction products with this enzyme yielded a 41-nt fragment (41-mer). Upon denaturing gel electrophoresis, the incorporation of labeled dNMPs into the repair patch within this fragment could be measured. Quantitative assessment of the repair product formed during the BER incubation could be obtained by double digestion with I and I and use of P-labeled dCTP as one of the nucleotide substrates. In this way, incorporation of labeled dCMP into the first dNMP position of the repair patch was used to quantify the molar amount of repair product formed. This digestion also produced a 16-mer fragment corresponding to the second through seventeenth incorporations by the LP BER sub-pathway, and should contain most of the LP BER product (A). In some cases, I digestion alone yielded a 25-nt (25-mer) labeled fragment reflecting accumulation of stalled/unligated BER intermediate, i.e. intermediate prior to the DNA ligation step in SN BER or the strand displacement step in LP BER. In the experiments shown subsequently, this material was considered as background since it did not represent complete repair. Digestion with other restriction enzymes is also illustrated in . Two fragments are produced by double digestion with I and I, a 6-mer representing the second through seventh incorporations in the repair patch and a 23-mer representing incorporations out to the 30th position in the repair patch (A). With the oligonucleotide substrates, a related strategy was used for analysis of BER reaction products. These oligonucleotide substrates had lesions and sequences identical to those in the plasmid substrates (). In the case of products formed in reaction mixtures with labeled dCTP, digestion with I yielded a 32-mer containing the first labeled dCMP incorporated (B), allowing assessment of the molar amount of product formed. The stalled or unligated BER intermediate was observed in some cases, and again, this material was considered as background. The I digestion also yielded a 23-mer fragment containing the LP BER incorporations (B). Typical results obtained with a laboratory reference tissue extract, BTNE, and the plasmid and oligonucleotide substrates described above are shown in . Repair of the uracil-DNA plasmid substrate and digestion with I resulted in the labeled 41-mer fragment; very little stalled 25-mer intermediate was observed (A). In contrast, with the THF-containing plasmid substrate, a significant band representing the stalled intermediate was observed with I digestion alone (A). This indicated accumulation of the intermediate at the step just prior to strand displacement synthesis in LP BER. The results obtained with the oligonucleotide substrates without restriction enzyme digestion were similar, except the stalled intermediate was more abundant (B). Double digestion of the plasmid products yielded the 25-mer and 16-mer fragments described above (C). The results illustrated that most of the uracil-DNA repair corresponded to SN BER, as expected from earlier results (). Analysis of the product formed on the THF-containing oligonucleotide substrate revealed that the amount of stalled intermediate was greater than the LP BER product, and with uracil-DNA the results were similar to those with plasmid substrate (D). The main goal of this study was to compare kinetic features of BER product formation with plasmid and oligonucleotide substrates. We measured the initial rates of BTNE-mediated BER for the two forms of substrate under conditions of substrate excess. The rates of product formation with various uracil-DNA substrate concentrations were linear over the first 10 min of incubation, as illustrated at one substrate concentration in the experiment shown in . Kinetic features for BTNE-mediated BER with the two substrates are summarized in . The maximal velocities () of BER product formation for the plasmid and oligonucleotide substrates were similar, but the for the plasmid substrate, pUN1, was slightly higher than that for the oligonucleotide substrate. The resultant efficiency (/) was slightly higher (3-fold) with the oligonucleotide substrate. These results indicate that similar rates of uracil-DNA repair can be obtained with the two forms of substrate, especially when they are used at concentrations high enough to support near maximal synthesis, i.e. ≫ . In some cases, it is desirable to measure the relative amounts of the SN and LP BER subpathways. Adjusting the sequence of the repair patch to eliminate ambiguity from incorporation of more than one labeled dNMP in each subpathway can facilitate this measurement. To accomplish this type of measurement, we constructed an alternate plasmid substrate termed ‘pPAL1’ (A). This plasmid was designed for use in BER assays along with [α-P]dCTP as the labeled nucleotide substrate. To illustrate use of this method, incubations were conducted with MEF extract. Double digestion of the reaction product with I and I yielded a 47-mer representing the entire excision repair patch and a small amount of stalled intermediate (B). Digestion with these enzymes plus I yielded a 22-mer with the first dNMP incorporated, labeled dCMP, and a 25-mer with labeled dCMP corresponding to the second nucleotide incorporated and potentially the 25th dCMP incorporated into the repair patch (A). Since the repair patch in LP BER is shorter than 25 residues, potential dCMP incorporation at the 25th position can be ignored. Therefore, quantification of [α-P]dCMP incorporated into the 22-mer and 25-mer fragments, respectively, yields the molar amounts of total BER product (SN and LP BER) and LP BER product, as illustrated in . In this case, the ratio of SN BER to LP BER was approximately 60:40 (C), as expected from earlier results obtained by different methods (,). As noted in , the plasmid and oligonucleotide substrates can be used for repair patch size assessment of LP BER reaction products. Such analysis involves restriction enzyme digestion and then the electrophoretic separation and quantification of various labeled fragments (P-labeled dNMP incorporation). Typical results from this type of analysis are illustrated in . In the experiment shown, we used the THF-containing plasmid, pUN2, () so that repair corresponded to the LP BER subpathway only, along with an extract known to be proficient in LP BER (i.e. from MEFs). Measurement of dCMP and dTMP incorporated in separate incubations was conducted (). In this case for LP BER, the amount of dCMP incorporation into the first dNMP position (i.e. 25-mer) is equal to the amount of dTMP incorporation into the second dNMP position. Therefore, to evaluate LP BER synthesis beyond the second nucleotide, the amount of dTMP incorporated into the second position of the 16-mer could be subtracted from total dTMP incorporation into the 16-mer; this synthesis in the 16-mer for dTMP incorporation was similar to that for dCMP incorporation (B and C). Further analysis of repair patch size is illustrated in . Typical results (A) and the strategy for analysis using P-labeled dCMP incorporation (B) were as follows: Restriction with I yielded a 41-nt fragment (41-mer) representing repair patch synthesis, i.e. from one up to 17 nucleotide incorporations (A, lane 1, and B). With I digestion, some 25-mer BER intermediate was also observed, reflecting stalling at this intermediate. Incorporation of the first nucleotide in the repair patch, dCMP, was measured after double digestion with I and I, yielding the 25-mer and the 16-mer (A lane 2). The 16-mer fragment reflected three LP BER dCMP incorporations (A, lane 2, and B). For LP BER synthesis at the 2nd through 7th nucleotides and longer repair patches, respectively, we measured dCMP incorporation into the corresponding fragments after double digestions with I and I, i.e. two fragments (6-mer and 23-mer) as illustrated in (A, lane 3, and B). Incorporation into the 6-mer was greater than that into the 23-mer, indicating that most of the LP BER corresponded to repair patches of 7 nt or less (C). We made use of a recently acquired method for obtaining large amounts of plasmid DNA after a preparative ligation step (). A plasmid with a single-strand 48 base gap was prepared in high yield and in pure enough form to enable annealing a 48-mer lesion-containing oligonucleotide into the gap. After ligation and purification of the lesion-containing plasmid DNA substrate, a relatively large amount of DNA was available for use in routine BER incubations (). The availability of such large quantities of plasmid substrate made it possible for us to compare steady-state kinetic values for BER obtained with an oligonucleotide substrate with those obtained with a plasmid substrate. Such a comparison had been unresolved and remained an important consideration. Some investigators maintained that plasmid substrates are more biologically relevant or useful than oligonucleotide substrates. On the other hand, others have held the opposite viewpoint asserting that plasmid substrates are of limited and specialized use because of the necessity to employ miniscule concentrations of substrate DNA and long incubation periods, both of which potentially confound interpretations on BER capacity measurements. Our results indicated that the two forms of substrate yielded similar values for rates of repair in a nuclear extract, provided the DNA substrate concentration was high enough in the reaction mixture. A precaution emerged from the observation that for plasmid substrate was higher than for oligonucleotide substrate. An explanation for this difference was not examined. However, on a practical level, when a substrate excess or steady-state approach is used to quantify extract-based repair synthesis, the difference between the two forms of DNA may need to be taken into account. In addition to characterizing rates of BER as a function of substrate form, we examined methods for analysis of repair patch length in the LP BER subpathway. Our initial experiments in this effort made use of phosphorothioate dNMP incorporation into the repair patch and subsequent exonuclease III digestion of products [data not shown, and ref. ()]. This approach yielded spurious results (data not shown). The main difficulty stemmed from the capacity of exonuclease III to digest into the repair patches containing phosphorothioate dNMPs, leading to an underestimate of repair patch length. On the other hand, attempts to reduce the digestion strategically (time or enzyme concentration) generally lead to an overestimate of repair patch length. Therefore, we concluded that reliable measurements of repair patch length could not be obtained with this approach. The approach described here, however, appeared to be reliable and was limited only by the technical requirements to obtain pure substrate preparations, restriction digestion of product molecules to completion, and attention to accurate quantification of the amount of labeled dNMP incorporated.

Proteins that bind DNA play a critical role in regulating gene structure, replication and expression in all organisms. Biochemical and structural analyses of proteins that bind specific DNA sequences have begun to provide insight into the molecular basis of both DNA binding and sequence-specific DNA recognition (). These analyses have identified protein folds for DNA binding, together with a few general rules for the protein-mediated recognition of specific DNA bases (,,). This work has also begun to suggest ways to modify the recognition specificity of existing, sequence-specific DNA-binding proteins. Two classes of site-specific DNA-binding proteins that have been the focus for efforts to engineer new DNA recognition specificities are the Type II restriction endonucleases (), and sequence-specific transcription factors (,). The homing endonucleases, a group of highly sequence-specific DNA-binding proteins, are also being investigated for recognition specificity engineering. Four different families of homing endonuclease proteins have been identified on the basis of protein sequence comparisons, and one or more families have been identified in all Kingdoms of life (,). The physiologic role of homing endonucleases is to target the lateral transfer of parasitic DNA elements known as mobile introns by making a highly sequence-specific DNA double-strand break in an intron-less recipient allele (,). The high site specificity of many homing endonucleases reflects a combination of long (15–40 bp) DNA target or ‘homing’ sites, together with a high degree of sequence specificity at most target site base-pair positions. A second intrinsic property of many homing endonucleases is tight coupling of site recognition to catalysis (). This is a particularly attractive feature of homing endonucleases, in contrast to other potential genome engineering reagents such as zinc finger nucleases (). High site specificity and tight coupling of site binding to catalysis may reflect the evolutionary history of many homing endonucleases: these two properties in concert permit the continued lateral transfer—and thus the persistence—of endonuclease-encoding mobile introns to related target sites, while minimizing spurious chromosome cleavage events (,). Our aim was to determine whether structure-guided protein engineering could be used to alter the DNA recognition specificity of the eukaryotic homing endonuclease I-PpoI, a member of the His-Cys box family of homing endonucleases. I-PpoI was originally identified as an open reading frame in a self-splicing mobile intron found in extrachromosomal copies of the 28S rRNA genes of Physarum polycephalum, a Plasmodial myxomycete slime mold (). It is the best characterized member of the His-Cys box family of homing endonucleases, one family in the ααβ-Me or His-Me endonuclease superfamily (,), and has not thus far been a focus for structure-guided design. The active form of I-PpoI is a 36 kDa homodimer that cleaves a 15-bp semi-palindromic DNA target site in the 28S Physarum rDNA locus, and in the corresponding large subunit rRNA genes of all eukaryotes [(); ]. Rare target sites may also exist outside the rDNA repeats (). We had previously determined high-resolution apo- and co-crystal structures of native and mutant forms of I-PpoI that allowed us to identify the molecular basis for high affinity site binding and cleavage (,). We had also identified amino acid changes that interfere with I-PpoI catalysis or site binding, together with DNA base-pair changes that disrupted target site cleavage by native I-PpoI protein (). These data were used to construct a yeast one-hybrid (Y1H) screening assay to identify I-PpoI protein variants able to bind a specific mutant target site in vivo. Biochemical characterization together with computational modeling were used to gain insight into the molecular basis for mutant site recognition and cleavage by variant I-PpoI proteins. Yeast reporter plasmids were constructed using the YEp24 two-micron plasmid vector or the integrating plasmid vector pRS404 (,). An I-PpoI-specific YEp24-lacZ reporter was constructed by inserting I-PpoI target sites into the SalI site upstream of a cyc5 promoter and lacZ gene. Target site inserts were prepared by annealing phosphorylated oligonucleotides (PPOSITES5 and PPOSITES6; all oligonucleotide sequences are given in Supplementary Data Table 1), or oligonucleotides that when annealed created three oriented copies of native or mutant I-PpoI target sites (oligonucleotides PPOX3_WT and _6 #1-4). Yep24-HIS3Ppo was constructed from the resulting plasmids by replacing the lacZ gene with a PCR fragment containing the budding yeast HIS3 gene. An integrating version of Yep24-HIS3Ppo was constructed by transferring an EcoRI fragment containing a multiple-cloning site, minimal promoter and HIS3 gene into the yeast TRP1 integration vector pRS404, and then inserting three copies of the native or +6C/–6G mutant I-PpoI target site as described earlier Yeast containing an integrated copy of pRS404-HIS3Ppo were constructed by transforming MfeI-cleaved linear plasmid DNA into yeast strain W1588-4C (MATa ade2-1 can1-100 his3-11, 15 leu2-3, 112trp1-1 ura3-1 RAD5), followed by selection on SD media lacking tryptophan (,). Selection for HIS3 reporter expression was performed by growth on SD medium lacking histidine, followed by a demonstration of growth suppression by 3 mM 3-aminotriazole. All reporter plasmids and strains were verified by DNA sequencing of the I-PpoI target site region. #text The RD1 and RD2 libraries were constructed by replacing the region encompassing I-PpoI residues 55–76 with an insert generated by PCR-mediated assembly of degenerate oligonucleotides. Five oligonucleotides (the 4 +6G_B oligonucleotides and PPOOADN were used to assemble the N-terminal portion of the I-PpoI H98A ORF. Two +6G_E oligonucleotides and PPOOADC (RD1) or PPOOADCdeg (RD2) were used to assemble the C-terminal portion of the I-PpoI ORF. PPOADCdeg included a 9 bp randomized ‘molecular bar tag’ sequence to allow different starting plasmids to be distinguished. Both PCR assembly products were gel purified by polyacrylamide gel electrophoresis (PAGE). The N-terminal assembly product was extended in a second PCR reaction using a pool of 27 (RD1) or 18 (RD2) +6_M oligonucleotides together with PPOOADN. The resulting product was gel-purified and amplified with recombination cloning primers AD70F and AD70R (kindly provided by Stan Fields, University of WA; ) to generate a full-length I-PpoI open reading frame insert, and then used for in vivo gap repair of PvuII + NcoI-cleaved pOAD-Ppo plasmid DNA. Ppo-AD libraries were characterized by sequencing plasmid DNA isolated from independent colonies. Seventy-four CR library plasmids were sequenced, 11 from plasmid preps and 63 from PCR products generated using PADSCREENF and PADSCREENB oligonucleotides. RD1 and RD2 were characterized by sequencing yeast colony PCR products. PCR and the sequencing were performed with PADSEQF and 882 oligonucleotide primers. The resulting forward and reverse sequences were aligned using Sequencher (). xref #text CR or RD library screening was performed by transforming CR or RD plasmid libraries, together with the +6C/-6G reporter plasmid Yep24-, into cells possessing an integrated copy of the pRS404- reporter plasmid. The resulting cells were then selected for growth on histidine-minus media followed by a visual screen for β-galactosidase activity. The number of independent co-transformants was determined by plating a small known fraction of the transformed yeast on SD–leucine–uracil plates. Library screening was performed by dense plating of transformed cells on SD–leucine–uracil–histidine plates supplemented with 3 mM 3-aminotriazole, followed by growth at 30°C for 7 days. Large colonies were then picked into 1 ml 96-well deep blocks containing SD–leucine–uracil media and grown for 7 days without shaking at 30°C. A colorometric screen for β-galactosidase reporter gene induction was used to identify Ppo-AD plasmids with site-binding activity. In brief, cells were pelleted, then resuspended and permeabilized in Z-buffer using SDS and chloroform (). The β-galactosidase substrate ONPG was added to 0.6 mg/ml, and wells were incubated at 30°C to monitor color change. Incubations were stopped between 2 and 4 h by adding NaCO prior to visual screening to identify positive wells by comparison against positive control (H98A Ppo-AD and native I-PpoI site lacZ reporter) or negative control (H98A Ppo-AD and mutant +6C/–6G site lacZ reporter) cells. Quantitative β-galactosidase activity assays were performed to further characterize several I-PpoI variants by using a modification of the screening assay as described above (). Cells were grown overnight at 30°C in SD–leucine–uracil media. Cell density was determined by measuring absorbance at 600 nm, and 2 ml of each culture was pelleted and resuspended in 0.8 ml of Z-buffer followed by the addition of 50 μl 0.1% SDS and 50 μl chloroform with vortexing for 30 s. Activity was measured by adding 160 μl of 4 mg/ml ONPG in Z-buffer followed by incubation at 30°C, and stopped by adding 0.4 ml NaCO after the development of yellow color. Cell debris was spun out of reactions prior to measuring absorbance at 420 nm and calculating units of β-galactosidase activity. All reported values represent an average of three independent determinations. Protein was purified from frozen cell pellets by thawing and resuspending cells in 1 ml of lysis buffer (50 mM NaHPO, 300 mM NaCl and 10 mM imidazole) containing 1 mg/ml lysozyme on ice, followed by the addition of protease inhibitors (1 mM PMSF, 1 μg/ml leupeptin and 1 μg/ml pepstatin) and sonication. RNAse A (10 μg/ml) and DNAse I (5 μg/ml) were added to cell lysates followed by incubation on ice for 15 min. Cell debris was pelleted at 4°C, and Ni-NTA resin (100 μl; Qiagen) was added to the cleared supernatant prior to incubation at 4°C on a roller for 1–2 h. Resin was pelleted from the supernatant in a benchtop centrifuge at 4°C, washed once with 1 ml of 20 mM imidazole buffer (50 mM NaHPO, 300 mM NaCl and 20 mM imidazole), and then washed twice with 1 ml 50 mM imidazole buffer (50 mM NaHPO, 300 mM NaCl and 50 mM imidazole). Protein was eluted with 3 × 50 μl washes of 50 mM NaHPO, 300 mM NaCl and 250 mM imidazole. Eluates were pooled to determine protein concentration by Bradford assay prior to the addition of glycerol to 50% (v/v) and storage at −20°C. Polyacrylamide gel electrophoresis of protein samples followed by Coommassie staining indicated that I-PpoI protein preparations were typically >75% pure with no other major contaminating bands. The binding affinity () and cleavage of I-PpoI target sites were determined as previously described (). The substrates for binding affinity assays were formed by annealing gel-purified PPOx1 oligonucleotide pairs in which (+) strand oligonucleotides had been end-labeled with P. values were determined from a minimum of two, or in most cases three, independent assays. Cleavage assays were performed using linearized plasmids containing either three copies of native or mutant I-PpoI target sites inserted into the SalI site of -, or single site PPOx1_Sal oligonucleotides pairs that had been annealed and inserted into pBlueScriptII KS(+) plasmid DNA (Stratagene). pRS404 plasmids were linearized with SnaBI, and pBluescriptII plasmids with XmnI, prior to performing cleavage assays. Modeling of the I-PpoI DNA interface of native and variant proteins was done using RosettaDesign (,). In brief, changes to the substrate DNA sequence and the amino acid sequence of the protein were simulated , and the DNA–protein complex was allowed to relax according to an energy function that mimics protein folding. I-PpoI variants E, G and T were modeled and visualized for inspection using the PyMOL molecular viewer (DeLano Scientific LLC). We identified I-PpoI target sites that were cleavage-resistant when challenged with native I-PpoI protein. We focused initially on 5-bp positions, ±3 to ±7, and on base-pair substitutions at these sites that were known to affect cleavage efficiency on the basis of previous homing site degeneracy, biochemical or functional analyses (,,). These base-pair positions are in a well-ordered part of the I-PpoI DNA–protein interface (A). Palindromic ±3 G>A/C>T and ±4 G>C/T>G base-pair substitutions were permissive of cleavage only at high protein concentrations (10 nM for the ±3 position; 1 μM for the ±4 position). In contrast, base-pair substitutions at positions ±5 (T>A/C>T), ±6 (A>C/T>G) and ±7 (G>A/C>T) completely inhibited cleavage at all protein concentrations tested (D; data not shown). Of the six possible base-pair changes from the native +6A/−6T target site, +6A>C/−6T>G substitutions most strongly inhibited site cleavage by native I-PpoI (C and D; data not shown). We chose this target site for re-engineering of the site recognition specificity of I-PpoI. A yeast one-hybrid (Y1H) screen was used to identify I-PpoI protein variants that bound the +6C/−6G target site to induce reporter gene expression. To establish this assay, we fused a Gal4p transcriptional activation domain to the N-terminus of catalytically inactive H98A I-PpoI to generate Ppo-AD. The H98A substitution was incorporated to abolish the catalytic activity of I-PpoI without disrupting high affinity site binding [A; (,,)]. A Ppo-AD negative control protein was constructed by fusing L116A I-PpoI to the Gal4p activation domain. L116A I-PpoI is catalytically inactive, and does not bind I-PpoI target site DNA (). Reporter activity was determined using a budding yeast or plasmid-borne bacterial reporter genes located downstream of 1 or more I-PpoI target site and a minimal promoter (B). Site binding was quantified by the induction of β-galactosidase activity, and was strongest when the reporter gene was flanked by 6 direct repeats of the native I-PpoI target site: 25-to-75-fold above the ‘no site’ or negative control background. The Gal4-L116A Ppo-AD negative control protein did not induce reporter expression above background on any target site or number of repeats (C; data not shown). We performed all subsequent library screens using reporter genes flanked by three target sites, as these reporters could be readily constructed and gave reporter activity well above background. In order to generate protein variants to screen against the mutant +6C/−6G target site, we made substitutions in the I-PpoI DNA–protein interface at residues that contacted base–pair position 6 or adjacent base–pair positions. This ‘contacting residue’ (CR) library was designed to allow all 20 amino acid residues at positions N57, R61, Q63, K65 and R74, and had a maximum potential complexity of 3.2 × 10 protein variants. The resulting CR library was difficult to construct, and thus not large enough to encompass all predicted variants. Upon experimental verification, the CR library was found to encode 1.6 × 105 different protein variants that each had an average of 4.7 amino acid substitutions (CR library, ). The CR library was screened by Y1H in two steps. We isolated yeast transformants that grew on histidine-minus plates containing 3-aminotriazole (3-AT), and then determined whether the same Ppo-AD plasmids could induce β-galactosidase expression. The Ppo-AD open reading frame of active variants was then sequenced to identify amino acid substitution(s) that conferred activity. Among the 2 × 10 colony transformants and 4600 His+ colonies screened, only one CR library transformant had activity on both reporter genes. This variant (variant A, ) had four amino acid substitutions at residue positions randomized during CR library construction. Variant A served as a starting point for the construction of two additional, rationally designed (RD) libraries that were used to further explore determinants of I-PpoI recognition and activity on native and +6C/−6G target sites. The first RD library allowed +6C/−6G contacting residue 63 to vary among seven amino acid residues: the native Q, or D, E, K, N, R or Y. Neighboring β-strand residues A55, N57, W62, Y64, R74 and G76 were substituted with residues that differed slightly in size and/or polarity from native residues found at these positions (RD1; ). The predicted maximum complexity of the RD1 library was 3.5 × 10. Sequencing indicated that 72% of RD1 Ppo-AD plasmids had intact open reading frames. Thus the completed RD1 library of 1.8 × 10 members was sufficiently large to include all RD1 design variants (RD1, ). screening of RD1 against +6C/−6G target site reporter genes yielded 352 His+ colonies, and 15 plasmids from this pool were verified by re-transformation and quantitative activity prior to sequencing. Five active Ppo-AD plasmids had the same Ppo-AD amino acid substitutions found in variant A (). Six different RD1 starting plasmids harbored the same Q63R and K65R substitutions (, variant E), as indicated by plasmid DNA sequence differences. Four other plasmids had the variant E Q63R and K65R substitutions together with 2 to 4 other substitutions at residues targeted during library construction (variants B, C, D and F; ). A second rationally designed library was constructed to determine whether additional residue substitutions could augment the activity of Ppo-AD proteins identified in the CR and RD1 libraries. The RD2 library design features were: (i) three positions were fixed as residues found in active variants of RD1 (63R, 64Y and 76G); (ii) residue 55 was an A, G, L or V, and residue 62 was an F or W, again as previously identified in active variants; (iii) contacting residues 57, 61, 65 and 74 were varied among eight amino acid residues commonly found in homing endonuclease DNA–protein interfaces (Y, E, D, R, K, Q, N, C; , RD2) and (iv) inclusion of a 9 bp post-C-terminal ‘molecular bar code’ to allow different RD2 plasmid molecules to be identified and distinguished. The RD2 library had a predicted maximum complexity of 3.2 × 10. Sequencing indicated that 67% of RD2 library plasmids had intact Ppo-AD open reading frames. Thus the completed RD2 library of 1.4 × 10 members was sufficiently large to include all designed RD2 variants. Thirty-five RD2 plasmids displayed activity against +6C/–6G reporter genes. These plasmids encoded nine different Ppo-AD proteins: seven Ppo-AD variants were unique to the RD2 library (variants G, H, J, K, L, M and N), whereas two variants had been previously identified in RD1 (variants D and E; ). Ppo-AD variants D, H, K and L were independently isolated, from 3 to 6 times each from RD2, as indicated by plasmid molecular bar code sequences. In addition to these library variants, we constructed nine I-PpoI variants (variants P to Y, ) by site-directed mutagenesis. This was done to determine the contribution of specific amino acid residues identified in the CR, RD1 and RD2 libraries to target site binding and cleavage. We reasoned that specific residue substitutions in library isolates might promote or interfere with site binding and cleavage, and thus wanted to assay individual substitutions in defined sequence contexts. These site-directed mutants were generated, in consequence, in either the native or variant A or C I-PpoI sequence contexts. In order to characterize the site-binding properties of Ppo-AD variants, we expressed and purified variant proteins from then determined their dissociation constants ('s) on native and on +6C/–6G target site DNAs. We also determined the ability of selected variants to induce reporter gene expression to allow a comparison of and site-binding properties (). The I-PpoI portion of each Ppo-AD fusion protein was subcloned into a pET11c expression vector containing an N-terminal six-histidine tag to facilitate purification. In order to assay site cleavage as well as binding by specific variant proteins, we used site-directed mutagenesis to revert the catalytically inactivating H98A substitution required to establish our Y1H assay. Dissociation constants for variant proteins were determined by electrophoretic gel mobility shift analysis using +6C/−6G and native target site DNAs. Fourteen of 22 variants displayed a higher affinity for +6C/−6G homing site DNA than did native I-PpoI, although this site-binding preference was modest (typically <5-fold; ). Seven of the remaining variants displayed no site-binding preference. One variant, T, displayed high binding affinity for both the native and +6C/−6G target sites and a 175-fold higher affinity for native target site DNA (). We also determined the binding affinity of native I-PpoI and four variant proteins on target site DNAs with base-pair substitutions at the ±6 position, the ±7 position and in one instance ±6/±7 position substitutions. The variants had amino acid substitutions at position 63 contacting the ±6 bp; at residue 74 contacting the ±7 bp; and at residues 55, 57 and 65 in adjacent regions of the DNA–protein interface. None of these variant proteins bound any mutant target site DNA with high site affinity or specificity (, and Supplementary ). Seven of 22 variant I-PpoI proteins cleaved native homing site DNA . Cleavage activity was determined on linear plasmid substrates that contained a single I-PpoI target site. Both native I-PpoI and variant T bound and cleaved native target site DNA, though did not cleave +6C/–6G target site DNA in the presence of Mg or Mn (A; data not shown). Other variant proteins with activity cleaved only native target site DNA (variants E, P, Q, S, U and Y; , A). In order to better understand the relationship between biochemical properties and activity of variant proteins, we quantified the ability of six variants to induce a reporter gene in budding yeast. Variants A, C, E, G, J and T were chosen for analysis as they represent a range of binding affinities. Expression plasmids for each variant as a Ppo-AD fusion were transformed into yeast containing a native or +6C/–6G target site reporter gene plasmid. β-Galactosidase activity was quantified for three independent colonies grown in liquid media and assayed on the same day. Protein variants C, G and J had higher activity on +6C/–6G than on native target site reporters. This paralleled the 1.8-to-5.3-fold higher binding affinity of these variant proteins for +6C/–6G sites ( and B). A graphical summary of site binding and site cleavage analyses, for all variants, is shown in (panels C and D). Structural modeling of base pair and residue substitutions in the I-PpoI DNA–protein interface was used to gain mechanistic insight into the properties of several I-PpoI variant proteins. When I-PpoI is bound to native target site DNA, Q63 makes an energetically favorable, glutamine:adenine contact with two hydrogen bonds to adenine 6. R74 makes a canonical contact with guanine 7. K65 makes an H-bond contact to guanine 9, whereas R61 makes a series of backbone contacts between base pairs 3 and 4 (A). Modeling indicates the +6C/–6G base-pair substitution inhibits site binding and cleavage by disrupting the glutamine:adenine 6 contact, and by forcing the Q63 residue to rotate 135° into and in part disrupt the adjacent R74:guanine 7 bp contact (A and B). Several residue substitutions found in variant G could in part restore mutant +6C/–6G target site binding. These were Q63R at the +6C/–6G base-pair-contacting position (present in 17 variants) K65R (present in 17 variants) and R61K (present in 4 variants; and , compare panels B and C). Modeling indicates that Q63R substitutions can in part restore mutant site-binding affinity by making a new contact at base-pair position 5, and by restoring an R74:guanine 7 canonical base-pair contact (C). Modeling also provides an explanation for why selected variants with Q63R substitutions, e.g. variant E, show preferential binding of the mutant +6C/−6G target site: when R63 packs in the native DNA interface, it leaves a small cavity adjacent to the native ±6 A/T base pairs. This cavity is partially filled by the cytosine in the +6C/–6G substrate (A and B; data not shown). These results indicate that the native contacting residue, Q63, acts as a gatekeeper at the ±6 A:T base pair by controlling contacts at this and adjacent base-pair positions. Variant T, which contains a K65R substitution, showed a marked increase in binding affinity for both native and +6C/–6C target sites. Modeling of this substitution revealed that K65 forms a single H-bond to the guanine base at base pair 9, whereas K65R substitution make two H-bonds to base pairs at positions 8 9 (compare panels B and C). These contacts are energetically more favorable than the single contact made by the native K65 residue (−1.22 units for R65 versus −1.09 units for K65; RosettaDesign analyses not shown). Neither K nor R65 residues contact target site position 6, and thus do not discriminate between native or mutant +6C/–6G target site DNAs. R61K substitutions decrease the DNA-binding affinity of I-PpoI, while increasing target site selectivity to favor the binding of mutant, +6C/–6G target site DNA (compare, e.g. variants E and G, ). Modeling indicates that the R61K substitution modifies backbone contacts in the native interface between base pair positions 3 and 4 (, compare panels B and C). These backbone positions are immediately adjacent to the scissile phosphate, and form part of the most deformed region of the I-PpoI:DNA substrate complex (). The selectivity of K61 variants for mutant target site DNA is likely explained by sequence-dependent conformation changes in the DNA–protein interface, as no new contacts are established with mutant target site DNA. We used a yeast one-hybrid (Y1H) assay to isolate and characterize variants of the I-PpoI homing endonuclease with altered DNA target recognition specificity. Variants were isolated using a mutant binding site target with symmetrical ±6 bp position substitutions that abolished site binding and cleavage by native I-PpoI. We reasoned that amino acid substitutions in the DNA–protein interface adjacent to this target site position, at contacting residue 63 alone or in conjunction with other substitutions, might restore high site-binding affinity and specificity . This strategy resembles Y1H screens that have been used to identify and characterize site-specific DNA-binding proteins from yeast and other organisms. One similar precedent reported for homing endonucleases was the use of a bacterial two-hybrid screen to explore the recognition specificity of the LAGLIDADG homing endonuclease PI-SceI (). The variant I-PpoI proteins we generated contained from 1 to 8 amino acid substitutions in the DNA–protein interface ( and ). Most variants had low site-binding affinities (100–1000 nM 's), and a modest (2-to-10-fold) preference to bind native or mutant target site DNA. One exception, variant T with only a K65R substitution, had high binding affinities for native and mutant target site DNAs that included a clear (<175-fold) preference for native site binding. Cleavage competence was assayed by reverting the catalytically inactivating H98A substitution required to establish the Y1H screen, and then assaying proteins for the ability to cleave native or mutant target site DNAs. We found little correlation among site binding, discrimination and cleavage properties of individual variant proteins ( and , panels C and D). One likely explanation for this is that I-PpoI must bind and bend substrate DNA in order to generate a productive substrate complex (,,). Our hypothesis is that active variants (e.g. variants U, S or P, A and ) retain the ability to both bind and bend site DNA to permit cleavage. Structure-based molecular modeling of I-PpoI variant proteins on both native and mutant target site DNAs was used to gain insight into the contribution of specific residue substitutions to the biochemical behavior of variant proteins. Modeling revealed that Q63R substitutions fail to confer ±6 mutant target site specificity by virtue of loss of a canonical glutamine:adenine contact at position 6, together with partial disruption of a canonical arginine:guanine contact at the adjacent base pair position 7. Q63R variants were nonetheless able to discriminate in favor of mutant +6C/–6G target sites by packing more favorably with mutant 6C than with the native 6A target site base, while making a high quality contact at the adjacent position 5 base pair (). K65R substitutions, in contrast, increase affinity for native ±6 mutant target sites by making an additional H-bond to bridge base pair positions 8 and 9 [ and (panels B and C)]. R61K substitutions, in contrast, appear to alter site binding by modifying DNA backbone contacts immediately adjacent to the scissile phosphate located between base pairs 2 and 3 [(); ]. This type of indirect readout may be strongly influenced by local, sequence-dependent DNA conformation (,). These sequence-dependent effects may be further amplified by DNA substrate deformation in this region of the I-PpoI substrate complex, and thus interfere with correct positioning of the scissile phosphate ( and ) (,). The Y1H screening assay we used employed the canonical two-hybrid reporters and that, respectively, confer a growth advantage or have readily detectable activity at low expression levels. The sensitivity of these reporters, in retrospect, permitted the identification of I-PpoI variants with modest activity and site-binding affinity ( and ). One explanation for the failure to recover I-PpoI variants with high affinity and specificity for +6C/–6G mutant target site DNA is that they were not present in our starting libraries. This may be the case for the initial randomized contacting residue (CR) library, which was too large to be exhaustively screened. The smaller rationally designed libraries (RD1 and RD2), in contrast, were exhaustively screened but may have been too small to encode a high affinity binding variant. The experimental and computational analyses described above indicate one productive approach to alter the DNA recognition specificity of I-PpoI and other homing endonuclease proteins. Computational DNA–protein interface design can be used to predict different residue substitutions that may confer high binding affinity and specificity for a mutant target site DNA. The resulting protein variants can then be rank-ordered on the basis of predicted DNA-binding energies, and further evaluated for structural quality by molecular modeling. This general approach has already been shown to work for single base-pair positions in I-MsoI, a member of the LAGLIDADG homing endonuclease family (), and should be applicable to His-Cys box proteins such as I-PpoI. A prerequisite for this engineering approach is a high resolution co-crystal structure. In addition to protein computational design, it should also be possible to improve the experimental selection or screening assays to identify variant proteins with high site-binding affinity and specificity. For example, the Y1H assay could be adapted to use reporter genes or selections that would require comparatively high levels of expression to identify active variants. Alternatively, a combination of positive and negative selection could be used to recover variants with high site-binding affinity and specificity (). Homing endonucleases remain the most attractive starting point for the generation of new, highly sequence-specific proteins for biology and medicine. They encode a wide range of different DNA recognition specificities, and display high site-binding specificity that is tightly linked to DNA cleavage. Homing endonucleases including I-PpoI have already been successfully expressed in human cells. They cleave their target sites with high site specificity (,), and thus are being used to promote site-specific recombination () and high resolution DNA double-strand break repair analyses (). The precedents outlined above indicate that it should be possible in the near future to generate many highly sequence-specific homing endonuclease variants useful for genome engineering, disease therapy or disease prevention. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Mammals have evolved complex adaptative mechanisms that enable cells to survive many stressful environmental conditions including amino acid limitation. The signal transduction pathway that is triggered in response to amino acid starvation is referred to as the amino acid response (AAR) (). The initial step in the AAR is the activation by uncharged tRNAs of GCN2 kinase which phosphorylates the α subunit of translation initiation factor eIF2 (eIF2α) on serine 51 (,). This phosphorylation decreases the translation of most mRNAs by inhibiting the delivery of the initiator Met-tRNA to the initiation complex. However, eIF2α phosphorylation also triggers the translation of specific mRNAs including the activating transcription factor 4 (ATF4). Once induced, ATF4 directly or indirectly induces transcription of specific target genes (). Among the genes induced via the GCN2/ATF4 pathway, the CCAAT/enhancer-binding protein homologous protein () encodes a ubiquitous transcription factor that heterodimerizes avidly with the other members of the C/EBP and jun/fos families (). The amino acid regulation of transcription involves a -acting element in the promoter that has been named amino acid response element (AARE) (). This element is essential for the induction of transcription by amino acid starvation and functions as an enhancer element. In the past few years, several functional AAREs have been described in other genes including () (,) and () (). The AARE sites of , and have a 9 bp core element (5′-/TT/CATCA-3′) but the sequences differ by one or two nucleotides between genes. It is now established that in amino acid-starved cells, a multiproteic complex is bound to the AARE sequences including a number of regulatory proteins such as ATF4 (,,), C/EBPβ (), ATF3 (,), activating transcription factor 2 (ATF2) (,) or tribbles-related protein 3 (TRB3) (). These factors are involved in either inducing or repressing transcription of target genes in response to amino acid starvation. Importantly, all of the known AARE sites bind ATF4 whereas the binding activity and the role of other AARE-binding factors appear to vary according to the AARE sequence and chromatin structure. For example, and sequences also bind ATF2 whereas the site does not (,,,). The key role of ATF4 in amino acid-regulated transcription has been clearly established in the past few years (,,,). It has been shown that (i) the expression of ATF4 and its binding to AARE sequences are increased following amino acid starvation, (ii) kinetic of ATF4 binding to AARE are similar between tested genes with a dramatic increase in the first hour after a single amino acid removal, sustained over the next two hours, (iii) in cells devoid of ATF4 expression, the induction of mammalian genes upon amino acid starvation is completely lost and (iv) when over-expressed, ATF4 by itself is able to activate the AARE-dependent transcription. One major role of ATF4 is to mediate part of cell response to stress signals such as ER stress or amino acid deprivation (). Both transcription and translation of ATF4 are selectively increased in stress conditions (,), resulting in the induction of many genes involved in amino acid metabolism or transport and in resistance to oxidative stress (). ATF4 belongs to the basic region/leucine zipper (bZIP) family of transcription factors, which also includes members of the Jun/Fos (AP-1) family (,). This factor is known to form heterodimers with members of AP1 and CCAAT/enhancer-binding protein (C/EBP) families () rather than proteins of the ATF/CREB family (). Heterodimerization of ATF4 represents a powerful means to regulate its transcriptional activity and consequently the expression of target genes. ATF4 also interacts with coactivators such as p300 and CBP (,) and with several general transcription factors such as TBP, TFIIB, RAP30 () and RPB3 (). In the context of the AAR, the heterodimeric partner of ATF4 on AARE remains to be identified. It has been suggested that ATF4 may also interact with one or more cofactors to make the promoter more accessible to the general transcription machinery () but these cofactors also remain to be identified. The present study was designed to identify proteins interacting with ATF4 and playing a role in the transcriptional activation of in response to amino acid starvation. Recent progress in mass spectrometric protein sequencing technology together with the rapid growth of protein and genome databases have made direct approaches to map protein–protein interactions feasible. Using a tandem affinity purification (TAP) tag approach, we identify p300/CBP-associated factor (PCAF) as a novel interaction partner of ATF4 in amino acid-starved cells. Our results provide evidence that PCAF acts as a coactivator of ATF4 and is involved in the enhancement of transcription following amino acid starvation. 2X-CHOP-AARE-TK-LUC was generated as previously described (). To express PCAF in mammalian cells, plasmids for wild-type (pCX-Flag-PCAF) and HAT-defective PCAF (pCX-Flag-PCAF ΔHAT) were provided by Chao-Zhong Song (University of Washington, Seattle, WA). To express PCAF , pCI-Flag-PCAF was kindly given by Rosemary Kiernan (Montpellier, France). The expression plasmid for the ATF4 cDNA was a gift of Irina Lassot (Institut Cochin, Paris, France). To generate the GST–ATF4 (amino acids 1–351) and GST–ATF4 (amino acids 1–100) fusion proteins, the corresponding regions of human ATF4 cDNA were amplified by PCR and inserted into the BamHI/EcoRI sites of pGEX-4T-1 (Amersham). Constructs including other ATF4 deletion mutants fused to GST (glutathione-S-transferase) were kindly provided by Florence Margottin-Goguet (Institut Cochin, Paris, France). The mammalian ATF4 expression plasmid used in the TAP technique (pZome-1-N-TAP-ATF4) was generated by inserting the full-length coding region of human ATF4 cDNA into the EcoRI site of pZome-1-N (Euroscarf, Germany). HeLa cells, mouse embryonic fibroblasts (MEF) and retroviral packaging cell line BOSC23 were cultured at 37°C in Dulbecco's modified Eagle's medium F12 (DMEM F12) (Sigma) containing 10% fetal bovine serum. When indicated, DMEM F12 lacking leucine was used. In all experiments involving amino acid starvation, 10% dialyzed calf serum was used. MEF deficient in ATF4 were kindly given by David Ron (Skirball Institute of Biomolecular Medicine, New York) (). Retroviral infection was performed as described (). BOSC23 cells were transfected with either pZome-1-N (mock) or pZome-1-N-TAP-ATF4. After 48 h of transfection, the medium containing retroviruses was collected, filtered, treated by polybrene (1 mg/ml) and transferred on ATF4 −/− MEF. Infected cells were selected with puromycin (2 mg/ml) for 3 weeks. The expression of TAP-ATF4 was analyzed by immunoblotting analysis with an anti-ATF4 antibody. The antibodies against ATF4 (sc-200), PCAF (sc-8999) and β-actin (sc-7210) were purchased from Santa Cruz Biotechnology, Inc (Santa Cruz, CA, USA). Nuclear extracts were prepared from HeLa cells and MEF as described previously (). TAP–ATF4 complexes were purified using a published procedure () with minor modifications. Nuclear extracts were prepared from thirty 150-mm plates of mock or ATF4–TAP-transfected cells and subsequently adjusted to IgG-binding conditions: 180 mM NaCl, 10 mM Tris–HCl pH 8.0, 0.2% NP-40, 0.5 mM dithiothreitol (DTT), complete protease inhibitors (Sigma), 10 mM β-glycerophosphate and 20 mM NaF. Diluted extracts were rotated overnight at 4°C with 100 μl of IgG matrix (Amersham Biotech), after which the beads were washed extensively in binding buffer. Washed beads were re-suspended in TEV cleavage buffer (10 mM Tris–HCl pH 8.0, 150 mM NaCl, 0.3% NP-40, 0.5 mM EDTA, 0.5 mM DTT), and 5–15 μl of recombinant TEV enzyme (Invitrogen) was added to the mixture. After 2 h of rotation at 16°C, the TEV eluate from the IgG column was recovered and adjusted to calmodulin-binding conditions: 150 mM NaCl, 45 mM Tris–HCl pH 8.0, 0.7 mM Mg-acetate, 0.7 mM imidazole, 2.5 mM CaCl, 0.2% NP-40, 10 mM β-mercaptoethanol and rotated for 2 h at 4°C with 50 μl of calmodulin affinity resin (Stratagene). After binding, sedimented beads were washed extensively with calmodulin-binding buffer. Bound proteins were recovered by boiling the calmodulin beads for 5 min in protein sample buffer. The protein complexes recovered from TAP purification were fractionated on a 7% SDS–polyacrylamide gel. Proteins were detected by silver staining. The protein bands on 1D gel were excised from gels using blade of scapel. The bands were washed with 100 µl of 25 mM NHHCO for 30 min, destained with 100 µl of 25 mM NHHCO/acetonitrile (v/v) twice 30 min and dehydrated in acetonitrile. Bands were completely dried using a speed vac before trypsin digestion. The dried gel volume was evaluated and three volumes of trypsin (V5111; Promega, Madison, WI, USA), 10 ng/µl in 25 mM NHHCO were added. Digestion was performed at 37°C during 5 h. The gels pieces were centrifuged and 8–12 µl of acetonitrile (depending of gel volume) were added to extract peptides. The mixture was sonicated for 5 min and centrifuged. For MALDI-TOF MS analysis, 1 µl of supernatant was loaded directly onto the MALDI target. The matrix solution (5 mg/ml -cyano-4-hydroxycinnamic acid in 50% acetonitrile/0.1% trifluoroacetic acid) was added immediately and allowed to dry at room temperature. A Voyager DE-Pro model of MALDI-TOF mass spectrometer (Perseptive BioSystems, Farmingham, MA, USA) was used in positive-ion reflector mode for peptide mass fingerprinting. External calibration was performed with a standard peptide solution (Proteomix, LaserBio Labs, Sophia-Antipolis, France). Internal calibration was performed using peptides resulting from auto-digestion of porcine trypsin. Monoisotopic peptide masses were assigned and used from NCBI database searches with the ‘Mascot’ and ‘Profound’ softwares ( and ). Fresh overnight cultures of BL21 pLysS strain transformed with GST-fused constructs were diluted 1:10 in LB medium containing ampicillin (100 μg/ml). Isopropyl-1-thio-β--galactopyranoside was added in growing exponential bacterial culture to a final concentration of 1 mM and incubated for 4 h at 30°C. Cells were re-suspended in STE buffer (10 mM Tris pH 8, 150 mM NaCl, 1 mM EDTA). After 10 000 centrifugation during 10 min at 4°C, pellets were frozen during 5 min, re-suspended in STE buffer containing 1 mM dithiothreitol and 10% sarcosyl. Lysates were sonicated for 1 min and clarified at 10 000 for 10 min at 4°C. The bacterial supernatant was rocked overnight at 4°C with glutathione-Agarose resin (Sigma) and beads were washed three times in PBS containing Triton X-100 (0.1%) and PMSF (1 mM). S-labeled PCAF protein was generated using the TNT T7-coupled reticulocyte lysate system (Promega) according to the manufacturer's instructions. S-labeled PCAF protein or 1 mg of nuclear extracts were incubated with the beads in binding buffer (20 mM HEPES pH 7.4, 125 mM NaCl, 0.1% Triton X-100, 2 mM DTT), 2 mM ethylenediamine tetra-acetic acid (EDTA), 10 μM ZnCl, complete protease inhibitors (Sigma) by rocking 2 h at 4°C. The glutathione-Agarose beads were then washed four times in PBS buffer containing 0.1% Triton X-100 and 160 mM NaCl and re-suspended in Laemmli buffer. Proteins were released from beads by boiling 5 min, and subjected to SDS–PAGE analysis. Fractionated proteins were visualized by western blot using anti-PCAF or anti-ATF4 antibodies or by autoradiography using a PhosphorImager and IMAGEQUANT software (Molecular Dynamics). Cells were plated in 12-well dishes and transfected by the calcium phosphate coprecipitation method as described previously (). One microgram of luciferase plasmid was transfected into the cells along with 0.05 µg of pCMV-βGal, a plasmid carrying the bacterial β-galactosidase gene fused to the human cytomegalovirus immediate-early enhancer/promoter region, as an internal control. Cells were then exposed to the precipitate for 16 h, washed twice in phosphate-buffered saline (PBS), and then incubated with DMEM F12 containing 10% fetal bovine serum. Two days after transfection, cells were harvested in 100 µl of lysis buffer (Promega) and centrifuged at 13 000 for 2 min. Twenty micro liters of the supernatant were assayed for luciferase activity (YELEN, Ensue La Redonne, France). For all the transfection experiments presented, a plasmid pCMV-βGal was used as an internal control. β-Galactosidase activity was measured as described previously (). Relative luciferase activity was given as the ratio of relative luciferase unit/relative β-Gal unit. All values are the means calculated from the results of at least three independent experiments performed in triplicate. Total RNA was prepared using a RNeasy mini kit (Qiagen) and treated with DNase I, Amp Grade (InVitrogen) prior to cDNA synthesis. RNA integrity was electrophoretically verified by ethidium bromide staining. RNA (0.5 μg) was reverse transcribed with 100 U of Superscript II plus RNase H Reverse Transcriptase (InVitrogen) using 100 μM random hexamer primers (Amersham Biosciences), according to the manufacturer's instructions. To measure the relative amount of human , and mRNA, primers were the following: (forward primer, 5′-cagaaccagcagaggtcaca-3′ reverse primer, 5′-agctgtgccactttcctttc-3′), (forward primer, 5′-aaccgacaaagacaccttcg-3′; reverse primer, 5′-acccatgaggtttgaagtgc-3′) and (Qiagen, QT00092267 #). All the primers yielded PCR products 200 bp in size. To control for RNA quality and cDNA synthesis, β-actin mRNA was also amplified with forward (5′-ctcgcaggtcaagagcaag-3′) and reverse primers (5′-gacagctgctccaccttctt-3′). To measure the transcriptional activity from the gene, oligonucleotides derived from intron 1 and exon 1 were used to measure the short-lived unspliced transcript (hnRNA, heterogenous nuclear RNA). This procedure for measuring transcriptional activity is based on that described by Lipson and Baserga (). The primers for amplification were: forward primer, 5′-aaggcactgagcgtatcatgt-3′; reverse primer, 5′-ctctcggacggtccctaact-3′. Quantification involved the use of standard curves that had been prepared with plasmids containing specific sequences of each gene. We cloned all the PCR products into the pGEM-T easy vector (Promega) according to the manufacturer's instructions. For the construction of standard curves, pGEM-T easy plasmids were prepared as 10-fold serial dilution in water, from 4 ng to 0.4 pg. PCR was carried out using a LightCycler™ System (Roche) as described previously (). LightCycler quantification software (version 3.5) was used to compare amplification in experimental samples during the log–linear phase to the standard curve from the dilution series of control plasmids. Relative results were displayed in nanograms of target gene per 100 ng of . Each experiment was repeated three times to confirm the reproducibility of the results. ChIP analysis was performed according to the protocol of Upstate Biotechnology, Inc. (Charlottesville, VA, USA) with minor modifications. Cells were seeded at 1 × 10/100 mm dish with DMEM F12 and grown for 24 h. Cells were transferred to fresh DMEM F12 12 h before transfer to either complete DMEM F12 or DMEM F12 lacking leucine for the time period indicated in each figure. Protein–DNA was cross-linked by adding formaldehyde directly to the culture medium to a final concentration of 1% and then stopped 8 min later by the addition of glycine to a final concentration of 0.125 M. Cross-linked chromatin was sonicated using a Vibra cell sonicator (Biobloc Scientific Technology) for 10 bursts of 30 s at power 2 with 1-min cooling on ice between each burst to obtain DNA fragments of an average of 400 bp. Extracts from 1 × 10 cells were incubated with 5 µg of antibody. A rabbit anti-chicken IgG was used as the nonspecific antibody control. The antibody-bound complex was precipitated by protein A-Agarose beads (Upstate Biotechnology). The DNA fragments in the immunoprecipitated complex were released by reversing the cross-linking overnight at 65°C and purified using a phenol/chloroform extraction and ethanol precipitation. Real-time quantitative PCR was performed by using a LightCycler (Roche) and a SYBR-Green-I-containing PCR mix (Qiagen), following the recommendations of the manufacturer. The immunoprecipitated material was quantified relative to a standard curve of genomic DNA. Primers used for human sequences: amplicon A, 5′-gcagcctaaccaaagacctg-3′ and 5′-ggaggcaacttgaccaaaag-3′; amplicon B (AARE), 5′-aagaggctcacgaccgacta-3′ and 5′-atgatgcaatgtttggcaac-3′; amplicon C, 5′-agtgccacggagaaagctaa-3′ and 5′-ccatacagcagcctgagtga-3′. Primers used for mouse sequences: AARE, 5′-gggcagacaagttcaggaag-3′ and 5′-atgatgcaatgtttggcaac-3′. The reactions were incubated at 95°C for 15 min to activate the polymerase, followed by amplification at 95°C for 15 s, 55°C for 20 s and 72°C for 20 s for 45 cycles. After PCR, melting curves were acquired by stepwise increases in the temperature from 65 to 95°C to ensure that a single product was amplified in the reaction. The results are expressed as the percentage of antibody binding versus the amount of PCR product obtained using a standardized aliquot of input chromatin. Samples are the means from at least three independent immunoprecipitations. SiRNA corresponding to PCAF mRNA (5′-ucgccgugaagaaagcgcadTdT-3′ and 5′-ugcgcuuucuucacggcgadTdT-3′) (1024864 #) and to control (1027280 #) were from Qiagen. Annealing was performed as described by the manufacturer: the complementary two strands (each 5 nmol) in 250 μl of siRNA suspension buffer (Qiagen) were heated 1 min at 90°C and then incubated for 1 h at 37°C. One day before transfection with siRNA, HeLa cells were plated in 6-well plates at 25% confluency. Then 30 pmol of siRNA were introduced into the cells using the calcium phosphate precipitation as described above. Forty-eight hours after transfection, the expression level of PCAF was analyzed by western blotting. Cells were lyzed in radioimmune precipitation assay buffer (50 mM Tris–HCl, pH 7.4, 150 mM NaCl, 1% Triton X-100, 0.1% SDS, 50 mM NaF, 2 mM NaVO, 100 nM acid okadaic, 25 mM β-glycerophosphate, 1 mM phenylmethylsulfonyl fluoride, protease inhibitor cocktail from Sigma), then proteins were resolved by SDS–PAGE and transferred onto a Hybond-P PVDF membrane (Amersham Biosciences). Membranes were blocked for 1 h at room temperature with a solution of 5% nonfat milk powder in TN (50 mM Tris–HCl, pH 8.0, 150 mM NaCl, 0.1% Tween-20). The blots were then incubated with primary antibody in blocking solution overnight at 4°C. Antibodies were diluted according to the manufacturer's instructions. The blots were washed three times in TN and incubated with horseradish peroxidase-conjugated goat anti-rabbit IgG (1:5000) (Santa Cruz, CA, USA) in blocking buffer for 1 h at room temperature. After three washes, the blots were developed using the enhanced chemiluminescence (ECL) detection system (Amersham Biosciences). To identify ATF4 interacting proteins in mammalian cells, we performed TAP coupled with mass spectrometry (). We developed a mammalian expression vector coding for a fusion protein consisting of amino acids 1–351 of ATF4 linked to the TAP tag (Supplementary Figure 1A). In this construct, the TAP tag consists of the protein A (Prot.A) and the calmodulin-binding peptide affinity sequences that are separated by the recognition sequence for tobacco etch virus (TEV) protease, permitting proteolytic elution of the fusion protein from the IgG affinity resin (,). The constructs expressing the ATF4 fusion protein (TAP-ATF4) or the tag alone (TAP) were stably transfected into ATF4 −/− mouse MEF. We chose ATF4-deficient cells because absence of endogenous ATF4 expression was expected to increase purification efficiency. Supplementary shows that in ATF4 −/− MEF expressing TAP proteins, the endogenous form of ATF4 was not detected in the absence of leucine (compare lanes 3–5 with lane 2). To check the functionality of TAP-ATF4 protein, a LUC reporter driven by two copies of the AARE was transiently transfected into ATF4 −/− MEF. Supplementary shows that this fusion protein activated the AARE-dependent transcription and produced about the same transcriptional response as obtained with the wild-type form of ATF4. Cells expressing TAP proteins were incubated for 2 h in control medium or in medium lacking leucine, and nuclear extracts were prepared and applied to dual affinity chromatography according to the TAP protocol (). Bands representing putative ATF4-binding proteins were purified by SDS–PAGE and analyzed by MALDI-TOF (data not shown). Comparison of the obtained peptide sequences with protein databases identified several proteins that had been previously linked to RNA transcription. Here we report the identification of p300/CBP-associating factor (PCAF) as a novel interaction partner of ATF4 in leucine-starved cells. To confirm that PCAF can be a partner of ATF4, pull-down assays were performed. Bacterially expressed GST-ATF4 (amino acids 1–351) fusion protein was immobilized on glutathione beads and later incubated with nuclear extracts from HeLa cells starved for 2 h with leucine. Immunoblot analysis revealed that PCAF present in nuclear extracts from leucine-starved HeLa cells was retained specifically with the full-length ATF4 (A). No interaction of PCAF with GST alone could be detected. These results show that the interaction observed by the TAP method can also be reconstituted . To find which domain of ATF4 is required for the interaction with PCAF, truncated ATF4 derivatives were used in GST pull-down assays (B). As shown in C, ATF4 deleted of residues 282–351 including the bZIP domain (lane 2) retained PCAF-binding capacity. By contrast, ATF4 deleted of residues 1–85 (lane 3) or other larger deletion mutants (lanes 4 and 5) did not interact with PCAF like GST alone (lane 6) used as a control, suggesting that the N-terminal region of ATF4 is required for the interaction with PCAF. From these results, we cannot exclude the possibility that the ATF4–PCAF interaction is not direct and requires specific accessory factors present in eukaryotic cell nuclear extracts. To demonstrate the direct interaction between PCAF and the N-terminus of ATF4, we monitored the binding of S-PCAF produced to full-length, N-terminal (amino acids 1–100) or ATF4 deleted of residues 1–85 fused to GST (D). Only the full-length and the 1–100 N terminal derivatives (lanes 2 and 3) showed consistent interaction with PCAF. These data demonstrate that ATF4–PCAF interaction occurs through a direct interaction involving the N-terminal region of ATF4. To further identify the role of PCAF in amino acid-regulation of transcription in human cells, we first examined the effect of leucine starvation on the PCAF mRNA and protein levels in HeLa cells. Kinetic analysis of mRNA level indicated that PCAF mRNA was not affected by amino acid starvation while ATF4 and CHOP mRNA were increased (A). Protein analysis showed that the expression of PCAF was not significantly affected by 1–2 h of leucine starvation. However, PCAF level was greatly reduced following 4–8 h of amino acid starvation while ATF4 level was markedly increased (B). Using a ChIP analysis, we had previously demonstrated that following amino acid starvation, ATF4 binds to the AARE sequence within the promoter (). To determine whether PCAF also targets the AARE, HeLa cells were incubated in control or leucine-free medium for 2 h and ChIP assays were performed with primer sets covering either the 5′ region (amplicon A), the AARE (amplicon B) or the first intron (amplicon C) of the gene (A). The results show recruitment of both PCAF and ATF4 to the AARE following 2 h of leucine deprivation (B). In addition, bindings of PCAF and ATF4 were not detected in the 5′ region or in the first intron of , confirming that both factors are specifically engaged on the AARE. We then investigated the kinetics of PCAF engagement on the AARE in response to amino acid starvation. PCAF recruitment increased slightly after 1 h of leucine deprivation, peaked at 2 h and fell within 2–8 h of amino acid deprivation (C). Comparison of these kinetics with those obtained for ATF4 binding reveals a similarity in the time courses of recruitment of these two factors. Also, by plotting the pre-mRNA content on the same graph, it is apparent that the engagement of PCAF and ATF4 closely paralleled the increase in transcription in the first 2 h. In previous studies, ATF4 was shown to be essential for induction in response to leucine starvation (). The results described above suggest that ATF4 may be involved in PCAF recruitment to AARE following amino acid starvation. To investigate the link between binding of ATF4 to the AARE and the recruitment of PCAF, ChIP experiments were performed in MEFs deficient in ATF4 and in the corresponding wild-type cells. The ChIP results obtained with wild-type MEFs are consistent with those described above with HeLa cells (A). By contrast, in cells lacking ATF4, the increase in PCAF binding to the AARE was lost. Protein analysis shows that the lack of ATF4 did not affect the level of PCAF expression (B). Taken together, these results demonstrate that ATF4 is essential for the recruitment of PCAF on the AARE following amino acid starvation. Several studies have shown that PCAF is a transcription coactivator with intrinsic acetylase activity (). Having established that ATF4 recruits PCAF on the AARE, we sought to determine whether PCAF functioned as a coactivator of ATF4 in AARE-dependent transcription. Cotransfection experiments were carried out in HeLa cells using a LUC reporter driven by two copies of the AARE and the expression plasmids for ATF4 and PCAF or their respective empty vectors. This assay revealed that PCAF stimulated ATF4-driven transcription but had no effect by itself on luciferase expression (). By contrast, an HAT-defective PCAF containing a deletion of amino acids 497–526 () failed to stimulate ATF4-driven transcription. These results demonstrate that PCAF functions as a coactivator of ATF4 and show that PCAF HAT activity is required for ATF4/PCAF synergistic activation of the AARE-dependent transcription. To assess the role of PCAF in the amino acid regulation of expression, we first measured the effect of leucine starvation on both mRNA content and AARE-dependent transcription in PCAF-deficient cells. We employed small interfering double-stranded RNA (siRNA) transfection to specifically inhibit the endogenous expression of PCAF. HeLa cells were transfected with either siPCAF or control siRNA, and then incubated with either control or leucine-free medium for 2 h. A shows that PCAF-siRNA transfection dramatically decreased the PCAF protein content but did not affect the increase in ATF4 expression. Lack of PCAF affected the response of transcription to leucine depletion: the induction of mRNA and the AARE-dependent transcription were significantly reduced. In control siRNA-transfected cells, the response to leucine starvation was not affected. We then examined the effect of over-expressing PCAF on the increase in mRNA in response to leucine starvation (B). HeLa cells were transiently transfected with PCAF or the empty vector and then incubated either with control or leucine-free medium for 2 h. Over-expression of PCAF protein leads to a significant increase in the amino acid inducibility of . Taken together, these results demonstrate that PCAF is required to obtain maximal induction of by leucine starvation. Mammalian cells have evolved complex cellular responses to stress conditions. Both transcription and translation of ATF4 are selectively increased in response to amino acid deprivation (), even when global protein synthesis is repressed, resulting in the induction of a wide variety of ATF4 target genes (). The data reported in the present study yield several novel findings regarding the mechanisms by which ATF4 activates gene transcription upon amino acid starvation: (i) we have found evidence that the N-terminal region of ATF4 interacts directly with PCAF in amino acid-starved cells, (ii) we demonstrate that PCAF is involved in enhancing the transcriptional response of by amino acid starvation, (iii) we establish that PCAF is recruited on the AARE in response to amino acid starvation and that ATF4 is essential for its recruitment and (iv) we show that PCAF enhances ATF4-driven transcription via its HAT domain. PCAF has been described as a coactivator that mediates the transcription of many genes (). Like a number of transcriptional coactivators, this factor possesses an intrinsic histone acetylase activity (,). The role of PCAF in transcription has been investigated in multiple studies, and its requirement as a HAT and coactivator has been described for nuclear receptor- (,) and growth factor-mediated () activations and for myogenesis () among other processes. Here we report evidence that PCAF functions as a coactivator of ATF4 in the transcriptional response of following amino acid starvation. Like several nuclear proteins such as CBP and p300, PCAF interacts directly with the ATF4 N-terminal domain, shown to be a transcriptional activation domain (). We also show that the HAT activity of PCAF is required for enhancing the activation of the AARE-dependent transcription by ATF4. PCAF preferentially acetylates lysine 14 of histone H3 but also less efficiently acetylates lysine 8 of histone H4 (). At present, the exact role of the HAT activity of PCAF in promoting the AARE-dependent transcription by ATF4 remains to be established. However, there are several lines of evidence suggesting that PCAF is not involved in histone acetylation at the promoter. First, we recently reported that the acetylation status of histone H3 remained unchanged at the AARE region within 1 h of removal of leucine from the medium while the acetylation of H4 is increased (). Second, in ATF4-deficient cells, where the recruitment of PCAF was completely lost, we had also previously shown that the level of histone H4 acetylation remained elevated (). Last, we now show that the HAT activity of PCAF is required to stimulate the transcriptional activity of ATF4 on a non-integrated AARE-reporter bacterial plasmid. In addition to histones, a number of transcription factors are also substrates for acetylation by nuclear HAT (). The consequences of acetylation on protein function range from one protein to another depending on where in the protein the acetylation takes place. Acetylation has been reported to modulate protein–protein interactions, inhibit nuclear export () and alter protein stability (). CBP and p300 were described as acetylating ATF4 in its bZIP domain and enhancing its transcriptional activity (,). In addition, Gachon () have found evidence that acetylation of ATF4 is mediated by p300 but not by PCAF. In our model, the target of the PCAF HAT domain in the transcriptional activation of upon amino acid starvation remains to be identified. The present experiments demonstrate that ATF4 binding is essential for the transitory recruitment of PCAF on the AARE following amino acid starvation. PCAF recruitment on the AARE falls within 2–8 h of leucine deprivation, while transcription is still increased. Therefore, ATF4-mediated PCAF recruitment is essential in enhancing the transcription of in response to a short period of amino acid starvation. It is possible that another cofactor may be involved in the ATF4-dependent response of during long-term amino acid deprivation. The drop in PCAF recruitment might be explained by the decrease in PCAF expression level observed following 4–8 h of amino acid starvation, while ATF4 is still bound to the AARE. The ubiquitin-proteasome degradation pathway plays an important role in transcription regulation to assure the controlled and timely termination of signaling by irreversible destruction of the activated transcription regulators. PCAF has been shown to be a target for the E3 ubiquitin ligase MDM2 (). However, the enhancement of the PCAF degradation by leucine starvation remains to be demonstrated. Our present findings provide evidence that PCAF is required to obtain maximal induction of transcription in response to leucine starvation. We have recently reported that following amino acid starvation, phosphorylation of ATF2 at the AARE occurs prior to ATF4 binding, histone acetylation, and increase in CHOP mRNA (). We have further shown that ATF2 is involved in promoting the modification of the chromatin structure to enhance transcription (). ATF2 was not identified in our ATF4-TAP screen. We have recently examined the formation of the ATF2/ATF4 heterodimer by translated proteins in gel shift assays. ATF2 and ATF4 do not form a heterodimer that binds the AARE sequence (data not shown). Therefore, it is unlikely that ATF2 and ATF4 interact on the promoter. Although it is well established that ATF2 can interact directly with p300 and CBP coactivators (,), the interaction of PCAF with ATF2 on the AARE remains to be shown. ATF4 may also participate in the trigger mechanism of transcriptional activation, promoting further recruitment of activator proteins such as PCAF to the promoter. Among the potential ATF4-binding proteins we have identified, PCAF is the only one with HAT activity. Using a ChIP approach, we have recently observed that p300 and CBP are present constitutively in the AARE-binding complex (data not shown). Whether these coactivators interact directly with ATF4 is unanswered. Further experiments will be required to study their role in the amino acid regulation of transcription. Taken together, the results demonstrate that following amino acid starvation there is a highly coordinated time-dependent program of interaction between a precise set of ATF subfamily members and coactivators leading to transcriptional activation of . ATF4 has been shown to be a master regulator of a number of amino acid-regulated gene transcription such as and (). Although and AARE sequences exhibit some structural and functional similarities, there are significant differences in the molecular mechanisms involved in the induction of following amino acid starvation and those described for . Using a ChIP approach, Chen () did not observe any significant recruitment of PCAF to the promoter in response to amino acid limitation. We show here that PCAF is recruited specifically to the AARE to enhance the ATF4 transcriptional activity. By contrast, in the context of the AARE sequences, we also observe that ATF4 did not require PCAF to activate transcription in response to amino acid starvation (data not shown). As suggested by Chen (), ATF4 may act as a recruiting factor for an unknown HAT activity making the promoter more accessible to RNA Pol II and the general transcription machinery, but it is clear that PCAF is not involved. It is possible that another ATF4-interacting factor not present in the AARE-binding complex may also be essential for PCAF recruitment on the AARE. All these data suggest that although most of the amino acid-responsive genes have AARE sites that are similar in sequence, the key regulator ATF4 and other distinct transcription factors and coactivators may be involved in modulating transcriptional activation. These differences in mechanism would permit flexibility among amino acid-regulated genes in the rapidity and magnitude of the transcriptional response for the same initial signal. Further insight into how the transcriptional machinery assembles at the amino acid-responsive gene promoter and modulates transcription will improve our understanding of the molecular steps required for nutritional control by the amino acid response pathway. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Mammalian genome-wide analyses are revealing an increasingly complex transcriptome (). While predictions concerning the number of human protein-coding genes declined from >100 000 to <30 000 since 2001, transcript number estimations followed an opposite trend (). Attempts to assemble hundreds of ESTs into clusters expected to map on the same locus, as in UniGene (), did not eliminate the discrepancy between the small number of protein-coding genes and the large number of detected transcripts. Massively parallel hybridization on already known sequence probes, as in classical microarray technologies, cannot explore the whole transcriptome complexity. For this purpose, new generations of high density arrays have been developed using probes which span a genome region at regular intervals, either overlapping or spaced at defined distances (,). Besides these new open strategies, methods based on sequence signatures (tags) such as serial analysis of gene expression (SAGE) also meet the requirements to provide fresh information on unknown transcripts. SAGE tags are extracted from the 3′ most 4-nt ‘anchoring site’ of cDNAs. The restriction enzyme that cuts cDNA at this topologically defined sites is usually NlaIII (CATG sites), but Sau3A1 (GTAC sites) may be used as well (). Starting from this site, stretches of 14 or 21 nt (respectively in conventional SAGE and in LongSAGE) are extracted using Bsmf1 or Mme1 as ‘tagging’ enzymes (,). Tags matching known mRNAs are readily identified and the individual frequency of each tag measures the expression level of its cognate mRNA. As the quality of analysis depends on the number of sequenced tags, SAGE was limited up to now by the cost and capacity of the Sanger technique. However, with the advent of new DNA sequencers, the flow rate of tag-based methods may grow by an order of magnitude with a substantial reduction of time and cost of analysis () and now it becomes realistic to analyze in parallel larger collections of tags. In addition to the tags of well-annotated mRNAs, SAGE experiments currently reveal tags unmatched to known transcripts. Their high number cannot be explained simply by sequencing errors or genetic diversity, and many of them are susceptible to reveal new transcripts. The problem is to map these unmatched tags directly on large genomes. For this purpose, we investigated a new strategy, which consists in building two SAGE libraries from the same biological sample, with tags respectively anchored on the two adjacent CATG and GATC sites located at the 3′-end of each cDNA. We developed a new algorithm for assembling these tandem tag pairs on the genome sequence, defining tag-delimited genomic sequences (TDGS). In a small-scale experiment, we checked the rate of success of this strategy on a sample of well-annotated mRNAs, and starting from previously unmatched tags, we evaluated its ability to reveal new transcripts. In a large-scale analysis, we assembled a collection of TDGS based on the whole set of publicly available human SAGE tags. We found that a part of them mapped on transcription sites also indicated by tiling arrays and in addition we detected novel transcribed loci. In conjunction with other high-throughput approaches, this tandem SAGE tags strategy may help to complete the annotation of genomics regions transcribed into polyadenylated [poly(A)] RNAs. SAGE data were collected from publicly available repositories [: Platforms: GPL4, GPL6 and GPL1485, , CAGP project (Sage genie): ]. The list of SAGE libraries is available (Supplementary Table 1). chromosome sequences (HG17, NCBI build 35) were retrieved from the UCSC Genome Bioinformatics site (). UniGene cluster-representative sequences were taken from the Hs.seq.uniq. file, retrieved by FTP from the National Center for Biotechnology Information site (). We used the UniGene built # 162 assembling 4.47 million sequences into 123 995 clusters and providing the same number of cluster-representative sequences. Since SAGE may detect several authentic transcripts from the same locus, we did not use more recent UniGene releases in which transcripts co-locating with known genes have been merged. Alu sequences were taken from RepBase Update () (). Venous blood from healthy donors was obtained from the Etablissement Français du Sang (Montpellier, France). Monocytes, isolated by adherence to culture flasks, were differentiated into >99% Monocyte Derived Macrophages (MDMs) as previously described (). Total RNA (50 µg) from 8.10 MDMs was extracted with Trizol™ (Invitrogen, Cergy Pontoise, France). Poly(A) mRNA was selected by hybridization to oligo (dT) 25-coated magnetic beads according to manufacturer's instructions (Dynal, Compiegne, France). CATG-tags were prepared using the I-SAGE kit (Invitrogen, Cergy Pontoise, France) and GATC-tags using a modified Sau3A1 SAGE procedure (). The sequences of 22 387 CATG-tags and 8221 GATC-tags determined by the Centre National de Séquençage (Evry, France) were analyzed for tag detection and counting using the C+tag software (Skuld-Tech, France). The virtual SAGE analysis of UniGene cluster-representative sequences was performed using the Preditag software (Skuld-tech, Montpellier, France, ) as described (). For each sequence, the tag expected to be observed in a SAGE analysis, i.e., the one originating from the first anchoring site starting from the 3′-end of the sequence, was registered as Rank 1 tag (R1). We also registered tags from upstream anchoring sites (R2, R3, R4) susceptible to reveal technical artifacts or alternative transcripts, and tags read on the opposite strand (AS1 to AS4), which may reveal antisense transcripts (). We performed this procedure for both CATG and GATC anchoring sites. From the previous set, we selected high quality R1 tags according to the following criteria: RefSeq annotation or mention of a full-length mRNA, known chromosomal location, absence of Alu sequence in the tagged site. Hereafter, a tag will be referred to as a C-tag if anchored on a CATG site (using NlaIII as anchoring enzyme) or as a G-tag if anchored on a GATC site (using Sau3A1). The algorithm used to assemble tag pairs on the genome is depicted in A. It takes as input two sets of experimental tags, one of C-tags and one of G-tags, and retrieves all combinations of successive 5′G-3′C and 5′C-3′G tag pairs on the genome. The algorithm follows three rules. First, each transcript must possess both restriction sites. Among RefSeq mRNAs, we found 4.6% lacking one of them. Second, both sites may be found in any order, implying that two sets of oriented pairs, 5′G-3′C and 5′C-3′G, must be generated. Third, each tag is anchored on the most 3′ restriction site. Therefore, if a G-tag is located in 5′ relatively to the most 3′ C-tag of the transcript, there is no intervening G-tag between them. This assertion holds for the processed transcript but not for genomic DNA, since tags may be located on distinct exons. Because 4-bp restriction sites are frequent, scanning introns will necessarily detect false-positive tags. To alleviate this problem, the genome is scanned using actually observed experimental SAGE tags, so that irrelevant sequences may be skipped over. Intronic 14-bp stretches will be registered only if they are fortuitously identical to real tags (C). For assembling G–C tag pairs, the chromosome sequence is read from the 5′ to the 3′-end. Each occurrence of CATG is searched with a variation () of the Boyer–Moore–Horspool algorithm (). Then it is checked whether CATG with the next 10 symbols matches a tag of the experimental list. This is performed with a hash table holding the variable parts of tags (the 10 nt suffix). Once a C-tag is located, the sequence is scanned again from 5′ toward 3′ and in the same way to find the 3′ most experimental G-tag preceding the C-tag. The chromosomal co-ordinates of this G–C tag pair is then recorded together with the sequence comprised between the two tags, which we call a TDGS. The search for the next pair resumes on the nucleotide position following the C-tag anchoring site. G–C pairs are assembled on both DNA strands and the search is iterated for C–G pairs in a similar way. With the larger sets of G-tags and C-tags (106 748 and 619 771 tags, respectively), the search on the full set of human chromosomes required <2 h on a Pentium processor at 1.5 GHz running Linux with 256 megabytes of main memory. The program is available at the following URL: . We retrieved tiling arrays data from the UCSC Genome Bioinformatics site (). We used transcriptional active regions (TARs) data from Affymetrix Transcriptome Project Phase2, Affymetrix Poly(A) RNA transfrags, Yale RNA TARs and Yale Maskless Array synthesizer experiments (,). We computed the number of TDGS that either strictly overlap a TAR, or are in a 500-bp vicinity of a TAR. We assembled two sets of 270 and 15 libraries built with NlaIII and Sau3A1, respectively as anchoring enzymes, associating local data and a large number of publicly available SAGE human libraries (Supplementary Table 1). This collection, assembling 13.7 million C-tags and 0.5 million G-tags, will be called UniSAGE hereafter. The total number of distinct C-tags registered in UniSAGE is 619 770, i.e. 59% of the number of all possible 10-nt combinations, largely exceeding the number of UniGene clusters and the small number of well-annotated mRNAs. A salient feature is that individual tag counts evenly decrease, from a large number of tags observed only once to a small number of highly abundant tags (). Similar distributions are observed in individual libraries and in the dataset obtained by summing all UniSAGE tag counts. As illustrated in , most unidentified tags are observed at low levels and tags matching RefSeq sequences are the most abundant ones. Among the set of tags expressed at low levels, distinguishing between biological and artifactual ones prompts for novel approaches. Several sources of inflation make rare biological tags indistinguishable from artifacts, including infidelities in reverse transcription and PCR, or inaccuracies in single-run sequencing. To estimate a global error rate, we simulated 1-nt errors on R1 tags of widely expressed genes. We sampled four proteins (EEF1A1 and three ribosomal proteins) for which C-tag variants were uniformly distributed at low level among libraries, thus providing no evidence of them being shared with other abundant transcripts. In each case, all 30 tags differing by 1 nt from the canonical tag were observed more than once in UniSAGE. As a whole, these variants accounted for 6–7% of R1 tag counts (mean 6.6%). Individual variant frequencies varied depending on the substitution position. However, as frequency distributions were similar in the four samples, we could derive an empirical curve fitting well with their mean distribution ( = 0.99). Applying this model to the whole UniSAGE dataset, we estimated the total number of 1-nt variants at nearly 160 000 C-tags. The lack of efficiency of the anchoring enzyme may also generate additional tags generated from upstream sites. R2 tags for widely expressed genes were usually found at low levels. Wide differences from gene to gene made problematic an estimation of their mean frequency. However, tentatively assuming that all transcripts counted at least 100 times in the sum of all libraries generate R2 tags, we may estimate that some 15 000 R2 C-tags are probably registered in UniSAGE. Genetic diversity also increases tag numbers. Single nucleotide polymorphisms (SNPs) either modify tag sequence or create new tags if the anchoring site itself harbors SNPs. This problem has already been investigated elsewhere, the analysis of 54 645 mRNAs from UniGene built #163 revealed 8.6% of SNP-associated alternative tags (). Altogether, these multiple causes of inflation explain far less than a half of the huge number of different tags registered in UniSAGE. It thus seems unavoidable to conclude that a large number of tags originate from authentic transcripts. table-wrap #text ext-link xref #text xref sup fig #text The algorithm was used to assemble the whole set of UniSAGE-registered C- and G-tags. We first tested its efficiency on a sample of 11 998 gene sequences, selected from the UniGene collection for providing high quality R1 tag pairs (A). The algorithm failed to found a chromosomal tag pair for 3110 of them (26%) and succeeded in identifying 8888 gene sequences (74%). Among them, 8189 (68%) were assigned to a unique genome site. In comparison, when searching for the subset of R1 LongSAGE tags (70 284 tags, 8% of all LongSAGE tags), 71% identify a unique genomic location. With the whole dataset of the LongSAGE tags (98 142 tags observed more than once), this percentage drops to 53% (A). Experimental tags considered as unmatched were collected according to the same criteria as described above. In UniSAGE, 321 498 C-tags and 49 103 G-tags fall in this category. Working on the subset of tags found at least three times in the sum of all libraries, the algorithm assembled 93 859 potential tag pairs on the genome. We evaluated the TDGS length of unmatched tags pairs and compared them to well-annotated TDGS obtained from high quality R1 tags pairs (). Well-annotated TDGS extend up to 6000 nt. Unmatched TDGS span from 4 to 20 000 nt; however, more than 84% do not exceed 6000 bases. Each tag being present several times in the genome can be involved in several pairs. A direct examination being unpractical in this case, we crosschecked the set of TDGS with tiling arrays data. We computed the proximity between each TDGS and transcriptionally-active regions (TARs) from tiling arrays and found 43 813 TDGS (47%) overlap a TAR, and 65 808 (70%) are located <500 bp away from a TAR. In the present work, we developed a new algorithm associating pairs of gene expression signatures to localize their position on the genome. This work was motivated by studies on a large SAGE dataset (UniSAGE) showing a discrepancy between the number of loci for well-annotated genes and the large number of potential transcripts suggested by the number of tags. Using the whole set of presently available NlaII and Sau3A1 individual SAGE tags (619 000 and 106 748, respectively), this algorithm, efficient enough to process on a standard desktop computer, predicted 93 859 potentially transcribed sites in the human genome. This observation corroborates independent evaluations based on the prediction of functional transcription units and on the experimental results of tiling arrays (). Genetic polymorphism and technical errors do not account for all unmatched SAGE tags. Apart from a non-specific natural transcription noise, they may reveal genuine transcripts justifying more thorough investigations. As an open method, SAGE may indeed reveal transcripts never observed before because they are expressed in rare physiological conditions or in unique cell types, such as the terminally differentiated macrophage studied in the present work. Allowing retrospective comparisons of large datasets, it enables to distinguish a uniformly distributed transcription noise from the controlled expression of tissue-specific transcripts. Its main limitation is to detect only Poly(A) transcripts. Other wet-lab methods exist for other kinds of RNA molecules (,). The algorithm described here may used for assembling any pair of tags irrespective of their length (14–21 bp) or position (5′ or 3′) on the transcript. For instance, it could be used in the context of the new method developed to form paired-end ditags (PETs) in which the 5′ and 3′ tags defining both ends of cDNAs are physically linked and sequenced together (); this method is valuable for accurate transcript demarcation but it requires full-length cDNAs, which may be difficult to synthesize. Moreover, our algorithm could use data simultaneously collected from different cDNAs tags technology [SAGE and derivates technologies as Massively Parallel Signature Sequencing (MPSS) ()] However, the approach investigated here, based on the current SAGE technology, is technically less demanding, particularly with the new DNA sequencers available today (see the Introduction section). While tags matching well-annotated mRNAs are easily recognized, most tags are in any case difficult to locate directly on the genome. Individual 14-bp tags cannot be mapped unambiguously since large genomes are statistically expected to contain multiple copies of any 14-bp stretch. Theoretically, a 21-bp tag should occur only once per genome if nucleotides were randomly distributed. However, the benefit of LongSAGE is limited because extending sequence length increases the rate of technical errors, the number of mismatches due to genetic polymorphism and the risk of tags being split in two exons. It must be stressed that among all 21-bp tags sequenced up to now, 89% have been observed only once and that only 8% of all tags correspond to well-known genes. Even in the case of well-annotated mRNAs, 15% of 21-bp tags cannot be mapped directly on the genome. As a whole, we found 66% of LongSAGE tags unassigned to the published human genome sequence, in agreement with other groups who found 70% on a preliminary draft version and 64.5% on a more recent release, respectively (,). The presence of repeated elements inserted in the genome limits the possibility to locate tags on a unique position. In the human genome, 71% of 21-bp tags are observed only once instead of 99.83% if nucleotides were distributed at random (). In the present work, we tested the algorithm with pairs of 14-bp classical SAGE tags, thus requiring perfect matches on 28 positions. For simplification, we put aside individual tags matching Alu sequences inserted in the 3′ non-coding region of multiple human mRNAs. Nevertheless, we still observed tags involved in multiple pairs. These repeats inflate the number of TDGS but cannot be rejected as erroneous since insertion elements and pseudogenes may actually be transcribed. Finally, we found 68% of TDGS matching a unique site, close to the 71% observed for individual 21-bp tags. Whatever the method, either based on physical hybridization (as in tiling arrays) or searches (in the present case), it is obvious that insertion elements complicate the interpretation of transcriptome data (). Using a test sample of experimental tags obtained from twin macrophage libraries, we detected 187 potentially new transcripts. Among them, 39 appeared as alternative transcripts of known genes, while 134 potential sites were found in intergenic regions or in antisense orientation of known genes. The hypothesis that a newly detected TDGS identifies a novel site is initially based on the presence of the two experimental tags at both ends of the sequence. At this stage, false-positive cases are unavoidable and additional data are needed to confirm the expression of the intervening sequence. The classical solution is to design primers in the region defined by a TDGS, perform RT–PCR, and sequence the resulting amplicon, as in GLGI method (). We confirmed by this the existence of five new transcription variants of known genes and 10 novel transcripts, i.e., 50% of candidates. In other cases, we did not detect amplicons or found complex patterns. Among technical causes of failure, natural catabolic products may inhibit amplification and primers may be captured by irrelevant transcripts. Nevertheless, it must be stressed that a positive rate of 50% offers the possibility to identify thousands of new biologically relevant transcription sites at the whole genome scale. Although individual validations provide definitive evidence, the interest of high-throughput strategies is lost in this time-consuming approach. Other strategies may help to select rapidly the best candidates. The size of TDGS is by itself informative and may be used to classify them, assuming that the shorter ones have a higher probability to match genuine transcripts. Another round of selection can be based on comparisons with independent datasets. The ENCODE project plans to use tiling arrays as a major tool for human genome annotation (). Here, we showed the possibility to connect efficiently both kinds of data, a task very difficult with classical microarray and conventional SAGE data. We found a 47% overlap between our TDGS collection and TARs. This result shows that the tandem SAGE strategy corroborates for a part the results of tiling arrays and enables to reveal new transcripts having escaped from other detection systems. As a whole, these results emphasize the importance to combine independent and complementary methods for thoroughly exploring the transcribed part of the genome. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

To protect the integrity of their DNA against the attacks from various endogenous and environmental sources, cells have evolved a genome surveillance network that carefully coordinates DNA repair with cell cycle progression. DNA double-strand breaks (DSBs) are considered the most toxic type of DNA damage. If left unrepaired or repaired improperly, they cause chromosomal aberrations, which may be lethal or result in oncogenic transformation. A prominent cellular response to DSBs is the focal assembly of a large number of DNA repair proteins and checkpoint proteins at the site of damage. In particular, in budding yeast, induction of HO endonuclease in G1 results in a Tel1-dependent histone H2A S129 phosphorylation, the subsequent retention of the checkpoint adaptor protein Rad9 to DSBs, and the phosphorylation of its N-terminal region by the Ddc2/Mec1 complex (). More generally, Rad9 is phosphorylated in a Mec1- and Tel1-dependent manner after cell treatment with various DNA-damaging agents [UV, γ-rays, MMS; (,)]. Subsequently, Rad9 binds to the checkpoint effector kinase Rad53, which transautophosphorylates and becomes active (,). Similarly, in fission yeast, after creation of DNA damage by either ionizing radiation (IR) or the site-specific HO endonuclease, the checkpoint adaptor Crb2 forms distinct nuclear foci at DSBs (). It is also hyperphosphorylated by the Rad3 protein kinase and is required for the activation of the effector kinase Chk1 (). In metazoa, upon exposure to DNA-damaging agents, 53BP1 undergoes a rapid relocalization to sites of DSBs and is phosphorylated by the ATM kinase in its N-terminal region (). All three proteins Rad9, Crb2 and 53BP1 possess two C-terminal BRCA1 C Terminus (BRCT) motifs. This motif is found in a number of proteins implicated in various aspects of cell cycle control, recombination and DNA repair (,). Crb2 and 53BP1 also exhibit a tandem tudor domain between their N-terminal region, rich in phosphorylation motifs, and the C-terminal BRCT domains (). Such a domain is predicted in the case of Rad9, despite the lack of sequence identity in this region between the three proteins (). Furthermore, , the tandem tudor domain of 53BP1 specifically binds with a micromolar affinity to H4K20me2, a histone H4 peptide dimethylated on K20 (,). , the interaction between 53BP1 and the correspondingly modified H4 is necessary for the accumulation of 53BP1 to DSBs (). A millimolar interaction is found when looking at the binding of Crb2 tudor region with the same H4K20me2 peptide by NMR (). Histone H4 K20 methylation by Set9 methyltransferase is required for formation of Crb2 foci in (). However, Set9 is not required to arrest division in response to DNA damage. Based on these properties, Rad9, Crb2 and 53BP1 are proposed to play similar roles in DNA damage signaling and repair. In particular, the current thinking is that Rad9 recognizes modified histones close to DSBs through its predicted tudor domains (,). Indeed, cells lacking the H3 methylase Dot1 or carrying a mutant allele of Rad9 (Y798Q) supposed to be defective in K79 dimethylated H3 binding are G1 checkpoint-defective and fail to phosphorylate Rad9 or activate Rad53. However, there are clearly some differences between metazoan 53BP1 and yeast Crb2/Rad9. For example, the BRCT motifs of Crb2 are required for both homo-oligomerization and foci formation at sites of damage (), whereas the regions sufficient for these functions lie upstream of the BRCT domains in 53BP1 (). Moreover, Rad9 and Crb2 play a major role in cell cycle checkpoint control, whereas 53BP1 has limited checkpoint functions (). It was recently proposed that 53BP1 might rather act as an adaptor in the repair of DSBs (). Here, we investigate the functional role of the predicted tudor region of Rad9. We confirm that this region is important for the resistance of to different stresses. We also analyze the molecular mechanisms linking Rad9 to chromatin. We determined the 3D structure of the proposed tudor region and assessed its binding properties . Our results show that ScRad9[754–947] indeed folds into a tandem tudor domain. However, the five-residue histone-binding cage found in 53BP1 is only partially conserved in Rad9. Moreover, the tandem tudor region of Rad9 does not directly recognize the K79 dimethylated histone H3 peptide or the K20 dimethylated histone H4 peptide reported to bind 53BP1. Our results rather support a mechanism in which Rad9 directly recognizes DNA at the site of damage. The yeast strain rad9Δ-L157 harbors a complete deletion of the gene and is described in (). To test the resistance of yeast cells to genotoxic stresses, overnight precultures grown in YPD medium (1% yeast extract, 2% Bacto peptone, 2% glucose) were diluted to an OD of 0.1 and grown for an additional 4–5 h. Tenfold dilutions were then spotted on YPD plates with or without drugs. Sets of YPD plates were irradiated by UV light at 120 J/m using a Stratalinker 1800 or X-irradiated (150 gray) using a X-irradiator 130 kV Faxitron at 2.5 Gy/min. To construct the plasmid YCp50-Rad9ΔTudor expressing the mutant Rad9 protein deleted for the tudor domain (aa 754–947), the fragments A (bp 2110–2262) and B (bp 2842–2887) of gene encoding the Rad9 sequences surrounding the tudor domain were amplified by PCR using the primers 5′-gatacaatag agatcggtga-3′ (A5′) and 5′-acttcagtat gcgtatttat gcccgaacct gtctcccctg-3′ (A3′) and the primers 5′-caggggagac aggttcgggc ataaatacgc atactgaagt-3′ (B5′) and 5′-cctgttctga tttcaccaga-3′ (B3′), respectively. The fragment AB was constructed by PCR sewing the fragments A and B, and amplified using the primers A5′ and B3′. The -marked plasmid YCp50-Rad9 () harboring the wild-type gene under the control of its own promoter was digested at a unique SnaBI site located within the tudor-encoding sequence and cotransformed into the yeast strain rad9Δ-L157 along with the fragment AB so as to promote its repair by homologous recombination with the fragment AB and consequently the deletion of the tudor-encoding sequence. Ura transformants containing the repaired YCp50-Rad9 plasmids were recovered and tested by PCR for the presence of the allele. Plasmids encoding Rad9ΔTudor were then isolated and verified by sequencing. Rad9Δ-L157 transformants containing either an empty vector, the YCp50-Rad9 or the YCp50-Rad9ΔTudor plasmids were grown to exponential phase in YPD, arrested in G1 with α-factor (0.5 μM final concentration), UV irradiated with 120 J/m and harvested at different time points after irradiation. Analysis of Rad53 phosphorylation was performed as described previously (). Preparation and purification of wild-type and mutant 53BP1 and Rad9 domains, as well as isotope enrichment with N and N/C follow previously published procedures (,). The peptides H3-K79, H3K79me2, metazoan H4-K20me3 and yeast H4-K20me2 were purchased from Peptide Speciality Laboratories GmbH (Heidelberg). The metazoan peptides H4-K20me2 and H4-K20me2-long were synthesized in the laboratory, purified by HPLC (cationic and reverse column) and checked by mass spectroscopy. The following peptides were obtained: Rad9 NMR samples were prepared in 50 mM MES buffer (pH 6) containing 50 mM NaCl in either 90%H2O/10% DO or in 100% DO. 1 mM EDTA, a protease inhibitor cocktail (SIGMA), 1 mM NaN and 1 mM 3-(trimethylsilyl)[2,2,3,3,-H]propionate (TSP) were added to the samples. All assignment experiments were performed at 30°C on a Bruker DRX-600 equipped with a triple resonance TXI cryoprobe. The H-N HSQC NOESY and H-C HSQC NOESY were recorded on a 800 MHz Varian spectrometer at the IBS in Grenoble, France. All spectra were processed with the programs Xwinnmr (Bruker) or NMRPipe () and analyzed using Sparky. NMR titrations were carried out by recording H-N HSQC experiments at 600 MHz, using N-labeled protein sample at concentrations of 0.2–0.5 mM. The peptides H3-K79me, H4-K20me3, H4-K20me2 and H4-K20me2-long were added up to 9-, 7-, 5- and 2 molar excess to the protein samples, respectively. Two different oligonucleotides were tested: a 10 bp oligonucleotide 5′-AACTCGAGTT-3′ (PROLIGO, Paris) and a 12 bp oligonucleotide 5′-CGATCAATTACT-3′ (EUROBIO, Courtaboeuf). Both oligonucleotides were annealed with their complementary strand prior to NMR experiments and added up to a 6- and 2-fold molar excess to the protein sample, respectively. Dissociation constants were estimated by fitting the titration curves with the Kaleidagraph software, using the following equation: , where is the weighted chemical shift displacement |Δ δ (H)| + 0.1 × |Δ δ(N)|, is the ligand concentration, is the initial concentration of the protein, Δ is the maximum variation of the weighted chemical shift displacements and is the estimated dissociation constant. All ITC measurements were recorded at 30°C with a MicroCal MCS instrument (MicroCal, Inc., Northampton, MA, USA). The mouse 53BP1 tudor fragment (1463–1617), H4-K20me2 and H4K20-me2-long were equilibrated in the same buffer containing 50 mM Tris/HCl at pH 7.2, and 50 mM NaCl. The protein (18–33 µM) in the 1.337 ml calorimeter cell was titrated by the peptide (generally 10–15 times more concentrated) by automatic injections of 5–10 µl each. The first injection of 2 µl was ignored in the final data analysis. Integration of the peaks corresponding to each injection and correction for the baseline were done using Origin-based software provided by the manufacturer. Curve fitting was done with a standard one-site model and gives the stoichiometry (), equilibrium binding constant () and enthalpy of the complex formation (Δ). Control experiments, consisting of injecting peptide solutions into the buffer, were performed to evaluate the heat of dilution. The 357, 211 and the 146 bp DNA fragments were generated by polymerase chain reaction with a thermostable DNA polymerase (Promega) using a PTC-100 PCR System (MJ Research, Inc.). The 357, 211 and the 146 bp DNA fragments, obtained from the BamH1 digest, the Dra I and BamH1 double digest, and the BamH1 and Dra I double digest of the plasmid pUC(357.4) () respectively, were used as templates. The 5′-GATCCTCTAGAGTCCGGCTAC-3′ oligonucleotide was used as sense primer for the 146 and the 357 bp fragments whereas the 5′-AAAGGGTCAGGGATGTTATGACG-3′ and the 5′- CCCGGGCGAGCTCGAATTCC-3′ oligonucleotides were used as antisense primers for the 146 and the 357 bp fragments, respectively. The 5′-AAATAGCTTAACTTTCATCAAGCAAG-3′ and 5′-CCCGGGCGAGCTCGAATTCC-3′ oligonucleotides were used as sense and antisense primers for the 211 bp fragment. The five DNA fragments of 35 bp or nucleotides long were obtained by annealing of oligonucleotides (see Table below) purchased from MWG or Eurobio. The 35 bp fragment was obtained by annealing oligonucleotide 1 and 2, the 35 bp fragment containing a GT mismatch was obtained by annealing oligonucleotide 1 and 3, the 35 bp fragment containing a nick was obtained by annealing oligonucleotide 1, 4 and 5, the 35 bp fragment containing a gap was obtained by annealing oligonucleotide 1 and 5, the 35 bp fragment containing one biotin was obtained by annealing oligonucleotide 2 and 6, the 35 bp fragment containing two biotins was obtained by annealing oligonucleotide 6 and 7, the single-stranded DNA of 35 nt was formed by oligonucleotide 1. Annealing was performed in 10 mM Tris–HCl pH 8, 1 mM EDTA, 200 mM NaCl, by heating at 90°C for 10 min followed by slow cooling at room temperature. Correct annealing was controlled on 8 or 10% native polyacrylamide gels. 5′ End labeling with P-ATP and T4 polynucleotide kinase was performed at 25°C for the 35 bp DNA fragments and at 37°C for the 146, 211 and 357 bp fragments according to standard protocols (). r e a c t i o n s i n w h i c h s t r e p t a v i d i n w a s p r e s e n t , t h e d o u b l e - s t r a n d e d 3 5 b p D N A w i t h o n e o r t w o b i o t i n s w a s p r e i n c u b a t e d w i t h t h e s t r e p t a v i d i n ( 0 . 5 m g / m l ) f o r 1 0 m i n p r i o r t o a d d i t i o n o f t h e p r o t e i n , i n o r d e r t o a l l o w c o n j u g a t i o n o f b i o t i n . P r o t e i n s d i l u t e d a t t h e i n d i c a t e d c o n c e n t r a t i o n s , i n b u f f e r c o n t a i n i n g 5 0 m M T r i s – H C l , p H 8 . 0 , 1 m M E D T A , 1 m M 4 - ( 2 - a m i n o e t h y l ) b e n z e n e s u l f o n y l f l u o r i d e , 1 m M D T T , 5 0 m M N a C l a n d 0 . 1 % t r i t o n X - 1 0 0 w e r e i n c u b a t e d w i t h r a d i o a c t i v e D N A f r a g m e n t s f o r 2 h a t 2 2 ° C . P r o t e i n – D N A c o m p l e x e s w e r e a n a l y z e d o n 6 % p o l y a c r y l a m i d e g e l s a t a n a c r y l a m i d e t o b i s - a c r y l a m i d e r a t i o o f 2 9 / 1 ( w / w ) , i n 0 . 5 × T E G [ 1 2 . 5 m M T r i s – H C l ( p H 8 . 4 ) , 9 5 m M g l y c i n e a n d 0 . 5 m M E D T A ] a s i n d i c a t e d i n t h e f i g u r e l e g e n d s . A f t e r o n e h o u r p r e - e l e c t r o p h o r e s i s , s a m p l e s w e r e l o a d e d o n t o t h e g e l s a n d r e s o l v e d a t 7 0 V b y a 1 – 4 h e l e c t r o p h o r e s i s d e p e n d i n g u p o n t h e s i z e o f t h e D N A . D N A r e t a r d a t i o n w a s d e t e c t e d b y a u t o r a d i o g r a p h y o f t h e d r i e d p o l y a c r y l a m i d e g e l s a t – 8 0 ° C u s i n g B i o m a x M R f i l m s ( K o d a k ) a n d a n i n t e n s i f y i n g s c r e e n . F o r a f f i n i t y m e a s u r e m e n t s , d r i e d p o l y a c r y l a m i d e g e l s w e r e e x p o s e d t o a p h o s p h o r s c r e e n , a n d m e a s u r e m e n t s o f t h e r a d i o a c t i v e s i g n a l s w e r e p e r f o r m e d w i t h a S T O R M 8 6 0 s c a n n e r ( A m e r s h a m B i o s c i e n c e ) u s i n g I m a g e Q u a n t s o f t w a r e ( G E H e a l t h c a r e ) . From the analysis of Rad9 and 53BP1 sequences, it was predicted that the fragment 754–947 of Rad9 corresponds to a tudor region similar to that of 53BP1 (). We showed that this fragment, which we will hereafter call ScRad9[754–947], is folded, contains 73% of β-sheet and 27% of random coil and has a heat transition midpoint of 52°C (). These structural characteristics are similar to those of the tandem tudor domain of mouse 53BP1. However, the Rad9 fragment is prone to aggregation. Various attempts to optimize its solubility by modifying its limits or by mutating its cysteines failed. In particular, ScRad9[754–931] showed the same aggregation propensity than ScRad9[754–947] and ScRad9[754–931,C789A,C812S,C853S] aggregated 1.5 times faster. Cys863 could not be mutated without strongly affecting the solubility of the Rad9 fragment in . Screening of buffer conditions was carried out to limit the aggregation of ScRad9[754–947] and led to the selection of MES 50 mM, NaCl 50 mM, 2 mM TCEP at pH 6, in which 50% of the fragment is aggregated after three days at 0.2 mM and 30°C. Under these conditions, the NMR N-HSQC spectra of ScRad9[754–931] and ScRad9[754–947] are superimposable, while the spectrum of ScRad9[754–931,C789A,C812S,C853S] shows frequency shifts at peaks further assigned to residues spatially close to the three mutated cysteines (Supplementary Figure S1A and B). All these fragments share a common fold characterized by a large dispersion of the H and N NMR signals. The 3D solution structure of ScRad9[754–947] was characterized by heteronuclear double and triple resonance NMR spectroscopy. Between residues 756 and 895, 81% of the backbone NMR signals were assigned; moreover, the NMR signals of 67 and 8% of the side chains were completely and partially assigned, respectively. Unassigned signals were clustered in residues K755, C789, I814, C863 and segments 776–777, 794–798, 824–830, 832–833, 896–897. In the C-terminal part of ScRad9[754–947], between residues 896 and 947, the NMR signals corresponding to regions 896–910 and 918–929 were also unassigned. Most of the unassigned fragments are in conformational exchange on a millisecond timescale, as only few peaks could correspond to these fragments in the 3D spectra. A heteronuclear 2D N→H nOe experiment was carried out in order to identify faster motions, i.e. picosecond to nanosecond timescale dynamics, in exposed loops or unstructured segments. This experiment shows that the backbone of residues 756–762 and 930–947, corresponding to the N- and C-terminal parts, is essentially unstructured (N→H nOe ≤0). Within and after the globular core, residues 806–811, 879–882 and 911–916 belong to highly flexible loops (0 < N→H nOe < 0.5). Analysis of the NMR frequencies by TALOS () gave access to 65 Φ and Ψ values, and observation of long-range nOes on H-N HSQC NOESY and H-C HSQC NOESY experiments provided 985 inter-residual H-H proximities within the globular core. Calculation of 3D structures consistent with these experimental data using CNS () enabled us to determine the global fold of region 762–896. Additional refinement took into account 42 hydrogen bond restraints corresponding to slowly exchanging amide protons. Nine hundred structures of the fragment 762–896 were calculated, and the ten structures of lowest energy were analyzed (Supplementary Table S1). These structures have a root-mean-square deviation around the average backbone structure of 1.7 Å. They consist of two β-barrels formed by residues 786–790 (β1), 797–806 (β2), 811–816 (β3), 819–824 (β4), 828–830 (β5) and residues 768–771 (β0′), 837–841 (β1′), 844–853 (β2′), 869–874 (β3′), 884–889 (β4′), 893–895 (β5′) (A). Such a tandem β-barrel structure is reminiscent of the 53BP1 and Crb2 tandem tudor structure (). Consistently, we obtained a significant structural alignment between Rad9 sequence and the sequences of the tandem tudor domains of Crb2 and 53BP1 (E). This alignment reveals that Rad9/Crb2 and Rad9/53BP1 sequences share only 18 and 15% of identical residues in the tudor region, respectively. However, the structural fit on the 60 Cα atoms of the 10 common β-strands (i.e. β1 to β5 and β1′ to β5′) yields 3.2 and 3.1 Å between Rad9/Crb2 and Rad9/53BP1, respectively (B and C). The individual β-barrels of Rad9 are particularly similar to those of the tudor folds of Crb2 and 53BP1, with Cα root-mean-square deviations within each barrel comprised between 1 and 1.5 Å. Moreover, the relative positioning of the β-barrels in Rad9 is close to that found for Crb2 and 53BP1. The major structural differences in the tudor region between Rad9 and its potential homologs Crb2 and 53BP1 reside in the poorly defined structure of loop 786–801 (β1β2) and in the slightly different positioning of one barrel relatively to the other (B and C). In fact, as region 824–829, which contains β5, is not assigned (as shown by the purple color of its backbone on A), it is not possible to demonstrate the presence of a β-sheet between β1 and β5. Similarly, because the region 789–798 is essentially unassigned (A), the 3D structure of loop β1β2 is unknown, and the presence of a β-sheet between β2 and β5′ cannot be proven. Finally, a C-terminal α-helix is found after β5′ in Crb2 and 53BP1, which interacts with β2, but the region corresponding to this helix is unassigned in Rad9. Thus, because several NMR frequency assignments in loop β1β2, in strand β5 and in the region following β5′ are lacking, it is not possible to describe the conformation of the interface between the two β-barrels in the β5/β1/β2/β5′ region. Conformational exchange on a microsecond to millisecond timescale at this interface is one possible explanation for the lack of NMR data. We tested if the relative positioning of the tudor folds observed in 53BP1 and Crb2 was consistent with the NMR data obtained on Rad9. Therefore, we calculated a set of 3D structures of the Rad9 fragment 762–896, using three additional hydrogen bond restraints linking the oxygen of Q790 (β1) to the nitrogen of D827 (β2), and the oxygen/nitrogen pair of F797 (β2) to the corresponding pair in L895 (β5′) (D). These structures are as consistent as the first set of structures with the experimental NMR data (Supplementary Table S2). The second β-barrel of these structures is structurally similar to that of the first set of structures. However, the conformation of the first barrel as well as the relative positioning of the two barrels is closer from Crb2: the structural fit on the 60 Cα atoms of the 10 common β-strands (i.e. β1 to β5 and β1′ to β5′) yields 2.7 and 2.4 Å between Rad9/Crb2 and Rad9/53BP1, respectively. Thus, the relative positioning of the tudor folds observed in 53BP1 and Crb2 is one of the possible relative positioning of Rad9 tudor folds. To analyze the functional importance of Rad9 tandem tudor region for resistance to DNA damage, cells complemented with either the wild-type gene or the allele deleted for the tudor-encoding sequence were tested for their resistance to various genotoxic stresses, including UV- and X-irradiation, camptothecin (CPT, an inhibitor of DNA topoisomerase I) and 4-nitroquinoline 1-oxide (4-NQO, a drug producing bulky base adducts of the type that is mainly repaired by the nucleotide excision repair system). As shown in A, the deletion of Rad9 tandem tudor region resulted in a hypersensitivity to DNA damage which was particularly pronounced in cases of UV-irradiation and camptothecin treatment. However, the allele retained much of the functionality of since cells were more resistant to DNA damage than mutants. To clarify further the biological role of the tandem tudor domain of Rad9, we analyzed Rad53 phosphorylation status after UV irradiation in wild-type, and cells. When asynchronously growing cells were UV irradiated, a slight delay in Rad53 phosphorylation was observed in and mutants 30 min after UV irradiation, but this defect disappeared 60 min after irradiation (data not shown). In contrast, when cells were synchronized in G1 with α-factor before UV irradiation and maintained in G1 thereafter, and cells exhibited a strong defect in Rad53 phosphorylation (B), which suggests that the presence of the tandem tudor domain in Rad9 is critical for the phosphorylation of Rad53 under these conditions. 53BP1 and Crb2 tandem tudor domains interact with a histone H4 peptide specifically dimethylated at K20 (H4K20me2) with 2.10 M and 10 M affinity, respectively (,). 53BP1 tandem tudor domain also interacts with a histone H3 peptide dimethylated on K79 (H3K79me2) with a 10 M affinity (). Interactions of 53BP1 and Crb2 with H4K20me2 seem to play a preponderant role in targeting these proteins to DNA damage foci (,,). Thus, we looked for similar interactions in the case of ScRad9[754–947] by performing NMR titrations. All the experiments were carried out in parallel on mouse 53BP1 fragment 1463–1617 (hereafter called Mm53BP1[1463–1617]) as a control, by performing NMR titrations and ITC studies. The peptides tested in this study were H3K79, H3K79me2, H4K20me2, H4K20me2-long and H4K20me3. First, we added each histone peptide to N-labeled ScRad9[754–947] and Mm53BP1[1463–1617] samples and followed the chemical shift perturbations of N and Hn nuclei by recording H-N HSQC experiments. Surprisingly, the Hn and N chemical shifts of ScRad9[754–947] were not modified by the addition of the following histone peptides: H3K79me2, yeast and metazoan H4K20me2, metazoan H4K20me3 (). Thus, our Rad9 fragment alone cannot bind to yeast H3K79me2 and H4K20me2 peptides and metazoan H4K20me2 and H4K20me3 peptides. On the opposite, when adding the same H3K79me2 or H4K20me3 peptides to Mm53BP1[1463–1617], several H and N chemical shifts moved from their native values to values corresponding to a bound state (). The exchange rate between the free and bound states of the 53BP1 fragment was fast on the NMR timescale, suggesting that the affinity between Mm53BP1[1463–1617] and the histone peptides is in the millimolar range. Fitting the variation of the weighted chemical shift displacements against the peptide concentration yielded a Kd value of 1.5 ± 0.5 mM and 1.7 ± 0.8 mM for H3-K79me2 and H4-K20me3, respectively. After addition of the H4K20me2 or H4K20me2-long peptides to the same 53BP1 fragment, the H-N peaks that shifted in the previous interactions either disappeared or shifted (). In these two latter cases, the exchange between the free and bound states of the 53BP1 fragment was intermediate to fast on the NMR timescale, suggesting that the affinity of Mm53BP1[1463–1617] for these histone peptides is in the micromolar range. NMR estimation of the affinity was confirmed by ITC measurements. A of 1.3 × 10 ± 1.6 × 10 M ( of 7.7 µM) with a stoichiometry of 1, a of –2.2 ± 0.1 kcal/mol, a of –7.1 kcal/mol and a of 4.9 kcal/mol was obtained for H4-K20me2 (Supplementary Figure S2A). A of 1.1 × 10 ± 1.6 × 10 M ( of 8.8 µM) with a stoichiometry of 1, a of –2.3 ± 0.1 kcal/mol, a of –6.9 kcal/mol and a of 4.6 kcal/mol was obtained for H4-K20me2-long (Supplementary Figure S2B). These values are similar to that previously measured by Botuyan () for the interaction between human 53BP1 fragment 1484–1603 (similar to mouse 1470–1589) and the K20 dimethylated H4 peptide fragment 12–25 (). Interestingly, the main driving term in the free energy of interaction is the positive entropic change, likely due to the dehydration of the binding interface upon complex formation. NMR titration using unmethylated H3K79 showed that this peptide does not interact with Mm53BP1[1463–1617], thus underlying the critical role of lysine methylation in the binding. Residues whose N and Hn chemical shifts are sensitive to H4K20me2 peptide binding are colored on the 3D representation of Mm53BP1[1463–1617] (Supplementary Figure S3). The peptide-binding region is arranged around an aromatic cage formed by W1495, Y1500, F1519 and Y1523, as found by Mer and co-workers (). In order to verify that these residues, as well as two close polar residues D1521 and C1525, are also involved in binding to the whole modified H3, several of them were mutated in Mm53BP1[1463–1617], and the binding to H3 from calf thymus was checked by GST-pulldown experiments. Mutations W1495A and D1521A completely abolished the binding, Y1500A and F1519A reduced the binding, and Y1523A and C1525A did not change significantly the GST-pull-down results (Supplementary Figure S4). 53BP1 and Crb2 tandem tudor domains were suggested not only to recognize modified histones, but also to bind DNA, in particular in the context of nucleosomes (,). However, if Mm53BP1[1463–1617] binds with a millimolar affinity to a short 10 bp oligonucleotide, as seen by NMR (), we could not detect any interaction with a double-stranded DNA of 146 bp by gel retardation assays (A). On the opposite, ScRad9[754–947] forms complexes with this 146 bp double-stranded DNA as demonstrated by delayed migration compared to naked DNA (B). In order to check if ScRad9[754–947] binds to a particular DNA structure, we measured its affinity for different 35 bp oligonucleotides: single, double-stranded containing a mismatch, a nick or a gap (; A and B). Clearly, ScRad9[754–947] recognizes with a similar affinity the double-stranded oligonucleotides with or without mismatch, nick or gap. Its affinity for one of the single-stranded oligonucleotide is 2.5 times lower. The shorter ScRad9[754–931] binds to the 35 bp double-stranded DNA with a twice lower affinity (C; ). The ScRad9[754–931,C789A,C812S,C853S] mutant does not recognize DNA anymore when tested by gel retardation assay (D). In parallel, surface plasmon resonance experiments confirmed that the ScRad9[754–931,C789A,C812S,C853S] mutant looses the DNA-binding capacity of wild-type ScRad9[754–931] (data not shown). Because various studies suggest that Rad9 intervenes in the case of DSBs, we tested DNA molecules of various lengths, in order to determine if ScRad9[754–947] binds preferentially at the ends of DNA. In the absence of any specificity, for the same concentration of DNA, the percentage of complexes should be inversely proportional to the length of the tested DNA fragment. Measurement of the affinity of ScRad9[754–947] for the 35 bp double-stranded DNA, as well as for three other double-stranded DNAs of different lengths, shows that the affinity does not depend on the tested DNA: calculation of the ratios between bound DNA and free DNA measured on the gels enabled us to estimate an apparent affinity constant of 0.4 (± 0.1) µM, 0.5 (± 0.1) µM, 0.3 (± 0.1) µM and 0.6 (± 0.1) µM for DNAs of respectively 357, 211, 146 and 35 bp (; B and E). Furthermore, we tested the binding of ScRad9[754–947] to a 35 bp double-stranded oligonucleotide biotinylated at one or two 3′-OH ends. Clearly, the presence of biotine does not modify the affinity of ScRad9[754–947] for the oligonucleotide (). However, after binding of streptavidin to one of the ends of the 35 bp DNA, the affinity drops by a factor of two [i.e. 1.4 (± 0.1) µM; F and ]. After binding of streptavidin to both biotinylated extremities, the gel shift pattern is largely modified (data not shown). These results are consistent with the hypothesis of Rad9 tudor domain binding to DSBs directly, but it must be stressed that obstruction by streptavidin might also have prevented the formation of the previously observed Rad9–DNA interactions. Finally, we observed by NMR the binding of a 10 bp oligonucleotide 5′-AACTCGAGTT-3′ or a 12 bp oligonucleotide 5′-CGATCAATTACT-3′ to an NMR sample of N labeled ScRad9[754–947]. Therefore, we added progressively the DNA to an NMR sample of ScRad9[754–947], and we followed the chemical shifts of labeled N nuclei and N attached H nuclei by recording H-N HSQC spectra. Fifteen HSQC peaks corresponding to the backbone nuclei (N, Hn) of A838, V839, F841, D842, V847, L872, K874, S878, L880, G882, T885, I887, K888, S892, I893 and three peaks corresponding to the side-chain nuclei (N, Hn) of N844, R854, N876 shifted after addition of DNA (A). All these changes correspond to residues located around the highly flexible loop β3′β 4′, in the second tudor fold and in the main positively charged region of ScRad9[754–947] (B and C). Fitting the variation of weighted chemical shift displacements against the oligonucleotide concentration for these peaks yielded a value comprised between 3 and 20 µM. xref #text We have shown that Rad9, as Crb2 and 53BP1, possesses a tandem tudor domain upstream from its BRCT region. This domain exhibits additional N-terminal β-strand and loops when compared to the published structure of 53BP1 and Crb2 tudor regions, and is highly dynamic both on a fast ns to ps timescale and on a slower μs to ms timescale. , it is not capable of recognizing the methylated histone peptides reported to bind 53BP1 tudor region. However, it possesses a large and positively charged loop β3′β4′ which recognizes DNA with a micromolar affinity and is conserved in most identified yeast Rad9 analogs. Thus, both the tudor regions of 53BP1 and Rad9 seem to recognize chromatin, however through different interactions: 53BP1 binds with a micromolar affinity to H4 dimethylated on K20 through a five-residue binding cage located on the first tudor fold, whereas Rad9 binds with a similar affinity to DNA through a positively charged region situated on the second tudor fold. The two binding regions are located on the same side of the tandem tudor domain; they are distinct but contiguous. In the case of Crb2, Botuyan () reported that the affinity of the tudor region for H4K20me2 is only in the millimolar range, but proposed that , the effective affinity is higher if one considers that Crb2 not only interacts with histones but also with DNA in the context of nucleosomes. Consistently, they noticed some affinity of the Crb2 tudor region for nucleic acids. Rad9 and 53BP1 tudor regions could similarly be involved in such multi contact interaction with chromatin. All these results underline that a common mechanism exists in the biological functions of Rad9, Crb2 and 53BP1: they bind chromatin through their similar tudor domains. As proposed by Kron and co-workers (), these transient interactions may allow the initial targeting of Rad9 to DSBs. Then, phosphorylation of H2A by Tel1 may induce a stabilization of the Rad9/chromatin interaction and promote Rad9 hyperphosphorylation and Rad53 transautophosphorylation by the Ddc2/Mec1 complex. However, the mode of chromatin recognition by Rad9, Crb2 and 53BP1 is different, which implies that the corresponding interactions might be differently regulated. It was proposed by several authors that Rad9 tandem tudor region interacts directly with H3 dimethylated on K79 (,). We show in this paper that our Rad9 fragment does not bind to such modified histone . Methylation of H3 K79 may play a more general role in coordinating various repair processes (). More recently, it was proposed that dimethylation of H4 K20 plays a critical role in the direct binding of yeast Rad9 homologs to chromatin (). This hypothesis is not supported by our study. However, Rad9 tandem tudor domain might recognize another region of H4, or the multimeric functional state of Rad9 might be critical for histone binding. Lastly, the targeting of yeast Rad9 homologs to DSBs might depend on indirect interactions with methylated histones. More quantitative binding studies and 3D structural information are needed in order to understand how Rad9 cooperates with other BRCT proteins in order to recognize chromatin at the vicinity of DSBs and thus prevent accumulation of DSBs. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Primases of bacteria and eukarya differ in sequence and structure suggesting that these enzymes are not evolutionary related although they perform the same function. Bacteria and most phages have a primase with the catalytic ‘toprim’ domain which is also found in topoisomerases IA and II (). In contrast the heterodimeric primases of the eukarya and archaea form a distinct group. The catalytic small subunits of these primases have similar amino acid sequences and adopt a different fold that is related to the RNA recognition motif (RRM). Currently there is no structure available for eukaryotic primases. However, the structures of three archaeal primases () have been determined and they possess the RRM fold consisting of four β-strands and two α-helices. Additionally a further β-strand (flange), which runs perpendicular to the strands of the RRM, is found in these enzymes. The catalytic residues, namely three acidic residues involved in chelating the catalytic metal ions and a catalytic histidine, are found in three neighbouring strands of the RRM and form the active site of these enzymes (). The replication protein of the archaeal plasmid pRN1 is a multifunctional enzyme with primase, DNA polymerase and helicase activities (). By deletion mutagenesis the N-terminal domain responsible for primase and DNA polymerase activity could be identified and has been named the ‘prim/pol’ domain as suggested by its two enzymatic functions. The amino acid sequence of the prim/pol domain has no detectable sequence similarity with characterized primases or DNA polymerases. Nevertheless the structure of this domain revealed a high structural similarity to the archaeal primases. In fact, the catalytic core of the enzyme with the RRM fold and the flange is very well conserved and the essential catalytic residues occupy the homologous positions. Point mutants of the active site residues further confirmed that the inferred active site residues are in fact directly involved in catalysis (). Based on the structure of the prim/pol domain, the inclusion of this domain into a superfamily of archaeo-eukaryotic primases was proposed by Iyer and coworkers (). The prim/pol domain (pfam PF09250) is found in ∼100 proteins. About one-third of these proteins are encoded by viruses, plasmids and bacteriophages. In these genomes, the prim/pol domain could be directly involved in the replication of the respective elements. The remaining instances are found in the genomes of bacteria and archaea and are likely to stem from integrated mobile elements which may no longer be functional. It is currently unknown how the plasmid pRN1, the prototypical member of a group of crenarchaeal plasmids, is replicated. The plasmid pRN1 shares three conserved open reading frames with the other members of the plasmid family pRN. The encoded proteins, two rather small DNA-binding proteins and the large replication protein with ∼900 amino acids, probably constitute the essential core for plasmid maintenance and replication. Our current model is that the helicase activity of the replication protein might unwind the origin and primers are synthesized subsequently. After that either the replication protein or the host replication machinery replicates the whole plasmid. In order to understand initiation of plasmid replication in more detail, we analysed the primase activity of the replication protein. We found that the primase is highly sequence specific and only primes DNA synthesis on templates that possess the trinucleotide GTG. The resulting primer begins with a ribonucleotide but is extended with seven deoxynucleotides. All oligodeoxynucleotides were purchased from metabion (Germany). α-P labelled deoxyadenosine 5′-triphosphate (3000 Ci/mmol), α-P adenosine 5′-triphosphate (3000 Ci/mmol) and γ-P adenosine 5′-triphosphate (5000 Ci/mmol) were obtained from Hartmann Analytic (Germany). The full-length protein and the deletion mutants were expressed from the plasmid pET28c (Novagen). The construction of the full-length protein () and the deletion mutant C255 () were described previously. The deletion mutants C370 and C526 were constructed by amplifying the part of the wild-type gene with primers, which contained restriction sites for NheI and HindIII, and cloning the fragments into pET28c. The expressed proteins possess an N-terminal hexahistidine tag. The deletion mutant N40-C370 was constructed similarly with primers with NcoI and XhoI restriction sites. This protein has a C-terminal hexahistidine tag and starts at amino acid 40. The purification was done for all proteins essentially as described before () with the following differences: the plasmids were transformed into Rosetta (pLys) (Novagen) and grown in 2 × YT at 22°C. The proteins were purified to a purity of >90%, dialysed against 25 mM sodium phosphate pH 7.0, 40% glycerol, 100 mM NaCl, 0.01% 2-mercaptoethanol and stored in aliquots at −20°C. The protein concentrations were determined by UV spectroscopy using their respective theoretical extinction coefficients. The primase assays were performed in a reaction buffer of 25 mM Tris–HCl pH 7.5, 1 mM DTT and 10 mM MgCl. In a typical primase reaction of 10 µl, 0.4 µM of protein was incubated with 4 µM of oligodeoxynucleotide and 1 mM ATP in the presence of 10 µM dNTPs supplemented with 0.6 nM [α-P] dATP for 10 min at 50°C. For the reactions that required a radioactive ribonucleotide, cold ATP was omitted and therefore the concentration of ATP was only between 6 and 10 nM. We determined the apparent Michaelis–Menten constant () and the maximal velocity () by varying the concentration of dATP, ATP and template in a primase assay with 40 nM enzyme. The reactions were incubated at 50°C for 45 min. In these assays, saturating concentrations of 100 µM dNTPs, 1 mM ATP and 6 µM template were used. As DNA templates we used oligodeoxynucleotides C and G (). To determine the kinetic parameters, varying concentrations of one of the components were added to the reaction mixes, while the other two components were kept constant at the saturating concentration. The reactions were stopped on ice, loading buffer was added and the samples were separated on 20% polyacrylamide/urea gels. After quantification using an InstantImager, the velocity was calculated and the apparent and were obtained from least-square curve-fitting. The apparent and values were derived from a minimum of three measurements. The oligodeoxynucleotides D and E, () were labelled with fluorescein at the 3′ end and were used at a concentration of 40 nM in the assays. The binding buffer consisted of 12.5 mM Tris–HCl pH 8.0, 1 mM MgCl, 100 mM KCl and 0.01% Tween. The binding isotherms were obtained by reverse titrations. To the mixture of binding buffer and DNA a small aliquot (6 µl) of protein, which had been dialysed against binding buffer was added. Then the protein concentration was gradually decreased by replacing 30% of the mix in the cuvette with a solution of DNA in binding buffer. The anisotropy was measured with excitation at 495 nm and emission was monitored at 526 nm with a cutoff filter at 515 nm. The integration time was 10 s. For each data point at least three measurements were collected. The dissociation constant was obtained by fitting the data with a single-site binding model. Our initial characterization of the primase activity of the replication protein ORF904 from plasmid pRN1 revealed that this enzyme synthesizes short (8 nt-long) primers in the presence of single-stranded M13 DNA. Primer synthesis requires the presence of ATP. In addition, dNTPs are much better incorporated than rNTPs () which is unusual for a primase. In order to understand the molecular mechanism of primer synthesis by this plasmid-encoded primase, we investigated in more detail the substrate requirements and the kinetics of primer synthesis. Initial studies used single-stranded M13 DNA to study the primase activity of ORF904. We wished to use better-defined single-stranded templates and tried all four homopolymers as substrates. Homopolymers did not yield detectable primase activity (data not shown). Next we tested various unrelated oligodeoxynucleotides. Surprisingly the products synthesized by ORF904 differed considerably. As can be seen in , a specific primer of 8 nt was synthesized only in the presence of single-stranded plasmid DNA or of substrate A (). The primer synthesis was only observed when 1 mM ATP is included in the reaction. In contrast, in the presence of two other oligonucleotides (30 and 32 nt, ) a rather long product is synthesized independently of ATP. The lack of ATP dependence and the product length suggest that this product does not stem from the primase activity of ORF904. Instead this activity appears to reflect a somewhat sequence-specific terminal deoxynucleotide transferase activity of the protein which is not considered further in this study. Some oligodeoxynucleotides, e.g. 39 nt (), did not serve as template for either activity. In control reactions without template DNA, no products are observed further underscoring that ORF904 requires a template for primer synthesis. We first defined the minimal substrate that supports primer synthesis. The run-off product of 20 bases () suggested that primer synthesis could start in the middle of substrate A close to the central GTG motif (). We therefore used oligodeoxynucleotides with deletions at both ends of substrate A. The length of the run-off products shortened, whereas the primer synthesis remained essentially unchanged (, lanes 1–3). Further shortening the 5′ end lead to primers with reduced lengths, but the primase activity was still high with these shorter templates (, lanes 3–8). Additionally, deletions at the 3′ end were prepared. When the 3′ end was further reduced relative to substrate B, the amount of primer synthesized decreased sharply (compare lane 12 with lanes 13 and 14). Our analysis therefore showed that the 12-bases-long substrate B with the sequence 5′-CTTCTTCTGTGC-3′ represents the minimal substrate which supports synthesis of the full-length primer. The experiment also demonstrated that the base 3′ to the central GTG motif is critical for primer synthesis and that the bases 5′ of the GTG determine the length of the primer or run-off product but do not appear to be important for the efficiency of primer synthesis. We next investigated which bases of substrate B make up the recognition site of the primase activity. As a starting point, we used an oligodeoxynucleotide similar to the minimal substrate (substrate B) where we exchanged the cytidines of the 5′ half with thymidines. This substrate has the advantage that the label ([α-P]dATP) is maximally incorporated during primer synthesis. On the basis of this substrate, we tested the importance of six base positions around the GTG motif bases by replacing each position with all four possible bases (B). For position 1 and position 2, all bases supported a high primase activity. In contrast, at position 3 only a guanosine was able to initiate primer synthesis. Likewise at positions 4 and 5 only a thymidine and a guanosine, respectively, were good primase substrates. The templates with the three other bases at these positions did not serve as substrates. At position 6, we could not detect a preference for any base. Replicate measurements of these primase assays allowed us to define a motif (C and Figure S1). The height at each position was calculated from the information content at each position (). This analysis shows that the trinucleotide GTG in positions 3–5 is an important recognition site for the primase activity of ORF904. Modifications of this motif were not tolerated, demonstrating highly specific sequence recognition by ORF904, and explaining why we did not observe primase activity with homopolymers and most oligodeoxynucleotides. It is possible that there are other recognition sites besides the GTG motif discovered by us. However, given the large variation in sequences tested in our laboratory we do not believe that this is very likely. The replication protein ORF904 consists at least of two domains, a prim/pol domain and a superfamily 3 helicase domain (A). The prim/pol domain is situated in the N-terminal part of the protein (amino acids 40–255) with the active site of DNA polymerization. The active site is responsible for primase and DNA polymerase activity and bears structural resemblance with the archaeal-eukaryal primases (see before). In order to determine which part of the protein could be responsible for the site-specific recognition of templates with a GTG motif, we tested various deletion mutants with substrates containing the recognition motif GTG. Full primase activity was observed with a deletion mutant from amino acid 1–370. In agreement with prior experiments, no primase activity was detected with shorter deletion mutants. Although the deletion mutant C370 was highly active in primase activity, it had no detectable primer extension activity under these experimental conditions. In contrast, the deletion mutant C526 (spanning amino acids 1–526) and the full-length protein were more active in elongating the primer (). Possibly the longer proteins bind better to the DNA substrate and primer elongation occurs more often. Significantly, all deletion mutants retained their strict ATP dependence for primer synthesis (data not shown), ruling out the possibility that ATP binding to the helicase domain is crucial for primer synthesis. We wanted to determine whether the preference for the GTG motif in primer initiation is linked to a higher affinity for substrates with this sequence. We determined the dissociation constants () by fluorescence anisotropy measurements using substrates D and E (). Substrate D was 17-nt long and contained the GTG motif and supported primer synthesis, substrate E contained the modified motif GAG with a single base exchange and therefore did not serve as template for primer synthesis. The oligonucleotide with the recognition sequence (substrate D) was bound with a of 225 ± 5 nM, whereas the affinity for DNA without intact motif was found to be 5-fold lower, with a of 1200 ± 80 nM (Figure S2). The differences in affinity therefore correspond to the observed sequence specificity of the primase reaction. Since the oligodeoxynucleotide with the mutated motif is still bound by the protein, however, other factors are presumably important in the sequence-specific initiation, as well. We have previously shown that the primer synthesized by ORF904 is alkali stable and DNase I sensitive, suggesting that it is made up in part or entirely of deoxynucleotides. We reconfirmed the former results by assembling primase reactions in the presence of dNTPs and rNTPs using either [α-P]dATP or [αP]ATP as label. As substrate we used substrate C in order to maximize incorporation of the label (). With [α-P]-dATP as label, an efficient primer synthesis was observed only in the presence of ribonucleotides and deoxynucleotides. Without ribonucleotides a primer was not formed, which is in agreement with our observation that ATP (or another ribonucleotide, see subsequently) is required for primer synthesis. In the presence of ribonucleotides but without cold deoxynucleotides, a small amount of primer was synthesized. In this reaction only labelled dATP was present without competing unlabelled dNTPs. Therefore the incorporation seen represented only a marginal amount of primer synthesized. The full-length primer synthesis was low because the low concentration of dATP effectively limited primer synthesis. Most of the radioactivity in this lane was found just above the unincorporated dATP. This species could be a dinucleotide formed between a ribonucleotide and dATP (see subsequently). When [α-P]ATP was used as label we observed incorporation, but ribonucleotides alone did not support complete primer synthesis, since a full-length primer was seen only when deoxynucleotides were present. Even in the presence of only labelled ATP, no primer synthesis was seen, thus reinforcing the conclusion that a primer or a truncation product consisting solely of ribonucleotides cannot be formed by the enzyme. These results strongly suggest that both ribo- and deoxynucleotides are required for synthesis and that both are incorporated into the primer. The amount of label incorporation (50% with [α-P]dATP versus 5% with [α-P]ATP, lanes dNTPs/rNTPs) also clearly shows that deoxynucleotides make up the majority of the primer. In principle, it is possible that the primase has only low discriminatory power and randomly incorporates ribonucleotides and deoxyribonucleotides with a clear preference for the latter. However such a stochastic behaviour does not explain the strict requirement for both bases for efficient primer synthesis. The dinucleotide could be formed of two ribonucleotides. However we do not observe a dinucleotide in the absence of deoxynucleotides. On the contrary, in the presence of [α-P]dATP and rNTPs a dinucleotide is formed (, lane 3). In addition the amount of α-ATP incorporation is only a tenth of the incorporation with α-dATP (5% versus 50%). With two ribonucleotides in the primer a ratio of 1:4 would be expected. We therefore consider the possibility unlikely that the dinucleotide is made up of only ribonucleotides. In a mixed dinucleotide the ribonucleotide could be the first or second base. With the use of [γ-P]ATP as label it is easy to differentiate between these two possibilities since the radioactivity will only be incorporated in the primer when ATP is the first base. This is indeed the case ( and see below). The experiment thus suggested that a dinucleotide consisting of a ribonucleotide at the first position and a deoxynucleotide at the second position is formed and then extended by the addition of further deoxynucleotides to the full-length primer. This interpretation also agrees quite well with the amount of incorporation (50% with [α-P]dATP versus 5% with [α-P]ATP as label), since the ratio would be expected to be 7:1 (with 7 dATP and 1 rATP in the full-length primer). We observed label incorporation when the reaction mixture contained 5 nM [γ-P]ATP, dNTPs, the template and the enzyme (, upper panel, lane 1). In a control experiment with excess unlabelled ATP (1 mM) no incorporation was seen, as would be expected. In contrast, ADP and AMP as well as AMPcPP (α, β-methyleneadenosine 5′-triphosphate) and γ-S-ATP (adenosine 5′-[γ-thio]triphosphate) did not compete for the labelled ATP and label incorporation was observed. Only the analogues AMPPcP (β,γ-methyleneadenosine 5′-triphosphate) and AMPPnP (adenosine 5′-(β,γ-imido)triphosphate) competed with the labelled ATP and prevented label incorporation. The behaviour of the non-hydrolysable ATP analogues is somewhat unexpected: The ATP is not hydrolysed during the primase reaction and the inability of the analogues to compete is therefore not related to their inability to be hydrolysed. Nevertheless the analogues exhibit different qualities regarding their effectiveness in primer initiation. AMPcPP and γ-S-ATP behave like ADP or AMP and are not accommodated by the ribonucleotide-binding site of the enzyme. It appears therefore that there are strict structural requirements for the first base. We reconfirmed that only ATP, AMPPcP and AMPPnP can serve as first base and support the synthesis of a primer by performing the primase assay in the presence of [α-P]dATP (, lower panel). Next we asked whether other ribonucleotides could also be accepted as first base of the primer. Since the first base is likely to base pair with the template, we performed these experiments with all four possible templates which differ in the position of the likely initiation site of primer synthesis. Results of the previous experiments imply an initiation at the first base 5′ to the GTG. We therefore varied the base in this position and also varied the ribonucleotide present in the reaction mixture. As can be seen in , ATP can serve as first base with all four templates. In the reactions with the ribonucleotides CTP or UTP, a large amount of primer is synthesized when the template contains the cognate base 5′ of the GTG. The other substrates yield less or no primer. In contrast to this situation, however, with GTP primer synthesis appears to be most efficient when a thymine was present at the position in question, rather than the cognate base cytosine. On the whole, these results show a clear dependency between the initiating ribonucleotide and the first template base and provide independent evidence that the primer synthesis initiates complementary to the base 5′ of the GTG recognition site. The fidelity of the primase reaction was assayed using a template that contained all four bases successively upstream of the GTG recognition sequence (substrate F, ). We then determined whether misincorporation occurs when the cognate base is absent from the reaction mix (A). For this experiment [α-P]dATP was used as a label, which will be incorporated at the second position of the expected primer. When ATP and dATP were present only a dinucleotide is formed, indicating that dATP was not misincorporated at the third position opposite the cytosine. Addition of dGTP to the reaction allowed the efficient synthesis of a trinucleotide along with a minor, barely visible, amount of a longer product of 7 nt. When the reaction mix contained additionally dCTP (and was therefore only lacking dTTP) the majority of product synthesized was a tetranucleotide and again, a minor amount of a longer product was visible. When all four dNTPs were present, a full-length primer (8 nt) and a run-off product of 10 nt were synthesized. This suggests that the sequence of the primer synthesized by ORF904 is predominantly a reverse complement of the template. We attribute the minor by-product that is shorter than the full-length primer to slippage (see subsequently). Additionally tests investigated whether dideoxynucleotides (ddNTPs) are incorporated by the primase. The same template was used to determine whether the addition of cognate ddNTPs leads to an elongation of the primer. When dATP and ddGTP were added, only a dinucleotide was observed, showing that ddGTP cannot be incorporated by the primase. The same results were obtained when the mix contained ddCTP in addition to dATP and dGTP; the product had the same length as if only dATP and dGTP were present (A), indicating that ddCTP is not incorporated either. Furthermore, an excess of dideoxynucleotides did not appear to compete with the incorporation of dNTPs (data not shown). In conclusion, neither ddNTPs nor non-cognate dNTPs are readily incorporated into the primer. The infidelity of primases can be caused by genuine misincorporation opposite of a template base or by slippage. As short primer/template duplexes are unstable, slippage could be the major contributor to primase infidelity. To assess the relative contribution of genuine misincorporation versus slippage, we took advantage of the strict sequence dependence of primer synthesis and analysed the fidelity at the second position of the primer with two different types of templates. To measure genuine misincorporation, we used a short primase template that allows the synthesis of a dinucleotide. Only when the cognate base was present in the template, a radioactive dinucleotide was formed (B). The lack of a radioactive dinucleotide for the other templates indicates that the misincorporation does not occur to an appreciable rate. We estimate that we would be able to detect misincorporation at least at a rate of 1/100 of the correct nucleotide. The second type of template allows the synthesis of a full-length primer. With these templates primer synthesis took place with relatively high efficiency, even when the cognate deoxynucleotide opposite the base at position 2 was not present in the reaction mixture (C, left). In all three cases the primers synthesized under these conditions were shorter than the primers synthesized when the deoxynucleotide is present. We therefore suggest that primer synthesis is possible in these cases because the ternary complex of primase, template and initiating ribonucleotide can move forward so that the second base of the primer is incorporated opposite the third position of the template. The slippage should be disfavoured when a longer primer has already been synthesized. In fact, when templates are used where the missing nucleotide must be incorporated at position 3, the synthesis of long primers is very inefficient and only dinucleotides are observed (C, right). Our data showed that the primase could initiate primer synthesis with an ATP regardless of the base 5′ of the GTG motif and then proceed to synthesize a primer made up of dNTPs. We investigated the kinetics of this reaction using substrates C and G (). With substrate C, the addition of ATP and dATP to the reaction will yield a full-length primer of 8 nt. We have measured the apparent and for the incorporation of dATP, ATP and for the template with this reaction. In the case of template G in the presence of dATP, only a dinucleotide is synthesized thus enabling us to study the kinetics of the formation of the initiating dinucleotide. With this template the apparent for dATP was determined ( and Figure S3). The apparent for dATP was ∼35 µM for the dinucleotide as well as for the full-length primer synthesis. This value is similar to the reported of a number of other eukaryotic primases, and of the primase, for rNTPs (,). At ∼150 µM, the for ATP is with somewhat higher than the for dATP whereas the for template is 200 nM. The maximal velocity values in the range of 0.2–0.4 pmol dATP/min correspond to a primer formation rate of about 1 full-length primer per minute per enzyme. A similar rate has been observed for DnaG when interacting with DnaB helicase () and for calf thymus primase (). In contrast the observed rate compares favourably to the synthesis rate of the primase (0.1 pmol dATP/min/µg which equals about 0.001 primers per minute per enzyme). The rate of dinucleotide formation (∼10 min) is higher than the rate of full-length primer synthesis (∼1.4 min), suggesting that the dinucleotide formation is not the rate-limiting step for full-length primer synthesis. Rather dinucleotide formation and primer extension by a single nucleotide appear to proceed with the same velocity (5–10 min). Although the extension of the primer seems to be a rather slow process the primase does not appear to dissociate from the DNA. The products of the primase reaction are mainly full-length primer. This was examined by measuring primase activity at two different dNTP concentrations (1 and 10 µM) over time (). Truncated products were observed mainly at the low dNTP concentration of 1 µM. Even at short incubation times and at 10 µM dNTP, which is a third of the , the majority of the products were full-length primers. Therefore even under non-saturating dNTP concentrations, the processivity of the primase to complete primer synthesis is high once a dinucleotide is formed. We have investigated the primer synthesis by the replication protein ORF904 from the cryptic plasmid pRN1 from using oligodeoxynucleotides as short defined templates. We found that there is a strict template dependence of the primase activity. One oligodeoxynucleotide yielded a short primer that was strictly dependent on ATP and displayed similarities to the primer that had been observed using single-stranded plasmid DNA as a template. This oligodeoxynucleotide was used to further investigate primer formation by ORF904. By mutating systematically the nucleotides of the oligonucleotide, the motif 5′-GTG-3′ was found to be of great importance. It is known that a number of bacterial and viral primases have preferred initiation sites and the recognition sites consist usually of a trinucleotide (). For example the DnaG protein requires 5′-CTG-3′ to initiate a primer (), the T7 primase 5′-GTC-3′ () and SP6 primase 5′-GCA-3′ (). Eukaryotic primases usually do not exhibit such sequence specificity but exceptions have been observed. The primase from mouse requires the trinucleotide 5′-CC(C/A)-3′ for efficient primer synthesis () and the viral herpes simplex helicase-primase has a somewhat relaxed dependence on a trinucleotide made up of 5′-pyr-pyr-G-3′ for the synthesis of longer primers (). The sequence specificities of archaeal primases have not been investigated extensively, but for the primase the highest primase activity has been reported with a thymine-rich bubble substrate (). The specificity of ORF904 for the GTG trinucleotide is very strict, which is rather unusual for primases. For example, the alteration of the respective recognition sequence usually leads to a decrease in the velocity of the primase reaction [e.g. mouse primase ()] or as is the case for herpes simplex primase to a decrease of the length of the synthesized primers and the primer synthesis rate (). Some primases, e.g. the T7 primase, do however display similarly stringent sequence requirements and do not initiate a primer in the absence of a recognition site (). Shortening of the primase template revealed that ORF904 also accepts very short templates: a dinucleotide primer can be formed with a minimal template length of six nucleotides, i.e. two nucleotides upstream and one downstream of the recognition motif GTG. From the lengths of the primer formed and run-off products obtained with shortened oligodeoxynucleotides, we inferred that the primer is likely to be initiated at the first base upstream of the recognition trinucleotide. Thus, no part of the recognition site is copied into the primer and the site is therefore entirely cryptic. In contrast, prokaryotic and viral primases initiate the primer at the central nucleotide of the recognition sequence () whereas eukaryotic primases have been found to start primer synthesis some distance upstream of their recognition motif as in the case of mouse primase (). The utilization of nucleotides during primer formation by ORF904 has been assessed using oligodeoxynucleotides with the recognition sequence. With γ-labelled ATP it could be shown that a ribonucleotide is incorporated as the first moiety of the primer. We could also show by exchanging the bases at the position of the template opposite the first nucleotide of the primer that the other ribonucleotides can also serve as first base of the primer. Contrary to ATP they are generally only efficiently incorporated when base pairing with the template at this position is possible. This preference for ATP is in accord with the observation that primases generally prefer to initiate a primer with a purine (,). Primases are known to accept NTPs with modified triphosphate groups as first base of a primer (,,) and likewise ORF904 does seem to be able to initiate a primer using certain modified ATP analogues. But the fact that AMPPcP and AMPPnP are incorporated while AMPcPP and γ-S-ATP are not indicates that there are certain other structural requirements that have to be met by the first nucleotide of the primer. The rest of the primer is exclusively made up of dNTPs. For other archaeal primases such as the primase from (), the primase from () and the small primase subunit from () it has been shown that dNTPs are also utilized to generate primers or are even the preferred substrates for primer synthesis. In these cases however, in contrast to the substrate requirements of ORF904, a combination of dNTPs and rNTPS is not required for efficient primer synthesis. It rather appears that the active sites of these archaeal primases are able to accept both rNTPs and dNTPs. For the primases the kinetic data suggests that an RNA primer is formed. On the other hand, the pyrococcal primases appear to prefer dNTP and might synthesize a DNA primer . Interestingly the activity of the small catalytic primase subunit of is modulated by the addition of the large primase subunit. If both subunits act together the efficiency to synthesize RNA primers is increased (). The ability to incorporate dNTPs has also been observed for bacterial primases such as DnaG from (). The physiological relevance of the dNTP incorporation is not known. The length of the primer that ORF904 synthesizes is 8 nt using single-stranded plasmid DNA or oligodeoxynucleotides as template. Longer products that are also observed seem to be the result of an elongation of the primer rather than an additional specific primase product. These primer multimers are usually synthesized as multiples of the unit-length primer () but we never observed any specific primase products at multiples of 8. It is not clear which part of the protein is responsible for the recognition of the GTG motif. By using different deletion mutants, we could show that the first 370 amino acids of ORF904 are sufficient for primer synthesis and exhibit a sequence specificity comparable to the full-length protein. The deletion mutant C255 (amino acids 1–255) that only comprises the prim/pol domain is not able to synthesize a primer either on single-stranded circular DNA or on oligodeoxynucleotides. Although DNA polymerase activity is seen for the deletion mutants C255 and C370, the primer synthesized by C370 is not efficiently elongated in the primase assays presented here. The deletion mutant C526 on the other hand elongates the primer as the wild type protein does, though it does not contain any part of the helicase domain, which is localized between amino acids 546 and 797. Therefore the central part of the protein from amino acids 370–526 seems to play a part in the observed elongation of a primer. For the T7 gene 4 product, which possesses a similar, but not evolutionary related, organization with an N-terminal primase and a C-terminal helicase domain, it could also be shown that the primase domain when expressed alone can synthesize primers with rates and sequence specificity as the full-length protein (). The substrate requirements of the prim/pol domain are very strict. Initiation is possible only with a ribonucleotide and some analogues and extension is only possible with deoxynucleotides. These findings are in line with models that postulate that priming requires two distinct sites: an initiation site and an elongation site (). For the formation of the initial dinucleotide, both sites of the primase are occupied and catalysis takes place: the nucleotide at the elongation site loses its pyrophosphate group, and its α-phosphate group reacts with the 3′ hydroxyl group of the nucleotide in the initiation site. Next the primase moves along the template, and the nucleotide in the extension site now occupies the initiation site and a new nucleotide can bind to the empty extension site. According to this model, the replication protein ORF904 accepts only ribonucleotides at the initiation site and only deoxynucleotides at the extension site. This behaviour is in sharp contrast to the archaeal primases investigated. These enzymes appear to accept ribo- and deoxynucleotides for initiation and extension and are therefore able to synthesize DNA, RNA and mixed primers. Very often primases display a rather low fidelity, which is usually explained by the fact that the synthesized primer is excised after the replication and replaced with DNA by a more accurate DNA polymerase. Low fidelity has been shown both for prokaryotic primases [ ()] and archaeal-eukaryotic primases [ () and Herpes simplex virus ()] that readily misincorporate ribonucleotides with rates as high as 1 in 7. In contrast the bovine primase has misincorporation rates between 1/200 and 1/1600 (). This distribution of tasks between DNA polymerases (high fidelity and processivity) and primases (less accurate initiation) is an elegant solution to the problem that priming is probably intrinsically more inaccurate because of the low stability of the duplex of a short primer with the template strand. For example, slippage of the initiating nucleotide or dinucleotide would lead to an incorrect primer sequence. Here we show that the replication protein ORF904 synthesizes a primer consisting of an initiating ribonucleotide followed by seven deoxynucleotides. Therefore in the context of the plasmid replication, it is improbable that the DNA primer synthesized by ORF904 can be recognized later and specifically removed. Accordingly we find that that the primase will only reluctantly incorporate non-cognate deoxynucleotides. We do not observe elongation of the initiating ribonucleotide to a dinucleotide in the absence of the cognate deoxynucleotide. Likewise elongation of short primers is not possible in the absence of the cognate deoxynucleotide. However we observe considerable slippage at the second but not the third position of the primer. Our results suggest that under conditions (availability of all dNTPs), misincorporation occurs only at the initiating ribonucleotide and possibly also at the deoxynucleotide at position 2 due to slippage. Our findings also have interesting evolutionary implications. According to a detailed study of Iyer and coworkers (), the archaeo-eukaryotic primase superfamily (including the archaeo-eukaryotic primases and the prim/pol domain proteins) might have a common ancestor with the family B DNA polymerases which are the replicative polymerases of the archaea and the eukaryotes. The specialization of the cellular replication machinery with separate primase and DNA polymerase activities conceivably did not exist at the invention of DNA as genetic material. It seems plausible that an enzyme capable of DNA polymerization was central to primordial DNA replication. We show here that the prim/pol domain is able to perform DNA polymerization with rather high fidelity, suggesting that an enzyme with similar properties could have been able to replicate simple DNA genomes. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

The early steps after translation initiation are critical for the efficiency of gene expression. In addition to the initiation codon in the mRNA, recognized by the initiator tRNA, the flanking sequences affect the rate of translation (). Upstream from the initiation codon there is usually a four to six nucleotide tract, named Shine–Dalgarno sequence () or SD, that pairs with a complementary, or anti-SD sequence, close to the 3′-end of the 16S RNA of the 30S ribosomal subunit. The SD region directs the ribosome to the initiation codon during translation initiation (). The downstream region (DR), the nucleotide sequence following the initiation codon also affects the efficiency of translation (). It is unlikely that the DR acts by pairing with a complementary sequence in the 16S rRNA as SD does (,). It has been suggested that the effect of the DR on gene expression stems from the base sequence in mRNA rather than from the encoded amino acid sequence in the protein. Comparison of DRs containing different iso-codons, thus generating an identical protein, could give significant differences of gene expression (). Particularly critical for gene expression is the nature of the codon next to the initiation triplet, the +2 position. Codon changes in +2 can affect gene expression by 15- to 20-fold (,). Also, the effect of the +2 codon on the gene expression can be modulated by the subsequent triplets (,). In general, a high adenine content of the +2 codon is associated with high gene expression (). The lysine codon AAA is the most common codon at +2 to +5 positions in reading frames and it usually determines efficient gene expression (,,). Indeed, the changes of gene expression due to variations in the content of adenines downstream of the initiation codon, correlates with changes in ribosome-binding strength, an association probably mediated by protein molecules (). Tandems of the low usage arginine codons AGA or AGG at different positions in the reading frame inhibit gene expression (). The longer and closer to the initiation codon is the tandem the stronger is the inhibition; starting up at the +10 codon, the closest assayed position (). The effect of low-usage codon tandems located farther downstream from the initiation codon is observed under conditions which favor tRNA limitation. For example, translation of a Z variant harboring contiguous AGA codons in the positions +352 and +353 is arrested at these codons upon the concurrent expression of a minigene that sequestered the cognate tRNA as peptidyl-tRNA (pep-tRNA) in a host defective for peptidyl-tRNA hydrolase () (). Also, depleting the pool of tRNA stops the ribosome movement and enhances tagging at tandem of AGA codons by the SsrA system ().The naturally occurring AGA and AGG codons in the positions +3 and +4 of the phage lambda gene modulate the expression by tRNA sequestration (,). It is likely that the expression of in cells may be limited by the elevated drop-off rate of pep-tRNA that overwhelms the Pth activity in the cell. We have investigated the effect of the substitutions of arginine codons at positions +2 and +3 on the expression of a reporter gene. Unexpectedly, CGU and CGC, the arginine codons more frequently used in bacteria, were deficient in supporting the gene expression whilst AGA and AGG, two of the less frequent arginine codons, were the most effective in wild-type bacteria. In spite these results, a variant substituted for the AGA AGA codons in a mutant strain was deficient in gene expression due to ribosome stalling at these codons. This indicates that the efficiency of translation does not necessarily correlate with the propensity of the pep-tRNAs to dissociate from the ribosomes. The nucleotide composition of the codons at +2 and +3 is a dominant factor in translation efficiency. Therefore, the deficiency of the CGU and CGC codons located at these positions, is due to the unfavorable base composition for translation rather than to an increased rate of abortive pep-tRNA dissociation from the ribosome. We carried experimental procedures on the K-12 strains P90C [] and its mutant P90C rap [P90C (rap) ::Tn10]. This mutation just expresses 10% of the normal Pth activity (). The plasmids used were ampicillin-resistant derivatives of pKQV4 () containing Z gene variants in the second and third codons of the ORF (). We also employed pDC952 which carries U, the gene for tRNA, cognate to the AGA codon () and pGREC, containing the gene of (,). Both constructs are chloramphenicol-resistant pACYC184-based derivatives. The cell cultures of the strains harboring the Z variant plasmids were grown at 37°C in Luria–Bertani (LB) medium containing ampicillin 100 µg/ml (Amp). The strains co-transformed with the Z variant plasmids and pDC952 or pGREC were grown in LB-Amp medium plus chloramphenicol 50 µg/ml (Cm). The Z gene from pLEX/Z plasmid (Invitrogen) was amplified by PCR and the final product cloned between the EcoRI and HindIII restriction sites of pKQV4 (). The second and third codons of the Z ORF were replaced by identical or combined pairs of all six arginine (AGA, AGG, CGA, CGG, CGC and CGT), leucine (CTA and CTC) and lysine (AAA) codons. The Z constructs were obtained by site-directed mutagenesis using a mutagenesis kit (Stratagene) and pairs of complementary oligonucleotides with the common sequence 5′-CAGAATTCATGNNNNNNCCCGTCGTTTTACAACG-3′ and 5′-CGTTGTAAAACGACGGGNNNNNNCATGAATTCTG-3′ where the tracts of Ns represent the above-mentioned codon substitutions and their complementary sequences. Fresh cell cultures at an OD of 0.3 were diluted to an OD of 0.1 in the same fresh medium preheated at 37°C. After 10 min, 1 mM of IPTG was added to induce β-galactosidase (β-Gal) synthesis from the Z variant plasmids. Samples were extracted at different times and the β-Gal activity was determined by employing a modified procedure of the Miller protocol (). The cellular density of the samples was measured at OD and immediately, 0.5 ml of each sample was mixed with 0.5 ml of Z buffer containing 30 µl of chloroform and 15 µl of 0.1% SDS. The mixes were vortexed for 30 s and then incubated at room temperature for at least 10 min. To start the reaction, 200 µl of ONPG solution (4 mg/ml) in Z buffer was added to the samples. The reaction was stopped by mixing 0.5 ml of 1 M sodium carbonate (the reaction time fluctuated depending on the velocity of β-Gal synthesis). All the samples were centrifuged at 10 000 r.p.m. for 10 min to sediment the cell debris and to measure the OD of the supernatants. With these data, we calculated the β-Gal activity in Miller units as previously indicated (). Reaction rates (Miller units/min) were calculated by linear regression from the β-Gal synthesis obtained at different incubation times for each Z variant. Pep-tRNA levels were measured by northern blot assays as previously described (). Briefly, cultures of the mutant transformed with Z variant plasmids were grown at 37°C to an OD of 0.3 in LB-Amp. Then, 1 mM IPTG was added and the cultures incubated for 40 min more. The cells were harvested at 4°C and the total tRNA was isolated under acidic conditions (). To estimate the fraction of pep-tRNA relative to total tRNA, aminoacyl-tRNA was hydrolyzed with copper sulfate in one of two aliquots. Four microgram of RNA from each sample were resolved overnight by acid/urea PAGE, transferred to Hybond-N nylon membranes (Amersham Biosciences) and hybridized to 5′-P end-labeled oligonucleotides. The radioactive signals were quantified using a Typhoon Scan (Amersham Biosciences). The amount of pep-tRNAs in the samples was estimated using the following formula:% pep-tRNA = c.p.m. of pep-tRNA × 100/ c.p.m. of uncharged-tRNA + c.p.m. of aminoacylated-tRNA + c.p.m. of pep-tRNA. The tRNA-specific oligodeoxyribonucleotide probes: 5′- CCTGCGGCCCACGACTTAG-3′, for tRNA; 5′-CCTGCAATTAGCCCTTAGG-3′, for tRNA; 5′-CCTCCGACCGCTCGGTTCG-3′, for tRNA; 5′-CCTGAGACCTCTGCCTCCGGA-3′, for tRNA; 5′-CCTGCGACCAATTGATTAAA-3′, for tRNA; 5′-CACCTTGCGGCGCCAGAA-3′, for tRNA; 5′-CCCGCACAGCGCGAACGCCG-3′, for tRNA were chemically synthesized according to the sequences reported by Dong (). Cultures induced with 1 mM IPTG were grown to an OD of 0.4 and harvested by centrifugation. Total RNA was extracted with hot phenol (65°C) from a 10 ml culture as described by Aiba (). About 30 μg of RNA was denatured in 40% formamide plus 5 μg/ml ethidium bromide solution at 65°C for 10 min. The RNA species were resolved by electrophoresis through 1.5% denaturing agarose gel containing 2.2 M formaldehyde and transferred to a nylon membrane (Hybond-N, Amersham Pharmacia Biotech). Hybridization was carried out at 42°C in 5× SSPE, 0.5% SDS, 100 μg/ml salmon sperm DNA, 0.1% bovine serum albumin, 0.1% Ficoll and 0.1% polyvinyl pyrrolidone and 5′-P end-labeled antisense oligonucleotide: 5′-CGTTGTAAAACGACGGGTCTTCTCATGAATTCTG-3′, for Z AGA AGA variant; 5′-CGTTGTAAAACGACGGGCCGCCGCATGAATTCTG-3′, for Z CGG CGG variant; or 5′-CGTTGTAAAACGACGGGTAGTAGCATGAATTCTG-3′, for Z CTA CTA variant. After 16 h incubation the membranes were rinsed twice at room temperature with 2× SSPE, 0.1% SDS, dried and exposed to X-ray film to develop the signal. It has been noticed that the presence of low usage AGA triplets in Z +2 and +3 codon positions (AGA AGA) induces a robust expression of the Z gene (). This was so in spite of reports that contiguous low usage codons near the initiation codon were deleterious for gene expression (,,,). In order to further examine this apparent paradox we analyzed the effects of AGA AGA and other pairs of identical arginine codons substituted in these positions of the Z ORF. By maintaining the N-terminal amino acid composition of the protein, variations in the β-Gal activity due to different N-terminal protein composition were eliminated. We generated the appropriate Z variants by site-directed mutagenesis on the construct shown in . Then, P90C a strain lacking in the operon, was transformed with the different constructs bearing each of the double substituted variants. The Z expression was induced with IPTG and the rates of β-Gal synthesis calculated (see Materials and Methods section). The results (A) showed that the AGA AGA Z variant generated the highest rate of β-Gal synthesis in the wild-type strain. The rate of β-Gal synthesis measured for the AGA AGA variant was nearly as high as that observed for a variant which contained a pair of AAA codons in +2 and +3 positions (A). This last pair of lysine codons has been reported as one of the most efficient pairs for gene expression (,) and one of the most frequently located at these positions (,). The variant harboring low-usage AGG codon, AGG AGG, followed expressing Z about one-third as fast as the AGA AGA construct. The variants with pairs of other identical low usage (CGA, and CGG) or common (CGC or CGU) arginine codons generated low levels of β-Gal activity. A similar result was observed using leucine codons. A pair of CUA leucine low-usage codons in these positions promoted Z expression comparable to that expressed by the AAA AAA and AGA AGA pairs (A). However, a variant substituted with the leucine common codons CUC CUC, was defective to promote gene expression. Thus, the differences of gene expression due to changes in the second and third positions of Z ORF cannot be assigned to the codon usage or to the relative abundance of the tRNA isoacceptors. We investigated whether the abortive translation process of pep-tRNA drop-off played a role in the scant Z expression observed for some of the variants. The appropriate constructs were transformed into the (rap) strain. The rates of β-Gal synthesis and the levels of cognate pep-tRNA accumulation were determined upon IPTG induction (A and ). The AGA AGA variant, which expressed β-Gal very efficiently in the wild-type cells, promoted poor β-Gal synthesis and high accumulation (70%) of pep-tRNA in the strain (see A, 4B, 4C and ). On the other hand, the AGG AGG variant, which expressed β-Gal moderately in wild-type cells, mediated 5-fold less expression in the cells and accumulated intermediate level (36%) of pep-tRNA (A and ). The high rate of the Z gene expression, especially that of the AGA AGA variant in the wild-type cells, is not incompatible with the high level of pep-tRNA drop-off under limiting Pth activity. Rather, this observation suggests either that drop-off does not occur in the wild-type cells or that, if it does, the level of Pth activity present readily hydrolyzes the released pep-tRNAs preventing starvation for free tRNAs. Associated to the Z expression inhibition, the Z AGA AGA variant also affected the grow rate of the mutant (D, pACYC). No comparable levels of accumulation of the corresponding pep-tRNAs were observed for the other variants assayed (). Unlike the AGA AGA and AGG AGG Z variants, the efficient expression of the AAA AAA and CTA CTA variants was only somewhat reduced in the strain (A) in agreement with the low levels of the pep-tRNA accumulated upon induction of these variants (). Therefore, it appears that there is no correlation between the high rates of protein synthesis in the wild-type cells and the accumulation of pep-tRNA in the cells. On the other hand, the poor β-Gal expression mediated by other Z variants (CGG CGG, CGC CGC, CGA CGA, CGT CGT or CTC CTC) in the wild-type cells was also accompanied by inefficient β-Gal synthesis and modest accumulation of the cognate pep-tRNA in the strain (A and ). Thus, for this last group of variants, the inefficient synthesis of β-Gal protein in wild-type cells may result from defective translation of the codons located in the +2 and +3 positions or defective interaction of the mRNA with the ribosome, but not from abortive translation. We investigated whether the favorable effect of AGA codons on the Z expression was associated with its location in the second, the third or both codon positions. Z variants containing combinations of the AGA codon in the +2 or +3 positions and each of the other five arginine codons placed in the reciprocal +3 or +2 positions were assayed. The appropriate constructs were transformed into the P90C strain and the rates of β-Gal synthesis determined upon IPTG induction (see Materials and Methods section). The results, shown in B, indicated that the different variants containing the combinations of AGA and the other arginine codons expressed broadly different levels of β-Gal activity which spanned almost a 10-fold range. But in all cases, the variants harboring the AGA codon in position +3 expressed the Z gene at higher rates than the corresponding variants harboring the AGA codon in position +2 (compare wild-type columns in B). The variants containing an AGA codon, either in positions +2 or +3, and any of the other five arginine codons in the alternative position, enhanced the rate of Z expression relative to those variants harboring identical pairs of non-AGA arginine codons (compare A and B). The variants carrying combinations of the AGA codon in position +2 and other favorable codons such as the CUA or AAA codons in location +3, promoted as high rates of Z expression in the wild-type cells as those mediated by Z variants carrying the pairs AGA AGA, CUA CUA and AAA AAA (compare A and B). These non-arginine favorable codons may share with AGA the enhancing effect with the CGN codons. From the results in , it appears that AGA and AGG are the arginine codons that mediate the highest accumulation of pep-tRNAs in the strain during expression of corresponding Z variants. We asked whether the position of the AGA codon, either in the positions +2 or +3 induces drop-off of the involved pep-tRNAs. The expression of Z variants containing alternative combinations between AGA and other arginine codons in positions +2 and +3 were assayed in the (rap) mutant. The results (B) showed, in general, that the rate of Z expressed by a variant in the wild-type cells was higher than the rate promoted by the same variant in the strain. These results are compatible with the notion that, under limiting Pth activity, the AGA codon and/or the accompanying arginine codon of the pair have a propensity to drop-off. Because the cells are defective in the regeneration of aminoacylable tRNAs from pep-tRNAs, they would be limited in their capacity to synthesize β-Gal. Then, the relative accumulation of pep-tRNA and the pep-tRNAs specific for the accompanying arginine codons was measured (, see Materials and Methods section). With the exception of the AGA CGT variant, all the Z variants carrying the AGA codon accumulated pep-tRNA at higher levels than 20%. The pep-tRNA specific for the accompanying non-AGA arginine codon in the pair was accumulated by the variants where AGA resided in the +3 codon position, but not in the +2 location (, compare lines 1 and 2, 3 and 4, etc.). These data argue that the AGA codon promoted translation of the accompanying arginine codon and that the pep-tRNAs of these codons accumulated in response to the drop-off at the subsequent AGA codon. Therefore, AGA in the positions +2 and/or +3 promoted translation of the associated arginine codon and facilitated the dissociation of pep-tRNA from the ribosome affecting the rate of Z expression under limiting Pth activity. The variants harboring AAA AAA or CTA CTA expressed Z efficiently in the wild-type cells and only slightly less well in the strain (A). This was observed also with variants bearing the codon combinations AGA AAA and AGA CUA (B). Thus, combinations of codons that express Z variants efficiently as identical pairs placed in +2 and +3 are also efficient in mixed combinations. In addition the levels of expression of these variants in the cells, suggest that they do not promote the accumulation of pep-tRNAs. Thus, it seems that the pep-tRNA accumulates in conditions where the codon subsequent to AGA is difficult to translate. It is assumed that starvation for an aminoacylated-tRNA due to the accumulation of pep-tRNA in cells induces ribosome stalling at the ‘hungry’ codons in the mRNA. The stalled ribosomes protect the associated mRNA against the ribonucleolytic activities of the cell (,). Therefore, it is expected that ribosomes stalled in the AGA codons, located next to the initiation codon, would protect a 5′-proximal segment of the mRNA. To test this prediction the Z variant harboring AGA codons in the positions +2 and +3 was expressed in the strain. Total RNA was extracted and submitted to northern blot analysis using an oligonucleotide-probe complementary to the 5′-end Z mRNA (see Materials and Methods section). The results showed that the expression of Z AGA AGA variant generated a short transcript containing the 5′-end sequences of Z mRNA (B). The expression of the same variant in the wild-type strain, that expressed the Z gene efficiently, yielded Z transcripts of a wide size range, mostly larger than the 5′-end fragment (A). The CTA CTA variant, that expresses the Z gene nearly as efficiently in the mutant as it does in wild-type cells, produced the wide size range pattern of Z transcripts in both strains. On the other hand, the Z CGG CGG variant, that was very ineffective to express Z (A), yielded no Z mRNA signals at all (). Likewise, no evidence of Z mRNA signals were observed for the low expressed variants carrying codons such as CGA CGA, CGC CGC, CGU CGU, CGA AGA or AGA CGA (data not shown). We suggest that these pairs of codons in positions +2 and +3 are defective in translation either because they fail to interact properly with the ribosome or because these codons are difficult to read by the specific tRNA. The presence of the 5′-end Z mRNA segment in the cells expressing Z AGA AGA variants (B and A, pACYC) correlated with accumulation of pep-tRNA (B, pACYC), deficiency of Z expression (C, pACYC) and reduction of the rate of cellular growth (D, pACYC). Cells supplemented with an excess of Pth protein or tRNA showed longer species of Z mRNA (A, lanes 2 and 3), did not accumulate pep-tRNA (B, lanes 4 and 6), restored β-Gal activity and rescued the cellular growth (C and D). These results support the notion that, under limited Pth activity, the production of β-Gal protein from the Z mRNA containing the AGA AGA codons was limited by the starvation for charged tRNA and the ribosome pause at the AGA codons in the Z mRNA. The presence of the 5′-proximal short mRNA was not observed in a preparation of the Z CTA CTA variant (). Instead, they produced the pattern of long Z transcripts associated with the high rates of Z expression (). In general, the presence of large-size Z mRNAs correlated with high rates of β-Gal synthesis, but low rates of β-Gal synthesis were compatible either with high levels of truncated Z mRNA, as in the AGA AGA variant in the cells, or no Z mRNA at all. The objective of the current study was to understand how the sequences of the early codons in ORFs affect gene expression in bacteria. A recent study shows that the low-usage AGA triplets in early codon positions induce a robust expression of the Z gene (). In order to know how these codons affect mRNA translation, we have analyzed their effect on the efficiency of Z expression, induction of pep-tRNA drop-off and Z mRNA concentration. The AGA AGA Z variant promoted the highest rate of β-Gal synthesis, the AGG AGG variant was fairly efficient and all the variants carrying identical pairs of CGN codons (CGU, CGC, CGA or CGG) were rather defective (A). The combination of one AGA codon, in either position +2 or +3, with any other arginine codon, in the alternate +3 and +2 positions, promoted Z expression efficiently (B). Interestingly, the high levels of Z expression induced by AGA codons in wild-type cells were compatible with the high rate of pep-tRNA drop-off in bacteria ( and ). On the other hand, it was difficult to assess whether the low rate of Z expression promoted by the CGN codons was accompanied by pep-tRNA drop-off ( and ). Therefore, in the analyzed cases, there is no correlation between gene expression and pep-tRNA accumulation. As expected, the level of Z expression corresponded with the concentration of Z mRNA accumulated (). These findings support and extend the notion that the sequence downstream the initiation codon affects the efficiency of gene expression. Possibly, the AGA AGA, unlike CGN CGN, facilitates the interaction of the ribosome with the mRNA (see subsequently). The efficiency of gene expression is affected by multiple factors relative to the early codon composition: secondary structure of mRNA, efficiency of mRNA association with the ribosomes, degree of ribosome pausing at specific codons, propensity of the different pep-tRNAs to drop-off, rates of codon reading by the tRNAs and codon context itself. Here, it is shown that the variant that harbors AGA AGA promote quite an efficient expression of the Z gene in bacteria. This expression is comparable to that attained by the Z variant that carries AAA AAA (A), a codon configuration frequently found in the highly expressed genes of (,). The efficient expression of the AGA AGA variant in wild-type cells occurred in spite of elevated rates of pep-tRNA drop-off. This was revealed by the high levels of pep-tRNA accumulated upon expression of the variant in a mutant deficient in Pth activity () (). It seems that the mechanism involved in enhancing translation efficiency by the AGA codons overwhelms the negative effect of the pep-tRNA drop-off that occurs at these codons. It is likely that in the mutant, the expression of Z leveled off soon after an initial burst of β-Gal synthesis due to starvation for the aminoacylable tRNA which was sequestered as pep-tRNA (C, inset). Accordingly, the overproduction of Pth and tRNA, conditions that increase the pool of aminoacylable tRNA in the cell, mitigated the defect in β-Gal synthesis (C). The propensity of pep-tRNA to drop-off did not depend on the presence of two contiguous AGA codons because the expression of variants carrying the AGA codon in combinations with the other arginine codons also resulted in the accumulation of pep-tRNA (). We did not observe any correlation between the rate of pep-tRNA drop-off ( and ) or the frequency of codon usage and the efficiency of Z expression (A and B) consistent with previous findings relative to codons located next to the initiation codon (). Instead, our results conform to the correlation between gene expression and adenine content of the early ORF sequences that has been proposed for different genes (,). Accordingly, the AGA AGA variant showed the highest rate of Z expression, the AGG AGG variant expressed Z at an intermediate rate and the CGN CGN variants (where N is adenine, see subsequently) expressed Z the poorest. It has been proposed that the early adenines enhance translation by increasing the rate of association of mRNA and ribosomes. Rather than a direct interaction between mRNA and 16S rRNA by sequence complementation, the association could be mediated by ribosomal proteins (). One could argue that the CGA CGA Z variant, expressed poorly in spite the fact that it contained as many adenines as the AGG AGG variant. However, the CGA variant may be a special case as CGA has been reported as a codon difficult to translate by its correlative tRNA (). As the defective expression of the CGN CGN variants was not related to the accumulation of pep-tRNA specific to these codons under limiting Pth activity (), we assume that the mRNAs from these variants are deficient in binding to ribosomes or in the formation of ternary complexes with aminoacyl-tRNAs. The possibility that these codons, , are difficult to translate when located in the positions +2 or +3 is unlikely because when they are next to an AGA codon they express Z efficiently (B). Again the exception was codon CGA. We considered the possibility that the different degrees of expression of the Z variants were explained by other mechanisms: translation reinitiation, formation of internal secondary structures in the mRNA, and the AG-rich arginine codons acting as secondary SD sequences. In wild-type cells, it is unlikely that re-initiation of translation downstream of the AGA tandem would explain Z expression because the released pep-tRNA would be readily hydrolyzed by Pth and ribosomal stalling would not be expected to occur. Under limiting Pth, however, stalling does occur as a consequence of pep-tRNA accumulation and reduction in the pool of charged tRNA. However, if reinitiation takes place in (rap) cells, it should be negligible because the β-Gal activity synthesized was very scant (A and C). To assess the formation of secondary structures between the different codons in +2 and +3 positions and the neighboring nucleotide sequences in the mRNAs of the variants, an informational search using an appropriate program was used (). The data did not reveal a consistent correlation between the degree of Z expression and the proposed stability of the generated structures (data not shown). Then, it was examined whether the efficient gene expression of the Z variants carrying the codon pairs AGA AGA and AGG AGG was due to these pairs acting as secondary SD regions (). However, if this assumption was true, the facts indicate that it did not correspond to a simple scheme because first, the Z variant carrying the codon sequence AGG AGG, which is a near consensus SD, expresses the gene less efficiently than the codon sequence AGA AGA, a less SD-like sequence and second, unlike authentic SD sequences that anchor the mRNA to the 16S rRNA, the pairs of codons AGA AGA and AGG AGG in the Z mRNAs are translated robustly as shown by the accumulation of the respective pep-tRNAs in cells defective for Pth activity (). Furthermore, the levels of the pep-tRNAs accumulated in the (rap) mutant upon expression of the variants, correlated with the efficiency of Z expression in the wild-type cells. Thus the AGA and AGG codons seem to affect the rate of Z expression mainly by their contribution to the translation rate of the Z mRNA rather than by acting as secondary SD regions. However, a mechanism invoking the transient association of the ribosome with the secondary SD region represented by these codons, followed by sliding back to the original SD sequence to start translation (,) cannot be ruled out by our results.

During DNA replication, recombination and repair, double-stranded DNA inevitably forms three- or four-way junctions, bubbles, flaps or broken ends with single-stranded extensions. These irregular structures must be processed correctly in order to successfully complete DNA metabolism and thereby maintain genome integrity. This task is accomplished by structure-specific endonucleases specialized in pruning downstream of branch, flap or bubble structures by incision at junctions between double- and single-stranded DNA (). Inactivation or malfunctioning of these enzymes causes genetic defects or cancer, underlying their importance in genome stability. A remarkable class of structure-specific endonucleases in humans is XPF. The protein family is characterized by the presence of the ERCC4 domain and consists of seven members (XPF, MUS81, ERCC1, EME1, EME2, FANCM and FAAP24) (). Only XPF and MUS81 have nuclease activity, which is mediated by the conserved core nuclease motif (ERKXD) (). Their catalytic function depends on heterodimer formation with the non-catalytic family members. XPF forms an obligate complex with ERCC1 and functions primarily in nucleotide excision repair (NER), a versatile pathway able to detect and remove a variety of DNA lesions induced by UV light and environmental carcinogens. The ERCC1/XPF heterodimer has additional roles in DNA interstrand cross-link (ICL) repair () and telomere maintenance (). The symptoms of the first patient with inherited ERCC1 deficiency () and of a patient with a novel XPF mutation () are distinct from the classical NER phenotype, and underscore the pleiotropic function of ERCC1/XPF. In contrast to eukaryotes, archaea have a single homolog of the XPF endonuclease that forms homodimers. The archaeal XPF minimally consists of the catalytic nuclease domain followed by a DNA-binding domain containing two consecutive helix-hairpin-helix motifs (HhH domain). Dimerization occurs between both the nuclease and HhH domains of each subunit (A) (). The strong preference of the archaeal endonucleases for 3′ flap DNA substrates is explained by the function of their individual domains. Structural data suggest a model for DNA binding where one HhH domain of the dimer binds to an upstream and the other to a downstream DNA duplex, sharply bending the DNA substrate and thereby allowing one active site of the dimeric nuclease domain to cleave the 3′ protruding single strand (). Combination of mutations in nuclease and HhH domains that do not disrupt the dimeric interfaces but impair the individual functions (DNA cleavage or DNA binding), provide evidence for this model (). The DNA substrate specificity is different for the human ERCC1/XPF heterodimer. The human enzyme is known to specifically incise hairpin, bubble or splayed-arm DNA substrates (,), which consist of only one duplex (upstream). Previous data indicated the role of the C-terminal ERCC1 ‘canonical’ HhH domain in engaging the upstream DNA duplex of a hairpin substrate (). This function requires heterodimerization with the C-terminal HhH domain of the human XPF partner to form a complex analogous to the archaeal HhH dimer interface (,). Previous structural data () and the data presented here illustrate the structural similarity between the human ERCC1 central domain and the archaeal XPF nuclease domain. Unlike the archaeal nuclease dimer interface, there is no evidence that the corresponding domains of human ERCC1 and XPF interact with each other in solution (A) (). The catalytic function of the ERCC1/XPF endonuclease is crucial for NER. NER operates through a ‘cut and patch’ mechanism by excising and removing a short stretch of DNA containing the lesion, and subsequently restoring the genetic information by repair synthesis using the undamaged strand as the template. The incision step involves the sequential and coordinated assembly of the DNA damage sensor XPC-HR23B, the transcription/repair factor TFIIH, the architectural protein XPA, the ubiquitous ssDNA-binding protein RPA, and the two structure-specific endonucleases ERCC1/XPF and XPG, responsible for the incisions 5′ and 3′ to the damaged site, respectively (). A main role in the progress of the reaction is attributed to XPA, which in complex with RPA probes the DNA helix conformations () and participates in multiple interactions with the other NER factors (). XPA is required for the recruitment of ERCC1/XPF in the NER pre-incision complex (,). Consistent with this regulatory function, studies with recombinant proteins and studies using a yeast two-hybrid system have demonstrated interactions between XPA and ERCC1 (). Here, we report the solution structure of the ERCC1 central domain (cERCC1) and investigate its interactions with XPA and DNA at the molecular level. Using biochemical studies and NMR titration experiments, we have identified two distinct interaction surfaces of cERCC1 that mediate XPA and DNA binding. Interestingly, the two interactions can take place simultaneously to yield a cERCC1/DNA/XPA ternary complex, which in turn explains the important role of ERCC1 in targeting its catalytic XPF partner to the NER pre-incision complex. ERCC1 central domain (cERCC1, residues 96 to 219) was PCR-amplified from a vector containing the full-length ERCC1 gene and subcloned into a pET28b (Novagen) expression vector between BamHI and XhoI sites. Downstream of the XhoI site a linker together with a his-tag has been engineered. XPA constructs were derived from a full-length XPA vector purchased from RZPD (clone ID IRAUp969B1273D6) and subcloned into a pLICHIS vector using the enzyme free cloning (EFC) strategy (). The same procedure was followed for the GST fusion proteins, either XPA or cERCC1 (pLICHISGST vector). As a result all the EFC constructs contain an N-terminal his-tag, while the cERCC1 construct bears a C-terminal his-tag. The nucleotide sequences of the cloned DNAs were confirmed by sequencing. The cERCC1-HIS, HIS-XPA, HIS-GST-XPA and HIS-GST-cERCC1 proteins were expressed in BL21(DE3) Rosetta cells (Novagen) and were subject to two-step purification as has been described (). When appropriate, HIS-GST-cERCC1 was cleaved with thrombin in the presence of 1 mM Ca, 50 mM Tris (pH 8.0) and 100 mM NaCl in order to obtain the untagged protein. Finally, the protein samples were exchanged either to 50 mM sodium phosphate (pH 5.5), 100 mM NaCl for the cERCC1 structure determination, or to 50 mM Tris (pH 8.0), 100 mM NaCl for the binding studies. Multidimensional NMR experiments were carried out at 290 K on Bruker AVANCE 600 and 900 NMR spectrometers equipped with TXI triple-resonance probes in 50 mM sodium phosphate (pH 5.5), 100 mM NaCl by using the cERCC1-HIS protein. Spectra were processed using the NMRPipe software package () and analyzed with Sparky (). The H, N and C resonance assignments are 97% complete and were made using standard triple resonance techniques, 3D NOESY-(N, H)-HSQC spectra, and 3D NOESY-(C, H)-HSQC spectra (both with a mixing time of 60 ms) (). The chemical shifts have been deposited at the BMRB with accession number 15240. Automatic NOE assignment and structure calculations were performed using the program CYANA version 2.0 (). Hydrogen bond restraints were defined when they were consistent with the secondary chemical shift data and expected NOE contacts and only for the helical parts of the protein. The set of 2669 NOE restraints determined by CYANA, together with restraints for 31 hydrogen bonds and ϕ and ψ torsion angle restraints derived from TALOS () were used in a water refinement run according to the standard RECOORD protocol () utilizing CNS (). Molecular images were generated with PyMol (). Coordinates have been deposited at the PDB with accession code 2jpd and the structural restraints at the BMRB with entry number 15240. All the DNA substrates were 5′ P-labeled with T4 polynucleotide kinase and purified on a polyacrylamide gel under native conditions. The nucleotide sequences for ss20, b10 and ds30 have been described before (). Per reaction, 100 fmol of substrate was incubated with the appropriate amount of ERCC1 protein in binding buffer (50 mM Tris, pH 8.0, 100 mM NaCl, 10% glycerol and 1 mg/ml BSA). After incubation for 30 min at 4°C, samples were loaded onto 7.5% native polyacrylamide gels containing 0.5× TBE and run at 4°C. Alternatively Electrophoretic mobility shift assay (EMSA) reactions were loaded on a 3.5% agarose gel containing 0.5% TBE, run at 4 or 20°C yielding essentially identical results. Gels were visualized and quantified using a phosphor imager (BioRad) as described before (). For depletion experiments, 20 µl of MagneHis beads (Promega) or GST agarose beads (Sigma), were extensively washed with binding buffer, and after buffer removal the EMSA reaction was added to the beads. After 15 min incubation with regular mixing, the beads were removed from the EMSA reaction using a magnet or centrifugation and the reaction mixture was loaded on acrylamide gel or, when appropriate, the substrate was added to the depleted reaction mixture. GST pull-down assay was performed and quantified as described before () using the indicated cERCC1 concentrations and 3–5 µg of GST proteins in 150 µl of 50 mM Tris, 100 mM NaCl, 10% glycerol, 1 mM DTT, 0.2 mM PMSF and 20 mg/ml of BSA (pH 8.0). The semi-quantitative experiments were performed in 50, 150, 500, 1500 and 5000 µl buffer with a constant amount of cERCC1 and GST-XPA. cERCC1 titrations with XPA and DNA were performed on a Bruker AVANCE 700 NMR spectrometer and were monitored with 2D H-N HSQC experiments. In all cases, the concentration of cERCC1 was 0.2 mM in 50 mM Tris (pH 8.0), 100 mM NaCl. XPA and DNA were dissolved in the same buffer and therefore salt and pH did not vary throughout the experiments. DNA binding was feasible only with the cERCC1 construct lacking the C-terminal his-tag, while XPA binding was identical, regardless of the his-tag presence. Normalized chemical shift changes were calculated by using the equation: δ = ([δ] + [δ/6]). We have determined the solution structure of the ERCC1 central domain (cERCC1) to elucidate the molecular details of its proposed functions. We have collected a large number of distance and dihedral angle restraints that yielded an ensemble of 20 conformers with very good convergence (C). The quality of the structure can be judged by the summary of the structural and restraint statistics given in . All the secondary structure elements are well defined including the short loops that join them. cERCC1 comprises a compact architecture and folds as a six-stranded β-sheet flanked by five α-helices on both sides. As described before (), this fold is reminiscent of the type II restriction endonucleases, to which the catalytic domain of XPF also belongs. Overall, the solution NMR structure of cERCC1 is in very good agreement with the crystal structure of the same domain (2a1i) (rmsd 1.1 Å for 108 C atoms). The only noticeable difference relates to the last few residues of helix α3 and the loop connecting this structure element with β5. The presence of a mercury atom (linked to C137) in the crystal structure may account for the substantial side-chain rearrangements within this loop. When compared to the crystal structures of the archaeal nucleases, helices α1, α2, and strands β1, β2 appear to be shorter in the solution structure of cERCC1. Solution cERCC1 and the crystal structures of archaeal XPF nucleases (2bgw and 1j23) contain the same number of secondary structure elements arranged in a very similar manner (rmsd 1.8 and 2.0 Å for and nucleases, respectively, for 108 C atoms), although in both cases the primary sequence homology is very low (B and D). Remarkably, conserved residues of XPF nuclease scattered in the primary sequence superimpose structurally with residues conserved in the ERCC1 sequence family ( and ). Whereas the conservation in XPF is directly related to the catalytic function, ERCC1 has preserved the same fold but lacks the essential residues for catalysis. The fold similarities, coupled with the obligate nature of the heterodimerization (), are in full agreement with the common origin of the two proteins (). Truncation studies of ERCC1 have mapped the XPA interaction site to a region between residues 91 and 118 (). From the cERCC1 structure and its compact fold, we predict that such truncations will have a devastating effect on the structural integrity of the central domain (96–219). For XPA, the ERCC1 interaction region seems to be located in a small stretch containing two highly conserved motifs rich in glycine and glutamic acid residues (72-ILEEE-84, conserved residues are underlined) (). We have prepared a HIS-GST-XPA construct (59–99) containing the conserved stretch, and examined its ability to interact with cERCC1 in a GST pull-down assay. Indeed, the GST XPA fusion protein was able to specifically bind cERCC1 (A), confirming that this domain of ERCC1 is sufficient for interactions with XPA. No binding of cERCC1 to GST bound agarose beads or uncharged agarose beads was observed (A, data not shown). This XPA peptide is unstructured () and appeared very sensitive to proteolysis both during overexpression in and as pure protein after extensive purification (A and B). We next performed a semi-quantitative GST pull-down assay () and estimated an apparent of 1 µM. Since we reached only 20% binding saturation at the two highest protein concentrations (3 and 10 µM of cERCC1) under these experimental conditions, we could not determine the binding constant more accurately (B). The cERCC1–XPA interaction specificity has been explored using NMR titration experiments. Titration with the HIS-XPA peptide causes extensive specific changes in the cERCC1 H-N HSQC NMR spectra. The titrations show slow exchange regime with respect to the NMR chemical shift timescale, in agreement with the determined affinity (C). We have been able to assign all the amide resonances of cERCC1 in the bound form (1:1 complex stoichiometry) by using 3D NOESY-(N, H)-HSQC spectra. The pattern of the backbone amide NOEs does not change substantially compared to the free cERCC1 protein. This indicates that the overall structure is maintained, and led us to conclude that the observed chemical shift perturbations arise from specific contacts with the XPA peptide. The most significant perturbations map to the surface of the V-shaped groove of the cERCC1 structure and involve many residues that compose the corresponding DNA cleavage site of the archaeal XPF endonucleases (D and ). Interestingly, among the largest chemical shifts of cERCC1 upon XPA binding are those of L139, F140 and L141. These ERCC1 residues correspond to the characteristic strongly conserved ERK catalytic triad of the XPF nuclease domain (). The rest of the significantly perturbed residues are polar or positively charged (Q107, N110, S142, R144, N147 and R156), are competent for hydrogen bonding with the XPA peptide and are more conserved than the surrounding regions of the cERCC1 V-shaped groove (). Additionally, of special interest are the amide signals of G109 and H149, which are not observed in the free protein spectrum due to fast exchange with the solvent at pH 8 (these amide resonances are observable at pH 5.5). However, both give sharp signals at the end of the titration, an indication that the solvent exchange in the free protein is quenched in the complex. These residues reside at the rim of the V-shaped binding site and may be involved in hydrogen bonding with the XPA peptide (). Although ERCC1 residues forming the XPA-binding site are different from the acidic and basic residues which form the active site in XPF (), there are no significant changes in the backbone fold and the spatial side-chain orientations in the structure remain unaffected. However, their different chemical properties result in a distinct charge and hydrophobicity distribution on the structure surface (). While XPF displays negative charge at the DNA cleavage site (important for divalent cation coordination), the equivalent XPA-binding site in ERCC1 is neutral or slightly positively charged. Therefore, the ERCC1–XPA interaction is a combination of hydrophobic and electrostatic interactions possibly involving the glutamic acid stretch of XPA () and two conserved positively charged residues (R106 and R156) of cERCC1. Given the proposed common origin of ERCC1 and XPF (), it can be expected that the degenerated catalytic site of ERCC1 could still bind DNA in a similar fashion as the archaeal XPF active site (). Because our ERCC1–XPA interaction data render this unlikely, we explored whether cERCC1 is able to bind DNA as has been suggested previously (). We performed an electrophoretic mobility shift assay (EMSA) where increasing amounts of cERCC1-HIS protein were added to ssDNA. While we saw a loss of free probe, we failed to see any complex formation under these conditions. If the same protein–DNA complex was loaded on a 3.5% agarose gel, we detected a weak smeary complex at the highest protein concentration (A). However, by using a HIS-GST-cERCC1 protein, we observed formation of the complex with ssDNA (B). This suggests that the C-terminal his-tag interferes with DNA binding. Therefore, we used the HIS-GST-cERCC1 protein, which we cleaved with thrombin to obtain the untagged protein. DNA binding with the untagged cERCC1 demonstrated a faster migrating complex with various ssDNA substrates (data not shown). HIS-GST-cERCC1/DNA complex formation is prevented by incubation of the reaction mixture with MagneHIS magnetic beads or glutathione agarose beads, which depletes HIS-GST-cERCC1 from the reaction mixture either prior to or after the addition of DNA. If beads were added after complex formation, a small but significant decrease in the amount of free probe was observed. Depletion of the HIS-GST tagged cERCC1 protein from the binding reaction confirms that the protein–DNA complex is indeed formed by the HIS-GST-cERCC1 protein (C). In agreement with earlier studies, under these conditions no binding was observed for dsDNA (), while fairly comparable binding affinities were found for both ssDNA and bubble substrates with 10 or 20 unpaired bases. We calculated an apparent of 2.5 ± 0.7 μM for the HIS-GST-cERCC1 protein bound to a bubble10 (b10) substrate in the presence of 100 mM NaCl (B). Corroborating our results, an equilibrium binding titration experiment using fluorescence anisotropy has demonstrated ssDNA binding for the same domain with an apparent of 10 μM (). The relatively small difference in affinity can be well explained by different salt conditions or substrate used. For the NMR titrations we performed a thrombin cleavage on HIS-GST-cERCC1, to remove the his-tag. The untagged cERCC1 H-N HSQC spectrum is identical to that of cERCC1-HIS, except for the two amides preceding the artificial his-tag (Supplementary Figure 1). Addition of the b10 substrate to this cERCC1 protein caused a limited number of specific perturbations that were unambiguously assigned. Again, the perturbed resonances exhibit slow exchange behavior in the NMR titrations and indicate contacts with the DNA. The affected resonances include the backbone amides of N99, I102, L132, K213, A214 and the side-chain amide of Q134 (D and E). Compared to the free-protein spectrum we only miss T211, which cannot be identified. The perturbation induced by the b10 DNA was complete at equimolar concentration with cERCC1. Chemical shift mapping reveals that all perturbations are in the vicinity of the C-terminus of the cERCC1 construct, consistent with the presence of a flexible tag interfering with DNA binding. The DNA-binding region resides at the surface of the structure where the N- and C-termini meet (F). The perturbed residues belong to well-defined structure elements. The only positively charged residue we identified (K213) is fully conserved in the ERCC1 proteins and may explain the dependence of DNA binding on salt concentration. All the other affected residues are less conserved than the residues involved in XPA binding. The composition of the affected residues together with the slow exchange behavior suggests a large contribution to the binding by hydrophobic interactions with the DNA bases. Summarizing, both the biochemical and the NMR experiments show that the ERCC1 central domain binds to DNA. The DNA-binding site we identified on cERCC1 by the NMR titrations suggests that cERCC1 contacts only a small part of the ssDNA, probably three to four bases. The situation is different in the full-length heterodimer, where other DNA-binding surfaces, such as previously established for the C-terminal ERCC1/XPF domains (,), assist in DNA recognition. To independently confirm the formation of a ternary cERCC1/ssDNA/XPA complex, we added the HIS-XPA peptide to the cERCC1-bubble10 protein–DNA complex and followed the chemical shift perturbations in the H-N HSQC spectra. As shown in , addition of an equimolar amount of the XPA peptide causes the same amide displacements as in the titration with the free protein. On the contrary, most amides influenced by the DNA binding remain unaffected by the presence of XPA (A). I102 and N110 were perturbed by both XPA and b10 DNA when the titrations were performed independently. In the ternary complex, however, I102 remains close to the position as in the DNA bound form. Conversely, the side-chain amide of N110, although affected by the DNA, adopts the distinct XPA bound position upon XPA addition. These data strongly indicate the formation of the ternary ERCC1/DNA/XPA complex. DNA and XPA-binding sites are distinct, and the two interactions can happen simultaneously (B). Since the catalytic activity has been preserved in the human XPF protein through strong conservation of the nuclease signature (), the human nuclease fold should be identical to that of the archaeal counterparts. In that sense, human XPF nuclease and human ERCC1 central domains are expected to exhibit the same fold. Moreover, the ERCC1 and XPF HhH domains feature a common architecture and come into tight association exactly in the same way that the HhH domains of both archaeal species do (,,). Therefore, in structural terms the human ERCC1 and XPF proteins share the essential architectural subunits observed in the short XPF homodimer of . The structural similarities within the XPF family provide additional evidence for the proposed common origin of the human ERCC1 and XPF proteins (). Because ERCC1 is absent in archaea, this gene is thought to have been acquired from an ancient XPF gene duplication in the eukaryal lineage. The ERCC1 and XPF genes have subsequently evolved by the process called subfunctionalization (,). This model suggests that after gene duplication, both copies may be reciprocally preserved through the fixation of complementary loss-of-subfunction mutations, which results in a partitioning of the tasks of the ancestral gene. From the functions present in the ancestral XPF protein (archaea), human ERCC1 retained the canonical HhH domain that acts as a DNA-binding domain (), while XPF retained the catalytic activity (). Once the separation of the ancestral subfunctions occurred in ERCC1 and XPF genes, only the heterodimeric protein complex could restore the original function. Additionally, due to genetic alterations, the second helix-hairpin-helix motif of human XPF degenerated, yet the fold remained crucial for stabilizing the corresponding intact domain of human ERCC1 (). Similarly, the central domain of ERCC1 lost the catalytic activity by sequence drift, but despite adoption of a novel dual function (XPA interaction and DNA binding), the 3D fold was preserved. We suggest a mechanistic model for the heterodimeric function based on the model for the homodimeric archaeal homolog () (,). Human ERCC1 corresponds to the archaeal protomer that binds with the HhH domain to the upstream duplex. Human XPF corresponds to the archaeal protomer that recognizes the downstream duplex. Since it cannot make contacts with the minor groove (), the specificity of the human heterodimer has shifted to splayed-arms substrates consisting of only one duplex. This does not exclude contribution of the XPF HhH domain to ssDNA interactions, as reported before (). We show here that the central domain of ERCC1 is also involved in ssDNA binding. Additional DNA interactions are likely for the nuclease of XPF, analogous to the archaeal case (). Furthermore, ssDNA binding by cERCC1 will be stimulated by specific interactions with XPA, already present in the NER pre-incision complex and possibly bound to the DNA via its own DNA-binding site. Most importantly, through these coordinated interactions of the ERCC1 domains the XPF protein will be positioned to cleave the 3′ protruding strand (limon), thereby retaining the polarity present in the archaeal homodimer. Our model underlines the significant role of ERCC1 in the context of the full-length heterodimer. XPF is the catalytic module but the ERCC1 domains guarantee that the enzymatic activity is targeted properly. The presence of multiple distinct DNA-binding surfaces within the ERCC1/XPF/XPA/RPA repair protein intermediate coordinates cleavage to occur only when the DNA damage is recognized correctly by the NER machinery. The duplication of the ancestral XPF gene within the eukaryal kingdom resulted in an obligate heterodimer through loss of function (with an altered HhH domain of XPF and a degenerated catalytic domain of ERCC1) and adoption of novel functions (ssDNA and XPA binding of cERCC1). This permitted additional quality control mechanisms through a more complicated molecular interaction network, mediated by the novel functional domains, and thereby improving the fidelity of DNA damage repair. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

DNA replication is a key step of cell cycle that ensures the complete duplication of genomic DNA prior to mitosis. Over the past 40 years, it has been evidenced that eukaryotic genomes replicate accordingly to an invariant temporal order (,). This has first been shown in the Myxomycete . Indeed, taking advantage of the natural synchrony of several million nuclei within a single plasmodium, the authors have carried out pulse-labeling experiments and showed that sub-fractions of replicating DNA are the same through successive S phases (,). More recently, the visualization of labeled replication foci within single cells strongly suggested that replicons remain associated within the same clusters throughout consecutive cell cycles (). Cytogenetic analyses of metaphase chromosomes also showed an invariant pattern of replication banding () and density shift experiments validated these results at the level of individual genes by defining their timing of replication (). In addition, replication timing and transcriptional status of genes have been correlated in many organisms. Indeed, active genes are often found to replicate early whereas inactive genes replicate later (,). Genome-wide analysis in human cells and in confirmed the connection between early replication timing and transcriptional activity (). However, this link is more obvious for large domains rather than at a small scale () and was not seen at all in budding yeast (). It was also shown that the temporal program of gene replication could change during cell differentiation or development, reinforcing therefore the concept of a co-ordination between replication and transcription (). Studies of the profilin genes in and the immunoglobin heavy chain locus in mammalian cells have clearly demonstrated that, during differentiation, replication of these loci is altered by a change in the pattern of origin activation (,). Nonetheless, recent reports suggest that replication timing is not strictly defined. Indeed, in mammalian cells, molecular combing of DNA molecules associated with FISH analyses showed that redundant origins fired randomly with no timing preference (). Stochastic firing of origins was also described in fission yeast (), although it was not confirmed by genome-wide analyses (). In human cancer cells, studies of chromosomes 21–22 replication using micro-arrays analyses and FISH have demonstrated that a fixed timing of replication could not be assigned to large set of DNA sequences (i.e. they were found to replicate early as well as late). This led the authors to propose a ‘pan-S-phase’ pattern as opposed to the classical fixed pattern of replication timing (). We have previously demonstrated by incorporation of bromodeoxyuridine that active genes are replicated early in the naturally synchronous plasmodium of (). However, we also found that the highly expressed gene is late replicated in plasmodia (,). Here, we used neutral bidimensional agarose gel electrophoresis (2D-gel) method () to determine whether gene late replication comes from its association to a late firing origin or from its long distance from an early firing origin. Surprisingly, replication forks were found on the locus at the onset of S phase and could be detected through half of S phase. We demonstrated that this observation could not be explained by random replication timing among nuclei of a plasmodium but rather by a very slow progression of the forks enhanced by fork stalling upstream the gene. Importantly, we also showed that the coding region of is actually early replicated because of its proximity to a replication origin activated at the onset of S phase. Here again, active transcription is thus related to early replication. Furthermore, our results also demonstrated that the origin is developmentally regulated in correlation with the gene activity and reinforced the concept of replication-transcription coupling in . We used M3CIV and TU291 strains of plasmodia. They were routinely grown in shaken liquid cultures as multinucleated microplasmodia that are not synchronous to each other. Five centimeter diameter synchronous plasmodia were obtained by coalescence of microplasmodia as previously described (). As plasmodium nuclei lack for G1 phase, monitoring of the 3 h S phase was made by mitosis detection on smears observed under phase contrast microscope. We used plasmodia after the second or third mitosis, indifferently. Once at the stage of interest, plasmodia were harvested and frozen in liquid nitrogen. We used LU352 strain of amoebae that were grown as described (). Nuclei were isolated from one plasmodium as previously described (). Nuclei were fixed with three volumes of ethanol and stored at −20°C. The number of nuclei within the sample was evaluated by measuring optic density (260 nm) of nuclei aliquot after lysis with 2 M NaCl, 5 M urea (). Typically, 10 nuclei were washed twice in isolation medium, then digested with 0.1 mg RNaseA for 30 min at 37°C and stained with 0.1 mg propidium iodide for 30 min at 37°C. Internal control for overlaying the curves were carried out in duplicate experiments in which isolated nuclei from different cell cycle stages were mixed during the washing step. Samples were analyzed with a FACSortTM (BD Biosciences, San Jose, CA, USA). For hydroxyurea (HU) treatment experiments, one half of the plasmodium was placed on 2 ml culture medium as control, while the other half was placed on 2 ml culture medium supplemented with 50 mM HU. The targeting of the drug in nuclei was estimated to ∼15 min, thus treatment durations reported in the text and should be supplemented of 15 min to get the actual treatment duration. For instance, by placing the plasmodium on HU medium from + 15 to + 60 min after the beginning of S phase, the drug effect was estimated to ∼30 min, from + 30 min to + 60 min. Macroplasmodial DNA was obtained from isolated nuclei and was embedded in agarose plugs as previously described (). Microplasmodia were pelleted (500 g, 5 min) and their nuclei were isolated (). Pellet of nuclei was resuspended in 50 mM Tris [pH 8], 50 mM NaCl, 25 mM EDTA, lysed with 1% sarkosyl and digested with proteinase K (200 μg/ml) overnight at 45°C. CsCl and ethidium bromide were added at a final concentration of 915 mg/ml and 1 μg/ml, respectively. The gradient was centrifuged 6 h at 70 000 rpm at 20°C with a Beckman NVT90 rotor in a Beckman ultracentrifuge LE-80. The DNA band was withdrawn with a syringe and dialyzed against 10 mM Tris [pH 8], 1 mM EDTA (TE) at 4°C during 3 days. Amoebal DNA was obtained from isolated nuclei and purified on CsCl equilibrium gradients as described (). 2D-gel analyses were performed as previously described (). Digestion of DNA before the second dimension was adapted from (). Briefly, after the first dimension, the lane of interest was sliced off and rinsed twice in 10 mM Tris [pH 8], 0.1 mM EDTA. The DNA was digested with 3000 U of restriction enzyme overnight, followed by two additional incubations with 2000 U during 2 h. The lane was then rinsed with TE and with electrophoresis buffer before inclusion in an agarose gel for the second dimension. DNA from plasmodia were denatured and analyzed on alkaline gel as previously described (), except that alkali treatment was made on the agarose plugs. After electrophoreses, agarose gels were transferred onto a nylon membrane (Gene Screen Plus, Perkin Elmer) (). RNA was extracted from plasmodia by solubilization in guanidium isothiocyanate and centrifugation onto a CsCl cushion as described (). RNA samples were analyzed by northern blot as previously described (). probe derived from a 720 bp EcoRI-PstI fragment corresponding to the partial 5′ end of cDNA (accession number X64708, nucleotides 9–728). probe is the 960 bp PvuII-PstI fragment derived from a genomic clone (accession number M38038, nucleotides 1358–2318). probe is the 979 bp HindIII-XhoI fragment derived from a genomic clone (accession number X07792, nucleotides 20 to 999). Agarose gel purified fragments were [α-P]dCTP labeled by random priming with Radprime kit (Invitrogen). Hybridizations were performed in Church buffer (0.5M sodium phosphate buffer [pH 7.2], 1mM EDTA, 1% bovine serum albumin, 7% sodium dodecyl sulfate) at 65°C overnight (). The membranes were prehybridized for 1 h in Church buffer and hybridizations were initiated by adding heat-denatured probe and 0.1 mg/ml heat-denatured salmon testes DNA. Washes were performed at 65°C in five successive bathes of 40 mM phosphate buffer [pH 7.2], 1 mM EDTA and 1% sodium dodecyl sulfate. Hybridization signals were obtained and quantified by storage phosphor imaging (Molecular Dynamics 400A) and ImageQuant software. To follow the replication of gene encoding a subtilisin-like protease, we carried out kinetic analyses. Plasmodia were harvested at specific time points through the 3 h of S phase. DNA samples embedded in agarose plugs were digested by restriction endonucleases, resolved in 2D-gel electrophoresis and hybridized with a specific cDNA probe. Surprisingly, in disagreement with our previous reports (,), we found that the 6.7-kb EcoRV-EcoRI fragment encompassing gene exhibited prominent Replication Intermediates (RIs) during the first hour of S phase (A). A transition from a bubble arc to a Y arc was observed 5 min after the beginning of S phase (+5′) and indicated that initiation takes place within the fragment (see subsequently). At + 10 min (+10′), the RIs were essentially composed of Y-shaped molecules. This partial Y arc persisted up to + 60 min and was significantly detected until + 90 min, when about 75% of genomic DNA synthesis is completed. Quantifying hybridization signals evidenced the broad temporal window of locus replication. Indeed, replicative arcs represented ∼20% of the total hybridization signal from + 5 min until + 60 min, decreased to ∼8% at + 90 min and lowered to <1% as late as + 120 min. To rule out that this large temporal window of replication is due to a lack of synchrony of our plasmodia, we carried out flow cytometry analyses (B). Nuclei were isolated at specific time points of the cell cycle and DNA content of each population of nuclei was measured after nucleic acid staining with propidium iodide. B shows that each population exhibited homogeneous pattern of DNA content and that this latter increased synchronously from 2C to 4C as nuclei progressed in S phase. Therefore the synchrony of nuclei within a plasmodium is a property of the whole S phase. Moreover, in previous studies we were able to pinpoint the replication timing of single copy DNA sequences within a 5–10 min period during S phase (,,,). Therefore, as an internal control of DNA samples used for analysis of replication kinetics, we re-hybridized the same blots with a probe derived from gene that replicates at the onset of S phase (). We detected in the 4.8 kb EcoRI fragment containing gene RIs at +5 min, with a level of about 65% of the hybridization signal (C). This demonstrated that much more molecules containing gene were engaged in replication than in the case of gene at this time point. In contrast, only a faint signal about (or “∼”) (3%) was detectable at +10 min and no replicative signal could be detected later on, in agreement with reported results (). The size difference between and containing restriction fragments could not explain these contrasted replication patterns. Therefore, direct comparison of the two loci indicated radically different temporal windows of replication: gene is replicated in less than 10 min whereas it takes ∼90 min to replicate the gene-containing fragment. We also detected X-shaped molecule signals for both loci (see open arrowheads in A and C) after the forks have reached both ends of the fragment (i.e. from +25 min to +60 min for and from +10 min to +40 min for ). Such molecules correspond to transient post-replicative joint DNA molecules involving sister chromatids (). These X-shaped molecules had a maximum of intensity at +10 min for and +60 min for . The delay in X-DNA apparition in the gene-containing fragment is consistent with a later period of replication. We observed that, in contrast to and other loci (,,,), the timing of replication of locus is extended. This unexpected long period of replication of locus may be explained either by slow progression of replication forks or by different replication patterns among the millions of nuclei contained within a single plasmodium. To distinguish between these possibilities, we first wanted to map the replication origin of the gene-containing replicon (A) and to determine the fork position on the locus early in S phase. We thus performed a series of 2D-gels to analyze different restriction fragments of DNA extracted 5 min after the onset of S phase. After probing with probe, we compared RI patterns in overlapping fragments for deducing the localization of replication origin (A). We found a bubble arc in fragment , and a bubble to Y arc transition in fragments -, indicating the firing of a bidirectional replication origin in these fragments. We located the origin at the middle of fragment , since it exhibited the most developed bubble arc (see the schematic extending bubble above the map in ). Consistently, the extent of the Y arc was more important in fragments and in which the origin would be less centered. Only Y arcs were detected in fragments , and , in agreement with an outside position of the origin. In the case of fragment , due to the origin position close to its extremity and due to its size, no bubble arc could be detected and only a nascent Y arc was revealed. These results are consistent with an origin positioned at the 5′ side of the gene. The observation of a partial Y arc when analyzing the origin-containing fragment strongly suggests that the origin is efficiently fired. Otherwise, a complete Y arc would be observed in addition to the bubble arc, as a result of passive replication of the locus in some nuclei (). In order to follow the progression of replication forks, we compared the RI patterns shown in A with those obtained from similar analyses performed with DNA prepared at +10 min (B). A slow evolution of fork distribution was revealed by the different mean positions of RIs along the replicative arcs. Indeed, at +10 min, for fragment , the maximal density of RIs was found at the end of the bubble arc. We also detected a faint terminal portion of a Y arc. For fragment , the bubble arc was then essentially converted in a Y arc, as a result of fork movement within the fragment. Similarly the major position of RIs had moved along the Y arc for fragments and , revealing the homogeneous displacement of replication forks. Knowing the origin position in a context of apparently smooth velocity for both forks, we could deduce which replication fork came out first of the restriction fragment. The downstream position of the origin in fragment implied that the rightward moving fork reached first the end. Importantly, as at +10 min the bubble arc had almost disappeared in fragment , and as fragments and were almost free of replication forks, we thus conclude that gene is replicated in early S phase. At +25 min, the fork movement was again evidenced by a change of RI mean position in the 2D-gels (B). Interestingly for fragment , we also detected a spot (star) close to the intersection of the Y arc with the diagonal of linear molecules that corresponds to accumulation of RIs of a 2X size, like observed in the kinetics (A). This pattern differed from those obtained at +5 and +10 min where steady progression of RIs along the bubble and the Y arc could be detected. Such RI accumulation at +25 min indicated a stalling of the leftward moving fork close to the upstream EcoRV site (see the striped rectangle above the map). This stalling did not correspond to an arrest of the fork but rather to a slowing down. Indeed, the spot marked by the star for fragment analysis spread on most of terminal portion of the Y arc for the shorter fragment . The accumulation of RIs at the apex of the Y arc for fragment confirmed the fork stalling and allowed to map pausing at the middle of the fragment. We also noticed that replication forks go through the stalling region since RIs were found on the last part of the Y arc for fragment . Thus, in agreement with kinetic analyses shown in A, we observed a low mobility of replication forks through 21 kb surrounding gene (). The slow removal of replication forks from the restriction fragments is enhanced by fork stalling upstream the gene, while the coding region is rapidly replicated. Our results also indicated that, within a plasmodium, the collection of replication forks progresses concomitantly at locus and argued against a randomly timed replication. We have shown that a replication origin close to gene is fired at the onset of S phase; however, we still observed on kinetics a particularly prominent Y arc (∼20% of the signal) at +25 min after the onset of S phase (A and B). The fact that this Y arc was never completed strongly suggested the conversion of bubble-containing fragments to single fork-containing fragments as a result of fork progression. To further confirm our assumptions and to rule out that these RIs originated from other origins activated in the vicinity of locus, we determined the direction of replication fork movement at locus in early S phase by using an adaptation of the 2D-gel method (). In this approach, DNA contained in the agarose lane from the first dimension was digested again before it was submitted to the second electrophoresis. The resulting RI patterns depend on the polarity of replication forks in the shortened fragments (). We used a plasmodium at +20 min in S phase, when a strong intensity of RI signals was found. Following HindIII digestion, three restriction fragments were obtained, two of them (H2 and H2*) resulting from a restriction fragment length polymorphism. We observed Y arcs for the three fragments (see control experiment in ), with an accumulation of RIs at the apex of the Y arc for H1 fragment. This corresponds to the stalling of replication forks close to the EcoRV site at this stage of S phase (B). The second digestion in the gel was performed with EcoRV. For each resulting fragment, we found only one derived pattern (see fork polarity experiment and interpretative scheme in ). For the upstream fragment H1-RV, a faint bubble arc and the end of a Y arc were observed, which implied that forks moved leftward. The spot appearing on the left at a size of 2X corresponds to forks stalling close to the upstream EcoRV site, since digestion with EcoRV converted stalled forks into linear fragments. Rightward moving forks replicated the downstream polymorphic fragments H2. At this stage of S phase, part of the RIs has gone beyond EcoRV site so that the resulting fragments were linear. Shorter RIs formed the vertical end of the Y arc originating from H2-RV 1X spot. These results showed that fork directions are identical among the population of nuclei and also that replication forks have the same polarity for both alleles. Importantly, fork direction is not the same in the HindIII fragments: in the H1 upstream fragment we detected leftward moving forks, while in the H2-H2* downstream ones we detected rightward moving forks (see the arrowheads under the map in ). This fork divergence confirms the presence of a replication origin coinciding with the 5′ region of the gene. Therefore this analysis reinforces our previous data and rules out the possibility of a distant origin whose firing would produce forks reaching the EcoRI-EcoRV fragment 20 min after the onset of S phase. Since we have shown that the persistence of RIs could not be the consequence of multiple initiation events among the nuclei, it was most likely due to a slow elongation rate of replication forks. To test this hypothesis, we measured the elongation rate of the replicon by analyzing the growing of nascent strands by alkaline gel electrophoresis (A). The natural synchrony of the plasmodium allows detection of the nascent strands of a single-copy replicon (). After probing with cDNA, short single stranded RI (stars) was observed from stages +7 to +40 min (A). At later stages, our electrophoresis procedure did not allow their separation from parental DNA. Although RIs were seen on 2D-gel at +5 min (A), they were not detected by alkaline gel electrophoresis at this stage due to lesser sensitivity of the latter method. We measured the mean size of RIs to calculate the mean rate of replication fork progression (C). Interestingly, the small size of RIs at earliest stages confirmed that the replication origin is close to the coding region. Furthermore, we measured a slow increase from 4 kb at stage +7 min to 17 kb at stage +40 min, that corresponds to an average rate of 0.4 kb/min/replicon. Comparison with the rate of elongation of replicon was performed by re-hybridization of the blot with gene (B). A stronger signal was obtained for nascent strands especially at earlier stages, suggesting a more acute firing of origin. nascent strands were detected slightly earlier, at +5 min, and had a size of 4 kb. The largest nascent strands that could be separated from parental DNA in these conditions were seen at +25 min with a size of 22 kb. Plotting the nascent strand lengths against time in S phase (C) revealed an average rate of 0.9 kb/min/replicon for replicon. Although the rate of elongation of replicon has been evaluated to be more than twice greater than the one of replicon, this value is in agreement with the canonical mean of 1.2 kb/min/replicon that has been calculated for (). We thus conclude that replicon is characterized by an unusually slow progression of replication forks. We also compared these data with the evaluation of fork rates obtained from our 2D-gel analyses of locus (). Indeed, the mean position of the signal on the arc of RIs indicated the mean location of forks within the fragment of interest. As indicated in D, alkaline gel and 2D-gel analyses gave consistent results. Both forks had covered each 2 kb as a mean after 5 min (i.e. at a speed of 0.4 kb/min/fork) and 3.0 kb after 10 min (i.e. 0.3 kb/min/fork). However, after 25 min the rightward moving fork had progressed over 6 kb (i.e. 0.24 kb/min/fork) whereas the leftward moving fork had progressed over less than 4 kb (i.e. 0.15 kb/min/fork) likely due to the stalling. Thus, the replication forks have an unequal rate. The accumulation persisted up to +60 min and implied that the replicon elongation is mostly unidirectional during this period. In order to evaluate the life span of RIs during the whole S phase, we studied RI pattern in an asynchronous nuclei population prepared from microplasmodia where all replication events were represented. Our aim was to compare intensities of and signals. We reasoned that, if the life span of RIs was longer than the one of RIs, we would expect stronger signal intensity for RIs since they were present during a longer period of S phase. On the opposite, a stochastic replication of locus at a normal rate would not give any difference between signal intensities for and loci in an asynchronous population. shows a comparison of the replication pattern of and loci obtained by 2D-gel analysis from DNA extracted from the same culture of exponentially growing microplasmodia. We could see on a single 2D-gel all the RIs appearing at any moment of S phase, in addition to the prominent 1X spot of non-replicating molecules. A similar transition from a bubble arc to a Y arc was detected for both genes. However, we obtained about 3–4-fold more RIs at locus, as compared to locus, meaning that the RIs life span is longer (the experiment was repeated four times). Clearly, this asynchronous population of nuclei, like plasmodium nuclei, exhibited a not fully expanded Y arc, demonstrating the efficient activation of the -linked origin. The higher intensity found for as compared to locus rules out the hypothesis of a random replication timing of the locus. To check that replication forks at locus are moving forks, we inhibited DNA replication with HU, a drug that prevents replication fork elongation. Drug treatments were performed at successive periods of the S phase on half of each plasmodium and the other half was used as a control. We tested fork movement from the onset of S phase up to +25, 15–60, 60–90 and 120–150 min (). 2D-gel analyses revealed delays of RI patterns for treated plasmodia as compared to the control, except for the latest period of treatment (120–150 min). Therefore, fork progression was impeded by drug treatment. Note that a bubble arc was still observed after a 60–90 min treatment, indicating that, in a non-negligible part of nuclei, both replication forks were active within the EcoRI-EcoRV fragment at least at +60 min. These results underline the delayed activation of the origin in a small proportion of nuclei, which is consistent with the faint bubble arc seen until +60 min on . We conclude that, despite the long kinetics of elongation of replicon, we detected on 2D-gels moving forks since they were sensitive to HU treatment. Therefore the slow replication of locus is due to slow progression of replication forks rather than arrests randomly distributed along the replicon or fork collapsing. The gene has been previously described as developmentally regulated during the two alternative stages of growth, the diploid multinucleated plasmodium and the asynchronous haploid uninucleated amoebae (). We used northern blot analysis to compare the steady state level of mRNA in our strains of plasmodia and amoebae. We detected with probe an abundant 1070 nt mRNA in total RNA from plasmodium (Pl in A). In contrast, a weak signal could be detected in the amoebae sample only upon much longer exposure (Am’ in A). This indicated that gene is highly expressed in the plasmodium. On the opposite, expression of gene is almost extinguished in amoebae. Quantification and normalization against the constitutively expressed actinC gene mRNA (upper band in A, Pl and Am) indicates a 1 to 500 ratio of mRNA abundance between these two stages, confirming the developmental regulation of gene expression. Therefore, considering our previous results, showing a variation of origin usage in the case of developmentally regulated profilin genes and (), we addressed the question of the origin usage during development. We compared the replication pattern of the gene in plasmodia and amoebae by 2D-gel analysis (B). In plasmodium, the 8.2 kb EcoRV fragment is replicated from an internal origin firing in early S phase and located at the center of the fragment, as deduced from the bubble to Y arc transition observed at +10 min (see above and B). In contrast, DNA prepared from amoebae exhibited only Y arcs when the same restriction fragment was analyzed (B). We previously showed that it is possible to detect a site-specific origin in this cell-type (). Therefore, these data demonstrated that the replication origin evidenced in the promoter region of gene in the plasmodium is inactive in the amoebae. We thus conclude that replicon is developmentally regulated and that usage of the origin upstream the coding region is correlated with transcriptional activity of the gene. We found that gene was early replicated from a bidirectional origin located in its promoter region (). However, the 2D-gel analysis of locus also showed a surprisingly long life span of RIs that persisted for half of the S phase (). It should be noticed that these kinetic data were obtained with plasmodia harvested in two consecutive cell cycles, showing the invariance of this feature over S phases. This unusual pattern can be explained by a slow progression of replication forks on the locus in early S phase ( and ) and also by a stalling of the leftward replication fork from +25 min to +60 min (). A slowing down of replication forks has been also found in upstream rRNA genes and downstream the histone H4-1 gene (,). In this latter case, it has been shown that forks are stalled in DNaseI hypersensitive regions (). It is also possible that the DNA sequence of locus might impede the replication forks. Indeed, particular sequence patterns such as trinucleotide repeats might reduce fork progression (). In fact, several examples of replication stalling have been described in eukaryotic cells. For instance, in , almost 1500 discrete sites were found to correspond to pauses caused by DNA sequences or by local protein-DNA complexes (). These sites include tRNA genes (), rDNA (), centromeric and subtelomeric regions (,). Clearly, variation of fork rate is not a rare event and DNA replication does not seem to be a steady process. RIs were found in restricted DNA fragments from plasmodia harvested in early S phase. Our previous studies showed that gene is contained in a late replicating DNA fragment (,). This discrepancy is explained by the usage of different methods of analyses. In a first study, the replication timing of locus had been analyzed by density-shift experiment following incorporation of bromo-deoxyuridine (). The downstream allelic 6 kb and 12 kb HindIII fragments had been studied and were found clearly enriched in the heavy-light DNA fraction only after 90 min in S phase. Gene dosage analyses had confirmed these density shift experiments: the same HindIII fragments were found at 2 copies per genome only after 90 min. This led us to conclude at a late replicating locus (). In the present work, 2D-gel technique favored detection of low amounts of RIs and allowed analyses of fork progression in a synchronous system. It revealed an ongoing and slow replication of locus during most of the first half of S phase ( and ). As a result, completion of replication of -containing HindIII fragments is achieved late in S phase. In the same vein, if the locus had been analyzed with microarrays, like DNA sequences from human chromosomes 21 and 22 (), it is likely that it would be found to replicate both with early and late DNA and would be considered as a ‘pan S phase’ replicating locus. The lack of strict timing of replication at locus raises the question of homogeneity of temporal order of replication among the millions of nuclei contained within a plasmodium. We argued that a random replication of locus would give a similar intensity of RI signal towards 1X signal at locus and locus in an asynchronous population of microplasmodia. It is not the case, so that a long life span of RIs is more consistent with our data (). In addition, homogeneous replication fork progression was evidenced by the analyses of RIs obtained from DNA at specific time points of S phase ( and ). Therefore, our results rule out random replication timing among the plasmodium nuclei. In contrast, they illustrate a local variation of replication timing within a replicon, since we demonstrated that the gene replicated early and surrounding DNA sequences replicated later. Replication timing of DNA sequences depends also upon how and when replication initiation occurs. The question of the nature of eukaryotic origins has been largely debated and two models seem to emerge: either specific origins fire at a fixed timing with a given efficiency, like it has been described in budding yeast, or redundant origins fire stochastically, a model that correlates with many observations made in metazoan cells (). Thus in egg extracts, random initiation has been described and it has been proposed that the frequency of initiation events increases during the S phase in order to ensure the completion of genome duplication (). In our mapping experiments at locus, transitions from a bubble to a Y arc were only detected in DNA fragments the most centered on the promoter region of gene (), which clearly demonstrated that replication initiates at a fixed origin linked to gene. Moreover, we did not find a mixture of bubble arc and complete Y arc throughout the S phase (), indicating that the origin activation is efficient. Accordingly, the bubble arc signal did not result from a rare event, since it has also been detected in microplasmodia despite their asynchrony (). Moreover, only the terminal portion of the Y arc was detected when analyzing this asynchronous population, showing the efficiency of origin firing. We also checked that no other origin was activated elsewhere within this locus in early S phase by determining replication fork directions (). In addition, 2D-gel analyses of overlapping restriction fragments spanning 21 kb around the gene did not show other initiation events or termination during S phase (; data not shown). These results argue for an efficient activation of a localized origin. Such origins have been described before in (,,), indicating that stochastic firing is not the rule in this organism. However, we detected a faint bubble arc signal from +10 min to +60 min (). HU treatment from +60 to +90 min showed that forks forming this bubble arc were still active at +60 min since the drug treatment delayed replication pattern (). These observations can be related to the low intensity of replication signals at the beginning of S phase, following 2D-gel and denaturing gel analyses ( and ). Altogether, these results suggest a delayed activation of the origin in a small part of the nuclei contained within a plasmodium or they reflect different replication patterns of the two alleles contained within each nucleus. Although our previous studies have clearly shown a concerted activation of allelic origins at other loci (,), such a different replication pattern between two alleles has been already described in (). From a gene dosage analysis, the authors found that the 2 allelic B1 and B2 alpha-tubulin loci replicate synchronously in early S phase, while A locus replicates later. Remarkably, A2 allele replicates in a prolonged period of mid-S phase and asynchronously from A1 allele, which replicates earlier. In this view, for locus, we can hypothesize simultaneous early activation of one allelic origin in all nuclei, while the other is activated progressively throughout the first hour of the S phase. Such distinct patterns of replication of the alleles were not obvious on 2D-gel. However individual quantification of allelic RI signals is not significant when allelic fragments are of a similar size. Furthermore, we did not find a restriction fragment length polymorphism that would allow unambiguous distinguishing of the replication timing of the two alleles. In these conditions, it is not clear whether alleles are replicating exactly synchronously or not. Nonetheless a delayed activation of origin certainly occurs in a non-negligible number of molecules. origin is developmentally regulated and origin activation correlates with transcriptional activity (). Such a modulation of origin firing has been previously reported for A and P loci in () and has also been described in other organisms (,). These observations indicate that eukaryotic origins are at least in part epigenetically defined and suggest a strong correlation between replication and transcription. This relationship has been previously observed on chromatin spreads from early S phase plasmodia: electron microscope investigation showed a tight linkage between active genes and early firing origins (). At the level of individual genes, we confirmed by 2D-gel mapping that efficient early firing origins are situated in the vicinity of abundantly transcribed genes. This was demonstrated for the constitutively expressed B and C actin genes, the developmentally regulated P profilin gene and the cell cycle regulated H4-1 and H4-2 histone genes (,,). In contrast, studies of weakly expressed genes revealed that they are replicated with different patterns. Inactive A profilin gene is passively replicated in mid-S phase (,). The weakly expressed B and E and topoisomerase II genes are replicated early in S phase since they are embedded in a cluster of early-activated replicons [(), unpublished data]. Yet, in these cases, the genes are not coincident with an origin but with a termination site. Finally, the A gene contains a replication origin in the promoter region, but this origin inefficiently fires in a large temporal window of mid-S phase (). Therefore, the association of an efficient early origin with a transcriptional promoter might be a unique property of highly expressed genes in . It remains to be determined whether the origins surrounding B, E and topoisomerase II genes could be coincident with active genes. Likewise, the transcriptional status of the locus is unknown; it would be of interest to investigate it in the region where replication forks are stalling. In metazoan, co-localization of active genes and origins has been often found and suggests that replication and transcription may share common regulation, perhaps as chromatin domain units (,). Several examples of origin specification in relation with transcription have been described at various loci like rRNA genes in embryos (), DHFR locus in hamster cells () and Hox genes in mouse cells (). In addition, deletion of DHFR promoter results in a modification of replication initiation activity at this locus (). This again suggests a coupling of these two nuclear activities. In this light, we propose that the replication of highly expressed genes is strictly regulated in . This tight control would involve the location of the active genes close to very early firing replication origins. Lower expressed loci or non-coding regions would be under a more relax control, so that the replication timing would be less defined or the origin efficiency would be reduced. The replication organization of locus may illustrate a transition in replication control stringency related with transcription level.

Diamond–Blackfan anemia (DBA) is a rare congenital disorder characterized by the defective differentiation of pro-erythroblasts, the precursors of red blood cells. Patients suffer severe anemia and display heterogeneous clinical features including malformations, growth failure and predisposition to cancer (,). Linkage analysis has revealed that a quarter of all DBA reported cases are connected to the heterozygous mutation of the gene encoding the ribosomal protein RPS19 (,). The RPS19 protein is a component of the 40S ribosomal subunit and belongs to a family of ribosomal proteins restricted to eukaryotes and archea. It is essential for yeast viability and for early stages of development in mice (,). Disruption as well as point mutations of the gene in yeast and human cells affect maturation of the pre-ribosomal RNA (pre-rRNA) and block production of the 40S ribosomal subunits (,). Why the mutation of a ribosomal protein primarily affects pro-erythroblast differentiation remains a central question. However, recent linkage of a two other ribosomal protein genes, and , to DBA (,) strongly supports the hypothesis that DBA is the consequence of a ribosomal disorder (,). Over 60 different mutations affecting the gene have been reported, including deletions, insertions, frameshifts, premature stop codons and missense mutations (). Some mutations, like very early stop codons or modification of the promoter clearly result in RPS19 haplodeficiency by hampering synthesis of RPS19 from the mutated allele. However, for more subtle mutations like missense mutations, the question arises as to whether they affect the folding of the protein or whether they are milder mutations affecting the function while preserving the overall fold. Since there is no homolog of RPS19 in bacteria, for which high-resolution structures of the small ribosomal subunit are available, the structure of RPS19 and its precise location within the 40S subunit remain unknown. The crystal structure of RPS19 from presented herein fills this gap and provides a rationale for the impact of RPS19 mutations in DBA. RPS19 cDNA was cloned into a pET-15b (Novagen) modified plasmid. The expression was carried out in BL21 (DE3) Rosetta cells (Novagen) at 15°C. Bacterial cells were sonicated and centrifuged for 30 min at 50 000 . The supernatant was heated up for 20 min at 50°C and centrifuged for another 30 min at 50 000 . After incubation and elution from the cobalt-affinity resin, the tag was cleaved from the protein by an overnight digest with 1/200 (w/w) ratio with TEV. A step of purification on Hi-S (Pharmacia) was carried out and the protein was eluted at about 600 mM NaCl. The protein was concentrated to up to 10 mg/ml in 50 mM Tris–HCl pH 7.5, 600 mM NaCl. Crystals were obtained at 20°C by the hanging drop vapor diffusion method by mixing equal amounts of the protein solution and of a reservoir composed of 30–36% PEG 2000 MME, 100 mM Tris–HCl pH 6.8–7.5 over a couple days to a size of 50 × 200 × 200 microns. They diffracted to 1.15 Å on synchrotron beamline and belonged to the space group 222 with cell dimensions = 32 Å, = 57 Å, = 82 Å and contained one molecule per asymmetric unit and 48% solvent. Native and derivative data were collected at the ESRF synchrotron and processed with XDS (). Data collection statistics are shown in . A HgBr2 derivative resulting from a 12-h soak in 2.5% (v/v) HgBr2 (prepared from a saturated solution) was collected at the LIII-edge of Hg. Derivative data set was combined with the most isomorphous native data set, SHELXD and SHARP were used to locate and refine the heavy atom site position (,). The resulting phases had an overall figure of merit (FOM) of 0.33 at a resolution of 2.0 Å. After solvent flattening with Solomon (SHARP), automatic building was carried out with ARP/wARP (). Out of a total of 150 residues, 125 were initially placed. Further building and refinement cycles were carried out with Coot and REFMAC (,). Last cycles of refinement were carried out including hydrogens and using individual anisotropic B factors. The final model has a good stereochemistry with an R-free value of 15.5% and a R-factor value of 13.7% (). Site-directed mutagenesis on yeast RPS19 was performed by PCR. For complementation experiments, the mutated alleles were subcloned into vector pFL38-Ps15 (URA3), downstream of the constitutive RPS15 promoter. The resulting plasmids were introduced in the Δ GAL-RPS19 strain (). Cells were cultured in liquid synthetic medium containing galactose, but no uracil, and spotted at different densities on agar plates using the same medium, with either galactose or glucose as the carbon source to modulate RPS19 expression. To evaluate incorporation of wild-type or mutated RPS19A into ribosomes, the corresponding open reading frames were also cloned in frame downstream the TAP tag coding sequence in the pFL38-Ps15 vector and expressed in Euroscarf strain Y06271 (::, RPS19B). Whole cell extracts were fractionated by ultracentrifugation on sucrose gradient for ribosome analysis. Two hundred milliliters of yeast culture were grown in YPD medium to an OD of 0.5 and cycloheximide was added at a final concentration of 100 μg/ml. After 10 min incubation, yeast cells were harvested by centrifugation at 5000 r.p.m and washed in 20 ml ice-cold buffer A [20 mM HEPES (pH 7.5), 10 mM KCl, 5 mM MgCl2, 1 mM EGTA, 1 mM DTT and 100 μg/ml cycloheximide]. Cells were broken with glass beads and re-suspended in 150 μl buffer A. The suspension was clarified by centrifugation for 5 min at 10 000 r.p.m. An amount of the extract corresponding to 1 mg of protein was loaded on a 10.5 ml 10–50% sucrose gradient in buffer A without cycloheximide and centrifuged for 12 h at 26 000 r.p.m in a SW41 rotor. A gradient collector (ÄKTAprime—Amersham Biosciences), in combination with a Pharmacia UV-detector LKB.UV-M II, was used to record the UV profile. Twenty 0.5 ml fractions were collected and 150 μl of each fraction were slot-blotted on nitrocellulose membrane to detect TAP-tagged RPS19A. The membrane was then incubated with peroxidase anti-peroxidase complexes (Sigma), which were revealed by chemoluminescence (ECL, GE Healthcare). After scanning of the film to obtain a digital image, the labeling intensity for each fraction was quantified by densitometry using MetaMorph (Universal Imaging). The UV profiles were superimposed with the blot quantifications to obtain . Ribosomal RNAs were co-immunoprecipitated with RPS19-TAP and analysed as published previously (,). Pre-rRNAs were detected on northern blots with probe D-A2 (5′-GAAATCTCTCACCGTTTGGAATAGC-3′) and A2-A3 (5′-ATGAAAACTCCACAGTG-3′). The structure of RPS19 from (RPS19) was determined by SIRAS on a mercury derivative and refined against the best native data set. The final model has an R-free value of 15.5% () and comprises residues 2 to 150 with the exception of two disordered loops between residues 37–44 and 79–84 (a). The structure of RPS19 is almost entirely α-helical and folds around a five α-helix bundle. Knowing that human RPS19 (RPS19) shares 36% identity and 57% similarity with RPS19 (RPS19) sequence (c), we may assume that they display a similar fold. In the rest of the text, RPS19 residues in and are labeled ‘Pa’, ‘Sc’ and ‘Hs’ respectively (). Fourteen amino acids of RPS19 are the targets of missense mutations in DBA patients. Although these mutations are spread along the entire primary sequence (c), the structure of RPS19 shows that most missense mutations cluster within or around the α-helix 3 (a). This α-helix, made of residues 50–65 of RPS19 (52–67 in human numbering), corresponds to the mutation hot spot. It is located at a central position in the structure where it bridges α-helices 1 and 6 on one side with α-helices 4 and 5 on the other side (a). The apolar side of α-helix 3 is engaged into hydrophobic interactions with residues of the neighboring α-helices, thus forming the hydrophobic core of the protein (b). Strikingly, three mutations on α-helix 3 (A57P, A61S/E, L64P), two on α-helix 1 (V15F and L18P) and two on α-helix 6 (G127E and L131R) affect residues involved in this hydrophobic core (green residues on ). Thus, although distant on the primary sequence, these seven amino acids are functionally related and are involved in the folding and the stability of the protein. In contrast, mutations P47L, W52R, R56Q, S59F, R62W/Q, R101H and G120S affect residues located on the surface of RPS19 (red residues on ). Remarkably, these residues show a much higher degree of inter-species conservation than the amino acids within the hydrophobic core (c). These residues are located within two highly conserved basic patches at the surface of the protein (). On the polar face of α-helix 3, four exposed mutated residues (W52, R56, S59 and R62) form the floor of a central basic groove (patch A on a–c). In addition, residues from α-helices 4 and 5 and from the β-sheet define the conserved patch B (d–f). It comprises residues R99 (R101), K100 (R102), Q103 (Q105), K113 (K115), G118 (G120) and R119 (R121). DBA mutations R101H and G120S affect this conserved surface. Interestingly, both conserved surfaces have a positive electrostatic charge: patch A harbors a localized positive charge (c), whereas patch B is embedded in a larger positive area (f). Based on these observations, we propose to subdivide DBA mutations into two classes: class I encompasses structural residues affecting the folding of the protein; class II mutations affect surface residues and would impair the function of RPS19 without altering its overall fold. It should be noted that the Val13 (V15F) and W50 (W52R) are partially involved in Van der Waals contacts with neighboring atoms and are also partially exposed to the solvent. Class I mutations affecting protein folding are predicted to render the protein unstable. Indeed, when expressed in COS cells, mutant RPS19 V15F and G127E were shown to be present at low level, in contrast to class II mutants R56Q, R62W whose level was similar to that of wild-type RPS19 (,). Along this line, a recent report confirms and extends these results by showing that class I mutants V15F, L18P, A57P, A61S/E, G127E and L131P are rapidly degraded by the proteasome when expressed in mammalian cells, whereas class II mutants P47L, W52R, R56Q, R62W, R62Q, R101H, are G120S are more stable (). Thus, the classification proposed here perfectly correlates with the instability observed for class I mutants expressed in cultured cells. In order to test whether the conserved surface areas are essential for the function RPS19, we performed functional complementation in yeast. RPS19 is encoded by two genes, and . DBA-like mutations, as well as systematic R to E mutations that alter the charge of these basic conserved areas, were introduced by site-directed mutagenesis into the gene. The ability of these mutant alleles to support growth was evaluated in a Δ Δ mutant strain conditionally expressing the wild-type RPS19A under control of a galactose promoter. As shown in , DBA mutations R57Q, R63Q and R102H had a drastic effect on RPS19 capacity to support cell growth in glucose, whereas mutation R63W partially supported growth. Similarly, mutation to glutamic acid of R57, S60 and R63 in patch A and R102, R122 in patch B, completely abolished cell growth. In contrast, the double mutation R130E-R134E, outside of the conserved patches, had little impact on viability. These results show the crucial role of basic patches A and B in RPS19 function. As already observed with this assay (), among class I mutations, I65P (L64P) was not viable, I15F (V15F) strongly affected RPS19 function, whereas substitution A62S (A61S) had little impact on cell growth, which was also the case for class II mutation G121S (G120S). Although affecting highly conserved residues, these two latter mutations may have more critical effects in human than in yeast. In order to further characterize the loss-of-function associated with mutation of surface residues, we tested whether class II mutations affected RPS19 incorporation into pre-ribosomes. We thus analysed association of RPS19 mutants with pre-ribosomal RNAs in yeast. TAP tagged wild-type and mutant alleles of were expressed in place of the endogenous gene in a Δ strain. Production of RPS19 in this strain is insufficient and pre-rRNA processing is partially defective (), which mimics RPS19 haploinsufficiency in DBA patients. Using the complementation assay described in , we checked that the RPS19-TAP protein remained functional (data not shown). When captured on IgG Sepharose, the wild-type RPS19-TAP and the G121S mutant co-precipitated with the 18S and 25S rRNAs (b, lower panel), indicating incorporation into mature ribosomes (not dissociated under our experimental conditions). Detection of the precursors of the 18S rRNA, by northern blot analysis with the D-A2 probe, showed co-precipitation of the 20S pre-rRNA (a, upper panel), the direct precursor to the 18S rRNA (b). This indicates the presence of these proteins in pre-40S particles. The 32S and 35S pre-rRNAs, which are part of the early 90S pre-ribosomes, were undetectable or close to background in the precipitated fractions, depending on the experiments. Mutations in basic patch A (R57E, R57Q, R63E) or basic patch B (R102E, R122E) reduced the amount of co-precipitated mature 18S and 25S rRNAs to background level (b and c, lower panel). A similar result was obtained with class I mutation I15F. In addition, these proteins did not co-purify with 20S pre-rRNA (b). This result was not correlated with differences in protein expression levels of the various mutants (d). Cleavage of the 32S pre-rRNA in the early 90S pre-ribosomal particles normally yields the 20S and 27S-A2 pre-rRNAs, included in the pre-40S and pre-60S particles, respectively (a, left panel). Deficiency in RPS19 was shown to block cleavage at site A2 (); alternative cleavage at site A3 produces the 21S and 27S-A3 pre-rRNAs, as illustrated in a (right panel). We thus considered the hypothesis that RPS19 mutants were incorporated into ill-matured pre-40S particles containing 21S pre-rRNA instead of 20S. As expected, a significant increase in the amount of 21S pre-rRNA, paralleled by a drop in the 27S-A2 pre-rRNA level, was detected in cells expressing RPS19-TAP with mutations R57E, R63E, R102E or R122E (c, ‘input’). However, the 21S pre-rRNA did not co-precipitate with mutated RPS19-TAP (c, ‘I. P.’). Thus, mutation of the basic patches A and B prevents incorporation of RPS19 into pre-ribosomal particles. These data were fully consistent with the analysis of RPS19 association with ribosomes on sucrose gradients. As shown in a, the wild-type TAP-RPS19 protein was mostly present in the fractions containing the 40S subunits, the 80S ribosomes and the polysomes, indicating very efficient incorporation into mature subunits. In contrast, mutations in conserved patches A and B resulted in a large amount of free protein at the top of the gradient, with no preferential association with mature ribosomes (a,b). Noticeably, part of these mutated forms of RPS19 was present in high-density fractions. The A62S and G121S mutants, as well as double mutant R130E/R134E (outside of the conserved basic patches), which all support growth in RPS19 depleted yeast, showed an association profile very close to that of wild-type RPS19 (c). The I15F mutation, which does not totally abrogate RPS19 function in yeast, showed partial association with mature ribosomes (c). In conclusion, we propose that DBA missense mutations primarily result in RPS19 haplodeficiency, by impacting either its folding and its stability, or its capacity to engage intermolecular interactions. The DBA mutation hot spot appears as a key structural element of RPS19 involved both in the constitution of the hydrophobic core and in a highly conserved surface. Impairment of ribosome biogenesis resulting from RPS19 haplodeficiency may have two direct consequences. First, a low rate of small ribosomal subunit production may limit the cell translation capacity. Second, alteration of ribosome biogenesis, and subsequently of the nucleolus organization (), may be perceived by the cell as a ‘nucleolar stress’ and favor cell cycle arrest () These phenomena, alone or in combination, could prevent differentiation of pro-erythroblasts (,,). In addition, our data do not exclude a dominant negative effect of the missense mutants, upstream or aside of association with pre-ribosomes, especially in the case of the more stable class II mutants. Presence of some mutants in high-density fractions of the sucrose gradient suggests that they may be engaged in multi-molecular complexes, different from ribosomes, and whose composition remains to be determined. Establishing genotype/phenotype relationships has proved difficult in DBA and this remains true when considering the relative clinical impacts of class I and class II mutations. Availability of RPS19 crystal structure should be decisive to design new strategies to understand RPS19 function and its role in erythropoiesis.

The proto-oncogene encodes for a 145–160 kDa tyrosine kinase receptor, which is especially expressed in mast cells, melanocytes and hematopoetic stem cells (,). The tyrosine kinase domain of c-kit has become an important molecular target for the treatment of gastrointestinal stromal tumors (GIST), and the small molecule kinase inhibitor Gleevec has become the most significant therapy for GIST, where it has made a major difference to survival rates (). Over-expression and/or mutation of c-kit may also play a significant role in several other cancers, including some leukaemias () and testicular cancers (). However, resistance to Gleevec occurs as a result of deactivating mutations in the kinase active site (,). These diminish binding and rapidly reduce the clinical effectiveness of the drug. Several 2nd-generation c-kit kinase inhibitors are currently being developed to overcome this resistance (), although it is possible that they in turn may produce new patterns of resistance mutations in the kinase active site. Selective gene regulation at the transcriptional level is a potential alternative to targeting a protein, the product of gene expression. One way in which this can be achieved is by the induction of higher-order G-quadruplex DNA structures () in a G-rich region such as a promoter sequence () by a small-molecule ligand. This has been demonstrated for the oncogene at the nuclease hypersensitivity element (NHE) III that is responsible for up to 90% of transcription (,). G-quadruplexes, which may have transient stability by themselves when embedded within the double-stranded DNA of a eukaryotic gene, may thus be stabilized further by a small-molecule ligand. The structure and topology of two DNA quadruplexes have been determined by NMR spectroscopy (,), as well as that of a ligand (TMPyP4) complex (). These are structurally complex parallel-stranded quadruplexes, with several strand-reversal loops and base-pair platforms. Two discrete G-rich quadruplex-forming sequences have been identified (,) in the human core promoter region (). These are within the nuclease hypersensitive region of the promoter, suggesting that they are not involved in a chromatin complex. Biophysical and 1-D NMR studies have shown that these individual sequences can both form G-quadruplex structures (,). One sequence, d(AGTAGA), which occurs 87-nt upstream of the transcription start site, forms a single G-quadruplex species in solution (). The occurrence of four tracts of three consecutive guanines (underlined), separated by linkers of either one or four residues initially suggested that the sequence forms a G-quadruplex structure with these G-tracts forming the G-tetrad core, and the linker sequences forming loops, analogous to the parallel-stranded structure of the human intramolecular telomeric quadruplex (). However, this proposed model was unable to explain the dramatic quadruplex destabilizing effect caused by mutations in the linker sequences (). The NMR-based solution structure of the G-quadruplex formed by this precise sequence in K solution has now been determined (), and shows that c-kit87 has an unprecedented G-quadruplex folding topology that involves 18 of the 22 nt in tertiary interactions (), and providing rationales for the mutant data (,). These four non-essential nucleotides are in bold in the above sequence. One of the ‘loop’ guanine bases is directly involved in G-tetrad core formation, contrary to expectations and despite the presence of four three-guanine tracts. There are also four loops; two single-nucleotide double-chain-reversal loops, a two-residue loop, and a five-residue d(AGGAG) stem-loop. The net result is a tertiary quadruplex structure with complex features absent in simpler quadruplexes such as the human telomeric parallel and antiparallel arrangements. In particular, the presence of two well-defined clefts in the structure that are defined by the stem-loop and the two-residue loop strongly suggest that the c-kit87 quadruplex could be a target for the design of selective small molecules that would serve to stabilize the structure within the context of the core promoter sequence, and thus down-regulate expression. The structure allows for straightforward continuation of a DNA sequence in both 5′ and 3′ directions, suggesting that it could be formed within the promoter region without undue steric constraint. The potential of c-kit87 as a therapeutic target raises the question of the degree of its sequence and structural uniqueness. This issue is addressed here using a combined bioinformatics, circular dichroism (CD) and molecular dynamics simulation approach. The Ensembl human genome core database () version 38 (NCBI build 36) was searched for sequences of the patterns: The search software was that developed for earlier quadruplex searches (). The positions of each hit within the chromosome and its relation to the surrounding genes, or the gene within which it occurred, was recorded and compiled into a mySQL database. This database was then queried so that the results could be ordered and grouped as desired. Where sequences occurred upstream relative to the transcription direction of a gene, the distance between the gene and the transcription start site was retrieved and in the case where one of the hit sequences occurred within a gene, it was noted in which intron, exon or untranslated region the sequence occurred. Searches for the mutated c-kit sequences which were previously examined () and shown not to form quadruplex structures, were also carried out, in the same way as described above. These sequences are: Additional variations of these mutated sequences were also investigated. The human genome was searched for the following sequences with systematic variations at the 5, 9, 11 and 18 positions: The c-kit upstream regions from various different species were obtained from the Ensembl web site (using Ensembl release 43). Upstream regions for the orthologues to the human c-kit sequence were found for macaque, rat, mouse, cow, opossum chicken and zebrafish. A multiple sequence alignment was carried out on these sequences using the CLUSTAL software package (). The c-kit87 and the ten mutant sequences were synthesized and hplc purified (Eurogentec), and were then used in this study: CD spectra for them were acquired on a Chirascan spectrometer (Applied Photophysics Ltd) at King's College London. All samples were prepared at 100 μM in 50 mM potassium chloride and heated to 95°C and slowly annealed overnight to room temperature. The samples were further diluted, with buffer to 1 optical density unit prior to data collection. UV absorbance and CD spectra were measured between 360 and 200 nm in a 10 mm path-length cell. Spectra were recorded with a 0.5 nm step size, a 1.5 s time-per-point and a spectral bandwidth of 1 nm. All spectra were acquired at room temperature and buffer baseline corrected. The concentrations of the above oligonucleotides were determined by using the absorbance value at 260 nm and the Beer–Lambert law. One of the experimental c-kit87 NMR structures (PDB accession code 2O3M) was arbitrarily chosen and used as a starting point for all calculations. Mutants occurring with high frequency were identified using the bioinformatics techniques outlined above. Structural modifications were made to the native c-kit87 model to generate 3D models from these mutant sequences, changing only the base; backbone conformations were not altered at all. This was carried out using the Insight suite of programs (). In all, ten mutants were constructed and are listed in . Molecular dynamics simulations were carried out using the ff99 forcefield in the AMBER v9.0 package (). Each system was equilibrated with explicit solvent molecules (TIP3P) using 1000 steps of minimization and 20 ps of molecular dynamics at 300 K. The entire systems were kept constrained, while allowing the ions and the solvent molecules to equilibrate. The systems were then subjected to a series of dynamics calculations in which the constraints were gradually relaxed, until no constraints at all were applied. The final production run was performed without any restrain on the complex for 10 ns and co-ordinates were saved after every 10 ps for analysis of their trajectories. The simulation protocols were consistent for all of the systems. Periodic boundary conditions were applied, with the particle-mesh Ewald (PME) method () used to treat the long-range electrostatic interactions. The solute was first solvated in a TIP3P water box (), the boundaries of which were at least at a distance of 10 Å from any solute atoms. Additional positively charged K counter-ions were included in the system to neutralize the charge on the DNA backbone. The counter-ions were automatically placed by the LEAP program throughout the water box at grid points of negative Coulombic potential. The final system had net zero charge. All calculations were carried out using the SANDER module, trajectories were analysed using the PTRAJ module from the AMBER9.0 suite and viewed using the VMD program (). The c-kit87 experimental structure contains three looped-out bases (A5, C9 and C11), which visual inspection () shows do not play a role in maintaining structural integrity, since they do not interact directly with any part of the folded structure. Other notable features of the structure are (i) a Watson–Crick base pair between A1 and T12 and (ii) the AGGAG stem loop, which contains two A … G base pairs. The middle guanine, G18 in this loop sequence, stacks on the end of the loop and is not involved in any hydrogen bonding with other residues. Changing A5 and C9 to thymines has been found () to produce a structure with the same topology as the native; changing C11 to thymine produces a mixture of structures with a similar topology but where one structure contains an A1–T11 base pair and the other an A1–T12 base pair, as in the native structure. Modification of G18 to T18 also maintains the topology. There are 64 possible combinations for the three 'flipped out' bases at the 5, 9 and 11 positions. A total of 61 sequence occurrences were found, corresponding to just 12 unique sequences (a). The relative frequencies with which different bases occur are not random, with sequences that have T, G and G substitutions at the 5, 9 and 11 positions being the most common type, of which 21 were found. The thymine substitution at the 5 position occurs in ca. 97% of the sequence hits and the 9 and 11 positions were most frequently guanines. Sequences closely similar to the c-kit87 sequence itself are exceptionally rare. Only one other sequence has an adenine at the 5 position, only one other sequence has a cytosine at the 9 position and of the five other sequences which have a cytosine at the 11 position only one has another of the substituted bases in common with the c-kit87 native sequence. Examination of substitutions at position 18 in addition to those at the 5, 9 and 11 positions, showed that although there are a further 192 possible sequence combinations, only nine more actually occur (b). Again none of these nine additional sequences have more than two of the substituted bases in common with the c-kit87 sequence, and only two sequences have two bases in common, and the remaining seven have none in common. Our previous analysis of quadruplex loop occurrences in the human genome () found that loops of sequence AGGA, and therefore sequences containing AGGAG, are highly over-represented. Out of the many thousands of loop sequences which were found when searching for potential quadruplex sequences, AGGA was the 14th most frequently found loop. Searches for alternative Watson–Crick base-pairing combinations between the A1 and T12 positions yielded only 21 further hits (from a possible 768 more sequences), the majority of which have T at the 1st position and an A at the 12th position (c). Again there were no other sequences which differ from the native c-kit87 in only the alternative 1–12 pairing, although one sequence differed in only the 1–12 pairing together with the 9 position. d and e shows that the alternative 1–12 base pairings G-C and C-G have even fewer sequence hits, with just six and four sequences found respectively. Again these were dissimilar to the native c-kit87 sequence. A search for occurrences of the mutated c-kit sequences used in the initial study () (none of which form a stable quadruplex structure), found no hits for two of the sequences examined, (AGGGAGGGCGCTGGGCGCTGGG and AGGGAGGGCGCTGGGCGGCGGG). There are seven occurrences of the third sequence, AGGGAGGGAGGAGGGAGGAGGG () in the human genome. Two instances of this high purine-content sequence occurred in a potential promoter region, two instances were close together within an intron, and the other two were not near any regions of biological importance as far as is known. The distances to putative transcription start sites (TSS) were then examined for all of these sequence variants ( and ). The majority do not occur within genes, but are distributed in non-coding regions. The c-kit87 native sequence, which is 34 bases upstream of the TSS, is by far the closest to its TSS. The next closest is ENSG00000185245 (coding for Platelet glycoprotein Ib alpha chain precursor) which appears 147 bases upstream of its transcription start site. The sequence which bears the greatest similarity to the c-kit87 sequence, differing only by a C in the 9th position, occurs upstream of the transcription start site of the gene ENSG00000136213 (coding for the protein carbohydrate sulfotransferase 12); however this sequence is far upstream, at ∼18.6 kb from the TSS). The remaining sequences are also, in general, located in quite remote positions. We have also examined the phylogenetic features of the c-kit87 sequence. shows the results of the multiple sequence alignment between the upstream sequences of several c-kit87 orthologues. The two non-mammalian sequences gave very dissimilar alignments to the rest of the species, however the mammalian sequences were similar enough to identify the relevant, orthologous upstream regions. The opossum and macaque sequences were identical to the human while the cow sequence differed by only one base, where a cytosine appears instead of guanine at position 21. The mouse and rat sequences are identical. However, they have an adenine inserted at the 2nd position and a deletion at 9 and 15 which seem to make it impossible for them to form quadruplex structures with the same topology as the human c-kit sequence. They remain guanine-rich however, so it is not impossible that they can fold into an alternative quadruplex topology. As a check on the sequence occurrences we have compared search results for different G-rich sequences, using the non-quadruplex-forming c-kit87 mutants 1 m, 2 m and 3 m. In total there were 38 hits for sequence 1m, none for sequence 2 m and two for sequence 3 m ( and ). One of the sequences appears 71 bases upstream of ensembl gene ENSG00000133466 (HGNC name: C1q tumor necrosis factor-related protein 6) and one occurs 228 bases upstream of the transcription start site of ensembl gene ENSG00000185985 (HGNC name: SLIT and NTRK-like family, member 2). The UV and CD spectra for the c-kit87 sequence and the ten mutants are shown in . All of the UV spectra are identical. The CD spectra all show the same pattern of minimum at 240 nm and maximum at 262 nm, although there are significant differences in peak heights. We have undertaken molecular dynamics simulations on the native c-kit87 structure and ten mutants, as detailed above and in . The root mean-square deviation (RMSD) over the course of a molecular dynamics simulation was used as a measure of the conformational stability of a structure or model during that simulation. The native c-kit87 NMR model and the mutant models examined here are extremely stable structures, as is evident from the stable and small RMSDs over the timescales of 10 ns simulations, starting from the initial structure. The maximum variance ranged between 1.6 and 2.2 Å for the native and mutant 5 respectively and is shown in . A more detailed picture of differences in residue mobility within and between simulations was obtained from graphs of the root mean-square fluctuation (RMSF) of residues relative to the average structure. The RMSF profiles of all the mutants are somewhat similar to that observed for the native structure. In particular, the peaks in the RMSF profile correspond to residues 5, 9 and 11 (a). The NMR structure shows that these three bases do not interact directly with any other part of the structure and hence do not play any role in stabilizing it. This is fully confirmed by the simulation of the native structure and of the 5, 9 and 11 mutants. The guanine bases which contribute to quartet formation are extremely stable, whereas the AGGAG loop (and nucleotide G18 in particular) shows significant flexibility. Interestingly, as predicted by our bioinformatics results, mutation of G18 to T18 results in the retention of the same topology as the native sequence. This can be explained by the overall flexible nature of the AGGAG loop, which would allow the G18T modification to be adopted into a similar folding topology. The G17–G18 stacking in the loop is similar to that found in the T4-T5 stacking adopted in the loop region of the crystal structure of the telomeric sequence GTG (40: PDB id 1JPQ). Mutants 3 (A5T) and 6 (C9T) also exhibit patterns of flexibility that are very similar to the native structure (b). This is in accord with the NMR studies where again these modifications produced a structure with same topology as the native. However, modification of C11 to T11 was found to produce a mixture of structures with A1–T12 base pairing (in the native structure) and A1–T11 base pairing (in the mutant). Examining the RMSF profile for the mutant-9 (C11T) simulation (c), we see that residue A1 has increased flexibility compared to the flexibility of residue A1 in the native structure simulation. Furthermore, the flexibility of residue C9 in mutant 9 is considerably reduced. A slight increase in flexibility of residue T12 is also observed, suggesting that some minor structural changes may have occurred during the course of the simulations. To investigate the dominant motions, principal components analysis was performed. By calculating the eigenvectors from the covariance matrix of a simulation and then filtering the trajectories along each of the different eigenvectors, it is possible to identify the dominant motions observed during a simulation, by visual inspection. Plotting the start and the end points of eigenvectors as arrows, highlights the direction of motion for a particular atom. Application of such an analysis to these simulations enabled us to identify the structural changes occurring between A1–T2 and T11. In order for the A1–T11 base pair to form, the A1–T12 base pair needs to be broken and T12 has to move out and pave the way for T11 to occupy its place (). This is clearly observed in the PCA analysis; however the timescale of the simulations are too short for these entire processes to be fully simulated. The 22-nt c-kit87 promoter sequence is unique within the human genome. Its fold and tertiary structure does not have precedent among known DNA quadruplexes. The present theoretical and experimental studies have shown that (i) none of the closely related sequences (encompassing all nucleotides not involved in the maintenance of structural integrity) occur immediately upstream (<100 nt) of a transcription start site, and (ii) that all of these sequences correspond to the same stable tertiary structure. The identity of the CD spectral maxima and minima indicate that all the ten related mutant sequences adopt the same overall fold as the native c-kit87 sequence; the differences in peak height can be ascribed to the sequence differences, although a detailed analysis is beyond the scope of this article. It is concluded that the c-kit87 tertiary structure may also be formed in a small number of other loci in the human genome, but the likelihood of these playing a significant role in the expression of particular genes is small. The c-kit87 quadruplex thus fulfils a fundamental criterion of a ‘good’ drug target, of possessing distinctive 3D structural features that are only present in at most a handful of other genes, with only one, that for platelet glycoprotein Ib alpha chain precursor (ENSG00000185245) also being in a likely core promoter region. The genome searches with mutant c-kit87 sequences that are known not to form quadruplexes, found a number of hits; two are close to transcription start sites, demonstrating the importance of knowledge of the folding behaviour. DNA is normally considered as a structurally homogeneous molecule, defined in its flexibility by the constraints of the double helix. The possibility of DNA forming higher order structures is not new, and triplexes and quadruplexes have long been postulated, especially in regulatory sequences. However, until now even these features have not been considered to possess a high degree of complexity and variation (though the structures of the c-myc quadruplexes do show features that are absent in previous quadruplex structures). The c-kit87 structure, involving 18 out of 22 nt in tertiary interactions, shows that non-duplex DNA sequences can adopt highly stable and complex arrangements. We are as yet far from knowing the rules governing these folds or the extent to which they may occur. Searches for potential quadruplex sequences in non-telomeric DNAs have always used a template pattern based on known quadruplex sequences and their topologies (,,,), in which four runs of guanine bases are separated by three distinct loop regions: G X G X G X G where = 3–5 and = 1–7. In lieu of structural data providing evidence that additional sequence patterns are valid, we suggest that this remains a reasonable assumption. Important caveats are (i) that a particular topology cannot be assumed purely on the basis of the sequence alone, and (ii) that the occurrence of a sequence does not necessarily mean that it corresponds to a stable or potentially stable quadruplex—as is the case with a number of the c-kit87 mutants (). The distinctly non-random distribution of particular bases at the non-essential 5, 9, 11 and 18 positions of the c-kit87 sequence is a surprising observation, which is being further examined experimentally and theoretically.

Translation initiation of capped mRNA is remarkably stimulated by the 3′ poly(A) tail. This synergistic effect on translation was attributed to interactions between the eukaryotic translation initiation factor (eIF) complex eIF4F and the poly(A)-binding protein (PABP) (). eIF4F consists of the cap-binding protein eIF4E, the ATP-dependent RNA helicase eIF4A, and eIF4G that serves as a scaffold for the binding of several proteins, including eIF4E, eIF4A and eIF3. In addition, eIF4G contains in its N-terminal part a binding site for PABP. The simultaneous binding of eIF4G to the cap-bound eIF4E and PABP associated with the poly(A) tail can bring the RNA ends in close proximity and circularize mRNA, thereby allowing ribosome recycling (,). Cytoplasmic PABP is an abundant protein of 70 kDa that has been recognized as a true translation initiation factor (). It contains four RNA recognition motifs (RRM) in its N-terminal domain (NTD) that are involved in RNA and eIF4G binding (). A proline- and glutamine-rich linker region connects the NTD to the C-terminal domain (CTD) that interacts with various proteins involved in both translation initiation (eIF4B, 60S ribosomal subunit) as well as termination (releasing factor eRF3). In addition, the CTD mediates PABP oligomerization that contributes to the cooperative binding to poly(A) sequences longer than 12 residues. Besides its function in translation, PABP controls mRNA stability and plays additional roles in the regulation of gene expression (,). Picornaviruses have a messenger-sense RNA genome of ∼7500 bases, with a large open reading frame that is flanked by 5′ and 3′ nontranslated regions (NTR) and terminated by a poly(A) tail of more than 60 residues. In place of the cap structure, the 5′ end of the viral genome is covalently attached to the small protein VPg that is the primer of RNA synthesis. Replication of most picornaviruses in cell culture is associated with cytopathology. These viruses have evolved various mechanisms to ensure efficient viral translation, often at the expense of host protein synthesis. Viral proteins rearrange cell membranes and viral proteinases impact the structure and localization of host cell proteins. Among the identified cellular targets, eIF4G is cleaved by poliovirus (PV) proteinase 2A or foot-and-mouth-disease virus proteinase Lb, resulting in the abrogation of cap-dependent translation of host mRNAs early in infection (). PABP cleavage by coxsackievirus B3 (CVB3) or PV proteinases 2A and 3C seems to occur later in the viral life cycle (). Overall, these studies demonstrated that cytolytic picornaviruses down-regulate host metabolism mostly by drastically inhibiting cellular translation, while viral translation continues via a cap-independent mechanism. As a member of the hepatovirus genus, the hepatitis A virus (HAV) is genetically and phenotypically distinct from the other members of the picornavirus family (e.g. enterovirus, rhinovirus, cardiovirus and aphthovirus). Whereas most other picornaviruses contain—besides the major proteinase 3C—a second proteolytic activity in their polyprotein, which is predominantly causing host shut-off, the HAV polyprotein lacks such an activity. The HAV 5′NTR of 734 residues comprises -acting elements involved in genome replication and translation initiation. Computer-assisted folding predictions and biochemical probing showed that the HAV 5′NTR forms extensive higher order structures which include six stem-loop domains [(), see for a simplified outline of essential parts of the HAV 5′ NTR]. The 5′ most terminal domain (bases 1 to 95) contains a hairpin and two pseudoknots and is followed by a pyrimidine-rich tract (pY1, bases 96 to 148). These parts of the 5′NTR are involved in viral replication (). The remainder of the 5′NTR (bases 155 to 734) functions as an internal ribosome entry site (IRES), allowing cap-independent translation initiation. Owing to its unique structure and due to diverse requirements for optimal activity, the HAV IRES (type III IRES) substantially differs from the IRES of other picornaviruses and that of the hepatitis C virus (,,,). The HAV IRES interacts with a number of host proteins, such as the poly(rC)-binding protein 2 (PCBP2) (), glyceraldehyde-3-phosphate dehydrogenase (GAPDH) (,), the polypyrimidine tract-binding protein [PTB; (,)] and eIF4GI (,). The role of these interactions is still elusive. Following uptake of the viral messenger-sense RNA into the cell, the viral polyprotein is synthesized through internally initiated translation at the IRES. The HAV proteinase 3C, that itself is a part of the polyprotein, catalyzes the subsequent liberation of the structural and nonstructural proteins. Together with host proteins, the latter assemble into the viral replication complex and viral RNA synthesis starts with the generation of a negative strand intermediate. As translation and negative strand synthesis are competing processes that proceed in opposite directions on the infecting viral genome, the translating viral RNA must be cleared of initiating or recycling ribosomes before RNA synthesis can begin (). In this context, the observation is interesting that polioviral genomes that have not been translated cannot function as template for RNA synthesis. The molecular mechanisms regulating this selection of viral RNA template functions are currently scrutinized. Most strikingly and in sharp contrast to the cytolytic picornaviruses, replication of most HAV strains is highly protracted and noncytopathogenic in cell culture. One way to ensure viral persistence in culture is the observed suppression of cellular antiviral defense mechanism by HAV (). As no effect on the host metabolism has been detected so far in HAV-infected cells, it appears that HAV remains in a state of constant competition with host translation and never reaches a privileged stage of protein synthesis (,,). Beyond competing with host protein synthesis, production of the HAV polyprotein clearly conflicts with HAV RNA synthesis at later phases of the viral life cycle. In particular, it is not clear how HAV translation is stalled to allow RNA synthesis to occur. It is tempting to speculate that once enough of the replication complex is formed, one or several of its components interfere with HAV translation initiation by affecting essential translation factor(s). Here, we present evidence that—unlike the enteroviruses—HAV proteinase 3C does not cleave eIF4G. However, HAV 3C cleaves PABP and and exploits its cleavage product. We show that the NTD of PABP has improved RNA-binding capacity to the pY1 in the HAV 5′NTR and specifically suppresses HAV IRES translation pointing to its regulatory function in halting viral protein synthesis. Plasmids pET15b-3ABCwt, pET15b-3ABCm and pET15b-3ABCµ that encode various precursor forms of HAV 3C were described before (). pET28-hPABP (kindly provided by M. Görlach) encodes the complete PABP with an N-terminal His-tag (). pET28-PABP1234 (kindly provided by G.J. Goodall) encodes the N-terminal domain (NTD) of PABP with four RNA-binding motifs and a N-terminal His-tag (). pET28-PABP-CT [kindly provided by M. Kiledjian; ()] encodes the His-tagged CTD. The HAV replicon (pT7-18f-Luc-A60) was described before (). The poliovirus replicon pRluc31 was kindly provided by R. Andino (). Luciferase-encoding replicon RNA was prepared according to the user manual of the T7 RiboMAX Large Scale RNA production system (Promega), after linearization of the HAV replicon cDNA with AgeI and the PV replicon with MluI. pHAV-IRES-luc encodes the firefly luciferase that is preceded by the HAV IRES (HAV nucleotides 44–736) (). pHAV-IRES-luc was linearized with NotI prior to transcription with T3 RNA polymerase. Radiolabeled RNAs was prepared as described in the manual of the MaxiScript™ transcription kit (Ambion), with 3 µl α-[P]-UTP (10 µCi/µl) and additional 2 µl UTP (0.05 mM) in a 20 µl volume. To generate 3′ NTR-A20 and 3'NTR-A60 transcripts, pT7-18f-(ΔP1-P3)-A20 and pT7-18f-(ΔP1-P3)-A60 were linearized with AgeI and used as template for T7 transcription (). pT7-18f-(ΔP1-P3) A0rbz was linearized with RsrII to generate the 3′ NTR-A0 transcript. To produce RNA1-94 and RNA95-148, pGEM1-HM175-1-95 and pGEM1-HM175-95-736 were linearized with EcoRI or SspI, respectively, and transcribed with SP6 RNA polymerase (). Radiolabeled RNA was purified, and dissolved in 50 µl RNase-free water. Plasmids pET28-hPABP, pET28-PABP1234 and pET28-PABP-CT were expressed in strain BL21 (DE3) pLysS as described (,,). The soluble proteins were purified using HisTrap chelating HP columns as recommended by the manufacturer (Amersham Biosciences, USA). The eluted proteins were concentrated and transferred into 50 mM Tris–HCl, pH 8.0, 50 mM NaCl, 15% glycerol using a centrifugal filter device (Amicon Ultra 30 000). Purified 3C of HAV and coxsackievirus B3 (CVB3; kind gift of R. Zell) were described previously (,). EMSA was essentially performed as described before (,,). [P]-labeled riboprobes (2.5 × 10 c.p.m.) were incubated with increasing amounts of purified PABP or NTD (50–700 nM) in 15 µl reaction buffer containing 5 mM HEPES, (-2-hydroxyethylpiperazine-′-ethansulfonic acid), pH 7.9, 25 mM KCl, 2 mM MgCl, 1.75 mM ATP, 6 mM dithiothreitol (DTT), 0.05 mM phenylmethylsulfonyl fluoride, 166 µg/ml of tRNA, and 5% glycerol. After 20 min at 30°C, the mixture was supplemented with 5 µl of sample buffer (1 mM EDTA, 0.25% bromophenol blue (BPB), 0.25% xylene cyanol, 50% glycerol) and analyzed by electrophoresis using a 6% nondenaturing polyacrylamide gel (PAGE). Electrophoresis was conducted in 0.5× Tris-borate buffer at 150 V for 30–90 min until the BPB marker had migrated to 2/3 of the gel length. The gel was scanned using a PhosphorImager (Fujifilm BAS 1000, Japan) and the image was analyzed with the analysis software PCBAS (Raytest, Isotopemessgeräte GmbH, Germany). The apparent equilibrium-binding constant (app. ) was calculated according to Lane (). Four-hundred nanogram-purified recombinant PABP and various amounts of HAV or CVB3 proteinases 3C (final concentration, 1–10 µM) were incubated at 37°C for 6–24 h in cleavage buffer (). The reaction was stopped by the addition of sodium dodecylsulfate (SDS)–PAGE sample buffer and the products were analyzed by SDS–PAGE, followed by immunoblot with anti-3C, anti-PABP or anti-His. For cleavage of PABP and eIF4G in cell fractions (S10, S200 and P200, see subsequently), 10 µl of the extracts were incubated with HAV 3C (7 µM final concentration). The cleavage products were detected by immunoblot using anti-eIF4G or anti-PABP. HAV strain 18f was propagated in the human hepatoma cell line Huh-7. For HAV infection, cells were inoculated for 3 h at 37°C with the soluble extract of HAV-infected cells at a multiplicity of infection (moi) of 1 tissue culture infectious dose/cell in OptiMEM (Invitrogen). Infected cells were incubated in Dulbecco's-modified Eagle medium (DMEM) with 2% fetal calf serum for the indicated time periods. The recombinant vaccinia virus vTF7-3 that encodes T7 RNA polymerase was amplified in COS-7 cells and the virus stock was plaque titrated by serial 10-fold dilutions (). The working dose of vTF7-3 as a helper virus to aid the expression of T7-promoted genes was characterized by quantification of luciferase activity after transfection of pT7-LUC (Luciferase T7 Control DNA, Promega), followed by infection with vTF7-3. For coexpression, Huh-T7 cells that constitutively express T7 RNA polymerase were used (). They were grown in DMEM in the presence of geneticin (G-418 sulfate, 400 µg/ml), penicillin (100 U/ml) and streptomycin sulfate (100 µg/ml). pET28-hPABP and plasmids encoding HAV 3C were cotransfected into 5 × 10 cells. The transfection mixture containing 1 µg cDNA and 8 µl Lipofectamin (Invitrogen) in 200 µl OptiMEM was pre-incubated for 30 min at room temperature and diluted with OptiMEM to 1 ml before applying to 80% confluent cells. After incubation for 3 h at 37°C, transfected cells were infected with vTF7–3 (moi ∼1). After 1 h at 37°C, the inoculum was replaced with DMEM containing 10% fetal calf serum and antibiotics. After incubation for 24–48 h at 37°C, the cells were scraped in 250 µl phosphate-buffered saline containing 0.05% Tween-20 and lyzed by three cycles of freeze/thawing. The clarified supernatant was used for reporter gene detection and/or for immunoblot analyzes with anti-PABP or anti-His-tag. To confirm the expression of viral proteinase 3C, the transferred proteins were also tested with anti-HAV 3C raised against the recombinant proteinase (). To determine PABP cleavage in the course of the HAV infection, extracts of infected cells were obtained as described above and analyzed by immunoblot with anti-PABP. After electrophoretic separation by SDS–PAGE (10 or 12% polyacrylamide) and transfer onto a nitrocellulose membrane (Protran, Schleicher & Schuell, Bioscience), the blots were probed either with monoclonal anti-His (Novagen), anti-PABP or anti-eIF4G1 directed against the peptide KKEAVGDLLDAFKEVN representing amino acid residues 523–538 of the N-terminus [(); kind gift of R.E. Rhoads]. Anti-PABP was raised against a synthetic peptide sequence (GIDDERLRKEFSPFGTC) in the RRM4 of human PABP (kind gift of R. Lloyd). The particle-specific enzyme-linked immunosorbent assay (ELISA) with the monoclonal anti-HAV 7E7 (Mediagnost, Germany) and its horseradish peroxidase conjugate were applied to detect viral particles as described (,). Cell extracts were prepared as described elsewhere (,). In brief, Huh-7 cells at 90% confluence were suspended and harvested by centrifugation (800, 4°C, 6 min). After washing with PBS, the cell pellet was re-suspended in 2 vol of hypotonic buffer (50 mM KCl, 25 mM HEPES, 1.6 mM MgCl, 1 mM DTT). The suspension was allowed to swell on ice for 15 min, before lyzing with 15 strokes of a Wheaton Douncer. Then 1/9 vol of 10× concentrated HNG buffer (25 mM HEPES, pH 7.5, 1 M potassium acetate, 30 mM MgCl, 30 mM DTT) was added. The debris was spun at 11 000, for 20 min at 4°C and the supernatant (S10) stored at −70°C. The extract concentration was >25 A U/ml. S200 and P200 (Ribo) were prepared as described (). The 50 µl translation mixture contained 25 µl S10 extract, 5 µl 10× translation mix (125 mM HEPES pH 7.3, 10 mM ATP, 2 mM GTP, 2 mM CTP, 2 mM UTP, 100 mM creatine phosphate, 0.2 mM amino acids, 1 mg/ml creatine phosphokinase), 5 µl salt mix (1 M potassium acetate, 30 mM MgCl, 2.5 mM spermidine), 1 µl methionine (1 mM), 40 U RNase inhibitor and 1 µg luciferase-encoding RNA. When the effect of PABP and its truncated versions was tested, PABP, NTD and CTD in native and heat-denatured form were added at the indicated amounts, before the mixtures (prepared in at least duplicate) were incubated at 30°C. Aliquots in duplicate were taken at 90 min, and luciferase activity was tested with the Luciferase Assay System (Promega) in the luminometer Lucy-3 of Anthos, Germany. Luciferase activity is expressed in relative light units (RLU). To evade the cells’ antiviral machinery early on in the viral life cycle, proteinases of some picornaviruses cleave eIF4G that serves as scaffolding protein in the cap-binding complex eIF4F (,). Whereas host translation is subsequently blocked, viral IRES-mediated translation proceeds and is even stimulated in the presence of cleaved eIF4G (). HAV replicates in a highly protracted and asynchronous fashion in cells. This particular replication feature, combined with low yields of viral progeny, was often posed as argument that specific viral effects on the host metabolism could not be identified in HAV-infected cells. To provide direct evidence for the unconfirmed observation that eIF4G remained intact in HAV-infected cells (), eIF4G cleavage by the one and only HAV proteinase 3C was directly assessed. To this aim, endogenous eIF4G of a S10 cell extract was subjected to treatment with purified viral proteinases. After incubation with HAV 3C for 6 h at 37°C, no cleavage products were detectable in the anti-eIF4G blot (A, lower panel, compare lanes 1 and 2). However, and in accordance with previously reported data (), eIF4G was almost completely cleaved by CVB3 3C (lane 3) under the same conditions. Moreover, numerous attempts to demonstrate eIF4G cleavage were unsuccessful. In no case, eIF4G cleavage products were detectable. These approaches included eIF4G analysis in HAV-infected cells and after vaccinia virus-mediated overexpression of HAV 3C and its proteolytically active precursor 3ABC. eIF4G was cleaved by CVB3 3C that was used as control, but not by HAV 3C (Figure 1S, in Supplementary Data). The resistance of eIF4G to HAV 3C and its precursor-mediated cleavage clearly is in line with the requirement of intact eIF4G for translation initiation by the HAV IRES (,). As eIF4G cleavage correlates with enteroviral cytopathology, lack of HAV 3C-mediated eIF4G cleavage is also consistent with the noncytolytic replication of HAV. Strikingly, HAV IRES-dependent translation does not only require complete eIF4G, but the entire eIF4F complex. This was clearly evidenced by the finding that 4E-binding protein interfered with HAV IRES activity, supposedly by sequestrating eIF4E and preventing its interaction with eIF4G (). Moreover, addition of cap-analog blocked the HAV IRES activity , indicating that translation initiation by the HAV IRES requires association of eIF4E with eIF4G and an empty cap-binding pocket of eIF4E (,). Combined with earlier observations, all evidence points to the notion that translation initiation by the HAV IRES requires essentially the same translation factors as capped host mRNAs and seems to compete with those (,,,). This notion presents a surprising paradox that the function of the 3'part of the HAV 5′NTR in translation is currently indistinguishable from cap-dependent translation, although it folds into the conformation of a genuine IRES and mediates the expression of a second cistron in a bicistronic construct (). Apparently, HAV cap-independent translation does not benefit from the advantage that is usually supplied by viral or cellular IRES. PABP is a translation initiation factor that together with eIF4G forms a bridge between the 3′ and 5′ ends of mRNA, including picornaviral RNA (,,,). The interaction between eIF4G and PABP is sufficient for mRNA circularization (). PABP cleavage has been implicated in the cytopathic and apoptotic degeneration of infected and non-infected cells (,). To investigate whether PABP is a direct substrate of HAV 3C, the same cleavage reaction used to identify eIF4G cleavage products was analyzed by immunoblot with anti-PABP (A, upper panel). Consistent with previously published data (,), CVB3 3C partially cleaved PABP, resulting in three products (lane 3, marked with asterisks). PABP cleavage by HAV 3C generated two products (lane 2, marked with arrowheads). The smallest HAV 3C cleavage product migrated with the same mobility as the smallest cleavage product of CVB3 3C, suggesting that HAV 3C also cleaves near or at the dipeptide sequence (Q/T) of PABP at position 415/416, which was proposed as site of cleavage by enteroviral 3C (). Cleavage at this site separates two essential functions of PABP: the NTD with four RRMs binds RNA, whereas the CTD interacts with various translation factors (,,,). For poliovirus, it was demonstrated that initiation factor- and ribosome-associated PABP was specifically targeted by the viral proteinase , whereas non-ribosome-associated PABP was mostly resistant to 3C cleavage (). However, no difference in PABP cleavability by HAV 3C was detectable when the S200 and P200 (Ribo) fractions of Huh-7 cell extracts were incubated with purified recombinant HAV 3C (Supplementary Figure 2S). Moreover, under no condition PABP cleavage was complete when CVB3 or HAV 3C was used, suggesting that compartimentalization or an alternative conformation (PABP complexed with proteins or RNA) may modulate PABP cleavage. Notwithstanding, PABP cleavage by HAV 3C was further analyzed to ensure that PABP was a direct substrate of HAV 3C. For this, purified recombinant PABP with an N-terminal His-tag was incubated for various times (data not shown) or with increasing concentrations of HAV 3C. In addition to anti-PABP (data not shown), anti-His was used in the immunoblot, in order to specifically demonstrate the N-terminal nature of the cleavage products. As depicted in B, purified PABP with an N-terminal His-tag was almost completely cleaved when HAV 3C was used up to a 10 µM concentration (lanes 1–4). Again, two N-terminally tagged cleavage products were detectable by anti-His (marked by arrowheads). The cleavage products that were derived from native and recombinant PABP and recognized either by anti-PABP and/or anti-His were indistinguishable (A and B, respectively), implying that they were C-terminally truncated. In HAV-infected cells, 3C co-exists with the precursor polypeptide 3ABC that also shows proteolytic activity with slightly different substrate specificity (,). To test whether this form of the viral proteinase was active and yielded different PABP cleavage products, proteolytically active 3ABCwt and 3ABCm, with non-cleavable 3A/3B and 3B/3C junctions, were co-expressed with His-tagged PABP and with the help of vaccinia virus vTF7-3. Proteolytically inactive 3ABCµ, which carries a mutation at the active site of 3C (C172A), was used as a control. The expression and cleavage products were detected by immunoblot with anti-His (C). As expected, 3ABCµ did not cleave PABP (lane 2). Both active proteinases, 3ABCwt (lane 1) and 3ABCm (lane 3), generated two PABP cleavage products (arrowheads) that corresponded to those produced by cleavage with the purified mature enzyme (A and B). The proteinase precursors with an N-terminal His-tag were also detected in the blot (empty arrow). Due to its autoproteolytic activity resulting in 3BC and 3C, lower amounts of unprocessed 3ABCwt (lane 1) were found, as compared to either 3ABCµ (lane 2) or 3ABCm (lane 3). The data suggest that mature proteinase 3C and its precursor 3ABC have the same PABP cleavage specificity. In order to evaluate the extent of PABP cleavage in HAV-infected cells, cell extracts of infected and non-infected Huh-7 cells were analyzed by immunoblot with anti-PABP (). To correlate viral replication with the extent of PABP cleavage, cell morphology was judged by light microscopy and HAV particle formation was determined in the same experiment. Compared to uninfected cells, no morphological changes were observed at any time during viral replication, underlining the noncytopathogenic replication of HAV in this hepatoma cell line (data not shown). Under the chosen infection conditions, the small PABP cleavage product of 41 kDa (arrowhead) was detectable starting 9 days post-infection (lane 6) when viral replication was actively proceeding. The cleavage product detected in infected cells comigrated with the polypeptide generated by PABP cleavage with purified HAV 3C and with the small cleavage product of CVB 3C (data not shown). Interestingly, only the small PABP cleavage product was detected in HAV-infected cells as compared to or cleavage with recombinant 3C (). Although viral particle accumulation still continued until the end of the experiment (indicated below the lanes), the relative small amount of the 41 kDa PABP cleavage product remained constant. Independent of the infectious dose used, the extent of PABP cleavage was always low and only detectable by the appearance of the cleavage product. The highly limited PABP cleavage is in accordance with the noncytolytic replication of HAV that does not shut off host protein synthesis. Based on results shown below, we speculate that HAV 3C targets only those PABP molecules that are bound to the viral poly(A) tail and associated with the translation complex. It is expected that the majority of PABP molecules, in particular those involved in host protein translation, remains intact due to inaccessibility to HAV 3C. It is also possible that the low concentration or short half-life of 3C is the reason for the incomplete cleavage (). Collectively, the and cleavage data clearly show that PABP is a direct substrate of HAV 3C. Based on its electrophoretic mobility and compared with the PABP products generated by enteroviral 3C cleavage, it can be concluded that HAV 3C removes the C-terminal third of PABP that mediates PABP oligomerization and recruits proteins involved in translation initiation and termination (). The exact locations of the HAV 3C cleavage sites within PABP await sequence analysis. Apparently PABP cleavage in HAV-infected cells is strictly limited to a particular portion of PABP molecules, suggesting that HAV 3C-mediated cleavage might support a viral function rather than concern host translation. Possibly, HAV 3C is only active on PABP molecules associated with the viral poly(A) tail. In fact, only a small portion of the viral genomes appears to be unpackaged and translationally active in HAV-infected cells (). In the next experiments, we tested the recently proposed hypothesis that 3C cleavage of PABP might be a precondition for viral translation arrest (). Similar to studies described elsewhere, transcripts containing the firefly luciferase preceded by the HAV IRES were translated using either the reticulocyte lysate (data not shown) or S10 extract that were pretreated with HAV 3C (). As under the experimental conditions used, not only PABP, but also PCBP and PTB were found to be cleaved, the effect of cleaved PABP on HAV translation could not be singled out under this experimental condition (data not shown). To evade the unintended cleavage of host proteins, we next tested the effect of purified NTD added to translation reactions primed with synthetic transcripts representing the HAV and PV replicon. The S10 extract of Huh-7 cells was supplemented with purified PABP, NTD and CTD and IRES translation was monitored by the reporter gene activity. Cap-independent translation of the HAV replicon was mostly unaffected by the addition of intact PABP or CTD (A, bars 3 and 4), but inhibited when NTD was present (bar 2). In the assays, the NTD and CTD concentration was 0.3 µM and thus in the range of endogenous PABP (). Under the same conditions, translation of the PV replicon RLuc31 was two times higher (right ordinate) and no suppression was observed (A, lanes 5–8). The inhibitory effect of NTD on HAV IRES translation was also detectable when a synthetic transcript derived from pHAV-IRES-luc and containing the firefly luciferase preceded by the HAV IRES was expressed in the S10 extract (B). In this experiment, the effect of native NTD was normalized by that of the heat-denatured protein set at 100%. Under these experimental conditions, NTD at concentrations equal to and higher than 0.3 µM specifically suppressed the function of the HAV IRES. No effect was observed when CTD was added to the system (data not shown). These findings clearly demonstrate that the N-terminal domain of PABP has a dominant negative effect on HAV IRES-mediated translation. They suggest that NTD might either change the conformation of the HAV IRES or of eIF4G or compete with eIF4G for binding to the IRES. HAV IRES-dependent translation competes with and precludes subsequent negative strand RNA synthesis on the same RNA molecule. For PV, several viral and host proteins in association with viral 5′ terminal RNA structures were proposed to be involved in RNA template switching from translation to replication (,,). In particular, cleavage of RNA-binding host proteins (PABP, PCBP, PTB, La autoantigen) might be implicated in this regulatory step (,,). In line with this concept, we were interested to search for a positive role of NTD in HAV RNA synthesis. So far, no system is available to directly study HAV genome replication. As binding to terminal RNA structures of the viral genome is an essential prerequisite for the role of a host or viral protein and therefore a surrogate feature for RNA synthesis, we compared the binding specificities and affinities of PABP and NTD to terminal RNA elements of the HAV genome. In this context, it is also interesting to note that biosynthesis of PABP is inhibited by autoregulatory binding of PABP to the 5′NTR of its own mRNA (). The functionality and specificity of complete and truncated recombinant PABP was first determined by their interaction with the HAV 3′ NTR, using EMSAs. The apparent equilibrium-binding constant (app. ) was determined in additional titration experiments (data not shown). No binding of recombinant PABP to the HAV 3′ NTR lacking the poly(A) tail was observed (A, lanes 1–4, app. >> 10 µM). Ribonucleoprotein (RNP) complex formation was seen when the RNA ligand contained a poly(A) tail of 14 (data not shown) or 20 residues (lanes 5–9, app. = 0.05 µM). PABP binding to the 3′NTR was even more enhanced when the poly(A) tail comprised 60 residues (lanes 10–13, app. < 0.05 µM). This app. was similar to what has been reported for the 3′NTR of PV with a poly(A) tail of 80 nt (). As expected, binding of multiple copies of PABP to the poly(A) tail was noticeable by the formation of RNPs with different mobilities. The NTD of PABP also interacted with the polyadenylated HAV 3′NTR, yet somewhat less efficiently (data not shown). Our results confirm the notion that cooperative PABP binding to the HAV 3′NTR is dependent on the poly(A) tail length and that binding requires the N-terminal RRMs of PABP. Surprisingly, initiation of enteroviral negative strand RNA synthesis is regulated by -acting RNA elements present at the distant 5′ end of the positive strand RNA genome (,,). Moreover, a protein bridge formed by PCBP and PABP was proposed to be required for PV negative strand synthesis (). In line with such a model, we were interested to test the possibility that C-terminally truncated PABP might be directly involved in a replication function mapping to the HAV 5′NTR. Both RNA structural elements comprising bases 1–94 and 95–148 (pY1) were previously found to be essential for HAV RNA synthesis () and were therefore tested for their direct interaction with PABP and NTD. Whereas RNA1–94 did not directly bind to either complete or C-terminally truncated PABP (B, lanes 1–3), incubation of purified NTD with radiolabeled pY1 yielded a RNP that was discernable by EMSA (lanes 4–6). In contrast to PABP (lane 5), NTD at the same concentration (0.7 µM) was able to shift the mobility of pY1 (lane 6). Based on the relative app. that was calculated for PABP and NTD, it was concluded that NTD was ∼10 times more efficient in binding pY1 (see Supplementary Figure 3S). The data indicate that removal of the CTD from PABP-enhanced binding to pY1 in the HAV 5′NTR, a structure pivotal in virus replication. Although speculative, the data provide the basis for the hypothesis that PABP cleavage mediated by HAV 3C might be involved in HAV template switching from translation to genome replication. In summary, we report that HAV 3C cleaves PABP, which was shown to mediate the synergistic effect of the poly(A) tail on IRES-dependent translation and substantially enhance protein synthesis (). We also show that the N-terminal fragment of PABP inhibits HAV IRES-dependent translation () and has an enhanced binding capacity to pY1 ( and 3S). Based on these new findings and a model proposed for poliovirus (), we suggest a model to explain the putative role of PABP cleavage in HAV RNA template switching (). In contrast to poliovirus that shuts off host translation by cleaving eIF4G, HAV translation constantly competes with host translation for initiation factors (e.g. eIF4G) and is therefore highly inefficient (,,). After adequate amounts of translation and polyprotein processing products have accumulated, PABP bound to the HAV poly(A) tail is specifically targeted by HAV 3C. In contrast to the complete protein, the resulting poly(A) bound N-terminal cleavage product of PABP (NTD) no longer bridges the poly(A) tail to the IRES, but rather to the 5′RNA structure pY1, which was shown to be essential for viral replication (). As a consequence, ribosomes cease to recycle in order to initiate HAV IRES translation. Subsequently, stalled protein synthesis gives way to viral RNA synthesis that uses the same RNA template as translation, yet in the opposite direction. To better understand the exact molecular mechanisms, in particular the possible involvement of other viral and/or cellular components, further studies are needed. Not depicted in our model is the possibility that truncated PABP sequesters and/or modifies the functional conformation eIF4G, such that ribosome re-initiation at the 5′IRES structure is no longer possible. As—to our current knowledge—the major components needed for HAV IRES-dependent translation are the same as for cap-dependent host translation (), it is likely that the model proposed for cellular translation arrest () holds also true for HAV translation and provides an example for the delicate viral–host interplay to the best of both partners. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Splicing of most polII transcripts is required prior to their export out of the nucleus (). The spliceosome is the macromolecular complex responsible for this intricate process. Early spliceosome assembly events are target of regulatory factors that alter splice site choice and/or modulate splicing activity, constituting a widespread mechanism of gene expression regulation (). Interestingly, processes as diverse as meiosis in yeast, dosage compensation in fruit flies or programmed cell death in humans are regulated in this manner (). Spliceosome assembly is initiated by formation of stable complexes containing U1 snRNP, pre-mRNA and non-snRNP factors (). In yeast, the earliest known splicing complexes are called commitment complexes because their formation targets the pre-mRNA substrate to the splicing pathway (). The mammalian E complex is the counterpart of the yeast commitment complexes (). During the formation of commitment complexes, U1 snRNA base pairs with the intron and exon sequences flanking the 5′ splice site (), whereas the cap-binding complex (CBC) binds to the pre-mRNA cap (). In addition, several protein components of the U1 snRNP make contacts with the pre-mRNA (,). These protein–RNA contacts stabilize pre-mRNA–U1 snRNP interaction and affect 5′ splice site selection (). Like its metazoan counterparts, the yeast U1 snRNP contains two classes of proteins: the Sm proteins shared with the U2, U4 and U5 snRNPs and the U1 snRNP-specific proteins (,). Interestingly, homologs of all three mammalian U1 snRNP-specific proteins (U1-A, U1-C and U1-70K) can be found in the yeast complex (Mud1p, yU1-C and Snp1p, respectively). However, the yeast U1 snRNP contains in addition seven specific proteins (Snu71p, Snu65p, Snu56p, Prp39p, Prp40p, Nam8p and yLuc7p) (,,). Among these proteins, only for Nam8p has been described a mammalian homolog, the apoptotic factor TIA-1 (,). We identified yLuc7p as component of the U1 snRNP by means of biochemical purification (referred to as Snu30p in our previous study) () and through a genetic screen causing synthetic lethality with CBC (). Analysis of yLuc7p mutants revealed that the composition of yeast U1 snRNP was altered in these strains and that the corresponding extracts were unable to support any of the defined steps of splicing unless supplemented with recombinant wild-type yLuc7p (). In addition, splicing of introns with non-consensus 5′ splice site or branchpoint sequences was defective in yLuc7p mutant strains. For reporters containing two competing 5′ splice sites, a loss of efficient splicing of the cap proximal splice site was observed in yLuc7p-deficient cells, analogous to the defect seen in strains lacking CBC. These results lead Fortes () to suggest that the loss of yLuc7p disrupts a U1 snRNP–CBC interaction normally contributing to 5' splice site recognition. Using a combination of RNA–protein cross-linking, and a strategy we called Directed Site-Specific Proteolysis (DSSP), we have now shown that yLuc7p contacts specifically exon 1 of the pre-mRNA through the first of its two zinc finger motifs. Modification of the RNA sequence contacted by yLuc7p affects the pre-mRNA splicing efficiency in a yLuc7p-dependent manner. Our data suggest that interaction of yLuc7p with the upstream exon stabilizes the pre-mRNA–U1 snRNP interaction. This is reminiscent of the mode of action of Nam8p, which facilitates intron recognition by binding to intron sequences following the 5′ splice site (). To assess whether the function of yLuc7p in splice site selection is conserved, we identified putative human homologs and cloned cDNAs encoding two of them (hLuc7A and hLuc7B2, respectively). Both proteins have RE and RS repeats characteristic of splicing factors. In addition, they have two zinc finger motifs similar to those present in yLuc7p and characteristic of RNA-binding proteins. By using antibodies raised against hLuc7A we show that this protein localizes in the nucleus of HeLa cells. Interestingly, these antibodies specifically precipitate U1 snRNA from HeLa extracts. Furthermore, overexpression of hLuc7A in HeLa cells affects splice site selection of an adenovirus E1A reporter, directing splicing selection towards the most 5′ distal site. Our results demonstrate that hLuc7A is a new U1 snRNP-associated splicing factor. In addition, our data support the formation of a wide network of protein–RNA interactions around the 5′ splice site by U1 snRNP associated-factors. This network allows for modulated 5′ splice recognition and contributes thereby to alternative splicing regulation. All yeast strains derive from the wild-type MGD353-13D (). BSY885 (yLuc7p-protA) and BSY761(Nam8p-TAP) were constructed following standard methodology (,). BSY747(yLuc7p-TAP) and BSY593(Nam8p-protA) have been previously described (,). The yLuc7p-HA-TEV-protA fusion inserted in a centromeric LEU2 plasmid was introduced by plasmid shuffling in a strain carrying a disrupted chromosomal copy of LUC7 to give strain BSY1069. An isogenic wild-type strain expressing the yLuc7p-protA fusion was constructed in parallel and named BSY1067. The deletion of the first zinc finger of yLuc7p (fusion yLuc7p-ΔZF-protA) was built following the same strategy, yielding strain BSY1077. All the RP51A derivatives used in cross-linking reactions were synthesized with a specific activity of 2 × 10 Ci/mol as previously described (). The constructs WT-B and 5′SSmut correspond to the plasmids pBS195 and pBS196, respectively (,) and were linearized with EcoRV prior to transcription. In order to obtain constructs Ex5′U (plasmid pBS1983), Ex3′U (pBS1982) and IntU (pBS1985), the sequence including from the T7 promoter to the EcoRV site from plasmid pBS195 was cloned by PCR between the EcoRI and XbaI sites of pUC19. The EcoRV site present in pBS195 was substituted by a PvuII site since this site lacks Ts upstream the site of cleavage and therefore no Us are incorporated upon transcription. These constructs were linearized prior to transcription with PvuII. Plasmid pBS983 () was used to build the reporters used in splicing assays. It contains a synthetic intron inserted upstream of the lacZ coding sequence designed to include unique restriction sites in the intron and flanking exons. Sequences A, G and H (A) were generated by PCR and introduced between the BamHI and KpnI sites of pBS983. Sequence H can potentially form a hairpin structure with Δ = −8.2 kcal/mol at 25°C. Duplications were obtained by cloning the XhoI–SacI fragment from each construct in the SalI–SacI sites of the same plasmids (,,). Constructs AA, AG, AH, GA, GG, GH, HA, HG and HH correspond to the plasmids pOP708 to pOP716. Plasmids overexpressing SF2 and hnRNPA1 in pCGT7 (tagged with a T7 epitope) and the expression reporter for adenovirus E1A were gifts from J. Caceres (). To overexpress hLuc7A in eukaryotic cells, the hLuc7A ORF was subcloned between the BamHI and XbaI sites of pCGT7. All the constructs were confirmed by sequencing. Extract preparation, commitment complex formation and cross-linking analysis were performed as described (). All pre-mRNAs used in cross-linking reactions were internally labeled with αP-UTP, and 4-thio-UTP substituted for cold UTP. (). Wash buffer in immunoprecipitations was IPP150 (10 mM Tris–Cl pH8.0, 150 mM NaCl, 0.1% NP-40). For TEV digestion, after immunoprecipitation the beads were sedimented in a picofuge and 40 µl of TEV cleavage buffer (IPP150 with 1 mM DTT and 1 mM EDTA) was added to the beads. Subsequently, samples were split in two and 1 µl of TEV enzyme (Gibco) was added to one of them. Samples were incubated 30 min at 37°C. Ten microliters of 3× SDS loading buffer was added directly after cleavage and proteins were analyzed by SDS–PAGE. Fragments coding for most of the hLuc7A protein, the N-terminal part containing the region of homology with yLuc7p, or part of the C-terminal extension containing the RS and RE repeats, were inserted in the pGEX2T′-6 plasmid (gift from V. Baldin) yielding plasmids pBS1905, pBS1907 and pBS1909, respectively. All three constructs were expressed in codon plus RIL cells (Stratagene). Rabbits (New Zealand White females) were immunized following standard procedures. HeLa cell nuclear extracts were prepared as described (). Fifty microlitres of beads were washed three times in IPP150 and incubated with 100 µl of serum (0J38 for N-terminal, 0410 for C-terminal) and 300 µl of IPP150 for 2 h at 4°C. The volume equivalent to 10 µl of beads was used for each immunoprecipitation experiment. The beads were sedimented in a picofuge and 75 µl of IPP150 plus 25 µl of nuclear extract were added. After 2 h at 4°C, the beads were extensively washed with IPP150. RNA was recovered and analyzed by primer extension using primers specific for U1 snRNA (CTGGGAAAACCACCTTCGTGATC) and U2 snRNA (AGGACGTATCAGATATTAAACTG). HeLa cells grown to 70% confluence in DMEM medium supplemented with 10% fetal calf serum were transferred to coverslips and incubated overnight to allow for attachment. Cells were washed with PBS, fixed with 2% formaldehyde in PBS for 5 min at 37°C and permeabilized by incubation in 0.1% NP-40 in PBS for 10 min at 37°C. After washing with PBS, cells were incubated in presence of hLuc7A antibody 03J0 (raised against the N-terminal fragment of hLuc7A) at 1:5000 dilution and anti-tubulin (T9026, Sigma) at 1:1000 dilution for 2 h at RT. Cells were extensively washed with PBS and incubated with the secondary antibody (anti-rabbit Alexa 488 and anti-mouse Alexa 568, both from Molecular Probes) at 1:100 dilution for 1 h at RT. Cells were extensively washed with PBS, stained with DAPI (at 5 μg/ml) and coverslips were mounted for observation under a confocal microscope (Axioplan, Zeiss) equipped with a digital color video camera (Leica, Model LEI-750TD). HeLa cells were grown in 60 mm Ø plates to 50% confluence in DMEM medium supplemented with 10% fetal calf serum. Cells were transfected with effectine reagent (Qiagen) using 300 ng of adenovirus E1A reporter, 170 ng of SF2 expression plasmid and 90 or 130 ng of hnRNPA1 or hLuc7A expression plasmids. pBluescript (Stratagene) was added to normalize the amount of DNA. Forty-eight hours post-transfection RNA was purified (Trizol, Gibco) and RT–PCR was performed as described (). yLuc7p has two zinc finger motifs of the types CCCH and CCHH, respectively. These motifs have been shown to act in other proteins as RNA-binding modules. Thus we tested whether yLuc7p binds the pre-mRNA in commitment complexes. We synthesized P-internally-labeled WT-B pre-mRNA, a derivative of the RP51A intron () with 4-thio-U substituted for normal U. This RNA was used to assemble commitment complexes in an extract prepared from a yeast strain expressing yLuc7p fused at its C-terminus with the protein A (yLuc7p-protA, this modification has no phenotypic consequence). Subsequently, the reactions were UV irradiated, treated with RNase T1 and immunoprecipitated by using IgG-coupled beads. Cross-linked proteins present in the precipitate were analyzed by SDS–PAGE and identified by autoradiography. As a positive control we used Nam8p-protA, a protein known to contact the pre-mRNA in the intron downstream from the 5′ splice site (). Nam8p–protA cross-linked to the functional WT-B pre-mRNA (, lane 1). The corresponding signal was strongly reduced when we used the 5′SSmut pre-mRNA (lane 2), which contains a mutation in the 5′ splice site sequence (GUAUU instead of GUAUGU) known to impair commitment complex formation (). , lane 3, shows the pattern of cross-linking obtained when an extract from a yLuc7p–protA tagged strain was used in commitment complex formation with the WT-B pre-mRNA. A protein migrating with the expected size of yLuc7p–protA cross-links to the pre-mRNA. This band was only detected when 4-thio-U substituted RNA was used for cross-linking (data not shown). Further experiments confirmed the identity of this protein as yLuc7p–protA (data not shown). The cross-linking signal is reduced to background level when the 5′SSmut substrate is used (, lane 4) suggesting that it is meaningful in terms of splicing. These results indicate that yLuc7p contacts the pre-mRNA during commitment complex formation. Similar results have been reported for other yeast U1 snRNP proteins such as Nam8p. Strikingly, yLuc7p had not been previously identified in a general analysis as a subunit of U1 snRNP contacting the pre-mRNA (). To determine the region of the pre-mRNA cross-linking to yLuc7p, we used a truncated version of the WT-B pre-mRNA () (A). The pre-mRNA branchpoint region and downstream sequences were found to be dispensable for this reaction, indicating that yLuc7p contacts the pre-mRNA during formation of the first commitment complex (CC1) in the vicinity of the 5′ splice site (data not shown). To narrow down the pre-mRNA region cross-linking to yLuc7p we used derivatives of the WT-B carrying substitutions such that non-essential U residues were replaced by A residues in specific regions of the pre-mRNA () Three substrates covering the exon 1 (the 5′ exon) and the intron upstream of the branchpoint were used for these experiments (A). Each substrate contained essential U residues in the 5′ splice site (GUAUGU). In addition, Ex5′U contained U residues in the 5′ half of exon 1 (nucleotides −42 to −19, where −1 is the nucleotide adjacent to the 5′ splice site, A), Ex3′U contained U residues in the 3′ half of exon 1 (nucleotides −18 to −1) and IntU contained U residues in the intron region (nucleotides +1 to +66) (A). Because 4-thio-U residues are essential for the detection of yLuc7p cross-linking to the pre-mRNA, appearance of a cross-linked species with one of these substrates would identify the region of contact between yLuc7p and the pre-mRNA. Cross-linking to all three substrates would indicate that yLuc7p contacts the pre-mRNA at the 5′ splice site itself. Control experiments demonstrated that these substrates, in the context of wild type, branchpoint and 3′splice site sequences (Materials and Methods section) formed commitment complexes with the same efficiency (Supplementary Figure 1). Commitment complexes were assembled in extracts containing yLuc7p–ProtA, yLuc7p–TAP and Nam8p–TAP (the latter containing fusions of yLuc7p or Nam8p to the TAP tag) () and cross-linking reactions were carried out as before. yLuc7p–TAP cross-linked specifically to the Ex5′U substrate but not to the Ex3′U or IntU pre-mRNA (B, compare lanes 3, 7 and 8). Consistently, yLuc7p-ProtA also cross-linked to the Ex5′U substrate (B, lanes 5 and 6). In contrast, Nam8p–TAP was only cross-linked to the IntU substrate (B, lanes 1 and 2), which was consistent with previous analyses (), and demonstrated that this substrate was competent for cross-linking. In order to confirm the identity of the cross-linked protein in the bands detected, half of the pellets in each cross-linking reaction were cleaved with the TEV protease. This enzyme specifically cleaves the TAP tag () releasing a ∼15-kDa ProtA fragment. As expected, the cross-linked species detected when we used Nam8p–TAP or yLuc7p–TAP tagged extracts were cleaved with the TEV protease (compare lanes 1 with 2, and 3 with 4). However, there was no difference in mobility between TEV-treated and untreated samples when extracts from yLuc7p–protA were used (lanes 5 and 6), indicating that the cleavage observed was specific for the TAP tag and there are no other TEV sites within yLuc7p. Similarly, non-specific background bands were not affected by TEV protease treatment (B, bands indicated with an asterisk). yLuc7p–TAP, yLuc7p–TAP cleaved by TEV protease, and yLuc7p–ProtA migrated at different positions, consistent with their predicted relative molecular weights (compare lanes 3, 4 and 5). These data confirm unequivocally the identification of the cross-linked species as yLuc7p. Moreover, this experiment demonstrates that, during commitment complex formation, yLuc7p contacts the 5′ half of exon 1 in the WT-B pre-mRNA, a region that in this construct is within 23 nt from the 5′ cap. Zinc finger motifs have been shown to act as DNA or RNA-binding modules (). Since yLuc7p has two zinc finger motifs, we wanted to know whether yLuc7p contacts the pre-mRNA through any of its zinc finger motifs. For this purpose, we adapted a strategy originally described for topological studies of membrane proteins (). We reasoned that insertion of a TEV protease cleavage site in a non-conserved loop at the surface of the yLuc7p protein would result in the synthesis of a functional protein. This would then allow DSSP of this polypeptide after cross-linking and identification of the peptide covalently linked to the radiolabeled RNA fragment. We first inserted the sequence coding for an HA epitope tag and a TEV site in regions predicted from phylogenetic comparison to be in a variable loop exposed on the surface of the yLuc7p protein and located between the two putative zinc fingers. These fusions were engineered in the context of a yLuc7p–ProtA fusion and next tested to determine whether they were functional and if they fully complemented a yLuc7p deletion. One such fusion, named here yLuc7p–HA-TEV–protA, was selected for further studies (A). Western blot of extracts obtained from this strain showed that yLuc7p–HA-TEV–protA is stable in yeast cells (B, lane 2). Commitment complex reactions were assembled using extracts derived from this strain and from the parental yLuc7p–protA strain. Samples were UV irradiated, immunoprecipitated with IgG-coupled beads and split in two. One half was treated with TEV protease while the other half was mock-treated. Proteins were analyzed by SDS–PAGE. Both yLuc7p–protA and yLuc7p–HA-TEV–protA cross-linked efficiently to pre-mRNA (C, lanes 5 and 6). No effect was observed after TEV treatment of the yLuc7p–protA sample (C, compare lanes 5 and 9). However, after TEV treatment, yLuc7p–HA-TEV–protA releases a fragment which is cross-linked to the pre-mRNA and has a mobility of ∼18 kDa, corresponding to the size of the N-terminal half of yLuc7p plus the HA tag (C, lane 10). Reproducible cross-linking to the C-terminal half (approximate MW 30 kDa) was not observed even though a weak and diffuse band with slower mobility (possibly a degradation product, shown in with an asterisk) was sometimes detected. This result demonstrates that the N-terminal part of yLuc7p cross-links to the pre-mRNA. Although our results suggest that in these conditions no interaction occurs through the C-terminal half of yLuc7p, due to the presence of the spurious band migrating close to 30 kDa in some of our experiments, we cannot exclude the possibility that the second zinc finger may bind cooperatively aiding in the U1snRNP–pre-mRNA interaction. To further demonstrate that yLuc7p makes contacts with the pre-mRNA through its first zinc finger we generated a deletion mutant of yLuc7p HA-TEV–protA (named yLuc7p-ΔZF), which lacked its first zinc finger (A). Cells expressing this protein as the sole source of yLuc7p were viable, indicating the first zinc finger is not essential. Western blot experiments showed that the protein lacking the first zinc finger was stable in yeast extracts (B, lane 3). Interestingly, cross-linking of the pre-mRNA to this protein was completely abolished (C, lanes 8 and 12). This lack of cross-linking was not due to problems with that particular sample since all samples cross-linked with similar efficiency as shown before immunoprecipitation (C, lanes 1–4). Altogether these results demonstrate that yLuc7p makes contacts with the pre-mRNA through its first zinc finger. Non-conserved sequences surrounding the 5′ splice site have been shown to be very important for efficient splicing (). Given that yLuc7p makes contacts with the pre-mRNA in a non-conserved region, we wanted to know whether the sequence in that region affected splicing efficiency and whether yLuc7p was involved in this process. For this purpose we turned to a sensitive assay based on the alternative choice between two 5′ splice sites competing for a single 3′ splice site (A) (,,). The reporters used in this experiment had two duplicated 5′ splice sites and upstream of each one (nucleotides −15 to −35, A) we introduced three different sequences: a sequence rich in Gs (G sequence), a sequence that could potentially form a hairpin, (H sequence) and a sequence with high content in As (A sequence). The resulting reporters were named using a two-letter code, where the first letter represents the sequence preceding (nucleotides −35 to −15, where −1 is the last nucleotide of exon 1, A) the upstream 5′ splice site, and the second letter represents the sequence preceding the downstream 5′ splice site (A). According to this nomenclature the construct named GH has a G-rich sequence preceding the upstream 5′ splice site and a potential hairpin sequence preceding the downstream 5′ splice site (A). These constructs were introduced into isogenic yLuc7p–protA and yLuc7p–ΔZF strains. Splicing efficiency was then examined by analyzing β-galactosidase activity that reports usage of the upstream versus the downstream 5′ splice site (,,). As shown in B, changing the sequence of the pre-mRNA had important effects in the efficiency of splicing. The G-rich sequence was preferred over the A-rich sequence (10-fold difference, B, compare lanes 9 to 11 and 15 to 17, note the logarithmic scale), and the potential hairpin H-sequence had an intermediate effect (compare, for example, lanes 7, 9 and 11). These effects are lost in the yLuc7p–ΔZF background (B, compare odd with even lanes for each construct) indicating that activation was mediated, at least to a significant extent, through the first zinc finger of yLuc7p. We conclude that the non-conserved sequence contacted by yLuc7p has an effect in splicing efficiency and that yLuc7p, through interaction with its first zinc finger, can modulate this effect. Although these results suggest that the primary factor determining splice site selection in this context is yLuc7p, we cannot exclude contribution of yLuc7p-independent effects to the observed splicing phenotype. Thus it is possible that additional factors, by contacting the pre-mRNA in the same region, could act together with Luc7p to modulate splice site selection. Database searches revealed expressed sequence tags (ESTs) corresponding to three human proteins sharing extended homology with yLuc7p (hLuc7A, hLuc7B1 and hLuc7B2) (). We cloned cDNAs for two of them, hLuc7B2 and hLuc7A (A). Both proteins have two zinc finger motifs similar to the yeast protein. Interestingly, the human proteins have extended C-terminal domains rich in arginine, serine and glutamate (and to a lesser extent lysine and aspartate) as independently noted by others (,). Similar repeats are present in a number of known splicing factors, the SR proteins (). Computer analysis using the cDNA sequence of hLuc7A also revealed several isoforms resulting from alternative splicing (data not shown) (B). N-terminal or C-terminal fragments of hLuc7A were expressed in and used to raise rabbit polyclonal antibodies. Antibodies against hLuc7A recognized a single band in HeLa nuclear extracts migrating in SDS polyacrylamide gels with an apparent MW of 58 kDa (C). The same protein was detected in extracts from a wide variety of human cell lines (lanes 1–5 and data not shown). In addition, faint signals corresponding to shorter proteins were detected with some cells (e.g. lanes 1 and 3). These smaller products may correspond to a splice variant or to post-translational modification (i.e. phosphorylation) or weakly cross-reacting polypeptides expressed from the related hLuc7A and/or hLuc7B2 genes. Antibodies against hLuc7A also cross-reacted with a protein present in egg extracts with similar mobility than hLuc7A (C, lane 7), and weakly with a protein present in S2 cell line extracts which has higher mobility. These results suggest that hLuc7A is expressed in different tissues and that it is present in metazoans from flies to humans. In order to know the subcellular localization of hLuc7A we used the antibodies raised against the N-terminal part of hLuc7A in immunolocalization studies. shows that anti-hLuc7A antibodies recognize a protein that co-localizes in HeLa cells with the DAPI signal for DNA. The hLuc7A signal does not overlap with the cytoplasm stained with an antibody specifically recognizing tubulin. Interestingly, the hLuc7A signal is not uniform in the nucleus. This is consistent with the speckled staining detected for other splicing factors in mammalian cells (). This result demonstrates that hLuc7A is a nuclear protein consistent with a role in pre-mRNA splicing. In yeast, yLuc7p is tightly associated to the U1 snRNP. This allowed purification of U1 snRNP by using a tagged version of yLuc7p (). Purified human U1 snRNP contains only three specific proteins U1-A, U1-C and U1-70k (). However it is possible that some other proteins interacting loosely and/or non-stoichiometrically with the U1 snRNP are not detected after the purification process. In order to know whether hLuc7A associates to U1 snRNP, we performed immunoprecipitation of HeLa nuclear extracts with antibodies against hLuc7A and assayed for the presence of U1 snRNA in the pellets. As positive control we used antibodies against the U1 snRNP protein, U1-A. As negative control an antibody that recognizes POP1, an RNase MRP component, was used. U1-A antibodies precipitated U1 snRNA (, lane 11). However, only background is seen with POP1 antibodies (, lane 10). Antibodies raised against the N-terminal or the C-terminal regions of hLuc7A specifically precipitated a significant amount of U1 snRNA but not U2 snRNA (, lanes 7 and 8). The pre-immune serum precipitated only background levels of U1 snRNA (, lane 9). Analysis of the supernatant confirmed that lack of co-precipitation did not result from RNA degradation (lanes 2–6) Western blot analysis also demonstrated that anti-hLuc7p antibodies co-precipitated specifically the U1-A protein (data not shown). These results show that hLuc7A associates specifically to U1 snRNP despite the fact that it does not copurify with it. hLuc7A can be depleted to 97% from HeLa nuclear extracts by using a mixture of antibodies generated against it. These extracts, when assayed in splicing by using several reporters () are 2- to 3-fold less active (data not shown). To further demonstrate a role for hLuc7A in pre-mRNA splicing, we decided to analyze the effect of overexpression of hLuc7A on splicing of a reporter . We used the system where HeLa cells are co-transfected with an adenovirus E1A reporter and overexpress SF2, which activates splicing from proximal (13s) sites (B) (,). We then analyzed the effect of co-transfected hLuc7A on splicing. As a positive control we co-transfected hnRNPA1, which activates distal sites (9s) and counteracts the effect of SF2. A and D, lane 1, shows the pattern of mRNAs obtained when E1A reporter is transfected with SF2 alone. The 13s, 12s and 9s mRNAs derived from splicing are shown. Co-transfection of hnRNPA1 activated splicing from the distal site detected as a reduction in 13s and a relative increase in 9s (compare A and D, lanes 1–3). hLuc7A overexpression acted in a similar way to hnRNPA1, by preventing reproducibly the splicing from the proximal sites (12s and 13s). Therefore overexpression of hLuc7A switched 5′ splice site utilization towards the more distal site, an effect similar to that produced by hnRNPA1. Taken together, our results demonstrate that hLuc7A is a new splicing factor. Our work provides insight in the role of yLuc7p in splicing, which we find remarkably similar to that of Nam8p. We previously showed () that Nam8p, a U1 snRNP component, binds the pre-mRNA during commitment complex formation in a region directly downstream of the 5′ splice site, where it stabilizes pre-mRNA–U1 snRNP interaction and helps in the formation of commitment complexes. We also showed that changes in the pre-mRNA sequence introduced in the non-conserved region contacted by Nam8p produced a striking Nam8p-dependent effect in splice site recognition and selection. Thus, Nam8p by stabilizing U1 snRNP–pre-mRNA interaction modulates splicing (). We now propose that yLuc7p acts in a similar way. yLuc7p contacts non-conserved sequences in the exon, and in the specific context of our synthetic constructs, in a region within 23 nt from the cap. We also showed that this contact occurs mainly through its first zinc finger. Interestingly, similarly to what happens with Nam8p, changes in the non-conserved pre-mRNA sequence bound by yLuc7p affect splice site selection, and this effect is dependent on the presence of its first zinc finger too. Fortes () showed that extracts from yLuc7p mutant strains display defects in all steps of splicing , and these defects can be rescued by adding recombinant yLuc7p. They also showed that yLuc7p mutants exhibit reduced splicing activity , and that yLuc7p is required for CBC–U1snRNP interactions. Based on our data, and in agreement with the data from Fortes (), we propose a model where yLuc7p contacts the pre-mRNA in a region close to the cap, where it is more likely to establish an interaction with CBC. yLuc7p acts by binding the pre-mRNA upstream the 5′ splice site and stabilizing pre-mRNA–U1 snRNP interaction in commitment complexes. Therefore, yLuc7p could directly or indirectly interact with CBC and mediate the CBC effect in splice site selection. Zhang and Rosbash () showed that another U1snRNP component, U1-C, contacts the pre-mRNA and stabilizes its interaction with the U1snRNA. Altogether, yLuc7p, U1-C, Nam8p, CBC and maybe other U1 snRNP proteins (ySnp1p/U1-70K, SmD1, SmD3 and Snu56p) () would interact with the pre-mRNA and produce a network of protein–RNA interactions keeping the pre-mRNA stably bound to the U1 snRNP. yLuc7p has two zinc finger motifs. We show that the first one cross-links the pre-mRNA and it is required for yLuc7p splicing activity. What is then the role for the second zinc finger? The U1 snRNP particle purified from a yLuc7p mutant strain appears completely disrupted, missing several proteins and it is inactive in splicing (). It is possible that yLuc7p acts as a bridge between the pre-mRNA and the U1 snRNA through its two zinc fingers, the first one binding to the pre-mRNA and the second one binding to U1 snRNA. We tried to generate a yeast strain lacking the second zinc finger but we did not succeed. Diploid cells integrated the mutation but after sporulation none of the spores harboring the deletion were viable. This indicates that the integrity of the second zinc finger is necessary for viability. Perhaps it is required to keep a minimal structure of the U1 snRNP. It is also possible that the second zinc finger contributes to stabilize globally the snRNP–pre-mRNA interaction by cooperatively helping the first zinc finger bind the pre-mRNA. Further experiments will be necessary to clarify this point. Three U1 specific proteins co-purify with human U1 snRNP (U1-A, U1-C and U1-70k) and all three have counterparts in yeast U1 snRNP. In contrast, yeast U1 snRNP has seven additional proteins (Snu71p, Snu65p, Snu56p, Prp39p, Prp40p, Nam8p and yLuc7p). However, a human homolog has been described only for Nam8p, the apoptotic factor TIA-1 (). Here we demonstrate that hLuc7A is a new splicing factor, homolog to yeast yLuc7p. Human hLuc7A is a nuclear protein expressed in several human cell lines ( and ). Importantly, antibodies against hLuc7A specifically precipitate U1 snRNA (), indicating that it is a U1snRNP component. Interestingly, hLuc7A affects splice site selection by activating splicing from the distal 5′ splice site (). Supporting our results, a recent report identified the new SR protein, SRrp53, as a protein interacting with hLuc7A (). In addition, hLuc7A was purified with the supraspliceosome, a macromolecular complex involved in pre-mRNA splicing (). Therefore, we conclude that hLuc7A is a new splicing factor. hLuc7 has three isoforms (hLuc7A, hLuc7B1 and hLucB2) derived from different genes, suggesting that it could be tissue specific or developmentally regulated. ESTs analysis shows that the pre-mRNA for hLuc7A undergoes alternative splicing, although we do not know yet the biological meaning of this variability. It is possible that hLuc7B1 and hLuc7B2 are also regulated by alternative splicing. The abundance of isoforms present in mammalian cells is remarkable when compared to the simplicity of the yeast system. This underscores the degree of complexity in the regulation of splicing in mammalian systems. hLuc7A has a C-terminal tail rich in Asp, Ser and Arg repeats. These repeats are present in the splicing factors known as SR proteins where they have been shown to act as protein–protein interaction modules, or to influence RNA–RNA interactions. The SR repeats in hLuc7A could act by recruiting other splicing factors to the pre-mRNA, or by stabilizing U1snRNP–pre-mRNA interactions. The fact that hLuc7A does not co-purify with human U1 snRNP () indicates that it remains loosely associated to this particle. This situation allows for more flexibility in the regulation of splice site selection. Thus, hLuc7A could first bind the pre-mRNA independently from U1 snRNP and, perhaps through interaction with U1-70k (via their RS domains) or by direct binding to U1 snRNA, could subsequently recruit the U1 snRNP to form the E-complex. In this sense, hLuc7A would act in a similar way than TIA-1, the human homolog of yeast Nam8p (). Interestingly, human and yeast Luc7p share 50% identity in the zinc finger region suggesting that the mechanism of action might be similar. More experiments will be necessary to confirm this hypothesis. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Traces of genetic material preserved within ancient specimens can provide a unique and important real-time record of the past (e.g. ). However, this record is compromized because ancient DNA (aDNA) is invariably damaged and degraded to some extent, initially by endogenous nucleases and microorganisms after death, and subsequently by hydrolysis and oxidation reactions that can fragment the DNA backbone and chemically modify bases (,). Spontaneous base-loss events, creating non-coding abasic sites (,), and certain base modifications (,) can block the amplification of aDNA templates by halting DNA polymerase-mediated primer extension. In contrast, other base modifications can create damage-derived miscoding lesions (DDMLs) which do permit polymerase extension but which have altered base-pairing properties, leading to altered sequences in newly amplified DNA (). Almost all aDNA studies to date have been PCR-based, as this method can generate sequence data from the trace amounts of DNA preserved in ancient specimens. However, PCR can generate incorrect sequence data from aDNA for a number of reasons. In addition to an intrinsic background rate of polymerase misincorporation errors, the altered base-pairing properties of endogenous DDMLs can cause considerable amounts of sequence variation in PCR-amplified aDNA (,). ‘Jumping-PCR’, where partially extended primers switch between different damaged and degraded aDNA template strands during the early cycles of PCR amplification, has been shown to create non-authentic, recombinant, sequences (). ‘Jumping-PCR’ artefacts may also be compounded by the tendency of many DNA polymerases to add a single nucleotide to the 3′-end of primer extensions in a non-template directed fashion (,). PCR can also generate additional, non-endogenous, sequence artefacts such as so-called ‘Type 1 damage’ (). PCR amplification of low copy number templates is known to create products with highly skewed representations (,). This means that sequence artefacts can easily come to dominate the products of PCR-amplified aDNA (,,,,). Sequence accuracy is therefore a major issue in aDNA research. These factors have been recognized to varying degrees. The overlapping ‘best-of-three’ PCR amplification and cloning strategy currently used when key ancient samples are amplified by standard or multiplex PCR (,,), explicitly accepts the inherent shortcomings of PCR-generated aDNA sequences and significantly increases the chances of correctly inferring the original endogenous DNA sequence. However, there are two essential prerequisites for a quantitative investigation into the molecular nature of aDNA damage and its effects on sequence accuracy. First, authentic endogenous DDMLs must be disentangled from other non-endogenous, PCR-generated, forms of sequence variation. Secondly, due to the complementary double-stranded nature of DNA, the template strand-of-origin of particular DDML base modification events must somehow be unambiguously specified. Exponential amplification from both strands of a DNA template is intrinsic to PCR and this prevents the strand-of-origin of particular base changes from being determined (). Together with the demonstrated potential for the generation of additional non-authentic sequence variation, these limitations of PCR-based methods have prevented full resolution of the molecular nature of DDMLs in aDNA. Although there has always been strong theoretical and biochemical evidence that C > U-type DDMLs are a major cause of Type 2 ‘damage’ (C > T/G > A) transitions in PCR-amplified aDNA sequences (e.g. ,,), there has also been considerable debate about the existence, or otherwise, of a genuine biochemical basis for Type 1 ‘damage’ (T > C/A > G) transitions. However, it has recently become clear that so-called Type 1 ‘damage’, observed at significant levels by some traditional PCR-based studies (e.g. ,), disappears once alternative techniques are employed (,; ), and this is now recognized as a non-endogenous, PCR-generated, phenomenon (). The potential role(s) of aDNA templates that are shorter than the target region in PCR amplifications is an issue that requires closer examination. Following the first cycle, only those initial primer extensions long enough to cover the entire target region could be utilized directly by both PCR primers. As we demonstrate however, as the PCR target length increases so does the proportion of shorter, abortive, primer extensions. These have the potential to contribute to the creation of recombinant and other non-authentic sequence artefacts in subsequent cycles. These findings raise questions about the widespread use of quantitative PCR (qPCR) methods to estimate the numbers of aDNA templates contributing to the products of PCR amplification reactions. qPCR results give no information about the number of templates below the target size that end up contributing to amplification products via ‘jumping-PCR’ and other PCR-generated mechanisms. PCR amplification from ancient extracts with template copy numbers estimated by qPCR to be in the tens-of-thousands have been shown to produce significant levels of non-endogenous ‘Type 1 damage’ artefacts (e.g. ). Therefore the widespread assumption that given a sufficiently high estimated starting number, endogenous DDML sequence diversity in aDNA templates will necessarily be reflected by the sequence variation within PCR-generated products simply cannot be sustained. As traditional PCR-based approaches have proven incapable of fully resolving the molecular nature of DDMLs, we have developed a novel SPEX-based approach () to generate detailed information about DDML base modifications in aDNA. In direct contrast to PCR, SPEX is an amplification methodology that specifically targets only one of the aDNA template strands at a locus-of-interest and imposes no predefined target length. This allows the production of first-generation copies of aDNA template molecules, with quantifiable (up to 40-fold or more) coverage from a single reaction. SPEX is shown to be capable of producing sequence data of unprecedented accuracy from aDNA, without the generation of additional, non-endogenous, sequence artefacts over and above a background rate of misincorporation errors common to polymerase-based methodologies. Recently, massively parallel metagenomic sequencing approaches have also been used to investigate aDNA damage. By inferring the sequences of individual single-stranded DNA (ssDNA) templates generated from aDNA via the 454-methodology (), independent studies concluded that in addition to C > U-type DDMLs, a substantial proportion of Type 2 transitions were due to modification(s) of G residues (by an unknown biochemical process) that caused them to be read as A by polymerases (,,). Here, we use SPEX to overcome limitations inherent to both traditional PCR- and current 454-based approaches. The ability of SPEX to rigorously distinguish between authentic aDNA and first-generation copied sequences, whilst simultaneously quantitatively generating highly accurate sequence data from designated loci, has enabled the molecular nature of DDMLs to be fully revealed for the first time. The main body of this study analyses data from SPEX amplification experiments on fifteen aDNA extracts, performed in a dedicated aDNA laboratory at the Henry Wellcome Ancient Biomolecules Centre, University of Oxford. DNA had previously been extracted and analysed for each specimen using well-established aDNA methods (as for ,). Each SPEX analysis focussed on sections of the mitochondrial control region where diagnostic SNPs or sequence ‘fingerprints’ were known to characterize individual specimens. Samples from three mammalian species (bison, human, Eurasian cave lion) were selected to cover a broad range of ages, environments, regions and types of site (Table S2). SPEX experiments using synthetic oligonucleotide templates were carried out independently at the Australian Centre for Ancient DNA (ACAD) in Adelaide. The SPEX strategy for amplification of aDNA is shown schematically in . All polymerase-based methodologies introduce a background level of polymerase misincorporation errors. However, the use of a single primer extension, followed by homopolymer tailing, avoided the potential creation of additional PCR-generated artefacts. Any deviations from the known underlying primary aDNA sequences in the first-generation copies of aDNA were quantitatively analysed. Partially nested SPEX primer sets are shown in Tables S1 and S2. SPEX can access genetic information from highly fragmented and damaged aDNA templates since, unlike PCR, it does not have a pre-defined target size based on the primer pair. SPEX primer extension continues until halted by aDNA template fragmentation or a polymerase-blocking lesion. After the primer extension stage, all aDNA templates were completely removed by Strepavidin bead-washes and the remaining, single-stranded, first-generation copies of aDNA target strands were then permanently ‘trapped’ by polyC-tailing. Nested PCR amplification was then used to amplify what were now effectively ‘modern’ ssDNA template molecules, with minimal risk therefore of subsequent ‘jumping-PCR’ events. A multi-cycle extension variant of SPEX was also examined. Sequence data was generated using SPEX from the following sequence positions for the bison, human and cave lion samples. Bison: equivalent position to 16 178 (single-cycle SPEX) or 16 175 (multi-cycle SPEX) upwards on the mitochondrial genome [Genbank V00654; ()]. Humans: from three parts of the human mitochondrial control region (according to the revised Cambridge Reference Sequence (), Genbank J01415.2); 16 223 upwards; 16 364 downwards and 16 267 downwards. Cave lions: from the equivalent to position 95 onwards on the partial mitochondrial sequence (Genbank DQ899910). Reactions were performed in 50 μl volumes with 1–5 μl of aDNA extract added to reactions comprising: 1 mg/ml rabbit serum albumin (RSA; Sigma) to help overcome polymerase inhibitors; 1× High Fidelity PCR buffer; 2 mM magnesium sulphate; 200 μM of each dNTP; 1.5 Units (U) Platinum DNA Polymerase High Fidelity (Invitrogen) and 0.2 μM of the appropriate 5′-biotinylated SPEX-1 primer (Tables S1 and S2). Synthetic oligonucleotide templates (Figure S3) were used at 0.25 μM. The thermocycling profile was: 95°C for 3°min, 53–57°C (depending on primer) for 1 min, 68°C for 10 min; then 4°C until bead washing. Reactions were performed in 50 μl volumes with 1.0–5.0 μl of aDNA extract added to reactions comprising: 1 mg/ml bovine serum albumin (BSA; Sigma) for non-bison extracts (RSA for bison extracts) to help overcome polymerase inhibtors; 1× AmpliTaq Gold buffer II; 2.5 mM magnesium chloride; 200 μM of each dNTP; 1.5 U AmpliTaq Gold (Perkin Elmer) and 0.2 μM of the appropriate 5′-biotinylated SPEX-1 primer (Tables S1 and S2). The thermocycling profiles were: 95°C for 5 min; followed by 50 cycles of 30 s at 95°C, 30 s at 54–60°C and 1 min at 72°C; then 4°C until bead washing. Aliquots of 20 μl of Streptavidin magnetic beads (New England Biolabs, S1420S) were pre-washed three times with 2× BW buffer (); resuspended in 50 μl 2× BW; mixed with the 50 μl SPEX primer extension reaction and rotated at room temperature for 30 min to immobilize biotinylated molecules to the beads; then a series of washes with 2× BW, 0.15 M NaOH and 1× Tris/EDTA (TE, pH 7.5) were carried out as described by Chen () to remove everything but biotinylated molecules. The beads were resuspended to 14 μl with 0.1× Qiagen buffer EB (10 mM Tris·Cl, pH 8.5). PolyC-tailing was performed for 1 h at 37°C in 20 μl reactions comprising: 15 U of Terminal Deoxynucleotidyl Transferase, Recombinant (rTdT) (Invitrogen, 10 533–065); 1× TdT reaction buffer; 500 μM dCTP; and the 14 μl of resuspended beads. Washes with 1× TE (pH 7.5) were carried out to remove everything but polyC-tailed, biotinylated molecules. The beads were resuspended to 15 μl with 0.1× Qiagen buffer EB. Resuspended beads were used in 50 μl reactions with: 1× High Fidelity PCR buffer; 2 mM magnesium sulphate; 200 μM of each dNTP; 1.5 U Platinum DNA Polymerase High Fidelity and 0.2 μM of the appropriate SPEX-2 (F & R) primers (Tables S1 and S2). The thermocycling profiles were: 2 min at 95°C; followed by 50 cycles of 30 s at 95°C, 30 s at 48–54°C and 1 min at 68°C; with a final extension of 10 min at 68°C. Excess primers and nucleotides were removed with QIAprep PCR purification columns (Qiagen). In order to ensure complete specificity prior to the extensive cloning and sequencing of SPEX amplicons required for quantitative aDNA damage analyses, a second round of partially nested PCR amplifications were performed. However, this additional step could be omitted for general SPEX experiments. Reactions were performed in 25 μl volumes comprising: a 100-fold dilution of the cleaned-up first-round products; 1× High Fidelity PCR buffer; 2 mM magnesium sulphate; 200 μM of each dNTP; 0.75 U ; and 0.2 μM of the appropriate SPEX-3 (F & R) primers (Tables S1 and S2). The thermocycling profiles were: 2 min at 95°C; followed by 35 cycles of 20 s at 95°C, 20 s at 54–57°C, and 1 min at 68°C; with a final extension of 10 min at 68°C. Excess primers and nucleotides were removed as above. All steps followed standard protocols, according to manufacturer's instructions where appropriate, and are described in detail as Supplementary Data. To examine the behaviour of the Platinum DNA Polymerase High Fidelity mix (commonly used in aDNA research) upon completing primer extension to either the physical end of a template molecule, or to an internal abasic site, simplified systems of HPLC-purified synthetic oligonucleotide templates were amplified by single-cycle SPEX and cloned and sequenced as above. These constructs used the same SPEX primers as the single cycle SPEX amplification of bison aDNA to allow a direct comparison between otherwise equivalent regions of ancient and non-ancient DNA. We re-analysed 1449 mitochondrial sequences from a data set of ssDNA molecules produced by the GS 20 Sequencing System (454 Life Sciences, Branford, CT) as previously described (,) to investigate whether Type 2 transitions are randomly distributed across the molecules. A Kolmogorov–Smirnov one sample test was used to test the hypothesis that the positions of 514 C > T and 231 G > A transitions were each distributed according to a uniform distribution along the length of the sequence traces. A -test and Wilcoxon rank sum test were also performed to compare the positions of the C > T and G > A transitions relative to one another, along each trace. These tests allowed us to accept or reject the null hypothesis that, respectively, the mean and median relative positions of the two types of mutation are the same. The summary data from this analysis was compiled and represented using a box-plot (). The distribution of C > U-type DDML events (observed as complementary G > A transitions on SPEX-derived first-generation copied aDNA sequences) was investigated for each of the six ancient bison extracts amplified with single-cycle SPEX using a generalized linear model (Poisson family, log link function) in the R statistical package (): testing for a random distribution of C > U-type events whilst taking into account the lengths of each molecule. We calculated the lack of fit to this model by treating the residual deviance as a χ random deviate with the residual number of degrees of freedom. We calculated (1-) where is the probability of obtaining a random deviate as large, or larger, than that observed. Small values of (1-) suggest a lack of fit of the model because of over-dispersion—a departure from a Poisson distribution such that damaged sites are clustering on particular strands within a sample, even when differences in length are taken into account. Samples that displayed a trend towards over-dispersion were parametrically modelled using the negative-binomial family version of the Generalized Linear Model (). Sections of the mitochondrial control region were amplified by single-cycle SPEX from six bison extracts covering a wide range of ages and environments (Table S2). A ‘total cloned’ data set (TCDS) contained all sequences obtained. A ‘conservative’ data set (CDS; Figure S1) comprised inserts with discrete lengths and/or primary sequences. Differences in SPEX length resulted from differences in the sites of aDNA template fragmentation or polymerase-blocking lesions. Differences in primary sequence reflected contributions from endogenous DDMLs, ‘non-directed’ polymerase activity at the 3′-end following primer extension and polymerase misincorporation errors. Many polymerases can add a single nucleotide in a non-directed manner when primer extension stalls at a non-coding lesion, such as an abasic site (), or halts after reaching the physical end of a template molecule (,). Fifty-two percent of 3′-terminal bases in the CDS did not match the known underlying template sequence (Figures S1 and S2), meaning at least 69% (Table S3) of SPEX events must have undergone non-directed nucleotide addition (NDNA) at the 3′-end. For this reason, bases at the 3′-terminal position (immediately 5′ to the polyC-tail) were excluded from all SPEX aDNA damage analyses. The single-cycle SPEX TCDS covered 10 644 bases from 548 amplicons, while the CDS covered 7654 bases from 337 discrete sequences (). Most CDS sequences were distinguishable on the basis of length (Figure S1). Of those with identical length, the vast majority of these differed due to G > A transitions (i.e. largely from endogenous C > U-type DDMLs on the template strand) and/or NDNA at the 3′-end. Only 5/337 of the CDS sequences differed from one another due solely to a non-C > U-derived transition. Therefore, we estimate that at least 98% of the SPEX amplicon sequences in the CDS were ultimately derived from discrete SPEX events on discrete aDNA template molecules (with <2% reflecting polymerase errors). Single-cycle SPEX amplification appears to be highly robust since, despite subsequent partially nested amplification steps, 548 sequenced clones did not produce a single example of cross-contamination between any of the six individual bison specimens (Figure S1). Multi-cycle SPEX gave essentially indistinguishable results to single-cycle SPEX (). As single-cycle SPEX primer extension is functionally equivalent to the first cycle of PCR, the observed lengths of first-generation copied aDNA were analysed to estimate what proportion would, or would not, have been available for subsequent exponential PCR amplification with a reverse PCR primer placed at defined distances away. When given the opportunity, PCR is known to preferentially amplify shorter targets (), meaning that there is likely to have been at least some drift towards SPEX amplicons representing shorter primer extension events. Nevertheless, a plot of apparent ‘template amplifiable size’ by PCR (i.e. the maximal SPEX-amplified aDNA product length) versus the percentage of all clones for each product length () clearly shows that as the size of the target PCR amplicon increased, the production of directly PCR-amplifiable primer extensions would dramatically decrease compared to the production of shorter, abortive, primer extensions. By analogy to the single-cycle SPEX data, the kinds of sizes targeted in most PCR-based aDNA studies so far e.g. (,,,,,) should also have led to the production of many times more abortive primer extensions than directly PCR-amplifiable ones during the initial PCR cycles. Abortive primer extensions like these would be available to contribute to the creation of recombinant ‘jumping-PCR’ artefacts in subsequent cycles. Moreover, the majority of these abortive primer extension events could be expected to undergo NDNA at their 3′-end. Therefore the potential contribution of molecular events like these to the generation of non-endogenous sequence artefacts in subsequent cycles during traditional PCR-based aDNA studies is an area that needs further investigation. Many polymerases can catalyse the non-templated addition of a single overhanging nucleotide (overwhelmingly A) to blunt-ended double-stranded DNA (dsDNA), following full primer extension to the physical end of a template molecule (,). The requirement for blunt-ended dsDNA is absolute as NDNA does not occur on ssDNA (). On the other hand, most known polymerase-blocking base modifications do not result in non-authentic nucleotides at the 3′-terminal position (). Physical blocks to polymerase extension (e.g. inter-strand crosslinks) should also lead to correctly-paired 3′ nucleotides. The high proportion of positions exhibiting a non-authentic 3′-terminal A (Figures S1 and S2; Table S3) might suggest full primer extension and NDNA following a fairly random and extensive fragmentation of aDNA template molecules. However at abasic sites, polymerases can also ‘non-instructionally’ add A as well as lower levels of G, T or C, nucleotides (). Purine (A or G) bases are known to be released from aDNA at a higher rate than pyrimidines (C or T) to create abasic sites (). If abasic sites played a significant role in NDNA, we would expect to find higher levels of non-authentic 3′-terminal bases opposite A and G sites on the complementary strand. All six bison specimens shared an identical sequence over the first 17 bases of single-cycle SPEX primer extension (Figure S1). Over this region, 62 non-authentic 3′-terminal A, G or T bases were observed opposite the sites of the eight purines on the complementary strand, while only 24 were observed opposite the sites of the nine pyrimidines (Figure S2). (The use of polyC-tailing prevents an analysis of NDNA events involving C.) These findings point towards a significant contribution from abasic sites. A less likely alternative, given the significant levels of non-authentic G and T bases at the 3′-terminal position, is that there is an elevated rate of strand breakage immediately 3′ to purine bases, due to some unknown mechanism. Single-cycle SPEX was used to amplify synthetic oligonucleotide templates (Figure S3) in a detailed analysis of the behaviour of the Platinum DNA Polymerase High Fidelity mix; both at an abasic site and when primer extension reached the physical end of a template. The level of NDNA on aDNA templates (69%) lies between the levels observed at an abasic site (94%) and following full primer extension (50%) in the test system (Table S3). Overall, the NDNA data is therefore consistent with the great majority of primer extension events on aDNA templates being halted due to either template fragmentation or an abasic site, with these being the major factors limiting the effective amplifiable size of aDNA. This evidence contradicts other studies that have argued that crosslinks play the major role (). However, the elevated levels of G and T at the 3′-terminal position with aDNA templates (Figure S2) suggests that the detailed picture may be more complicated than the test system and that other non-coding lesions, as well as abasic sites, may also play a role. The single-cycle SPEX sequences provided no evidence for so-called Type 1 (T > C/A > G) ‘damage’ events (; ). If the previously proposed A > HX (hypoxanthine) lesion () plays a role in aDNA, then significant levels of complementary T > C transitions should have been observed on first-generation copied SPEX sequences. These results concur with the findings of recent large-scale 454-based aDNA studies, which also precluded the possibility of ‘jumping-PCR’-type events, and similarly produced no evidence of Type 1 transition artefacts (,,). G > A transitions (from C > U-type DDMLs on the original aDNA template strand) account for >90% of observed base changes in the single-cycle SPEX sequence data ( and S1; ). The remaining <10% are distributed between the other 11 possible transitions and transversions with an overall level of ∼2.5 × 10 base changes per nucleotide per cycle - an error rate comparable to that observed with Platinum High Fidelity on non-aDNA templates (,). A comparison between the percentage of base changes over the first 17 bases of primer extension, for both bison control region aDNA and an equivalent synthetic oligonucleotide, also strongly suggests that aside from C > U-type DDMLs, base changes in aDNA are at background levels consistent with polymerase misincorporation errors (Figure S4). The multi-fold coverage of first-generation copies from a known strand of origin provided by SPEX clearly suggests C > U-type base modification events are the only significant cause of authentic endogenous DDMLs in aDNA. aDNA damage studies using traditional PCR with either Platinum High Fidelity or AmpliTaq Gold polymerase systems have often produced strikingly different findings (cf. ,,,,). Multi-cycle SPEX was used with AmpliTaq Gold to perform repeated cycles of single primer annealing and extension with: five bison extracts (at one mitochondrial locus); four human extracts (at three mitochondrial loci—each with key identifying SNPs to monitor modern contamination) and three Eurasian cave lion extracts (at one mitochondrial locus). and show the results for all three groups of samples (11 577 nucleotides from 446 discrete sequences for the CDS and 14 051 nucleotides from 1121 independently cloned sequences for the TCDS). All species and loci examined again showed ∼90% of observed base changes were G > A transitions (due to C > U-type DDMLs in template aDNA), with an overall spectrum of base changes from multi-cycle SPEX using AmpliTaq Gold essentially indistinguishable from single-cycle SPEX using Platinum High Fidelity. The spectrum of observed G > A transitions on discrete single-cycle SPEX first-generation copied strands (Figure S1) strongly suggests that particular individual aDNA templates had undergone multiple, clustered, C > U-type DDML event ‘hits’. Statistical analyses confirm that the distribution of G > A transitions is non-random, with three of the four most highly damaged extracts (BS143; BS477; BS569) exhibiting clustering (over-dispersion) of hits onto certain strands independent of sequence length ( = 0.02, 0.09 and 0.01, respectively). The dispersion parameter [θ] of the negative binomial distribution fitted by GLM () reported for all three samples shows low values of θ and relatively narrow standard errors (0.89, SE 0.54; 1.49, SE 1.06; 0.83, SE 0.37, respectively), indicating a better fit than to a Poisson model (which is a special case of the negative binomial, when θ tends to infinity). Further investigation into possible mechanisms for the creation of this intra-molecular clustering of C > U-type DDML base modification events is required. The conclusions about aDNA damage processes reached from the SPEX data directly contradict those reached by two recent large-scale 454-based aDNA damage analyses (,). High-throughput pyrosequencing on the 454 platform can generate sequence data from thousands of individual ssDNA molecules derived from ‘enzymatically polished’, adapter-tagged, DNA (). In contrast to SPEX, 454-based aDNA damage analyses generated both C > T and G > A Type 2 transitions (,). Since the sequence data was generated from ssDNA templates, both studies independently concluded that in addition to C > U-type events, distinct DDMLs must also be causing some G residues to be read as A. To further investigate this apparent contradiction over the existence of endogenous ‘G > A’ DDMLs, we examined whether an inability of the 454 approach to clearly distinguish between regions of authentic aDNA sequence and regions of sequence derived from first-generation copied aDNA might be an issue. The specific combination of the conditions used in the pre-PCR ‘polishing’ steps of 454-based studies so far and the damaged, fragmented, nature of the aDNA template molecules suggested a potential mechanism (Figure S5). A significant proportion of ssDNA starting templates would originally have been double-stranded aDNA templates with 3′ recessed ends, filled in by T4 DNA polymerase () or both T4 DNA polymerase and the Klenow fragment of DNA polymerase I (,). Due to its strong strand displacement activity, the Klenow enzyme would also extend from any single-stranded breaks (SSBs) or nicks with 3′-OH ends within dsDNA (), displacing ‘downstream’ 3′ regions of endogenous aDNA and replacing these with a newly synthesized first-generation copy of the complementary aDNA template strand (Figure S5B). The subsequent ‘nick repair’ step (,,,) by the strand-displacing Bst DNA polymerase would similarly replace 3′ regions downstream of suitable SSB sites in cases where Klenow had not been used (). As seen with SPEX-derived sequence data, first-generation copied aDNA naturally produces high levels of G > A transitions derived from authentic C > U-type DDML events on the template strand (, S1 and S4). If these mechanisms were responsible for creating the G > A transitions observed in 454-based aDNA studies, then there are several explicit, testable, predictions. First, under the model in Figure S5, the 5′ > 3′ direction of DNA synthesis should strongly skew G > A transitions towards the 3′ ends of individual 454-derived ssDNA templates in a highly non-random distribution. Secondly, since authentic aDNA should always be 5′ to newly synthesized first-generation copies of aDNA in enzymatically modified molecules, then all damage-derived C > T transitions should be 5′ to all G > A transitions in any sequence which contained both. On the other hand, genuine DDMLs of G residues should produce a distribution of G > A transitions wholly independent of the distribution of C > T miscoding lesions. Kolmogorov–Smirnov one sample tests on the relative positions of 514 C > T and 231 G > A transitions from 1449 454-derived mitochondrial sequences (Table S4) allowed us to reject the null hypothesis that their relative positions are uniformly distributed along the DNA strand (C > T, D = 0.1904, < 0.001; G > A, D = 0.2014, < 0.001). The two sided -test and Wilcoxon rank sum test performed on the relative 5′ > 3′ positions of both C > T and G > A transitions also allowed us to reject the null hypothesis that the mean and median relative positions of the two types of mutation are the same ( = −10.96, nCT = 515, nGA = 231, < 0.001; W = 32337, nCT = 515, nGA = 231, < 0.001, Wilcoxon test). G > A transitions are skewed towards the 3′ end with a median location of 67.8% from the 5′ end (). As this effect evidently occurred in enough molecules to also similarly skew the distribution of authentic endogenous C > U-type DDMLs (resulting C > T transitions have a median location of 33.6% from the 5′ end; ), a significant proportion of double-stranded aDNA templates must originally have possessed extendable recessed 3′-ends and/or SSBs (Figure S5). An analogous reciprocal skewing of Type 2 transitions is also observed with 454-derived sequences from a Neanderthal specimen (). To graphically illustrate this effect, Figure S6 shows all 17 sequence reads which had both C > T and G > A transitions, but otherwise had a 100% match to the mitochondrial consensus sequence (i.e. no other transitions, transversions, indels, etc). In 16 of these 17 reads, all C > T transitions are 5' to all G > A ones. This result is also significantly non-random: G = 45.73, df = 1, < 0.001. We assume that the single exception to the predicted pattern (1/24 G > A transitions) is due to a polymerase misincorporation error, although this cannot be proven. The well-characterized samples and loci chosen for this study provided a stringent test for SPEX, as any generation of either endogenous DDML or artefactual sequence changes could easily be identified and quantified. Aside from C > U-type DDML events ( and S4), SPEX amplification of aDNA produced both a spectrum and level of sequence differences typical of the background level of polymerase misincorporation errors on non-ancient specimens. SPEX-amplified sequences also provided a simple means to estimate the minimum number of aDNA templates contributing to product sequences, thereby permitting the molecular nature of miscoding and other lesions to be assessed on a quantifiable basis. With standard PCR methods, the lengths of target amplicons are pre-defined by the primer pair. In order to gain as much data as possible, most phylogenetic and aDNA damage studies so far (e.g. ,,,–,,) have tended towards the analysis of amplicons that are larger, sometimes significantly larger, than the smallest possible amplifiable fragments from given extracts (but that are nevertheless ‘reliable’ according to currently accepted aDNA criteria; e.g. ,). makes it clear that unless the PCR primer pair directly abutted one another, then during the initial cycles of PCR-based aDNA studies like these the numbers of primer extensions of directly PCR-amplifiable length would generally be greatly exceeded by the numbers of short, abortive, primer extensions. As Figure S1 and Table S3 demonstrate, the majority of primer extensions also undergo the addition of non-authentic 3′-terminal bases and any of these products could serve as potential protagonists in ‘jumping-PCR’ events in subsequent PCR amplification cycles. Whether these processes play a role in creating PCR-generated sequence artefacts, like so-called ‘Type 1 damage’ (observed at significant levels in several PCR-based studies of aDNA damage; e.g. ,,), is currently unclear and requires further analysis. However, this class of artefact is strikingly absent from SPEX- or 454-derived sequences, where ‘jumping-PCR’ should not be an issue. It has long been observed that analysing the products of a single PCR amplification from aDNA can lead to wholly incorrect inferences about the underlying endogenous sequence (e.g. ,,). One explanation of this phenomenon might be that the absolute number of initial primer extensions of directly PCR-amplifiable length was small or zero (for the particular target size) in amplifications like these. Primer extension steps that created only one or a small number of molecules that traversed both PCR primer binding sites, perhaps containing authentic endogenous DDMLs or polymerase-generated/‘jumping-PCR’ artefacts, could then undergo a form of positive selection and come to dominate the exponentially amplified products (cf. ,). However, aDNA extracts with high estimated copy numbers (according to qPCR) can still generate significant levels of non-endogenous, PCR-generated, ‘Type 1 damage’ (e.g. ). Therefore, perhaps the relative proportions of intact, directly PCR-amplifiable, templates versus fragmented, damaged, templates may be key. Further investigation is required. Until now, a 3-fold redundancy PCR amplification and cloning strategy has been employed to attempt to generate credible consensus sequences from key ancient samples (e.g. ,), but this approach is both labour and sample intensive and has been shown to be fallible even with high-quality, frozen, aDNA templates (). Overall, traditional and multiplex PCR can probably be relied upon to produce correct consensus sequences over the great majority of nucleotide positions in non-human samples by the ‘best-of-three’ strategy, provided that enough suitable aDNA templates are available and appropriate care is taken (e.g. ,,,,). However, despite many years of effort, the inherent features of the methodology discussed above have meant that no PCR-based study has been able to fully resolve the molecular nature of DDMLs. Unlike PCR, single-cycle SPEX synthesizes a first-generation copy from only one of the aDNA template strands, thereby precluding any ‘jumping-PCR’-type mechanisms. Multi-cycle SPEX also did not exhibit any obvious indications of these kinds of artefacts (e.g. repeated characteristic DDML motifs in SPEX amplicons of different lengths or enhanced levels of base changes not due to endogenous DDMLs). Multi-cycle SPEX produced a spectrum of transitions and transversions indistinguishable from single-cycle SPEX (). The linear mode of multi-cycle SPEX primer extension amplification (as opposed to the exponential mode of PCR amplification) meant that the single SPEX-1 primer should have remained at vast molar excess to aDNA targets and extended primers throughout the reaction. The absence of a reverse PCR primer in the multi-cycle SPEX primer extension stage meant there was no potential for the positive selection of ‘jumping-PCR’-type events (). Another potential source of non-authentic sequence diversity is the cloning of heteroduplexes. When 50 or 60 PCR cycles are used (e.g. ,), high levels of exponentially amplified product can drastically reduce primer-to-template ratios during the final cycles, favouring self-annealing of complementary PCR-amplified strands over productive primer-template binding (). With heterologous starting sequences, the subsequent cloning of heteroduplexes has been shown to allow the mismatch repair system to generate further non-authentic sequence microvariation (). As PCR-amplified aDNA is known to have extensive sequence variation, due to both endogenous DDMLs and PCR-generated artefacts, this issue should not be neglected by PCR-based studies. SPEX minimizes potential hetroduplex formation prior to cloning by amplifying a wide range of insert sizes for only 35 cycles. Quantitative damage analyses on both the CDS and TCDS for both single- and multi-cycle SPEX amplified sequences from three separate species support the same two conclusions. First, that Type 1 (T > C/A > G) ‘damage’ transitions are non-endogenous, PCR-generated, sequence artefacts. Secondly, C > U-type base modification events appear to be the only DDMLs present at significant levels in ancient DNA. Comparing SPEX-amplified sequences from bison aDNA and an equivalent synthetic oligonucleotide template also emphasizes that C > U-type DDMLs occur at a remarkably consistent level (∼11–12% per site), regardless of local sequence context (Figure S4). Therefore, the increased accuracy of this quantitative SPEX sequence data provides no support for the DDML ‘hotspots’ inferred by some traditional PCR-based aDNA studies (e.g. ,). Recent 454-based aDNA studies (,,) argued that a currently unknown DDML must be causing some G residues to be read as A during PCR amplification from individual ssDNA templates. The quantitative demonstration of the predicted highly non-random distribution of G > A transitions towards the 3′ ends of 454-derived sequences, coupled with the skewing of C > T transitions towards the 5′ ends, strongly supports the hypothesis that G > A transitions are generated during the pre-PCR ‘enzymatic polishing’ and/or subsequent ‘nick repair’ steps, from C > U-type DDML events on the complementary aDNA strand (Figure S5). Re-interpreted in this way, 454-generated metagenomic sequence data supports the central finding from the SPEX aDNA studies that C > U-type base modification events are effectively the sole cause of authentic DDMLs in aDNA. This identifies significant methodological issues for 454-based aDNA studies, as 454-derived sequence variation does not reflect the authentic underlying pattern of DDMLs in an aDNA extract. The 454 sequence traces contain DDML sequence variation from of the original aDNA template strands (in the form of variable and unquantifiable proportions of 5′ regions of authentic, endogenous, aDNA and 3′ segments of first-generation copied DNA derived from the complementary strand). Until input ssDNA templates can be unequivocally produced from single strands-of-origin in 454-based studies this will remain a key issue. Theoretically, multiple overlapping traces could allow correct consensus sequences to be inferred, enabling Type 2 miscoding lesion transitions (whether observed as C > T or G > A) to be clearly discounted. However, with most aDNA specimens, current 454-based methodologies appear unlikely to regularly generate a sufficient depth-of-coverage to allow the accurate SNP-typing of key sites in this way. SPEX has shown why almost 20 years of PCR-based approaches have not been able to fully resolve the molecular basis of DDMLs. Traditional PCR and current 454-based aDNA studies cannot unambiguously resolve the template strand-of-origin for any particular endogenous Type 2 DDML. Moreover, the production of significant levels of non-endogenous PCR-generated sequence artefacts, such as so-called ‘Type 1 damage’ in some PCR-based investigations (e.g. ,,), clearly demonstrates that any firm inferences and conclusions about authentic endogenous DDMLs from these studies are now questionable. In contrast, PCR-based strategies using the ‘best-of-three’ approach are likely to yield correct consensus sequences most of the time, particularly in studies of well-preserved ancient specimens with reasonably high template copy numbers. The development of the SPEX approach to aDNA has allowed the processes of aDNA damage to be disentangled from PCR-generated sequence artefacts, and revealed the molecular nature of DDMLs. Although much work remains to be done before SPEX could be more widely used in high-throughput situations, a far greater and quantifiable, depth-of-coverage could potentially be achieved compared to other current aDNA methodologies. Sequence data of unprecedented accuracy can be produced from single aDNA target strands with only a single aliquot of extract, a simple system and no specialized equipment. By allowing C > U-type base modification events to be unambiguously identified as the sole significant cause of DDMLs in ancient specimens, SPEX also shows that potential miscoding lesions at key sites could be avoided altogether in future SNP-typing studies by simply targeting the appropriate aDNA strand. This could reduce SNP-typing errors in aDNA studies down towards the theoretical limit of the background rate of polymerase misincorporation errors and, at the same time, introduce quantifiable genotyping from many other kinds of low copy number, damaged, DNA such as forensic, environmental or fixed clinical samples. p p l e m e n t a r y d a t a i s a v a i l a b l e a t N A R O n l i n e . P . B . d e s i g n e d a n d d e v e l o p e d t h e S P E X a p p r o a c h t o a D N A d a m a g e a n d p e r f o r m e d t h e a D N A e x p e r i m e n t s w i t h s u p p o r t f r o m A . C . T h e S P E X - d e r i v e d s e q u e n c e d a t a w a s a n a l y s e d b y P . B . , P . E . , J . J . S . a n d A . C . P . B . p r o p o s e d t h e m e c h a n i s m f o r t h e o r i g i n o f 4 5 4 - d e r i v e d G > A t r a n s i t i o n s . P . E . p r o v i d e d a n c i e n t h u m a n e x t r a c t s , s e q u e n c e d a t a a n d a p p r o p r i a t e t a r g e t S N P s . J . J . S . , P . E . a n d M . B . c a r r i e d o u t s t a t i s t i c a l a n a l y s e s o n S P E X a n d 4 5 4 - d e r i v e d d a t a . R . B . p r o v i d e d a n c i e n t E u r a s i a n c a v e l i o n e x t r a c t s a n d s e q u e n c e d a t a . J . A . g e n e r a t e d S P E X d a t a f r o m s y n t h e t i c o l i g o n u c l e o t i d e t e m p l a t e s . P . B . , A . C . , P . E . J . J . S . a n d M . r o t e t h e m a n u s c r i p t .

Precise transgene expression dosing in mammalian cells has become a cornerstone for synthetic biology (), (pre)clinical gene therapy studies (), drug discovery (,), biopharmaceutical manufacturing () as well as for numerous applications in functional genomic research (). Research and development of inducible mammalian transgene expression systems has focused on clinically licensed small-molecule inducers to minimize pleiotropic side effects in cell culture, animal models or ultimately in clinical trials (). Several antibiotics (), steroid hormone analogs () and immunosuppressive substances (,) have been successfully used to modulate transgene expression and . Immunosuppresive drugs such as rapamycin can trigger transgene expression by conditional heterodimerization of FKBP, fused to a DNA-binding domain, and FRB, fused to a transactivation domain, which form a chimeric transactivor inducing transcription from specific promoters (). Steroid hormone-based systems capitalize on the cytosolic sequestration of nuclear hormone receptors engineered to contain heterologous DNA binding and transactivation domains by endogenous heat-shock protein 90 [Hsp90; ()]. Addition of steroid hormone analogs (for example, tamoxifen) displaces Hsp90 which results in nuclear translocation of the transactivator and induction of cognate promoters (,). Antibiotic-dependent expression systems [tetracyclines (), macrolides (), streptogramins ()] take advantage of engineered prokaryotic repressors which activate or repress synthetic target promoters in an antibiotic-responsive manner (). The prototype tool for heterologous mammalian gene regulation is the tetracycline-responsive expression system (known as the TET system) which consists of an -derived tetracycline-dependent transactivator (tTA, a fusion of the TetR repressor and the VP16 transactivation domain) that binds and activates tetracycline-responsive promoters (P, heptameric TetR-specific operator sites [] linked to a minimal version of the human cytomegalovirus immediate early promoter) (). In the absence of tetracycline antibiotics, tTA binds and triggers P-driven transgene expression. In the presence of tetracycline, tTA- binding is abolished and transcription remains shut down (). Despite solid regulation characteristics, ongoing concerns about long-term side effects of antibiotics (,), as well as the rapid emergence of antibiotic resistance in bacteria exposed to subclinical environmental antibiotic concentrations (), have prevented the use of the TET systems in gene therapy and biopharmaceutical manufacturing scenarios. Despite advantages of using clinically licensed drugs for therapeutic transgene expression fine-tuning, there are ongoing concerns about side effects associated with their prolonged administration in a clinical setting (,,). For example, sustained administration of immunosuppressive drugs increases the susceptibility to bacterial infections (), persistent intake of steroid hormone analogs may trigger different side effects including pancreatitis (,) and continued ingestion of antibiotics compromises the intestinal flora () and increases the risk for the emergence of multidrug-resistant pathogenic bacteria (). Therefore, the design of novel transgene control modalities responsive to side-effect-free trigger molecules remains a current priority (,). In synthesis of activated vitamin H (biotin) as well as biotinylation of target proteins is catalyzed by the dual-function biotin ligase BirA: (i) BirA activates biotin (biotinyl-5′-adenylate) and transfers the biotin moiety to an acceptor lysine within a specific signal sequence (). (ii) Furthermore, intracellular biotin concentrations are feedback-controlled by biotinyl-5′-adenylate which triggers binding of BirA to a specific 40 bp operator site, thereby inhibiting transcription of the divergently oriented A-BFCD operon and preventing synthesis of biotin (). Several synthetic biotinylation signals have been recently described [Avitag signal, GLNDIFEAQIEWHE, target lysine in bold print, ()] and used in combination with BirA for and purification of biotinylated target proteins (), chromatin immunoprecipitation, as well as for detection of protein–DNA complexes (,). Capitalizing on the high affinity of biotin to (strept)avidin [10 M; ()], we have designed a chimeric transrepressor, which is assembled from tTA-Avitag and streptavidin-KRAB (krueppel-associated box protein of the human gene) components by biotin-triggered heterodimerization, and capable of tTA-mediated induction of P in a vitamin H-adjustable manner. Precise transgene expression fine-tuning using a non-toxic vitamin may represent an important advancement for the gene therapy and biopharmaceutical manufacturing communities. (P-tTA-AT-pA) encoding the tetracycline-dependent transactivator [tTA; ()]—Avitag (AT) fusion (tTA-AT) under the control of the simian virus 40 promoter (P) was constructed by PCR-mediated amplification of the VP16 transactivation domain from pWW35 () using oligonucleotides OWW18 (5′-GTACGAATTCCCACCatgccccgccccaagctcaa-3′, annealing sequence in lower case) and OWW428 (5′-GGATCGCGGCCGCTTAcccaccgtactcgtcaattcc-3′, Avitag sequence underlined, annealing sequence in lower case, HindIII restriction site in italics) and subsequent ligation of the VP16-AT module (BssHII/HindIII) into pSAM200 (). (P-SA-KRAB-pA) encoding streptavidin (SA) fused to the krueppel-associated box protein of the human gene (KRAB) was constructed by PCR-mediated amplification of streptavidin from pWW801 [P-TetR-SA-pA, P, simian virus 40 promoter; TetR, tetracycline-dependent repressor; ()] using oligonucleotides OWW434 (5′-GGATCCACCatggctagcatgactggtggac-3′, annealing sequence in lower case, EcoRI restriction site in italics) and OWW433 (5′-GGATCATGGCTGTACGCGGActgctgaacggcgtcgagc-3′, annealing in lower case, BssHII restriction site in italics) and subsequent ligation of the SA module (EcoRI/BssHII) into pWW43 (). (P-IFN-β-pA) harboring beta interferon (IFN-β) under the control of the tetracycline-responsive promoter (P) was constructed by excising IFN-β (EcoRI/HindIII) from pWW430 () and ligating it into pMF111 [P-SEAP-pA, ()], thereby replacing SEAP (human placental secreted alkaline phosphatase). (P-BirA-pA-P-neo-pA, P, human cytomegalovirus immediate early promoter; P, simian virus 40 promoter; neo, neomycin resistance gene) encoding the biotin ligase BirA has been described previously (), in brief, BirA was excised (EcoRI/SpeI) from pGEM-SD2 [Strouboulis,J., unpublished data, ()] and ligated (EcoRI/XbaI) into pMF150 [P-PIP-pA-P-neo-pA ()]. BirA harbors an N-terminal hemagluttinin (HA) tag. The triple-transcript vector pTT-Bio [pWW1091, P-SA-KRAB-P-tTA-AT-P-BirA-pA, P, P, synthetic promoter ()] was constructed in a three-step procedure: (i) SA-KRAB excised from pWW944 (EcoRI/HindIII) was ligated into the corresponding sites of the first multiple cloning site (MCS) of pCF263 [P-MCS-P-MCS-P-MCS-pA, (,)] resulting in plasmid pWW1089. (ii) BirA excised from pWW804 (SpeI/PmeI) was ligated (SpeI/SwaI) into the third MCS of pWW1089 resulting in plasmid pWW1090. (iii) Finally, tTA-AT excised from pWW938 (NotI) was ligated in sense orientation (NotI) into the second MCS of pWW1090 thereby resulting in pTT-Bio. Wild-type Chinese hamster ovary cells (CHO-K1; ATCC CCL-61) were cultivated either in biotin-containing or biotin-free ChoMaster® HTS medium (Cell Culture Technologies, Gravesano, Switzerland) supplemented with 5% fetal calf serum (FCS, PAN Biotech GmbH, Aidenbach, Germany, Cat. No. 3302, Lot. No. P231902) or 10% biotin-free serum replacement (KOSR, Invitrogen, Carlsbad, CA, USA, cat. no. 10828-028), respectively, and 1% of a penicillin/streptomycin solution (Sigma, St. Louis, MO, USA, Cat. No. 4458). Human embryonic kidney cells [HEK293-T ()] were cultivated in Dulbecco's modified Eagle's medium (DMEM) medium supplemented with 10% FCS or 10% serum replacement and 1% penicillin/streptomycin solution. All transfection protocols were optimized for a well of a 24-well plate. : 30 000 CHO-K1 were seeded 12 h before transfection (0.5 ml biotin-free HTS medium supplemented with 5% FCS). An aliquot of 12 µl 0.5 M CaCl solution containing 1.2 µg total plasmid DNA (for co-transfections, equal amounts of individual plasmids were used) were mixed with 12 µl phosphate solution (50 mM HEPES, pH 7.05, 280 mM NaCl, 1.5 mM NaHPO), vortexed for 5 s, incubated for 25 s at room temperature to allow formation of the transfection precipitate and transferred to 0.4 ml ChoMaster® HTS medium containing 2% FCS. The culture medium was replaced by the transfection mixture and the plates were centrifuged for 5 min at 1200 × prior to a 90 min incubation period and a 30 s glycerol shock (0.5 ml 15% glycerol in ChoMaster® HTS medium containing 2% FCS). After washing once in 0.5 ml biotin-free ChoMaster® HTS medium, the cells were cultivated in biotin-free or biotin-containing ChoMaster® HTS medium supplemented with 10% KOSR serum replacement or 5% FCS as indicated. 40 000 HEK293-T were seeded 12 h before transfection (0.5 ml DMEM, 10% FCS). Twenty microliter of 0.25 M CaCl containing 0.6 µg total DNA (for co-transfections, equal amounts of individual plasmids were used) were mixed with 20 µl 2× HBS solution (100 mM HEPES, 280 mM NaCl, 1.5 mM NaHPO, pH 7.1), incubated for 20 min at room temperature to allow formation of the transfection precipitate, which was transferred dropwise to the cell culture and centrifuged onto the cells (5 min at 1200 × ). Transfected cells were incubated for 90 min at 37°C before they were washed once in FCS-free DMEM and then cultivated in DMEM containing 10% biotin-free KOSR. CHO-K1 were first co-transfected with plasmids pWW938 (P-tTA-AT-pA), pWW944 (P-SA-KRAB-pA) and pWW804 (P-BirA-pA, also carrying a constitutive expression cassette conferring resistance to G418) at a ratio of 15:15:1, selected for 12 days in ChoMaster® HTS medium containing 5% FCS and 400 µg/ml G418 and subjected to single-cell cloning. Individual clones, which functionally express pWW938- and pWW944-encoded genes, were then co-transfected with pMF111 (P-SEAP-pA), pWW804 (P-BirA-pA) and pPUR (conferring constitutive resistance to puromycin; Clontech, Palo Alto, CA, USA) at a ratio of 15:15:1 followed by selection for 10 days in ChoMaster® HTS medium containing 5% FCS, 400 µg/ml G418 and 15 µg/ml puromycin and single-cell cloning. Individual cell clones were assessed for biotin-triggered SEAP production and CHO-SEAP was chosen for further studies. BirA (HA-tagged) and tTA-AT (TetR-VP16-Avitag) were quantified by resolving cleared cell lysates on a 10% SDS–polyacrylamide gel followed by western blot analysis using anti-HA-tag (Santa Cruz Biotechnology, Santa Cruz, CA, USA, Cat. No. sc-805) or anti-VP16 (Santa Cruz, Cat. No. sc-7545) as primary antibodies for detection and horseradish peroxidase-couple anti-IgG antibodies for chemiluminescence-based visualization (ECL plus, GE Healthcare, Piscataway, NJ, USA, Cat. No RPN2132) with a Chemilux CCD camera (Intas, Göttingen, Germany). Chemiluminescence signals were indicated as integrated optical density (IOD). Streptavidin-KRAB was quantified by incubating cleared cell lysate (40 µl) with 1 µg/ml FITC-biotin (60 µl) (Fluka, Buchs, Switzerland, Cat. No. 53608) for 30 min before measuring fluorescence intensity. Fluorescence of FITC-Biotin bound to streptavidin is quenched and can be used to quantify streptavidin (). Fluorescence quenching is indicated in relative fluorescence units (RFU). SEAP production was quantified as detailed by Schlatter and coworkers () In brief, the cell culture supernatant, collected 48 h after seeding/transfection, was heated to 65°C for 30 min to inactivate endogenous phosphatases and then centrifuged for 2 min at 14 000 × to remove cell debris. An aliquot of 80 µl supernatant were mixed with 100 µl SEAP buffer (20 mM homoarginine, 1 mM MgCl, 21% (v/v) diethanolamine, pH 9.8) and the reaction was started by addition of 20 µl pNPP (120 mM p-nitrophenylphosphate in SEAP buffer) and the light absorbance timecourse scored at 405 nm. SEAP levels were calculated according to Lambert-Beer's law using the slope of the timecourse and the specific absorption coefficient of the reaction product p-nitrophenolate ( = 18 600 cm M). IFN-β production was profiled by ELISA (PBL Biomedical Laboratories, Piscataway, NJ, USA, Cat. No. 41400-1A) and the cell number was determined either by a Casy™ Counter TTC (Schärfe System GmbH, Reutlingen, Germany) or by using a colorimetric WST-1-based assay (Roche Applied Science, Rotkreuz, Switzerland, Cat. No. 11644807001). All experimental data represent average values derived from three independent experiments with SDs indicated by the error bars. Cells were cultivated in a BioWave® 50SPS bioreactor (Wave Biotech AG®, Tagelswangen, Switzerland) equipped with Wave Bag® 2LOpt for optical pH and DO control of the 1 l culture. The bioreactor was operated at a rocking rate of 16 min, a rocking angle of 5° and an aeration rate of 50 ml/min with inlet gas humidification to prevent evaporation of the medium (HumiCare® 200, Gruendler Medical, Freudenstadt, Germany). Two milliliter samples were withdrawn at the indicated points in time. Samples for assessment of SEAP production were centrifuged (2 min, 14 000 × ) and the supernatant was stored at −20°C until analysis. Cell density was quantified using the colorimetric WST-1 kit (Roche Applied Science, Rotkreuz, Switzerland, Cat. No. 11644807001). CHO-K1 transgenic for pWW938, pWW944, pWW804 and pMF111 were encapsulated in 400 µm alginate-PLL (poly--lysine)-alginate beads using the Inotech Encapsulator Research IE-50R (Inotech Biotechnologies Ltd., Basel, Switzerland) according to the manufacturer's protocol and at the following specific settings: 0.2 mm nozzle, 20 ml syringe at a flow rate of 315 units, nozzle vibration frequency 1108 s, voltage for bead dispersion 800 V. Female OF1 (oncins france souche 1) mice were obtained from Iffa-Credo and kept on a biotin-free diet (Société SAFE, Augy, France) for two weeks, whereas the control mice were fed the same diet supplemented with 4 mg/kg biotin. Biotin deficiency was monitored using the forced swim test (). 700 microliter of ChoMaster® HTS medium containing 50% capsules (2 × 10 cells, 200 cells/capsule) were injected intraperitoneally into the mice. When indicated, biotin was dissolved in 0.9% NaCl and administered by intraperitoneal injection 1 h after capsule implantation. Blood samples were collected retroorbitally and serum was produced using microtainer SST tubes (Beckton Dickinson, Franklin Lakes, NJ, USA, Cat. No. 365968). Serum of control mice (implanted with capsules and kept with or without biotin supplementation) did not show any detectable SEAP levels. All experiments involving animals were conducted according to European Community legislation (86/609/EEC), and have been approved by the French Republic (No. 69266310) and performed by M.D-E. at the Institut Universitaire de Technologie, IUTA, F-69622 Villeurbanne Cedex, France. D(+)-biotin (Acros Organics, Geel, Belgium, Cat. No. 23009) was prepared as a 10 µM stock solution in HO and was typically used at a final concentration of 100 nM unless stated otherwise. At concentrations above 10 µM, biotin was dissolved as stock solution in DMSO. Avidin (Pierce, Rockford, IL, USA, Cat. No. 21121) was prepared as a 100× stock solution in phosphate-buffered saline (PBS; 20 mM NaHPO, 150 mM NaCl, pH 7.2). The concentration of avidin was expressed as biotin-binding capacity. Tetracycline (Sigma, Cat. No. T7660) was prepared as a 1 mg/ml stock solution in HO and used at a final concentration of 2 µg/ml. For determination of biotin stability, ChoMaster® containing 100 µM biotin and 5% FCS was incubated for one week at 37°C and biotin samples were quantified daily using a HABA (2-(4′hydroxyazobenzene) benzoic acid)-based biotin quantification kit (EZ biotin quantification kit, Pierce, Rockford, IL, USA Cat. No. 28005). No significant decrease in biotin concentration could be observed. In pigs, biotin shows a half-life time of up to 22 h (). In order to enable precise expression dosing by vitamin H, we have converted proven tetracycline-responsive expression system into a vitamin H-adjustable transcription control modality (A). In the presence of biotin (+Biotin) constitutive expression of the biotin ligase BirA [pWW804; P-BirA-pA; (,)] biotinylates tetracycline- [tTA, TetR-VP16; ()] dependent transactivators, which have been fused to the synthetic biotinylation signal (Avitag, AT; (); pWW938, P-tTA-AT-pA, C-i) thereby promoting biotin (B)-dependent heterodimerization with the krueppel-associated box protein of the human gene (KRAB; SA-KRAB; pWW944, P-SA-KRAB-pA) which has been fused to streptavidin (SA) (A and C-ii). Since KRAB-mediated silencing overrides the VP16-based transactivation capacity of tTA-AT, the chimeric transrepressor (tTA-TA-B-SA-KRAB) binds and represses the established tetraycline- (P; pMF111, P-SEAP-pA) responsive promoter (C-iii). At high biotin concentrations (C-iv), all avitag and streptavidin sites are expected to be saturated by free biotin thereby competitively inhibiting SA-AT heterodimerization and de-repressing target gene expression. Biotin deprivation (C, −Biotin) prevents or abolishes heterodimerization of chimeric transrepressors thereby enabling classic tTA-mediated transgene expression which can be repressed by addition of regulating tetracycline antibiotics (C, +TET) independent of biotin availability. Expression of all biotin-responsive regulatory genes has been validated by western blot analysis or fluorescence quenching (B). Vitamin H-controlled transgene expression was validated by co-transfection of CHO-K1 with pWW804 (P-BirA-pA), pWW944 (P-SA-KRAB-pA), pWW938 (P-tTA-AT-pA) and pMF111 (P-SEAP-pA) followed by SEAP quantification 48 h after cultivation in the presence or absence of biotin and the regulating antibiotic tetracycline (A). Cells exhibited high-level SEAP expression in the absence of regulating agents and repressed transgene expression in the presence of biotin or tetracycline. The biotin-regulated gene expression system also mediated conditional SEAP expression in human embryonic kidney cells (HEK293-T) (B) and both vitamin H and tetracycline could be used to control the production of the multiple sclerosis therapeutic beta interferon (IFN-β; pWW732, P-IFN-β-pA) (C). Specificity of the biotin-controlled expression system was assessed by selective omission individual components including (i) the biotin ligase BirA (pWW804), (ii) SA-KRAB (pWW944), (iii) BirA and SA-KRAB (pWW804, pWW944), (iv) BirA, SA-KRAB, tTA-AT (pWW804, pWW944, pWW938) or (v) by replacing the avitag-containing tTA-AT (pWW938) by the original tTA (pSAM200, P-tTA-pA). Exclusion of either BirA, SA-KRAB or Avitag abolished biotin-responsive transgene repression while tetracycline-responsive expression control remained intact (D). In the absence of either tTA-AT or tTA no SEAP expression could be observed indicating that all components are essential and sufficient for biotin-controlled gene expression (D). In all of these experiments, SEAP production reached identical maximum levels in the absence of biotin and tetracycline, indicating that the AVITAG fusion does not impact tTA function (tTA-AT, pWW938, 2.9 ± 0.3 U/l; tTA, pSAM200, 2.7 ± 0.1 U/l). To provide biotin-controlled gene regulation in a easy-to-handle two-vector format (), we have cloned all components (SA-KRAB, tTA-AT, BirA) into a triple-transcript expression format, in which the first cistron (SA-KRAB) is transcribed from a simian virus 40 promoter (P) while subsequent cistrons two (tTA-AT) and three (BirA) are under control of compact synthetic promoters (P) (,) (pTT-Bio; P-SA-KRAB-P-tTA-AT-P-BirA-pA). Co-transfection of pTT-Bio with the response vector pMF111 into CHO-K1 cells resulted in high-level SEAP expression in the absence of biotin while addition of biotin and/or tetracycline shut off SEAP production (E). Exposure of CHO-K1 containing components mediating biotin-controlled SEAP expression (pWW938, pWW944, pWW804 and pMF111) to increasing vitamin H concentrations followed by scoring of SEAP production 48 h after transfection revealed adjustability over 2 logs of biotin concentrations. Maximum SEAP production occurred below 1 nM of biotin and SEAP repression to background levels occurred at 100 nM of biotin (A). Biotin-dependent expression kinetics were assessed by cultivating CHO-K1 cells transfected with plasmids pWW938, pWW944, pWW804 and pMF111 for 68 h in the presence (100 nM) or absence of biotin which resulted in linear induction profiles between 5 and 53 h after transfection (B). In order to validate biotin-responsive expression technology in a prototype gene therapy scenario, we microencapsulated CHO-K1 engineered for biotin-controlled production of the human model glycoprotein SEAP into coherent alginate-poly--lysine-alginate capsules and injected them intraperitoneally into mice kept on a biotin-free diet. After implantation, the mice were injected with either biotin (100 µg/ml) or doxycycline (100 mg/kg) and the serum levels of SEAP were profiled after 48 h. SEAP profiles confirmed maximum product protein production in the absence of biotin and repressed protein production in the presence of biotin or doxycycline (A). Relative regulation profiles were similar compared to identical subcultures maintained suggesting that the biotin control system retains its regulation characteristics (B). The higher basal expression in the presence of doxycycline or biotin (A) compared to (B) may result from lower bioavailability of the inducer . However, since biotin- and doxycycline-mediated and repression profiles are comparable, the regulation performance of both systems may be considered equivalent. Extensive characterization of the biotin-responsive transcription control system for biopharmaceutical manufacturing requires availability of a stable cell line for long-term studies. We have therefore constructed a stable CHO-K1-derived cell line, which is tetratransgenic for pWW938 (P-tTA-AT-pA), pWW944 (P-SA-KRAB-pA), pWW804 (P-BirA-pA) and pMF111 (P-SEAP-pA) and produces SEAP in a biotin and tetracycline-responsive manner. Five out of 40 stable cell clones were cultivated for 48 h in biotin- and/or tetracycline-containing medium and profiled for SEAP production. All cell clones showed high SEAP production when grown in the absence of regulating molecules (−biotin, −tetracycline), while either addition of biotin or tetracycline abolished SEAP production (A). The different expression levels observed for individual cell clones may result from random chromosomal integration and transcription-modulating impact of adjacent genetic elements. Likewise, transgene expression differences also exist between stable and transiently transfected populations, which produce the product protein from expression vector episomes (,). Since CHO-SEAP clone no. 1 (CHO-SEAP) showed the best regulation profile it was chosen for further characterization. Expression stability of CHO-SEAP was confirmed by continuous cultivation of the cell line for 90 days, which did not reveal any statistically significant decrease in SEAP production (day 0, 0.67 ± 0.5 U/l; day 90, 0.65 ± 0.6 U/l). Biotin-responsive SEAP production kinetics for which CHO-SEAP was seeded in biotin-free or biotin-containing medium ( = 0) revealed a steady increase over time when cultured in the absence of an inducer, while biotin addition resulted in background SEAP levels only (B). Reversibility of biotin-responsive gene expression induction was assessed by cultivating initial CHO-SEAP populations in either biotin-free or biotin-containing medium. After 31 and 100 h of cultivation, the cell population was split and the biotin status maintained or reversed as indicated. Sequential induction and repression of SEAP production was followed over a one-week period (C). Biotin-controlled SEAP expression was reversible and the induction or repression kinetics were not influenced by the biotin cultivation history of the CHO-SEAP population (C). The dose-response profiles recorded over 7 biotin concentration logs revealed unique expression characteristics: (i) gradual ON-to-OFF expression profiles between 1 and 100 nM, reflecting increasing heterodimerization of transsilencer and transsilencer-mediated repression of tTA-AT-driven P transcription. (ii) Full OFF state at biotin concentrations ranging from 100 to 1000 nM, reflecting maximum transsilencing capacity. (iii) graded OFF-to-ON expression at biotin levels above 100 µM, suggesting that free biotin levels exceeding the ligation capacity of BirA may bind and sequestrate SA-KRAB thereby competitively inhibiting its heterodimerization with biotinylated tTA-AT (D). While all aforementioned experiments were conducted in biotin-free medium, standard cell culture media and serum additives contain biotin at an average level of 50 nM which fully represses the biotin control system (D). Dose-response characteristics of CHO-SEAP cultivated in standard serum-containing cell culture medium and exposed to increasing biotin concentrations resulted in a gradual increase in SEAP production, confirming that the biotin control modality operates as an inducible expression system under standard culture conditions (A). Furthermore, expression kinetics in the presence of 2000 µM biotin increased steadily, whereas the absence of additional biotin resulted in background expression only (B). A further strategy to induce the biotin control system is to sequester repressing biotin in standard medium by addition of avidin. Cultivation of CHO-SEAP in the presence of increasing avidin concentrations resulted in a gradual increase in SEAP production, thereby enabling adjustable transgene expression by administration of an exogenous protein (C). Expression kinetics in the presence or absence of avidin increased steadily over time (D). However, the overall regulation performance using avidin as an inducer was lower compared to biotin-triggered expression control (A and C), suggesting that avidin is unable to fully compete with the thermodynamically favored enzymatic biotinylation of tTA-AT. Despite intensive research no significant side effects have been discovered for biotin, neither following oral uptake nor after injection (). Therefore, this FDA-licensed molecule represents an ideal inducer for any bioprocess application requiring precisely timed expression control (,,). We have therefore evaluated the compatibility of vitamin H-responsive gene expression with standard bioreactor operation by cultivating CHO-SEAP for 53 h in a BioWave® bioreactor using 1 l of standard biotin-containing (non-supplemented) culture medium before adding 2 mM biotin to trigger SEAP production. While CHO-SEAP grew exponentially during the entire period, SEAP expression remained silent until shortly after addition of biotin, thereby validating the suitability for timely controlled biotin-responsive gene expression in a prototype biopharmaceutical manufacturing scenario (). Adjustable transgene expression technology is fundamental to prototype gene therapy scenarios (,,), functional genomic research (,), synthetic biology (,) and forefront biopharmaceutical manufacturing strategies (,,,). The novel vitamin H-responsive expression technology exhibits several advantages regarding standard operation and regulation performance compared to established systems which are typically triggered by side-effect-prone immunosuppressants, hormones or antibiotics (,,,,): Vitamin H (biotin) is the first inducing agent with no or only low known side effects (e.g. to our knowledge, no LD has ever been described). As biotin is a natural component of cell culture media, its removal or supplementation to native levels is not expected to require additional validation or licensing efforts for biopharmaceutical manufacturing as would be the case for pharmacologically active inducers such as hormones, immunosupressors and antibiotics. When comparing biotin-adjustable regulation characteristics to classic expression systems, biotin-adjustable transgene control exhibits unique band pass filter-like expression characteristics using a single inducer molecule: (i) within the low biotin concentration window (0–2 nM), transcription is fully induced, (ii) it is maximally repressed between 100 and 1000 nM biotin and (iii) again entirely turned on beyond 100 µM. This exclusive feature enables sequential induction, repression and re-induction using increasing doses of vitamin H, an expression pattern which has so far only been achieved in bacteria using a sophisticated gene network (). The unique biotin-dependent band pass filter characteristic could be explained by a competition of tTA-AT and SA-KRAB for free biotin. At low biotin concentrations (0–100 nM), binding of biotin to tTA-AT [via the intermediate biotin–BirA complex, ()] is thermodynamically favored [Δ = −360 kJ/mol for an amide bond (in trans) versus Δ = −80 kJ/mol for the non-covalent binding of biotin to SA-KRAB (,)], resulting in biotinylated tTA-AT and subsequent transcriptional shutdown after binding to SA-KRAB. At high biotin concentrations (>100 µM) free biotin is expected to saturate all tTA-AT and SA-KRAB biotin-binding sites thereby preventing dimerization with biotinylated tTA-AT. While basic research scientists may appreciate the extended regulation bandwidth emulating developmental expression patterns (,), the biopharmaceutical manufacturing could welcome the first opportunity to reverse the expression state of difficult-to-produce protein therapeutics without media exchanges and filtration operations. Owing to the use of TetR as the DNA-binding component, the biotin-responsive gene expression technology is compatible with standard tetracycline-responsive promoters, established cell lines and mouse strains (,,,) and is unique in integrating two input signals: biotin and tetracycline. This dual-input sensitivity permits override of the biotin-controlled transgene expression by administration of tetracycline which may represent a useful safety feature in future cell engineering strategies. Since biotin is a physiological component the synthetic control modality may be used to convert endogenous biotin concentrations into a therapeutic transgene response. Scenarios to treat biotinidase-associated metabolic disorders by triggering therapeutic biotinidase expression at low pathologic biotin concentrations and shutting it off when physiologic biotin concentrations are reached, could now become a reality (). Despite the ubiquitous presence of biotin in living systems, interference-free transgene control can still be achieved by (i) addition of biotin-sequestrating avidin or by (ii) administration of excess biotin (>1000 nM) thereby taking advantage of the high-biotin regulation window of the aforementioned band-pass filter. Considering all of the design and performance characteristics in mammalian cells, mice as well as during standard bioreactor operation, we are convinced that biotin-controlled transgene expression will foster advances in basic research, gene therapy scenarios as well as in biopharmaceutical manufacturing.

Helicases are ubiquitous enzymes that harness the energy of nucleotide hydrolysis to translocate on a nucleic acid strand. Helicases participate in most transactions involving nucleic acids, including DNA replication, recombination, repair, transcription, protein synthesis and ribonucleoprotein assembly and remodeling (,). Pif1p is the founding member of a DNA helicase family conserved from yeast to humans (,). The helicase was originally discovered by its effects on maintenance of yeast mitochondrial DNA (,). In the absence of Pif1p, yeast cells lose mitochondrial DNA at high rates, which generates respiratory-deficient (petite) cells. Pif1p was rediscovered by its nuclear role in the synthesis of telomeric DNA by telomerase (,), the specialized reverse transcriptase that maintains chromosome ends in most eukaryotes (). The nuclear and mitochondrial isoforms of Pif1p, which originate from the same mRNA, are generated by differential codon start usage (,). The effects of the nuclear isoform of Pif1p on telomerase have been extensively studied, both and , Pif1p inhibits telomerase-mediated telomere lengthening and telomere addition at double-strand breaks (,,,). addition is rare in wild-type cells, but is increased 600–1000-fold in Δ cells (,). Telomere length is inversely related to Pif1p levels: reduced Pif1p results in long telomeres, and Pif1p over-expression results in modest telomere shortening (,). The effects of Pif1p on telomere length and telomere addition are telomerase dependent (,,,), suggesting that Pif1p affects telomerase directly. In support of this possibility, assays show that Pif1p removes telomerase from DNA ends, which limits telomerase nucleotide addition processivity (). The role of Pif1p in the nucleus is not limited to telomeres. deletion has a modest effect on ribosomal DNA (rDNA) replication by inhibiting fork movement past the replication fork barrier (RFB) (). Pif1p may have a more general role in chromosomal replication as its deletion suppresses the lethality associated with loss of the helicase/endonuclease Dna2p, which functions in Okazaki fragment maturation (). Pif1p may also function to resolve recombination intermediates as Pif1p over-expression suppresses the Sgs1p-induced damage that occurs in the absence of the topoisomerase Top3p (). Understanding how Pif1p functions in the cell requires a detailed analysis of its enzymological properties. Several characteristics of the enzyme have already been reported. Using an enzyme purified from yeast mitochondria or a recombinant protein produced in insect or bacterial cells, Pif1p was shown to be a distributive 5'–3' DNA helicase (,,,), whose activity is stimulated by forked substrates (). Pif1p is also able to unwind RNA–DNA hybrids if the substrates contain a 5' single-stranded DNA overhang but does not unwind RNA–RNA hybrids (). Here we show that Pif1p preferentially unwinds RNA–DNA hybrids, and that the enzyme is synergistically stimulated by the presence of a forked structure. Recombinant Yeast RPA purified from was a generous gift from P. Sung (). The nuclear form of yeast Pif1p fused at its N-terminus to a 6-histidine tag was purified essentially as described (). Briefly, the DNA coding for the nuclear form of Pif1p (amino acids 40–859) was cloned into vector pET28 (Novagen) and transformed into Rosetta strain, which is a derivative of the BL21(DE3) strain (Novagen). Protein expression was induced at 23°C by addition of 1 mM IPTG to a mid-log phase culture in rich media, and the culture was further grown for 16 h. Pif1p was purified by affinity chromatography to the 6-histidine tag, followed by cation exchange chromatography on a resource 15S column (GE Healthcare). Fractions containing pure Pif1p as assessed by Coomassie staining were pooled, concentrated and stored at –80°C in 25 mM HEPES pH 7.0, 100 mM NaCl, 25 mM (NH4)SO, 25 mM Mg(OAc), 1 mM DTT, 50% glycerol. Oligonucleotides substrates were synthesized and purified by high pressure liquid chromatography by Integrated DNA Technology (IDT). These oligonucleotides were designed as random sequences containing a high GC content and unable to form stable secondary structures. For formation of the oligonucleotide substrate, the top (short) strand was labeled with γ-P-ATP using polynucleotide kinase (New England Biolabs), according to standard protocols. Top and bottom strand oligos were mixed at equimolar ratios and annealed by slow cooling from 95 to 25°C in 10 mM MgSO, 10 mM Tris pH 7.5. Substrates were subsequently gel purified and stored in 10 mM Tris pH 8.0 at −20°C. Concentration of each substrate was calculated based on the specific activity of the labeled oligonucleotide. Duplexes were prepared as described above for helicase substrates but in the following buffer: 0.15 M NaCl, 5 mM MgCl, 10 mM cacodylate pH 7.0. Thermal melting profiles were determined in a computer-driven AVIV 14DS spectrophotometer equipped with a thermoelectrically controlled cell holder for cells of 1 cm path length. Melting profiles were obtained by measuring absorbance at 260 nm every 0.5°C over the indicated temperature range. values (±0.5°C) were determined from the midpoint of the transition (inflexion point). All of the melting profiles were measured against a solvent blank. Helicase assays were carried out by incubating various amounts of Pif1p and 1 nM nucleic acid substrate at 25 or 35°C, as indicated. Unless otherwise noted, standard reaction buffer was 20 mM Tris pH 7.5, 50 mM NaCl, 100 µg/ml bovine serum albumin, 2 mM DTT, 5 mM Mg and 4 mM ATP. For kinetic studies, reactions were started by addition of ATP. 5 µl aliquots were withdrawn at indicated times and the reactions stopped by addition of 2 µl quenching/loading buffer (6% Ficoll, 13 mM EDTA pH 8.0, 300 nM unlabeled top strand, 0.05% bromophenol blue and 0.05% xylene cyanol). Reaction products were loaded on a 12% polyacrylamide (20:1 acry:bis-acrylamide ratio) non-denaturing gel and resolved by electrophoresis at 4°C and 10 V/cm in TBE 1× buffer. Gels were dried and scanned with a storm PhosphorImager (Molecular Dynamics) and quantified using ImageQuant software (GE Healthcare). Data were fitted using PRISM software (Graphpad Software) to the single order exponential law () = (1 − ), where describes the amplitude of the reaction and the apparent turnover rate of the reaction. For single cycle kinetics, the reaction was started with ATP and 500 nM single-stranded 36 bp trap DNA oligonucleotide of the following sequence: 5′-CGTAATCATGGTCATAGCTGTTTCCTGTGTGAAATT. Pif1p was incubated at various concentrations (ranging from 5 nM to 1 µM) in the presence of 50 pM of the indicated γ-P labeled substrate at 25°C in the following buffer: 20 mM Tris pH 7.5, 10 mM NaCl, 100 µg/ml bovine serum albumin, 2 mM DTT and 5 mM Mg, ±4 mM AMPPNP (Sigma) or ATP-γS (Fluka). After 30 min incubation at room temperature, binding reactions were supplemented with loading buffer (6% ficoll, 0.05% bromophenol blue and 0.05% xylene cyanol) and resolved by electrophoresis at 4°C on a 6% (40:1 acry:bis-acrylamide ratio) non-denaturing gel run at 10 V/cm. Gels were dried, scanned and quantified as described above. Percentage of binding was calculated as the ratio between bound and (bound+unbound) radiolabeled substrate. Five micromolar of Pif1p was incubated at 4°C for 3 h in a buffer containing 10 mM Tris pH 7.5, 50 mM NaCl, 4 mM MgCl and 5% glycerol. Aggregated protein was removed by centrifugation at 23 000 for 30 min. The supernatant was loaded on a 5 ml 15–40% glycerol gradient equilibrated with 10 mM Tris pH 7.5, 50 mM NaCl, 4 mM MgCl and centrifuged for 12 h at 151 693 g in a sw55ti rotor (Beckman Coulter). 250 μl fractions were collected and analyzed for the presence of Pif1p by western blot using an anti-His monoclonal antibody (Novagen). Molecular weight markers (Sigma) were resolved on an identical gradient run at the same time. Linearity of the gradient was verified by running a blank gradient and measuring diffraction index of each 250 µl fraction with a Bausch & Lomb Abbe-3L Refractometer. Pif1p was overexpressed in bacteria and purified to near homogeneity using a two-step procedure involving a cobalt affinity column and cation exchange chromatography () (a). In a recent study, we showed that Pif1p is able to displace short DNA or RNA strands, provided that the loading strand is made of DNA. This result, along with our demonstration that Pif1p inhibits telomerase and (,) raised the possibility that Pif1p inhibits yeast telomerase by unwinding the RNA–DNA hybrid formed between the telomerase RNA, TLC1 and the end of the TG telomeric DNA. In order to compare the activity of Pif1p on RNA–DNA hybrids versus DNA–DNA substrates, we first determined if Pif1p unwinding of these two types of substrates had the same optimal salt requirements. For this experiment, we used as a substrate a 40-mer DNA oligonucleotide to which was annealed a complementary 20-mer oligonucleotide made of either DNA or RNA, leaving a 20-nt 5′ single-stranded overhang (substrates D20 and R20, and a). The experiments were conducted in conditions in which the enzyme was in excess, as significant unwinding was not observed if the substrates were present in excess or in the presence of a DNA trap (data not shown). Both DNA–DNA (closed circles) and RNA–DNA (open circles) hybrids were optimally unwound at 50 mM NaCl, and no activity was detected at 200 mM NaCl (b and c). Therefore, the optimal salt concentration for unwinding both RNA–DNA and DNA–DNA hybrids is 50 mM NaCl. At all salt concentrations, a larger fraction of the RNA–DNA substrate compared to the DNA–DNA substrate was unwound. From the previous experiment, it seemed that Pif1p was more potent at unwinding the RNA–DNA substrate than the DNA–DNA substrate (). To characterize this preference more quantitatively, we designed a family of RNA–DNA and DNA–DNA substrates of various sizes. These substrates contain a 13-, 20- or 40-nt-long duplex region with the same 5' single-stranded 20-nt overhang (, a). The different substrates were designed to have high G+C content but an otherwise random sequence and predicted not to form stable secondary structures at 35°C. Not only was the overhang identical on all of the substrates, but in addition, longer substrates contained the sequence of the shorter substrates at the 5′ end of the duplex region (a). To ensure that the differences in unwinding rates were not due to differential stability of the substrates, we measured the thermal melting curves of all the substrates. RNA–DNA hybrids of high GC content are predicted to be more stable in solution than their DNA–DNA counterparts (). As expected, RNA–DNA hybrids were consistently as stable as or more stable than their DNA counterpart (b). The six substrates were incubated with ATP, 50 mM NaCl, and excess Pif1p (60:1 enzyme/substrate ratio), which corresponded to saturating amounts of the enzyme for these substrates (data not shown). The displacement of the P-labeled top (short) strand of the substrate was monitored by electrophoresis in a non-denaturing polyacrylamide gel and quantified by phosphorImager analysis of the dried gels (a). Efficient unwinding was observed for each RNA–DNA substrate. Even the longest RNA–DNA substrate, R40, was 60% unwound by the 3-min time point (a and b). In contrast, only the 13-mer DNA substrates were fully unwound during the 60-min time course (a and b; D13). In each case, Pif1p unwound the RNA–DNA hybrid more efficiently than its DNA–DNA counterpart, and the difference in the amplitude of the reaction and in the turnover rate between the two types of substrates increased with the size of the hybrid (c and d). Indeed, the difference in the turnover rates between the RNA–DNA and DNA–DNA substrates was respectively 2-fold for the 13-mer, 13-fold for the 20-mer and 75-fold for the 40-mer. Pif1p was almost completely unable to unwind the 40-mer DNA substrate in these conditions even after 60 min (a and b; D40). When the reactions were started in the presence of a single-strand DNA trap, none of the substrates was unwound at a quantifiable level, which made it impossible to measure single cycle kinetic rates on these substrates. Thus, although the activity of Pif1p is strongly stimulated by RNA–DNA substrates, the enzyme is poorly processive on both DNA–DNA and RNA–DNA substrates. The activity of Pif1p purified from yeast mitochondria is stimulated by forked substrates (). Here we determine if this stimulation is also observed with RNA–DNA substrates. The substrates used for this experiment are based on the 20-mer hybrids D20 and R20 (, a). To generate forked substrates, a 10-mer poly(dA) segment was added to the 3' end of the labeled (top) strand of the D20 substrate to generate fD20. Similarly, a 10-mer poly(rA) was added to the 3' end of the top strand of the R20 substrate to generate fR20 (, a). At 35°C, fD20 and fR20 were rapidly and completely unwound (<20 s, data not shown). This result confirmed that Pif1p activity is stimulated by forked substrates, but in these conditions the unwinding of the substrates was too fast to allow quantitative analyses. Lowering the temperature to 25°C slowed the reaction kinetics and allowed quantitative comparison of the unwinding of the RNA–DNA and DNA–DNA forked substrates in presence of saturating amounts of Pif1p. At 25°C, both fD20 and fR20 were readily unwound (b). However, the RNA–DNA hybrid was unwound at a faster rate [18.3 versus 2.1 min, compare fR20 (−) to fD20 (−), c]. We then tested if the increased unwinding rate for forked substrates was accompanied by a change in the ability of the enzyme to unwind these substrates under single cycle condition. In the presence of a 500-fold excess of a 36-nt-long single-strand DNA trap, which prevents the enzyme from re-associating if it dissociates from the substrate during the course of the unwinding reaction, no unwinding of fD20 was observed (b, fD20, left panel). In contrast, ∼20% of fR20 was unwound under these conditions (b, fR20, right panel; c). This result provides two mechanistic insights. First it indicates that, in contrast to what is observed with DNA–DNA substrates, the enzyme is able to unwind the 20-mer forked RNA–DNA hybrid without dissociating. Since no unwinding of the unforked substrates was observed in single turnover experiments (data not shown), this result also suggests that the presence of a RNA–DNA fork helps stabilize the enzyme–substrate complex upon binding and/or during the initial steps of unwinding. A possible explanation for the increased activity of Pif1p on RNA–DNA hybrids and on forked substrates could be that Pif1p binds preferentially to the substrates that are more easily unwound. Using a quantitative gel retardation assay, we compared the binding of Pif1p to forked and unforked substrates comprised of either DNA–DNA or RNA–DNA hybrids in absence of nucleotide, in presence of AMP–PNP, a nonhydrolyzable ATP analog () or in presence of the poorly hydrolyzable ATP analog ATPγS. In absence of nucleotide, Pif1p bound about as well to the fR20, fD20 and D20 substrates. The for binding to these substrates was, respectively, 1.0 × 10 M (D20), 1.3 × 10 M (fD20) and 5.2 × 10 M (fR20). Pif1p bound most efficiently to the unforked RNA–DNA R20 substrate ( = 1.7 × 10 M), although this was not the most readily unwound substrate. These data indicate that preferential binding in absence of nucleotide does not account for the preferential unwinding of forked substrates, at least not in a way that is measurable by gel retardation (compare squares to circles, a). However, Pif1p does show preferential binding to RNA–DNA substrates compared to their DNA–DNA counterparts, although this 2- to 5-fold preference is relatively small (a, compare circles to circles and squares to squares). In presence of saturating amount of AMPPNP (b) or in presence of ATPγS (data not shown), the binding affinity decreased significantly by more than 10-fold for all substrates except for the forked DNA substrate fD20. Thus, under conditions containing nucleotide analogs, which are the most similar to conditions used for unwinding assays, a DNA–DNA substrate had the highest binding affinity rather than the RNA–DNA substrates that were more readily unwound. It has been shown for a number of helicases, including the yeast Srs2p and the human RecQ helicases, RECQ1, WRN and BLM, that addition of a sequence non-specific single-strand DNA-binding protein, such as SSB or the eukaryotic replication protein A (RPA), significantly enhances their ability to unwind longer substrates (,). The mechanistic explanation for this effect is that SSB or RPA prevents the re-annealing of partially unwound substrates by trapping the unwound region of the substrate in single-strand form. If, during the unwinding reaction, the DNA strand of the DNA–DNA substrate had a higher tendency than the RNA oligonucleotide to re-anneal to the opposite DNA strand, it would lead to an apparent slower unwinding rate for DNA–DNA substrates. We carried out the unwinding reaction of the 40-mer substrate (1 nM) in the presence or absence of saturating amounts of yeast RPA (5 μM) at 30°C and 100 nM Pif1p. As seen for other helicases, Pif1p unwinding activity was stimulated by RPA (). RPA stimulation of Pif1p was detected with both DNA–DNA and RNA–DNA substrates, although stimulation of unwinding of the RNA–DNA substrate was relatively modest (a and b). However, RPA did not alleviate the preference for RNA–DNA unwinding. Therefore, the relative inefficiency of Pif1p for unwinding DNA–DNA substrates is not due to a preferential re-annealing of the displaced DNA strand, but is due to an intrinsic lower processivity of the enzyme when unwinding DNA. Pif1p purified from mitochondria has been shown to exist as a monomer in solution (). We analyzed the nuclear recombinant Pif1p by glycerol gradient sedimentation. Pif1p was incubated in the standard reaction buffer at high concentration (5 µM) before being sedimented in a 15–40% glycerol gradient. The position of Pif1p was revealed by western blotting using an anti-Histidine tag monoclonal antibody (). Pif1p eluted between carbonic anhydrase (29 kDa) and alcohol dehydrogenase (ADH, 150 kDa). The predicted molecular weight for His-tagged nuclear Pif1p is 95 kDa. Although Pif1p fractionated over a broad range in this assay, none of the protein could be detected above 150 kDa. This result suggests that the recombinant nuclear form of Pif1p exists as a monomer in solution, similar to what has been observed for mitochondrial Pif1p or for Pfh1p, the homolog of Pif1p (,). In this study, we show that Pif1p has a preference for unwinding RNA–DNA hybrids. This preference was manifest at several levels. First, the rate of unwinding of a given RNA–DNA substrate was always faster than the rate of unwinding of the comparable DNA–DNA substrate. In the case of the linear 40-mer substrates, the rate of unwinding of the RNA–DNA substrate was almost two orders of magnitude higher than for the DNA–DNA 40-mer substrate (). Second, the fraction of RNA–DNA substrate that was unwound was consistently higher. Indeed, Pif1p was able to unwind efficiently a 40-mer RNA–DNA substrate, while <5% of a 40-mer DNA–DNA substrate was unwound even in the presence of a 60-fold excess of Pif1p (). Third, even though the unwinding reactions for RNA–DNA and DNA–DNA substrates shared the same optimal salt concentration, the RNA–DNA unwinding reaction was more resistant to increasing ionic strength in the physiological range (). The ability to unwind RNA–DNA hybrids has been reported for several DNA and RNA helicases, such as the DNA repair helicase UvrD (), the RNA helicase NPH-II () and several replicative DNA helicases, including DnaB, the MCM protein and the MCM4, 6, 7 complex (). However, of these helicases, only UvrD also exhibits preferential unwinding of RNA–DNA hybrids (). Therefore, the preference for unwinding this type of substrate does not appear to be a common property of helicases. Indeed, Pif1p's preference for RNA–DNA hybrids contrasts with the emerging view on the mechanism of unwinding for several helicases that suggests that most helicases are not particularly sensitive to the chemical identity of the displaced strand. For example, studies with the bacteriophage T4 Dda helicase, a 5′–3′ SFI superfamily member, show that the rate-limiting step for unwinding is relatively insensitive to the chemical nature of the displaced strand (). Similar conclusions have been reached for NPH-II, a 3′–5′ SFII RNA helicase (). Therefore, the mechanism of nucleic acids recognition and/or unwinding by Pif1p are likely different from that of these prototypical SF1 and SF2 superfamily helicases. Pif1p is also stimulated by forked substrates (). Here we show that the stimulatory effect of the fork structure occurs with both DNA–DNA and RNA–DNA substrates (). At 35°C, both the forked RNA–DNA hybrids and the forked DNA–DNA substrates were unwound very rapidly, in <20 s. In comparison, the unwinding of R20 was complete in 1 min under identical experimental conditions (). Therefore, we can estimate that the presence of the fork increases the apparent kinetic rate of unwinding by at least 5-fold. Since Pif1p binding to forked substrates was not significantly higher than its binding to conventional tailed substrates, both in absence of nucleotide and in presence of the nucleotide analog AMPPNP () or ATPγS, this substrate specificity is probably not established by differential binding. However, it is possible that the fork structure increases the stability of the helicase–substrate complex and/or triggers the transition between the binding form of the enzyme and the translocating form upon ATP binding. A detailed analysis of the effects of nucleotide binding on nucleic acids binding by Pif1p will be needed to address this question. When the unwinding of the two forked substrates was done at 25°C, the unwinding of both forked substrates was slower, but the unwinding of fR20 was still nine times faster than unwinding of fD20. The stimulatory effect of the fork was observed with both DNA–DNA and RNA–DNA substrates (). This result indicates that the stimulation provided by the forked structure is additive to the stimulation provided by the top strand being made of RNA. The combined stimulatory effect of the forked structure and the RNA strand on Pif1p activity has an important consequence for Pif1p helicase activity as Pif1p was able to unwind the forked 20-mer RNA–DNA hybrid in a single turnover reaction (). Interestingly, the calculated rate of unwinding for the single turnover reaction was ∼3 times lower that the calculated rate for multiple cycle conditions (5.2 min versus 18.3 min; ), which indicates that the rate-limiting step is likely to be different in the two reaction conditions. Pif1p is a poorly processive helicase . An intriguing possibility is that enzyme stalling and/or dissociation from substrates accounts for the low rate of unwinding in single cycle conditions. In multiple cycle conditions, alignment of multiple Pif1p monomers ‘pushing’ each other forward along the substrate could have a stimulatory effect on the overall kinetic rate of the reaction. Increased activity when multiple helicases are bound to a substrate has been shown with the Dda helicase, another monomeric helicase, in a different experimental setting (). A detailed study of Pif1p kinetic mechanism is necessary to resolve this question. Since Pif1p does not bind RNA (), an interesting possibility is that during the unwinding of DNA–DNA substrates, Pif1p interacts with both strands of the duplex. If the interaction with the upper DNA strand counteracts the unwinding reaction, whether it is achieved by translocation of the enzyme or another mechanism that remains to be identified, this interaction could result in lower unwinding rates for DNA–DNA hybrids. This scenario could arise if the active form of Pif1p is a dimer with each monomer attached to a different DNA strand. However, we did not detect self-association of the protein even when the enzyme was incubated at 50 times the concentration that was used in helicase assays (), suggesting that nuclear Pif1p is active as a monomer. If Pif1p is active as a monomer and dissociates from the bottom strand, it could potentially re-associate with either the lower or the partially unwound upper strand if it is made of DNA, which would also slow down unwinding of the DNA–DNA substrate. Taken together, our results indicate that Pif1p is sensitive to both the structure of the duplex region of the substrate, preferring forked to linear duplexes, and to the chemical nature of the top strand, preferring to displace a RNA strand. How do these data inform our understanding of Pif1p function ? The best studied function of Pif1p is its inhibition of telomerase lengthening of existing telomeres and telomere addition to double-strand breaks (,,). The preference of Pif1p for unwinding forked RNA–DNA hybrids supports the hypothesis that Pif1p inhibits telomerase by unwinding the RNA–DNA hybrid formed between telomerase RNA and the telomeric DNA end. Indeed, the TLC1-RNA annealed to the telomeric single-stranded DNA resembles a forked RNA–DNA hybrid (a) (). In addition to its telomeric functions, nuclear Pif1p helps maintain the RFB in the rDNA (). Given its preference for forked RNA–DNA substrates () and its association with rDNA (), Pif1p might sit at a fork stalled at the RFB and act on the rRNA transcript in a manner that prevents the transcript from running into the RFB (b). Nuclear Pif1p also acts in cooperation with the helicase/endonuclease Dna2p. Dna2p is a helicase/nuclease that is required for Okazaki fragment maturation (). While Δ cells are dead, Δ Δ are viable (although not at high temperatures) and have improved repair and replication capabilities. It has been proposed that Pif1p helps extend the DNA flap generated during Okazaki fragment maturation (), a proposal consistent with Pif1p's preference for forked substrates. Pif1p is also required in cells lacking topoisomerase III activity, and this requirement is dependent upon an active Sgs1p helicase and homologous recombination (). Again, this role can be explained by Pif1p acting on the non-linear DNA–DNA hybrids that arise during recombinational repair. Finally, Pif1p is also important for maintenance of mitochondrial DNA (,), especially in the presence of oxidative damage or the intercalating agent ethidium-bromide (). The increased mitochondrial DNA breakage detected in ethidium bromide treated cells occurs at discrete sites, suggesting that Pif1p action is required only at specific loci in the mitochondrial genome (). Pif1p might carry out its role at non-telomeric sites by unwinding forked DNA–DNA hybrids. However, it is tempting to speculate from its substrate preferences that Pif1p acts primarily or even exclusively on RNA–DNA hybrids. One interesting possibility is that Pif1p participates in the removal of the RNA–DNA hybrids that form during transcription and which, if not removed, lead to the formation of R-loops, which can trigger replication arrest, fork breakage and recombination (). Because replicative helicases from various organisms are able to unwind RNA–DNA hybrids (), in most situations, replicative helicases are likely able to remove most RNA–DNA hybrids that are present in front of the replication fork. Nonetheless, it is possible that particular sequences in the genome are prone to the formation of stable RNA–DNA hybrids that require a dedicated helicase like Pif1p for efficient processing. This model would reconcile the observation that Pif1p acts at discrete loci of the genome (,) with its preference for RNA–DNA hybrids (this study). The fact that the enzyme is involved in maintenance of mitochondrial and rDNA, two loci that contain highly transcribed genes, is compatible with this hypothesis. Given the evolutionary conservation of the PIF1 family helicases (,,), it will be interesting to determine if other members of the family also act preferentially on RNA–DNA hybrids, as a first step to identify their substrates.

Recombination between homologous DNA molecules is a key element of proper genome maintenance and duplication. Homologous recombination is an important pathway for accurate repair of DNA double-strand breaks, is required for generating genetic diversity during meiosis, and is essential for overcoming difficulties in replication. The core mechanistic steps of homologous recombination consist of joint molecule formation and strand exchange between DNA molecules. In homologous recombination, one DNA partner is first processed to yield long single-stranded regions. This single-stranded DNA (ssDNA) is then coated by RecA-type recombinase proteins, forming a nucleoprotein filament (,). RecA-type recombinases include, in addition to bacterial RecA, archaeal RadA, and eukaryotic RAD51. These recombinases assembled into nucleoprotein filaments on ssDNA are the catalytic core of homologous recombination responsible for aligning homologous sequence and driving strand exchange between the ssDNA in the filament and homologous double-stranded DNA (dsDNA). RAD51, RecA, and RadA all form structurally similar helical nucleoprotein filaments, despite their limited amino acid sequences conservation (). These recombinases do have highly conserved ATPase domains (,) and ATP binding is essential for filament formation (). Structural characterization of these nucleoprotein filaments has almost exclusively been performed using static methods, which resulted in emphasizing regularity. However, even static methods reveal variation in filament structure, often depending on the nucleotide cofactor bound, indicating flexibility and possibly dynamic rearrangements (). The effect of different nucleotide cofactors on filament structure can be explained by the location of their binding site at the interface between adjacent monomers in the DNA-bound filament (,,). Electron-microscopy studies revealed that RAD51 filaments have approximately six monomers per helical turn, and each monomer covers 3 nucleotides (nt) or base pairs (bp) (,,). Similar to RecA, the length of the DNA substrate within the RAD51-DNA filament is maximally extended by 50% compared to the length of B-form DNA (,). For RecA, the kinetics of filament formation has been extensively characterized (). In particular most recently using time-resolved spFRET () and fluorescently labeled RecA (), a detailed quantitative picture on nucleation and growth of RecA filaments has been obtained. Due to pentameric nucleation but monomeric growth there are conditions, where nucleation is relatively inefficient but filament extension very efficient (here referred to as cooperative filament formation) under which RecA forms very long (>10 kb) continuous filaments. In contrast, RAD51 filament formation is less well characterized. Recent work has shown that RAD51 filaments are irregular in many conditions indicating possible dynamic rearrangements (,,). It has been suggested in those works that the dynamic nature of RAD51-ssDNA filaments may be an important aspect of strand exchange activity (,). Analysis of the kinetics of human RAD51 filament formation has been done only with short oligonucleotides (). We set out to determine from similar kinetic measurements the parameters of dynamic filament formation, specifically nucleation and extension rates, on a quantitative level. For this we use magnetic tweezers, which allow monitoring the real-time dynamic behavior of RAD51-DNA filament formation and disassembly on individual molecules of both ss- and dsDNA of several kilobase in length. Here, we measured human RAD51 filament formation for a set of controlled reaction conditions that are known to allow or inhibit filament dissociation (,) in order to decouple filament growth and dissociation. We compared data sets with extensive Monte Carlo simulations from which we could deduce kinetic parameters of the reaction. Binding of RAD51 to both ss- and dsDNA in the presence of ATP was not strongly cooperative, defined here as the ratio of filament-extension to nucleation rates. At any concentration of RAD51, nucleation played a dominant role in filament formation, in contrast to RecA for which lower concentrations allow infrequent nucleation events followed by rapid filament extension. Removal of RAD51 from the buffer resulted in filament disassembly at multiple positions along the tethered dsDNA molecule, if ATP hydrolysis was allowed. Nucleoprotein filament formation on ssDNA was more complex and dynamic, due to the more prominent role of dissociation on ssDNA. Our results indicate that the RAD51 filaments consist of multiple short filament patches. The human RAD51 protein was purified as described (). Nicked 8 kb dsDNA with biotinylated and digoxigenin-modified ends was made as described previously (). A ssDNA construct was prepared as follows. First a double-stranded 8660 bp DNA fragment was produced by PCR, using λ-DNA as a template and primers sequences 5′-AACTCAGCTCACCGTCGAACA and 5′-AAAAGAAATTCCCTCAAATGGACGCCGGATGAC, which was 5′ biotinylated. After purification, the PCR-fragment was digested close to the non-biotinylated end using ApaI at 30°C for 1 h. After this, the ApaI-treated extremity was ligated to a 700 bp PCR fragment containing several digoxigenin-modified dUTP bases and an ApaI ligatable end. The resulting DNA construct was incubated with superparamagnetic 2.8 μm streptavidin-coated beads (Dynal, Oslo) for 30 min at room temperature. Subsequently the bead-bound DNA was denatured by addition of 50 μl 0.1 M NaOH for 30 min at room temperature, in which the biotinylated single DNA strand bound to the magnetic beads was separated from the non-biotinylated complementary strand. Using an external magnet a pellet of the magnetic beads was formed followed by removal of the supernatant. The magnetic beads were re-suspended in 10 mM Tris–HCl (pH 8.0) and 1 mM EDTA, and deposited in the flow cell. A magnetic tweezers setup was used in these experiments as described (). By using image processing, 5 nm position accuracy of the bead was obtained in all three dimensions (). To exclude the effect of thermal drift, all positions were measured relative to a 3.2 μm polystyrene bead (Bang Laboratories, Carmel, IN) fixed to the bottom of the flow cell. Polystyrene beads as well as DNA constructs carrying a magnetic bead at one end were anchored to the bottom of a flow cell as described elsewhere (). The force-extension curve of single DNA molecules was measured (B). After conformation of the correct contour and persistence lengths, experiments were started by addition of RAD51. All measurements were carried out at 25°C. The flow-cell final volume was ∼100 μl. All reactions were done in 25 mM Tris–HCl (pH 7.5), 5 mM MgCl or CaCl, 25 mM KCl, and 1 mM DTT. RAD51 and ATP [final concentrations 187 nM RAD51 (unless stated otherwise) and 1 mM ATP] were added into the cell. Interaction of RAD51 with the tethered DNA molecule was monitored through measurement of the height of the magnetic bead. Binding of RAD51 onto a DNA substrate was modeled using Monte Carlo simulations. A similar simulation has previously been done for RecA (). A 1D array was used to mimic the DNA substrate containing a number of elements equivalent to the number of nucleotides or base pairs of the DNA molecule of interest. In the case of ssDNA the array contained 8660 elements, and for dsDNA 8000. RAD51 filament assembly was described by nucleation followed by growth that extended the nucleation point. Simulations were done with various binding units consisting from 1 to 5 RAD51 monomers. Upon binding of RAD51 to the DNA substrate, the protein was assumed to cover 3 nt or base pairs per monomer, and to elongate the binding site length to a value corresponding to 50% compared to the base–base spacing in B-form DNA. Varying the elongation of the binding site between 45 and 55% only changed the obtained increase in end-to-end distance and not the rates for nucleation and filament extension. Nucleation was allowed to occur at any point along the entire molecule. In the Monte Carlo simulations, the nucleation step was simulated as follows: a value was randomly extracted from a uniform distribution yielding a value between 0 and 1. If this value was smaller than a given threshold corresponding to the set nucleation rate for the entire molecule, a RAD51 oligomer was bound. The binding location was deduced from a second random number between 0 and 1 that was extracted from a uniform distribution that was multiplied by the number of elements in the 1D array. Binding of a -mer occurred only when this site plus the following 3 − 1 sites were not covered by another protein, to account for the fact that each protein covered 3 nt or base pairs (cf. A). Subsequently, we evaluated all nucleation sites where filament extension could occur. For each site, a value was extracted from a uniform distribution and compared to a given threshold corresponding to the set rate of filament extension for a single filament patch. If this value was smaller than the threshold, the filament was extended by an oligomeric-binding unit if the next 3 nucleotides or base pairs were not already covered by protein. Extension was only permitted into the direction of higher numbers in the 1D array. The probabilities for nucleation and growth per time step were taken so small, that the chance of two binding events within a single Monte Carlo step was negligible. The threshold values, which are rates expressed in units (Monte Carlo step), convert into kinetic rates expressed in s by adjusting the time axis of the Monte Carlo growth curve to the experimental growth data. For every Monte Carlo step, the end-to-end distance of the DNA substrate can be calculated by multiplication of all the protein-covered elements in the 1D array by 0.51 nm, and the uncovered ones by 0.34 nm. To account for the entropic coiling of DNA at the specific stretching force applied, the resulting length is decreased according to an amount expected for a worm-like-chain polymer (). This means that in the force regime probed here, the end-to-end distance is smaller than the crystallographic contour length. Whereas our simple modeling involved filament extension and disassembly in a unidirectional fashion, essentially the same results are found if extension and disassembly occur in both directions, albeit with two slightly different values for the rates that change by a factor up to 2. In those cases where disassembly was considered, we additionally allowed dissociation to occur after the filament extension step. At each end of a filament patch opposite to the filament-extension end (i.e. towards lower numbers in the array), a value was extracted from a uniform distribution and if this value was smaller than the threshold set by the dissociation rate, the monomer dissociated and a vacancy was created. Alternatively, dissociation was considered at all monomer sites, where the above procedure was extended to all bound RAD51 proteins. In the case of ssDNA, disassembly was modeled in two steps. In the first step, the monomer remained bound, but the contour length of the DNA substrate changed from 0.51 into 0.459 nm [a change of the helical pitch from 92 to 82 Å (,)] presumably due to ATP hydrolysis. A value was extracted from a uniform distribution and if this value was smaller than the threshold set by the hydrolysis rate, the monomer underwent this conformational change. This was followed by dissociation presumably through the release of ADP, incorporated as described above. The Monte Carlo simulated profiles at different ratios between nucleation and filament extension were fit to the experimental data by adjusting the time axis of the Monte Carlo growth using a least-squares method (A). For every data set, a best value for the ratio and the time axis was extracted this manner. The average values with the standard deviation mentioned in the main text were calculated from multiple experiments. italic fig #text We have quantitatively described the association and dissociation of RAD51 with ss- and dsDNA using single-molecule measurements and Monte Carlo simulations. The obtained rates are summarized in . Though direct comparison with RecA is difficult due to variation in reaction conditions and assays, RAD51 has apparently a higher nucleation rate and lower extension rate per filament patch. This is based on the behavior of RecA binding to dsDNA as measured in recent single-molecule experiments done under similar conditions of protein concentration and nucleotide cofactor (). From those data, we can extract an extension rate at 170 nM RecA of 2–7 monomers s, ∼10-fold higher than the extension rate we measure for RAD51 (). The highest nucleation rate at 200 nM would correspond to 6 × 10 s bp for a RecA pentamer, or 10-fold lower than we measure for RAD51. Comparing the ratio of extension per filament patch to nucleation makes RAD51 filament formation much less cooperative than RecA filament formation, under these conditions. The behavior of RAD51 that we observe is well described with a multimeric-binding unit for both nucleation and extension per filament patch. We measured DNA length profiles that imply extension in steps equivalent to binding of multimers of on average RAD51 coordinated monomers where = 4.3 ± 0.5. A variety of multimeric forms in this range have been described for RAD51 (,). Furthermore, recent studies reported that RecA nucleates on both ss- and dsDNA as a pentamer (,). RecA filament extension was, however, suggested to occur by monomeric association extracted from a novel analytical tool based upon hidden Markov modeling (). In contrast to this result, our finding for RAD51 indicates the DNA length changes during extension per filament patch involves multimers. RAD51 forms complexes with both ss- and dsDNA during the course of homologous recombination. RAD51 functions in eukaryotic cells in coordination with many other proteins including the so-called recombination mediators, favoring RAD51 assembly onto ssDNA in need of homologous recombination and disfavoring assembly elsewhere (). The behavior of RAD51 with DNA that we describe here, provides a baseline for assessing the detailed effect other recombination proteins have on filament assembly and stability. The flexible RAD51 interface domain, a proposed site of molecular hand-off events (), may be important in determining multimer addition, limiting filament growth rate or effecting filament stability. It will also be important to determine if the described effects of the BRCA2 domains are due specifically to changes in RAD51 filament nucleation, growth, or disassembly rates (). The intrinsic dynamics of RAD51 filament assembly and disassembly are likely important aspects of homology search and strand-exchange. We observe intermittent assembly dynamics and partial coverage, for RAD51 especially on ssDNA. When ATP hydrolysis is active, interactions between RAD51 and ssDNA are much more dynamic than with dsDNA. With respect to this different behavior of RAD51 on ss- and dsDNA, it is interesting to note that, in contrast to the situation in bacteria, proteins capable of disassociating RAD51 filaments from dsDNA play an important role in mammalian homologous recombination (). Most notably, the disassembly rate of RAD51 monomers from ssDNA and dsDNA is very different. Our experiments reveal that during the continuous interplay of assembly and dissociation, short RAD51 filament patches are formed on DNA. These RAD51-DNA patches are separated by tracks of bare DNA that are too short to allow additional RAD51 association (). Force-extension measurements showed that RAD51-DNA was more flexible than RecA-coated DNA by a factor of 4. The greater flexibility of RAD51-DNA complexes supports the presence of bare DNA gaps. The presence of short interrupted RAD51 filaments explains some aspects of RAD51 strand-exchange reactions. For instance, RAD51 cannot bypass short regions of heterology (). Though heterology bypass by RecA depends on ATP hydrolysis and therefore a dynamic filament, even where ATP hydrolysis is active, the average length of RAD51 patches, and the resulting joint molecules, would be too short to stabilize longer stretches of mis-paired bases. The arrangement of RAD51 as interrupted filaments is less prone than RecA to inhibition of strand exchange due to topological constraints from non-productive pairing interactions requiring a stiff recombinase filament (). It is of interest to speculate about possible mechanistic advantages of the RAD51-DNA filament arrangement we have obtained. Because ssDNA is highly flexible, the bare DNA gaps likely will act as hinges, creating a quite flexible RAD51-DNA structure. A stiff continuous nucleoprotein filament where association with the target only occurs at the filament ends allows only a single functional interaction point. The overall flexibility and dynamics of the RAD51 structure will promote multiple sites for potentially functional interactions between the filament segments and the target dsDNA. Thus a flexible filament could be an advantage in identifying homology because one diffusion-limited encounter would sample multiple regions of the target dsDNA. This advantage in homology search may not be evident in strand exchange assays because they do not report on homology search but rather detect the products of strand exchange, which may still be hindered by filament interruptions. The tracks of bare ssDNA in between RAD51 patches can also act as swivels to release topological constraints during joint-molecule formation, which is not possible for long continuous nucleoprotein filaments (). p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Mosquito species of the complex represent the major vectors of human malaria and they pose an enormous burden on global health and economies. Every year 300–500 million people are infected by malaria and over a million people die as consequence of parasite infections (). While many insect pests have long been successful targeted with population control measures such as insecticides or release of sterile males (), for others, including , classical control measures have largely failed to deliver long-term solutions. Disease endemic countries often do not have the economic resources and the logistics to sustain control efforts like the massive and prolonged use of insecticides. New control strategies that are affordable, easy to implement and sustainable are desperately needed. This global health problem has prompted an unprecedented effort aimed at generating new molecular tools and a better understanding of the biology and the genetics of Anopheline mosquitoes that culminated in the sequencing of the genome () and development of gene transfer technology for a series of vectors species (,). These molecular advances have made it possible to express genes that can block the transmission of in model systems () or express traits facilitating the implementation of sterile insect techniques for vector control (). The translation of these achievements in suitable control measures still represents a major scientific and technical challenge. Genetically modified mosquitoes carrying a desired trait such as malaria refractoriness would need to be released on a gigantic scale given the vast numbers of these insects and the wide areas that are inhabited by vectors of human tropical diseases. Therefore, a mechanism must be developed to spread the desired genetic modification from a few laboratory-reared mosquitoes to a large fraction of the wild-type vector population (,). Naturally occurring ‘selfish’ genetic elements that have non-mendelian inheritance mechanisms and spread through populations even when they provide no benefit to the host organism () have been proposed to transform wild-type mosquito populations. Homing endonuclease genes (HEGs) are highly specific DNA endonucleases found in some viruses, bacteria and eukaryotes. The endonuclease promotes the movement of its encoding DNA from one allele to the other by creating a double-strand break (DSB) at a specific, long (15–40 bp) target site in an allele that lacks the HEG. Homologous DNA repair then copies the HEG to the cut chromosome in a process called ‘gene conversion’ (,). The observation that HEGs can be engineered to cleave novel DNA sequences () offers a multitude of opportunities to utilize these elements for mosquito control. For example, HEGs could be used to disrupt genes regulating the ability of mosquitoes to function as efficient vectors for parasites, or to drive recombinant refractory genes through a mosquito population, rendering them unable to transmit malaria. Alternatively, HEGs designed to target an essential mosquito gene or a gene required for female fertility could be utilized to introduce a genetic load on the population leading to population size reduction or collapse (). More recently, it has been suggested that a harmful selfish element put under the control of a promoter which is active in individuals susceptible to infection but inactive in refractory individuals should drive alleles causing refractoriness through the population (). Finally, HEGs could be used to bias the sex ratio towards males, using an endonuclease that targets X-linked sequences and is expressed during male spermatogenesis from the Y chromosome (). To investigate the feasibility of using HEGs as driving genetic element in mosquitoes, we have analysed the activity of ectopic HEGs in both cells and embryos, using experimental systems that are highly predictive of behaviour of mobile genetic elements (,). We determined the activity of two well-characterized HEGs, I-PpoI (a member of the His-Cys box family of endonucleases from the slime mold ) () and I-SceI (a LAGLIDADG class endonuclease originally isolated from mitochondria) (), both of which have been used in a variety of organisms (including ) to induce DNA DSBs (). We systematically analysed the nature of HEG-mediated integration and recombination events in and the effect of expressing these endonucleases on cell proliferation. The target plasmids pBC/SacRB S1/S2/S3 and pBC/SacRB P1 were constructed as follows: pBC/SacRB () was cut with SalI and PstI, ends filled in with T4 DNA polymerase and religated to remove a redundant EcoRI site. Linkers were created by annealing 5′ phosphorylated oligonucleotides (S1: AATTCATTACCCTGTTATCCTAG and AATTCTAGGGATAACAGGGTAATG; S2: AATTCGATAGGGATAACAGGGTAATTG and AATTCAATTACCCTGTTATCCCTATCG; S3: AATTCAATTACCCTGTTATCCCTACCG and AATTCGGTAGGGATAACAGGGTAATTG; P1: AATTCCGCTACCTTAAGAGAGTCG and AATTCGACTCTCTTAAGGTAGCGG), which were cloned into the now unique EcoRI site in the SacRB CDS. Selection against SacRB was performed in LB agar lacking NaCl containing 15% sucrose (w/v) and chloramphenicol (25 μg/ml). To create pSL-SacRB, the minimal 1.5 kb tetracycline (Tet) resistance cassette from pBR322 was amplified with the primers TTCAAGAATTCTCATGTTTGACAG and ATGAATTCTGCTAACCAGTAAGGCAACC and cloned into pBC/SacRB with EcoRI. The Tet cassette replaces the I-SceI site and disrupts the gene. The 3.4 kb SacRB cassette was moved to pSLfa1180fa () using XhoI/XbaI to create pSL-SacRB. To create pDR-CMV-GFP, the CMV promoter from pEGFP-Ppo was amplified with the primers AAAGGGCCCTAGTTATTAATAGTAATCAATTACGGGGTCATTAG and AAAGAATTCGATCTGACGGTTCACTAAACCAGCTC and cut with ApaI and EcoRI. The fragment was ligated into partially ApaI and EcorI digested pDR-GFP (). This replaces the 3′ part of the chicken β-actin promoter with CMV. To remove the remaining sequences of the chicken β-actin promoter, the resulting vector was cut with SnaBI and religated. PSL-Act-EGFP was constructed in pSLFa1180fa to contain the 2.5 kb 5C promoter driving EGFP (BamHI/XbaI fragment) and the Hsp70 terminator. Cells from the stable anchorage-dependent cell line, Suakoko 4 (Sua 4.0) (Müller,H.M. , 1999) were cultured in Schneider's medium (Invitrogen) supplemented with 10% FCS (Invitrogen) and 200 U/ml penicillin and 200 μg/ml streptomycin sulphate (Invitrogen) in a cooled incubator at 27°C. HEG activity assays were performed by lipid-mediated transient transfections (Effecten, Qiagen) of 1–3 × 10 cells with 2 μg/ml (culture volume) recipient plasmid and 4 μg/ml donor plasmid. When necessary, cells were heat-shocked for 1 h at 41°C, 24 h post-transfection to induce expression of I-SceI from the Hsp70 promoter on the pP[v+, 70 I-SceI] plasmid. Total DNA was extracted 48 h post-transfection (Promega Wizard Genomic DNA purification Kit) and re-suspended at 25–40 μl. This preparation was used to transform DH5α strain. adult females (G3 strain) were allowed to deposit their embryos 72 h after a blood meal on a moist filter paper. Embryos were injected 60–120 min after oviposition, essentially as described (). Embryos were injected with a mixture of I-SceI donor plasmid pP[v+, 70 I-SceI] (400 μg/ml), Tet donor plasmid pSL-SacRB (500 μg/ml) and HEG target plasmid pBC/SacRB S1 (500 μg/ml). Between 150 and 200 embryos were microinjected and 12 h later embryos were heat-shocked for 1 h at 41°C and left to recover for 2 h. Low molecular weight DNA was extracted (), re-suspended in 20 μl and 150 ng of the recovered DNA was used to transform DH5α strain. Genomic DNA was digested with ClaI in the presence and absence of I-PpoI. As a probe we used a 2 kb rDNA PCR fragment amplified from genomic DNA using the primers GCCGAAGCAATTAGCCCTTAAAATGGATG and CACCAGTAGGGTAAAACTAACCTGTCTCACG. The probe was P labelled using the High Prime DNA labelling kit (Roche) and purified with ProbeQuant™ G-50 columns (GE Healthcare). For primer extension, genomic DNA was digested with HincII (a.k.a. HindII). The reaction was performed essentially as described () using the 5′ P-labelled primer rPrex GTTAATCCATTCATGCGCGTCACTAATTAG and vent (exo-) polymerase (New England Biolabs). The reaction products were resolved on a 6% denaturing polyacrylamide gel. Results for both experiments were visualized using a FUJIFILM-FLA-5000 Phosphoimager (Fuji Photo Film Co. Ltd, Stamford, CT, USA). For digestions, we used commercially available I-PpoI (Promega) and I-SceI (New England Biolabs) enzymes. Sua 4.0 cells were transfected with either of the two endonuclease plasmids (4 μg/ml) together with pIB/V5-His (2 μg/ml), conferring resistance to blasticidin (Invitrogen). Forty-eight hours post-transfection, blasticidin was supplemented to complete medium at 50 μg/ml. Cells were incubated in blasticidin for an initial proliferation period of 5 days, at which point they were harvested and re-seeded at 1.5 × 10 cells/ml and grown for a further 5 days. Transfections were performed in triplicates and cell numbers for each were counted in four replicates. Sua 4.0 cells were transfected with pDR-CMV-GFP (2 μg/ml) in the presence or absence of donor plasmid pP[v+, 70 I-SceI] (4 μg/ml). Twenty-four hours post-transfection, cells were heat-shocked for 1 h at 41°C. Gene conversion was measured by fluorescence activated cell sorting (FACS) 72 h post-transfection using a Beckman FACS Calibur; 50 000 size-dependent gated events were analysed for GFP fluorescence. Data were analysed using the FlowJo software package. Transfection efficiency was assessed by parallel co-transfections with an 5C-DsRed plasmid. Sua 4.0 cells were transfected with pEGFP-Ppo, pEGFP-Ppo H98A and the control pSL-Act-GFP, and 48 h later cells were fixed for 5 min in 4% paraformaldehyde and permiabilized for 10 min with 0.1% Triton X-100. Nuclei were stained with DAPI (2 ng/μl) and actin filaments with Alexa546-phalloidin (1 U/ml, Invitrogen). Cell micrographs were taken at ×40 magnification using a Zeiss widefield microscope. To assess the functionality of HEGs in cells, we developed a rapid and efficient reporter system to score HEG-mediated activity (site specific cleavage and homologous recombination events) in both cells and embryos. This assay, based on the interplasmid transposition assay utilized in insect cells and embryos to assess the activity of transposable elements (), utilizes two sets of plasmids (a). A donor plasmid directs the production of either I-SceI [pP(v+,70I-SceI)] or I-PpoI (pEGFP-PpoI), and the target plasmid (pBC/SacRB) contains the bacterial suicidal gene , coding for levansucrase, engineered to contain either the I-SceI or the I-PpoI recognition sequences. catalyses the hydrolysis of sucrose and the synthesis of levans, high-molecular-weight fructose polymers that accumulate in the periplasmic space, and are toxic to gram negative bacteria (,). HEG cleavage of its recognition sequence results in the inactivation of genes, which is then detected in bacteria transformed with plasmid DNA recovered from transiently transfected cells and injected embryos. Negative selection against the functional gene in medium containing sucrose is accompanied by positive selection for the chloramphenicol (Cam) resistance marker, which is also present on pBC/ (). The I-SceI or I-PpoI recognition sites were inserted in the SacRB sequence after Glu67, downstream of the signal peptide required for secretion (), at a position predicted not to be essential (). We assessed the effect of inserting several recognition site variants in different frames on SacRB activity. None of the inserted amino acid variants (pBC/SacRB S1: , pBC/SacRB S2: DI, pBC/SacRB S3: NI for I-SceI; pBC/SacRB P1: for I-PpoI) interfered with SacRB function as inferred by the continued inability of bacteria transformed with these vectors to grow on Cam supplemented with 15% sucrose (data not shown). We therefore concluded that the gene is tolerant to amino acid insertions at this position and a variety of recognition sequences of natural and reengineered HEGs could be inserted and tested using this approach. We transfected the plasmids pP[v+,70I-SceI] and pEGFP-Ppo together with their corresponding target pBC/SacRB S1 and pBC/SacRB P1 into Sua4.0 cell lines (a). The plasmid pP[v+,70I-SceI] () expresses the I-SceI ORF with an N-terminal SV40 NLS and hemagglutinin (HA) tag under the control of the Hsp70 promoter, which directs a significant inducible expression of I-SceI in Sua4.0 cells (d, lanes 2 and 3). The plasmid pEGFP-Ppo expresses the EGFP-I-PpoI fusion protein containing a SV40 NLS (a and d) under the control of the CMV promoter. An inactive variant pEGFP-Ppo H98A (Raymond Monnat personal communication) was used as a control. Cells were heat-shocked for 1 h at 41°C 24 h post transfection and DNA extracted after an additional 24 h. Under these experimental conditions, we observed that the target plasmids were cut by the corresponding endonuclease (b). Digestion of DNA extracted from transfected cells with purified endonucleases revealed the presence of a fraction of endonuclease resistant plasmids (b, lanes 6–10, white arrow), possibly generated by non-homologous religation of HEG-mediated cleavage events. To recover and analyse non-homologous end joining (NHEJ) events generated in mosquito cells, total DNA (which includes plasmid DNA) extracted from transfected cells was used to transform bacterial cells plated on Cam or Cam/Suc selective media. Compared to control experiments (target plasmid only), transfection of mosquito cells with I-SceI- or I-PpoI expressing plasmids increases the number of Cam/Suc resistant clones by 15- and 8-fold, respectively (c). By comparing the colony numbers on Cam and Cam/Suc plates of bacteria transformed with recovered DNA from cells, we found that ∼0.5–2% of recovered plasmids allowed growth on Cam/Suc as compared to Cam alone. This indicates that although HEG cleavage is efficient (b) only a small number of these plasmids are subsequently religated. Plasmids from Cam/Suc-resistant bacterial cells were isolated and digested with BamHI and HindIII endonucleases. Only plasmids showing a 1.8 kb band, indicating the presence of an intact gene (a), were analysed by sequencing (). This step was undertaken to ensure that sucrose resistance is due to HEG-mediated disruption of SacRB, as we occasionally observed the growth of Cam/Suc-resistant bacteria in the absence of a HEG expression vector (c). Digestion of plasmids with BamHI/HindIII in the presence of I-SceI or I-PpoI showed that all recovered clones were also resistant to endonuclease cleavage (data not shown). The sequence of the regions surrounding the HEG cleavage sites was analysed in plasmids recovered from Cam/Suc-resistant bacteria (). All 27 sequenced clones showed deletion events of variable sizes, ranging from 1 to 80 bp, close to the predicted HEG cleavage site. We did not observe any nucleotide insertions, contrary to reports of NHEJ repair products in human cell lines (,). Copying of HEGs from one allelic site to the other requires that cleavage of the target site is followed by gene conversion events using the HEG-containing allele as a repair template. We therefore adapted our reporter system to test for the occurrence HEG-induced gene conversion events in mosquito cells. We used plasmid pDR-CMV-GFP () (a), which contains the CMV promoter 5′ of a non-functional eGFP gene. In this gene, 11 bp of the GFP CDS have been deleted and replaced by an I-SceI site which also introduces two stop codons. The plasmid also carries a second promoterless and inactive GFP locus, which lacks 220 bp of the GFP C terminus, functioning as a repair template. Cells were transfected with pDR-CMV-GFP or co-transfected with pDR-CMV-GFP and pP[v+,70I-SceI]. As expected, transfection of mosquito cells with pDR-CMV-GFP alone did not result in a high rate of spontaneous repair of the GFP gene as shown by the low frequency (0.02%) of fluorescent cells (b), whereas we found that co-transfection with pP[v+,70I-SceI] increased the number of GFP+ cells 50-fold (0.99%). Adjusted for transfection efficiency, GFP positive cells represent 0.15% (pDR-CMV-GFP) and 10.9% (pDR-CMV-GFP + pP[v+,70I-SceI]) of transfected cells. Co-transfection with the I-SceI expressing plasmid increases the average and peak intensity of fluorescence of GFP+ cells (b, lower right panel and data not shown), thus suggesting the presence of a higher number of repaired plasmids per cell in these experiments. We also investigated HEG activity in embryos. Briefly, I-SceI donor plasmid pP[v+,70I-SceI] and pBC SacRB S1 were co-injected into preblastoderm embryos. I-PpoI was not used in these experiments due to its toxic effect on insect cells (see subsequently). One day post-injection embryos were heat-shocked for 1 h at 41°C and low molecular weight DNA was extracted 2 h later. Selection for target plasmids modified by NHEJ was performed as described above for cells. As previously observed in mosquito cells, co-injection of the plasmid pP[v+,70I-SceI] increased the number of Cam/Suc-resistant bacteria colonies ∼10-fold: 0.8% of the recovered pBC/SacRB S1 pool were able to grow on Cam/Suc, a similar frequency to that obtained from cells. We analysed the sequence of the I-SceI recognition site in 20 plasmids recovered from injected embryos (b). In contrast to what we observed in mosquito cells, both deletion and insertion events were found. One of the recovered plasmids contained a 66 bp insertion homologous to an genomic sequence. In order to test for HEG-induced gene conversion events in embryos, we used the plasmid pSL-SacRB. This plasmid contains a Tet resistance cassette inserted into the gene on a pSL backbone (a and c). Therefore, this plasmid shares two regions of homology with the pBC target plasmids which flank the Tet resistance cassette but contains a different backbone (c). As shown in c and d, the outcome of a perfect gene conversion event would generate a plasmid that contains both Cam and Tet resistance cassettes but a non-functional gene. Plasmids pBC/SacRB S1, pSL-SacRB and pP[v+,70I-I] were co-injected into pre-blastoderm stage embryos and DNA extracted as described above. As a control a second set of injections was performed excluding the pP[v+,70I-SceI] plasmid. In control experiments, where only the HEG target plasmid was injected we obtained six Cam/Suc bacterial colonies and none of these were able to grow when plated on Tet. To demonstrate that gene conversion cannot be achieved by the bacteria alone, bacteria were transformed with the injection mixture and plated on Cam, Cam/Suc, Tet as well as Cam/Suc/Tet. Whereas bacteria transformed with injection mix can grow on Cam or Tet, no colonies were obtained on Cam/Suc or Cam/Suc/Tet plates. From injection experiments performed with all three plasmids (pBC/SacRB S1, pSL-SacRB and pP[v+,70I-I]) we obtained 98 bacterial colonies growing on Cam/Suc in the initial selection round. Of these 11 were also able to grow on Tet (11.2%). Further analysis revealed that two of them were unable to grow on ampicillin (Amp) (the Amp resistance gene is found in the pSL-SacRB but not in the pBC/SacRB S1 backbone). We analysed these two colonies by restriction endonuclease digestion (e) and sequencing. The sequence analysis revealed the occurrence of perfect gene conversion events, in which the Tet cassette had been inserted into pBC/SacRB S1 as predicted. The nine clones that were able to grow on Amp were resistant to cleavage with I-SceI but they appeared to be larger than the expected product. All nine plasmids were cleavable by AscI, a rare cutting enzyme that flanks the SacRB cassette (a and c), indicating that larger parts of pSL-SacRB had been transferred during the process of homologous recombination. I-PpoI mediates homing of intron 3 (Pp LSU 3) in the extrachromosomal nuclear rDNA of the acellular slime mold (). This region of the 28S rDNA is highly conserved in eukaryotes and I-PpoI has been shown to cleave human rDNA repeats (). In order to determine whether an I-PpoI site is present in rDNA, the gene of was assembled from the sequence of 31 cDNA sequences obtained from the AnoEST database. Sequence analysis indicated the presence of the full 29 bp I-PpoI recognition site. The assembled gene sequence was used to design a set of primers (rDfwd and rDrev, a) that amplify a 2 kb fragment of the gene containing the I-PpoI site (a). The PCR product from strains KWA and G3 was digested with I-PpoI to confirm the presence of the site in both strains (data not shown). To determine whether I-PpoI can bind and cleave chromosomal rDNA repeats , Sua 4.0 cells were transfected with pEGFP-Ppo or pEGFP-Ppo H98A plasmids that express a functional and an inactive version of I-PpoI respectively. Genomic DNA was extracted 24 and 48 h post-transfection and digested with ClaI or ClaI and I-PpoI (a). ClaI is predicted to cut 1.5 kb up and 1.4 kb downstream of the I-PpoI site. Southern blot analysis using the 2 kb 28S rDNA PCR product as a probe indicated that chromosomal 28S rDNA was cut efficiently by I-PpoI but not I-PpoI H98A (b). The probe hybridized to several fragments, the shortest of which corresponds to the expected 2.9 kb ClaI fragment. All but one of these fragments were cleaved by I-PpoI (b). This seems to indicate some heterogeneity in the 28S genes and we obtained a similar result with another enzyme combination (data not shown). Presumably this is a result of retrotransposon insertions downstream of the I-PpoI site (). The appearance of resistant bands, retrieved when using I-SceI in cells (b, lanes 7 and 10), was not observed when analysing genomic rDNA cleaved with I-PpoI (b, lanes 7–10). To confirm the result obtained by Southern blotting, we performed a primer extension analysis using primer rPrex, which binds 21 bases downstream of the I-PpoI cleavage site (a). c shows that I-PpoI expression causes premature stops of the primer extension at a position corresponding to the I-PpoI site. These data together indicate that I-PpoI is able to efficiently cut mosquito rDNA in transfected cells. In several mosquito species including , the 28S rDNA genes are clustered as tandem repeats on the X chromosome. The use of I-PpoI offers therefore the possibility to selectively disrupt the X chromosome. While I-PpoI cleaves the essential genes, I-SceI is not predicted to have a target site in the genome. To study the effect of expressing these two HEGs on the viability of mosquito cells, we transfected Sua 4.0 cells with pP[v+,70I-SceI], pEGFP-Ppo, pEGFP-Ppo H98A or pSL-Act-EGFP, together with the plasmid pIB/V5-His, which confers resistance to the translation inhibitor blasticidin S. Cell proliferation was assessed as a measure of growth in medium containing 50 μg/ml blasticidine, which will kill all cells that do contain the pIB/V5-His plasmid. The results of these experiments are shown in a. While neither I-SceI nor mutant I-PpoI H98A expression interferes with cell proliferation, I-PpoI expression leads to growth arrest. Expression of the I-PpoI-GFP fusion protein seemed to induce nuclear fragmentation (d) as observed by fluorescent microscopy. It appears that the nucleoli of cells transfected with pEGFP-Ppo (b) are disintegrating. This can be explained by the fact that ribosomal DNA repeats which are cut by I-PpoI form the nucleolus organizer regions of the cell. Whereas natural HEGs have evolved under constant selection pressure towards reduced toxicity and are selected to cut only at their native homing sites, engineered HEGs might be less specific. Experiments with designed zinc finger nucleases have shown that expression of these enzymes is often accompanied by general toxicity (). Our assay confirms that I-SceI, although highly active, is not toxic when expressed in cells. I-PpoI on the other hand cleaves the essential rDNA genes and induces proliferation arrest, as has been described for human cells (). The HEG activity, GFP reporter and cell proliferation assays described can be applied to the initial assessments for the suitability of any engineered HEG candidates designed against genes. Our results demonstrate that two different HEGs, I-SceI and I-PpoI, when expressed in cells and embryos, retain their ability to recognize and cleave their cognate target sequences. Analysis of I-SceI and I-PpoI function using an interplasmid activity assay in cultured mosquito cells showed that, in the absence of a template for homologous recombination, recovered target sites contained deletions ranging from 1 to 80 bp created by HEG-induced cleavage and subsequent NHEJ repair. Sequences recovered from embryos injected with the I-SceI gene contained both insertions (1–66 bp) and deletions (2–62 bp). It is not yet known whether the observed bias towards deletions in cells reflects a feature of biology or whether it is a consequence of our plasmid-based selection system. All targeted sites analysed were resistant to subsequent I-SceI and I-PpoI cleavage as nucleotide deletions or insertions in the recognition site significantly impair endonuclease activity (). Recently, it has been shown that HEGs can be engineered to confer new sequence specificities (), thus offering the possibility of using these highly specific nucleases for gene disruption and gene therapy (,). Our findings indicate that engineered HEGs could be used to selectively disrupt mosquito genes either in the germ line or in a tissue-specific manner depending on the spatial and temporal expression pattern of the driving promoter. It has been suggested that the ‘selfish’ genetic behaviour of HEGs could be exploited to rapidly spread a genetic modification affecting mosquito vectorial capacity from a few individuals to an entire population (,). This process would require both DNA cleavage and repair via homologous recombination at the targeted site using the HEG-carrying allele as template. To determine the feasibility of such an approach, we determined whether repair of HEG-mediated cleavage occurred by homologous recombination when appropriate template sequences are present. A functional study carried out in mosquito cells showed, that in the presence of I-SceI, a plasmid carrying a transcriptionally silent and incomplete eGFP coding sequence could function as template to repair (as inferred by the gain of fluorescence) another non-functional eGFP sequence disrupted by an I-SceI cleavage site. The frequency of homologous repair events in mosquito cells is comparable to that previously described in mammalian cell lines (). To extend this observation, we adapted the SacRB-based interplasmid assay to score homologous recombination events. We used for this purpose a relatively large conversion cassette containing a Tet resistance transcription unit (1.5 kb), in an attempt to simulate the size of a complete HEG element. While no recombinant clones were observed in the absence of the HEG, the precise transfer of the Tet cassette between two plasmids was found in 2 out of 98 events. We also observed nine recombination events showing larger insertions encompassing regions of the pSL plasmid flanking the Tet cassette. We speculate that these events arose by imprecise extensions during homologous recombination. Results from these experiments suggest an overall frequency of homologous repair in embryos of ∼10%. Our Tet selection system and GFP reporter system only allow us to recover events that have resulted in the complete insertions of the Tet/GFP sequences. The actual rate of homologous repair might be higher as the occurrence of putative recombination events that resulted in the transfer of smaller fragments of Tet or GFP can be expected but were selected against. Also, in none of our experiments were the repair donor fragments flanked by HEG half sites, a fact that might reduce the rate of homologous repair. In our experiments, DNA DSBs are induced by the activity of HEGs. Efficient homologous repair in Ag55 cells has also been demonstrated by the transfection of linear recombination substrates (). Together, these findings indicate in two highly predictive experimental systems that HEG would retain their gene-driving activity in mosquitoes. Furthermore, HEG-mediated cleavage of a target gene in embryos followed by gene conversion using a repair template provided could represent an alternative approach to traditional transposon or integrase-mediated germ line mosquito transformation. We have shown that I-PpoI specifically recognizes and cleaves a vital target sequence in cells inducing cell proliferation arrest and presumably cell death. The I-PpoI recognition site is located in the 28S rDNA gene within domain V of the 28S rRNA, a region that forms the peptidyl transferase centre of the ribosome. This sequence is among the most conserved sequences in the entire eukaryotic kingdom. In some mosquito species, including at least two members of the complex (), the rDNA repeats are exclusively located in the centromeric region of the X chromosome. This arrangement raises the possibility of using I-PpoI to selectively attack the X chromosome: if specifically expressed during male meiosis, I-PpoI could distort the sex ratio towards males by selectively incapacitating X-carrying spermatozoa (). Natural driving Y chromosomes in and have been described and can produce extreme sex ratios of more than 90% males (). Although the molecular details of how these distorters act are unknown, cytological evidence suggests that they are associated with breaks in the X chromosome during male meiosis I (,). These observations are relevant for the development of malaria vector control measures. A male sex distorter combining the specificity of I-PpoI together with its potential genetic drive activity if placed on the Y chromosome could knock down a wild-type Anopheline population in a few tens of generations ().

Living organisms are sensitive to DNA damage such as that caused by UV light or chemical mutagens and it is essential for genome stability that the DNA damage is accurately repaired. Many mechanisms are responsible for this repair, but damage that remains during DNA synthesis causes DNA replication to stall. To avoid such stalling there is a specific set of post-replicative repair enzymes that allows bypass of the damaged DNA and continuation of replication. A critical component of this post-replicative repair system is the Rad18 protein. This protein performs one of the earliest steps in the damage recognition and bypass. It acts as an E3 ligase for monoubiquitination of PCNA, together with its cognate E2, Rad6 (). The monoubiquitination of PCNA allows switching from normal replicative polymerases to Y-family translesion polymerases (,). These polymerases are recruited to ubiquitinated PCNA to allow bypass of the damaged lesion (). Monoubiquitination can also be followed by polyubiquitination by Ubc13/Mms2 as E2 and Rad5 as E3, leading to an error-free repair pathway that involves recombination with the newly synthesized strand (,). The importance of Rad18 is notable, since mice and chicken DT40 cells deficient in Rad18 show sensitivity to various DNA-damaging agents and enhanced genomic instability as seen by increased spontaneous sister chromatid exchange (). Rad18 is also involved in homologous recombination, cell-type-specific processes such as somatic hypermutation, and in S-phase it plays a role in single-strand break repair (). In addition, Rad18 was found to play a role in meiosis (). In agreement with this, the Rad18 protein is found in the nucleus of many different cell types, with the highest abundance in testis (,). Ubiquitin E3 ligases are multidomain proteins that confer the target specificity on a ubiquitin modification system, bringing the ubiquitin conjugating E2 enzyme and the target together. The ubiquitin conjugation activity of Rad18 on PCNA has been reconstituted (,) and was dependent on ubiquitin E1, Rad6 as E2 enzyme, ubiquitin, Mg and ATP, as well as on PCNA loaded onto DNA by the RFC complex. Under these conditions, the enzyme can efficiently monoubiquitinate all three monomers of PCNA. In this process, Rad18 must interact with the E2, Rad6 (), and the target, PCNA () and it was also found to interact with Pol-eta () and single-stranded DNA (). Rad18 utilizes multiple domains for these interactions. At its N-terminus Rad18 contains a Ring domain, common in ubiquitin E3-ligases, where the Ring domains interact with the cognate E2. In the case of mouse Rad18 (), it was shown that the E2 Rad6 interacts with the Rad18 Ring domain, and that mutations in this region result in increased sensitivity to DNA-damaging agents. A second region, in the C-terminal domain of Rad18, is also involved in Rad6 binding (,,). Deletion of the peptide 340–395 results in loss of Rad6 interaction , although localization to damage was not affected (). An unusual C2HC Zinc-finger (ZnF, residues 201–225 in human Rad18), has been identified in Rad18, which was suggested to have the potential to bind to DNA (,). A sequence that contains this ZnF, between residues 83 and 248, is important for dimer formation according to two-hybrid studies performed in yeast () and mammalian Rad18 (,). This self-association is disrupted in a mutant, where one of the Zn-binding cysteines is replaced by a phenylalanine (C207F). These experiments () gave rise to the suggestion that the ZnF domain is critical for dimerization. The C207F mutation also interferes with auto-monoubiquitination and localization of the Rad18 protein to CPD damage on DNA (), reinforcing the notion that the ZnF region is of critical importance for Rad18 function. A SAP domain (243–282 in human Rad18) is located just C-terminal of the ZnF region. This is a domain type named after SAF-A/B, Acinus and Pias, the three proteins where it was first identified (). In these proteins the SAP domains are involved in DNA interaction. In Rad18, this domain was found to be important for localization to pol-eta-containing foci (). Regions of the protein that include the ZnF and SAP domain are sufficient for localization to DNA damage even in the absence of DNA replication but mutation of C207F interferes with this localization. In the C-terminus of human Rad18 Watanabe () identified a region (401–445) that is important for Pol-eta binding. Upon UV-damage Rad18 and Rad6 are critical for formation of pol-eta-containing foci on stalled replication forks (), but other translesion polymerases, such as Rev1 play an additional role in these complexes. Altogether considerable knowledge on the Rad18 domain structure has been acquired, but relatively few experiments have been performed on the purified mammalian proteins. Here we use biophysical methods to study protein–protein interactions and map functional domains on mammalian Rad18 proteins. We show that Rad6/Rad18 forms stable dimers of heterodimers and our ubiquitination reactions and mapping of Rad6 correlates well with published data. Surprisingly we find that only the SAP domain, and not the ZnF domain, is capable of DNA recognition . In contrast, we find that the ZnF domain binds to ubiquitin, with an affinity similar to other ubiquitin interaction motifs. Bicistronic PET25b plasmid (Novagen) was constructed containing both genes expressed via one promotor but with two separate ribosome-binding sites (rbs). hRad18 was encoded untagged after the first rbs already present in the plasmid (NdeI/NcoI), while hHR6B was given a His6 N-terminal tag and its own rbs encoded in the reverse Rad18 PCR primer (AflIII/XhoI). A three-point ligation resulted in the dicistronic vector PET25b—hRad18—His6hHR6B. Alternatively, the genes were cloned separately into pET28a (Rad6) and pET22b (Rad18). The isolated Ring domain was cloned into pGEX4T, and the ZnF, SAP and ZnF-SAP domains were cloned into pETM30 (generous gift from Arie Geerlof). Mutagenesis reactions were carried our using a Quikchange kit (Stratagene) according to manufacturer's protocol. Plasmid DNA sequences were confirmed by in-house sequencing. Overnight pre-cultures were used to inoculate large cultures of LB broth at 37°C. When cell densities reached an absorbance of 1.0 at 600 nm, they were induced with a final concentration of 0.3 mM IPTG and 100 μM ZnCl and left to express at 15°C overnight (typically 15–17 h). Cleared cell lysates were incubated with Talon beads (Rad6/Rad18 complexes) or GST beads (GST-ZnF, GST-SAP, GST-ZnF-SAP and GST-RING) (Clontech) and washed with 10-column volumes of 50 mM Tris pH 8.0, 150 mM NaCl, 2 μM ZnCl and 5 mM β-ME. Further purification was achieved by Heparin binding for Rad6/Rad18 and finally size exclusion chromatography for all proteins. The single plasmid system PET25b—hRad18—His6hHR6B yielded ∼2 mg of pure protein complex per liter of culture, with an ∼10-fold excess of hHR6B, despite being encoded as the second gene. Wheat E1 (1 μM) was pre-incubated with ubiquitin (100 μM), ATP (10 mM) and MgCl (10 mM) at room temperature in reaction buffer (25 mM Tris pH 8.0, 150 mM NaCl, 2 μM ZnCl and 5 mM β-ME) for 5 min. Rad6/Rad18 (10 μM) and PCNA (35 μM) were added and the reaction incubated at 30°C for 1 hour, or longer when stated. All concentrations are final. Samples were boiled in reducing and denaturing loading buffer and run on 12% SDS–acrylamide gel. Forty-two micromolar samples of Rad18/Rad6 were incubated at 4°C for 30 s with a 4-fold excess of PCNA in 25 mM Tris pH 8.0, 150 mM NaCl, 5 mM β-ME, 2 μM ZnCl. Fifty microliters of this mixture, Rad6/Rad18 or PCNA were analyzed on a 2.4 ml SMART GF Superdex 200 column (S200), PC 3.2/30 (Amersham Biosciences). Alternatively, equivalent experiments were performed with GST-RING and Rad6, alone or in combination, and analyzed on a Superdex 75 (S75), PC 3.2/30 column. Fractions were collected and analyzed by electrophoresis on a 15% SDS-PAGE gel. MALLS experiments were performed at 4°C on a Mini-Dawn light scattering detector (Wyatt Technology) online with a Superdex S75 10/30 column in 25 mM Tris pH 8.0, 150 mM NaCl, 2 μM ZnCl, 5 mM β-ME buffer. Refractive index and light scattering detectors were calibrated against toluene and BSA. Ubiquitination sites were identified from in-gel digested protein bands by online reverse-phase nanoscale liquid chromatography tandem mass spectrometry (RP-nanoLC-MS/MS) using an AB-MDS Sciex QSTAR pulsar quadrupole time of flight mass spectrometer. The extracted tryptic peptide mixtures were auto-sampled onto a RP-C18 column packed in a nanoLC emitter, separated with a 60 min linear gradient from 4–40% MeCN in 0.5% AcOH and eluted directly into the mass spectrometer via a nano-electrospray interface. The mass spectrometer was operated in the data-dependent mode for automatically switching between MS and MS/MS acquisition. Peptide ubiquitination sites were identified via protein database searching of the resulting tandem mass spectra using Mascot. Twenty-five microliters of glutathione-beads were incubated at 4°C with either buffer (25 mM Tris pH 8.0, 150 mM NaCl, 2 μM ZnCl, 5 mM β-ME), 500 μg GST or 500 μg GST-ZnF and excess protein was washed out with buffer. The beads were then incubated with 500 μg of ubiquitin for 10 min before washing with buffer. The beads and flow-through solutions were analyzed by electrophoresis on a 15% SDS–PAGE gel. Native polyacrylamide gels were run with a final concentration of 200 μM GST-cleaved ZnF with 0, 10, 20, 40, 60, 80, 100, 140 and 200 μM ubiquitin. A control run was performed with final concentration of 100 μM GST with 0, 5, 10, 20, 30, 40, 50, 70 and 100 μM ubiquitin. After a short (30 min) incubation at 4°C, the samples were loaded on a 10% native polyacrylamide gel with a glycerol-based native loading buffer, run at 50 V for 1.5 h at room temperature and stained with Coomassie brilliant blue. A 500 nM hRad18 + hRad6 was pre-incubated with 0.1 nM P-labeled 35T DNA for 7 h at 4°C and either cold 35T or cold ds 20-mer was added in concentrations of 0, 0.01, 0.03, 0.1, 0.3, 1 and 10 μM and left overnight at 4°C. The data shown was with S471A Rad18 as this construct was slightly more stable, although wild-type Rad18 bound with a similar affinity (data not shown). Native loading buffer was added and the samples were loaded on a 4% native polyacrylamide gel at 4°C at 75 V for 45 min. The gel was dried in a slab-dryer and exposed on a phospho-imager plate for 1 h. All experiments were performed at 10°C on a Biacore T-100 instrument (Biacore AB). A specific DNA-binding surface was prepared by binding biotinylated 40-T or gel-filtered 40-mer double-stranded DNA to a SA sensor chip to a density of 20–100RU, while GST-ZnF was immobilized using an amine coupling procedure. GST-ZnF, GST-SAP or ubiquitin in 25 mM Tris, pH 8.0, 125 mM NaCl, 2 μM ZnCl, 5 mM β-ME and 0.05% [vol/vol] surfactant p20 were injected over the sensor chip at 30 μl min with a 60 or 150 s association phase followed by a 10 min dissociation phase. Binding was also tested at higher flow rates and showed no change in interaction characteristics. The sensor surface was regenerated using a 60 s pulse of 0.2 M glycine-HCl, pH 2.0 followed by a 60 s pulse of 0.05% SDS. Standard double referencing data subtraction methods were used before analysis of kinetics. Curve fitting and other data analyses were performed using Biaevaluation software (Biacore AB). In order to study the functional domain architecture of the Rad6/Rad18 complex, we have expressed this protein complex in (A, constructs used). We used both the mouse and human versions of the protein in a bicistronic co-expression construct in (), with a His-tag on the Rad6 protein for initial purification on Talon beads. The Rad6/Rad18 complex was found to be soluble and could be purified away from a significant excess of Rad6 using heparin affinity and gel-filtration chromatography (B). This excess of Rad6 was also seen when expressing Rad6 and Rad18 from independent vectors, although to somewhat lesser extent (). To determine whether the complex is functional, we established an assay for PCNA ubiquitination. In the presence of ubiquitin, ubiquitin-activating enzyme E1, magnesium and ATP, we could detect significant ubiquitination of PCNA as seen by analysis on SDS-PAGE and confirmed on western blot against PCNA and ubiquitin (C). The reaction appears to proceed in a stepwise manner, with partially modified trimers more abundant than tri-ubiquitinated PCNA. The activity is presumably suboptimal, since our PCNA is not loaded onto DNA (,). Nevertheless, the reaction is dependent on the presence of E1 and E2/E3. Moreover the complex is capable of modifying all three sites on PCNA trimers, as indicated in native PAGE (D), while peptide mapping of the ubiquitin site detected K164 as the only modified site (data not shown). Thus our assay recapitulates many of the significant features of the PCNA modification and it can serve as a functional test of enzymatic activity. We expressed a series of Rad18 deletion constructs as complexes with Rad6 using either the polycistronic system or a two-plasmid system, in order to map the site of interaction of the ligands of Rad18 onto the complex. One complex, Rad18 with Rad6 was produced as a side product of expressing the full-length mouse and human Rad6/Rad18 protein complex. This was the result of a modified Kozak sequence, which apparently acts as an alternative translation start site at methionine 312, as confirmed by N-terminal sequencing. We tested various complexes in the activity assay and we showed that Rad18 is still fully active (data not shown) and Rad18 has maintained substantial activity on PCNA (E). Thus the proposed Pol-eta interaction domain () is not essential for the PCNA ubiquitination activity by Rad6/Rad18. Using biochemical and biophysical analysis we were able to unravel the protein recognition roles of various Rad18 domains and define their function. We have confirmed the previously published Rad6 interaction domains and shown that the functional unit of Rad18 activity is a dimer of Rad6-Rad18 heterodimers. We have shown that the N-terminal half of the Rad18 protein binds to PCNA by size-exclusion chromatography. Our mapping shows that the DNA recognition region in mammalian Rad18 is not localized to the ZnF domain, but to the SAP domain (). Only fragments containing the SAP domain were able to bind DNA, as demonstrated in an electrophoresis mobility shift assay and the ZnF alone fails to do so. The SAP domain binds to DNA as shown by surface plasmon resonance. The affinity is ∼1 μM, as previously reported, and mutagenesis studies revealed that the mode of DNA binding may be conserved with other SAP family members (). This role is not unusual for SAP domains, as these are found in other DNA-interacting proteins such as KU and PARP, the histone mRNA 3′ exonuclease, Saf-A and Pias. The importance of this domain for DNA targeting fits with the data of Nakajima who found that the SAP domain of Rad18 is important for the formation of pol-eta containing foci (). Rather than DNA binding, we clearly demonstrate that the Rad18 ZnF domain plays a critical role in binding ubiquitin. This capacity is in line with other zinc fingers that form ubiquitin-binding motifs. The affinity that we measure is in the range determined for UBMs. The precise role of such ubiquitin-binding motifs is under much debate. Apparently this region is not critical for PCNA modification (), thus implying a role at a different step in the Rad18 function. The effect of the C207F mutation on the dimerization and autoubiquitination () could not be reproduced . Nevertheless, it is possible that ubiquitin binding is a step required for autoubiquitination, which subsequently regulates Rad18 localization and degradation. The identification of the domains responsible for the component interactions of Rad18 with DNA, ubiquitin, Rad6 and PCNA will be useful for designing mutants of Rad18 with altered ligand-binding properties to study the role of Rad18 further. Meanwhile these domains should not be envisaged as beads on a string. The presence of multiple domains that interact with Rad6 and PCNA make this obvious. The role of the dimer of Rad6/Rad18 is also of interest. Various different modes of interaction can be envisaged in which the multiple Ub and DNA-binding domains interact in different ways, and it will be interesting to see how these communicate with each other in time and space to execute the Rad18 function in translesion DNA synthesis. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

RNAP is an obligatory processive enzyme that must complete synthesis of the entire RNA chain since the transcripts that are released prematurely cannot re-enter transcription cycle. In bacteria, even in the absence of the tightly condensed chromatin, RNAP still encounters many roadblocks that either stall it temporarily or trigger RNA release. DNA-bound proteins, DNA lesions and various nucleic acid signals that induce pausing, arrest and termination () can hinder RNAP progression along the template. Even at saturating substrate concentrations , RNAP is moving in leaps, with its fast movement along the template punctuated by pauses (). Pausing plays numerous regulatory roles, is an obligatory step in termination pathways, and likely controls the overall rate of RNA chain elongation (). RNAP is capable of making very long RNA chains (30 000 nt long in bacteria) but its rate is rather modest compared to DNA replicases: in , elongating RNAP (a complex of αββ′ω subunits) moves at 20–90 nt/s () whereas the replication fork advances 1000 nt/s (). This relatively inefficient operation of RNAP does not represent the limit of its catalytic potential since ‘fast’ substitutions in the β and β′ subunits that significantly increase its overall rate have been described (). An attractive explanation rests on an assumption that the relatively slow rate of transcription is necessary for efficient regulation of gene expression where it provides for timely recruitment of, and response to regulatory factors, attenuation control, as well as determines folding pathways of the nascent RNA. Moreover, in bacteria transcription and translation are coupled, imposing additional restrictions on the speed that RNAP can attain without placing the nascent RNA in danger of release by Rho, which terminates the untranslated messages (). In other words, a catalytically perfect RNAP would leave little room for regulation and likely uncouple transcription and translation, while much slower RNAP would not be nimble enough to keep up with sustaining the RNA pool as it adapts to changing environmental and physiological conditions. Indeed, while different ‘fast’ and ‘slow’ viable alleles of RNAP have been isolated, they alter the apparent elongation rate by less than 3- to 5-fold in each direction (,,), whereas mutations coding for much faster or slower enzyme variants are lethal (,,). As substitutions that constitutively change the overall rate of RNA chain elongation appear to have a negative impact on fitness and are being removed by natural selection, the stage is set for transient alteration of RNAP kinetic properties by regulatory proteins. A subset of such factors (known as antiterminators) reduces pausing and termination (in other words, confers a fast phenotype) thereby helping RNAP transcribe long operons. These proteins use different nucleic acid targets during recruitment: λN binds the nascent RNA structure, λQ is recruited to the double-stranded DNA near the promoter, RfaH is recruited to the single-stranded non-template (NT) DNA strand during elongation (). The sites on RNAP to which these proteins bind are likely also distinct: we have recently concluded () that RfaH binds to the β′-subunit clamp helices (β′ CH), whereas the target sites for λ regulators are still unknown but are thought to be quite different (,). Yet all antiterminators share the ability to accelerate RNAP, suggesting that they induce similar changes in the transcription elongation complex (TEC). To date, the changes that lead to the ‘antitermination’ modification of the RNAP have not been characterized in detail, and the molecular mechanism(s) by which elongation factors or substitutions in RNAP make the enzyme faster or slower is not known: they may control nucleotide addition at every template position by affecting the common rate-limiting step (which has not been elucidated for RNAP), or influence the TEC isomerization into off-pathway states at pause and termination sites (). We have proposed that at a pause site RNAP isomerizes into a state in which nucleotide addition is slowed due to transient changes in the active site architecture (), and from which different classes of pause and termination complexes arise (). We further speculated that substitutions in RNAP may alter its propensity towards the isomerization into the slow state. In a fast RNAP, the productive alignment of the 3′ RNA end in the active site, and consequently nucleotide addition, is favored. In contrast, a slow RNAP is more likely to lose the 3′end from the active site and enter a paused state, escape from which can be delayed by two orders of magnitude. Antiterminators may act in the same regulatory pathway, switching RNAP into the fast state. Slow, pause-prone enzymes should then be hypersensitive to modification by antiterminators, whereas fast RNAPs should appear resistant to further acceleration. To test this hypothesis, we have determined effects of the RfaH on RNA chain elongation by enzymes from an expanded panel of fast and slow RNAPs, including many previously uncharacterized kinetic variants. RfaH is recruited to the TEC at specific sites (called ) and is required for expression of several long operons (). Reports from several labs indicated that RfaH acts as an antiterminator both and (), although the exact mechanism of its action remains elusive. RfaH structure and its binding site on the TEC are known (). RfaH increases the overall elongation rate , reduces pausing at mechanistically distinct regulatory sites (known as class I and II, ), and facilitates bypass of some terminators (,). Unlike other well-studied antiterminators (,), RfaH does not require any accessory proteins (e.g. NusA or NusG), and its action is not dramatically affected by addition of the cellular extract. These features allow us to dissect RfaH effect on the TEC in a highly purified model system. Our results indicate that RfaH acts primarily to reduce pausing: it fails to further accelerate the already fast mutants, as well as the wild-type RNAP transcribing under pause-free conditions, but is particularly effective with the enzymes that are prone to pausing because of amino acid changes or substrate deprivation. In contrast, RfaH cannot correct those slow phenotypes, which are due to the defects in the elementary catalytic steps rather than to the off-pathway events like pausing. We also show that the enzymes traditionally regarded as fast are better characterized as pause-resistant. We discuss our results within the framework of the current structural analysis of bacterial TECs. pIA238 encodes the wild-type ORF in pET28 (Novagen); the protein carries a His tag followed by a thrombin cleavage site (). pIA777 is a derivative of pET36b(+) (Novagen) that encodes RfaH N-domain–TEV–C-domain–[His]. Plasmid pVS10 () encodes the wild-type core RNAP. Other plasmids used to purify RNAP variants were: , β′Y795A; , β′T786V; , βH625Y; , βQ513P; , β′F773V; , βD675A; , β′Δ; , β′S793F; , β′L672D,V673D; , βS1105A; , βM1107A; , β′ΔGly. Sequences of all plasmid constructs were verified at the OSU PMGF center and will be made available upon request. All general reagents were obtained from Sigma and Fisher; NTPs and [αP]-NTPs, from GE; PCR reagents, restriction and modification enzymes, from NEB. Oligonucleotides were obtained from Integrated DNA Technologies. DNA purification kits were from Qiagen and Zymo Research. The full-length RfaH, RfaH N-domain, and RNAPs were purified as described in (). βP560S,T563I RNAP was purified as described in (). For pause assay with RNAP variants and full-length RfaH, linear DNA template generated by PCR amplification (30 nM), holo RNAP (40 nM), ApU (100 μM), and starting NTP subsets (1 μM CTP, 5 μM ATP and GTP, 10 μCi [αP]-CTP, 3000 Ci/mmol) were mixed in 100 μl of 20 mM Tris–acetate, 20 mM Na-acetate, 2 mM Mg acetate, 5% glycerol, 1 mM DTT, 0.1 mM EDTA, pH 7.9. Reactions were incubated for 15 min at 37°C. RfaH was added to 50 nM where indicated (for 3 min at 37°C) and transcription was restarted by addition of nucleotides (10 μM GTP, 150 μM ATP, CTP and UTP) and rifapentin to 100 μg/ml at 37°C. Samples were removed at 10, 20, 30, 40, 60, 90, 120, 180, 240, 360, 480, 600, 720 s and after a final 5-min incubation with 200 μM GTP, quenched by addition of an equal volume of STOP buffer (10 M urea, 20 mM EDTA, 45 mM Tris–borate; pH 8.3), and loaded on 8% denaturing urea/acrylamide (19:1) gels in 0.5× TBE. The gels were dried and analyzed using Storm 820 and ImageQuant (GE). Pause half-life (the time during which half of the complexes re-enter the elongation pathway) was determined by non-linear regression analysis. Termination assays were performed as described in (). Assays with the N-domain of RfaH were performed in 20 mM Tris–HCl, 14 mM MgCl, 20 mM NaCl, 5% glycerol, 1 mM DTT, 0.1 mM EDTA as described in (). Linear pIA226 DNA template generated by PCR amplification (100 nM), holo RNAP (120 nM), ApU (100 μM), N-domain (120 nM) (or storage buffer) and starting NTPs (1 μM GTP, 5 μM ATP and UTP, 10 μCi [αP]-GTP, 3000 Ci/mmol) were mixed in 25 μl of buffer TGC (20 mM Tris–Cl, 20 mM NaCl, 2 mM MgCl, 5% glycerol, 1 mM DTT, 0.1 mM EDTA, pH 7.9), and incubated for 15 min at 37°C. Halted A26 complexes were purified by gel filtration through AutSeq50 spin columns (GE) equilibrated in TGC, and diluted 2-fold. Reactions were initiated by the addition of 1/10 vol of 250 μM PP in TGC, samples were removed at times shown in , and quenched. We first tested RfaH effect on two model enzymes, the slow RpoB8 (βQ513P) and the fast RpoB5101 (βP560S,T563I), whose elongation, pausing and termination properties are well known (,): B8 pauses and terminates more efficiently and elongates RNA less rapidly than the wild-type RNAP, whereas B5101 displays the opposite phenotypes. We first employed the standard single-round termination assay on T, a typical intrinsic terminator from the hemolysin operon that responds to RfaH () and (), to test if these enzymes differ in their response to RfaH. We used the pIA416 linear transcription template on which the T was positioned downstream from a strong T7A1 promoter and a canonical element (A). On this template, radiolabeled transcription complexes can be halted at position G37 when transcription is initiated in the absence of UTP, with ApU dinucleotide, ATP, GTP and α–[P]CTP. The halted G37 complexes can then be chased upon addition of all four NTP substrates and RfaH. We found that RfaH (at 50 nM) decreased termination by the wild-type RNAP by ∼2-fold, but was much more effective with the slow RNAP (∼3-fold effect), and less effective (∼1.4-fold effect) with the fast variant (A). The termination assay lends support to our hypothesis but does not allow us to distinguish whether RfaH fails to bind to the βP560S,T563I TEC or is unable to trigger the post-recruitment RNAP modification. We then used the pIA349 template () that allows to monitor both RfaH recruitment to the site and its post-recruitment effect at the pause site (B); this template is identical to pIA416 except for the sequence downstream of the element. The element induces RNAP pausing at position 43 with nearly 100% efficiency ; when present, RfaH reduces pausing at U43 but dramatically delays RNAP at a site located 2 nt downstream (C45). This characteristic delay at the C45 position can be used to ascertain RfaH recruitment to the NT DNA. When halted G37 complexes were formed with wild-type RNAP and chased in the presence of RfaH, the half-life of the P was reduced ∼2.7-fold (B and A). This effect was independent of the source and method used for RNAP purification: the same results were obtained with the chromosomally-encoded RNAP purified from MRE600 cells by the standard procedure () or expressed from the multi-cistronic vector and purified using chitin-binding domain-intein tag on β′() or hexahistidine tag on β′ () during the first, affinity purification step. When the wild-type RNAP was replaced by fast or slow variants, RfaH failed to reduce pausing at the P site by βP560S,T563I, and was particularly effective with the βQ513P enzyme (B and A). Addition of RfaH produced the delay at C45 by all three enzymes, suggesting that it was recruited to the TEC in a similar fashion. We extended this analysis to include a panel of enzymes that were obtained during previous studies of resistance to antibiotics (,,,), analysis of species-specific differences in response to regulatory signals () and the mechanism of substrate selection (); these RNAPs displayed a range of elongation rates . All these RNAPs were purified through several chromatographic steps, and were free from accessory proteins. We excluded from this analysis a number of enzymes with gross defects in transcription, such as the very slow β′N458D () or β′R1106A (). We used the rate of RNAP escape from the site (in the same assay system as in B) to evaluate the post-recruitment effects of RfaH; RfaH also reduces the pause efficiency () but this parameter is difficult to measure accurately, particularly with the slow RNAPs that pause at the upstream sites more prominently, and thus arrive at the site asynchronously. We found that RfaH acted similarly on the wild-type enzyme and a variant with the ‘wild-type’ overall elongation rate (β′T786V) while failing to accelerate the fast enzymes (βH526Y/RpoB2, βD675A, β′F773V, β′ΔSI3, and βP560S,T563I) to the same extent (the defects varied, though). The slow enzymes fell into two categories: four variants (β′S793F, β′Y795A, βQ513P, and β′L672D,V673D) displayed augmented response, whereas two others (βS1105A and βM1107A)—reduced response to RfaH (A). The summary of our findings is presented in B, where the positions of these RNAP substitutions are shown in the context of the model of RfaH bound to the TEC. The observation that RfaH cannot further accelerate those RNAPs that are already fast is consistent with the proposed switch. However, there remain two trivial explanations: (i) the RfaH-binding site is altered by substitutions or (ii) RfaH does not become recruited to the fast elongation complexes because these TECs move through the signal too rapidly. Several observations argue against the first possibility. First, substitutions that confer fast phenotypes are located in different regions, far apart from each other, and in the regions that are not accessible from the enzyme surface in the context of the TEC (B). Second, RfaH-binding site is located on the tip of the β′ CH (), 50 Å away from the closest fast substitution. Third, RfaH still delays the fast enzymes immediately downstream from the site (B). Finally, RfaH affects the kinetic parameters of the fast enzymes (see subsequently). In contrast, RNAPs with substitutions or deletions in the β′CH possess ‘wild-type’ elongation properties but are totally defective in both the recruitment of RfaH at the site and the post-recruitment response to RfaH at the downstream sites (). To exclude the second possibility, we utilized the isolated N-terminal domain of RfaH in place of the full-length protein. The N-domain contains all the DNA- and RNAP-binding determinants and retains all the elongation enhancement properties of RfaH (). However, it no longer requires the site for binding to the TEC because its RNAP-binding site is always exposed (see Discussion section)—thus, the N-domain binds to (presumably) all TECs formed by the wild-type RNAP (in particular, the halted G37 in B; data not shown) and has ample opportunities to become recruited to the G37 TECs formed by the fast enzymes, provided that the RfaH-binding site remains intact. We found that the N-domain accelerated the wild-type RNAP but not the two fastest mutants, β′F773V and β′ΔSI3 (); however, the N-domain was clearly recruited to the altered TEC since it delayed all three RNAPs at the C45 site similarly. We conclude that RfaH binds to the fast RNAPs but fails to accelerate them further. If RfaH works as an anti-pausing factor, its failure to accelerate the fast enzymes could be due to their inability to pause. Indeed, all the templates that we have tested for RfaH effects on elongation thus far contained pause sites ( and ), at which the wild-type RNAP pauses with high efficiency, but the fast enzymes do not. A simple prediction of this mechanism would be that if the wild-type RNAP transcribed a template devoid of any strong pauses, particularly at saturating concentrations of substrate NTPs, it would also become resistant to RfaH action because pausing would no longer determine the overall elongation rate. Since the site is in itself a pause, we utilized an -less template pIA146 (A) that encodes the gene fragment and the -independent N-domain. While it is impossible to find a truly pause-less natural DNA template, the gene is the best model system known: it is devoid of strong regulatory pauses and has been used extensively in studies of RNA chain elongation (,). We monitored the overall elongation rate by accumulation of 1225 nt run-off RNA (B). At 1 mM NTPs, the N-domain failed to accelerate wild-type RNAP, and the mean rate was actually reduced from 30 to 25 nt/s (C). However, wild-type RNAP was modestly (1.7-fold) accelerated by the N-domain at subsaturating (25 μM) NTPs. The net effect of the RfaH N-domain on wild-type RNAP (D) was the reduction in and, more importantly, the 1.7-fold increase in the / ratio that approximates the second order rate constant for substrate binding. The increase in the / ratio was more pronounced (2.1-fold) with the slow βQ513P RNAP, which was accelerated by the RfaH N-domain even at 1 mM NTPs; apparently this enzyme was still limited by substrate availability under such conditions. Interestingly, the fastest RNAPs in our collection, β′F773V and β′Δ SI3, were actually slower than the wild type at 1 mM NTPs, and were not accelerated by N-domain at all NTPs concentrations tested (from 25 μM to 1 mM; D and data not shown). We conclude that the fast phenotype of these enzymes is due to their inability to pause rather than the higher intrinsic elongation rate. Consequently, it appears that RfaH acts by reducing pausing rather than by increasing the elongation rate, as it fails to act under the conditions that already suppress pausing. Molecular motions that underlie the formation of the paused state are not known in detail, but this isomerization likely occurs from the pre-translocated state and involves some rearrangements of the active site (see Discussion section). The increase in NTP concentration is known to reduce pausing by shifting the TEC into the post-translocated state. RfaH increases / ratio, thus effectively mimicking the increase in the NTP concentration, and may therefore also act to stabilize the post-translocated state either directly or indirectly, by increasing the TEC′s affinity for substrate NTPs during elongation. To distinguish between these possibilities, we examined the effect of RfaH on pyrophosphorolysis, a reversal of the nucleotide addition reaction that occurs in pre-translocated TEC (A) in the absence of NTPs. We used halted A26 complexes that exist predominantly in pre-translocated or even backtracked state and are sensitive to PP—or Gre-mediated transcript cleavage, respectively (). We formed halted radiolabeled A26 complexes in the absence or in the presence of the RfaH N-domain, and then challenged these complexes with low concentrations of PP (25 μM) for 10–180 s at 37°C (A); as reported previously, A26 complexes were highly sensitive to cleavage, generating 25- and 24-mer products within 10 s. We found that complexes formed with the wild-type RNAP were stabilized against PP-induced cleavage by the N-domain (C). This effect was dependent on specific RfaH–RNAP interaction because RNA cleavage in the complexes formed with the β′ ΔCH enzyme, which lacks the RfaH contact site and is unable to respond to either full-length RfaH or N-domain (), was not altered in the presence of the N-domain. The apparent protection against pyrophosphorolysis is consistent with the hypothesis that RfaH stabilizes the post-translocated state of the TEC. These results are inconsistent with the direct effect of RfaH on NTP binding: if RfaH acted primarily to increase the substrate affinity, it would facilitate forward and reverse reactions similarly at limiting substrate concentrations (such as used here) because RNAP contacts to the β and γ phosphates (pyrophosphate moiety) constitute the major fraction of the NTP interactions in the substrate-bound TEC (). In summary, our data argue that RfaH increases NTP binding indirectly by favoring the particular (post-translocated) TEC conformation that serves as a target for NTP, thereby facilitating the catalysis in the forward (nucleotide addition) direction. xref #text S u p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Several recent structural studies indicate that a number of protein folds have been repeatedly deployed as scaffolds for a biochemically diverse set of interactions with nucleic acids. Some notable examples of such folds are the RNA recognition motif (RRM)-like fold, double ψ-beta barrel (and the related EI barrel), β-grasp, S5-like fold, HhH (helix-hairpin-helix) and HTH (helix-turn-helix) [for further details see the SCOP database ()]. These folds are not only found in proteins that passively interact with nucleic acids, but also form the catalytic domains of several key enzymes involved in nucleic acid metabolism, such as nucleic acid polymerases, pseudouridine synthases, topoisomerases, RNA phosphatases and nucleases (). Detection of conserved folds and the characterization of common structural features shared by different representatives of a fold often illuminate several functional aspects of the proteins in which they are found (). In particular, such studies are useful in interpreting nucleic acid–protein interactions, predicting the active sites of enzymes that operate on nucleic acids, and uncovering the evolutionary history of complex biochemical functions observed in extant organisms(). While these nucleic acid binding domains display folds spanning the entire structural spectrum, certain generic structural classes are frequently encountered amongst them. These include small β-barrel folds (e.g. double ψ-beta barrel and the related EI barrel), several two-layered α+β folds (e.g. RRM-like, β-grasp and S5-like fold) and simple helical bundles (e.g. HhH and HTH) [see SCOP database ()]. Though, the conserved core of these nucleic acid binding domains are small compact structures, they might show several elaborations in the form of insertions and extensions that are associated with acquisition of diverse biochemical activities. Furthermore, some structures show signs of having been assembled from simpler structural units that usually need to form obligate dimers in order to attain stability. In particular, such a pathway has been invoked to explain origins of some β-barrel folds, like the 6-stranded double ψ-beta barrel (DPBB) and the EI-barrel folds, which are found in several ancient domains with major roles in nucleic acid binding and metabolism (,,,). The former domain, amongst other contexts, forms the catalytic domain of both DNA- and RNA-templated RNA polymerases, while the latter domain is found in translation elongation factors (,,). Both of these domains have been derived from duplication of the same 3-stranded precursor, followed by dimerization. Despite being assembled from a common ancestral precursor, the two folds have very different dimerization patterns of the monomer units: in the DPBB the two units interlock to form the two characteristic ψ-loops, whereas in the EI barrel they are placed adjacently without any cross-over (). The β-clip fold found in the SET methyltransferase domain, and sandwich-barrel hybrid motif fold also found in the RNA polymerases, are other comparable examples of assembly of barrel-like folds from simple 3-stranded elements (,,). The small size of many nucleic acid binding folds makes identification of their members, through entirely automatic methods, difficult. This difficulty is further compounded by the possibility of circular permutation, insertions and alternative structural arrangements seen in folds potentially evolving from accretion of simple structural elements. However, given the relatively small number of globular non-helical folds in the protein universe, these could be identified using a combination of transitive structural and topological similarity searches, and case-by-case analysis of individual folds. We were especially interested to explore if structural themes analogous to those observed in the small β-barrel folds might also be operational in small, ancient α+β folds, specifically those with functions related to nucleic acid metabolism. In particular, we sought to identify simple α+β folds with internal symmetries that might have been potentially assembled through duplication in the manner of the above-mentioned β-barrel folds. Using the above-stated multi-pronged approach, we present the discovery of a small two-layered α+β fold constructed from simple units with two strands and a helix, showing multiple topological variants emerging from different circular permutations. This fold appears to have been utilized in diverse biochemical contexts in key roles related to nucleic acid and nucleotide metabolism. Its characterization laid out in this article helps in understanding the substrate (nucleotide, nucleic acid or protein) interaction and evolution of various proteins containing this fold. The non-redundant (NR) database of protein sequences (National Center for Biotechnology Information, NIH, Bethesda, MD) was searched with the BLASTP program (). Profile searches were conducted using the PSI-BLAST program () with either single sequences or multiple alignments as queries, with a profile inclusion expectation (e) value threshold of 0.01. Searches were iterated until convergence. For queries and searches containing computationally biased segments, the statistical correction option built into the BLAST program was used. Multiple alignments were constructed using the MUSCLE and/or T-COFFEE programs (,), followed by manual adjustment based on PSI-BLAST hsp results and information provided by solved three-dimensional structures. All large-scale sequence and structure analysis procedures were carried out with the TASS software package (V. Anantharaman, SB and LA, unpublished results), a successor to the SEALS package (). Protein structures were visualized using the Swiss-PDB viewer () and cartoons were constructed with the PyMOL program (). Protein secondary structure predictions were made with the JPRED program (), using multiple alignments as queries. Phylogenetic analysis was carried out using a variety of methods including maximum-likelihood, neighbor-joining and minimum evolution (least squares) methods (). Maximum-likelihood distance matrices were constructed using the TreePuzzle 5 program () and were used as input for the construction of neighbor-joining with the Weighbor program (). Structure similarity searches were conducted using the standalone version of the DALI program called DaliLite (,) with the query structures scanned against local current version of PDB that has all chains as separate entries. The structural hits for each query was collected and parsed for congruence of strand orientation with the template structure (L3-I, PDBID: 1JJ2, chain B, 80-190).This was further confirmed by visual examination of each structure. The interacting residues of various proteins of the fold with their interacting molecules have been deduced using custom-written PERL scripts. The scripts encode interacting distance cut-off values of 5.0 and 3.5 Å between appropriate atoms in 3D for deducing the hydrophobic and polar interactions, respectively. These inferred interactions were further examined manually using Swiss-PDB viewer for confirming the contacts between residues of the fold and atomic groups of interacting partners. In the search for novel domains of unknown provenance, we surveyed the folds in the SCOP database () for uncharacterized globular inserts. In the archaeo-eukaryotic ribosomal protein L3 from the ribosomal large subunit (50S) (), which is classified as an EI barrel fold (,,,), we observed an insert (PDB: 1JJ2 chain B residues 80–190) folding into a distinct un-classified α+β domain. Examination of this domain showed that it formed a two-layered structure with a 4-stranded β-sheet, and two α-helices packing against one of the faces (). The topology of the L3 insert (L3-I) domain indicated that it was comprised of a tandem repeat of two β–β–α units. When viewed from the exposed face of the sheet, the strands show a characteristic down-up-down-up polarity (). Comparison of the topology with other two-layered α+β folds with β–β–α elements, such as () Ribonuclease PH fold, () DcoH fold and () homing endonuclease and glucose permease fold in the SCOP database showed that the L3-I was distinct from those folds (). The topology of the L3-I domain also showed that the N- and C-termini were juxtaposed, potentially allowing circularly permuted versions of such a fold to exist. In order to identify other domains with an equivalent fold, we set up a search procedure incorporating multiple criteria: () structure similarity searches of a local, current version of the PDB database were initiated with the DaliLite program. These searches were conducted transitively to account for the possibility of extreme structural divergence acting on a fold of relatively small size. () Results of these searches were filtered such that the recovered modules completely mapped on the L3-I fold and did not overlap with any previously characterized globular fold. () The resulting hits were further constrained for equivalence of strand polarity with the query, and not just topology (to account for circularly permuted versions). () Iterative sequence similarity searches, using the PSI-BLAST program, were set up with each true positive recovered in the above structure similarity searches to identify sequence homologs and the phyletic patterns of the concerned domains. The results of the procedure were represented as a network and true positives form a completely connected graph (A), which was not reproduced with a comparably high degree of inter-connectivity using other 4-stranded, two-layered folds as starting points. As a result we identified 11 different domains containing an equivalent fold, namely: () L3-I (PDB: 1JJ2 chain B residues 80–190); () siRNA silencing repressor of Tombusviruses (CIRV p19; PDB: 1RPU, chain A); () The GYF domain (PDB 1L2Z, chain A; 1WH2); () Enhancer of rudimentary proteins (ER; PDB: 1WWQ); () one domain of the tRNA Wybutosine biosynthesis enzyme Tyw3p, typified by the SSO0622 protein from (PDB: 1TLJ, chain A, residues 53–102 and 147–172); () the two related globular domains of AMMECR1 (PDB: 1WSC, chain A); (7a) the N-terminal domain of ATP-grasp enzyme superfamily (PDB: 1WR2, chain A, residues 37–116) (7b) the related domain from the DNA/RNA ligase-type nucleotidyltransferases (1V9P chain B, residues 2085–2119 and 2255–2293); () C-terminal DNA-interacting domain of DinB-like (Y-family) DNA polymerases (PDB: 1JX4, Chain A, residues 241–341); () The DNA-recombination proteins of the NinB family from λ and other related caudate bacteriophages (PDB: 1PC6 A and B); () the ribosomal protein L1 (1DWU, chain B) (). Visual examination of the above structures, when positioned equivalently as shown in , confirmed their structural congruence, strongly indicating the presence of a shared fold in these proteins. All versions are unified by the presence of a sheet in which the strands show a characteristic down-up-down-up polarity as was first noted in L3-I (). We refer to the exposed surface of the sheet as the open face, and the one packed against the two helices as the obscured face (). Accordingly, we termed this fold as RAGNYA after certain key proteins in which it was detected, encompassing its major structural variations (ibosomal protein L1 and L3, TP grasp modules, YF domain, inB, -family DNA polymerases, MMECR1). Four distinct structural variations were found in the above-mentioned 11 domains with the RAGNYA fold. Not surprisingly, three of the four major variants of the fold are related by circular permutations that result from a connection of the juxtaposed N- and C-termini of the original configuration (i.e. L3-I), and corresponding generation of new termini elsewhere in the fold (B). The first variant, typified by the original configuration observed in the L3-I domain, is additionally represented by the globular domain of the Enhancer of rudimentary proteins (ER) (,), tombusvirus p19 proteins (,), GYF domains (,) and tRNA Wybutosine biosynthesis enzyme Tyw3p (,). The second variant is characterized by a circular permutation resulting in the connection of the N- and C-termini of the first version, and concomitant generation of new N- and C-termini either just N-terminal to strand-2 or just C-terminal to strand-3 of the first version (B). This variant is represented by the C-terminal domain of the DinB-like (Y-family) DNA polymerases (,), the N-terminal domain of classical ATP-grasp module (,) and in the AMMECR1 proteins (,). The third major version has a single representative in the form of the ribosomal protein L1 (). It is typified by a circular permutation that connects the N- and C-termini of the original L3-I-like configuration while generating new termini just C-terminal to either strand-2 or to strand-4 in the original topology (B). The fourth variant represents a ‘broken-up’ version of the fold, in that it is comprised of a dimer of two identical subunits. Each monomer contributes a unit of two strands and one helix for the assembly of a complete fold ( and B). This version is currently only represented by the NinB proteins of lambdoid bacteriophages (,). The configuration of the monomeric subunit of these proteins is a simple β–α–β unit, which is effectively equivalent to the internal repeats seen in the second variant of the fold (B). Some representatives of these basic variants show additional elaborations in the form of domain insertions and extensions, as well as further duplications and permutations. Insertions of other domains into the fold are seen in versions found in Tyw3p and the ribosomal protein L1 (A). While the insertion is in an equivalent position in both of these versions, the inserted domains themselves are unrelated, implying that they occurred independently. In Tyw3p, the insert is an α+β globular domain that assumes a topology similar to the SHS2 domain () (A). The core RAGNYA domain and the insert domain together with an N-terminal extension form a pseudo-symmetric structure with a large C-shaped cleft which contains the catalytic residues required for Wybutosine synthesis in its center (see later for details). In the case of the ribosomal protein L1, the insert is a catalytically inactive version of the TOPRIM domain, which assumes a 4-stranded form of the Rossmannoid fold (,). This domain is held away from to the core RAGNYA domain by means of an extended linker and forms an independent surface for interaction with other proteins in the ribosomal subunit (D). AMMECR1 is a two domain protein that appears to have arisen from duplication of the entire RAGNYA domain. However, it additionally displays a higher order circular permutation that has resulted in the first strand of the first RAGNYA domain being permuted to the extreme C-terminus of the protein (E). This permutation results in the first RAGNYA domain effectively acquiring a topology comparable to the form seen in ribosomal protein L1. The sheets of the duplicated RAGNYA domains face each other at an angle greater than 90° resulting in a deep cleft that superficially resembles the situation in Tyw3p. The classical ATP-grasp modules have an N-terminal RAGNYA fused to a C-terminal domain related to protein kinases and PIPK C-terminal domains (,) (B). RNA/DNA ligases and the closely related capping enzymes have a peculiar version of the ATP-grasp module, wherein the two internal β–α–β repeats of the RAGNYA domain flank the kinase-like domain of the ATP-grasp module respectively at the N- and C-termini (C). This configuration could have arisen through: () a circular permutation of the classical ATP-grasp module resulting in N-terminal β–α–β unit of the original module being displaced to the C-terminus. () Alternatively, the kinase-like domain might have been secondarily inserted into the RAGNYA domain between the two β–α–β units. However, in both the classical ATP-grasp module and the nucleic acid ligases, the critical phosphate-binding lysine and base interacting residue are found in the second strand of the N-terminal β–α–β unit (B and C) (). If there was indeed a circular permutation in the ligases, then the β–α–β unit containing these residues would have been at the C-terminus. Hence, the presence of the equivalent lysine in the N-terminal β–α–β unit in both versions argues for the kinase-like domain being inserted into the middle of the RAGNYA fold in the nucleic acid ligases (C). Interestingly, this version of the RAGNYA domain is distorted due to a C-terminal extension which assumes an extended configuration and is incorporated as an additional stranded inserted in the middle of 4-stranded sheet in the core RAGNYA fold ( and C). Of the eleven distinct domains with the RAGNYA fold, seven have been shown to directly interact with either RNA or DNA. The L3-I, Tombusvirus p19 and ribosomal protein L1 interact with double-stranded (ds) regions of rRNA or siRNA-mRNA duplexes, tRNA Wybutosine biosynthesis enzyme Tyw3p with tRNA, the family Y DNA polymerase C-terminal domains and phage NinB proteins interact with DNA and the RNA/DNA ligases interact with both nucleotides and either RNA or DNA. In classical ATP-grasp modules, as well as nucleic acid ligases, the RAGNYA fold interacts with ATP. Other members of the fold, namely the enhancer of rudimentary proteins and the GYF domain have been shown to be parts of nucleoprotein complexes involved in pre-mRNA splicing and transcription or DNA replication, respectively, but there is no evidence for their direct interaction with nucleic acids (,). Structures with bound substrates are available in the case of the L3-I, Tombusvirus p19, ribosomal protein L1, Y-family DNA polymerase C-terminal and the two versions of the ATP-grasp module (). Examination of these structures reveals a common mode of substrate interaction for the RAGNYA domains, with the open face of the sheet being primarily involved in contacting the nucleic acid or nucleotide (). In the case of bound nucleic acids both hydrophobic and polar contacts are made with the backbone and the bases. For example, the RAGNYA domain of the Y-family DNA-polymerase appears to be involved in the binding of damaged DNA close to the site of abasic lesions (,). In this family the conserved basic residues from the N-terminus of strand-1 and end of strand-4 contact phosphates of the dsDNA's backbone, whereas a conserved arginine from the N-terminus of strand 4 participates in binding bases close to the site of abasic lesion. NinB proteins function in a similar capacity to the bacterial RecFOR complex, downstream of the λ-exonuclease in the early stages of recombination of λ-like phages. They have been shown to bind strongly to ssDNA and weakly to dsDNA (,). The use of different gapped substrates with ssDNA and dsDNA segments have suggested that the DNA binds across a surface cleft, whose base is formed by the open face of the RAGNYA fold in NinB. The clear preference for ssDNA as against dsDNA is atypical, given that most other nucleic acid binding members of the RAGNYA fold interact with dsDNA or dsRNA. Examination of the crystal structure of NinB revealed that the predominantly α-helical C-terminal domain of NinB obscures a part of the open face of the RAGNYA fold that is available for interaction in the other representatives of the fold. Hence, the version of the fold in NinB appears to have sufficient space only to accommodate ssDNA, thereby explaining its preferential binding properties. A comparable mode of substrate interaction (with nucleic acids, nucleotides and peptides), using the open face of the sheet, has also been observed in other structurally distinct two-layered folds with comparably sized β-sheets, such as the RRM-like, β-grasp and the S5 folds (,,). Furthermore, the preservation of a common mode of substrate binding in the RAGNYA fold, irrespective of circular permutation or constitution from separate 2-strand-1-helix elements (NinB), strongly suggests that this mode of interaction is the preferred binding mode preserved throughout the fold. These observations on the binding mode also help in predicting the mode of interaction of Tyw3p with its tRNA substrate. Tyw3p is an enzyme required for the synthesis of the modified base 2-methylthio--isopentenyladenosine or wybutosine (yW) in the anticodon loop of phenylalanine tRNA (). Based on the precedence of the other RNA–protein interactions seen in this fold, we propose that the Tyw3p would probably bind the dsRNA of the anticodon stem and present the anticodon loop to the catalytic residues. The insert domain seen in the Tyw3p is additionally likely to cooperate with the RAGNYA fold by forming a ‘roof ’ over the bound anticodon stem. Despite the common mode of substrate interaction used by most members of the RAGNYA fold, it has often been utilized for very distinct biochemical functions. A careful analysis of the structures and the underlying sequence conservation pattern revealed the different adaptations that emerged in functionally distinct versions of the fold. In Tyw3p, the nucleic acid ligases/capping enzymes, the ATP-grasp enzymes and AMMECR1 the fold has been adapted to perform enzymatic functions in very distinct ways. Tyw3p catalyzes the fourth step of the six-step synthesis of yW, in which an AdoMet donor provides a methyl group for the methylation of the available nitrogen in the central ring of the tricyclic yW precursor (). Superposition of the conservation pattern of the Tyw3p proteins on to the structure of the ortholog of Tyw3p shows that the three blocks of nearly universally conserved residues are spatially closely clustered (see Supplementary information). The first of these is an asparate derived from the N-terminal extension, the second is a motif of the form [ST]xSCxGR that lies in the junction between the SHS2-like insert and the RAGNYA fold, and the third is a conserved histidine from the end of the strand-2 of the RAGNYA fold (see A and Supplementary information). The polar nature of these conserved residues and their spatial clustering strongly indicate that they constitute the active site of Tyw3p (A). Furthermore, this predicted active site lies close to ‘one edge’ of the protein (A), suggesting that this location allows interaction with the target guanine 37 in the anticodon loop, when the anticodon stem is bound along the open face of the RAGNYA domain as proposed earlier. The presence of the single absolutely conserved cysteine suggests that the methyl transfer reaction catalyzed by this enzyme is probably very different from that catalyzed by the Rossmann fold methyltransferases, like Trm5p, which catalyzes the first step of yW synthesis (). It is likely that the cysteine actually receives the methyl group from the AdoMet cofactor and then relays it to the target Nitrogen atom on the yW precursor. Thus, emergence of a set of residues in the core RAGNYA fold, as well as in the associated insert and N-terminal extension, which were located on the ‘edge’ of the structure, appears to have given rise to a highly specific RNA modifying enzyme on the ancestral platform provided by the core RNA-binding domain. Interestingly, in addition to the superficial similarity to Tyw3p in its C-shaped structure with a deep cleft, the AMMECR1 protein also shows a comparable set of nearly absolutely conserved residues in the form of two motifs. The first of these is a RGChG (where ‘h’ is any hydrophobic residue) signature in the middle of strand-2 of the first RAGNYA domain, and a DxRa signature (where ‘a’ is any aromatic residue) at the beginning of the helix-2 of the same domain (see Supplementary information). These conserved residues form a spatially close group, when mapped on to the structures of the AMMECR1, indicating that they are likely to constitute the active site of the AMMECR1 proteins (E). In particular, the thiol group of the absolutely conserved cysteine of the AMMECR1 protein projects into the central cleft and is potentially available for a catalytic reaction (see E and Supplementary information). The second RAGNYA fold forms the floor of the cleft and does not appear to contribute any obvious catalytic residues. Hence, it appears likely that the duplicated RAGNYA folds have specialized, with the second one probably being involved mainly in substrate contact, while the first provides the catalytic residues. Thus, a similar shape and potential set of catalytic residues appear to have convergently emerged in the Tyw3p and AMMECR1 proteins. While the reaction catalyzed by AMMECR1 remains uncharacterized, examination of contextual information derived from phyletic profiles, gene neighborhood, domain fusion and protein interaction network analysis hint certain definitive possibilities. AMMECR1 is highly conserved in archaea and eukaryotes and sporadically found in certain bacterial lineages (see later for details). The strong archaeo-eukaryotic phyletic pattern is indicative of a role in core cellular functions, including RNA metabolism. In the yeast protein–protein interaction network it appears to belong to complexes including RNA transport and processing proteins such as Nup114p, Soh1p, Yra1p and Jsn1p (). In prokaryotes it shows a persistent gene-neighborhood association with a conserved radical SAM enzyme related to MiaB and a ring-opening dioxygenase. It also shows gene fusions with the gene for latter enzyme (see Supplementary information). This observation implies that AMMECR1 catalyzes a reaction in the same pathway as the radical SAM enzyme and the dioxygenase, perhaps transferring an organic radical on the conserved cysteine. Notably, Tyw1p, an enzyme prior to Tyw3p in the yW biosynthetic pathway is also a radical SAM enzyme involved in production of one of the rings of yW (). Thus, the combined contextual information points to the possibility that AMMECR1 might catalyze an as yet uncharacterized RNA base modification, like Tyw3p. RAGNYA domain in both versions of the ATP-grasp module has a set of common interactions with the nucleotide substrate. These primarily include a phosphate contact using a conserved lysine and a base interaction via hydrophobic or polar contacts mediated by an equivalently positioned residue, both from strand-2 (B and C). However, they also possess unique additional phosphate contacts. In the case of the classical ATP-grasp module this contact comes from a conserved basic residue in the N-terminal helical extension to the core RAGNYA fold, while in nucleic acid ligases it is from the C-terminal extension, which is inserted as a strand into the sheet of the RAGNYA domain (C). Furthermore, many members of the classical ATP-grasp version of the RAGNYA domain (e.g. glutathione synthetase) have a glycine-rich loop between the two β–α–β units of the fold, which provides additional contacts with the phosphates, in a manner reminiscent of the glycine rich loops seen in several other NTP-binding domains (). In both cases, the RAGNYA domain is stacked against the kinase-like fold with the bound nucleotide in between them. This suggests that the two domains might have originally functioned as stand-alone partners; with the RAGNYA domain supplying the chief nucleotide contacts and the kinase-like domain providing other key catalytic residues. This is consistent with the observation that the kinase-like domain has been independently linked either to the C-terminus or inserted between the two β–α–β units of the RAGNYA in two versions of the ATP-grasp module respectively. This is also in agreement with observations reported in previous studies, which suggest that the kinase-like domain has similarly partnered with other globular domains in classical eukaryote-type protein kinases and PIPKs (,). It appears likely that the ancestral version of the ATP-grasp RAGNYA domain had a single conserved lysine for phosphate contact and a base-contacting position on the strand-2 (B and C), which were further augmented by the additional innovations for phosphate contact as described earlier. This adaptation of the RAGNYA domain for enzymatic functions again parallels the similar deployment of the RRM-like fold as a scaffold for the catalytic activities of numerous enzymes functioning in nucleotide and nucleic acid metabolism (). Prominent examples of such RRM-like fold enzymatic domains include pseudouridine synthases, nucleic acid polymerases and nucleotide diphosphate kinases. In all these domains the exposed face of the sheet plays some role in substrate-binding, just as in the RAGNYA domains (). While most characterized members of the RAGNYA fold interact with nucleotides or nucleic acids, the version in the GYF domain is currently known to interact mainly with peptides (,). The only structurally characterized interaction, namely that with proline-rich peptides, occurs in a ‘side-on’ fashion (), via the edge of the domain, involving several conserved hydrophobic residues that are part of the domain's hydrophobic core (). This interaction appears to be important for the function of several eukaryotic GYF domains that exhibit preferential binding for different types of proline-rich peptides (,). Our discovery of bacterial GYF domains (see later for details) provides several additional leads regarding the ancestral-binding properties and functions of this domain: () In both eukaryotic and bacterial versions the aromatic and hydrophobic residues of the core are strongly conserved suggesting that interaction with proline-rich peptides via these residues is a conserved feature. () Several bacterial proteins with GYF domains also contain flanking proline-rich stretches suggesting an interaction between these and the GYF domains. () The bacterial versions are often fused to transmembrane (TM) domains or occur in conserved gene-neighborhoods encoding adjacent TM domain proteins ( and see Supplementary information). Many bacterial GYF domains are also found fused in the same polypeptide with tetratricopeptide repeat domains suggesting a role in protein–protein interactions (). These observations taken together suggest that the role in protein–protein interactions via a ‘side-on’ contact with proline is likely to be an ancestral specialization of both bacterial and eukaryotic GYF domains. However, recent studies have suggested that the GYF domain also interacts with U5-15k protein in the U5 ribonucleoprotein complex independent of proline-rich sequences (). Furthermore, the GYF domains contain a conserved position (almost always tryptophan in the bacterial versions) in the strand-3 whose side chain is exposed on the open face, which could potentially mediate contact with a substrate through hydrophobic interactions (see Supplementary information). These observations suggest that the GYF domain might contain an uncharacterized interaction mode involving the open face, as seen in other RAGNYA domains. The most conserved versions of the GYF domain in eukaryotes are found associated with the U5 RNP complex. In bacteria, conserved gene-neighborhoods and gene fusions suggest a functional association with a potential nucleic acid binding protein with Zinc-ribbons similar to TFIIB and PriN’ (see and Supplementary information). This might suggest that the GYF domain was originally derived from ancestral nucleic acid binding RAGNYA domain proteins, with a function shift for peptide interactions in nucleoprotein complexes. Subsequently, it appears to have been utilized more widely in peptide-binding contexts in other nucleoprotein complexes. The enigmatic, highly conserved ER protein has been identified in several independent protein-interaction screens to associate with the RNA polymerase complex, and as the DNA-polymerase associated protein PDIP46 (). Thus, like the GYF domain, it might represent another case of secondary adaptation of the RAGNYA domain for protein–protein interaction. Genetic studies have implicated the ER protein in regulation of pyrimidine biosynthesis, cell-cycle progression and transcriptional regulation (,,), but its exact role is yet to be uncovered. Examination of the structure and conservation pattern of the ER proteins from diverse eukaryotes suggests that the open face of the RAGNYA domain contains several patches of polar residues that could be potentially critical for its interactions with other proteins (see Supplementary information). To understand better the early history of the fold, we compared the phyletic patterns of all domains containing it (). Most groups appear to have deep evolutionary histories—the ribosomal protein L1, three families of the classical ATP-grasp module, and at least one family of nucleic acid ligase are conserved in most groups of organisms, and include representatives from all the three superkingdoms of Life (bacteria, archaea and eukaryotes). This suggests that they trace back to the last universal common ancestor (LUCA) of all extant life forms. Of the nucleic acid ligases, the archaeo-eukaryotic clade universally conserves an ATP-dependent DNA ligase, whereas the bacterial clade universally contains a NAD-dependent form. This suggests that they possibly diverged from each other in early evolution from a precursor present in LUCA. Three families of the classical ATP-grasp domain that might potentially trace back to LUCA are the carbamoyl phosphate synthetase, pyruvate phosphate dikinase and the ribosomal protein S6 α- glutamate ligase (RimK). These observations imply that by the time of LUCA not only had the ATP-grasp module diversified into its classical and nucleic acid ligase-like versions, but the classical version had itself further radiated to occupy very distinct functional niches related to amino acid and nucleotide metabolism, protein modification and pyruvate metabolism. Thus, the cooperation between the RAGNYA domain and the kinase-like domain, and the distinct associations between these two domains (fusion or domain insertions) had all taken place prior to LUCA. Additionally, presence of a distinct dsRNA-binding version in the form of the L1 protein implies that the differentiation between the RNA-binding and nucleotide-binding versions had also occurred in this period. A possible corollary of this long pre-LUCA history of the RAGNYA domain is that it first emerged in the RNA world itself, in the form of a generic nucleic acid/nucleotide-binding domain. It subsequently appears to have further differentiated into specialized nucleotide-binding versions as in the ATP-grasp domain and RNA-binding versions. The situation in the L3-I domain is more complicated—the ribosomal protein L3 itself is present throughout the three superkingdoms of life, but both structural comparisons and sequence similarity searches detect the L3-I domain only in the archaeo-eukaryotic orthologs. An examination of the bacterial L3 ortholog reveals that an insert is present in the equivalent region, which appears to be poorly structured in comparison to the L3-I domain. Nevertheless, at least two extended regions and one helical segment can be identified in this insert suggesting that it could have emerged through the loss or degeneration of part of the original L3-I domain. Thus, the L3-I module was potentially present in the ancestral L3 protein, and emerged as a part of the radiation of RAGNYA fold domains prior to LUCA. New RNA-binding roles appear to have been acquired later in evolution as suggested by the Tyw3p protein (in the archaeo-eukaryotic lineage) and perhaps in the AMMECR1 protein. The tombusvirus siRNA repressor appears to be a late virus-specific innovation perhaps acquired from the structurally closely related L3-I domain of the eukaryotic host. The distribution of the Y-family DNA polymerases with the RAGNYA domain in all the three superkingdoms might imply their presence in LUCA. However, their sporadic distribution in archaea, along with the lack of a clear signal for vertical evolutionary relationship between the versions from the three superkingdoms raises the possibility of a later origin and dispersal through lateral gene transfers, especially amongst the prokaryotes. The DNA-binding NinB versions of the RAGNYA domains are distributed only in lambdoid siphoviruses, a few podoviruses or their prophage remnants in bacterial genomes. In many lambdoid siphoviruses of low GC Gram-positive bacteria we observed fusions between their NinB ortholog and a nuclease domain of the EndoVII (HNH) fold (see Supplementary information) (,). This implies that NinB might collaborate in different phages with unrelated families of nucleases (e.g. lambda exonuclease and HNH) in genome recombination. The potential secondary reassembly of the version of the RAGNYA fold in the Y-family DNA polymerases and the presence of an equivalent stand-alone monomeric unit in NinB suggest that these DNA-binding versions might share a more recent ancestral monomeric precursor. Until now the peptide-binding version of the RAGNYA fold, the GYF domain was found only in eukaryotes, including the basal most eukaryotic lineages. However, using a sequence profile constructed from eukaryotic representatives of the GYF domain we were able to detect bacterial homologs with significant -values (e.g. RB6375, gi: 32474220 from was recovered with e = 10 in iteration 3). Conversely, reciprocal searches initiated with bacterial proteins (e.g. gi: 84704887, PB2503_12664 from ) recovered eukaryotic GYF domains with significant -values ( < 10) within six iterations. As a result of these searches we identified numerous bacterial GYF domain proteins from diverse bacterial lineages including planctomycetes-chlamydia, bacteroidetes, proteobacteria, firmicutes, actinobacteria and cyanobacteria (see Supplementary information). This wide distribution in bacteria is also accompanied by considerable domain architectural diversity greater than that observed in eukaryotes (). However, no GYF domains were found in archaea. Given the widespread presence in bacteria, especially α-proteobacterial lineages that spawned the eukaryotic mitochondrion (), it is likely that the GYF domain first arose in bacteria, and was transferred to the ancestral eukaryote perhaps during the mitochondrial endosymbiosis. Interestingly, its recruitment to roles in spliceosomal complexes, like the U5 snRNP, and in endosomal-trafficking proteins like RME8 suggest that the acquisition of the GYF domain from the bacteria might have played an important role in the emergence of quintessentially eukaryotic systems (,). We present the identification of a new fold with nucleic acid, nucleotide or peptide-binding properties, shared by 11 distinct functionally diverse protein domains. The RAGNYA fold is characterized by the presence of an internal symmetry constituted by two topologically identical units, each with two strands and one helix. This structural peculiarity of the fold has resulted in it displaying at least three different circular permuted versions, and one case of re-assembly from two monomeric units derived from different polypeptides. In spite of this, the fold retains a distinctive exposed β-sheet with a unique strand polarity, and most members of the fold bind nucleic acids or nucleotides via this face. The RAGNYA fold appears to have been utilized as a scaffold on multiple occasions in the generation of novel enzymatic activities, as exemplified by the ATP-grasp enzymes, the nucleic acid ligases, the Tyw3p AdoMet-dependent tRNA modifying enzyme and the experimentally uncharacterized AMMECR1 enzyme. Based on our analysis of this fold we present functional predictions that help in explaining the reaction mechanisms and substrate-binding of the Tyw3p enzyme, and the possible function of the AMMECR1 protein. We also provide evidence that the two versions of the ATP-grasp module have been assembled independently either via a C-terminal fusion to a kinase-like domain or insertion of the kinase-like domain into the RAGNYA domain. Analysis of the RAGNYA domain also helped us to identify the common denominator in the nucleic acid-binding mode of the Y-family polymerase C-terminal DNA binding domain, phage NinB proteins, the siRNA repressor of Tombusviruses and ribosomal proteins L1 and L3. We also identify for the first time the bacterial GYF domains, which might lead to better understanding of the different modes of interaction of this version of the RAGNYA fold with peptide substrates. We hope that the results presented here open up new avenues for the experimental investigation of this diverse group of proteins unified by a subtle, yet functionally significant structural feature. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Aminoacyl-tRNA synthetases (aaRS) catalyze the esterification of an amino acid to the 3′-end of its cognate tRNA in a two-step reaction. In the first step, the amino acid is activated in the form of an enzyme-bound aminoacyl–adenylate intermediate, while the second step involves the transfer of the amino acid to the 3′-end of its cognate tRNA to produce an aminoacyl-tRNA (). Based on two distinct ATP-binding cores, the 20 aaRSs are equally divided into two classes (,). In most cases, formation of the aminoacyl-adenylate does not require the presence of its cognate tRNA and can be performed by the isolated catalytic domain of some synthetases (,). However, for three class I synthetases, arginyl-, glutaminyl- and glutamyl-tRNA synthetase (ArgRS, GlnRS and GluRS, respectively), as well as the exceptional class I lysyl-tRNA synthetase (LysRS), tRNAs are absolutely required to accomplish the first step of the reaction. The requirement of tRNA in aminoacyl-adenylate formation has been confirmed for ArgRS from different sources () and is likely a common feature of this enzyme. An intact 3′-terminal adenosine of its cognate tRNA is also required, because arginine activation cannot be induced by periodate-treated tRNA (,,,). Proper tRNA structure is also necessary for arginine activation (). For GluRS and GlnRS, amino acids also cannot be activated without their cognate tRNA (). The activator function also requires the integrity of the 3′ terminus of the tRNA, and chemical modification of this terminus abolishes this activity (,,). Exceptionally, several bacteria and archaea have the class I-type LysRS instead of the class II-type LysRS (,). Class I LysRS is structurally related to GluRS (), and does not catalyze lysyl-adenylate formation in the absence of tRNA (). However, the biological significance of requiring tRNA to activate amino acids remains puzzling. The fact that ArgRS requires its cognate tRNA to activate arginine has led to the suggestion that the arginyl–adenylate intermediate does not exist in the arginylation reaction (), although Kern and Lapointe (,) later demonstrated that GluRS charges tRNA with glutamate in two steps and that the glutamyl–adenylate intermediate exists, despite the fact that tRNA is needed for the first step. Tryptophanyl-tRNA synthetase (TrpRS) is considered a class I aaRS, and together with all other synthetases except for ArgRS, GluRS, GlnRS and class I LysRS, can activate an amino acid substrate smoothly in the absence of cognate tRNA (). We recently showed that the ATP-PPi exchange reaction of a mutated TrpRS can be modulated by its inactivated tRNA (). Eukaryotic TrpRS differs from prokaryotic TrpRS in containing an appended N-terminus. For human TrpRS, the N-terminal domain is composed of an appended peptide and an eukaryote-specific patch (E82–K154) containing a β1–β2 hairpin structure adjacent to the ATP-binding pocket and the KMSAS loop (). Deletion of this hairpin severely reduces aminoacylation activity, suggesting that human TrpRS has a more complicated aminoacylation mechanism than TrpRS. At the β1–β2 hairpin of human TrpRS (1R6T from the Protein Data Bank), the backbone of V85 and V90 form a hydrogen bond. The side chains of the two valines protrude into the catalytic domain, which looks like a buckhorn (A). In this study, we mutated V85 and V90 and unexpectedly found that the mutant could no longer activate tryptophan even though its tryptophanylation activity still proceeded normally. Furthermore, we confirmed that the tryptophan activation of can be partially rescued by the addition of bovine tRNA. We constructed further mutants at V85 to further investigate and found that the V85E mutation also caused a switch to a tRNA-dependent mode of amino acid activation, just like ArgRS, GluRS, GlnRS and class I LysRS. Structural analysis showed that V85 can interact with I311 through a hydrophobic interaction to affect tryptophan activation, which was confirmed by I311 mutations. The results indicate that an aaRS that does not normally require tRNA for amino acid activation can be switched to a tRNA-dependent mode. Bovine tRNA was expressed in JM109 and human TrpRS and its mutants were expressed in BL21-CodonPlus (DE3)-RIL. BL21-CodonPlus (DE3)-RIL was purchased from Stratagene (La Jolla, CA, USA). The bovine tRNA gene was cloned into the pGEM-9Zf (–) plasmid obtained from Promega (Madison, WI, USA) and was a kind gift from Dr Xue Hong at the Hong Kong University of Science and Technology. The plasmid pTrc-hTrpRS containing the human gene was also a gift from Dr Xue Hong. The human gene expression vector pET24a (+) was purchased from Novagen. TrpRS protein sequences from archaebacteria and eukaryotes were downloaded from the aaRS database () (). Abbreviations of the species are as follows: E-Hs, ; E-Bt, ; E-Ms, ; E-Dm, ; E-Ce, ; E-Sp, ; E-Ec, ; A-St, ; A-Su, ; A-Pa, ; A-Pf, ; A-Pe, . The first letter E in the abbreviation means eukaryote, and A means archaea. Multiple sequence alignments were carried out with Clustal X 1.83 software (ftp-igbmc.u-strasbg.fr/pub/ClustalX/). The pdb files of human TrpRSs (1O5T, 1R6T, 1ULH and 2DR2) were downloaded from the Protein Data Bank (). Structure comparisons and figure production were carried out with Swiss-PdbViewer 3.7 software, which was downloaded from . The operation pathway of Swiss-PdbViewer 3.7 software for structure comparison was Fit\Magic Fit\CA (carbon alpha) only\. Expression and purification of human TrpRS were carried out as described previously (). Concentrations of purified proteins were determined with the Bradford reagent (). Mutagenesis was carried out by PCR as described previously (). Mutant human genes were verified by DNA sequencing. Abbreviations of the mutations were as follows: V85A, valine 85 to alanine; V85E, valine 85 to glutamate; V85K, valine 85 to lysine; V85L, valine 85 to leucine; V85S, valine 85 to serine; V90A, valine 90 to alanine; V90S, valine 90 to serine; V85A/V90A, V85A/V90A double mutations; I311V, isoleucine 311 to valine; I311E, isoleucine 311 to glutamate. The bovine tRNA used for the enzyme activity assay was expressed and purified as described previously (). The oxidation of tRNA by sodium periodate was carried out as previously described (,) with some modifications. Bovine tRNA was treated with 10 mM sodium periodate in 1 ml of 150 mM sodium acetate (pH 5.3), incubated at 4°C in the dark for 1 h. Glucose was then added to a final concentration of 10 mM to remove excess sodium periodate. The tRNA was incubated at 37°C for 30 min and then precipitated with ethanol. The ATP-PPi exchange reaction was assayed at 30°C in a reaction mixture containing 100 mM Tris–HCl pH 7.8, 10 mM potassium fluoride, 10 mM magnesium chloride, 4 mM ATP, 0.02 μCi [γ-P]ATP, 4 mM sodium pyrophosphate and 2 mM tryptophan in total volume of 20 μl. Reactions were initiated by the addition of 0.2 μM enzyme. At each time point, samples were quenched on ice. One microliter of the reaction mixture was spotted onto PEI cellulose TLC plates (purchased from Merck, Germany) and developed in 1 M KHPO and 1 M urea to separate the ATP and PPi. The radioactivity was revealed and quantified using a PhosphorImager™ (Molecular Dynamics, Little Chalfont, Bucks, UK). For mutants and , the tRNA-dependent ATP-PPi exchange reaction was assayed. Wild-type bovine tRNA and oxidized bovine tRNA were added to the PPi exchange reaction mixture to a final concentration of 60 μM. The other components for the reaction were identical to that described above. The aminoacylation assay was also carried out at 30°C in a mixture containing 4 mM ATP, 0.8 mM DTT, 1 μCi -[5-H]tryptophan, 12 μM -tryptophan, 8 mM MgCl, 80 mM Tris–HCl, pH 7.5 and 12 μM purified bovine tRNA in a total volume of 50 μl. Reactions were initiated by the addition of 50 nM enzyme. At each time point, samples were quenched on ice and 20 μl aliquots were spotted onto filter paper disks, which were washed three times with ice-cold 5% trichloroacetic acid containing 0.05% tryptophan and with cold anhydrous ethanol, dried and transferred into vials for determination of radioactivity. For all kinetic assays, the concentration of tRNA varied from 0.2 to 12.8 μM. Each reaction was repeated at least four times under the same conditions. The / for aminoacylation was calculated from Eadie–Hofstee plots. All data were fitted to the Michaelis–Menten equation. Based on a structural analysis of human TrpRS, we found that there were two adjacent valines in the middle of the β1 and β2 sheet, V85 and V90, whose side chains both approach the substrate-binding pocket (A). The β1–β2 hairpin is essential to the aminoacylation activity of human TrpRS (), and therefore these two valine residues may be important for enzyme activity. Based on sequence alignment, we found that V85 and V90 are conserved in eukaryotes and archaebacteria (B). Therefore, we constructed mutations at these two positions, V85 and V90, to generate the and variants. In the ATP-PPi exchange reaction, the and mutants showed decreased, but visible tryptophan activation activity, with having the lowest enzyme activity (A). Furthermore, when V85 was mutated to serine, no ATP-PPi exchange activity was detected, even at a very high enzyme concentration (5.1 µM) (A). Unexpectedly, the aminoacylation activity of the mutant remained comparable to the other three mutant enzymes (B) and could not catalyze the tryptophanylation reaction in the absence of ATP (B). This suggested that the aminoacylation reaction of the mutant proceeds normally and that the mutation does not change the recognition of ATP. These results implied that the mutant might be able to activate tryptophan only in the presence of tRNA. To investigate whether the mutant assumes a tRNA-dependent ATP-PPi exchange activity, wild-type or sodium periodate-treated bovine tRNA was added to the ATP-PPi exchange reaction mixture. Unexpectedly, both the wild type and the sodium periodate-treated bovine tRNA were able to partially rescue the ATP-PPi exchange activity of the mutant (C). Sodium pyrophosphate can greatly inhibit the tryptophanylation reaction. After adding the wild-type tRNAs to the ATP-PPi exchange reaction, the tryptophanyl-tRNA could hardly be detected (data not shown). Thus, the tRNA-independent ATP-PPi exchange reaction for human TrpRS is switched to a tRNA-dependent mode by the single mutation. However, in contrast to ArgRS, GlnRS and GluRS (,,,,), the integrity of the 3′ terminus of the tRNA was not necessary for the ATP-PPi exchange activity of the mutant (C). Therefore, it is reasonable to conclude that human TrpRS is switched to a ‘nonproductive’ form by mutation but can be induced to a ‘productive’ form by tRNA binding. Comparing A B, we conclude that V85 mutations have a greater impact on tryptophan activation than the tryptophanylation reaction. To further investigate the role of the V85 residue in the ATP-PPi exchange reaction, we constructed three more mutations at this position including and . Relative to the V85E and V85K mutations, the mutant did not show a large decrease in tryptophan activation (A). However, the and mutants lacked the ATP-PPi exchange activity of human TrpRS under normal conditions (A), just like the mutant (A). In the tryptophanylation reaction, the mutant was able to acylate bovine tRNA with very high efficiency (B). Similar to the mutant, the mutant has low but noticeable activity (B). In contrast, the mutant had barely detectable aminoacylation activity (B). In the presence of oxidized bovine tRNA, the TrpRS was unable to catalyze the ATP-PPi exchange reaction, but the wild-type bovine tRNA could still serve as an activator (C), which suggests that the hydroxyl groups of the 3′ adenosine of tRNA can also activate human TrpRS. For the mutant, wild-type tRNA could slightly rescue its PPi exchange activity while tRNA treated with sodium periodate could not (C). Thus, we conclude that mutation of V85 to a hydrophilic amino acid such as serine, lysine and glutamate abolishes the normal ATP-PPi exchange activity of human TrpRS (A and A). On the other hand, replacing V85 with a neutral amino acid such as alanine, or a hydrophobic amino acid such as leucine, did not produce the same result (A and A). Moreover, the more hydrophobic the side chains at the V85 position, the higher the enzyme activity. The single mutation switches human TrpRS from a tRNA-independent to a tRNA-dependent mode of tryptophan activation (). For the mutant, the tRNA molecule, not the hydroxyl group at the 3′ end, serves as an activator, which is different from ArgRS, GluRS and GlnRS (C). The mutant was activated to catalyze the ATP-PPi exchange reaction only in the presence of wild-type bovine tRNA (C). This suggests that the 2′ or 3′ hydroxyl group at the 3′ adenosine of tRNA also can act as an activator, similar to ArgRS, GluRS and GlnRS. The mutant had barely detectable enzyme activity. Based on the recent co-crystal structure of human TrpRS and bovine tRNA, we concluded that V85 should be close to a tRNA acceptor stem when tRNA is bound to human TrpRS (,). Thus, the phosphate groups of tRNAs should be easily affected by the V85 mutation. To verify our hypothesis, we calculated the of tRNA for the and mutants. For the mutant, the value was 2.14 μM, similar to 1.30 μM for wild-type human TrpRS (). The value was 8.50 μM for and 0.294 μM for (). Therefore, it is reasonable to conclude that the mutant that leaves the side chain uncharged has little effect on tRNA binding, so not only wild type but also oxidized tRNA is able to properly bind to the enzyme to prompt the ATP-PPi exchange reaction. Because the side chain of glutamate is negatively charged, the mutation would push out the negatively charged phosphate groups of tRNAs, weakening tRNA binding. The mutant, with a positively charged side chain, can attract the negatively charged phosphate groups of tRNAs. As a result, tRNA may be bound to human TrpRS at an incorrect position when V85 was mutated to a lysine. Therefore, the fact that the mutant has the lowest acceptor activity is reasonable. We next calculated the for tRNA and the values of all the above mutations (). These varied dramatically at the β1–β2 hairpin, with ranging from 0.294 to 24.8 μM and ranging from 9.44 × 10 S to 5.11 S. The values suggest that V85 and V90 can interact with tRNA when tRNA binds to human TrpRS. Thus, it is logical that charge variations of the side chains at V85 and V90 have the greatest impact on tRNA acceptor activity. Based on the crystal structure of the full-length human TrpRS (1R6T), we found that V85 is far away from the substrate-binding pocket (A) and therefore could not directly affect tryptophan activation. However, the side chain of V85 is only 4.41 Å away from the side chain of I311 (A). I311 is a residue in the conserved AIDQ sequence. A310 and D312 interact with the ribose of ATP through hydrogen bonds to stabilize ATP binding (). Combining all findings, we concluded that V85 may affect I311 through hydrophobic interactions between their side chains to affect tryptophan activation. Furthermore, we compared the crystal structure of unliganded human mini-TrpRS (1ULH), which contains the β1–β2 hairpin structure at one monomer, with the unliganded crystal structure of T2-TrpRS (1O5T) in which the β1–β2 hairpin structure is hydrolyzed. Both the mini-TrpRS and T2-TrpRS are truncated forms of human TrpRS at the N-terminus with mini-TrpRS encoding residues 48–471 and T2-TrpRS encoding residues 94–471. We found that the AIDQ sequence of mini-TrpRS assumes a compact form (A), while the AIDQ sequence of T2-TrpRS adopts a loose form (A). When bovine tRNA binds to T2-TrpRS (2DR2), as expected, its AIDQ sequence switches to the compact form (B). Therefore, it is reasonable to conclude that, at least partially, the interaction of V85 with I311 can induce the AIDQ sequence to an active compact form. If V85 is mutated to a hydrophilic amino acid that disrupts the hydrophobic interaction between V85 and I311 (such as serine, glutamate and lysine), the conformational change in the AIDQ sequence might not be induced correctly. Alternatively, because the 3′ end of tRNA and the AIDQ sequence interact (), tRNA can substitute for the function of the β1–β2 hairpin. To confirm our hypothesis, we mutated I311 to valine and glutamate and determined their enzyme activity. As we had expected, when the isoleucine was mutated to acidic glutamate, the normal ATP-PPi exchange activity of human TrpRS was abolished (A), but the mutant enzyme could still catalyze the tryptophanylation reaction with decreased efficiency (B). Wild-type bovine tRNA, not the oxidized form, was able to promote the ATP-PPi exchange reaction (C), just like the V85E mutation. Compared with wild-type human TrpRS, the I311V mutation had little effect on both the ATP-PPi exchange reaction and tryptophanylation reaction relative to the large effect of the mutant on both reactions (A and B). Therefore, the above inference should be true. Relative to the mutant, the mutant has even lower aminoacylation activity. The mutant might interfere with the interaction between the A76 ribose of bovine tRNA and AIDQ sequence (). We have found that the V90 mutations had little effect on the ATP-PPi exchange reaction. Mutants of human TrpRS at position 90 only showed reduced ATP-PPi exchange activity. Based on structural analysis of the full-length human TrpRS (1R6T), we found that the side chain of V90 is farther away from the side chain of I311 than that of V85. Thus, V90 should only act as a subsidiary residue to enhance the hydrophobic interaction between V85 and I311. Therefore, we constructed a double mutant of human TrpRS, . As expected, the behavior of was similar to the mutant (). We also calculated the and for the double mutant, which were 2.23 μM and 0.0471 S, respectively (). Previous studies have shown that tRNA oxidized by sodium periodate can form Schiff bases with the ε-NH of lysine residues close to the binding site (,), irreversibly inhibiting enzyme activity. To clarify whether the oxidized bovine tRNA can form Schiff bases with human TrpRS, we pre-incubated the and mutants with oxidized tRNA for 10 min, 20 min, 30 min and 40 min, and then performed the aminoacylation assay (see Materials and Methods section for aminoacylation conditions). After 10 min, reactions were stopped and enzyme activity was determined. The results showed that oxidized tRNA did not inhibit the activity of the enzymes (Supplementary Figure S1). Therefore, Schiff bases do not form between oxidized tRNA and TrpRS. ArgRS, GluRS, GlnRS and Class I LysRS all share the common feature that is the absolute requirement of tRNA in the amino acid activation step. For these aaRSs, the tRNA is postulated to serve as the enzyme activator in the first step and as the substrate in the second step of aminoacylation. Except for these four enzymes, the other class I aaRSs do not require the presence of tRNA to activate amino acid, such as TrpRS. Here, we combined mutations to illustrate that human TrpRS can be switched to a tRNA-dependent mode to activate tryptophan by single or double mutation(s) in the β1–β2 hairpin and AIDQ sequence. These mutant enzymes were all tested at a high concentration (>2.5 μM) and were unable to activate tryptophan in the absence of tRNA. This finding was unexpected. These mutations abolished the PPi exchange activity of TrpRS, which can be partially rescued by the addition of tRNA. For ArgRS, GluRS and GlnRS, comparisons of the tRNA-free and tRNA-bound enzyme crystals have revealed tRNA-induced enzyme conformational changes (), which should be responsible for amino acid activation. The situation for human TrpRS should be the same and was confirmed by structural comparison (). On the other hand, for wild-type human TrpRS, V85 should perform the same function, at least partially, as tRNA in the tryptophan activation reaction (). Conformational changes in human TrpRS induced by V85 would presumably require specific molecular interactions. We have identified an interface between the eukaryotic-specific patch and the catalytic center, which might be responsible, at least in part, for molecular interactions that would facilitate tryptophan activation. This site is defined in human TrpRS by the eukaryotic-specific patch that contains V85 at the β1–β2 hairpin and I311 at the conserved AIDQ sequence which directly interacts with ATP, based on the crystal structure of full-length human TrpRS (). The side chains of V85 and I311 are as little as 4.41 Å apart. A comparison of the X-ray crystal structures of T2-TrpRS and mini-TrpRS (both not bound by substrates) illustrates that the β1–β2 hairpin is responsible for the ‘productive’ form of human TrpRS (). We concluded that, at least partially, the compact form of the AIDQ sequence of human TrpRS is ensured by hydrophobic interactions between the side chains of V85 and I311, because the side chain of V90 is much farther away from the side chain of I311 than that of V85. V90 should be a minor residue to interact with I311 by hydrophobic interaction, which was confirmed by the double mutant. A310 and D312 are known to be able to form hydrogen bonds with the ribose of ATP to stabilize ATP binding. Therefore, if V85 is mutated to a hydrophilic amino acid, the hydrophobic interaction between V85 and I311 will be disrupted and the AIDQ sequence might assume a loose form. As a result, the hydrogen bond interaction between ATP and the AIDQ sequence will be disrupted and ATP may not be able to bind TrpRS stably. Thus, mutated human TrpRS cannot catalyze tryptophan activation until tRNA is also bound. For V85E and I311E, the oxidized tRNA cannot rescue their PPi exchange activity, but wild-type tRNA still can (C and C). After treatment with sodium periodate, the 2′ and 3′ hydroxyl groups of tRNA are oxidized to aldehyde groups. This should not result in any large changes in the hydrogen bonds between the A76 of tRNA and residues D312 and Q313 of the AIDQ sequence (), otherwise the tRNA oxidized by sodium periodate would not be able to bind to human TrpRS properly and rescue the PPi exchange activity of the mutant and the double mutants. These results imply that the 2′ or 3′ hydroxyl group also can serve as an activator, just like ArgRS, GluRS and GlnRS. As enzyme activity proceeds, the 2′ or 3′ hydroxyl group of A76 will consequentially interact with the Trp–adenylate intermediate to accept activated tryptophan. The interaction between tRNA and Trp-adenylate may stabilize ATP binding allowing the human TrpRS to activate tryptophan, even though V85 or I311 was mutated to glutamate. Furthermore, the sequence alignments show that V85 and V90 are conserved in eukaryotes and archaebacteria, including the proline and tryptophan in the β1–β2 hairpin (B). Therefore, the β1–β2 hairpin structure of human TrpRS should be conserved in eukaryotes and archaebacteria and its function should be the same as that in human TrpRS. Wild-type human TrpRS or mini-TrpRS does not require its cognate tRNA to activate tryptophan (). However, we can switch its tRNA-independent PPi exchange activity to a tRNA-dependent mode by introducing mutations within the eukaryote-specific patch (E82–K154), which is not found in prokaryotic enzymes. It raises the question of whether the tRNA-independent PPi exchange activity of human TrpRS is an ancestral feature or has been more recently acquired with the acquisition of the eukaryote-specific patch. Our results cannot give a definite answer, although it seems that the partition between tRNA-independent and tRNA-dependent PPi exchange activity within a single aaRS can shift dramatically with relative evolutionary ease. Based on sequence alignment, we recently found a TrpRS from , a hyper-thermophilic archaebacterium. This TrpRS has eukaryotic features, but the eukaryotic-specific patch is completely absent in this enzyme. This archeabacterial TrpRS could provide further evidence to define the evolutionary connection between TrpRS and ArgRS-like aaRSs. For ArgRS, GluRS, GlnRS and class I LysRS, the biological significance of requiring tRNA as a cofactor to activate amino acid is unclear. However, a previous study showed that tRNA binding can help these four class I aaRSs specially recognize their cognate amino acid. The co-crystal structure of GluRS and tRNA with or without glutamate demonstrated that the amino acid-binding site of GluRS is incomplete and the presence of the cognate tRNA facilitates glutamate binding by the enzyme (). In the absence of tRNA, GluRS binds not only -glutamate but also noncognate amino acids such as -glutamate, -aspartate and -glutamine (). The binding of the noncognate amino acids is eliminated by tRNA binding to the enzyme. Similarly, the co-crystal structure of GlnRS indicated that the tRNA itself is involved in the Gln-AMP-binding site (). For ArgRS, the and ArgRSs can bind the cognate arginine only in the presence of tRNA (,). A similar phenomenon was also reported for class I LysRS (). The recently discovered dual specificity enzyme, prolyl-tRNA synthetase (ProRS), can acylate not only proline but also cysteine, although it can only recognize cysteine in the presence of Trna (,). The existence of aaRSs that can catalyze the synthesis of more than one aminoacyl-tRNA is assumed to be an important step in the evolution of these enzymes (). tRNA-dependent amino acid recognition may be a relic of synthetase evolution. However, a recent paper by Carter () showed that the minimal TrpRS catalytic domain from can still catalyze the tryptophan activation step without tRNA and also mischarge tRNA. Therefore, the biological significance of tRNA-dependent amino acid recognition remains puzzling. Otherwise, is the tRNA-dependent amino acid recognition inherent to class I aaRSs? Maybe we can get the answers in the future. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Werner syndrome (WS) is a rare, autosomal recessive disease characterized by early onset and increased frequency of many phenotypes normally associated with human aging including graying and loss of hair, wrinkling and ulceration of skin, cancer, atherosclerosis, cataracts, osteoporosis, diabetes and hypertension (,). Intriguingly, all of these phenotypes result from loss of function of a single gene product, WRN, belonging to the RecQ family of DNA helicases () that includes the prototype RecQ in , Sgs1 in , Rqh1 in and four other family members in humans. Importantly, the highly cancer-prone Bloom syndrome (BS) is caused by mutations in human RecQ family member BLM (), while Rothmund–Thomson (RTS), RAPADILINO and Baller–Gerold syndromes are caused by different mutations in family member RECQL4 (). The RECQL4-related syndromes are collectively characterized by abnormalities in skeletal development and skin pigmentation (poikiloderma), but RTS also shows an elevated incidence of osteosarcoma. At the cellular level, loss of function of a RecQ family member generally results in increased spontaneous and damage-induced chromosomal aberrations, suggesting crucial functions for these proteins in maintaining large-scale genome stability. In agreement with this notion, WRN-deficient cells have higher frequencies of chromosomal deletions, insertions and translocations and are more sensitive to selected DNA damaging agents (including replication inhibitors, topoisomerase I inhibitors and interstrand crosslinking agents) than cells derived from normal individuals (). Moreover, primary fibroblasts from individuals with WS rapidly undergo senescence in culture, apparently as a result of the inability to properly maintain their telomeres (). Accumulation of senescent cells or the cumulative loss of cells by apoptosis is almost assuredly the cause of premature aging in WS; these mechanisms are also postulated to play a role in normal aging (). Thus, the WS phenotypes may point to specific tissues in which these mechanisms may be at work during the development of aging phenotypes in normal individuals. All RecQ helicases including WRN are highly homologous within defined amino acid sequence motifs that are also identifiable but less conserved in a larger group of enzymes from helicase superfamilies 1 and 2. These sequence motifs anchor a domain that, in RecQ helicases, uses the energy derived from ATP hydrolysis to unwind DNA with a 3′ to 5′ polarity defined by the strand upon which the enzyme translocates. WRN and other RecQ helicases unwind short duplexes and a variety of unusual DNA structures including forks, D-loops, G-quartets and triplexes (). Both WRN and BLM also readily branch migrate Holliday junctions (,). Recently, a number of RecQ helicases, including the human WRN, BLM, RECQ1, RECQL4 and RECQ5β proteins, have been shown to facilitate annealing of complementary DNA strands (). Under certain circumstances, the unwinding and annealing activities of WRN, BLM or RecQ5β can function coordinately to achieve strand exchange (,). Taken together, biochemical studies suggest that some RecQ helicases are structurally designed to act on three- or four-stranded replication or recombination intermediates. Importantly, WRN is the only human RecQ homolog to have a nuclease domain in its N-terminal region () that confers a 3′ to 5′ exonuclease activity that is particularly robust on complex DNA structures and thus similar to the specificity of its unwinding activity (). Although each of these DNA-dependent activities has been independently examined , it remains unclear whether and how they might act together in a specific DNA metabolic pathway to help maintain genome stability. Although multiple DNA repair systems are present in all cells, encounters between replication forks and persistent DNA damage cannot be completely avoided. Recent investigations indicate that cells have evolved important pathways to respond to and overcome replication fork blockage caused by lesions in the DNA template or other circumstances (). It has been proposed that the initial step in dealing with a blocked replication fork involves its regression—a process by which the parental strands re-anneal and the daughter strands are paired to generate a Holliday junction or so-called ‘chicken foot’ structure (A). Following Holliday junction formation, several alternative pathways might be employed for removing or circumventing the obstacle and restarting replication. With each pathway, eventual re-establishment of a functional replication fork is crucial for maintaining genomic stability and permitting cell survival. Because of the cellular phenotypes caused by RecQ deficiencies, they are often postulated to participate in pathways responding to fork blockage (). More specifically, WRN-deficient cells show an extension of S phase and specific replication abnormalities including asymmetry in the normal bidirectional progression of replication forks, suggesting difficulty in overcoming obstacles to replication (,). They are also hypersensitive to compounds such as hydroxyurea, topoisomerase inhibitors and interstrand crosslinking agents that inhibit replication fork progression (,,,). Moreover, immunofluorescence studies in normal cells demonstrate that WRN is present in some replication foci and is actively recruited to these foci by treatment with certain genotoxic agents (,,). If WRN or other RecQ helicases act in pathways that respond to replication fork blockage, loss of their function might cause sporadic replication fork collapse, leading to generation of double-strand breaks, elevated genomic instability and an increased likelihood of cell death. At the minimum, the regression of blocked replication forks to form Holliday junctions would entail unwinding of both parental–daughter duplex arms and pairing of the nascent daughter strands with concomitant re-annealing of the parental strands. Accurate completion of this complex process would be facilitated by an enzyme that possesses both unwinding and strand annealing capability such as WRN, or perhaps another RecQ helicase. Some exonucleolytic processing of either the leading or lagging daughter strand may also be involved in the fork regression process. A preliminary report from our laboratory has shown that BLM and an exonuclease-deficient WRN mutant, WRN-E84A, can catalyze fork regression (). In this study, a series of replication fork substrates have been used to determine the effect of leading arm structure on the fork regression capabilities of WRN and BLM. Our results show a pronounced effect of leading arm structure on the efficiency of regression mediated by WRN-E84A and BLM. Importantly, the 3′ to 5′ exonuclease activity of wild-type WRN enhances regression on a number of these structures through limited degradation of the leading daughter strand. Thus, our results indicate that the multiple enzymatic activities of WRN act together to mediate regression of replication forks. Furthermore, we demonstrate (on another model fork substrate) that WRN can mediate fork regression to form the Holliday junction or ‘chicken foot’ structure characteristic of a fork regression process. A function of WRN to specifically regress blocked forks during replication fork repair would be highly consistent with the specific genomic instability phenotypes associated with WS. Wild-type WRN (WRN-wt), WRN-E84A, and WRN-K577M were overexpressed in insect cells and purified essentially as described previously (), except that 0.1% Nonidet P40 (NP40) was included in all liquid chromatography buffers. WRN-E84A contains a point mutation in the conserved nuclease domain that inactivates its 3′ to 5′ exonuclease activity (); this mutant retains DNA unwinding and annealing activities (,). WRN-K577M contains a point mutation inactivating its unwinding activity but still retains exonuclease and annealing activities (,,). Recombinant human BLM, purified after overexpression in yeast as described previously (), was provided by Joanna Groden (Ohio State University). The Holliday junction resolvase RusA was purified as previously described (), except that RusA overexpression in was at 25°C and the lysis step was performed in 1.5 M KCl. UvrD was provided by Steven Matson (University of North Carolina) while both PriA and Rep were from Ken Marians (Sloan-Kettering); these proteins were purified by previously described methods (,) Standards of known concentration were used to determine protein concentrations using the Bradford assay and/or SDS–PAGE. All proteins were stored at −80°C prior to use. Nucleotide sequences of gel-purified oligonucleotides (Midland Certified Reagent Company, Midland, TX) are specified in . The 3′ ends of the 70lag, 70lead and 30lag oligomers were modified with phosphate groups that block the 3′ to 5′ exonuclease activity of WRN-wt and WRN-K577M (). For construction of short fork substrates with both the lagging parental and leading daughter strands labeled, the 70lag, 21lead, 24lead, 27lead, 30lead and 32lead oligomers were 5′ end-labeled with P-γ-ATP and T4 polynucleotide kinase, 3′-phosphatase free (Roche Molecular Biologicals, Indianapolis, IN) and unincorporated nucleotides were removed using standard procedures. In an initial annealing step to form parental-daughter partial duplexes, labeled 70lag was heated to 90°C and slow-cooled with excess unlabeled 30lag, while unlabeled 70lead was treated similarly in individual reactions with excess labeled 21lead, 24lead, 27lead, 30lead or 32lead. The resulting lagging and leading parental–daughter partial duplexes were then mixed together at 37°C for 18 h. The long fork substrate was prepared similarly, except it contained radiolabels on both the lagging daughter (82lag) and leading parental (122lead) strands. Three-stranded forks were also prepared from sequential high and low-temperature annealing reactions, but without one of the leading or lagging daughter strands. Double-stranded substrates were prepared from single-step annealing reactions. Single-stranded oligonucleotides used for markers and in annealing reactions were simply labeled and gel-purified. After separation by native 8% polyacrylamide gel electrophoresis (PAGE), all DNA substrates were excised, extracted into TEN buffer (10 mM Tris, pH 8.0, 1 mM EDTA and 10 mM NaCl), and stored at 4°C prior to use. All enzymatic assays were conducted in WRN reaction buffer (40 mM Tris–HCl, pH 7.0, 4 mM MgCl, 0.1% NP40, 100 ug/ml bovine serum albumin and 5 mM dithiothreitol); fork regression, exonuclease and helicase assays also contained ATP (1 mM) unless otherwise indicated. For these assays, labeled fork (regression), partial duplex (helicase) or oligomeric (annealing) DNA substrate (50–200 pM) was pre-incubated for 5 min at 4°C with enzyme (WRN-E84A, WRN-K577M, WRN-wt, BLM, UvrD, Rep or PriA) at the concentrations in figure legends, then transferred to 37°C for the indicated times. In annealing reactions, complementary single-stranded oligomer (50 pM) was added just prior to incubation at 37°C. For potential detection of Holliday junctions during regression assays with long fork substrate, RusA (10–40 nM) was added 1 min into the 37°C incubation. Reactions (or aliquots thereof) were stopped by adding either one-sixth volume of helicase dyes (30% glycerol, 50 mM EDTA, 0.9% SDS, 0.25% bromphenol blue and 0.25% xylene cyanol) or an equal volume of formamide dye (95% formamide, 20 mM EDTA, 0.1% bromphenol blue and 0.1% xylene cyanol) for analysis by native 8% PAGE or denaturing 14% PAGE, respectively. Specific DNA species (daughter duplexes and RusA-generated products) identified on native PAGE were excised and extracted using a gel extraction kit (Qiagen) then re-analyzed by denaturing 14% PAGE. DNA products on native and denaturing gels were visualized and quantitated using a Storm 860 phosphoimager and ImageQuant software (GE Healthcare). In fork regression assays, radioactivity associated with individual DNA species was measured. For specific kinetic experiments (F), the amounts (as a percentage of total radioactivity) of each DNA species were determined and directly compared. To calculate enzyme-mediated generation of DNA products, the percentage of each product with respect to the total (molar) amount of original fork substrate in that reaction was quantitated following subtraction of background levels of respective DNA species in reactions without enzyme. For analysis of WRN exonuclease activity during regression reactions, the amounts of radioactivity associated with intact and digested products of the leading daughter strand were determined and the percentage of each length product with respect to the total radioactivity derived from the leading daughter strand in that lane was quantitated. This data for each product derived from the leading daughter strand from individual lanes is comparatively presented for daughter duplexes extracted from native PAGE (C). Alternatively, the percentages of each product formed after 5 min of digestion are compared with the amount of the respective product in the undigested substrate (0 min time point), and the percent change over time for each product is plotted (B). Our earlier experiments () indicated that WRN and BLM could act on a model fork structure with homologous arms to generate both parental and daughter duplexes, consistent with the possibility that these enzymes might regress replication forks . We next wanted to determine whether and how the precise structure at the fork junction might influence these novel regression activities. To this end, a series of model four-stranded replication fork substrates was constructed from individual oligomers (for details, see ‘Materials and methods’ section and for nucleotide sequences). These short fork substrates (A) contained a 38 bp parental duplex region, a lagging parental–daughter arm of 30 bp plus a 2 nt single-stranded gap at the fork junction, and a leading parental–daughter arm with a parental strand region of 32 nt but, on individual fork substrates, the leading daughter strand varied in length from 32 to 21 nt resulting in single-stranded gaps of 0–11 nt at the fork junction. Individual short fork substrates are identified below by their leading daughter strand (32lead fork) and/or the size of the single-stranded gap on the leading arm. Importantly, lagging and leading parental–daughter arms of these substrates were entirely homologous (except for 5 non-complementary nt on each parental strand precisely at the fork junction included to prevent spontaneous branch migration), permitting pairing between daughter strands and re-annealing of parental strands to form Holliday junction intermediates and eventually produce both parental and daughter duplexes (A). With the exception of the leading daughter strand, the other 3′ ends of these substrates were modified to block the 3′ to 5′ exonuclease activity of WRN. These short fork substrates were radiolabeled on the 5′ ends of both the lagging parental strand (70lag) and the leading daughter strand (21lead, 24lead, 27lead, 30lead or 32lead) to facilitate identification of multiple DNA products. In assays on this series of replication fork substrates, we initially used an exonuclease-deficient protein, WRN-E84A, to avoid potential degradation of the leading daughter strand that might complicate interpretation of our results. To determine the influence of leading arm structure on regression activity, WRN-E84A was incubated with individual fork substrates with gaps of 0, 2, 5, 8 and 11 nt on the leading arm at the fork junction. Our fork substrates contain three duplex regions potentially subject to unwinding when treated with a DNA helicase such as WRN. Specifically, forward unwinding of the parental duplex region of the fork would yield two parental–daughter partial duplexes (PDs), while unwinding of either parental–daughter arm would yield a three-stranded fork and a displaced daughter strand. However, experiments performed with these substrates with low (sub-equimolar to a 3-fold molar excess) concentrations of WRN-E84A produced, within 5 min, primarily two species that co-migrated with markers for the parental (70 bp) and daughter duplexes (B). In comparison, other DNA species (some present in low amounts in substrate preparations) including three-stranded forks, parental–daughter partial duplexes, and leading daughter strands were not produced in significant amounts by WRN-E84A. The parental duplex could be the result of a fork regression event but also might be generated from spontaneous annealing of the partially hybridized parental strands following unwinding of both parental–daughter duplex regions. However, the daughter duplex could only arise from the unwinding of both parental–daughter arms of the fork combined with annealing of the daughter strands and thus specifically reflects fork regression. Importantly, the generation of daughter duplex (as well as parental duplex) products was observed for all fork substrates but was dramatically increased using the fork substrate (21lead fork) with an 11 nt gap as compared to fork substrates with smaller gaps (B). Higher concentrations of WRN-E84A mediated increased conversion to daughter duplexes for each substrate, but the relative efficiencies of regression were preserved between substrates. Quantitation of data obtained over a wider range of WRN-E84A concentration (C) demonstrated clearly that daughter duplex formation preferentially occurred when the fork substrate contained an 11 nt gap. Using a fork with a smaller gap of 8 nt reduced daughter duplex formation precipitously, while further shortening of the gap lowered the efficiency of this reaction further. This data was corroborated by kinetic experiments performed using a fixed concentration of WRN-E84A on each substrate. In these assays (D), the daughter duplexes formed were clearly detectable and increased linearly with time but were relatively modest for fork substrates with gaps of 0, 2, 5 and 8 nt. In contrast, daughter duplex formation from the substrate with an 11 nt gap was markedly higher at each time point (reaching about 60% conversion by 5 min) than for the substrates with smaller gaps. It is notable that generation of the leading daughter strand product is minimal over the same time frame for each substrate (D, inset), again suggesting that the daughter duplex formation occurs through direct coordination between unwinding of both parental–daughter arms and pairing of the daughter strands. Taken together, our data indicates that daughter duplex formation (indicative of fork regression) by exonuclease-deficient WRN-E84A occurs much more readily on replication fork substrate with a larger (11 nt) gap on the leading strand than on forks with smaller gaps. The efficiency of regression by WRN-E84A drops considerably when the gap is shortened to 8 nt and decreases further on substrates with even smaller gaps. Our results on these substrates confirm our earlier observation (using a structurally different fork substrate) that WRN catalyzes a reaction reminiscent of fork regression (). Moreover, they suggest that, although WRN-E84A has a certain amount of structural flexibility, the efficiency of this reaction is determined by structure of the leading arm at the fork junction. Since daughter duplex formation catalyzed by WRN-E84A on 21lead fork substrate containing an 11 nt leading arm gap was so much more efficient than on other substrates, a more in-depth analysis of WRN-E84A action on this substrate was performed. The amount of each detectable DNA species from a kinetic experiment on this substrate (E) was determined at each time point. Then, the contribution (expressed as percentage) of each DNA species to the total radioactivity was plotted over time (F). Before the beginning of the reaction, the fork substrate contained 91.6% of the total radioactivity with no other individual species contributing more than 2.5%. After initiation of the WRN-E84A-mediated reaction, only the amounts of four-stranded fork, daughter duplex and parental duplex changed significantly; at any time point, not one of the other DNA species (lagging parental strand, parental-daughter partial duplexes, three-stranded fork or leading daughter strand) ever contributed more than 6.3% to the total radioactivity. Most notably, the amount of fork substrate decreased dramatically with time while the amounts of parental and daughter duplex increased (E and F). Importantly, the combined increases in parental and daughter duplex products () almost exactly reflected the decreases () in fork substrate between individual time points (F). Thus, it can be concluded that, during the course of this reaction, the radioactivity in the fork substrate (labeled on one parental and one daughter strand) was distributed in a concerted manner between the parental and daughter duplex products without significant generation of other intermediates. This strongly suggests that WRN-E84A, in a reaction mimicking fork regression, catalyzes direct and coordinated conversion of our short fork substrate with the 11 nt gap on the leading arm to parental and daughter duplexes. In all likelihood, this mechanism also applies to WRN-mediated (and possibly BLM-mediated) action on other, less favorable, fork substrates. The most straightforward analysis of these results is that WRN-E84A produces daughter (and parental) duplex from fork substrates by coordinately unwinding the parental–daughter arms and annealing the leading and lagging daughter strands by an intramolecular strand exchange reaction. However, we wanted to determine whether daughter duplex formation might occur by strand exchange between independent DNA molecules. To this end, we examined the action of WRN-E84A on two different three-stranded forks, one containing only the lagging daughter strand and the other only the leading daughter strand. When WRN-E84A was incubated with only one of these three-stranded forks, no formation of daughter duplex was possible and, indeed, only other DNA species (leading daughter strand, parental duplex and parental–daughter duplexes) are produced (Supplemental , ). As expected, production of daughter duplex () is observed when WRN-E84A is incubated with four-stranded fork containing both leading and lagging daughter strands (Supplemental , ). If WRN-E84A is incubated with both three-stranded forks simultaneously, a daughter duplex might conceivably occur via intermolecular strand exchange between the different forks. When this experiment was performed (with concentrations of both three-stranded fork substrates either half or equal to that of the four-stranded fork in the positive control), no daughter duplex was produced; instead, we observed only DNA species corresponding to those formed by unwinding of individual three-stranded forks (Supplemental , ). This result indicates that WRN-mediated intermolecular strand exchange does not detectably occur between these three-stranded replication forks under the same conditions in which daughter duplex is produced from a four-stranded replication fork substrate. Thus, daughter duplex formation from a four-stranded fork by WRN-E84A likely occurs by an intramolecular strand exchange mechanism, a concept even more strongly supported by our experiments showing that WRN can regress our longer fork substrates to form Holliday junctions (). Although our kinetic experiments showed little or no production of free leading daughter strands at the near equimolar WRN-E84A concentrations that mediated daughter duplex formation, we wanted to more thoroughly examine whether daughter duplex formation was a concerted process or the result of possibly independent unwinding and annealing steps. Thus, annealing reactions were performed both with complementary daughter and parental oligomers. Although WRN-mediated annealing of 80-mers can be achieved in reactions with or without ATP (), these reactions were performed without ATP to minimize potential unwinding of duplex products. In a concentration-dependent manner, WRN-E84A annealed the parental oligomers to generate the 70 bp parental duplex (Supplemental , ), confirming its previously reported annealing capability (). However, at WRN-E84A concentrations equal to and significantly higher than needed for fork regression, enzyme-mediated annealing of lagging daughter oligomer (30lag) to any of the leading daughter oligomers was not detected (Supplemental , ). These experiments and others (A. Machwe, unpublished results) indicate that WRN does not facilitate annealing of free oligomers when both are relatively short (<30 nt). A similar effect of oligomer length on BLM-mediated annealing has been previously reported (). More importantly, the inability of WRN-E84A (and BLM) to anneal these oligomers when free in solution demonstrates that formation of daughter duplex from our fork substrates does not occur by independent enzyme-mediated unwinding and annealing steps. Instead, daughter strands appear to be paired while still associated with the fork, a process likely facilitated by some structural property inherent in WRN and BLM. These results indicate that daughter duplex formation results from an intimate linkage between unwinding of the parental–daughter arms and pairing of the daughter strands. Together with our analysis of the production of daughter and parental duplexes from 21lead fork (F), WRN-mediated conversion of our fork substrates to daughter (and parental) duplexes reflects a fork regression process. With regard to protein specificity, it is worthwhile to determine whether this fork regression function is limited to WRN (and BLM) or perhaps common to other enzymes with helicase activity. Since the UvrD, Rep and PriA proteins have been implicated in the response to replication fork blockage in (,), we examined the action of these helicases on our replication fork substrates. When increasing concentrations of these proteins were incubated with the 21lead, 27lead and 32lead fork substrates containing leading arm gaps of 11, 5 and 0 nt, respectively, no significant production of daughter duplex (reflective of fork regression) was observed (Supplemental ), in contrast to our results with WRN-E84A. Both UvrD and Rep produced almost exclusively parental duplex and free leading daughter strand. This pattern suggests that UvrD and Rep displace both daughter strands, allowing the already linked parental strands to completely re-anneal but (unlike WRN) without pairing of the displaced daughter strands. The behavior of PriA varied depending on fork structure. On forks with 11 and 5 nt leading arm gaps, PriA not only produced parental duplexes and free leading daughter strand but also generated a significant amount of parental–daughter partial duplexes, the latter due to forward unwinding of the fork. However, on 32lead fork without a leading arm gap, PriA generates predominantly a three-stranded fork (without concomitant release of labeled leading daughter strand). This indicates that PriA preferentially unwinds the lagging daughter strand on forks without a leading arm gap, in agreement with previous biochemical analyses (). Thus, fork regression is not a common property of all helicases, but limited to a subset that includes WRN and BLM. Notably, RecG can also regress our model replication fork substrates (data not shown), as would be expected from previous studies (). Our data above indicate that fork substrates containing larger single-stranded gaps on the leading arm are superior structures for regression by exonuclease-deficient WRN-E84A. Since an increase in gap size within these forked structures could be accomplished by degradation from the 3′ end of the leading daughter strand, we hypothesized that the 3′ to 5′ exonuclease activity of WRN-wt might also participate in the fork regression process. To initially test the merit of this theory, we compared the action of exonuclease-proficient WRN-wt with that of WRN-E84A on our fork substrates with different gap sizes on the leading arm. Wild-type BLM, which does not contain exonuclease activity, was also examined to directly compare its action with that of both WRN proteins. First, we measured unwinding by WRN-wt, WRN-E84A, and BLM on a 27 bp partial duplex substrate (70lag/27lag, with both 3′ ends modified to block WRN exonuclease) containing a 3′ overhang of 43 nt (A) to determine the amounts of each enzyme needed for comparable activity within the linear range of unwinding. Notably, the same molar concentrations of WRN-E84A and WRN-wt catalyzed nearly equal levels of unwinding; BLM-mediated unwinding was also of similar strength on this 3′ overhang substrate. Then, the action of these enzymes (using concentrations with the same relative unwinding strength) was directly compared on our fork substrates containing leading strand gaps of 11, 8, 5, 2 and 0 nt. These experiments confirm that each of our replication fork substrates can be acted upon by both WRN-E84A and WRN-wt to primarily yield parental duplex and daughter duplex products as well as minor amounts of leading daughter strand (B). As judged by formation of daughter duplex, the regression activities of WRN-wt and WRN-E84A on fork substrate with an 11 nt gap are comparable. However, as the gap size on the leading arm decreases, daughter duplex formation becomes much more efficient with WRN-wt than with WRN-E84A (C). This difference is particularly notable on the 27lead fork with a 5 nt leading arm gap (B, , and C). A titration of both proteins on this fork substrate also demonstrated that WRN-wt is consistently much more efficient than WRN-E84A in forming the daughter duplex reflective of the fork regression process (D). BLM is less efficient at fork regression than WRN-wt on all forks except the substrate without a gap on the leading arm (for which regression activity is very weak for each protein), as evidenced by consistently lower daughter duplex formation and higher production of free leading daughter strand (B and C). In assessing the action of WRN-wt on these fork substrates, two other differences from WRN-E84A (and BLM) were observed. In reactions containing WRN-wt, there was a faint smear extending down from the leading daughter strand band, reflecting minor generation of shorter leading daughter strands. More importantly, for the fork substrates with gaps of less than 11 nt, daughter duplexes produced by WRN-wt migrated slightly faster than those formed by WRN-E84A or BLM, indicating that one or both strands that compose this DNA species had been altered by WRN-wt. As the other strands in the fork substrates are blocked at their 3′ ends, these effects appear to be due to the 3′ to 5′ exonuclease activity inherent in WRN-wt specifically on the leading daughter strand during the course of the reaction. Importantly, this data could imply that the exonuclease activity of WRN-wt may digest the leading daughter strand prior to or during formation of daughter duplexes, particularly on replication fork substrates with gaps on the leading arm shorter than 11 nt. The above analysis of WRN-wt action on our replication fork substrates suggests that its 3′ to 5′ exonuclease activity clearly participates in production of faster-migrating daughter duplexes. Since the 3′ ends of other strands in our fork substrates are blocked, WRN-mediated degradation is likely targeted specifically to the leading daughter strand. However, the timing of this exonucleolytic digestion with respect to other enzymatic events (i.e. regression) was unclear. One way to approach this question was to determine whether (and how) WRN exonuclease activity might digest possible reaction products that contain the leading daughter strand. Thus, the action of WRN-wt on both isolated lead daughter strand oligomers (21lead, 24lead, 27lead and 30lead) and daughter duplex substrates (21lead/30lag, 24lead/30lag, 27lead/30lag and 30lead/30lag) as well as our replication fork substrate (27lead fork) with a 5 nt gap was examined by denaturing PAGE. The 27lead fork substrate was utilized here and subsequently due to the obvious benefit that the exonuclease activity of WRN-wt provides for the regression efficiency on this substrate (B–D). In this experiment (as well as later ones), the labeled lagging parental strand (70 nt) of the fork substrate that is modified at its 3′ end is not detectably digested by WRN exonuclease (, ). Importantly, at enzyme concentrations comparable to those used in regression assays above, the exonuclease activity of WRN-wt could not detectably digest either pre-formed daughter duplexes (, ) or isolated leading daughter strand oligomers (, ). In dramatic contrast, the leading daughter strand was extensively digested in the context of the intact 27lead fork substrate (, ). This result on single-stranded oligomers is in agreement with earlier studies showing that oligomers of <30 nt are very poor substrates for WRN exonuclease (). These experiments on isolated daughter duplexes and daughter strand oligomers are highly relevant to the timing of WRN exonuclease activity in our reactions containing short fork substrates. Specifically, once daughter duplexes are formed or leading daughter strands are displaced, they are not subject to WRN exonuclease activity. Therefore, digestion of the leading daughter strand of the fork substrate by WRN-wt must have occurred before or concomitant with (but not after) the formation of daughter duplex. To further investigate a possible relationship between digestion of the leading daughter strand and regression, the effect of ATP on WRN exonuclease activity was examined. Reactions with 27lead fork substrate and either WRN-wt or the ATPase- and helicase-deficient WRN-K577M mutant were analyzed in parallel for regression and exonuclease activity in the presence or absence of ATP. For these experiments, concentrations of WRN-wt and WRN-K577M that yielded approximately equal levels of exonuclease activity on the leading daughter strand within this fork substrate in the absence of ATP were used. When DNA products of kinetic assays over 5 min with and without ATP were analyzed by native PAGE, formation of daughter (and parental) duplexes occurs readily with WRN-wt in the presence of ATP but not detectably in its absence; moreover, WRN-K577M does not catalyze formation of daughter duplex regardless of the presence or absence of ATP (A, ). WRN-wt and WRN-E84A also do not catalyze formation of daughter duplex in the presence of the weakly hydrolyzable analog ATPγS (data not shown). These results demonstrate that formation of daughter duplex indicative of fork regression requires the ATPase and helicase activities of WRN. When these same reactions were analyzed by denaturing PAGE, some degradation of the leading daughter strand by WRN-wt and WRN-K577M is observed whether or not ATP is present (A, ). However, digestion of the leading daughter strand is much more pronounced with WRN-wt in the presence of ATP in comparison to each of the other conditions, as indicated both by the direct digestion patterns (A, ) and by the quantitation of relative changes in leading daughter strand length ranging from 27 nt (undigested) to 17 nt after 5 min (B). While the other conditions have remarkably similar digestion profiles, reactions containing WRN-wt in the presence of ATP show, by 5 min, a much higher level of exonucleolytic activity on the leading daughter strand by two criteria. They have 1) a much more dramatic (approximately 2.5-fold greater) loss of signal associated with undigested leading daughter strand (27 nt) and 2) a concomitantly increased generation of shorter leading daughter strand products with lengths between 26 and 19 nt (B, ). Also notable in this digestion pattern are elevated amounts of 19- and 20-mers compared to several longer leading daughter strand products (see next paragraph for further clarification). Since only reactions including WRN-wt with ATP showed both formation of daughter duplex and enhanced exonuclease activity on the leading daughter strand, these results suggest that optimal WRN exonuclease activity may be connected to partial unwinding of the leading daughter strand and perhaps regression of the fork substrate. These experiments not only confirm that regression activity specifically requires the ATPase and helicase activities of WRN, but also indicate that the exonuclease activity of WRN on the leading daughter strand of replication fork substrates is positively correlated to its ability to perform regression. Finally, we tested how the 3′ to 5′ exonuclease activity of WRN processed the leading daughter strands of fork substrates with different leading arm gaps in relation to the specific formation of daughter duplexes. For these experiments, fork substrates with leading arm gaps of 2–11 nt were incubated with WRN-wt and total DNA products were analyzed in parallel by native and denaturing PAGE. In addition, daughter duplex products from native PAGE were excised, extracted, and then analyzed by denaturing PAGE alongside the total DNA products from the regression reaction. As before, native PAGE (A) showed that WRN-wt generated daughter and parental duplexes along with some short single-stranded products from these substrates. Significantly, the migration of daughter duplexes generated from fork substrates with gap sizes of 2–8 nt was slightly faster than their respective (undigested) daughter duplex markers (A, , respectively). Analysis of total DNA products by denaturing PAGE shows that WRN-wt degrades the labeled leading daughter strand of each substrate, although the number of nucleotides removed decreases as the gap size increases (B, ). Conspicuously, for each fork substrate, there is substantial degradation until the leading daughter strand reaches 19–20 nt, but digestion beyond that point decreases precipitously. In this analysis of total products, the leading daughter strand could be associated with daughter duplex, displaced leading daughter strand and/or four-stranded fork, but not to any significant extent with other DNA structures. Parallel analysis of the extracted daughter duplexes ( in A and B) shows that the leading daughter strand component of these duplexes is strictly degraded in a range extending down to 19 nt (B, ). Thus, the lengths of leading daughter strand specifically associated with the daughter duplex product reflects a defined subset of the WRN exonuclease activity occurring during the entire reaction. Keeping in mind that WRN exonuclease does not act on the daughter duplex once formed (), this result indicates that the exonuclease activity of WRN-wt digests the leading daughter strand of each fork substrate, decreasing its length and thereby increasing the leading arm gap size to as much as 13 nt, prior to or concomitant with unwinding and pairing of daughter strands to yield the (faster-migrating) daughter duplexes. To scrutinize this further, the relative amounts of radioactivity associated with each band present in lanes corresponding to the extracted daughter duplexes produced from each fork substrate were determined and directly compared (C). This analysis shows that, for each substrate, a minor portion of the daughter duplexes contained undegraded leading daughter strands—i.e. 15.1, 9.5, 19.1 and 25% (C, ) for the daughter duplexes from the fork substrates with 2, 5, 8 and 11 nt gaps, respectively. Thus, digestion of the leading daughter strand (prior to or concomitant with daughter duplex formation) occurs during the vast majority of fork regression events catalyzed by WRN-wt. The most favored length of the leading daughter strand within the daughter duplex product was, by far, 20 nt followed closely by 19 nt, regardless of the leading strand gap size in the original fork substrate. Although the length of the leading daughter strand in these daughter duplexes is certainly not uniform, there is a definite preference for a leading daughter strand length of 19 or 20 nt corresponding to a gap size of 13 or 12 nt on the original fork substrate. When considered in combination with our comparison of WRN-E84A and WRN-wt on fork substrates with different leading arm gaps (), these findings convincingly demonstrate that the exonuclease activity inherent in wild type WRN enhances the regression of forks with small (2–8 nt) leading strand gaps, apparently by generating a better structure for the regression process. Furthermore, our data thus far indicate that the favored structure for regression by the coordinated helicase and annealing activities of WRN is a fork with a leading strand gap of at least 11–13 nt. The experiments above demonstrate that WRN (and BLM) can readily produce a daughter duplex from model replication forks with homologous parental–daughter arms of limited length. However, the parental-daughter arms are essentially continuous, extending back to the origin of replication. Thus, during the process of fork regression, the daughter strands would remain associated with the parental strands resulting in a Holliday junction or ‘chicken foot’ structure. Such a structure is believed to be the key intermediate in pathways that cope with blocked forks and restart replication (). To determine whether WRN could generate Holliday junction intermediates, a longer, more complex replication fork substrate was designed. Precisely at the fork junction, this substrate (A, ) contained 5 nt of non-complementarity on each arm (again to prevent spontaneous branch migration) as well as the optimal size gap (12 nt) on the leading arm for regression by WRN that obviated the need for its exonuclease activity. Most importantly, both parental–daughter arms of this substrate were significantly longer and contained homologous sequences proximal to the fork junction () but non-homologous sequences (30 bp on each arm) at the distal ends (). Theoretically, this design would allow initiation of fork regression with branch migration and daughter strand pairing up to the point of heterology, at which further regression might be inhibited. Nevertheless, if regression was initiated, the fork should be converted, at least transiently, to a Holliday junction with limited mobility (A, ). The high efficiency of WRN-mediated regression of fork substrates with similar-sized leading arm gaps ( and ) made us hopeful that we could detect transient formation of such Holliday junction structures by using known resolvases. Thus, consensus sequences (5′-↓CC-3′) for cleavage by RusA, a Holliday junction-specific resolvase, were included on each strand within the homologous but not the heterologous regions of the parental–daughter arms. Notably, 5′-CC-3′ sequences were also present in the parental duplex region, serving as an internal control for the Holliday junction specificity of RusA. Unique Rsa I and Xmn I sites were included precisely at the junctions between homologous and heterologous regions on the leading and lagging arms, respectively. For this substrate, radioactive labels were placed at the 5′ ends of the lagging daughter strand (82lag) and the leading parental strand (122lead). Notably for this series of experiments, the specific activity of the former was approximately 18 times that of the latter, influencing the relative intensities of individual DNA products. Moreover, this particular placement of radiolabels determines the nature of DNA products that can be observed following treatment of both intact fork substrate with WRN-E84A plus RusA (A, ) and restricted fork with only WRN-E84A (A, ). We initially wanted to determine whether this longer fork substrate could be acted on by WRN to yield a daughter duplex product in a manner identical to the substrate above. To this end, this substrate was restricted with Rsa I and Xmn I (A, ) to release the heterologous (and unlabeled) regions of both arms; the resulting fork with entirely homologous leading and lagging arms was purified. When this substrate was treated with WRN-E84A, two new DNA products (B, ) appeared with migration consistent with that of the potential daughter duplex (a 42 bp plus 10 nt 5′ overhang) and the parental duplex (92 bp); note that the daughter duplex appears more intense due to the higher specific activity of the lagging daughter strand. As above, production of both species required ATP hydrolysis and occurred at extremely low WRN concentrations without detectable amounts of single-stranded products (data not shown). These experiments confirm that this restricted fork could be regressed by WRN-E84A in the same manner as the shorter fork substrate (21lead fork) that contained a similar structure at the fork junction. These findings also suggest that the length of the homologous arms in our substrates does not significantly affect WRN's regression function, as the homologous arms of the restricted fork are approximately twice as long as those in our short fork substrates. Next, the unrestricted long fork substrate was treated with WRN-E84A in the presence or absence of RusA. Over the same range of concentration used above, WRN-E84A alone did not generate a product resulting from pairing of the daughter strands (C, ), in contrast to our results on forks with completely homologous arms. This suggests that the heterology on the distal ends of each arm inhibits WRN-mediated generation of ‘daughter duplex’ in a manner that cannot be overcome by unwinding activity at these limiting concentrations of WRN-E84A. Thus, if fork regression intermediates are formed by WRN-E84A from this substrate, they revert (by enzyme-mediated or spontaneous reversal of branch migration) back to the fork structure during the reaction and/or electrophoresis. Small amounts of parental–daughter duplexes were detected, consistent with weak forward unwinding of this fork. In order to detect whether Holliday junction intermediates are transiently generated, RusA was added one min following the start of a 37°C incubation of substrate with WRN-E84A. Addition of RusA to WRN-containing reactions yielded new DNA products () when analyzed by native PAGE (C, ). The more prominent product migrated a little slower than a 77 bp marker, while the other migrated just slightly above a 122/82 partial duplex marker, consistent with RusA resolution of a Holliday junction formed from this fork substrate—i.e. a 72 bp duplex with a 10 nt 5′ overhang (more prominent because it contains the labeled strand with higher specific activity) and a 122 bp duplex, each putatively containing a nick (see A, ). In an important control, RusA without WRN-E84A does not detectably act on the fork substrate (C, ). These results are consistent with the previously characterized specificity of RusA for Holliday junctions and also indicate that WRN-E84A is generating the Holliday junction structure for RusA cleavage. Importantly, RusA cleavage required not only WRN-E84A but also ATP hydrolysis, as cleavage was not detected in the absence of ATP or in the presence of the poorly hydrolyzable analog ATPγS (D), demonstrating that WRN ATPase and helicase activities were necessary to convert the fork substrate to a Holliday junction intermediate. Similar reactions were analyzed to determine whether these new DNA products indeed corresponded with RusA-specific cleavage events on Holliday junctions generated by WRN-dependent fork regression. If a Holliday junction were generated by fork regression, positioning of RusA consensus (5′-↓CC-3′) sequences in the labeled lagging daughter (82lag) and leading parental (122lead) strands within the homologous regions on the arms of this substrate would cause RusA to cut 30 and 70 nt from the labeled 5′ ends of the respective strands (see A, ). When reactions containing WRN-E84A and/or RusA were analyzed by denaturing PAGE, neither WRN-E84A alone nor RusA alone detectably produced any change to the labeled (122lead and 82lag) strands of the fork substrate (E, ). In contrast, reactions that contained WRN-E84A followed by RusA showed new bands that co-migrated with 30 and 70 nt markers (E, ). To further clarify the nature of the events catalyzed by WRN-E84A plus RusA, the new DNA products detected by native PAGE (E, ) were individually excised, extracted, and subjected to denaturing PAGE. This analysis indicated that the lower product (a) contained both cleaved and uncleaved 82-mer (E, ). The upper product (b) contained both cleaved and uncleaved 122-mer with also some 82-mer (E, lane 11), the latter apparently derived from lagging parental–daughter duplex that nearly co-migrates by native PAGE with this RusA cleavage product. Regarding this outcome, it is important to reiterate that CC sequences were also present on the unlabeled (122lag and 72lead) strands within the homologous regions. Thus, we conclude that after WRN-E84A mediates fork regression, RusA can cleave either both labeled or both unlabeled strands of the resulting Holliday junction (see A, , respectively). While each set of cleavages generates different locations of nicks, analysis by native PAGE shows only two new products, a 72 bp duplex with 10 nt 3′ overhang plus a 122 bp duplex, regardless of whether the nicks are present in the labeled or unlabeled strands. When these products are extracted and analyzed by denaturing PAGE, concerted RusA cleavage of the labeled strands yields shorter single-stranded products (30- and 70-mers) while concerted cleavage of the unlabeled (undetectable) strands leaves the labeled strands intact (82- and 122-mers). These results clearly demonstrate that WRN-E84A regresses this fork substrate to generate a Holliday junction or ‘chicken foot’ structure that is cleaved by RusA at its specific recognition sequences. Replication fork blockage is such a common event that cells have evolved specialized pathways to handle these situations. Regression of a blocked replication fork is theorized to be the initial step in dealing with these serious challenges to completion of DNA replication, genome stability and cell survival (). Fork regression would involve pairing of nascent daughter strands and re-annealing of parental strands to form a Holliday junction or ‘chicken foot’ intermediate. Following fork regression, obstacles to replication might be addressed by alternative pathways including repair of a blocking lesion and reverse branch migration to regenerate a forked structure, a strand-switching replication step in which the lagging daughter strand serves as template followed again by reverse branch migration, bypassing the blocking lesion and re-establishing the fork or resolution of the Holliday junction to generate a double-strand break that could initiate recombinational pathways to restore a functional replication fork. A preliminary report from our lab established that the human RecQ helicases WRN and BLM have the ability to regress a specific replication fork substrate (). In this study, we have examined the ability of WRN to act on several model replication fork substrates, including a series of short fork substrates containing structural differences at the fork junction and another longer substrate that allowed Holliday junction formation and detection during a potential regression reaction. Importantly, fork regression by WRN on our long fork substrate directly forms a Holliday junction that is detected using the RusA resolvase (). Fork regression requires the ATPase and helicase activities of WRN, as it does not occur in reactions lacking ATP or including the poorly hydrolyzable analog ATP-γ-S or using the WRN-K577M mutant that lacks both ATPase and unwinding activity. Most notably, our experiments with shorter fork substrates indicate a specific role for the 3′ to 5′ exonuclease activity of WRN during fork regression—i.e. controlled digestion of the leading daughter strand to generate a more favorable structure for regression and thereby increase the efficiency of this process. These findings suggest a novel role for the exonuclease activity of WRN during fork regression that operates in coordination with its unwinding and annealing activities. Thus, all of the DNA-dependent activities of WRN may cooperate to promote replication fork regression. Enzymatic reactions containing WRN-E84A or WRN-wt showed a concentration- and time-dependent conversion of our short fork substrates to both parental and daughter duplex products (see and A). Daughter duplex formation is specific for a regression event as it requires both unwinding and pairing of the physically unlinked daughter strands (A). Although generation of free leading daughter strand product is detectable in reactions containing relatively high WRN concentrations, several lines of evidence demonstrate that daughter duplex formation occurs through an intimate linkage between unwinding of parental–daughter arms and pairing of daughter strands. First, WRN does not anneal the free daughter oligomers (due to their short length) that comprise daughter duplexes (Supplemental ), indicating that annealing does not independently follow unwinding of the parental-daughter arms. Second, at limiting concentrations of WRN-E84A (), there is almost exclusive production of daughter and parental duplexes without significant generation of three-stranded forks or free leading daughter strands. Furthermore, free leading daughter strands are not significantly produced at any time during our kinetic experiments (D–F and A), strongly suggesting that daughter duplexes are formed from the fork substrates without release of daughter strands. Most convincingly, a quantitative analysis of WRN-E84A-mediated regression of 21lead fork substrate over time demonstrates that the parental and daughter duplexes are produced directly and essentially simultaneously from intact four-stranded fork (F). Lastly, pairing of daughter strands during regression is a process mediated by WRN (or BLM) but not all unwinding enzymes, as other helicases such as UvrD and Rep unwind both daughter strands of our short fork substrates but do not produce daughter duplexes. These results indicate that daughter duplex formation does not simply occur spontaneously during daughter strand unwinding. Taken together, our results suggest that, during regression, the partially unwound daughter strands are juxtaposed by WRN in a manner that promotes their pairing to form a Holliday junction that, for our short fork substrates, is subsequently converted to daughter and parental duplexes. This is further supported by the WRN-dependent conversion of our long fork substrate to Holliday junctions () in which the daughter strands are paired while still associated with the parental strands. It seems very likely that WRN-mediated fork regression occurs through coordination between its unwinding function and its previously reported strand annealing activity (). The results with our short fork substrates indicated that the leading arm structure had a major influence on the efficiency of fork regression. Specifically, daughter duplex formation by exonuclease-deficient WRN-E84A was greatly enhanced when the single-stranded gap on the leading arm increased to 11 nt from ≤8 nt ( and ). Although WRN-wt was similarly effective as WRN-E84A in daughter duplex formation from substrate with an 11 nt gap, it was much more efficient on fork substrates with shorter leading strand gaps (), suggesting that WRN's 3′ to 5′ exonuclease activity might be digesting the leading daughter strand to increase the gap size. In addition, daughter duplexes formed by WRN-wt migrated slightly faster by native PAGE than those formed by WRN-E84A, in agreement with putative 3′ to 5′ processing of the leading daughter strand. Analysis by denaturing PAGE confirmed that WRN-wt was specifically degrading the leading daughter strand during these regression reactions (). However, the exonuclease activity of WRN-wt could not detectably digest daughter duplexes (or displaced leading daughter strands) once formed (), indicating that digestion of the leading daughter strand was occurring prior to or concomitant with formation of daughter duplex. Further analysis indicated that, during regression, exonucleolytic digestion of the leading daughter strand by WRN-wt was essentially limited to a defined range, regardless of the original length of this strand in the fork substrate. The observed preference for digestion to a length of 20 or 19 nt corresponds to generation of a 12 or 13 nt single-stranded gap on the leading arm, respectively (B and C). Taken together, these results indicate that WRN exonuclease activity promotes regression on substrates with gaps shorter than 11 nt by digesting the leading daughter strand to increase the leading arm gap size. The exonuclease-deficient WRN-E84A protein is unable to alter the gap size and thus acts efficiently only on our short fork substrate with an 11 nt gap. By this reasoning, the optimum structure for regression by WRN, without assistance from its exonuclease function, is a fork with a leading arm gap of at least 11–13 nt. This conclusion is further supported by the highly efficient regression and Holliday junction formation by WRN-E84A on the longer fork substrate containing a leading arm gap of 12 nt (B and C). These experiments suggest WRN exonuclease activity may participate in processing the leading daughter strand during regression of blocked replication forks, but its degree of involvement may depend on the precise structure at the fork junction and perhaps the spatial relationship between leading and lagging daughter strands. The efficiency of WRN-mediated regression on forks with ≥11 nt leading arm gaps suggests that WRN may preferentially act on forks in which leading strand synthesis is blocked while lagging strand synthesis continues, leaving a single-stranded gap on the leading arm. During fork regression, WRN exonuclease activity may further digest the leading daughter strand, a processing step that may simply generate the optimum structure for enzyme-mediated pairing of the daughter strands. However, there may be other advantages to this arrangement. On this series of substrates, WRN-mediated regression only occurs efficiently when the leading daughter is at least 10–12 nt shorter than the lagging daughter strand. If this structural relationship was maintained (or perhaps further exaggerated) , fork regression would always yield a Holliday junction containing a single free end with a 5′ overhang. Since formation of Rad51-mediated filaments occurs on 3′ overhangs (), this structure might inhibit recombination, perhaps in favor of alternate, less error-prone pathways such as repair of the blocking lesion and reverse branch migration to re-establish a viable replication fork or strand switching synthesis and reverse branch migration with concomitant bypass of the obstacle (on the parental leading strand) that originally impeded fork progression. Recent studies in suggest that lesion repair might be the first alternative attempted following fork blockage caused by DNA damage (). It is tempting to speculate that enzymatic regression processes have evolved to at least initially favor less error-prone pathways such as repair or strand switching. Cells from WS patients have a higher rate of spontaneous RAD51 foci formation than normal cells, supporting the idea that recombination might be utilized more often when WRN is non-functional (). Importantly, the notion that WRN regresses replication forks to specifically generate intermediates that suppress instead of promote recombination would be consistent with the hyperrecombination phenotypes of cells that have lost WRN function. Fork structures with heterologous arms have previously been shown to be excellent substrates for WRN helicase (). Here, we demonstrate that WRN specifically regresses forks with homologous arms. With regard to its action on fork structures, WRN binds in the vicinity of the fork junction as judged by DNase I footprinting (Machwe , unpublished results). It is quite relevant that WRN-wt and WRN-E84A regression activities are highly efficient when the structure of the fork substrate is favorable. As judged by daughter duplex formation, regression of the fork substrate with an 11 nt gap on the leading arm is readily detectable after 5 min using sub- and near-equimolar levels of enzyme compared to substrate (B, ). Furthermore, at low concentrations, WRN is more effective at catalyzing regression of this fork substrate than unwinding a 27 bp partial duplex (). WRN also efficiently regresses our longer fork substrate containing a 12 nt gap at near-equimolar concentrations (B and C). Although these experiments cannot determine the stoichiometry of WRN with respect to the DNA substrate, the requirement for unwinding of both parental-daughter arms during regression might imply at least a dimeric structure. Irregardless, the efficiency by which WRN catalyzes this multi-faceted regression reaction suggests that it is particularly suited to this task. It is noteworthy that BLM, another human RecQ helicase that possesses unwinding and strand pairing activities and DNA substrate specificity similar to WRN, also performs fork regression (,). Like WRN, BLM produces Holliday junctions from our long unrestricted fork substrate that are recognized and cleaved by RusA (data not shown). However, BLM appears to be consistently less efficient in fork regression of each of our short fork substrates than WRN-wt (A and B). Although RecG can also regress model forks as previously reported (), other helicases (including UvrD, Rep and PriA implicated in resolution of fork blockage in ) tested thus far could not perform this function, suggesting that fork regression capability appears relatively limited to a small group of helicases that includes WRN and BLM. Thus, some RecQ helicases are structurally designed to catalyze fork regression by combining their helicase and strand pairing activities. Notably, two other human RecQ family members, RecQ1 and RecQ5β, have also been shown to possess both DNA unwinding and annealing activities (,). It may be relevant in a physiological context that BLM and other human RecQ helicases do not possess exonuclease activity and alone cannot modify the structure at the fork junction in the way that WRN-wt can. We speculate that WRN is preferentially involved in regression of blocked forks, while BLM may be more likely to participate (in combination with topoisomerase IIIα and BLAP75) in other DNA transactions, such as double Holliday junction resolution to prevent crossing over during homologous recombination (). RecQ helicases have previously been postulated to play roles in resolution of replication fork blockage (,,). The finding that both WRN and BLM can readily regress model replication forks greatly strengthens this hypothesis. However, formation of Holliday junctions resulting from replication fork regression is somewhat speculative and further proof is needed of specific genome maintenance pathways utilizing fork regression and under what circumstances they are implemented. It is also noteworthy that WRN has been hypothesized to participate in several other pathways that preserve genome stability (such as telomere maintenance). Despite these caveats, potential involvement of WRN in replication fork regression is consistent with previous findings regarding WRN and certain properties of WRN-deficient cells. WRN-deficient cells are slower than their normal counterparts in completing S phase and show asymmetric replication fork progression, consistent with an inability to properly resolve fork blockage (,). They are also hypersensitive to agents that severely inhibit DNA replication including hydroxyurea, which depletes deoxynucleotide pools, topoisomerase inhibitors that induce strand breaks and DNA-protein crosslinks and interstrand crosslinking agents (such as mitomycin C and cisplatin) that prevent unwinding of the parental strands (,,,). Furthermore, in normal cells, WRN migrates rapidly to sites of DNA synthesis following treatment with these and other DNA damaging agents (,), findings that suggest that WRN is recruited to sites where replication is blocked. Our results suggest the reason for this relocalization—i.e., WRN is brought to blocked replication forks to catalyze their regression as part of a pathway that maintains genome stability. It is possible that, although they lack exonuclease activity, other human RecQ helicases such as BLM can partially compensate for loss of WRN function in fork regression. However, if such redundancy exists, it is likely imperfect and there may be situations in which WRN-deficient cells are still compromised in dealing with blocked replication forks. As a result, cells lacking WRN are hyperrecombinant and accumulate chromosomal abnormalities that are almost assuredly responsible for the increased cancer incidence of WS patients. Alternatively, in response to these DNA metabolic problems some cell types may trigger apoptosis or permanent cell cycle exit (senescence). With time, the cumulative loss of either cells by apoptosis or reduction in proliferative capacity within a tissue may cause the premature aging phenotypes of WS. Although more research is needed to confirm these hypotheses, a specific function in fork regression is highly consistent with existing knowledge regarding WRN and WS. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Genetic variation occurs on multiple levels, from single nucleotide polymorphisms (SNPs) to larger events involving contiguous blocks of DNA sequence that vary in copy number between individuals. The structural diversity in the human genome is much higher than previously assumed, and attracts an increased interest within the genetics community. It is now becoming increasingly clear that submicroscopic variations are major contributors to genetic diversity and human disease (,). The interest in copy-number variation (CNV) has led to the establishment of a number of analytical methods, using either global or targeted approaches. Microarray-based comparative genome hybridization (array-CGH) is a commonly used global approach to CNV detection (,), enabling genome-wide scans for detection of novel CNVs. CGH arrays are manufactured with different resolution and coverage, using different approaches to probe genomic samples, ranging from BAC clones to short oligonucleotides attached to the array surface (). High-throughput SNP analysis can also be employed for CNV-detection, as revealed by long stretches of apparently homozygous loci or unusual heterozygous signal ratios (,). Although global array-based approaches can provide high resolution data on CNVs in individuals, there remains a need for simple, cost-efficient, accurate methods to validate and test candidate CNVs across larger populations. One established targeted approach for CNV analysis is quantitative PCR (qPCR) (). However, this technique requires setting up a large number of replicate reactions to score individual deletions and duplications, and is generally not suitable for multiplexing. Similarly, fluorescence hybridization (FISH) is a labor-intensive technique which is not usually highly multiplexed, though it is well-established in diagnostics laboratories. Examples of multiplexed targeted copy-number analysis approaches are Quantitative multiplex PCR of short fluorescent fragments (QMPSF) (), multiplex amplifiable probe hybridization (MAPH) () and multiplex ligation-dependent probe amplification (MLPA) methods (). In MLPA, which has become perhaps the most commonly used one, up to 40 loci can be analyzed in parallel. Here, we present an approach based on the selector technique (), called multiplex ligation dependent genome amplification (MLGA). In contrast to MLPA, genomic DNA is amplified rather than probe molecules, and a single probe is required for each target instead of two. This leads to increasing reaction kinetics and decreasing probe amplification background. Furthermore, these shorter probes are easily manufactured by conventional oligonucleotide synthesis. These properties allow for cost-efficient design of custom MLGA assays with a short turnover time. This is demonstrated in an accompanying paper, where a candidate duplication was verified, sized, and diagnosed in a very cost-efficient approach (Salmon Hillbertz,N.H.C. ., Nat. Genet. in press). A set of 14 human target genes were chosen on five different chromosomes (). Sequences for each target were collected from the Ensembl database (, assembly NCBI 36, Oct 2005). These sequences were processed in the PieceMaker software () to generate a set of restriction fragments using a restriction enzyme of choice. A single fragment for each target sequence was then chosen in such a way that the fragments in each pool were between 100 and 400 nt in length with each fragment having a different length, with a minimum size difference of 6 nt. Selector probes serving as templates for the circularization of each chosen target fragment () were designed using the ProbeMaker software (). Each selector consists of two synthetic oligonucleotides; a target-specific selector probe (70–74 nt), and a universal vector oligonucleotide (34 nt). Oligonucleotides were synthesized by DNA Technology A/S, Denmark (). The central part of each selector probe is complementary to the vector oligonucleotide so that hybridization between the two generates the recognition sequence for the III restriction enzyme and a universal primer pair site for parallel PCR amplification. The ends of the selector probes (18–20 nt each) have sequences complementary to the ends of the restriction fragments targeted for selection. Six genomic DNA samples were extracted from blood (Flexigene, Qiagen), collected with the appropriate permissions from individuals diagnosed with Down syndrome, and admitted to the Department of Clinical Genetics, Uppsala University. DNA samples were also extracted from the aneuploid cell cultures NA04626, NA01416 and NA06061 (Coriell Cell Repositories) with 3, 4 and 5 X-chromosomes. Pooled samples of male and female DNA from Promega (cat

The formation of spatial patterns of cells is a recurring theme in developmental biology. The mechanisms underlying these pattern formation processes often include directed, morphogenetic cell movements. An example is provided by cells of the bacterium , which display two types of morphogenetic cell movements depending on their nutritional status, and which result in the formation of two distinct structures, colonies and fruiting bodies (). In the presence of nutrients, colonies form, and the cells at the edge spread coordinately outward. In the absence of nutrients, the spreading behavior is constrained and cells aggregate to construct multicellular fruiting bodies. The cells in the fruiting bodies differentiate into spores. The formation of these two spatial cell patterns depends on the ability of the cells to actively move and to regulate their motility behavior. The rod-shaped cells move by gliding in the direction of their long axis, and active movements are restricted to solid surfaces. Two distinct motility systems are involved: the social (S)- and the adventurous (A)-motility systems (). Generally, mutations only affect one of the two systems; however, mutations in , which encodes a member of the Ras superfamily of small GTPases (), abolish the activity of both systems (). The S-motility system depends on type IV pili (Tfp) and is the equivalent of twitching motility in and species (). In , this Tfp motility is dependent on cell–cell contact and is active when cells are within contact distance of each other (). Tfp localize to the leading cell pole (), and a motive force is generated by retraction of Tfp (; ; ). The A-motility system provides cells with the ability to move as single cells. It is currently unknown how motive force is generated in the A-motility system. However, two models have been suggested for force generation in this motility system. In one model, the A-motility system functions similarly to junctional pore complexes in gliding cyanobacteria, which generate force by polyelectrolyte secretion (; ). In , structures equivalent to the junctional pore complexes, called A-motility nozzles, are present at both poles, and force generation has been suggested to involve polyelectrolyte secretion from nozzles at the lagging pole (). An A-motility model involving polymer export is supported by the requirement of a large number of genes for A-motility that encode proteins involved in polymer synthesis and export (; ). In an alternative model, force generation involves multiple adhesion complexes assembled at the leading cell pole and distributed along the cell body (). These complexes are defined by the AglZ protein and are thought to function in a manner similar to that of focal adhesion complexes in eukaryotic cells (). As cells move over a surface, they periodically stop and then resume gliding in the opposite direction, with the old lagging pole becoming the new leading pole (). Regulation of the cell reversal frequency is critical for establishing both types of morphogenetic cell movements (). The reversal frequency is regulated by the Frz two-component system (). Analysis of single-cell behavior as well as colony expansion rates suggests that the A- and S-motility systems generate motive force in the same direction (; ) and that the directionality of the two engines changes synchronously during reversals (; ). The mechanism underlying a direction change in the S-motility system involves an Frz-dependent switch of the pole, at which Tfp assemble and with the FrzS protein relocating from the old leading to the new leading cell pole (). In the A-motility system, the change in directionality involves the Frz-dependent relocation of the AglZ protein from the old leading cell pole to the new leading cell pole (). To further the understanding of the A-motility system and the mechanisms underlying polarity switching, we focused on the equired fr otility esponse regulator (RomR). We identify RomR as essential for A-motility. RomR localizes in a bipolar, asymmetric pattern with a large cluster at the lagging pole. In synchrony with cell reversals, the large RomR cluster relocates in an Frz-dependent manner to the new lagging pole. Our data suggest that the large RomR cluster stimulates the A-motility machinery at the lagging pole and that dynamic RomR localization is essential for polarity switching of the A-motility system. Moreover, we show directly that RomR and FrzS oscillate between the cell poles independently but synchronously, thus ensuring the synchronous polarity switching of the two motility systems. While analyzing the ORF , which is required for fruiting body formation in (Freymark , unpublished), we also analyzed the downstream ORF, . The deduced MXAN4461 protein encodes an uncharacterized response regulator, RomR (). RomR consists of an N-terminal receiver domain of two-component systems (residues 1–116) and a C-terminal output domain (residues 117–420) that can be subdivided into a Pro-rich region (residues 117–368) and a Glu-rich tail (residues 369–420) (). Database searches revealed that several bacteria belong to the δ-proteobacteria, which like , encode a protein with a domain structure similar to that of RomR (). The function of these response regulators is unknown. To understand the function of RomR in , we created an insertion mutation in which an gene was inserted at codon 30 in (::) in the fully motile strain DK1622, which serves as the wild-type strain in this work. The resulting strain, SA1128, was analyzed further. Strain SA1128 (::) was indistinguishable from the wild type with respect to growth in rich medium and in chemically defined A1 minimal medium (data not shown). We tested whether SA1128 was deficient in motility by examining colony spreading on 1.5% agar, which favors motility by means of the A-motility system (). Under these conditions, the wild-type strain formed large colonies with rafts of cells and single cells at the edge. Strain SA1128 formed small colonies, and rafts of cells but no single cells were observed at the edge (). When analyzed on 0.5% agar, which favors motility by means of the S-motility system, SA1128 formed colonies that were only slightly smaller than those of wild-type cells (data not shown). These observations suggested that the mutation caused an A-motility defect. We investigated this hypothesis by introducing the mutation into strains containing mutations that inactivated either the A-motility system (AS) or the S-motility system (AS). The mutation did not interfere with S-motility in the AS strain, but it abolished A-motility in the AS strain (). To conclusively determine whether is required for A-motility, we analyzed movement of single cells on 1.5% agar by time-lapse microscopy. In these recordings, SA1128 did not display any single-cell movement, whereas wild-type cells displayed normal single-cell movements (data not shown). Thus, the mutation results in an A-motility defect. To determine whether the mutation caused a defect in fruiting body formation, cells were exposed to starvation. Wild-type cells had completed fruiting body formation at 72 h, whereas SA1128 cells formed many small, abnormally shaped fruiting bodies (). Moreover, the sporulation frequency of SA1128 was only 3% that of wild-type cells. We then integrated a allele including the native promoter by site-specific recombination at the chromosomal Mx8 attachment site in SA1128, yielding strain SA2272. In this complementation experiment, the allele corrected the A-motility defect as well as the developmental defects caused by the mutation (). In contrast, the integrated vector pSWU30 alone (strain SA2210) did not correct these defects ( and not shown). Immunoblot analysis using polyclonal rabbit antibodies against full-length RomR confirmed that RomR accumulated at similar levels in SA2272 and wild type (). Taken together, these observations show that RomR is required for A-motility. We speculate that the defect in fruiting body formation results from the defect in the A-motility system. Given that RomR plays a role in A-motility, we hypothesized that RomR might function in a spatially confined manner. To test this idea, we used the green fluorescent protein (GFP) as a fluorescent marker in localization studies. A allele including the native promoter was integrated at the chromosomal Mx8 attachment site in the wild type and in strain SA1128 (::), giving rise to strains SA2273 and SA2271, respectively. RomR-GFP corrected the A-motility defect caused by the mutation in strain SA2271 (::, ) () and did not interfere with the activity of wild-type RomR in strain SA2273 (, ) (data not shown). Moreover, single cells of SA2271 cells moved with the same speed (3.4±0.2 μm/min) as single cells of wild type (3.3±0.9 μm/min). These results provided evidence that the fusion protein is fully functional. Immunoblot analysis using antibodies against RomR and GFP confirmed that RomR-GFP (calculated molecular mass 71.3 kDa) accumulated at a level similar to that of RomR in wild-type cells and that degradation was negligible (). Fluorescence microscopy showed that RomR-GFP localized to the cell poles in both SA2271 (::, ) and SA2273 (, ) (). In both strains, 90% of the cells (=200) had an asymmetric RomR-GFP distribution, with a large and a small polar cluster; the remaining 10% displayed a bipolar, symmetric localization pattern. To verify the asymmetric distribution, we determined the localization of native RomR in the wild type by immunofluorescence microscopy using affinity-purified RomR antibodies. A bipolar, asymmetric RomR localization pattern was observed; as expected, RomR was not detected in mutant SA1128 (::) (). To clarify whether the large RomR cluster localized to the leading or lagging cell pole, two different approaches were used. First, RomR-GFP localization was determined relative to Tfp, which are localized to the leading cell pole. In 88% of the cells (=32) of SA2271 (, ) stained with the fluorescent dye Cy3, the large RomR-GFP cluster was localized to the pole opposite to that containing Tfp, that is, the lagging cell pole (). Second, in time-lapse fluorescence microscopy of SA2271 (, ) and SA2273 (, ) cells moving on a thin agar pad, the bright RomR-GFP cluster was always detected at the lagging cell pole (; ). In these time-lapse recordings, only cells separated from other cells by at least one cell length were scored to ensure that cells moved only by means of the A-motility system. We observed the same RomR-GFP localization pattern in a Δ, mutant (SA2289), which lacks Tfp-dependent motility owing to an in-frame deletion of the gene, which encodes the Tfp subunit (). Thus, in cells moving by means of the A-motility system, the large RomR cluster is at the lagging pole. To resolve whether RomR localization changes during a reversal, RomR-GFP location was analyzed during reversals in single cells of SA2271 (, ). Thirty-two reversals were observed, and all reversals were accompanied by a switch in localization of the large RomR-GFP cluster from the old lagging pole to the new lagging pole (; ). Quantification of the fluorescence intensity of the RomR-GFP clusters during reversals revealed the following order of events (data for a representative cell are shown in ). Initially, the fluorescence intensity of the cluster at the lagging pole was greater than that of the cluster at the leading pole, and the cell moved in one direction. The fluorescence intensity of the two polar clusters then became similar, as the intensity of the cluster at the lagging pole decreased and the intensity of the cluster at the leading pole increased. At the same time, the cell stopped moving. As the intensity of the cluster at the old lagging pole continued to decrease and the intensity of the cluster at the old leading pole continued to increase, the cell began to move in the opposite direction. Similar observations were made in single cells of the mutant SA2289 (Δ, ::, ), which harbors only an active A-motility system (data not shown; ). To determine the mechanism underlying dynamic RomR localization, we followed RomR-GFP in single cells of strain SA2271 (::, ). Cells treated with 25 μg/ml chloramphenicol to inhibit protein synthesis displayed a motility pattern and a RomR-GFP localization pattern similar to that of untreated cells (data not shown). This suggests that the mechanism underlying dynamic RomR localization involves the transfer of RomR between the poles and not proteolysis at the old lagging pole accompanied by localization of -synthesized protein at the new lagging pole. To test whether the two domains in RomR have specific functions in RomR localization, genes encoding full-length RomR, the receiver domain plus 24 additional amino acids (residues 1–140 of RomR), and the output domain (residues 117–420 of RomR) were each expressed separately from the promoter in the mutant. The genes encoding these three proteins fused to GFP (RomR-GFP, receiver-GFP, and output-GFP) were also expressed from the promoter in the mutant. In immunoblots, neither the receiver (calculated molecular mass 15.1 kDa) nor the receiver-GFP (calculated molecular mass 42.2 kDa) was detected by anti-RomR antibodies. As the receiver-GFP protein was detected by anti-GFP antibodies (), the receiver-GFP protein was stably synthesized and the anti-RomR antibodies did not recognize the receiver domain. In the strain encoding the output domain, a larger protein (39 kDa) than expected (29.3 kDa) was detected by the anti-RomR antibodies. We attribute this difference to an abnormal mobility of the output domain in the SDS–polyacrylamide gel owing to the unusual sequence of the domain. The output-GFP protein with the expected size (calculated molecular mass 56.4 kDa) was detected by both the anti-RomR and anti-GFP antibodies. Both output domain proteins accumulated at slightly lower levels than full-length RomR (). Full-length RomR (SA2059) as well as RomR-GFP (SA2058) restored the ability of mutant cells to move as single cells and, thus, corrected the A-motility defect in the mutant (; ). However, neither the receiver domain (SA2244) nor the receiver-GFP protein (SA2259) corrected the A-motility defect in the mutant (; ). Moreover, the receiver-GFP protein was homogeneously distributed throughout the cells and failed to segregate to the poles (; ). The output domain (SA2256) as well as the output-GFP protein (SA2260) restored the ability of mutant cells to move as single cells (; ). The output-GFP protein localized in a bipolar, asymmetric pattern, with 88% of the cells having a large cluster at the lagging pole and 12% having a large cluster at the leading pole (). Cells only rarely reversed, and these rare reversals were not accompanied by a switch in the localization of the output-GFP protein (). These results suggest that the output domain is a pole-targeting determinant, that the receiver domain is involved in dynamic RomR localization, and that dynamic RomR localization is required for reversals. For many response regulators, it has been shown that phosphorylation of a conserved Asp residue in the receiver domain is required for activity (). To test genetically whether phosphorylation of RomR contributes to RomR function and localization, genes encoding two RomR mutant proteins in which the phosphorylatable Asp (D53) in the receiver domain had been substituted were expressed from the promoter in the mutant. In RomR, D53 was substituted with Asn, resulting in loss of the ability to be phosphorylated. In RomR, D53 was substituted with Glu; in several response regulators, this substitution partially mimics the phosphorylated state (). The genes encoding RomR and RomR fused to GFP were also expressed from the promoter in the mutant. Immunoblots with anti-RomR and anti-GFP antibodies confirmed that all four proteins accumulated at levels similar to that of RomR, when wild-type was expressed from the promoter (). All four mutant proteins restored the ability of mutant cells to move as single cells (; ). RomR-GFP (SA2062) localized in a bipolar, asymmetric pattern, with the large cluster at the lagging pole in all cells observed (; ). In contrast to cells synthesizing RomR-GFP, cells synthesizing RomR-GFP did not reverse direction, and RomR-GFP did not relocate between poles (). RomR-GFP (SA2060) also localized in a bipolar, asymmetric pattern, with the large cluster at the lagging pole in all cells observed (; ). But these cells displayed a 1.5-fold higher reversal frequency than cells synthesizing RomR-GFP (cf. SA2058 in , and all reversals were accompanied by relocation of the large RomR cluster from the old to the new lagging pole (). The opposite reversal phenotypes of RomR-GFP and RomR-GFP suggested that RomR partially mimics the phosphorylated state of RomR, that phosphorylation of D53 is crucial for reversals, and that dynamic RomR localization depends on phosphorylation. The Frz two-component system regulates the cell reversal frequency, and mutants only rarely reverse (). The correlation between reversals and pole-to-pole transfer of RomR-GFP suggested that the Frz system regulates dynamic RomR localization. To investigate this hypothesis, the gene encoding RomR-GFP was expressed in an mutant from the promoter (SA2070). In this strain, RomR-GFP localized in a bipolar, asymmetric pattern in all cells, and all cells harbored the large RomR-GFP cluster at the lagging pole (; ). Single cells of SA2070 did not reverse direction, and RomR-GFP did not relocate between poles (). These data show that the Frz system is dispensable for bipolar, asymmetric RomR localization but required for dynamic RomR localization. To test whether the Frz system promotes RomR relocation by inducing phosphorylation of RomR, we introduced plasmids encoding RomR-GFP and RomR-GFP into a :: mutant containing the mutation. RomR-GFP (SA2068) localized in a bipolar, asymmetric pattern, with the large RomR cluster at the lagging pole in all cells (; ). Moreover, as expected, these cells did not reverse direction, and RomR-GFP did not relocalize from pole to pole (). RomR-GFP (SA2054) localized in a bipolar, asymmetric pattern, with the large RomR cluster at the lagging pole in 71% of cells; in the remaining cells, RomR-GFP localized in a bipolar, symmetric pattern (; ). Importantly, cells containing RomR-GFP frequently reversed direction; in cells with an asymmetric pattern, all reversals were accompanied by relocation of RomR-GFP, whereas in cells with a symmetric pattern, no relocation was observed (). Thus, RomR-GFP, which likely mimics the phosphorylated state of RomR, bypasses the Frz system for reversals. These data suggested that RomR acts downstream of the Frz system to induce reversals in the A-motility system and that the Frz system induces RomR relocation by inducing RomR phosphorylation. The MglA protein is important for the activity of both motility systems in and has been implicated in the control of the reversal frequency (). To test whether MglA is required for correct RomR localization, we introduced the allele into DK3685, which contains the mutation and does not accumulate MglA (), giving rise to strain SA2042. Strikingly, RomR-GFP localized in a unipolar pattern in 90% of the cells (; ). We determined at which pole RomR-GFP localized by staining Tfp with Cy3. Surprisingly, 85% of the cells (=32) contained RomR-GFP and Tfp at the same pole (). Time-lapse microscopy of SA2042 cells was used to monitor the dynamic behavior of RomR-GFP. In these experiments, the cells did not move. This is in contrast to a previous report in which cells with a deletion of as well as the upstream gene, which codes a protein that stabilizes MglA, were reported to reverse at a high frequency (). Under all conditions tested, including those used by Spormann and Kaiser, and also using the Δ strain used by these authors, we were unable to observe movement of mutant cells. Notably, RomR-GFP did not undergo pole-to-pole transfer in the mutant. These observations suggested that MglA is required for establishing the correct polarity of the two motility systems and for RomR pole switching. The two motility systems in generate motive force in the same direction (; ) suggesting that the two systems switch polarity in synchrony during cell reversals. The FrzS protein, which is required for the full function of Tfp (), localizes in a bipolar, asymmetric pattern, with a large cluster at the leading pole and a small cluster at the lagging pole (). When cells reverse direction, the large FrzS cluster in parallel relocates in an Frz-dependent manner from the old to the new leading pole. To test whether FrzS is required for correct RomR localization or vice versa, we analyzed RomR-GFP localization in a Δ mutant and FrzS-GFP localization in a mutant. In a Δ mutant, RomR-GFP localized in a pattern similar to that in cells, and all reversals were accompanied by RomR-GFP relocation (SA2268) (; ). Likewise, FrzS-GFP localized in the same bipolar, asymmetric pattern in (SA2028) and (SA2041) cells (). Moreover, 13 out of 15 reversals (25 cells observed) in the strain and 9 out of 10 reversals (25 cells observed) in the mutant were accompanied by FrzS-GFP relocation from the old to the new leading pole (data not shown). Thus, RomR and FrzS localize to the poles and relocate independently. To determine whether relocation of RomR and FrzS occurred synchronously, a strain synthesizing FrzS-GFP and RomR fused to monomeric DsRed protein (RomR-mDsRed) (SA2036) was constructed. In a strain containing only RomR-mDsRed, localization was similar to that of RomR-GFP (data not shown). In all SA2036 cells observed (=50), the large FrzS-GFP and RomR-mDsRed clusters localized to opposite poles, with the large FrzS-GFP cluster at the leading and the large RomR-mDsRed cluster at the lagging pole (). In 10 reversals observed (25 cells observed), the large FrzS-GFP and RomR-mDsRed clusters switched poles within 30 s (data for a representative cell are shown in ). Thus, FrzS-GFP and RomR-mDsRed oscillate in synchrony. italic #text Construction of strains and plasmids, growth conditions, and motility and development assays are described in . A list of strains used is given in . For phase-contrast and fluorescence microscopy, steady-state cultures of cells were grown to a density of 7 × 10 cells/ml in liquid CTT medium at 32°C, transferred to a microscope slide with a thin 1.0% agar pad buffered with A50 buffer (10 mM MOPS, pH 7.2, 10 mM CaCl, 10 mM MgCl, 50 mM NaCl), and immediately covered with a coverslip. After 30 min at room temperature, cells were observed in a Leica DM6000B microscope, using a Leica Plan Apo × 100/NA 1.40 phase-contrast oil objective, and visualized with a Leica DFC 350FX camera. For fluorescence microscopy, a Leica GFP filter (excitation range 500–550 nm, emission range 450–490 nm) was used for visualizing GFP proteins, and fluorescein-conjugated antibodies and a Y3 filter (excitation range 530–560, emission range 570–650 nm) were used for visualizing RomR-mDsRed- and Cy3-stained cells. Images were recorded and processed with Leica FW4000 V1.2.1 software. Processed images were arranged in Adobe Photoshop 6. For time-lapse recordings, cells were treated as described and imaged at 30-s intervals for 10 min, and images were processed as described. All cells analyzed from the time-lapse recordings were separated from other cells by at least one cell length to ensure that cells moved only by means of the A-motility system. Fluorescence signals were quantified using the region measurement tool in Metamorph 7.0r2 (Visitron Systems). Immunofluorescence microscopy was as described (). Briefly, cells were grown as described and fixed with 2.6% paraformaldehyde and 0.008% glutaraldehyde for 20 min on freshly prepared poly--lysine-treated 12-well diagnostic slides (Erie Scientific Company). Cells were permeabilized with 0.2 μg/ml lysozyme for 4 min and probed with affinity-purified, rabbit polyclonal anti-RomR antibodies at 4°C overnight in PBS buffer (137 mM NaCl, 2.7 mM KCl, 10 mM NaHPO, 1.8 mM KHPO, pH 7.4) supplemented with 2% BSA. Fluorescein-conjugated goat anti-rabbit antibodies (Perbio Science) were used as a secondary antibody. Slow Fade Anti Fade Reagent (Molecular Probes) was added to each well, and cells were visualized and imaged as described. To stain Tfp, the procedure of was adapted. Briefly, cells were grown on 1.5% agar plates supplemented with 1% CTT, scraped off the agar, and resuspended in 100 μl labeling buffer (50 mM KPO, pH 8.0, 5 mM MgCl, 25 μM EDTA). Cells were harvested by gentle centrifugation and gently resuspended in labeling buffer. This step was repeated three times. Cy3 from one vial (Amersham Biosciences) was dissolved in 250 μl labeling buffer and added to the cells, and the mixture was incubated for 1 h at 20°C. Cells were washed three times in labeling buffer, spotted onto a glass slide, covered with a coverslip, and visualized as described above. Purification of RomR for generating anti-RomR antibodies and immunoblotting is described in .

For eukaryotic chromosome segregation to occur correctly, the two copies of the chromosome (sister chromatids) need to be physically connected. This is accomplished by the ‘cohesin' complex (), which links sister chromatids from their generation in S-phase until the onset of anaphase. By metaphase, the two sister chromatids become attached to microtubules emanating from opposite spindle poles (bi-orientation). A central component ensuring the bi-orientation of sister kinetochores is the ‘chromosomal passenger complex' (CPC) that is composed of Aurora-B, INCENP, Borealin/Dasra and Survivin (). In anaphase, cohesion between the sister chromatids is released, which is accomplished by cleavage of the Scc1/Rad21 subunit of cohesin by the protease separase (), a process that is under control of the mitotic spindle checkpoint (). Similar principles govern meiotic chromosome segregation (; ), but in contrast to mitosis, two rounds of chromosome segregation follow only one round of DNA replication in order to generate haploid gametes. Several modifications allow this two-step process (): homologous chromosomes become connected via chiasmata, which result from crossover recombination during meiotic prophase, and sister kinetochores on each chromosome adopt a side-by-side rather than back-to-back conformation (). These two mechanisms allow the recognition of homologous chromosomes as entities destined for opposite poles during anaphase-I. In addition, only cohesion between chromosome arms is lost during anaphase-I, which allows the separation of homologs; cohesion at the centromere is preserved depending on Shugoshin proteins (), and sister chromatids therefore stay connected so that they can be properly segregated during meiosis-II. For meiosis-II, kinetochores are again in a back-to-back position, and chromosome segregation is very similar to mitosis. In all model eukaryotes that have been studied, Aurora-B kinases are required for the proper bi-orientation of sister chromatids in mitosis. In the absence of Aurora-B, syntelic (both sister kinetochores attached to the same spindle pole) or merotelic (one kinetochore attached to two opposing spindle poles) attachment of chromosomes occurs with increased frequency (; ; ; ). These malattachments escape the surveillance by the mitotic spindle checkpoint, indicating that Aurora-B kinases are also required for proper checkpoint function. Budding and fission yeast have a single Aurora kinase, Ipl1 and Ark1, respectively. These single Aurora kinases are thought to be homologous to Aurora-B and presumably interact with INCENP ( () Pic1) and Survivin ( Bir1) homologs (; ; ). Whereas Ipl1 has been shown to be required for the proper bi-orientation of chromosomes (; ; ), a role for Ark1 in regulating chromosome attachment has not been demonstrated. Fission yeast cells that lack Ark1 fail to divide the chromatin during anaphase, but nevertheless proceed to septation, resulting in a cut (‘cell untimely torn') phenotype (). Mutants in the condensin complex, which is required for compaction of chromatin during mitosis, display a similar phenotype (), and Ark1 and Bir1 are indeed required for the correct intranuclear localization of condensin subunits during mitosis (; ). In meiosis, Aurora-B kinases have been shown to regulate cohesion in and . In the worm, the Aurora-B kinase AIR-2 promotes segregation of homologous chromosomes, presumably by phosphorylation-dependent removal of meiotic cohesin, which contains the meiosis-specific subunit Rec8 that replaces Rad21 (; ). In , Aurora-B seems to be required to preserve centromeric cohesion beyond meiosis-I, depending on MEI-S332, a member of the Shugoshin family of proteins (). The only indication for a role of Aurora in controlling chromosome attachment in meiosis comes from budding yeast, where very recent work has shown that Ipl1 is required for the bi-orientation of homologous chromosomes (; ). Here, we examined the role of the fission yeast Aurora kinase Ark1 in chromosome segregation during mitosis and meiosis. We find that Ark1 is required to promote the bi-orientation of chromosomes in mitosis and to prevent or correct syntelic and merotelic attachment. Furthermore, Ark1 is necessary for the bi-orientation of homologs in meiosis-I. However, notably different from budding yeast (), we describe that fission yeast Aurora is required for the monopolar attachment of sister chromatids in meiosis-I, and acts in a different pathway from the Moa1/Rec8 mono-orientation pathway that has been defined. To address whether Ark1 has a role in the bi-orientation of chromosomes, we examined chromosome segregation in haploid fission yeast, in which chromosome 2 was marked with GFP close to the centromere (-GFP; ) using a temperature-sensitive allele of (; ) or analog-sensitive versions of Ark1 (, , see Materials and methods), which can be inactivated by specific inhibitors (4-amino-1-tert-butyl-3-(1′-napthyl)pyrazolo[3,4-d]pyrimidine (1NA-PP1) or 4-amino-1-tert-butyl-3-(1′-napthylmethyl)pyrazolo[3,4-d]pyrimidine (1NM-PP1)). When Ark1 was inactivated, centromere segregation was perturbed in half or more than half of the cells (). In about 40% of cells, one of the sister chromatids did not move entirely toward one pole in anaphase (‘lagging'), and in 11–25% (depending on the allele), both sister centromeres moved to the same pole. This is indicative of merotelic and syntelic chromosome attachment, respectively. To obtain further insight into chromosome movement in -mutant cells, we observed chromosome segregation in living cells at the restrictive temperature using -GFP and the spindle pole body (SPB) marker Sid4-GFP (). We found that even before anaphase, alignment of chromosomes on the mitotic spindle was defective, since the -GFP mark was found close to one of the spindle poles in about 60% of the cells at restrictive temperature (, and data not shown). In anaphase, sister chromatids of chromosome 2 missegregated in 28% of the cells. In a few cases (3% of all cells) the sister chromatids stayed entirely at one pole during anaphase. In the other cases, sister chromatids were lagging behind on the spindle during anaphase, with about two-thirds eventually moving to the correct and the remainder moving toward the incorrect pole (). Cells with mutations in condensin fail to segregate the bulk of chromatin very similar to -mutant cells, and we therefore wanted to exclude that the chromosome segregation defect observed after Ark1 inhibition is merely a consequence of the condensation defect. When we observed -GFP segregation in temperature-sensitive condensin-mutant cells (), we also found some failure in centromere segregation (), which might be attributed to the disturbed structure of the chromosomes (). However, this defect in centromere segregation was less pronounced than in -mutant cells, although the chromatin condensation defect in these two strains was similar (). Furthermore, in the presence of low concentrations of the microtubule-destabilizing substance TBZ, cells exhibited considerable missegregation (about 35%) even at the permissive temperature when chromosome condensation was largely normal (). In contrast, in wild-type or -mutant cells, TBZ only caused a very minor increase in missegregation (). This suggests that Ark1 is needed to establish proper microtubule–kinetochore attachment independent of its role in chromosome condensation. Ark1 also functions in the mitotic spindle checkpoint (; data not shown). However, even the additional deletion of the checkpoint gene + in cells leads to a weaker defect in chromosome segregation than the one observed in cells (). Taken together, these results indicate that Ark1 plays a specific role in promoting proper chromosome segregation beyond its role in the mitotic checkpoint and in chromosome condensation. In both budding yeast and metazoans, Aurora kinases act in the bi-orientation of chromosomes by correcting improper, syntelic attachment of chromosomes to the same spindle pole (; ; ). When we inactivated Ark1 in an otherwise unperturbed mitosis, segregation of both sisters to one spindle pole was rare (). This could be because in an unperturbed mitosis, initial syntelic attachment is rare or because syntelic attachment can still be corrected in -mutant cells. We therefore increased the frequency of misattachment by first arresting cells in mitosis without microtubules, using a cold-sensitive tubulin mutant and then releasing to permissive temperature (; ; ). Because -mutant cells do not arrest in mitosis under these conditions (; data not shown), we could only perform the experiment with the allele, which is functional when cells are grown without inhibitor and thus allows arrest by . Shortly before release, we inhibited Ark1-as3 by the specific inhibitor 1NM-PP1, which inactivates the kinase within 10 min (data not shown and ). When checking anaphase spindles 10 min after the release, both -GFP marks were found close to one edge of the spindle in about 20% of cells. No such missegregation was observed in wild-type cells treated with inhibitor or in cells grown without inhibitor (), indicating that its occurrence depends on the inactivation of Ark1. To determine whether this indeed represents initial misattachment that fails to correct, we filmed Ark1-inactivated cells being released from the arrest. Among those cells that we could image from the start of mitosis, about half had both sister centromeres located entirely at one pole throughout prometaphase and anaphase (; ). These data indicate that Ark1 is required to correct syntelic misattachment of chromosomes (also see ). To confirm in a different setting that malattachment of chromosomes cannot be corrected in -mutant cells, we used a temperature-sensitive mutant of fission yeast cohesin, (). Sister chromatids precociously detach from each other in the absence of cohesin, but remain competent to attach to microtubules. However, in the absence of cohesion between sister chromatids, kinetochore–microtubule attachment cannot be stabilized because of a lack of tension. Therefore, attachment remains unstable, which in budding yeast depends on Ipl1 (). In live-cell microscopy experiments, the instability of sister chromatid attachment in cells was exemplified by an occasional switching of at least one sister chromatid from one spindle pole to the other (, arrows). When Ark1 was additionally inactivated, such switching was abolished (), indicating that Ark1 is required to keep sister chromatid attachment unstable in cells. This result is consistent with Ark1 being needed to correct kinetochore–microtubule attachments that fail to generate tension through bi-orientation. However, the fact that sister chromatids were not very motile when Ark1 is inhibited () could also indicate a more profound defect in kinetochore–microtubule dynamics. Inactivation of condensin did not seem to have the same effect as Ark1 inactivation (), which again suggests that Ark1 regulates chromosome attachment independent of condensin. To study the requirement of Ark1 in meiosis, we used an ‘ shut-off' () strain, where the + gene is under control of the mitosis-specific promoter of the + gene. In this strain, the level of Ark1 protein in meiosis was largely reduced, but minor amounts of Ark1 were still observed at centromeres and midspindles (). For some experiments, we therefore either additionally or alternatively used the allele together with the specific inhibitor. The most prominent phenotype observed after Ark1 depletion in mitosis is a failure to condense chromosomes, which leads to a defect in nuclear division (). We similarly found a defect in chromosome condensation during meiosis in cells (). Nevertheless, four distinct, albeit often unequally sized, nuclei formed after meiosis-II in cells (). However, in about 50% of cells, the two meiosis-II spindles formed extremely close to each other and often in what seemed to be one nucleus (), indicating a failure of nuclear division during meiosis-I. To assess this phenotype better, we arrested cells after meiosis-I using the mutation (). Under these conditions, more than 80% of + cells but only 30% of cells formed two nuclei (). Nevertheless, the mononucleated cells seemed to have undergone anaphase-I, because Rec8-GFP was largely removed from chromatin (). The failure in nuclear segregation after Ark1 depletion presumably reflects a failure in condensin function, since mutants in the condensin subunit also show a slightly increased number of cells with one nucleus when arrested by the mutation at a semi-permissive temperature (data not shown). To address whether Ark1 has any role in meiotic chromosome segregation, we examined the segregation of -GFP in and cells during anaphase-I. When Ark1 was inhibited or depleted, both homologous chromosomes 2 segregated to the same pole in about 30–40% of anaphase-I cells (). In another 30% of cells, at least one of the homologs was lagging on the anaphase-I spindle. These segregation problems could be caused either by a failure to bi-orient bivalents or by a failure to join homologous chromosomes through chiasmata. Since the intergenic recombination frequency between the + and + locus on chromosome 1 was similar in + and strains (data not shown), the latter is unlikely. Therefore, these data suggest that Ark1 is required for the bi-orientation of connected homologs during meiosis-I (also see ). have shown that specific features of the chromosome and not of the spindle determine the special chromosome segregation of meiosis-I. In fission yeast, genetic tricks can be used to create mitosis-like chromosomes during meiosis-I. Deletion of the gene for the meiosis-specific cohesin subunit Rec8 causes a failure in recombination and therefore chiasmata generation. Additionally, sister kinetochores in each homolog are faithfully bi-oriented presumably because geometry at the centromeric region is mitosis-like and sister chromatids remain cohered by mitotic cohesin complexes, which persist into meiosis (). Similarly, mitosis-like chromosomes can be generated by deleting +, which is required for recombination, and +, the gene product of which is required for the mono-orientation of sister kinetochores in concert with Rec8 (). In both these genetic backgrounds, Ark1 shut-off caused missegregation of sister centromeres in meiosis-I (), with the effect being greater in a Δ Δ strain than in a Δ strain (see ). Thus, during meiosis-I, Ark1 can promote the equational segregation of sister chromatids (), or the bipolar segregation of homologs (), depending on the chromosome structure. We suggest that Ark1 promotes the bi-orientation of any two kinetochore-containing entities that are connected. This is in accordance with findings by , who showed that budding yeast Ipl1 ensures the bi-orientation of two separate kinetochores on an unreplicated plasmid. When we tested the segregation of homologs during anaphase-I, we found that frequently at least one of the homologs was lagging and often the -GFP signal of lagging homologs split in two (). We therefore hypothesized that the sister kinetochores on one homolog, which normally attach to only one spindle pole, were pulled in opposite direction. To verify this assumption, we labeled only one of the homologous chromosomes 2 with GFP and determined the segregation pattern. Indeed, in about 12% of anaphase-I cells, the GFP signal split in two when Ark1 was depleted (), indicating that sister chromatids were pulled to opposite poles and separated precociously. These data suggested that in the absence of Ark1, sister kinetochores erroneously become attached to opposite spindle poles during meiosis-I. In accordance, we found that a visible separation of sister centromeres could already be observed during metaphase-I (). We therefore considered the possibility that Ark1 is required for the localization and function of Moa1, which prevents bi-orientation of sister chromatids at meiosis-I (). However, visualization of Moa1-GFP did not reveal any difference between wild-type and cells during meiosis-I (). Similarly, no significant difference in Moa1 localization was observed by chromatin immunoprecipitation (ChIP; ). Furthermore, the localization of Rec8 at the centromeric central core, which is important for mono-orientation of kinetochores (; ), was intact in the cells (). Thus, the mono-orientation defect caused by the reduction of Ark1 is likely different from the one caused by the absence of Moa1 or Rec8, suggesting that Ark1 and Moa1/Rec8 influence sister kinetochore mono-orientation in meiosis-I through distinct mechanisms. Cells in which + is deleted show entirely equational segregation of sister chromatids in meiosis-I if recombination is abolished by Δ (see ). In this situation lagging chromatids can be observed in anaphase-I. Their appearance is completely suppressed by deleting +, the meiosis-I-specific protector of centromeric cohesion (). In contrast, neither deletion of + nor the additional deletion of + leads to completely equational segregation in cells (, and data not shown). When contemplating the reason for the mono-orientation defect in cells, we noticed the high frequency of lagging chromosomes or lagging sister chromatids during anaphase-I ( and ). Given that lagging chromosomes may originate from merotelic attachment and Ark1 is involved in its correction (), we envisaged that the primary defect in cells preventing monopolar attachment could be the inability to correct merotelic attachment of a unified pair of sister kinetochores (). In Δ Δ cells, all sister centromeres eventually segregated at anaphase-I even in the presence of Sgo1, implying that microtubule-mediated pulling on bi-oriented sister kinetochores can overcome Sgo1-mediated protection (; ; ). The tension on homologs that are attached in a merotelic manner is expected to be less, because some microtubules on both sister kinetochores likely attach to the same pole (, ‘+'). This reduced tension might not be sufficient to overcome protection by Sgo1. This hypothesis makes the key prediction that deprotection of sister chromatid cohesion by Δ would increase the equational segregation of sister centromeres in cells, different from Δ Δ cells (see ). Consistent with this scenario, the ratio of cells in which sister centromeres were moving entirely to opposite poles in anaphase-I increased from ∼1% in to 17% in Δ cells (). Furthermore, lagging sister chromatids could be observed in cells even after + deletion, indicating that single chromatids were attached in a merotelic manner ( and , ‘Δ'). Thus, we suggest that Ark1 promotes monopolar attachment of sister kinetochores at meiosis-I most likely by correcting merotelic attachment of paired sister kinetochores, a mechanism that is fundamentally different from that of Moa1 and Rec8, which are thought to promote the side-by-side orientation of sister kinetochores by fostering cohesion in the central core region of the centromere (). Recent reports indicated that the Shugoshin protein Sgo2 interacts with the CPC protein Bir1 and is required for the full recruitment of the CPC including Ark1 to centromeres (; ). Cells depleted of Sgo2 exhibit non-disjunction of homologs in about 20% and equational segregation of sister centromeres in ∼5% at meiosis-I (; ; ; ; and , and data not shown). Since localization of Ark1 at centromeres is reduced in Δ cells at meiosis-I (; ), the non-disjunction of homologs in Δ cells was attributed to defects in Ark1 function. Based on the results described above, we hypothesized that the mono-orientation defect in Δ cells originates from the inability to correct merotelic attachment of paired sister kinetochores, like in cells. Supporting this assumption, an attempted separation of sister centromeres at anaphase-I was observed in about 16% of Δ cells, which is very similar to that in cells (). Moreover, deletion of + increased the equational segregation of sister centromeres of chromosome 2 at meiosis-I to 12% (; also see ), which is again very similar to that in cells. Thus, we conclude that defects in meiotic chromosome segregation in Δ cells are mostly caused by reduced Ark1 function. In contrast to the chromosome segregation function, the role of Ark1 in promoting chromosome condensation is not shared by Sgo2 (). Previous data supported the view that Bub1 acts upstream of Sgo1 and Sgo2 (; ). Fittingly, Ark1-GFP localization was perturbed during metaphase-I to a similar extent as in Δ cells by deletion of + (). Furthermore, we found that deletion of + did not enhance the mono-orientation defect of Δ cells (), indicating that Bub1 and Sgo2 work in a single pathway. These results suggest that the defects in mono-orientation after Bub1 depletion might be partly caused by the inability to correct merotelic attachment of paired sister kinetochores, similar to the situation in or Δ cells (see Discussion). Aurora kinases are highly conserved throughout eukaryotes, and have been implicated in proper chromosome segregation in several organisms (). Noticeably, however, in fission yeast, a well-studied model in mitosis research, it was unknown whether the single Aurora kinase, Ark1, has any role in regulating the proper attachment of chromosomes during mitosis. Here, we demonstrate that Ark1 inhibition causes misattachment of chromosomes to the mitotic spindle (). As in budding yeast (), Aurora seems necessary to destabilize syntelic attachment that fails to create tension (). In addition, lagging chromatids occurred with high frequency when Ark1 was inhibited (). Those might arise because attachment to microtubules is weak or dysfunctional, or because the corresponding kinetochore is attached to both spindle poles (merotelic attachment). We favor the latter hypothesis, since most sister chromatids in cells move to the spindle poles in anaphase-A with velocities comparable to those observed in wild-type cells (data not shown), indicating that there is no general problem with attachment or microtubule-dependent anaphase movement. In vertebrate cells, where kinetochore–microtubule attachment can be visualized directly, it has been shown that Aurora-B is required to suppress merotelic attachment, possibly by destabilizing the faulty attachment (; ). We consider it an additional possibility that Aurora is required to build the kinetochore in a way that favors attachment of all microtubules on one kinetochore to the same pole. It has been proposed that fission yeast Pcs1, which is a homolog of one of the components of the budding yeast monopolin complex, is required to clamp together single microtubule-binding sites on one kinetochore, thus favoring their attachment to one pole (). Indeed, deletion of + causes kinetochores to attach in a merotelic manner (; ). Thus, Ark1 might be required for Pcs1 function. Our preliminary experiments nevertheless failed to detect an influence of Ark1 on Pcs1 localization (data not shown). In addition to its functions in regulating kinetochore attachment, budding yeast Ipl1 is a component of the ‘NoCut' pathway, which prevents abscission in the presence of spindle-midzone defects (). In our experiments it was evident that an equatorial microtubule ring, which normally forms during mid or late anaphase-B in the plane of cell division (; ), was formed precociously when Ark1 was inhibited (), thus also implying Ark1 in the regulation of cytokinesis. Since Ark1 promotes sister chromatid bi-orientation in mitosis, but sister chromatids have to mono-orient during meiosis-I, it was unclear how Ark1 would influence chromosome segregation in meiosis-I. We found that Ark1 normally promotes the bi-orientation of homologs in meiosis-I, but if the morphology of the bivalent is disrupted and mitosis-like chromosomes are created, Ark1 promotes the bi-orientation of sister chromatids (). This indicates that the molecular mechanism of Ark1 function is the same in mitosis and meiosis, and the different outcome is determined by the structure of the bivalent. In contrast to mitosis, where Ark1 inhibition causes only about 10–20% sister chromatid co-segregation, depletion of Ark1 in meiosis causes a more pronounced co-segregation of homologous chromosomes (30–40%). The most likely explanation is that tension-controlled correction of attachment is more important in meiosis, because the two pairs of sister kinetochores on a bivalent are not as tightly coupled as the sister kinetochores on a mitotic chromosome, which might favor their syntelic attachment (). Very recently it has been demonstrated that the budding yeast Aurora kinase, Ipl1, is similarly required for homolog bi-orientation and sister chromatid bi-orientation of artificial mitosis-like chromosomes in meiosis-I (). Because of the high conservation of Aurora functions in all eukaryotes, we expect that this will also hold true for metazoans. In , the Aurora-B kinase AIR-2 seems to be required to efficiently remove cohesin complexes containing Rec8 from chromosome arms (; ). In contrast, we find that Ark1 is not essential for Rec8 removal during meiosis-I in fission yeast (). In time-lapse movies of fission yeast expressing Rec8-GFP, one can clearly observe the solubilization of Rec8 at the onset of anaphase-I, presumably at the moment when it is cleaved and removed from chromatin (). Although this step is not as easy to discern when Ark1 is inhibited, because chromosomes are less condensed, it is clear that this solubilization also happens fairly efficiently if Ark1 is inhibited and Rec8 is subsequently degraded with kinetics similar to wild-type cells (). Nevertheless, it is possible that Ark1 facilitates but is not essential for Rec8 cleavage (see ). We show that fission yeast Ark1 is required for the faithful mono-orientation of sister chromatids in meiosis-I (). In fission yeast, mono-orientation of sister kinetochores is regulated by Moa1 and Rec8, which may cooperatively promote the formation of a side-by-side structure of sister kinetochores through cohesion of the centromeric central core region (). The depletion of Moa1 together with Rec12 or of Rec8 entirely disrupts the mono-orientation of sister chromatids at meiosis-I. The mono-orientation defect in cells is less pronounced, and Moa1- or Rec8-localization is not disrupted when Ark1 is depleted (; ). This indicates that Ark1 and Moa1 act in separate pathways to promote mono-orientation. Our experiments suggest that in the absence of Ark1, sister kinetochores on one homolog become attached in a merotelic manner so that they are torn apart at anaphase-I, even though they have the proper side-by-side configuration that favors mono-orientation (). Thus, the complicated chromosome segregation defects in meiotic cells depleted of Ark1 can be explained by the well-recognized role of Aurora in correcting malattachment of chromosomes. Syntelic and merotelic attachment of bivalents in meiosis-I provokes non-disjunction of homologs and precocious sister separation, respectively. In budding yeast, it has been proposed that the two pairs of sister kinetochores on a bivalent only attach to one microtubule each () and one sister kinetochore may thus be inactivated. Consequently, merotelic attachment might not be possible, which would explain the unperturbed mono-orientation in Ipl1-depleted cells () despite the otherwise similar function of Ipl1 and Ark1. There is, however, controversy in the literature whether the depletion of Ipl1 causes a mono-orientation defect (). In any case, our data clearly indicate that in fission yeast, both sister kinetochores are active in meiosis-I and can attach to microtubules. As attachment of both kinetochores to microtubules at meiosis-I is observed in several organisms (; ), the mechanism we identified here may be functional in other eukaryotes as well. Our finding that Ark1 is needed for the bi-orientation of homologs and the mono-orientation of sister kinetochores in meiosis-I provides a crucial clue to solve the enigma why Sgo2 and Bub1 are required for monopolar attachment as well as proper homolog disjunction in meiosis-I. In either Δ or Δ cells, centromeric localization of Ark1 is reduced at meiosis-I (). This is consistent with previous findings that Bub1 acts upstream of Sgo2, which in turn plays a crucial role to load the CPC to centromeres (; ; ). Since Moa1 localization is intact in either Δ or Δ cells (data not shown), it is reasonable to assume that perturbation of mono-orientation in these cells may originate from the reduced Ark1 activity at centromeres. Indeed, the defects of Δ cells in monopolar attachment at meiosis-I resemble those of cells (). Because Δ cells are defective in both Sgo1 and Sgo2 localization to centromeres (; ), one would expect that Δ phenocopies the + + double deletion. However, the frequency of equational segregation is significantly higher in Δ cells (∼30%) than Δ Δ cells (12%) (; ). Since in mitosis Δ cells show a higher number of lagging chromosomes than Δ cells (; S Kawashima and Y Watanabe, unpublished results), we suggest that Bub1 has functions that go beyond Sgo2 regulation both in mitosis and meiosis. Whatever the nature of these additional functions is, our results argue that Bub1 depletion perturbs monopolar attachment by generating merotelic attachment of paired sister kinetochores, like Sgo2 or Ark1 depletion. All strains used in this study are listed in in . To generate the allele (Ark1-Leu166Ala; ), the gene was PCR-mutagenized from a strain into which a hygromycin-resistance cassette (hygR) had been integrated 400 bp 5′ of the + open reading frame (ORF). The hygR- construct was integrated in a wild-type strain at the endogenous locus. The allele rendered the cells sensitive to 5 μM 1NA-PP1 or 5 μM 1NM-PP1 (both from TRC, North York, ON, Canada). The strain was created from by additionally mutating Ser229 to Ala. Both the and the strain used in this study contain the additional amino-acid mutations Gln28Arg and Gln176Arg, which were unintentionally inserted during the first PCR mutagenesis. Because the strain without addition of inhibitor grows similar to a wild-type strain and is not benomyl-sensitive like other -mutants, it is unlikely that the two additional mutations affect the functionality of Ark1. The strain is benomyl-sensitive even when grown without inhibitor. To create the or allele (see ), a kanamycin- or hygromycin-resistance cassette and about 1000 bp of promoter region from the + gene were integrated 5′ of the respective ORF by PCR-based gene targeting (). To visualize microtubules, CFP-Atb2 or mCherry-Atb2 were expressed from the promoter in the pREP81 plasmid (+), or mCherry-Atb2 was expressed from the endogenous locus by integration of the promoter and the mCherry-coding region upstream of the + ORF. To tag Rec8 with GFP, we modified plasmid pFA6a-GFP(S65T)-hphMX6 () by integrating 950 bp from the 3′-end of the + ORF upstream of GFP and 350-bp genomic sequence 3′ of the + ORF downstream of GFP. The plasmid was linearized by I digest and integrated in a wild-type strain. The strains or alleles not mentioned above have been described (; ; ; ; ; ; ). Medium for mitotic cultures was YEA (YE with additional 50 mg/l adenine) or minimal medium (MM) containing 5 g/l NHCl and additional nutrients if required (). To synchronize cells in mitosis (), we arrested cells in S-phase by incubation in 12 mM hydroxyurea for 4.5 h at 25 or 30°C depending on the strain. The arrest was released by washing the cells twice with fresh, warm medium before reculturing. Cells carrying the mutation () were arrested in mitosis by incubation at 19°C for 6 h and subsequently released by shifting to 32°C. To observe cells in meiosis, cells were first grown to logarithmic phase. If the or promoter should be induced by thiamine depletion, cells were grown in MM containing 5 g/l NHCl and, if necessary, 200 mg/l leucine and 50 mg/l adenine for about 14 h at 30°C. Cells were washed, collected and spotted on sporulation agar (SPA; ) to which leucine or adenine had been added if necessary. After a further 7–8 h of incubation at 30°C, cells were observed directly or fixed by methanol at −80°C. To observe cells in meiosis, the cells were first incubated in MM with 5 g/l NHCl for 8–9 h, and then washed and incubated in MM without NHCl for 4–5 h before spotting on a plate with synthetic sporulation agar (SSA; SSL with agar; ) containing 5 μM 1NA-PP1. Cells were observed after 7–10 h of incubation at 30°C. For immunostaining, cells were fixed with paraformaldehyde. To stain microtubules, we used the mouse anti-tubulin TAT1 antibody (kind gift from K Gull) at a dilution of 1:200, followed by an Alexa568-coupled anti-mouse secondary antibody (Invitrogen) at 2 μg/ml. To stain DNA, methanol-fixed cells were washed, resuspended in PEM buffer (100 mM PIPES, 5 mM EGTA, 5 mM MgCl, pH 6.9) and stained by 1 μg/ml Hoechst 33342 or 1 μg/ml DAPI. Images were acquired on a Zeiss AxioImager microscope (Zeiss, Jena, Germany) with MetaMorph software (Molecular Devices Corporation, Downingtown, PA). Typically, a Z-stack of about 4-μm thickness, with single planes spaced by 0.25–0.4 μm, was acquired and subsequently projected to a single image. To compare signal intensities, all images were taken with the same exposure conditions and processed similarly. Live-cell recordings were performed on a DeltaVision RT system (Applied Precision, Issaquah, WA) equipped with a heating chamber. For imaging mitosis in -mutants, cells were grown in liquid medium at permissive temperature, transferred to a glass-bottom culture dish (MatTek, Ashland, MA) coated with lectin and incubated on the microscope stage at the restrictive temperature (34°C) for at least 1 h, before starting image acquisition. To image cells in an release, cells were transferred from a liquid culture at 19°C to a glass-bottom culture dish coated with lectin, which was placed in the microscope chamber heated to 32°C. Image acquisition was started immediately. Images usually were acquired with the Z-sweep acquisition (OAI) feature and deconvolved using softWoRx software. Kymographs were assembled with Adobe Photoshop and Image Ready software. The procedure was carried out essentially as described previously (; ). Anti-Moa1 polyclonal antibodies, anti-GFP polyclonal antibodies (Living Colors Full-length A.v. Polyclonal Antibody, Clontech) and anti-Cnp1 polyclonal antibodies were used for IP (). DNA prepared from whole-cell extracts or immunoprecipitated fractions was analyzed by quantitative PCR with ABI PRISM7000 (Applied Biosystems) using SYBR Premix Ex Taq (Perfect Real Time; Takara). The primers used for PCR were described previously (; ).

To test whether human DCs could alter the clonogenic growth of human myeloma cell lines, we plated tumor cells alone or with purified monocytes or monocyte-derived DCs in methylcellulose cultures. Plating tumor cells alone in this assay results in the growth of discrete tumor colonies with an efficiency of ∼1% of cells plated. The addition of DCs to these cultures led to a greater number of tumor colonies in a dose-dependent manner compared with tumor cells alone or cocultures with monocytes (). This growth-promoting effect of DCs on human tumors is not restricted to myeloma, as the clonogenic growth of two other tumors tested (lymphoma and breast cancer) was also enhanced (). In contrast, DCs had only a minor impact on the growth of glial tumors. The enhanced number of tumor colonies was mostly evident as an increased number rather than size of individual colonies, suggesting an effect on cloning efficiency or survival (). Therefore, interactions of tumor cells and DCs can directly promote the clonogenicity of several but not all human tumors. The phenotype of tumor colonies in the clonogenic assay was monitored by flow cytometric detection of CD138 and CD11c, and the presence of myeloma cells was identified by the presence of cells expressing the appropriate cytoplasmic Ig light chain (λ light chain in the case of U266 cells; ). Flow cytometry data were also confirmed by immunofluorescence microscopy (). The majority of the tumor cells in these cultures had a typical plasma cell phenotype with the expression of CD138 and light chain restriction. However, the culture of tumor cells in the presence of DCs led to a mild but consistent increase in the proportion of cells lacking CD138, a marker of terminally differentiated plasma cells (). Upon immunofluorescence microscopy, tumor cells grown in the presence of DCs were somewhat smaller in size with less cytoplasm compared with clonogenic cultures of tumor cells grown alone (). This is also evident as lower forward scatter of these tumors on flow cytometry (). Therefore, DC-mediated enhancement of myeloma clonogenicity is associated with an altered phenotype of tumors. A prior study has suggested that the CD138 subpopulation of MM cell lines is enriched for the clonogenic growth in serial replating assays (). Thus, we tested whether this altered phenotype was associated with enhanced clonogenicity in serial replating assays. Cells initially cultured with DCs had higher cloning efficiency in replating assays, suggesting that the observed alteration in phenotype is associated with the enhancement of clonogenicity (). DC-mediated enhancement of tumor clonogenicity in this system required short-range interactions between tumor cells and DCs, as it was not evident when the two cell populations were separated in a Transwell (). To optimize cell–cell contact, we also evaluated the initial coculture of DCs and tumor cells in suspension culture for 24 h before plating them in methylcellulose. Adding this step led to a further increase in tumor colonies compared with direct cocultures, supporting the need for cellular proximity (). DCs express several molecules that are implicated in B cell differentiation as well as costimulatory molecules and integrins that may be important for the observed effects (). We focused on two of these pathways involving TNF superfamily members. Both B cell–activating factor (BAFF)–APRIL (a proliferation-inducing ligand; references –) and receptor activator of NF-kB (RANK)–RANK ligand (RANK-L) pathways () have been previously implicated in the survival of myeloma cells. Blockade of BAFF–APRIL-mediated interactions with TACI-Fc (transmembrane activator calcium modulator and cyclophilin ligand interactor-Fc chimera) or blockade of RANK–RANK-L–mediated interactions with osteoprotegerin (OPG; reference ) led to the inhibition of DC-mediated enhancement of tumor clonogenicity in both myeloma cell lines tested (, c and d). This was also associated with less enrichment of the CD138 subpopulation in these cocultures (). TACI-Fc or OPG did not alter the clonogenicity of tumor cells alone. We were unable to demonstrate a synergy between these ligands under the conditions tested. Therefore, DC-mediated enhancement of myeloma clonogenicity is mediated, in part, by RANK-L and BAFF–APRIL-mediated interactions. To gain insights into the early events during tumor–DC interaction, we cocultured DCs and tumor cells for 24 h before separating tumor cells by FACS sorting to >99% purity and plated them in clonogenic assays without DCs. Tumor cells from these short cocultures demonstrated enhanced clonogenic growth compared with mock-sorted tumor cells cultured alone under similar conditions (). This suggested that even short-term interactions between tumor cells and DCs can alter the behavior of tumor cells. Pilot microarray experiments suggested B cell lymphoma 6 (BCL6) as one of the major genes consistently up-regulated in tumor cells purified from these cocultures (unpublished data). A prior study has suggested an important role for BCL6 in survival and self-renewal of germinal center B cells (). Thus, we analyzed the expression of BCL6 in these tumor cells by real-time RT-PCR (TaqMan) to confirm these results. Coculture of tumor cells with DCs was associated with an induction of BCL6 messenger RNA (mRNA) in sorted tumor cells compared with tumors cultured alone (). This was also confirmed at the protein level by immunofluorescence microscopy. U266 tumor cells cultured alone do not express BCL6; however, clear intranuclear staining for BCL6 was observed in tumor cells cocultured with DCs (). As TACI-Fc and OPG inhibited the DC-mediated enhancement of tumor clonogenicity, we also tested their effect on BCL6 up-regulation. The addition of TACI and OPG led to the modest but detectable inhibition of DC-mediated BCL6 mRNA up-regulation in tumor cells (). Therefore, the short-term coculture of DCs and tumor cells is associated with an up-regulation of BCL6 on tumor cells. To extend these observations on cell lines to primary cells from patients, we obtained bone marrow samples from patients with myeloma and preneoplastic gammopathy (MGUS). A prior study has shown that clonogenic growth of tumor cells is enriched in CD34CD138 subpopulations (). Bone marrow mononuclear cells (MNCs) from myeloma ( = 9) or MGUS ( = 3) patients were separated into CD138 and CD34CD138 subpopulations and cultured in the presence or absence of autologous or allogeneic DCs in clonogenic assays. The addition of DCs to these subpopulations led to a more than twofold enhancement of tumor colonies from the CD34CD138 fraction in 4/5 patients (two MGUS and three MM) using autologous DCs and 7/11 patients (two MGUS and nine MM) using allogeneic DCs (). There were no major differences between MGUS and myeloma samples. To further assess their clonogenicity, tumor cells harvested from some of these assays were replated in fresh assays. Tumor cells could be successfully passaged in serial assays from both subpopulations (). Importantly, coculture with DCs allowed clonogenic growth even from the purified CD138 subpopulation from myeloma patients, which normally does not grow well in vitro by itself (). Therefore, DCs lead to the enhanced clonogenic growth of primary tumor cells from myeloma patients, and this coculture system may be a useful model system for the growth of primary myeloma cells. Most human tumors recruit diverse immune cells, including DCs, to the tumor bed. However, infiltration of human tumors by DCs has previously been interpreted largely in the context of their immunologic functions (, ). Prior studies have shown that both myeloma tumors in patients and mouse plasmacytomas are extensively infiltrated by DCs (, ), accounting for up to 10% of all nucleated cells within these lesions. Our data suggest the possibility that tumor-infiltrating DCs may provide a niche to support the clonogenic growth of human myeloma without the need to invoke a viral infection (). DC-mediated enhancement of tumor clonogenicity may also involve other tumor types, such as lymphoma and breast cancer. Interestingly, a recent study of gene expression profiles of lymphoma has shown that the presence of DC signature in lymphoma portends an adverse outcome, which is consistent with these results (). However, the involved mechanisms, which are only studied for myeloma here, may differ between different tumor types. In our studies, blockade of the RANK–RANK-L pathway by OPG or blockade of BAFF–APRIL-mediated interactions via TACI-Fc led to the inhibition of DC-mediated enhancement of the clonogenicity of human myeloma. These data are consistent with prior studies showing the importance of both of these pathways in myeloma biology and point to DCs as an important source of these ligands (, , ). However, our data does not exclude the possibility that other molecules such as integrins or costimulatory molecules commonly expressed on DCs may also be important in DC–myeloma interactions. Additional mechanisms of the DC-mediated regulation of myeloma growth may include the potential contribution of tumor-associated DCs as precursors to new blood vessels () or osteoclasts (). Indeed, a recent study has shown that osteoclasts can also support the growth of myeloma cells in vitro (). The culture of U266 cells with DCs led to an increased proportion of Igλ cells lacking CD138, a marker of terminal plasma cell differentiation, as well as induction of BCL6 expression in tumor cells. Suppression of BCL6 is a critical feature of normal plasma cell differentiation. These data are reminiscent of the findings of a previous study that observed the reactivation of the B cell program after exogenous expression of BCL6 in myeloma cell lines (). Together, these data suggest that the differentiation state of myeloma cells is plastic and can be modified by cues provided by DCs in the tumor bed. To our knowledge, these data provide the first evidence that DCs can directly impact the clonogenic growth of human tumors. Therefore, the recruitment of DCs into tumors may impact not just the host immune response but also the biology of the tumor itself. In a prior study, we have shown that the effector function of tumor-infiltrating T cells correlates with favorable clinical features in MGUS versus myeloma (). Thus, the immune system may be a two-edged sword, with distinct components capable of both supporting and inhibiting tumor growth. Identification of tumor-infiltrating DCs as potential contributors to tumor progression also provides the rationale for specifically targeting this interaction as a novel approach for the therapy of human cancer. The human MM cell lines ARK (gift from J. Epstein, University of Arkansas, Little Rock, AR) and U266 (American Type Tissue Culture) were cultured in complete medium consisting of RPMI 1640 (Cellgro), 2 mM -glutamine, 20 μg/ml gentamicin sulfate, and 10% FBS. Other tumor cell lines used were NCEB1 (mantle cell lymphoma; gift of O. O'Connor, Memorial Sloan Kettering Cancer Center [MSKCC], New York, NY), MCF-7 (breast cancer; gift of P. Livingston, MSKCC), and U251 (glioma; gift of R. Puri, Food and Drug Administration, Bethesda, MD). Bone marrow and blood specimens from patients with myeloma and MGUS were obtained after informed consent that was approved by The Rockefeller University Institutional Review Board (IRB). Peripheral blood samples were obtained from healthy donors after informed consent as approved by The Rockefeller University IRB or were purchased from the New York Blood Center. PBMCs were isolated by density gradient centrifugation (Ficoll-Paque Plus; GE Healthcare). DCs were generated from purified blood monocytes as described previously (). In brief, monocytes isolated using CD14 microbeads (Miltenyi Biotec) were cultured in the presence of 20 ng/ml GM-CSF (Immunex) and 10 ng/ml IL-4 (R&D Systems). DCs were used on days 5–6 of culture. Clonogenic growth of tumor cell lines was evaluated by plating tumor cells (50,000 cells/ml) in methylcellulose containing 5% leukocyte-conditioned media (Methocult; Stem Cell Technologies, Inc.) using a protocol modified from a previous study (). Cells were plated in 35-mm tissue culture dishes in quadruplicates and incubated at 37°C and 5% CO. Colonies consisting of >40 cells were counted by microscopy 2–3 wk after plating. To assess the impact of monocytes/DCs on tumor clonogenicity, tumor cells were mixed with purified CD14 monocytes or monocyte-derived DCs (Mo-DCs) at varying ratios before plating in Methocult. Tumor growth was monitored weekly. For cell contact–dependent assays, DCs were suspended in 2% IMDM and were separated from U266 cells by a Transwell insert. Control inserts had 2% IMDM only. For some experiments, tumor cells and DCs were cultured in the presence or absence of either 1 μg/ml TACI-Fc (R&D Systems) or 0.5 μg/ml OPG (R&D Systems). CD28-Fc protein (R&D Systems) was used as a control. Bone marrow MNCs were isolated from marrow samples using density gradient centrifugation. For clonogenic assays on primary tumor cells from patients, CD138 and CD138 fractions were isolated from bone marrow MNCs using CD138 microbeads (Miltenyi Biotec) and AutoMACS (Miltenyi Biotec). The CD138 fraction was further depleted of normal hematopoietic progenitors using CD34 microbeads (Miltenyi Biotec). The resulting fractions, CD138, and CD138CD34 cells (5 × 10/ml) were plated with or without Mo-DCs at a ratio of 1:2 in Methocult as described above for cell lines. Tumor colonies were scored at 2 wk of culture. The phenotype of tumor cells was confirmed by immunofluorescence microscopy. For replating assays, cells were harvested from the clonogenic assays, washed, and replated at original cell concentration with or without DCs at a tumor/DC ratio of 1:2. Colonies were scored after 2 wk of culture. Tumor colonies harvested from clonogenic assays were analyzed for the cell surface expression of CD138-PE, CD11c-APC, and intracellular κ or λ light chain–FITC (BD Biosciences) by flow cytometry. Cytospins were made on the poly-lysine–coated (Sigma-Aldrich) multiwell slides (Carlson Scientific). Cells were fixed with acetone and stained with primary and secondary antibodies for 30 min at room temperature. Primary antibodies CD138 (PE), Igλ, and Igκ (BD Biosciences) and the secondary antibody AlexaFluor488 goat anti–mouse IgG1 (Invitrogen) were used at 1:30 and 1:200 dilutions, respectively. Acetone-fixed cytospins of tumor cells were also stained for BCL6 mAb (clone P1F6+PG-B6p; Lab Vision Corp.) followed by AlexaFluor488 goat anti–mouse IgG1 (Invitrogen) and CD138 (PE) using the protocol described previously with few modifications (). Slides were evaluated using an epifluorescence microscope (AX70; Olympus) with a motorized stage to allow 0.5-mm optical sections imaged with a cooled CCD camera (C4742-95; Hamamatsu) and analyzed with MetaMorph software (Universal Imaging Corp.). RNA was extracted from cells by using the RNeasy Mini Kit (QIAGEN). BCL6 expression was quantified by using Assays-on-Demand primer probes from Applied Biosystems. RT-PCR was performed by using EZ PCR Core Reagents (Applied Biosystems) according to the manufacturer's instructions. A BCL6-expressing plasmid (provided by K. Calame, Columbia University, New York, NY) was used as a positive control. The samples were amplified and quantified on a sequence detection system (PRISM 7700; Applied Biosystems) by using the following thermal cycler conditions: 2 min at 50°C, 30 min at 60°C, 5 min at 95°C, and 40 cycles of 15 s at 95°C followed by 60 s at 60°C. , a housekeeping gene, was used to normalize each sample. The data were analyzed, and samples were quantified by the software provided with the Applied Biosystems PRISM 7700. Data from different experimental groups were compared using the Students' test, and significance was set at P < 0.05.

Diabetes mellitus is associated with slowly progressive changes in the brain []. Neuropsychological studies show that patients with type 2 diabetes mellitus have mild to moderate impairments in attention and executive functioning, information processing speed and memory (for reviews see [, ]). Patients with type 2 diabetes also show changes on brain magnetic resonance imaging (MRI), such as cortical and hippocampal atrophy [, ]. We have recently shown that cognitive dysfunction in patients with type 2 diabetes was associated with white matter lesions (WML), (silent) brain infarcts and to a lesser extent with atrophy []. The determinants of changes in cognition and abnormalities on brain MRI of patients with type 2 diabetes are uncertain []. Some studies report associations with hypertension [, , , ], but this was not supported by others [, , ]. Associations between impaired cognition and chronic hyperglycaemia have also been noted []. Studies on other diabetic complications may provide leads for potentially relevant determinants. Complications like nephropathy, retinopathy and neuropathy are thought to be due to hyperglycaemia-induced microangiopathy [, ], with additional involvement of hypertension and macrovascular disease [–]. Since atherosclerosis and hypertension are established risk factors for age-related cognitive decline and brain MRI changes in the general population [–], we hypothesised that the combined effects of atherosclerotic macrovascular disease, chronic hyperglycaemia and hypertension are involved in the development of cognitive impairments in patients with type 2 diabetes. The aim of the present study was to identify possible metabolic and vascular determinants of cognitive dysfunction and changes on brain MRI in patients with type 2 diabetes. Given the uncertainty about these determinants, an exploratory design was chosen. A detailed neuropsychological examination and brain MRI were obtained from a large cross-sectional sample of type 2 diabetes patients and related to different measures of glucose metabolism, vascular risk factors, microvascular complications and macrovascular disease. The Utrecht Diabetic Encephalopathy Study aims to identify determinants of cognitive impairment in patients with diabetes []. Therefore, patients were not selected for the presence or absence of diabetic complications, co-morbid conditions (e.g. hypertension) or exposure to other risk factors (e.g. smoking). For inclusion patients had to be 55 to 80 years of age, functionally independent and speakers of Dutch, with a minimal diabetes duration of 1 year. Exclusion criteria for all participants were: a psychiatric or neurological disorder that could influence cognitive functioning; a history of alcohol or substance abuse and dementia; and, for the control group, a fasting blood glucose ≥7.0 mmol/l []. Participants with a history of stroke who were still fully functionally independent were classified as eligible. To increase statistical power for within-group analyses in the type 2 diabetes group, twice as many patients as controls were enrolled. The neuropsychological examination tapped the major cognitive domains in verbal and non-verbal ways. Eleven tasks were administered in a fixed order, taking about 90 min to complete. These tasks were divided into five cognitive domains, as described previously []: attention and executive functioning; information processing speed; memory; abstract reasoning; and visuoconstruction. For analysis the test scores were standardised into scores for each of the five domains, based on the means of the whole group. The mean performance from each participant across the domains is expressed as the composite cognitive score. Premorbid IQ was assessed with the Dutch version of the National Adult Reading Test. To control for possible effects of mood disturbances or affective disorders a Beck depression inventory [] was performed. WML were rated according to the Scheltens scale [] with slight modifications []. Periventricular WML (PWML) were rated on a severity scale (0–2) at the frontal and occipital horns and the body of the lateral ventricle on both sides (sum score 0–12). For the rating of deep (subcortical) WML (DWML) the brain was divided into six regions: frontal, parietal, occipital, temporal, basal ganglia and infra-tentorial. Per region the size and number of WML were rated on a scale ranging from 0 to 6. The total score thus ranged from 0 to 36. Cortical atrophy was evaluated by the frontal interhemispheric fissure ratio and the Sylvian fissure ratio []. Subcortical atrophy was evaluated by the bifrontal ratio and by the bicaudate ratio []. These ratios were converted to scores: a cortical atrophy score (mean of frontal fissure ratio and Sylvian fissure ratio) and a subcortical atrophy score (mean of bicaudate ratio and bifrontal ratio). Blood was drawn by venepuncture to assess HbA, fasting glucose and insulin levels. Insulin resistance was estimated with the homeostasis model assessment of insulin resistance (HOMA-IR). The HOMA-IR is calculated as fasting glucose (mmol/l) × fasting insulin (mU/l)/22.5 []. Because insulin was expressed in pmol/l we used the formula fasting glucose (mmol/l) × fasting insulin (pmol/l)/(22.5 × 6.945) []. Following mydriasis with phenylephrine and tropicamide, single-field photographs were taken of both eyes with a 50-degree retinal camera (Zeiss FF 450, Carl Zeiss B.V., Sliedrecht, the Netherlands), centred on the macula. Retinopathy was rated on slides, according to the diabetic retinopathy severity scale (grades 1–7) as used in the Wisconsin Epidemiologic Study of Diabetic Retinopathy []. Photocoagulated eyes were rated at grade 5 or higher (severe non-proliferative diabetic retinopathy). Ratings were performed by two investigators, blinded to patient characteristics. In case of disagreement (2%), a third investigator was involved and a consensus was made. Retinopathy was defined as a grade of 1.5 or higher. Neuropathy was rated with the Toronto Clinical Neuropathy Scoring System [], with a slight modification. A sensory test for temperature was not performed, so that the maximum score was 18 points (severe polyneuropathy) instead of 19. A score of 0–5 indicated no neuropathy, 6–8 indicated mild neuropathy, 9–11 moderate neuropathy and ≥12 severe neuropathy. Neuropathy was defined as a score of ≥6. Several composite measures of macrovascular disease were defined. ‘Any peripheral arterial disease’ was defined as current complaints of intermittent claudication (assessed with the Rose questionnaire []) or a history of surgery or endovascular treatment for arterial disease of the legs or the abdominal aorta. ‘Ischaemic heart disease’ was defined as a history of myocardial infarction or surgery or endovascular treatment for coronary artery disease. ‘Any vascular event’ was defined as a history of myocardial infarction or stroke, or a history of operative or endovascular treatment for coronary, carotid or peripheral (legs, abdominal aorta) artery disease. Carotid intima-media thickness (CIMT) was measured in both common carotid arteries as described previously [] with an ATL Ultramark 9 (Advanced Technology Laboratories, Bothell, WA, USA) equipped with a 10-MHz linear-array transducer. Scanning was performed at three different longitudinal projections (anterior-oblique, lateral and posterior-oblique). The CIMT was measured in a 1 cm section proximal to the beginning of the dilatation of the carotid bulb in all three projections, in both carotid arteries. CIMT was calculated as the average of these six measurements. CIMT readings were not available in six type 2 diabetes patients and one person in the control group. The differences between patients and the control group were examined with test for means, Mann–Whitney was used for non-parametric data and χ test for proportions. In the text and tables, data are shown as mean ± SD or proportions, unless stated otherwise. Within the type 2 diabetes population, cognition (five domains) and brain MRI findings (cortical and subcortical atrophy scores, PWML, DMWL and infarcts) were related to the different measures of glucose, insulin and lipid metabolism, and to microvascular complications and macrovascular disease by linear or logistic regression analyses, adjusting for age, sex and estimated IQ. In order to limit the number of analyses the ‘composite cognitive score’ was used as the primary cognitive outcome measure in the regression analyses. For significant associations, post hoc tests were performed per domain. Secondary analyses were performed with information processing speed, the domain most markedly affected by type 2 diabetes. The results were essentially the same as for the composite cognitive score (data not shown). In the regression analyses, B values >0 indicate that a variable is associated with more severe MR abnormalities; for cognition B values <0 indicate that a variable is associated with more pronounced performance impairments. For the between and within-group analyses,  < 0.05 was considered statistically significant. All variables that reached a significance level of  ≤ 0.1 in the adjusted univariate risk factor analyses were included in a multivariate model that also included age, sex and estimated IQ. The age, sex, level of education and estimated IQ in the groups were comparable (Table ). Detailed neuropsychological and MRI data have been reported previously []. In short, performance of patients with type 2 diabetes was worse than that of the control group across all five cognitive domains, with statistically significant differences on attention and executive functioning (difference mean scores 0.23 [95% CI 0.03, 0.43];  = 0.02), information processing speed (0.40 [0.17, 0.63];  = 0.001) and memory (0.20 [0.05, 0.36];  = 0.01). Patients with type 2 diabetes had more pronounced cortical atrophy (difference mean scores 0.62 [95% CI 0.33, 0.91];  < 0.001) and subcortical atrophy (0.38 [0.07, 0.68];  = 0.01). They also had more severe DWML (controls, median [range]: 5 [0, ]; type 2 diabetes: 7 [0.5, 27.5];  = 0.02), but PWML severity in the two groups was similar (control: 6 [, ]; type 2 diabetes: 6 [, ];  = 0.13). Patients with type 2 diabetes also had more (silent) cerebral infarcts than controls (type 2 diabetes 22/113, control 4/54;  = 0.06). HbA, fasting glucose and insulin levels were higher (all  < 0.01) in patients with type 2 diabetes than in the control group. BMI was similar in both groups. Only a small proportion (6%) of type 2 diabetes patients had ever experienced a severe hypoglycaemic event (Table ). In the regression analyses within the type 2 diabetes group, HbA levels were significantly related to cognition (composite score: B [per % HbA]: −0.07 [−0.14, 0]  = 0.047; post hoc per domain: information processing speed: B [per % HbA]: −0.15 [95% CI: −0.27, −0.2],  = 0.02; abstract reasoning: B: −0.15 [−0.29, −0.01],  = 0.04). Elevated fasting insulin levels and HOMA-IR were related to increased DWML severity (B [per 10 pmol/l insulin]: 0.14 [0.04, 0.26],  = 0.009; B [HOMA-IR]: 0.21 [0.04, 0.39],  = 0.02). Table  shows that patients with type 2 diabetes had higher systolic blood pressure ( < 0.01) and pulse pressure ( < 0.05) than controls. They also had hypertension more often ( < 0.01). Total cholesterol was lower in the type 2 diabetes group ( < 0.01), but the proportion of individuals taking lipid-lowering drugs was higher in that group ( < 0.01). There were no statistically significant differences between type 2 diabetic patients and the control group in the proportion of participants who smoked or had dyslipidaemia (Table ). In the regression analyses within the type 2 diabetes group there were no statistically significant associations with the composite cognitive score. Non-significant trends ( ≤ 0.10) were observed for associations between both hypertension and current smoking and impaired cognitive performance (hypertension B: −0.19 [−0.38, 0],  = 0.053; smoking B: −0.21 [−0.43, 0],  = 0.051), and between the use of lipid-lowering drugs and better performance (B: 0.15 [−0.03, 0.32],  = 0.10). Reanalysis with cut-off values for hypertension of 140/90 mmHg made the association with the composite cognitive score less strong. Mean arterial pressure was associated with more severe PWML (B [per 10 mmHg]: 0.28 [0.03, 0.53],  = 0.03). The use of lipid-lowering drugs (statins in all but one patient) was associated with less severe MRI abnormalities (PWML: B: −0.68 [−1.25, −0.12],  = 0.02; cortical atrophy: B: −0.36 [−0.69, −0.03],  = 0.03). These effects were not affected by additional adjustment for the actual cholesterol levels (Table ). In 20 patients with type 2 diabetes and 8 control individuals it was not possible to perform retinal photographs. Mostly due to logistical reasons, overnight urine samples could not be obtained from 21 diabetic patients and 13 controls. Patients with type 2 diabetes had more retinopathy and neuropathy than the control group (both  < 0.01). Although albuminuria was more common in the type 2 diabetic than in the control group, this difference was not statistically significant (Table ). In the regression analyses within the type 2 diabetes group, there were no statistically significant associations with the composite cognitive score. Retinopathy was associated with more pronounced cortical atrophy (B: 0.48 [0.11, 0.85],  = 0.01) (Table ). Patients with type 2 diabetes were more likely to have had intermittent claudication ( < 0.01) or a history of ischaemic heart disease ( < 0.01). There was no difference between the two groups in the CIMT (Table ). In the regression analyses within the type 2 diabetes group, a history of ‘any vascular event’ and the presence of brain infarcts on MRI were associated with an impaired composite cognitive score as follows: vascular event: composite score B: −0.25 (−0.44, −0.05),  = 0.01; post hoc per domain: information processing speed B: −0.46 (−0.80, −0.12),  = 0.008; and memory B: −0.23 (−0.41, −0.06),  = 0.01; infarct on MRI: composite score B: −0.28 (−0.50, −0.06),  = 0.01; post hoc per domain: information processing speed B: −0.77 (−1.14, −0.39),  < 0.001; and abstract reasoning B: −0.41 (−0.82, 0.01),  = 0.06. A history of ‘any vascular event’ was also associated with more pronounced DWML (B: 2.0 [0, 4.1],  = 0.05) and with an increased occurrence of infarcts on MRI (odds ratio: 2.9 [1.1, 7.9],  = 0.04). Patients with a (silent) infarct on MRI tended to have more pronounced PWML (B: 0.7 [0, 1.4],  = 0.06) and cortical atrophy (B: 0.51 [0.10, 0.92],  = 0.02) relative to type 2 diabetes patients without infarcts on MRI (Table ). Exclusion of patients with a history of stroke attenuated the association between ‘any vascular event’ and the cognitive score, but not that between ‘any vascular event’ and DWML. For the majority of risk factors the values of the standardised regression coefficients ß were similar in the single risk factor and multivariate models, indicating that interaction between factors was limited. Values for ß for individual factors varied between 0.15 and 0.30, indicative of modest associations (Table ). In the multivariate model, hypertension and a history of vascular events were associated with worse cognitive performance and statin use with better performance. Retinopathy and brain infarction on MRI were associated with more severe cortical atrophy and statin use with less atrophy. A higher insulin level was associated with more DWML, brain infarction on MRI with more PWML and statin use with less PWML. Overall, macrovascular disease (history of macrovascular events or infarct on MRI) were most consistently associated with the different outcome measures (Table ). Patients with type 2 diabetes had more cortical and subcortical atrophy and more DWML than control participants and their overall performance in the five cognitive domains was worse. As expected, patients with type 2 diabetes had more microvascular complications, more macrovascular (atherosclerotic) disease and more hypertension than the control group. In multivariate regression analyses within the type 2 diabetes group, hypertension and a history of vascular events were associated with worse cognitive performance, while statin use was associated with better performance. Retinopathy and brain infarcts on MRI were associated with more severe cortical atrophy and statin use with less atrophy. Insulin level and brain infarcts were associated with more severe WML and statin use with less severe WML. Cognitive function in patients with type 2 diabetes has been studied extensively (for reviews see [, ]). Performance in the domains verbal memory and information processing speed, and probably also executive functioning and non-verbal memory, is moderately impaired. Our results are in keeping with these findings. Thus far, relatively few studies have specifically addressed brain MRI abnormalities in patients with type 2 diabetes. In agreement with our observations, modest cortical and subcortical atrophy and symptomatic or asymptomatic infarcts have been found more often in type 2 diabetes patients than in control individuals [, , ]. Results of previous studies on the association between type 2 diabetes and WMLs are less consistent []. This might be due to the study populations involved and the use of relatively insensitive WML rating scales []. Chronic hyperglycaemia might be a determinant of cerebral changes in patients with type 2 diabetes. In the present study, HbA levels were related to the composite cognitive score, but only in de univariate analysis. Moreover, retinopathy, which is generally considered to be a consequence of chronic exposure to hyperglycaemia [], was related to cortical atrophy. Previous studies on cognition in patients with type 2 diabetes have also reported an association with HbA levels [, , ]. The relation with fasting blood glucose or duration of diabetes is, however, inconsistent [, ]. No previous studies have provided detailed data on the association between glycaemic control and MRI changes in type 2 diabetes. Studies in type 1 diabetes mellitus, however, have shown an association between diabetic retinopathy (as a proxy of chronic hyperglycaemia) and both brain atrophy [, ] and cognitive functioning []. There are no previous studies on the relation between insulin levels and cerebral complications in type 2 diabetes. The association with WML severity, observed by us in the present study, is of particular interest in the light of recent studies in the general population, which link insulin to vascular abnormalities and degenerative changes in the brain [, ]. Previous studies in the general population indicate that risk factors for vascular disease, such as hypertension, dyslipidaemia, increased BMI and smoking, are associated with an increased risk of cognitive decline and dementia and with brain MRI changes, including WML (e.g. [–]). Previous studies on the modulating effect of hypertension on cognitive function in type 2 diabetes show conflicting results [, , , ]. In the present study, hypertension was related with impaired cognitive performance and mean arterial pressure with PWML severity. To our knowledge, the relation between other vascular risk factors and both cognition and brain MRI in patients with type 2 diabetes has not been examined previously. The reverse association between the use of statins and both cognition and MRI findings is intriguing. Nevertheless, this observation cannot be regarded as proof of a possible treatment effect. It should be noted that the association between statin use and both cognition and age-related brain MRI changes in the general population is still being debated []. The present findings will need to be confirmed by further studies. Macrovascular atherosclerotic disease appeared to be the most consistent determinant of impaired cognition and brain MRI abnormalities in the type 2 diabetes patients in the present study. We have not found any previous studies that presented detailed data on the relation between macrovascular disease and cerebral changes in people with type 2 diabetes. In the general population, however, several studies have shown that macrovascular atherosclerotic disease is associated with age-related cognitive impairment and changes in brain MRI. In a large cross-sectional study, for example, previous vascular events, presence of plaques in the carotid arteries and presence of peripheral arterial atherosclerotic disease were negatively associated with cognitive performance []. In another study, the association between the number of cardiovascular disease conditions and cognitive impairment appeared to show a ‘dose–response’ relationship []. With regard to brain MRI changes, a history of stroke or myocardial infarction has been associated with the presence of WML [] and plaques in the carotid artery with PWML [, ]. The strength of our study is that we combined detailed data on cognitive function and brain MRI with detailed data on metabolic and vascular risk factor clusters, thus allowing an accurate assessment of the relation between these factors. Possible limitations include patient selection, the cross-sectional design and the large number of explanatory variables addressed. With regard to patient selection, we aimed to obtain a representative sample of functionally independent patients with type 2 diabetes from the general population. Although the rather demanding testing protocol may have deterred patients with relatively severe mental or physical limitations, the prevalence of microvascular and macrovascular disease, hypertension and smoking habits, as well as the level of metabolic control in our study sample is comparable with those found in other population-based studies in the Netherlands [–]. To minimise the effects of lifestyle and socioeconomic factors, control participants were recruited from the direct environment of the type 2 diabetic patients. Consently, the prevalence of risk factors such as hypertension and high BMI was higher than would be expected in the general population in the Netherlands. If anything, this would have decreased the differences in cognition and MRI ratings between the groups. The cross-sectional design of our study precludes inferences about causal relationships. Moreover, the cognitive and imaging outcome measures were probably influenced by a large number of factors, some of which are specific to type 2 diabetes mellitus (e.g. chronic hyperglycaemia, diabetes treatment) and some not (e.g. age, hypertension, atherosclerosis). Our exploratory analysis included a large number of explanatory variables, which has certain drawbacks. First, different explanatory variables might be interrelated. The relatively small regression coefficients and effect sizes affect the evaluation of these interrelations and limit statistical power. This may also explain why some of the variables that reached statistical significance in the univariate analyses dropped out of the multivariate model. Nevertheless, the multivariate analysis as presented in Table  does indicate which variables were the strongest independent determinants of cognition and MRI abnormalities in the model used. The second drawback is that the large number of regression analyses can lead to type I errors. Nevertheless, we feel that this first detailed study of cognition and brain MRI in type 2 diabetes patients in relation to metabolic and vascular risk factors does provide important leads that could be further evaluated in future studies. Such studies should: preferably have a longitudinal design; include assessment of cognition and brain MRI in relation to chronic hyperglycaemia and atherosclerotic vascular disease; and allow the assessment of potential confounders (e.g. hypertension). Type 2 diabetes is associated with modest impairments in cognition, as well as with atrophy and vascular lesions on MRI. This ‘diabetic encephalopathy’ is a multifactorial condition, for which atherosclerotic (macroangiopathic) vascular disease is an important determinant. Chronic hyperglycaemia, hypertension and hyperinsulinaemia may play additional roles. p

To address this and other confusion that have arisen in recent classifications, a special interest subgroup meeting was held at the 2003 meeting of the American Society for Cell Biology. As a result of this meeting and other feedback from kinesin researchers worldwide (available for review as an “Archived Web Discussion” at ), we determined that a single nomenclature must be adopted to facilitate communication among researchers. Here, we propose a standardized nomenclature for all kinesin families (). To minimize confusion, the names of individual sequences will remain unchanged. The formal kinesin nomenclature is shown in and was constructed using the following logic and rules: This was the name given to the first described superfamily member (from the Greek “kinein,” to move; ; ). Its use will avoid past confusion encountered by having families called “kin” (which can be confused with kinase), “kif” (an acronym for “kinesin superfamily”), and KLP (an acronym for kinesin-like protein). This avoids the problems with using Roman numerals for database searches, and helps us to steer clear of the structural and functional misinterpretations that could continue to arise if letters of the alphabet (e.g., C, N, I, and M) were to be used. For example, the Kinesin-14 family is made up of two large subfamilies, designated Kinesin-14A and Kinesin-14B (previously referred to as C-I and C-II, respectively; ). Examples include Kinesin-1 (the first kinesin discovered was a member of this family), Kinesin-2 (the holoenzyme for these family members has been referred to as “kinesin II” in past publications; ), and Kinesin-4 (named for mouse KIF4, the family's founding member; ). The numbers associated with Hirokawa's class designations have been retained for Kinesin-1, -3, -6, -7, and -8. This number was derived by determining the consensus monophyletic groups conserved among past phylogenetic analyses (; ; ; ; ; ) in conjunction with examining phylogenetic trees generated by S.C.D. using a dataset that includes many protistan kinesin sequences (manuscript in preparation) and a phylogenetic tree generated by H. Miki, Y. Okada, and N. Hirokawa using a dataset made up of 608 publicly available kinesin sequences (which can be viewed as a part of the “Archived Web Discussion” at ). It is important to note that the naming system outlined here is based primarily on molecular systematic analysis, as are the accepted systems for other cytoskeletal gene families (e.g., the myosin and actin gene families; ; ; ). This criterion prevents classification of small groups of sequences from individual species or closely related groups of organisms as families or subfamilies, and it will help keep the number of recognized families small enough to be manageable. Sequences not grouping consistently within a family (based on measures of phylogenetic consistency such as high bootstrap support) will be called “orphan kinesins” (). It should be noted that in order for new groups of kinesins to become recognized as a new kinesin family, the group of sequences must conform to all rules set forth herein. In practice, there are two acceptable ways to identify the family to which a sequence belongs. The first uses the formal name of the appropriate family followed by one of the family's former names in parentheses (and a reference to an appropriate publication). The second uses only the formal name. Examples of acceptable usage follow. Human Eg5, a member of the Kinesin-5 family (previously referred to as BimC by ), is involved in establishing the bipolar spindle. Human Eg5, a member of the Kinesin-5 family, is involved in establishing the bipolar spindle. We recommend that such a statement be included within the introduction of any publication. However, it is not necessary to call the sequence itself by the name of the family to which it belongs: individual sequence names will remain unchanged to reduce needless confusion. ext-link #text

Establishing specific gene expression programmes during cell differentiation requires controlled activation and silencing of large numbers of genes. Once these patterns of expression or repression have been set up, they must be kept in place to maintain the identity of the differentiating cell. Chromatin is known to play a key role in these processes. It can exert its effects both at the local level, with single nucleosomes modulating access of factors to promoters and enhancers, and on a more global level through large-scale condensation of megabase regions or even entire chromosomes (; ). Chromatin-mediated silencing of gene expression is strongly influenced by epigenetic modifications of the core histones, which include methylation and acetylation of lysine residues and phosphorylation of serine residues. These modifications can directly affect chromatin packaging as well as influence the binding of chromatin proteins and transcription factors (reviewed in ). The presence of epigenetic modifications can alter the structure of gene expression domains that include the gene and surrounding sequences, and when these modifications extend over larger regions, they facilitate packaging of the DNA into visibly condensed heterochromatin. Constitutive heterochromatin, which is present throughout the cell cycle, is relatively gene poor and enriched for repetitive satellite sequences. Facultative heterochromatin is quite different in that it is the result of a developmentally regulated condensation of gene-rich euchromatic regions, which are decondensed and transcriptionally active in some cell types. Some of the most dramatic examples of facultative heterochromatin are observed during the final stages of cell differentiation. Terminally differentiated cells that exhibit this type of widespread heterochromatinisation include rod photoreceptor cells (), neutrophils (), nucleated erythrocytes () and plasma cells (). It seems likely that formation of facultative heterochromatin during cell differentiation involves at least some of the silencing mechanisms that affect gene expression locally. This suggests that there may be a point during cell differentiation when these local gene-silencing events reach a critical mass that is sufficient to give rise to visibly condensed facultative heterochromatin. Methylation of histone H3 lysine 9 (K9) has been shown to be involved in epigenetic silencing of euchromatic genes and also marks the constitutive heterochromatin that is found close to centromeres (; ; ; ). An important function of methylated H3K9 groups is to act as a recognition site for heterochromatin protein 1 (HP1), which binds to the methylated lysine via the chromodomain region (). HP1 was originally identified as a modifier of position effect variegation in and has been shown to associate strongly with constitutive heterochromatin. There is also evidence that it is recruited directly to some gene promoters where it participates in transcriptional silencing (; ; ). Three HP1 proteins have been identified in mammals (α, β and γ). HP1α is mainly associated with constitutive heterochromatin, HP1β is present both on pericentric heterochromatin and euchromatin, whereas HP1γ is predominantly euchromatic (). Binding of HP1 to H3K9 has been shown to be affected by the presence of a phosphate group at serine 10 (S10ph). Phosphorylation of H3 S10 during G2/M has been found to prevent HP1β from binding to the adjacent H3K9me residue. As a result, HP1β is released from chromatin at the onset of mitosis (; ). The kinase responsible for S10 phosphorylation is the Aurora B kinase, a component of the chromosomal passenger complex, which modulates chromosome structure and segregation at mitosis by promoting HP1 displacement from the chromosomes, and chromosome alignment and attachment to the microtubules of the mitotic spindle (). The other subunits of the complex, INCENP, Survivin and Borealin, are non-enzymatic and are involved in regulating and targeting Aurora B to its substrates. In non-transformed cells, Aurora B has, until now, been considered to be highly cell-cycle regulated and to be involved primarily in protein phosphorylation during mitosis. The double histone H3 tri-methylated K9/phosphorylated S10 (H3K9me3/S10ph) modification generated by Aurora B is known to be widely distributed on mitotic chromosomes, and inhibition of S10 phosphorylation has been shown to interfere with chromosome condensation during mitosis (; ; ). The double modification has been proposed to be a marker of M phase (), but it has not been shown previously to be involved in modulating chromatin structure outside mitosis. Here we show that in addition to its functions during mitosis, the Aurora B kinase mediates formation of the double H3K9me3/S10ph modification independently of the cell cycle during cell differentiation. In terminally differentiated postmitotic plasma cells, this results in displacement of HP1β from facultative heterochromatin. We also use microarray analysis to demonstrate the presence of domains of H3K9me3/S10ph at silent genes in differentiated cells. Our results suggest that binary modifications can play an important role in modulating the effects of specific histone modifications at different stages of development. To identify epigenetic markers that are involved in long-term silencing of gene expression during cell differentiation, we initially screened facultative heterochromatin in terminally differentiated bone marrow plasma cells for the presence of candidate marks that might play a role in silencing. The results of this analysis revealed an unexpected correlation between the presence of visible heterochromatin and the binary K9me3/S10ph modification on histone H3. The double modification was detected by immunofluorescence (IF) using an antibody that specifically recognises these modifications when they are present in combination on the same histone H3 molecule (). The specificity of the antibody was confirmed by western blotting (), peptide ELISA and peptide competition, which showed that it did not crossreact with tri-methyl K9 (K9me3) alone, phospho-S10 (S10ph) alone, di-methyl K9/phospho-S10 (K9me2/S10ph) or tri-methyl K27/phospho-S28 (K27me3/S28ph) (). Analysis by IF using this antibody showed strong staining of facultative heterochromatin in 100% of bone marrow plasma cells. A particularly striking aspect of this result is the fact that plasma cells are postmitotic, whereas the double modification had previously been thought to be associated with the G2/M phase of the cell cycle and displacement of HP1 from mitotic chromosomes. This finding led us to perform a detailed analysis of the role of the double H3K9me3/S10ph modification in heterochromatin formation and epigenetic marking of silent genes during cell commitment and differentiation. The model systems that were used for the analysis were differentiation of mesenchymal stem cells and terminal differentiation of B cells into plasma cells (). The multipotent mesenchymal stem cell line C3H10T1/2 can be induced to differentiate into osteocytes, chondrocytes and adipocytes and has been used extensively as an model for mesenchymal stem cell differentiation (). In this study, the cells were differentiated into osteocytes by treatment with thyroxine and dexamethasone (). Plasma cells were generated by differentiation of primary splenic B cells (see Materials and methods and ). Terminal differentiation of B cells into plasma cells is accompanied by widespread gene silencing and formation of facultative heterochromatin as the cell becomes specialised for the production and secretion of large amounts of antibody (). The differentiation system () offered the advantage of providing enough plasma cells for chromatin analysis. Because of the known link between the presence of the double modification and mitosis, the cell division status of the cells was determined before and after differentiation by measurement of bromodeoxyuridine (BrdU) incorporation (). The activated B cells were clearly cycling (75% BrdU), whereas the plasma cells were almost entirely postmitotic (3% BrdU). The undifferentiated C3H10T1/2 mesenchymal stem cells were also actively cycling (89% BrdU) and became largely postmitotic after differentiation (8% BrdU). The mesenchymal stem-cell and B-cell differentiation systems were used to analyse the distribution of the double modification before and after cell differentiation and exit from the cell cycle. Immunostaining was carried out using the anti-H3K9me3/S10ph antibody. Staining of the undifferentiated mesenchymal stem cells showed a strong punctate staining for the double modification in 23% of the cells. The punctuate staining coincided with the DAPI-dense pericentromeric heterochromatin (). The larger size of these cells identified them as being in G2 and this was confirmed by FACS analysis of the level of the double modification in cells that were gated according to DNA content (). The results of the immunostaining and FACS are in agreement with the observation by of a global increase in H3K9me3/S10ph that coincides with the onset of G2/M. The staining was much lower in the remaining cells and many of the cells showed little signal. The variation in the signal that was observed in the dull cells is likely to reflect progression through the different stages of the cell cycle. A quite different result was obtained when the cells were induced to differentiate along the osteogenic pathway. The differentiated postmitotic cells showed staining in the nuclei of all the cells analysed (). The staining in the differentiated cells was found at the pericentromeric heterochromatin and also as a more diffuse signal that appeared to extend over much of the rest of the chromatin. In the activated B-cell cultures, 27% of the cells showed a strong punctate staining for the double modification. Once again, the larger size of the positive cells and the FACS analysis identified them as being in G2 (). A smaller proportion (16%) showed an intermediate staining of the pericentromeric heterochromatin, but the majority of the cells (56%) were negative for the double modification. The differentiated plasma cells showed a very strong nuclear staining in approximately 60% of the cells, despite the fact that these cells were clearly postmitotic. In the positively staining cells, the signal for H3K9me3/S10ph was particularly intense on the large masses of DAPI-dense facultative heterochromatin, but was also present as a more diffuse staining on the rest of the chromatin (). The observation that around 40% of the cells were negative for the double modification could be related to the fact that these are short-lived plasma cells in which a proportion of the cells are likely to have chromatin that is already undergoing pre-apoptotic changes. It should be noted that bone marrow plasma cells, which have been shown to be long-lived (), showed a strong staining for the double modification in all the cells analysed (). The staining pattern for H3K9me3/S10ph was compared with the pattern obtained using an antibody that was specific for H3K9me3 alone (). There was no difference in the level and distribution of the single modification between undifferentiated and differentiated mesenchymal stem cells. The differentiated plasma cells gave a lower H3K9me3 signal compared to activated B cells. This may reflect the very high levels of the double H3K9me3/S10ph modification that were observed in plasma cells. Analysis of the double and single modifications by western blotting gave results that broadly supported the conclusions from the immunostaining. In particular, the plasma cells showed a much higher global level of the double modification than activated B cells (). The modification was completely absent from resting B cells. Significant levels of the double modification were also present in differentiated postmitotic C3H10T1/2 cells (). The levels were somewhat lower than the level in undifferentiated cells, but this was likely to reflect the fact that these cells were cycling and therefore included significant numbers of G2/M cells. The levels of the single H3K9me3 and H3S10ph modifications detected by western blotting showed only small variations between differentiated and undifferentiated cells (). The lower level of the H3K9me3 single modification that was observed by immunostaining of plasma cells compared with the level measured by western blotting ( and ) could be due to reduced accessibility of the modification to intracellular antibody staining, suggesting that it may be located within the heterochromatin mass. We next tested whether the histone H3K9me3/S10ph mark in plasma cells and differentiated mesenchymal stem cells correlates with the presence of the Aurora B kinase, which is known to be responsible for generating the phospho-S10 moiety of the double modification during mitosis (). Immunostaining of postmitotic plasma cells showed intense staining for Aurora B, which coincided with the large DAPI-dense regions of facultative heterochromatin (). More diffuse staining was also observed on the rest of the chromatin. This pattern correlates closely with the distribution of the H3K9me3/S10ph staining (). The proportion of plasma cells that stained positively for Aurora B (54%) was similar to the proportion that was positive for the double modification. Aurora B was also present in 100% of differentiated mesenchymal cells where it was concentrated into dot-like nuclear foci that did not coincide with constitutive pericentromeric heterochromatin (). Activated B cells and undifferentiated mesenchymal stem cells both gave a pattern of Aurora B staining that was similar to the H3K9me3/S10ph distribution with staining of the pericentromeric heterochromatin observed in the G2 cells and largely absent from G1 cells (). This distribution was confirmed by FACS analysis of the levels of Aurora B in G1 and G2/M cells (). Western blotting analysis also showed that Aurora B is present in postmitotic plasma cells and differentiated mesenchymal cells (). The analysis of undifferentiated mesenchymal stem cells was also carried out on cells that were sorted into G1 and G2/M populations and showed that the level of Aurora B is higher in differentiated cells than in undifferentiated G1 cells (). These results indicate that Aurora B persists in differentiated postmitotic cells and provide support for the idea that it is involved in the epigenetic regulation of cell differentiation. To directly test whether Aurora B is responsible for generating the double modification during differentiation, -differentiated plasma cells were treated with hesperadin, which specifically inhibits Aurora B activity (). Inhibition of the kinase resulted in a drastic reduction in the amount of H3K9me3/S10ph modification that was observed by western blotting (). This leads us to conclude that phosphorylation of H3 S10 by Aurora B is responsible for increasing the level of double H3K9me3/S10ph modification in postmitotic plasma cells. The presence of phosphorylated S10 has the potential to affect binding of proteins that recognise the adjacent methylated K9. One possible candidate for such an effect is HP1. Phosphorylation of S10 has been shown to modulate the binding of HP1 proteins to methylated histone H3K9, although different effects have been reported and there have been conflicting results for binding of different HP1 family members to di- and tri-methyl H3K9 (; ). We used surface plasmon resonance (SPR) to examine the effect of S10 phosphorylation on the binding of HP1β to di- and tri-methyl H3K9. The SPR assay made use of a recombinant 6 × His-HP1β protein and H3 peptides containing covalent modifications. Phosphorylation of S10 markedly decreased the interaction of HP1β with both K9me3 and K9me2 peptides (). This result supports the idea that S10 phosphorylation forms part of a binary switch that regulates binding of HP1β to chromatin. We next set out to investigate whether the presence of the double H3K9me3/S10ph modification affects the binding of HP1β to chromatin during cell differentiation. Immunostaining was used to determine whether HP1β is associated with the facultative heterochromatin of plasma cells (). In activated B cells, HP1β showed the expected localisation to pericentromeric heterochromatin clusters. However, when the immunostaining was carried out on -differentiated plasma cells, HP1β was depleted from the heterochromatic regions and was present more abundantly in regions of the nucleus that stained less brightly with DAPI. The cytoplasmic HP1β staining in plasma cells was mainly due to crossreactivity of the secondary anti-rat antibody with mouse immunoglobulins (Igs) (). Western blot analysis showed that HP1β is still present in plasma cells, thereby excluding the possibility that the absence of heterochromatin staining is due to loss of HP1β from the cell (). To test whether the results obtained with the -differentiated cells reflected the distribution of HP1 in plasma cells , we assessed HP1β localisation in the nuclei of short- and long-lived plasma cells isolated . The short-lived plasma cells were isolated from spleens of LPS-immunised mice and the long-lived plasma cells were obtained from the bone marrow of mice that had been immunised with ovalbumin (see Materials and methods). Interestingly, the two types of cells gave different results. The short-lived splenic plasma cells had a similar HP1β distribution to the -differentiated cells, whereas in the long-lived bone marrow cells, the localisation of HP1β to the heterochromatin was restored (). Analysis of the subnuclear localisation of HP1β in differentiated C3H10T1/2 cells showed that it is not visibly delocalised from heterochromatin (data not shown). This can be explained by the lower levels of the double H3K9me3/S10ph modification in these cells compared with plasma cells. To establish whether Aurora B activity is responsible for HP1β release from heterochromatin in short-lived plasma cells, we analysed the effect of the Aurora B inhibitor hesperadin on HP1β distribution in plasma cell nuclei. Incubation of -differentiated plasma cells with hesperadin for 1.5 h resulted in relocalisation of HP1β to the heterochromatic foci in 29% of the cells (), compared with 1% of untreated cells. This result demonstrates that Aurora B is involved in modulating the interaction of HP1β with chromatin outside mitosis. The comparison between different types of plasma cells also suggests that the binding of HP1β to heterochromatin is part of a complex phenomenon, with other factors coming into play during the transition from short- to long-lived plasma cells. The association of the double H3K9me3/S10ph modification with facultative heterochromatin in plasma cells suggests that it could be an epigenetic marker for gene silencing. To address this question directly, a custom oligonucleotide tiling microarray was used to analyse the modification at the level of individual genes and to relate it to transcriptional activity. The microarray covered a 2-Mb gene-rich region of mouse chromosome 3, with an average resolution of 100 bp (). The 67 genes in the region have a variety of expression patterns ranging from ubiquitous to highly tissue specific. Expression of all of the genes was measured in C3H10T1/2 cells before and after differentiation by quantitative real-time PCR (data not shown). Chromatin from undifferentiated and differentiated C3H10T1/2 cells was precipitated using the α-H3K9me3/S10ph and α-H3K9me3 antibodies and the DNA was hybridised to the microarray. Inspection of the pattern obtained for H3K9me3/S10ph revealed a strong enrichment for the double modification on repressed genes in differentiated cells (defined by average positive log IP/input ratio) (, yellow shadowed domains). In contrast, a generally negative log ratio for K9me3/S10ph was found at transcriptionally active genes (purple shadowed domains). A striking feature of the pattern of enrichment of the double modification was that it appears to define domains that can extend across several adjacent repressed genes ( and ). The pattern is dependent on differentiation, as enrichment for H3K9me3/S10ph on repressed genes was not observed in undifferentiated C3H10T1/2 cells ( and , light-blue trace). In contrast, the highest levels of the single H3K9me3 modification are observed in undifferentiated cells, with peaks of enrichment present on active and silent genes (, light-green trace). When the cells were differentiated, it was no longer possible to detect significant levels of the single modification (, dark-green trace) and this was mirrored by the increase in the level of the double modification. This result suggests that H3K9me3 becomes widely modified by the presence of S10ph during cell differentiation, with the double H3K9me3/S10ph modification forming domains across silent genes. Conventional chromatin immunoprecipitation (ChIP) using quantitative real-time PCR was also carried out on activated B and plasma cells and was compared with the results from undifferentiated and differentiated mesenchymal cells. The results show increased levels of the double modification at silent genes in plasma cells compared with activated B cells (). Taken together, the data from the microarray and the PCR–ChIP analysis indicate that the H3K9me3/S10ph modification becomes progressively enriched at repressed genes as cells differentiate and suggest that it forms part of a mechanism for epigenetic marking of silent chromatin in differentiated postmitotic cells. The results of this study add a new dimension to the biology of the Aurora B kinase by showing that it can act as a regulator of the epigenetic status of differentiated postmitotic cells. Previous work had shown that Aurora B is tightly regulated in cycling cells and phosphorylates proteins that are critical for the transit through mitosis (). Aurora B expression is upregulated during the transition from S to G2, peaks in G2/M and is minimal in interphase (). The observation that deregulation of Aurora B expression is associated with uncontrolled cell division in many types of cancer has reinforced the perception that the Aurora B kinase is primarily involved in cell-cycle regulation (). However, our results provide evidence that Aurora B also has important functions in non-dividing cells. In dividing cells, Aurora B is known to be responsible for a transient peak of histone H3S10 phosphorylation, which generates the double H3K9me3/S10ph modification at the G2/M phase of the cell cycle. This leads to the release of HP1β and to chromatin changes necessary for chromosome segregation (; ). Our study demonstrates that the double H3K9me3/S10ph modification is not restricted to mitotic cells, but is also found at high levels in differentiated postmitotic cells and that its presence depends on the activity of Aurora B. A particularly striking feature of plasma cell differentiation is the formation of large clusters of facultative heterochromatin as the nucleus shrinks and the cell specialises to become an antibody factory. The double H3K9me3/S10ph modification is strongly associated with these heterochromatic regions. It is interesting that plasma cells show no detectable increase in the levels of H3K9me3 without the accompanying S10ph despite there being a substantial increase in the amount of heterochromatin in these cells compared with activated B cells. This suggests that the presence of S10ph modification has an important role in modulating the epigenetic marking of facultative heterochromatin by H3K9me3 during cell differentiation. Previous studies have implicated H3 S10 phosphorylation in gene activation (reviewed in ). Our results suggest that the combination of S10 phosphorylation with K9 methylation can reverse this effect and turn the modification into a silencing mark. Microarray and conventional ChIP analysis of the distribution of the double H3K9me3/S10ph modification revealed a developmentally regulated association of the modification with gene silencing. Differentiation of mesenchymal stem cells is accompanied by the formation of domains of enrichment for the double modification that can extend across several adjacent silent genes. There is a corresponding decrease in log ratios and in the number of enrichment peaks observed for the single H3K9me3 modification, which suggests that it is converted into the double modification by addition of the phosphate group at S10. The highest levels of the double modification were present in terminally differentiated plasma cells, as measured by IF, western blotting and ChIP of individual genes. These results suggest that the complexity of epigenetic regulation can be increased during cell differentiation, as the readout from single modifications is subject to progressive modulation by the addition of new modifications to the same histone molecule. One effect would be to influence the binding of chromatin proteins, either by creating new binding sites or by preventing proteins from binding to existing sites. A key observation in this study is the finding that HP1β is displaced from facultative heterochromatin in short-lived plasma cells and that this displacement is reversed by inhibition of Aurora B. HP1 is a major component of constitutive heterochromatin, which is conserved in a wide range of organisms from fission yeast through to mammals. It binds to methylated H3K9 (; ) and recruits the Su(Var)39 histone methyltransferase (), thereby promoting further H3K9 methylation and reinforcing the process of heterochromatinisation. HP1 also recruits histone deacetylases, cohesin and the CAF1 chromatin assembly complex (reviewed in ; ). Although HP1 has been thought to be associated mainly with gene silencing, recent studies have shown that it is associated with transcriptionally active chromatin puffs on polytene chromosomes () and that it binds within the transcribed regions of a number of active genes (). Together with studies that have revealed the presence of methylated H3K9 within active transcription units (), these results point to a dual role for HP1 in activating and silencing transcription. This conclusion has received further support from studies in showing that Swi6/HP1 recruits the chromatin opening protein Epe1, which in turn makes heterochromatin more accessible to transcription by RNA polymerase II, a necessary precursor to RNAi-mediated heterochromatinisation (). Displacement of HP1 from chromatin would, therefore, be expected to have complex effects on activation and silencing of transcription. These results raise the possibility that displacement of HP1 could be important for long-term silencing of gene expression in differentiated cells. A second effect of displacing HP1 could be to allow binding of other, as yet, unidentified chromatin proteins that would recognise the double modification and would be involved in maintaining transcriptional silencing and formation of facultative heterochromatin. This idea is supported by the observation that HP1 is absent from facultative heterochromatin in chicken erythrocytes (). However, the absence of HP1 from chicken erythrocyte heterochromatin appears to be due to complete loss of HP1 from the nucleus rather than a histone modification displacing HP1 from chromatin. The following antibodies were used in western blot analysis, ChIP and IF: α-histone H3 (Abcam ab1791), α-H4 (Upstate 07-108), α-H3K9me3/S10ph (Abcam ab5819) (peptide ELISA to confirm the specificity of this antibody was carried out as described previously; ), α-H3K9me3 (Upstate 07-442), α-Aurora B (Abcam ab2254), α-HP1β (anti-M31 rat monoclonal MCA1946; Serotec), α-mouse IgM-Texas red (Vector) and α-mouse IgG-Texas red (vector), α-β-tubulin (Santa Cruz Biotechnologies) and α-actin (Sigma). Control antibodies for ChIP were nonspecific rabbit IgG (Santa Cruz Biotechnologies) and α-FLAG antibody (Sigma). Secondary antibodies for western blot studies were α-rabbit IgG-horseradish peroxidase-conjugated antibody (Sigma) and α-rabbit IgG-alkaline phosphatase-conjugated antibody (Sigma). Resting B-cell isolation and B-cell activation were carried out as described previously (). The activated B cells were positively selected with biotinylated anti-CD19 antibody (BD Biosciences Pharmingen) conjugated to streptavidin beads (CELLection Dynabeads; Dynal) and released from the beads before further analysis. C3H10T1/2 cells were expanded in complete expansion medium, DMEM medium supplemented with 10% FCS, 100 U/ml penicillin, 100 μg/ml streptomycin, 2 mM glutamax, 2 mM non-essential amino acids (Sigma) and 2 mM sodium pyruvate. Osteogenic differentiation was induced as described in by culturing the C3H10T1/2 cells in IMDM supplemented with 10% FCS, 10% HS, 100 U/ml penicillin, 100 μg/ml streptomycin, 2 mM glutamax, 20 mM β-glycerolphosphate, 50 ng/ml thyroxine, 1 nM dexamethasone and 0.5 μM ascorbic acid (all from Sigma) for 3 weeks. To confirm differentiation, an aliquot of the cells was fixed for 2 min with 1% formaldehyde and then stained for phosphatase activity using OPD (Dako) as a chromogenic substrate. Real-time PCR was used to confirm expression of the osteogenic markers alkaline phosphatase and osteocalcin. For BrdU incorporation studies, C3H10T1/2 cells were incubated in media with 10 μM BrdU for 2 h. Spleens from 6- to 10-week-old CBAB6F1 mice were disaggregated and the single-cell suspension was centrifuged on a Ficoll cushion to remove erythrocytes and cultured for 7 days in RPMI supplemented with 15% FCS, 50 μM 2-mercaptoethanol, 50 μg/ml gentamicin, 25 μg/ml LPS, 5 ng/ml IL-5, 5 ng/ml IL-6 and 10 ng/ml IL-10. On day 7, the cells were again centrifuged on a Ficoll cushion to remove dead cells and the cells obtained from the interface were blocked with 10% goat serum and 5 μg/ml anti-FC receptor γ III/II (BD Biosciences Pharmingen). The cells were then incubated with streptavidin beads (CELLection Dynabeads; Dynal) conjugated with biotinylated α-syndecan-1 antibody (BD Biosciences Pharmingen). Following magnetic selection, the cells were washed and released from the beads by digestion with DNaseI and incubated with α-B220-conjugated magnetic beads (Dynal) to deplete the B220 cells. After selection, the cultures were typically >97% α-syndecan-1/>98% B220. For plasma cell isolation from bone marrow, mice were immunised with ovalbumin, followed by a second immunisation after 3 weeks. Bone marrow was collected 3 days after the second immunisation. Plasma cells were isolated from spleen 7 days after immunisation with LPS. Plasma cell-enriched cell preparations from bone marrow and spleen were obtained by centrifugation on a Ficoll cushion followed by positive selection with α-syndecan-1 antibody conjugated to magnetic beads. For BrdU incorporation assays, activated B and plasma cells were incubated in media with 10 μM BrdU for 6 h. IF analysis was carried out as described previously (). Secondary antibodies used for fluorescent detection were Alexa 488 anti-rabbit and Alexa 488 anti-rat (both from Molecular Probes). Samples were mounted in Vectashield supplemented with DAPI (10 μg/ml) and images were collected by confocal microscopy using a TCS-SP1 microscope (Leica Microsystems) and Metamorph 4.0 software. ChIP analysis of unfixed chromatin or formaldehyde-fixed chromatin was carried out as described previously (). For plasma cells, the ChIP was carried out in the presence of SL2 cells according to the carrier ChIP protocol described previously (). Unfixed chromatin (150 μg) was immunoprecipitated with the following antibodies: 20 μl of anti-diacetylated histone H3 (Upstate 06-599), 100 μl α-H3K9me3/S10ph (Abcam ab5819), 20 μl α-H3K9me3 (Upstate) and as control 37 μl nonspecific rabbit IgG (Santa Cruz Biotechnology). For the fixed chromatin method, 500 μg of fixed chromatin was immunoprecipitated and the same ratio of chromatin to antibody was used that was described for the unfixed chromatin procedure. The input and immunoprecipitated samples were phenol/chloroform-extracted and resuspended in 100 μl of TE. For quantitation, 2 μl of the input and the immunoprecipitated DNA (IP) samples were amplified by real-time PCR using SYBR–green mix (Bio-Rad) and a DNA Engine Opticon system (MJ Research Inc.). PCRs were carried out in duplicate. Primer sequences can be obtained on request. Microarrays were generated by NimbleGen Systems by tiling 50-bp oligonucleotides across a region of 2 Mb of mouse chromosome 13 (Chr3: 87462375–89534891) with 100-bp resolution. High-frequency repeats were excluded by repeat masking. DNA (4 μg) from immunoprecipitation and input was amplified using the GenomePlex WGA kit (Sigma). After amplification, the DNA was tested for enrichment at control loci and compared to the unamplified DNA by real-time PCR. The amplified input and pull-down DNA were labelled with Cy3 and Cy5 by random priming and hybridised to the microarray by NimbleGen. Two immunoprecipitations were performed for each antibody. Input and pull-down signal intensities, scaled log ratios and enrichment peaks were provided by NimbleGen. Enrichment peaks were calculated using a permutation-based algorithm (see NimbleScan User's Guide at , default algorithm). This estimates the false discovery rate (FDR) for each peak, which is equal to the probability of finding a peak of comparable significance by chance. Significant peaks are characterised by low FDRs and correspond to four or more probes within a 500-bp sliding window with signals above the cutoff value (cutoff values are a percentage ranging from 90 to 15% of a hypothetical log ratio maximum, which is the mean+6(standard deviation)). The assays were performed on a Biacore X instrument (Biacore, Milton Keynes, UK) with Biacore HBS-EP buffer (10 mM Hepes pH 7.4, 0.15 M NaCl, 3 mM EDTA, 0.005% Tween 20) at a flow rate of 10 μl/min. Anti-6 × His monoclonal antibody (Abcam; 6500 response units, RUs) and control anti-RNA polII monoclonal antibody (kind gift of Laszlo Tora; 6700 RU) were immobilised using the Amine Coupling Kit (Biacore) on flow cells 1 and 2, respectively, of a CM5 sensor chip. 6 × His-HP1β (300 nM) was injected simultaneously over flow channels 1 and 2 followed by injection over both flow channels of 50 μM solutions of peptides for 2 min. After the injection, the dissociation of the complex was followed for 180 s. Non-phosphorylated peptides were also injected at lower concentrations for comparability to phosphorylated peptide binding. The difference between the increase in RU between flow channels 1 and 2 was used to measure the specific binding of the peptides to His-HP1β. At least two injections were carried out for each peptide. Data were analysed using Bioevaluation 3.2 software (Biacore). Live logarithmically growing undifferentiated C3H10T1/2 cells (2 × 10) were stained for 30 min at 37°C/5% CO with the DNA-intercalating dye Hoechst 33342 (10 μg/ml). The cells were sorted by FACS into G1 and G2/M fractions on the basis of DNA content. Sorting was carried out with a FACS Vantage SE sorter (Becton Dickinson).

Several mouse models recapitulate the cardinal features of human oligodendrogliomas (ODGs) (, , ; ; ; ). Chronic signaling and overexpression of the ligands and receptors of platelet-derived growth factor (PDGF) signaling pathways are particularly common in human ODG (, ; , ; ). We have demonstrated that expression of PDGF alone in nestin-positive progenitors is sufficient to induce ODG development (). Alterations that disrupt stem/progenitor cell dynamics and cell cycle arrest may affect the progression of ODG. Cells become committed to undertake a cell cycle when sufficient cyclin-dependent kinase (cdk) activity accumulates during the G1 phase. When cdk activity remains below the threshold level needed to initiate progression through the cell cycle, cells exit the cell cycle, and in some cases differentiate. Thus, interactions between the cyclins and their catalytic partners the cdks are critical in driving the cell cycle forward. Cyclin–cdk complexes are regulated by a group of proteins known as cdk inhibitors (cki). Cki are classified into two subfamilies: the Ink4 subfamily (p15, p16, p18, and p19), which specifically target cdk4 and cdk6 and disrupt the cyclin D–cdk complex; and the Cip/Kip subfamily (p21, p27 and p57), which inhibit cyclin–cdk2 complexes. Cki are considered tumor suppressors because their binding to cdks generally inhibits cell proliferation; however, cki can also increase cell motility, reduce apoptosis, modulate receptor tyrosine kinase signaling, and alter the activity of a host of transcription factors and chromatin remodeling enzymes (; ; ; ; ; ; ; , ). Subcellular location might affect these interactions and roles (; ; ; ). For example, cytosolic p27 can modulate rho activity and affect cell migration (; ; ). p21 and p27 can also bind nuclear cyclin D–cdk4 complexes, preventing their crm1-dependent export into the cytoplasm, where cyclin D1 would be degraded in a ubiquitin-dependent process (; ). Any of these interactions might provide an oncogenic advantage to a proliferating cell; however, because of the lack of suitable genetic systems that allow manipulation of the incipient cells that give rise to solid tumors, none of these have been demonstrated to be responsible for the tumor-promoting effects associated with cki expression. We have used an ODG model where disease is induced by infecting nestin-TvA mice with RCAS vectors expressing PDGF to probe the role that cki play in tumor development (). This system takes advantage of the fact that avian retroviruses (RCAS) cannot infect mammalian cells, unless they express the avian retroviral receptor (TvA) (). Transgenic mice expressing TvA under the control of the nestin promoter (Ntv-a) allow the infection of oligodendrocyte progenitor cells (OPC) and some earlier neural/glial lineage cell types (, ). Our studies of introducing PDGF into nestin-positive progenitors and driving the development of growth-factor-induced ODG highlighted a critical role for the cyclin-binding domains of p21, and we showed that a mutant of cyclin D1 which accumulates in the nucleus and binds cdk4 in p21-deficient cells could bypass the requirement for p21 during tumor development. Mutants that did not bind cdk4 or did not accumulate in the nucleus did not bypass the requirement. Together, these results establish genetic proof that p21 can act through the cyclin D1–cdk4 complexes to support tumor growth, and establish the utility of using the RCAS/TvA system for delineating the types of contributions that a protein makes to tumor development. Both a low p27 staining index and a high p21 staining index are associated with poor prognosis in human ODG (; ). Furthermore, both of these proteins play a role during the development of the oligodendroglial lineage (, ; ; ). Consequently to understand if these cki contributed in a causal manner to the pathogenesis of ODG, we crossed mice with targeted deletion of or a mutation of p27, , which removes the N-terminal 51 amino acids of the protein preventing cyclin–cdk interaction, onto an Ntv-a background and infected newborn wild-type, heterozygous, and deficient mice with a single intracranial injection of 10 DF-1 cells producing RCAS-PDGF-HA (). Ntv-a-negative littermate controls did not develop tumors regardless of their genotype. Whereas p27 loss was associated with enhanced progression of tumors, as expected of a tumor suppressor (Wendy See, EH, Marilyn Resh, and AK, manuscript in preparation), the loss of p21 unexpectedly reduced the development of tumors (). There was no evidence of morbidity in mice infected with RCAS-PDGF-HA during the 12-week period. All mice were killed at the end of the 12 weeks, and two had a tumor visible by gross histology. Because the tumor suppressive properties of p27 are well established, the rest of this report focuses on the role of p21. Our results delineating the tumor suppressor function of p27 will be reported elsewhere. Similar to gliomas in wild-type animals, tumors in the and animals were composed of small cells with round nuclei and scant cytoplasm. As shown in the , we observed intrafascicular queuing in white matter tracts of the corpus callosum, subpial infiltration, and perivascular and perineuronal satellitosis in areas of cortical invasion as described previously. The tumors were positive for oligodendrocyte markers, sox10 and olig2 (; ), and negative for a neuronal marker, NeuN (). The tumor size did not correlate with genotype; large and small tumors were observed in all genotypes. These tumor characteristics were consistent with a diagnosis of ODG. We noted that the Ki67 index, a marker for proliferating cells, was approximately two-fold lower in the two tumors arising in the mice (per × 400 field: tumor A, 36±25, <0.001; tumor B, 59±35, <0.05) compared to the average level seen in five randomly chosen wild-type mice (per × 400 field: 86±53). High standard deviations are typical because of the diffuse nature of ODGs. Furthermore, apoptotic indices, measured by the percentage of cells staining positively for cleaved caspase 3, were approximately 3–4-fold higher in the two tumors that arose in the mice (per × 400 field: tumor A, 6±2; tumor B, 8±4) compared to wild-type mice (per × 400 field: 1±1). Thus p21 might support proliferation or reduce apoptosis (or both) during the development of growth-factor-induced ODG. To ask if these effects of p21 deficiency were seen in the ‘progenitor' cells that give rise to the tumors, we isolated OPC, made them quiescent in culture, and subsequently induced them to enter the cell cycle by exposing them to PDGF. p21 status affected the growth, proliferation, and apoptosis of quiescent progenitor cells exposed to PDGF in culture. PDGF induced cell growth when added to purified wild-type cells at a concentration between 10 and 50 ng/ml (; ), but not when added to OPC (). At 18 h after the addition of PDGF, the percent of BrdU-positive cells did not increase in the culture, although it did in the wild-type culture (). Additionally, apoptotic indices measured immunohistochemically by cleaved caspase 3 (), by nuclei morphology (), or by immunoblotting extracts for cleaved caspase 3 or cleaved PARP () were all elevated in the p21-deficient culture. Consequently, p21 levels could affect the growth of progenitors responding to unbalanced PDGF signaling by acting through either the proliferative or apoptotic pathways or both. As shown in the accompanying , we confirmed that p21 deficiency did not affect the appearance of nestin-positive cells in the brains of neonatal mice or the expression of PDGF from the retroviral vector. Infection with RCAS-3xFp21 alone did not result in tumor formation (0/15). animals coinfected with both RCAS-PDGF-HA and RCAS-3xFp21 developed tumors (). The majority of dually infected mice (80%) were killed before 4 weeks of age because of symptoms of intracranial pathology such as hydrocephalus and dehydration. All dually infected mice had moderate to high grade anaplastic ODGs, as determined by gross histological examination (). The grading system we used is described in the accompanying . The three tumors that were selected for additional immunohistochemical analysis were all positive for the oligodendroglial lineage markers, olig2 and sox10, and negative for the neuronal lineage marker, neuN. Viral infection was also confirmed using interphase fluorescent hybridization (FISH) to detect RCAS-DNA. In one animal, 73 of 178 cells from a tumor region were positive, and in the other 67 of 144 cells were positive. Greater than 95% of the cells from the tumor region were positive for a reference marker that specifically reacts with the X-chromosome. In addition, as shown in the , FLAG-p21 protein accumulated in the nucleus, consistent with its localization in human ODG. Using the RCAS/TvA system allows us the opportunity to introduce multiple gene products into tumor cell progenitors. Thus, we thought it might be suitable for analyzing, at a genetic level, the importance of various protein–protein interaction domains. Specifically, we assessed the need for the carboxyl portion of p21, the amino-terminal domain, and the Cy motifs. All constructs were tagged with a 3xFLAG epitope at the amino terminus. RCAS-Np21 contained the first 32 amino acids (aa) of mouse p21 and was predicted to interact with procaspase 3 (). RCAS-Cp21 contained aa 33–160 and was predicted to interact with a variety of transcription factors and chromatin remodeling proteins (), as well as the apoptotic regulator ask1 (). Both fragments contained a Cy motif (). The relative expression of each mutant was assessed by immunoblotting extracts from the viral producer lines (). Both the RCAS-Cp21 and RCAS-Np21 constructs supported gliomagenesis when introduced with RCAS-PDGF-HA into mice. Tumor latency, incidence and grade were similar to those seen in p21-deficient mice infected with RCAS-3xFp21 and RCAS-PDGF-HA (). Mice lacking the Ntv-a transgene, regardless of p21 status, did not develop disease regardless of which virus or combination of viruses they were infected with. We confirmed expression of nuclear FLAG-tagged Np21 and Cp21 proteins in the tumors by FLAG immunohistochemistry. Mutation of the cyclin binding or Cy element, in either fragment, reduced complementation activity (). Tumors that arose in animals infected with constructs containing mutated Cy elements were of lower grade (). Thus, p21 expression facilitates the progression of the disease. To further test the requirement for the Cy element, we generated eight single amino acid substitution mutants in and around (aa 12–25) in Np21, and introduced these as RCAS vectors into p21-deficient animals with RCAS-PDGF-HA (). Our analysis of these single base alanine substitutions and another multisite mutation in Cy1 (HRSK-to-AASA) revealed a role for this element in tumor initiation or progression. We also found evidence that there may be additional rate-determining functions in regions outside what we defined as the Cy element. Mutation of D6 slowed tumor development, and mutation of S2 abrogated complementation activity. We think this might reflect a second domain. An analysis of protein folding suggests that neither the S2A and D6A mutation converted the flexible linker of the Cy element region, bordered by S14 and D25, to a more rigid helical fold as observed in the other noncomplementing mutants (). To test if the Cy elements were sufficient to account for the p21 requirement in PDGF-induced ODG, we generated additional RCAS constructs containing full-length p21 harboring either a single mutation at the Cy1 or Cy2 site, or a double mutation at both sites (). The p21Cy2 and p21Cy1Cy2 mutants were expressed at levels similar to full-length p21 (). The expression of the p21Cy1 mutant was much lower than the others and thus we could not analyze its effect (data not shown). Of the 11 p21Cy2-infected animals, we were able to grade 10 tumors (one animal died and that tumor was necrotic which prevents reliable grading). Of the four animals that died within 6 weeks, three had low-grade tumors and one had a moderate-grade tumor. Although low-grade tumors generally do not affect survival, occasionally these cells infiltrate and destroy critical brain structures leading to morbidity. Of the remaining six animals that lived into the 12th week, three had no tumor, two had low-grade tumors and one had a moderate-grade tumor. We could grade all 11 p21Cy1Cy2-infected animals. Of the five animals that died in the first 9 weeks, three had moderate-grade tumors and two had low-grade tumors. Of the remaining six animals that lived into the 12th week, two were negative and four had low-grade tumors. No mouse, regardless of p21 status, developed tumors when injected with an RCAS vector encoding p21 or any of the p21 mutants used in this study alone. PDGF was required to drive tumor development and p21, through its Cy domains, appears to affect progression. Our data implicate the Cy element in the function of p21 during PDGF-induced ODG development. This domain is necessary for the association of p21 with cyclin–cdk complexes (). Evidence from our lab and others has suggested that binding of p21 and p27 to cyclin D–cdk4 complexes does not inhibit their kinase activity in proliferating cells (; ; ), but rather can interfere with nuclear export () and cytosolic ubiquitin-dependent protein turnover (). In addition to cyclin–cdk interaction, we have found that the Cy element is also required for the interaction of p21 with two proteins involved in receptor and endosome trafficking (YL, Hediye Erdjument-Bromage, Paul Tempst, and AK, unpublished data). To determine which type of interaction was more likely to account for the p21 requirement during ODG, we looked at both changes in cyclin D and associated kinase activity (see ) and PDGF receptor density at the cell surface (see ). Cyclin D1 accumulated in the nucleus of tumor cells induced by PDGF in wild-type mice, as it did in any tumor that arose when p21-deficient mice were reconstituted with different alleles of p21 (). Furthermore, in PDGF-transformed glial progenitors, p21 was nuclear () and bound to cyclin D–cdk4 (). p21 immunoprecipitates contained an Rb kinase activity (). In p21-deficient cells, the amount of cyclin D–cdk4 complex and cyclin D-associated kinase activity were reduced (), even though p27, a related cdk inhibitor, was present (), suggesting that there might be a division of function between Kip-family members in this cell type. Furthermore, the amount of the D-type cyclins and cdk6 were reduced in PDGF-transformed progenitor cells (). The half-life of cyclin D was reduced five-fold, from approximately 60 min in wild-type cells to 12 min in p21-deficient cells. p21 is an inhibitor of cdk2 activity. While the amount of cyclin E, cyclin A and cdk2 were only modestly affected by p21 deficiency (), cdk2- and cyclin A-associated kinase activity was increased approximately 60% in p21-deficient cells (). There was no change in cyclin E-associated kinase activity (). Thus, in PDGF-transformed glial cells, p21 status clearly affected the accumulation of nuclear cyclin D. We also looked at the expression of PDGF receptor on the cell surface. The density of PDGF receptors on the cell surface was similar in both wild-type and p21-deficient cells (). This reduces the likelihood that p21 deficiency is affecting PDGF receptor accumulation at the cell surface by acting in the receptor trafficking or endosome-sorting pathway. Together, these data suggest that the requirement for the Cy element was likely to reflect its role of stabilizing nuclear cyclin D1–cdk4 complexes. To evaluate the ability of p21 to stabilize the cyclin D–cdk4 complex during PDGF-induced tumor development, we asked if enforcing expression of functional cyclin D1–cdk4 complexes would be sufficient to overcome the effect of p21 deficiency on tumor growth. Simply overexpressing wild-type cyclin D1 did not suffice to drive nuclear accumulation in p21-deficient glial cells; however, two cyclin D1 mutants were previously shown to accumulate in cki-deficient cells, and we confirmed this in our p21-deficient glial cells as well. Mutation of Thr286 to Ala (cycD1T286A) blocks phosphorylation at this residue and prevents nuclear export (). Mutation of Thr156 to Ala (cycD1T156A) largely, although not completely, accumulates in the cytosol where it binds cdk4 but fails to activate it (). Thus, using these two mutants, we could test the relative importance of accumulated cyclin D1–cdk4 complexes. cDNAs encoding these mutants were cloned into RCAS vectors and their subcellular localization was confirmed by immunofluorescence. Almost all the cycD1T286A protein accumulated in the nucleus, and the cycD1T156A accumulated almost exclusively in the cytosol (). Approximately 15% of the cycD1T156A-expressing cells also had some nuclear staining, consistent with previously published work (). Although cycD1T286A is a weak oncogene in a lymphoma model (), we did not observe this in our model probably because of the short time frame in which animals were maintained post-infection. Similarly, expression of cycD1T156A alone did not induce tumors in mice. We subsequently examined the ability of these mutants to promote ODG when introduced into mice with RCAS-PDGF-HA. Expression of these mutants in the DF-1 producer cells was approximately equivalent ( inset). p21-deficient mice infected with cycD1T286A had reduced survival compared to those infected with cycD1T156A (). Survival of cycD1T286A animals was slightly better, although not statistically significant, than those reconstituted with full-length p21. Greater than 50% of the p21-deficient mice infected with cycD1T286A developed moderate grade ODGs within 5 weeks and most of the mice were dead by 10 weeks with moderate- to high-grade glioma. Consistent with the reduced nuclear accumulation of cycD1T156A, the rate of tumor progression was significantly slower in mice infected with the RCAS-cycD1T156A mutant. Of the 10 p21-deficient mice infected with cycD1T156A, only two died within the first 6 weeks and these had low-grade tumors. Four more mice died over the next four weeks, all with low-grade tumors. The tumors that arose in both cycD1T286A- and cycD1T156A-expressing animals were ODGs as judged by histological appearance of the cells, positive staining for olig2 and negative staining for NeuN. cycD1T286A was able to bind to cdk4, and did not bind cdk2 or cdk6 (). This suggests that enforced nuclear accumulation of cyclin D1–cdk4 complexes bypassed the requirement for p21. Nuclear cyclin D can interact with both Kip-family members and nuclear hormone receptors (; ). Association with Kip-family members depends on the ability of the cyclin to bind to cdks, whereas association with nuclear hormone receptors is prevented by cdk binding (). Thus, we wanted to determine whether the ability of cycD1T286A to complement the p21 deficiency was also dependent on its ability to bind to cdk4. To accomplish this, we generated an RCAS vector expressing another cyclin D1 mutant, cycD1T286A/K114E, which did not bind to cdk2, 4, or 6 (), but accumulated in the nucleus of all cells (). This cycD1K114E mutant was originally characterized as a non-cdk-binding protein (). We also observed cytosolic accumulation in about 30% of these cells where protein accumulated in the nucleus as well. Alone, when infected into mice or wild-type mice, this mutant did not promote tumor development by 9 weeks, when this experiment was ended. However, unlike cycD1T286A, cycD1T286A/K114E did not support the development of ODG induced by RCAS-PDGF (). The one mouse that died during the sixth week was tumor free. Consequently, we conclude that the nuclear cyclin D1–cdk4 is the most likely target of p21 responsible for the tumor-promoting effect of p21. Our understanding of the roles that proteins play in hematological malignancies is further advanced than our understanding in solid tumors. Germline mutations are widely used to study both hematological malignancies and solid tumors; however, in hematological malignancies our ability to isolate stem cells from mice, genetically manipulate these , and reintroduce them into syngenic animals allows us to determine the role of each protein or pathway in a natural setting. Our understanding of the pathways impacting solid tumor development is commensurately poorer because the cell of origin of many solid tumors remains a mystery and an allograft may not recapitulate the environment in which tumors evolve. The cki family of proteins was originally identified by their ability to bind to G1 cdks, ultimately inhibiting kinase activity and preventing progression through the G1-S transition. However, we have begun to appreciate that this is only one biochemical activity, and growth suppression is only one role that these proteins have. For example, p21 and p27 play distinct roles in the growth and differentiation of OPC (; ; ). These proteins are also useful prognostic markers in ODG, albeit in a reciprocal fashion with high p27-staining indices associating with good prognosis, and high p21-staining indices associating with poor prognosis (; ). In our studies we found that p21 facilitates the development of PDGF-induced ODG in mice. Thus, p21 makes a contribution to tumor progression. It is ‘oncogenic.' How common are tumor-promoting activities? While there is an abundance of examples where cki are growth suppressive, there are a smaller number in which an ‘oncogenic' role is consistent with the data. The biochemical activities of cki might reflect the specific cell types or conditions in which they are studied; thus, ‘oncogenic' activity might be restricted to certain cell types or carcinogenic insults. In a Pten/Nkx3.1-deficient prostate model (), an MMTV-erbB2/neu mammary model (), and an MMTV-Wnt1 mammary model (), the complete absence of p21 or p27 reduces tumor development, suggesting that at least some level of p21 or p27 might be required for tumor progression under these conditions. In these three studies the cki was nuclear. A more recent study in a p27ck(−) knock-in animal model suggested a cdk-independent function promoting stem cell expansion and tumor development, and p27ck(−) protein was both nuclear and cytoplasmic (). In addition to ODG, there are suggestions for p21 ‘oncogenicity' in other human cancers as well, including prostate (; ; ), cervical (; ), breast (), squamous cell carcinoma (), and tall-cell and well-differentiated papillary thyroid cancer (RG, BS and AK, unpublished data). Consequently, a growth- or tumor-promoting role is not unusual, but our understanding of it at the molecular and cellular levels is largely based on inferences drawn from subcellular localization and evaluation of the affect of protein levels on proliferation and apoptosis. Genetic evidence validating such notions has been elusive. What biochemical activities of p21 and p27 might be important for ‘oncogenicity'? cki are found in multiple protein complexes, sometimes operating in distinct subcellular locations (; ; ). These features might account for their ‘oncogenic' role (; ; ). Some of these interactions occur when the cki are in the cytosol. In neuronal cells and mouse embryo fibroblasts, cytoplasmic p27 interacts with rhoA to affect cell migration (; ). Cytoplasmic p27 can also interact with grb2 (). Reducing cytosolic p27 inhibits cancer cell motility and tumorigenicity by affecting rho and akt signaling pathways (). Binding of cytosolic p21 to procaspase 3 (, , , ; ; ; ) or ask1 (; ) can desensitize tumor cells to apoptotic stimuli. Conversely, nuclear roles should be considered. As mentioned previously, nuclear p21 and p27 facilitate tumor development in the Pten/Nkx3.1, MMTV-Wnt1, and MMTV-erbB2/neu models. Nuclear cki can promote the accumulation of cyclin D–cdk4 (; ), and p21 can interact with a surfeit of transcription factors and chromatin remodeling proteins (; , ). However, establishing that a particular interaction is responsible, , in a developing tumor is a considerable challenge. Furthermore, given the cornucopia of possible interactions, it is unlikely that a single mechanism explains its role in all tumors. In the studies presented here we have shown that p21 accumulates in the nucleus of ODG tumor cells and in glial cells stimulated by PDGF signaling. We have shown that this is associated with the accumulation of nuclear cyclin D1 and formation of cyclin D–cdk4 complexes, and increased proliferation and reduced apoptosis. Most importantly, by using somatic cell engineering, we established that p21 acts cell autonomously to promote tumor development, and this depends on the Cy element. Through this element, p21 interacts with cyclin–cdk complexes, and interacts with components of the receptor trafficking and endosome sorting machinery. Nevertheless, the status of p21 had no effect on the accumulation of PDGF receptors at the cell surface, and we were able to bypass the effect of p21 deficiency by enforcing accumulation of functional cyclin D1. Mutants of cyclin D1 that fail to accumulate in the nucleus but bind cdk4, or that accumulate in the nucleus but fail to bind cdk4 were both unable to support tumor development. All together, this suggests that p21 promotes ODG by stabilizing cyclin D1–cdk4 in the nucleus. Although this mechanism has been suggested before, specifically for p27 in the Pten/Nkx3.1 and MMTV-erbB2 models, and for p21 in the MMTV-Wnt1 model, this is the first time that a genetic proof has been used to assess the veracity of this model. Nevertheless, our approach to identify protein domains will also benefit from further biochemical refinement. For example, it was surprising that the ability of the p21Cy2 and p21Cy1Cy2 mutants were comparable, albeit there was a ‘cy-dose' dependency to the onset of morbidity. We expected that the p21Cy2 mutant, with an intact Cy1 element, would support tumor development, just like Np21. The fact that it does not suggests that its interactions with other proteins in the cell could affect its availability to associate with cyclin D–cdk4, which is unaffected (data not shown). Additionally, overexpression of the mutants from a heterologous promoter might allow ‘weak' alleles to have functional affect. In the absence of a genetic analysis, suggestions based on knowing where a cki accumulates and the effect of its absence on proliferation and apoptosis might be incorrect. For example, it is difficult to reconcile the suggestion that p21 supports cyclin D–cdk4 accumulation in the MMTV-Wnt1 model, when later demonstrated that cyclin D1 was not required in this model. Ultimately, identifying the correct mechanism is critical for providing insight into how to modulate p21 levels for therapeutic gain. RCAS vectors were propagated in chicken DF-1 cells (ATCC, CRL-12203), cultured as suggested by ATCC. Only DF-1 cells that had been in culture for less than six passages after transfection with RCAS-viral cDNA were used for infections. PDGF-transformed glial progenitors were generated by infecting whole brain cultures of either or mice with RCAS-PDGF-HA viral supernatants obtained from infected DF-1 cells, and maintained in DMEM supplemented with 10% fetal bovine serum. The FLAG-p21 expression plasmid was constructed by cloning the mouse p21-coding sequence into pcDNA3 (Invitrogen). FLAG-tagged p21 deletion constructs were generated by PCR-based DNA mutagenesis. FLAG- or myc-tagged cyclin D1 expression plasmids were obtained by cloning mouse cyclin D1 into either p3XFLAG-CMV14 (Sigma) or pCMV-myc (CloneTech), respectively. Mutation was carried out using an site-directed mutagenesis kit (CloneTech). All FLAG-tagged cDNAs were subcloned into RCAS vectors using the Gateway recombination system. All mutations were confirmed by sequencing both DNA strands. The RCAS-PDGF-HA expression plasmid employed was described previously (). DF-1 cells producing RCAS viruses were trypsinized, suspended in ∼50 μl of media, and placed on ice before injection as described previously (). An aliquot of these cells was taken to make an extract to allow the detection of vector expression by anti-FLAG immunoblotting (). To measure proliferation, cells were grown on coverslips and incubated for 90 min in medium containing 65 μM BrdU, and subsequently fixed with 4% paraformaldehyde and stained with anti-BrdU antibodies as described previously (; ). To measure apoptosis, 2∼3 × 10 cells were analyzed using an Annexin V-FITC apoptosis detection kit according to the manufacturer's instructions (BD Pharmingen). Brain touch imprints or metaphase spreads on glass slides were air-dried, fixed in 3:1 methanol/glacial acetic acid at −20°C for 20 min, air dried, then stored at −20°C. FISH was performed as described (). Mouse DX-Was70 was used as reference probe. RCAS probes were labeled with Digoxigenin-dUTP (Roche) and X chromosome probes were nick translation labeled with Spectrum orange-dUTP (Vysis). Two hundred cells were scored for the analysis. Areas of overlapping cells were excluded from analysis. Additional Materials and methods can be found in the .

Heterochromatin is a conserved feature of eukaryotic chromosomes and plays an important role in chromosome segregation, genomic stability and gene regulation. In the fisson yeast , heterochromatin is formed at centromeres, telomeres and the mating-type () locus (). Centromeres are composed of a central domain (), which has a specialised chromatin structure associated with the histone H3 variant Cnp1/CENP-A, flanked by heterochromatic outer repeats () (). At centromeres, tRNA genes () and the IRCs () have been implicated in confining heterochromatin. At the mating-type locus, the and silent donor loci and the K region are packaged into heterochromatin constrained by the IR-R and IR-L barrier elements which recruit TFIIIC (, ; ). In regions of silent chromatin, histones are generally underacetylated (; ) and are methylated at lysine 9 of histone H3 (H3K9me) by the histone methyltransferase (HMTase) Clr4, a member of the highly conserved Suv39 family (). The H3K9 methylation is a binding site for the chromodomain proteins: Swi6, Chp1 and Chp2 (; ; ; ). Transcription of the outer repeats by RNA polymerase II (RNAPII) generates noncoding RNA transcripts that are processed into small interfering RNAs (siRNAs) by the RNaseIII-like ribonuclease Dicer (Dcr1). siRNAs associate with the RNA-induced Initiation of Transcriptional Silencing (RITS) complex, which consists of Chp1, Argonaute (Ago1) and Tas3. The RITS complex uses the siRNAs to target it to homologous chromatin for silencing (; ; ). Mutants in RNAi pathway proteins such as ΔΔ and , the second largest subunit of RNA polymerase II, lose centromeric silencing (reviewed by ). However, RNAi is dispensable for the maintenance of heterochromatin at the locus (; ). Previously, we proposed that protein Epe1 and other members of the JmjC domain family are 2-OG/Fe(II)-dependent dioxygenases that may act as histone demethylases (). Recently, several JmjC domain proteins have been demonstrated to have this activity (reviewed by ). Epe1 is distributed across all the major heterochromatic domains and certain meiotic genes (). The observation that Epe1 blocks heterochromatin from forming beyond the IR-L barrier at the locus lead to the proposal that Epe1 is a negative regulator of heterochromatin (). Loss of Epe1 leads to the downregulation of genes that are known to be upregulated in cells with defective silent chromatin, suggesting that Epe1 counteracts silencing of repressed genes (). It has also been suggested that Epe1 directly facilitates the access of RNAPII to centromeric repeats and that Epe1 has a role at heterochromatin boundaries by facilitating transcription of the IRC boundary elements (). Here we show that contrary to previous reports, predicted Fe(II)- and 2-OG-binding residues are required for Epe1 function, suggesting that Epe1 is a 2-OG/Fe(II)-dependent dioxygenase. We also demonstrate that Epe1 acts at the chromatin level to prevent heterochromatin domains from both expanding and contracting. We initially identified Epe1 as an Swi6 interacting protein in a yeast two-hybrid screen. The Epe1 cDNA obtained corresponded to the region spanning from amino acid 652 to the C-terminus, indicating that the region containing the JmjC domain of Epe1 is not required for the interaction with Swi6 (). Consistent with this and the observations of others (; ), GFP-tagged Epe1 was found to colocalise with Swi6 at heterochromatin. This localisation is dependent on Swi6, Clr4 and Rik1 (). As Epe1 is localised to heterochromatin, we investigated its role in heterochromatin stability using marker genes inserted within and outside centromeric heterochromatin at centromere 1 (). Genes placed within the centromeres are transcriptionally silenced due to the formation of H3K9 methylation/Swi6-dependent heterochromatin (; ). In the case of the marker gene, this silencing results in restricted growth on selective plates that lack uracil (−URA) and good growth on counter-selective plates that contain 5-fluoroorotic acid (FOA). Genes inserted in the distal extremity of are less silent (sites 3 and 4: ; ) and genes inserted in the euchromatin immediately adjacent to are expressed well (sites 1 and 2: ). Deletion of the gene encoding Epe1 (Δ) results in enhanced silencing of markers inserted at the extremities of the outer repeat (sites 3 and 4), indicated by increased growth on FOA. In addition, loss of Epe1 causes significant silencing of the normally fully expressed marker genes in adjacent euchromatin (sites 1 and 2; ). Chromatin immunoprecipitation (ChIP) analysis was performed to examine the level of H3K9me2, a well-characterised histone modification associated with silent chromatin. In Δ cells, high levels of H3K9me2 are found at the normally euchromatic region outside of the centromere (). This agrees with previous observations showing that in Δ cells, silent chromatin extends into nearby euchromatic regions and results in gene silencing (; ). Loss of Epe1 promotes spreading of silent chromatin from heterochromatin domains into euchromatin (; ; ). However, the effect of Δ on heterochromatin integrity has not been tested. To address this, we examined the effect of the Δ mutation on silencing of an gene inserted within the centromeric outer repeats (). In wild-type cells, this gene (I) is strongly repressed, resulting in the formation of red colonies on plates with limiting adenine, due to the accumulation of a red adenine precursor. Conversely, Δ cells that lack the histone H3K9 methyltransferase are unable to assemble heterochromatin, resulting in full expression of the marker and the formation of white colonies. Unusually, Δ colonies exhibit variegation, resulting in white, pink and red colonies (). The state of silencing of switches frequently in an Δ population, so that when white colonies are replated they often give rise to red and pink colonies and (). Thus, although these Δ isolates are genetically identical, they display distinct metastable silent and expressed states, which must reflect epigenetic differences in centromeric heterochromatin integrity. We also tested whether loss of Epe1 affects silencing of a gene located within centromeric heterochromatin (I; ). In wild-type strains, this gene is strongly silenced, allowing good growth on FOA and low levels of growth on media lacking uracil. In contrast, the silencing in Δ cells variegates (data not shown). Colonies with the gene in the transcriptionally silent or active states were picked from FOA or −URA plates and their ability to grow was assessed when challenged with selective (−URA) or counter selective (FOA) medium. Δ colonies expressing the gene (from −URA plates) were consistently able to sustain this active state, allowing better growth on −URA plates than similarly preselected wild-type colonies (). In reciprocal experiments, FOA-resistant Δ cells selected for silencing of I gene (from FOA plates) were less capable than wild type in sustaining this repressed state resulting in more growth on −URA (). Together, this indicates that silent chromatin at centromeres is less stable in the absence of Epe1. We show that loss of Epe1 causes not only spreading of heterochromatin into euchromatic regions, but also the destabilisation of heterochromatin within the centromere. The destabilisation of silencing observed in the absence of Epe1 is inconsistent with the previously proposed role for Epe1 as a factor that acts to prevent heterochromatin spreading past specific boundary elements (; ). It is well established that mutants with defective centromeric heterochromatin, such as Δ, Δ and Δ, have chromosome segregation defects; they display lagging chromosomes on late anaphase spindles and are sensitive to microtubule-destabilising drug thiabendazole (TBZ) (; ). If loss of Epe1 leads to disruption of heterochromatin, then Δ cells would be expected to display similar chromosome segregation defects. Δ colonies in which I was silent (red) or expressed (white) were replated in a serial dilution assay on plates containing 15 μg/ml TBZ. Cells derived from white Δ colonies consistently displayed greater TBZ sensitivity than wild type; however, genetically identical red Δ colonies were not very TBZ sensitive compared to wild type (). This indicates that TBZ sensitivity covariegates with the silent/expressed state, implying that cells with less intact silent chromatin are more prone to chromosome mis-segregation events. Δ cells exhibit lagging chromosomes at an elevated frequency compared to wild-type cells. A higher incidence of lagging chromosomes was observed in cultures derived from white I colonies than their genetically identical red/pink relatives (). This indicates that the white-expressed state caused by loss of Epe1 is incompatible with normal chromosome segregation and is consistent with Epe1 being required for centromeric heterochromatin integrity and centromere function. Therefore, white Δ colonies, in a manner similar to Δ cells, exhibit defective centromeric heterochromatin, which results in loss of silencing, lagging chromosomes and TBZ sensitivity. Epe1 is required to restrict domains of heterochromatin, and in its absence heterochromatin spreads into surrounding euchromatin (; ; ). However, contrary to this, our analysis demonstrates that Epe1 is required for normal heterochromatin integrity since loss of Epe1 destabilises silencing at centromeres and causes chromosome segregation defects. It is surprising that these seemingly opposing effects could be caused by absence of the same protein. A possible explanation for this phenotype is that expansion of a silent domain disrupts silencing at more internal sites, perhaps by titrating away essential components of heterochromatin. An alternative explanation is that in the absence of Epe1, the silent chromatin domains oscillate, either expanding into euchromatin or retreating to allow alleviation of silencing. To address these possibilities, a strain was constructed with a in a normally nonsilent euchromatic site ((I):) and, on the same side of the centromere, an gene within the silent region (I; ). Wild-type cells silence the centromeric gene, forming red colonies and express the euchromatic gene. Some Δ colonies form an extended heterochromatin domain, silencing the euchromatic gene (FOA colonies). The majority of these Δ colonies are red or pink (), indicating that in the absence of Epe1 centromeric heterochromatin is not disrupted when silent chromatin extends into neighbouring euchromatin. Also, when centromeric heterochromatin is disrupted in Δ cells, silencing does not spread into the euchromatin. This is demonstrated by white (: expressing) Δ cells which when replated showed good expression of the euchromatic gene (poor growth on FOA). However, occasionally, white Δ colonies with disrupted centromeric silencing gave rise to a few colonies on FOA plates, however, these were red/pink rather than white. This again indicates that in Δ cells, repression of outside the normal silent domain requires silencing to be intact in the adjacent outer repeats (). A similar experiment was performed with a strain in which the gene was inserted at the equivalent position in the euchromatin on the opposite side of the centromere to the marker (). Again, when Δ cells were plated on FOA to select colonies in which heterochromatin has spread over the gene, the colonies formed were red or pink, indicating that is repressed on the other side of the centromere (). Therefore, silencing must be maintained across the outer repeats on the left-hand side of the centromere to allow the expansion of the heterochromatin domain on the right-hand side. Together, these data indicate that in Δ cells the spreading of heterochromatin beyond the normal centromeric domain does not destabilise silent chromatin within the centromere. It also indicates that Δ mutants require intact heterochromatin on both left and right centromeric repeats in order to form an extended heterochromatin domain. Therefore, loss of Epe1 leads to a more erratic form of silent chromatin, allowing heterochromatin to oscillate, retreating or extending over greater distances than observed in the wild-type cells. Epe1 has been proposed to act at boundaries because peaks of Epe1 localisation have been found to coincide with, and promote the transcription of IRC elements (). Moreover, IRC elements have been demonstrated to act as boundary elements (). However, if Epe1 functions only at boundary elements, loss of Epe1 would be expected to have no effect on an ectopically silenced locus where no known boundary elements are present. The ectopic silencer strain contains a 1.6 kb fragment from the outer repeat of centromere 3 (L5) inserted at the locus () and has been shown to efficiently silence an adjacent marker gene but not the gene 1.3 kb downstream of the ORF. H3K9me2, Swi6 and Chp1 are associated with this ectopically silenced and this silencing is dependent on RNAi and heterochromatin components. Thus, the silent chromatin formed at this ectopic site (:L5-) is indistinguishable from that found at the centromeric repeats themselves (; ). In wild-type strains, silencing of this reporter allows good growth on FOA relative to −URA plates (). However, some Δ cells display increased growth on −URA relative to wild type (). This suggests that loss of Epe1 destabilises heterochromatin at the ectopic silencer. ChIP analysis shows that in Δ cells with disrupted silencing, H3K9me2 decreases and H3K9 acetylation increases on the marker (). Conversely, we examined whether deletion of allows spreading of heterochromatin at this ectopic site to silence the gene that resides 1.3 kb downstream of (). In wild-type cells, this gene remains expressed, resulting in white colonies. In contrast, Δ cells containing the same form a significant number of red and pink colonies on nonselective plates. The frequency of these red/pink ( repressed) colonies increases when cells with a silent gene are selected on FOA plates (). Thus, in the absence of Epe1, silent chromatin can extend further from the L5/centromeric repeat fragment and silence both and genes. When Δ mutants are grown on −URA media to select for cells that are expressing the gene, the frequency of colonies in which heterochromatin has spread from the L5 to silence the downstream is very low. This suggests that in Δ mutants, heterochromatin spreads in a contiguous and directional fashion and is consistent with previous data (). This analysis indicates that at an ectopic silencer, where there is no boundary between adjacent and genes, Epe1 is required both for robust silencing and to counteract heterochromatin spreading. As at , in the absence of Epe1, silencing variegates and heterochromatin domains fluctuate. Therefore, although Epe1 may indeed have a role at boundaries, Epe1 does not act solely at boundaries. Epe1 clearly acts both to prevent spreading at sites that lack known boundary elements and to prevent disruption of heterochromatin, suggesting that Epe1 has a direct role in regulating the extent and integrity of heterochromatin domains. It has been proposed that Epe1 acts in the RNAi pathway to recruit RNAPII to centromeric repeats, and thereby promoting the production of noncoding transcripts (). An alternative possibility is that Epe1 acts independently of RNAi and functions at the chromatin level, for example by directly modifying a heterochromatin factor. To determine if Epe1 resides in the RNAi pathway, we examined hallmark criteria that distinguish between RNAi factors and heterochromatin factors. RNAi components are required both for the establishment and maintenance of silencing at centromeres. Thus, reintroduction of the histone H3 lysine 9 methyltransferase Clr4 into ΔΔ cells does not allow the reestablishment of silent chromatin (). However, when Clr4 is reintroduced into a ΔΔ double mutant, heterochromatin is formed (). Thus, functional Epe1 is not an absolute requirement for the establishment of the silent state at centromeres. The RNAi machinery has been shown to be dispensable for maintenance of silencing at the mating-type locus (; ). Silencing at the mating-type locus was examined using a marker gene inserted 150 bp distal to (RV):; ). In wild-type cells and Δ mutants, the marker at this site is strongly silenced, resulting in poor growth on −URA plates. However, we find that in some Δ and ΔΔ colonies, silencing is alleviated giving better growth than wild type on −URA plates (). This indicates that loss of Epe1 causes variable silencing at the locus with some colonies exhibiting disrupted silencing. Δ cells have also been shown to form extended domains of heterochromatin at the locus (). Therefore, it is possible that at the locus, silencing oscillates in a similar fashion to that observed at centromeres. Defective silencing of RV in Δ cells is consistent with Epe1 not acting in the RNAi pathway, this suggests that Epe1 acts to regulate silent chromatin and/or is a component of silent chromatin itself. To determine if in the absence of the RNAi pathway, loss of Epe1 can still result in the formation of extended heterochromatin domains, we examined silencing in Δ mutants and cells bearing a mutation in Rpb2. mutants lack centromeric siRNA and loses centromeric silencing due to the inability of the mutant RNAPII to recruit RNAi components (). We examined and Δ strains with markers inserted in the centromeric heterochromatin of (I and I respectively). Silencing in the Δ double mutant variegates, so colonies were preselected on −URA and FOA. Analysis demonstrates that an Δ mutant can suppress the silencing defect of an mutant (). Stronger silencing (FOA) is observed in the Δ double mutant compared to the mutant alone. Mutations in are also able to suppress the loss of centromeric silencing observed in the RNAi-deficient Δ mutant, the enzyme responsible for siRNA generation (). Therefore, in the absence of Epe1, heterochromatin spreads along the centromeric repeats even without intact RNAi. Previous analyses have demonstrated that a moderate level of H3K9me2 methylation persists in an RNAi-deficient background (). The most plausible explanation for our observations is that in the absence of Epe1, silent chromatin can extend outwards using this residual H3K9 methylation as a nucleation point for the expansion of silent chromatin along the chromatin fibre without RNAi components or siRNAs. We next examined centromeric heterochromatin formed in ΔΔ and an ΔΔ double mutant. ChIP analysis demonstrates that in Δ cells, the level of H3K9me2 is reduced below wild-type levels. Compared to Δ, H3K9me2 is significantly increased in the ΔΔ double mutant (). This indicates that Δ cells can maintain H3K9 methylation in the absence of RNAi. Furthermore, to confirm that in the absence of the RNAi pathway Δ mutants can form extended heterochromatin domains, silent chromatin was examined at a synthetic telomere. At this synthetic telomere, heterochromatin is established independently of the RNAi pathway as it is composed of terminal TTACAG repeats but lacks the proximal telomere-associated repeats through which RNAi mediates silencing (). The synthetic telomere was created adjacent to the gene on the minichromosome Ch16. In wild-type cells, the gene juxtaposed to the synthetic telomere exhibits variegated expression resulting in red, pink, white and sectored colonies (). To assess silencing of the gene, red (repressed) colonies were replated. The red Δ colonies maintain the red silent state more effectively than red wild-type cells (). This suggests that loss of Epe1 allows more robust heterochromatin to form at the synthetic telomere, consistent with an extended silent domain. Again, no known boundary exists between telomere repeats and the gene. Together, these data indicate that in the absence of Epe1, heterochromatin can expand () and be disrupted () in the absence of functional RNAi. Noncoding RNAPII transcripts derived from the centromeric outer repeats are processed by the RNAi pathway to produce siRNAs. To determine if Epe1 affects or is required for the production of these noncoding centromeric RNAs, transcript levels were assessed by Northern blot and RT–PCR. As expected, the level of centromeric transcript in the wild type is low, but transcripts accumulate in Δ cells. We observe that in ΔΔ double mutants, the levels of transcript observed are significantly reduced compared to that of the Δ background (). This is consistent with previous observations (). However, in ΔΔ cells (and also in Δ), the levels of centromeric transcript detected are inversely correlated with the level of phenotypic silencing. Strains containing a marker within the centromeric heterochromatin (I) were grown in media containing FOA to select for the repressed state and media lacking uracil to enrich for cells in which heterochromatin is disrupted. In ΔΔ cells, more transcripts accumulate when silencing is disrupted than when silencing is intact (). This suggests that in Δ cells, the stochastic loss and formation of heterochromatin over the centromeric repeats (and the transcript promoters) regulates the amount of centromeric transcription. Therefore, the overall reduction in centromeric transcript detected in ΔΔ cells is caused by loss of regulation of heterochromatin. This suggests that the effect Epe1 has on centromeric transcription is indirect and provides an alternative explanation for the reduced levels of RNAPII associated with the heterochromatic repeats and IRC elements observed in Δ cells (). Consistent with a misregulation of heterochromatin in mutants causing reduced centromere repeat transcription, Northern analyses of siRNAs homologous to centromeric repeats revealed that siRNAs levels are variable but lower in Δ cells compared to the wild type (). siRNA levels do not cause the variegation in silencing as siRNA levels are not higher in cells with extended heterochromatin domains, than in cells with disrupted heterochromatin (). Also, the low level of siRNAs is not the cause of the disruption of silencing, because Δ cells have defective silencing at the locus where siRNAs are not required to maintain silencing (). Therefore, since we have demonstrated that in Δ cells the expansion and disruption of heterochromatin is independent of the RNAi pathway, the erratic behaviour of heterochromatin observed must be due to defective regulation of heterochromatin rather than reduced siRNA levels. Previously, we demonstrated that the JmjC domain of Epe1 can be modelled on the structure of FIH (Factor inhibiting HIFα) (). FIH is a member of the 2-OG/Fe(II)-dependent dioxygenase superfamily, which bind Fe(II) using the consensus amino-acid residues HXD/EXH (). Epe1 contains a variant of this motif in which the second histidine is replaced with tyrosine (HXEXY), therefore, we predicted that Epe1 coordinates Fe(II) with the residues H297, E299 and Y370. The structural model also suggested that Epe1 would interact with its co-substrate 2-OG via K314 along with two additional amino acids (). More recently it has been suggested that Epe1 is not an active dioxygenase enzyme. This was proposed due to the lack of histone demethylase activity (data not shown; ; ) and Epe1 overexpression studies. Overexpression of either wild-type protein or Epe1 mutated in a predicted Fe(II)-binding residue causes the disruption of centromeric heterochromatin (). This suggested that the critical H297 residue, predicted to bind Fe(II), is dispensable for Epe1 function. However, this interpretation neglects the possibility that the defective silencing observed is due to Epe1 overexpression rather than Epe1 activity. To investigate further, wild-type Epe1 and the mutant proteins Epe1-H297A and Epe1-K314A (), with defective Fe(II)-binding and 2-OG-interacting residues, respectively, were overexpressed from the promoter on a high copy plasmid. The Epe1-H297A and Epe1-K314A proteins are stable and expressed at a similar level to the wild-type Epe1 protein (). Overexpression of the wild-type Epe1, Epe1-H297A or Epe1-K314A in a strain containing the normally strongly repressed I marker, alleviated silencing, so that mainly white colonies were formed compared with the red colonies formed with the empty plasmid control (). Similarly, in a strain bearing I, overexpression of wild-type or the Epe1 point mutants resulted in the majority of colonies exhibiting increased growth on −URA plates, consistent with defective centromeric heterochromatin formation (). Therefore, the H297 and K314 residues are not required for the disruption of silencing observed when Epe1 is overexpressed. However, to determine if these critical residues are really required for Epe1 function, the same H297A and K314A alterations were made in the open reading frame of endogenous gene expressed from its native promoter. Interestingly, the and mutants have phenotypes that are indistinguishable from Δ. Like Δ cells, both and cells exhibit variegated expression of I, resulting in red, pink and white colonies (). Moreover, and cells have extended centromeric chromatin domains, H3K9me2 can spread from the centromere () resulting in silencing of a marker gene located in a normally expressed euchromatic site as indicated by increased growth on FOA (). These analyses clearly demonstrate that the activity of the Epe1 protein is abolished by the H297A and K314A mutations. This indicates that the predicted Fe(II)-binding and 2-OG-interacting residues are essential for Epe1 function. Furthermore, this is consistent with Epe1 being an active enzyme of the 2-OG/Fe(II)-dependent dioxygenase superfamily. The disruption of silencing observed when Epe1 is overexpressed is therefore not due to the enzymatic activity of Epe1. These analyses contradict previous reports that suggested that these residues are not important for Epe1 function and that Epe1 is not a 2-OG/Fe(II)-dependent dioxygenase. Here we have demonstrated that Epe1 regulates the stability of heterochromatin domains. Our analyses have show that loss of Epe1 causes silencing to variegate with some Δ cells forming extended silent chromatin domains ( and ) while other genetically identical Δ cells have destabilised heterochromatin (, and ). These silent and expressed states are metastable (). We propose that in Δ cells, silent chromatin domains oscillate, expanding into the surrounding euchromatin or contracting to cause alleviation of silencing (). We suggest that Epe1 acts directly to prevent the oscillation of heterochromatin domains rather than via boundary elements. Although Epe1 may have a specific role at heterochromatin boundaries, our data plainly demonstrates that Epe1 stabilises heterochromatin in the absence of known boundary elements ( and ). Epe1 could either act in the RNAi pathway or at the chromatin level to regulate the stability of heterochromatin domains. Our analyses show that Epe1 is not a component of the RNAi pathway and therefore must function at the chromatin level. Unlike RNAi factors, Epe1 is not required for the establishment of centromeric heterochromatin (). Also, loss of Epe1 causes variegated silencing at the locus (), where RNAi is dispensable for maintenance of heterochromatin (). In fact, in Δ cells, heterochromatin can spread independently of RNAi (). We postulate that residual pockets of H3K9me2 that remain in RNAi mutants () may act as nucleation sites from which, in the absence of Epe1, heterochromatin can spread. Therefore, the erratic behaviour of heterochromatin observed is due to aberrant regulation of heterochromatin rather than by an RNAi defect. Our data suggest that Epe1 does not regulate RNAPII (), but that Epe1 regulates the integrity of heterochromatin and therefore indirectly effects access of RNAPII to centromeric chromatin. Heterochromatin may spread along fibres in a transcription/RNAi-coupled manner. Or, in an alternative model, spreading might be caused by the polymerisation of chromatin factors in a step-wise fashion, for example, a nucleation site of Swi6, bound to H3K9me2, could recruit a histone deacetylase and Clr4, allowing H3K9 methylation of adjacent nucleosomes and binding of additional Swi6 (reviewed by ). Our results demonstrate that in the absence of Epe1, heterochromatin can spread or collapse without active RNAi, therefore, suggesting that Epe1 may prevent heterochromatin from spreading and collapsing via the step-wise assembly mechanism. We propose that Epe1 dampens the natural tendency of silent chromatin to assemble or disassemble. Thus, in the absence of Epe1, minor fluctuations in the extent of silent chromatin domains remain unchecked and the process is unregulated, resulting in frequent expansion–contraction of the silent domain (). Epe1 is likely to be an active 2-OG/Fe(II)-dependent dioxygenase. Contrary to other reports, we have demonstrated that a predicted iron-binding residue (H297) and 2-OG-binding residue (K314) are essential for the activity of Epe1 (). However, no histone demethylase activity can be detected for Epe1 (unpublished observation; ; ). Although Epe1 could be a histone demethylase, the lack of activity leads us to propose an alternative mechanism for Epe1. It is possible that Epe1 acts analogously to another JmjC domain protein, FIH, to which Epe1 has strong structural homology (). FIH is a protein hydroxylase that hydroxylates Asn803 of HIF and prevents it binding to the histone acetylase p300 (). We propose that Epe1 could be a protein hydroxylase that affects the stability of a heterochromatin protein, or protein–protein interaction, to regulate the extent and stability of heterochromatin domains. Hydroxylation of a heterochromatin factor could regulate the stability of silent chromatin, effectively buffering the extent of heterochromatin formed adjacent to a nucleation site. There are many potential substrates for Epe1. As Epe1 interacts with Swi6 (; ; ), it is possible that Epe1 hydroxylates Swi6. However, it is equally possible that Epe1 prevents silent chromatin from oscillating by hydroxylating other components of silent chromatin. For example, Epe1 could directly regulate the activity or stability of the Clr4 methyltransferase, histone deacetylases or another heterochromatin protein. The media and standard genetic procedures used were described previously (; ). Epe1 was deleted by homologous integration of to replace the ORF. The ∷KanMX4 mutant was derived from a diploid strain obtained from Bioneer (Korea). The I (I): strain was constructed by transformation of the HI fragment of the p-otr1(I)-ura4 plasmid () into cells containing I. The successful integration of at the I site was determined by PCR and Southern blot. Strains or single colonies were spotted in either a 10- or five-fold dilution onto the appropriate plates and incubated for 4 days at 32°C. To assess the sensitivity to TBZ, serial dilutions were spotted onto YES+15 μg/ml TBZ. For the analysis at the mating-type locus, h colonies were identified for analysis by their brown colour when stained with iodine. Cells were fixed and stained as previously described (). For lagging chromosome analysis, 100 late anaphases (spindle >5 μm) for each strain were analysed. Details of microscopy were described previously (, ). Whole-cell extracts were prepared from logarithmically growing cells. Cells were harvested, resuspended in trichloroacetic acid and vortexed with beads. The acid-soluble proteins were boiled in SDS–PAGE loading buffer and used for immunoblotting. Blots were probed with an anti-Epe1 antibody and anti-BIP antibody as a loading control. ChIP was performed as described () except for the following modifications. For H3K9me2 ChIP, cells were fixed with 1% PFA for 15 min at room temperature. Cells were lysed using a bead beater (Biospec products) and sonicated using a Bioruptor (Diagenode) sonicator for a total of 15 min (30 s ON and OFF cycle). Monoclonal H3K9me2 antibody (1 μl) was used per ChIP. Multiplex PCR products were separated on 1.7% agarose gels and post-stained with ethidium bromide. Quantitation of bands was performed using the Kodak EDAS 290 system and 1D Image Analysis Software (Eastman Kodak). For RT–PCR, cells were resuspended in 10 mM Tris–HCl, pH 7.5, 10 mM EDTA, pH 8, 0.5% SDS and lysed with the addition of phenol:chloroform 5:1, acid washed beads and vortexed for 30 min at 65°C. The aqueous phase was chloroform extracted and the RNA ethanol precipitated. The RT–PCR was performed using SuperScript III Reverse Transcriptase (Invitrogen). For northern blots, RNA was extracted by resuspending cells in 50 mM Tris–HCl pH 7.5, 10 mM EDTA pH 8, 100 mM NaCl, 1% SDS, lysing by the addition of phenol:chloroform 5:1, acid washed beads and vortexing for 30 min at 4°C. The soluble fraction was extracted with phenol/chloroform and ethanol precipitated. Centromeric transcripts were precipitated with 10% polyethylene glycol 8000 and 0.5 M NaCl on ice for 30 min followed by centrifugation. siRNAs were precipitated by addition of ethanol and sodium acetate and incubation at −20°C for 3 h. Transcripts were run on a 1% agarose 6% formaldehyde gel. siRNA samples were run on an 8% polyacrylamide gel. To check for loading, siRNA gels were cut above the xylene cyanol band and stained with ethidium bromide. siRNA and transcript gels were blotted by capillary transfer onto Hybond-NX (Amersham) and UV crosslinked. Transcript gels were probed with a P-labelled PCR product homologous to the centromeric repeats. siRNA gels were probed with a PCR product homologous to the repeats, and as a loading control an oligonucleotide homologous to a snoRNA. Epe1 was cloned into the pDONR201 entry plasmid (Invitrogen). Point mutations were introduced into the Epe1 entry plasmid using the QuikChange Site-Directed Mutagenesis Kit (Stratagene). The wild-type and mutated Epe1 ORFs were LR recombined into a expression plasmid under control of the promoter. The wild-type and point mutated Epe1 ORFs were checked by sequencing. A PCR product containing the point mutation was obtained by amplification of the mutated gateway entry plasmid. Primers were used with 80 base pair of homology to the region surrounding Epe1. The mutant DNA fragment was obtained using a two-step PCR protocol, creating a product with 200 base pair of homology either side of the ORF. PCR products were transformed into or strains. Colonies were screened for replacement of the marker and point mutants were checked by sequencing.

In the summer of 1985, Vallee recalls seeing work on axonal transport in the squid axoplasm from the labs of Michael Sheetz and Ray Lasek at the Woods Hole Marine Biology Laboratory. “The thinking about axonal transport was all over the place before that,” he says. Observations of fast axonal transport argued against a passive mechanism, but no one had found a mechanism to support theories like cytoplasmic streaming along the MTs. “But now,” says Vallee, “there was good evidence that there might be specific molecules responsible for transport.” The Sheetz lab identified and named the new molecule kinesin and showed it could move both MTs on glass and axonal organelles along MTs (Vale et al., 1985a). At about this time, Bryce Paschal joined Vallee's lab at The Worcester Foundation for Experimental Biology (Shrewsbury, MA) as a graduate student. A few months into his Ph.D., Paschal had second thoughts about graduate school and took a leave of absence, but he continued working with Vallee as a technician. He began working on a project to test whether kinesin had an MT-stimulated ATPase activity. He purified kinesin, tested column fractions for ATPase activity, and noticed that he got two peaks of activity—one that tracked to kinesin, but another in fractions containing MT-associated protein 1C (MAP 1C), previously identified in the lab. Another lab reported that kinesin's activity was dependent on MTs (Kuznetsov and Gelfand, 1986), so Paschal turned his attention to MAP 1C. Trace MAP 1C had always been found in the lab preparations of calf brain MTs. Preliminary characterization had shown that MAP 1C was insensitive to proteolysis, unlike other MAPs (Bloom et al., 1984). In 1982 Vallee had developed a new protocol for MT preparations using taxol, which was such a potent promoter of MT assembly that preps could be made in the absence of ATP or GTP nucleotide (Vallee, 1982). In this situation, MAP 1C became much more abundant. Nucleotide-sensitive MT association “was very characteristic of a motor protein,” says Vallee. In previous preps using nucleotide, he realized, “all of us had been throwing milligrams and milligrams of motor proteins down the drain. It explained why these proteins had not popped up before.” Thinking they were on the trail of the elusive cytoplasmic dynein, Vallee wanted more definitive proof. The best test was to do scanning transmission EM on the protein itself and compare its structure directly to that of flagellar dynein. The first images, produced early in the project, were conclusive. “There was no question that this thing was dynein,” says Vallee. Paschal set to work to show that MAP 1C acted as an MT motor. He took on the tricky MTs-on-glass motility assay and demonstrated that, in a kinesin-free prep, MAP 1C could translocate MTs in a unidirectional manner (Paschal et al., 1987a). Anterograde transport by kinesin had been demonstrated (Vale et al., 1985b). “There was no indication that kinesin could mediate bidirectional transport, but decades of neurobiology had established the retrograde movement of proteins,” says Paschal, now at the University of Virginia (Charlottesville, VA). Using flagella that have a defined polarity, he showed that MAP 1C and kinesin moved the axonemes in opposite directions (Paschal and Vallee, 1987), and that MAP 1C was the retrograde motor. The clincher was the publication of EM pictures showing that MAP 1C was in fact a two-headed cytoplasmic dynein (Vallee et al., 1988). Paschal went on to show that flagellar dynein isolated from sea urchin sperm behaved similarly in his MT motility assays (Paschal et al., 1987b). In all, it was a banner year for him, with four major publications that largely solved the vexing question of how cells moved things along MTs in two distinct directions. It was definitely worth the 5 a.m. drives to Cambridge, MA to pick up calf brains from a slaughterhouse and, “needless to say,” says Paschal, “I decided to go back to graduate school.”

Deliberate scientific fraud exists, but in modern microscopy a far greater number of errors are introduced in complete innocence. As an example of a common problem, take colocalization. Upstairs in the lab, a researcher collects a predominantly yellow merged image on a basic microscope, naturally interpreted as colocalization of green and red signals. But on the confocal microscope, there is no yellow in the merged images. How can this be? Many factors contribute. Here, I take the reader through the imaging process, from sample preparation to selection of the imaging and image-processing methods. Throughout, we will be on the look-out for problems that can produce misleading results, using colocalization as the most common example. Because one short article cannot be an exhaustive “how to” guide, I have also assembled a bibliography of a few highly recommended textbooks and microscopy web sites, which readers should consult for more extensive treatments of the critical issues introduced here. “Garbage in = garbage out” is the universal motto of all microscopists. A worrying tendency today is to assume that deconvolution software or confocal microscopes can somehow override the structural damage or suboptimal immunolabeling induced by poor sample preparation. The importance of appropriate fixation, permeabilization, and labeling methods for preserving cellular morphology or protein localization is well known to electron microscopists (), but often underestimated in optical microscopy (). Many labs use one standardized protocol for labeling with all antibodies, irrespective of whether the targets are membrane- or cytoskeleton-associated, nuclear or cytosolic. However, inappropriate fixation can cause antigen redistribution and/or a reduction in antigenicity. It is therefore important to test each antibody on samples fixed in a variety of ways, ranging from solvents such as methanol to chemical cross-linking agents such as paraformaldehyde and glutaraldehyde (for protocols see ; ), although glutaraldehyde fixation often reduces antigenicity and increases background autofluorescence. Consult textbooks for notorious pitfalls: phalloidin labeling is incompatible with methanol fixation, while microtubules are inadequately fixed by formaldehyde. Moreover, certain cell types, such as yeast cells, require specialized fixation protocols (). Permeabilization is also critical in achieving a good compromise between antigen accessibility and ultrastructural integrity. Specific detergents will produce different effects (for example, Saponin treatment produces smaller holes in membranes than Triton exposure), and it is also important to test the effects of pre-, simultaneous, or post-fixation permeabilization. Be aware that tissue processing, and particularly “air drying” steps, may introduce tissue distortions that will affect dimensions and measurements. Many sample preparation problems are of course avoided by imaging living cells, though live cell work introduces a whole range of new potential artifacts (see Important considerations for live cell imaging). Of the many types of homemade and commercial mounting media, no one product is ideal for all applications. Mounting media that harden (often containing polyvinyl alcohol) are useful for long-term sample storage and are preferred for imaging using a wide-field (compound) microscope because the sample flattens as the mountant hardens. For that very reason, however, those that remain liquid (typically glycerol-based) are preferable when three-dimensional (3D) information is desired. These require a sealant around the coverslip for stability and to prevent desiccation. Anti-fade agents are used to suppress photobleaching, but an anti-fade that is incompatible with specific fluorochromes can quench their signal significantly and/or increase background fluorescence. Consult the mountant's manufacturer for compatibility information because the anti-fade's identity may not be revealed in the datasheet. For GFP and its derivatives it is advisable to avoid anti-fades altogether, unless the sample is also labeled with a fluorochrome prone to photobleaching. Reports differ as to whether nail varnish, when used as a coverslip sealant, reduces GFP fluorescence, but users should be aware of the potential problem. A nondetrimental alternative sealant is VALAP, a 1:1:1 mixture of Vaseline, lanolin, and paraffin. w s t u d e n t s a n d p o s t - d o c t o r a l r e s e a r c h e r s w i l l h a v e t h e o p p o r t u n i t y t o c h o o s e t h e m i c r o s c o p e t h e y w i l l u s e , o r t o i n f l u e n c e t h e s e l e c t i o n o f s p e c i f i c c o m p o n e n t s f o r p u r c h a s e . H o w e v e r , t h e r e a r e c e r t a i n f a c t o r s t h a t u s e r s c a n c o n t r o l , a n d t h e y s h o u l d c o n s i d e r t h e s e c h o i c e s w h e n c o n f i g u r i n g t h e m i c r o s c o p e f o r t h e i r o w n e x p e r i m e n t s . The objective lens is the most critical component of a microscope and yet few researchers grasp the differences between specific objective classes. For example, most scientists can tell you the magnification of an objective lens, but few will know its numerical aperture (light-gathering ability). Yet it's the numerical aperture (NA) that determines the resolving power of the lens (), while magnification is only then useful to increase the apparent size of the resolved features until they can be perceived by the human eye. Thus, a 40× 1.3 NA objective lens will be able to resolve far finer details than a 40× 0.75 NA lens, despite their similar magnification. The intensity of the signal also increases steeply with increasing NA (). Therefore, the objective's NA, as well as its magnification, should always be provided in the Materials and methods section of publications. Why would anybody then choose an objective of lower NA? The answer is that other features of the objective may prove more critical for a particular sample or application. For example, NA is proportional to the refractive index of the immersion medium, thus oil immersion objectives can have a higher NA than water immersion objectives, and dry objectives have the lowest NA. But for certain applications water immersion objectives have distinct advantages over oil (see section “The problem of spherical aberration”) and high NA also comes at the expense of reduced working distance (how far the objective lens can focus into your sample), which may be problematic for thicker specimens. Other important factors to consider include design for use with or without coverslips, corrections for flatness of field and for chromatic aberrations, and transmission of specific wavelengths (particularly UV or IR light) (for detailed explanations see ; ). It is important to consider how resolution will affect colocalization analysis. We consider two fluorochromes to be “colocalized” when their emitted light is collected in the same voxels (3D pixels). If the distance separating two labeled objects is below the resolution limit of the imaging system, they will appear to be colocalized. Thus, users may “see” colocalization using a low resolution imaging system where a higher resolution system might achieve a visible separation of labels that are in close proximity but are not actually colocalized (). The NA of the objective lens, good refractive index match, and appropriate sampling intervals (small pixel sizes) will all affect resolution, and consequently, colocalization analysis. Note also that colocalization never indicates that two proteins are actually interacting, but only that they are located within close proximity. Colocalization can only be claimed in the certain absence of “cross-talk” (or “bleed-through”) between selected fluorochromes. Choosing fluorochromes with well-separated excitation and emission spectra is therefore critical for multiple labeling. Consider the use of any two fluorochromes together. If their excitation peaks overlap, the wavelength of exciting light selected for the first may also excite the second, and vice versa. If their emission spectra also overlap, the fluorescence emitted by each may pass through both the emission filter selected for the first channel and that selected for the second. Thus one fluorochrome may also be detected in the other's detection channel, a phenomenon known as cross-talk or bleed-through. Be particularly suspicious of cross-talk if your two fluorochromes appear to be 100% colocalized. Certain fluorochromes, such as Cyanine 3, are excellent for single labeling but can be problematic for multiple labeling because of spectral overlap with green emitters like fluorescein or Alexa Fluor 488. Conversely, Alexa Fluor 594 is well separated from standard green emitters, but is shifted too close to the far-red region to be useful for most green/red/far-red triple imaging (Rhodamine Red-X is better suited to this). It pays to stock a range of secondary antibody conjugates or dyes in order to tailor the combination toward specific protocols. Moreover, the brighter and more stable fluorochromes that are continually being developed may prove vastly superior to the reagents your lab has used for the past 20 years! It is equally important to consider which filter sets are available on your microscope before selecting your fluorochromes. Long-pass filter sets, collecting all emissions past a certain wavelength, are generally less useful for multiple labeling than band-pass filters, which collect emissions in a specific range (), and the narrower the range of the band-pass filter, the better it can separate fluorochromes with close emission spectra. Single-labeled controls should always be used to assess bleed-through. On confocal microscopes an additional test involves collecting images with each laser line deactivated in turn (you should now see no emission in that laser line's corresponding detection channel, unless there is cross-talk). Some cross-talk problems can be overcome on confocal microscopes by the use of sequential scanning (also known as multitracks or wavelength-switching line scans). In this mode, rather than exciting the sample with multiple laser lines at once and collecting the emissions simultaneously, first one laser line is activated and its corresponding emission collected, followed by the second laser line and its corresponding emission. However, this will not solve the problem if there is significant overlap between both excitation and emission spectra. Equally problematic is the overlap of specific fluorescence with background autofluorescence, particularly in plant tissues, in animal tissues rich in highly autofluorescent proteins such as lipofuscin and collagen, and in cultures containing large numbers of dead or dying cells. Unlabeled samples are necessary to establish the levels and locations of autofluorescence, and narrow band-pass filters maximize the collection of specific signal compared with autofluorescence. Modern spectral imaging systems can be invaluable for separating specific fluorescent signal from autofluorescence, as well as for separating fluorochromes with extensive spectral overlap ( C). #text The size of the confocal pinhole aperture determines the thickness of the optical section; that is, the thickness of sample slice from which emitted light is collected by the detectors. In most laser scanning confocal systems the pinholes have an adjustable diameter. Small pinhole diameters give thinner optical sections and therefore better z-axis resolution, which is important for colocalization analysis. However, the signal intensity is decreased, so when z-axis information is not required, or photobleaching is a problem, a larger pinhole diameter may be preferred. Stating either the pinhole diameter or the optical section thickness in publications facilitates a more informed discussion of 3D localization (including colocalization). Confocal images are generally collected using a pinhole aperture setting around 1 Airy Unit, a diameter that achieves a good balance between rejection of out-of-focus light and signal collection. For multicolor imaging it is critical to achieve the same optical section thickness in all channels, which is accomplished by adjusting the pinhole size for the different wavelengths. Be aware that regular maintenance to ensure alignment of the pinholes is critical, as a poorly aligned pinhole can result in lateral shift and a “double” image where the same pattern is visualized in consecutive z-sections. The property of different wavelengths of light being focused to different positions within your sample is known as chromatic aberration. This can lead to an apparent lack of colocalization in the image stack of fluorochromes that are colocalized in the actual sample. Thus all microscopists need to be aware of this phenomenon. Lateral (xy-axis) chromatic aberrations are generally corrected within the microscope, but note that full compensation is only achieved with proper matching of optical components. Some manufacturers use only the tube lens to impart corrections, whereas others use the objectives; thus, combining objectives and microscopes from different manufacturers can introduce aberrations. Lateral chromatic shifts can also be caused by mechanical shifts between different filter cubes or dichroic mirrors. Corrections for axial (z-axis) chromatic aberrations are more difficult. Objective lenses are corrected for chromatic aberrations across a certain wavelength range, the extent of which depends on the type and age of the objective (improved lenses are developed every year). Most users are unaware that the majority of objective lenses currently in use are fully corrected across only the (approximately) green-to-red range of emission wavelengths. Thus, two fluorochromes outside these ranges (such as DAPI and Cy5) could be focused to z-positions several hundreds of nanometers apart, even if their targets are colocalized in the actual specimen. When compounded by further aberrations such as spherical aberration, they could appear well over a micrometer apart in the z-axis, and thus in different z-slices in your image stack (). How do we check for chromatic aberration? One option is to image the tiny (e.g., 0.1-μm diameter) multicolor “Tetraspeck” beads available from Molecular Probes and see whether the different colors of each bead show up in the same z-position or not. Another method is to use two secondary antibodies, both directed against the same primary antibody but conjugated to different fluorochromes (those used in your double-labeling experiment), and see whether the signals are superimposed in the z-axis or whether one always appears below the other. Ruling out severe chromatic aberrations in your microscope set-up by these methods permits you to be more confident of your data interpretation. When aberrations are found, try using fluorochromes closer together in emission wavelength. Alternatively, certain software programs will permit you to “shift” the image in one channel relative to the other (applying the exact shift calculated from multicolor bead images). Spherical aberration describes the phenomenon whereby light rays passing through the lens at different distances from its center are focused to different positions in the z-axis. It is the major cause of the loss in signal intensity and resolution with increasing focus depth through thick specimens. Spherical aberrations occur as light rays pass through regions of different refractive index (for example, from the coverslip to the tissue, or between regions of different refractive index within the sample itself). The effects include a reduction in intensity and signal-to-noise in the plane of interest and distortions in the 3D image, with fine features appearing smeared out along the z-axis (). The aberrations become worse as you focus deeper into the sample. Corrections for spherical aberrations within the objective lens itself are only effective when certain preconditions are met. Thus, aberrations are increased by factors such as the use of the wrong coverslip thickness or type of immersion oil, too thick a layer of mounting medium, the presence of air bubbles in the immersion or mounting medium, or simply a temperature change. Note that most objective lenses for high resolution fluorescence work are calibrated for use with 0.17-mm thick glass, to which no. 1 1/2 coverslips correspond most closely. The specimen should be mounted on or as close to the coverslip as possible (avoid multiwell slides with a nonremovable gasket that places the coverslip many micrometers from the cells). The coverslip must also be mounted flat, as an angled coverslip will result in distorted optical properties (remove excess mounting medium by briefly placing small pieces of torn filter paper against the edge of the coverslip after mounting). In selecting an objective lens for imaging thicker samples, you need to consider the balance between the effects of spherical aberrations and NA on your signal intensity. A high NA oil immersion lens may be optimal for use with thin specimens, because the glass coverslip, whose refractive index is matched to that of the immersion oil, becomes the predominant sample component. However, a water immersion lens, despite its lower NA, may achieve better images from a thick, largely aqueous specimen due to the better match of refractive index between immersion medium and sample. Methods of minimizing spherical aberrations range from the development of objectives with adjustable correction collars to the use of immersion oils with differing refractive indices (; and for a detailed and highly readable explanation of optical aberrations and their practical correction see ). Appropriate acquisition settings are critical for obtaining meaningful and quantifiable data as well as “pretty” images. You must distinguish between acquiring all information in the raw data, and later presenting the data in a way that conveys the result more clearly. All settings should be established using the real sample (or a positive control), and then the negative control is imaged using identical settings. The “autoexpose” or “find” functions should never be used for negative controls, as the camera or detectors will attempt to compensate for the low signal. First, keep the acquisition settings constant between specimens to be compared quantitatively and particularly between sample and control. Second, it is important to distinguish between an image that is useful for visualization alone, and an image from which meaningful quantitative data can be extracted. For quantitative microscopy the exposure time and/or gain (brightness) and offset (by which pixels below a certain threshold are defined as being black) should be adjusted to use the entire dynamic range of the detectors. Too high a gain results in saturated pixels, which cannot be quantified because the dataset is clipped at the maximum end of the dynamic range. Conversely, an inappropriately large offset, often used to hide “background” cellular fluorescence, clips the data at the minimum end of the dynamic range and again prohibits quantitative measurements. More significantly, how do you distinguish nonspecific background signal from a low, ubiquitous level of your protein with real biological significance? Most researchers lean toward a bright, high contrast image, and thus will invariably saturate their images. To avoid this, use acquisition software features such as an “autohistogram” display, or a “range indicator” or “glow scale” look-up table, to establish the settings more objectively. Once the correctly acquired data is saved (and always stored in the raw format for future reference!), brightness and contrast or scaling adjustments can then be applied for a more visually pleasing presentation. How do you avoid saturation with samples containing some particularly bright but other very weak regions? The collection of more grayscale levels (12-bit instead of 8-bit data) will help. You can then present two pictures showing different scalings applied to the same image, adjusted for the bright features in one and the finer, less intense details in the second. In more severe cases, image each area using two different acquisition settings, then present the two images side by side. Great care must be taken to ensure adequate sampling (pixel/voxel dimensions) of your data, in all axes. According to the Nyquist sampling theorem, your spatial sampling intervals must be more than two times smaller than the smallest resolvable feature in your specimen. If this sampling requirement is not met, you will have gaps in your data and also spurious features can be introduced into your image by a process called “aliasing” (for explanation see ). Thus, most confocal microscope software packages suggest the use of a z-interval around half the optical slice thickness (which is usually calculated for you). This is sufficient for detecting all resolvable features, although the use of even smaller z-intervals is advantageous for deconvolution or for creating smoother volume renderings. In the xy-plane you should aim to relate the pixel size to the resolution of the system by adjusting the optical zoom setting (if available) or by using binning (a process by which the signals from neighboring pixels are combined into one value). The use of smaller pixel spacing than this, known as “oversampling,” results in longer acquisition times that can cause greater photodamage to your samples. The image frame resolution (e.g., 1024 × 1024) should be set high enough to submit images for publication at a suitable size while maintaining 300 dpi resolution (). Note that the use of an optical (real) zoom during image acquisition, to magnify specific features, will avoid the “pixilated” appearance of low magnification images to which digital zoom has subsequently been applied. Beyond a certain optical zoom, however, the user will enter “empty magnification,” where that objective's maximum resolution has been reached and so no additional information is being obtained. One of the greatest microscopy challenges is the choice of which cell(s) to present as a “typical” image. You may have preconceived ideas concerning your protein's localization, and subconsciously scan the sample to find the cell most closely fitting your expectations. In some cases this is a valid approach—for example to search for microinjected cells, for the sole expressers of a gene or for cells at a particular stage of the cell cycle. But there is a strong risk of focusing in on one cell and ignoring 10,000 strikingly different ones around it. The more passionate we are about our experiment, the more we must doubt our ability to be truly objective. So ask an unbiased colleague to blind label the samples or to help collect or evaluate the data. The use of a motorized stage to image multiple, random positions can also help avoid bias. Samples containing cells with varying expression patterns or morphology should be presented as a low magnification overview beside high magnification views of representative, contrasting regions. Most importantly, a statistical analysis of cell numbers exhibiting particular characteristics will strengthen your data interpretation. Transient transfections are particularly problematic for localization studies. A common mistake is to seek out transfected cells displaying the strongest expression levels, but here the high concentration of expressed protein may interfere with the balance of other proteins or cellular processes. Weak expressers are generally a better choice, in particular those showing limited signal localization. When available, antibodies to the endogenous protein can be used to assay for a normal distribution pattern. Aberrant localization may also be indicated by the abnormal distribution of a partner protein that should colocalize with the tagged component. You must decide how to present your data in the most appropriate form. With 3D or 4D data this typically involves a choice between a single z-slice or a projection of multiple slices. A single slice must be presented when colocalization and/or z-resolution are in question, but a projection may better illustrate the continuity of a 3D network. A merged image is often inadequate for demonstrating colocalization. A green-emitting fluorochrome and a red-emitting fluorochrome could be completely colocalized, but if one is brighter than the other the merged image may not appear yellow. Colocalization is better demonstrated using the “line profile” function included in many software packages, where an intensity plot for each channel is created along a line drawn across the image. Algorithms are available for calculating the degree of colocalization, but take care when establishing parameters such as threshold levels (for practical tips and caveats for colocalization studies see ; for detailed methods of quantifying colocalization see ; ). All images should be presented with scale bars. Many software packages include automatic scale bar calculation and pasting onto exported images. You can also image a stage micrometer to calculate the total magnification of a given system, which will be the product of the magnification of the objective and of other components such as the tube lens and relay optics to the camera. Quantifying image data is necessary for the transition from anecdotal observation to an actual measurement. Quantitation is also an important means of avoiding subjective bias and presenting the overall pattern of the data. It is rarely straightforward in practice, requiring stringent acquisition conditions. Images to be quantified should be acquired and exported in 12-bit or higher grayscale format, rather than the standard 8-bit (or 24-bit color) format suitable for most image presentation. Image processing can then be used, before quantification, to correct aberrations that have been introduced into the image stack during acquisition. You need to be aware of which image processing manipulations are consistent with quantification, and which are not. Constrained iterative 3D deconvolution algorithms, for instance, maintain the total signal intensity within an image stack, whereas nearest neighbor algorithms are subtractive and therefore do not. Relative quantitation, such as comparing the signal intensity between one region of interest and another, or between the sample and a control (assuming constant acquisition settings), is simpler than absolute quantitation, but even this assumes a number of prerequisites such as even illumination across the entire field. Calibration slides (made from colored plastic and available from companies such as Applied Precision, Chroma Technology Corp., and Molecular Probes) can be imaged to determine irregularities in illumination and apply corrections. When calculating changes in signal intensity over time you must compensate for general photobleaching as well as for temporal fluctuations in laser power or lamp illumination. Laser power can be particularly volatile immediately after switching on the system, so a warm-up period of 30–60 minutes is recommended. A monitor diode or photosensor (if available on the system) and/or standardized samples are useful for normalizing experiments for fluctuations in excitation intensity. Absolute quantification presents a greater challenge, requiring the researcher to have a good understanding of both the spectral and physical properties of the specific fluorochrome/fluorescent molecule and the appropriate choice of microscope optics and settings (). Important properties of the fluorochrome that must be taken into consideration include the extinction coefficient, the quantum yield, the photobleaching rate and properties, the chromophore folding kinetics, and the pH sensitivity (this can substantially affect measurements of proteins moving into and out of subcellular compartments). The most critical components of the fluorescence microscope to consider for quantitative imaging are the objective lens (including its NA and its spectral transmission properties), the emission filter, and the detector (). An emission filter that is well matched to the spectrum of your fluorescent probe will result in a better signal-to-noise ratio. A narrow band-pass filter is usually preferable to maximize collection of specific signal while minimizing the contribution of autofluorescence. Linear detectors (including the majority of cooled CCD cameras and photomultiplier tubes) will facilitate quantitation better than nonlinear ones (such as intensified CCD cameras). Standardized samples of known fluorochrome concentration can be used to establish appropriate gain and offset settings for the detectors. Saturation of the fluorophore, which occurs particularly when using laser excitation, also introduces nonlinearity into the measurements, making calibration of the system very difficult. Thus, it is recommended to use the lowest laser power that gives a sufficient signal-to-noise ratio. Wide-field microscopy is often inappropriate for quantitation because you collect emitted light from the whole sample depth without knowing the thickness of each cell or structure. The application of 3D deconvolution algorithms to an image stack can overcome this problem for thin samples, but not for thick or highly fluorescent samples. Confocal microscopy is generally more quantification-friendly for samples over 15–20 μm in depth because of the defined optical section thickness. However, deeper focal planes will show reduced signal intensity due to absorption and scatter, necessitating further, more complex corrections. So far this article has concentrated on basic image acquisition. This next section will highlight a few danger areas associated with some more complex techniques used to monitor the kinetics of protein trafficking or protein–protein interactions in living cells. The most common technique for monitoring protein kinetics is fluorescence recovery after photobleaching (FRAP). Many confocal and deconvolution microscope systems have incorporated remarkably user-friendly FRAP routines into their acquisition software. Unfortunately, interpreting the data is not always as simple. It is essential to “normalize” for general photobleaching by monitoring control cells that were not targeted. Furthermore, be aware that excitation light bright enough to bleach fluorescent molecules in a short time period can severely disrupt cellular ultrastructure. For quantitative FRAP, you must decide in advance which of the numerous available models will be used for analyzing the recovery curves, as this choice may affect the experimental design (). Fluorescence (or Forster) resonance energy transfer (FRET) describes the nonradiative transfer of photon energy from a donor fluorophore to an acceptor fluorophore when they are less than 10 nm apart. FRET thus reveals the relative proximity of fluorophores far beyond the normal resolution limit of a light microscope. However, since FRET also relies on additional prerequisites, such as a certain relative orientation of donor and acceptor (), an absence of FRET cannot always be interpreted as the fluorophores being more than 10 nm apart. Positive signals can also be misleading, as FRET may occur where any two noninteracting proteins are highly concentrated in localized areas. Two common methods for measuring FRET include sensitized emission and acceptor photobleaching, and the most appropriate for your experiment will depend on factors such as the need for time-lapse imaging, the ability to bleach regions of interest, and laser line availability. In sensitized emission studies, rigorous positive and negative controls are required to correct for factors such as cross-talk and direct excitation of the acceptor by the donor's excitation. When working with live cells, the rules for optimal imaging of fixed cells drop in priority. Phototoxicity and photobleaching become your biggest enemies and efforts focus on keeping the cells alive and behaving in a “normal” physiological manner (see the invaluable “Live Cell Imaging—A Laboratory Manual”, R.D. Goldman and D.L. Spector [eds.], for detailed coverage). This requires appropriate environmental conditions (temperature, media, CO, and possibly perfusion) and also optical considerations (such as the reduced phototoxic effects of longer excitation wavelengths). A major issue during time courses is the focal drift caused by thermal fluctuations in the room. Environmental chambers that enclose the entire microscope are generally more thermo-stable than smaller imaging chambers, but the latter can be more convenient for perfusion or for experiments requiring rapid temperature shifts. Another serious challenge lies in acquiring images fast enough to capture rapid biological events and to accurately portray dynamic structures. This is particularly tricky when using multiple probes, as the labeled target may move in the time required to switch between filter positions. Solutions to this include the use of a simultaneous scan mode on a confocal microscope (in the absence of cross-talk), or positioning an emission splitter in front of a CCD camera to collect both signals simultaneously on the camera chip. With fixed cells you typically maximize your signal-to-noise ratio via longer exposure times in wide-field systems, or in confocal microscopes by using higher laser power or by increasing the time the laser dwells on each pixel (using averaging or slower scan speeds). Optical zoom and Nyquist sampling are applied for optimal image quality. Approaches to minimize photobleaching in live cell imaging include a reduction in exposure time or laser power and pixel dwell time (you can compensate by using higher gain), increased pinhole diameters, the use of lower magnifications, and sub-optimal spatial sampling in xy- and z-axes. Binning your signal and the use of faster camera readout rates will enable more rapid imaging. The consequence of these compromises may include poorer resolution and reduced signal-to-noise ratios. As the imaging proceeds, how do you avoid misleading data by checking for normal physiological behavior of your cells? Here are a few clues: Have they maintained their typical morphology, or are they shrunken, blebbing, rounded up, or coming off the coverslip? Are cells and organelles still moving around at a customary rate? Acquiring simultaneous or sequential transmitted light and fluorescent images is an excellent way of assessing this; If the sample is returned to the incubator after the experiment, will the cells carry on dividing or the embryos survive? Continuing cell division is perhaps the most critical indication of healthy cells. Given the complexities discussed above, how can we all share the responsibility for ensuring that published imaging data is an accurate representation of the truth? Researchers need to learn enough about specimen preparation and the available imaging equipment to establish appropriate settings and collect optimal images. The wealth of modern information resources (see online supplemental material, available at ) enables all users to grasp at least basic microscopy concepts. Central imaging facilities can provide more advanced information required for specific applications and can help to ensure the use of appropriate imaging systems. The researcher's supervisor should rigorously critique both the raw and processed data, and needs to appreciate that high quality microscopy requires a significant time investment. Manufacturers must strive to design hardware that provides constant imaging conditions, and software that includes user-friendly tools for image analysis, and to ensure that researchers purchase the most appropriate equipment for their needs. Finally, scientific journals should set stringent guidelines, and manuscript reviewers must be critical of imaging data presentation, to guarantee that publications contain sufficient experimental detail to permit the readers to properly interpret the images and to repeat the experiments. Only by such a collective effort can we strive to present the true picture. For a more extensive list of useful microscopy resources, including highly recommended textbooks, web sites, and practical courses, please see the online supplemental material, available at .

Spindle formation relies on intricate spatial and temporal control of microtubule (MT) dynamics and coordinated organization by motor proteins (for review see ). Mitotic chromosomes play an active role in this process by stabilizing MTs in their vicinity and by forming attachments at their kinetochores that facilitate their metaphase alignment and anaphase segregation. However, the molecular mechanisms linking dynamic MTs to chromosomes are poorly understood. Stabilization of MTs by mitotic chromosomes is most apparent and essential in systems that lack MT nucleation centers (centrosomes), but increasing evidence suggests that this is a conserved process operating in many cell types (; ; ; ; ). Using meiotic egg extracts is a useful way to study this phenomenon, as chromatin-coated beads are sufficient to induce spindle assembly in the absence of centrosomes and kinetochores (). Dynamic MTs generated by chromatin are organized by MT-based motor proteins, which may contribute to chromatin–spindle interactions (). A fundamentally different kind of MT connection occurs at the kinetochore, where plus ends of a MT bundle form a stable yet dynamic attachment capable of coupling MT depolymerization to chromosome movement. A variety of kinetochore-associated proteins have been implicated in this process, including dynein, kinesin 13 (mitotic centromere-associated kinesin [MCAK]/XKCM1), the chromosomal passenger complex, and kinesin 7 (centromeric protein E [CENP-E]). However, it is poorly understood how the kinetochore–MT interface mediates chromosome movements and which factors are involved. A class of MT-associated proteins that concentrate at MT plus ends has emerged as a potential key player in chromosome–MT interactions during mitosis. These plus end–tracking proteins or +Tips, such as the cytoplasmic linker protein 170 (CLIP-170) and adenomatous polyposis coli (APC), localize to kinetochores during mitosis and have been suggested to participate in MT–kinetochore attachments (; ; ; ). CLIP-associated proteins (CLASPs) have also been identified and have been shown to associate with kinetochores independently of MTs. Mutant analysis and RNA interference of the version, multiple asters/Orbit, revealed that it is required for chromosome alignment, kinetochore–MT attachment, and maintenance of spindle bipolarity (; ; ). Intriguingly, a study using photobleaching and microsurgery suggested that CLASP is involved in MT polymerization at plus ends essential for MT poleward flux (). Further evidence supporting a role for CLASP in mitosis results from studies in human cells and embryos (; ), but the molecular mechanisms behind CLASP protein function remain unclear. To investigate the role of CLASP in spindle assembly and chromosome segregation in egg extracts, we cloned the homologue Xorbit (Fig. S1, available at ). Consistent with Orbit/CLASP localization in and mammalian cells, Xorbit associates with spindle MTs, spindle poles, and kinetochores during metaphase in egg extracts and shifts to the central spindle in late anaphase (Fig. S2; ; , ). To assess the mitotic processes for which Xorbit is required, α-Xorbit antibody was used to quantitatively (>98%) deplete the protein from extracts arrested in metaphase of meiosis II (cytostatic factor–arrested [CSF] extract; A). Spindle assembly reactions were performed by cycling CSF extract containing sperm nuclei through interphase to allow DNA and centrosome replication and then cycling back into metaphase (). Although mock-depleted extracts yielded predominantly bipolar spindles with chromosomes congressed at the metaphase plate, Xorbit-depleted extracts generated aberrant spindles with severe chromosome alignment defects ( B). The average spindle length after Xorbit depletion was significantly shorter than controls (19.6 ± 3.8 μm and 31.8 ± 4.2 μm, respectively; = 100; three experiments), and ∼20% of all structures were monopolar. We conclude that Xorbit depletion causes a metaphase phenotype similar to CLASP inhibition in other organisms (; ; ). Previous domain analysis of human CLASP revealed a highly conserved CT domain capable of localizing to kinetochores (Fig. S1; ). To potentially distinguish among Xorbit's different mitotic functions, we expressed the CT domain fused to GST (GST-CT) and tested its effects on spindle formation. In contrast to the GST control, spindles that formed in the presence of GST-CT displayed severe chromosome alignment defects, and many spindles appeared shorter in length ( C), which is a phenotype similar to Xorbit depletion. By immunofluorescence with an α-GST antibody, Xorbit GST-CT localized to spindle MTs and concentrated at kinetochores and spindle poles similar to endogenous Xorbit ( D). These results provide further evidence that the mitotic defects observed upon Xorbit depletion are specific. Because GST-CT did not prevent the association of endogenous Xorbit with chromosomes by biochemical analysis (not depicted), we speculate that it affects Xorbit's molecular interactions. In addition to chromosome alignment defects indicative of a role at the kinetochore, the small spindle phenotype observed upon Xorbit inhibition suggested that Xorbit might also influence nonkinetochore MTs. To examine Xorbit's role in chromatin-induced MT stabilization, spindles were assembled around DNA-coated beads (). In mock-depleted extracts, MT stabilization and organization generated bipolar spindles, whereas in Xorbit-depleted extracts, this process was severely inhibited ( A). No MT polymerization was observed at ∼98% of all bead clusters ( = 300; three experiments), suggesting that Xorbit plays a crucial role in chromatin-induced MT formation. Interestingly, GST-CT addition did not cause such a severe phenotype but predominantly resulted in small, distorted spindles (∼60% of all bead spindles; = 300; three experiments; B). This differs from the quite similar effect of Xorbit depletion and GST-CT addition on sperm spindle assembly, suggesting that endogenous Xorbit can still perform part of its function in the presence of GST-CT and that this function is redundant with that of centrosomes and/or kinetochores. One intriguing possibility is that Xorbit also acts at MT minus ends through nucleation/stabilization in addition to stabilizing dynamic plus ends. Consistent with a potential role at MT minus ends, Xorbit localizes to the poles of both bead and sperm spindles (Fig. S2 and not depicted). To determine whether Xorbit plays a general role in regulating MT dynamics, MT asters were generated in CSF extracts by adding either centrosomes as nucleation sites or DMSO, a MT-stabilizing agent that induces aster formation. No effect on aster assembly was observed by fluorescence microscopy upon Xorbit depletion in either assay ( C and not depicted). To confirm these observations more quantitatively, DMSO asters were pelleted and MT polymer levels were evaluated by α-tubulin immunoblotting, which revealed no obvious difference between mock- and Xorbit-depleted extracts ( D). These data suggest that Xorbit does not exert a global effect on MT dynamics but plays a role specifically in chromatin-driven MT assembly. Previous studies and our characterization of Xorbit depletion establish an important function for Xorbit/CLASP in spindle assembly, but its function in later stages of mitosis has not been closely examined (; ; , ). Therefore, we investigated the effects of Xorbit depletion on chromosome segregation and spindle dynamics during anaphase by adding calcium or an activated version of calcium/calmodulin-dependent kinase II to mock- or Xorbit-depleted cycled spindles arrested in metaphase. In control extracts, sister chromatids separated and segregated to opposite spindle poles, and kinetochore MTs shortened and spindle poles separated. Surprisingly, in Xorbit-depleted extracts, spindle MTs depolymerized within a few minutes of anaphase induction ( A). Time-lapse fluorescence microscopy further illustrated the dramatic spindle MT disassembly at anaphase onset in the absence of Xorbit ( B and Videos 1 and 2, available at ). To determine whether the MT depolymerization was a consequence of aberrant metaphase spindle assembly or the result of a specific requirement for Xorbit during anaphase, we added the GST-CT construct to wild-type metaphase spindles and concomitantly triggered anaphase. Although the metaphase spindles were robust and normal in size, their MTs rapidly depolymerized upon anaphase onset when Xorbit was inhibited ( A). In contrast, if GST-CT was added without triggering anaphase, spindles persisted and gradually became small and distorted, as in C (not depicted). To examine whether the depolymerization resulted from defects in kinetochore–MT interactions or because all spindle MTs require Xorbit activity to persist during anaphase, we compared the effects of GST-CT on chromatin bead spindles that were arrested in metaphase or induced to enter anaphase. Whereas GST-CT addition caused spindle distortion in metaphase, it caused complete MT depolymerization within 10 min of anaphase onset ( B). We conclude that Xorbit plays an essential role in MT stabilization during anaphase. At the kinetochore, CLASP has been proposed to promote tubulin subunit addition at kinetochore fibers, driving poleward MT flux (), and our results are consistent with such a role for Xorbit/CLASP during metaphase. In anaphase, however, polymerization at the plus end stops and kinetochore MTs shorten, partly because of flux/minus end disassembly and partly because of plus end depolymerization (). Our results expand the role of Xorbit as a plus end–stabilizing factor that protects both kinetochore and nonkinetochore MTs from uncontrolled disassembly while attached chromosomes segregate. To investigate potential binding partners of Xorbit that could account for its mitotic phenotypes, we performed pull-down experiments with GST-CT and GST as a control in and extracts ( A). Mass spectrometry analysis revealed a homologue of CLIP-170 as a specific GST-CT–interacting 250-kD protein. CLASP is known to interact with CLIP-170, which is preferentially associated with MT plus ends and is implicated in vesicle transport, but so far this interaction has only been reported in interphase (). Another high molecular mass protein that specifically pulled down with GST-CT was identified as kinesin 7 or CENP-E, a kinetochore-associated motor protein required for chromosome alignment (; ; ). We confirmed this interaction with an α–CENP-E antibody, which recognized a specific band above 300 kD in CSF extracts and GST-CT pull-downs ( B). CENP-E did not coimmunoprecipitate with Xorbit antibodies, nor were its levels altered in Xorbit-depleted extracts (unpublished data), but it is likely that the COOH-terminal antibody disrupted the interaction between Xorbit and CENP-E. Furthermore, CENP-E localization to kinetochores in Xorbit-depleted extracts was not impaired ( C), suggesting that chromosome alignment defects in the absence of Xorbit reflect a functional interaction between the two proteins through the Xorbit CT domain. In the presence of GST-CT, CENP-E levels at kinetochores appeared slightly elevated, and aggregates of the two proteins often formed near spindle poles. These foci did not contain another kinetochore component, BubR1 (; arrows), indicating that the aggregates were not kinetochores and further supporting a biochemical interaction between CENP-E and GST-CT. CENP-E is believed to somehow link the leading kinetochore to shrinking MT plus ends independent of ATP hydrolysis and to move the trailing kinetochore toward the growing MT plus end during antipoleward movement (; ; ; ). We propose that Xorbit supports CENP-E in coupling MT dynamics to chromosome movement either by a direct tethering activity or through its association with CENP-E. It is possible that Xorbit also functions through other factors during congression such as cytoplasmic dynein, because Xorbit interacts with CLIP-170, a dynein regulatory factor that has also been implicated in chromosome alignment ( A; ). Thus, Xorbit could function by physically linking chromosomes to dynamic plus ends during congression or by targeting/regulating necessary motor activities. In summary, Xorbit plays an essential role in multiple aspects of spindle MT dynamics in egg extracts. This differentiates Xorbit/CLASP from other +Tips that have been examined to date for roles in spindle assembly and function. For example, although APC is implicated in chromosome missegregation (; ; ), the depletion of APC from egg extracts caused subtle defects in spindle morphology, except under conditions in which kinesin 13/XMCAK was codepleted (; ). In contrast, EB1 influences global MT dynamics by promoting plus end polymerization (). Our study highlights Xorbit/CLASP as a critical factor linking chromosome segregation to MT dynamics during cell division. Full-length Xorbit was generated by PCR from a cDNA library (gift from P. Budde, Cell Press, Boston, MA) with primer sequences derived from EST clones homologous to human CLASP. GST-CT encoding the 282 CT amino acids fused to GST was expressed in , purified by glutathione Sepharose 4B chromatography, and dialyzed into XB (10 mM Hepes, pH 7.7, 1 mM MgCl, 0.1 mM CaCl, 100 mM KCl, and 50 mM sucrose). α-Xorbit antibody was generated against amino acids 799–1,457 fused to GST (Covance Research Products) and was affinity purified. CSF extract was prepared, and spindle assembly reactions with replicated sperm chromosomes were performed as previously described (). Extracts were driven into anaphase with calcium solution (4 mM CaCl, 100 mM KCl, and 1 mM MgCl) or an activated version of CamKII (CamKII plasmid was a gift from M. Doree, Centre National de la Recherche Scientifique, France) expressed in reticulocyte lysate. Centrosome/DMSO aster formation and spindle assembly around chromatin-coated beads were assayed as previously described (). For MT pelleting assays, DMSO aster reactions were diluted with 30% glycerol in BRB80 (80 mM K-Pipes, pH 6.8, 1 mM MgCl, and 1 mM EGTA) and spun down through a cushion of 60% glycerol/BRB80. Pellets were analyzed by immunoblotting with a monoclonal α-tubulin antibody (1:5,000; E7; Developmental Studies Hybridoma Bank) using standard techniques. 10 μg α-Xorbit antibody or random rabbit IgG (control; Sigma-Aldrich) was coupled to 50 μl protein A–Dynabeads (Dynal) and used to immunodeplete 150 μl CSF extract. Depletion was assessed by immunoblotting with 1 μg/ml α-Xorbit antibody. To inhibit Xorbit, GST-CT was added to the extract at a final concentration of 1.8 μM (approximately endogenous Xorbit levels) either upon entry into mitosis or at the onset of anaphase. As a control, 1.8 μM GST was added. Spindle reactions were spun onto coverslips and fixed as previously described (). α-Xorbit and α–CENP-E antibody were used 1:3,000, α-BubR1 antibody was used at 1:5,000 (α–CENP-E and α-BubR1 antibodies were a gift from D. Cleveland, University of California, San Diego, La Jolla, CA) and α-GST antibody (gift from T.J. Maresca, University of California, Berkeley, Berkeley, CA) was used at 1:500. Images were collected with a fluorescence microscope (model BX51; Olympus) with a dry 40× NA 0.75 objective, a cooled CCD camera (model OrcaII; Hamamatsu), and MetaMorph software (Molecular Devices). Images in Figs. S2, 1 D, and 5 (C–E) were taken on an imaging station (DeltaVision; Applied Precision) equipped with an inverted microscope (model IX70; Olympus) with a 60× NA 1.35 oil immersion lens (Olympus) and a cooled CCD camera (model Photometrics; Roper Scientific). Images were taken with a Z-stack size of 0.2 μm and deconvolved, and stacks were projected into a single plane. Images were processed using Adobe Photoshop. 100 μl of or extract (a gift from M. Blower, University of California, Berkeley) incubated with 7 μM GST-CT was added to 60 μl glutathione Sepharose 4B slurry and rotated at 4°C. Extract was removed, and the slurry was washed once with XB, three times with XB + 100 mM KCl, once with XB, eluted with 25 μl 2× SDS sample buffer, and analyzed by immunoblotting with α–CENP-E antibody (1:5,000) or mass spectrometry. For mass spectrometry analysis, protein bands were cut into small pieces, washed in 25 mM ammonium bicarbonate in 50% acetonitrile, reduced with DTT, alkylated with iodoacetamide, and digested with Sequencing Grade Modified Trypsin (Promega). Peptides were extracted with 5% trifluoroacetic acid (TFA)/50% acetonitrile and combined. Recovered peptides were purified with μ-C18 ZipTips (Millipore) and eluted. About 1 μl of the peptide mixture from each gel band was combined with an equal volume of matrix solution and allowed to dry on the MALDI target. The matrix solution used was a 10-mg/ml solution of α-cyano-4-hydroxycinnamic acid in 0.1% TFA/50% acetonitrile. Mass spectra were acquired on a MALDI-TOF mass spectrometer (Reflex III; Bruker). Proteins were identified by searching the SwissProt or NCBInr databases using MS-Fit, a program of ProteinProspector (University of California, San Francisco; ). Fig. S1 shows a domain schematic and sequence alignment comparing Xorbit with human CLASP1 and 2 and a second clone. Fig. S2 shows immunofluorescence localization of Xorbit in extract spindles in metaphase and anaphase. Videos show time-lapse fluorescence microscopy of spindle MTs after anaphase was induced in mock- (Video 1; total time of video is 20 min) or Xorbit-depleted (Video 2; total time is 12 min) extracts, performed using the wide-field Olympus microscope set-up described above. Frames were collected every 30 s and are displayed at a rate of 10 frames/s. Online supplemental material is available at .

Little is known about how microtubules (MTs) overlap and function in living cells to promote haploid nuclear fusion, or karyogamy. The budding yeast provides a genetic model system to study nuclear congression, the process in which haploid nuclei are moved toward each other (). MTs are nucleated from the spindle pole body (SPB), and plus ends elongate into the cytoplasm ( A; ; ; ). A MT plus end protein complex is formed to orient the nucleus and maintain the attachment of dynamic MT plus ends to the shmoo tip ( B; ; , ). Attached MT plus ends switch between polymerization and depolymerization phases of dynamic instability, producing nuclear oscillations toward and away from the shmoo tip (). At the onset of cell fusion, MT plus ends from oppositely oriented mating cells are in proximity to one another ( C), ultimately facilitating MT–MT interactions. Nuclear oscillations cease after MT interactions are established, and MTs switch into a persistent depolymerization state during nuclear congression (). Once MTs have drawn both nuclei into proximity, karyogamy can begin. A number of proteins bind MT plus ends and are required for karyogamy. The minus end–directed MT motor protein Kar3p concentrates at plus ends and is required to maintain depolymerizing MTs at the shmoo tip in addition to functioning in nuclear congression (; ). Nuclear translocation and orientation to the shmoo tip before cell fusion are actin dependent. Kar9p associates with the MT plus end–binding protein Bim1p, the budding yeast EB1 homologue, and the type V myosin Myo2p to link MTs to the polarized actin cytoskeleton (). Bik1p, the human CLIP-170 orthologue, binds MT plus ends to stabilize MT length and is critical for karyogamy (; ; ). Both Bik1p and Kar9p are transported to the MT plus end by the kinesin-like protein Kip2p, but no role for Kip2p in karyogamy has been described (; ; ). The main hypothesis for nuclear congression in living cells is a “sliding cross-bridge” mechanism in which, after cell fusion, MTs from opposite SPBs are thought to elongate past each other, producing a bundle of overlapping MTs of opposite orientation ( D; ). Kar3p, through its minus end–directed motility, is thought to cross-link the overlapping MTs and pull the SPBs together (; ; ). In addition to sliding, the MTs are proposed to shorten as the SPBs come together. An unexplained puzzle in the sliding cross-bridge model is what coordinates MT depolymerization with sliding, because MT shortening occurs as the two SPBs and attached nuclei come together. The sliding cross-bridge model proposes that Kar3p depolymerizes MTs from the minus end at the spindle poles, although this was based on early in vitro studies (; ). Thus far, fluorescent marks on MTs indicate that both polymerization and depolymerization occur solely at the plus ends (, ; ). Additionally, a recent in vitro study demonstrated that Kar3p is a plus end MT depolymerase (). These data suggest that proteins at the plus ends regulate polymerization and depolymerization and could both tether dynamic plus ends to the shmoo tip and perform nuclear congression. In the sliding cross-bridge model, plus ends should be found near the SPBs during nuclear congression. An alternative model for nuclear congression arises from the proximity of plus end–binding proteins on MTs at the shmoo tip before cell fusion ( E). In the plus end model, linkage of MTs from opposite SPBs occurs when plus end complexes interact. MT depolymerization would provide the force to pull the nuclei together. This model predicts that plus end complexes remain concentrated at the site where MTs from oppositely oriented SPBs contacted each other after cell fusion. To determine by what mechanism nuclear congression occurs, MTs and plus end–binding proteins were analyzed in living cells. Before nuclear congression, Kar3p, Bik1p, and Kip2p were required for the anchorage of MT plus ends to the shmoo tip. After cell fusion, MT plus ends interacted stochastically to drive nuclear congression. Bik1p and Kar3p localized to oppositely oriented MT plus ends that interacted near the site of cell fusion in wild-type cells. As nuclear congression occurred, the positions of the plus ends were unchanged as SPBs moved inward. By analyzing karyogamy mutants, our data suggested that Kar3p was required to initiate MT plus end interactions, whereas Bik1p promoted persistent MT interactions during nuclear congression. Kar9p contributed to the fidelity of nuclear congression by guiding plus ends toward each other. These data support a model in which oppositely oriented MTs interact and depolymerize at their plus ends to draw opposing nuclei together in . Kar9p, Bik1p, and Kar3p are required for karyogamy after cell fusion (; ; ). Δ mutants have the most severe nuclear congression defect, followed by Δ and Δ cells (). To test whether defective nuclear congression was preceded by a defect in nuclear orientation to the shmoo tip, GFP-Tub1p–marked SPBs and MTs were examined. In wild-type cells, nuclear orientation occurred when the SPB was inside or near the base of the shmoo tip ( A and Table S1, available at ). As expected, Δ cells had a nuclear orientation defect (), and the SPB was positioned in the cell body distal to the shmoo tip (Table S1). However, nuclear orientation to the shmoo tip occurred in the absence of Bik1p and Kar3p ( and Table S1). Additionally, Kip2p was not required for nuclear orientation ( and Table S1; ). Therefore, the nuclear congression defects characterized for Δ and Δ mutants do not result from a general defect in nuclear orientation. After nuclear orientation in wild-type cells, MTs remain attached to the shmoo tip (). The persistence of MT plus end attachment at the shmoo tip was measured by the percentage of time that continuous GFP-Tub1p fluorescence extended from the SPB to the shmoo tip in time-lapse records. In wild-type cells, the degree of persistence was 100%, indicating that the attachment of one or more MTs was continuously maintained at the shmoo tip ( A, Table S1, and Video 1). The attachment was not persistent in Δ cells when MTs switched to depolymerization ( B and Table S1; ; ). Similarly, the degree of persistence was reduced in Δ and Δ mutants (; Video 1, and Table S1). In these cells, detachment occurred when MTs switched to depolymerization. Therefore, like Kar3p, Bik1p is required to maintain depolymerizing MT plus ends at the shmoo tip. During mitosis, Bik1p localizes to both growing and shortening MT plus ends in the cytoplasm (). In pheromone-treated cells, Bik1p-3xGFP localized predominately to MT plus ends in the shmoo tip and marked growing and shortening MT plus ends (). The incorporation of new plus ends into the shmoo tip bundle could increase the fluorescence intensity over time (; and Video 2). Bik1p-3xGFP fluorescence intensity at the shmoo tip accumulated when the distance from the SPB to the shmoo tip decreased ( B). Conversely, the fluorescence intensity diminished when the SPB–shmoo tip distance increased ( C). This suggests that Bik1p may anchor shortening MT plus ends to the shmoo tip similarly to Kar3p (). A critical difference between the sliding cross-bridge and plus end models is the position of MT plus ends during nuclear congression. To determine the distribution of MT plus ends, we examined Bik1p-3xGFP or Kar3p-GFP during nuclear congression. In 82% of cells, these proteins localized as a focus in between both SPBs before and during nuclear congression ( = 23/28 cells). Using Bik1p-3xGFP to label plus ends, we acquired single plane images at 1-s intervals. Before nuclear congression, MT plus ends concentrated at the shmoo tips, and newly nucleated MTs could elongate and become incorporated into the shmoo tip bundle (Video 3). Once shortening was activated, SPBs moved in toward the position of the initial plus end interactions ( A and Video 3). The two sets of plus ends joined into a single Bik1p-3xGFP focus that persisted after nuclear congression began ( A, 286–391 s; arrows). SPBs moved in toward the Bik1p-3xGFP focus at 1.08 ± 0.72 μm/min ( = 6 cells), and nuclear congression could be completed in as little as 2 min. During nuclear congression, newly nucleated MTs could incorporate into or be released from the Bik1p-3xGFP focus but were not seen to cross over toward the other SPB (Video 3). Kymographs demonstrated that the two SPBs moved in toward the site of plus end interactions during nuclear congression ( = 5 cells; A, bottom). To ensure that Bik1p-3xGFP was labeling MT ends during nuclear congression, we mated GFP-Tub1p–expressing cells with Bik1p-3xGFP cells ( B) and imaged them in three dimensions over time. A single Bik1p-3xGFP focus was observed at the site of MT tip interaction ( B, 30–90 s; arrows). Occasional spreading of the Bik1p-3xGFP signal from a distinct focus to a more diffuse localization along the MT was visible at later time points ( B, 117.5–167.5 s). The position of the Bik1p-3xGFP focus did not significantly change as SPBs moved inward ( B, bottom). Kar3p-GFP also localized as a single focus between the two SPBs in mating cells (Fig. S1, available at ). Thus, during nuclear congression, MT plus ends from opposing SPBs interact, and depolymerization likely drives the SPBs together for nuclear fusion. If the plus end depolymerization model is the predominant mechanism for nuclear congression, the zone of MT overlap should be small or undetectable ( A and Video 4). Line scans of GFP-Tub1p during nuclear congression showed two peaks of fluorescence that corresponded to the SPBs with no detectable overlap zone ( B, left panels). Additionally, measurements of the fluorescence intensity before and during nuclear congression were analyzed ( C). If MTs slide past each other before nuclear congression, the fluorescence should be additive. However, after MT plus end interactions occurred ( C, graph; arrow), the GFP-Tub1p fluorescence did not increase as SPBs moved inward ( C, graph). As a positive control, MT overlap in the central spindle during anaphase of mitosis was analyzed. Line scans of the central spindle displayed three peaks: two representing SPBs ( B, graphs; arrows) and one at the midzone ( B, right panels). This demonstrates that overlap between one to two MTs () can be detected. The lack of MT overlap and the localization of Bik1p and Kar3p to a single focus indicate that plus end linkages are the predominant anchorage mechanism for nuclear congression. The karyogamy defect in the nuclear orientation mutant Δ is not as severe as other karyogamy mutants (). One possible explanation is that in the absence of nuclear orientation, oppositely oriented MT plus ends use a stochastic search-and-capture mechanism to interact and promote nuclear congression. To test this hypothesis, Δ cells with separated SPBs were examined after cell fusion ( and Video 5). MTs were seen to grow and shrink in the cytoplasm (, – min), and lateral MT interactions, which were visible in the same focal plane, did not move SPBs together (, 7.5–13 min). In contrast, nuclear congression did occur when MT tips contacted each other (, 17–18.5 min; = 5/6 cells). Therefore, MT plus end interactions, but not orientation to the shmoo tip, are required for nuclear congression in Δ cells. Bik1p is required for the formation or stability of MTs in mating cells () and is delivered to MT plus ends by Kip2p. Bik1p-3xGFP localized predominantly to the SPB in Δ cells with diminished localization at presumptive MT plus ends ( = 117/118 cells; A and Video 6). The low level of Bik1p-3xGFP at plus ends in Δ cells is not sufficient to promote persistent attachment of MTs to the shmoo tip (Table S1). However, despite the shorter length of MTs, there is no mating defect in Δ cells (Table S2; ). Thus, MT length is not the critical parameter for nuclear congression, and reduced levels of Bik1p on the plus ends appear sufficient to support karyogamy but not persistent attachment to the shmoo tip. To ensure that the delivery of Bik1p to the plus end was specific to Kip2p, Δ cells were also examined. Bik1p-3xGFP labeled both polymerizing and depolymerizing MT plus ends in the shmoo tip as well as the SPB of Δ cells ( A and Video 7), suggesting that the shmoo tip attachment and karyogamy defects observed in Δ cells do not result from Bik1p mislocalization. If MT length is not a critical factor governing nuclear congression, Bik1p could be required to promote persistent interactions between MTs. Alternatively, Bik1p could be a factor that directly links plus ends to promote nuclear congression. In Δ mutants expressing GFP-Tub1p, MTs were short and rapidly depolymerized back to the SPBs, limiting the ability of oppositely oriented MTs to interact (Fig. S2, available at ). This phenotype resulted in a large fraction of cells (10/13 cells) that did not perform nuclear congression. In those cells where nuclear congression did occur, MTs appeared to interact without rapidly shortening back to the SPBs ( B and Video 8). Despite the instability of MT interactions in Δ cells, the MTs could remain associated long enough to draw the opposing SPBs together ( B, 11 and 14 min). These data suggest that Bik1p promotes persistent MT–MT interactions or contributes to plus end linkage during nuclear congression. Kar3p may be the key component in MT plus end interactions during nuclear congression. In bilateral crosses of Δ mutants, the MTs were longer than in wild-type cells, and MTs did not interact to perform nuclear congression ( A and Video 9). Unlike Δ cells, MT plus ends passed each other without interacting in Δ mutants ( A, 0–29 min). No MT plus end interactions were observed when Δ strains expressing Bik1p-3xGFP crossed to GFP-Tub1p were imaged (Fig. S3), and nuclear congression was rarely seen in Δ bilateral crosses (18/19 cells with no congression). Thus, Kar3p is required to promote the persistent interaction of oppositely oriented MT plus ends during nuclear congression before the switch to coordinated MT depolymerization occurs. cells contain a constitutive point mutation in that results in rigor binding of the motor head to the MT and generates a semidominant mating defect (; ). In the sliding cross-bridge model, Kar3p should act along the length of MTs to promote MT interactions. kar3-1p localizes along the length of the MT instead of concentrating at the plus end (; ). If plus end interactions drive nuclear congression, failing to concentrate Kar3p at the plus ends may prevent nuclear congression from occurring. When was the only source of Kar3p in the cell, MTs did not interact, and nuclear congression did not occur (8/8 cells with no congression; see A for a representative example). Therefore, rigor-bound Kar3p is not sufficient for nuclear congression. In contrast, mating cells to wild-type cells resulted in a single bridge of GFP-Tub1p fluorescence between the two SPBs in ∼50% of cells ( B). The rigor-bound prevented complete MT depolymerization in these cells but did not prevent the persistence of MT–MT interactions (10/11 cells with no congression). A single focus of plus ends, visualized by Bik1p-3xGFP, did not form between the two SPBs ( = 6 cells; C and Video 10). Bik1p-3xGFP redistributed from a single focus in the wild-type cell before cell fusion to a diffuse localization along the MTs as cross-linking occurred after cell fusion ( C, 3.5–5.5 min). This indicates that when kar3-1p bound to the MT lattice encounters Kar3p, interactions are no longer restricted to the plus end. Thus, Kar3p localization at the plus end initiates MT–MT interactions during nuclear congression. #text Media composition and genetic techniques are described elsewhere (). Geneticin (Invitrogen) or hygromycin B (CellGro) were used at a concentration of 300 μg/ml. α-Factor (Sigma-Aldrich) resuspended in distilled water was used at a final concentration of 8 μg/ml. 5-Fluoroorotic acid (Toronto Research Chemicals) was used at a concentration of 1 g/L. strains and plasmids used in this study are listed in . strains and the Bik1p-3xGFP plasmid were provided by D. Dawson (Tufts University, Boston, MA) and D. Pellman (Dana Farber Cancer Center, Boston, MA), respectively. Deletion of genes was performed using the pFA6::MX vectors (; ). GFP-Tub1p () and CFP-Tub1p (provided by M. Segal, University of Cambridge, Cambridge, UK) was linearized with StuI before transformation, whereas Bik1p-3xGFP was linearized with NsiI before integration (). Cells were grown to early to midexponential phase in YPD (yeast extract/peptone/glucose) or appropriate selective media at 32°C except for Δ strains, which were grown at 25°C. All subsequent manipulations were performed at 32°C. For pheromone treatment, MATa cells were collected by centrifugation and resuspended in YPD supplemented with α-factor. Cells were incubated for 90–120 min, collected, and resuspended in distilled water before imaging. All images were acquired using spinning disk confocal microscopy as previously described () except where noted. Wide-field images were acquired with a 100× NA 1.4 differential interference contrast objective on an upright microscope (Eclipse E-600; Nikon) or an inverted microscope (TE-2000; Nikon). Image acquisition was performed as previously described (). The epifluorescence exposure time (2 × 2 binning) was 300–400 ms, whereas the differential interference contrast exposure time was 100–250 ms. Five-plane Z-series of 0.50-μm steps were acquired every 7–120 s and compiled into a single maximum projection image for each time point. Imaging processing and fluorescence intensity measurements were performed in MetaMorph software (Universal Imaging Corp.) as previously described (). γ adjustments for image presentation were performed in MetaMorph after data analysis was completed, and any brightness or contrast adjustments were performed in CorelDRAW10 (Corel Co.). Nuclear orientation to the shmoo tip was scored as wild-type if the GFP-Tub1p–marked SPB was either within or near the base of the mating projection. Aberrant nuclear orientation was recorded when the SPB was in the half of the cell body distal from the shmoo tip. The amount of time GFP-Tub1p fluorescence extended from the SPB to the shmoo tip during time-lapse imaging was recorded as the degree of persistence. Images were calibrated before analysis, and distances were recorded from projections of compiled 5-plane Z-series using either the single line tool or the calipers tool in MetaMorph to a linked Microsoft Excel spreadsheet. Nuclear congression was scored as defective if SPBs were visibly separated in the fluorescence image >0.5 μm after cell fusion occurred in still images. Successful nuclear congression in living cells was noted when both SPBs migrated toward each other and associated persistently. Rarely, time-lapse videos did not record SPB fusion and bud formation, possibly introducing a slight overestimate in the percentage of successful nuclear congression. 10 videos are included that display shmoo tip attachment (Video 1), Bik1p-3xGFP localization in the shmoo tip (Videos 2, 6, and 7), nuclear congression in wild-type cells (Videos 3 and 4), and MT behavior in karyogamy mutants (Videos 5, 8, 9, and 10). Additionally, three supplemental figures show Kar3p-GFP localization in wild-type cells (Fig. S1), MT behavior in Δ cells (Fig. S2), and Bik1p-3xGFP localization in Δ cells after cell fusion (Fig. S3). Table S1 shows measurements of nuclear orientation and cytoplasmic MT attachment to the shmoo tip. Table S2 shows nuclear congression efficiency among karyogamy mutants. Online supplemental material is available at .

The existence of distinct nuclear and cytoplasmic compartments is dependent upon the presence of a selective barrier called the nuclear envelope (NE). The NE consists of several structural elements (; ), the most prominent of which are the inner and outer nuclear membranes (INM and ONM, respectively). In most cells, these two membranes are separated by a regular gap of ∼50 nm, which is known as the perinuclear space (PNS). Periodic annular junctions between the two membranes form aqueous channels between the nucleus and the cytoplasm that accommodate nuclear pore complexes (NPCs) and, therefore, permit the movement of macromolecules across the NE. In addition to its connections to the INM at the periphery of each NPC, the ONM also exhibits numerous continuities with the ER, to which it is functionally related. In this way, the INM, ONM, and ER form a single continuous membrane system. Similarly, the PNS represents a perinuclear extension of the ER lumen. The final major structural feature of the NE is the nuclear lamina. This is a relatively thin (∼50 nm) protein meshwork associated with the nuclear face of the INM. The major components of the lamina are the A- and B-type lamins (). These are members of the larger intermediate filament family, and, like all intermediate filament proteins, they feature a central coiled-coil flanked by nonhelical head and tail domains (). The lamins are known to interact with components of the INM as well as with chromatin proteins. In this way, the lamina provides anchoring sites at the nuclear periphery for higher order chromatin domains. In mammalian somatic cells, there are two major A-type lamins (lamins A and C) encoded by a single gene, (in mice). The B-type lamins, B1 and B2, are encoded by two separate genes (, ; , ). Although B-type lamins are found in all cell types, the expression of A-type lamins is developmentally regulated (; ). Typically, A-type lamins are found in most adult cell types but are absent from those of early embryos. Mutations in the gene have been linked to a variety of human diseases (), many of which are associated with large-scale perturbations in nuclear organization. These observations have reinforced the view that the lamina is an important determinant of nuclear architecture and has an essential role in the maintenance of NE integrity. Despite their numerous connections, the INM and ONM are biochemically distinct. Proteomic studies have revealed the existence of at least 50 integral membrane proteins that are enriched in NEs. Many of these appear to reside within the INM (). Proteins become localized to the INM via a process of selective retention (; ; ). In this scheme, membrane proteins synthesized on the peripheral ER or ONM may gain access to the INM by lateral diffusion via the continuities at the periphery of each NPC using an energy-dependent mechanism (). However, only those proteins that can interact with nuclear or INM/lamina components are retained and concentrated. The recent identification of ONM-specific integral membrane proteins has raised some puzzling issues (; ; ). In particular, what prevents ONM proteins from simply drifting off into the peripheral ER? This question was originally addressed in , where the localization of Anc-1, a very large type II ONM protein involved in actin-dependent nuclear positioning, was shown to be dependent upon Unc-84, an INM protein (). Localization of Unc-84 itself was found to be dependent upon the single lamin (). Based upon these findings, and proposed a novel model in which Unc-84 and Anc-1 would interact across the PNS via their respective lumenal domains. In this way, Unc-84 would act as a tether for Anc-1. In mammalian cells, two giant (up to 1 MD) actin-binding proteins have been identified (variously termed NUANCE, nesprin 2 Giant [nesp2G], nesprin 1, enaptin, Syne 1, syne 2, and myne 1) as integral proteins of the ONM (; ; ; ; ). These belong to a very large family of proteins encoded by the nesprin 1 and 2 genes () and consist of a bewildering variety of alternatively spliced isoforms. , Anc-1, and Syne homology). This domain comprises a single transmembrane anchor and a short segment of ∼40 residues that resides within the PNS. One of the defining features of Unc-84 is a region of homology consisting of ∼200 amino acids with Sad1p, a protein that is associated with the spindle pole body (). This region of homology is known as the SUN domain (Sad1p and UNc-84) and is believed to extend into the PNS. Mammalian cells also contain several SUN domain proteins. At least one of these, Sun2, has been shown to be an INM protein with the appropriate topology in which the SUN domain is localized in the PNS (). It is tempting to speculate, based upon the model of , that SUN domain proteins function as tethers for ONM-associated nesprins in mammalian cells. Recently, have shown that localization of nesp2G to the ONM is dependent upon an interaction with another mammalian SUN domain protein, Sun1. In this study, we provide evidence that Sun1 is inserted into the INM in such a way that its SUN domain, like that of Sun2, faces the PNS. In this way, we can conclusively demonstrate that Sun1 does indeed have the appropriate orientation, as assumed by , for its COOH-terminal domain to interact with the nesp2G KASH domain. The NH-terminal region of both Sun1 and 2 face the nucleoplasm and interact with lamins. Surprisingly, our results indicate that Sun1 has a very strong preference for prelamin A. Sun1 is the only nuclear membrane protein described to date that exhibits such binding activity. This raises the distinct possibility that Sun1 may be involved in the targeting and assembly of newly synthesized lamin A. Finally, we demonstrate that Sun1 and 2 share some degree of functional redundancy and that both of these proteins cooperate in tethering nesp2G in the ONM. This tethering involves the establishment of molecular interactions that span the PNS and contributes to the remarkably regular spacing that is observed between the ONM and INM. Based upon our findings and upon those of , we are able to conclude that Sun1 and 2 function as key links in a molecular chain that connects the actin cytoskeleton via giant nesprin proteins to nuclear lamins and other components of the nuclear interior. We now refer to this assembly as the LINC complex (linker of nucleoskeleton and cytoskeleton). We have previously shown that Sun2 is an INM protein featuring an NH-terminal nucleoplasmic domain and a significantly larger COOH-terminal domain that is localized within the PNS (). This lumenal region of Sun2 contains a COOH-terminal SUN domain that is also found in Unc-84. The SUN domain is found in at least two additional mammalian proteins ( A): Sun1 (GenBank/EMBL/DDBJ accession no. ) and Sun3 (GenBank/EMBL/DDBJ accession no. ). Sun1 transcripts are present in a variety of tissues and cell types ( B). A comparison of Sun1 sequences in GenBank reveals that it exists in multiple, alternatively spliced isoforms. This conclusion is supported by Northern blot analysis, which reveals at least four or five discrete Sun1 transcripts in different tissues. We have not, however, surveyed these tissues for Sun1 isoforms. Immunofluorescence experiments using a polyclonal antibody raised against recombinant human Sun1 suggests that like Sun2, Sun1 is localized largely, if not exclusively, to the NE ( C). This is consistent with the appearance of Sun1 as well as Sun2 in a proteomic screen for NE-specific membrane proteins (). To further address the issue of Sun1 localization, we fused an HA epitope to the NH terminus of the largest isoform of mouse Sun1. After transfection into HeLa cells, the exogenous protein was found by immunofluorescence microscopy to be enriched at the NE ( D). At high expression levels, although still concentrated in the NE, HA-Sun1 began to appear in the peripheral ER ( D). Together, these observations confirm that Sun1 is a nuclear membrane protein. Similar experiments with HA-tagged Sun3 revealed a distribution that was more typical of an ER protein (not depicted). Furthermore, Northern blot analysis suggests that Sun3 is found primarily in the testis (not depicted). Consequently, all of our subsequent experiments focused on Sun1 and 2. The primary structure of mouse Sun1 reveals no NH-terminal signal sequence. However, two distinct hydrophobic domains, H1 and H2, are present ( A). H1 lies between residues 241 and 258, whereas the second and larger domain, H2, lies between residues 356 and 448. The presence of extended hydrophobic regions clearly raises questions concerning the topology of Sun1 within the nuclear membranes. An earlier study on Sun2 has shown that the SUN domain resides within the PNS (). Is this also the case for Sun1? To address this, we used in vitro transcription/translation of Sun1 (tagged and untagged) both in the presence and absence of microsomes ( B). Digestion of Sun1 translation products with proteinase K revealed the existence of a 65–70-kD protected fragment in samples containing microsomes. Proteinase K digestion in the presence of Triton X-100 to permeabilize the microsomal membranes resulted in the complete loss of the protected fragment. Given the size of this protected fragment and the location of potential transmembrane domains, the most reasonable orientation for Sun1 would place its COOH terminus, including its SUN domain, within the microsome lumen. By extension, the SUN domain should thus reside within the PNS in vivo. This orientation is supported by additional experiments described below. To examine the roles of the two hydrophobic segments in Sun1 membrane anchoring, we took advantage of naturally occurring splice isoforms. Searches of GenBank reveal several mammalian Sun1 cDNAs lacking sequences within the NH-terminal domain. Comparisons with genomic sequences indicate that at least one splice isoform of Sun1 lacks exons 6–8, corresponding to a deletion of residues 222–343 ( C). This particular Sun1 isoform is missing the first hydrophobic segment, H1. When this isoform (Sun1Δ6–8) was tagged with HA and transfected into HeLa cells, its localization at the NE was found to be indistinguishable from full-length Sun1 ( D). Translation of Sun1Δ6–8 in vitro in the presence of microsomes revealed a protease-resistant fragment identical in size with that derived from full-length Sun1 ( C). This finding presents us with two conclusions. First, it proves that the protected fragment must be derived from the COOH-terminal portion of the molecule. Second, if this first hydrophobic segment in Sun1 were to represent a transmembrane domain, its removal should logically alter the topology of the subsequent membrane-spanning sequences of Sun1 (i.e., H2) within the ER or nuclear membranes. This might potentially lead to the flipping of lumenal segments of newly synthesized Sun1 to the cytoplasmic aspect of the ER membrane or failure to insert H2 sequences into the membrane. Evidently, this does not happen in any detectable way. Therefore, our conclusion is that given this first hydrophobic sequence in Sun1 is dispensable with respect to membrane insertion, it does not function as an obligate membrane-spanning domain. Further secondary structure analyses of Sun1 using HMMTOP () indicate that the larger hydrophobic segment, H2, is capable of spanning the ER/nuclear membranes three times. Although this conformation for H2 still requires biochemical confirmation, it suggests that Sun1 is a multispanning protein in which the NH-terminal domain faces the cytoplasm, whereas the COOH-terminal domain (including the SUN domain) is localized to the lumenal space. To further examine Sun1 distribution and orientation in vivo, we prepared a form of mouse Sun1 that was tagged with an HA epitope at the NH terminus and a myc epitope at the COOH terminus (, HA-Sun1–myc). These analyses took advantage of the fact that low concentrations of digitonin can be used to selectively permeabilize the plasma membrane of cells while leaving the nuclear membranes and ER intact (). For these experiments, HeLa cells expressing HA-Sun1–myc were fixed with formaldehyde and permeabilized on ice for 15 min with 0.003% digitonin. The cells were then labeled with either rabbit anti-myc or rabbit anti-HA. After PBS washes to remove unbound antibodies, the cells were refixed, permeabilized with Triton X-100, and further incubated with mouse antibodies against either HA or myc. In this way, the first tag could be assayed for accessibility at the nuclear surface, whereas the second tag could be used to define the localization and expression level of HA-Sun1–myc. As shown in , the myc tag was never visible at the nuclear surface (or at any other location) after digitonin permeabilization regardless of the expression level of HA-Sun1–myc. The HA tag was also undetectable at the nuclear surface after digitonin permeabilization of cells expressing low levels of HA-Sun1–myc. At high expression levels, however, the tag was detectable at the nuclear surface and was associated with a cytoplasmic reticular structure corresponding to the peripheral ER. Clearly, the NH-terminal, but not the COOH-terminal, domain of ER-associated Sun1 is exposed to the cytoplasm. Altogether, these findings are consistent with the view that Sun1 is a component of the INM and that its COOH-terminal domain resides with the PNS. The aforementioned data, as well as our previously published study (), indicate that the NH-terminal domains of Sun1 and 2 are exposed to the nucleoplasm and, consequently, are accessible for interaction with nuclear components. Given the role that such interactions play in the appropriate targeting of INM proteins, it is not surprising that the lumenal domain of Sun1 is entirely dispensable with respect to Sun1 localization ( A). In this respect, Sun1 mimics Sun2 (). Additional experiments suggest that Sun1 and 2 share overlapping interactions. Overexpression of HA-Sun1 in HeLa cells causes displacement of endogenous Sun2 from the NE ( B). However, the converse is not the case (not depicted). The implication is that Sun1 and 2 share a subset of interactions that are required for Sun2 localization. Sun1 retention must be dependent upon additional binding partners. With what proteins might the Sun1 and 2 nucleoplasmic domains interact? Our initial thoughts were that the Sun proteins might associate with lamina components. To address this, we first adopted an in vitro approach to determine whether A- and B-type lamins could interact with the NH-terminal domain of either protein. We prepared a GST fusion protein containing the first 165 amino acids of the Sun2 sequence (Sun2N165). This represents most of the Sun2 nucleoplasmic domain. A similar, although slightly larger (the first 220 amino acids), fusion protein was also prepared using Sun1 sequences that encompassed the bulk of the nonalternatively spliced region of the nucleoplasmic domain (Sun1N220). This region of Sun1 exhibits significant sequence similarity to the NH-terminal domain of Sun2 over a region of ∼120 residues ( C), although Sun1 does display a unique 50-residue NH-terminal extension. Both of these fusion proteins as well as a GST control were used in pull-down experiments using in vitro translated lamins as targets. Four lamin species were used in these experiments ( D): lamin B1 (LaB1), lamin C (LaC), full-length lamin A (FL LaA), and mature lamin A. Full-length lamin A contains a CaaX motif and should be farnesylated in the reticulocyte lysate (). However, it does not undergo detectable COOH-terminal proteolysis because the microsome-free in vitro translation mix lacks the appropriate processing enzymes. The mature lamin A cDNA contains a premature stop codon at position 647. In this way, it mimics processed (i.e., mature) lamin A. As shown in D, GST-Sun2N165 bound all four lamin species, although the interaction with lamins B1 and C appeared barely more than the background observed with GST alone. Similarly, GST-Sun1N220 was also found to interact with all four lamin species. As was the case with Sun2N165, the interactions with lamins B1 and C were relatively weak. However, Sun1N220 showed a very strong preference for unprocessed (full-length lamin A) versus mature lamin A. To determine whether these in vitro interactions might have any relevance in vivo, we prepared HA-tagged versions of both Sun1N220 and Sun2N165. Upon introduction into HeLa cells, both proteins accumulated within the nucleoplasm ( E), although a significant cytoplasmic pool was always present. We also prepared a form of lamin A (preLaA) containing an L647R mutation. This lamin A mutant is cleavage resistant and, therefore, retains its farnesylated COOH-terminal peptide. Cotransfection of HeLa cells with preLaA along with either HA-Sun1N220 or HA-Sun2N165 lead to a dramatic decline in the nucleoplasmic concentration of both Sun protein fragments coincident with recruitment to the nuclear periphery. Lamin B1, on the other hand, had no such effect. These results indicate that the Sun proteins do indeed have the capacity to interact with lamin A in vivo. To determine whether this interaction with lamin A is required for Sun2 retention at the NE, we first performed immunofluorescence experiments on fibroblasts derived from both wild-type and -null mouse embryonic fibroblasts (MEFs). Although Sun2 was detected at the NE of all wild-type cells, in the majority of -null MEFs, Sun2 was dispersed throughout cytoplasmic membranes ( A). At first, these results do indeed implicate A-type lamins in the appropriate localization of Sun2 within the INM. However, there is clearly a minority of -null cells in which Sun2 is fully retained at the NE ( A, inset). Furthermore, loss of Sun2 from the INM is not reversed simply by introducing lamin A and/or C by transfection into -null MEFs (not depicted). Evidently, although A-type lamins could contribute to Sun2 localization, they are not the only determinants. This suggestion is reinforced by experiments in HeLa cells in which we eliminated A-type lamins by RNA interference (RNAi). After 48–72 h of RNAi treatment, lamin A/C was undetectable in many cells. However, the NE localization of Sun2 was barely affected ( C). In contrast to Sun2, we could find no evidence whatsoever for any lamin-dependent localization of Sun1. We took the approach of introducing HA-Sun1 into both wild-type and -null MEFs. In either case, exogenous Sun1 was always found at the nuclear periphery ( B). Similarly, in HeLa cells depleted of A-type lamins by RNAi, endogenous Sun1 always remained concentrated at the NE ( C). Regardless of A-type lamin content, we have never observed cells in which Sun1 is substantially mislocalized. Thus, although Sun proteins demonstrably interact with A-type lamins, this interaction is not required for their localization in the INM. Studies in have shown that the prototype SUN domain protein Unc-84 is required for the appropriate localization of Anc-1 in the ONM (; ). Giant nesprin family members are also known to localize to the ONM (; ; ) and, like Anc-1, feature a COOH-terminal KASH domain. Therefore, we examined the role that mammalian SUN domain proteins might play in the localization of one of these nesprin proteins (nesp2G; ∼800 kD). To accomplish this, we raised an antibody against the NH-terminal actin-binding domain (ABD) of nesp2G. The affinity-purified antibody recognized a very large (>400 kD) protein on Western blots of HeLa cell lysates ( A). At longer exposure times, lower molecular mass bands appeared, possibly corresponding to smaller nesprin 2 isoforms () or to degradation products. Immunofluorescence microscopy using digitonin permeabilization revealed that the anti-nesp2G antibody decorated the cytoplasmic surface of the NE ( B, mock). This labeling pattern was abolished by RNAi treatment of cells using nesprin 2–specific SmartPool oligonucleotides. Clearly, our antibody recognizes a very large ONM-associated nesprin 2 isoform, which is almost certainly nesp2G. Permeabilization of non–RNAi-treated cells with Triton X-100 yielded additional intranuclear labeling (unpublished data). This confirms a previous study suggesting that smaller nesprin 2 variants reside within the nucleus (). For all of our subsequent experiments, we used digitonin permeabilization to ensure that we were looking exclusively at nesp2G that was localized in the ONM. To explore the roles of Sun1 and 2 in nesp2G localization, we first adopted an RNAi-based approach. When we depleted HeLa cells of either Sun1 or 2 (), we could find little effect on the localization of nesp2G at the ONM (). However, codepletion of Sun1 and 2 led to the elimination of nesp2G from the ONM (). Quantitative analysis indicated an 80% reduction in the number of cells with detectable NE-associated nesp2G after codepletion of Sun1 and 2 versus mock RNAi treatment ( E). Ultrastructural analyses revealed changes in NE morphology in cells codepleted of both Sun proteins. Mock-treated cells displayed the usual uniform spacing between the ONM and INM of ∼50 nm. In the double RNAi-treated cells, however, the ONM was clearly dilated with obvious expansion of the PNS to 100 nm or more ( F). If SUN and KASH domain proteins form a molecular link across the PNS (), it should be possible to use a dominant-negative approach to break this linkage. In this strategy, we used almost the entire lumenal domain of Sun1 tagged at its NH terminus with HA (HA-Sun1L), which we introduced in soluble form into the lumen of the ER and PNS ( A). To accomplish this, we fused the signal sequence and signal peptidase cleavage site of human serum albumin onto the NH terminus of HA-Sun1L to yield SS–HA-Sun1L. To prevent its secretion, we fused a KDEL tetrapeptide to the COOH terminus of SS–HA-Sun1L, forming SS–HA-Sun1L–KDEL ( A). Synthesis of this chimeric protein in vitro in the presence of microsomes yielded a protein product of the appropriate size, which was resistant to digestion by proteinase K ( B). Clearly, the signal sequence directs HA-Sun1L to the microsome lumen. The shift up in molecular weight of latent HA-Sun1L–KDEL is likely a result of NH-linked glycosylation (Sun1 has two potential glycosylation sites in its lumenal domain). Upon transfection into HeLa cells, HA-Sun1L–KDEL was found to accumulate intracellularly within the peripheral ER and PNS. Examination of the distribution of nesp2G in transfected cells revealed that it was completely eliminated from the ONM ( C). EM analysis of these cells exposed clear dilation of the ONM and expansion of the PNS ( D). This phenotype is indistinguishable from that associated with Sun1/2 codepletion by RNAi ( F). In further experiments, we took advantage of a cDNA encoding a chimeric protein in which GFP is fused to the NH terminus of the nesprin 2 KASH domain (). This fusion protein localizes to the NE with the GFP exposed to the cytoplasm/nucleoplasm. The KASH domain is integrated into the NE, with its 40-residue COOH-terminal domain residing within the PNS. We prepared a HeLa cell line harboring GFP-KASH under the control of a tetracycline-inducible promoter (HeLaTR GFP-KASH). In the absence of tetracycline, GFP-KASH is present at low levels and localizes exclusively to the NE ( A, −Tet). After tetracycline induction, large amounts of GFP-KASH may be observed in both the NE and peripheral ER ( A, +Tet). Introduction of SS–HA-Sun1–KDEL into these cells leads to the complete loss of NE-associated GFP-KASH. Indeed, all of the GFP-KASH, regardless of expression level, appears to be recruited into vesicular structures, potentially as a prelude to degradation ( B, top; arrows). Conversely, when full-length Sun1 is introduced into tetracycline-induced cells, it leads to the recruitment of GFP-KASH from the peripheral ER to the NE ( B, middle and bottom; arrows). This is exactly what one would predict if Sun proteins function as tethers for KASH domain proteins. A similar effect was also observed when HA-Sun2 was introduced into the tetracycline-induced cells ( C), with GFP-KASH recruited to and stabilized at the NE. Altogether, these results can only be interpreted in terms of lumenal interactions between SUN domain and KASH domain proteins. Furthermore, they suggest that both Sun1 and 2 can interact in vivo with the nesprin 2 KASH domain, which is consistent with our RNAi results indicating that both SUN domain proteins contribute to nesp2G anchoring. To further define Sun1/2–KASH interactions, we set out to identify Sun–KASH complexes both in vivo and in vitro. For the former, we performed immunoprecipitation analyses of Sun2. As shown in E, a high molecular weight protein recognized by our antibody against nesp2G was found to coimmunoprecipitate with Sun2 from HeLa cell lysates. As a complement to these experiments, we synthesized SS–HA-Sun1–KDEL and GFP-KASH in vitro. Interactions between these proteins were then analyzed by immunoprecipitation using antibodies against GFP. As revealed in D, only when the SUN and KASH domain proteins were cosynthesized in vitro in the presence of microsomes could we detect HA-Sun1–KDEL in immunoprecipitates performed with the anti-GFP antibody. Together, all of these data provide strong evidence for the interaction, either direct or indirect, between Sun1/2 and nesp2G. Such an interaction spanning the PNS provides an obvious mechanism for the Sun1/2-dependent tethering of nesp2G in the ONM. We have shown that Sun1 is an INM protein with an NH-terminal nucleoplasmic domain of ∼350 amino acids and a larger COOH-terminal domain of ∼500 amino acids, including the SUN domain, that resides in the PNS. In this way, the topology of Sun1 matches that of another INM protein, Sun2, to which it is related. Based upon structural predictions, it is likely that Sun1 possesses three closely spaced transmembrane domains between residues 356 and 448. A separate hydrophobic region, H1, that is situated closer to the NH terminus does not appear to function as a membrane anchor. This conclusion is based upon the behavior of naturally occurring splice isoforms that lack this hydrophobic sequence. A third mammalian Sun protein, Sun3, is also an integral membrane protein with a lumenal COOH-terminal SUN domain and a relatively small cytoplasmic NH-terminal domain (unpublished data). In this way, Sun3 conforms to the general topological organization of other Sun family members. The organization of the lumenal domains of Sun1 and 2 bears some comment. The membrane proximal sequences of both proteins are predicted to form a coiled-coil. The implication is that these proteins may form homodimers. Given the number of residues within the Sun1 coiled-coil region, this could potentially project ∼25–30 nm into the PNS and would terminate in a pair of globular SUN domains. The coiled-coil domain of Sun2 is of a similar size. In both cases, the overall conformation of the Sun protein lumenal domain would be that of a flower on a stalk, which could potentially bridge the gap between the lumenal faces of the INM and ONM. The exact mechanism by which Sun1 and 2 are localized to the INM has yet to be resolved, although it is likely to involve the type of selective retention that has been observed for other INM proteins. What is clear is that the lumenal domain of both proteins is entirely dispensable for appropriate localization. This is exactly the reverse of what is observed for nesprin proteins (including nesp2G) of the ONM, where the lumenal and transmembrane domains (comprising the KASH domain) are essential for their retention at the nuclear periphery (). On the nucleoplasmic side of the INM, the Sun1 and 2 NH-terminal domains contain regions of similarity within the first ∼200 amino acid residues. This common NH-terminal region interacts, to a greater or lesser extent, with A-type lamins. In the case of Sun2, there is some evidence that A-type lamins might contribute to Sun2 localization in the INM. However, whether this requires a direct interaction with A-type lamins is less clear. Certainly, the concentration of Sun2 in the INM is A-type lamin independent in some cells. Furthermore, even in -null MEFs in which Sun2 is mislocalized, the mere reintroduction of A-type lamins fails to recruit Sun2 to the INM, at least within a period of ∼24 h (unpublished data). It seems more likely to us that A-type lamins may have indirect effects on Sun2, perhaps by altering the accessibility of chromatin proteins with which Sun2 might interact. In the case of Sun1, there is no evidence that lamins play any role in its localization to the INM. However, Sun1 displays an extremely robust interaction with prelamin A. Newly synthesized lamin A undergoes extensive COOH-terminal processing (). This involves farnesylation of the COOH-terminal CaaX (single letter code where C is cysteine, a is any amino acid with an aliphatic side chain, and X is any amino acid) motif followed by endoproteolysis to remove the aaX residues and carboxy methylation of the farnesyl cysteine. Once incorporated into the nuclear lamina, a second cleavage event after Y646 yields mature lamin A (). This cleavage of prelamin A at Y646 abolishes any strong interaction with Sun1. Because prelamin A exists only transiently in normal cells, it seems unlikely that its interaction with Sun1 could contribute to Sun1 localization. In our opinion, it is more likely that Sun1 might function in the organization of newly synthesized lamin A within the nuclear lamina. This suggestion is currently under investigation. have shown that the SUN domain protein Unc-84 is required for the localization of Anc-1 in the ONM. They have proposed a model in which the lumenal domain of Unc-84, which itself is retained in the INM through interactions with the single lamin, forms a complex with the lumenal KASH domain of Anc-1. In this way, Unc-84 and Anc-1 would provide links in a molecular chain that spans the PNS and connects the actin cytoskeleton to the nuclear lamina. Because similar SUN and KASH domain molecules are widely represented in the animal kingdom, we attempted to determine whether the Unc-84/Anc-1 paradigm was applicable in mammalian systems. We used a combination of RNAi and a dominant-negative form of Sun1 to test this model. We found that both Sun1 and 2 contribute to the tethering of nesp2G in the ONM. Elimination of either Sun protein by RNAi had little or no effect on nesp2G localization in HeLa cells. However, codepletion of Sun1 and 2 leads to the loss of nesp2G from the ONM. This was accompanied by separation of the ONM and INM, leading to expansion of the PNS. The implication here is that links between the Sun proteins in the INM and KASH proteins in the ONM help to maintain the remarkably regular spacing of the nuclear membranes. This view was reinforced by the findings that overexpression of a soluble form of the lumenal domain of Sun1 (SS–HA-Sun1L–KDEL) induced essentially the same phenotype: loss of nesp2G from the ONM and expansion of the PNS. In a complementary series of experiments, SS–HA-Sun1L–KDEL expression was also found to lead to the loss of GFP-KASH from the NEs of HeLa cells. This effect can only be accounted for by perturbation of lumenal interactions. Conversely, overexpression of full-length Sun1 (or Sun2) leads to the recruitment of GFP-KASH to the NE. All of these results are predictable on the basis of SUN domain proteins functioning as translumenal tethers for KASH domain proteins. Our final experiments demonstrated the existence of SUN–KASH complexes. Immunoprecipitates of Sun2 from HeLa extracts were found to contain nesp2G. Similarly, in vitro translation of SS–HA-Sun1L–KDEL and GFP-KASH leads to the formation of HA-Sun1L–KDEL–GFP-KASH complexes provided that microsomes were present in the translation mix. While this manuscript was in preparation, published a study that demonstrated a similar interaction between SUN domain and KASH domain proteins. Their results suggest, however, that rather than interacting with the SUN domain itself, the KASH domain actually bound to a region of the polypeptide chain that is proximal to the SUN domain. This region is present in our Sun1L-based dominant-negative mutant. All of our data suggest that the two Sun proteins are the key to the appropriate localization of nesp2G in the ONM. In contrast to previously published findings (), we could find no evidence of a role for A-type lamins. This is not surprising given that in our HeLa cells, the localization of Sun1 and 2 appear relatively insensitive to A-type lamin expression (or depletion). reached exactly the same conclusion with respect to Sun1. However, in -null MEFs, Sun2 is frequently lost from the NE. Given that expression levels of Sun1 appear to vary somewhat between different tissues, it is conceivable that in at least some cell types, nesp2G localization to the ONM might be sensitive to A-type lamin expression. Altogether, our findings and those of are entirely consistent with the model proposed by in which SUN and KASH domain proteins form a link across the PNS ( F). In addition, in , a similar mechanism may well operate in the tethering of Zyg-12, a NE protein that is required for dynein-mediated centrosome positioning (). As well as tethering ONM proteins, our data would suggest that SUN–KASH linkages further contribute to the structural integrity of the NE in maintaining the precise separation of the two nuclear membranes. Furthermore, given that the giant nesprins are actin-binding proteins, the SUN–KASH links provide direct molecular connections between the actin cytoskeleton and the nuclear interior. We now refer to this molecular chain as the LINC complex. Many studies have documented mechanical coupling between the nucleus and the cytoplasm. used microneedle-mediated deformation of the cytoplasm of cultured cells to demonstrate mechanical connections between integrins, cytoskeletal filaments, and nucleoplasm. More recently, were able to show that fibroblasts derived from -null mouse embryos have impaired mechanically activated gene transcription. In related studies, have shown that these same cells exhibited reduced mechanical stiffness and perturbations in the organization of the cytoskeleton. The existence of the LINC complex provides a basis for these various observations in that it may integrate the nucleus into a protein matrix that includes the cytoskeleton, extracellular matrix, and cell–cell adhesion complexes. This mechanical link not only provides structural continuity within and between cells, but it also allows for a direct physical signaling pathway from the cell surface to the nucleus, potentially facilitating rapid and regionalized gene regulation. HeLa cells and MEFs, both +/+ and −/− (), were maintained in 7.5% CO and at 37°C in DME (GIBCO BRL) plus 10% FBS (Hyclone), 10% penicillin/streptomycin (GIBCO BRL), and 2 mM glutamine. The following antibodies were used in this study: the monoclonal antibody against lamins A and C (XB10) has been described previously (). The monoclonal antibodies 9E10 and 12CA5 against the myc and HA epitope tags were obtained from the American Type Culture Collection and Covance, respectively. Rabbit antibodies against the same epitopes were obtained from AbCam. Rabbit antibodies against Sun1 and 2 were raised against GST fusion proteins as previously described (). The chicken antibody against the ABD of nesp2G was raised against an ABD-GST fusion protein by Aves Labs, Inc. Affinity purification of the IgY was performed in two stages. In the first step, an affinity column was prepared consisting of GST cross-linked to glutathione agarose (Sigma-Aldrich) using 40 mM dimethyl pimelimidate in 0.2 M borate buffer, pH 9.0, for 1 h at 4°C. 5 ml IgY solution was passed over this column (1-ml bed volume), and the flow through was collected. This flow through was applied to a second 1-ml column prepared from ABD-GST that was also cross-linked to glutathione agarose. Antibody bound to the column was eluted at pH 2.8 in 0.2 M glycine-HCl. The antibody eluate was neutralized with 3 M Tris, pH 8.8, and stored at 4°C with 1 mM sodium azide. Secondary antibodies conjugated with AlexaFluor dyes were obtained from Invitrogen. Peroxidase-conjugated secondary antibodies were obtained from Biosource International. Cells were grown on glass coverslips and fixed in 3% formaldehyde (prepared in PBS from PFA powder) for 10 min followed by a 5-min permeabilization with 0.2% Triton X-100. The cells were then labeled with the appropriate antibodies plus the DNA-specific Höchst dye 33258. For experiments involving selective permeabilization, the cells were first fixed in 3% formaldehyde. This was followed by permeabilization in 0.003% digitonin in PBS on ice for 15 min (). The cells were then labeled with appropriate primary and secondary antibodies. For certain double-label experiments, a single primary antibody was applied after the digitonin permeabilization. After removal of unbound antibody with three PBS washes, the cells were refixed for 5 min (in 3% formaldehyde) and subjected to a further permeabilization step in 0.2% Triton X-100. The second primary antibody was then applied followed by appropriate secondary antibodies. Specimens were observed using a microscope (model DMRB; Leica). Images were collected using a CDC camera (CoolSNAP HQ; Photometrics) linked to a Macintosh G4 computer running IPLab Spectrum software (Scanalytics). Cells grown in 35-mm petri dishes were fixed in 3% glutaraldehyde and 0.2% tannic acid in 200 mM sodium cacodylate buffer for 1 h at room temperature. Postfixation was performed in 2% OsO for 20 min. The cells were dehydrated in ethanol, lifted from the culture dish using propylene oxide, and infiltrated with Polybed 812 resin. Polymerization was performed at 60°C for 24 h. Silver-gray sections were cut using an ultramicrotome (Leica) equipped with a diamond knife. The sections were stained with uranyl acetate and lead citrate and examined in an electron microscope (model 7000; Hitachi). HeLa cells were depleted of Sun1, Sun2, nesprin 2, and lamins using appropriate SmartPool oligonucleotide duplexes (Dharmacon). Cells were exposed to each siRNA in the presence of OligofectAMINE (Invitrogen) precisely as recommended by the manufacturer. Cells were subjected to siRNA treatment for periods up to four days. However, most of our analyses were performed after 2–3-d treatments. Cells (siRNA- or mock-treated) grown in 35-mm tissue culture dishes were washed once in PBS and lysed in a buffer containing 50 mM Tris-HCl, pH 7.4, 500 mM NaCl, 0.5% Triton X-100, 1 mM DTT, 1 mM PMSF, and 1:1,000 CLAP (10 mg/ml in DMSO each of chymostatin, leupeptin, antipain, and pepstatin). The lysate was centrifuged for 5 min in an Eppendorf centrifuge at 4°C. Proteins in the supernatant were precipitated by the addition of TCA to a final concentration of 10%. The precipitate was washed with ethanol/ether and solubilized in SDS-PAGE sample buffer. Protein samples were fractionated on polyacrylamide gels (7.5, 10, or 4–15% gradient as appropriate; ) and transferred onto nitrocellulose filters (usually BA85; Schleicher and Schuell) using a semidry blotting apparatus manufactured by Hoeffer Scientific Instruments Inc. Filters were blocked and labeled with primary antibodies and peroxidase-conjugated secondary antibodies exactly as previously described (). Blots were developed using ECL (GE Healthcare) and exposed to X-OMAT film (Kodak) for appropriate periods of time. In vitro translations were performed in 25-μl reaction volumes using the TNT T7 coupled transcription translation system (Promega). Each translation reaction contained 20 μl TNT master mix and was programmed with 1 μl plasmid DNA at a concentration of 0.1 μg/μl. Labeling of translation products was accomplished by the inclusion of 10 μCi S Translabel (MP Biolabs). Where appropriate, up to 3 μl of canine pancreatic microsomes (Promega) was also included in each reaction. Translation reactions were assembled on ice before incubation at 30°C for 90 min. At the end of this period, translation mixes were further processed for in vitro binding studies (see In vitro pull-down…proteins) or were subjected to digestion with proteinase K in order to define Sun protein topology. Proteinase K digestions were performed on ice for periods of up to 1 h. Each digestion mix (10-μl total volume) contained 5 μl of the complete in vitro translation reaction, 1 μl proteinase K (from a 1-mg/ml stock solution), 1 μl of 10× compensation buffer (containing 0.5 M sucrose, 50 mM Tris-HCl, pH 7.6, and 200 mM potassium acetate), and, where appropriate, 1 μl of a 10% solution of Triton X-100. Termination of digestion was accomplished by the addition of 100 μl of 10% TCA to precipitate the proteins. Precipitates were washed in ethanol/ether, air dried, and dissolved in 25 μl SDS-PAGE sample buffer by incubation at 37°C. A mouse Sun1 cDNA (IMAGE clone ID 5321879) was obtained from Invitrogen. To generate Sun1 tagged at the NH terminus with an HA epitope, Sun1 cDNA flanked by 5′Sal1 and 3′Afl2 restriction sites was amplified by PCR using primers 5′-GAACGTCGACTTTTCTCGGCTGCACACGTACACC-3′ and 5′-CTGGCTTAAGCTACTGGATGGGCTCTCCG-3′. The PCR product was digested with Sal1 and Afl2 and inserted downstream of an HA tag sequence in the vector pCDNA3.1(−). This vector was prepared from pcDNA3.1(−) containing HA–lamin A () by digestion with Xho1 and Afl2. The resulting plasmid was pcDNA3.1(−)HA-Sun1. The soluble Sun1 lumenal domain construct SS–HA-Sun1L–KDEL, which was targeted to the ER and PNS, was prepared in three stages. The first step involved ligation of a double-stranded oligonucleotide encoding the entire NH-terminal signal sequence of human serum albumin, which was ligated into pcDNA3.1(−) between Nhe1 and Apa1 sites to yield pcDNA3.1SS. In the second intermediate step, the 5′ end of HA–lamin A was amplified by PCR using the pair of primers 5′-AATTGGGCCCGCTTACCCTTACGATGTACCG-3′ and 5′-ATATCTTAAGCAGCGCATCCGCCAGCCGGCTC-3′. The 787-bp PCR product was ligated downstream of the signal sequence in pcDNA3.1SS between the Apa1 and Afl2 sites to yield pcDNA3.1SS-HALaA5′. For the final step, the lamin A sequences were excised using Xho1 and Afl2. To prepare the Sun1 lumenal domain sequence incorporating a KDEL motif, PCR was performed using mouse Sun1 cDNA as a template and using the primers 5′-AGAGGGTCGACGATTCCAAGGGCATGCATAG-3′ and 5′-CTGGCTTAAGCTACAACTCATCTTTCTGGATGGGCTCTCCGTGGAC-3′. The resultant 1,403-bp product was cut with Sal1 and Afl2 and was ligated into the Xho1–Afl2 cut vector. The resulting plasmid was pcDNA3.1SS–HA-Sun1L–KDEL. All enzymes were obtained from New England Biolabs, Inc. Plasmids were introduced into HeLa cells using the Polyfect reagent as described by the manufacturer (QIAGEN). Transfections were normally performed in six-well plates. In brief, 1.5 μg plasmid DNA was combined with 100 μl of serum-free medium and 12 μl Polyfect. After a 10-min room temperature incubation, this mixture was combined with 600 μl of complete medium. The entire volume was then placed on the cells with an additional 1.5 ml of complete medium. The cells were then returned to the tissue culture incubator for 12–24 h. At the end of this period, the cells were processed as appropriate. GST-Sun1 and Sun2 fusion proteins were prepared using the plasmids pGEX-4T3Sun1N220 and pGEX-4T3Sun2NP. These plasmids were created by amplifying 5′ sequences of Sun1 and 2 by PCR. The Sun1 PCR product encoded the first 222 amino acids of the NH-terminal domain, whereas the Sun2 sequences encoded the bulk of NH-terminal domain of 165 amino acids. These PCR products were inserted into pGEX4T3 (GE Healthcare) between BamHI and EcoRI sites. The primers used for Sun1 were 5′-CGCGGATCCGACTTTTCTCGGCTGCAC-3′ and 5′-CCGGAATTCTTAGCGTGGTTTGAGAGTCCT-3′, whereas the Sun2 primer pair consisted of 5′-CGCGGATCCTCCCCGAAGAAGCCAGCGCCTCACG-3′ and 5′-CCGGAATTCTTAGGAGCCCGCCCGTGAGACGGC-3′. A single colony of Bl-21 cells transformed with either plasmid was grown overnight in 10 ml Luria-Bertani containing 100 μg/ml ampicillin and was induced with 0.1 mM IPTG for 4 h at 37°C. The cells were harvested by 15-min centrifugation at 4°C and at 3,200 in an Eppendorf table-top centrifuge. The bacterial pellets were resuspended by trituration in 1 ml of lysis buffer consisting of STE (150 mM NaCl, 10 mM Tris, pH 8.0, and 1 mM EDTA) containing 5 mM DTT and 0.25% sarkosyl (-laurylsarcosine). The suspension was sonicated to achieve maximum cell breakage and was centrifuged at maximum speed in a microcentrifuge for 10 min at 4°C. The supernatant was then transferred to a fresh microcentrifuge tube containing 30 μl of a 50% suspension in PBS of swollen glutathione agarose beads. The cleared bacterial lysate and beads were then incubated with continuous mixing at 4°C for 1 h. At the end of this period, the beads were washed three to five times with ice-cold STE and twice with ice-cold binding buffer (50 mM Tris, pH 7.4, 100 mM NaCl, 0.1% Triton X-100, and 1 mM DTT). 1.5 μg of plasmid DNA (pcDNA3.1HA–lamin A, –mature lamin A, –lamin B1, and –lamin C) was included in 25 μl TNT-coupled transcription/translation mixes containing 10 μCi S Translabel and incubated at 30°C for 90 min. 1 ul of each reaction was retained for the analysis of total translation products, whereas the remainder was incubated with 10 ul GST–agarose beads in 600 μl of binding buffer (containing 10 μg/ml chymostatin, leupeptin, antipain, pepstatin, and 1 mM PMSF) for 30 min at room temperature with constant mixing. After a low speed centrifugation at 800 , the supernatant was split into three tubes containing 5 ul GST–agarose, GST-Sun1N220, or GST-Sun2NP beads. The suspensions were then incubated for 45 min at room temperature with constant mixing. Finally, the beads were washed three to five times with binding buffer containing 1 mM DTT. After the final wash, the binding buffer was replaced with 20 μl SDS-PAGE sample buffer. The samples were subsequently fractionated by SDS-PAGE. The gels were stained with Coomassie blue R-250, impregnated with Amplify (GE Healthcare), dried, and exposed to X-OMAT film (Kodak). Double-stranded DNA probes consisting of the 5′ (650 bp) and 3′ (860 bp) fragments of mSun1 and the full-length human Sun3 (1,050 bp) were generated by PCR. Incorporation of [P]dCTP into the PCR products was accomplished by random priming using the Rediprime II Random Primer Labeling System (GE Healthcare) using 15 ng (in 45 μl Tris-EDTA buffer) denatured DNA. A mouse multiple tissue Northern blot (BLOT-2; Sigma-Aldrich) containing 2 ug polyA+ (per lane) RNA isolated from 10 different mouse organs (brain, heart, liver, kidney, spleen, testis, lung, thymus, placenta, and skeletal muscle tissues of BALB/c mice) was hybridized independently with each of the prepared probes, including one against glyceraldehyde-3-phosphate dehydrogenase (GAPDH; 1.4 ng/ml in 6 ml PerfectHyb Plus hybridization buffer) for 17 h (Sigma-Aldrich). Between each hybridization, the blot was stripped of the probe according to the manufacturer's instructions.

Cell polarization is important for processes as diverse as cell movement, axonal outgrowth, secretion of hormones, and cell differentiation. Polarization requires the vectorial delivery of secretory vesicles to, and their subsequent fusion with, the plasma membrane. The yeast is an excellent model organism to study mechanisms of polarization because it propagates through the polarized outgrowth of a bud. In , the polarized transport of secretory vesicles to the bud tip or, later in the cell cycle, to the mother-bud neck, depends on the actin cytoskeleton and the actin-based myosin motor protein Myo2p (; ). An important factor in this transport event is Sec4p, the founding member of the Rab branch of the Ras superfamily of small GTPases (; ). Rab proteins are so-called molecular switches, which cycle between an “on” (GTP bound) and “off” (guanosine 5′-diphosphate [GDP] bound) state. This GTPase switch is influenced by guanine nucleotide exchange factors (GEFs), which trigger the binding of GTP, thus activating the Rabs, and GTPase-activating proteins, which accelerate hydrolysis of the bound GTP to GDP, inactivating the GTPase (for review see ; ). Rabs are thought to accomplish their function by binding specific effector proteins in their GTP-bound state (for review see ; ). Sec4p was found in a screen for mutants that block exocytosis and accumulate secretory vesicles at a restrictive temperature (). Mutants of , the Sec4p GEF, show an accumulation of vesicles randomly distributed throughout the cell, implying that the activation of Sec4p by Sec2p directs the polarized delivery of secretory vesicles (). Supporting the view that activated GTP-Sec4p promotes Myo2p-dependent movement of secretory vesicles along actin cables, Sec4p was found to coimmunoprecipitate with Myo2p (). Although mutations in tightly block secretion (), mutations in actin () and depolarize secretion and cell surface growth but do not block secretion (; ). These results imply that Sec4p has at least one function in addition to its role in polarized vesicle delivery. A clue toward one such function came from the identification of Sec15p as a Sec4p effector (). Sec15p is a subunit of the exocyst, an octameric complex required for tethering secretory vesicles to the plasma membrane in preparation for fusion (; ). Sec8p, another exocyst subunit, was recently found to coimmunoprecipitate with Sec4p, suggesting that the entire exocyst complex acts downstream of Sec4p (). Furthermore, Sec4p is required for the assembly of the exocyst complex (). It has become increasingly clear that Rab GTPases interact with a variety of effectors in order to fulfill their functions in the cell (for review see ; ). Combined with the evidence that Sec4p regulates both the transport of secretory vesicles to and fusion with the plasma membrane, it seems likely that there are still more Sec4p effectors to be found. The identification of additional effectors would provide important new insights into the molecular function of Sec4p. Lethal giant larvae (lgl) was first identified as a tumor suppressor gene in the fly (for review see ; ). mutant flies were found to develop malignant tumors in the larval brain and imaginal discs that appear to result from a loss of cell polarity. Subsequently, homologues of lgl have been identified in many organisms, ranging from yeast to humans (for review see ). Because lgl family members are often found to be associated with the actin cytoskeleton, it has been argued that the observed polarity defects in cells bearing mutations are caused by defects in the actin cytoskeleton (; ; for review see ). However, data are accumulating that lgl family members function in polarized membrane traffic. lgl is required for the targeting of proteins to the basolateral membrane (), and the yeast homologues of lgl, Sro7p, and Sro77p (also known as Sop1p and Sop2p, respectively) are redundantly required for exocytosis (). Sro7p and Sro77p are 55% identical (; ). Both proteins were originally identified as high-copy suppressors of Δ (), a Rho GTPase required for actin cytoskeleton polarity and polarized exocytosis (; ; ). In this study, we describe the identification of Sro7p as an effector of Sec4p. Sro7p from yeast extracts as well as purified Sro7p interact specifically with the GTP-bound form of Sec4p. Sro7p was found to coimmunoprecipitate with Sec4p, demonstrating that the interaction we observed in vitro also occurs in vivo. Moreover, we found that Sro7p, Sec4p, and the t-SNARE Sec9p can form a ternary complex, suggesting that Sec4p regulates SNARE function through Sro7p. In agreement with this, genetic analysis shows that Sro7p shares a function with the exocyst downstream of Sec4p. We used an affinity purification approach to identify potential Sec4p effectors. GST-tagged Sec4p was purified from bacteria (). GST-Sec4p bound to glutathione beads was loaded with either GTPγS or GDP and, subsequently, incubated with wild-type yeast extract. As a control, GST bound to beads was also incubated with extract. The beads were washed several times, and bound proteins were eluted with high salt (see Sec4p affinity chromatography for details). Despite the presence of background proteins ( C, GST lane), differences could be readily detected between the protein bands in the GST-Sec4p-GTPγS and GST-Sec4p-GDP affinity chromatography samples ( C). The protein mixture of each sample was TCA precipitated and subjected to mass spectrometry analysis. Common background proteins were identified by their abundance in both samples and by a survey of published literature (). By definition, an effector binds more strongly to the GTP-bound form of a GTPase than to the GDP-bound form. One measure of the relative abundance of a protein in a protein mixture examined by mass spectrometry is the percentage of residues in the protein sequence that are represented by at least one peptide. We considered all proteins with an at least threefold higher coverage in the sample retrieved from the GTPγS-bound versus the GDP-bound form of Sec4p to be potential Sec4p effectors. Among the identified proteins meeting these criteria was Sro7p, a yeast member of the lgl family of proteins. As shown in , Sro7p displayed fivefold higher mass spectrometry coverage when Sec4p was GTPγS bound compared with its GDP-bound form. To validate these findings, we repeated the affinity chromatography with an extract from yeast cells expressing an integrated HA-tagged allele of Sro7p. The tag did not interfere with the functionality of the protein, as demonstrated by its ability to suppress the cold sensitivity of an Δ Δ double mutant and the salt sensitivity of an Δ mutant strain ( A and not depicted). Western blot analysis confirmed that Sro7-HAp migrates at its expected molecular weight ( B). An extract of this strain was incubated with GST-Sec4p in its different nucleotide-bound states or in the nucleotide-free state, GST-Ypt1p and GST. The latter two were used as specificity controls. Ypt1p is another member of the Rab GTPase family that is required for ER-to-Golgi transport in (). To test whether this method indeed distinguishes the different nucleotide-bound states and the nucleotide-free state of Sec4p, we used antibodies to probe for Sec2p and Sec15p. As mentioned above, Sec15p is the only previously documented effector of Sec4p, which was shown by two-hybrid analysis to bind preferentially to a hydrolysis-deficient (GTP locked) allele of Sec4p (). Sec2p is the exchange factor of Sec4p, which preferentially binds to the nucleotide-free state of Sec4p and, with lesser affinity, to the GDP-bound form of Sec4p (; ). As shown in C, Sec15p binds specifically to the GTPγS-bound form of GST-Sec4p (α-Sec15), whereas Sec2p binds with the highest affinity to the nucleotide-free state of GST-Sec4p (α-Sec2), thus confirming the nucleotide specificity of the method. To detect Sro7-HAp, these samples were analyzed by Western blotting with an α-HA antibody. Confirming the result obtained by mass spectrometry, Sro7-HAp was found to bind much more efficiently to the GTPγS-bound form of GST-Sec4p than to its GDP-bound or nucleotide-free state ( C, α-HA). Moreover, Sro7-HAp does not bind to GST-Ypt1p or GST ( C, α-HA), establishing the specificity of the Sro7p–Sec4p interaction. Approximately 2–4% of total Sro7-HAp was found to bind to GTPγS-Sec4p in this assay. The amount of Sro7p bound to GTP-Sec4p in this assay exceeds the amount of bound Sec15p ( C, compare α-Sec15 with α-HA). Interestingly, the same amount of Sro7-HAp binds to GTPγS-Sec4p when only half the amount of extract is used (not depicted), suggesting that the amount of activated Sec4p available for Sro7p binding is limiting in this assay. Altogether, our data indicate that Sro7p from yeast extract binds preferentially to the GTP-bound form of Sec4p, confirming the result obtained by mass spectrometry. We next determined whether purified Sro7p and Sec4p bind to each other in the absence of other proteins. Because Sro7p cannot be purified from bacteria (unpublished data), it was purified from yeast using a multistep procedure (see Purification of full-length Sec9 and Sro7 for details). This purification resulted in an Sro7p preparation that appears homogeneous by SDS-PAGE and Coomassie staining ( A) and dissociated from its binding protein, Sec9p (; unpublished data). As shown in B, purified Sro7p binds to GST-Sec4p preferentially in the presence of GTPγS. The amount of Sro7p bound to GST-Sec4p ranges from ∼5 to 20% of total Sro7p bound to GTPγS-Sec4p, and 1/20–1/10 of this amount bound to the GDP-bound or nucleotide-free Sec4p ( B). The observed binding is specific because Sro7p does not bind to any of the nucleotide-bound and -free forms of GST-Ypt1 or GST alone ( B and not depicted). Thus, purified Sro7p binds to Sec4p in its activated state. Given the strength of the observed interaction ( B) and the purity of Sro7p ( A), we conclude that this interaction is very likely to be direct because any copurifying factor would be substoichiometric. To test whether this interaction takes place in vivo, coimmunoprecipitation experiments were performed. Because of the low abundance of Sro7p (; ), Sro7p and HA-tagged Sec4p or, as a control, HA-tagged Ypt1p, were cooverexpressed in yeast. The HA-tagged Rab proteins were immunoprecipitated using an α-HA antibody, and coimmunoprecipitating Sro7p was detected by Western blotting with an α-Sro7p antibody. As shown in , Sro7p coimmunoprecipitates with HA-Sec4p. This interaction is specific because only a background amount of Sro7p is found to interact with Ypt1p; the signal requires the cooverexpression of both Sec4p and Sro7p; and no signal can be detected when the antibody is omitted from the precipitation reaction (). These data further support the conclusion that Sro7p binds directly to Sec4p because no third protein was overexpressed. Quantification of the interaction revealed that ∼2% of total Sro7p bound to GST-Sec4p (immunoprecipitated amount set to 100%). This relatively low amount might reflect the predominantly inactivated state of Sec4p in a yeast lysate. Together, our in vivo and in vitro data establish Sro7p as an effector of Sec4p. Sro7p and its paralogue, Sro77p, belong to the lgl tumor suppressor family (; ; ). Both proteins have been found to localize to the plasma membrane and to be required for exocytosis in yeast (; ; ). Moreover, it has been shown that Sro7p binds to the plasma membrane t-SNARE Sec9p and that this interaction is required for Sro7p's function (; ). Although no direct, GTP-specific interaction of Sec4p with SNAREs was previously found (), our data suggested the possibility that GTP-Sec4p indirectly signals to SNAREs via Sro7p. To explore this possibility, we performed in vitro binding assays using recombinant GST-Sec4p, Sec9p, and purified Sro7p. As shown in , Sec9p interacts with Sec4p, but only in the presence of GTPγS and Sro7p. Under the conditions used in this assay, ∼5–10% of the input Sro7p binds to Sec4p. Only background binding of Sec9p to GTPγS-Sec4p was observed when Sro7p was omitted from the binding assay, suggesting that Sec9p binds to Sec4p through Sro7p and implying that Sec4p and Sec9p use different binding sites on Sro7p (). The specificity of this interaction was demonstrated by the fact that no significant binding of Sro7p or Sec9p to GDP-Sec4p or GST was detected (, GST). Therefore, our data imply that GTP-Sec4p, Sro7p, and Sec9p are able to form a ternary complex. Our data suggest that Sec4p might play a role in SNARE regulation via its effector, Sro7p. The other known effector of Sec4p is Sec15p, a member of the exocyst complex (). The exocyst is required for the tethering of secretory vesicles to the plasma membrane (; ), a step that precedes the final SNARE-mediated fusion of those vesicles with the plasma membrane. If Sro7p and the exocyst act in converging pathways, each downstream of Sec4p, the overexpression of Sro7p might be expected to compensate for the loss of exocyst function, and, conversely, the loss of Sro7p would exacerbate the phenotype of exocyst mutants. As shown in A, overexpression of on a 2μ plasmid partially rescues the temperature sensitivity of a Δ mutant. In agreement with this, it has recently been published that the overexpression of rescues the exocytosis defect of the Δ mutant (). Interestingly, however, the level of suppression achieved by overexpression of is not as strong as that achieved by 2μ ( A, compare Δ 2μ with Δ 2μ ). We found that the deletion of increases the doubling time of a yeast strain about twofold compared with wild-type yeast at 30°C in synthetic complete (SC) minimal medium (225 ± 21 min vs. 125 ± 7 min; B). Overexpression of rescues this growth defect to nearly wild-type levels (145 ± 7 min), but Δ cells overexpressing still display a strong growth defect (190 ± 14 min; B). Similarly, 2μ does not suppress the lethality of Δ on yeast peptone dextrose (YPD) media ( C; ). One interpretation of these data is that Sec4p signals to other effectors in addition to Sro7p and the exocyst to achieve vesicle fusion. In support of our suggestion of converging functions for Sro7p and the exocyst, deletion of in a Δ strain further aggravates its growth defect ( A). We observed that the doubling time of a Δ Δ strain (360 ± 28 min) is almost twice that of a Δ single mutant strain (225 ± 21 min; B). This growth phenotype is accompanied by a significant drop in exocytosis in the Δ Δ mutant of ∼10% compared with the Δ mutant (73 ± 7% vs. 60 ± 5% of secreted invertase; C). This further implicates Sro7p and the exocyst in interrelated functions on the exocytic pathway. Because we observed a synthetic growth defect in the Δ Δ mutant strain, we decided to test whether the deletion of would have a similar influence on the growth of other exocyst mutant strains. For that purpose, we used a collection of temperature-sensitive late secretory mutant strains (, , , , , , , , , and ). We observed that , , , and in combination with Δ displayed slower growth compared with the single mutant strains at the permissive temperature of 24°C (Fig. S1, available at ). Interestingly, the most striking feature we observed is that combining a deletion of with almost all of the heat-sensitive mutants lead to a synthetic growth defect at 14°C ( and Fig. S1). Double deletion of and its paralogue has been found to lead to cold sensitivity of the resulting double mutant strain (; ), which demonstrated that their gene products share a common function. We found that the double mutants of the Rab GTPase Sec4p () or its GEF Sec2p () with Δ display synthetic lethality at 14°C ( and Fig. S1). This result further indicates a common function of Sro7p and Sec4p. The fact that Δ causes , a heat-sensitive mutant of the SNARE-interacting protein Sec1p, and , a mutant of the t-SNARE Sec9p, to become cold sensitive ( and Fig. S1) adds genetic evidence for a role of Sro7p in SNARE function in yeast. Interestingly, we also found that the deletion of leads to cold sensitivity in only a subset of the temperature-sensitive exocyst mutant strains ( and Fig. S1). These data might either reflect the relative strength of these mutant alleles at low temperature or indicate that only those subunits share a function with Sro7p and, therefore, add further evidence for a functional specialization of exocyst subunits within the complex (). Altogether, our genetic data support a possible role for Sro7p in transferring the signal of the Rab GTPase Sec4p to SNARE function in yeast. Although Sec3p is the only nonessential exocyst protein (), a previous study revealed that and also could be deleted if either Sec4p or Sec1p were overproduced (). Given our data that overproduction of Sro7p rescues the growth defect of a Δ mutant strain, we hypothesized that 2μ might also be able to overcome the lethality of Δ and Δ mutant strains. As shown in , overexpression of on a 2μ plasmid indeed rescues the lethality of both Δ ( A) and Δ ( B) mutant strains. Because we found that Sro7p is an effector of Sec4p with an exocyst-related function, it appeared likely that Sro7p would be required for the Sec4p-mediated rescue of Δ and Δ mutant strains (). Indeed, the deletion of reduces the growth of both Δ and Δ strains rescued by the overexpression of from a 2μ plasmid (; compare Δ 2μ or Δ 2μ with Δ Δ 2μ or Δ Δ 2μ ). We found that the doubling times of the Δ Δ 2μ (240 ± 21 min) or Δ Δ 2μ (242 ± 31 min) yeast strains are significantly increased compared with the yeast strains without additional deletion of (190 ± 14 min and 185 ± 21 min, respectively; ). In striking contrast, deletion of was found to only have minor influences on the growth of both Δ and Δ mutant strains rescued by overexpression of from a 2μ plasmid on either solid or liquid media (, compare Δ 2μ or Δ 2μ with Δ Δ 2μ or Δ Δ 2μ ). Therefore, these data provide genetic evidence that Sro7p functions downstream of Sec4p in an exocyst-related function. They further imply that although Sec4p signaling is upstream of Sro7p and the exocyst, Sec1p functions downstream of both (summarized in the model in ). Our results are also consistent with previous data that showed that Sec4p and Sec1p use different mechanisms for suppression of exocyst deletion mutations (). Given the fact that both Sro7p and the exocyst are Sec4p effectors and that the overexpression of suppresses the phenotypes of three different exocyst deletions, it appeared possible that Sro7p would be able to completely bypass the exocyst, which would imply that Sro7p and the exocyst have identical functions. We assessed this possibility by investigating whether the overexpression of (or, if necessary, in combination with ) would rescue the lethality of a Δ strain. Sec15p is the subunit of the exocyst that directly interacts with Sec4p (). It has been previously shown that the overexpression of and was able to suppress the temperature sensitivity of a mutant strain (Fig. S2 A, available at ; ; ; ). However, under all tested conditions (different media and temperatures), neither 2μ nor 2μ in combination with 2μ were able to suppress the lethality of a Δ strain (Fig. S2 B and not depicted). These genetic data indicate that Sro7p shares some, but not all, functions with the exocyst in yeast. Sec4p is a member of the Rab GTPase family that plays important roles in the yeast secretory pathway (; ). Only one effector for Sec4p was known before this study: the exocyst subunit Sec15p (). We described the identification by affinity chromatography of a second Sec4p effector, Sro7p, which is a member of the lgl family of tumor suppressors (; ; ). We have shown that Sec4p and Sro7p interact in vitro and in vivo and that the interaction requires Sec4p to be in its activated, GTP-bound state (–). Our data establish Sro7p as an effector of Sec4p and provide further support for a role of lgl family members in membrane traffic (see Introduction). Several previously published reports support our findings. Sro7p and its paralogue Sro77p were originally identified as high-copy suppressors of Rho3p (; ), which is a Rho GTPase required for actin cytoskeleton polarity and polarized exocytosis in yeast (; ; ). Another protein found in this screen (Sro6p) was subsequently identified as Sec4p. These genetic data suggest that the identified proteins positively influence cell polarity and/or polarized exocytosis, as had been shown for Sec4p (; ). Subsequently, Sro7/77p have also been found to be required for exocytosis in yeast (). Additionally, the overexpression of Sro7p was found to suppress the cold sensitivity of a mutant (), which suggested that Sro7p might act downstream of Sec4p function. Genetic data presented in this study indicate that the function of Sro7p partially overlaps with that of the exocyst (–), an eight-subunit vesicle-tethering complex required for tethering secretory vesicles to the plasma membrane (; ). We found that the overexpression of suppresses the growth defects of three different exocyst deletion mutants (Δ, Δ, and Δ; and ), extending recently published data (). We also showed that the deletion of impairs the growth and secretory function of a Δ strain (), which further implies that Sro7p and the exocyst function in concert. We recently established that the overexpression of either Sec4p or Sec1p can bypass the inviability of Δ or Δ strains (). Interestingly, we now demonstrate that the deletion of in Δ or Δ strains reduces growth only when is overexpressed (). If is overexpressed, no significant difference in growth can be detected upon deletion (). These data provide genetic evidence that Sro7p acts downstream of Sec4p in its capacity as a suppressor and demonstrate that Sro7p is involved in a Sec4p- and exocyst-related function (). Moreover, they further strengthen the findings by that indicate that Sec4p and Sec1p use different mechanisms in their suppression of the two exocyst mutants. Although , and can each act as a high-copy suppressor of ΔΔ, and Δ there is no simple hierarchy in the efficiency of suppression. Thus, suppresses Δ and Δ better than does ( and ), but is somewhat better at suppressing Δ than is (). Furthermore, suppresses Δ much better than does , but suppresses Δ much better than does (). Although this pattern is presently difficult to interpret, it does support the notion that different subunits of the exocyst fulfill distinct functions (), possibly in tethering and regulation of SNARE function. As mentioned above, the Sec15p subunit of the yeast exocyst is also a Sec4p effector (). If Sro7p and the exocyst have purely redundant functions downstream of Sec4p, the overexpression of one effector should overcome the loss of the other. Contrary to this prediction, we found that overexpression of either alone or in combination with did not suppress the lethality of a Δ mutant under all conditions tested (Fig. S2 B and not depicted). Thus, these data indicate that Sro7p and the exocyst have interrelated but not identical functions. In agreement with this, the overexpression of individual exocyst subunits did not suppress the salt sensitivity of the Δ mutant (Fig. S3, available at ) or the cold sensitivity of an Δ Δ mutant (). Although it might simply be that more than one subunit (i.e., a subcomplex) is required for suppression, another explanation for this finding is that Sro7p and Sro77p have important cellular functions beyond their role in exocytosis. In agreement with this, overexpression of either of the t-SNAREs, and , or the SNARE regulator failed to suppress the Δ Δ phenotype (). Sro7/77p have been shown to interact biochemically with the yeast type II myosin Myo1p, an interaction that appears to play a role in the remodeling of the actin cytoskeleton (), which is consistent with data concerning lgl family members from other organisms (see Introduction). Another explanation for the failure of overexpression of single exocyst subunits to suppress the salt or cold sensitivity of Sro7/77p mutants could be that Sro7p and Sro77p act downstream of some, but not all, aspects of exocyst function (; ). Rab GTPase activity has previously been implicated in the regulation of SNARE complex assembly (; ). Furthermore, genetic evidence suggested that plasma membrane SNAREs act in response to Rabs because the overexpression of has been found to suppress the growth defect of a mutant strain (). Nonetheless, a direct interaction of activated Sec4p and the exocytic SNAREs could not be detected (). We provide evidence in this study that GTP-bound Sec4p interacts with the plasma membrane SNAREs via Sro7p. The Sec4p effector Sro7p was found to interact with the t-SNARE Sec9p (), and recent data indicate that this interaction is important for Sro7p's function in secretion (). We have demonstrated that Sec9p can associate with Sec4p, but only in the presence of GTPγS and Sro7p (). This interaction provides a link between Rab signaling and SNARE function in yeast exocytosis. Together, two convergent signaling pathways from the Rab GTPase Sec4p appear to lead to vesicle fusion in yeast (). Secretory vesicles carrying activated Sec4p transmit a signal to Sec15p, which leads to exocyst assembly and vesicle tethering (). A recently documented association of Sec1p with the exocyst () potentially links this pathway to SNARE function. In a second branch of the pathway, Sec4p transmits a signal through the lgl family member Sro7p. Because Sec4p, Sro7p, and the t-SNARE Sec9p can assemble into a ternary complex, we propose that Sro7p conveys the signal from the Rab GTPase to SNARE function in yeast. These interactions may normally occur after vesicle tethering by the exocyst. There appears to be crosstalk between these two pathways because Sro7p has recently been shown to associate with the exocyst subunit Exo84p (). Further studies will be necessary to explore these proposals. strains used in this study are listed in . Standard techniques and media were used for growth, mating, sporulation, tetrad dissection, and yeast transformation (; ). Most yeast strains were created by mating appropriate parent strains. The resulting diploids were sporulated and dissected, and appropriate spores were picked. Strain NY 2587 was created by using a PCR-based method (). NY 2592–2595 were made by transforming combinations of plasmids pNB 529, pNB 833, or pNB 829 and either pNB 530 or pNB 1246 into NY 1210. Overexpression of HA-Sec4p, HA-Ypt1p, and Sro7p was confirmed by Western blotting. NY 2602 and 2603 were created by transforming pNB 142 or pB 745 into the appropriate diploid yeast strain before dissection. Strains NY 2609 and NY 2610 were made by streaking diploids (NY 2590 × NY 2478 or NY 2590 × NY 2476) onto SC medium containing 5-FOA (1 mg/ml final concentration) to select cells that lost the 2μ plasmid. Subsequently, these diploid cells were transformed with plasmid pB 745, were sporulated and dissected, and the appropriate spores were picked. For genetic analysis, at least two different spores were analyzed per experiment. For dot spot analysis, stationary yeast cultures were diluted to an OD of ∼0.3, and 10-fold dilutions were spotted onto YPD media or SC plates and incubated at the indicated temperatures for 2–3 (25°C and above) or 7–8 (14°C) d. Standard techniques were used for plasmid construction. The ORF was amplified by PCR from genomic DNA and subsequently inserted into plasmid pGEX5X-1 (GE Healthcare) to create the GST- expression vector (pNB 1245). The plasmid carrying GST- was described previously (). The plasmid carrying behind the promoter (pNB 1246) was created by inserting the ORF into plasmid pNB 530 ( promoter, terminator, and marker). The plasmids carrying (pNB 833) and (pNB 829) behind the promoter and the control plasmid pNB 529 have been described previously (). Plasmid pB 745 (2μ ) was described previously (). 2μ plasmids pNB 142, 680, 807, 690, 216, 328, 685, 148, 887, and 888 were used to overexpress , , , , , , , , or , respectively (P. Novick collection). The α-Sro7p antibody was described previously (). Clones 3F10 (rat monoclonal; Roche) and 16B12 (mouse monoclonal; Convance) were used as antibodies against the HA epitope. Rabbit polyclonal antibodies were used for the detection of Sec2p or Sec15p (P. Novick collection). A goat α-GST antibody (Sigma-Aldrich) was used to detect GST fusion proteins. 27 liters of wild-type yeast culture were lysed by homogenization in a microfluidizer (Microfluidics Corporation) in 140 ml of nucleotide-binding (NB) buffer (20 mM Hepes, pH 7.2, 100 mM KCl, 5 mM MgCl, and 1 mM DTT) containing 1 mM PMSF, 5 μg/ml pepstatin A, and complete protease inhibitor EDTA-free cocktail (Roche). Triton X-100 was added to 1%, and unbroken cells and debris were eliminated by centrifugation at 10,000 for 25 min at 4°C. The extract was dialyzed against NB buffer overnight. The concentration of the extract was adjusted to 40 mg/ml. GST fusion proteins were purified from BL21 strains (Novagen) as described previously (; ; ). For nucleotide loading, GST-Sec4p or GST-Ypt1p were incubated with a 200-fold excess of either GTPγS or GDP in NB for 2 h at 30–37°C. To obtain the nucleotide-free state, the proteins were incubated in NB buffer supplemented with 10 mM EDTA. Full-length Sec9p tagged with COOH-terminal His tag was purified from as described previously (). Sro7p with an NH-terminal protein A/tobacco etch virus (TEV) tag was isolated from lysates prepared from overexpressing yeast strains using affinity chromatography with IgG Sepharose beads. It was then eluted by cleavage of the protein A tag with TEV protease and subsequently purified by ion exchange chromatography to apparent homogeneity based on SDS-PAGE analysis. The protocol was adapted from . In brief, 100 μl GTPγS- or GDP-bound GST-Sec4p or GST-containing beads (∼750 μg of protein) were incubated with 25 ml of wild-type yeast extract (40 mg/ml) for 2 h at 4°C. After several washings, bound proteins were eluted with elution buffer (20 mM Hepes, pH 7.2, 1.5 M NaCl, 20 mM EDTA, and 1 mM DTT) supplemented with 5 mM of the opposing nucleotide. The proteins were TCA precipitated, dried, and analyzed by mass spectrometry. For the experiment in , 2.5 μl of loaded beads (∼15 μg) were incubated with 1 ml of a 20-mg/ml yeast extract (lysed using a French press; Sim-Amico Spectronic Instruments). Bound proteins were analyzed by Western blotting. Samples were suspended in 8 M urea and 100 mM Tris, pH 8.5, reduced with 100 mM TCEP, and cysteines were alkylated with 55 mM iodoacetamide. Lys-C was used to digest the proteins for 4 h at 37°C at a concentration of 1 μg/100 μl. CaCl was added to ensure tryptic specificity at 1 mM, and trypsin was used to digest the samples further at 1 μg/100 μl. The digests were then analyzed by μLC/μLC-MS/MS using an ion trap mass spectrometer (LCQ Deca; ThermoElectron). Multidimensional chromatography was performed online according to using the following salt steps of 500 mM ammonium acetate: 10, 25, 35, 50, 65, 80, and 100%. Tandem mass spectra were collected in a data-dependent fashion by collecting one full MS scan (m/z range = 400–1,600) followed by MS/MS spectra of the three most abundant precursor ions. The collection of resulting spectra was then searched against a database of yeast ORFs obtained from the genome database (release date 08/27/04) using the SEQUEST algorithm (). Peptide identifications were organized and filtered using the DTASelect program (). Filtering criteria for positive protein identifications in the Smt3p purification were the identification of two unique, fully tryptic peptides with Xcorr values >2.0 for +1 spectra, 2.2 for +2 spectra, and 3.75 for +3 spectra. Glutathione beads carrying 6 μM GST-Sec4p or GST were washed with 20 mM Tris, pH 7.5, 100 mM NaCl, and 1 mM DTT and were incubated in 20 mM Tris, pH 7.5, 100 mM NaCl, 5 mM EDTA, and 1 mM DTT in the presence of either 100 μM GTPγS, 100 μM GDP, or no nucleotide for 15 min at 25°C. Then, MgCl was added to a final concentration of 25 mM and incubated 45 min at 25°C. Binding assays were performed in binding buffer (20 mM Tris-HCl, pH 7.4, 140 mM NaCl, 5 mM MgCl, 1 mM DTT, and 0.5% Triton X-100). The final concentrations of GST/GST-Sec4p in the binding reactions were 4 μM, whereas concentrations of Sro7p or Sec9p were 1 μM. Samples were incubated at 4°C for 1 h. The beads were washed four times with binding buffer, and bound proteins were subjected to Western blot analysis. Sro7p and Sec9p were detected with rabbit α-Sro7 or α-Sec9 antibodies and α-rabbit IgG conjugated to AlexaFluor680 and were analyzed on an Odyssey Infrared Imaging System (LI-COR). For the experiment in , Sro7p–Sec9p mixtures (or Sro7p or Sec9p alone) were preincubated on ice for 1 h to allow the formation of binary complexes before their addition to the coated glutathione beads. Yeast strains 2,592–2,595 were grown overnight in yeast peptone media containing raffinose at 25°C to an OD of ∼0.4. Production of HA-Sec4p, HA-Ypt1p, and Sro7p was then induced by the addition of 2% galactose for 45 min. Cell pellets were resuspended in immunoprecipitation buffer (20 mM Hepes, pH 7.2, 150 mM KCl, 1 mM DTT, 6 mM MgCl, and 1 mM EDTA) containing 1 mM GTP, 1% NP-40, and protease inhibitors (1 mM PMSF, 2 μg/ml pepstatin A, 2 μg/ml chymostatin, 2 μg/ml aprotinin, 2 μg/ml leupeptin, and 2 μg/ml antipain). 2 g zirconia-silica beads were added, and the cells were lysed by homogenization in a Mini-Bead Beater (Biospec Products). Unbroken cells and debris were eliminated by centrifugation at 10,000 for 20 min at 4°C. The concentration of the yeast extracts was adjusted to 1 mg/ml, 30 μl of 50% protein G–Sepharose beads were added to 500 μl of extract, and the reaction was incubated for 90 min at 4°C. Beads were pelleted, 7.5 μl rat α-HA antibody was added to the supernatants, and the reactions were incubated overnight at 4°C. 6 μl of 50% protein G–Sepharose beads were added and incubated for another 30 min. The beads were pelleted, washed several times with immunoprecipitation buffer, and bound proteins were subjected to Western blot analysis. The monitoring of invertase secretion was performed as described previously (). Data were digitalized using a scanner (HP Scanjet 4570c; Hewlett Packard). Fig. S1 shows the growth of exocytic temperature-sensitive mutants combined with the deletion of compared with the single mutants and wild-type yeast at 24 and 14°C (data are summarized in ). Fig. S2 shows that the overexpression of and suppresses the growth phenotype of a mutant strain (A) but not the deletion of (B). Fig. S3 shows that the overexpression of single exocyst subunits does not suppress the salt sensitivity of an Δ strain.

Peroxisome biogenesis is a complex process involving >20 conserved peroxins (). Improper assembly of peroxisomes results in metabolic defects, such as the inability to perform fatty-acid oxidation, impairment in development, lethality in plants and mammals, and severe diseases in humans (). Import of matrix proteins (cargoes) occurs by two pathways, depending on the type of peroxisomal targeting signal (PTS) present on the cargo (). Most cargoes are targeted by a COOH-terminal tripeptide, the PTS1. An unrelated signal, the PTS2, is an NH-terminal nonapeptide with a loose consensus sequence used by a smaller subset of proteins including the β-oxidation enzyme β-ketoacyl CoA thiolase (Fox3p) in yeasts (). Targeting of PTS1 and PTS2 proteins to peroxisomes requires binding to soluble receptors, Pex5p and Pex7p, respectively, in the cytosol. Evidence supports an “extended shuttle” mechanism, where the soluble receptors are translocated together with the cargo and then recycled back to the cytosol after cargo unloading in the peroxisomal lumen (; ). After receptor–signal interaction in the cytosol, both pathways converge by binding to the “docking complex” at the peroxisomal membrane (Pex13p, Pex14p, and Pex17p). E3-like peroxins (Pex2p, Pex10p, and Pex12p) containing really interesting new gene (RING) domains are also necessary for cargo import (). Two AAA ATPases (Pex1p and Pex6p) and, in lower eukaryotes, an E2-like protein (Pex4p) are required for later steps of import (; ). Finally, in lower eukaryotes, an intraperoxisomal peroxin (Pex8p) was proposed to bridge the docking and the RING subcomplexes in a larger structure, the importomer (). In higher eukaryotes, targeting of PTS2 proteins by Pex7p requires the long isoform of the PTS1 receptor Pex5L (; ; ). In yeasts and fungi, PTS2 import does not involve Pex5p but requires other PTS2 auxiliary proteins. possesses redundant auxiliary proteins (Pex18p and Pex21p; ), but other organisms (, , and ; ; ; ) contain only one, named Pex20p. These auxiliary proteins are not involved in PTS1 import but share with mammalian Pex5L a common motif for interaction with Pex7p, suggesting a replacement of these auxiliary proteins in higher organisms by Pex5L through domain swapping (; ). The necessity and the equivalence of these auxiliary proteins in PTS2 protein targeting to peroxisomes is poorly understood and varies among organisms. Pex20p interacts directly with thiolase in a PTS2-independent fashion and helps in its oligomerization before translocation (), whereas Pex20p binds PTS2 sequences but does not assist thiolase oligomerization (). None of these interactions is observed for the homologues Pex18p and Pex21p (). In addition, there are conflicting reports concerning the ability of the latter to dock at the peroxisomal membrane. Finally, in view of the ability of Pex7p to enter peroxisomes, it is unclear whether the auxiliary peroxins are translocated during the import process. Overall, both the function and the properties of Pex20p-like proteins required further study. We functionally characterized Pex20p from and studied its subcellular localization and the regulation of its dynamics. Our results suggest that Pex20p behaves as a cycling peroxin. We propose a model for the dynamics of Pex20p during its import cycle involving a ubiquitin-dependent recycling mechanism. Putative PTS2 auxiliary peroxins of were investigated using a functional Pex7p–tandem affinity purification (TAP) construct. Pex7p-TAP was purified from oleate-grown cells after treatment of the extract with 0.5% digitonin. Mass spectrometry on the purified fraction and comparison of the data to the draft genome sequence of from Integrated Genomics revealed several proteins. These included the PTS2 protein Fox3p, the docking peroxin Pex14p, and a protein encoded by the ORF (16% of sequence covered), with 25% overall identity to Pex20p (). Cloning and sequencing of the gene showed a 969-nt ORF encoding a predicted protein of 323 residues (available from GenBank/EMBL/DDJB under accession no. ). Alignments of the predicted protein with Pex20p from other species revealed several conserved motifs (Fig. S1, available from ), including the putative Pex7p interaction domain found in other Pex20p-like proteins and in the long isoform of human PTS1 receptor (Pex5L), and three diaromatic pentapeptide motifs: Wxxx(F/Y) (; ). Based on these homologies and the data in –, the gene is referred to as and its product as Pex20p. The involvement of Pex20p in peroxisome biogenesis was analyzed by gene deletion. Although the strain grew at the wild-type level on glucose, it failed to do so on oleate ( A), suggesting its involvement in peroxisome biogenesis. This also indicates that no other genes of redundant function with exist in . Growth of in methanol medium also requires functional peroxisomes; however, because the β-oxidation enzyme Fox3p is the only known PTS2 cargo in and is not required for methanol degradation, the PTS2 receptor mutant grows in methanol (). Growth of like that of cells, was unaltered on methanol ( B), indicating that Pex20p is dispensable for import of PTS1 cargoes. The ability of the strain to import proteins into peroxisomes was monitored by fluorescence microscopy. The blue fluorescent protein (BFP) fused to a COOH-terminal PTS1 and a PTS2-targeted monomeric red fluorescent protein (mRFP) served as markers for each pathway. cells displayed a defect in PTS2 protein import because PTS2-mRFP remained cytosolic, whereas targeting of BFP-PTS1 to peroxisomes was not affected ( C). In agreement with data presented later indicating lack of direct interaction between Pex20p and thiolase, we suggest that Pex20p has a general role in peroxisomal import of PTS2-containing proteins, including but not limited to thiolase. After subcellular fractionation of a postnuclear supernatant (PNS), only a small amount of Fox3p was found in the organelle pellet (P200) of cells ( D), whereas it was more abundant in the P200 of the mutant strain complemented with a tagged version of Pex20p ( + HA-Pex20p). The subcellular location of the PTS1 cargo catalase was not affected. Neither overexpression of Pex20p in , nor of Pex7p in , restored growth in oleate (unpublished data), suggesting that these peroxins, although involved in the same pathway, have nonoverlapping functions. Finally, growth of the Δ Δ mutant on oleate was complemented by PpPex20p to the same extent as it was by the endogenous ScPex18p ( E). Therefore, Pex20p functionally substitutes for the PTS2 auxiliary proteins Pex18p and Pex21p. Collectively, these data demonstrate that Pex20p is the PTS2 auxiliary protein from . Because most interactions of Pex20p-like proteins were studied in artificial heterologous systems (; ), we investigated these in a homologous context. HA-Pex7p coimmunoprecipitated with some Pex20p ( A). The amount of Pex20p interacting with Pex7p in wild-type cells was lower than that in cells ( A), indicating that the Pex20p–Pex7p complex accumulates when docking is prevented. However, because PTS2 cargoes are also cytosolic in this docking mutant, this observation may reflect a role of the cargo in stabilizing the Pex20p–Pex7p interaction. Interactions of Pex20p with various proteins were examined by coimmunoprecipitation using an NH-terminal 3xHA-tagged version of Pex20p that complemented the strain on oleate. Coimmunoprecipitation of thiolase with HA-Pex20p was dependent on the presence of Pex7p ( B), suggesting that Pex7p mediates the Pex20p–thiolase interaction. In this respect, PpPex20p behaves like ScPex18p, which had no direct interaction with thiolase (). The amount of thiolase recovered in the immunoprecipitate was higher in the absence of Pex14p (unpublished data), presumably because the Pex20p–Pex7p–thiolase complex accumulates in the cytosol in the absence of docking to peroxisomes. We then investigated potential interactions with members of the docking complex, as the ability of Pex20p-like proteins to dock to peroxisomes is poorly documented. HA-Pex20p coimmunoprecipitated with members of the docking complex (Pex13p, Pex14p, and Pex17p; C). The interaction was particularly strong with Pex14p and Pex17p. In cells, Pex17p did not interact with HA-Pex20p ( C), showing that the Pex20p–Pex17p interaction observed was mediated by Pex14p. Interestingly, the interaction of HA-Pex20p with Pex13p was unaffected by the absence of Pex14p ( C). The interactions identified in this paper demonstrate that Pex20p is found in the cytosol as a complex with Pex7p and thiolase, with Pex20p interacting indirectly with thiolase through Pex7p. Also, Pex20p docks at the peroxisomal membrane through independent interactions with Pex13p and Pex14p. As judged by the yeast two-hybrid technique, Pex7p interacted with the full-length Pex20p construct ( A) as well as with its COOH-terminal half (residues 146–323), predicted to contain the Pex7p binding site (Fig. S1). Mutation of the conserved Ser residue present within this region (S280F) disrupted the interaction with Pex7p ( A) as described earlier (; ; ). Further truncations (constructs spanning either aa 146–260 or 260–323) abolished the interaction (unpublished data), suggesting that residues 276–296 are insufficient for Pex7p binding. Although none of the Pex13p constructs were suitable for yeast two hybrid (unpublished data), Pex14p interacted with Pex20p(1–146) and Pex20p(80–146) ( B). Interaction between Pex20p and Pex14p has been described, but this was bridged by Pex7p (; ). This is not the case here because Pex14p and Pex7p interact through different regions of Pex20p. The diaromatic pentapeptide (Wxxx[F/Y]) repeats present in Pex5p from various organisms bind to Pex14p and Pex13p (; ). The Pex14p-interacting region (aa 1–146 of PpPex20p) contains three such motifs (aa 89–93, 102–106, and 141–145). Site-directed mutagenesis was performed on each of these sites (W89G, W102G, and W141G). The construct Pex20p(80–146; W89G) failed to interact with Pex14p, whereas Pex20p(80–146; W102G) and Pex20p(80–146; W141G) still interacted with Pex14p ( C). Therefore, only the first Wxxx(F/Y) motif of PpPex20p is crucial for Pex14p interaction, with the other motifs being either not involved or redundant. Only the NH-terminal fragment (aa 1–170) of Pex14p interacted with Pex20p ( D). Interestingly, this region is involved in Pex5p binding (), suggesting that the same region of Pex14p interacts with Wxxx(F/Y)-containing proteins, such as Pex5p or Pex20p. Finally, although Pex7p interacted with Fox3p, there was no interaction between Pex20p and Fox3p (unpublished data). Although interactions between peroxins Pex18p and Pex21p and Fox3p are mediated by Pex7p in the yeast two-hybrid system (; ), Pex20p interacts with Fox3p in (; ) and presumably in (). As indicated in B, thiolase was coimmunoprecipitated with HA-Pex20p but only in the presence of Pex7p. The interaction data ( E) indicate a strong resemblance between Pex20p and Pex5p interaction maps because both proteins interact with Pex14p through the same motifs. Furthermore, the domain involved in Pex7p binding possesses strong similarities with that of human Pex5L. Pex20p and Pex7p functions appear to be tightly linked, but the precise function of Pex20p in PTS2 import is still unclear. We assessed its role in the various steps leading to thiolase import into the peroxisome, namely receptor–cargo binding, receptor docking to the peroxisomal membrane, and receptor–cargo translocation. Thiolase coimmunoprecipitated with HA-Pex7p independently of Pex20p ( A), indicating that Pex20p is not essential for Pex7p–thiolase interactions. From the aforementioned data ( and ) showing interactions of Pex20p with members of the docking complex and work on Pex18p (), we assumed that Pex20p might help Pex7p in its docking to peroxisomes. Instead, we observed that Pex13p, Pex14p, and Pex17p coimmunoprecipitated with HA-Pex7p, regardless of the presence of Pex20p ( B). A similar conclusion was made in where Pex7p still interacted with the docking peroxins even in the absence of Pex18p or Pex21p (). However, contrary to a previous study (), the interaction between Pex20p and the docking subcomplex was independent of Pex7p ( B). In addition, the subcellular distribution of Pex7p between the cytosol and the organelle pellet did not change drastically in the absence of Pex20p ( C). Conversely, the distribution of Pex20p in the supernatant and pellet fractions was not altered significantly by the presence or absence of Pex7p (unpublished data). Collectively, these data show for the first time that association of Pex7p or Pex20p with the peroxisomal docking subcomplex can be independent of the other and that each protein does not significantly affect the peroxisomal localization of the other. Surprisingly, the protease protection assay performed on the P200 fraction from cells showed that Pex7p was protease resistant, as was the PTS1 enzyme catalase ( D). This suggests that Pex7p translocates into peroxisomes independently of Pex20p, a feature that has not been described previously and whose physiological role is unknown. Our results contrast with those obtained in where Pex7p depends on Pex18p and Pex21p for its peroxisomal localization (). Both and are deficient in thiolase import and unable to grow on oleate. We investigated whether the organelle-associated fraction of thiolase in cells () reflects a low efficiency, or rather an absence, of import. As shown in D, thiolase was protease sensitive. Therefore, although Pex7p was imported into the peroxisomes of cells, thiolase remained on the surface of peroxisomes. Because also fails to import thiolase into peroxisomes (), we conclude that Pex20p functions in the translocation of the Pex7p–cargo binary complex, although Pex7p alone does not require Pex20p to go into peroxisomes. A polyclonal antibody to Pex20p ( A) was used to study Pex20p distribution after cell fractionation. One third of the cellular Pex20p pool could be pelleted, whereas two thirds were cytosolic ( B). This membrane-associated Pex20p sedimented at the same density as peroxisomes in a density gradient ( C), whereas the rest of it was cytosolic and at the top of the gradient. A protease protection assay performed on the P200 fraction from + Pex20p-GFP cells showed that Pex20p-GFP behaved like the intraperoxisomal peroxin Pex8p, which is protected from external protease, unless detergent was added ( D). Therefore, Pex20p behaves as both a cytosolic and peroxisomal peroxin, similar to the cycling peroxins Pex5p and Pex7p (; ; ). We established the requirements of Pex20p peroxisomal localization using a functional, COOH-terminal GFP-tagged version of Pex20p, driven by its own promoter. As shown in , much of the Pex20p-GFP was cytosolic in wild-type cells, although a signal was detected in structures that colocalized with a peroxisomal membrane marker (Pex3p-mRFP). Absence of Pex14p led to mislocalization of Pex20p-GFP to the cytosol (), suggesting that the presence of the docking complex is a prerequisite for the peroxisomal localization of Pex20p-GFP. Surprisingly, in any of the RING peroxin mutants (, Δ, and Δ), Pex20p-GFP accumulated in a bright dot that colocalized with a marker containing a peroxisomal membrane PTS (mPTS)–mRFP (). These data strongly indicate that Pex20p-GFP does not require the RING complex to locate to peroxisomes; instead, it appears that the RING peroxins are involved in relocating Pex20p from the peroxisome to the cytosol. While the localization of Pex20p was being addressed in mutants of the late steps of import (namely , , and ), we were surprised to observe that no Pex20p was detectable in these strains ( A) as previously described for Pex5p (; ). This down-regulation of Pex5p is conserved between plants (), and humans (; ), but the underlying mechanism is unknown. Other mutants contained amounts of Pex20p that were comparable to those of wild-type cells, although a decrease of Pex20p levels in the strain was noted, which remains unexplained. The low steady-state level of Pex20p in recycling mutants was not affected by further deletions affecting the vacuolar proteases Pep4p and Prb1p (unpublished data). Therefore, Pex20p and Pex5p are regulated through similar mechanisms during the import cycle. To explain these common regulatory features, we hypothesized that sequence similarities in both proteins would confer a similar regulation. Alignment of Pex20p and Pex5p sequences revealed a conserved domain involving 25 out of 35 amino acids at their NH termini ( B), including a conserved lysine residue. Because no function was assigned to this domain, we assessed its importance using truncated proteins. NH-terminal deletions of Pex20p were expressed in cells under the control of the endogenous promoter, and the ability to grow on oleate was checked. Deletion of the first 16 residues did not affect growth, but an effect was observed for further deletions (Δ1–19, Δ1–22, and Δ1–31; C), although the proteins were expressed (not depicted). Mutation of the conserved lysine present in this domain (Pex20p-K19R and Pex5p-K22R) had no effect on the protein function ( D), suggesting that other residues within this domain may be essential or that the protein structure is affected when the whole region is missing. To finally address whether the NH-terminal domain shared by Pex5p and Pex20p is the basis for the common down-regulation observed in late-steps mutants, we expressed Pex20p-K19R in the Δ double mutant strain and observed that this mutation rendered the protein stable ( E), whereas Pex20p was undetectable in cells. Also, when Pex20p-GFP was expressed in , Δ, or , only small amounts of the fusion protein were detected, showing that Pex20p and Pex20p-GFP behave similarly. However, the steady-state level was comparable to that of the wild-type strain when Pex20(K19R)-GFP was expressed instead ( F). Therefore, K19 is essential for Pex20p down-regulation in , , or mutants. In the recycling mutants of , ubiquitylated species of Pex5p are detected (; ; ). We investigated whether the down-regulation of Pex20p observed in these mutants of results from an unusually fast degradation by the ubiquitin–proteasome system (UPS). Because a nearly complete absence of Pex20p is noticed after overnight induction in oleate ( A), we studied the steady-state level of Pex20p at earlier time points. Interestingly, after only 6 h of induction ( G), no change in Pex20p-GFP steady-state level was noted in the strain, but higher molecular mass bands were detected. Noticeably, these bands depended on the presence of the K19 residue ( G). These results define a new, essential NH-terminal domain in Pex20p, which is conserved in Pex5p and whose conserved lysine is essential for Pex20p down-regulation in recycling mutants, likely via the UPS. Because K48-branched polyubiquitylation of a protein acts as a signal for its degradation by the UPS, we investigated whether constitutive overexpression of the ubiquitin mutant, Ub (K48R), in wild-type cells would affect Pex20p regulation. First, we observed that it affected the ability of the strain to grow on oleate medium but not on glucose medium ( A). This suggests that polyubiquitylation is essential for peroxisome biogenesis, perhaps by interference with the action of Pex4p. Surprisingly, we also observed higher molecular mass species of Pex20p in crude extracts from this strain, whose presence was dependent on the K19 residue ( B), mimicking the situation obtained in cells in the early stages of induction ( G). We assessed whether these species were polyubiquitylated forms of Pex20p. Denatured extracts of wild-type cells coexpressing Pex20p-GFP and myc-tagged Ub(K48R) were immunoprecipitated with a monoclonal anti-GFP antibody. This allowed recovery of Pex20p but also of higher molecular mass species that were also immunodecorated with the anti-myc antibody, indicating they are truly ubiquitylated forms of the protein ( C). Also, omission of the proteasome inhibitor MG-132 and the isopeptidase inhibitor -ethylmaleimide during sample preparation did not allow detection of these species (unpublished data), supporting the idea that these bands represent ubiquitin conjugates of Pex20p-GFP that would normally be degraded by proteasomes. When we addressed the subcellular localization of polyubiquitylated species of Pex20p-GFP by differential centrifugation, they were found exclusively in the pellet (P200) fraction ( D). Overexpression of Ub(K48R) in wild-type cells led to a dramatic accumulation of Pex20p-GFP in the organelle pellet. This observation was confirmed by fluorescence microscopy experiments, where Pex20p-GFP colocalized with peroxisome remnants when Ub(K48R) was overexpressed ( E). At least part of the pelletable Pex20p was protected from external protease, unlike Pex17p ( F). We exploited the apparent lack of down-regulation of Pex20p(K19R) in recycling mutants () to study its subcellular localization. Differential centrifugation analysis and fluorescence microscopy experiments showed a sharp increase in the amount of peroxisome-associated Pex20p(K19R)-GFP () in these mutants as compared with wild type. Pex1p, Pex6p, and Pex4p are thus essential for the proper distribution of Pex20p between the organelles and the cytosol, analogous to their proposed role in recycling of Pex5p (; ; ). It was recently proposed that the NH terminus of human Pex5p is required for its recycling to the cytosol (). We therefore investigated the effect of an NH-terminal truncation of Pex20p-GFP on its subcellular localization. Pex20(Δ1–19)p-GFP, the longest truncated construct that fails to complement ( C), accumulated nearly exclusively in the organelle pellet as determined by differential centrifugation ( C) and colocalized in fluorescence microscopy with Pex3-mRFP ( D). Deletion of this domain might abolish the function because the cytosolic redistribution of Pex20p is affected. Our characterization of PpPex20p confirms the necessity of this class of proteins for Pex7p-mediated peroxisomal import of PTS2 cargoes because cells fail to grow on oleate and have a PTS2 import defect (). However, the few studies (; ) on the Pex20p from other species were either limited or done in artificial heterologous systems. Our systematic studies of the location, interactions, and steps of thiolase import into peroxisomes reveal new insights regarding the role of PpPex20p (). Pex20p might stabilize a thiolase–Pex7p complex before import or, more likely, act as a chaperone to facilitate its translocation across the peroxisomal membrane. Interestingly, Pex7p was translocated into peroxisomes even in cells ( D). This raises the existence of futile cycles in which Pex7p could be translocated without cargo and is consistent with our previous conclusion that cargo-binding mutants of Pex7p were partially peroxisomal like wild-type Pex7p (). Our experiments support a model in which the peroxisomal import of PTS2 is mediated by the docking, import, and recycling steps of Pex20p itself (; discussed on the next page). Pex20p interactions with other peroxins resemble those of the PTS receptors. It interacts with members of the docking complex ( and ), especially Pex14p, through its Wxxx(F/Y) repeats whose presence was noted in PTS2 auxiliary proteins (; ; ), but we show for the first time their actual involvement in docking to peroxisomes. This provides a structural clue to the question of why both PTS pathways converge at this docking site: they possess related motifs allowing interactions with the same peroxins. Lack of Pex14p did not prevent interaction with Pex13p ( C), nor did it prevent docking of Pex20p to organelles (unpublished data). Thus, Pex20p possesses two docking sites on peroxisome membranes, as noted for Pex7p and Pex5p (; ; B). Although Pex18p and Pex21p interact with Pex13p and Pex14p in a Pex7p-dependent fashion (), our data show that Pex20p docks to peroxisomes independently of the PTS2 receptor ( B). Several lines of evidence indicate that a fraction of Pex20p is peroxisome associated, with some of it being present inside peroxisomes (or fully embedded in the membrane; ). Interestingly, Pex20p also interacts with the intraperoxisomal protein Pex8p (unpublished data), as described in (). In the context of the extended receptor shuttling model, the dual localization of Pex20p to both the cytosol and peroxisomes suggests that it too is a shuttling peroxin, like Pex5p and Pex7p (; ). Interestingly, the E3-like RING peroxins (Pex2p, Pex10p, and Pex12p) were not required for the peroxisomal localization of Pex20p (). Their absence or deletion of the NH-terminal 19 amino acids of Pex20p (; and ) led to an increase in peroxisome-associated Pex20p, and the protein was inaccessible to the (cytosolic) ubiquitin-dependent degradation pathway. This indicates a role for the RING peroxins and this NH-terminal sequence in Pex20p relocation to the cytosol ( A), rather than in Pex20p translocation to the matrix. Similarly, the absence of peroxins involved in the late steps of protein import (the E2 Pex4p and the AAA ATPases Pex1p and Pex6p) caused a mostly peroxisomal localization of Pex20p () when ubiquitin-dependent degradation was abolished by the Pex20p-K19R mutation (). However, in these same mutants, peroxisome-associated Pex20p was susceptible to ubiquitylation and degradation (– G), most likely on the cytosolic side of the peroxisomal membrane. Therefore, Pex1p, Pex4p, and Pex6p are not involved in Pex20p import into peroxisome but rather in its recycling from peroxisomes to the cytosol ( B). Epistasis analysis of Pex20p stability is consistent with the action of RING peroxins before that of recycling peroxins (unpublished data). This dependence of Pex20p recycling on Pex4p, Pex1p, and Pex6p is remarkably similar to that for Pex5p (; ). During completion of this paper, a study was published that indicates a role of Pex1p and Pex6p in the recycling of ubiquitylated Pex5p from the peroxisomal membrane (), in agreement with our data on Pex20p. Additionally, both Pex5p () and Pex20p () need their NH-terminal regions for recycling. This underlines the many similarities between Pex5p and Pex20p dynamics during the import cycle. Pex20p steady-state level, like that of Pex5p (; ), decreases in recycling mutants cultured overnight in oleate medium (). Among yeasts, this down-regulation of Pex5p is peculiar to . Instead, ubiquitylated species of Pex5p accumulate in these mutants of (; ; ). At an earlier time point (6 h after induction; G) higher molecular mass species (likely ubiquitin conjugates) of Pex20p-GFP are actually detected. K19R mutation in Pex20p prevents both the appearance of these additional Pex20p species and Pex20p down-regulation (, E–G). Therefore, in this degradation is also likely to happen through the UPS. In conclusion, Pex20p is probably degraded by a quality-control mechanism triggered by the absence of recycling ( B), as suggested for ScPex5p (; ). We call this the peroxisomal receptor accumulation and degradation in the absence of recycling (RADAR) pathway (). We observed that interfering with K48-branched polyubiquitylation phenocopies the absence of the late-steps peroxins ( G; and ). It was intriguing to see polyubiquitylated species of Pex20p appear when Ub(K48R) was overexpressed (, B–D), a condition that should reduce polyubiquitylation. However, because Pex20p degradation in these mutant backgrounds is likely to happen via the UPS ( G), Ub (K48R) slows down this process and causes the accumulation of ubiquitylated species less susceptible to proteasome degradation, leading to a balance between the generation of ubiquitylated species by the RADAR pathway and their stabilization after interference with K48-branched polymerization. This allowed us to detect polyubiquitylated species of Pex20p, with K19 being the target residue (). In these conditions, both Pex20p and its ubiquitin conjugates were in and on peroxisomes (), with its ubiquitylated forms being more susceptible to protease than the nonubiquitylated form ( F). These observations are summarized in our working model (). Understanding the links between ubiquitin-mediated degradation and the import of peroxisomal proteins will be required for a better understanding of peroxisome biogenesis. The strains used included PPY12 (; ), mutants (; ), and (this study). Cells were routinely grown in YPD (1% yeast extract [YE], 2% Bacto Peptone, and 2% glucose), YPM (1% YE, 2% Bacto Peptone, and 0.5% methanol), YNB (0.17% YNB, 0.5% [NH]SO, and 2% glucose), or YNO (0.05% YE, 0.25% [NH]SO, 1 mM MgSO, 20 mM NaHPO, 4 mM KHPO, 0.02% Tween 40, and 0.2% oleic acid). Oleate induction was overnight (18 h) unless otherwise indicated in the figures. Media were supplemented with 20 μg/ml of histidine and arginine as needed. Δ Δ (UTL7a: Mata, -, , , , and ) was a gift of W.-H. Kunau (Ruhr-Universität, Bochum, Germany) and was grown on YNB medium. Oligonucleotides used are presented in Table S1 (available at ). The G418 resistance cassette was amplified (KanMX.d/KanMX.r) from pFA6a-KanMx6 and cloned at KpnI–BamHI in pBluescript II KS+ (Stratagene), creating pSEB44. The 5′ flanking region of the ORF was amplified from PPY12 genomic DNA (pPEX20.d/5′20.r, KpnI–blunt) and cloned at KpnI–SmaI sites of pSEB44, resulting in pSEB46; the 3′ flanking region of the ORF was amplified (3′20.d/3′20.r, XhoI–blunt) and was further cloned in pSEB46 (XhoI–EcoRV) to create pSEB47. The disruption cassette was amplified (pPEX20.d/3′20.r) from pSEB47 and transformed into the PPY12 strain. G418 clones were screened by PCR and product size analysis. The -based Matchmaker yeast two-hybrid system (CLONTECH Laboratories, Inc.) was used. The cloning strategy involved PCR amplification of peroxins from genomic DNA (oligonucleotides with XmaI–SalI sites; Table S1), in-frame cloning at the XmaI–SalI sites of pGAD-GH or pGBT9 and sequencing. AH109 strain was cotransformed and selected on complete synthetic medium lacking Leu and Trp (CSM-Leu-Trp). Independent cotransformants were patched on CSM-Leu-Trp and replica plated on CSM-Leu-Trp, CSM-Leu-Trp-His, and CSM-Leu-Trp-His + 50 mM 3-aminotriazole (3-AT; Sigma-Aldrich). Site-directed mutagenesis was performed with primers (20W89.d/20W89.r, 20W102.d/20W102.r, 20W141.d/20W141.r, and 20S280F.d/20S280F.r), sequencing, and excision/recloning into the original vector. Interactions were judged by the transcriptional activation of the gene (growth on CSM-Leu-Trp-His + 3-AT). coding sequence was amplified (Y2H20.d/Y2H20.r, XmaI–SalI) and cloned at the XmaI–SalI sites of pCu416, creating pSEB41. was amplified from genomic DNA (CuPex18.d/CuPex18.r, XhoI–XbaI) and cloned (XhoI–XbaI) in pCu416, creating pSEB49. Serial dilution of YNB-grown cells were spotted on YNO agar plates. Cells were grown overnight on YPD medium, precultured on YPD for 10 h, and transferred overnight into YNO. Cells were homogenized as described previously (), except that the last centrifugation was performed at 200,000 to ensure pelleting of peroxisome remnants in mutants (). Cells were broken as for subcellular fractionation but without protease inhibitors. Pellets of a 200,000- centrifugation (see previous section) were resuspended in ice-cold Dounce buffer () to a protein concentration of 1 mg/ml, and 8 aliquots of 50 μg were taken. Freshly prepared proteinase K (Sigma-Aldrich) was added to all tubes (20 μg) after addition of Triton X-100 (0.125% final concentration) where specified in the figures and incubated for the indicated times. Trichloracetic acid (10% final concentration) was added to stop the reaction. Proteins were precipitated overnight in ice, washed three times with acetone, and resuspended in lysis buffer, and 10 μg of each reaction was loaded. The ORF was amplified with its promoter (pPex20.d/20-GFP.r, KpnI–PstI) and cloned upstream the GFP coding sequence of pWD3 (a gift of W. Dunn, University of Florida, Gainesville, FL) at KpnI–PstI sites, creating pSEB48. The vector was linearized with SalI and inserted at the locus. Pex20(Δ1–19)-GFP was constructed by mutagenesis on pSEB48 using the primer pair 20G(Δ1–19).d/20G(Δ1–19).r, resulting in pSEB149. Cells were grown on YPD and switched to YNM or YNO when in exponential phase. Other constructs included p:BFP-PTS1 (a gift of W. Dunn), p:PTS2-mRFP (pKSN39) and p:mPTS-mRFP (pKSN7; gifts of K. Noda, University of California, San Diego, La Jolla, CA), and p::PEX3-mRFP (pJCF215; a gift of J-C Farré, University of California, San Diego). Copper induction was with CuSO (100 μM final concentration) 2 h before observation. Images were captured on a motorized fluorescence microscope (AxioSkop 2 plus; Carl Zeiss MicroImaging, Inc.) coupled to a cooled charge-coupled device monochrome camera (AxioCam MRM; Carl Zeiss MicroImaging, Inc.) and processed using the AxioVision software. The coding sequence was amplified (5SHA20/3PHA20, SacI–PstI) and cloned at SacI–PstI sites in a pIB2-based vector (constitutive GAP promoter; ) containing a triple HA tag (pIB2-HA), creating pIB2-HA-PEX20 (a gift from I. Suriapranata, University of California, San Diego). HA-Pex7p construct was obtained from W. Snyder (University of California, San Diego). Vectors were cut with SalI and integrated at the locus. Transformants were grown overnight on YPD and transferred for 5 h on YNO before extraction. Immunoprecipitation of HA-tagged proteins was performed as follows. Cells (8 ODs) were broken with glass beads in 200 μl IP lysis buffer (50 mM Hepes-KOH, pH 7.5, 0.5 M NaCl, 0.5% NP-40, 10% glycerol, 1 mM EDTA, and protease inhibitor cocktail) and centrifuged twice (14,000 , 10 min). Monoclonal anti-HA antibody (Covance) was added to the supernatant (6 μL/ml of lysate) and incubated overnight with the extract. 25 μL of GammaBind beads (GE Healthcare) were added and incubated for 2 h. Beads were washed twice (1 ml) with the lysis buffer for 10 min and three times (1 ml) with the wash buffer (50 mM Hepes-KOH, pH 7.5, 150 mM NaCl, and 1 mM EDTA) and finally boiled in 50 μL SDS loading buffer. Loading was as follows: input and unbound fraction, 0.2 OD equivalent; immunoprecipitate, 1 OD equivalent unless otherwise indicated. Immunoprecipitations of TCA-lysed cells for the detection of ubiquitylated Pex20p were performed as described in using the anti-GFP monoclonal antibody from Roche Applied Sciences. A chicken polyclonal antibody was generated against aa 1–146 of Pex20p. Affinity-purified antibodies were further purified by incubation with crude acetone powder extract of methanol-grown cells and used at 1:1,000 dilution for immunoblotting. Cells were grown overnight on YPD, precultured on YPD for 10 h, and transferred overnight in YNO. Cells were disrupted with glass beads as for immunoprecipitations. The supernatant of the 14,000 centrifugation was considered the raw extract. Proteins were assayed, and 20 μg were loaded on each lane. The promoter was amplified from genomic DNA (pPex20.d/r, KpnI–SmaI) and cloned in pIB1 (), creating pSEB95. Sequences encoding truncated versions of Pex20p were amplified using a forward primer (Pex20-7.d, -11.d, -16.d, -22.d, and -31.d; SmaI) in combination with the reverse primer Y2H20.r (SalI) and cloned (SmaI–XhoI) downstream of p in pSEB95 (pSEB101, -103, -105, -122, and -108, respectively). Myc-Ub(K48R) was a mutagenized version of pTK132 (p:myc-Ub from A. Koller, University of California, San Diego) and was provided by I. Suriapranata. 6xHis-myc-Ub and 6xHis-myc-Ub(K48R) were created by PCR amplification of myc-Ub from YEp105 or pTK132, respectively (6xHis-Myc.d/Ub.r, EcoRI–KpnI), and cloning into pJCF215 (pIB2-based vector containing the GAPDH promoter, in which the marker gene was replaced by ; a gift from J.-C. Farré) was done to create pSEB127 and -128. Fig. S1 shows an alignment of PpPex20p with its homologues from various organisms. Table S1 shows the oligonucleotides used. Online supplemental material is available at .

The emergence of the earliest neural cells during mammalian development and the mechanisms that govern this process remain incompletely characterized. Such cells are likely to be neural precursors or stem cells, though the ontogeny of the neural stem cell (NSC), which can be isolated from embryonic and adult forebrain (; ), has not been fully elucidated. During development, neural cells arise from the ectodermal germ layer, which also produces epidermis. According to the classical model of this process, conceptualized largely from amphibian embryology studies, nascent embryonic ectoderm receives a positive signal from a specialized group of dorsal mesodermal cells, termed the organizer, which instructs the adjacent ectodermal cells to adopt a neural fate (; ; ). The structural equivalent of the organizer in amniotes is the node. It was thought that organizer/node–derived signals were necessary for the process of neural induction and that in their absence the ectoderm would adopt an epidermal fate. More recent data have challenged the validity of this classical model. Low-density cultures of dissociated ectodermal cells, in the absence of organizer tissue, were found to differentiate into neural cells (; ; ). Furthermore, undissociated ectodermal explants expressing a dominant-negative receptor for activin (a member of the TGFβ family of growth factors), which effectively inhibited signaling of multiple TGFβ-related molecules (; ), were shown to become neural when cultured in vitro (). Signaling molecules secreted from the organizer tissue, such as Noggin, Chordin, and Follistatin, were found to exert potent neuralizing effects (; ; ) and were thus initially thought to represent the instructive neuralizing signal. However, the mechanism by which they promoted neural differentiation of ectodermal cells was not entirely consistent with the existing positive induction model. The neuralizing effects of these factors were found to depend on inhibitory interactions with bone morphogenic proteins (BMPs), which are members of the TGFβ family of molecules that strongly inhibit neural differentiation (; ; ). Thus, their mechanism of action appeared to be through prevention of BMP binding to their cognate receptors on ectodermal cells. These findings led to the development of the currently more widely accepted model, the default model, which states that each individual ectodermal cell has an intrinsic default program to become a neural cell (). In the context of the intact embryo, this default program is being actively suppressed by ubiquitously expressed BMPs. Thus, the organizer tissue does not provide a positive inductive signal but rather secretes factors that antagonize BMP signaling, thereby disinhibiting the default neural program in proximal ectodermal cells. Several subsequent studies have challenged the default model of neural fate acquisition. For example, experiments in chicks have suggested that BMP inhibition may not be sufficient to induce neuralization (; ). However, it is uncertain how complete the BMP inhibition was in these studies, and it is possible that the activity of individual BMPs was insufficiently suppressed to allow neuralization to occur (and/or that some BMP subtypes or other neural inhibitors escaped blockade). It has also been suggested that other factors, such as FGF and Wnt signaling, are involved with neural specification in several vertebrates (; ; , ), though whether they are required for the initial neural fate change or for the later expansion of this neural population is currently unresolved. Further, their mechanism of action may be through modulation of BMP gene transcription (), consistent with a model of BMP inhibition–mediated neuralization. There are currently few published studies examining the neural default model in mammalian cells, and there is controversy over whether such a default neural mechanism exists in mammals. In an effort to determine whether a default mechanism underlies neural fate specification from uncommitted mammalian precursors, we undertook studies using mouse embryonic stem (ES) cells, which are derived from the inner cell mass (ICM) of the blastocyst-stage embryo and represent a model of the earliest pluripotent mammalian cell (; ). ES cells are capable of generating entire viable mice in vivo () and are able to produce most, if not all, cell types in vitro (; ; ). Use of ES cells to investigate neural determination can potentially provide many insights into the developmental process. Importantly, though, their use in assessing a default fate specification mechanism allows us to explore a more basic and fundamental issue, i.e., how an uncommitted, pluripotent mammalian cell will self-organize in the absence of extrinsic instructive or inhibitory signals and what cellular configuration/fate will result. The standard methodology for the in vitro differentiation of ES cells typically involves the formation of embryoid bodies (EBs; ), which are formed by aggregation of ES cells in the presence of serum and in the absence of leukemia inhibitory factor (LIF), a cytokine necessary for maintaining ES cells in an undifferentiated state. EBs contain many different cell types that are fated to produce cells of all three primary germ layers. Because there is complex intercellular signaling between the multiple cell types of an EB and they are generated in the presence of serum with its host of undefined factors, EB formation precludes a direct analysis of the mechanisms regulating the differentiation of a specific cell lineage. To assess a default state, we wanted to isolate single ES cells and minimize any exposure to extrinsic factors that might be either instructive or inhibitory to cell fate specification. Therefore, we used a system of chemically defined serum- and feeder layer–free culture conditions coupled with low cell densities (to abrogate intercellular signaling). In a previous study, we reported that these conditions appeared to favor neural determination of ES cells (). Further, a novel colony-forming primitive NSC population arose under these conditions, one with characteristics intermediate to those of ES cells and forebrain-derived “definitive” NSCs. Here, we demonstrate default neural fate acquisition by ES cells, a process shown to be independent of potential instructive factors. FGFs were found to be important for the proliferation but not the generation of the default pathway–derived primitive NSCs. Further, we provide evidence that the default neural fate pathway specifically gives rise to primitive NSCs and that primitive NSC mortality resulting from a survival challenge, which could be mitigated by survival factors or genetic interference with apoptosis, was responsible for limiting the persistence and proliferation of these cells. To assess the potential default fate of ES cells, we removed any factors that might be either instructive or inhibitory to cell fate specification. Therefore, single dissociated R1 ES cells were plated at low cell densities (≤10 cells/μl; 2,600 cells/cm) in chemically defined serum- and growth factor–free media. As we reported previously (), ES cells placed in these minimal conditions rapidly acquired a neural identity, with >90% of cells initiating expression of nestin, an intermediate filament protein associated with neural precursors (), within 4 h ( A). The neural precursor identity of these cells was supported further by their expression of Sox1, one of the earliest transcription factors expressed in cells committed to the neural fate (; C). Undifferentiated ES cell colonies did not exhibit expression of these markers (). At 4 h, no significant cell mortality was observed and <1% of the plated cells had proliferated, indicating that single ES cells began a direct transition to neural cells, without requirement for cell division. After an additional 20 h, most (77.5 ± 1.3%) of these ES-derived neural cells did not survive, as these minimal conditions were not very supportive. However, 99.3 ± 0.2% of the remaining viable cells expressed nestin and Sox1 and maintained some expression of Oct4, a transcription factor expressed in ES cells (; ). To verify that the observed up-regulation of nestin was not a nonspecific stress response, STO fibroblasts were placed in identical conditions; however, RT-PCR and immunostaining did not show any nestin expression either before or after culture in minimal media (unpublished data). Further, 24 h after cell plating, 90.4 ± 1.7% of the surviving cells expressed neurofilament-M (NFM) and 47.5 ± 3.6% expressed the early neuronal marker β-tubulin (). RT-PCR analysis confirmed expression of nestin and Sox1 ( I). Additional confirmation of neural lineage commitment after 24 h was evidenced by the rapid down-regulation of brachyury and GATA-1 (mesodermal markers), as well as HNF3β, HNF4, and GATA-4 (endodermal markers), which were detectable in ES cells ( J). At 3 d, most surviving cells maintained a neural precursor identity, expressing nestin and Sox1 (). Cells with more elaborate neural morphologies were also evident, with the observation of neuronal and glial cells (). Cells with very advanced morphology, and positive for a neural cell subtype marker, typically displayed down-regulation of nestin expression (faint or absent immunostaining) and were Oct4 (unpublished data). By 7 d, viable cells were not observed, indicating that the exiguous nature of the media conditions was not supportive enough for the maintained survival of the neural cells. These data indicate that in the absence of extrinsic signals, ES cells rapidly begin to acquire a neural precursor identity, consistent with a default mechanism for an ES cell to transition directly into a neural cell. Is this neural transition truly occurring by default? A protein component of the media formulation, transferrin, has been suggested to be required for ES cell neural fate initiation (). We show that nestin and Sox1 expression was established after 4 h, even when ES cells were cultured in PBS alone, ruling out any requirement for media components in an instructive capacity (). After a 24-h culture period in PBS, very few cells remained viable, though they all expressed nestin and Sox1 (). The fragmented nuclei represent dead cells, which typically did not display immunoreactivity for the neural markers. Though not exogenously required, autogenously produced FGFs have been proposed to be essential for ES cell neuralization, suggesting an instructive process (). To the contrary, pharmacological inhibition of FGF signaling using the FGF receptor kinase antagonist SU5402 (5 and 10 μM) in our minimal conditions did not prevent the rapid acquisition of neural markers by ES cells (, E–H). Similarly, ES cells harboring deletion of the FGF receptor-1 gene (; ) displayed the typically observed neural markers (, H–K). Further, SU5402 treatment or deletion did not prevent the advanced neural morphologies and marker expression found at 3 d in culture (unpublished data). These data strongly suggest that neither media components (including transferrin) nor FGF signaling is required for ES cell neuralization, supporting the default nature of this transition. What is the nature of the neural cells emerging from this default pathway? The early expression of nestin and Sox1 indicates that they are neural precursors. However, these markers are expressed both in true NSCs as well as in more restricted neural progenitor cells. Our previous work introduced a novel neural precursor arising under these conditions, the primitive NSC (), identified with an assay analogous to the neurosphere (NS)-formation assay used to identify definitive forebrain-derived NSCs. When the cytokine LIF was included in our minimal condition assay, a very small percentage (0.18 ± 0.01%) of the initially plated ES cells became neural cells that proliferated over 7 d to form clonally derived floating sphere colonies, termed primitive NSs ( A). LIF was not required for early ES cell neuralization, as it could be added after 4 h without any decrement in primitive NS production (101 ± 4% of control). Rather, LIF appeared to prevent further differentiation down the neural lineage and thereby maintained these neural cells in an undifferentiated, proliferative state. This interpretation is supported by the findings that LIF addition after 2 d did not enable any primitive NS generation (unpublished data), indicating that there was an LIF-responsive temporal window, after which the neural cells had progressed to a non–LIF-responsive, unproliferative, and more differentiated state. These primitive NSs expressed both nestin and Sox1 (), similar to NSs generated by definitive NSCs, indicating that they are composed of neural precursors. The primitive NS cells also maintained some expression of Oct4 (by immunostaining and RT-PCR), as well as nanog (by RT-PCR), another transcription factor expressed by ES cells (; unpublished data). The neural precursor character of the primitive NSs was confirmed by expanded RT-PCR gene expression analysis. Various additional neural genes were expressed by primitive NSs, including Sox2, Sox3, Neurogenin1, NeuroD, Pax6, Nkx2.2, Mash1, Musashi-1, Otx2, and HoxB1, strongly supporting their neural nature (Fig. S1, available at ). Primitive NSs did not express CK-17 (an epidermal marker), nor did they express the mesodermal markers brachyury and GATA-1 or the endodermal markers HNF3β, HNF4, and Pax4. These multi–germ layer markers are typically found in EBs (), and their absence further verifies that the primitive NS cells are specified to the neural lineage. Transcripts for most of these genes were found at detectable levels in undifferentiated ES cells, suggesting that there may be widespread, low-level, promiscuous gene expression in the uncommitted state and that differentiation may involve the down-regulation of certain genetic programs with the maintenance and up-regulation of others. The primitive NS cells could be maintained in an undifferentiated state by subcloning/passaging in serum-free conditions to produce secondary, tertiary, and successive spheres. Interestingly, secondary and subsequent sphere formation was dependent on the addition of exogenous FGF2, similar to the FGF2-dependent proliferation of forebrain-derived NSCs. When transferred to conditions promoting differentiation, individual primitive NSs differentiated to produce neurons, astrocytes, and oligodendrocytes ( D), the three neural lineage cell types. Cells negative for differentiation markers maintained a neural precursor identity, expressing nestin and Sox1 ( E). Differentiation of the passaged spheres was similar to that of the primary primitive NSs (unpublished data). Though the cells present after in vitro differentiation of primitive NSs appeared limited to differentiated and immature/precursor neural cells, the maintained expression of some ES-like genes (Oct4 and nanog) in the undifferentiated primitive NS state suggested that cells constituting the primitive NS were not yet absolutely committed to the neural lineage. This is supported by our previous finding that primary primitive NSs could still manifest broader lineage potential when placed in the appropriate developmental environment, i.e., when used to generate morula-aggregation embryo chimeras (). However, when the primitive NSs were passaged to produce secondary spheres (with FGF2 and without LIF), efficient chimera production was not observed (similar to our results obtained using NSs from forebrain-derived adult NSCs), suggesting that the secondary spheres were now fated exclusively to the neural lineage (; unpublished data). Further, expression of Oct4 and nanog was not found by RT-PCR in the secondary spheres (unpublished data). Thus, a novel neural precursor, the primitive NSC, arose in these default conditions. Though not required for early default neural fate acquisition, autogenously produced FGFs were found to be necessary for primitive NS formation. Inclusion of 5 μM SU5402 in the primitive NS assay virtually eliminated primitive NS generation (reduced by 98 ± 1%; A), and −− ES cells displayed dramatically diminished (by 91 ± 1%) primitive NS production ( B). This suggests that autogenous FGF signaling was important for the proliferation and/or survival of primitive NSCs. The proliferation interpretation is supported by results obtained from rescue experiments. When SU5402 was included in the primitive NS assay, minimal proliferation was observed by 3 d. If the drug was then removed and the assay was continued for 7 d, primitive NSC proliferation resumed and there was a large recovery of primitive NS formation ( C). Similarly, if the drug was maintained for a full 7 d and then removed, a substantial recovery of primitive NS formation was observed ( C). Further, if SU5402 was added on day 3 of the assay (at which point the cells have established their neural transition and primitive NSCs are proliferating), proliferation was impaired and there were consequently fewer primitive NSs observed after 7 d ( D). As FGF signaling was dispensable for the default neural fate switch in the short-term (4 and 24 h) experiments discussed previously, this suggests that primitive NSCs are still formed in the transition from ES cells, though their proliferation to form primitive NSs is dependent on signaling by autogenously produced FGFs. Sonic Hedgehog signaling, reported to be important in the regulation of some neural progenitors (), was found to be dispensable, as Sonic Hedgehog inhibition with cyclopamine had no inhibitory effect on primitive NS formation (unpublished data). The early nestin neural cells present after 4 h in minimal conditions subsequently underwent a survival challenge because of the exiguous nature of the media, resulting in extensive cell mortality (77.5 ± 1.3% by 24 h). In the absence of LIF, mortality progressively increased over 7 d, until no viable cells remained. In the presence of LIF, a few of the neural cells retained a proliferative character, surviving and propagating to form primitive NSs, though the vast majority of the early nestin cells still died. Accordingly, primitive NSCs may have been produced but did not form primitive NSs because of the early survival challenge that they experienced. Thus, increasing cell viability should enhance primitive NS production. The survival factor -acetyl--cysteine (NAC), which has multiple modes of action but is thought to act primarily through its antioxidant effects (), promotes survival of various neural cell types (; ; ). Inclusion of NAC dose-dependently increased primitive NS formation (up to ∼35-fold), with toxic effects observed at very high concentrations ( A). We also explored the cAMP–protein kinase A pathway as a more physiological modulator of cell survival. cAMP signaling has been shown to enhance cell survival in several different neural cell systems (; ; ). Exogenous application of a membrane-permeable cAMP analogue, 8-(4-chlorophenylthio) (pCPT)-cAMP, was able to dose-dependently augment primitive NS formation (up to ∼25-fold), with toxic effects observed at exceedingly supraphysiological concentrations ( A). Concentrations at which peak effects were observed were used in all subsequent experiments (1 mM NAC and 100 μM pCPT-cAMP). When added simultaneously, the effects of NAC and pCPT-cAMP were additive, if not synergistic, enhancing primitive NS formation (up to ∼100-fold) such that ∼20% of the initially plated ES cells became primitive NS–forming primitive NSCs ( B). This also suggests that the two compounds have distinct mechanisms of primitive NSC survival promotion (leading to subsequent primitive NS formation). RT-PCR analysis of primitive NSs derived in each or both factors demonstrated the expression of the neural precursor markers without detection of the multi–germ layer markers (discussed previously). Addition of NAC and/or pCPT-cAMP after 4 h of the primitive NS assay (at which point the cells have already undertaken their neural transition) revealed no decrement in their ability to augment primitive NS formation (NAC addition at 4 h was 98 ± 3% of NAC added from start, and pCPT-cAMP addition at 4 h was 101 ± 3% of pCPT-cAMP added from start), suggesting that they were not effecting the NS increase through an instructive role in early differentiation. Additionally, this demonstrated that the survival-promoting effects of NAC and cAMP were not directly on ES cells but rather on their nestin clonal neural derivatives. Viability assessment showed that NAC and pCPT-cAMP did indeed increase cell survival under minimal conditions. At 24 h, NAC increased survival of the initial 4-h nestin cells from 22.5 ± 1.3 to 44.8 ± 2.3% (P < 0.01), whereas pCPT-cAMP increased survival to 43.6 ± 2.4% (P < 0.01). Though the survival factor–induced increases in viability relative to the control media were observed over the course of the complete 7 d, progressive mortality was still observed. Even with both factors present in the primitive NS assay, there was still 34.4 ± 2.6% mortality for the initial 4-h nestin cells after 24 h and 54.3 ± 2.9% mortality by 3 d. There were very few cells that survived past the 7-d assay that were not in a primitive NS, suggesting that only cells that were in the context of a proliferating sphere were effective at long-term survival in these conditions. It is worth noting that although the survival factors greatly increased cell viability at 1 and 3 d in minimal media, the rapid acquisition of neural identity and the frequencies of marker-positive cells (discussed previously and in ) were not altered (not depicted). Thus, these results indicate that the primitive NSC is the primary identity of cells derived from the default pathway but that cell mortality limits the number that can survive to form clonal primitive NSs and thereby be detected by our assay. The ability of NAC and pCPT-cAMP to facilitate primitive NS formation decreased dramatically and progressively when the compounds were added during the derivation of secondary and tertiary spheres ( C). This suggests that the extensive survival challenge that limited primitive NS generation occurred only with the primary ES–derived primitive NSCs and not with the subsequently passaged NSCs, which are similar to forebrain-derived definitive NSCs (which also did not exhibit increases in NS formation with the survival factors; unpublished data). This effect highlights a fundamental difference between primitive NSCs and the passaged, more mature definitive NSCs. When individual primary spheres derived in the presence of pCPT-cAMP were passaged back into control media for secondary sphere formation, they generated 2.3 ± 0.2 times (P < 0.01) more secondary spheres compared with primary spheres that had been derived in control media. Spheres derived in NAC did not display any increased secondary sphere formation when passaged back into control media. This suggests that pCPT-cAMP may have had an additional effect, increasing the number of symmetric divisions that the primitive NSC underwent, thereby expanding the stem cell numbers within the primary sphere. Alternately, there may have simply been a prolonged survival enhancement in the pCPT-cAMP–derived primary sphere cells, even when they were placed back into media without pCPT-cAMP. Primitive NSs derived in NAC and/or pCPT-cAMP did not display any notable alterations in differentiation (unpublished data). Pharmacological modulation of the endogenous pathways showed that the cAMP pathway positively regulated and the cyclic guanosine monophosphate (cGMP) pathway negatively regulated primitive NSC survival and primitive NS formation (see online supplemental material). To further substantiate that cell mortality limited the number of primitive NSCs that survived to proliferate and form clonal primitive NSs, we used ES cell lines with a survival advantage conferred by mutations in apoptotic signaling pathway components. () and () are components of the more common apoptotic pathway (), whereas () is a mediator of an alternate, independent apoptotic pathway (). The −/− (), −/− (), and −/Y () ES cells displayed substantially increased primitive NS formation relative to wild-type ES cells, with a more pronounced effect observed for the and mutants ( A). Heterozygotes for and displayed an intermediate effect, indicative of a gene dosage effect ( A). Differentiation of primitive NSs generated by these ES cell lines was similar to wild type (unpublished data). This demonstrates that, similar to the effect of survival factors, survival promotion through interference with apoptosis enhanced primitive NS formation. This also suggested that the primary apoptotic pathway responsible for mediating cell death in these cultures was the – apoptotic pathway, although there was still some contribution from the pathway. If the primitive NS–promoting effect of the survival factors NAC and pCPT-cAMP was achieved through a reduction in primitive NSC death mediated by these apoptotic pathways, then the survival factor effect should be reduced in these mutant ES cell lines. Typical robust stimulation was observed for the wild-type line, which was much reduced, though still considerable, in the mutant ( B). Strikingly, the effects of NAC and pCPT-cAMP were drastically reduced in the −− and −− ES cells to the extent that there was no longer any significant effect of NAC and only an approximately twofold stimulation by pCPT-cAMP ( B). This attenuation or occlusion of the survival factor effects in these mutant ES cell lines verifies that NAC and pCPT-cAMP promote primitive NS formation through a reduction in apoptotic cell death, primarily mediated by the more dominant pathway. We have previously demonstrated that TGFβ-related signaling negatively regulates ES cell neuralization and the basal number of primitive NSs that form in our assay (). To determine how survival promotion interacted with interference of the TGFβ pathway, the effects of NAC and pCPT-cAMP were assessed in an ES cell line that contains a mutation in , a key downstream signaling molecule in multiple TGFβ-related pathways (). The −− ES cells displayed increased primitive NS generation relative to wild-type ES cells under both control media conditions as well as in the presence of the survival factors such that in the presence of both NAC and pCPT-cAMP ∼35% of the initially plated ES cells became primitive NSCs that were able to survive and proliferate to form primitive NSs ( C). These results provide further support for the notion that the default pathway does indeed give rise specifically to primitive NSCs (though TGFβ signaling hinders this default to primitive NSCs), which then experience a separate survival challenge limiting the number of primitive NSCs that form primitive NSs. The present study demonstrates that in the absence of extrinsic influencing signals, ES cells will rapidly undergo a direct neural conversion, transitioning into neural precursor cells. This neuralization does not depend on any instructive factors but rather occurs by a default mechanism. The default neural pathway specifically gives rise to primitive NSCs, and because of the exiguous nature of the default conditions, primitive NSC mortality is primarily responsible for limiting the persistence, proliferation, and detection of these cells. Accordingly, increasing primitive NSC viability in these conditions with exogenous survival factors, activation of the endogenous cAMP pathway, or genetic interference with apoptosis enhanced primitive NS formation. Studies using human ES cells have supported the concept of default neural differentiation (; ), although clonal analyses were not performed. Other work examining the neural conversion of mouse ES cells without EB formation reported that most cells adopted a neural precursor phenotype in serum-free conditions; however, they found that widespread neuralization did not occur until at least 3 d (). Further, suggest that the media component transferrin and autocrine FGFs were required positive instructive factors, in contrast to a pure default model. The present results demonstrate that neural specification commences over a matter of hours, even when ES cells were placed in PBS alone, ruling out any necessity for media components in an instructive capacity. Certainly, cells cannot be maintained long-term in PBS alone, as nutritive factors are required in a permissive role. Further, we provide evidence against any positive inductive role for autogenously produced FGFs by demonstrating that neither pharmacological FGF inhibition (using the same inhibitor as ) nor genetic deletion of the FGF receptor-1 effected any reduction in neuralization. Rather, such FGF signaling functions to allow the proliferation of the default pathway–derived neural precursors. Experiments of used cell densities up to ∼20 times higher than those used in the present studies and in some cases allowed low-density plated ES cells to grow as colonies overnight in normal ES media (increasing the effective cell density) before removal of LIF and serum. These conditions are likely to facilitate the density-dependent accumulation and signaling of secreted neural inhibitors (e.g., BMPs), as we demonstrated previously (), potentially explaining why their neuralization was delayed. They also reported the gradual increase in expression of BMP antagonists (e.g., noggin and follistatin) over 1–5 d in their conditions, which would then provide the BMP antagonism necessary to allow the default program to manifest. Furthermore, the extensive viability and exponential cell expansion in their conditions is in stark contrast to our findings, indicating that their high cell densities and/or richer media formulation produced an environment less minimal than that used in the current study. The severely minimal nature of our culture conditions was necessary to effectively minimize potential extrinsic influences and thereby assess a default state. However, this was clearly responsible for the extensive cell mortality observed. In attempts to evaluate the default state, one encounters a catch-22. At some point, progressive removal of extrinsic factors will dispense with constituents essential for default fated cell survival, and there will be no experimental output to assess. Such factors are required for neuralization not in an instructive capacity but rather in a permissive, supportive one. In interpretation of our results, one might suggest that we were merely providing conditions that selected for the survival of neural cells while nonneural cells perished, and thus the default state was not assessed. Two observations argue against this interpretation. First, in the initial 4 h after ES cell plating, no significant mortality was observed, yet the vast majority of individual ES cells had already begun their neural transition. Second, with the use of both survival factors in the 24-h assay, viability was increased from 23 to 65%; however, a multiple marker–based analysis showed that there was no difference in the uniform cell neuralization and that there was no expression of other germ layer markers. If the conditions used were selecting against the survival of nonneural lineage cells, they should have been apparent when survival was so dramatically increased. The primitive NSCs described here exhibited characteristics (i.e., proliferative sphere formation, gene expression, and differentiation potential) that are analogous to forebrain-derived NSCs. Although it is clear that the primitive NSCs commenced neural lineage specification, it was apparent that they retained vestiges of ES cell features, suggesting that the commitment was not yet absolute. This was evidenced by the broader differentiation potential observed in blastocyst chimera experiments (). Passaging of the primitive NSs yielded more committed neural precursors, increasingly similar to NSs formed by forebrain-derived NSCs (e.g., they acquired dependence on exogenous FGF2, ceased expression of ES cell markers Oct4 and nanog, and lost blastocyst chimerism potential). Recent work from our lab has demonstrated that LIF-dependent sphere-forming cells can be isolated directly from the embryonic day (E) 5.5–7.5 mouse epiblast (). These embryo-derived spheres possessed similar characteristics to the ES-derived primitive NSs and were found to give rise to FGF2-dependent, NS-forming, definitive NSCs upon passaging in vitro (). Thus, in vivo there are two distinct sphere-forming NSC populations that are present at different stages of development from E5.5 to adult. In addition, corresponding sphere-forming NSC populations could be sequentially derived from ES cells. This suggests the in vivo relevance of an ES cell–based model describing the ontogeny of NSCs (). Pluripotent ES cells derived from the E3.5 ICM will commence a direct default transition in the absence of inhibitory influences (e.g., TGFβ-related signaling) to yield LIF-dependent primitive NSCs, similar to those that can be isolated from the epiblast (). The primitive NSs produced can be passaged to give rise to definitive, FGF2-dependent NSCs, analogous to the FGF2-dependent NSCs that can be isolated from the E8.5 neural plate (). The in vitro–derived definitive NSCs can be extensively passaged to demonstrate long-term self-renewal, consistent with the effective isolation of definitive NSCs from the adult remnant of the embryonic brain germinal zone throughout the lifetime of the animal. Is the propensity of ES cells to default to a neural identity developmentally relevant? ES cells have properties similar to the E3.5 ICM cells from which they are derived. Our data suggest that the ES cells can directly become neural without a discernible ectodermal intermediate phase. The developmental dogma states that neural tissue arises from the ectodermal germ layer, which is not defined until gastrulation (E6.5–8.5). ES cell differentiation has been suggested to recapitulate these stages (). It is possible that the neural tendency of the E3.5 cells never has the opportunity to manifest in vivo because of extensive antineural signaling via BMPs until after gastrulation, when node regions are formed to disinhibit the neural fate in what is now specified ectoderm. However, recent work in both and chicks has suggested that there may be some neural-fate acquisition even before gastrulation (; ). This may be attributable to preectodermal ICM-like cells that begin to express their default neural tendency after ineffective neural inhibition. In the context of the intact embryo, even with its milieu of secreted BMPs, it is conceivable that transient interruption or inefficiency in BMP delivery or activity would occur, allowing a cell or cells to escape neural inhibition and rapidly initiate the default neural program. In addition, the present data suggest that mouse ES cells can transition directly into neuronal cells; however, neurons are not produced in vivo until approximately E10. It is possible that neural precursors present in vivo before this stage are competent to form neurons but do not do so because of the presence of factors like LIF, which have been shown to prevent neural precursors from further differentiation down the neural lineage and maintain the undifferentiated primitive NSC state. In addition to providing insights into the mechanisms of neural development, default fate studies using ES cells allow exploration of what is perhaps a more basic and fundamental issue, i.e., how an uncommitted, pluripotent mammalian cell will self-organize in the absence of extrinsic instructive or inhibitory signals and what cellular configuration/fate will result. Our data suggest that such uncommitted cells tend to self-organize in the configuration of a NSC, i.e., the default fate. What is meant by the term default? Clearly, such analyses begin with the definition of a system, and the default state represents the configuration that the system will autonomously acquire in the absence of extrinsic inputs to the system. However, this means that there can be different levels of analysis depending on the definition of the initial system. In the context of ES cell default neural fate, a fairly gross level of analysis would simply state that no other tissues or cells are required for an ES cell to become a neural cell, whereas all other factors (such as environmental/media components and endogenous/autocrine signaling) are included within the system. The present data would clearly satisfy this level of analysis. A finer level of analysis would state that a single ES cell would acquire a neural identity with only permissive molecules present (in the media) and without autocrine inductive signals. The present data seem to partially satisfy this level of analysis as well, as we have ruled out requirement for exogenously supplied instructive molecules and autogenous FGF signaling. However, we cannot yet definitively demonstrate that no other potential instructive endogenously supplied (and perhaps autocrinely acting) factors are present. An even finer level of analysis would be at the genomic level, with the system defined as the entire genome. Though comprehensive and definitive analysis of the genomic system in isolation is currently impossible (as the genome only exists in an active form in the context of living cells), a default neural state at this level would predict that the highest genes within the neural genetic program (i.e., a potential “master neural genes”) would have only repressor elements within its regulatory regions. In this model, signals extrinsic to the genome would not be necessary to activate the neural genetic program; rather, they could serve only to inhibit it. Transcriptional repressors have been identified that are responsible for inhibition of the neural program, e.g., neuron-restrictive silencing factor/RE1-silencing transcription factor (NSRF/REST; ). In vivo inhibition of NSRF/REST leads to derepression of neural genes in both neural and nonneural tissues (). Thus, it is possible that NSRF/REST, as well as other transcriptional repressors, is required to prevent the default neural genetic program from being activated in all cells. ES cells were maintained on mitotically inactivated mouse embryonic fibroblast feeder layers in DME media containing 15% FCS and LIF (1,000 U/ml). The main ES cell line used was R1, as well as E14K, − (), +−−− (), +−−− (), and −− () ES cells (provided by J. Rossant, A. Nagy, T. Mak, and J. Penninger, University of Toronto, Toronto, Canada). NSCs were isolated from the forebrain ventricular subependyma of adult CD1 mice (Charles River Laboratories) and cultured/maintained using the NS assay as previously described (). ES cells were dissociated into a single-cell suspension and plated at ≤10 cells/μL on laminin/polyornithine–coated culture plates (Nunclon) in chemically defined, serum-free media (, ), which consisted of DME/F-12 (1:1; Invitrogen) with 0.6% -glucose, 5 mM Hepes, 3 mM NaHCO, 2 mM glutamine, 25 μg/ml insulin, 100 μg/ml transferrin, 20 nM progesterone, 60 μM putrescine, and 30 nM sodium selenite. PBS used for cell incubation in the short-term neural conversion experiments included 20 mM glucose. Uncoated dishes were also tested, and though cells exhibited greatly decreased adhesion, no differences in the neural conversion assays were observed. Cells were fixed for immunocytochemical analysis or cells were collected for RT-PCR analysis at various time points over the course of a 7-d culture period. For assessment of the survival of the 4-h nestin cells, ES cells were initially plated at 50 cells/well (48-well plate). At various time points thereafter, trypan blue exclusion was used to determine the number of viable cells or cell colonies (each representing the survival of a single 4-h nestin cell). Survival factors (Sigma-Aldrich) were included as indicated in the text. A single-cell suspension of ES cells was plated at ≤10 cell/μL in uncoated 24-well culture plates (Nunclon) in serum-free media (as described in the previous section) containing LIF (1,000 U/ml). Over a 7-d culture period, clonal, floating sphere colonies (>75-μm diam) of neural precursors formed (primitive NSs; see online supplemental material for more details). For passaging, spheres were individually or bulk dissociated into single cells by trypsin/EDTA (Invitrogen) incubation followed by manual trituration. Cells were then replated in the same media with inclusion of 10 ng/ml FGF2 (Sigma-Aldrich) and 2 μg/ml heparin (Sigma-Aldrich) and allowed to form subsequent spheres for 7 d. Survival factors and other pharmacological agents were from Sigma-Aldrich, except SU5402 (Calbiochem), and were included as described in the text. For differentiation, individual spheres were removed from growth factor–containing media and transferred to Matrigel-coated plates in the same media formulation as noted previously (without growth factors) supplemented with 1% FCS and cultured for 7 d. Undifferentiated and differentiated NSs were fixed for immunocytochemical analysis or were collected for RT-PCR analysis. Fixation and immunocytochemical analysis of cells was performed as described previously for NSs (). Primary antibodies used included anti-nestin rabbit polyclonal (1:2,000; a gift from R. McKay, National Institutes of Health Research, Bethesda, MD), anti-nestin rabbit polyclonal (1:50, a gift from R. Goldman, Northwestern University, Chicago, IL), anti-nestin mouse monoclonal (1:1,000; Chemicon), anti-Oct4 mouse monoclonal (1:500; BD Biosciences), anti-Sox1 rabbit polyclonal (1:150; Chemicon), anti-NFM rabbit polyclonal (1:200; Chemicon), anti-tubulin isotype III (β-tubulin) mouse monoclonal (IgG; 1:500; Sigma-Aldrich), anti-glial fibrillary acidic protein (GFAP) mouse monoclonal (IgG; 1:400; Sigma-Aldrich), and anti-O4 mouse monoclonal (IgM; 1:100; Chemicon). Nestin immunoreactivity in the short-term (1–3 d) assays and in primitive NSs was verified by immunostaining with at least two nestin antibodies. Secondary antibody–only wells were processed simultaneously using the identical protocol, except that solutions did not contain primary antibodies. All secondary-only controls were negative for staining. Cell nuclei were counterstained with the nuclear dye DAPI or Hoechst 33258. Fluorescent images were visualized using a motorized inverted research microscope (IX81; Olympus) with 20×/0.40 and 40×/0.60 objectives at room temperature and captured using Olympus MicroSuite Version 3.2 image analysis software (Soft Imaging System Corp.). Images were prepared using Photoshop 6.0 (Adobe) software. Total RNA was extracted from cells using an RNeasy extraction kit (QIAGEN) with inclusion of DNase to prevent genomic DNA contamination. RT-PCR was performed using a OneStep RT-PCR kit (QIAGEN) in a GeneAmp PCR System 9700 (Applied Biosystems) according to kit instructions. See Table S3 (available at ) for details of primer sequences and cycling conditions. Data are expressed as means ± SEM unless specified otherwise. Statistical comparisons between two groups were performed using a test where appropriate or by one-way analysis of variance with Dunnett's posttest for comparing groups to control or the Bonferroni posttest for comparisons between groups where an acceptable level of significance was considered at P < 0.05. Fig. S1 shows expanded RT-PCR analysis of primitive NSs, confirming their neural precursor nature. Fig. S2 shows that cAMP pathway activity promotes primitive neurosphere production, whereas cGMP pathway activity impairs it. Table S1 shows actual raw sphere counts obtained in one experiment, and Table S2 shows normalized sphere counts. Table S3 shows gene primer sequences and expected product sizes for RT-PCR. The Results section describes studies in which the cAMP and cGMP pathways were modulated. These studies showed that the cAMP pathway positively regulated and the cGMP pathway negatively regulated primitive NSC survival and primitive NS formation. The Materials and methods section contains additional details regarding the primitive NS assay and the counting and normalization of data. Online supplemental material is available at .

genes play key roles during development. Members of this family of paired box/homeodomain transcription factors regulate the contribution of progenitor cells to different tissue types (). and its paralogue have been implicated in the specification of cells that will enter the myogenic program. In the absence of both Pax3 and -7, there is a major deficit in skeletal muscle, with arrest of myogenesis occurring during later embryonic and fetal development (). Cells in which the genes are activated become incorporated into other tissues or die in the double mutants. Normally, Pax3/7-positive skeletal muscle progenitor cells, which are derived from the central dermomyotome region of the somites (; ), activate the myogenic regulatory genes and differentiate into skeletal muscle fibers or remain as a proliferating reserve cell population within the muscle mass (; ; ). In late-stage fetal muscle, these cells begin to adopt a satellite cell position (; ), suggesting that this somite-derived population also provides the progenitor cells of postnatal skeletal muscle (). In these cells, the expression of and results in muscle cell determination. During the formation of early embryonic skeletal muscle in the somite, Myf5 and Mrf4 play a critical role in myogenic progenitors, which at this stage are derived from the edges of the dermomyotome (; ; ). Pax7 is not expressed in these cells in the mouse, where Pax3 is present. Early myogenesis occurs in the mutant; however, in a triple /() mutant no skeletal muscle forms and MyoD, which is required for skeletal muscle determination in the absence of Myf5 and Mrf4, is not expressed (). Therefore, Pax3, together with Myf5/Mrf4, regulates the activation of . Consistent with this conclusion, is up-regulated in embryos in which PAX3-FKHR, which acts as a strong transcriptional activator, has been targeted to an allele of (). Pax3 is essential for the survival of cells at the edges of the dermomyotome, particularly to those located hypaxially, where it is also required for the delamination and migration of muscle progenitor cells to other sites where skeletal muscle will form, such as the limbs (; ; ). When the coding sequence is targeted to the gene, Pax7 can substitute for Pax3 function in the trunk, but not in the limbs, suggesting that after the duplication of a common gene, which is present before vertebrate radiation, the functions of Pax3 and -7 diverge in response to the requirements of appendicular muscle formation (). Satellite cells, the myogenic progenitor cells of postnatal muscle, lie under the basal lamina of muscle fibers in a quiescent state until they become activated, proliferate, and form new skeletal muscle, which occurs during postnatal growth and in response to damage (). Myogenic regulatory genes are expressed during this process; is already expressed in quiescent satellite cells (), and is expressed as the cells become activated and subsequently differentiate with the expression of (). : double mutants have not yet been examined in this adult context because of the perinatal lethality of the original mutant. However, in the absence of MyoD, muscle regeneration is less efficient and the balance between proliferation and differentiation of myosatellite cells appears to be affected (). The most striking result, however, came from the examination of mutant mice (). Pax7 is present in satellite cells, and in its absence muscle regeneration is severely affected. Satellite cells were not observed in the mutant, leading to the proposal that Pax7 is essential for the specification of adult muscle progenitor cells (). However, it has recently been shown that satellite cells are present in the mutant, although in decreasing numbers as the mice mature, and it has been suggested that their proliferation is compromised in the absence of Pax7 (). Pax7 is present in quiescent satellite cells and during their activation, but is down-regulated when they differentiate. A proportion of activated satellite cells remain undifferentiated, retain Pax7 expression, and are thought to reconstitute the satellite cell pool (; ). Therefore, Pax7 appears to play a predominant role in adult muscle progenitor cells. The presence of Pax3, however, has been noted after satellite cell activation, leading to the proposal that it is implicated in their proliferation (). allele can be detected in quiescent satellite cells (). We now investigate the role of Pax7 in relation to Pax3, which we show is expressed in the quiescent satellite cells of a major subset of skeletal muscles. We show that both Pax3 and -7 control activation, and therefore regulate myogenesis in the adult. However, we demonstrate that in the absence of Pax7 satellite cells are progressively lost postnatally because of apoptosis accompanied by cell cycle defects. Pax7 has a critical antiapoptotic function in activated satellite cells for which Pax3 does not compensate. These results underline the critical role of upstream regulators of tissue formation and regeneration in assuring progenitor cell survival. Because Pax3 plays a key role during the onset of skeletal myogenesis in the embryo, we investigated its status in adult muscle in relation to Pax7. Analysis of adult mice in which the gene is targeted with reporters () revealed the presence of β-galactosidase (β-gal)–positive cells in adult skeletal muscle. The number of such cells varies between muscles. They are particularly abundant in the diaphragm ( A), but are much less frequent in hindlimb muscles, with the exception of the gracilis muscle ( B). In contrast, ∼50% of forelimb muscles express . As in the embryo (), expression of is not detectable in head muscles. Most ventral trunk muscles are positive, with a striking juxtaposition in the rib area, where intercostal muscles are mainly negative, whereas body wall muscles such as the serratus caudalis dorsalis are positive ( C). This difference is confirmed by semiquantitative RT-PCR analysis of Pax3 versus -7 transcripts in the tibialis anterior hindlimb muscle compared with the diaphragm ( D). The Pax3 protein is also present, as shown by Western blot analysis of different muscles ( D'). -expressing cells are found to be associated with muscle fibers (). The Pax3/7 protein is transcriptionally active in adult muscle, as indicated by activation of the P34 reporter transgene in which Pax3/7 binding sites regulate expression ('; ). These Pax3-expressing nuclei are present in satellite cells, as shown by coimmunolocalization of β-gal with the satellite cell markers CD34 and M-cadherin and by the inclusion of β-gal–positive cells within the basal lamina of the muscle fiber (, H–K'). Because Pax7 is present in satellite cells (), the question of Pax3 expression in relation to Pax7 was addressed. In the diaphragm, the majority of Pax7-positive satellite cells also coexpress Pax3. About 15% of the cells only label with Pax7, and, occasionally, cells that express only Pax3 are also detected (<3%; , L–N'). We therefore conclude that Pax3, like Pax7, is expressed in quiescent satellite cells and that the frequency of this event varies between muscles. There is no direct relation to fiber type () because the diaphragm muscle (type IIX, IIA, and I fibers) is positive, whereas in the hindlimb the soleus (type I and IIA) is negative. Similarly, in the hindlimb the gastrocnemius (mostly type IIB) is negative, whereas other fast muscles (type IIA and IIB) in the trunk and forelimb have Pax3-expressing satellite cells. We next examined the expression of Pax3 and -7 in primary cultures of satellite cells prepared from different muscles (). mice, which are Pax3 (β-gal)-positive, also express MyoD and down-regulate Pax3 (β-gal) in differentiated muscle fibers (, A–C). mice (), where expression of myogenin, which marks the onset of differentiation, is associated with rapid reduction in Pax3 (GFP) expression (, G–L). MyoD-positive cells in which Pax3 (β-gal) is low () have probably activated . MyoD ( C), or later myogenin (), is expressed in most cells marked by DAPI staining. In cultures from trunk (, D–F) and diaphragm muscle (not depicted), most nondifferentiating satellite cells coexpress Pax3 (β-gal) and -7. mice this is not the case and many colonies only express Pax7 (). Some colonies (≤20%) express both genes (, P–R). The distribution of these two types of colonies from different muscle sources is quantified in S. To confirm that is not activated in satellite cells that only express , we isolated these cells using flow cytometry. Based on the isolation of GFP-positive satellite cells from the diaphragm muscle, we had previously established the gating window that contains these cells (), which is shown as a boxed area (R2) in (T and U). When satellite cells isolated on this basis from the diaphragm muscle are cultured and reanalyzed by flow cytometry, they remain Pax3 (GFP)-positive ( T). However, when Pax3-negative satellite cells are sorted from the muscle of the lower hindlimb and cultured, no GFP-positive cells were found in the R2 window after re-sorting by flow cytometry ( U). These results demonstrate that activated satellite cells from Pax3-negative muscles, do not activate this gene in cell culture. In the genetic hierarchy that regulates the onset of myogenesis in the embryo, Pax3 activates (; ), and Pax7 can replace Pax3 in this function (). Therefore, we investigated the myogenic activity of the two Pax proteins in adult muscle by infecting cultured satellite cells with adenoviral vectors expressing wild-type or dominant-negative forms of Pax3 and -7, together with a GFP reporter. The dominant-negative proteins contain the repression domain of the engrailed transcription factor () fused to the NH-terminal region of the Pax sequence, which retains its DNA-binding domain. An initial series of experiments was performed with satellite cells isolated from a Pax reporter line, , in which the transgenic mice express β-gal from an reporter that is regulated by multimerized Pax3/7 binding sites (). Overexpression of Pax3 or -7 resulted in no obvious increase in reporter activity, which was already expressed at a high level in the Pax3/7-positive satellite cells. However, the dominant-negative versions of both of these factors (Pax3DN and -7DN) resulted in down-regulation of the reporter ( A). With decreasing levels of Pax3DN or -7DN, no significant difference in the repression exerted by either dominant-negative Pax was detectable (not depicted). We conclude that Pax3 and -7 bind to the consensus site with similar affinities and that their dominant-negative forms compete effectively with both endogenous proteins, which normally function as transcriptional activators in these adult muscle cells, as in the embryo (). When dominant-negative forms of Pax3 or -7 were expressed in satellite cell cultures, the level of the myogenic factor Myf5 was not markedly affected ( B), whereas MyoD was down-regulated in infected cells ( C). As for the Pax3/7 reporter transgene, similar dosage effects were seen for both Pax3DN and -7DN, with reduction, but not elimination, of MyoD in cells in which the adenoviral expression vector was expressed at a lower level ( C, arrowheads). These results are quantified in D. In contrast to the striking down-regulation of MyoD seen with the dominant-negative constructs, overexpression of wild-type Pax3 or -7 is compatible with MyoD expression ( E). Because myogenesis still occurs in the absence of MyoD ( F and not depicted), we investigated whether this is regulated by Myf5, which is expressed independently of Pax3 or -7 in the satellite cell cultures. and mice and infected with Pax3DN and -7DN adenoviral vectors ( F). Unlike infected cells from heterozygous mice, mutant cells do not express myogenin or differentiate in the presence of the dominant-negative Pax vectors. This demonstrates that either Pax3/7, acting via MyoD, or Myf5 are required for the myogenic differentiation of adult satellite cells. Because Pax3-expressing satellite cells are found in adult muscles, their potential contribution to muscle growth and regeneration was investigated in the mutant mouse. We first evaluated whether Pax3 continues to be expressed in diaphragm and trunk muscles. This is the case as shown in at postnatal day 3 (P3). Western blots show that Pax3 is present, although at a reduced level ( A), and immunohistochemistry confirms that Pax3 is expressed in satellite cells ( B). muscle at P2 ( C) and, indeed, even at P10 cultures from the diaphragm contain MyoD-positive cells that differentiate into myofibers expressing troponin T ( D), although such cells are much rarer (≤15% of wild type). Similarly, single fiber experiments with the extensor digitorum longus hindlimb muscle (Fig. S1, available at ) show that satellite cells are still present in the mutant, but that their number is reduced to ∼10% of normal levels at P10. The number of myonuclei is also reduced by ∼50% at this stage (Fig. S1 F) and muscle fibers are smaller (not depicted), consistent with the reduced size of mutant mice (; ). muscles and compared with heterozygotes at P2 and P11 ( E). At P2 there is only a small reduction (∼18%) in the numbers of satellite cells in trunk, diaphragm, or hindlimb muscles. By P11, this reduction is striking (∼83%) and is similar for all three muscle sources, whether Pax3 is expressed in most satellite cells (e.g., diaphragm) or not (hindlimb). This demonstrates that Pax3 cannot compensate for Pax7 function during the postnatal development of skeletal muscle. This is not because of a failure in myogenesis in the mutant satellite cells. is still expressed and, as in wild-type cultures (), its expression is inhibited by a dominant-negative form of Pax3 (). Because the numbers of satellite cells fall during postnatal development in mutant mice, and muscle fiber size is reduced, it is possible that activated satellite cells, which contribute to muscle growth, do not proliferate normally. and mice, plated at low density so that the number of cells per colony could be monitored. A reduction of 25–30% in the number of cells per myogenic colony was observed ( A), indicating that, in the absence of Pax7, proliferation is affected and that this is also the case for colonies from a muscle, such as the diaphragm, which expresses Pax3. To determine whether some cells have withdrawn from the cell cycle, a Ki67 antibody was used, which marks proliferating cells in all phases of the cell cycle (). There was no difference between colonies from mutant or heterozygous mice, where the proportion of proliferating cells is concerned ( B). We next examined progression of cells through the cell cycle, using a cyclin A antibody that marks cells in the S and G phases (). There were 33% more cyclin A–positive cells in the mutant myogenic colonies ( C). One possible explanation for the increase in the proportion of proliferating cells in the S and G phases of the cycle in the mutant is that some cells exit the cycle during G1 and immediately undergo apoptosis, resulting in a reduction in the number of cells per colony observed in the absence of Pax7 ( A). In the colony assay, we did not detect a significant difference in the numbers of dying cells using standard markers of apoptosis (not depicted). This may be because such cells detach immediately. Alternatively, there may be a cell cycle defect independent of apoptosis, such as a cell cycle arrest in G, resulting in a slower progression through the cell cycle. To investigate the survival of satellite cells in postnatal muscle in vivo, in the absence of Pax7, we used an antibody to the activated form of caspase-3 to label cells undergoing apoptosis (). Coimmunohistochemistry was performed with an antibody to desmin, which marks activated satellite cells as they assume a myoblast phenotype (), as well as muscle fibers. In the postnatal trunk muscle of mutant mice, activated caspase-3–labeled cells are observed, whereas in control mice labeled cells are not detected (, A–D). These results are quantitated in E. The decrease observed from P0 to P6 reflects the decreasing numbers of satellite cells in the mutant. These cells also express desmin, suggesting that they correspond to activated satellite cells, probably contributing to the postnatal growth of muscle (, arrowheads). The identification of these cells was confirmed by labeling with a laminin ( C) or β-gal antibody ( D). The latter detects transcription, which marks satellite cells. During postnatal development, apoptotic cells were detected in all trunk and limb muscles examined in the mutant, whereas they were very rare in the muscles of normal mice ( E). Therefore, we conclude that Pax7 has an antiapoptotic function and that in its absence satellite cells die, despite the presence of Pax3. To compare the roles of Pax3 and -7 in protecting against apoptosis, wild-type satellite cells were transfected with GFP-marked adenoviral vectors coexpressing a dominant-negative form of Pax3 or -7. These cells were analyzed by flow cytometry on the basis of GFP expression, and their viability was measured by propidium iodide (PI) staining, which detects dying cells (). Such an experiment is shown in A for cells from hindlimb muscle infected with a dominant-negative form of Pax3 or -7, which led to 71% of dying cells in the GFP-positive population when Pax7DN was expressed. The results of these experiments are summarized in B. Whereas Pax7DN led to substantial cell death, the Pax3DN-expressing virus at similar or sixfold higher multiplicities of infection did not show any effect on these cells, relative to control values. Similar results were seen in primary cultures from young mice (P7; not depicted) to those shown here for 3–4-wk-old animals (). No such apoptotic effect was observed when muscle cultures were infected with adenoviral vectors expressing Pax7. mice infected by Pax7DN a similar extent of cell death was observed (not depicted). Furthermore, Pax7DN (or Pax3DN) did not provoke cell death in nonmyogenic cells, such as the OP9 bone marrow stromal cell line (not depicted). When cells isolated from the diaphragm ( C) were similarly infected, cell death caused by Pax7DN was about half of that witnessed in cells from hindlimb muscle. Equivalent levels of Pax3DN showed no effect, but in contrast with the observations on the hindlimb ( B), when the concentration of Pax3DN was increased, some cell death was observed. These results suggest that Pax3 can have a limited antiapoptotic effect in the muscles in which it is expressed. In keeping with this conclusion, more myogenic colonies are present in cells isolated from the diaphragm of mutant mice at P4, although in both diaphragm and hindlimb preparations, this number was strikingly lower than with the wild type ( D). At later stages no difference was detectable. #text () and () mice were genotyped on the basis of the “plotch” phenotype (). mice () were genotyped by PCR, using the following primers: DPax7Ex2A: CTTggCCAAggCCgggTCAATCAgCTTggTggg; RLacZ3: AAATTCAgACggCAAACgACTgTCCTggCC; and RPax7Ex3C: gATggACCCAgTCTCCTgATATCggCACAg. The wild-type band was amplified using DPax7Ex2A/RPax7Ex3C (800 bp), whereas the mutated allele was amplified using DPax7Ex2A/RLacZ3 (500 bp). Cells were prepared from muscle tissue of mice at different time points after birth by enzymatic dissociation, as previously described (; ). Cells were plated on gelatin-coated dishes in a 1:1 mixture (vol/vol) of Ham's F12 and DME (GIBCO BRL) containing 20% (vol/vol) fetal calf serum (AbCys) and 2% (vol/vol) ultroser (Biosepra). This medium, which supports both the proliferation and differentiation of muscle cells (), was used in all experiments. To allow the formation of colonies of muscle cells, primary cultures were plated at a density of 100 and 200 cells/cm. When plated under these conditions cultures were highly enriched in myogenic colonies. The number of mononucleated cells to be plated was determined by counting after labeling an aliquot with 5 μg/ml of the DNA dye bis-benzimide (Hoeschst). Single fiber preparation and culture were performed according to . Cells were treated as previously described (). In brief, after fixation with 4% (wt/vol) paraformaldehyde and permeabilization with 0.2% (wt/vol) Triton X-100, cells were incubated with antibodies diluted in PBS containing 0.2% (wt/vol) gelatin. All incubations were at room temperature. For immunofluorescence, cells were mounted in mowiol (Calbiochem) after the staining of DNA with 5 μg/ml bis-benzimide in the penultimate PBS wash. Antibodies used were as follows: Myf5, rabbit polyclonal (), at a 1:1,000 dilution; MyoD, either a rabbit polyclonal (Santa Cruz Biotechnology, Inc.), at a 1:200 dilution, or a mouse monoclonal (clone 5.8A; DAKO), at a 1:200 dilution; troponin T, mouse monoclonal (clone JLT12, Sigma-Aldrich), at a 1:200 dilution; desmin, mouse monoclonal (clone D33; DakoCytomation), at a 1:200 dilution; laminin, rabbit polyclonal (Sigma-Aldrich), at a 1:200 dilution; M-cadherin, mouse monoclonal (clone 12G4; Nanotools GmBh), at a 1:200 dilution; antiactive caspase-3, rabbit polyclonal (BD Biosciences), at a 1:250 dilution; cyclin A, a rabbit polyclonal (gift from A. Fernandez and N. Lamb, Institut de Génétique Humaine, Montpellier, France), at a 1:200 dilution; Ki67, a mouse monoclonal (BD Biosciences), at a 1:100 dilution; Pax7, mouse monoclonal (Developmental Studies Hybridoma Bank), at a 1:100 dilution; Pax3, mouse monoclonal (provided by M. Bronner-Fraser, California Institute of Technology, Pasadena, CA), at a 1:100 dilution; β-gal, either rabbit polyclonal (Invitrogen), at a 1:4,000 dilution in cell culture experiments, or another rabbit polyclonal used on sections (provided by J.-F. Nicolas, Institut Pasteur, Paris, France), at a 1:500 dilution, or mouse monoclonal (clone Gal13; DakoCytomation), at a 1:100 dilution. Secondary antibodies were coupled to a fluorochrome, either Alexa 488 or 594 (Invitrogen), at a 1:250 dilution. For X-Gal (Roche) staining, single fibers were fixed for 30 min with 4% paraformaldehyde in PBS, on ice. Fibers were rinsed twice with PBS, and then stained with X-Gal, using 0.4 mg/ml X-Gal in 2 mM MgCl, 0.02% NP-40, 0.1 M PBS, pH 7.5, 20 mM KFe(CN), and 20 mM KFe(CN) for 4–16 h at 37°C, with shaking. Fibers were rinsed in PBS, postfixed overnight in 4% paraformaldehyde, and mounted after washing in PBS-buffered mowiol with DAPI. Similar conditions were used for X-Gal staining of sections. X-Gal staining of cells was performed after a 5-min fixation in 4% paraformaldehyde in the same solution used for fibers, but without NP-40. RNA extracts were prepared from tibialis anterior and diaphragm muscles of 8-wk-old mice, using TRIzol (Invitrogen). Reverse transcripts were generated using Power Script reverse transcriptase (BD Biosciences). The primers for Pax3 were DPax3-740 (TGCCCTCAGTGAGTTCTATCAGC) and RPax3-1100 (GCTAAACCAGACCTGCACTCGGGC), which generate a 360-bp PCR fragment. The primers for Pax7 were Dpax7-140 (TGGAAGTGTCCACCCCTCTTGGC) and RPax7-650 (ATCCAGACGGTTCCCTTTGTCGCC), which generate a 510-bp PCR fragment. Three other primer pairs were used and gave similar results. PCR products were separated on 1.5% agarose gels, using standard techniques, and revealed by UV light (Image Master CVS; GE Healthcare). Protein extracts were prepared and analyzed by Western blotting, as previously described (). The antitubulin antibody (clone 5HI; BD Biosciences) was used at a 1:2,000 dilution and the Pax3 antibody at a 1:400 dilution. Adenoviral vectors were generated using standard molecular biology techniques. In brief, dominant-negative Pax3 and -7 constructs were made by fusing in-frame sequences encoding the engrailed repression domain (298 amino acids; ) to the first 340 amino acids of murine Pax7 (to generate the Pax7DN construct) or the first 374 amino acids of Pax3 (to generate the Pax3DN construct). Dominant-negative activity of Pax7DN and -3DN was verified by cotransfection in 293 cells with a plasmid containing polymerized Pax3/7 binding sites () in front of a thymidine kinase (Herpes virus) minimal promoter followed by a reporter gene. Before their introduction into the adenovector, Pax7DN and -3DN were cloned into the Adtrack-CMV shuttle vector and recombinant adenoviruses with a GFP reporter sequence were generated as described previously (). High-titer viral stocks were prepared by repeated infection into the packaging cell line 293T. Viruses were purified by CsCl banding followed by passage through a 2.5-ml Sephadex G25 column (GE Healthcare) for desalting and stored in aliquots at −80°C. The titer of each preparation was determined after infection of 293 cells by limiting dilution of virus and detection of GFP expression. mice have permitted us to define parameters for isolating adult muscle progenitor cells both from Pax3-expressing muscles (e.g., diaphragm) and non-Pax3–expressing muscles (e.g., lower hindlimb muscles; ). In the case of the latter, this was on the basis of the size and granularity of CD34 cells. GFP-positive muscle progenitor cells from diaphragm muscle and GFP-negative muscle progenitor cells from lower hindlimb muscles, isolated by flow cytometry, were maintained in culture for 6 d as proliferating cells before a second flow cytometric analysis for detecting the presence of GFP-positive cells. Fig. S1 presents an analysis of satellite cells present on single fibers isolated from the extensor digitorum longus hindlimb muscle of Pax7+/− and Pax7−/− mice at P10. Online supplemental material is available at .

Muscle satellite cells are thought to represent the only population of committed myogenic progenitors in adult skeletal muscle. The activation of muscle satellite cells generates proliferative myogenic precursor cells that differentiate to repair and replace damaged fibers (). The transcription factors regulating the specification and differentiation of satellite cell–derived myogenic progenitors is analogous to the molecular mechanisms regulating embryonic myogenesis (). Pax7, a paired-box transcription factor, is specifically expressed in quiescent and newly activated satellite cells. skeletal muscle demonstrates a requirement for Pax7 in the function of the satellite cell lineage (). mice appear normal at birth but fail to thrive and subsequently die at 2–3 wk from unknown causes (; ). myofibers is attributable to the lack of satellite cell fusion during the postnatal growth of muscle (). muscle suggests that embryonic and fetal myogenesis is unaffected and identifies a unique requirement for Pax7 in the satellite cell lineage. Recent work has identified stem cell populations capable of giving rise to satellite cells during regeneration. Muscle-derived side-population cells separated on the basis of Hoechst dye exclusion give rise to satellite cells after intramuscular or intravenous injection (; ). In addition, bone marrow–derived cells similarly have the capacity to engraft skeletal muscle and form myogenic satellite cells after whole-body irradiation and transplantation (; ; ; ). More recently, we demonstrated that endogenous CD45 muscle-derived cells give rise to Pax7 myogenic cells in response to Wnt proteins during regeneration (). Furthermore, ectopic expression of Pax7 in CD45/Sca1 cells is sufficient for their myogenic specification (). Together, these data support the notion that a developmental relationship exists between adult stem cells and Pax7-dependent satellite cell myogenic lineages. Pax3, a paralogue of Pax7, is critical for the delamination and migration of the somitic muscle progenitor cells to the limb buds, as mutant mice lack limb muscles (; ; ). Despite these distinct functions of Pax3 and -7 in muscle development (), recent studies have begun to elucidate a common playground for these paralogues. Specifically, a novel population of Pax3/Pax7 double-positive (Pax3/Pax7) stem cells were identified in the dermomyotome of the embryonic somites (; ; ; ). Proliferating Pax3/Pax7 cells were observed to persist throughout embryonic and fetal development and later to give rise to a subset of muscle satellite cells. In the absence of muscle environment or Pax3/Pax7 expression, these somitic stem cells apoptose or adopt alternative nonmuscle lineages. Interestingly, Pax3 expression in satellite cells is mostly down-regulated before birth, although a subset of satellite cells appears to express Pax3 (; ; ). In this study, we set out to determine the relative role of Pax7 and -3 in postnatal muscle growth and regeneration. mice in C57/B6 background appear normal at birth but fail to thrive and subsequently die at 2–3 wk of age (; ), preventing them from further postnatal analysis. Therefore, carrying a knockin of a β-galactosidase (β-gal) cassette ( ) were backcrossed nine generations into the 129Sv/J genetic background, where some mutant mice were viable. 129Sv/J mice allow us to examine not only muscle growth and regeneration in the absence of Pax7 expression but also the function of Pax3 independent of Pax7 in postnatal muscles. muscle was greatly compromised, suggesting an essential role for Pax7 in these processes. Furthermore, we identified a novel population of Pax3-expressing myogenic progenitors in the interstitial space of adult skeletal muscles. Together, these results demonstrate an essential role for Pax7 in the productive formation of myogenic progenitors during postnatal growth and regeneration of skeletal muscle. muscles during the postnatal period in the 129Sv/J genetic background. mice appeared normal during embryonic development until birth, as demonstrated by the normal body weight as compared with control littermates at postnatal day (P) 0 (). mice failed to maintain normal postnatal growth. At P3, mice were only two thirds of the weight of wild-type littermates, and at P10, they were less than half the weight of wild-type littermates. and wild-type littermates (). mice was at least partially attributable to decreased skeletal muscle growth as revealed by the significant size/weight decreases in tibialis anterior (TA) and other skeletal muscles (; unpublished data). mice resulted from reduced myofiber size or number or both, we enumerated the total number of myofibers at muscle mid-belly and determined the myofiber cross-sectional area. The total fiber number in both extensor digitorum longus (EDL) and soleus muscles was not different between and wild-type littermates at P10 (). mice. More precisely, the cross-sectional areas of Types II and I myofibers (253 ± 16 and 289 ± 15 μm, = 4, respectively) in the mutant soleus were ∼1.5- and 1.8-fold smaller than those of wild-type fibers (374 ± 25 and 512 ± 29 μm, = 4, respectively) at P10, whereas no significant difference in the cross-sectional area of Type I myofiber was detected at P0 ( : 205 ± 37 μm, = 2; : 192 ± 29 μm, = 2). mice directly resulted from defects in postnatal growth. mice in both soleus and EDL muscles (), suggesting a deficiency in (or lack of) satellite cell function to supply myonuclei to the growing myofibers. skeletal muscles is severely retarded because of inadequate myonuclei increase and myofiber growth. muscle during aging, we analyzed the muscle phenotype of aging mice. mice displayed prominent kyphosis (curvature of the spinal column), which is typical of extensive muscle wasting and a hallmark of aging (; ). Evidence for muscle wasting was strikingly depicted by a progressive decline in the number of TA myofibers from 2–3 mo of age ( C). mice was decreased to one third of that in muscle ( C). mice ( D). TA, the number of newly regenerated myofibers, as indicated by the centrally located nuclei (5 ± 1 per cross section, = 8), remained comparable to that of wild-type littermates (4 ± 1, = 7). mice. TA muscles further demonstrated the defective regenerative response during aging (). muscle fibers did not display signs of extensive damage, suggesting that the loss of muscle did not result in a reduction in fiber integrity. fibers were resistant to Evans blue dye incorporation (unpublished data), and serum creatine kinase levels were normal at all ages studied (216 ± 92 U/liter [ = 2] and 579 ± 179 U/liter [ = 6] in ; 328 ± 206 U/liter [ = 2] and 1148 ± 500 U/liter [ = 6] in at P3 and in adults, respectively). mice is accelerated as a function of age, and the accelerated fiber loss is not adequately compensated because of an impairment in muscle regenerative ability. muscle, we injected cardiotoxin (CTX) into TA muscles to chemically induce injury. TA regained the overall normal appearance of regenerated muscle characterized by numerous centrally nucleated regenerating myofibers (). There were >700 regenerating fibers per TA cross section at both 10 d ( = 2) and 1 mo ( = 3) after injury, without appreciable deposition of calcium, adipose, or fibrotic tissues (, A–D). TA displayed a severe regeneration deficit, with only rare centrally nucleated myofibers observed at 10 d (9 ± 6 fibers/TA cross section, = 3) and 1 mo (61 ± 50 fibers/TA cross section, = 4) after injection (). TA failed to mature and remained significantly smaller than those in controls even 1 mo after injury (compare B with F [arrows]). TA had been replaced by extensive adipose tissue ( F, arrowheads), fibrotic tissue ( G, arrowhead), and deposition of calcium ( H). Similar results were obtained after CTX injection in gastrocnemius muscles (unpublished data). mice appears to be the most striking regeneration deficit reported in any mouse model. mice after physically (crush) induced injuries. TA and gastrocnemius muscles ( K, arrows, and not depicted) contained only rare centrally nucleated myofibers with extensive calcium deposition ( L) and fibrosis (not depicted) 10 d after crush. littermates were fully regenerated with large centrally nucleated fibers ( I and not depicted) and without calcium deposits ( J). Together, these experiments indicate that Pax7 is necessary for efficient muscle regeneration. muscle at 10 d and 1 mo after injury did, however, suggest a limited capacity for muscle differentiation. mutant expressed embryonic MyHC and high levels of Desmin (), confirming their newly regenerated state. and muscles contained BrdU-labeled nuclei at 10 d after injury, indicating that proliferating cells had differentiated and fused into these myofibers (). muscle, only 3.5% of BrdU-labeled nuclei were found within muscle fibers, compared with 45% of BrdU-positive nuclei within myofibers in regenerated muscles. myofibers were multinucleated and extended over several hundred micrometers (unpublished data). myofibers observed after muscle injury were regenerated rather than surviving myofibers. It has been reported that satellite cells are formed in the absence of Pax7 and that these cells have some capacity for myogenic function (). satellite cells and whether these cells were capable of following the myogenic developmental program. mice. In mice, all β-gal cells also coexpressed Pax7 ( A) and vice versa, validating the absolute specificity of β-gal labeling. Rare β-gal cells were detected on EDL fibers isolated from younger (P25) mice ( B). myofibers typically displayed a small round shape indicative of activation by the single fiber isolation procedure, most β-gal cells displayed a morphology characterized by a large cell body with long filopodium-like processes (). mutant fibers (1.1 ± 0.2/fiber, = 28) was ∼7% of that of the control (15.9 ± 2.2/fiber, = 25) at P25. mutant muscle (6.9 ± 1.3/microscopic field under 20×, = 7) was less than half that of the control (16.0 ± 3.8/field, = 5; Fig. S1, available at ). β-gal cells were capable of proliferation upon activation, single fibers were isolated and cultured in horse serum–coated Petri dishes to prevent attachment of fibers. myofibers displayed characteristic large aggregates of proliferative satellite cell–derived myoblasts that expressed MyoD and/or Pax7 (). fibers ( = 29). Occasionally, pairs of nuclei that expressed low levels of MyoD were found within arrested single-blobbing cell bodies (). We did not detect Caspase3 expression in these cells (unpublished data), suggesting that Pax7 mutant satellite cells underwent an abortive mitosis and failed to complete cell division. We next examined whether the Pax7-remnant cells on isolated single myofibers still retained typical satellite cell features. or fibers coexpressed Pax7 and Syndecan4 () or CD34 ( H), suggesting that Syndecan4 () and CD34 () are reliable markers of satellite cells. Specifically, 8.7 ± 1.9 Syndecan4 cells/fiber and 3.9 ± 1.8 CD34 cells/fiber were detected in 6–8-wk-old adult fibers ( = 16 fibers/3 mice for Syndecan4; = 35 fibers/2 mice for CD34). fibers ( = 56 fibers/3 mice for Syndecan4; = 65 fibers/2 mice for CD34). fibers were negative for syndecan4 ( C) and CD34 (, K–M). fibers when care was deliberately not taken during single fiber preparation. These cells were uniformly negative for β-gal immunostaining (, C–E and N). Notably, these syndecan4, β-gal cells were also positive for the endothelial marker CD31 (, D–G). musculature, these cells have lost the characteristics and function of wild-type satellite cells. fiber ( = 76 fibers/3 mice) yielded any mononuclear cells (mean of 1.85 ± 0.34 cells/fiber) after several days of attached culture in growth medium. fibers were uniformly negative for the myoblast-specific markers MyoD, Myf5, and Desmin (unpublished data). fibers ( = 60 fibers/3 mice) gave rise to an average of 11.8 ± 0.97 cells/fiber. myoblasts were generated by all fibers (unpublished data). mutant mice. hind limb muscle, single-cell suspensions were prepared from hind limb muscles and analyzed after 15 h in growth conditions or after an additional 3 d in differentiation medium. muscles yielded some myogenic cells expressing MyoD () or Myf5 ( D). myogenic cells were recovered at an extremely low frequency (∼1/150; ) compared with wild-type muscle (363,012 ± 34,247 [ = 2] and 2,961 ± 1,096 [ = 4] MyoD cells and 242,132 ± 5,843 [ = 2] and 1,277 ± 578 [ = 3] Myf5 cells/g of hind limb muscle for and , respectively). TA 3–4 d after CTX injection (unpublished data). cells isolated from limb muscle formed small colonies of 6–20 MyoD and Desmin myoblasts after 10–20 d compared with proliferative bursts of >500 myoblasts in cultures (unpublished data). cells to expand in culture suggested a profound proliferative deficit under standard myoblast growth conditions as compared with myoblasts isolated from wild-type littermates. Even when cultured in the presence of high concentrations of growth factors (5% chick embryo extract; 50 ng/ml of stem cell growth factor; 1 μg/ml of insulin; Methocult M3434 [StemCell Technologies, Inc.] or in Matrigel-coated dishes), myogenic cells did not proliferate. myogenic cells underwent myogenic differentiation and expressed MyHC after culture in low-mitogen medium after 3–5 d ( A). or muscles, an average of 311.3 ± 49.3 ( = 6) nuclei/20× field were found within MyHC cell bodies. muscle, an average of 3.0 ± 1.5 ( = 11) nuclei were found within MyHC cell bodies. muscle, all MyoD cells expressed MyHC after 6 d of culture in differentiation conditions. In contrast, in the control, ∼10% of MyoD cells did not express MyHC. mice were plated on Matrigel and cultured for 3 d in growth media followed by 1 d in differentiation media. EDL muscle yielded large numbers of MyoD-, Myf5-, and MyHC-expressing cells together with formation of large myotubes ( C). EDL myofiber bundles gave rise to low numbers of MyoD- and Myf5-expressing cells that were capable of forming myotubes ( D). single fibers free of interstitial tissues did not yield any Myf5-, MyoD-, or MyHC-expressing cells (see Absence of functional satellite cells…). myogenic cells resides in an alternate anatomical location and support the contention that myogenic cells represent an interstitial myogenic-progenitor population distinct from the satellite cell compartment. To investigate the possibility that the Pax7-independent myogenic cells in adult muscle represent a Pax3-dependent lineage, cell cultures from limb and diaphragm were analyzed for Pax3 expression by immunohistochemistry. diaphragm (unpublished data) and limb ( B) muscles yielded similar numbers of MyoD/Pax3 cells ( B). The majority of wild-type MyoD cells did not express Pax3 ( B, arrows) but expressed Pax7 (not depicted). muscle, virtually all MyoD cells coexpressed Pax3 ( B, arrowheads), with very few MyoD/Pax3 cells detected. muscle. To demonstrate the presence of the Pax3 myogenic cell population in wild-type animals, we performed RNase protection assay, in situ hybridization, and immunostaining on wild-type muscles. By RNase protection assay, primary myoblasts isolated from adult limb muscles and expanded over several weeks expressed extremely low levels of mRNA, and we were unable to detect Pax3 protein by Western blot analysis (unpublished data). In addition, mRNA was below the limit of detection in C2C12 myoblasts and myotubes. limb musculature outside the basal lamina (, A–E and G–K). muscle were uniformly observed in a sublaminar position (). Limb muscle of wild-type mice at P1 was also examined and we found a similar interstitial localization of Pax3 cells and sublaminar positioning of Pax7 cells (Fig. S2, available at ). diaphragm (). Immunolabeling of the basal lamina confirmed the location of the Pax3 cells outside the satellite cell location in the diaphragm (unpublished data). and muscle (1.6 ± 0.5 and 1.1 ± 0.5 Pax3 cells/section in and by in situ hybridization, respectively). These analyses reveal the presence of myogenic Pax3 cells in the interstitial environment in a niche distinct from the satellite cell compartment in both mutant and wild-type musculature. Finally, we tested for whether the rare Pax3 myogenic cells have any biological function in vivo. TA and gastrocnemius muscles were examined for expression of Pax3 and MyoD 2 d after CTX-induced regeneration. or mice ( and not depicted). Within 2 d after CTX injection in mice, ∼84% Pax3 cells expressed MyoD ( = 74 cells from four 20× microscopic images taken from regenerative regions); ∼11% of the MyoD cells were also Pax3 ( = 569 cells from four regenerative regions; ). In the regenerating muscles of mice, ∼29% Pax3 cells coexpressed MyoD (; = 80 cells from five regenerating regions). and muscles, the total number of MyoD cells in regenerating muscle was only 5% of that in muscles, a result consistent with the impaired muscle regeneration in mutant mice. regenerating muscles ( = 38 cells from five regenerative regions), ∼61% coexpressed Pax3, suggesting that the majority of MyoD cells were derived from the Pax3 myogenic cells. In addition, the frequency of Pax3 cells in the regenerating region (>10 cells/field) was higher than in the resting tissue (1–2 cells/section; see the previous paragraph), suggesting that Pax3 myogenic cells are preferentially recruited to the regenerating region and/or Pax3 expression is up-regulated during regeneration. Together, these results indicate that these interstitial Pax3 cells found in limb musculature likely represent a novel lineage of myogenic progenitors that is distinct from the Pax7 satellite cell lineage of myogenic cells. mice clearly establishes an essential requirement for Pax7 during postnatal muscle growth and regeneration. Importantly, these experiments demonstrate that Pax7 is required for presatellite cells to express genes normally associated with functional satellite cells located in the sublaminar niche. Moreover, Pax7 is also required for the productive formation of committed myogenic precursor cells from sublaminar satellite cells. Furthermore, our experiments have identified a novel population of interstitial Pax3-expressing myogenic progenitors in adult limb musculature. mice. muscle regeneration have also been reported in a mixed C57/BL/Sv129 genetic background (). mice in the mixed background. Our experiments do not support these findings. mice with loss of fast myofibers, calcium deposition, and chronic degeneration associated with aging. The specific loss of fast muscle fibers can be attributed to the deficient fiber growth, as manifested by the reduced fiber size and myonuclei number per fiber, susceptibility of fast fibers to damages, and inefficient regeneration mechanisms to replace damaged fibers. muscle was rapidly replaced by an inflammatory infiltration, together with large calcium deposits and extensive adipose infiltration. Only rare regenerated myofibers were formed. Lastly, our results demonstrate that Pax7 is required for the normal function of satellite cells and the productive formation of myogenic precursor cells. Therefore, our study unequivocally demonstrates that -dependent myogenic cells, which comprise the satellite cell compartment, are required for the efficient regeneration of skeletal muscle. myofibers suggested the presence of Pax7-independent myogenic progenitors within the muscle. muscle at a frequency of ∼1/150 as compared with wild-type muscle. However, our in vivo data suggests that Pax3 myogenic progenitors may give rise to as much as 5% of MyoD cells during regeneration. myofibers were analyzed to conclusively demonstrate that these myogenic cells were not associated with myofibers and therefore were not satellite cells. Specifically, control EDL myofibers contained ∼9–10 Syndecan4 satellite cells per fiber. fibers at a frequency of 1/150, one satellite cell would be observed for every 16 fibers. From our experiments, no Syndecan4 cells were detected on a total of 197 fibers examined either by direct immunohistochemistry or following in vitro culture. fibers, these cells did not express the satellite cell markers CD34 and Syndecan4 and were incapable of on-fiber proliferation in culture, suggesting that they were incapable of productively forming myogenic precursor cells. mice were not associated with myofibers, arguing against a possible contribution from any residual Pax7-deficient satellite cells. Recent findings have demonstrated the presence of resident or circulating stem cells in the muscle (; ; ; ; ; ; ; ; ; ). Our results suggest that these stem cells are insufficient to compensate for the loss of satellite cell function in the absence of Pax7 or suggest that Pax7 is a key player in the myogenic specification of most, if not all, adult stem cells. mice, by our previous studies showing that Pax7 expression is necessary for the myogenic differentiation of CD45 muscle-derived stem cells (), and by recent studies showing that embryonic stem cells giving rise to satellite cells apoptose or undergo nonmyogenic differentiation in the absence of Pax7 and -3 (). The molecular function of Pax3 and -7 in myogenic specification remains to be fully defined. Genetically, and function upstream of and , and Pax3-FKHD is capable of activating MyoD transcription in fibroblasts (). In addition, several recent studies have reported that newly activated satellite cells and proliferating myoblasts coexpress Pax7 and MyoD in vitro (; ; ). Together, these findings support the notion that Pax3/Pax7 directly or indirectly activates the transcription of the myogenic regulatory factors. Moreover, ectopic expression of Pax7 leads to enhanced proliferative and survival potential of myoblasts (). Likewise, transient activation of Pax3 expression in response to Notch signaling has been reported in cultures of primary myoblast (), resulting in enhanced proliferation of these cells. Interestingly, overexpression of Pax7 also seems to down-regulate MyoD and promote cell-cycle withdrawal from the proliferating state, therefore playing a critical role in the maintenances of the satellite cell pool (; ). Our present data demonstrate that satellite cells lacking Pax7 undergo arrest when forming myogenic cells. Pax7 appears to function at multiple levels, including activation of the myogenic program, stimulating the proliferation and survival of myogenic progenitors, and the self-renewal of satellite cells. Our experiments unequivocally demonstrate that Pax7-deficient satellite cells that survive are incapable of giving rise to functional myogenic progenitors. Therefore, we conclude that Pax7 is required for sublaminar satellite cells to either form functional myogenic daughter cells or maintain a renewable satellite cell pool. Importantly, these data imply that myogenic commitment or specification occurs when satellite cells undergo an asymmetric cell division to form a myogenic factor–expressing daughter cell. mice uniformly expressed Pax3. Immunolabeling and in situ hybridization revealed the presence of Pax3 cells in interstitial locations outside the basal lamina of muscle fibers and not in the sublaminar satellite cell niche. The interstitial Pax3 cells did not express MyoD in undamaged muscle. However, after injury, the Pax3 cells rapidly increased in number and were found to express MyoD. Expression of or knockins has recently been reported in postnatal diaphragm in a subset of Pax7-expressing satellite cells, but expression was not detected in satellite cells in most hind limb musculature (; ; ), as opposed to our observation that they are located outside muscle fibers. In our experiments, we only detected Pax3 protein in rare cells located outside of the basal lamina and not in satellite cells. This discrepancy may be the result of the different labeling techniques used and the different thresholds of detection between immunostaining of endogenous Pax3 versus bacterial enzymes or GFP with long half-lives. It is also possible that the extralaminar Pax3-expressing cells represent presatellite cells that require Pax7 to become functional sublaminar satellite cells. However, the presence of these cells in both wild type and mutant muscle, and the absence of detectable Pax3 protein in Pax7-deficient LacZ cells, supports the notion that the Pax3 cells represent a novel myogenic lineage. Future studies applying immunolabeling to the knockin mice may help clarify the discrepancy. Together, our experiments suggest that the interstitial Pax3 cells represent a novel myogenic lineage that is distinct from the sublaminar Pax7 satellite cell compartment. The normal role of these Pax3-expressing myogenic progenitors in adult muscle remains to be established. It is interesting to speculate that Pax3-expressing myogenic cells have a specialized role in adult muscle. For example, it is possible that these cells have a role in the formation of muscle spindles, neuromuscular junctions, myotendinous attachments, or other muscle specializations. Future studies characterizing the expression, differential activity, and developmental role of Pax3 and -7 in postnatal myogenesis are thus required. Mice carrying a targeted reporter allele of provided by P. Gruss (Max Plank Institute for Biophysical Chemistry, Göttingen, Germany; ) were backcrossed into the 129Sv/J genetic background for nine generations to generate 129Sv/J-inbred mice carrying the allele. In these mice, the gene was knocked out by insertion of a β-gal gene and a neomycin cassette into the first exon of . The muscle phenotype was indistinguishable between C57BL/6- or 129Sv/J-inbred backgrounds and a C57BL/6 × 129Sv/J-outbred background. Mouse serum was prepared by coagulation and centrifugation of blood samples and assayed by the Department of Biochemistry at the Children Hospital of Eastern Ontario. and littermates were used for all regeneration experiments. However, no difference was found in the regeneration efficiency of age groups tested. We therefore pooled all data together in the results. For CTX-induced muscle regeneration, mice were killed at 10 d or 1 mo after injection of 25 μl CTX (10 μM; Latoxan) into the TA or the gastrocnemius muscles. For cell proliferation assays, 30 mg/kg BrdU (Sigma-Aldrich) was injected i.p. on days 4, 6, 7, and 8 after CTX injection. For crush injury, TA or gastrocnemius muscles were crushed using large forceps. Experiments were performed under University of Ottawa regulations for animal care and handlings. Single myofibers were isolated by collagenase digestion as previously described () and subsequently plated in Matrigel-coated chamber slides for attached culture or suspended in 60-mm Petri dishes coated with horse serum to prevent fiber attachment. Myofiber bundles, which contained ∼10–50 myofibers and surrounding tissue, were prepared and plated in chamber slides as per single myofiber culture but without extensive trituration. Primary myoblasts were isolated from hind limb or diaphragm muscles and cultured as previously described (). For all injury and regeneration experiments, TA muscle was isolated, cut at mid-belly, embedded in optimal cutting temperature compound (Tissue-Tek)/20% sucrose, and frozen in liquid nitrogen for cryosections or fixed in 4% PFA and processed for paraffin sections. Cryosections were used for immunohistochemistry staining, and paraffin sections were used for hematoxylin–eosin, Alizarin red S (calcification), or Van Gieson's (fibrosis) staining. For postnatal fiber size analysis, lower hind legs were cryosectioned and total fiber number and fiber size were analyzed on transverse sections (10 μm) of the mid-belly soleus and EDL muscles, using NIH Image software. Fiber types were identified immunohistochemically with antibodies specific for MyHC subtypes (see next section) as described previously (). In brief, cryosections, single myofibers, or cultured cells were fixed in 2–4% PFA, quenched with glycine (100 mM glycine, 0.2% Triton X-100, and 0.1% sodium azide in PBS), and blocked in PBS containing 2% BSA, 5% goat serum, and 0.2% Triton X-100. Tissues or cells were then incubated with the primary antibodies diluted in the same blocking solution and finally with biotinylated secondary antibodies or fluorophore-conjugated secondary antibodies. The primary antibodies used were as follows: mouse monoclonal anti-Pax3 (IgG2a; a gift from C. Ordhal, University of California, San Francisco, San Francisco, CA; available from the Developmental Studies Hybridoma Bank), rabbit anti-Pax3 (Active Motif and Zymed Laboratories), Pax7 (Developmental Studies Hybridoma Bank), rat anti-laminin (Qbiogene), laminin (DakoCytomation), Desmin (DakoCytomation), CD34 (BD Biosciences), MyoD (5.8A; BD Biosciences), rabbit anti-MyoD (C-20; Santa Cruz Biotechnology, Inc.), Myf5 (Santa-Cruz Biotechnology, Inc.), MyHC (MF-20; Developmental Studies Hybridoma Bank), embryonic fast MyHC (F1.652; Developmental Studies Hybridoma Bank); MyHC IIb (BF-F3; Deutsche Sammlung von Mikroorganismen und Zellkulturen), MyHC I (A4.840) and IIa (A4.74; Developmental Studies Hybridoma Bank), anti–β-gal (Invitrogen), and chicken anti-Syndecan4 (gift of B. Olwin, University of Colorado, Boulder, CO; ). BrdU detection was performed using the BrdU in situ detection kit (BD Biosciences) following the manufacturer's protocol. The secondary antibodies used were Alexa488 anti-mouse IgG2a; Alexa568 anti-mouse IgG1 (Invitrogen); and Fluorescein-, Rhodamine-, or Cy5-conjugated antibodies (Chemicon). Nuclei were counter-stained with DAPI. adult diaphragms were processed for in situ hybridization as previously described (). A cDNA template was used for synthesis of sense and antisense digoxigenin-labeled riboprobes (). After in situ hybridization, sections were reacted with an antibody to laminin (DakoCytomation) and a Rhodamine-conjugated secondary layer (Chemicon). Errors quoted are SEM throughout. Data were analyzed using unpaired tests. Asterisks indicate significance at P < 0.05 and P < 0.01 throughout. Samples were visualized with a microscope (Axioplan2; Carl Zeiss MicroImaging, Inc.), and images were acquired using a camera (AxioCam; Carl Zeiss MicroImaging, Inc.) and the Axioview 3.1 software (Carl Zeiss MicroImaging, Inc.). Digital fluorescent images were captured at room temperature with a 20× (plan-apochromat, air, NA 0.75) or 63× (plan-apochromat, oil, NA 1.40) objective using the least possible exposure to minimize bleaching. The final displaying levels were subsequently adjusted similarly for all figures using Axioview or Photoshop (Adobe) software. Fig. and muscles at P0. Fig. S2 shows Pax7 and Pax3 cells in the limb muscle of a wild-type mouse at P1. Online supplemental material is available at .

The major role of p53 is to promote cell cycle arrest, programmed cell death, and cell senescence, and it is mutated in more than half of the primary human tumors (; ). p53 executes its function mainly by activating the transcription of target genes involved in these cellular events, in which the binding of p53 to a specific DNA sequence is necessary (; ). p53 can also act as a transcription repressor (). In some cases, the repression is independent of DNA binding (; ; ; ). p53 can be brought to the promoter regions of these genes through other DNA-binding proteins, such as the CAAT-binding factor (), where p53 may interfere with the assembly of the transcription–initiation complex () or recruit transcription repressors like histone deacetylase (; ). Hence, p53 can repress gene expression using complex and diverse mechanisms, most of which are not well understood. In addition, the biological significance of p53-mediated repression remains unclear (). mice did not reveal any developmental abnormalities (; ; ). embryos were more susceptible to teratological reagents (; ), yet, the molecular mechanisms through which p53 regulates cell differentiation and mouse development remain largely unknown. We have previously established that the nonreceptor tyrosine kinase c-Abl plays a positive role in murine osteoblast differentiation and bone development (). Given the functional relationship between c-Abl and p53, we decided to study the role of p53 in osteoblast differentiation and bone remodeling. Osteoblast is a cell of mesenchymal origin that is responsible for bone formation and can support osteoclast differentiation. Over the last 10 yr, a growing body of knowledge has emerged regarding the transcriptional control of osteoblast differentiation and function. Runx2 is the earliest determinant of osteoblast differentiation, and its expression defines a bipotential cell type called an osteochondroprogenitor (). Downstream of Runx2, other osteoblast-specific transcription factors have been identified. One is osterix, which has mainly been studied in a developmental context until this point (). The other is ATF4, which controls both osteoblast differentiation and function (; ). We report that p53 negatively regulates osteoblast differentiation and function by repressing the expression of osterix. mice, and elevated expression of p53 inhibited osterix expression and the promoter activity in a p53-binding–independent manner. Moreover, p53 deficiency conferred the osteoblasts with an increased ability to promote osteoclastogenesis, most likely through the up-regulation of macrophage-colony stimulating factor (M-CSF). This study adds to our current knowledge of osteoblast differentiation, osteoblast-supported osteoclastogenesis, bone remodeling, and the developmental role of the tumor suppressor p53. To determine whether p53 has a role in the skeleton, we analyzed bones from 3–4-mo-old wild-type and p53 mice. No evident abnormality was observed in the gross development of the skeleton. Histological analysis of hematoxylin-eosin–stained long bone sections indicated that the growth plates were not significantly altered either ( A). mice and their littermate controls revealed that mice showed a modest but consistent increase in bone mineral density ( B). mice had a 29% increase in the number of trabecular bones and a 40% increase in the volume of trabecular bones (, C–E). A slight increase was detected in the volume of cortical bones (45.7 ± 2.8% [bone vol/tissue vol] for +/+ vs. 52.4 ± 2.7% for −/− of femurs, = 8, P = 0.035). mice consistently showed a marked increase in the mineral apposition rate and bone formation rates at the periosteal surface (, F–I), as well as in the trabecular bones (, J–L), but not at the endosteal surface (not depicted). mice also showed a marked increase in the osteoblast surface and the number of osteoblasts per bone surface (). Together, these results established that p53 exerts a negative influence on osteoblast differentiation and/or function. mice, we first wanted to determine whether the bone marrow contains the same number of osteoprogenitor cells as in control mice. Total bone marrow cells were counted and directly plated in a 35-mm plate. and control mice ( A), suggesting that p53 deficiency did not significantly alter the differentiation potential of mesenchymal cells into osteoblasts. mice. osteoblasts expressed more ALP ( B), with the p53 heterozygous culture displaying an intermediate phenotype. Quantitative assays for ALP (enzyme activities normalized to total protein levels or the number of osteoblasts) confirmed the results obtained from histological staining (). During the second to third week in culture, osteoblasts start to express osteocalcin (), an osteoblast-specific gene that is a marker of fully differentiated osteoblasts and is involved in osteogenesis. osteoblasts expressed maximal osteocalcin from day 0 of culture, suggesting that p53 negatively regulates osteocalcin expression ( E). osteoblasts and retroviral expression of p53 in control cells inhibited their differentiation (). mice (unpublished data). These results indicate that the biological function of p53 in cells of the osteoblast lineage is to inhibit or slow down their differentiation. and control osteoblasts and found that in the absence of p53 osteoblasts doubled at a higher rate ( I). mice exhibit an increased number of osteoblasts in vivo. cells (). This may explain why p53 osteoblasts have a shorter cell cycle. cells the levels of p16 were up-regulated, whereas cyclin-dependent kinase 4 and cyclin D1 were down-regulated ( , J and K). This may reflect a compensation mechanism for the loss of p53. osteoblasts with a retrovirus and found that cell proliferation rates were down to those of control osteoblasts (not depicted), yet these osteoblasts still exhibited enhanced differentiation ( L), suggesting that enhanced differentiation caused by p53 deficiency is independent of cell proliferation. To determine the molecular basis of p53 action in osteoblast, we next asked whether it could influence expression or function of the known osteoblast-specific transcription factors such as Runx2, Osterix, and ATF4. and control osteoblasts at different stages of differentiation and used for RT-PCR assays. No dramatic up-regulation of Runx2 was observed during differentiation of mutant or control cells. and control osteoblasts (; and not depicted). osteoblasts showed a significant elevation of osterix at the basal level ( A). and control osteoblasts, expression of osterix peaked rapidly at day 2 and subsided to basal levels later on. osteoblasts displayed a much higher peak value than control cells. This up-regulation of osterix could be a cell response to bone morphogenetic proteins (BMPs) that are secreted by osteoblasts. The results suggest that the effect of p53 on osteoblast differentiation is downstream of Runx2 and upstream of osterix. Note that p53 levels started to increase at day 4 during differentiation, and the significance of this regulation warrants further investigation ( B). mice (). In situ hybridization using RNA probes generated from the osterix coding sequence confirmed the results ( F and not depicted). Overall, these results suggest that one function of p53 is to inhibit osterix expression and, hence, osteoblast differentiation. Two experiments further support that p53 plays a negative role in osterix expression. Primary mouse embryonic fibroblasts (MEFs), when stimulated with BMP2, start to express osteoblast-specific markers. MEFs compared with control cells, in response to a short-term treatment with BMP2 ( A). osteoblasts ( B). These results confirmed that p53 is a negative regulator of osterix. To test whether p53 represses the transcription of osterix, a 6.0-kb DNA fragment upstream of the transcription start site of osterix gene () was fused to the luciferase gene in pGL3 basic vector. Coexpression of p53 led to a dosage-dependent repression of promoter activity in C2C12 and MC3T3-E1 cells ( C and not depicted). p53R273H, a mutant with no DNA-binding ability, still significantly repressed promoter ( D), suggesting that DNA binding may not be required. Moreover, p53S15A, a mutation of a phosphorylation site important for transactivation of p53 target genes in response to DNA damage, could repress the promoter. On the other hand, p53 mutants such as p53Δ62-91 (the proline-rich domain deleted), p53M246I (point mutation in the DNA-binding domain), and p53L22Q/W23S (two point mutations that abolish p53 transactivation activity), which were reported to have dramatically reduced the ability to repress other p53 target genes (; ; ), failed to repress the promoter ( D). These results suggest that p53 might repress promoter activity with a common mechanism. The following observations suggest that p53 represses the promoter independent of DNA binding. First, no canonical p53-binding site was present in the 6-kb promoter sequence. Second, serial deletion experiments indicated that all fragments (6, 4, 2, 1.0, 0.5, 0.3, and 0.13 kb upstream of the start site) could still be repressed by p53 ( E). This 0.13-kb fragment contains mainly the TATA box–like sequence and is a minimal sequence that retains some activity in a reporter assay. Third, chromatin immunoprecipitation assays demonstrated that p53 was not directly associated with the 2-kb sequence of the osterix promoter even though its binding was detectable at the p21 promoter (unpublished data). These results suggest that p53 repressed osterix transcription independent of its binding to the upstream activation sequence. We also found that p300, among several transcription factors and coactivators, could activate the promoter, and that this activation was repressed by p53 as well ( F). Because p300 is expressed in the cells at a limiting concentration, it is possible that p53 subjugates this coactivator to repress gene transcription, as proposed previously (; ). To date, p53 has been reported to inhibit the transcription of many genes like . Yet, the molecular mechanisms behind the repression are not well understood. This prevents us from establishing the exact role of p53 in the repression of transcription. osteoblasts. We first tested whether osterix overexpression could lead to enhanced differentiation. and control osteoblasts were infected with retroviruses expressing osterix, selected against puromycin for 3 d, and then used for ALP or osteocalcin assays (, A–E). We observed a retroviral dose-dependent increase in osterix levels (), which was likely caused by multiple infections in a single cell. An up-regulation in ALP activities and osteocalcin was also seen (). osteoblasts, they expressed similar levels of ALP and osteocalcin as cells (). osteoblasts might be at least partially mediated by the increased expression of osterix. osteoblasts. osteoblasts and osteoblast differentiation markers were analyzed (, F–J). Different combinations of siRNA species reduced osterix to different levels (), leading to decreased expression of ALP and osteocalcin (, F–H and J). The results suggested that osterix was necessary for osteoblast differentiation and that the effect of p53 on osteoblast differentiation might be at least partially mediated by osterix. Increased bone mass could be attributed to reduced bone resorption, in addition to increased bone formation. Further analysis revealed an unanticipated result. mice showed a onefold increase in the bone resorption surface and the number of osteoclasts (). mice also showed an increase in the excretion of urine deoxypyridinoline cross-links ( C). mice have increased bone resorption in addition to increased bone formation. mice only showed modest osteosclerotic phenotypes despite bone formation rates being nearly double that of control mice. and control bone marrow monocytes (BMMs) in the presence of M-CSF and receptor activator of NFκB ligand (RANKL), which are necessary and sufficient for osteoclast proliferation and differentiation. and control BMMs () nor did we observe a significant difference in the bone resorption activities of the and control osteoclast cultures based on pit formation on dentine slices (). These results suggest that p53 has no cell-autonomous effect on the differentiation and the function of osteoclasts. osteoblasts led to increased osteoclastogenesis and bone resorption in vivo. It has been well established that osteoblasts control osteoclast differentiation by synthesizing M-CSF, RANKL, and osteoprotegerin (OPG, a RANKL decoy receptor; ; ). To prove this, we cocultured primary calvarial osteoblasts ( or control) with BMMs ( or control) in the presence of 10 M dihydroxyvitamin D3, and stained for tartrate-resistant acid phosphatase (TRAP)–positive cells (). Primary osteoblasts were plated at a high density so that the plates would become confluent overnight. This was to eliminate any possible effects of growth disparity of osteoblasts. These TRAP-positive osteoclasts were much smaller, compared with those formed in the presence of M-CSF and RANKL ( D), with a small multinucleated portion, similar to what has been observed in previous studies (). osteoblasts exhibited a marked increase in the number of TRAP-positive osteoclasts regardless of the genotype of BMMs, compared with osteoblasts. When only cells with ≥3 nuclei were counted, a 2.5-fold increase was observed for monocytes (both and ) cultured on osteoblasts, compared with those cultured on control osteoblasts ( J). and monocytes behaved similarly when cultured on osteoblasts or wild-type osteoblasts, confirming that p53 did not have a cell-autonomous effect on osteoclastogenesis (). mice might be attributable to the enhanced activities of osteoblasts. osteoblasts, but not of RANKL or OPG ( K and not depicted). M-CSF controls both the proliferation and the differentiation of osteoclasts (; ). osteoblasts may be attributable to the elevated expression of osterix. Mice deficient for c-Abl show signs of osteoporosis, and the mutant osteoblasts show defects in differentiation and survival against oxidative stress (, ). Because p53 is a c-Abl–interacting protein and genetically interacts with c-Abl during cell proliferation and apoptosis (; ), we studied their relationship in bone development. Unfortunately, compound homozygous mice for c-Abl and p53 were very difficult to obtain, probably owing to the embryonic and/or postnatal lethality of these mice. Nevertheless, we isolated calvarial osteoblasts from four 20-d embryos of double homozygous mice and control littermates, and compared their differentiation potential. osteoblasts in ALP expression and in mineral deposition (, A–C). osteoblasts ( D). These data suggest that p53 functions downstream of c-Abl in the process of osteoblast differentiation. We provide genetic evidence that p53 plays a negative role in postnatal bone development, with a cell-autonomous effect on osteoblastogenesis. osteoblasts. Moreover, p53 deficiency confers the osteoblasts a greater osteoclastogenic capacity, without directly affecting osteoclast differentiation or resorption. mice. Thus, the osteosclerotic phenotype is a net result of the direct effect of p53 on osteoblast action combined with an osteoblast-mediated effect on osteoclasts. This osteoblast-supported osteoclastogenesis might explain why most of the osteosclerosis models associated with enhanced osteoblast function do not exhibit a huge increase in bone mass (, ; ; ). The conclusion that p53 plays a negative role in osteoblastogenesis is supported by the findings that p53 might directly regulate the expression of osteocalcin (; ). However, our results are not in agreement with a recent study stating that p53 did not affect osteoblast differentiation and mouse bone formation, although they did show that p53 deficiency rescued the bone loss induced by mechanical unloading (). The discrepancy could be caused by the genetic background (C57BL/6 in our studies, but not mentioned by ) or the age of the mice used (4 mo in our studies vs. 2 mo in the Sakai study), and warrants further investigation. mice. We also provided genetic evidence to support that p53 functionally acts downstream of c-Abl in postnatal bone development. Our findings indicate that p53 plays a role in osteoblast differentiation without directly affecting the differentiation of osteoclasts in a cell-autonomous manner. The significance of the link between p53 and bone development is underscored by the recent findings that p53 cooperates with TGF β–BMP pathways to positively regulate early development of (; ), which is in contrast to the negative role for p53 in bone homeostasis. One explanation for the opposite roles of p53 in relationship to the TGF–BMP pathways could be the timing of these two events. Mesoderm differentiation occurs at an early stage when p53 levels are high, whereas osteoblast differentiation occurs at a much later stage when p53 levels start to decline (). The inhibition of osterix by p53 may provide a mechanism to block bone development in early embryos. Several lines of experiments suggest that p53 inhibits osteoblast differentiation as a result of repressing the expression of the lineage-specific transcription factor osterix. First, the in vivo and in vitro data indicate that osterix, but not other tested transcription factors, was elevated in the absence of p53. Second, when p53 was up-regulated, osterix expression was repressed and osteoblast differentiation was impeded. Third, BMP2-induced osterix expression was enhanced in the absence of p53. osteoblast differentiation could not be effectively repressed by c-Abl deficiency, consistent with sustained expression of osterix under this condition. Moreover, p53 was found to inhibit the promoter activity of osterix, whereas some mutant forms of p53 failed to do so. Finally, we found that the osterix levels are an important determinant in osteoblast differentiation, as the overexpression of osterix led to enhanced differentiation and the knocking down of osterix led to reduced differentiation in p53-deficient osteoblasts. Our findings might provide the first example in which p53-mediated gene repression has a physiological impact on postnatal development of the mouse. Our results suggest that p53 might do so by inhibiting osterix promoter activity by the minimal promoter independent of its binding to the upstream activation sequence. There is an increasing number of genes that are controlled by the core promoter sequence, including the TATA box (minimal promoter), and it is also becoming evident that general transcription factors, such as the thyroxine-binding protein (TBP) and TATA-binding protein-associated factors, which are associated to the core promoter sequences, can selectively regulate the transcription of certain genes (; ). Moreover, p53 also forms a complex with p300/cyclic AMP response element–binding protein, which interacts with the basal transcription machinery (). Our results suggest that p53 might repress expression by repressing p300. Still, exactly how p53 regulates transcription of warrants further investigation. Our findings indicate that p53 has a cell-autonomous role in osteoblast differentiation and proliferation. It is also reported that Li-Fraumeni syndrome patients develop osteosarcoma as a component tumor and that p53 is mutated in 24 to 42% of osteosarcoma (; ). Osteosarcoma is usually developed from osteoblasts and is related to periods in life with rapid bone growth. These observations suggest that p53 plays an important function in bone growth. More interestingly, osterix has been implicated in osteosarcoma development, as osterix is down-regulated in both human and mouse osteosarcoma cell lines and transfection of osterix into osteosarcoma cells inhibits their growth (). It will be interesting to determine whether down-regulation of osterix is related to the p53 status in osteosarcoma cells. In many tumor cell lines, p53 is mutated but its expression is greatly up-regulated. Depending on the nature of the mutations, up-regulated p53 may repress osterix expression. For example, a hot-spot mutation, R273H, can still repress osterix promoter activity. Osterix is an Sp1-like transcription factor that has been studied in the early development of mouse. It is essential for osteoblast maturation. Osterix has three Zinc-finger domains and is able to bind to consensus Sp1-binding sites (). We show that p53 deficiency results in the elevation of osterix and enhanced osteoblastogenesis. Cell culture studies confirmed that osterix is sufficient and necessary for osteoblast differentiation. Our results also suggest that osterix regulates the expression of M-CSF to control osteoclastogenesis. M-CSF actually contains in its promoter an Sp1-binding site, which mediates the effects of estrogen on M-CSF expression and osteoclastogenesis (). It is likely that osterix directly controls M-CSF expression. Recent studies revealed that transcription factors involved in osteoblast differentiation also regulate osteoclast differentiation through controlling RANKL and OPG expression. In differentiated osteoblasts, Wnt pathway and TCF1/4 were found to regulate the expression of OPG and osteoclast differentiation (). ATF4, a cyclic AMP response element–related transcription factor that is essential for osteoblast differentiation and function, also regulates the expression of RANKL and, subsequently, osteoclast differentiation (). In summary, p53-deficient mice show an osteosclerotic phenotype, which is a net result of increased bone formation and increased bone resorption. p53-deficient osteoblasts exhibit accelerated proliferation, enhanced differentiation, and increased osteoclastogenic activities. Enhanced differentiation can be mediated by an elevation of osterix expression, and improved osteoclastogenesis can be mediated by increased expression of M-CSF, which is also induced by elevated osterix. These findings suggest that p53 might control bone remodeling by modulating expression of osteoblast-specific transcription factor osterix. (The Jackson Laboratory) and mice were crossed to C57BL/6 six times. Primary osteoblasts were isolated from newborn pups or 20-d-old embryos, as previously described (). The cells were amplified and frozen for future use. Bone marrow stromal cultures were extracted from 3–4-mo-old mice. Both calvarial osteoblasts and stromal cells were cultured in α-MEM supplemented with 15% FCS (HyClone). For osteoblast differentiation, cells were cultured in differentiation medium (α-MEM medium containing 50 μg/ml ascorbic acid and 10 mM β–glycerol-phosphate) and were re-fed every 3 d. MEFs were isolated following a previously described standard protocol (). For osteoclast differentiation, the bone marrow of 3–4-mo-old mice was flushed and the monocyte fraction was isolated by centrifugation on a Ficoll plus lymphocyte separation medium gradient (ICN Biomedicals), washed, seeded at 7.5 × 10 cells/well of 96-well plates, and cultured for 7 d in differentiation medium (α-MEM containing 10% FCS [Invitrogen], 30 ng/ml M-CSF [R&D Systems], and 50 ng/ml soluble recombinant RANKL [Sigma-Aldrich]). TRAP staining was performed using an acid phosphatase kit (Sigma-Aldrich). Osteoclast resorption function was assessed by a pit formation assay on dentine slices (OsteoSite). Monocytes were cultured for 2–3 d in the presence of 30 ng/ml M-CSF and 50 ng/ml of soluble recombinant RANKL, counted, and plated onto dentine slices that were preincubated with serum for 2 h. After 7 d, the dentine slices were sonicated in 0.5 M ammonium hydroxide, stained with either Gill's Hematoxylin or Toluidine blue for 2 min, washed with water, and photographed under a light microscope (Eclipse TE 200; Nikon). The resorbed areas were measured using a densitometry system and were normalized to the number of osteoclasts in the well. Urine levels of deoxypyridinoline cross-links were determined using commercial kits (Quidel Corporation) and were normalized to urine creatinine following the manufacturer's protocols. For coculture studies, primary calvarial osteoblasts were plated at 5 × 10 per well in 24-well plates. Upon confluency, osteoclast progenitors (BMMs; 5 × 10 per 24 wells) were then plated and the cultures were complemented with 10 M 1, 25 (OH) vitamin D. Multinucleated cells appeared after 6–10 d and were counted. Cells were lysed in TNEN buffer (50 mM Tris, 150 mM NaCl, 5 mM EDTA, 0.5% NP-40, and 0.1% Triton X-100) supplemented with 1 mM NaF, NaVO, 1 mM PMSF, and 1 μg/ml of aprotonin, leupeptin, and pepstatin. Protein concentration was determined using an assay (Bio-Rad Laboratories). Proteins were resolved by SDS-PAGE and transferred to polyvinylidene difluoride membranes (Millipore). Anti-osterix was generated by injecting rabbits with a synthesized peptide (Biogenes GMBH). Anti-p53 and -p19 antibodies were purchased from Oncogene Research Products, anti-p21 antibodies were purchased from BD Biosciences, anti-p16 and -p27 antibodies were obtained from Santa Cruz Biotechnology, Inc., and anti-actin antibodies were obtained from Sigma-Aldrich. The osterix promoter (fragments ranging from 0.13 to 6.0 kb from the start of transcription) was cloned into pGL3 (luciferase basic vector; Promega). Increasing amounts of p53 expression constructs, the promoter plasmid (pGL3-Osx-Luc), and renilla plasmid (internal control) were cotransfected into C2C12 or MC3T3-E1 cells. Cells were harvested 24 h later, washed with PBS, and lysed with reporter lysis buffer (Promega). TBP, p300, and p53 are expressed under the control of cytomegalovirus promoter (in pcDNA3, pXJ, and pCMV, respectively). p53 mutants (gifts from K. Löhr and M. Dobbelstein, University of Marburg, Marburg, Germany) were described previously (). The luciferase activities were measured following the manufacturer's procedures and were normalized against the renilla activity. All transient expressions in this assay were performed in triplicate. For siRNA knocking down experiments, the following siRNA oligomers (Dharmacon) were used: siGENOME on-target duplex 1, sense: GCCAUACGCUGACCUUUCAUU; siGENOME on-target duplex 2, sense: GGGCACAGCUCGUCUGACUUU; siGENOME on-target duplex 3, sense: CAACACACCUACUCCUUGGUU; siGENOME on-target duplex 4, sense: GCAGGCAUCUCACCAGGUCUU. The control siRNA oligomer used was a nontargeting negative siRNA control pool that was transiently transfected into primary osteoblasts following the manufacturer's procedures (Dharmacon). After 3 d, cells were harvested for ALP, osteocalcin, and Western blot analysis. To express osterix in osteoblasts, a retroviral vector was constructed with osterix ORF cloned into the pMSCVpuro vector (BioMed Diagnostics). The viruses were produced by transfecting platE cells following a standard protocol (). Osterix retroviral supernatant (different dilutions) was then used to infect primary cells, followed by 3-d selection against 5 μg/ml of puromycin before being harvested for Western blot, ALP, and RT-PCR assays. Based on the level of GFP expressed in the virus, the minimum amount of retroviral supernatant that can infect ≥80% of the cells was set at 100%. To express osterix at different levels, different dilutions (50, 100, 150, and 200%) of retroviral supernatant were used. Total mRNA was isolated from osteoblasts or MEF cells growing on 60- or 100-mm dishes using TRIzol reagent (GIBCO BRL) and used for Northern blot analysis as described (). and wild-type littermates were isolated and homogenized in TRIzol. The extracts were frozen at −80°C for 1 d and thawed for RNA extraction. Total RNA was subjected to DNase treatment (Ambion) and quantitated. 5 μg of total mRNA was reverse transcribed into complementary DNA (cDNA) using AMV (Roche) reverse transcriptase. The total reaction was used in the PCR to assay for the presence of osterix, Runx2, or actin with the following primers: osterix (197 bp): forward, 5′-TGAGGAAGAAGCCCATTCAC-3′; reverse, 5′-ACTTCTTCTCCCGGGTGTG-3′. Runx2 (113 bp): forward, 5′-TGGCAGCACGCTATTAAATC-3′; reverse, 5′-TCTGCCGCTAGAATTCAAAA-3′. β-Actin (104 bp): forward, 5′-AGATGTGGATCAGCAAGCAG-3′; reverse, 5′-GCGCAAGTTAGGTTTTGTCA-3′. PCR was performed for 30 cycles of denaturation (94°C for 30 s), annealing (57°C for 30 s), extension (72°C for 1 min), and one cycle of final extension (72°C for 10 min), which was just enough to detect the PCR products of osterix and Runx2. and control cells than RT-PCR. and wild-type mice were used in the presence of specifically designed osterix primers in a 20-μl reaction mix with TaqMan MGB probe (FAM dye-labeled). The osterix expression levels were normalized by GADPH as an internal control in real-time PCR analysis according to the manufacturer's instructions (Applied Biosystems). RT-PCR (negative images of gels) or Western blot results were scanned with a Molecular Dynamics scanning densitometer. The relative levels of mRNA or protein of interest were then determined by measuring the intensity of the corresponding bands. mice and their control littermates and were normalized to the constitutive expression of the housekeeping genes. The assays were performed as previously described (). Primary osteoblasts were plated at a high density so that the plates would become confluent overnight. This was to eliminate any possible effects of disparity in osteoblast growth rates. The relative ALP activity is defined as millimoles of -nitrophenol phosphate hydrolyzed per minute per milligram of total protein (units). α-S-UTP–labeled single-stranded RNA probes were prepared using an RNA labeling kit (DakoCytomation) following the manufacturer's procedure with slight modifications. A 200-bp mouse osterix cDNA fragment was used to generate antisense and sense probes. All mice were labeled with 15 mg/kg of calcein subcutaneously (Sigma-Aldrich) twice in an interval of 9 d before death. The right femur of each animal was dissected free of soft tissue and used for measurement of femoral bone density by a dual energy x-ray absorptiometer. The right tibia was dissected and cut into three equal parts. The right proximal tibia and tibial shaft were fixed in 70% ethanol solution for 2 d and immersed in Villanueva Osteochrome Bone Stain (Polysciences, Inc.) for 5 d. The specimens were dehydrated by sequential changes of ascending concentrations of ethanol (70, 95, and 100%) and xylene and embedded in methyl methacrylate (Eastman Organic Chemicals). Frontal sections of the proximal tibia were cut at 5 μm using a microtome (model RM2155; Leica) and cross sections of the tibial shaft proximal to the tibiofibular junction were cut at 40 μm using a diamond wire Histo-Saw machine (Delaware Diamond Knives, Inc.). All sections were coverslipped with Eukitt (Calibrated Instruments, Inc.) for static and dynamic histomorphometric analysis. Right femoral bone mineral content (BMC) and bone mineral density (BMD) were determined using a dual-energy X-ray absorptiometer (model QDR-1000W; Hologic). The machine was adapted for an ultra-high-resolution mode with line spacing of 0.0254 cm, resolution of 0.0127 cm, and a collimator of 0.9 cm diam. The bones were placed in a Petri dish. To simulate soft tissue density surrounding the bones, tap water was poured around the bones to achieve a depth of 1 cm. Results are given for BMC and for area; area BMD is calculated as BMC/area. In addition to results for total femur, the distal and mid-region of the femur were analyzed as subregions. Coefficients of variation for these measurements in our laboratory are 0.8, 1.0, and 0.6%, respectively. Staining of cell culture plates for ALP or mineralization shown in , , and was photographed using a digital camera (model Coolpix 995; Nikon). Micrographs shown in were visualized on a microscope (Eclipse TE200; Nikon) with Plan Fluor objectives (4×, 0.13 NA; 10×, 0.25 NA; 20×, 0.40 NA; 40×, 0.55 NA) or a dissecting microscope (model SMZ645; Nikon), which were connected to the previously mentioned digital camera. Micrographs shown in and were visualized and captured as described in the previous section. The images were acquired and processed using Photoshop 6.01 (Adobe). For Western blots, Northern blots, and DNA gels, the images were acquired from films or Kodak papers with a scanner (Canoscan N1240U; Canon) and processed using Photoshop 6.01. Each experiment was repeated with three or more mutant and control mice. Statistical analysis was performed using an unpaired test (STATISTICA software; StatSoft, Inc.). P values were provided for all in vivo results. Significant association was defined when P < 0.05 compared with control.

Ca is a versatile second messenger that mediates a wide range of cellular processes, including cell division and apoptosis (). Under physiological conditions, cytoplasmic Ca is maintained at a low level, and it is the elevation of cytoplasmic Ca that generates Ca signals. Elevated Ca transmits information by activating Ca-sensitive effectors, including phosphatases and kinases. The Ca elevation involved in signal transduction is often in the form of repetitive Ca spikes or oscillations (). The information-processing capability of Ca signaling is enhanced by modulation of the frequency, amplitude, and spatial properties of Ca elevations. This in part explains how a simple messenger such as Ca can regulate diverse cellular processes. In T cells, Ca signals mediate a variety of responses to T cell receptor (TCR) activation, including cell proliferation and apoptosis (; for reviews see ; , ; ). As in all nonexcitable cells, the T cell Ca response begins with the release of Ca from the ER through inositol 1,4,5-trisphosphate (InsP)–dependent Ca channels (InsP receptors). The resulting cytoplasmic Ca elevation is amplified by Ca entry through Ca-release–activated Ca channels on the plasma membrane, producing either a transient Ca elevation or Ca oscillations (,; ; for review see ). The Ca signal is then transduced through Ca/calmodulin–mediated activation of the protein phosphatase calcineurin, which dephosphorylates and thereby activates the nuclear factor of activated T cells (NFAT; for review see ; ). NFAT is a transcription factor that activates the interleukin-2 promoter, increasing cell proliferation. Activation of calcineurin, and hence NFAT, is sustained more efficiently by Ca oscillations than by a transient elevation of Ca, whereas other Ca responses (e.g., nuclear factor kB and c-Jun NH-terminal kinase activation) are preferentially activated by transient Ca elevation (, ). The importance of Ca oscillations in T cell signaling is increasingly recognized, including evidence that Ca oscillations regulate thymocyte motility during positive selection in the thymus (). We recently reported that the antiapoptotic protein Bcl-2 () interacts with InsP receptors on the ER and inhibits InsP-mediated Ca efflux (). As a consequence, Bcl-2 dampens the cytoplasmic Ca elevation induced by an antibody to the CD3 component of the TCR complex. These findings are intriguing in view of the known role of Ca in signaling apoptosis (for reviews see ; ; ), but an inhibitory effect of Bcl-2 on InsP-mediated Ca elevation would seem incompatible with the wide range of physiological processes governed by InsP-mediated Ca signals. Would not Bcl-2 interfere with Ca signals that regulate physiological processes required for cell function and survival? A possible clue to this dilemma was provided by earlier work indicating that Ca responses after TCR activation vary according to the strength of TCR activation (). Typically, strong signals induced by a high concentration of anti-CD3 antibody trigger a single transient elevation of cytoplasmic Ca, whereas weaker signals induced by a low concentration of anti-CD3 induce Ca oscillations (). Our previous experiments demonstrating an inhibitory effect of Bcl-2 on anti-CD3–induced Ca elevation used a high concentration of anti-CD3 antibody that induced a transient Ca elevation rather than Ca oscillations. Therefore, in the present work, we investigated the effect of Bcl-2 on Ca signals induced over a broad range of anti-CD3 concentrations. This led to the discovery that Bcl-2 differentially regulates Ca signals according to the strength of TCR activation. Thus, Bcl-2 inhibited the transient Ca elevation induced by a high concentration of anti-CD3 antibody, without interfering with Ca oscillations induced by a low concentration of anti-CD3 antibody. Accordingly, Bcl-2 inhibited Ca-mediated apoptosis after strong TCR activation but did not inhibit NFAT activation after weak TCR activation. Therefore, by selectively regulating Ca signals according to the strength of TCR activation, Bcl-2 discriminates between proapoptotic and prosurvival Ca signals. The WEHI7.2 T cell line corresponds to an immature double-positive stage of T cell differentiation, as WEHI7.2 cells express both CD4 and -8 antigens and are sensitive to glucocorticosteroid-induced apoptosis. Consistent with this stage of development, Bcl-2 is virtually undetectable in WEHI7.2 cells. In earlier work, Bcl-2–positive and –negative clones were derived by stably transfecting WEHI7.2 cells with an expression vector encoding full-length human Bcl-2 or an empty vector, respectively. The full characterization of the clones used in this work has been reported previously (). All findings reported here are based on comparison of three Bcl-2–positive and three Bcl-2–negative clones. Findings were consistent among the different clones; therefore, data from individual clones have been pooled, unless otherwise noted. Throughout this paper, cytoplasmic Ca was measured at a single-cell level by digital imaging. An initial series of Ca measurements was performed to determine the dose–response relationship between anti-CD3 concentration and cytoplasmic Ca response patterns in a Bcl-2–negative clone ( A). In this experiment, a transient elevation of Ca was defined as only one or two Ca elevations reaching at least twice the basal level of Ca, whereas sustained Ca oscillations were defined as three or more Ca spikes at least twice the basal level of Ca and separated by at least a 30-s interval. The percentage of cells responding with a transient Ca elevation was maximal at 20 μg/ml anti-CD3 antibody and declined progressively with increasing antibody dilution ( A). Conversely, the percentage of cells developing Ca oscillations increased progressively as anti-CD3 antibody concentration was reduced. Thus, there is a reciprocal dose–response relationship for transient elevations of Ca versus Ca oscillations after TCR activation. Based on these initial findings, subsequent studies used 20 μg/ml as representative of a high concentration of anti-CD3 antibody and 2, 0.75, and 0.33 μg/ml as representative of low concentrations of anti-CD3 antibody. Representative Ca traces comparing Bcl-2–negative and –positive cells at a high anti-CD3 concentration (20 μg/ml) are shown in . Bcl-2–negative cells responded to high anti-CD3 with an almost synchronous elevation of cytoplasmic Ca as shown in A, where multiple single-cell Ca traces collected in a single experiment (i.e., a single microscope coverslip) are plotted together. In contrast, only a small proportion of Bcl-2–positive cells responded with a detectable Ca elevation ( E). The inhibitory effect of Bcl-2 on anti-CD3–induced Ca elevation in these experiments is also illustrated by averaging the Ca response of multiple cells (). Characteristics of Ca responses at a single-cell level are illustrated by the Ca traces in (C, D, G, and H). Though the number of responding cells and the amplitude of Ca elevations differed markedly between Bcl-2–negative and –positive cells, the shape of the Ca elevation was similar. In a small percentage of cells, secondary Ca elevations were observed (e.g., H), but as noted in B, sustained oscillations were infrequent. Representative Ca traces comparing Bcl-2–negative and –positive cells at lower concentrations of anti-CD3 antibody (2, 0.75, and 0.33 μg/ml) are shown in . In contrast to the broad elevation of Ca triggered at 20 μg/ml anti-CD3, the Ca responses at low concentrations of anti-CD3 antibody were typically in the form of narrow spikes. Also, in contrast to the inhibitory effect of Bcl-2 on Ca responses at high anti-CD3, Bcl-2 did not inhibit the Ca spikes evoked by low concentrations of anti-CD3. This is illustrated by the representative traces shown in . In show repetitive Ca spikes, or oscillations, detected in both Bcl-2–negative and –positive cells over the range of anti-CD3 concentrations tested, whereas A and E show cells where only one or two Ca spikes were detected. Note that the Ca oscillations induced by low anti-CD3 were irregular in terms of both amplitude and frequency. This is characteristic of Ca oscillations in T cells, as reported previously, and is in contrast to the more uniform pattern of Ca oscillations observed in nonlymphoid cells (for reviews see ; ). Also, spontaneous Ca oscillations were not detected in control experiments where anti-CD3 was not added to cells. The responses of cells to either high (20 μg/ml) or low (2, 0.75, or 0.33 μg/ml) anti-CD3 in a large number of experiments are summarized in (B and C). These data indicate that oscillations are much more frequent at the low than at the high concentration of anti-CD3 ( B). Moreover, these data confirm that Bcl-2 markedly inhibits Ca responses to strong TCR activation (20 μg/ml anti-CD3 antibody), based on a significant (P ≤ 0.01) reduction in both the percentage of cells that respond ( B) and the amplitude of Ca elevations in those cells that do respond ( C). In contrast, Bcl-2 did not inhibit Ca responses to weak TCR activation (2, 0.75, or 0.33 μg/ml anti-CD3 antibody), based on the percentage of cells that respond ( B) and the mean amplitude of Ca spikes ( C). Interestingly, at low anti-CD3 in Bcl-2–positive cells, there was a small but insignificant (P > 0.10) reduction in the percentage of cells developing a transient Ca elevation (i.e., one or two Ca spikes; B, left) and a small but insignificant (P > 0.10) increase in the percentage of cells developing sustained Ca oscillations (i.e., three or more Ca spikes; B, right). Moreover, the mean amplitude of Ca spikes at the lowest anti-CD3 concentration (.33 μg/ml) was higher in Bcl-2–positive than in Bcl-2–negative cells ( C), although this difference was not statistically significant (P > 0.10). As noted in D, a major difference between the Ca elevations induced at high versus low anti-CD3 antibody concentrations was in the duration (i.e., width) of the individual Ca peaks. This is illustrated by the representative Ca traces in and and is also documented quantitatively in D. The mean peak width was 4 min at 20 μg/ml anti-CD3 antibody but <1 min at 2 μg/ml anti-CD3 antibody. Moreover, Bcl-2 did not alter the width of individual Ca elevations (either transient induced by high anti-CD3 or transient and oscillatory at low anti-CD3; D). Detailed comparisons of Ca oscillations induced by low concentrations of anti-CD3 antibody in Bcl-2–negative and –positive cells are summarized in . To quantitatively compare the oscillatory frequency in Bcl-2–positive and –negative cells, Ca traces from numerous experiments were separated into successive 5-min time periods and the number of Ca spikes during each period was logged, a method that has been described previously (; ). Overall, the frequency of Ca oscillations induced by 2 μg/ml anti-CD3 antibody appeared higher in Bcl-2–positive than in Bcl-2–negative cells, although differences reached statistical significance only during the 15-min time interval (P = 0.01) and borderline significance during the 10-min time interval (P = 0.057; A). The frequency of oscillations also appeared to be higher in Bcl-2–positive cells at 0.75 μg/ml anti-CD3 antibody, but differences were not statistically significant at any of the time intervals ( B). To analyze oscillatory frequency by a different method, the time interval between Ca spikes was measured in multiple Ca traces at 2 μg/ml anti-CD3 antibody and, based on these data, mean and mode frequencies were calculated. The mean frequency of Ca oscillations in Bcl-2–negative cells was 5.2 ± 0.6 mHz, whereas the mode frequency was 7.3 ± 0.55 mHz. Mean frequency was lower than mode frequency because of the presence of low-frequency spiking detected in a small proportion of the Ca traces. The mean frequency was not significantly different in Bcl-2–negative and –positive cells ( C), but the mode frequency was significantly higher in Bcl-2–positive cells ( D). Thus, consistent with the analysis in A, this analysis suggests an increased frequency of Ca oscillations in Bcl-2–positive compared to Bcl-2–negative cells. The total duration of Ca oscillatory runs (i.e., time duration from initial to final Ca spikes) appeared longer in Bcl-2–positive than in Bcl-2–negative cells, although this difference was of borderline significance (P = 0.05; E). WEHI7.2 cells adhered loosely to coverslips, limiting the rate at which anti-CD3 antibody could be perfused onto cells. Therefore, to estimate latency period, the initial Ca response to 2 μg/ml anti-CD3 antibody was recorded in cell suspensions fluorometrically ( F). Based on these data, the latency period was on the order of 2 min and was the same in Bcl-2–negative and –positive cells. Although the preceding analyses suggest that Bcl-2 may slightly increase the frequency and duration of Ca oscillations, the major conclusion from these data is that Bcl-2 does not inhibit Ca oscillations induced by low concentrations of anti-CD3 antibody. Consistent with this finding, Bcl-2 did not inhibit NFAT activation ( G). The mean amplitudes of Ca elevations induced by high and low concentrations of anti-CD3 antibody were compared in C. Although the mean peak amplitude of the Ca elevation induced by 20 μg/ml anti-CD3 was similar to the mean peak amplitude of Ca spikes induced by 2 μg/ml anti-CD3, the net effect of amplitude and width of Ca peak is a much more substantial Ca elevation after high anti-CD3 than after low anti-CD3. Furthermore, analysis of the peak Ca elevation induced by 20 μg/ml anti-CD3 on an individual cell basis illustrates a broad range of Ca elevations detected in Bcl-2–negative cells (). 27% of the Ca spikes in Bcl-2–negative cells were >200 nM, an arbitrarily chosen threshold level, whereas only 7% of Ca spikes in Bcl-2–positive cells were higher than this level after treatment with 20 μg/ml anti-CD3 antibody. Therefore, Bcl-2 not only reduces the percentage of cells that elevate Ca in response to high anti-CD3 ( B) but also dampens Ca elevations in the cells that do respond by preventing very high peak Ca elevations. In contrast to the wide distribution of Ca elevations observed at 20 μg/ml anti-CD3 antibody in Bcl-2–negative cells, only 6 and 1% of Ca spikes were >200 nM at 2 and 0.75 μg/ml anti-CD3 antibody in Bcl-2–negative cells (). Even fewer cells elevated their Ca to >200 nM at these low concentrations of anti-CD3 antibody in Bcl-2–positive cells ( C). Thus, although Bcl-2 did not reduce the percentage of cells that responded to low concentrations of anti-CD3 antibody by developing sustained Ca oscillations ( B) and did not reduce the mean amplitude of these Ca elevations ( C), it does appear to have set a threshold level above which the Ca does not elevate even in response to weak TCR activation. To investigate the contribution of high Ca elevations to anti-CD3–induced apoptosis, cells were treated with 20 μg/ml anti-CD3 antibody and sorted by flow cytometry into two different populations based on relative levels of cytoplasmic Ca ( A). The cells were then placed in culture, and the percentage of apoptotic cells was measured 24 h later. A significantly higher percentage of cells in the high Ca population underwent apoptosis, compared to cells in the low Ca population ( B). Conversely, reducing extracellular Ca concentration, a condition that partially prevents Ca elevation after treatment with high anti-CD3 ( A), inhibited apoptosis ( B). Thus, the induction of apoptosis in Bcl-2–negative cells is Ca dependent, and the percentage of cells undergoing apoptosis is proportional to the peak amplitude of Ca elevation after treatment with a high concentration of anti-CD3 antibody. Consistent with these findings, Bcl-2 inhibited apoptosis induction by high anti-CD3 (), in accordance with the dampening effect of Bcl-2 on anti-CD3–induced Ca elevation described in preceding experiments (). Treatment with low anti-CD3 did not induce apoptosis ( C), consistent with the evidence that high Ca elevation (≥200 nM) was much less common after treatment with low anti-CD3 than it was after treatment with high anti-CD3 (). Interestingly, the percentage of apoptotic cells was lower after treatment with low anti-CD3 antibody than it was in untreated cells ( C). This suggests that treatment with low anti-CD3 may have a prosurvival action, in contrast to the proapoptotic action of high anti-CD3. To address the question of how Bcl-2 differentially regulates Ca signals induced by high versus low concentrations of anti-CD3 antibody, we used small interfering RNA (siRNA) to reduce levels of all three InsP receptor subtypes in WEHI7.2 cells (). The siRNA olignucleotide pools were introduced into cells by electroporation with a transfection efficiency of 78 ± 2% (mean ± SEM). The siRNA-mediated reduction in InsP receptor levels was documented by Western blotting ( A). Substantial reduction in all three InsP receptor subtypes was reproducibly achieved ( B). This reduction in InsP receptors significantly inhibited the Ca elevation induced by 20 μg/ml anti-CD3 antibody (). But the siRNA-mediated reduction in InsP receptor levels did not inhibit Ca responses to 2 μg/ml anti-CD3 antibody. This is documented by representative Ca traces (), by analysis of the percentage of responding cells ( G), and by amplitude analysis ( H). Thus, consistent with the inhibitory effect of Bcl-2 on Ca responses to strong but not to weak TCR activation, the former appears to be more dependent on InsP receptor expression levels than the latter. The principal finding in this study is that Bcl-2 differentially regulates Ca signaling in T cells according to the strength of TCR activation. Bcl-2 was found to inhibit the cytoplasmic Ca elevation induced by a high concentration of anti-CD3 antibody but did not inhibit Ca elevation induced by low concentrations of anti-CD3 antibody. This finding evolved as a natural extension of our earlier investigation into the effect of Bcl-2 on InsP receptor–mediated Ca release from the ER (). Those studies initially used fluorometric measurements of Ca elevation in response to a relatively high concentration of anti-CD3 antibody. It was this Ca elevation that we found to be inhibited by Bcl-2. Although digital imaging was eventually used in addition to fluorometry in these earlier studies, a high concentration of anti-CD3 antibody was adhered to throughout. Thus, only the effect of Bcl-2 on the transient Ca elevation induced by a relatively high concentration of anti-CD3 antibody was investigated. The present study was undertaken based on the prediction that Ca oscillations would be detected if a lower concentration of anti-CD3 antibody were used. This prediction was based on evidence that high concentrations of cell surface receptor agonists generally induce transient elevations of Ca, whereas low concentrations of agonist are more likely to induce sustained oscillations (). Consistent with this paradigm, it was previously reported that strong TCR activation induces primarily a transient Ca elevation, whereas weaker TCR activation induces primarily repetitive Ca spikes, or oscillations (). As anticipated, the Ca response pattern in WEHI7.2 cells underwent a transition from transient Ca elevation to sustained oscillations as anti-CD3 antibody concentration was decreased ( A). The Ca oscillations induced by anti-CD3 antibody were irregular in their amplitude and frequency (). This is characteristic of Ca oscillations in T cells, as reported previously by others, and is in contrast to the more uniform pattern of Ca oscillations observed in nonlymphoid cells (for reviews see ; ). The irregularity of anti-CD3–induced Ca oscillations necessitated a large number of experiments to objectively compare oscillatory responses in Bcl-2–negative and –positive cells (). The only significant differences were an increase in oscillatory frequency () and duration of oscillations ( E) in Bcl-2–positive cells. The oscillatory patterns induced in Bcl-2–negative and –positive cells were indistinguishable in all other respects, including the percentage of cells that developed oscillations ( B), amplitude ( C), width of Ca spikes ( D), and latency period ( F). It has been reported that NFAT is optimally activated by Ca oscillations (; ; for reviews see ; ). Therefore, NFAT activation was measured as a convenient “readout” of the Ca oscillations induced by anti-CD3 in WEHI7.2 cells. Consistent with the finding that Bcl-2 did not inhibit anti-CD3–induced Ca oscillations, Bcl-2 did not inhibit NFAT activation ( G). The finding that Bcl-2 selectively inhibits the transient Ca elevation induced by high anti-CD3 without interfering with Ca oscillations induced by low anti-CD3 is relevant to the role of Bcl-2 in regulating apoptosis, as WEHI7.2 cells undergo apoptosis after treatment with high anti-CD3 but not when treated with low anti-CD3 ( C). Moreover, apoptosis induction by high anti-CD3 was Ca mediated (), and the percentage of cells undergoing apoptosis was proportional to the level of Ca elevation (). These findings are consistent with evidence that apoptosis induction after TCR activation is triggered by InsP receptor–mediated Ca elevation (; ). Thus, by selectively repressing the Ca elevation induced by strong TCR activation, Bcl-2 inhibits apoptosis without interfering with physiological Ca signals induced by weak TCR activation. The present findings are intriguing in light of the known role of Bcl-2 in T cell development. T cells located in the thymic cortex are TCR positive and both CD4+ and CD8+ (“double positive”). At this stage of T cell development, Bcl-2 expression is low and cortical thymocytes are highly susceptible to apoptosis induction after TCR activation by antigen or anti-CD3 antibody (; ; ; ). When T cells mature and migrate to the thymic medulla, they remain TCR positive but become either CD4+CD8− or CD4−CD8+ (“single positive”). Bcl-2 expression is increased at this stage of development, and as a consequence, single-positive thymocytes are less susceptible to apoptosis than are cortical thymocytes (; ). A strong correlation has been demonstrated between Bcl-2 expression and susceptibility to Ca-induced apoptosis during T cell development (). Thymocytes at the earliest stage of development (TCR−CD4−CD8−), the stage during which thymocytes relocate from bone marrow to thymus, express high levels of Bcl-2 and are resistant to Ca-mediated apoptosis, whereas thymocytes in the next stage of development (TCR+CD4+CD8+) are highly susceptible to Ca-mediated apoptosis. It is in this stage of development that thymocytes undergo either negative or positive selection. Strong ligation of the TCR (e.g., by self-peptide–major histocompatibility complex [MHC] complexes) induces negative selection, whereas weak ligation of the TCR (e.g., by foreign antigen–MHC complexes) induces positive selection (for review see ; ). Double-positive thymocytes from Bcl-2–transgenic mice accumulate excessively because of reduced negative selection and are resistant to anti-CD3–induced apoptosis (for review see ). Therefore, the role of Bcl-2 expression during T cell development may be to regulate when and where negative selection occurs. Decreased Bcl-2 expression in double-positive cortical thymocytes, but not in earlier or later stages of T cell development, may limit negative selection to the cortical region of the thymus and to this stage of development. Elevated Bcl-2 expression at earlier (double negative) and later (single positive) stages of T cell development may dampen Ca transients produced by strong TCR engagement while permitting repetitive Ca oscillations that signal cell proliferation and survival. The mechanism by which Bcl-2 differentially regulates Ca elevation after strong but not weak TCR activation is not entirely understood. In our earlier work (), the inhibitory effect of Bcl-2 on anti-CD3–induced Ca elevation appeared to be mediated at the level of the InsP receptor rather than in upstream TCR signaling pathways. This conclusion was based on two experimental strategies in which upstream TCR signaling pathways were bypassed. In one strategy, we found that Bcl-2 inhibited Ca elevation induced by a cell-permeant InsP ester. In the other strategy, we found that Bcl-2 inhibited ER Ca release induced by adding InsP to cells in which the plasma membrane had been permeabilized by digitonin. In addition, Bcl-2 appeared to interact with InsP receptors, based on results of blue native gel electrophoresis and coimmunoprecipitation (). Finally, purified Bcl-2 inhibited InsP-gated single-channel opening when microsomal membrane fractions containing InsP receptors were incorporated into planar lipid bilayers (). Therefore, the collective evidence that Bcl-2 interacts with InsP receptors and inhibits InsP-mediated Ca release from the ER raises the question of whether the induction of Ca oscillations by low concentrations of anti-CD3 antibody is InsP receptor independent or at least requires far fewer functional InsP receptors than does the elevation of Ca induced by a high concentration of anti-CD3 antibody. To address this question, we used siRNA to reduce InsP receptor levels in WEHI7.2 cells (). This procedure inhibited Ca elevation induced by strong TCR activation but did not inhibit the induction of Ca oscillations by weak TCR activation. In contrast, Ca responses evoked in HeLa cells by both high and low concentrations of ATP or histamine were repressed by InsP receptor type 1 knockdown (). Thus, mechanisms of Ca oscillation formation after TCR activation and G protein–coupled receptor activation may differ. Our findings indicate that Ca responses initiated by weak TCR activation are generated independent of InsP receptor–mediated Ca release or that only a relatively small proportion of the full InsP receptor complement is required to initiate Ca signals in response to weak TCR activation. In future studies, the mechanism of how Bcl-2 regulates InsP receptor function will be addressed in greater depth. In preliminary studies, we found that Bcl-2 overexpression decreases InsP receptor phosphorylation in WEHI7.2 cells. Moreover, it has recently been reported that Bcl-2 interacts with InsP receptors in a manner that is dependent on the Bcl-2 phosphorylation state and may regulate Ca dynamics in the ER through regulation of InsP receptor phosphorylation (; ). Others have reported that in neuronal cells Bcl-2 shuttles calcineurin to InsP receptors and regulates Ca release from internal stores (,; ). Therefore, one hypothesis is that strong TCR signals enhance InsP receptor phosphorylation, enhancing InsP-induced Ca release, and that Bcl-2 dampens the Ca response to strong TCR activation by mediating dephosphorylation of InsP receptors. Although untested, this theory is consistent with evidence that phosphorylation regulates the Ca channel activity of InsP receptors (; ; ; ; ). In summary, we previously discovered that the known antiapoptotic protein Bcl-2 interacts with InsP receptors and inhibits InsP-induced Ca release from the ER in T cells. In this paper, we report that Bcl-2 selectively inhibits Ca elevation induced by high but not low anti-CD3. As a consequence, Bcl-2 represses the transient elevation of Ca associated with apoptosis induction after strong TCR activation but does not interfere with Ca oscillations that activate NFAT after weak TCR activation. The capacity of Bcl-2 to differentially regulate Ca signals induced by strong versus weak TCR activation allows Bcl-2 to selectively inhibit apoptotic Ca signals without interfering with Ca signals that mediate cell proliferation and survival. EGTA and standard reagents were purchased from Sigma-Aldrich. Fura-2–AM and Hoechst 33342 were purchased from Invitrogen. Hamster anti–mouse CD3e ɛ chain monoclonal antibody (clone 145-2C11) and mouse anti–hamster IgG1 monoclonal antibody were obtained from BD Biosciences. Mouse monoclonal antibody NFATc2 was obtained from Santa Cruz Biotechnology, Inc. WEHI7.2 cells were cultured in DME supplemented with 10% bovine calf serum, 2 mM -glutamine, and 100 μM of nonessential amino acids. Transfection procedures, isolation of Bcl-2–positive and –negative clones, and the characterization of these clones were reported previously (). Methods of Ca imaging, described in detail previously (), were used here with only minor modifications. In brief, cells adhered to poly--lysine–coated coverslips (35-mm coverslip dishes; MatTek Corp.) were loaded with 1 μM fura-2–AM for 45 min at 25°C in extracellular buffer (ECB; 130 mM NaCl, 5 mM KCl, 1.5 mM CaCl, 1 mM MgCl, 25 mM Hepes, pH 7.5, 1 mg/ml BSA, and 5 mM glucose). The buffer was replaced with fresh ECB and the incubation was continued for 45 min at 25°C to permit deesterification. Culture dishes were mounted on the nonheated stage of an inverted microscope (CKX41; Olympus) equipped with a 20× fluor objective. Excitation light was alternated between 340 and 380 nm by a filter wheel (Sutter Instrument Co.), with 0.8- and 0.2-s exposure times, respectively, and emitted light was filtered at >510 nm and collected with an intensified charge-coupled device camera (12-bit VGA; Cooke). Anti-CD3 antibody was gently added to buffer overlaying the coverslip so as not to disturb cells loosely adherent to the coverslip. The video signal was digitized using InCyt Im2 software (Intracellular Imaging) and subsequently processed using Excel (Microsoft). To determine R, cells were perfused with ECB deficient in Ca and supplemented with 4 mM EGTA and 10 μM ionomycin. R was obtained by perfusing cells with ECB supplemented with 4 mM CaCl and 10 μM ionomycin. for fura-2 of 220 nM, by the equation of . The measurement of Ca concentration in cell suspensions by fluorometry using fura-2–AM have been described in detail previously (). Cells were treated with 2 μg/ml anti-CD3 antibody for various time periods at ambient temperature, after which they were placed on ice, pelleted, and resuspended in RIPA buffer (1% Triton X-100, 0.1% SDS, 50 mM Tris, pH 7.6, 150 mM NaCl, and 200 mM DTT) supplemented with Complete mini protease inhibitors (Roche) and Phosphatase inhibitor cocktails I and II (Sigma-Aldrich). Cell extracts were resolved by electrophoresis on 7% SDS–polyacrylamide gels under reducing conditions. The separated proteins were transferred to Immobilon-P PVDF membranes (Millipore) and incubated with anti-NFATc2 antibody at a dilution of 1:500, followed by incubation with horseradish peroxidase–conjugated goat anti–mouse IgG and visualized by the ECL Western blotting detection reagent (GE Healthcare). Western analysis for InsP receptors was performed as described previously (). Protein samples were extracted as in the preceding method and resolved (60 μg/lane) through 4–20% gradient gels (Bio-Rad Laboratories). The antibodies for types 1 and 3 InsP receptors were purchased from EMD Biosciences and BD Biosciences, respectively. The antibody for type 2 InsP receptor was a gift from R. Wojcikiewicz (State University of New York Upstate Medical University, Syracuse, NY). The antibody for actin was obtained from Sigma-Aldrich. Secondary antibodies were obtained from GE Healthcare. Cells were stained with Hoechst 33342 (final concentration 10 μg/ml), and typical apoptotic nuclear morphology was detected by epifluorescence microscopy using a microscope (Axiovert S100; Carl Zeiss MicroImaging, Inc.) equipped with a 63× oil/1.4 NA plan apochromat objective (Carl Zeiss MicroImaging, Inc.) and a filter cube (model XF23; Omega Optical; excitation = 485 nm, emission = 535 nm). Images were taken on a charge-coupled device camera (ORCA C4742-95-cooled; Hamamatsu) operating with Simple PCI software (Compix, Inc.). Cells (1 million/ml) were loaded with 5 μM calcium green–AM (Invitrogen) for 45 min at 37°C in ECB. The cells were then pelleted and resuspended in ECB at the same concentration and incubated at room temperature for 30 min to allow dye deesterification. The cells were then pelleted and concentrated to 5 million/ml. The cells were then analyzed and sorted on a flow cytometer (Epics Elite; Beckman Coulter). Calcium green fluorescence was measured after dye excitation with a 488-nM argon laser, and emitted light collection was measured through a 525-nM band-pass filter. The cells were initially run through the flow cytometer for 1 min to assess basal cytosolic Ca, and 20 μg/ml anti-CD3 antibody was then added. The cells were gated and sorted into two populations: cells with a high level of Ca elevation and cells with a low level of Ca elevation. The sorted cells were pelleted and resuspended in fresh culture medium and 20 μg/ml anti-CD3 antibody was re-added, and 30 min later an equal concentration of anti–hamster IgG was added. Apoptosis was measured 24 h later, as described in the previous section. The negative control, siCONTROL Non-Targeting siRNA Pool, and siGENOME SMARTpools for all three subtypes of InsP receptor were purchased from Dharmacon. After suspension in 1× siRNA buffer, SMARTpools were added at a concentration of 1 μM each to 0.2-cm cuvettes containing 5 million WEHI7.2 cells suspended in 200 μl Opti-MEM I (Invitrogen). Cuvettes were then subjected to a single 140V 10-ms–wave pulse from a GenePulser Xcell (Bio-Rad Laboratories), and the contents of the cuvette were immediately added to fresh media. Cells were grown in culture after transfection for 48 h before use in experiments. Transfection efficiency was measured by transfecting siGLO Cyclophilin (Dharmacon) at a concentration of 1 μM. After 30 min, cells were pelleted and then resuspended in phosphate-buffered saline. Cells were visualized by fluorescence microscopy, with excitation at 546 nm, and at least 200 cells were counted in three separate experiments to determine the percentage of transfected cells. Comparisons were made using the two-tailed test for two samples, assuming equal variance.

The protooncogene encodes a transcription factor, Myc, which forms an obligate heterodimeric complex with a partner protein, Max (; ). The complex can both activate and repress transcription. It activates transcription upon direct binding to specific DNA sequences, termed E-boxes, which are found in the promoters of a large group of Myc-induced genes, including both protein-coding and ribosomal RNA genes (). The Myc–Max complex represses transcription when it is tethered to DNA via other transcription factors, such as the zinc finger protein Miz1 (). The ability to bind to and activate transcription from E-boxes is required for multiple biological functions of Myc (); however, it is less clear which functions of Myc require complex formation with Miz1. Using a loss-of-interaction screen in yeast, we have previously mapped the Myc–Miz1 interaction surface to the “outside” of the helix-loop-helix domain (). This analysis identified a point mutant of Myc (MycV394D) that has lost the ability to bind to Miz1, but not to Max, in vivo and is not recruited to Miz1-binding sites on DNA (; ). MycV394D is capable of E-box–dependent activation of reporter plasmids, but does not repress Miz1-activated transcription (). Extensive array analyses showed that MycV394D is fully able to activate transcription of endogenous Myc target genes, but fails to repress a large set of genes that are repressed by wild-type Myc (; unpublished data). Surprisingly, MycV394D induces apoptosis and cell cycle entry in serum-starved fibroblasts with an efficiency similar to wild-type Myc (). Furthermore, the mutant is able to transform primary rat embryo fibroblasts together with an activated allele of Ras, suggesting that binding to Miz1 is not required for these properties of Myc (unpublished data). One group of genes that is repressed by Myc via Miz1 encodes the cell cycle inhibitors p15Ink4b (; ), p21Cip1 (; ; ), and p57Kip2 (). Of these inhibitors, p15Ink4b is induced by TGF-β and mediates the TGF-β–induced arrest of proliferation (). A second class of genes that is repressed by Myc encodes proteins involved in cell–cell adhesion, in the actin cytoskeleton, and in adhesion to the ECM (; ; ). Whether Miz1 is involved in their regulation is unknown. A tissue in which Myc-mediated repression of gene expression is important is the epidermis. The epidermis is maintained throughout adult life by a stem cell population (). When cells exit from the stem cell compartment, they undergo a few further rounds of division, during which time they are known as transit-amplifying cells. Thereafter, they undergo terminal differentiation along several distinct lineages, forming the interfollicular epidermis, sebaceous gland, and hair follicle (). Activation of Myc in cultured human epidermal cells stimulates cells to become transit-amplifying cells and to undergo terminal differentiation (). Activation of Myc in the basal layer of transgenic mouse epidermis leads to an increase in proliferation, which may reflect the cell's departure from the stem cell into the transit-amplifying cell compartment and a stimulation of differentiation into interfollicular epidermis and cells of the sebaceous gland (; ; ; ). Repression of gene expression by Myc is potentially important in triggering these events (; ). First, ectopic expression of Myc represses p15in4kb and renders cultured keratinocytes resistant to growth inhibition by TGF-β (; ), suggesting a potential mechanism for the increase in proliferation resulting from Myc activation in vivo (; , ). Second, keratinocyte adhesion to the underlying basement membrane is a negative regulator of terminal differentiation () and repression of cell adhesion genes by Myc may promote exit from the stem cell compartment and differentiation by disrupting adhesive interactions with the local microenvironment (; ; ; ). We now show that endogenous Miz1 is highly expressed in the basal and suprabasal layers of mouse epidermis and that multiple genes involved in cell–cell and cell–basement membrane adhesion of keratinocytes are regulated by Myc through binding to Miz1. Our data show that Myc regulates cell adhesion by binding to Miz1 and may help to elucidate how Myc can regulate the epidermal stem cell compartment. In situ hybridization of embryonic day (E)15.5 mouse embryos has shown that mRNA is preferentially expressed in multiple epithelia, including skin, the olfactory epithelium, and epithelia of the gastrointestinal tract (). To extend these findings, we stained sections of murine epithelia with a monoclonal antibody (10E2) that is directed against Miz1. We observed strong nuclear staining in all epithelia that we analyzed, including paw skin, back skin, and tongue (, A–C). In stratified epithelia, nuclear staining was most intense in basal and immediately suprabasal layers, and the staining intensity declined in more differentiated cell layers (, A–C). Two controls were performed to ascertain the specificity of staining: first, no staining was observed in sections lacking the primary antibody (unpublished data); second, staining was abolished by preincubation of the antibody with a bacterial lysate expressing a GST-Miz1 fusion protein, but not by preincubation with an equal amount of nonrecombinant lysate (). We conclude that Miz1 is highly expressed in mouse epithelia. To analyze the function of Myc and Miz1 in these cells, we isolated keratinocytes from newborn mice and infected them with recombinant pBabe-puro retroviruses () encoding either wild-type human Myc or MycV394D (Introduction; ). As a control, cells were infected with empty pBabe-puro viruses. Infected cells were selected in puromycin, and pools of resistant cells were used for further analysis. Immunoblotting showed that pools of cells infected with either Myc or MycV394D viruses expressed equally elevated levels of Myc protein, relative to control-infected cells ( G). To test whether Myc affects keratinocyte proliferation, we labeled exponentially growing cells for 1.5 h with BrdU and determined the percentage of cells incorporating BrdU by immunofluorescence ( E). In these experiments, we did not observe a reproducible difference between the different cell populations in the absence of TGF-β (unpublished data). Upon addition of 100 pM TGF-β, control cells rapidly exited from the cell cycle ( E). FACscan analysis revealed that addition of TGF-β led to an arrest in both the G1 and the G2 phase of the cell cycle ( F). The response of cells expressing MycV394D was indistinguishable from that of control infected cells; in contrast, cells expressing wild-type Myc showed a significantly delayed exit from the cell cycle ( E). Surprisingly, even cells expressing wild-type Myc were not completely resistant to TGF-β, in contrast to what has been observed in established cell lines (; ). Immunoblotting revealed that addition of TGF-β decreased the amount of not only endogenous but also of the retrovirally expressed Myc proteins; whether this is caused by enhanced proteolysis or inhibition of translation is currently unknown ( G). We suggest that this down-regulation accounts for eventual cell cycle exit, even for cells expressing wild-type Myc. Myc and Miz1 have both been implicated in the expression of the cell cycle inhibitors p15Ink4b and p21Cip1, both of which can be induced by TGF-β in different cell types (; ). Immunoblotting of cell extracts revealed that addition of TGF-β had no effect on the expression of p21Cip1 in primary mouse keratinocytes ( G). In contrast, TGF-β up-regulated expression of p15Ink4b in control cells and in cells expressing MycV394D, but not in cells expressing wild-type Myc. This regulation of p15Ink4b is in accordance with the altered proliferation behavior of cells expressing wild-type Myc. Similarly, an RT-PCR analysis revealed that expression of Myc, but not of MycV394D, inhibited up-regulation of mRNA in response to TGF-β ( H). Analysis of multiple cell cycle regulatory genes revealed that addition of TGF-β suppressed the expression of , , and , which are genes that are expressed in the S phase of the cell cycle. Expression of Myc, but not of MycV394D, maintained expression of these genes even in the presence of TGF-β. Because these genes are regulated by E2F factors, which act downstream of p15Ink4b, we suggest that their regulation is an indirect consequence of the regulation of mRNA by Myc. In contrast, TGF-β had no influence on expression of cyclin mRNAs expressed in the G1 (with the possible exception of cyclin D1) and G2 phases of the cell cycle, consistent with the observation that the addition of TGF-β led to an arrest in both the G1 and the G2 phases of the cell cycle ( F). We also considered the possibility that the differential effects of Myc and MycV394D might be mediated through another member of the Myc network of proteins. Therefore, we measured the expression of , , , , , and by RT-PCR in these cells ( I). Consistent with earlier findings, the addition of TGF-β led to an up-regulation of gene expression (). Importantly, neither expression of Myc nor of MycV394D had an effect on expression of any the tested mRNAs, suggesting that the differential effects of both proteins on cell proliferation are not indirect consequences of regulation of another Myc network protein. Finally, we found that the effects were specific for the response of keratinocytes to TGF-β because neither Myc nor MycV394D had any effect on Ca-induced cell cycle exit or expression of p21Cip1 and p15Ink4b (unpublished data). Together, the data show that the binding of Myc to Miz1 is required for inhibition of p15ink4b expression and for maintaining keratinocytes in cycle in the presence of TGF-β. To identify epidermal genes that are regulated by Myc via binding to Miz1, we performed a microarray analysis of RNA isolated from either control keratinocytes or from cells expressing Myc or MycV394D both before and 12 h after addition of TGF-β. For this analysis, we used a 22,500 complementary DNA (cDNA) array. The analysis identified 160 genes, which were induced by twofold or more upon the addition of TGF-β to control infected cells. Comparison of the expression of these genes to cells expressing either wild-type Myc or MycV394D revealed three distinct expression profiles ( A and Table S1, available at ). The induction of the first class of genes ( = 54), exemplified by , was unaffected by either Myc or MycV394D, confirming previous suggestions that Myc does not generally inhibit signal transduction by TGF-β (; ). A second group of genes ( = 49), including , was inhibited by expression of both Myc and MycV394D ( A). Finally, we found 67 genes, including , which were induced by TGF-β in control cells and in cells expressing MycV394D, but not in cells expressing wild-type Myc, arguing that they were repressed by Myc via binding to Miz1 ( A). These results were confirmed by a second array experiment from independently generated cell pools (unpublished data) and by an RT-PCR experiment for several of the differentially regulated genes ( B). Functional annotation revealed that almost 40% of the genes in this latter group encoded components of the ECM (e.g., tenascin c), integrin ECM receptors (e.g., integrins subunits β4 and α6), and proteins involved in cell–cell interactions (e.g., α-catenin; C). Our previous work had shown that Myc represses a group of functionally similar genes in mouse skin in vivo (). In this analysis, we had also identified genes encoding proteins of the cytoskeleton as targets for repression by Myc, some of which are not present on the array used in the current analysis. Thus, we performed additional RT-PCR experiments and found that at least one of these genes, , is repressed by wild-type Myc, but not by MycV394D ( B). Finally, the group included two collagen genes, which were expressed at low levels in the isolated keratinocytes although they are predominantly expressed in the dermis in vivo ( and ). The results suggest that binding of Myc to Miz1 is required for repression of a set of genes that regulate the adhesive properties of mouse keratinocytes. To test whether any of these genes are direct targets of Myc and Miz1, we performed chromatin-immunoprecipitation experiments. Because the available antibodies do not precipitate mouse Miz1, we used chromatin isolated from HeLa cells for these experiments. Chromatin was prepared and precipitated with secondary reagents alone, with control antibodies, or with antibodies directed against Myc or Miz1, respectively ( A). The known binding sites for Miz1 are close to the transcription start sites (; ); therefore, we performed PCR using primer pairs that span 1 kb around the transcription start sites of the indicated genes. In vivo binding of Miz1 and Myc was detected at the start sites of the integrin α6 (), β1 (), and β4 () genes, as well as to procollagen I α2 () and galectin-1 () genes ( A). To demonstrate specificity of binding, we used target genes that are activated by Myc and detected binding of Myc, but not of Miz1, to sequences surrounding E-box elements in the nucleolin (), proliferating cell nuclear antigen (), and prothymosin-α genes (; A). As a negative control, we used primers surrounding the start site of the β-tubulin () gene, which is regulated neither by Myc nor by Miz1. To exclude the possibility that binding was restricted to HeLa cells, we repeated the experiment for several of the genes using chromatin from an established human keratinocyte line (HaCaT) and obtained identical results ( B). We conclude that at least some of the genes regulating cell adhesion in keratinocytes are direct target genes of Myc and Miz1. To confirm the differential effects of Myc and MycV394D at the protein level, primary human epidermal keratinocytes were transduced with retroviral vectors encoding inducible alleles of Myc and MycV394D (termed MycER and MycV394DER, respectively) or, as a control, the empty vector (pBabe). As previously reported (; ), cell surface levels of β1 and α6 integrins decreased upon activation of MycER with 4-hydroxy-tamoxifen (4-OHT; ). In contrast, when MycV394DER was activated, α6 integrin levels increased relative to the empty vector control cells ( A) and β1 integrin levels remained unchanged ( B). In addition to reducing integrin expression, Myc impairs keratinocyte spreading and migration and this correlates with decreased expression of components of the actin cytoskeleton (). Immunoblotting showed that activation of MycER led to decreased galectin-1 and adducin levels, whereas activation of MycV394D had either no or a smaller effect ( C). Activation of either Myc or MycV394D resulted in increased expression of nucleolin, a known target for transactivation by Myc ( C; ). Consistent with these observations, chromatin immunoprecipitations (ChIPs) showed that MycER, but not MycV394DER, bound to the start site of both β1 and α6 integrin genes in vivo in these cells; in contrast, both MycER and MycV394DER bound to the E-box of the nucleolin promoter ( D). These results are consistent with the expression array of primary mouse keratinocytes and suggest that Myc represses expression of integrins (α6, β4, and β1), galectin-1, and adducin through binding to Miz1, whereas transactivation of E-box element genes is not dependent on Miz1. To assess the effects of these alterations on ECM adhesion, we measured the spreading of infected keratinocytes on collagen-coated dishes (). When plated in serum free medium (FAD), both control keratinocytes and keratinocytes expressing MycV394DER spread extensively (). In contrast, cells expressing wild-type Myc were significantly less spread out (). As previously reported, EGF reduced spreading of control keratinocytes (; ). In contrast, spreading of keratinocytes expressing MycV394DER was not reduced by the addition of EGF (, FAD+EGF), suggesting that MycV394DER may act as a dominant-negative inhibitor of EGF-induced cell contraction. Myc activation inhibits cell motility and wound healing in vivo and in vitro (; ). To analyze whether MycV394DER impairs motility in keratinocytes, we performed motility assays using time-lapse microscopy. Human keratinocytes were pretreated with 4-OHT, plated on collagen-coated dishes and filmed for 36 h. Surprisingly, keratinocytes expressing either MycV394DER or MycER showed a similar reduction in motility, relative to control cells (Videos 1–3, available at ). Whereas the reduced motility of MycER cells correlated with decreased spreading and lamellipodia formation, cells expressing MycV394DER were more highly spread than control cells ( C). It is likely that the reduced motility of keratinocytes expressing MycV394DER is attributable, at least in part, to the increased expression of α6β4, relative to control cells (). EGF is known to stimulate keratinocyte motility () and the reduced motility of MycV394D-expressing cells may also reflect their inability to contract when stimulated with EGF (). Expression of integrin β1 has been linked to exit from the stem cell compartment (; ). To test whether Miz1 might have a role in Myc-induced exit from the stem cell compartment, we measured the proliferative potential of individual keratinocytes in a clonogenic assay ( D). Consistent with earlier observations (), expression of MycER resulted in a reduced colony number and a higher proportion of abortive colonies, suggesting that Myc promotes exit from the stem cell compartment in vitro. In contrast, expression of MycV394DER resulted in a higher number of colonies and a higher proportion of proliferative cells, relative to control infections. These findings are consistent with the differences in regulation of integrin gene expression and suggest that Miz1 may have a role in Myc-induced exit from the stem cell compartment in vivo. In skin, proliferating keratinocytes are in contact with the basement membrane, and disruption of adhesion to the basement membrane inhibits proliferation and promotes terminal differentiation (). In vivo, the activation of MycER in cells of the basal layer and their progeny leads to the appearance of terminally differentiated cells in the basal layer of interfollicular epidermis, potentially because of disrupted cell adhesion to the basement membrane (). Interfollicular epidermis can be reconstituted in culture by growing human epidermal cells on de-epidermized dermis (DED) obtained from human breast skin (). Therefore, we used DED cultures to compare the effects of Myc and MycV394D on terminal differentiation (). Stratification of control keratinocytes on DEDs resembled that of normal epidermis in vivo, with distinct basal, spinous, granular, and cornified layers ( A). The epidermis formed by keratinocytes infected with MycER in the presence of 4-OHT was more disorganized, with a striking accumulation of cornified layers (). DEDs reconstituted with keratinocytes infected with MycV394DER showed a similar accumulation of the differentiated layers to MycER ( C). In contrast, the morphology of the basal layer of epidermis reconstituted with keratinocytes expressing MycER or MycV394DER could easily be distinguished. The basal layer of epidermis reconstituted with wild-type Myc was loosely packed and disorganized, whereas the basal layer of MycV394D-expressing epidermis was densely packed, with highly polarized cells in contact with the basal lamina (, D–F). Staining with antibodies directed against E-cadherin (, J–L) and α-tubulin (unpublished data) showed that cells in the basal layer of DEDs reconstituted with control keratinocytes or cells expressing MycV394DER appeared more polarized compared with cells expressing MycER. Staining with antibodies directed against Myc clearly revealed the presence of the ectopically expressed MycER and MycV394DER proteins and established that the observed differences in morphology were not caused by differences in expression of both proteins (, M–O). To analyze the effects on cellular differentiation, we stained sections with antibodies against involucrin, a marker of terminal differentiation (). DEDs reconstituted with keratinocytes infected with MycER show increased numbers of involucrin-positive cells in the basal layer, indicating premature differentiation (; ). In contrast, expression of involucrin in DEDs reconstituted with MycV394DER was restricted to the suprabasal layers, similar to DEDs reconstituted with control keratinocytes, strongly suggesting that induction of premature differentiation by Myc requires binding to Miz1 ( I). We conclude that Myc binding to Miz1 is required for premature terminal differentiation and disorganization of the basal layer. The lack of premature differentiation of human keratinocytes expressing MycV394D suggested that binding to Miz1 is required for this effect of Myc. To provide direct evidence that repression of integrin expression contributed to a phenotype of mice in which Myc is activated in the epidermal basal layer, we crossed K14MycER mice with mice that express the human β1 integrin subunit under the control of the K14 promoter (K14β1 mice; ). We chose β1 integrin for this experiment because both Myc and Miz1 bind to its promoter in vivo and because expression of this integrin has been linked to exit from the stem cell compartment in cultured human keratinocytes (). The phenotype of K14β1 mice was indistinguishable from wild-type mice ( A, B, E, F, I, J, M, N, Q, and R). In the presence of the β1 integrin the Myc-induced increase in differentiated cells of the sebaceous gland was reduced (), but the sebaceous glands were not entirely normal because the number of proliferating cells at their periphery was increased (). The β1 integrin did not prevent Myc-induced proliferation and indeed Ki67 labeling was slightly higher in the double transgenic mice than in K14MycER mice (, I–L). Importantly, however, transgenic expression of β1 integrin resulted in a normalization of keratin 10 and of involucrin expression and strongly reduced the number of cells in the basal membrane expressing either keratin 10 or involucrin (, M–P and Q–T); in addition, expression of the β1 transgene restored polarity to the basal layer (). In contrast, overexpression of β1 integrins did not prevent the delay in wound closure resulting from Myc activation (, U–X; and not depicted). We conclude that Myc–Miz1-mediated down-regulation of integrin expression is responsible for the loss of polarity and premature differentiation of cells in the basal layer of the epidermis. Several previous observations have linked the Myc oncoprotein to the control of cell adhesion. In particular, down-regulation of several integrins is observed in different cell types expressing either deregulated (c-)Myc or N-Myc (; ; ; ; ). Conversely, genetic ablation of endogenous (c-)Myc up-regulates multiple genes encoding cell adhesion proteins, arguing that regulation of cell adhesion by Myc is a physiological function during normal development (). We show here that in keratinocytes regulation of cell adhesion by Myc occurs through the Myc–Miz1 complex. This notion is supported by four main arguments. First, endogenous Miz1 is highly expressed in the basal layer of multiple epithelia. Second, a microarray analysis identifies multiple genes that encode proteins of the ECM, cell surface receptors for ECM proteins, and proteins that are involved in cell–cell or cell–matrix adhesion as genes that are repressed by wild-type Myc, but not by MycV394D, a point mutant of Myc that is unable to bind to Miz1 (). Third, ChIP reveals that Myc and Miz1 bind in vivo to the start sites of the genes we tested. In particular, the integrins α6, β1, and β4, which mediate the adhesion of keratinocytes to the basement membrane and which are critically required for the integrity of the epidermis in vivo, are direct target genes of both Myc and Miz1. These findings are consistent with a previous study demonstrating in vivo binding of Miz1 to the start site of the integrin α2 promoter (). Fourth, spreading of keratinocytes in vitro on collagen, as a measure of α2β1-mediated adhesion to the ECM, is inhibited by wild-type Myc, but not MycV394D. Together, our data demonstrate that regulation of keratinocyte adhesion is a direct function of Myc that is exerted through binding to Miz1. In epidermis, cell adhesion is tightly linked to the control of terminal differentiation and departure from the cell cycle (for review see ). Activation of Myc in keratinocytes results in exit from the stem cell into the transit-amplifying compartment, which is associated with an increase in proliferation and stimulation of sebocyte and interfollicular epidermal differentiation in vivo (; ). Comparison of epidermis reconstituted on DED by Myc- or MycV394D-expressing keratinocytes and of transgenic epidermis, in which Myc is activated alone or in combination with β1 integrin overexpression, allows us to dissect the different roles of Myc in this process. The disruption of the basal layer and premature interfollicular epidermal differentiation can be attributed to Miz1-dependent repression of gene expression by Myc. In contrast, introducing the V394D mutation in Myc or overexpressing β1 integrin does not normalize the increase in sebocyte differentiation, arguing that repression of integrin expression is not required for these events. These data imply that the interaction between Myc and Miz1 must be tightly controlled during normal skin differentiation; in particular, expression levels of endogenous Myc are high in basal layers, where we suggest that Miz1 is active to drive integrin expression. Although we do not know how exactly the interaction between Myc and Miz1 is regulated, we have recently characterized an E3-ligase, HectH9, that ubiquitinates and activates free Myc–Max complexes, but not Myc bound to Miz1, through K63-linked ubiquitination, suggesting that ubiquitination by HectH9 may have a role in regulating the interaction (). It should be noted that elevated expression of integrins and of multiple TGF-β–induced transcripts are markers of stem cells in human interfollicular epidermis and in human and mouse hair follicles (). Down modulation of adhesive interactions by Myc has been suggested to promote exit of both epidermal and hematopoietic stem cells from the stem cell niche (; ; ; ). Our results suggest that the interaction of Myc with Miz1 may play an important role in this process. Finally, Myc is a classic protooncogene and it is worth noting that disruption of cell adhesion may also be critical for the oncogenic effects of Myc (). For example, activation of Myc not only promotes growth and proliferation but also induces highly invasive tumors in a transgenic model of pancreatic tumorigenesis using the insulin promoter (). In addition, amplification of is closely correlated with the progression from the in situ to the invasive stage in human breast carcinomas (). Therefore, we suggest that disruption of cell adhesion by Myc through Miz1 could contribute to the genesis of a wide range of tumors; genetic models to address this hypothesis are currently being generated. 1 d postpartum Friend leukemia virus B (National Institutes of Health [NIH]) mice were killed by decapitation, and keratinocytes were isolated according to . For cell culture medium, we used Eagle's minimum essential medium with Earle's BSS without CaCl (Cambrex), supplemented with 10 M choleratoxin (Calbiochem), 0.4 μg/ml hydrocortisone (Sigma-Aldrich), 0.75 mM aminoguanidine nitrate (Sigma-Aldrich), and 2 ng/ml EGF (ICN Biomedicals) and with antibiotic and antimycotic agents (10,000 U/ml penicillin, 10,000 μg/ml streptomycin, and 25 μg/ml amphotericin B; Invitrogen). FCS (Sigma-Aldrich) was decalcified using Chelex 100 (Bio-Rad Laboratories) and was added in a concentration of 8% to the medium. This essentially calcium-free medium was then substituted with 60 μM CaCl. To obtain conditioned medium, we cultivated primary dermal fibroblasts in this medium for 48 h. Isolated keratinocytes were cultivated at 37°C in a humidified chamber equilibrated with 5% CO, using equal parts of conditioned and unconditioned medium. Cells were plated at 10 cells/cm on collagen IV (Fluka)–coated plastic dishes (1 μg/cm) and cultured for 2–3 d until they reached 60–80% confluency. Keratinocytes were infected with retroviral supernatants for 7 h, rinsed two times with PBS, and placed in keratinocyte medium. After 2 d, selection with 1 μg/ml puromycin (Sigma-Aldrich) was started. 2.5 d later, cells were incubated with 100 pM TGF-β1 (Sigma-Aldrich) for 12 or 24 h. J2-3T3 cells were cultured in DME containing 10% donor calf serum. Primary human keratinocytes were isolated from neonatal foreskin and cultured in the presence of a feeder layer of J2-3T3 cells in FAD medium (one part Ham's F12 medium, three parts DME, and 1.8 × 10 M adenine) supplemented with 10% FCS and a cocktail of 0.5 μg/ml hydrocortisone, 5 μg/ml insulin, 10 M cholera toxin, and 10 ng/ml EGF, as described previously (). Keratinocytes were cultured on a dermal equivalent of dead DED from adult breast skin (). Fourth or fifth passage keratinocytes were seeded on DED and grown at the air–liquid interface for 14 d. Retroviral supernatants were generated by transient transfection of Phoenix packaging cells and stable infection of AM12 cells. Keratinocytes were infected by coculture with retroviral producer cells as described previously () and used within one or two passages after infection. Activation of the steroid-inducible constructs was performed by adding 200 nM 4-OHT (Sigma-Aldrich) to the culture medium. Total cellular RNA was isolated with the RNAeasy reagent (QIAGEN) and a subsequent DNase digestion was included. RNA was amplified with MessageAmp RNA amplification kit (Ambion) and labeled with CyScribe cDNA post-labeling kit (GE Healthcare). All procedures were performed according to the manufacturer's instructions. RNA quality was controlled by agarose gel electrophoresis. Each experiment was performed as a sandwich hybridization, i.e., instead of a coverslip, a second microarray slide was used. This provides a replicated measurement for each hybridization that can be used for quality control and to reduce technical variability. Spot intensities were extracted from a scanned image with ImaGene 3.0 (BioDiscovery, Inc.). Software parameters like “signal range” or “spot detection threshold” were optimized for maximum reproducibility before the image analysis of our experiment. For each spot, median signal and background intensities for both channels were obtained. To account for spot differences, the background-corrected ratio of the two channels were calculated and log transformed. To balance the fluorescence intensities for the two dyes, as well as to allow for comparison of expression levels across experiments, the raw data was standardized. We used a spatial- and intensity-dependent standardization to correct for inherent bias on each chip (the lowess scatter-plot). To compare different cell types and conditions, the replicated chips were averaged. To find differently expressed genes, the differences between different cell lines with and without TGF-β treatment was calculated for each gene. First-strand cDNAs of amplified RNA were synthesized with M-MLV Reverse Transcriptase (Invitrogen) and random hexamer primers (Promega). For each PCR amplification, aliquots were taken after different cycles to determine the linear range of the amplification. Primer sequences are available in Table S2 (available at ). ChIP assays were performed as described previously () using the following antibodies: anti-Miz1 (C-19), anti–c-Myc (N-262), and anti-Gadd45 (H-165). All were obtained from Santa Cruz Biotechnology, Inc. Immunoprecipitated DNA was amplified by PCR using primers specific for the transcription start site of the galectin, procollagen Iα2, integrin α6, and integrin β4 promoters and for the E-boxes of pcna, nucleolin, prothymosin-α, and β-tubulin promoters. For each promoter, PCR reactions were performed with a different number of cycles or with dilution series of input DNA to determine the linear range of the amplification; all results shown fall within this range. Primer sequences are available in Table S2. Tissue samples were either fixed in 4% PBS-buffered formalin and embedded in paraffin or were placed in optimal cutting temperature compound (Miles Scientific) and frozen in isopentane surface cooled with liquid nitrogen. 5-μm sections were used for hematoxylin and eosin staining, immunohistochemistry, and immunofluorescence. Paraffin sections of tissue samples were microwaved in antigen retrieval solution (10 mM citrate buffer, pH 6) for 15 min in a microwave oven. After being cooled to room temperature, slides were washed in distilled water, incubated in 3% HO for 10 min, and washed three times in PBS. After a 45-min incubation in 10% normal goat serum in PBS, the undiluted 10E2 antibody was applied and slides were incubated overnight at 4°C. Slides were washed three times in PBS and incubated for 30 min at room temperature with a biotinylated secondary antibody (goat anti–mouse immunoglobulin G; DakoCytomation) at a 1:500 dilution. After three washes in PBS, a streptavidin–biotin–peroxidase complex (DakoCytomation) was applied for 30 min at room temperature. Slides were washed three times in PBS again and incubated with 3-amino-9-ethylcarbazole (Zymed Laboratories) at room temperature under microscopic control. After a final wash in distilled water, slides were mounted in Mowiol (Hoechst). Primary human keratinocytes were labeled with α6 integrin antibody conjugated to FITC (BD Biosciences) or anti–β1 integrin (MB1.2; ) for 30 min at room temperature with rotation. Cells were then analyzed using a FACscalibur II sorter and Cell Quest FACS analysis system (BD Biosciences). For cell cycle analysis, keratinocytes were harvested and washed with PBS, fixed in 70% ethanol on ice for 30 min, washed twice with PBS, and treated with 100 μg/ml RNase. After staining with 50 μg/ml propidium iodide, the DNA content was analyzed by flow cytometry. Primary murine keratinocytes were plated on coated 24-well dishes and cultured as described in Cell culture, transfection, and retroviral infection. After incubation with TGF-β for 0, 12, or 24 h, BrdU was supplied to the keratinocyte medium 1.5 h before the termination of incubation. Keratinocytes were fixed in 4% PBS-buffered paraformaldehyde for 30 min at room temperature, washed three times in PBS, and incubated for 10 min in 10 mg/ml glycine. After three washes in PBS, cells were incubated with 3% HO in methanol for 20 min at room temperature and washed three times in both PBS and PBS/0.1% Tween. 2 M HCl/0.1% Triton X-100 was applied for 5 min at room temperature, and then cells were washed three times in PBS and blocked in 3% BSA/0.1% Tween in PBS for 45 min at room temperature. The BrdU antibody (DakoCytomation; diluted 1:50 in 3% BSA/0.1% Tween) was applied and cells were incubated over night at 4°C. Further staining was performed as described as in Immunohistochemistry. Cells were counterstained with DAPI. To obtain the percentage of BrdU-labeled cells, a minimum of 3,500 cells from each sample were counted. To analyze motility, cells were cultured in complete FAD medium and incubated with 4-OHT for 24 h, harvested, and cultured on collagen-coated dishes in the presence of 4-OHT (Becton Dickinson). The cells were kept humidified at 37°C with 5% CO and videotaped for 48 h. Frames were taken every 4 min using inverted microscopes (models IMT1 or IMT2; Olympus) driven by Broadcast Animation Controllers (BAC 900; EOS Electronics [AV], Ltd.) and fitted with monochrome charge-coupled device cameras and video recorders (models M370 CE and PVW-2800P, respectively; Sony). Recordings were digitized and the sequence of all frames was run on a PC. For the clonogenicity assays, retrovirally infected keratinocytes were treated with 4-OHT for 3 d. The cells were harvested and 10 keratinocytes were plated per 35-mm dish coated with type IV collagen (Becton Dickinson). Mitotically inactivated J2-3T3 feeders were added; after 10 min, the dish was rinsed with PBS to remove nonadherent cells and then cultured for 14 d. Cells were fixed with 4% paraformaldehyde and stained with 1% rhodamine B and 1% Nile blue (BDH). Colonies were viewed using a dissection microscope (model 3DZ; Wild) and scored as described previously (). The K14β1 transgene cassette was generated by inserting the human β1 integrin subunit cDNA () into the K14 expression cassette (gift of E. Fuchs, The Rockefeller University, New York, NY) previously used to generate K14MycER transgenic mice (). The K14β1 transgene was injected into the male pronucleus of day 1 fertilized (CBA × C57BL/6) F1 mouse embryos. Six founder lines were obtained, all of which showed cell surface expression of the human β1 integrin subunit on the surface of cells in the epidermal basal layer, outer root sheath, and periphery of the sebaceous gland. None had a spontaneous skin phenotype. The line with the highest β1 integrin expression level (3428B.3) was selected for further analysis. It was crossed with K14MycER mice (2184C.1; ). Dorsal epidermis was treated with 1 mg/ml 4-OHT, three times per week, and animals were killed after 7 d. In some experiments, mice received two 3- or 5-mm-diam punch biopsy wounds that were treated with 4-OHT the day after wounding and harvested at day 7. Table S1 shows a summary of results from the microarray experiment. Videos 1–3 are time-lapse movies documenting the motility of human keratinocytes transduced with either control, MycER, or MycV394DER viruses. Table S2 shows the primer sequences for RT-PCR and ChIP experiments. Online supplemental material is available at .

Mouse vascular morphogenesis begins in the yolk sac (YS) on embryonic day (e) 6.5, when endothelial cells (ECs) differentiate from mesenchymal-derived precursors (angioblasts) in the blood islands of the extraembryonic mesoderm. ECs subsequently coalesce into a honeycomb-shaped primitive capillary plexus. This de novo formation of blood vessels is called vasculogenesis and occurs slightly later in the embryo proper. By e8.5, the dorsal aortae, cardinal veins, and the surrounding primitive vascular plexus emerge. To assemble a mature vascular network composed of a hierarchy of arteries, arterioles, capillaries, postcapillary venules, and veins, the primitive vasculature undergoes a profound remodeling starting at e8.5 involving vessel expansion and regression (; ). Angiogenesis, the growth of new capillaries from preexisting blood vessels, plays an essential role in vascular expansion (; ; ). Two mechanisms are involved in forming new vascular segments. Sprouting angiogenesis is the extension of new vessels from existing capillaries into avascular tissues. It is thought that tip cell migration results in a decrease in EC-to-cell contacts, triggering EC proliferation and generation of new blood vessels (). Thus, EC migration and proliferation are crucial events in sprouting angiogenesis. Intussusceptive angiogenesis is a process in which new vessels are formed by the insertion and extension of translumenal tissue pillars. It is hypothesized that these processes require no immediate proliferation of ECs; rather, they depend on the rearrangement of existing cells through the regulation of cell migration and adhesion (). Vascular regression, or pruning, is an essential remodeling process in which excess ECs or vascular segments are eliminated to construct an efficient network (). It is believed that only a minority of blood vessels formed during embryonic development remain through adulthood (). The two most prominent examples of vascular regression are the primordial aortic arches and paired dorsal aortae, which exist only transiently in mammals (). The precise morphological events, along with the underlying cellular and molecular mechanisms, remain obscure. Targeted mutagenesis of many signaling pathways (e.g., VEGF, angiopoietin, ephrinB2, Notch, or TGF-β) results in a defective primitive vascular network, suggesting that they are essential for vascular remodeling (). However, these signaling pathways often elicit an array of biological effects, and the precise cellular function and intracellular signaling events that mediate such function within ECs are unclear. FAK is a ubiquitously expressed protein–tyrosine kinase that mediates integrin, growth factor, and mechanical stress signaling (; ; ). Localized to focal adhesions, it interacts with integrin-associated proteins, such as paxillin and talin, and elicits downstream signaling. Activation of FAK via tyrosine phosphorylation occurs when cells are grown on integrin ligands or are stimulated by certain growth factors. The importance of FAK in vascular morphogenesis is evident because of its abundant expression in the vasculature at the time of critical vascular development (). Furthermore, -null embryos die at e8.5 with multiple defects, including a disorganized cardiovascular system (). embryos indicates that the absence of does not prevent EC differentiation (). However, -null embryos fail to form vascular networks, suggesting that FAK functions in the subsequent angiogenesis and vascular remodeling. In vitro experiments have also demonstrated that FAK relays angiogenic signaling from multiple pathways, including those of VEGF and angiopoietin-1, and mediates EC migration, survival, and proliferation (; ; ; ; ). It is well recognized that conclusions drawn from in vitro experiments can be inconsistent with in vivo findings. For example, in ECs as well as in other cell types, FAK is critical for in vitro cell proliferation (; ). embryos or in a brain-specific conditional deletion mutant (; ), raising doubts about whether FAK is important for cell proliferation in vivo. Furthermore, studies in different cell types reveal differential effects of FAK on cell migration. For example, -null fibroblasts exhibit an increase in focal adhesion contacts and a decrease in cell migration (; ). However, HeLa cells expressing reduced levels of FAK by short inhibitory RNA demonstrate an increase in cell motility (), raising the possibility that FAK may elicit diverse effects in different cell types. Thus, it is imperative to dissect FAK's role in specific cell lineages in an in vivo environment. In this study, we selectively deleted in ECs during mouse embryogenesis and investigated the subsequent effects on vascular morphogenesis. We also monitored EC behavior using time-lapse microscopy of embryonic explants and isolated primary ECs. Our work, with its comprehensive in vivo and in vitro evidence, demonstrates that in ECs is required for vascular maintenance and that its deletion severely compromised EC survival and spreading, leading to vascular regression. To delete in ECs, we generated vascular-specific Cre mice using the promoter/enhancer (). To screen and characterize the mice, we used either Z/AP or Rosa26R reporter mice, in which AP or β-galactosidase activity, respectively, is activated only after Cre-mediated recombination (; ). From nine founder lines, we selected one that efficiently mediated gene excision. Cre was active at e7.5 in the blood islands of the extraembryonic mesoderm ( A), the earliest vascular cells, and in the dorsal aortae at e8.5 ( B). At e11.5, Cre was active in all vessels examined, including the vitelline vessels ( C). Histological evaluation revealed that Cre was active in ECs along with some blood cells ( D). These results demonstrated that Cre-mediated gene excision occurred specifically in early vascular progenitor cells, differentiated ECs, and some hematopoietic cells. . Littermates missing any of the required alleles were used as controls throughout the study. allele, two loxP sites flank the exon encoding the second kinase domain, which results in ablation of FAK protein expression but does not affect the expression of FAK-related nonkinase (FRNK; ). allele and the absence of FAK expression in ECs, we stained cells isolated from embryos because in situ identification of -null ECs was obscured by adjacent perivascular cells still expressing FAK. We performed double staining with anti–P-Y397FAK () and anti-CD31 (an EC marker; ). FAK was present in the focal adhesion contacts of both CD31 and CD31 cells ( E). Although non-ECs from a mutant still expressed FAK, most CD31 cells from the mutant did not express FAK ( F). To assess the time course of the –mediated FAK depletion, we quantitatively analyzed endothelial FAK expression at different gestational stages. FAK was absent in ∼70% of ECs at e8.5, ∼90% at e9.5, and ∼95% at e10.5 ( G). began before e8.5 and resulted in the nearly complete depletion of FAK protein in CD31 cells by e10.5. To verify the loss of FAK expression, we performed double immunostaining with antibodies (Abs) specific to the COOH terminus of FAK and CD31 in purified ECs from e9.5 embryos and found similar FAK deletion (′–G). Furthermore, using this Ab, we examined lysates of e9.5 embryos by Western blotting. Despite the presence of other cell types, we detected a significant reduction in FAK expression in the mutant ( H), further validating efficient deletion. Expression of FRNK was low, undetectable by Western blotting (unpublished data), and was unchanged by immunoprecipitation (IP; H). In contrast, the expression of Pyk2 (proline-rich tyrosine kinase 2), a FAK-related kinase, was unchanged in embryonic lysates ( H). We also performed triple immunostaining with Abs to CD31, FAK, and Pyk2 and found no obvious increase in Pyk2 signal in mutant ECs without FAK (Fig. S1, available at ). Because FAK is deleted in early development, we predicted that our mutants would die early in gestation from lethal vascular defects. However, at e9.5, no detectable gross abnormalities were observed ( and unpublished data). At e10.5, ∼10% of the mutants were dead, and the remaining live mutants were readily identified by severe vascular defects. At e11.5, no live mutant embryos were recovered (). To determine the embryonic phenotype more precisely, we examined the mutant embryos at e10.5. The mutant YSs lacked the blood-filled vascular tree typically observed in the controls. Instead, hemorrhage was apparent in both the amniotic and YS cavities ( A). Although major structures were present in the embryo proper, the mutants were slightly smaller and showed patches of sequestered blood (reflecting dilated vessels) primarily in the upper trunk and head regions ( B). To examine any endothelial abnormalities, we performed whole-mount immunostaining using anti-CD31 and found abundant CD31 ECs organized in severely defective vascular structures. The mutant YSs lacked the hierarchical vitelline vascular pattern seen in the controls ( C), and only remnants of major vessels were observed ( D). The mutant capillary plexuses lacked the intricate network structure seen in the controls; instead, the microvessels were irregularly shaped, frequently dilated, and flattened with a sheetlike appearance and thin, spiky connections ( D and not depicted). We observed a distinct internal carotid artery that was well connected to a homogenous network of head capillary plexuses in the control ( E) but not in the mutants. Instead, as in the mutant YS, vessels appeared flat and fused to the surrounding widened and sinusoidal capillaries, leading to a great variation in the size of the capillaries and intercapillary spaces ( F). Because the NE lacks mesenchymal cells, its vascularization is dependent entirely on the invasion of vessel branches generated from preexisting vessels in the surrounding tissues by sprouting angiogenesis (). We observed a complete absence of blood vessels in the NE at e10.5 by CD31 staining, indicating that the -null mutants were defective in sprouting angiogenesis into the NE (). Because major vessels were severely defective at e10.5, we questioned whether they had never developed or had degenerated. To distinguish between these possibilities, we analyzed the vasculature at e9.5 by CD31 and smooth muscle α-actin (SMαA) double immunostaining. At e9.5, the mutant embryos were morphologically indistinguishable from the controls when viewed under a dissecting microscope (unpublished data). The appearance of mutant YS vasculature was surprisingly similar to that of controls, with comparable branching patterns and recruitment of smooth muscle cells, except that the vitelline arteries were slightly less elaborate (). Staining for SMαA revealed that smooth muscle cells were recruited to vitelline arteries in both the controls and mutants (). In the e9.5 embryo proper, CD31 and SMαA double staining revealed no apparent differences between the head vasculature of the mutants and the controls (unpublished data). In addition, the major embryonic vessels (including the dorsal aortae, aortic arches, and cardinal veins) and cardiac chambers in the mutants had developed comparably (). To further verify these findings and to examine vessels that were not visible on the surface, we analyzed 100-μm–thick cross sections of CD31-stained embryos (), finding again that the dorsal aortae and cardinal veins were comparable. However, we noticed fewer small vessels in the mutant cross section ( H), suggesting that vascular defects had begun to develop. Given that is also expressed in hematopoietic cells in mice, we looked for defects in blood cells. Upon gross examination, blood was present in e9.5 mutant embryos and YSs, similar to the controls. Furthermore, we isolated e9.5 circulating hematopoietic cells and found no significant difference in the number of cells isolated from the controls and mutants (unpublished data). This result suggests that no obvious hematopoietic abnormalities had developed at this stage. Altogether, these phenotypic findings at e9.5 indicate that subtle vascular defects had begun to develop and that they are primary effects of the loss of FAK in ECs. To identify the cellular defects during vascular development, we focused on the highly reproducible irregular microvessel sizes seen in the e10.5 mutant YSs. To investigate this phenotype further and to avoid any secondary effects that might have contributed to it, we examined the mutants at e9.5, before other detectable abnormalities. High magnification microangiographs of mutant YSs showed wider capillary diameters, fewer small intercapillary spaces, and larger intercapillary areas that were often associated with incomplete vascular sprouts (). To confirm these observations quantitatively, we analyzed the YS microangiographs morphometrically. The mutant YSs had a 10-μm (25%) increase in mutant capillary width that was accompanied by a reduction in the number of intercapillary areas smaller than 200 μm and an increased number of intercapillary spaces that were larger than 2,000 μm in diameter (). Branch points representing network complexity were significantly reduced in the mutants ( I). To separate any influence of the heart, blood, or other organs that might contribute to the vascular phenotype, we analyzed capillary morphogenesis in embryonic explants ex vivo. Microangiography of allantoic explants () and para-aortic splanchnopleural mesoderm (P-Sp) explants () revealed vascular phenotypes similar to those in the YSs. Dilation of capillaries and intercapillary spaces was observed along with a reduction in network complexity. A quantitative assessment revealed an approximate threefold increase in capillary width, which is a significant reduction in intercapillary spaces <200 μm, an increase in intercapillary spaces >2,000 μm, and a reduction in the number of branch points (, J–L). Because the explant phenotypes resembled the in vivo capillary defects, they likely represent the primary effects of deletion in ECs. To identify defects in vascular development, we performed time-lapse microscopy on P-Sp explants carrying an additional allele in which GFP is driven by an EC-specific promoter (). Because the P-Sp explants were taken from e8.5–9.5 embryos, most ECs had lost FAK. To ensure that Cre was active, we used the Rosa26R reporter and confirmed that explant vascular networks were positive for Cre activity (unpublished data). The most striking differences involved vessel contraction. The control explants (; and Video 1, available at ) retained a high level of network integrity. In contrast, the mutant explants (; and Video 2) underwent a process in which ECs contracted and clustered with each other, resulting in an irregular network composed of wider and smaller vessels, which is reminiscent of the e10.5 YS microvessels. In addition, we analyzed endothelial sprouts and networks in 286 videos and found apparent vessel deterioration: 51.8% of mutant vessels regressed versus 30.1% of control sprouts, and 51.1% of mutant networks regressed versus 18.8% of control networks (). Furthermore, sprout and network growth was reduced in the mutants (35.1 vs. 56.3% and 23.9 vs. 45.9%, respectively; ). Together, these results suggest that ECs lacking FAK lead to reduced vessel growth and increased vessel regression. To determine the cellular mechanism of the vascular defects, we analyzed cell survival, proliferation, and migration. We monitored 4,240 control and 3,895 mutant ECs in P-Sp explants over 8 h by time-lapse microscopy. The percentage of dying cells in the mutants was more than double that in controls (5.8 vs. 2.4%; A). To examine whether there was a differential effect in sprout tip versus nontip cells, we monitored 78 control and 108 mutant tip cells. Again, the overall percentage of dying cells in the mutants was more than double that of controls (21.8 vs. 8.6%). To verify EC apoptosis in vivo, we performed TUNEL along with CD31 staining on YS sections, which contain fewer cell types and allow for a clearer identification of ECs. We detected apoptotic ECs in both controls and mutants ( B), but quantitative analysis revealed a twofold increase in apoptotic ECs in the mutants ( C). We did not detect any significant changes in EC proliferation in the explants ( A). To confirm this result, we quantified EC proliferation in vivo by injecting BrdU into pregnant mice. Cells were subsequently isolated from e9.5 embryos. The number of BrdU CD31 cells was slightly, but insignificantly, reduced in the mutant, whereas the number of BrdU CD31 cells remained the same ( B). Therefore, the loss of did not provoke significant change in EC proliferation but reduced EC survival, which is likely responsible for the reduced vascular growth and increased vascular regression. To investigate EC migration in the context of vessel morphogenesis, we examined ECs lining the vessels in P-Sp explants. We monitored the migration path of ECs over a 300-min period at 30-min intervals. From 13 control and 16 mutant ECs analyzed, six representative pairs are shown in A, demonstrating no obvious difference. We quantified the velocity (distance/interval) of 33 control and 55 mutant ECs that were analyzed over a period of 8 h and found an average velocity of 10.7 ± 8.7 μm versus 12.2 ± 11.56 μm, respectively, indicating no major difference in the migration distance of mutant ECs within a vessel and a subtle, insignificant increase in velocity. Because previous reports of cell migration of FAK-deficient cells were analyzed in isolated cells and not in an organ context (; ), we examined ECs that did not belong to a vessel structure but were scattered in the P-Sp explants. The migration paths of six representative pairs from 26 control and 34 mutant ECs analyzed at 28-min intervals over a 280-min period are shown in B, demonstrating an increase in cell migration of the mutant ECs. Consistent with this result, the average velocity of 10 control and 15 mutant ECs analyzed over an 8-h period was 7.6 ± 6.6 μm versus 16.7 ± 12.8 μm, respectively. We also monitored (at 10-min intervals over 8 h) the migration of isolated embryonic ECs, which were identified by DiI-Ac-LDL uptake or Tie1-GFP, grown on fibronectin (FN). Control ECs exhibited extensive membrane ruffling and lamellipodia formation at the migrating front ( A and Video 3, available at ). In contrast, mutant ECs were poorly spread and lacked membrane ruffling and lamellipodia formation ( B and Video 4). Spiky, thin cell protrusions extended at the cell periphery in a random fashion, compromising cell polarity. Neither the mutant nor the control clusters of ECs migrated a significant distance (unpublished data). The single mutant cells were actively moving, although they lacked the stable directional migration seen in the controls. To quantify cell migration, we tracked the cell center and found that single mutant ECs traveled ∼50% longer distances compared with the controls (unpublished data). These results demonstrate that FAK-deficient single ECs exhibit a defective, but faster, locomotion on FN in the presence of serum and growth supplements. To further identify the cellular defects resulting from FAK depletion in ECs, we examined the actin cytoskeletal structure. Phalloidin staining revealed typical actin stress fibers throughout the control ECs ( C). FAK-deficient ECs showed fewer and abnormal stress fibers that instead resembled cortical actin bundles ( D). fibroblasts exhibit an increased number of focal adhesion contacts (), we stained our mutant ECs for paxillin, which is a component of focal adhesion contacts and an anchor for actin filaments. In contrast to the abundant, well-spread focal adhesions throughout the control cells, fewer focal adhesions that were peripherally confined (even aggregating together) were detected in the mutants (). These results suggest that the loss of leads to abnormal actin cytoskeleton structure and focal adhesion organization in ECs. To determine the kinetics of cell spreading, we examined cells on FN and found a marked reduction in cell spreading at both 2 and 20 h after plating ( G). To ascertain whether this was a matrix-specific result, we plated cells on laminin (LM)-coated and poly--lysine (PLL)–coated plates and found similar defects in cell spreading (). Therefore, the loss of FAK compromised EC spreading through both ECM-dependent and independent mechanisms. In summary, these results suggest that the loss of leads to abnormal cell spreading without decreasing cell migration in ECs. a s c e r t a i n t h e c e l l u l a r a n d m o l e c u l a r m e c h a n i s m o f v a s c u l a r m o r p h o g e n e s i s , w e d e l e t e d F A K s p e c i f i c a l l y i n E C s . T h i s g e n e t i c a l t e r a t i o n l e a d s t o v a s c u l a r d e t e r i o r a t i o n a n d e m b r y o n i c l e t h a l i t y b e f o r e e 1 1 . 5 . M u t a n t E C s e x h i b i t e d r e d u c e d c e l l s p r e a d i n g a n d s u r v i v a l , a b e r r a n t ( b u t n o t r e d u c e d ) m i g r a t i o n , a n d n o r e d u c t i o n i n p r o l i f e r a t i o n . T h i s w o r k d e m o n s t r a t e s t h a t F A K p l a y s a c e l l - a u t o n o m o u s f u n c t i o n i n E C s t h a t i s e s s e n t i a l f o r v a s c u l a r f o r m a t i o n a n d m a i n t e n a n c e , d e l i n e a t i n g c e l l s u r v i v a l a s a m a j o r E C e f f e c t o f F A K i n v i v o . Generation of the transgenic (Tg) mice followed a previously described procedure (). Founder mice were identified by PCR genotyping, and the line was maintained in an FVB/N background. and mice, which were maintained in a mixed background and genotyped as described previously (; ). The mice have been previously described (). All animals were treated in accordance with the guidelines of the University of California San Francisco (UCSF) Institutional Animal Care and Use Committee. LacZ staining was performed as previously described (; ). For AP staining, samples were fixed in 4% PFA overnight (o/n) and stained with BM purple according to the manufacturers instructions (Roche). Rat anti–mouse CD31, mouse anti-Pyk2, and mouse anti-Y397 phospho-FAK were obtained from BD Biosciences. Rabbit anti-FAK was purchased from Upstate Biotechnology. All secondary Abs were obtained from Jackson ImmunoResearch Laboratories. Cy3-conjugated mouse anti–mouse SMαA and rabbit anti-FN were purchased from Sigma Aldrich. Mouse anti-BrdU was obtained from the Developmental Studies Hybridoma Bank. Mouse anti–mouse paxillin was obtained from Zymed Laboratories, and AlexaFluor488-conjugated phalloidin was purchased from Invitrogen. Embryos and/or YSs from e8.5–10.5 were isolated into ice-cold PBS containing 1% FBS and 100 μg/ml penicillin and streptomycin and were treated with 200 U/ml collagenase III (Worthington Biochemical Corp.) in PBS for 15 min at 37°C. 10 μg/ml DNaseI (Sigma Aldrich) was added for the last 5 min. Cells were pelleted and plated on plates or glass coverslips coated with 10 μg/ml FN or 5 μg/ml LM for 12 h at 4°C or 0.1 mg/ml of 300-kD PLL hydrobromide for 5 min at RT and blocked with 0.2% BSA for 30 min. To purify ECs, cell pellets were resuspended in buffer with 2 μg/ml CD31 and incubated for 30 min followed by binding to Dynabeads M450 (Dynal) for an additional 30 min at 4°C with rocking. Cells were washed with buffer (0.1% BSA and 2 mM EDTA in PBS lacking Ca/Mg, pH 7.4). ECs were separated using a magnet and were plated on plates coated with 10 μg/ml FN. Cells were cultured in F-12 medium supplemented with 15% FBS, 0.1 μg/ml EC growth supplement (Collaborative Biomedical Products), and 100 μg/ml penicillin and streptomycin. Hematopoietic cells were dissociated by gentle mechanical disruption of e9.5 YSs, passed through a 45-μm strainer, and counted. ECs isolated from e9.5 YSs were identified after a 1–3-h incubation with 2.5 μg/ml DiI-Ac-LDL (Biomedical Technologies). Cells were viewed with a time-lapse imaging system (Intelligent Imaging Innovations). ECs were maintained in a humidified 5% CO mixture at 37°C. EC migration time-lapse recording frames were taken every 5–10 min continuously for up to 24 h. Tissues were fixed in 4% PFA o/n, washed in PBS, and incubated in blocking solution (2% BSA plus 0.1% Triton X-100 in PBS) with primary Abs in blocking solution. After washing with blocking solution, they were incubated with the secondary Ab in blocking solution. All incubations were at 4°C for 12 h with gentle rocking. Five mutant and five littermate whole-mount control YSs were stained for CD31. Capillary images were taken in areas devoid of large vessels. 50 capillary diameters and intercapillary spaces were measured for each sample. 12 control and 4 mutant P-Sp explants were analyzed using the entire explant area after CD31 staining. Measurements were made with the ruler and pencil tools of Slidebook software (Intelligent Imaging Innovations). Diameters were measured from edge to edge at the point in the vessel located equidistant from its adjacent branches. Histology and immunohistochemistry were performed as previously described () using the Abs listed above. Rehydrated paraffin sections were blocked with 5% donkey serum in PBS for 2 h at RT and incubated with the primary Ab at 4°C o/n followed by secondary Ab incubation for 1 h at 4°C. After being washed with PBS, samples were mounted with Vectashield containing DAPI (Vector Laboratories). Cells grown on coverslips were fixed in 4% PFA/PBS for 20 min and permeabilized in 0.1% Triton X-100/PBS for 3–10 min at RT before blocking in 1% BSA for 1 h, after which the same staining procedure was used. To examine apoptotic ECs, e9.5 YSs were fixed in 4% PFA o/n, embedded in optimal cutting temperature, and frozen sectioned at 8 or 10 μm. Apoptosis of ECs was detected by double immunofluorescent staining for CD31 and subsequent TUNEL using a Fluorescein In Situ Apoptosis Detection Kit (Intergen Company). CD31 ECs were counted, and the ratios of TUNEL CD31 to total CD31 ECs were obtained. Statistical analysis was performed using the test. Allantoic explants were cultured according to a previously published protocol () except that 50% FCS was used instead of 50% rat serum. Allantoises were isolated at e8.0 at the four to six–somite stage and cultured for 54 h on FN-coated dishes. Explants were stained with anti-CD31 as described in Immunofluorescent staining of sections and isolated cells. For P-Sp explants, the P-Sp region of an embryo was dissected and placed into a 12well dish on top of a confluent OP9 stromal cell layer according to methods described previously (). Explants were cultured in RPMI 1640 medium supplemented with 10% FBS, which was changed daily. Tg ECs were imaged as described above (in EC spreading and migration analysis) every 5–10 min for up to 5 d. For EC migration analyses, the XY coordinates of single ECs and capillary ECs were tracked. Western blotting and IP was performed as previously described () using a polyclonal anti-FAK Ab or a polyclonal anti-Pyk2 Ab. IP was performed to detect FRNK using a monoclonal anti-FAK Ab (clone 2A7) that targets the COOH terminus of FAK and FRNK. EC and capillary structure dynamics were monitored in Tg control (Video 1) and mutant (Video 2) P-Sp explants. Chemokinetics of isolated e9.5 control (Video 3) and mutant (Video 4) YS ECs 18 h after plating on FN is also shown. There was no obvious induction of Pyk2 expression in FAK-null ECs (Fig. S1). Online supplemental material is available at .

For a decade, it had been known that the apical and basolateral membranes of epithelial cells had different lipid compositions (), and specifically that glycolipids are enriched apically. In 1981, the tight junction was proposed as the barrier that kept these two membrane populations distinct (). Playing off a finding that different viruses budded from the different poles of cultured epithelial cells (), van Meer and Simons showed in 1982 that the envelopes of those viruses contained different lipid compositions (). At about the same time, others had shown that heterogeneous lipid domains existed (Karnovsky et al., 1982) and that glycosphingolipids clustered () in both model membrane systems and biological membranes. This suggested that lipid domains might affect membrane functions and structure, but real evidence of a biological role was lacking. Because plasma membrane lipids are synthesized intracellularly, van Meer and Simons reasoned that lipid sorting must take place to set up the epithelial cell membrane domains. The NBD–ceramide probe offered a handy way to start on the project because, once it was inside cells, it would be converted into two distinct lipid probes: an NBD–sphingomyelin and an NBD–glucosylceramide, analogues of a basolateral and apical lipid, respectively. The conversion occurred in the Golgi and then the fluorescent probes could be followed to plasma membrane destinations. While he was a post-doc with Simons at the EMBL in Heidelberg, Germany, van Meer quantified the sorting using the NBD-labeled probes. Using Madin-Darby canine kidney (MDCK) epithelial cells grown on filters, he used “back exchange” with serum albumin applied to either side of the filter to extract and measure the fluorescent lipids that sorted to either the apical or basolateral side (). He found that the NBD–glucosylceramide was enriched two to four times on the apical membrane, whereas the NBD–sphingomyelin was equally distributed between the apical and basolateral sides. This process quantitatively accounted for the in vivo lipid distribution. “This was the first piece of evidence that we were on the right track,” says Simons. He notes that, until this point, lipid microdomains had been reported as biophysical phenomena, but the cell had not previously been caught in the process of actively setting up these differences. From this paper, a model emerged that would be the first tip of a lipid microdomain, or raft, iceberg. The paper provided data that lipids are potentially sorted in the Golgi complex. Based on the physical properties of glycolipids, which suggested they could associate with each other, “the apical sorting platform idea took form,” says Simons. The lipids wound up on the exoplasmic leaflet of the plasma membrane where they could not diffuse past tight junctions, so it seemed logical that they might be synthesized on the lumenal leaflet of the Golgi and transported to the plasma membrane via vesicles. It was at this point that the authors touched ever-so-briefly on the topic of lipid subdomains. Lipid “sorting,” they stated, “must involve the lateral segregation in this leaflet of lipids into those areas of the membrane that will bud to form transport vesicles destined for either the apical or basolateral plasma membrane domain. The factors involved in this segregation process are unknown.” The idea that lipid microdomains might exist within a continuous “fluid” membrane was radical then and continues to be controversial now. The following year, Michael Lisanti, Enrique Rodriguez-Boulan, and colleagues showed that six glycosyl-phosphatidylinositol (GPI)–anchored proteins followed the same apical distribution pattern as the NBD–glycolipid (). But it wasn't until 1992, van Meer says, that a “really crucial paper sent the thing off.” Deborah Brown and Jack Rose discovered that cold detergent extraction allowed the isolation of membranes enriched in glycosphingolipids and GPI-anchored proteins (). These detergent-resistant membranes (DRMs) literally floated like rafts to the top of the preparations, and the simple technique “opened the field up to experimental analysis,” says van Meer. A few years later, Brown and colleagues proposed that the structure of the DRMs was due to a separate, less fluid membrane phase regulated by tight packing of fatty acyl chains and cholesterol (Schroeder et al., 1994). Many later studies on DRMs showed that raft-associated proteins were important for cell signaling and that movement of those proteins into the DRM fraction was associated with signal activation (for review see ). But the harsh in vitro techniques of isolating DRMs and depleting membranes of cholesterol have brought much criticism to the field about artifactual or at least, nonphysiological, results (). Pinning down what rafts look like and how they act in real time in vivo has been slippery. Most researchers agree that, if rafts exist, they are extremely small (25–100 nm) and transient. The technology for observing lipids in unperturbed, living cells has yet to catch up. Ken Jacobson says “the raft concept looks good from our work on model membranes” showing that raft domains depend on cholesterol density and that GPI-anchored proteins partition into rafts (Dietrich et al., 2001). “But what is, if any, the in vivo correlate?,” he asks. “The membrane is a liquid structure and we are still learning how to derive structural information. There has to be lateral heterogeneity, but figuring out how to really prove that in a compelling way still remains the challenge.” Brown suggests that raft formation might be regulated so that “the membrane composition is poised at the brink of raft formation and you need to flip a switch.” These stabilized rafts almost certainly function in membrane trafficking, virus budding, and signal transduction.

The αIIbβ3 integrin is expressed on platelets and platelet precursors, megakaryocytes. Integrin αIIbβ3, when in a resting state, does not bind plasma fibrinogen. However, upon platelet stimulation by agonists such as thrombin, intracellular signals are generated that change the conformation of αIIbβ3 to an active state via “inside-out” signaling (for review see ). Activated αIIbβ3 is competent to bind soluble ligands, such as fibrinogen or von Willebrand factor, which link platelets together in aggregates. Although it is known that activation of αIIbβ3 requires the integrin cytoplasmic tails (; ; ), the role of the αIIb tail in this process is not well understood. Previously, we identified calcium and integrin binding protein 1 (CIB1; also known as CIB [] and calmyrin []), which binds to the integrin αIIb cytoplasmic tail. CIB1 is an EF-hand–containing, calcium binding protein that interacts with hydrophobic residues within the membrane-proximal region of the αIIb cytoplasmic tail (; ; ; ). Although CIB1 is expressed in a variety of tissues including platelets, its potential interaction with other integrin α or β subunits to date has not been reported (; ; ). However, CIB1 also interacts with several protein kinases, such as p21-activated kinase 1 (PAK1; ) and FAK (). Because CIB1 is one of a few proteins known to bind directly to the αIIb cytoplasmic tail, we hypothesized that CIB1 may modulate platelet αIIbβ3 activation. To determine whether CIB1 affects αIIbβ3 activation, we used differentiated megakaryocytes from murine bone marrow because megakaryocytes, unlike platelets, are amenable to direct genetic manipulation. However, like platelets but unlike many cell lines, mature megakaryocytes express αIIbβ3 and activate this integrin in response to agonists (; ; ), making them a suitable model system for studying platelet integrin regulation. We provide evidence that CIB1 is an inhibitor of agonist-induced αIIbβ3 activation, most likely via competition with talin binding to αIIbβ3. CIB1 has been shown to interact with the αIIb cytoplasmic tail by multiple approaches (; ; ; ) with an affinity of ∼0.3 μM (). We find that endogenous CIB1 coimmunoprecipitates with αIIbβ3 from both resting and agonist-activated platelets, with an increased apparent association in activated platelets ( ), in agreement with the purified protein studies of . However, the role of CIB1 in regulating αIIbβ3 function has been unclear. To address the role of CIB1 in αIIbβ3 activation, a well-characterized megakaryocyte model system (; ; ) was used. Stimulation of mature murine megakaryocytes with protease-activated receptor 4 activating peptide (PAR4P) significantly increased fibrinogen binding over basal levels to unstimulated megakaryocytes (agonist-induced binding is shown as percent over basal binding, which was subtracted from total binding). The PAR4P-induced fibrinogen binding was completely blocked by an anti-αIIbβ3 function-blocking mAb, 1B5 (Fig. S1 A, available at ), in agreement with , further confirming the use of fibrinogen binding as a specific marker of αIIbβ3 activation in megakaryocytes. Fibrinogen binding to unstimulated megakaryocytes was not affected by either the 1B5 mAb or by divalent cation chelation with EDTA (Fig. S1 A), indicating no basal αIIbβ3 activation. We then asked whether CIB1 affects agonist-induced αIIbβ3 activation. Megakaryocytes overexpressing either EGFP or CIB1-EGFP were stimulated with a PAR4P, followed by three-color flow cytometric analysis to gate on large, live cells expressing GFP fluorescence (see Materials and methods). Protein overexpression was confirmed by Western blotting (, inset) and by fluorescence microscopy (Fig. S1 E). We found that CIB1-EGFP completely inhibited agonist-induced fibrinogen binding compared with either EGFP alone or untransduced megakaryocytes () but did not inhibit fibrinogen binding to megakaryocytes exposed to 1 mM MnCl (Fig. S1 C), which directly activates αIIbβ3 independent of agonist-induced, inside-out signaling. These data suggest that CIB1 negatively regulates agonist-induced αIIbβ3 activation. In addition to binding the αIIb tail, CIB1 also interacts with the serine/threonine kinase PAK1 (; Fig. S2 A, available at ). Because platelets () and megakaryocytes () express PAK1, we asked whether CIB1 inhibits αIIbβ3 activation via a direct interaction with αIIb or indirectly via PAK1. We therefore overexpressed a CIB1 mutant (CIB1 F173A-EGFP) that does not bind αIIb () but retains binding activity to PAK1 (Fig. S2 A). Previous analysis of this mutant by circular dichroism indicated minimal change of CIB1 structure (), and yeast two-hybrid analysis confirmed that the mutant does not bind mouse αIIb (Fig. S2 B). Although the level of CIB1 F173A overexpression relative to endogenous CIB1 and percent of cells transduced was comparable to that of wild-type CIB1 ( [inset] and Fig. S1 E), the CIB1 F173A mutant was unable to suppress PAR4P-induced αIIbβ3 activation (). In addition, expression levels of αIIbβ3 and basal fibrinogen binding to unstimulated megakaryocytes were comparable in megakaryocytes expressing CIB1 F173A-EGFP versus CIB1-EGFP (Fig. S1, B and D). These data suggest that a direct interaction between CIB1 and the αIIb tail is critical for suppression of αIIbβ3 activation. To further determine whether CIB1 suppresses integrin activation by a direct or indirect mechanism, we tested its ability to suppress activation of αV integrins because we previously determined that CIB1 does not interact with the αV cytoplasmic tail (; ) and because megakaryocytes express the αV integrin subunit (Fig. S3 A, available at ). We found that neither CIB1-EGFP nor CIB1 F173A-EGFP had an effect on αV integrin activation as detected with WOW-1, a mAb that selectively recognizes activated αVβ3 and to a lesser extent, activated αVβ5 (), compared with untransduced megakaryocytes or megakaryocytes expressing EGFP alone (Fig. S3 B). These results suggest that CIB1 selectively inhibits the activation of αIIbβ3, most likely via a direct interaction with the integrin. To determine the role of endogenous CIB1 in agonist-induced αIIbβ3 activation, we reduced CIB1 levels by RNA interference. Introduction of murine CIB1–specific small interfering RNAs (siRNAs) into megakaryocytes resulted in a consistent knockdown of endogenous CIB1 protein levels by 40–60% (). We observed a statistically significant increase in fibrinogen binding to megakaryocytes with reduced CIB1 expression, relative to cells transfected with a human CIB1 siRNA control or untransfected cells, at two different concentrations of PAR4P (). This increased fibrinogen binding was not attributable to changes in αIIb expression because flow cytometric data indicated comparable expression levels of αIIbβ3 integrin in all transfected groups (Fig. S3 C). Moreover, no enhancement of basal fibrinogen binding in the absence of agonist was observed in CIB1-depleted cells (Fig. S3 D). Thus, a consistent correlation was observed between reduced CIB1 expression and increased fibrinogen binding to agonist-stimulated megakaryocytes. These data therefore complement the CIB1 overexpression studies and indicate a negative regulatory role for CIB1 in αIIbβ3 activation. Our data showing that CIB1 is an endogenous inhibitor of αIIbβ3 activation are in apparent contradiction to a study showing that CIB1 activates αIIbβ3 (). In this study, a CIB1 peptide introduced into platelets blocked agonist-induced αIIbβ3 activation. It was proposed that this blockage occurred because the peptide displaced endogenous CIB1 from αIIb, implying that CIB1 activates αIIbβ3. However, this study did not show a direct interaction between the CIB1 peptide and αIIb. Moreover, these results may be interpreted as an ability of this peptide to bind αIIb and mimic the inhibitory function of intact CIB1. We next examined the localization of CIB1 and αIIbβ3 in resting and activated megakaryocytes by immunofluorescence. Confocal images of nonstimulated megakaryocytes showed CIB1 colocalizing with αIIb at the cell periphery ( , left and inset). Upon agonist stimulation (PAR4P) in the absence of added fibrinogen, we observed a potential increase in membrane colocalization of CIB1 with αIIb (, middle and inset) that did not reach statistical significance (). However, this trend is in agreement with our coimmunoprecipitation experiments () and . In contrast, upon agonist stimulation in the presence of soluble fibrinogen, CIB1 colocalization with αIIb decreased considerably (), as shown by distinct areas of nonoverlapping staining of CIB1 and αIIb (, right and inset), suggesting a loss of the CIB1–αIIb interaction upon ligand occupancy of αIIbβ3. These results suggest that the CIB1–αIIb interaction is dynamically and spatially regulated by agonist stimulation and ligand occupancy. In addition to the effects of CIB1 on inside-out signaling, the significant redistribution of CIB1 upon fibrinogen binding to activated αIIbβ3 suggests that CIB1 may also become available to mediate outside-in signaling events. In this regard, it has been reported that CIB1 contributes to outside-in signaling via αIIbβ3 (). To determine a molecular mechanism by which CIB1 inhibits αIIbβ3 activation, we asked whether CIB1 affects binding of the integrin-activating protein talin with both the αIIb cytoplasmic tail and the intact integrin heterodimer. Talin is a cytoskeletal protein recently shown to play a critical role in activating several integrins via interaction of the talin head domain (THD) with β cytoplasmic tails, including β3 (; ); in addition, talin also binds to the αIIb cytoplasmic tail (). Using recombinant CIB1 and THD in solid-phase binding assays, we found that both CIB1 and THD bound to immobilized αIIb cytoplasmic tail peptide in a direct, saturable manner ( ), with THD binding to immobilized αIIb peptide at a slightly higher apparent affinity. To determine relative affinities of CIB1 and THD for soluble versus immobilized αIIb tail, equimolar concentrations of soluble CIB1 and THD were incubated with increasing concentrations of soluble αIIb cytoplasmic tail peptide before addition to immobilized αIIb cytoplasmic tail peptide (). Although CIB1 binding was significantly inhibited at concentrations of <1 μg of soluble peptide per well, no inhibition of THD binding to immobilized αIIb cytoplasmic tail peptide was observed at these concentrations, suggesting that CIB1 has a higher relative affinity than THD for soluble αIIb. These results also suggest that THD has a higher relative affinity than CIB1 for the immobilized αIIb tail peptide. Furthermore, competitive binding assays showed that increasing concentrations of soluble CIB1 almost completely inhibited THD binding to immobilized αIIb cytoplasmic tail peptide (). Similarly, soluble THD inhibited CIB1 binding to immobilized αIIb cytoplasmic tail peptide (). Consistent with THD having a higher affinity for immobilized αIIb cytoplasmic tail peptide, ∼25 nM THD inhibited 50% of CIB1 binding to αIIb cytoplasmic tail peptide, whereas this concentration of CIB1 had little effect on THD binding to αIIb cytoplasmic tail peptide. We then asked whether CIB1 interferes with THD binding to purified, activated αIIbβ3 heterodimer; CIB1 maximally inhibited ∼50% of the binding of solution phase αIIbβ3 to immobilized THD () because higher concentrations of CIB1 had no further inhibitory effect (unpublished data). These data are consistent with a study showing that an anti–αIIb tail antibody also maximally inhibits ∼50% of αIIbβ3 binding to immobilized talin, with the remaining binding attributed to the β3 tail (). Moreover, the mutant protein CIB1 F173A, which does not bind the αIIb tail, did not inhibit αIIbβ3 binding to immobilized THD (). These results suggest that CIB1 inhibits αIIbβ3 activation at least in part by competing with talin for direct binding to the αIIb cytoplasmic tail in platelets and megakaryocytes. These data may also indicate that CIB1 prevents a functional engagement of talin with αIIbβ3 via a CIB1-associated conformational change in αIIb that directly affects β3 so that bound talin cannot activate αIIbβ3. The finding that CIB1 overexpression completely inhibits αIIbβ3 activation but only partially inhibits talin interaction with αIIbβ3 also raises the possibility that talin binding to β3 alone is insufficient to activate αIIbβ3. Our data demonstrate that CIB1 is an endogenous inhibitor of agonist-induced αIIbβ3 activation. We propose that in the resting state, CIB1 is associated with a portion of αIIbβ3 molecules ( ). Agonist stimulation promotes talin association with the majority of integrin cytoplasmic tails (; ; ; ), resulting in αIIbβ3 activation and fibrinogen binding (). However, we also predict that during agonist-induced activation, talin cannot bind to αIIb and/or properly engage β3 within the CIB1-associated αIIbβ3 molecules (). The numbers of CIB1-associated complexes may increase during agonist-induced activation based on our coimmunoprecipitation data () and the observed trend toward increased colocalization (), thus implicating a role for CIB1 during inside-out signaling events that regulate integrin αIIbβ3 activation. Consequently, this portion of CIB1-occupied αIIbβ3 would be unable to undergo agonist-induced activation and fibrinogen binding (). Furthermore, the redistribution upon soluble fibrinogen binding () and decreased relative colocalization of CIB1 with αIIbβ3 suggests that CIB1 may also be regulated by and participate in outside-in signaling via αIIbβ3. Because a decrease in endogenous CIB1 levels does not induce spontaneous integrin activation (Fig. S3 D), our model further predicts that CIB1 exerts its effect on αIIbβ3 during agonist stimulation to limit the extent of activation, as opposed to maintaining αIIbβ3 in a resting state in unstimulated cells, which may instead be regulated by properties intrinsic to the integrin (). In conclusion, our results indicate that CIB1 is a negative regulator of agonist-induced αIIbβ3 activation, thus providing a mechanism for the precise control of αIIbβ3 activation in megakaryocytes. Although megakaryocytes are not platelets, they are platelet precursors and share many similarities, suggesting that the function of CIB1 extends to platelets. It will be of interest in future studies to determine whether endogenous CIB1 levels in platelets correlate inversely with platelet reactivity, a known risk factor for coronary artery disease (). The Sindbis expression system was obtained from Invitrogen. The human CIB1 gene or a mutant human CIB1 gene (F173A; ) was fused to the EGFP gene at the COOH terminus of CIB1 and cloned into the pSinRep5 vector. The virus was produced in BHK cells for megakaryocyte transduction. Megakaryocytes were derived from bone marrow cultures of C57BL/6J mice, and flow cytometry was performed as described previously (). Differentiated megakaryocytes were transduced with various viral constructs for 20 h and collected in modified Tyrode's buffer with 1 mM CaCl and 1 mM MgCl () at 10 cells/ml. Overexpression level of CIB1-EGFP and CIB1 F173A-EGFP were quantified via densitometry using the software Quantity One (Fluor-S Multimager; Bio-Rad Laboratories) and adjusted as fold over endogenous CIB1 expression. 50 μl of the megakaryocyte suspension was mixed with agonist PAR4P (GYPGKF) and soluble Alexa Fluor 546–conjugated fibrinogen (15 μg/ml final concentration; Invitrogen) at RT for 30 min and diluted with chilled Tyrode's buffer containing propidium iodide at a final concentration of 1 μg/ml. Cells were immediately analyzed on a flow cytometer (FACStar Plus; Becton Dickinson). Live, EGFP-positive megakaryocytes were measured for Alexa Fluor 546 fibrinogen binding in the FL2 channel. Data were collected as mean fluorescence intensities using Summit software (DakoCytomation). Basal fibrinogen binding is defined as the mean fluorescence intensity of megakaryocytes with Alexa Fluor 546 fibrinogen but without agonist stimulation. The test was used in statistical analyses in all experiments. Murine CIB1–specific siRNAs were generated with the Silencer siRNA construction kit (Ambion; ). Two 21-base sequences were developed that target sites 5′-AAGGAGCGAAUCUGCAUGGUC-3′ and 5′-AAGCAGCUGAUUGACAAUAUC-3′ of the mRNA transcript as well as a control human CIB1–specific siRNA (5′- AAGUG- CCCUUCGAGCAGAUUC-3′), which has no homology to murine CIB1 or to any sequence in the mouse genome. Megakaryocytes were transduced with siRNAs at 200 nM (according to the manufacturer's protocol; Mirus), incubated at 37°C for 48 h, and subjected to flow cytometry and Western blotting. Microtiter wells (Immulon 2 HB; Dynex Technologies) were coated with and without 5 μg/well of full-length human αIIb cytoplasmic tail peptide (LVLAMWKVGFFKRNRPPLEEDDEEGQ) or 2.5 μg/well of purified THD and blocked with 3% BSA. CIB1 or THD (50 μl) were incubated for 1 h, and binding was detected with a chicken anti-CIB1 polyclonal antibody (pAb) or mouse anti-talin (clone 8d4; Sigma-Aldrich). For competition binding assays, 10 nM of soluble CIB1 or THD was incubated with increasing concentrations of soluble αIIb cytoplasmic tail peptide, THD, or CIB1 before addition to wells containing immobilized αIIb peptide. Binding of CIB1 or THD was detected using an anti-CIB1 or -talin antibody, respectively. Binding of RGD-purified αIIbβ3 () to immobilized THD ± CIB1 or CIB1 F173A proceeded for 2–3 h. Integrin αIIbβ3 binding was detected with a mAb, 11G1, which recognizes the intracellular portion of αIIb and does not overlap with the CIB1 binding site. Cultured murine megakaryocytes on poly--lysine were stained with antibodies against CIB1 and αIIb. CIB1 localization was detected with a chick pAb, and αIIb was recognized by rabbit anti-αIIb pAb. Confocal images were captured by Fluoview software (Olympus) with a Fluoview 300 laser scanning confocal imaging system configured with a fluorescence microscope (Olympus; IX70) fitted with a Plan Apo 60× oil objective. The images were assembled () in Photoshop 7.0 (Adobe). To quantify the relative colocalization of multiple images ( = 4), we calculated the RColoc value (Pearson's correlation coefficients, per image set, for pixels above the calculated thresholds for the images). Pearson coefficient is a statistical appraisal of how well a linear equation describes the relationship between two variables for a measured function and is commonly used in image colocalization analysis. Fig. S1 characterizes the overexpression of wild-type CIB1 and mutant CIB1 F173A and the fibrinogen binding to megakaryocytes expressing these fusion proteins. Fig. S2 shows the binding characteristics of wild-type CIB1 and mutant CIB1 F173A to PAK1 and integrin αIIb. Fig. S3 shows integrin αV expression and activation and also shows effects of CIB1 depletion on integrin αIIbβ3 expression and basal activation. Online supplemental material is available at .

Regulated gene expression is an essential feature of tissue-specific differentiation involving transcription factor binding and an interplay of chromatin modifications. It has been proposed that the position of genes within the nucleus may be another important facet of the regulation of differential gene expression, where certain areas of the nucleus are repressive and others support or enhance transcription (for reviews see ; ). There are also compelling arguments against any requirement for genes to be at a specific location within the nucleus for transcription to proceed. Phosphorylated RNA polymerase II and global nascent RNA can be visualized throughout the entire volume of the nucleus, with no evidence that the periphery, apart from the nuclear rim, is a repressive environment to transcription or that the interior of the nucleus is enriched in transcription sites (; ). The whole nucleoplasm appears to be accessible to protein complexes (; ), and even mitotic chromatin has been shown to be accessible to transcription factors and chromatin proteins (). To address whether nuclear organization has a functional role, we have chosen to follow the behavior of the coregulated α- and β-globin genes in primary human erythroblasts during terminal erythroid differentiation (). Transcription from these genes is regulated in a tissue-specific manner, so that balanced amounts of protein from both genes are produced over the course of several days, as erythroblasts undergo differentiation. This balance is achieved despite the fact that the α- and β-globin genes lie on separate chromosomes and in very different chromatin contexts. The α-globin genes are subtelomeric on human chromosome (HSA)16 in a gene-rich, GC-rich domain of constitutively open chromatin () that was early replicating in all of the cell types examined (). In contrast, the β-globin genes are located on HSA11, in an AT-rich region of late-replicating, tissue-restricted genes, and become early replicating and sensitive to DNase1 digestion only in erythroblasts (; ; ). The β-globin genes associate with pericentromeric heterochromatin in cycling lymphocytes, where they are not expressed, but sit apart from heterochromatin in proerythroblasts, whereas the α-globin genes never associate with heterochromatin, irrespective of cell type or transcriptional status (K.E. ). Furthermore, the two genes occupy different regions of the nucleus; HSA16 is most frequently located in the middle and inner zone in lymphoblastoid cells, whereas HSA11 is more commonly found at the nuclear periphery (). The α-globin genes are also located outside their chromosome territory in lymphoblasts far more frequently than the β-globin genes (). The α- and β-globin genes, therefore, represent very different characteristics when silent. If positional or organizational changes are necessary for the regulated and balanced expression of these genes, then they should be apparent during the course of erythroid differentiation, when the globin genes become highly expressed. In this study, we have monitored the changes in positioning and expression of the α- and β-globin genes at successive stages of terminal erythroid differentiation. We have characterized precisely when the globin genes are transcribed during differentiation and compared these findings with their positions. When actively transcribing, the human α- and β-globin genes are frequently located very close together, but not colocalized, at splicing factor–enriched nuclear speckles. This proximity of globin genes is not observed to the same degree in the mouse. We wanted to characterize organization within the nuclei of primary differentiating erythroblasts, which fully recapitulate terminal erythroid differentiation, rather than using long-established and karyotypically abnormal cell lines. To obtain reproducible populations of erythroblasts, we induced erythroid differentiation in primary mononuclear cells and then sorted erythroblasts at different stages of differentiation based on cell surface markers. Mononuclear cells from buffy coat were cultured through a two-phase liquid culture system () that induces circulating erythroid burst-forming unit cells to expand and terminally differentiate to erythrocytes. Cells were harvested for sorting at 4, 6, 8, and 14 d after the addition of erythropoietin in the second phase of culture. Cells were sorted for CD36 (glycoprotein IV receptor; days 4 and 6 only), CD71 (transferrin receptor), and glycophorin A (GPA). The appropriate gates were selected to provide populations of early erythroid colony-forming unit (CFUe)–enriched day 4, proerythroblast day 6, intermediate day 8, and late day 14 stages of differentiation ( ). The gates chosen were based on those described by for mouse erythroblasts. The morphology of cells within each population showed restricted variation and were greatly enriched and distinct compared with unsorted cells (). CFUes and proerythroblasts were the only fractions that were difficult to distinguish based on morphology alone. Gated fractions from days 4 and 6 were plated out for culture of CFUes. This indicated that the human CFUe day 4 population is as enriched for CFUes (86%; unpublished data) as we are able to prepare from mouse erythroblasts (>80%; ). To evaluate nuclear organization in relation to transcription in these primary human cells, we needed to establish the pattern of transcription from the globin genes during the course of erythroblast differentiation. We analyzed the percentage of transcribed loci for α- and β-globin using RNA FISH in all successive fractions of differentiating erythroblasts ( = 187–372), together with lymphoblasts ( = 100) and activated T lymphocytes ( = 100; ). Transcription from both α- and β-globin genes peaked in intermediate erythroblasts, with nascent transcript signals scored for 73% of α-globin loci and 77% of β-globin loci. The percentage of signals declined in subsequent fractions, to almost nothing in late erythroblasts. For both α- and β-globin, there are significant differences in the proportion of transcribing loci between intermediate erythroblasts and all other cell populations (P < 0.02). We also undertook three-dimensional (3D) DNA FISH on cells from the same cultures of erythroblasts. We were able to detect 91% of α-globin genes by DNA FISH, whereas we only detected nascent transcript at 3% of the possible sites in late erythroblasts. This suggests that probes and antibodies used in detection are not hindered in accessing sites within very condensed nuclei and, therefore, that our scores for the percentage of RNA FISH signals observed are an accurate reflection of the percentage of transcribing sites. The human α- and β-globin gene loci are located in very different chromosomal contexts (), and we already know that they exhibit different positions in respect to their chromosome territories when silent (). We examined the location of the human globin genes in relation to their chromosome territories in human primary intermediate erythroblasts, where the globin genes are maximally expressed, and in human lymphoblasts and embryonic stem (ES) cells, where the globin genes are transcriptionally silent. We hybridized probes for the α- and β-globin genes, together with whole chromosome paints for HSA16 and HSA11, respectively, and scored the proportion of FISH signals sitting away from their territory ( ). The cells analyzed were fixed in methanol-acetic acid rather than in PFA, as this compares with previous studies involving other loci (; ). We found that 52% ( = 500) of human α-globin genes sit away from their HSA16 territory in lymphoblasts, which concurs with an earlier study (). We found a similar frequency in ES cells (42%; = 100) and also in intermediate erythroblasts where globin genes are active (45%; = 200). The human β-globin genes did not localize away from their HSA11 territories with high frequency in any of the cell types examined (5 to 13%; = 100–500). Therefore, although the human α- and β-globin genes show significantly different positioning in respect to their chromosome territories (P < 0.02), we did not find any evidence that this correlates with their transcriptional status. In addition, we know that the chromosomal contexts of the human and mouse α-globin genes are very differen The human α-globin genes are located very close (162 kb) to the HSA16 short-arm telomere, whereas in the mouse, the α-globin genes sit close to the HSA11 centromere in a region of lower gene density (). Therefore, we also looked at the location of the murine globin genes in respect to the main body of their chromosomes in anemic mouse spleen, where the globin genes are active in the majority of cells, and in murine ES cells and primary activated lymphocytes, where they are silent (). None of the three cell types, ES cells ( = 350), lymphocytes ( = 100), and spleen ( = 150), had a high frequency of α-globin sitting away from the murine chromosome (MMU)11 territory (8, 7, and 8%, respectively). In addition, the mouse β-globin genes did not often localize away from their MMU7 territories (4–5%; = 100–350). As in human nuclei, we found no correlation of transcriptional status with the positions of the murine globin genes, in respect to their chromosome territories. We did find that the α-globin genes rarely localize away from their chromosome territory in the mouse, as opposed to human cells, where they are frequently found outside their territory (P < 0.01 for human intermediates, compared with mouse spleen). In contrast, the β-globin genes, which are not subtelomeric in human or mouse (), rarely localized outside their territories in any cell type. The position of the human α-globin genes very close to the HSA16 telomere may mean that they are less frequently packaged into the main body of the chromosome. We hybridized a pool of 24 cosmids from a 2-Mb contig around the human α-globin genes and found that this region can extend to 3.5 μm, which can be half the diameter of a lymphoblast nucleus ().We hybridized four selected cosmids from the contig, together with an HSA16 paint, to lymphoblast nuclei () and found that probes for sites 874, 292, and 181 kb away from the telomere showed progressively more frequent positions outside the HSA16 territory (28, 40, and 57%, respectively; = 400 for each). This suggests that the human α-globin genes are not on a loop of chromatin protruding out of the territory, but rather are on a terminal piece of chromatin that frequently extends away from the body of the chromosome. We next examined the positioning of homologous and cotranscribed α- and β-globin genes relative to each other, during terminal erythroid differentiation. All cells were scored for whether signals were separate, physically juxtaposed, or actually colocalized with overlapping fluorescent signals. Measurements between signals scored as juxtaposed did not exceed 1 μm, with a mean distance of 0.48 μm and a standard deviation of 0.19 μm. Signals separated by >1 μm were not scored as juxtaposed. There was an average incidence of 3% colocalization of homologous α-globin signals and of α- and β-globin signals, whereas homologous β-globin FISH signals were hardly ever colocalized. A degree of random colocalization can occur between probes in interphase nuclei (), and there were no significant differences in the degree of colocalization in different cell types or in line with transcription. Therefore, all percentages given in this section represent a pool of juxtaposed and colocalized scores. We have described juxtaposed and colocalized signals as “associating”—this term does not imply a physical interaction, merely a physical proximity. First, we looked at association between the two homologous signals of the α- and β-globin genes. In nonexpressing lymphoblasts ( = 500) and lymphocytes ( = 100), the association between globin homologues in 3D FISH preparations was a little higher for α–α associations, at 10% of nuclei scored, than for β− associations, at 3 and 4%, respectively. In erythroblasts ( = 200–700), the level of β− associations remained unchanged, but α–α associations increased and peaked in intermediate erythroblasts at 33% ( = 700; ). The level of α–α associations in intermediate erythroblasts was significantly different (P < 0.005) than those scored in lymphoid and other erythroid cell populations. We also treated lymphoblasts with trichostatin A (TSA), which inhibits deacetylation and induces transcription from the α-globin genes (>10-fold increase; Garrick, D., personal communication). There was no change in the frequency of α–α associations in treated cells (8%; = 100; ), suggesting that such association is not an essential requirement for transcription to proceed. The increased association cannot be caused by changes in nuclear volume during differentiation for two reasons. First, intermediate erythroblasts have a nuclear volume similar to that of lymphoblasts (unpublished data) but very different levels of association of the α-globin genes. Second, the association levels of the β-globin genes do not change significantly in the different cell types scored. We also looked at the association of the cotranscribed α- and β-globin genes. We looked specifically at RNA FISH preparations, where we knew that all signals represented a gene in the process of transcription, and we used unsorted cells for this analysis, to minimize any potential disruption to nuclear organization. 7 d after the addition of erythropoietin, an erythroid culture contains both proerythroblasts and intermediate erythroblasts. These two stages of differentiation can be distinguished visually on the basis of size and the amount of cytoplasmic RNA FISH signal because intermediate erythroblasts accumulate globin mRNA in the cytoplasm, which can be detected with probes covering exons of the α- or β-globin genes. We scored the percentage of association of homologous loci in cells with two clear α- or β-globin transcript signals and the percentage of α- and β-globin association in cells with all four signals present. The α–α associations rose from 20.5% ( = 400) in proerythroblasts to 30% ( = 600) in intermediate erythroblasts, whereas β−β associations hardly changed, at 5% ( = 300) and 6% ( = 400), respectively. We found that α- and β-globin genes were associating in 21% ( = 100) of proerythroblast nuclei, where all four genes were active, increasing to 49% ( = 250) in intermediate erythroblasts (), where a few nuclei were also observed with three (two α and one β) globin loci associating. The increase in the number of associating globin loci mirrors the increase in the percentage of transcribing loci. The fact that the degree of association was greater in intermediate erythroblasts than in proerythroblasts in cells where all four globin loci have a nascent transcript signal (;P < 0.0025 for both α−α and α−β associations) implies that association is more likely to occur when globin transcription is most frequent, as we have demonstrated for the intermediate stage of erythroid differentiation. To confirm that this association was specific to the globin genes, we asked whether α-globin would associate with other chromosome regions. We chose the subtelomeric probe for HSA2 short arm to cohybridize, via DNA FISH, with α-globin to lymphoblasts and intermediate erythroblasts because this would occupy a more peripheral location in the nucleus, like the β-globin genes. We found no difference in association of α-globin and the 2p subtelomere in the two cell types (13%; = 100 for each). We went on to look at the association between globin genes in the mouse, given the different chromatin context of the murine α-globin genes, and found a very different picture than in human cells. By analysis of RNA FISH signals in mouse intermediate erythroblasts, we found that the percentage of α−α associations was 11% ( = 200) and of β−β associations was also 11% ( = 200). Association between α- and β-globin genes was equal to 13% of cells with all four globin loci transcribing ( = 244). Therefore, the high degree of association we have identified between active globin genes in human erythroblasts does not occur in the mouse (; P < 0.05, comparing human and mouse α−α and α−β associations). There were significantly more β−β associations scored in mouse than in human erythroid cells (P < 0.02). After identifying the association between globin gene loci, we explored whether this was occurring within any specific nuclear subcompartment. Proteins recognizing different nuclear bodies and globin gene DNA were detected by immuno-FISH to different cell types, and the number of gene signals directly contacting or sitting away from the nuclear body were recorded ( = 100–300 cells for each hybridization). We found no indication that the globin genes were associating in erythroid cells at PML bodies, detected with PML antibody, or at nucleoli, detected with upstream binding factor antibody or a nucleolar-organizer region DNA probe (unpublished data). We did find that the globin genes had increased contact with SC35-positive nuclear speckles (). The frequency of contact between α-globin genes and speckles was 26% in lymphoblasts and 23% in lymphocytes. This level of contact may be caused, in part, by the transcriptional activity of housekeeping genes in the region surrounding the α-globin genes. It was higher in early erythroblasts, and peaked in the intermediate erythroblast population at 87% before falling again in the late erythroblast population ( ). There was a significant difference between the proportion of α-globin loci contacting nuclear speckles in lymphoblasts and in erythroid intermediates (P < 0.02). We established that this was not attributable to differences in the size of the SC35 compartment by comparing the relative volume occupied by the SC35 signal in lymphoblast and intermediate nuclei (lymphoblasts, 0.064 ± 0.009; intermediates, 0.061 ± 0.008). Fewer β-globin than α-globin genes contacted SC35-positive nuclear speckles in lymphoblasts, but the percentage was greater in erythroid cells, peaking at 58% in intermediate erythroblasts (). We found that there was a good correlation between the percentage of genes transcribing in each erythroid population and the percentage of genes contacting nuclear speckles for both α-globin (R = 0.89) and β-globin (R = 0.95; ). The number of genes located at speckles clearly increased as the amount of transcription from the globin genes increased, although the degree of this contact was more marked for the α-globin genes. We looked at the contact of actual transcripts with speckles by RNA immuno-FISH, which does not involve a denaturing step, so that the 3D architecture of the cell is well preserved. We found that 100% ( = 81) of α-globin transcripts were located at speckles in intermediate erythroblasts. However, this association with large nuclear speckles cannot be essential for transcription in general because only 67% ( = 52) of β-globin transcripts were located at speckles in the same experiments. We have shown that the globin genes associate with nuclear speckles, and with each other, when transcribing. It seemed likely that these associations were related, thus, we undertook three-color immuno-FISH hybridizations, where we detected the α- and β-globin genes together with the SC35 splicing factor speckles. We were able to demonstrate that the association between α-globin genes, and between α- and β-globin genes, occurred at nuclear speckles in intermediate erythroblasts (100%; = 43; ); of these, 82% of associating α-globin genes and 78% of associating α- and β-globin genes were contacting the same nuclear speckle. Thus, the human globin genes exhibit a remarkable degree of association at the same speckle when transcribing. s s u e - s p e c i f i c g e n e e x p r e s s i o n i s i n f l u e n c e d b y t h e a v a i l a b i l i t y o f t r a n s c r i p t i o n f a c t o r s a n d t h e e p i g e n e t i c m o d u l a t i o n o f h i g h e r o r d e r c h r o m a t i n s t r u c t u r e . C h r o m a t i n a t t h e g l o b i n g e n e l o c i i s m o d i f i e d i n c o m m i t t e d e r y t h r o i d p r o g e n i t o r s , b u t t h e g e n e s t r a n s c r i b e a t h i g h l e v e l s s e v e r a l c e l l c y c l e s l a t e r , i n i n t e r m e d i a t e e r y t h r o b l a s t s . O u r a i m h a s b e e n t o u n d e r s t a n d w h a t c o n t r i b u t i o n t h e p o s i t i o n i n g o f t h e g l o b i n g e n e s w i t h i n t h e n u c l e u s m a y m a k e t o t h e i r t r a n s c r i p t i o n a l r e g u l a t i o n d u r i n g t e r m i n a l e r y t h r o i d d i f f e r e n t i a t i o n . B y c o m p a r i n g r e s u l t s i n h u m a n a n d m u r i n e c e l l s , w e h a v e s h o w n t h a t t h e p o s i t i o n i n g o f t h e g l o b i n g e n e s i s n o t d e t e r m i n e d s i m p l y b y t h e i r t r a n s c r i p t i o n a l s t a t u s , b u t i s a l s o d e p e n d e n t o n t h e u n d e r l y i n g c h r o m o s o m a l c o n t e x t . Primary human differentiating erythroblasts were obtained using a two-phase liquid culture system, as previously described (). In brief, mononuclear cells were obtained from peripheral blood buffy coat by centrifugation on a gradient of Ficoll-Hypaque and seeded at 2.5 × 10 cells/ml in α-minimal essential medium with 10% fetal calf serum, 1 μg/ml cyclosporin A, and 10% conditioned medium from cultures of the 5637 bladder carcinoma cell line for growth factors. The cells were cultured at 37°C for 6–7 d (phase I), and the nonadherent cells were washed and reseeded in fresh medium with 30% fetal calf serum, 1% deionized BSA, 1 U/ml of human recombinant erythropoietin, 10 M β mercaptoethanol, 10 mol/l dexamethasone, 0.3 mg/ml holotransferrin, and 10 ng/ml of human stem cell factor. Cells were cultured (phase II) for a further 4 (CFUe), 6 (proerythroblast), 8 (intermediate erythroblast), or 14 d (late erythroblast). Progression of the cells through terminal erythroid differentiation was monitored by May-Grunwald Giemsa (MGG)–stained cytospins. Concerns that the sorting process would disrupt organization within nuclei were addressed by returning aliquots of sorted cells to culture, where their ability to proceed to terminal erythroid differentiation in a normal manner was confirmed. Lymphoblastoid cell lines were established from normal individuals and cultured in RPMI with 10% fetal calf serum. Early passages were used. TSA (Sigma-Aldrich), when used, was added at 1 μM for the last 8 h of culture. Human activated T lymphocytes were obtained from whole blood by culture for 3 d in RPMI/10% fetal calf serum in the presence of 2% phytohemagglutinin M (GIBCO BRL) and 20 U/ml interleukin-2. Mouse-activated B lymphocytes were obtained from a suspension of spleen cells cultured in the same manner as mononuclear cells, but with 50 ng/ml lipopolysaccharide as mitogen. ES cells were cultured as described (; ) and harvested by standard cytogenetic techniques for two-dimensional FISH. ES cells were provided by A. Smith (mouse; The Institute for Stem Cell Research, The University of Edinburgh, Edinburgh, Scotland) and P. Andrews (human; University of Sheffield, Sheffield, England). Suspensions of mouse intermediate erythroblasts were obtained from adult mouse spleen treated with 1-acetyl-2-phenylhydrazine for 4 and 5 d (), under license from the Oxford University Local Ethical Review Process. Cells were sorted on a fluorescence-activated cell sorter (MoFlo; DakoCytomation) to purify populations of cells at specific stages of differentiation, as described previously (), but based on human cell surface markers CD36, CD71, and GPA. The antibodies used were anti–CD36-PE (BD Biosciences; days 4 and 6), anti–CD71-FITC (DakoCytomation), anti–GPA-APC (BD Biosciences; days 4 and 6), and anti–GPA-RPE (DakoCytomation; days 8 and 14). Probes used for RNA FISH were either pools of hapten-labeled oligonucleotides (Eurogentec) designed from within the human and mouse α- or β-globin gene introns (sequences provided by P. Fraser, The Babraham Institute, Cambridge, England), nick-translated plasmids covering the human α-globin gene pRA.03 (a cloned 1.3-kb PstI fragment), or the human β-globin gene pN1β7 (a cloned 1.8-kb XbaI fragment). The globin DNA FISH probes used were cosmid HSGG1 () or BAC RP11-344L6 () for human α-globin; cosmid 202 (a gift from B. Morley, Imperial College School of Medicine, London, England) or BAC pBeloBac11 () for human β-globin; BAC 14567 (Incyte) for mouse α-globin; and BAC 357K20 (Research Genetics) for mouse β-globin. Additional human FISH probes used were a subtelomeric clone for HSA2p (J. ), cosmids 439A6, 419C1, and GG1 from a 2-MB contig of HSA16p13.3 (), cosmid 330H2 (a gift from N. Doggett, Los Alamos National Laboratory, Los Alamos, NM), BAC clone dJ1174A5 for NOR (courtesy of M. Rocchi [Universita' di Bari, Bari, Italy] and P. Finelli [University of Milan, Milan, Italy]), and whole chromosome paints for HSA11 and HSA16 (Cambio-Biosys). 2 μg of BAC, cosmid, or plasmid DNA were labeled, as previously described (). RNA FISH was performed essentially as described previously (), but using 0.02% pepsin. DNP-labeled oligonucleotides were detected with rat anti-DNP (MONOSAN), donkey anti–rat Cy3, and goat anti–horse Cy3 (Jackson ImmunoResearch Laboratories); digoxigenin (DIG)-labeled probes were detected with antibody layers of sheep anti-DIG FITC (Roche) and rabbit anti–sheep FITC (Vector Laboratories); biotinylated probes were detected with Avidin-Cy3.5 (GE Healthcare). Cells were mounted in Vectashield (Vector Laboratories) with 1 μg/ml DAPI counterstain. For RNA FISH analysis of sorted cell populations, sorted fractions were put back into culture for 6 h to ensure reestablishment of transcription patterns. In test cultures, this time period proved to be an adequate interval for recovery of transcription, but short enough to avoid further differentiation. FISH was also performed on cells where the 3D structure was preserved by PFA fixation. Cells were washed and allowed to settle on poly--lysine–treated coverslips for 10 min. The coverslips were fixed in 4% PFA for 15 min and permeabilized in 0.2% Triton X-100 in PBS for 12 min, at RT. RNA was removed with 100 μg/ml RNase in 2×SSC at 37°C for 1 h, and cells were denatured in 3.5 N HCl for 10 min at RT and neutralized in ice-cold PBS. Probes with Cot-1 DNA were denatured in hybridization mix as the previous section at 95°C for 10 min, preannealed at 37°C for 20 min, and hybridized to the denatured cells at 42°C for 48 h. Cells were washed twice in 2× SSC at 37°C for 30 min, once in 1× SSC at RT for 30 min, and blocked in 3% BSA in 4× SSC at RT for 30 min. DIG-labeled probes were detected with antibody layers of sheep anti-DIG FITC (Roche), followed by rabbit anti-sheep FITC (Vector Laboratories), both at RT for 30 min. Biotin-labeled probes were detected with antibody layers of avidin Cy3.5 (GE Healthcare) and biotinylated anti-avidin (Vector Laboratories). Coverslips were washed between layers in SSCT (4× SSC with 0.05% Tween 20), rinsed in PBS, and mounted in Vectashield with 1 μg/ml DAPI or 200 nM TOPRO3 (Invitrogen) counterstain. Cells were washed and allowed to settle on poly--lysine–treated coverslips for 10 min. Cells were fixed in 2% PFA for 15 min and permeabilized in 0.2% Triton X-100 in PBS for 12 min at RT. Nonspecific sites were blocked using 5% normal goat serum/5% fetal calf serum in PBS at RT for 30 min. Antibodies were prepared in blocking solution at the following concentrations: rabbit anti-PML (sc-5621; Santa Cruz Biotechnologies, Inc.) 1:100; rabbit anti–upstream binding factor (serum 6'2; a gift from M.Valdivia, Universidad de Cádiz, Cádiz, Spain), 1:1,000; mouse anti-SC35 (Sigma-Aldrich) 1:500. The secondary antibodies used were goat anti–mouse Cy5 (Jackson ImmunoResearch Laboratories) and horse anti–mouse Texas red (Vector Laboratories). Coverslips were mounted in Vectashield with DAPI, as in the previous section. Immunofluorescence was first performed essentially as described in the previous section, with the exception that cells were fixed in 4% PFA and permeabilized in 1% Triton X-100. After immunofluorescence, cells were postfixed in 4% PFA for 15 min at RT, followed by RNA removal by RNase in 100 μg/ml 2× SSC (Sigma-Aldrich) at 37°C for 1 h. Cells were denatured in 3.5 N HCl at RT for 20 min and neutralized with three brief washes in ice-cold PBS. Probes were prepared as described in the 3D DNA FISH section. Hybridization was performed overnight at 37°C. Washes and detection of DNA probes were performed as described in the 3D DNA FISH section. Immuno RNA FISH was adapted from . Cells were fixed in 3.7% PFA in PBS at 37°C for 10 min, permeabilized in 0.5% Triton X-100 in PBS on ice for 6 min, and blocked in 1% BSA at RT for 15 min. Primary and secondary antibodies were prepared in blocking solution, as described in the Immunofluorescence section. After protein detection, the slides were postfixed in 3.7% PFA in PBS at RT for 10 min. Slides were washed in PBS, rinsed in 2× SSC, and hybridized with probes, as described in RNA FISH. Detection of probes is also as described in the RNA FISH section. MGG-stained cytospins were imaged on a microscope (BX60; Olympus) with a Q Imaging camera and OpenLab (Improvision) software. The chromosome territory images were taken on a BX60 microscope with a PSI MacProbe v4.3 image analysis package (Applied Imaging) and a Sensys charge-coupled device camera (Photometrics). All other fluorescent preparations were analyzed with a confocal microscope (Radiance 2000; Bio-Rad Laboratories) system on a microscope (BX51; Olympus). All images were taken through a 100× objective (UplanF1; Olympus), numerical aperture 1.3, using Lasersharp software (Bio-Rad Laboratories). Z stack images were acquired at 0.2-μm intervals. Contrast-stretch and gamma adjustments were made using Photoshop (Adobe) only for display in figures. Distance measurements between associating FISH signals used LaserPix software (Bio-Rad Laboratories). Estimations of nuclear volume were made by applying the volume formula for an oblate ellipsoid to maximal radii obtained from confocal Z stacks of nuclei. Volume density of the speckles was calculated using stacks of images that cover the entire volume of the nuclei. In each plane, we measured the area covered by the speckles and the area of the nuclei. The volume density of the speckles was calculated as the sum of speckle areas divided by the sum of the nuclear areas for each nucleus. This was repeated for 10 lymphoblast and 10 intermediate erythroblast nuclei.

Elucidation of spatial organization of chromosomes in the cell nucleus and the consequences on chromosome metabolism is an area of intense research. In many organisms, chromosomes adopt a preferential spatial arrangement (; ). This is also observed to a certain extent in the nuclei of the yeast , with centromeres found at one pole of the cell, the nucleolus—and therefore rDNA—at the opposite pole, while the 32 telomeres cluster in 4–8 foci close to the nuclear periphery (; ). Two out of these three chromosomal regions, namely telomeres and the rDNA locus, are characterized by their ability to epigenetically silence PolII-driven markers inserted therein (; ). This is also true in the silent mating cassettes and (). Apart from the rDNA, which follows specific silencing rules (), these domains are found at the nuclear periphery where pools of silent information regulator (Sir1-4) proteins, essential for establishment and maintenance of telomeric and silencing, are concentrated (). In certain experimental systems, delocalization of silencing factors can repress nontelomeric or silencer-deficient loci (, ; ; ), and Sir-mediated repression of an locus with a defective silencer is improved if the locus is brought to the nuclear periphery (). However, anchoring to the nuclear envelope and silencing can be separated (; ; ). Anchoring of telomeres is mediated by at least two partially redundant pathways requiring either yKu or Sir4p and their interacting partner Esc1p (; ; ). Surprisingly, neither of these proteins is an integral membrane protein, although indications exist that Esc1p might behave as a membrane protein (; ). In , the nuclear pore complex (NPC) has been recently shown to be involved in chromatin anchoring through establishment of a nonsilenced domain within the silent mating-type locus , association with highly transcribed genes, and repression of telomeric domains (; ; ). In addition to its role in telomere anchoring, yKu plays a major role in double-strand break (DSB) repair. In eukaryotes, DSBs are repaired by two pathways: homologous recombination (HR) and nonhomologous end joining (NHEJ). NHEJ involves a Ku heterodimer-dependent rejoining of the two chromosome ends; HR requires a homologous region and the concerted action of proteins of the Rad52p epistasis group. HR can occur as a gene conversion (GC) event or as break-induced replication in which one portion of the chromosome arm is entirely replicated from a homologue (; ; , ). In the nucleus, the sites of DSB repair are characterized by the aggregation of repair proteins into foci that can recruit multiple DSBs simultaneously as evidenced both in yeast and mammals (; ). It is, however, not known what defines the nuclear position of the DSB foci, beside the DSB itself (). The dual involvement of yKu in telomere positioning and DSB repair, together with the recruitment of repair proteins in determined foci, suggest that links, investigated here, might exist between DNA repair and spatial positioning. To affect spatial positioning, we mutated the Nup84 complex of nuclear pore proteins, a conserved core complex formed by seven distinct entities, including Nup84p, Nup120p, Nup133p, and the carboxy-terminal part of Nup145p, Nup145Cp, this having been previously shown to be involved in telomere clustering (; ; ). DSBs were generated by in vivo expression of the I-SceI endonuclease in yeast strains containing the recognition site at defined positions. Repair mechanisms have been shown previously to differ according to the position of the DSB on a chromosome (). In haploid yeast, surviving colonies are rare and cells have exclusively repaired DSB by NHEJ, unless the DSB occurred in subtelomeres (Ricchetti et al., 2003). In these domains, in addition to NHEJ, other repair events are observed, such as break-induced replication, telomere addition, and addition of exogenous sequences. These account for a three- to fourfold higher frequency of colony formation (, ). In this work we show that deletion of several components of the Nup84 complex affect telomere positioning, and that this is correlated with a defect in telomeric silencing. We also show that mutants of this complex, but not nucleoporin mutants peripheral to this complex, like Δ, affect repair of a DSB when it is subtelomeric but not when it occurs in non-subtelomeric domains. Furthermore, we confirm that telomere tethering and subtelomeric DSB repair were linked by analyzing another mutation, Δ. Altogether our results suggest that the NPC, through the Nup84 complex, plays an essential role in chromosome positioning, and this defines a perinuclear space required for efficient subtelomeric repair. We have chosen to tag the native left end of chromosome XI-L with binding sites for TetR-GFP fusion ( ; Fig. S1, available at ). The method is based on fluorescence imaging of labeled telomere in living cells through three-dimensional (3D) focus stacks and subsequent image analysis, correcting both image and nuclei distortions (see Materials and methods). We have analyzed the 3D position of the labeled telomere in the nuclear space in a wild-type strain, in mutants of the Nup84 complex, Δ, Δ, Δ, and Δ, as well as in Δ and Δ, the mutants previously shown to define two redundant pathways for anchoring of telomere VI-R and XIV-L at the periphery (). The quantification of the effect on telomere localization is shown in and corresponding cumulative distribution function (CDF) of radial distances in Fig. S1. For quantification, position of the locus was mapped into two concentric spheres of equal volume, the periphery being determined by the mean position of Spc29-GFP fusion protein (see Materials and methods). As expected, in the wild-type strain, telomere XI-L is located close to the periphery in 65.3% (±5.16; = 427) of cells (values for a 95% confidence interval and the number of analyzed nuclei are given). Strikingly, the deletion of the carboxy-terminal encoding part of , or deletion of , , or leads to the relocation of telomere XI-L in the nuclear interior with only 54.12% (±4.8; = 312), 48.37% (±3.44; = 283), 45.27% (±7.6; = 323), and 44.83% (±6,61; = 313) of tagged telomere XI-L remaining in the peripheral volume, respectively. These values are close to the ones observed in cells were the telomere anchoring pathway has been inactivated by deletion of either (45.11% ± 3.47; = 596) or (40.70% ± 5.04; = 322). Position of telomere XI-L was also analyzed in the ΔΔ double mutant () and found to be in the peripheral volume of 38.86% (±6.7; = 261) of the mutant cells. All together, our results show that telomere XI-L is preferentially found in the outmost external volume of the nucleus and that its localization at the nuclear periphery requires both yKu70p and Sir4p, a situation reminiscent to the one observed for chromosome VI-R and XIV-L (; ). Furthermore, our results show that a functional Nup84 complex is required for the peripheral localization of this particular chromosome end and suggest that the NPC might be involved in telomere anchoring. Having shown that telomere XI-L was delocalized in Nup84 complex mutants, we investigated the efficiency of survival and repair to DSBs in these mutant backgrounds. To generate DSBs, we expressed the I-SceI endonuclease in vivo in haploid strains carrying a marker flanked by two inverted I-SceI cutting sites at defined positions ( and ). Cell survival is monitored by comparing the number of colonies formed after plating on inducing (galactose) versus noninducing (glucose) conditions, whereas efficiency of DSB formation is followed by measuring the percentage of [URA] colonies after replica plating (see Materials and methods). In a wild-type haploid strain, most of the cells are unable to form colonies after induction of DSBs on galactose, but survival increases by a factor of 20 when the DSB is in subtelomeres (). We analyzed survival to both a cut 3.5 kb from the left telomere (L1 in ) and a cut in a central region, 64 kb from the left telomere (C5) of chromosome XI in cells mutated in the Nup84 complex and in , a nucleoporin peripheral to this complex. ; ). We found that cell survival is comparable between wild type, Δ, Δ, and Δ strains ( and ). On the contrary, in the case of a subtelomeric DSB, whereas in a wild-type strain cell survival reaches ∼4% ( and ; ), it decreases by 93% and 71% in Δnup145C and Δnup84 strains, respectively, and by 42% and 41% in Δnup120 and Δnup133, while it is only partially diminished in the Δnup170 strain (20%; ). Therefore, in all Nup84 complex mutants tested here a defect in cell survival after generation of a DSB was observed, with survival rates of cells bearing a DSB in a central and in subtelomeric region of the same chromosome becoming similar in Δ and to a lesser extent in Δ mutants, indicating that in these mutants the dependency of the survival rate on the location of the DSB along the chromosome is abolished. The decrease in cell survival could reflect either a problem in repair efficiency or some checkpoint-related problem in response to DNA damage. The latter was excluded because the Δ mutant adapts and recovers subtelomeric DSB like a wild-type strain, whereas the checkpoint Δ mutant does not (see Fig. S2, available at ; and ). We therefore analyzed the mechanisms by which the DSBs were repaired. 200 colonies that formed on galactose-containing medium were screened by PCR for possible repair by NHEJ. The absolute frequency of repair was calculated by scoring the number of positive PCRs relative to the number of plated cells (in noninducing conditions). In a wild-type strain, 0.21% (±0.039) of plated cells were repaired by NHEJ; this value remaining constant along the chromosome (; and our unpublished data). The absolute frequency of NHEJ drops to 0.014% (±0.002) in Δnup145C mutant cells when DSBs are generated close to the telomere, but not when the cut is central ( and unpublished data). This shows that a component of the Nup84 complex is required for efficient repair by NHEJ in subtelomeres. However, when the frequency of repair events is calculated among the survivors, similar values are obtained in wild-type and Δnup145C mutant strains (respectively 29% ± 7 and 25% ± 7), indicating that when repair of the DSB allows colony formation, the ratio of NHEJ to other mechanisms is the same in both backgrounds. 21 of these subtelomeric NHEJ events that occurred in Δnup145C mutants were sequenced, and repair event corresponded in all cases to NHEJ involving only minor rearrangements ( ). The majority of the cases (60%) correspond to a ligation between the two cut sites with insertion of two nucleotides, as already observed in a wild-type context (type II in and in ). The remaining cases (40%) correspond to a major resection of both I-SceI cutting sites, which was not previously described (therefore named type VII in ). We then analyzed the nature of non-NHEJ events at the L1 locus. The absolute frequency of these events (i.e., the number of events calculated on the total number of plated cells) is 4% in the wild-type strain, and drops to 0.25% in the Δnup145C strain (). Comparison of the Southern blot patterns (as in ) of repaired chromosomes in wild-type and Δnup145C cells shows only minor variations in the frequencies of different types of repair, and no novel pattern in the mutant (unpublished data). The observation that DSB repair is efficient in NPC mutants when the DSB is generated in a central region of a chromosome argues against a defect in nuclear transport of enzymes involved in the repair pathways. Nonetheless, we designed several experiments to confirm this. First, we studied cell survival and DSB repair using a similar experimental setup, but in which expression of I-SceI endonuclease is driven from the gene, flanked by two I-SceI cutting sites, integrated on chromosome XV at the central locus (). In this system, the presence of an mutated allele inserted into () 31.7kb from telomere V-L allows visualization of GC events. In these strains, induction of I-SceI expression leads to loss of the marker in 95% of cells. As was observed for a DSB generated in a central domain of chromosome XI, cell survival is similar in both wild-type and Δnup145C strains ( and ). Frequencies of repair either by NHEJ or GC are also unchanged, indicating that, in the Δnup145C strain, the repair machinery is functional and has access to a central chromatin domain ( and ). We next asked how a subtelomeric DSB can be repaired when homologous sequences are present. We thus constructed a diploid strain homozygous for the Δ deletion and heterozygous for the cassette inserted at the subtelomeric L1 position. Cell survival after induction of the DSB was high in both wild-type and homozygous Δnup145C (91.5% ± 6.9 and 84% ± 11.4) with a cutting efficiency of 100%. This indicates that the inability to repair a subtelomeric DSB observed in a haploid Δnup145C strain is overtaken in a diploid strain. Finally, we used an assay that relies on the in vivo repair of a linearized plasmid required for its propagation in yeast cells (; ). Because the DSB is generated in a region that is not homologous to the yeast genome, repair operates only through NHEJ. We determined the number of transformants recovered after transformation with a linearized and purified ARS-CEN-URA3 plasmid (pRS316; ) relative to an uncut control both in wild-type and Δnup145C strains (). Restrictions leaving either 5′ overhangs (HindIII) or blunt ends (SmaI) were chosen. No differences in efficiency recovery were observed in both strains for these two types of ends, although SmaI linearized plasmid was less efficiently repaired (WT 2.1% ± 0.7; Δnup145C 2.9% ± 0.2) than HindIII (WT 24.2% ± 7.8; Δnup145C 16.5 ± 7.8), as expected (). All the above results indicate that the repair machinery is functional in Δnup145C, and therefore deficiencies in subtelomeric repair cannot be explained by a general deficiency of repair in the mutant. Because repair of a subtelomeric DSB is affected when members of the Nup84 complex are mutated, we analyzed whether silencing of subtelomeric chromatin is affected in these mutants. XI-L subtelomere was shown to be one of the most efficiently silenced subtelomeres (). Transcriptional silencing of a marker integrated into this subtelomeric region is measured by counting the fraction of cells that are able to grow in the presence of 5-fluoro-orotic acid (5-FOA), a toxic analogue of uracil. Reverse growth of 5-FOA–resistant clones on medium lacking uracil verifies that resistance is due to subtelomeric silencing rather than mutation of the gene. Compared with the wild-type strain, disruption of Nup84 complex (i.e., Δ, Δ, Δ, and Δ) reduces silencing with variable strength, leading to very little or partial cell growth on 5-FOA ( ). On the contrary, silencing of native subtelomere XI-L remains intact when other nucleoporin genes (i.e., Δ and Δ) are disrupted. Furthermore, silencing was restored when the corresponding wild-type nucleoporin was used for complementation, as shown for , , and (). These results show that defects in telomere position effect (TPE) are specific for the Nup84 complex. In addition, transcomplementation of silencing defects by the Δ mutant, lacking the amino-terminal encoding part of Nup133p, which only affects NPC clustering (), indicates that NPC distribution is not involved in the silencing defects observed for native telomere XI-L. We then asked whether defect in silencing corroborated with a delocalization of Sir3p. This was monitored by microscopy using a Sir3-GFP fusion protein; addition of the GFP moiety to the COOH terminus of Sir3p having no effect on TPE of native subtelomere XI-L (Fig. S2 A). In a wild-type strain Sir3-GFP localizes in few dots (4–6) located at the nuclear periphery (Fig. S2 B). In Δnup145C or Δnup84 mutants, a nucleoplasmic staining is visible and Sir3-GFP localizes to an increased number of foci (6–8), indicating a partial delocalization of Sir3-GFP in Nup84 complex mutants that may be responsible for the defect in TPE. We have established above (see previous paragraphs) Introduction or see previous paragraph that the Nup84 complex is required for telomere XI-L localization, subtelomeric DSB repair, and silencing. To define whether requirement of peripheral localization for efficient repair of subtelomeric DSBs is a general phenomenon or a specificity of the Nup84 complex, we decided to analyze a mutant external to NPCs, with no known role in repair, that would disrupt telomere positioning. We have chosen Esc1p, a nuclear periphery protein that interacts with Sir4p and tethers native telomere XIV-L and truncated telomere VI-R (see Introduction and ; ). , only 47.90% (±6.3, = 311) of tagged telomere XI-L are found in the outmost external volume of Δ nuclei, indicating that Esc1p is also required for tethering this telomere end. Strikingly, cell survival to a subtelomeric DSB is decreased by ∼50% in an Δesc1 mutant (). Thus, the lack of telomeric positioning at the nuclear periphery is again correlated with inefficient subtelomeric DSB repair. Furthermore, we have found that deletion of induces only a minor reduction on silencing efficiency of subtelomere XI-L, as previously published for other telomeres (; ; ). This result allows the separation of the function of chromatin silencing with the one of chromatin repair and suggests that only chromatin positioning at the periphery is important for subtelomeric DSB repair. In this work we show that a central subcomplex of the NPC has a key role in the functional organization of the nuclear periphery. This is manifested by its capacity to anchor telomere XI-L at the nuclear periphery, to maintain a repressive environment for subtelomere silencing, and to participate in subtelomeric chromatin repair. These results suggest that telomere anchoring, and may be clustering, are important for the repair of DNA breaks occurring in these regions and thus for chromosome integrity. Our data corroborate and extend previous observations obtained for localization of native telomeres (,; ; ). Most of these in vivo studies rely on quantitative analyses of the position of a fluorescent locus in the focal plane. Our analyses take into account the position of the locus in the volume of the nuclear space, and thus give an unprecedented level of positional knowledge (see also ; ; ; ). We were able to show that tel XI-L behaves like native telomere XIV-L and VI-R, because it localizes at the periphery in a yKu70p- and Sir4p-dependent fashion, although the relative dependencies during the cell cycle for each of these proteins has not been addressed here. In addition, we show that anchoring of telomere XI-L requires the Nup84 subcomplex. Whether Nup84-dependent localization is restricted to chromosome XI-L or whether this is a shared pathway for other telomeres remains to be established, variable behaviors among native ends being probable. It is intriguing that, although required for peripheral tethering of telomeres, neither yKu70p nor Sir4p are proteins of the nuclear envelope. Sir4p was shown, however, to interact with Esc1p, a protein required for plasmid partitioning, that may interact with the nuclear envelope (; ). Nuclear pores, through the Nup84 subcomplex, now appear as serious candidates for nuclear envelope targets for telomeres that could either link yKu70p, as combined mutations of yKu and Nup84 suggest, Esc1p, or an as yet undiscovered protein(s). The role of nuclear pores in telomere anchoring has been debated. On one hand, Δ has been described as a mutant that induces delocalization of telomeres and disintegration of telomeric clusters as measured by FISH with subtelomeric Y′ probes (), a genetic element present on two thirds of subtelomeres in the sequenced strain S288C and absent in telomere XI-L studied here (); on the other hand, colocalization between clustered nuclear pores (in a Δ background) and the telomere-binding proteins Rap1p or Sir4p foci has rarely been observed, arguing against anchoring by the pore (). However, the localization of individual native telomeres in nuclear pore mutants has not been addressed in any of these studies. We show that mutants of the Nup84 complex are defective in telomere tethering. Localization of a functional GFP-tagged version of the Sir3p silencing factor suggests, however, that this protein is partially delocalized in Nup84 complex mutants, with an increased number of visible Sir3p foci; because Sir3p binds to telomeres, it is indicative of a splitting up—but not disappearance—of telomere clusters. Whether individual, untethered telomeres remain clustered with Sir3p, Rap1p, Sir4p, or Y′ subtelomeric foci is an open question that requires first of all the determination of the telomeres constituent of each focus. All these results suggest, however, that clustering and tethering may be uncoupled, a notion that is reinforced by the observation that the clusters of telomeres can be displaced from nuclear envelope in Δ mutants (). Telomeric anchoring is not the only function of the inner nuclear envelope. It is well established that periphery is important for the establishment of a transcriptionally repressive environment. Transcriptional repression appears to be due to a specialized chromatin structure found in in the regions close to the nuclear periphery; i.e., the chromatin close to telomeres and to the silent mating type cassettes and . Current models suggest that TPE involves binding of Rap1p to TG1-3 telomeric repeats that in turn helps to recruit Sir proteins required for subsequent proximal histone deacetylation (for reviews see ; ; ; ). It is proposed that local enrichment of Sir proteins at the nuclear periphery is helped by telomere clustering. We establish here that at least for the native subtelomere XI-L, the Nup84 complex has a role in repression that does not depend on NPC clustering. This is in agreement with our previous finding that Seh1p, which also belongs to the Nup84 complex, affects subtelomeric silencing (). Effects on silencing were, however, variable among Nup84 complex mutants, reflecting the relative functional importance of each entity, maybe linked to differential positioning of these proteins inside the complex (). Paradoxically, NPCs have recently been shown to also play a role in protecting chromatin from transcriptional repression (; ; ). Boundary proteins that prevent the spreading at include the NPC protein Nup2p (). Using genomic localization analysis, Mlp1p, Mlp2p, Nup2p, and Nup60p, but not Nup84p and Nup145Cp, were preferentially found associated with highly transcribed genes that tend to have Rap1p binding sites (). On the other hand, -dependent transcriptional activation involves members of the Nup84 complex (). This indicates that NPCs, through the binding of the multifunctional regulator Rap1p, have a dual role toward the transcriptional state of the chromatin and may influence organization of the nuclear periphery by partitioning active and inactive regions. Another possibility could be that NPCs are not all equivalent toward the transcriptional status of a chromosome. In any case, interaction of genes with NPCs offers a new, and not fully understood, level of gene regulation. Previous work by showed that DSB repair efficiency depends on the position of the DSB along the chromosome, an increased rate of survival being observed when DSBs were subtelomeric, due to NHEJ and the existence of additional mechanisms of repair including HR between repeated sequences shared by different subtelomeres. The increase in cell survival after a subtelomeric DSB was still low, possibly as a result of both limited length and level of sequence identity between these repeated sequences, or as a result of chromatin composition in that region (). Telomere tethering and/or clustering was also proposed to act as barriers to recombination between telomeric and internal chromatin in other work (). In fact, we have shown that release from the periphery does not improve cell survival to a subtelomeric DSB, but on the contrary diminishes the capacity to form colonies after DNA damage. We excluded that this decreased rate of survival is due to a defective checkpoint that would impede cellular recovery to the DSB. The observed decrease in cell survival might be due to a deficient DSB repair specific to subtelomeric regions. We do not know, however, at which precise level repair is affected; it could either be a defect in repair kinetics or in other processes like the choice of the partner (). Subtelomere tethering at the nuclear periphery appears as a requirement for their capacity to repair genomic injuries occurring therein, an observation reminiscent of the proposed role of telomeres clustering in facilitating recombination between subtelomeres of other organisms (; ; ). It is noteworthy that three mutants of the Nup84 complex, namely Δ, Δ, and Δ were found to be synthetically lethal with genes involves in replication (Δ) and repair (Rad52 pathway) (). This genetic interaction was correlated with an increased number of Rad52-YFP foci. Because spontaneous DSBs arising during mitotic DNA synthesis may be repaired though recombination, it was proposed that Nup133p may be important for this process. Our data extend this observation, showing that the Nup84 complex, through Nup84p, Nup120p, Nup133p, and Nup145Cp, is indeed important for repair, but restricted to the subtelomeric domain. Whether Rad52-YFP foci correspond to recruitment of the repair machinery at the periphery remains an open question. Nup84p, Nup133p, and Nup120p nucleoporins were also found in genome-wide screens searching for DNA damage pathways (resistance to γ radiations, sensitivity to MMS or bleomycin; (; ). It was shown in particular that recombination between donors and the mating type cassette, allowing for repair of a DSB initiated by HO, was impossible in a Δnup84 mutant (). This was interpreted as the putative consequence of a defect in nuclear transport of repair enzymes. In light of our results, we favor a hypothesis in which repair of or , which are subtelomeric, could not operate because the nuclear periphery is perturbed in the Δnup84 mutant. It should be noted that most of the constituents of Nup84p complex, when deleted, lead to a defect in NPC positioning in the plane of the nuclear envelope (; ; ; ). It is possible that NPC clustering is in part responsible for the defects observed here. However, studies with Δnup145C were performed at 25°C, a permissive temperature at which not only nuclear transport is normal, but also NPC clustering is not complete (unpublished data) and NPC clustering was excluded as a cause of increased Rad52 foci (). Altogether, this suggests that telomere anchoring through the Nup84 complex is necessary for the efficient repair of DSBs occurring in these regions. The finding that deletion of , which encodes a nonchromatin protein associated with the nuclear periphery (; ), affects both telomere tethering and efficiency of subtelomeric DSB repair, substantiates our findings on the functional importance of chromatin organization at the nuclear periphery. Furthermore, as previously shown, chromatin anchoring is found to occur independently of transcriptional repression in Δesc1 mutant (; ); this allows the uncoupling of silencing and the chromatin modifications inherent to it, and the defects observed here in subtelomeric repair. Despite the importance of spatial chromatin organization, little is known about the molecular principles that define its organization. A key role for nuclear envelope proteins is however substantiated, in particular in organisms in which mitosis is open; nuclear lamina proteins and the NPC proteins, through the mammalian orthologue of the Nup84 complex, vNup107, were shown to initiate the first steps of post-mitotic chromatin reassembly (; ). Another role for NPCs in defining nuclear environments favorable for DSB repair is highlighted here, maybe through nonrandom chroma- tin organization as is suggested for telomere clustering, origin firing, and pairing between silenced and nonsilenced mating type cassettes. . Nucleoporin, , , and disruptions were introduced either by transformation with kanMX or a PCR fragment allowing for single step disruption (), or by mating and sporulation. Deletion of was defined according to . All gene deletions were verified by Southern blot by using wild-type genes as probes. Tagging of telomere XI-L for in vivo localization was performed by integrating 112 repeats at 3.5 kb from telomere. The yeast integrative plasmid pRS306 () containing 112 42-bp fragments including the 19-bp palyndromic 2 sequence (hereafter referred to as repeats) was initially obtained from K. Nasmyth (Research Institute of Molecular Pathology, Vienna, Austria) (). Gene natMX4 coding for nourseotricin resistance was linked to the repeats, by cloning a BglII-AatII natMX4 containing fragment from pAG25 () into pRS306-112. The resulting plasmid was linearized by StuI and used to transform strain FYBL1-8B/Ai0611 (), allowing for insertion of repeats at insertion and size of the repeats were checked by Southern blot. Digestion by EcoRV of modified YIplac128 plasmid () allowed integration of SV40NLS-TetR::GFP at Δ locus. The strain was verified by Southern blot using as a probe and disrupted either for , , , or by transformation or for , or by mating. Haploid cells tagged by repeats on chromosome XI-L (see ) were grown to exponential phase at 25°C in 5ml of YPD. When concentration reached 1–2 × 10 cells/ml, cells from 1 ml aliquots were washed in synthetic complete medium and resuspended into 500 μl of the same medium. 5 μl were spotted onto slides, covered by a thin layer of agar (1.5% agar in synthetic complete medium). Images were captured with an UltraView RS Nipkow-disk confocal system (Perkin-Elmer) and controlled by PE-viewer software. A 100× objective (Plan Apo, 1.4 NA, oil immersion; Carl Zeiss MicroImaging, Inc.) was used. The pixel size is 65.8 nm. Stacks with a slice spacing of 200 nm were taken using a laser excitation of 488 nm with an acquisition time of 200 ms each. An analysis of each projected stack was performed with QuIA software. This software relies on the quantitative detection of the labeled chromosomal locus via its 3D coordinates (X, Y, Z) in the nuclear space by a multi-resolution method using a 3D wavelet transformation, allowing to correlate information from different sub-bands of analysis and to select spot-like structures only within the nuclear volume (). The output of the detection step are the coordinates of the center of the chromosomal locus and of the nucleus, from which a first estimate of their relative distance is computed, before correction of Z distortion. TetR-GFP background allows for nucleoplasm detection and calculation of R, i.e., the mean of all radii on the median focal plane, assuming the nucleus is spherical. In wild-type cells, border of TetR-GFP background coincides with Nup49-GFP staining. Z-distortion and Z-detection artifacts are corrected thanks to the values obtained from localization of a GFP fusion with Spc29p, a single pole body (SPB) protein. In yeast, the SPB is embedded in the nuclear envelope. Without any Z distortion, the distance between the SPB and the center of nucleus, normalized by the R value (d/R), should be constant whatever the Z coordinate of the SPB. On the contrary, the observed distribution d/R increases with the distance along the Z-axis of the SPB from the median plane of the nucleus. This distribution is best fitted by the second-degree polynomial curve: d/R = 0.3309(Z − Z) + 0.0129(Z − Z) + 0.8457. This equation was used to determine the compensatory term to be removed from all the measured radial distances between the barycentre of locus and of the nucleus (d) and normalized by the R value (dtetO/Rest). Thus the corrected ratio, termed relative 3D position (R3Dp) is d/R = d/R − (0.3309(Z − Z) + 0.0129(Z − Z)). In each strain, percentages of cells according to their R3Dp are then either distributed in two zones of equal volume (i.e., 50% of the volume each, the outmost external one defined as the peripheral one) or presented as a cumulative distribution function (CDF), allowing for a Kolmogorov-Smirnov test that determines the statistical differences between two distributions. Note that the depth of the most external volume of a sphere divided in two equal volumes is similar to the depth of the most external area of a circular surface divided in three equal areas (see ). Analyses were performed on at least 200 nuclei in at least three independent experiments. The average nuclear radius in mutant cells versus WT cells was measured and shown to be similar between strains (0.97 μm ± 0.02, in WT and Δesc1; 0.98 μm ± 0.03 in Δsir4 and Δnup133; 0.99 ± 0.11 in Δnup145C; 1.02 ± 0.04 in Δku70; 1.07 ± 0.04 in Δku70 Δnup145; 0.92 ± 0.02 in Δnup120 and 1.15 ± 0.13 in Δnup84). In Δnup84 mutant, cell morphology is more variable, but distorted cells do not exceed 10% of the entire population. Silencing of a marker inserted at the core X of chromosome XI-L () was scored after growth of strains in nonselective conditions (YPD) to mid-log phase. 10-fold serial dilutions were made and spotted or plated to either SC-URA or plates containing 5-FOA. The percentage of cells giving rise to 5-FOA resistant colonies was calculated after 2–3 d of growth. Wild-type, Δnup145C, Δnup84, Δnup133, Δnup120, Δnup170, or Δesc1 strains containing the cassette flanked by two inverted I-SceI cutting sites at different positions along chromosome XI were transformed by the I-SceI expression plasmids, either pPEX7 () or pPEX14 (this paper). Both plasmids carry the I-SceI ORF controlled by a galactose-inducible promoter and a marker. pPEX14 corresponds to a version of pPEX7 in which a marker is inserted, allowing for efficient selection of transformants on SC-TRP. Induction of DSB was performed as in . Experiments were repeated at least five times with at least two independent pPEX transformants each time. Analyses of repair events at the molecular level were done according to . The frequency of repair events was calculated as the ratio of the number of these events divided by the number of colonies growing on galactose. The absolute frequencies were determined in each strain by dividing the previous value by the number of plated cells on glucose medium. Fig. S1: Nup84 complex is required for tethering of tel-XIL at the periphery: CDF values. Fig. S2: Δnup145C mutant is not defective in cell cycle arrest due to subtelomeric DSB. Fig. S3: Nup84 complex partially affects Sir3p localization. Online supplemental material available at .

The early steps in biogenesis of most secreted and membrane-bound proteins take place in association with the ER (). A dedicated machinery of chaperones and enzymes located within the organelle's lumen and on its membrane accomplish this important task (). Homeostasis in the ER is maintained by several signal transduction pathways that couple the load of newly synthesized ER client proteins to the organelle's capacity to cope with that load. These signal transduction pathways are collectively referred to as the ER unfolded protein response (UPR), and they are activated by ER stress, a departure from equilibrium between client proteins, and the organelle's capacity to cope with their load (; ). Three distinct ER-associated proteins are known to initiate signaling in the UPR. Two of these, IRE1 and ATF6, are concerned exclusively with activating genes whose products enhance the secretory machinery's capacity to cope with client proteins (; ; ). The third upstream component of the UPR, PERK, is uniquely responsible for coupling stress on the luminal side of the ER to rates of protein synthesis on the cytoplasmic side (, ). In the absence of PERK, unregulated protein synthesis overwhelms the capacity of the ER's luminal machinery, leading to unmitigated ER stress, protein misfolding, and cellular dysfunction and death (, ). Cells with physiologically high levels of ER client protein flux are especially dependent on the PERK-mediated arm of the UPR, and its inactivation adversely affects the function of organs with high secretory rates (; ). PERK is a type I ER membrane protein consisting of a stress-sensing luminal domain connected by a transmembrane segment to a cytoplasmic effector domain. The latter consists of an unusually large protein kinase domain with a single known substrate, serine 51 of the α subunit of eukaryotic translation initiation factor 2 (eIF2α; ; ; ). The phosphorylation of eIF2 converts it to an inhibitor of the GTP exchange factor eIF2B and, thereby, attenuates the recycling of eIF2 and the rate of translation initiation (). In addition to this mechanism for regulating ER client protein load, PERK-dependent eIF2α phosphorylation also promotes the transcriptional activation of a large number of UPR target genes (, ). This occurs through translational up-regulation of the transcription factor ATF4 (; ; ) and other less well-defined mechanisms (). Thus, PERK-mediated eIF2α phosphorylation has a pervasive role in both the translational and transcriptional components of the UPR (). In the compensated ER, PERK's stress-sensing luminal domain is maintained in a repressive complex with chaperones. As ER stress develops and the balance between chaperones and unfolded client proteins is disturbed, PERK's luminal domain is released from its repressive complex with chaperones. This leads to oligomerization of PERK in the plane of the ER membrane and promotes the transautophosphorylation of a large number of residues in PERK's kinase domain (). Several of these residues lie in PERK's activation loop, and their phosphorylation is likely to play a part in the conventional mechanism by which the catalytic activity of protein kinases is activated. However, other phosphorylated residues are located on PERK's unusually large kinase insert loop (). The latter is a feature conserved in PERK homologues from diverse species; however, the importance of the kinase insert loop to PERK function has not been previously explored. In this study, we report that PERK activation and transautophosphorylation selectively recruit its substrate, the nonphosphorylated eIF2 complex, to the kinase and that this mechanism of substrate recruitment, which requires the large kinase insert loop, is indispensable for eIF2α phosphorylation in vivo. PERK activation is associated with extensive transautophosphorylation, which leads to marked retardation in the protein's mobility on SDS-PAGE (; ). These activation-dependent changes in PERK mobility were observed both in endogenous PERK, which were activated by treating cells with the ER stress–inducing agent thapsigargin, and in an artificial PERK derivative, Fv2E-PERK, in which the protein's COOH-terminal kinase domain has been fused to a drug-dependent oligomerization domain (). Treatment of Fv2E-PERK–expressing cells with the oligomerization-inducing ligand, the otherwise inert drug AP20187, elicited a similar marked shift in the kinase's mobility ( ). A similar activity-dependent shift in PERK mobility was observed in a bacterially expressed fusion protein of PERK's kinase domain and GST (GST-PERK) because of constitutive dimerization through the GST portion. After cleavage of the GST moiety, the wild-type active PERK kinase domain migrated on an SDS-PAGE above the 100-kD marker with an apparent molecular mass of 105 kD, whereas a point mutation (lysine 618 to arginine) predicted to disrupt the kinases' ability to coordinate the γ phosphate of ATP migrated as a 85-kD protein (). The role of phosphorylation in affecting these marked differences in protein mobility was demonstrated by the observation that dephosphorylation of wild-type PERK in vitro reduced its mobility to that of the inactive mutant (). These phosphorylation-dependent changes in mobility on SDS-PAGE were also reflected in the protease sensitivity profile of the purified bacterially expressed proteins. Both the inactivating point mutation and the dephosphorylated PERK kinase had a similar protease sensitivity profile consisting of several protease-resistant fragments (, asterisks) that were not observed in the active wild-type protein. Overall, the active kinase was more protease sensitive, which is consistent with the known tendency of active kinases to assume a more open conformation that promotes ATP's access to the active site (). These observations suggested that PERK autophosphorylation leads to marked conformational shifts. To examine the possible impact of these conformational shifts in the kinase on substrate recruitment, we incubated cell lysates with immobilized, purified, bacterially expressed wild-type, K618R mutant and in vitro dephosphorylated wild-type GST-PERK and measured the recruitment of eIF2. Wild-type GST-PERK strongly bound the eIF2 complex in the lysate, as reflected by the large amount of eIF2α and eIF2β immunoreactivity that remained associated with the resin after extensive washes. In contrast, the mutant PERK and the dephosphorylated wild-type PERK had no measurable interaction with eIF2, which was recovered in the flow-through of the binding reaction ( ). A nuclear magnetic resonance–based structural analysis reveals that eIF2α is a bipartite protein with globular COOH- and NH2-terminal domains (NTDs) connected by a flexible linker. The phosphorylated residue serine 51 is found on a relatively flexible loop extending from the globular core of the NTD (). To further characterize the activity-dependent interaction between PERK and its substrate, we sought to determine whether it could be reconstituted with the NTD of eIF2α (eIF2α-NTD). First, we established that eIF2α-NTD is a substrate for PERK; wild-type GST-PERK readily phosphorylated eIF2α-NTD in vitro but failed to phosphorylate a point mutant in which serine 51 was replaced by aspartic acid (). In the absence of substrate, P is incorporated into the wild-type kinase through transautophosphorylation. Interestingly, both eIF2α-NTD and the nonphosphorylatable S51D mutant inhibit this transautophosphorylation, likely through the masking of the relevant sites on PERK. Next, purified bacterially expressed eIF2α-NTD tagged on its COOH terminus with six histidines was immobilized on a Ni–nitrilotriacetic acid (NTA) affinity column. Lysates of untreated and AP20187-treated Fv2E-PERK–expressing cells were passed over this affinity resin, and the binding of Fv2E-PERK was measured by immunoblotting. Only the activated low-mobility PERK bound to the eIF2α-NTD affinity matrix (). These observations indicate that the NTD of eIF2α specifically interacts with the activated PERK kinase domain. To examine the impact of eIF2α phosphorylation on its interaction with the kinase, we combined bacterially expressed eIF2α-NTD with immobilized active GST-PERK in the presence of various concentrations of ATP. We then separated the bound complex from the free proteins, which were then resolved by SDS-PAGE. The amount of eIF2α-NTD that associated with PERK was markedly diminished by the addition of ATP, and the protein emerging in the flow-through had a lower mobility and reacted with a monoclonal antibody that selectively detects phosphorylated eIF2α (). To further characterize these apparent differences in avidity of active PERK for its substrate (nonphosphorylated eIF2α) and its product (phosphorylated eIF2α), we mixed tracer amounts of P-phosphorylated eIF2α-NTD with the nonphosphorylated protein, bound the mixture to immobilized GST-PERK, and challenged the complex with increasing concentrations of salt. The nonphosphorylated eIF2α, which was detected by staining, eluted from the resin at a peak of 225 mM KCl, whereas the phosphorylated eIF2α-NTD, which was detected by autoradiography of the same gel, eluted at a lower salt concentration (peak of 125 mM KCl). The phosphorylation mimetic mutation serine 51 to aspartate also destabilized the complex with PERK, eluting at a similar low salt concentration in a parallel experiment (). These observations indicate that active PERK preferentially recruits its substrate, the nonphosphorylated eIF2α, and that the substrate, once phosphorylated, loses its attraction for the kinase. Most serine and threonine residues phosphorylated in active PERK are found on the kinase's large insert loop (). A model of PERK (based on the crystalographically determined structure of the related kinase GCN2; ) predicts that this large kinase insert loop, which is modeled as a single globular domain in scale with the PERK kinase, will extend from the surface of the NH2-terminal lobe at a distance from the catalytic site and, thus, is likely to be dispensable for catalytic activity ( ). To test this notion, we deleted the large kinase insert loop of PERK and compared the catalytic activity of the wild-type and Δ-loop versions of bacterially expressed GST-PERK. First, we measured the catalytic activity in terms of the incorporation of radiolabeled phosphate into the kinase itself in a transautophosphorylation reaction. The reaction exhibited typical Michaelis-Menten kinetics with a measured Km for ATP of 317 ± 70 nM for the wild-type kinase and a lower Km of 152 ± 17 nM for the Δ-loop mutant (). The kinetics of eIF2α-NTD phosphorylation by both the wild-type and Δ-loop generated nonlinear Lineweaver-Burke plots, indicating non–Michaelis-Menten kinetics. Hill plot analysis yielded Hill coefficients of 2.5 for both enzymes, indicating positive cooperation, which, furthermore, is consistent with an apparent dimeric or possibly multimeric configuration of the active PERK kinase domain as revealed by gel filtration (Fig. S1, available at ). ) and agreed with similar measurements conducted on the related eIF2α kinases PKR and HRI (; ). Furthermore, like the wild-type protein, GST-PERK Δ-loop was able to phosphorylate the intact eIF2 complex when added to cell lysates (). Based on these observations, we conclude that PERK Δ-loop is uncompromised in its catalytic activity. To determine whether deletion of the kinase insert loop impacted the activation-dependent conformational shifts in the kinase domain, we compared the protease digestion profiles of purified bacterially expressed GST-PERK Δ-loop with that of the same protein after de-phosphorylation in vitro. A shows that most of the protease-resistant fragments observed in the dephosphorylated wild-type full-length kinase domain and in the full-length inactive K618R mutant () were missing in the dephosphorylated PERK Δ-loop (). The presence of a common 27-kD phosphorylation-dependent trypsin-resistant fragment suggests that the active wild-type and Δ-loop proteins share a common core. Having established that the kinase insert loop contributes substantively to the activation-dependent changes in PERK's kinase domain, we sought to determine whether it also influences substrate recruitment. Therefore, we compared the binding of eIF2 in cell lysates to immobilized wild type and the Δ-loop mutant GST-PERK. Under identical conditions, the wild-type protein bound all of the eIF2 in the lysate, whereas the Δ-loop mutant was as ineffective as the catalytically dead K618R mutant GST-PERK (). These differences in avidity of the wild-type and mutant kinase were also reflected in the salt sensitivity of eIF2α-NTD phosphorylation. Whereas the phosphorylation of eIF2α-NTD by wild-type GST-PERK was maintained at salt concentrations of up to 250 mM, the GST-PERK Δ-loop's activity toward its substrate declined significantly between 50 and 150 mM of salt (). Salt had a similar negligible effect on the two kinases' catalytic activities, as reflected by the undiminished incorporation of label into the kinases themselves. We concluded that although dispensable to catalytic activity, PERK's large kinase insert loop contributes substantively to substrate recruitment in vitro. A stable interaction between phosphorylated residues on PERK's kinase domain and its substrate predict that substrate binding would protect these residues from dephosphorylation in vitro. To test this hypothesis, we radiolabeled GST-PERK by autophosphorylation in the presence of γ-[P]ATP and incubated the purified, radiolabeled protein with bacterially expressed λ-phage phosphatase in the absence or presence of wild-type eIF2α-NTD or in the presence of a mutant S51D substrate. PERK dephosphorylation was significantly more attenuated by the wild-type substrate than by the binding-defective mutant ( , compare A with B). Furthermore, an important fraction of the phosphorylated residues protected by the substrate are likely located on the kinase's insert loop, as the substrate had a very modest protective effect when applied to radiolabeled PERK that lacked the loop (). To test whether the recruitment of substrate through interaction with phosphorylated residues within the insert loop was independent of kinase activity, we mutated the serine and threonine residues of this domain in the catalytically inactive K618R GST-PERK to aspartates, mimicking their phosphorylation. We then repeated the in vitro pull-down experiments with eIF2α-NTD in the presence and absence of ATP (). Wild-type GST-PERK recruited its substrate, releasing phosphorylated product in the presence of ATP. The K618R mutation abolished this interaction, but recruitment could be restored by phosphomimetic mutations within the insert loop, which failed to restore kinase activity. These experiments support the interaction of substrate with phosphorylated residues within the insert loop of PERK. To determine whether the kinase insert loop's role in substrate recruitment in vitro is mirrored by its activity in vivo, we constructed oligomerization-activatable 6His-tagged versions of wild-type Fv2E-PERK, inactive K618R mutant, and Δ-loop mutant Fv2E-PERK and transduced these into mammalian cells. Both the wild-type and Δ-loop proteins were activated by adding AP20187 compound to the culture media, as reflected by their shift in mobility, whereas the K618R mutant was inactive as expected ( ). Isolation of the kinase and associated cellular proteins by Ni-NTA affinity chromatography showed that eIF2 bound only to the activated wild-type kinase but not to the two mutants. Furthermore, levels of phosphorylated eIF2α increased selectively in the drug-treated cells transduced with the wild-type protein but not in cells transduced with the two mutants (). These observations indicate that in cells, activation-dependent substrate recruitment and eIF2α phosphorylation depend on PERK's kinase insert loop. This study has uncovered a novel mechanism for substrate recruitment by the activated eIF2α kinase PERK (summarized in ). In its nonphosphorylated inactive form, PERK has a low affinity for its substrate. Upon activation, PERK undergoes phosphorylation-dependent conformational changes that markedly increase its affinity for the substrate, the nonphosphorylated form of eIF2. This change in affinity depends on the unusually large kinase insert loop, which is dispensable for catalytic activity itself. Phosphorylation of the substrate results in a marked decline in its affinity for the kinase and likely promotes substrate dissociation. The phosphorylated eIF2 product is free to migrate away from the kinase and to interact with and inhibit the guanine nucleotide exchange factor, thereby attenuating protein synthesis. The kinase insert loop, which connects strands 4 and 5 of the large β sheet in the NH-terminal lobe of all known protein kinases, is highly variable in size (). However, to date, no function has been specifically attributed to this portion of any known kinase. The tyrosine kinase domain of the platelet-derived growth factor receptor has a large insert loop that undergoes tyrosine phosphorylation upon receptor activation, but it connects structural elements in the COOH-terminal lobe () that are distinct from those connected by PERK's loop. PERK has one of the largest kinase insert loops of all known serine/threonine protein kinases. Although the primary amino acid sequence of the kinase insert loop is poorly conserved among PERK homologues, its unusual size is a conserved feature and one that it also shares with the related eIF2α kinase GCN2. Phosphopeptide mapping indicates that several transautophosphorylation events on PERK take place on residues in the kinase insert loop (). The presence of numerous phosphorylated residues on the loop in a context lacking sequence conservation and the marked salt sensitivity of the interaction between active PERK and eIF2α, which is further weakened when the latter is phosphorylated, suggests that eIF2 recruitment relies heavily on ionic interactions. Our data are consistent with a simple model whereby autophosphorylation rearranges the structure of the loop exposing the phosphorylated residues to the solvent. The substrate serine 51 of eIF2α is exposed on a loop (residues 47–67) containing six arginines and a lysine (). These exposed residues likely create a positively charged surface that may interact with the negatively charged phosphorylated residues on the PERK kinase insert loop, recruiting the substrate. Our experiments also indicate that more than one molecule of eIF2α-NTD is able to dock on activated PERK in vitro (Fig. S2, available at ), suggesting the existence of multiple substrate-binding sites on the activated enzyme. Phosphorylation of serine 51 would tend to reduce these charge interactions either by repelling the negatively charged residues on the phosphorylated kinase or by neutralizing intramolecular interactions with the basic residues on eIF2α's phosphorylation loop; either mechanism would account for the reduced affinity of the kinase for its phosphorylated substrate. The nearly identical catalytic features of the wild-type and Δ-loop kinases ( and ) argue against a specific role for the loop in orienting the substrate for catalysis and suggest that the recruitment mechanism described here may rely on determinants in eIF2α that are dispensable for the phosphorylation reaction itself. In this regard, it is noteworthy that mutation of the three arginine residues immediately COOH-terminal to serine 51 on eIF2α's phosphorylation loop (RRR→VKA) had only a modest inhibitory effect on the ability of the mutant protein to serve as a substrate for GCN2 in yeast or for PKR in vitro (). Our findings are easier to reconcile with a sequential mode of action whereby the phosphorylated loop facilitates substrate recruitment to PERK and hands it off to the catalytic core of the enzyme. Phosphorylation of the substrate disfavors rebinding and, thereby, promotes unidirectional substrate flow. This proposed mechanism of substrate recruitment helps explain how PERK, which is a relatively scarce enzyme, can nonetheless rapidly phosphorylate significant amounts (∼20%) of its abundant substrate eIF2α in ER stressed cells (). Many signaling modules consist of a kinase and downstream effectors held together in a preformed complex; however, eIF2 has a catalytic role in translation, which likely precludes a persistent association with the kinase. Furthermore, recent evidence suggests that in yeast, eIF2B, the GTP exchange factor that is inhibited by eIF2α phosphorylation, is located in discrete cytosolic foci through which eIF2 must traffic (). These considerations favor a mechanism whereby the substrate is rapidly recruited to the kinase solely when the latter is activated and then is rapidly released upon phosphorylation. The expression plasmids PerkKD-pGEX4T-1 and PerkKD(KA)-pGEX4T-1 encoding GST fusion proteins of mouse PERK residues 537–1,114 wild type and the kinase-dead K618A mutant have been previously described (). PerkKD(K618R mutant)-pGEX4T-1 has incorporated a K618R mutation in the PERK cDNA by PCR. PerkKD(ΔL)-pGEX4T-1 was generated by excising residues 643–886 of wild-type PERK. Tobacco etch virus (TEV)–cleavable GST fusion proteins were generated by transfer of the coding sequences of wild-type, K618R mutant, and Δ-loop PERK into pGV67 (a gift from B. Nolan, Yale, New Haven, CT). Mammalian expression vectors for 6His-tagged Fv2E-PERK wild type, K618R mutant, and Δ-loop were generated by transferring the coding sequence form Fv2E-Perk-pBABE () into pcDNA1/Amp containing a 6His cassette. eIF2α-NTD encoding residues 1–185 of human eIF2α with three solubilizing mutations was bacterially expressed from codon optimized vector 2aOPTx3M(1–185)pET-30a(+) (). eIF2α-NTD S51D was generated by transferring residues 4–185 from 2aOPTs51dx3M(4–302) () into pET-30a(+). λ-Phage phosphatase cDNA was generated by PCR from phage DNA and cloned into the expression vector pDUET (Novagen). Mutation of 57 serine and threonine residues within the insert loop to aspartic acid was performed by means of custom gene synthesis (GenScript Corp.). Stable clones of CHO cells expressing Fv2E-PERK have been described previously (). 293T cells were transfected with His6-FV2E-PERK–expressing plasmids and, 48 h later, were treated with AP20187 (ARIAD). Lysates were prepared in 0.5% Triton X-100, 100 mM NaCl, 20 mM Tris, pH 7.4, 1 mM DTT, 1 mM PMSF, 4 μg/ml aprotinin, 2 μg/ml pepstatin A, 10 mM terasodium pyrophosphate, 15.5 mM β-glycerophosphate, and 100 mM NaF. Bacterially expressed GST fusion proteins were first cleaved with TEV protease to liberate the GST portion. Thereafter, they were incubated with serial dilutions of sequencing grade trypsin over the range 20 ng/ml to 5 μg/ml (1047841; Roche) for 30 min at room temperature in 100 mM Tris, pH 8.0. Total eIF2α was detected with a mAb to human eIF2α (a gift of the late E. Henshaw), and phosphorylated eIF2α was detected with an epitope-specific antiserum (RG0001; Research Genetics). eIF2β was detected with a mAb (a gift from S. Kimball, Pennsylvania State University, University Park, PA). Protein disulfide isomerase was detected with a mAb (SPA891; StressGen Biotechnologies). PERK was immunoprecipitated and detected with polyclonal antisera as previously described (). Myc-epitope was detected with a mAb (9E10). Bacterially expressed GST-PERK proteins were incubated with Mg-ATP at the indicated concentrations and spiked 1:200 with γ-[P]ATP (MP Biochemical, Inc.). Reactions were terminated by the addition of TCA, and insoluble protein was bound to glass microfiber filters for quantification by liquid scintillation counting. Data was fitted to Lineweaver-Burke double reciprocal plots. In experiments to determine the Km for ATP during eIF2α-NTD phosphorylation, eIF2α-NTD was at a concentration of 20 μM. In experiments to determine the K0.5 for eIF2α-NTD, the ATP concentration was at 100 μM. Bacterially expressed GST-PERK or GST-PERK Δ-loop was dephosphorylated by incubation with λ-phage phosphatase. Subsequent incubation with γ-[P]ATP led to rephosphorylation and the incorporation of P; unincorporated label was removed by gel filtration and dialysis. Samples of labeled kinase were then incubated at 37°C in the presence of limiting concentrations of λ-phage phosphatase with or without substrate addition. Samples were taken at intervals, and protein was precipitated by TCA. Free, released P was measured by liquid scintillation counting. Unless specified otherwise, copurification experiments were performed in buffer containing 0.1% Triton X-100, 2 mM MgCl, 20 mM Tris, pH 7.4, and NaCl 150 mM. Glutathione–Sepharose beads and Ni-NTA agarose beads were washed three times in the same buffer. When Ni-NTA beads were used, 10 mM imidazole was included during binding and 20 mM during washes. Binding took place at 4 or 30°C for 30 min. Fig. S1 shows that active wild-type PERK cytosolic domain migrates in oligomeric form during size exclusion chromatography, whereas an inactive mutant remains monomeric. Fig. S2 shows the saturability of eIF2α-NTD binding to PERK and suggests that its stoichiometry is not 1:1. Online supplemental material is available at .

In eukaryotes, the 26S proteasome handles the majority of regulated proteolysis and is pivotal for the proper functioning of the cell (; ). One important function of selective proteolysis is to remove misfolded proteins. For example, in the ER, misfolded proteins are eliminated by a stringent quality-control process termed ER-associated protein degradation (ERAD; ; ). Only properly folded proteins are allowed to proceed to their destination to carry out their physiological functions. Most proteins that are targeted to the proteasome for degradation are first modified by the ubiquitin (Ub) system (; ). Specifically, successive Ub molecules join to form a Ub chain on the substrates through the concerted actions of several enzymes, including a Ub-activating enzyme (E1), a Ub-conjugating enzyme (E2), and a Ub protein ligase (E3). The ubiquitylated substrate is then delivered to and degraded by the 26S proteasome. Many components involved in the recognition and Ub conjugation of ERAD substrates have been identified, such as E2s and E3s (; ). How the ubiquitylated proteins are transferred to the proteasome remains elusive (). Two interacting proteins, Png1 and Rad23, are suspected to play important roles in the degradation of ERAD substrates (for review see ). Png1 is a highly conserved protein that resides mainly in the cytosol but also in the nucleus (; ). Functional studies suggest that Png1 is the primary, if not the only, deglycosylating enzyme in the cytosol (; ; for review see ). Many ERAD substrates are -glycosylated in the ER (for reviews see ; ). Computer modeling suggests that the bulky N-linked oligosaccharide must be cleaved off the substrate before its degradation to maintain the efficiency of the proteasome (). Coincidentally, ERAD substrates (e.g., class I myosin heavy chains [MHCs]) are found to be deglycosylated in the cytosol upon the inhibition of proteasome activity (). Png1 preferentially deglycosylates misfolded proteins in vitro (; ) and in cell extracts upon the overexpression of Png1 or glycoproteins (). However, a definitive ERAD substrate that requires Png1 for its degradation in vivo has not been identified. The turnover of glycosylated ERAD substrates, including a mutant carboxypeptidase Y (CPY), T cell receptor α chain, and class I MHC, was not drastically affected in yeast Δ mutant, the Png1 small interfering RNA cell lines, or cells treated with the Png1 inhibitor (; ; ). These results raise significant doubts about the requirement of Png1 in proteolysis in vivo. Rad23 belongs to a family of proteins that contains both the Ub-associated (UBA) domain and a Ub-like (UBL) motif (; ; ). Notably, the UBA motif binds specifically to Ub chain/conjugates in vivo and in vitro (; ; ; ). The UBL motif directly binds the proteasome (; ). The loss of Rad23 leads to the stabilization of a Ub fusion degradation (UFD) substrate (; ) and the cell cycle inhibitors Sic1 and Far1 (), and the homologues of Rad23 are involved in the degradation of the Cdk inhibitor Rum1 () and the tumor suppressor p53 (). The stabilized substrates in the mutant cells are fully ubiquitylated, suggesting that Rad23 functions at a postubiquitylation but preproteasome step (; ). Importantly, Rad23 is required for the formation of the proteasome–Ub conjugates complex (; ). Therefore, Rad23 has been proposed to facilitate the substrate transfer to the proteasome. The mechanism underlying the substrate specificity of Rad23 remains poorly defined. In degrading two ERAD substrates (Deg1-Sec62 and Hmg2), Rad23 binds Ufd2, which is a Ub chain elongation factor, and together they couple substrate ubiquitylation and degradation (; ). However, the role of the Png1–Rad23 pathway in ERAD is far from clear (for review see ). In this paper, we identify the first in vivo substrate of the Png1- and Rad23-dependent pathway as glycosylated protein ricin A chain (RTA), an Hrd3–Hrd1–dependent ERAD substrate. We show that the XPC binding (XPCB) domain–mediated Png1–Rad23 interaction is important for the degradation of glycosylated RTA. Furthermore, we find that both the Ub chain binding activity of Rad23 and the deglycosylation activity of Png1 are required for efficient degradation of glycosylated RTA, suggesting that the Png1–Rad23 complex couples substrate deglycosylation and degradation. Interestingly, Ufd2, another Rad23 binding protein, is required for the degradation of Deg1-Sec62 but not RTA. Our results suggest that Rad23 binds various cofactors (e.g., Ufd2 and Png1) to regulate distinct proteolytic substrates. Finally, we more generally propose that the substrate selectivity of Ub binding proteins (e.g., Rad23, Rpn10, and Cdc48) may be determined by various protein–protein interactions. Rad23 interacts with Png1 in yeast and mouse (; ). Mutation analysis indicated that the COOH-terminal fragment of Rad23 containing its substrate-recognition domain (i.e., UBA) binds Png1 (). If the Ub chain binding (UBA) domain directly binds Png1, Png1 may be a proteolytic substrate of Rad23. To test for this possibility, we examined the stability of Png1 in wild type and mutant. We found that Png1 is stable in both wild-type and mutant cells (), suggesting that Png1 and Rad23 may form a stable complex in regulating ERAD. To define the role of the Png1–Rad23 complex, it is critical to determine the domain of Rad23 responsible for Png1 binding. Derivatives of Rad23 containing various functional domains () were tested in the GST pull-down assay for interaction with Png1 (). Specifically, Rad23, Rad23, and Rad23 were separately fused to the COOH terminus of GST and purified from (). Consistent with a previous study (), the Rad23 fragment binds Png1. However, the COOH-terminal UBA domain alone is not sufficient for Png1 binding (). To further examine the binding between Png1 and Rad23 in vivo, derivatives of Rad23 () were tested for interaction with Png1 by the two-hybrid assay (). Interestingly, our findings reveal that the XPCB domain (amino acids 250–307) is sufficient for the interaction with Png1 (), which is consistent with a recent structural analysis of the Png1–Rad23 complex (). The XPCB domain binds Rad4 (called XPC in mouse; ) and is essential for the functioning of Rad23 in DNA repair (). Note that Rad4 is not part of the Png1–Rad23 complex (). The results suggest a novel function of the XPCB domain outside of DNA repair. Because Png1 is not a substrate of Rad23, we hypothesized that the Png1–Rad23 complex may regulate substrate proteolysis. What substrates do Rad23 and Png1 share? The UBA/UBL proteins Rad23 and Dsk2 were recently shown to be required for the efficient turnover of the ERAD substrates CPY () and Deg1-Sec62 (). However, we could not detect obvious stabilization of CPY in Δ cells and cells lacking and under the assay condition (unpublished data), and Deg1-Sec62 does not require Png1 for its degradation (see ). The ERAD pathway can be exploited by viruses (e.g., human cytomegalovirus) and toxins (; ). We found that RTA is a substrate of Png1. Ricin is a plant protein toxin consisting of A and B subunits (; ). RTA is the catalytic subunit of ricin, which inhibits protein synthesis. Ricin uses multiple endocytic routes to enter cells and is then transported into the ER. Separated from the B subunit, RTA is degraded by the cytosolic proteasome in yeast, plants, and mammals (; ). RTA degradation in yeast requires the Sec61 translocon and the proteasome, suggesting that RTA is an ERAD substrate (). Because RTA is glycosylated (), we wanted to test whether glycosylated RTA is a substrate of the Png1–Rad23 complex in yeast. We appended Flag epitope to a misfolded RTA with a short deletion surrounding its active site to eliminate RTA-induced toxicity. RTA was fused to the yeast Kar2 signal sequence to be targeted to the ER. In wild-type yeast cells, we detected two immunoreactive bands corresponding to RTA modified with no (g0 form) or one (g1 form) sugar chain ( , lanes 1 and 2). To ascertain that the g0 form does not contain untranslocated RTA, RTA was expressed in the mutant that is defective in the cotranslational translocation of ER proteins (). We detected an endoglycosidase H (EndoH)–resistant form of RTA located between the g1 and g0 bands (, lanes 3 and 4), which was likely the product of defective translocation, suggesting that RTA is efficiently translocated in wild-type cells (, lanes 1 and 2). To identify the components critical for the degradation of RTA, we evaluated RTA degradation in cells lacking or , which define two major ERAD pathways in yeast (; ). Although Doa10 is a single-component E3, Hrd3 and -1 (a RING [really interesting new gene] finger–containing protein) form an E3 complex. We found that glycosylated (g1) but not nonglycosylated (g0) RTA was stabilized in the mutant, suggesting that the g1 and g0 forms of RTA are degraded by different pathways (). Glycosylated RTA was also stabilized in cells lacking (unpublished data), indicating that glycosylated RTA is a substrate of the Hrd1–Hrd3 E3 complex. Furthermore, the Ufd1–Cdc48–Npl4 complex has been proposed to extract ubiquitylated proteins out of the ER (). We found that both the g1 and g0 forms of RTA were stabilized in mutant cells (). We next investigated RTA turnover in yeast cells lacking , , and/or . Significantly more (g1 + g0) RTA proteins accumulate in cells lacking or (), suggesting that RTA is a proteolytic substrate of Png1 and Rad23. The half-life of glycosylated RTA is ∼15 min in wild-type cells, compared with ∼45 min in Δ and >90 min in Δ (). Importantly, the degradation kinetics of the g0 form is similar in wild-type, Δ, and Δ Δ cells (), suggesting that the degradation of the g1, but not the g0, form of RTA is impaired in Δ and Δ Δ cells. ). Glycosylated RTA clearly requires Png1 for its degradation (). We note that the disappearance of glycosylated RTA is faster in Δ than that in the Δ mutant (), suggesting that Rad23-independent Png1 function may exist. Further, the g0 form is degraded faster in Δ than in wild-type or Δ cells (). Therefore, some of the g1 RTA may be converted to g0 instead of being degraded by the Png1–Rad23 pathway. The parsimonious interpretation of the available data is that there are two pools of Png1, Rad23 bound and Rad23 free, which define two parallel pathways for the disappearance of g1 RTA. More specifically, the degradation pathway of g1 RTA is mediated by the Png1–Rad23 complex, and the deglycosylation pathway is performed by Rad23-free Png1. As a result, some of the g0 RTA in wild-type and Δ cells is derived from g1 RTA. This would explain the faster disappearance of g0 RTA in Δ than in wild-type or Δ cells (). It is important to note that the profile of g0 RTA is similar in wild-type and Δ cells (), suggesting that an increased amount of Rad23-free Png1 in Δ cells does not enhance the deglycosylation of g1 RTA. (In , we show that Png1 stability is not altered in Δ cells.) Therefore, the pool of Rad23-free Png1 that deglycosylates g1 RTA in wild-type cells is likely in complex with other proteins or at a location different from the cytosolic Png1–Rad23 complex (). This model of two parallel Png1 pathways could also explain the slower disappearance of only g1 RTA in Δ cells than in wild-type cells () as the degradation pathway, but not the deglycosylation route, for g1 RTA is abolished in Δ cells. RTA degradation proceeds normally in cells lacking (unpublished data). Deletion of did not further stabilize RTA in Δ cells (). It is worth noting that other previously tested Rad23 substrates (e.g., Ub-V-βgal, Hmg2, and Deg1-Sec62) are more stable in double-mutant cells than in either single mutant, suggesting that Rad23 and Dsk2 have redundant roles (; ). Our results show that Rad23 and Dsk2 can play nonoverlapping functions in substrate proteolysis. Although computer modeling suggests that bulky sugar chains may clog up the proteasome (), and ERAD substrates are found to be deglycosylated in cytosol upon the inhibition of the proteasome activity (), it remains to be demonstrated that deglycosylation is indeed a prerequisite of proteasome-mediated degradation. In fact, an inhibitor of Png1 activity blocks deglycosylation but not degradation of class I MHCs (). To examine the role of Png1-mediated deglycosylation in ERAD, we used a His218Tyr mutation that abolishes the enzymatic activity of Png1 (). The Png1 His218Tyr mutant interacts with Rad23 (). We found that RTA degradation in Δ cells is restored by wild-type but not mutant Png1 (), indicating that the enzymatic activityof Png1 is essential for its function in ERAD. Is Ub chain binding activity required for the functioning of Rad23? It is conceivable that Rad23 can bring substrates processed by Png1 to the proteasome through the UBL domain without the help of the UBA domain (). In this case, Ub chains attached to RTA are not required for proteasome targeting but rather are important for other functions, such as facilitating substrate translocation (). To establish the role of the UBA domains of Rad23 in glycoprotein turnover, we introduced the double mutation L183A and L392A into the UBA domains of Rad23 to inactivate its Ub binding ability (; ). The Rad23 mutant still binds Png1 and Rpn1 (), suggesting that the mutation specifically eliminates its Ub binding activity. The plasmid bearing the wild type or mutant was cotransformed with a plasmid expressing Flag-RTA to Δ cells. Wild-type but not mutant Rad23 fully restored the degradation of glycosylated RTA in Δ cells (), suggesting that the Ub chain binding activity of Rad23 is critical for its functioning in degradation. Our results support the essential role of Ub chain binding activity of Rad23 for substrate proteolysis in vivo. To establish the significance of the Rad23–Png1 complex, we looked for mutations in Rad23 that would alter its binding to Png1. We compared sequences of XPCB domains of various Rad23 to identify highly conserved residues among them (). We constructed eight variants of the XPCB domain of Rad23, each containing a mutation in one of the conserved residues (). These mutations were introduced into full-length Rad23 linked to the Gal4 DNA binding domain. The two-hybrid assay was used to determine their interactions with Png1 (XPCB domain dependent), Rad4 (XPCB domain dependent), Rpn1 (UBL domain dependent), Ufd2 (UBL domain dependent), Ub (UBA dependent), and Rad23 (UBA-dependent self-dimerization; ). Four mutations—Q267V, Q267Y, N272A, and N297G—did not affect the bindings to Png1 and Rad4 (unpublished data). Three mutations—D261A, P273G, and L280A—reduced but did not eliminate the interactions with both Png1 and Rad4. These three mutations may significantly alter the overall structure of Rad23 and were not further analyzed. Interestingly, the L276Q mutation specifically abolished its interaction with Png1 but maintained the bindings with other partners—Rad4, Rpn1, Ufd2, Ub, and Rad23 ()—indicating that the Png1 binding is separable from these other interactions. The defective binding of the Png1 and Rad23 L276Q mutant was also confirmed by coimmunoprecipitations (). Therefore, the L276Q mutant was further characterized to determine the specific role of the Png1–Rad23 complex in various Rad23-dependent pathways. To understand the function of the Png1–Rad23 complex in vivo, the mutation L276Q was introduced into full-length Rad23 under the control of its own promoter on a low-copy plasmid to avoid potential artifacts caused by overexpression. Rad23 is known to function as a DNA damage recognition factor in nucleotide excision repair through the binding of Rad4 (; ). We examined the UV sensitivity of Δ cells expressing the allele. Consistent with the unaltered binding between Rad4 and the Rad23 mutant (), UV resistance was restored in Δ cells expressing the Rad23 mutant (), suggesting that the Rad23 mutant is sufficient in fulfilling the function of Rad23 in the DNA repair. The UBA/UBL proteins Rad23 and Dsk2 are important for yeast survival under several stress conditions. Cells lacking both and are known to be unviable at 37°C (). We found that the L276Q mutant restored the growth of Δ Δ cells at 37°C (), suggesting that the Png1–Rad23 interaction is not required for cell growth at higher temperature. Rad23 is also known to regulate the UFD pathway in yeast through its UBL domain–mediated association with Ufd2 (). Is the XPCB domain required for a functional UFD pathway? The plasmids bearing variants were cotransformed with a plasmid expressing the UFD substrate Ub-V-βgal () to Δ cells. We used the LacZ assay to gauge the effects of the XPCB mutation on the intracellular concentration of the UFD substrate (). Δ cells had much higher levels of β-gal activity than wild-type cells (). The expression of in the Δ cells restored the low levels of β-gal activity (). Interestingly, lower levels of β-gal activity were detected in Δ cells expressing the mutant (), suggesting that Png1 binding is not important for the degradation of the UFD substrate. Next, we examined the effect of the L276Q mutation on the degradation of RTA by measuring the stability of RTA in yeast cells with pulse-chase assay. Glycosylated RTA is stabilized in Δ cells (). The expression of wild-type but not L276Q mutant in the Δ cells restored rapid degradation of glycosylated RTA (), suggesting that the amino acid residue L276 is essential for the proteolytic function of in ERAD. Differential effects of L276Q on two distinct proteasomal substrates, RTA and UFD, support that Rad23 uses multiple means in regulating various substrates (see the following paragraph). Our combined results suggest that the XPCB domain–mediated Png1 interaction is essential for the function of Rad23 in the degradation of glycosylated RTA but not in DNA repair, cell survival at 37°C, and the UFD proteolytic pathway. Combined ( and ), our results suggest that the Png1–Rad23 complex can couple substrate deglycosylation and degradation. Recently, Ufd2, a Ub chain elongation factor, was shown to be required for the efficient turnover of two membrane-associated ERAD substrates, Hmg2 and Deg1-Sec62 (). We previously demonstrated that Ufd2 binds Rad23 and the resulting complex is important for the functioning of Rad23 in the UFD pathway (). Does Png1 bind Ufd2? We did not detect the interaction between Ufd2 and Png1 (). Furthermore, Ufd2 is not critical for the degradation of glycosylated RTA ( ). We found that Deg1-Sec62 is also glycosylated in vivo (). We evaluated Deg1-Sec62 degradation in cells lacking , and we found that the degradation of Deg1-Sec62 was unaltered in Δ cells (), indicating that not all glycoproteins are degraded by the Png1 pathway. Our results suggest that Rad23 regulates the degradation of distinct substrates through its interactions with various cofactors that are involved in specific proteolytic pathways. In this paper, we show that glycosylated RTA is an in vivo substrate of the Png1–Rad23 degradation pathway. Furthermore, we demonstrate that the efficient degradation of glycosylated RTA also requires the association between Png1 and Rad23, which in turn binds ubiquitylated substrate and removes -glycans from the substrate. The physiological significance of Png1 in ERAD is not clear because the in vivo degradation of misfolded glycoproteins was not significantly altered in cells defective of Png1 activity (; ). Because many ERAD substrates are deglycosylated before their degradation (; ; ), Png1 has long been suspected to play a critical role in this process (for reviews see ; ). Png1 can remove -glycans from several glycosylated proteins in vitro (; ) and in vivo when Png1 is overexpressed (). In a recent review, Png1 was proposed to regulate glycoprotein turnover universally (). However, unaltered degradation of glycosylated class I MHCs in cells treated with a Png1 inhibitor and in Png1 small interfering RNA cell lines challenged the requirement of Png1 in glycoprotein turnover (; ). Our results demonstrate that Png1 plays a key role in the degradation of a subset of glycosylated ERAD substrates. We also show that not all glycoproteins are degraded by the Png1 pathway (), suggesting that Png1 is not universally required for glycoprotein turnover. It will be of interest to determine how these Png1-independent glycoproteins are degraded, such as the regulators involved and whether other deglycosylating activities are required. The Png1–Rad23 complex directly couples protein deglycosylation and degradation ( and ), thereby ensuring rapid turnover of misfolded glycoproteins and maintaining more efficient proteasomes. At the molecular level, what is the biological function of the Png1–Rad23 interaction? Nuclear magnetic resonance studies indicate that Rad23 contains four structured regions: UBL, XPCB, and two UBA domains (; ). Further, the intramolecular interactions between the UBL element and the UBA domains keep Rad23 in a closed conformation. Interestingly, the interaction with the proteasome induces Rad23 to adopt an open conformation () that is active for proteolysis. Rad23 exists in a stable complex with Png1. It is tempting to speculate that binding of Png1 may also open up the conformation of Rad23, which in turn facilitates its bindings to the proteasome and/or ERAD substrates. We propose that the XPCB domain–mediated Png1–Rad23 interaction facilitates not only substrate recognition of Rad23 and/or Png1 but also the direct transfer of deglycosylated ERAD substrates to the proteasome, which binds the UBL domain of Rad23 ( ). Rad23 also binds Ufd2, a Ub chain elongation factor. The Ufd2–Rad23 association is important for the degradation of UFD substrates (), the transcription factor Spt23, and two ERAD substrates (i.e., Hmg2 and Deg1-Sec62; ). The Ufd2–Rad23 interaction likely couples substrate ubiquitylation and degradation (; ). At present, it remains to be determined why some ERAD substrates (e.g., RTA) require Png1 for their degradation, whereas other Rad23-regulated ERAD substrates (e.g., Deg1-Sec62 and Hmg2) are degraded by the Ufd2 pathway. It is possible that the nature of glycosylation (e.g., the attachment site, structure, and number of glycans attached) may influence the substrate degradation. There exists a wide array of ERAD substrates, and it will be important to understand the factors that divert various glycoproteins to distinct ERAD pathways (; ). One obvious contributing factor is the localization of the substrates because RTA is soluble and Deg1-Sec62 is embedded in the ER membrane. Recent findings suggest that at least two checkpoints are used to sort out ERAD substrates to different degradation pathways based on the location of the misfolded domain (e.g., membrane, lumen, or cytosol) and the topology of the protein (; ). We are systematically determining the involvements of Rad23 and Png1 in the turnover of these different types of ERAD substrates. Our results not only reveal how Rad23 regulates the degradation of various substrates but also may affect the studies on other Rad23-like proteins. An increasing number of proteins have been shown to bind Ub and/or Ub chains in yeast and human. Many of these proteins play important regulatory functions in diverse cellular pathways because cells without them exhibit different phenotypes, suggesting distinct substrate selectivity (; ). How do these Ub binding proteins achieve their substrate specificity? It is worth noting that in regulating two distinct Rad23-dependent proteolytic substrates (i.e., RTA and Deg1-Sec62), Rad23 uses the XPCB domain and the UBL motif to interact with two key players, Png1 and Ufd2, in these pathways (). Our results indicate that these interactions are essential for the functioning of Rad23 in these processes (; ). More generally, various protein–protein interactions may be used to facilitate the functions of Rad23 and other Ub binding proteins (e.g., Rpn10 and Cdc48) in Ub-mediated processes. Rad23 binds Rad4/XPC in nucleotide excision repair (; ; ; ). We have also uncovered a novel function of the XPCB domain of Rad23 in Ub-mediated proteolysis. Rad23 and Png1 are highly conserved from yeast to human. Two Rad23 homologues, hHR23A and -B, exist in humans. Interestingly, XPC is mainly found in complex with hHR23B instead of -A (). It is possible that Png1 and/or Ufd2 mainly associate with one of the two homologues. PJ69-4A (Δ Δ ) was used for two-hybrid assays. Isogenic strains W303-1A, Δ, and Δ ( ) were previously published (). Yeast strains YHR114 () and YHR132 () were constructed by replacing with in strains W303-1A and Δ, respectively. Cultures were grown in rich (yeast extract/peptone/dextrose [YPD]) or synthetic media containing standard ingredients and 2% glucose (SD medium), 2% raffinose (SR medium), or 2% raffinose + 2% galactose (SRG medium). Yeast strains lacking , , , or were obtained from Open Biosystems. and strains were obtained from E. Johnson (Thomas Jefferson University, Philadelphia, PA) and D. Ng (National University of Singapore, Singapore, Singapore). The plasmids containing wild type and its derivatives or His6-tagged Png1 and Png1-1 (H218Y) mutant were previously described (; ). was amplified by PCR to incorporate the Flag epitope and cloned to the 3′-end of the promoter in pRS414Gal1 for its expression. A nontoxic allele of RTA () was amplified by PCR to incorporate the Flag epitope at its COOH terminus and fused downstream to the yeast Kar2 signal sequence to pRS416Gal1 vector to construct an ER version of RTA. Mutations in the XPCB domain or UBA domains of Rad23 were obtained using the Quick Change mutagenesis kit (Stratagene). Yeast cells carrying plasmids that expressed Flag-tagged Png1 or RTA from the P promoter were grown at 30°C to an OD of ∼1 in SR-ura medium with auxotrophic supplements and 2% raffinose as the carbon source. Expression of Flag-Png1 or Flag-RTA was induced with galactose for 1 h and then repressed by the addition of 2% glucose. Samples were withdrawn at the indicated time points and harvested by centrifugation. Proteins were extracted and processed for immunoprecipitation with Flag antibody (Sigma-Aldrich), followed by SDS–9% PAGE, as described previously (). Immunoblots were probed with anti-Flag antibody (Sigma-Aldrich) and goat anti–mouse HRP conjugate and were developed using ECL reagents (GE Healthcare). The stable protein Rpt5 was used as a loading control to ensure that an equal amount of extracts was used in expression shutoff experiments. Pulse-chase analysis was done as described previously (). Yeast cells carrying plasmids that expressed Flag-tagged RTA or Flag-tagged Deg1-Sec62 from the P promoter were grown at 30°C to an OD of ∼1 in SRG medium with auxotrophic supplements and 2% raffinose and galactose as the carbon sources. Cells were harvested, washed with 0.8 ml SRG, resuspended in 0.4 ml SRG, and labeled for 8 min with 0.16 mCi of S-Express (PerkinElmer), followed by centrifugation and resuspension of cells in SD medium with 4 mM methionine and 2 mM cysteine. 0.1-ml samples were taken at the indicated time points and processed for immunoprecipitation with Flag beads (Sigma-Aldrich), followed by SDS-PAGE and autoradiography. The amount of proteins was quantified by phosphorimager analysis. GST fusion proteins were purified as previously described (). GST fusion protein or GST alone (∼2 μg) was mixed with yeast extracts containing His6-Png1 in 200 μl of binding buffer () and incubated with 10 μl (bed volume) of glutathione-agarose beads (Sigma-Aldrich) for 2 h at 4°C. The beads were washed three times with the binding buffer, followed by SDS-PAGE of the retained proteins and immunoblotting with anti-His6 antibody (GE Healthcare). W303-1A (wild-type) cells carrying either pRS414Gal-Png1 (expressing Flag-tagged Png1) and pYes2 vector, pYes2-Ufd2 (expressing myc-tagged Ufd2) and p414Gal vector, or pYes2-Ufd2 and pRS425Gal-Rad23 (expressing Flag-tagged Rad23) were grown in the galactose-containing SG medium to an OD of ∼1, followed by preparation of extracts, immunoprecipitation with beads linked to specific antibodies indicated, SDS–8% PAGE, and immunoblotting, separately, with anti-Flag (Sigma-Aldrich) and anti-myc (Covance).

xref italic sup #text To determine at which stage of neuronal development Stau2 is expressed, we performed semi-quantitative RT-PCR on RNA extracted from cultured hippocampal neurons at different days in vitro (DIV) (see Fig. S1 A, available at ). Stau2 mRNA could be detected at all stages examined, indicating that Stau2 is expressed throughout neuronal development. We then performed loss-of-function analyses using RNAi in polarized hippocampal neurons to investigate the function of Stau2. Two 19mer oligonucleotides directed against different regions of the Stau2 cDNA were cloned into the pSUPER vector (). Their expression yields short hairpin RNAs (shRNAs) that are subsequently converted into small interfering RNAs (siRNAs). Cotransfection of HeLa cells with either of the two plasmids, si2-1 (unpublished data) or si2-2 (Fig. S1 B), together with Stau2-EYFP significantly down-regulated Stau2-EYFP expression as assessed by fluorescence microscopy. In contrast, neither of these plasmids affected the level of the paralogous protein, Stau1-EYFP (Fig. S1 B). Furthermore, a control Stau2 siRNA with 5 bp substitutions (mismatch siRNA) did not affect the level of Stau2-EYFP expression (unpublished data). We then determined the level and the specificity of Stau2 down-regulation in neurons by Western blot analyses. Primary hippocampal neurons were transfected by nucleoporation () before plating with plasmids yielding shRNAs against Stau2 (si2-2) and an unrelated protein, CDC10 (siCDC10, negative control). As additional controls, neurons were either mock treated (citrine) or transfected with pSUPER vector expressing the mismatch sequence of si2-2 (mis) ( ). Neurons were allowed to develop for 3 d and processed for Western blot analysis. Only siRNAs directed against Stau2 significantly down-regulated the expression of three Stau2 isoforms (62, 59, and 52 kD), whereas all other control plasmids did not (). The same results were obtained by using short interfering oligos (Dharmacon) (Fig. S1 C). Importantly, we did not observe any compensatory change in Stau1 expression levels upon Stau2 down-regulation (, bottom). To verify down-regulation of Stau2 in mature cells, 15 DIV neurons were cotransfected with plasmids encoding cyan fluorescent protein (ECFP) and si2-2 RNA, and immunostaining was performed for Stau2 3 d after transfection. Both Stau2 shRNAs, si2-2 and si2-1, reduced or completely abolished Stau2 expression in neurons (, and unpublished data). Stau2 signal in dendrites was strongly reduced in Stau2 down-regulated neurons compared with untransfected cells (, enlarged insets). Neither mismatch shRNA against Stau2 () nor shRNA against red fluorescent protein (RFP) (unpublished data) affected Stau2 expression. Moreover, the expression level of Stau1 remained unchanged (, middle), suggesting that this RNAi approach targeting Stau2 was specific. The double-stranded RNA-binding protein Staufen plays an essential role in oogenesis, early embryonic patterning, and in establishing cellular asymmetry in neuroblasts (). Despite a plethora of information on the roles of Staufen in , the functions of the mammalian homologues are still largely unknown. Stau2 is preferentially expressed in the central nervous system and here particularly in neurons (), where it is present during all stages of neuronal development and in mature neurons (Fig. S1 A and unpublished data). Previous work has implicated Stau2 in dendritic mRNA transport, which is assumed to occur solely in the cytoplasm (; ; ; ; ). A more detailed analysis, however, provided the first evidence that both Stau1 () and two isoforms of Stau2, Stau2 and Stau2 (; ), shuttle between the nucleus and the cytoplasm using separate export pathways, suggesting distinct roles for the different Stau2 isoforms (). Aside from these indirect observations, no direct function has been assigned to any of the mammalian Staufen proteins. Because they are assumed to play a role in dendritic mRNA transport, a process generally assumed to contribute to synaptic plasticity (), we decided to investigate the function of the brain-specific Stau2 in polarized hippocampal neurons using an RNAi approach. Down-regulation of Stau2 caused a loss of dendritic spines and the appearance of extended filopodia; a drastic reorganization of the actin cytoskeleton in dendrites; a significant reduction in β- RNA expression level in both the cell body and in dendrites of Stau2 down-regulated mature neurons; and an attenuation of excitatory synaptic transmission due to a decreased postsynaptic responsiveness. Dendritic spines are morphological specializations that protrude from the main shaft of dendrites contacting presynaptic nerve terminals (). Although their precise function is still unclear, the generally accepted hypothesis has been put forward that “spines create a micro compartment with a range of properties that enable them to operate as a multifunctional integrative unit” (). Dendritic spines can thereby act as semi-autonomous chemical compartments to segregate postsynaptic responses (e.g., elevated calcium) to spatially and temporarily modify molecules and to newly synthesize proteins upon synaptic activation. The formation and maintenance of dendritic spines are important for both neurogenesis and neuronal activity in the mammalian brain (; ). Considerable progress has been made toward identifying the molecules that might control dendritic spine growth and maturation (; ). Principally, the newly identified proteins fall into four categories: membrane proteins (e.g., glutamate receptors, cell adhesion molecules); scaffold proteins (e.g., PSD95, Shank, Homer); cytoskeletal-binding/regulating proteins (e.g., actin, drebrin, spinophilin, SPAR); and cytoplasmic proteins (e.g., kinases/phosphatases and other enzymes). Recently, two other RNA-binding proteins have been implicated in dendritic spine morphogenesis, the Fragile X mental retardation protein (FMRP) and translocated in liposarcoma (TLS) (; ). Both in Fragile X mental retardation patients () and in adult knockout mice (), unusually long and thin dendritic spines with increased density were observed. The putative functional consequences of this phenotype, e.g., on learning and memory, are currently under investigation. The increased dendritic spine density may be attributed to the absence of an activity-dependent translational repression by FMRP (). TLS, a component of heterogeneous nuclear ribonucleoprotein complexes and a nucleocytoplasmic shuttling protein, accumulates in spines at excitatory synapses upon mGluR5 activation. Hippocampal neurons derived from knockout mice, which die shortly after birth, also exhibited abnormal spine morphology: lower spine density and filopodia-like thin and long cytoplasmic protrusions (). A third RNA-binding protein, the zipcode binding protein 1 (ZBP1), is involved in localization of the β- message to growth cones of developing neurons and to dendrites of mature hippocampal neurons, respectively. This protein has been found to translocate from dendritic shafts into dendritic spines upon synaptic activity (). Whether it is involved in dendritic spine morphogenesis, however, is unclear at present. Our studies provide first evidence that another conserved RNA-binding protein, the brain-specific Stau2, is crucial to dendritic spine formation and maintenance, suggesting that the function of RNA-binding proteins is essential to this process. How could a defect in Stau2 cause impairment in dendritic spine morphogenesis? We have recently shown that Stau2 assembles into RNPs that move along microtubules into dendrites localizing in the proximity of the dendritic spines (; see also ). Stau2 particles, however, do not colocalize with PSD95, a postsynaptic marker, but instead are restricted to the dendritic shaft (see ). It is, however, possible that synaptic stimulation causes a translocation of Stau2 RNPs from the dendritic shaft to dendritic spines, as reported for TLS, FMRP, and ZBP1 (; ; ). This will be subject to future investigation. It is interesting to note that the longest Stau2 isoform, Stau2, displays a greater ability to rescue the observed loss of dendritic spines in Stau2 down-regulated neurons. This ties in with previous findings indicating what might be distinct roles of the different Stau2 isoforms in mammalian cells. The 62-kD isoform has been shown to preferentially accumulate in the nucleolus upon down-regulation of exportin-5 (). Because exportin-5 has been shown to be the export factor for microRNAs from the nucleus (; ), this suggests a possible involvement of Stau2 in microRNA trafficking and translational control at the synapse (). There is currently very little information on the composition of endogenous mammalian Stau2 RNPs. This is in sharp contrast to the more ubiquitously expressed Stau1 RNPs, for which both interacting proteins and putative cargo RNAs have recently been reported (; ; ; ; ). To our knowledge, not a single RNA has been identified as a bona fide cargo for Stau2. Our findings that down-regulation of Stau2 results in a reorganization of the actin cytoskeleton in dendrites and also affects the levels of β- mRNA in both the cell body and the dendrites allow us to draw two important conclusions. First, there is a yet to be identified link between Stau2 and the actin cytoskeleton. It is particularly interesting to note in this context that F-actin, consisting of β- and γ-isoforms of actin, is highly concentrated in dendritic spines and that the actin cytoskeleton () plays an important role in activity-dependent blockage of dendritic spine motility (). Second, Stau2 may bind to β- RNA, thereby influencing either its stability and/or its dendritic transport. It is tempting to speculate that the down-regulation of Stau2 may cause a reorganization of the (dendritic) actin cytoskeleton by either affecting the stability of Stau2-interacting transcripts (e.g., β-) or by controlling the translation of transcripts coding for key players in the observed actin dynamics. There are several lines of evidence that support this notion. A Stau1-specific mRNA decay pathway has recently been discovered as a means for cells to down-regulate the expression of transcripts bound by Stau1 (). In addition, Staufen functions in translational derepression of mRNA once the mRNA has been localized to the posterior pole (, ). Furthermore, a recent study provided first direct evidence that Stau1 may play a similar role in mammals by facilitating translation initiation (). The challenge will now be to characterize a possible molecular mechanism of how Stau2 may affect the β- mRNA stability and/or regulate local translation of β-actin. The link between Stau2 and the actin cytoskeleton may provide the basis for the functional alterations observed after down-regulation of Stau2. On one hand, the morphology of dendritic spines clearly depends on the presence of F-actin () and actin depolymerization leads to a loss of spines (). On the other hand, the geometry of spines is a major determinant of the responsiveness of synaptic non-NMDA glutamate receptors (). Thus, the reduction of mEPSC amplitudes correlates well with the molecular and morphological changes of Stau2 down-regulation and thus unveils this protein as a regulator of synaptic efficacy. Primers complementary to two distinct regions of rat Stau2 were cloned BglII and HindIII into both the pSUPER and pSUPERIOR vectors (Oligoengine; ). The sequences of the primers (Table S1, available at ) used in this study are available upon request. pEYFP-N1 vectors (Clontech/Invitrogen) expressing cleavage-resistant forms of Stau2 and Stau2 were created by mutagenesis using the QuickChange site-directed mutagenesis kit (Stratagene) according to the manufacturer's protocol. siRFP and misStau2 shRNAs served as negative controls. The GFP-PSD95 construct () was provided by Dr. M. Dailey (University of Iowa, Iowa City, IA) with the permission of Dr. D. Bredt (University of California, San Francisco, San Francisco, CA). To avoid cotransfection of too many plasmids, PSD95 was cloned in frame with the GFP cDNA into the pSUPERIOR vector that also contains the Stau2 shRNA (si2-2). The citrine (an EYFP variant) plasmid was provided by Dr. Virginie Georget (CNRS, Monpellier, France). The photo-activatable (PA) GFP () was provided by Dr. Jennifer Lippincott-Schwartz (National Institutes of Health, Bethesda, MD). The mutant version of Stau2 (dn2) was created as described in . RT-PCR was performed on total RNA isolated from cultured hippocampal neurons at different days in vitro (0, 3, and 23) using the RNeasy Kit (QIAGEN). The cDNAs were then amplified in the same PCR reaction and different numbers of PCR cycles were tested to ensure that amplification was not at the level of saturation. HeLa cells were cultivated and transfected as described in . Rat hippocampal neurons were cultured () and transiently transfected () with the plasmid expressing citrine together with pSUPER or pSUPERIOR vectors. Neurons were fixed with 4% PFA 3 d after transfection. For PSD95 experiments, 15 DIV neurons were cotransfected with vectors coding for GFP-PSD95 and si2-2, mis-Stan2, or dn2 in the ratio of 1:5 to avoid overexpression of GFP-PSD95. The following antibodies (incubation at RT for >1 h) were used: immunopurified rabbit anti-Stau1 antibodies (1 μg/ml); anti-Stau2 antibodies (1 μg/ml); monoclonal anti-PSD95 (1:1,000; Sigma-Aldrich) and polyclonal anti-synapsinI (1:1.500; Chemicon International). Cy3-coupled goat anti–mouse and anti–rabbit IgG antibodies (1:2,000) were used as secondary antibodies (Dianova). Phalloidin staining was performed as described in . Fluorescent images were acquired () using Axiovert 200M, Axiovert 100TV, and Axiophot microscopes (all Carl Zeiss MicroImaging, Inc.) equipped with the following objectives: 40× PlanApo oil immersion, 1.2 NA, or 63× PlanApo oil immersion, 1.4 NA (both Carl Zeiss MicroImaging, Inc.) and the following CCD cameras: Olympus F-View2 (Soft Imaging System), Spec-10 LN-1300 and CoolSnap HQ (both Princeton Instruments/Roper Scientific) and the following software: MetaMorph 6.3 (Universal Imaging Corp.) or AnalySIS Five (Soft Imaging System) and assembled using Adobe Photoshop 7.0. Pictures were not modified other than adjustments of scaling, levels, brightness, and contrast. To determine the length of dendritic spines, 15–30 EYFP-positive dendrites were randomly selected for each condition and the number and length of all protrusions was manually determined using MetaMorph 5.6. For GFP-PSD95–expressing cells, 10 cells per condition were analyzed. Only GFP-PSD95–positive puncta that reside beside the dendritic shaft were counted and expressed as puncta per μm dendrite length using Microsoft Excel. Cells expressing high levels of GFP-PSD95 were discarded. For statistical analysis, test was applied using Microsoft Excel, P values <0.001 (***) were considered as highly significant. The experimenter was not aware of the experimental conditions. To quantify β- mRNA levels, the fluorescence intensities of si2-2 and mis-transfected cell bodies were measured and normalized to the intensities of adjacent, untransfected neurons using MetaMorph 5.6. Three independent experiments with 15 cells per condition were evaluated. Dendritic particles were manually counted and the numbers per cell compared between si2-2 and mis-transfected neurons. mEPSCs were determined in whole-cell patch-clamp recordings at room temperature (20–24°C) on neurons at 18 DIV using an Axopatch 200B amplifier and the Pclamp 6.0 hard- and software (Axon Instruments; see ). The bathing solution contained (in mM) NaCl, KCl, CaCl, MgCl, glucose, and Hepes, and was adjusted to pH 7.4 with NaOH. Tetrodotoxin (TTX; 0.5 μM) and bicuculline methiodide (30 μM) were added to suppress action potential propagation and miniature inhibitory postsynaptic currents, respectively. Neurons were continuously superfused using a DAD-12 (Adams and List) application system. Electrodes were pulled from borosilicate glass capillaries (Science Products) using a Flaming-Brown puller (Sutter Instruments) to yield tip resistancies of 4.5–5.5 MΩ and were filled with a solution containing (in mM) KCl, CaCl (1.6), EGTA, Hepes, Mg-ATP, and Li-GTP, adjusted to pH 7.3 with KOH. 3 d after exposure to plasmids, mEPSCs were recorded from either transfected or nontransfected cells present within the same culture dish for periods of time sufficiently long to obtain at least 25 consecutive events. Thereafter, 10 μM cyano-2,3-dihydroxi-7-nitroquinoxaline (CNQX) was applied, which blocks all mEPSCs in the presence of Mg (). mEPSCs were evaluated using the Mini Analysis Program (Synaptosoft Inc.) and detection thresholds were adjusted for each cell by analysis of traces obtained in the presence of 10 μM CNQX. mEPSCs and their inter-event intervals were tested for normal distribution by a Kolmogorov-Smirnov test and then compared by a one-way analysis of variance. The results show arithmetic means±SEM, and P values below 0.05 were taken as indication of statistical significance. Endogenous β- mRNA was detected by two different approaches. First, a 500-bp RNA probe complimentary to bases 21–520 of rat β- mRNA (NM 031144) was used according to . After hybridization at 52°C, cells were blocked and the probes were detected using rhodamine-labeled anti-Dig Fab fragments (Roche) according to the manufacturer's instructions. Alternatively, a mix of four antisense oligonucleotides was applied (). These oligonucleotides were then modified with digoxigenin at the 3′-end by terminal transferase (according to the manufacturer's instructions; Roche). FISH was performed using a GAPDH antisense probe (, nucleotides 4–1233) as described in . Neurons were transfected by nucleoporation (Amaxa) with a total of 3 μg DNA (ratio of the shRNA and citrine was 1:1) or Dharmacon oligos alone according to the manufacturer's specifications (program O-03) with the following modifications: neurons were plated directly in growth medium and petri dishes were coated with 0.1 mg/ml poly-L-lysine in borate buffer (). Fig. S1: (A) Semi-quantitative PCR performed on RNA extract from E17 hippocampal neurons at different DIV. (B) RNAi for Stau2 in HeLa cells. (C) Western blot of nucleoporated neurons. (D) Quantitative analysis of the length of all protrusions. Fig. S2: Integrity of nuclei in transfected neurons. Table S1: List of oligonucleotides used in this study. Online supplemental material available at .

Skeletal muscle tissue is composed of multinucleated fibers that arise at defined periods of embryogenesis from fusion of myoblasts (). In particular, embryonic and fetal myoblasts, originating from different waves of myo-blasts (; ), give rise to primary (at about embryonic day [E] 11–12) and secondary (at about E15–16) fibers, respectively (). Subsequently, muscle masses undergo extensive growth in the fetal and postnatal period, and this growth is supported by specialized cells, the satellite cells, situated within a niche between the plasmalemma and the basal lamina of fibers (). Throughout myogenesis, a fine balance among proliferation, differentiation, and fusion is required for the correct formation of the definitive muscle units (; ). Many positive and negative signals responsible for the regulation of such a fine balance have been identified, acting at both embryo/fetal stages and postnatally. These include transcription factors, such as MyoD, Myf5, MRF4, and myogenin, as well as extracellular agonists and antagonists, such as members of the insulin-like growth factor (IGF) and TGFβ families, FGF, hepatocyte growth factor, and bone morphogenetic protein (BMP) and its antagonists follistatin and chordin (; ). The short-lived messenger nitric oxide (NO) regulates key functions of adult skeletal muscle, such as the activity of neuromuscular synapses, excitation–contraction coupling, vasodilation, glucose uptake, mitochondrial function and biogenesis, glycolysis, and phosphocreatine breakdown (; ; ; ; ; ; ; ). The possibility that NO plays a role in skeletal myogenesis is suggested by the observations that it participates in satellite cell activation (; ) and that its synthesizing enzymes, the NO synthases (NOSs), are developmentally regulated and may contribute to the myogenic program activated by IGF-II (; ; ; ). The precise role of NO in myogenesis and the signaling pathways acting downstream of it are, however, not known. In the present study, we investigated these aspects, both in vitro and in vivo, at different phases of myogenesis. Our results show that NO directly stimulates myoblast fusion through the up-regulation of follistatin, defining for the first time a link between NO and another key player in adult and embryonic myogenesis. We also found that the action of NO is limited to a defined time window and is mediated through a tightly regulated activation of guanylate cyclase and generation of cyclic guanosine monophosphate (cGMP), a physiological effector of NO (). Maintenance of cGMP signaling by treatment with 8 Br-cGMP leads to an increased fusion process with generation of hypertrophic myotubes and myofibers in vitro and in vivo. Overall, our results indicate a pivotal role of NO/cGMP in regulating myoblast fusion during muscle development. -nitro--arginine methyl ester (L-NAME), which is a broad spectrum inhibitor of NOS (). The fusion index was measured after 24, 48, or 72 h. , DETA-NO increased, whereas L-NAME decreased, the fusion index in concentration-dependent ways. These effects were specific because the corresponding amine, DETA, did not yield significant effects and the action of L-NAME was inhibited by 5 mM -arginine (unpublished data). The stimulation by DETA-NO and the inhibition by L-NAME were clearly detected after 24 h and were increased throughout the differentiation program, becoming statistically significant after 48 and 72 h of culture (). At 72 h, almost no fusion events were observed among myosin-expressing satellite cells in the presence of 5 mM L-NAME (). The fact that the effects of DETA-NO and L-NAME on fusion were already present after 24 h of treatment suggests that these events took place at an early stage in the differentiation program. To establish the time window of the NO action, 5 mM L-NAME and 50 μM DETA-NO (yielding a concentration of 120 ± 5 nM, = 5, measured with a NO electrode; ) were added to differentiating satellite cells at different time points. The fusion index was assessed after 72 h. As shown in , both DETA-NO and L-NAME were maximally effective in enhancing and preventing fusion, respectively, when added at the beginning of the differentiation process. The compounds were progressively less effective when added at later time points and almost completely ineffective when added after 16 h. Consistently, we found that the differentiation process was accompanied by an early increase in NOS activity that peaked at 8 h and decreased thereafter, returning to basal levels after 48 h (). We found that satellite cells express the endothelial (NOS III) and muscular (NOS Iμ) variants of the neuronal NOS. The levels of expression of NOS III and Iμ were unchanged throughout the differentiation process (), indicating that the changes in NOS activity were the consequences of activation and inhibition of enzyme activity and not of changes in protein expression. NOS II expression was never detected throughout the time window analyzed (unpublished data). Satellite cells cultured in vitro are asynchronous, with a small portion of them still persisting as undifferentiated cells even after several days in differentiating medium (; ). The increased fusion of satellite cells triggered by NO may be attributable to an effect of this messenger on the process itself or secondary to NO-dependent recruitment of the undifferentiated cells into terminal differentiation (). This should result in an increase in MyoD expression and a concomitant decrease in cell proliferation. To discriminate between these possibilities, we examined the effects of NO on these processes. As shown in , exposure of satellite cell cultures for 24 h to either 5 mM L-NAME or 50 μM DETA-NO changed neither the expression of the myogenic marker MyoD nor that of the DNA polymerase δ cofactor (PCNA), which is expressed during the S phase of the cell cycle. We consistently found no increases in either proliferation, measured by counting the number of nuclei in all cells, or the number of nuclei present in myosin-positive cells (unpublished data). This excludes the possibility that the formation of hypernucleated myotubes with increased size was a consequence of an increased number of undifferentiated myogenic cells recruited to terminal differentiation, suggesting that NO acts directly as an inducer of myoblast fusion. The process whereby myogenic cells generate myofibers is the consequence of the initial fusion between two myoblasts and the subsequent fusion of new cells to the initial two-cell myotube (). As shown in , 5 mM L-NAME increased, whereas 50 μM DETA-NO decreased, the percentage of mononucleated cells. In addition, DETA-NO increased the formation of binucleated cells and multinucleated myotubes, whereas L-NAME reduced the number of myotubes. We next analyzed the dependence of the effects of NO on the activation of guanylate cyclase and generation of cGMP, an important signaling event mediating several physiological actions of NO (). , differentiation of satellite cells in culture is accompanied by generation of cGMP, occurring in a 4–24-h time window, consistent with the time window of the effect of NO (). The generation of cGMP was NO dependent because exposure of differentiating cells to 5 mM L-NAME or 50 μM DETA-NO administered for 30 min at each of the time points indicated in inhibited or enhanced, respectively, generation of cGMP. Of importance, the ability of DETA-NO to increase the cyclic nucleotide was significantly higher when it was administered during the first 12 h of the differentiation process. It thereafter declined to a point similar to that of nondifferentiated cells (, compare time 72 h with time 0). These results suggest that in satellite cells sensitivity of guanylate cyclase to NO is regulated and its activation favored in the initial phases of differentiation. Because the levels of expression of the two guanylate cyclase subunits α and β did not change with time (), it appears that such regulation occurs through posttranslational events. To assess the role of cGMP generation in satellite cell fusion, we studied the effect of the cell membrane–permeant cGMP analogue 8 Br-cGMP (0.3–5 mM) and the guanylate cyclase inhibitor 1- oxadiazolo [4,3-α]quinoxalin-1-one (ODQ; 3 μM), which were administered in the 4–24-h time window in which these cells generate cGMP. Satellite cell fusion was measured after 72 h. As shown in (C–E), 8 BrcGMP and ODQ mimicked the effects of DETA-NO and L-NAME, respectively. In addition, the fusogenic effect of DETA-NO, although insensitive to L-NAME, was abrogated by ODQ. ODQ, however, did not affect fusion stimulated by 8 Br-cGMP. These results clearly indicate that the effect of NO on satellite cell fusion depends on activation of guanylate cyclase and generation of cGMP. The results depicted in suggest that NO signaling is regulated not only at the level of its generation by NOS but also downstream of it. The results in , showing the cGMP dependence of the effect of NO and the time-dependent changes in the sensitivity of guanylate cyclase to NO, indicate that this regulatory step downstream of NOS activity is at the level of guanylate cyclase. We examined whether the tight regulation of the NO/cGMP signaling during the differentiation process is needed to prevent nonphysiological overgrowth of the myotube. We investigated this possibility by both in vitro and in vivo approaches. To mimic deregulated cGMP signaling in satellite cells, differentiation was performed in the continuous presence of 0.3–3 mM 8 Br-cGMP. In addition, satellite cells were plated at high density (3 × 10 cells/cm) to favor fusion. After 48 h of differentiation under these conditions, satellite cells gave rise to distinctly hypertrophic myotubes ( ). Hypertrophy induction was concentration dependent, as it was the increase in myosin expression (). The myotubes in 8 Br-cGMP–treated cultures were considered to be hypertophic because their mean nuclei number and fiber dimension () and the total amount of myosin () were significantly higher compared with controls and were increased by the cyclic nucleotide in a concentration-dependent way. Such hypertrophy and increase in myosin were not observed in satellite cells differentiated in the presence of 50–300 μM DETA-NO, even at the higher concentrations tested (; and not depicted). Satellite cell differentiation occurs through signaling pathways similar but not identical to those observed during embryonic skeletal muscle development (). We therefore evaluated the role of the NO–cGMP pathway and of its deregulation in embryonic and fetal myogenesis. Presomitic mesoderm (PSM) or the most caudal somites (I–III) from the myosin light chain promoter 1/3 fast (MLC1/3F)–nLacZ 9.5-d postcoitum transgenic mouse embryos, in which the LacZ gene is under the transcriptional control of MLC1/3F (), were grown as explants in cultures in the presence or absence of 50–300 μM DETA-NO, 1–3 mM 8 Br-cGMP, or 5 mM L-NAME. After only 2 d in cultures, differentiated mononucleated myocytes emerged from the PSM explants, and these increased in number after an additional 2 d (). Neither 50–300 μM DETA-NO nor L-NAME had relevant effects on the time of appearance and the number of myocytes after 4 d at any concentration tested (, D and E; and not depicted); in contrast, 3 mM 8 Br-cGMP triggered formation of myotubes, an event that does not occur physiologically in PSM explant cultures (), indicating that the cyclic nucleotide was able to break the fusion restriction in the PSM and myotome (). The effect of 8 Br-cGMP was concentration dependent (). Similar results were observed in somite explants (unpublished data). To study myogenesis during late embryonic and fetal development, MLC1/3F-nLacZ pregnant females were treated with or without 8 Br-cGMP (3 g/kg body weight) from gestation day 10 to either 12.5 or 15.5. Embryos were then recovered, and myogenic cells were revealed by LacZ staining. As shown in , 8 Br-cGMP–treated embryos showed enhanced LacZ staining, indicating an increased level of myogenesis at both time points considered. These results clearly indicate that continuous presence of cGMP increases myotube size and results in muscle hypertrophy and that the tight regulation of its concentration is required for a normal myogenic process. We were interested in identifying the molecular effectors of the NO/cGMP signaling and to assess whether molecules known to play a role in muscle hypertrophy, such as IGF-I (; ), interleukin 4 (IL-4; ), or follistatin (), were involved. We thus performed semiquantitative RT-PCR on RNA isolated from differentiated satellite cells, PSM explants, or muscle from embryonic and fetal stages that were treated or untreated with 3 mM 8 Br-cGMP, 50 μM DETA-NO, or 5 mM L-NAME using primer specific for IGF-I, IL-4, follistatin, myostatin, and skeletal muscle–specific MLC1/3F ( ). In parallel, we examined the expression of IGF-I, IL-4, follistatin, myostatin, and sarcomeric myosin heavy chain by Western blotting (). Quantitative assessment and statistical analyses of the results obtained are shown in Fig. S1 (available at ). Although L-NAME had no effects in all conditions, we observed an increased level of myosin light chain transcripts and protein from 8 Br-cGMP– and DETA-NO–treated cultures and embryos. In satellite cells, this was accompanied by a marked increase in follistatin expression at both the mRNA and protein levels. Of importance, in PSM explants and embryos, follistatin expression was increased only in the presence of 8 Br-cGMP, whereas DETA-NO had no significant effect ( and Fig. S1). Myostatin protein levels were not significantly affected, even if there was a decrease in mRNA levels. We also observed that the mRNA and protein levels of IGF-I were slightly increased by 8 Br-cGMP. These changes, however, were not significant ( and Fig. S1). No changes in the levels of IL-4 were detected (unpublished data). These results suggest that increased generation of follistatin is relevant to the NO/cGMP signaling during myoblast differentiation. Follistatin has recently been described to be a central mediator of the fusogenic effects exerted by deacetylase inhibitors on myoblast fusion into preformed myotubes through a pathway distinct from those used by either IGF-I or IL-4 (). In particular, regulation of follistatin by deacetylase inhibitors, such as trichostatin A (TSA), appeared to be cooperatively activated by MyoD, nuclear factor of activated T cells (NFAT), and cAMP response element binding protein (CREB; ). To investigate whether the NO–cGMP pathway activated the follistatin promoter by the same pathway, we used the myogenic cell line C2C12, which was also used in . Similar to satellite cells, C2C12 myoblasts gave rise to hypertrophic myotubes when cultured in the constant presence of 8 Br-cGMP ( ). This was accompanied by concentration-dependent increases in the levels of follistatin (). To study the effect on follistatin transcription, the fol- listatin promoter linked to the luciferase gene (Fs-Luc) was transfected in C2C12 cells (). 8 Br-cGMP activated transcription of Fs-Luc (). To establish whether the activation of follistatin promoter by cGMP was mediated by MyoD, CREB, and NFAT, Fs-Luc–transfected C2C12 were differentiated in the presence of 8 Br-cGMP and/or transfected with the negative regulator of MyoD; Id1 (); or the dominant-negative form of CREB, A-CREB (), or VIVIT, which is a peptide that blocks NFAT-dependent transcription (). We observed that the 8 Br-cGMP–dependent activation of Fs-Luc was inhibited in the presence of any of these inhibitors (). Thus, NFAT, MyoD, and CREB mediate the effects of 8 Br-cGMP on follistatin transcription. Previous studies in other cell types showed that CREB and NFAT are activated by cGMP through a protein kinase G–dependent phosphorylation (; ; ). Accordingly, we found that 3 μM each of two structurally unrelated protein kinase G inhibitors, KT5823 and 8-(chlorophenylthio)guanosine 3′,5′- cyclic monophosphorothioate ([Rp]-8-pCPT-cGMPS; ), prevented the induction of Fs-Luc transcription by cGMP (). In smooth muscle, cGMP may act through protein kinase A, an enzyme that plays a role in selected myogenic pathways (; ). The possible involvement of protein kinase A in mediating the effects of cGMP, however, was excluded by the lack of inhibitory effects by 3 μM of the protein kinase A inhibitors KT5720 and (Rp)-8-pCPT-cAMPS (). We then studied whether the muscle hypertrophic effect of 8 Br-cGMP was exclusively dependent on the activation of follistatin transcription or whether IGF-I was also involved, in view of the small but detectable changes in the levels of IGF-I transcripts induced by 8 Br-cGMP (). We cultured both satellite and C2C12 cells in differentiating condition in the presence of 8 Br-cGMP and of either antibodies directed against follistatin to neutralize its activity () or the phosphatidylinositol 3′ kinase inhibitor , which inhibits IGF-I signaling in muscle (). In addition, we used the small interference RNA approach already shown to silence follistatin expression in C2C12 cells (). The neutralizing follistatin antibodies inhibited the action of 8 Br-cGMP without significant effects on its own in both satellite () and C2C12 cells (). Cell transfection with the small |interfering RNA against follistatin consistently inhibited the effect of 8 Br-cGMP in C2C12 cells (Fig. S2, available at ). These results indicate that the effect of cGMP is mostly mediated by the activation of follistatin. Consistent with this, did not modify the effects of 8 Br-cGMP while inhibiting those of IGF-I in both satellite cells and C2C12 ( and not depicted). To assess the cell specificity of the effect of cGMP on follistatin, we investigated whether it also increased follistatin expression in cells belonging to nonmuscle lineages, namely embryonic carcinoma cells (P19) undergoing neuronal differentiation, adult (NIH 3T3) and embryonic (10 T1/2) fibroblast cell lines, mesoangioblast-derived smooth muscle progenitors (D351; ), and primary cultures of cardiomyocytes. 8 Br-cGMP had no effects on follistatin levels in any of these cells, whereas it consistently increased follistatin induction in C2C12 cells used as control (Fig. S3, available at ). Skeletal muscle mass growth and maintenance is controlled principally through the regulation of myofiber size both during embryogenesis and in postnatal life. This regulation occurs through two main and distinct mechanisms. One relies on the regulation of the cytoplasmic volume associated with individual myonuclei, and this pathway appears to involve regulation of protein synthesis and of protein degradation (; ). The second mechanism involves the control of the number of myonuclei within a myofiber. In this case, the growth of the muscle fiber during fetal and postnatal development depends on the addition of single cells, embryonic or fetal myoblast and satellite cells, respectively, that must be instructed when to divide and when to differentiate by either fusing with preexisting fibers or among themselves to generate a new fiber (). Myoblast fusion is a multiple-step process involving cell migration, alignment, recognition, adhesion, and membrane fusion (; ), and several molecules have been identified as playing a role in one or more of these processes, including IL-4, IGFs, integrins, and metalloproteases (; ; ; ; ). The mechanisms and signaling pathways underlying the role of these molecules controlling myoblast fusion, however, have yet to be fully elucidated. We demonstrate that the generation of NO is crucial to myoblast fusion in mammals. We found that the action of NO has several important characteristics. (a) It appears to have an effect at critical stages of pre- and postnatal muscle development life. (b) It works through the same signaling pathway at all stages, i.e., activation of guanylate cyclase, with generation of cGMP and induction of follistatin. (c) It is specific to the fusion process itself because NO did not affect cellular differentiation and/or proliferation. This defines for the first time a common trigger for fusion and is the first evidence indicating that the fusion process may take place through the same activating process in both the embryo and the newborn. Another important aspect of the pathway to muscle fusion activated by NO emerging from our results is that it is a regulated process. We have studied it in detail in satellite cells. We found that the regulation of NO effects on muscle fusion occurred at two distinct early steps of the signaling cascade, i.e., the enzymatic activities of NOS and guanylate cyclase, which were regulated in the absence of detectable changes in protein levels. The regulation of guanylate cyclase appears to be particularly important because activation of the enzyme could not be increased even by administration of exogenous NO. We have not yet identified which events among the ones proposed to induce desensitization of guanylate cyclase to NO, e.g., phosphorylation by protein kinases or even a direct action of NO itself (; ), play a role in the desensitization observed here. The physiological relevance of this event, however, is clear because deregulation of cGMP signaling leads to muscle hypertrophy both in satellite cells and in the embryo. More strikingly, myoblasts differentiating from the PSM, known to be incompetent for fusion (), acquire such competence in the presence of cGMP, suggesting that the NO–cGMP pathway not only is crucial to stimulating fusion but also may confer competence to it. Whether and how this occurs when the mononucleated myocytes of the myotome are incorporated into newly formed primary fibers remains to be studied. Of the three NOS isoforms, murine (and human) skeletal muscles express the constitutive NOS Iμ and III, whereas expression of the inducible NOS II is clearly detected only in the presence of inflammation or other pathological conditions (; ). Accordingly, we found that satellite cells and skeletal muscle from embryos (unpublished data) express NOS Iμ and III. However, it is conceivable that both enzymes may play a role because both NOS Iμ and III are activated by increases in intracellular calcium concentrations, and NOS III is also activated by Akt (), i.e., signaling events triggered by many myogenic stimuli (; ). In addition, no obvious defects in muscle development have been reported in NOS I or III knockout mice. Also interesting in this respect is the observation that expression and activities of both NOS Iμ and III are developmentally regulated (; ). The identification of a link between follistatin and the NO/cGMP–dependent fusion adds important new information to skeletal muscle biology. Follistatin is a protein that interacts with and regulates the biological activities of several TGFβ family members, including BMP-4, BMP-7, and activins (; ). Follistatin has been also found to block the activity of myostatin, a negative regulator of skeletal muscle mass, thus leading to muscle hypertrophy (). Therefore, the connection of this protein with NO/cGMP identifies an endogenously activated pathway for follistatin induction that can be activated by many myogenic stimuli and defines the role of NO in the process of fusion. The pathway of follistatin induction by NO/cGMP was found to involve MyoD, NFAT, and CREB. Previous work has demonstrated that both CREB and NFAT are directly activated by NO/cGMP through protein kinase G–dependent phosphorylation (; ; ; ). The fact that inhibition of protein kinase G prevented follistatin induction by cGMP is consistent with these results and further confirms the role of CREB and NFAT. We found that expression of MyoD is not affected by NO/cGMP; whether this transcription factor is activated by NO/cGMP through phosphorylation, similar to other MyoD activating stimuli, remains to be established. The biological significance of each of these transcription factors in mediating the effect of NO/cGMP, however, clearly emerges from our results because each of them was found to be necessary to follistatin induction. Indeed, specific inhibition of MyoD, NFAT, or CREB was sufficient to prevent the transcriptional function of cGMP. Furthermore, these results suggest that the transcriptional effects of NO/cGMP may be more complex than previously envisaged. The fact that MyoD, CREB, and NFAT appear to be necessary to mediate the transcriptional effect of NO/cGMP resembles the situation already described for stimulation of myoblast fusion by the deacteylase inhibitor TSA (). The similarity between the action of TSA and NO/cGMP and the recent evidence showing that TSA is able to up-regulate the expression of NOS III in nonendothelial cells (; ) suggests that NO/cGMP is involved in regulating the process of acetylation. Interestingly, the effect of NO/cGMP, similar to that of TSA (), was restricted to cells of skeletal muscle origin. This cell specificity is intriguing, and the mechanisms beyond it need to be investigated further. In conclusion, the link emerging here among the NO/cGMP signaling and follistatin induction, an important event during prenatal and adult myogenesis, suggests that the role of this messenger may be broader than previously envisaged and at the crossroad of different signaling pathways central to skeletal muscle development and regeneration. The action mediated through cGMP/follistatin might also synergize with other known ac- tions of NO, such as activation of satellite cells (; ). In addition, the cGMP-dependent induction of follistatin might interact with other cGMP-independent actions of NO that may play a role in muscle development, such as inhibition of cytochrome oxidase and control of mitochondrial respiration and -nitros(yl)ation (; ). The following reagents were purchased as indicated: ODQ and DETA-NO from Alexis Italia, KT5823 and KT5720 from Calbiochem, and (Rp)-8-pCPT-cGMPS and -cAMPS from Biolog. The following mAbs were purchased as indicated: anti–NOS-Iμ and anti–NOS-III mAbs from BD Biosciences, anti–glyceraldehyde 3-phosphate dehydrogenase (GAPDH) mAb from Biogenesis, anti-PCNA mAb from Santa Cruz Biotechnology, Inc., anti-MyoD mAb from DakoCytomation, anti–sarcomeric myosin MF20 mAb from Developmental Studies Hybridoma Bank, anti-follistatin mAb (AF669) from R&D Systems, and anti-myostatin (ab996) and anti IGF-I (ab12517) antibodies from AbCAM. In immunofluorescence analysis, primary antibodies were detected by appropriate FITC-conjugated secondary antibodies (Southern Biotechnology Associates, Inc.). In immunoblot analysis, primary antibodies were detected by chemiluminescence with appropriate horseradish peroxidase–conjugated secondary antibodies, all purchased from Bio-Rad Laboratories. Cell culture media and sera were purchased from Cambrex. -[H]-arginine and the kits for the enhanced chemiluminescence and the cGMP radioimmunoassay were obtained from GE Healthcare. Bicinchoninic acid–based assay was obtained from Perbio; C2C12 cells were obtained from American Type Culture Collection; and the anti-desmin polyclonal antibody, L-NAME, 8 Br-cGMP, , IGF-I, and the Hoechst dye and other chemicals were obtained from Sigma-Aldrich. DETA-NO was always prepared fresh by dissolving it at pH 7.4 for 20 min before addition to the cell samples. Under these conditions, DETA-NO generates NO constantly and at defined concentrations (). Satellite cells were isolated from 3–5-d-old mice as described previously () with some modifications. In particular, after 3 d of isolation, proliferating myoblasts were harvested, counted, and plated on tissue culture plastic dishes coated with 1 mg/ml type I collagen. After 2 d of proliferation in growth medium, myogenic cells accounted for >90% of the cultures as revealed by anti-desmin immunostaining assay. Preparation showing <90% myogenic cells were discarded. Myoblasts were shifted to differentiating medium in the presence or absence of drugs as indicated in Results. Growth medium contained Iscove's modified Dulbecco's medium supplemented with 20% FBS, 3% chick embryo extract (), 100 U/ml penicillin, 100 μg/ml streptomycin, and 50 μg/ml gentamycin. Differentiation medium contained Iscove's modified Dulbecco's medium supplemented with 2% horse serum, 100 U/ml penicillin, and 100 μg/ml streptomycin. Fusion index was determined as the number of nuclei in sarcomeric myosin–expressing cells with more than two nuclei versus the total number of nuclei. C2C12 cells were cultured in DME supplemented with 15% FBS, 100 U/ml penicillin, and 100 μg/ml streptomycin and differentiated in DME supplemented with 2% horse serum, 100 U/ml penicillin, and 100 μg/ml streptomycin as described previously (). Immunofluorescence on cells and explants was performed as described previously (), using the following antibodies: MF20 (1:3) and anti-desmin (1:400). Cells were washed free of medium and solubilized by direct addition of a preheated (to 80°C) denaturing buffer containing 50 mM Tris-HCl, pH 6.8, 2% sodium dodecyl sulfate, and a Complete protease inhibitor cocktail (Roche) and immediately boiled for 2 min as previously described (). Alternatively, muscle tissues from embryos were dissected and homogenized in 50 mM Tris/HCl, pH 7.4, 1 mM EGTA, 1 mM EDTA, 1% Triton X-100, and Complete protease inhibitor cocktail and centrifuged (1,000 ) for 20 min at 20°C to discard cellular debris. For secreted proteins analysis, supernatants were collected and concentrated as described previously (). After addition of 0.05% bromophenol blue, 10% glycerol, and 2% β-mercaptoethanol, samples were boiled again and loaded onto 10% SDS–polyacrylamide gels. After electrophoresis, polypeptides were electrophoretically transferred to nitrocellulose filters (Schleicher & Schuell) and antigens were revealed by the appropriate respective primary and secondary antibodies (). The time course of NOS activity was assayed in intact cells by measuring the conversion of -[H]-arginine into -[H]-citrulline as described previously (). In brief, the reaction was performed in 145 mM NaCl, 5 mM KCl, 1 mM MgSO4, 10 mM glucose, 1 mM CaCl, and 10 mM Hepes, pH 7.4. 10 μCi/ml -[H]-arginine (10 μM) was added at various time points, and the reaction was stopped after 5 min by washing with ice-cold PBS supplemented with 5 mM -arginine and 4 mM EDTA. 0.5 ml of 100% cold ethanol was added to the dishes and left to evaporate before a final addition of 20 mM Hepes, pH 6.0. L-NAME–treated cells were run in parallel as control of specificity. Separation of -[H]-citrulline from -[H]-arginine was performed by DOWEX 50X8-400 chromatography. -[H]-citrulline formed was normalized to protein content and evaluated by the bicinchoninic acid procedure. At each time point, satellite cell cultures were incubated for 30 min at 37°C either in growth or in differentiation media with 0.5 mM of the phosphodiesterase inhibitor 3-isobutyl-1-methylxanthine supplemented with 50 μM DETA-NO, 5 mM L-NAME, or vehicle. The reaction was terminated by rapid medium removal and washing with ice-cold PBS and was lysed by the addition of ice-cold trichloroacetic acid (final concentration: 6%). After ether extraction, cGMP levels were measured using a radioimmunoassay kit and normalized to protein content. PSM and most of the five caudal somites were dissected, together with fragmentation of the neural tube from MLC1/3F E9.5 embryos, and cultured as explants, as described previously (). Differentiation was continued for 2–4 d in the presence of various drugs, as described in Results. Before the immunofluorescence assay, X-Gal staining was performed according to standard protocols (). MLC1/3F pregnant females were treated with or without 8 Br-cGMP (3 g/kg body weight; administered in drinking water) from gestation day 10 to 12.5 or 15.5. Embryos were then recovered, and myogenic cells were revealed by X-Gal staining (see Embryo explants culture). Animals were housed in the pathogen-free facility at the Stem Cell Research Institute (Milan, Italy) and treated in accordance with the European Community guidelines and with the approval of the Institutional Ethical Committee. 1 μg RNA was collected from cells, dissected embryos, or tissues using RNeasy mini (or micro) kit (QIAGEN) and was converted into double-stranded cDNA by reverse transcription using the cDNA synthesis kit Thermoscript RT-PCR system (Invitrogen) according to the manufacturer's instructions. cDNA was then amplified using the following primers: follistatin forward, CTCTTCAAGTGGATGATTTTC, and reverse, ACAGTAGGCATTATTGGTCTG; GAPDH forward, TGAAGGTCGGAGTCAACGGATTTGGT, and reverse, CATGTGGGCCATGAGGTCCACCAC; IGF-1 forward, GTGGATGCTCTTCAGTTCGT, and reverse, ACACTCCTAAAGACGATGTT; IL4 forward, AACCCCCAGCTAGTTGTCATCCTG, and reverse, CATCGAAAAGCCCGAAAGAGTCTC; MLC3F forward, GATCACCTTAAGTCAGGT, and reverse, GCAACGCTTCTACCTCTT; myostatin forward, AGCCTGAATCCAACTTAGGC, and reverse, GGTGCACAAGATGAGTATGC. 500 bp of the rat follistatin proximal promoter linked to the luciferase was derived from the 2.8-Kb rat follistatin promoter–luciferase construct (). Id, VIVIT, and A-CREB expression vectors have been described before (). The transfections were performed with the FuGENE6 reagent (Roche). Luciferase assay on cell lysates was performed as described previously (). The results are expressed as means ± SEM. Statistical analysis was performed using a two-tailed test for unpaired variables. Asterisks and crosses in the figure panels refer to statistical probabilities, measured in the various experimental conditions as detailed in the figure legends. Statistical probability values of <0.05 were considered significant. Images in fluorescence have been taken on a microscope (Eclipse E600; Nikon; plan fluor 4×/0.13, 10×/0.33, 20×/0.50, and 40×/0.75) and phase-contrast images of embryos on stereomicroscope (SMZ1500; Nikon; high-resolution plan apochromatic 1× WD54, eyepiece lens CW 10×/22). All images were acquired using a digital camera (DXM1200; Nikon) and the acquisition software ACT-1 (Nikon), imaging medium, PBS buffer, and room temperature. Images were assembled in panels using Photoshop 7.0 (Adobe). Images showing double fluorescence (FITC and Hoechst) were first separately acquired using the different appropriate filters and then merged with Photoshop 7.0. Fig. S1 shows the densitometric quantification of RT-PCR and Western blots analysis shown in . Fig. S2 shows that inhibition of follistatin expression by RNA interference abolishes the effect of NO/cGMP on fusion-induced muscle hypertophy in C2C12. Fig. S3 shows that the NO/cGMP effect on follistatin induction is specific to skeletal myoblasts. Supplemental text describes the methods used in the RNA interference and in the culture of cardiomyocytes and of C3H10 T1/2, NIH 3T3, P19, and D351 cells. Online supplemental material is available at .

The STAT (signal transducer and activator of transcription) family is comprised of seven members in mammals, of which Stat3 is the most pleiotropic member implicated in several biological processes (). Specifically, it has been established as an oncogene by cell transformation and tumor formation capacities in gain-of-function studies (; ). In addition, constitutive Stat3 activation has been reported in a multitude of human cancers (). The oncogenic potential of Stat3 was initially attributed to a positive influence on cell proliferation and/or protection from cell death (for review see ). However, a multifunctional role for Stat3 in promoting tumor growth is emerging, with Stat3 implicated in mediating cell motility and invasiveness, nutrient supply through augmenting angiogenesis, and in masking tumors from the body's innate immune defenses (; ; ). In addition to tumorigenesis, Stat3 knockout in mice causes early embryonic lethality (). Furthermore, conditional knockout mice revealed a critical role for Stat3 in migrating keratinocytes involved in wound healing of the skin (). These studies, in addition to conditional Stat3 ablation studies in other tissues (; ), indicate that Stat3 function extends beyond its tumorigenic potential. Despite the wide range in function, the molecular events leading to Stat3 activation remain relatively straightforward. The canonical Stat3 activation pathway involves cytokine-induced receptor subunit multimerization and activation of receptor-associated Janus tyrosine kinases at the intracellular domain (). Janus tyrosine kinase phosphorylation of the receptor subunits subsequently recruits latent Stat3 from the cytoplasm via Src homology 2 domain interaction (). In turn, Stat3 is phosphorylated, dimerizes, and nuclear translocates to regulate gene expression. Several Stat3 target genes have been identified, including cyclin D1, c-myc, Bcl-2, and Bcl-X (), which may be critical in Stat3-mediated tumorigenesis by regulating cell cycle and survival. However, it remains unclear how Stat3 regulates cell migration. In addition, it is unclear whether cytoplasmic Stat3, the majority of Stat3 population in normal cells, plays any role independent of its transcriptional activity. Stathmin, also known as oncoprotein 18, is a small (18 kD) ubiquitous phosphoprotein localized predominantly to the cytosol. The related stathmin-like family members include super cervical ganglion protein 10 (SCG10), SCG10-like protein (SCLIP), and RB3. These stathmin-related proteins share highly homologous tubulin-binding stathmin-like domains (SLDs) at their COOH terminus (). In addition, they possess a variable membrane-anchoring NH-terminal region that localizes the proteins predominantly to the Golgi apparatus (). Stathmin was originally identified as a key factor involved in regulating cell proliferation (). More recently, stathmin's ability to bind α/β-tubulin heterodimers to facilitate the depolymerization of microtubules (MTs; ) was identified as a principle mechanism of action associated with the control of mitotic spindle assembly as well as other processes involving MT dynamics (; ). Of particular relevance to this study are previous reports of the contribution of stathmin to cell motility (; ). However, the exact mechanism by which stathmin mediates this process remains unclear. Because of their compartmental restrictions, the functions of stathmin-like proteins are likely to be more specified compared with stathmin, although this also remains largely uncharacterized. In this study, we present evidence of Stat3 interacting with stathmin. Moreover, our results reveal that Stat3 may modulate MT dynamics and cell migration through a direct functional interaction with stathmin in the cytoplasm. We initially identified SCLIP as an interacting protein of Stat3 in a yeast two-hybrid screen of a mouse brain library using the COOH-terminal portion of Stat3 (aa 395–770) as bait. The interaction between exogenously expressed proteins was subsequently confirmed in mammalian COS-1 cells (unpublished data). Next, we tested for an interaction with Stat3 among other members of the stathmin family. We observed that myc-tagged SCLIP and stathmin but not myc-SCG10 was shown to associate with endogenous Stat3 in PC12 cells ( ). Because the biological functions of stathmin are better characterized compared with SCLIP, we focused our study on stathmin. As a result of the high expression of stathmin in neuronal cells, the endogenous association with Stat3 was tested in PC12 and NSC34 cells. The results showed that endogenous stathmin coimmunoprecipitated with the Stat3-specific antibody but not rabbit IgG in both cell lines (). Finally, we sought to determine whether the interaction between Stat3 and stathmin was direct. GST-stathmin was expressed and purified from , and its interaction with a baculovirus-produced His-tagged Stat3 was examined. shows that His-Stat3 (97 kD) and GST-stathmin (46 kD) are the major proteins in the Coomassie blue–stained gel. Association between His-tagged Stat3 and GST-stathmin was then determined by a pull-down assay with glutathione–Sepharose beads. In , a marginal amount of Stat3 was found associating with the GST. However, substantially greater amounts of Stat3 were pulled down by GST-stathmin. To exclude the possibility that stathmin associated with the His-tag nonspecifically, we showed that both GST alone and GST-stathmin did not interact with an irrelevant His-tagged protein, His-COMT (catechol-O-methyltransferase; ). Altogether, these results demonstrate a direct physiological association of Stat3 with stathmin. To determine the function of Stat3 association with stathmin proteins, we first investigated the effect of modulating stathmin expression on Stat3 signaling. Overexpression of myc-stathmin or short inhibitory RNA (siRNA) down-regulation of endogenous stathmin in PC12 cells did not significantly affect Stat3 transcriptional activity stimulated by interleukin (IL)-6, as determined by a luciferase reporter assay ( ). In addition, stimulated Stat3 Y705 and S727 phosphorylation remained unchanged during overexpression (not depicted) or knockdown of stathmin protein (). Finally, we showed that stathmin overexpression did not prevent Stat3 nuclear translocation stimulated by IL-6 (). Collectively, these experiments demonstrate that stathmin does not augment or perturb ligand-stimulated Stat3 signaling in terms of phosphorylation, capacity of nuclear translocation, and promoting transcription. We next investigated whether an interaction with Stat3 influenced stathmin activity by first looking at the MT network in murine embryonic fibroblasts (MEFs) genetically depleted of Stat3 (). Staining for α-tubulin revealed radial arrays of MTs in the wild-type (WT) MEF cells ( , i). Strikingly, the MT array in ΔSt3 MEFs appeared disorganized, with thin, discontinuous filaments and few MTs radiating from the centrosome to the plasma membrane (, iv). Quantitative analysis of the α-tubulin staining in randomly selected cells indicated that the fluorescence intensity in ΔSt3 MEFs (40 ± 14 intensity units/μm; = 12) were ∼40% of that in WT MEFs (94 ± 18 intensity units/ μm; = 12; not depicted). Further analysis with anti-acetylated α-tubulin revealed dramatically reduced levels of this MT subset in ΔSt3 MEFs compared with WT cells (, ii and v). Furthermore, we compared MTs at the same cell cycle stage by arresting cells at the G/S transition with hydroxyurea (HU). Arresting WT MEFs with HU resulted in substantially increased acetylated MTs (, iii). In contrast, ΔSt3 cells displayed a marginal increase after HU treatment (, vi). As posttranslational acetyl modification correlates with increased MT stability (), these results indicate that the ΔSt3 MEFs are deficient in stabilized long-life MTs. We further confirmed this biochemically. Initially, we examined the total expression of α-tubulin and acetylated α-tubulin in MEF cells. Densitometric quantitation of α-tubulin bands, normalized for glyceraldehyde-3-phosphate dehydrogenase (GAPDH) expression, indicated that there was not a marked difference between ΔSt3 and WT MEFs. In contrast, acetylated α-tubulin levels after normalizing for GAPDH expression were ∼35% lower in ΔSt3 MEFs compared with WT. Stat3 was also confirmed to be absent in ΔSt3 MEFs (). As MTs are denatured during the preparation of total cell lysates (TCLs), we next investigated the levels of tubulin polymers specifically by harvesting cells in a MT-stabilizing buffer and performing differential sedimentation of polymerized tubulin from free tubulin. These results indicated that the amount of α-tubulin polymers in the pellet fraction of ΔSt3 MEFs was substantially reduced with a concomitant increase of free α-tubulin in the soluble fraction when compared with WT cells (, top). Similarly, the amount of acetylated α-tubulin polymers in ΔSt3 MEF cells was reduced (, second panel). Reprobing with an antibody for GAPDH indicated equivalent protein loading between samples (, third panel). Densitometric analysis indicated similar total levels of tubulin (i.e., soluble and pellet fractions combined) between the MEF cells (not depicted). In addition, Stat3 in WT MEFs was predominantly located in the soluble fraction and did not pellet down with MTs (, fourth panel), suggesting that Stat3 is not directly associated with MTs in vivo. To confirm the specificity of this method, we pretreated WT MEFs with taxol or nocodazole before harvesting. As shown in (bottom), the majority of α-tubulin in taxol-pretreated cells was found in the pellet. Conversely, no α-tubulin was found in the pellet of nocodazole-pretreated cells. This confirms the polymerized state of tubulin found in the pellet fractions. To exclude the possibility that the differences in tubulin polymer content was caused by a change in cell cycle state in the ΔSt3 cells, we measured the DNA content of ΔSt3 compared with WT MEFs by propidium iodide staining and cell sorting. The percentages of cells in the various cell cycle states did not differ markedly between the two cell types (). Finally, as MTs influence cell motility, we assessed ΔSt3 MEF cell migration and demonstrated a severely reduced capacity to migrate into an in vitro wound track (). This is in agreement with previous reports of Stat3 function in regulating cell motility (). Collectively, our data demonstrate that Stat3 ablation in MEF cells results in a perturbed MT network and impaired cell migration. To determine whether Stat3 expression could promote MT formation, we reintroduced Stat3 into ΔSt3 MEFs by retroviral infection. Reexpression of Stat3 restored the level of acetylated MTs in ΔSt3 MEFs to a level comparable with WT cells ( ). In addition, the amount of polymerized α-tubulin in retrovirally infected ΔSt3 MEFs was elevated above that of uninfected cells (or ΔSt3 MEFs infected with retrovirus vector) and was comparable with that in WT MEFs (). Reprobing for Stat3 confirmed reexpression in cells infected with Stat3 retrovirus only, and reprobing for GAPDH demonstrated equivalent protein loading. To further characterize Stat3 regulation of MT stability in other cell types, endogenous Stat3 was down-regulated in MCF-7 cells by transient expression of a human Stat3 siRNA sequence. As a control, transfection with the empty siRNA vector did not affect Stat3 expression or acetylated MTs (, i–iii). In contrast, MCF-7 cells with reduced Stat3 expression exhibited a decrease in the amount of acetylated MTs (, iv–vi; arrows). Finally, we demonstrated that the expression in ΔSt3 MEFs of WT Stat3 or a Stat3 mutant (Y705F) previously shown to be deficient in transcriptional activity () induced a normal MT network morphology as detected by antibodies against α-tubulin () and acetylated α-tubulin (). In contrast, a transcriptionally active Stat3 mutant (Stat3C; ) did not markedly rescue the MT network in these cells (). This suggests that the effect of Stat3 on MTs may be independent of the transcriptional activity of the dimer. In support of this, we also found that ligand-induced activation of Stat3, as indicated by Y705 or S727 phosphorylation, had little effect on the polymerized MT mass in WT MEFs (Fig. S1 A, available at ). Likewise, complete disruption of the MT network did not perturb ligand-stimulated Stat3 tyrosine phosphorylation (Fig. S1 B). In total, these results indicate that the cytosolic expression of Stat3 promotes MT formation, which may be independent of its nuclear activity. We next sought to identify the specific mechanism through which Stat3 regulated MT assembly. We could not find a direct association between Stat3 and either α- or β-tubulin in WT MEFs (Fig. S2 A, available at ). In addition, the MT network was not required for Stat3–stathmin interaction, as nocodazole pretreatment did not prevent coimmunoprecipitation of stathmin with Stat3 (Fig. S2 B). We next investigated whether the Stat3–stathmin interaction was involved in modulating MTs. We first determined the region of stathmin that was required for Stat3 interaction by deleting various amino acids from the NH or COOH terminus of stathmin ( ). These mutants were then expressed in WT MEFs, and their interaction with endogenous Stat3 was analyzed. Stathmin deletion mutants truncated from the COOH terminus (aa 1–120 and 1–90) displayed a marked reduction in association with Stat3 in comparison with full-length or NH-terminally (aa 30–150 and 60–150) truncated stathmin (). This indicates that Stat3 binds to the COOH-terminal region of stathmin, which is also involved in binding tubulin. Surprisingly, although Stat3 bound to the full-length SCLIP, its binding capacity to the SLD of SCLIP (SCLIP-SLD, aa 38–180) was significantly reduced (). This suggests that the NH terminus of SCLIP may be also required for Stat3 binding. However, because of the loss of membrane-anchoring cysteine residues, it remains unclear whether Golgi localization or the intrinsic presence of the SCLIP NH terminus is required for the Stat3 association. We next investigated Stat3 contribution to MT depolymerization induced by stathmin overexpression in WT compared with ΔSt3 MEFs. In addition, given the differences in Stat3 binding, the effect of SCLIP-SLD overexpression was simultaneously assessed for comparison. We found that the MT network remained largely intact in WT MEFs overexpressing stathmin (, i–iii). This is in agreement with the previously reported effects of stathmin overexpression on the MT network of A6 endothelial cells (). However, the reduced MT network in ΔSt3 MEFs was further depolymerized by stathmin in ΔSt3 cells as indicated by the dispersed punctate α-tubulin staining in the positively transfected cell (, iv–vi). This different effect of stathmin is more clearly displayed in the enlarged images (, compare xiii with xiv). In contrast to stathmin, we observed that regardless of Stat3 expression, both WT and ΔSt3 MEFs overexpressing SCLIP-SLD showed certain degrees of MT depolymerization (, vii–xii). The different behavior of stathmin and SCLIP-SLD was not caused by the different expression level because their expression was comparable in both WT and ΔSt3 MEF cells (). These results suggest that endogenous Stat3 may prevent MT disassembly induced by stathmin but not SCLIP-SLD overexpression via a functional interaction. A key factor in a mechanism where Stat3 promotes MT formation through an antagonistic interaction with stathmin is the presence of sufficient Stat3 protein in relation to stathmin. Therefore, we determined the absolute intracellular levels of Stat3 and stathmin protein in PC12 cells. There was roughly 2.5 times the number of cellular Stat3 to stathmin (Fig. S3, A–E; available at ). In MEFs, the estimated ratio of Stat3 versus stathmin was even higher (not depicted). These results indicate that intracellular Stat3 is present in sufficient amounts and is capable of preventing the disassembly of MTs induced by an interacting partner (stathmin) but not the noninteracting SCLIP-SLD protein. Next, we down-regulated endogenous stathmin expression by siRNA to further investigate its contribution to disrupting MTs in ΔSt3 cells. We first demonstrated that stathmin knockdown did not overtly affect the level of polymerized α-tubulin in WT MEFs ( ). However, transient expression of increasing amounts of stathmin siRNA induced an increase in polymerized α-tubulin levels in ΔSt3 MEFs (). Densitometric quantitation indicated that in WT MEFs expressing control siRNA, polymerized α-tubulin in WT MEFs represented 61 ± 0.4% of total α-tubulin (). In contrast, polymerized tubulin was found to make up only 17 ± 3.8% of total tubulin in ΔSt3 MEFs expressing control siRNA. This increased to 39 ± 1.9% in ΔSt3 MEFs transfected with the highest amount of stathmin siRNA used, which was significantly elevated (P < 0.005), compared with cells expressing control siRNA (). As a control, the level of polymerized α-tubulin in WT MEF cells expressing similar amounts of stathmin siRNA was not markedly changed (57 ± 3.9% of total tubulin). These results indicate a contribution of stathmin to the loss of polymerized tubulin in ΔSt3 MEFs. The contribution of stathmin activity to the migratory properties of MEFs was then examined. First, we showed that the expression of stathmin siRNA had no apparent effect on the ability of WT MEFs to migrate into a wound track (, i–iii). This correlated with the lack of effect of the stathmin siRNA on MT levels in the normal fibroblast (). In contrast, the deficiency in ΔSt3 MEF migration into a wound track was partially reversed by down-regulating stathmin (, iv–vi). Quantitative analysis indicated 30–70% (P < 0.05 and P < 0.001, respectively) restoration in migration into the wound track by ΔSt3 MEFs transfected with increasing concentrations of stathmin siRNA (). As an increased stathmin protein expression had minimal effects on WT MEF cell migration (Fig. S4, available at ), these results suggest that excessive stathmin activity in a Stat3-null background is at least partially responsible for attenuating cell migration. In support of this, we observed that expression of a constitutively active mutant of stathmin (tetraA) prevented the migration of WT MEF cells into the wound track (Fig. S4). This correlated with the ability of stathmin tetraA to disassemble MTs in positively transfected WT MEF cells (not depicted). Altogether, our results point to stathmin as a possible downstream target of Stat3 in maintaining the MT array and normal cell migration. We next sought to confirm that Stat3 could directly attenuate stathmin activity by using recombinant proteins in an in vitro tubulin assembly assay. GST-Stat3, GST-stathmin, and GST–SCLIP-SLD were produced in and purified by binding to glutathione–Sepharose beads. GST-Stat3 was further purified by gel filtration, whereas stathmin and SCLIP-SLD was generated by removing the GST tag by proteolytic cleavage. The final recombinant proteins were resolved by SDS-PAGE and stained by Coomassie blue ( ). First, we demonstrated the normal kinetics of MT assembly, which included an initial nucleation stage followed by rapid polymerization that reached steady-state equilibrium at a ΔOD of 0.2–0.25 within 60 min under standard conditions (, control). Although including Stat3 did not affect the normal reaction, in the presence of stathmin, tubulin polymerization was severely inhibited () to a level similar to a negative control containing nocodazole (not depicted). In addition, the inhibitory effect of stathmin on tubulin assembly was demonstrated to be dose dependent (). Next, we investigated the effect of GST-Stat3 in reversing stathmin attenuated MT assembly using different molar ratios of GST-Stat3 versus stathmin. We first showed that as a control, GST alone did not inhibit the activity of stathmin (). Subsequently, we observed that although relatively low levels of GST-Stat3 (1:30 ratio to stathmin) had little effect, GST-Stat3 in a 1:5 molar ratio to stathmin restored MT polymerization to the same maximal level in comparison with the control (). These results indicated that GST-Stat3 reversed stathmin inhibition of MT polymerization in a dose-dependent manner. However, it was noticed that the normal polymerization kinetics were not completely restored. In the presence of GST-Stat3 and stathmin (1:5), the time required to reach maximum polymerization was >70 min compared with <40 min in the presence of tubulin alone (, control). Because of the low expression levels of the recombinant protein, a greater amount of GST-Stat3 could not be used without altering the total reaction volume and, subsequently, normal MT assembly kinetics. The antagonistic function of Stat3 was further demonstrated by adding GST-Stat3 at ∼30 min after the commencement of polymerization in the presence of stathmin. Upon adding Stat3, a reversal of stathmin that inhibited MT polymerization was immediately evident (), indicating that Stat3 could attenuate the interaction between stathmin and tubulin. As a control, GST had no such effect (). Finally, we showed that SCLIP-SLD, which does not interact with Stat3, can also inhibit MT assembly to a similar extent as stathmin but could not be attenuated when coincubated with GST-Stat3 (). These data demonstrate that Stat3 regulates the MT polymerization via direct interaction with stathmin. Indeed, in these MT polymerization reactions, GST-Stat3 directly interacted with stathmin, as shown by GST pull-down assays (). To further address the effect of stathmin and Stat3, MTs were assembled using rhodamine-labeled tubulin, and the products were examined by fluorescence microscopy. In the control reaction, large amounts of long MT strands were formed after 30 min of incubation time (, i). In the presence of taxol, tubulin assembled into many short MT aggregates (, ii). This confirms the previously defined mechanism of taxol action in promoting MT assembly through stabilizing short tubulin polymers (). This indicates that, in vitro, taxol promotes a rapid absorbance increase in the turbiditometric assay through MT nucleation rather than elongation. As expected, incubating stathmin with tubulin prevented MT assembly, as shown by the lack of MTs after 30 min at 37°C (, iii) and the observation of only a few long MT strands after 60 min (, iv). However, preincubating stathmin with GST-Stat3 significantly rescued the assembly of MTs by 30 min, which improved further after 60 min (, v and vi). Furthermore, unlike taxol-assembled MTs, MT strands rescued by GST-Stat3 were long and comparable with untreated control MTs (, compare vi with i). These results, together with those shown in , indicate that stathmin inhibits the de novo formation of MTs. These results also clearly demonstrate a direct role for Stat3 in antagonizing the MT-destabilizing effect of stathmin to promote MT assembly. In this study, we identified the MT-destabilizing protein stathmin as a novel Stat3-interacting protein. To date, previously identified Stat3-interacting proteins have been concerned primarily with modulating Stat3 activity (; ; ). This does not appear to be the case with stathmin, as manipulating the expression of stathmin failed to perturb the phosphorylation, nuclear translocation, or transcriptional activity of Stat3 (). Instead, we present evidence that the association of Stat3 with stathmin may be involved in modulating the MT-destabilizing activity of the latter. Embryonic fibroblasts lacking Stat3 clearly exhibited a disrupted MT network with a lower level of polymerized and acetylated tubulin, which was reversed upon the reintroduction of Stat3 ( and ). These results provide clear evidence that Stat3 is necessary for maintaining proper MT organization. However, there does not appear to be an obvious effect on mitotic MTs, as spindle formation and cell division do not appear perturbed in ΔSt3 MEFs (not depicted). Stat3 had not been reported to directly associate with MTs. Our study supports this, as the pelleting of polymerized tubulin does not bring down Stat3 (). In addition, there was no obvious change in total tubulin protein () or mRNA levels (not depicted) in ΔSt3 cells. This suggests that Stat3 does not directly modulate MT stability or tubulin expression. In contrast, our study indicates that Stat3 binds the COOH-terminal tubulin-binding domain of stathmin (), which is thought to contribute significantly, although not solely, to its MT-destabilizing activity (; ; ). This suggests that Stat3 may compete for the tubulin-binding site to antagonize the MT-destabilizing activity of stathmin. In support of this, we show that recombinant Stat3 competitively binds and interferes with stathmin's activity in vitro (), implicating stathmin as a direct target of Stat3 in maintaining MT morphology. Furthermore, tubulin polymerization is reestablished in ΔSt3 cells by stathmin down-regulation (). In addition, stathmin overexpression disassembled MTs in ΔSt3 but not WT MEFs (). Although an increase in stathmin expression would be expected to result in MT disassembly in the normal cells, the extent of perturbation exerted on the MT array by exogenous stathmin depends largely on the level of expression achieved (; ). In contrast, we found that a similar expression level of SCLIP-SLD destabilized MTs irrespective of the presence of Stat3 in vivo (). This was further confirmed by the ability of recombinant Stat3 to attenuate the activity of stathmin, but not SCLIP-SLD, in vitro (). As the MT-destabilizing kinetics between these two highly homologous domains are similar in vitro (; ) and given their similar expression levels in vivo (), the main differentiating factor remains the expression of Stat3 in the MEF cells. However, as stathmin down-regulation only partially reversed the deficiency in MT levels and cell migration (), stathmin-independent mechanisms perturbed by the loss of Stat3 remains a possibility. Collectively, we conclude that a direct binding of stathmin by Stat3 serves to maintain the interphase MT array and support cell migration. During the preparation of this manuscript, a similar study reported an interaction between stathmin and p27, a cyclin-dependent kinase inhibitor protein (). Although unclear, it remains possible that both proteins may be involved in regulating stathmin, depending on the cellular context. It is also interesting to note that p27 expression is regulated by Stat3 (). In conjunction with this report, our current study supports the notion that the MT-destabilizing activity of stathmin may be regulated by protein–protein interaction in addition to serine/threonine phosphorylation. However, given that the nonphosphorylatable stathmin-tetraA constitutively active mutant could still disassemble MTs (not depicted) and halt migration in WT MEF cells (Fig. S4), serine phosphorylation clearly remains a key negative regulatory mechanism in stathmin function. The loss of Stat3 severely affected the ability of MEFs to migrate into a wound track, which supports previous reports of Stat3 as a critical regulator of keratinocyte and ovarian cancer cell motility (; ). As stathmin down-regulation promoted the migration of ΔSt3 MEFs (), inhibition of stathmin by Stat3 may, at least, be partially involved in maintaining cell motility in fibroblasts. We have also reported that the down-regulation of stathmin by siRNA in WT MEFs did not lead to a noticeable effect on MT polymer levels or cell migration (). Although the reason for this is unclear, we can postulate that the majority of stathmin in the interphasic cell may not be required for maintaining the MT array and may be kept inactive by negative regulatory mechanisms. In support of this was our observation that an increased expression of WT stathmin did not significantly perturb the MT array () or migration (Fig. S4), suggesting that existing negative regulators of stathmin are sufficient for maintaining normal cell function. This is consistent with our observations that the wholesale loss of a negative regulator such as Stat3 would be detrimental to both the MT array and cell migration. Some have made the reciprocal observation that an elevated expression of stathmin contributes to promoting cell migration (; ). However, this is counterintuitive to evidence that localized MT stabilization promotes cell migration through the establishment of cell polarity and preservation of the leading edge (; ). Furthermore, a reduced level of stathmin–tubulin interaction has been described at the leading edge (). In addition, a mild nocodazole treatment has also been shown to severely reduce fibroblast locomotion (). Thus, excessive stathmin activity would likely be antimigratory. In confirmation of this, the expression of a constitutively active stathmin mutant prevented the migration of WT cells into the wound track (Fig. S4). However, although seemingly paradoxical, a chronic loss of stathmin activity, as is the case with stathmin-null fibroblasts (), may also be detrimental to cell migration as a result of a reduction in MT dynamics. In support of this, an analogous effect with taxol treatment also leads to a halt in cell motility (). Thus, it appears that relative stathmin activity, as opposed to expression levels themselves, may be the more critical determinant in maintaining cell migration. Collectively, we propose that Stat3 regulation of stathmin activity is required to maintain MT dynamics to promote normal cell movement. Stat3 function has traditionally been described in regard to its nuclear gene-regulating activity (for review see ; ). Although our results do not exclude a contribution by the transcriptional activity of Stat3, the notion that the Stat3–stathmin interaction is at least partially responsible for regulating MT assembly represents a key cytoplasmic function for Stat3. In support of this, fractionation studies have shown that after activation, an estimated 30% of Stats maximally and transiently accumulate in the nucleus (; ) with a substantial amount of Stat3 remaining cytoplasmic. In addition, we have determined that intracellular Stat3, relative to stathmin, is present in sufficient amounts to antagonize the MT-destabilizing protein by direct interaction. Furthermore, the cytoplasmic expression of the Stat3 Y705F mutant was sufficient to promote the formation of acetylated MTs (). Given that perturbation of the MT network did not affect Stat3 activation (Fig. S1) and that stimulating Stat3 activity did not change MT polymer mass, the cytoplasmic and nuclear functions of Stat3 appear to be independent of one another. In conclusion, our study has described a role for cytoplasmic Stat3 in mediating MT dynamics through functional interaction with a key MT-destabilizing protein, stathmin. MTs are central components of the cellular architecture required for proliferation, differentiation, and migration. Our work presents a novel mechanism through which Stat3 may mediate these processes in conjunction with its nuclear transcriptional activity. Stat3 was cloned into the pGBKT7 plasmid containing the GAL4 DNA-binding domain and was used as bait in a yeast two-hybrid screen of a fetal mouse brain cDNA library (CLONTECH Laboratories, Inc.) as previously described (). After elimination of false positive clones by a β-gal colony lift-filter assay, plasmids were isolated from positive colonies, transformed into DH5α, purified, and sequenced. Stathmin, SCG10, and SCLIP were amplified from a human fetal brain cDNA library (CLONTECH Laboratories, Inc.) by PCR with specific primers and subcloned into the pXJ40-myc vector with BamH1 and XhoI restriction enzymes. SCLIP-SLD and the truncated mutants of stathmin were subsequently generated by PCR, and the products were reinserted into pXJ40-myc. Stat3C was constructed from pRC-CMV-Stat3 by site-directed mutagenesis as previously described (). All constructs were subjected to restriction digest and full sequencing analysis. The GST-Stat3 fusion protein and the baculovirus-produced Stat3 were purified as previously described (; ). GST-Stat3 was further purified from cleavage products by fast protein liquid chromatography gel filtration through a Superdex HR200 column (GE Healthcare) equilibrated with 80 mM Pipes, pH 6.8, 1 mM EGTA, 5 mM MgCl, 10% (vol/vol) glycerol, and 1 mM benzamidine. The column was calibrated with the markers dextran blue (2,000 kD), femtin (440 kD), and cytochrome C (14 kD). Fractions corresponding to full-length GST-Stat3 were pooled and concentrated before use. The GST-stathmin and GST–SCLIP-SLD expression plasmid were constructed by inserting stathmin and SCLIP-SLD cDNA fragment into pGEX-6P-1 (GE Healthcare) vector with BamH1 and XhoI restriction enzymes. Plasmids were transfected to , and the expression of GST fusion proteins was induced with 100 mM IPTG for 4 h. GST fusion proteins were purified by binding to glutathione–Sepharose beads, eluted with 10 mM glutathione and 50 mM Tris-HCl, pH 8.0, and dialyzed into 80 mM Pipes, pH 6.8, 1 mM EGTA, 5 mM MgCl, and 10% (vol/vol) glycerol. GST was then removed by Precision Protease cleavage according to optimized conditions (GE Healthcare). MCF7, NSC34, and MEFs were maintained in DME containing 10% FBS and supplemented with penicillin/streptomycin and L-glutamine. PC12 cells were similarly cultured but with an additional 5% (vol/vol) horse serum. MTs were polymerized by 10 μM taxol and depolymerized by 10 μM nocodazole for 4 h at 37°C before cell lysis. Transient transfections were performed with LipofectAMINE 2000 (Invitrogen) according to the manufacturer's instructions. Cells were grown on glass coverslips overnight to subconfluency. Subsequent to cell treatments, cells were fixed in 4% (wt/vol) PFA, permeabilized with 0.2% (vol/vol) Triton X-100 in PBS, and blocked with 10% (vol/vol) FCS in PBS. Cells were then stained with primary antibodies diluted (1:100) in 1% (wt/vol) BSA in PBS and detected with Cy2/Cy3-conjugated secondary antibodies (1:400 in 0.1% [wt/vol] BSA in PBS). The cells were then mounted and examined on a confocal microscope (MRC1024; Bio-Rad Laboratories) using 40× NA 1.00 and 100× NA 1.25 oil immersion objectives (Carl Zeiss MicroImaging, Inc.) on an upright microscope (Axioplan2; Carl Zeiss MicroImaging, Inc.). Images were collected and processed using Lasersharp 2000 and Adobe Photoshop version 6.0, respectively. Assay of cell migration into a wound area was performed as previously described () with slight modifications. In brief, WT and ΔSt3 MEF cells were grown to subconfluency on collagen (type I)-coated culture dishes (Iwaki). The cells were starved of serum for 24 h and treated with mitomycin C for 90 min to arrest cell proliferation. A wound track was then introduced by scraping the cell monolayer with a yellow pipette tip. Cells were cultured for a further 24–48 h before phase-contrast analysis. The siRNA expression vector containing a 21-nt target sequence for human Stat3 was constructed by insertion into pBS/U6 vector (a gift from Y. Shi, Harvard Medical School, Boston, MA) in two separate steps as previously described (). The first annealed oligonucleotide pairs (5′-GGCGTCCATCCTGTGGTACAA-3′ and 5′-AGCTTTGTACCACAGGATGGACGCC-3′) were digested with ApaI (blunted) and HindIII. The inverted motifs that contained the six-nt spacer and five Ts (5′-AGCTTTGTACCACAGGATGGACGCCCTTTTTG-3′ and 5′-AATTCAAAAAGGGCGTCCATCCTGTGGTACAA-3′) were then subcloned into the intermediate plasmid (HindIII and EcoRI sites) to generate pBS/U6-hu Stat3 siRNA. A Basic Local Alignment Search Tool search of all target sequences showed no significant sequence homology with other genes. All siRNA expression vectors were confirmed by sequence analysis of the target insert. The murine stathmin synthetic siRNA (Santa Cruz Biotechnology, Inc.) was used according to the manufacturer's instructions. The in vivo assay of polymerized tubulin was performed as previously described (). In brief, MEF cells were scraped into a MT-stabilizing buffer (0.1 M Pipes, pH 6.9, 2 M glycerol, 5 mM MgCl, 2 mM EGTA, 0.5% Triton X-100, and protease inhibitors) supplemented with 4 μM taxol to maintain MT stability during isolation. The supernatant containing solubilized tubulin was clarified by centrifugation (20,000 for 45 min) and separated from the pellet containing sedimented polymerized tubulin. The pellet was washed once in MT-stabilizing buffer before being denatured in laemmli buffer. The in vitro kinetics of MT assembly was measured using the Tubulin Polymerization Assay kit (Cytoskeleton, Inc.) according to the manufacturer's instructions. In brief, the assembly of 300 μg tubulin in 100 μl tubulin assembly buffer (80 mM Pipes, pH 6.8, 0.5 mM EGTA, 2 mM MgCl, 1 mM GTP, and 10% vol/vol glycerol) into MTs was started by incubating at 37°C. Absorbance at 340 nm was then measured every minute for up to 90 min in a temperature-controlled 96-well microtitre plate spectrophotometer (Genios; Tecan). Unless otherwise stated, recombinant proteins were coincubated for 10 min before the addition to tubulin. For fluorescence studies, 2 μg/μl tubulin was polymerized with rhodamine-labeled tubulin in a 4:1 ratio at 37°C for 30 or 60 min. MTs were then immediately spotted on a glass slide, an equal volume of Gel Mount (Biomeda) was added, and a coverslip was mounted. Images were then collected on a confocal microscope (MRC1024; Bio-Rad Laboratories) using a 100× NA 1.25 oil immersion objective (Carl Zeiss MicroImaging, Inc.) on an upright microscope (Axioplan2; Carl Zeiss MicroImaging, Inc.). Fig. S1 shows that the MT polymerization state is independent of Stat3 activation. Fig. S2 shows that Stat3 does not associate with tubulin and that its binding to stathmin does not require the MT network. Fig. S3 indicates the quantification of Stat3 and stathmin protein in PC12 cells, and Fig. S4 shows that constitutively active stathmin attenuates cell migration. Online supplemental material is available at .

The ability of fibroblasts to perceive extracellular stress from and to transmit contractile force to the ECM is crucial for regulating their activity during connective tissue remodeling (; ). After tissue injury, changes in the composition, organization, and mechanical property of the ECM, together with the action of growth factors (), stimulate fibroblasts to migrate and to develop in vivo stress fibers, which are initially composed of cytoplasmic actins (). The resulting increase in matrix tension, in conjunction with the action of TGFβ (), is a prerequisite for the de novo expression of α-smooth muscle actin (α-SMA; ; ), which hallmarks myofibroblast differentiation. Myofibroblast populations in three-dimensional collagen gels exhibit increasing levels of α-SMA expression with increasing mechanical load (; ). Stressing wound granulation tissue fibroblasts in vivo by splinting full-thickness wounds enhances expression of α-SMA that is down-regulated after releasing tension by removing the splint (). We have previously shown that α-SMA generates high contractile activity in stress fibers (, ) that are preformed from β-cytoplasmic actin filament bundles in wound granulation tissue fibroblasts () and in spreading cultured myofibroblasts (; ). In this study, we demonstrate that focal adhesion (FA) size controls α-SMA localization by regulating stress fiber tension. FAs are central checkpoints in transforming extracellular mechanical cues into cellular responses and in transmitting contractile forces to the ECM (; ; ). Cultured myofibroblasts form 8–30-μm-long “supermature” FAs (suFAs; ; ), exhibiting a specific molecular composition by coexpressing vinculin, paxillin, tensin, and integrins αvβ3 and α5β1 (; ). This is in contrast to smaller “classical” FAs (2–6-μm-long) of α-SMA–negative fibroblasts that do not contain significant levels of tensin and α5β1 integrin or to fibrillar adhesions that are generally negative for vinculin, paxillin, and αvβ3 integrin (; ). It has been proposed that suFAs are particularly efficient in promoting tissue contraction (; ) by providing high adhesion to the ECM (; ); however, their role in controlling myofibroblast function and contractile activity has not been assessed yet. We show for the first time that recruitment of α-SMA to preexisting stress fibers requires a critical tension that is only generated upon formation of suFAs on sufficiently rigid substrates. We further demonstrate that suFAs exert significantly higher force per unit area compared with classical FAs of α-SMA–negative fibroblasts. Hence, establishment of suFAs is a central checkpoint in the mechanical feedback loop of intracellular contractile activity and extracellular tension because of its control of α-SMA recruitment to stress fibers. We have previously shown that the formation of β-cytoplasmic actin stress fibers precedes the recruitment of α-SMA during myofibroblast spreading (). To test whether this compartmentalization of α-SMA is controlled by the tension transmitted at FAs, we cultured differentiated rat embryonic fibroblast (REF)-52 myofibroblasts for 12 h on polydimethylsiloxane (PDMS) substrates ranging from the approximate stiffness of normal soft tissue to that of plastic ( ). The percentage of cells with clearly visible α-SMA stress fibers was manually quantified, and the morphology of all FAs was assessed using a segmentation algorithm () on fluorescence images, which were obtained after Triton X-100 (TX-100) permeabilization. On PDMS exhibiting a Young's modulus between 16 and 780 kPa, ∼90% of the myofibroblasts localized α-SMA to stress fibers that were anchored at suFAs with a length of 8.5 ± 3.6 to 9.9 ± 3.1-μm-long (). With decreasing substrate stiffness, suFAs were reduced to classical FAs with a length of 5.0 ± 2.2 μm at 11 kPa and 1.5 ± 1.4 μm at 9.6 kPa (). Concomitant with the reduction of suFAs to classical FAs, α-SMA was completely lost from stress fibers in 97% of the cells on 9.6 kPa substrates (); a detergent-free fixation procedure revealed the accumulation of stress fiber–derived α-SMA in neo-formed cytosolic rods (). Importantly, phalloidin- () and β-cytoplasmic actin–positive () stress fibers were formed on all substrates and protein expression levels remained unchanged (unpublished data), suggesting that myofibroblasts can modulate their contractile function by selectively redistributing α-SMA. To relate PDMS substrate compliance to the stiffness experienced by tissue myofibroblasts, we measured the Young's modulus of rat wound granulation tissue using atomic force microscopy (AFM; ). Tissue stiffness increased from 18.5 ± 5.4 kPa in 7-d-old wounds containing only fibroblasts with α-SMA–negative stress fibers () to 26.5 ± 12.7 kPa in 8-d-old wounds, when α-SMA starts to occur in stress fibers. In 9-d-old granulation tissue, stiffness further increased to 29.4 ± 17.5 kPa, together with increased appearance α-SMA–positive stress fibers in myofibroblasts (). In late granulation tissue (12-d old), when the wound is almost closed, substrate stiffness even reached a modulus of 49.3 ± 28.5 kPa. Hence, α-SMA localization in stress fibers in vivo and in vitro only occurs at a critical substrate stiffness of ∼20 kPa. One possible explanation for the observed redistribution of α-SMA to the cytosol is that the small FAs formed on compliant substrates limit tension within stress fibers at a level that is nonpermissive for α-SMA recruitment. To test this hypothesis, we microcontact printed (μCP) regular arrays of fibronectin (FN) islets on rigid planar surfaces (Fig. S1 A, available at ), exhibiting the characteristic length of classical FAs (2-, 4-, and 6-μm-long) and suFAs (10- and 20-μm-long). Only by considering the elongated shape of FAs (width, 1.25 μm) we achieved almost complete occupation of up to 20-μm-long islets by single adhesions after a 12-h culture ( ). This is in contrast to squares or circles that are only partially occupied when islet diameter exceeds ∼1 μm or that allow the formation of multiple FAs (; ; ). When plated on 20- () and 10-μm-long islets (), stress fiber formation and α-SMA incorporation () were indistinguishable from cells grown on continuous substrates (). On 6-μm-long islets, however, α-SMA was absent from stress fibers in 74 ± 3% of the cells () and in 96 ± 8% of the cells grown on 2-μm-long islets (; cells per condition = 400). ). Loss of α-SMA from stress fibers on islets ≤6-μm-long was also achieved observed using FN, collagen I, and vitronectin alone, in combination at 10–100 μg/ml (unpublished data), and after reducing islet spacing from 6 to 4 and 2 μm (Fig. S2). This indicates that the effect of islet size on α-SMA localization overrides the influence from ECM composition, ECM density, and total adhesive surface. Using a detergent-free staining procedure (), we identified neo-formed rodlike structures as cytosolic reservoir for stress fiber–derived α-SMA (); rods were not formed on 10–20-μm-long islets (). Western blotting confirmed the dissociation of α-SMA from the TX-100–insoluble cytoskeleton after growth on small adhesion islets. ). In contrast, β-cytoplasmic actin and all other FA proteins tested persisted in the TX-100–insoluble cytoskeleton. To further evaluate whether FA size controls α-SMA localization in stress fibers in a tension-dependent manner, we treated myofibroblasts on 10–20-μm-long islets for 30 min with ROCK inhibitor Y27632 (2 μM; and Video 1, available at ) and the myosin II inhibitor blebbistatin (10 μM; Video 1). At these low concentrations, α-SMA redistributed from stress fibers to rodlike structures, whereas β-cytoplasmic actin distribution was not affected (). In contrast, inducing myofibroblast contraction on 2–6-μm-long islets with lysophosphatic acid and thrombin did not lead to α-SMA incorporation into stress fibers, but frequently resulted in cell detachment (unpublished data). To evaluate whether the dynamic enlargement of classical FAs to suFAs can rescue α-SMA localization in stress fibers, we developed a method to perform covalent μCP of ECM proteins onto extendable PDMS membranes ( and Fig. S1 C). Identical with printing on plastic (), stress fibers lost staining for α-SMA, but not for β-cytoplasmic actin, when myofibroblasts were grown for 12 h on membranes exhibiting islets of 4- and 6-μm-long (unpublished data). After gradually stretching the substrate to 135%, a significant fraction of myofibroblasts (43 ± 7%) reintegrated α-SMA into stress fibers within 6 h when 6-μm-long classical FA islets were enlarged to 8.1-μm-long suFA islets (). However, the same stretch applied to myofibroblasts grown on 4-μm-long islets and still producing classical 5.4-μm-long FAs () was not permissive for α-SMA stress fiber incorporation (6 ± 5%; ). Collectively, these results strongly suggest that FA size limits the maximum tension developed in stress fibers, as well as α-SMA recruitment. To quantify the level of intracellular tension that is developed upon suFA formation, we measured the forces exerted at suFAs in comparison with classical FAs. REF-52 fibroblasts and myofibroblasts expressing GFP-tagged β3 integrin and paxillin were grown for 4 d on deformable silicone substrates micropatterned with fluorescent markers ( and Fig. S1 D). Forces at FAs were calculated from the substrate deformation (, dot displacement) after relaxing cells with cytochalasin D (Video 3, available at ). In both cell types, force increased with FA area (). However, suFAs (>7.5 μm; , red) exerted an average stress of 12.5 ± 2.5 nN/μm, compared with 3.8 ± 5.1 nN/μm of FAs ≤7.5 μm (, yellow and blue) in the same myofibroblasts. The stress at small myofibroblast FAs was comparable to that of classical FAs of α-SMA–negative fibroblasts (3.1 ± 5.3 nN/μm; ).The high standard deviation for FAs ≤7.5 μm derives from the largely varying forces of focal complexes (≤1 μm; , yellow). In summary, suFAs allow the generation of approximately fourfold higher forces per unit area compared with classical FAs. To decipher the molecular basis of the significant higher stress resistance of suFAs and to test whether adhesion size dictates suFA composition, we produced islets of no particular ECM by “negative” printing of poly-L-lysine-g3.5-poly(ethylene glycol) (PLL-g-PEG) in serum-free conditions. By these means we confined myofibroblast adhesion to the nonprinted regions (Fig. S2 B), leading to the loss of α-SMA from stress fibers on islets <6 μm long (unpublished data) as observed on protein islets (). On 10- and 20-μm-long islets, myofibroblasts coexpressed β3 integrin and paxillin ( ) with β1 integrin (unpublished data), tensin, and vinculin () in suFAs at the cell periphery. In contrast, growth on 2–6-μm-long islets induced the redistribution of the fibrillar adhesion marker tensin () and β1 integrin (unpublished data) from the cell periphery to the cell center, whereas classical FA markers β3 integrin () and vinculin () remained specifically in the cell periphery. Paxillin was partly redistributed to the cell center, but also remained in peripheral FAs (). A similar redistribution of matrix adhesion components occurred after switching culture substrates from 780 to 9.6 kPa (Fig. S3, available at ). Moreover, the protein tyrosine phosphorylation level on suFA islets was significantly higher compared with growth on classical FA islets ( and F ). In particular, phosphorylation, but not total protein expression of paxillin and FAK, significantly decreased in total extracts from myofibroblasts on small islets (). Hence, adhesion size appears to control the molecular composition and phosphorylation of FAs in a tension-dependent manner. In this study, we identify α-SMA as a mechanosensitive protein that is rapidly recruited to β-cytoplasmic actin stress fibers under high tension. To develop this critical tension, stress fibers need to be anchored at sites of suFA, permitting an approximately fourfold higher stress development compared with classical FAs. We further show that limiting FA size alone interrupts the mechanical feedback loop of α-SMA–mediated contraction and increasing matrix stiffness, which is characteristic of the persisting myofibroblast activity in fibrocontractive diseases (; ; ). The specific recruitment of α-SMA to preformed stress fibers is consistent with the observation that β-cytoplasmic actin filament bundle formation precedes incorporation of α-SMA during myofibroblast spreading (). In the absence of this organization template, α-SMA accumulates in detergent-soluble rods that appear to function as transient cytosolic reservoirs ( ; ). We show that reducing stress fiber tension by reducing substrate stiffness, FA size, and myosin contraction leads to similar α-SMA rod formation. The α-SMA–specific NH-terminal sequence AcEEED likely contributes to the tension-dependent distribution of α-SMA because cytoplasmic delivery of this sequence removes α-SMA from stress fibers without affecting β-cytoplasmic actin localization (; ) and induces α-SMA rod formation (). We propose a mechanosensitive element within stress fibers that alters its affinity for the AcEEED sequence or its stress fiber association with changing levels of tension. Potential candidates are cofilin and gelsolin, which colocalize with α-SMA in rods (), and SM22 and MLC, which partly dissociate from stress fibers after tension release in our conditions. It is conceivable that stress fiber contractile function is rapidly adapted to a new mechanical challenge by modulating its molecular composition. This is supported by the recently demonstrated tension-dependent localization and turnover of α-actinin and MLC () and zyxin (), and by the partial loss of nonmuscle MHC after tension release in our experiments. During wound healing, α-SMA–mediated myofibroblast contraction is delayed until the ECM has been sufficiently remodeled by α-SMA–negative fibroblasts (; ). We determined an elastic modulus of ∼20 kPa of granulation tissue at the time of first appearance of α-SMA–positive stress fibers in myofibroblasts. This corresponds well with the observation that α-SMA recruitment to stress fibers of cultured myofibroblasts requires a significantly higher culture substrate stiffness of ∼16 kPa than neoformation of β-cytoplasmic actin stress fibers at 2–6 kPa (; ). We have used AFM to determine the elastic modulus of granulation tissue and of PDMS substrates at the micrometer level, which is more relevant for cell mechanoperception at focal adhesions than global tissue compliance (,). Despite a rather high variation in AFM measurements because of tissue heterogeneity, our values range in the same order of magnitude compared with macroscopic measurements () of fibrotic (myofibroblast-populated) tissue, which can exhibit elastic moduli up to 80 kPa, comparable with the elastic modulus of ∼50 kPa of late granulation tissue in our measurements. The elastic modulus of normal soft tissue is in a considerably lower rage of 1–20 kPa (; ). A stiffer matrix allows formation of larger FAs and FA size was previously shown to increase with external mechanical load in two-dimensional culture (; ). We propose that fibroblasts “sense” their mechanical microenvironment by assessing the level of intracellular tension, which is limited by the size of stress fiber anchors (). Consequently, restricting myofibroblast adhesions to the length of classical FAs (≤6 μm) by μCP on a stiff matrix limits stress fiber tension and results in the loss of α-SMA, similar to growth on a substrates with low (9.6 kPa) elastic modulus. Most interestingly, fibroblasts also appear to detect application of gradual stretch from changes in FA size. In our experiments using μCP on extendable membranes, the resulting formation of suFAs, but not the percentage of stretch, promotes α-SMA reorganization. FA size-dependent mechanosensing is of physiological relevance during the comparably slow remodeling and stiffening of three-dimensional matrices; in mechanically restrained collagen gels, fibroblasts first form pointlike adhesions that mature into classical FAs () and subsequently into suFAs (; ). This slow tension increase may escape detection by mechanosensitive membrane channels (; ). Previous studies have determined a linear relation between the size and the force of fibroblast FAs, with an average stress of ∼5.5 nN/μm (; ). We confirm this observation by considering only FAs ≤7.5 μm, exerting a mean stress of 3–4 nN/μm independent from myofibroblast differentiation. However, by surpassing the size threshold of suFAs (7.5 μm), stress exponentially increases to about fourfold, providing a new quality of force exertion. From the minimum size of α-SMA–permissive islets (8 × 1.25 μm) and the average stress exerted by suFAs (12.55 nN/μm), we estimate a minimum local force of ∼125 nN for promoting recruitment of α-SMA to stress fibers. How is this higher force per unit area achieved? One explanation is that the characteristic colocalization of classical FA markers vinculin and αvβ3 integrin with fibrillar adhesion markers tensin and α5β1 integrin in suFAs creates a particularly solid matrix anchor. The recruitment of new components to larger adhesion sites may be achieved by simply creating space or by opening cryptic binding sites (; ; ; ). Accordingly, restricting the length of myofibroblast FAs to ≤6 μm results in the separation of classical FA and fibrillar adhesion markers. Moreover, we observed a significant loss of FAK from TX-100–resistant suFAs. This is consistent with the strong reduction in protein tyrosine phosphorylation of highly phosphorylated suFAs, in particular of paxillin and FAK. FAK appears to be central for the mechanoresponses triggered at classical FAs (; ; ; ) and it is involved in adhesion-dependent myofibroblast differentiation (). Interestingly, paxillin is a substrate of FAK in suFA-related three-dimensional adhesions but not in classical FAs (). Reduction of tension-dependent phosphorylation may explain the “shuttling” of paxillin from suFAs to classical FAs after reducing FA islet size and substrate compliance in this study, and after inhibiting α-SMA–mediated myofibroblast contraction (). Consistently, paxillin is specifically recruited to classical FAs after stretch (). We propose the following model on how FA supermaturation provides a crucial checkpoint in controlling α-SMA localization in stress fibers and, thus, myofibroblast function in physiological conditions (). (a) The degree of matrix organization and stiffness determines FA size, which limits the level of intracellularly generated tension. (b) Surpassing a critical adhesion size and intracellular tension leads to the engagement of fibrillar adhesion proteins to classical FAs, presumably by FAK-mediated phosphorylation. (c) The resulting suFAs provide significantly stronger matrix anchors that permit generation of an approximately fourfold higher stress fiber tension. (d) This high tension is pivotal to permit α-SMA incorporation. In vivo, this checkpoint may ensure that enhanced myofibroblast contraction only occurs when the tissue has been sufficiently remodeled and stiffened for effective force exertion. REF-52s were cultured in DME (Invitrogen) containing 10% FCS; all major results were confirmed using primary rat lung myofibroblasts that were obtained and cultured as described previously (). 5 ng/ml TGFβ1 and 250 ng/ml TGFβ-RII (both R&D Systems) were added to the culture medium for 5 d; cytochalasin D (Sigma-Aldrich) was used at 0.1 μM, blebbistatin (Calbiochem) was used at 10 μM, Y27632 (Calbiochem) was used at 2 μM, and lysophosphatic acid (Sigma-Aldrich) was used at 10 μM. For FA force analysis, REF-52s were stably transfected with β3 integrin–GFP (gift of C. Ballestrem and B. Wehrle-Haller, University of Geneva, Geneva, Switzerland; ), full-length GFP-paxillin (gift of B. Geiger, Weizmann Institute of Science, Rehovot, Israel; ), and α-SMA–GFP () using Fugene 6 (Roche) and cultured for 4 d in Ham's F12/5% FCS (± TGFβ) on micropatterned substrates mounted as observation chambers (). Cells were permeabilized for 5 min with 0.2% TX-100 in 3% PFA and fixed with 3% PFA/PBS for 10 min. This fixation procedure was used for standard fluorescence, to quantify FA size with a segmentation algorithm () and to evaluate α-SMA association with stress fibers by counting only cells with clearly visible α-SMA stress fibers. To preserve detergent-soluble α-SMA rodlike structures, cells were fixed with 1% PFA/PBS for 10 min, followed by a 3-min treatment with MeOH at −20°C (); this technique was not compatible with FA protein staining. Primary antibodies were used according to . For secondary antibodies we used TRITC- and FITC-conjugated goat anti–mouse IgG1, IgG2b, and IgG2a (Southern Biotechnology Associates, Inc.); Alexa Fluor 350–, 488–, and 568–conjugated goat anti–rabbit antibodies; and Alexa Fluor 647–conjugated mIgG2a antibodies (Invitrogen). F-actin was probed with Phalloidin–Alexa Fluor 350/488/568 (Invitrogen) and DNA with DAPI (Fluka). Images were acquired with an inverted microscope (Axiovert 200M; Carl Zeiss MicroImaging, Inc.) equipped with a spinning disk confocal head (Nipkow CSU10; Yokogawa Electric Corp.) and charge-coupled device camera (CoolSNAP-HQ; Roper Scientific). FA area was quantified from confocal images by implementing a segmentation algorithm () in Matlab 7.0 software (The MathWorks). Movies were acquired using a water immersion objective 63×, NA 0.9 (Leica), mounted on an upright confocal microscope (model DM RXA2, with a laser scanning confocal head, model TCS SP2 AOBS; Leica) and equipped with a heating stage and CO incubation chamber. Fluorescent photoresist dots (Fig. S1 D) were excited at 543 and 594 nm, and emission was detected at 610–800 nm. All figures were assembled using Photoshop (Adobe). Silicone substrates of an 80-μm thickness were produced by mixing PDMS curing agent and base (Sylgard 184; Dow Corning) in ratios between 1:5 and 1:80. Young's modulus was determined according to and confirmed with AFM (see the next section). Micropatterned silicone substrates were created similar to a recently published method () with important modifications (Fig. S2 D). Our substrates were produced with an elastic modulus of 23 kPa to promote the formation of suFAs and α-SMA stress fiber incorporation, which was not achieved using 12-kPa substrates (). 23-kPa substrates were sufficiently compliant to obtain deformations in the micrometer range (∼10 pixels), which are required to apply correlation-based optical flow analysis. In brief, the cell-generated deformation field in micropatterned elastic substrates was retrieved by maximizing a cross-correlation parameter (; ; ) to match a subimage of the substrate under cell traction to its undistorted equivalent after addition of cytochalasin D. Our method relates to a finite-element approach () but with the improvement that the computed forces are restricted to the zones of effective force transmission, i.e., FAs. We have implemented a new regularization method considering each FA as a tessellation of pointlike forces because the assumption of one pointlike force per classical FA () was not applicable to large suFAs. An iterative approach was used to successively apply a local regularization to each FA in contrast to performing a unique regularization on the whole image (). All computational work, including image processing, was performed with Matlab. AFM was used to relate the elastic modulus of PDMS substrates (see the previous section) with the compliance of rat wound granulation tissue using a method established for polyacrylamide gels and sections of pig artery (). A total of 12 female Wistar rats (200–220 g) were used. After shaving the skin, full-thickness 20 × 20-mm wounds, including the cutaneous muscle, were made using surgical scissors in the middle of the dorsum on the first day of the experiments and were allowed to heal spontaneously. Rats were killed by CO anesthesia and granulation tissue was harvested 7, 8, 9, and 12 d after wounding, as previously described (). Fresh tissue was sectioned into ∼150-μm-thick slices that were wet mounted under pyramid-tip AFM cantilevers (Vecco USA; spring constant, 60 pN/nm). First, force-indentation profiles were produced on a 100 × 100-μm area using an AFM (model XE-120; PSIA, Inc.) and surface topography images were produced to exclude small vessels from the indentation measurements (). Second, 10 force-indentation curves were produced from each sample and the elastic modulus was fitted with a conventional Hertz cone model (); indentation was performed at rate of 2 μm/s. Sections from the same tissue were processed for immunofluorescence and controlled for α-SMA expression and stress fiber formation. PDMS stamps for μCP () were obtained from silicon wafer molds (Fig. S1 A) and incubated for 30 min with different protein solutions (10–100 μg/ml), quick-dried, and put in conformal contact with substrates for 1 min. Nonprinted regions were passivated with 0.1 mg/ml PLL-g-PEG (). Negative stamps were produced from positive stamp molds (Fig. S1 B). For PLL-g-PEG stamping, both the substrate and the stamp were treated for 15 s with oxygen plasma. To provide stretchable PDMS membranes with adhesion islets, we developed a new method of covalent μCP (Fig. S1 B); alternative protocols did not provide sufficient protein absorption to sustain cell traction forces () or did not reach the required resolution (). Myofibroblasts were grown for 12 h on μCP membranes and subjected to uniaxial linear stretch of 135% in five steps every 5 min. To assess the association of proteins with the cytoskeleton, fractions of TX-100–insoluble cytoskeletal and TX-100–soluble cytosolic proteins were produced () and run on 10% SDS gels, together with total cell lysates. For subsequent Western blotting, we used HRP-conjugated secondary antibodies goat anti–mouse and goat anti–rabbit (Jackson ImmunoResearch Laboratories) detected by ECL (GE Healthcare). The ratio between all digitized band densities of one blot was quantified (ImageQuant V3.3; Molecular Dynamics) and normalized to housekeeping vimentin expression. Experiments were performed at least three times unless otherwise stated. Mean values are presented ± SD and tested by a two-tailed heteroscedastic test. Differences were considered to be statistically significant at P ≤ 0.01. Fig. S1 gives an overview on the different micropatterning methods used in this study. Fig. S2 demonstrates that the incorporation of α-SMA into stress fibers remains restricted to islets >6-μm-long, even after increasing islet density. Fig. S3 shows that suFAs disassemble into classical FAs and fibrillar adhesions after myofibroblast growth on compliant substrates. The redistribution of α-SMA from stress fibers into cytosolic rods after inhibition of cell contraction is demonstrated in Videos 1 and 2. Video 3 visualizes suFA disassembly and cell relaxation after cytochalasin D treatment on a micropatterned deformable substrate. Online supplemental material is available at .

Endocytic proteins such as dynamin, amphiphysin, and epsin, which directly bind and deform liposomes into tubules in vitro, play critical roles in membrane fission and curvature during clathrin-mediated endocytosis (; ; ; ; ; ; ). Dynamin is required for some forms of clathrin-independent or caveolae-mediated endocytosis (). These proteins interact directly with membrane phosphoinositides via lipid-binding domains, such as the pleckstrin homology (PH) domain in dynamin, the Bin-amphiphysin-Rvs (BAR) domain in amphiphysin, and the epsin NH-terminal homology (ENTH) domain in epsin. The BAR domain is proposed to drive membrane curvature (). The actin cytoskeleton is critical for many fundamental cellular processes such as cell morphology, motility, and cytokinesis (; ). Growing evidence indicates that the actin cytoskeleton plays an important role in endocytosis (; ; ; ). Actin regulatory proteins such as neural Wiskott-Aldrich syndrome protein (N-WASP), cortactin, and Abp1 bind to endocytic proteins such as syndapin, dynamin, and intersectin and are recruited to endocytic active zones (; ; ; ; ; ). However, the role of the actin cytoskeleton in endocytosis is poorly understood. Recent work has revealed that both invagination and scission of clathrin-coated vesicles and local actin polymerization are highly coordinated, resulting in the efficient formation of coated vesicles (, ). The FER-CIP4 homology (FCH) domain is found in the pombe Cdc15 homology (PCH) family protein members and is highly conserved from yeast to mammals (; ). Most PCH proteins have the Src homology 3 (SH3) domain at the COOH terminus. PCH family members, including CIP4; formin-binding protein 17 (FBP17); Toca-1; syndapins/PACSINs; cdc15; and proline-serine-threonine phosphatase–interacting proteins (PSTPIPs), are known to be involved in cytoskeletal and endocytic events (; ; ; ; ; ; ). Syndapins/PACSINs and FBP17 are implicated in endocytosis by their abilities to bind to dynamin via their SH3 domain (; ). In particular, FBP17 induces tubular membrane invagination, suggesting that this protein generates the membrane curvature necessary for dynamin-dependent endocytosis (). In this regard, syndapins/PACSINs have been predicted to be potential BAR domain–containing proteins (). Interestingly, several PCH family members have been shown to bind to both WASP/N-WASP and dynamin, indicating that the PCH family is involved in actin cytoskeleton reorganization associated with membrane fission or protrusion (; ; ). All PCH proteins possess a highly conserved region that includes and extends beyond the FCH domain. The conserved region includes a predicated coiled–coil region, suggesting that this region is a novel functional domain. However, the exact functions of this region are unknown. We term this region the extended FC (EFC) domain and show that the EFC domain binds to phosphatidylserine (PS) and phosphatidylinositol 4,5-bisphosphate (PIP2). The EFC domain shows weak homology to the BAR domain, and the EFC domain alone tubulates liposomes in vitro. Importantly, the EFC domain–containing protein FBP17 is directly involved in EGF internalization, including plasma membrane invagination and actin polymerization, via recruitment of dynamin-2 and N-WASP. Expression of FBP17 has been shown to induce plasma membrane tubulation in COS-1 cells (). We identified the specific domain of FBP17 critical for membrane tubulation. By expression analysis of various deletion constructs in COS-7 cells, we found that 1–300 aa, comprising a sequence longer than the FCH domain, is required for membrane tubulation ( ). Expression of 1–300 aa, but not of 1–250 aa, deformed membranes into tubules that were longer than those induced by the full-length protein (). Alignment of these sequences in FCH domain–containing proteins revealed that the sequence homology extended beyond the FCH domain and included a predicated coiled–coil region of ∼100–200 aa ( and Fig. S1, available at ). Interestingly, this region showed weak homology to the BAR domain (Fig. S1). Residues inside of the helix packing of the BAR domain are well conserved (; Fig. S1). As expected, the coiled–coil region contributed to dimer formation of this region as determined by chemical cross-linking analysis (unpublished data). The basic amino acid residues of the BAR and the ENTH domain are essential for binding and tubulation of liposomes (; ; ). To determine if the basic amino acid residues of this extended domain are involved in the binding to liposomes, we performed mutagenesis analysis. Some sites composed of basic residues are conserved in the extended domain (). Therefore, we mutated these basic amino acid to Gln. Interestingly, expression of the K33Q + R35Q, K51Q + K52Q, or R113Q + K114Q mutant did not induce membrane tubulation in cells compared with that of wild-type extended domain (). In contrast, the K138Q + R139Q or K171Q + R173Q mutant induced tubulation similar to that of the wild-type extended domain (). Expression of other PCH proteins including CIP4 (1–300 aa), PSTPIP1 (1–415 aa), and PSTPIP2 (1–337 aa) also resulted in membrane tubulation in cells ( ). However, expression of FER (aa 1–300) resulted in little tubulation (). In FER, amino acid residues corresponding to K33, K51, and K52 of FBP17 are not conserved, but those corresponding to R113 and K114 are conserved (). K33 and K51, but not R113/R114, are conserved between the BAR domain and the extended domain (Fig. S1). The relation between sequence conservation and tubulation ability indicates that this region is evolutionally and functionally conserved and is related to the BAR domain. Because a portion of this region includes the FCH domain, we termed the entire conserved region the EFC domain, which is located at the NH terminus (). We next examined whether the EFC domain can bind to lipids because domains that deform membrane interact directly with membrane lipids (). We performed liposome sedimentation assays and found that the EFC domain of FBP17 bound most strongly to PIP2-containing liposomes ( ). Brain lipids (Folch fraction) or PIP3 also bound moderately to the EFC domain. Brain lipids are rich in PS (∼50% of total lipids), and the EFC domain of FBP17 bound to PS-containing liposomes (). However, it did not bind to lysophosphatidic acid, lysophosphocholine, or sphingosine-1-phosphate (unpublished data). The dissociation constant ( ) of the EFC domain for PIP2 binding was determined with a dual polarization interferometer and was found to be 395.3 nM, comparable to those of the PLCδ1 PH and the epsin ENTH domains, which are known to bind PIP2 with high affinity (; ; ). We then tested whether the tubulation activity of the EFC domain in cells correlates with its lipid-binding ability. Mutants with reduced tubulation activity showed decreased lipid-binding ability compared with that of wild-type EFC, indicating that direct interaction with phosphoinositides is necessary for membrane tubulation in cells (). The EFC domains of other proteins also bound to PIP2. For example, the EFC domains of CIP4, FER, PSTPIP1, and PSTPIP2 also bound strongly to PIP2 and to brain lipids (). These results indicate that the lipid-binding ability of the EFC domain is evolutionally conserved and that the EFC domain comprises a novel phosphoinositide-binding module. Expression of the EFC domain strongly induced membrane tubulation in cells, and it showed high affinity for phosphoinositides. We next examined whether it can deform artificial liposomes composed of brain or synthetic lipids, including PIP2. The PLCδ1 PH domain, which binds to PIP2 with high affinity, did not alter the shape of brain liposomes ( ). In contrast, the EFC domain of FBP17 induced tubulation of brain liposomes (), as did the BAR domain of amphiphysin2/Bin1 (Amph2; ; ). The tubulation of liposomes by the EFC domain and by the BAR domain was confirmed by electron microscopy (). The diameter of tubules induced by the EFC domain of FBP17 (∼200 nm) was larger than that of tubules induced by the BAR domain of Amph2 (∼50 nm; ). Because the EFC domain bound strongly to PIP2, it was incubated with PIP2-supplemented brain liposomes (although brain lipids themselves contain a small amount of PIP2). Increased PIP2 concentration enhanced the tubulation of brain lipids by the EFC domain; very long liposome tubules, up to ∼60 μm in length, were formed upon the addition of PIP2 (). To confirm the involvement of PIP2 on liposome tubulation by the EFC domain, we performed a series of experiments. The EFC domain did not induce tubulation of synthetic liposomes composed of phosphatidylethanolamine (PE)/phosphatidylcholine (PC)/phosphatidylinositol (PI) (). Addition of PIP2 to these liposomes reenabled the ability of the EFC domain to induce the tubulation (). Interestingly, addition of the PLCδ1 PH domain at a one- or threefold molar excess to that of the EFC domain decreased liposome tubulation by the EFC domain (). Furthermore, overexpression of the PLCδ1PH domain in cells strongly inhibited the tubulation (), whereas high concentrations of wortmannin, which is a PI 3-kinase inhibitor, did not affect tubulation at all (not depicted). Because the PLCδ1PH domain competed with the EFC domain, these data indicate that the binding of the EFC domain to PIP2 may enhance the membrane tubulation in cells and in vitro. We next examined the liposome tubulation by EFC mutants defective in the plasma membrane tubulation in cells. The K33Q + R35Q, K51Q + K52Q, and R113Q + K114Q mutants, which showed reduced lipid-binding ability, induced little tabulation (). In addition, the EFC domains of CIP4 and PSTPIP1 strongly induced liposome tubulation (). The EFC domain of FER also induced tubulation; however, the efficiency was less than that of other EFC domains (). Decreased tubulation in vitro may reflect the lack of tubulation in cells expressing the EFC domain of FER (). These data indicate that the EFC domain alone can induce membrane tubule formation by interacting directly with the lipid bilayer. FBP17 is thought to play a role in endocytosis (). Therefore, we studied the involvement of the EFC domain of FBP17 in the endocytosis of EGF. The uptake of Texas red–labeled EGF was decreased in cells overexpressing FBP17 ( ). When we observed cells with strong FBP17 expression, Texas red–labeled EGF remained on the surface and associated with tubules, as reported previously (). In contrast, EGF was internalized and observed as dots within the control cells (). Approximately 25% of FBP17-overexpressing cells showed a clear dots staining (, histogram). SH3 domain–deleted FBP17 also inhibited the endocytosis of EGF (). We next performed a quantitative assay that measures internalization of biotinylated EGF as a function of total bound biotinylated EGF. A small, but reproducible, reduction of internalization of EGF in total cells was observed upon expression of FBP17 under ∼20–30% of transfection efficiency (). Interestingly, this decrease appeared to be dependent on the entire EFC domain because the 1–56-aa deletion mutant did not inhibit endocytosis (). We next examined whether endogenous FBP17, CIP4, and Toca-1 are involved in endocytosis by RNA interference (RNAi) using small interfering RNA (siRNA). We confirmed that each siRNA selectively reduced the amount of FBP17 mRNA, Toca-1 mRNA, or CIP4 protein. (). In cells with reduced FBP17 or CIP4 expression, the uptake of Texas red–labeled EGF was reduced compared with that in cells treated with control siRNA by microscopic analysis (). Moreover, double depletion of FBP17 and CIP4 further reduced EGF internalization (). We next performed a quantitative assay that measures internalization of biotinylated EGF as a function of total bound biotinylated EGF. Cells with reduced FBP17 or CIP4 expression exhibited reduced EGF internalization compared with cells treated with control RNAi (). However, reduction of Toca-1 expression proved less effective. Double depletion of FBP17 and CIP4 more effectively reduced EGF internalization (). Importantly, decreased EGF internalization in response to RNAi treatment began after 5 min of EGF application to cells, indicating that these proteins are involved in the early steps of endocytosis. These data suggest that FBP17 and CIP4 play redundant roles in EGF internalization. Recent studies have shown that FBP17 binds to dynamin-2 and to N-WASP (; ), both of which play important roles in endocytosis (; ; ; ). Therefore, we examined whether FBP17 is colocalized with dynamin-2 and/or N-WASP. FBP17 that was expressed at a low level hardly induced tubulation. Instead, low-expressed FBP17 was localized to the plasma membrane of COS-7 cells in a punctate pattern ( and Fig. S2, A and B, available at ). FBP17 colocalized with dynamin-2 at the plasma membrane (Fig. S2 A). Moreover, FBP17 colocalized with EGF at the plasma membrane before the recruitment of Rab5, which is a marker protein for clathrin-coated vesicles/early endosomes (; Fig. S2 B), indicating that FBP17 functions in the early endocytic events, such as invagination. Importantly, FBP17 also colocalized with N-WASP at the plasma membrane at sites where actin polymerization was induced (). N-WASP is known to activate the Arp2/3 complex to induce actin polymerization. FBP17 also colocalized with Arp2 only at the cell periphery (). The SH3 domain of several EFC domain–containing proteins has been reported to bind to N-WASP (; ; ). The SH3-deleted FBP17 that was expressed at a low level was localized to the plasma membrane in a manner similar to wild-type FBP17, but did not colocalize with N-WASP, and actin polymerization was no longer observed (). Thus, FBP17 may function in endocytosis as an N-WASP activator, inducing actin polymerization via activation of the Arp2/3 complex. We next examined whether FBP17 activates the N-WASP–Arp2/3 complex, leading to actin polymerization. FBP17 activated the N-WASP–Arp2/3 complex–dependent actin polymerization in vitro (), supporting the notion that FBP17 recruits N-WASP to the plasma membrane via the SH3 domain and induces actin polymerization in an N-WASP–Arp2/3 complex–dependent manner. We examined the affinities of the SH3 domains of FBP17 and other PCH family proteins for N-WASP and dynamin-2. The SH3 domains of CIP4, FBP17, Toca-1, and PSTPIP1 bound to N-WASP and dynamin-2 ( ). In pull-down assays, N-WASP was more concentrated in SH3 domain precipitates than dynamin-2. (). The SH3 domains of CIP4, FBP17, Toca-1, and PSTPIP1 appear to bind to N-WASP and dynamin-2 in a similar manner. N-WASP and dynamin-2 form a protein complex in the early stage of endocytosis (). FLAG-tagged N-WASP was expressed with or without GFP-tagged FBP17. The amount of dynamin-2 in anti-FLAG immunoprecipitates was significantly increased upon expression of FBP17 (), indicating that dimerized FBP17 physically links N-WASP and dynamin-2 (). However, many other proteins bind to the SH3 domains of these proteins (). These unidentified proteins may play important roles in regulating FBP17, CIP4, and Toca-1. We next examined the localization of proteins involved in endocytosis in membrane tubulated by FBP17. EGFR was colocalized with FBP17 in invaginating tubules ( ). N-WASP was also localized in these invaginating tubules, whereas SH3-deleted FBP17 was not colocalized with N-WASP, but still induced membrane tubulation (). Moreover, N-WASP colocalized with dynamin-2 in these tubules (). Combined with the immunoprecipitation analysis, these data indicate that these proteins are able to form a functional complex. Recent studies have indicated that actin polymerization has critical roles in the fission of endocytic vesicles away from the plasma membrane (; ). Moreover, inhibition of actin polymerization increases deep invaginations at the plasma membrane (). Thus, we examined the relationship between actin polymerization and tubular invagination induced by FBP17. Overexpression of FBP17 induced tubules that did not merge with the actin cytoskeleton ( ). Interestingly, coexpression of FBP17 and N-WASP, which is expected to strongly induce actin polymerization (), decreased the plasma membrane tubulation induced by FBP17 (). Instead, FBP17 localized in a punctate pattern at the plasma membrane, where it colocalized with N-WASP and cortical actin structure (). Importantly, the protein complex including dynamin-2 and N-WASP was observed upon expression of both N-WASP and FBP17 (). Tubules induced by the SH3-deleted FBP17 were not significantly attenuated by overexpression of N-WASP (). Consistently, inhibition of actin polymerization by treatment of latrunculin B increased FBP17-induced plasma membrane tubulation (unpublished data). Low-expressed FBP17 exhibited a punctate pattern of localization, but latrunculin B treatment caused the elongated tubular structures by low expressed FBP17 (). Therefore, the degree of actin polymerization downstream of FBP17 appears to be important for tubulation. Importantly, a previous paper pointed out the importance of dynamin in tubulation; mutants of dynamin defective in binding to FBP17 enhanced the tubulation (). These data suggest that these tubules might be created by the deformation of endocytic vesicles, for which fission by the coordination of dynamin-2 and actin polymerization is required. In this study, we identified a novel domain, the EFC domain, which is related to the BAR domain. Half of the EFC domain was previously characterized as an FCH domain, but an additional sequence is required for interaction with the membrane. Our results provide the first evidence that the EFC domain of FBP17 directly binds to the membrane and deforms protein-free liposomes into tubules. Moreover, the EFC domains of other PCH family proteins, such as CIP4, FER, PSTPIP1, and PSTPIP2, also strongly bind to and tubulate liposomes ( and ). Conservation of both amino acid sequence and function indicate that the EFC domain is a membrane tubulation module that is dependent on lipid binding. The SH3 domain of FBP17 and that of other EFC domain–containing proteins bind to dynamin-2 and N-WASP. Dimerized FBP17 recruited N-WASP and dynamin-2 simultaneously ( and ). N-WASP and dynamin preferentially bind to PIP2 (; ). The EFC domain of FBP17 binds to PIP2 preferentially (). Therefore, these proteins form a functional complex at the PIP2-rich plasma membrane. Thus, FBP17 may provide links between membrane invagination, fission, and actin polymerization, mediated by FBP17, dynamin-2, and N-WASP, respectively (Fig. S3, available at ). We found that FBP17 is required for EGF endocytosis by both knockdown and localization studies (Fig. S2 and ). Importantly, both the EFC domain and the SH3 domain of FBP17 were required for the recruitment of N-WASP and dynamin-2 to the plasma membrane. Because tubulation induced by FBP17 expression is enhanced by decreasing the affinity of dynamin to FBP17 (), and because increased amounts of N-WASP decreased tubulation (), the lack of appropriate binding partners for FBP17 may induce tabulation, presumably via the inhibition of fission of endocytic vesicles. Consistent with this idea, low-expressed FBP17 formed tubular structures when actin polymerization was blocked by latrunculin B treatment (). FBP17-induced tubulation was also enhanced by the inhibition of actin polymerization (unpublished data). The Arp2/3 complex was colocalized with FBP17-induced tubules only at the cell periphery (), also indicating that the lack of proper actin polymerization machinery caused the tubulation. Therefore, actin polymerization that occurs downstream of FBP17 may be essential for fission of endocytic vesicles, together with dynamin. Actually, the actin cytoskeleton is essential for the internalization step of endocytosis at the plasma membrane in yeast and mammals (; , ; ). Dynamin plays an established role in the fission of endocytic vesicles (; ). Thus, imbalanced recruitment of dynamin-2 or of the actin polymerization machinery, including N-WASP, may induce the tubulation associated with FBP17 and other EFC domain–containing proteins. Other unidentified proteins that associated with the SH3 domains of PCH family proteins may also play important roles in endocytosis. Although the EFC domain alone was able to deform membrane in vitro, it is still unclear whether the EFC domain alone can deform plasma membrane into tubules in cells because the amount of dynamin-2 or N-WASP or the inhibition of actin polymerization affects tubulation induced by the EFC domain (). Thus, it is possible that the EFC domain senses the curvature of endocytic vesicles for recruitment of N-WASP and dynamin-2. In this regard, it is interesting that the diameter of tubules induced by the EFC domain in vitro was larger than that induced by the amphiphysin BAR domain (; ). Differences in diameters may relate to differences in functions. It has been reported that some PCH protein family members are involved in membrane-coupling processes, such as endocytosis, cell movement, and cytokinesis (; ; ; ; ). Proteins containing the EFC domain may be involved in these shape changes involving both the membrane and cytoskeleton. Human FBP17 complementary DNA (cDNA) was cloned in pEGFP-C1 (CLONTECH Laboratories, Inc.). Myc-tagged FBP17, Myc-tagged NH-terminal region deletion mutant (lacking 1–56 aa), and SH3 domain–deleted mutant (lacking 552–609 aa) were cloned in pEF-BOS plasmid vector. Human FBP17 (1–250, 1–300, 1–340, and 1–380 aa), mouse CIP4 (1–300 aa), FER (1–300 aa), PSTPIP1 (1–415 aa), PSTPIP2 (1–337 aa), BAR domain (1–286 aa) of amphiphysin2/Bin1, PH domain (130 aa) of PLCδ1, SH3 domains of FBP17 (553–609 aa), CIP4 (487–543 aa), Toca-1 (483–539 aa), and PSTPIP1 (362–415 aa) were obtained by RT-PCR. These sequences were confirmed, and then subcloned into pEGFP-C1 or pGEX vector (GE Healthcare). FBP17 (1–300 aa) were also subcloned into pCMV HA (CLONTECH Laboratories). The presence of GFP, Myc, and HA tag did not inhibit the tubulation of membrane in vivo. The pGEX-ENTH domain construct was made as described previously (). Mutagenesis was performed by PCR with mutated primers using Quikchange site-directed mutagenesis kit (Stratagene). Expression and purification of GST fusion proteins were performed using standard protocols. GST tag was cleaved from proteins by preScisson protease (GE Healthcare), but the presence of GST tag did not inhibit the binding or tubulation of liposomes. COS-7 cells and A431 cells were cultured in DME containing 10% FCS. Transfection was performed using Lipofectamine 2000 (Invitrogen) according to the manufacturer's protocol. Transfected cells on coverslips were fixed in 3.7% formaldehyde for 15 min, permeabilized with 0.1% Triton X-100 for 5 min, and immunostained with various antibodies. Texas red–conjugated EGF (Invitrogen) uptake was examined as previously described (). Human EGF was purchased from Invitrogen. Polyclonal anti–N-WASP antibody was used as described previously (). Monoclonal anti-FLAG antibody and latrunculin B were obtained from Sigma-Aldrich. Monoclonal anti-Myc antibody, polyclonal anti-EGFR, and anti–dynamin-2 antibodies were obtained from Santa Cruz Biotechnologies, Inc. Monoclonal anti–dynamin-2 and anti-CIP4 were purchased from BD Biosciences. Polyclonal anti-GFP antibody was purchased from MBL International Corporation. For visualization of actin filaments, Alexa Fluor 647–conjugated phalloidin (Invitrogen) was incubated with fixed cells for 30 min. Secondary antibodies conjugated with Alexa Fluor 488, 568, and 647 were obtained from Invitrogen. Liposome binding assay was performed as described previously (). Liposomes were prepared as follows. PE/PC liposomes consisted of PE (70%), PC (20%), and 10% PI or various phosphoinositides. Brain liposome was made of total bovine brain lipids (Folch fraction 1; Sigma-Aldrich). PE, PC, PS, and PI were obtained from Sigma-Aldrich. PI3P, PI4P, PI5P, PIP2, PIP2, and PIP3 were purchased from Cell Signaling Technology. PIP2 was obtained from Cell Signaling Technology and Echelon Bioscience, Inc. These lipid mixtures were resuspended at 1 mg/ml in 0.1 M sucrose, 20 mM Hepes, pH 7.4, 100 mM KCl, and 1 mM EDTA. To remove aggregated proteins, purified proteins were subjected to centrifugation at 70,000 rpm for 15 min at 4°C in a TL 100 rotor (Beckman Coulter). 5 μg of proteins were incubated with 100 μg of liposomes in 100 μl of buffer for 15 min at RT and centrifuged at 60,000 rpm for 15 min at 25°C. Supernatants and pellets were subjected to SDS-PAGE and stained with Coomassie brilliant blue. The intensity of protein bands was measured using Image J software (National Institutes of Health). Lipid overlay assay was performed using PIP strip membrane purchased from Echelon Biosciences, Inc. The interaction of the EFC domain with PIP2 using a dual polarization interferometer was investigated with Analight Bio 200 (Farfield Sensors, Ltd.) as described previously (). values were calculated from curve fitting. Liposome tubulation assay was performed as previously described, with some modifications (). Liposomes containing 95% brain lipid and 5% rhodamine-conjugated PE (rhodamine-PE); or 85% brain lipid, 5% rhodamine-PE, and 10% of the indicated lipid; or 70% PE, 15% PC, 5% rhodamine-PE, and 10% of the indicated lipid were suspended in 0.3 M sucrose, and large liposomes (up to 3 μm in diameter) were formed by vortexing. 0.1 mg/ml of purified proteins were incubated with 0.2 mg/ml of liposomes in buffer (20 mM Hepes, pH 7.4, 100 mM KCl, and 1 mM EDTA) at RT for 2 min and immediately examined using confocal microscopy (Bio-Rad Laboratories). In vitro liposome tubulation assay was also performed by electron microscopy, as described previously (). Stealth RNAi was purchased from Invitrogen. The siRNA sequence targeting as follows: FBP17, 5′-CCCACTTCATATGTCGAAGTCTGTT-3′; CIP4, 5′-GCAGTTGGAAGAACGCAGTCGTGAA-3′; Toca-1, 5′-GGCGCACAGAGTGTATGGTGAATTA-3′; and control random siRNA sequence, 5′-CCCTCGCACTGAGTCACCTTTGATT-3′. A431 cells were transfected with 10 μl of 20-μM siRNA and 4 μl of Oligofectamine reagent (Invitrogen) in a 6-well plate. After 24 h, a second transfection was performed, and the cells were cultured for 72 h and subjected to RT-PCR or various other experiments. The mRNA from RNAi-treated A431 cells was isolated using TRIZOL reagent (Invitrogen). A first-strand cDNA was synthesized from the mRNA by SuperScript first-strand synthesis system for RT-PCR (Invitrogen). The PCR amplification was performed using the first-strand cDNA. Amplification condition was as follows: 25 cycles of 30 s at 95°C, 30 s at 60°C, 60 s at 72°C, and a final extension of 10 min at 72°C. Actin was prepared from rabbit skeletal muscle and monomeric actin was purified using gel filtration on Superdex 200 (GE Healthcare) in G buffer (2 mM Tris-HCl, pH 8.0, 0.2 mM CaCl, 0.5 mM DTT, and 0.2 mM ATP). Purified actin was labeled with N-(1-pyrene) iodoacetamide (Invitrogen). Purification of Arp2/3 complex and N-WASP and in vitro actin polymerization assay were performed as described previously (). EGF endocytosis assays using biotinylated EGF were performed as described previously (), but with modifications. A431 cells were treated with siRNA as described in RNAi and RT-PCR. After 72 h, cells were starved with serum-free DME for 16 h. Cells were then incubated with 20 ng/ml of biotinylated EGF (Invitrogen) for 30 min at 4°C and moved to a 37°C incubator for the indicated times. To measure internalized EGF, cells were washed two times with ice-cold acid buffer (10 mM HCl and 150 mM NaCl, pH 2.0) to remove surface-bound EGF. To measure total bound EGF, cells were washed two times with ice-cold PBS instead of the acid wash. Cells were lysed in PBS containing 1% Triton X-100. Cell lysates were incubated with an ELISA plate coated with anti-EGF antibody (polyclonal; Abcam plc) at RT for 1 h. After being washed with PBS containing 0.05% Tween 20, the plate was incubated with streptavidin-peroxidase (Sigma-Aldrich) at RT for 30 min and bound, and biotinylated EGF was detected by reaction with its substrate, orthophenyl enediamine dihydrochloride. The absorbance was measured at 492 nm in an ELISA plate reader (Bio-Rad Laboratories). All photographic images were taken through a microscope (ECLIPSE E600; Nikon) with a confocal microscopy system (Radiance 2000; Bio-Rad Laboratories) at RT. Fluorochromes used include Alexa Fluor 488, 546, and 647 or rhodamine-labeled phalloidin (all Invitrogen). A 60× oil immersion objective, NA 1.40 (Nikon) was used. Images were assembled with Photoshop (Adobe). In each plate, photographs were cropped and each fluorochrome was adjusted identically for brightness and contrast to represent the observed images. The samples were electrophoresed in SDS-PAGE gels, stained with Coomassie brilliant blue, or transferred to polyvinylidene difluoride membrane, blocked with 5% nonfat dry milk in PBS and 0.1% Tween 20, incubated with primary antibodies, and then incubated with alkaline phosphatase–conjugated goat immunoglobulin secondary antibodies (Promega), followed by incubation with NBT/BCIP substrate (Roche). GST pull-down assay and immunoprecipitation were described previously (). Resulting gels or blots were scanned with a calibrated densitometer (model GS-710; Bio-Rad Laboratories) and quantified with Image J software. Fig. S1 shows an alignment of EFC and BAR domains. Fig. S2 shows colocalization between FBP17 and dynamin-2 or EGF at the plasma membrane. Fig. S3 shows a possible mechanism of involvement of FBP17 in EGFR internalization. Online supplemental material is available at .

Ca-triggered exocytosis depends on the coordinated action of several proteins to guide vesicles through a sequence of steps that begins with the docking of a vesicle to the plasma membrane and culminates with membrane fusion (; ). Among the most important proteins in this process are the synaptic SNAREs, which are members of a large class of membrane trafficking proteins. The synaptic SNAREs are dedicated to neurotransmitter and hormone release and include syntaxin (Syx), synaptobrevin (Syb), and synaptosome-associated protein of 25 kD (SNAP-25; ). These three proteins can induce slow fusion when reconstituted into liposomes (; ), and this fusion becomes Ca sensitive with the inclusion of the Ca sensor protein synaptotagmin (). Clostridial neurotoxins specifically cleave SNARE proteins to inhibit synaptic transmission (; ), and these reagents provide some of the most compelling evidence that SNAREs are essential for neurosecretion. SNAREs engage each other through long α-helical domains, termed SNARE motifs, to form a highly stable complex that is referred to as the SNARE complex (; ; ). The crystal structure of the heterotrimeric SNARE complex has revealed a parallel four-helical bundle with one helix from Syx, one from Syb, and two from SNAP-25. The core of this bundle comprises layers of mostly hydrophobic amino acid side chains (). Mutations that disrupt complex formation generally reduced fusion efficiency, supporting the view that the SNARE complex plays an important role in exocytosis (; ; ; ; ). It has been difficult to ascertain when the SNARE complex assembles during exocytosis and whether it forms during an early priming step or a later fusion step. In squid nerve terminals, clostridial neurotoxins and soluble SNARE fragments do not reduce the number of docked vesicles even as they poison synaptic transmission (). Genetic manipulations of the SNARE proteins generally reduce or block synaptic transmission (; ; ) without altering docking (; ; ), suggesting a role for SNARE proteins after vesicle docking. The Syx transmembrane segment is a structural component of the fusion pore (; ), and fusion pore stability is altered by manipulations of Syb () and SNAP-25 (). These results imply a role for SNARE proteins during the final fusion reaction. Drawing on the structure of the SNARE complex and its unusual stability, investigators have proposed that it assembles in a vectorial fashion from the membrane-distal portion to the membrane-proximal portion. This “zipping” of the SNARE complex would then perform an essential function in overcoming repulsive forces between the lipid bilayers of the vesicle and plasma membranes (; ; ; ; ). Clostridial neurotoxins cleave only uncomplexed SNAREs (), and their effectiveness in unstimulated chromaffin cells indicates that a large amount of uncomplexed SNARE protein is present in the absence of Ca (). At the crayfish neuromuscular junction, botulinum neurotoxin B, which binds to the membrane-proximal region of Syb, also blocked exocytosis without stimulation, but tetanus toxin, which binds to a membrane-distal part of Syb, only inhibited exocytosis after stimulation. These findings led to propose that the SNARE complex partially assembles at the membrane-distal end before Ca influx. Additional results showing differential effects in chromaffin cells of antibodies against SNARE proteins also support a partially assembled complex (). However, experiments in permeabilized PC12 cells argue that the SNARE complex assembles after Ca addition and before fusion (). In this study, we attempted to define more precisely what the synaptic SNARE proteins are doing in specific stages of exocytosis. SNARE motif mutants were expressed in PC12 cells, and exocytosis was investigated with carbon fiber amperometry. Kinetic analysis revealed multiple effects of these mutations on kinetic processes both before and after fusion pore opening. These results indicate that the SNARE complex undergoes conformational transitions both to open and then to dilate the fusion pore. Depolarization of cells transfected with cDNA encoding wild-type synaptic SNARE proteins elicited robust secretion similar to that seen in control cells transfected with vector lacking an insert ( ). The secretion in control cells was ∼50% higher than in previous studies from this laboratory () as a result of modified culture conditions (see Cell culture). Spikes in amperometric current traces signal the exocytosis of single vesicles, and we used the frequency of these spikes as a basic kinetic measure of release. Cumulative spike plots in provide a readout of the time course of secretion. In untransfected cells, control cells, and cells overexpressing SNAP-25, Syb (), synaptotagmin I, synaptotagmin IV (), NSF, or α-SNAP (), there was a lag of ∼2 s before secretion became vigorous. In contrast, cells overexpressing Syx showed an almost immediate response with a lag of only tens of milliseconds and a steep initial slope (, inset). The speed of secretion was quantified by measuring the frequency of spikes in the first two seconds of a recording from each cell and then averaging over cells. This initial rate of secretion was virtually identical to the initial slope of the corresponding cumulative spike plots in , but by averaging the frequency over cells, we obtained a useful estimate of the error (see Statistics). The initial rate of secretion in cells transfected with Syx cDNA was more than twice as high as in control cells or in cells transfected with Syb or SNAP-25 cDNA, and this difference was highly significant (). After the initial lag, secretion usually reached a maximum rate between 2.5 and 7.5 s. The frequency of spikes in this time interval provided a measure of the sustained rate of secretion, which, when averaged over cells, was virtually identical to the maximum slope of the cumulative spike plots in . These rates were roughly similar for cells overexpressing the wild-type SNARE proteins (). Syb and SNAP-25 produced small, insignificant increases in the sustained rate, and Syx reduced this rate along with the total number of fusion events (evident in the plateaus in ). However, the difference between the sustained rates from control and Syx experiments was not significant. The small enhancement with SNAP-25 is consistent with findings in chromaffin cells (). We note that after the KCl depolarization ended at 6.5 s, secretion continued at a steady rate for the duration of the recording in control cells. In contrast, secretion slowed within 4 s after ending the application of depolarizing solution in cells overexpressing any of the SNARE proteins (). To assess the effects of SNARE proteins on the stability of the fusion pore, we focused on the foot signal that precedes a spike (). These prespike feet (PSF) reflect catecholamine flux through open fusion pores, and their durations provide a measure of fusion pore stability. PSF duration distributions are generally exponential (; ; ) with a time constant equal to the mean PSF duration ( ). For cells overexpressing wild-type SNAREs, all mean PSF durations were within 7% of 1 ms, and none of the values differed significantly from the control value (). (Note that these values are shorter than the value of 1.38 ms reported previously for the vector control []. As noted in Cell culture, this change was observed after we started using coated dishes for cell culture). Because PC12 cells express all of these proteins endogenously (), these results suggest that the stability of fusion pores during exocytosis is not substantially altered by changing the ratios of these proteins. The PSF amplitude also was not altered by overexpressing the different SNARE proteins (see ). The constancy of fusion pore properties in the presence of widely varying ratios of SNARE proteins suggests that their actions on open fusion pores depend on a complex with a fixed stoichiometry. To investigate SNARE complex function during exocytosis, we prepared a series of mutations in the hydrophobic core of the SNARE motifs of Syx, Syb, and the NH and COOH chains of SNAP-25 (NH and COOH termini, abbreviated as SNAP-25N and SNAP-25C, respectively). and include many double point mutations to alanine in adjacent layers, as in . When alanine was the native residue, it was mutated to leucine, isoleucine, valine, aspartate, or tryptophan. In this study, we refer to mutations by protein/chain and layer according to . The specifics can be seen in . All of the mutations were intended to reduce the stability of the SNARE complex by either creating holes (replacing large side chains by alanine) or by introducing bulk to disrupt packing or charge to favor solvation. Many of our mutations have been characterized previously (; ) and subjected to stability analysis (; ; ; ). Previous voltage clamp measurements had shown that mutations in the Syx membrane anchor had no effect on Ca channels (). For this study, we measured Ca current in PC12 cells for two Syx mutants (layers −4/−3 and +4/+5), two SNAP-25 mutants (NH-chain layer +2/+3 and COOH-chain layer −1/−2), and one Syb mutant (layer +3/+4) along with the wild-type proteins. Mean peak Ca currents ranged from −59.5 to −66.9 pA and were indistinguishable from controls (Fig. S1, available at ). The Syx layer +4/+5 mutant has been implicated in the modulation of heterologously expressed Ca channels (), but no effects on endogenous Ca channels were seen with this mutant. The following two sections show that the mutants we tested have strong effects on exocytosis, and the failure of these mutants to alter Ca channels indicates that these mutants are altering exocytosis directly. These mutants, together with wild-type proteins, were also examined by immunocytochemistry to evaluate expression and localization (Fig. S2). Syx and SNAP-25 clearly were localized to the edges of cells, which is consistent with a plasma membrane location. Syb showed a punctal appearance and colocalized with another vesicle protein, synaptotagmin. Mutant SNAREs showed the same localization as the overexpressed wild-type proteins. The kinetics of release was studied by measuring the initial and sustained rates in cells expressing these mutants after the analysis of wild-type proteins described above (). Most of the core mutations in either the NH or COOH chain of SNAP-25 reduced secretion ( ). All of the SNAP-25N mutations except the layer −7/−6 mutation reduced the initial rate of secretion significantly (, top right). Small reductions were seen in the sustained rate of secretion as well, but none of these changes were significant except for the layer −2/−1 mutation, in which the effect was only significant by the less stringent test (P = 0.02; , bottom right). Most of the SNAP-25C mutations reduced the sustained rate significantly (, bottom right), but these mutations only slightly reduced the initial rate (the −5/−4 layer mutation effect was significant by the test; P = 0.007; , top right). SNAP-25C mutations were effective in reducing secretion along the entire length of the SNARE motif, which is in contrast to the experiments of , in which mutations in the hydrophobic layers followed a trend of stronger rescue of secretion toward the membrane-distal end. Neither leucine nor tryptophan at layer +7 reduced either rate significantly, but only the tryptophan results are plotted in . Thus, the NH and COOH chains of SNAP-25 had opposite selectivities in their influences on the initial and sustained rates of secretion. As with SNAP-25, most mutations in Syb reduced secretion (, left). Syb mutations more strongly affected the initial rate (, top right) than the sustained rate (, bottom right). Most of the changes in the initial rate were statistically significant. The changes in the sustained rate were not, except for the layer +3/+4 mutant. Mutations in Syx also reduced secretion (, left). Like the Syb and SNAP-25N mutations, most of the Syx mutations significantly reduced the initial rate (, top right) and had smaller, insignificant effects on the sustained rate (, bottom right). The Syx layer −7/−6 mutation reduced the sustained rate by an amount that was significant only by the test (P = 0.02). The Syx layer −4/−3 mutation was unique in enhancing the initial rate, and this effect was large and highly significant. The Syx layer 0 mutation reduced the initial rate but enhanced the sustained rate, which is in contrast to the effects of the layer 0 mutation of SNAP-25C, which reduced both rates in parallel. We next examined the effects of these mutations on PSF to determine how the SNARE complex influences fusion pores. The selected traces show long PSF ( ) and are representative. When SNARE complex mutations altered the PSF duration, it almost always became longer, suggesting these mutations tended to enhance fusion pore stability. All four chains had some sites where mutations stabilized open fusion pores. However, these sites were limited to certain parts of each chain. In Syx, only mutations in layers −7/−6, −4/−3, and −2/−1 increased PSF duration significantly by one-way ANOVA. The Syx layer +4/+5 and layer +7 mutations had small effects that were only significant by the test (P = 0.04 and 0.02, respectively). Syx and Syb mutations with greater effects tended to cluster at the two termini. In SNAP-25N, four mutations in sequence from layer −5 through layer +3 significantly increased the PSF duration. The only mutation that reduced PSF duration anywhere in the SNARE complex was in SNAP-25N at layer −7/−6, and this effect was small (∼15%) but significant. The SNAP-25N layer −7/−6 mutation reduced the PSF duration and increased secretion (), and although both changes are opposite to those seen with the majority of mutations (secretion rates generally fell and PSF durations became longer), the changes still showed the same parallel. Mutations in the COOH chain of SNAP-25 that increased the PSF duration fell between layers −7 and +2. However, the layer 0 mutation had no effect. In general, the same layers in the NH and COOH chains of SNAP-25 changed the PSF duration in a similar manner, but this pattern did not extend to Syx and Syb. Many of the mutants tested in this study were previously reported to form SNARE complexes with lower melting temperatures (see supplemental material, available at ). Other mutants are homologous in design and should also destabilize SNARE complexes. To evaluate our results in terms of SNARE complex stability, we performed linear regression analysis on the initial rate of secretion, the sustained rate of secretion, and PSF duration against the published melting temperatures for wild-type SNAP-25 and seven SNAP-25 mutations (Fig. S3, A–C). No significant correlations were evident. Mutants that destabilize the SNARE complex had varied efficacies in rescuing norepinephrine secretion in PC12 cells (). This lack of correlation with melting temperature suggests that if the complete disassembly that occurs during an in vitro melting experiment is a physiologically relevant transition, it occurs during processes other than those studied here. The changes in the rate processes in exocytosis seen here are more likely to reflect perturbations of other forms of structural transitions in complexes of SNARE proteins. Overexpressing wild-type Syx accelerated the initial rate of secretion and nearly eliminated the initial lag between depolarization and maximal release (). Because of this remarkable enhancement, additional mutations outside the Syx SNARE motif were tested. Deleting the large NH-terminal H domain of Syx (ΔH) reduced the initial rate significantly compared with wild-type Syx, reducing the rate to a value close to that of control cells ( ). Thus, the H domain is essential to the initial burst of secretion produced by wild-type Syx. ΔH did not alter the sustained rate or the PSF duration (). Deleting the Syx transmembrane domain (ΔTM) or increasing the length of the linker between the SNARE motif and the TM (Linker) reduced secretion (). Both mutations profoundly reduced the initial rate () and had weaker effects on the sustained rate, but the effect of Linker was significant (). These results agree with studies in chromaffin cells () and liposome fusion assays (). Effects on the PSF duration were small, but the increase in PSF duration produced by linker was significant (). The goal of our experiments with SNARE motif mutations was to determine the role of the SNARE complex in secretion. Because NSF and α-SNAP disassemble the SNARE complex (; ), these proteins provide another way to perturb the SNARE complex. Overexpressing these proteins alone or together did not alter secretion rates ( ). The slight reduction in the initial rate was not significant (). The PSF duration was not altered by NSF alone, but α-SNAP either alone or together with NSF increased the PSF duration (). This result is consistent with the results of SNARE motif mutations on PSF duration and may also be relevant to the slowing of release at a squid synapse by an NSF peptide (). It is notable that two independent methods of perturbing the SNARE complex, overexpressing SNARE mutants and overexpressing NSF/α-SNAP, both stabilized fusion pores. The Syx transmembrane segment forms the part of the fusion pore through the plasma membrane and contains residues that influence the flux through open fusion pores. Mutations in this region typically alter flux by ∼15%, with changes as high as 23% (; ). Outside of the membrane anchor, mutations throughout the SNARE complex completely failed to alter the PSF amplitude. In these measurements from cells transfected with 39 constructs, PSF amplitude was typically within 5% of control, with the largest change being 10%. One-way ANOVA indicated that none of the values differ significantly ( ). This demonstrates a striking specificity of influences on fusion pore flux and strengthens our previous conclusion that the Syx membrane anchor forms part of the fusion pore. xref fig #text Norepinephrine release was recorded with a VA-10 amperometry amplifier (ALA Scientific) using 5 μm carbon fiber electrodes polarized to 650 mV (; ). Cells were bathed in a solution containing 150 mM NaCl, 4.2 mM KCl, 1 mM NaHPO, 0.7 mM MgCl, 2 mM CaCl, and 10 mM Hepes, pH 7.4. Secretion was evoked using a Picospritzer (General Valve Corp.) to eject a solution similar to the bathing solution but with 105 mM KCl and 5 mM NaCl. Amperometry spikes were analyzed with the aid of a computer program written in our laboratory by P. Chang (University of Wisconsin, Madison, WI). Cumulative spike counts were based on spikes with peak amplitudes ≥2 pA (∼10× the baseline root mean square noise) and were plotted versus time at intervals of 250 ms. PSF were measured for spikes with amplitudes from 20 to 100 pA. The PSF parameters were extracted based on the criteria of . For the determination of mean PSF duration, values were combined across cells, as this quantity shows no significant cell/recording dependence (Wang et al., 2006). PSF lifetimes were sorted into 0.25-ms bins, and the distributions were fitted to a single exponential with Origin software (Microcal). The means obtained in this way were indistinguishable from the arithmetic mean after correction for missed events. The standard errors from these fits were identical to those obtained from a calculation of the arithmetic mean (). Measurements of kinetic parameters were stable during the course of these studies. For example, dividing our wild-type Syx data on PSF duration into two groups from a 3-yr period gave 0.97 ± 0.06 ms ( = 355) for the first half of the study and 1.06 ± 0.15 ms ( = 204) for the second half. PC12 cells were cultured as previously described () and transfected using an electroporator (ECM 830; BTX). The pIRES2-EGFP vector (CLONTECH Laboratories, Inc.) was used to overexpress proteins of interest. GFP fluorescence identified cells that were successfully transfected (efficiency of ∼30%). A previous study has shown that the overexpression of Syx mutants in this system generally produced levels of protein in transfected cells that exceed the levels of the endogenous protein by ∼10-fold (). Cells were transferred to dishes coated with collagen I and poly--lysine (BD Biosciences) and loaded with 1.5 mM norepinephrine and 0.5 mM ascorbate 14–16 h before experiments. Amperometry was performed 48–96 h after transfection. We note that growing cells in collagen I and poly--lysine–coated dishes improved cell adherence and altered secretion. Cells cultured in these coated dishes produced ∼50% more spikes and had briefer PSF durations (see Results). Every transfection with mutants was performed together with a transfection of the wild-type protein, and both mutant and wild type were studied in parallel. Wild-type proteins were subcloned into the pIRES2-EGFP vector using pairs of restriction enzymes. Single and double point mutations were generated using QuikChange site-directed mutagenesis PCR (Stratagene). Syx was truncated by removal of the membrane anchor 23 amino acids from I266 (ΔTM) or the NH-terminal H domain to I186 (ΔH). The Syx-linker construct (provided by J. Rothman, Columbia University, New York, NY) contains 11 extra amino acids with a sequence of LGGSGGSGGSK between K265 and I266L (). NSF and α-SNAP were provided by P. Hanson (Washington University, St. Louis, MO; ; ). All constructs were subcloned into the pIRES2-EGFP vector and verified by sequencing. Rates of secretion were determined for individual cells and averaged. Standard errors for the initial and sustained rates and for PSF amplitudes were calculated using the number of cells, recognizing that cell-to-cell variation is the principle source of error in such measurements (). Statistical analysis was performed with Instat for one-way ANOVA and Prism (GraphPad Software, Inc.) for two-way ANOVA. One-way ANOVA was applied to groups of measurements displayed together in the panels of figures presented in this study. When this test rejected the hypothesis that all of the measurements within that group had the same mean, Dunnett's post-test was used to identify measurements from mutants that differed from the wild-type protein. Because Dunnett's test is conservative, we also applied the test, and, in five cases, this test detected significant changes when Dunnett's test failed. These are mentioned in the text. We note that this is a reasonable number of uncertain cases of statistical significance for >100 evaluations performed in this study on different rate processes. The first part of the online supplemental data presents Ca current data that was obtained with the whole-cell patch-clamp. Fig. S1 shows that the Ca currents were identical in cells transfected with wild-type protein and selected mutants. The second part of the online data examines protein localization. Fig. S2 presents fluorescence immunostaining that shows that the overexpressed wild-type and mutant proteins are found in the correct places. Syx, SNAP-25, and selected mutants of these proteins appear on the edges of cells; Syb and a mutation show a punctal appearance and colocalize with synaptotagmin. Fig. S3 shows linear regression analysis of kinetic parameters versus the SNARE complex melting temperature to make the point that the kinetic perturbations are not correlated with the stability of SNARE complexes in vitro. Online supplemental material is available at .

SNAREs are required for membrane fusion in the eukaryotic secretory pathway (; ; ). The concerted assembly of SNARE subunits is carefully regulated at many levels by intrinsic protein conformations and extrinsic regulatory proteins. Characterization of both the molecular properties and assembly of the SNARE complex is imperative to understand mechanistic details of membrane fusion. SNARE complex assembly at the plasma membrane begins with a binary association between the syntaxin component (the t-SNARE heavy chain) and the SNAP25 homologue (t-SNARE light chains), resulting in a functional t-SNARE complex. In the case of the yeast plasma membrane homologues (Sso1p or Sso2p and Sec9p), the formation of this binary complex (three SNARE domains) is rate limiting for the overall process of SNARE complex assembly (). Although the subunit composition of the yeast plasma membrane t-SNARE complex is clearly one Sso1p or Sso2p and one Sec9p (; ), the stoichiometry of the neuronal counterpart is debated. Increasing evidence suggests that four SNARE domains form a t-SNARE complex with two syntaxin1A proteins and one SNAP25 in vitro (; ; ). The functional consequences of a four-stranded t-SNARE complex remain unclear because this species has yet to be demonstrated in vivo. However, most t-SNARE complexes that form on internal membranes use three different proteins to form a functional t-SNARE (). In this case, one syntaxin family member serves as a t-SNARE heavy chain, and two nonsyntaxin proteins provide t-SNARE light chain function. The v-SNARE, imbedded in the vesicle membrane in vivo, associates with the t-SNARE complex to complete the ternary complex. In all known instances, a single, membrane-integral protein provides v-SNARE function. High resolution crystal structure determination of a stable proteolytic fragment of the neuronal ternary SNARE complex showed that the assembled ternary complex is a parallel ∼12-nm, four-stranded helical bundle with one helix contributed by syntaxin1A, one from vesicle-associated membrane protein, and two helices from SNAP25 (). Syntaxins exhibit various conformations that are an intrinsic part of SNARE complex formation. Biophysical characterization of SNARE proteins in various free and complexed states has yielded important conformational information (; ; ; ). Free syntaxins are almost entirely α-helical, whereas SNAP25 and Sec9p as well as the v-SNAREs, VAMP2 (vesicle-associated membrane protein 2), and Snc1/2p are unstructured in solution (; ; ; ). Secondary structure is induced in t-SNARE light chains when they associate with the syntaxin component during t-SNARE complex formation. Similarly, α-helical structure is induced in the v-SNARE as it enters the ternary complex (,; ). One of the first indications that the various conformational states of syntaxin1A are functionally important came from studies examining the interactions of the SNARE recycling machinery, SNAP and NSF, with syntaxin1A. Upon ATP hydrolysis, NSF promoted a conformational change in syntaxin1A (referred to as syntaxin* in ) that made it refractory to further SNARE binding. The physical basis for this change is likely mediated through the binding of an NH-terminal domain back onto a COOH-terminal segment, which prevents further protein–protein interactions (). Structural analysis has confirmed this association between the NH and COOH termini of syntaxins (; ). Although the conformational gymnastics of syntaxins are well documented, the precise in vivo role for the various states remains undetermined. All syntaxins appear to have a large NH-terminal regulatory domain (NRD; also called the H domain; ). The NRD exhibits inhibitory functions in vitro. Binding studies have documented that the binary association between the syntaxin heavy chain and the t-SNARE light chains are adversely affected by the presence of this sequence (; ). The strength of the interaction between the NRD and the syntaxin core SNARE domain appears to be much stronger for yeast Sso1p than for neuronal syntaxin1A. Removal of the NRD in Sso1p results in a 3,000-fold increase in the SNAP25 homologous region of Sec9p (Sec9c) binding (), whereas similar experiments with syntaxin1A and SNAP25 showed a much more modest sevenfold increase in t-SNARE complex formation in the absence of the syntaxin1A NRD (). Additionally, the NRD region of syntaxin 1A is completely dispensable for in vitro fusion (), yet the NRD of Sso1p is required for plasma membrane SNARE function in vivo for unknown reasons (). This study further characterizes the function of the Sso1p NRD in vivo using chimeric proteins that alter the NH-terminal sequence of Sso1p. We found that replacing the NRD of Sso1p with the homologous sequence from the neuronal plasma membrane SNARE, syntaxin 1A, did not restore function. However, when the t-SNARE complex was made intramolecular by physically linking the SNAP25 homologous region of Sec9p to Sso1p without the NRD sequence, in vivo function was restored. Conformation of an intramolecular t-SNARE was demonstrated by a single point mutation in the Sec9c portion of the t-SNARE chimera (Q468R) that abrogated Sso1p function. These results suggest that the NRD is primarily involved in facilitating t-SNARE complex formation or preventing inappropriate associations with the H3 domain of Sso1p in vivo. We began by confirming that, like syntaxin1A, the NRD of Sso1p was not required for in vitro fusion. Recombinant His8-Sso1p and GST-Sso1p-ΔNRD proteins were used to form t-SNARE complexes with GST-Sec9c in detergent, which were subsequently reconstituted into phosphatidylcholine/phosphatidylserine liposomes by detergent dilution and dialysis (). shows that Sso1p lacking the NRD sequence (closed circles) fuses comparably to full-length Sso1p (open circles) with fluorescent Snc1p liposomes. Recent work with proteoliposomes produced by detergent-assisted insertion into preformed liposomes yielded a similar result (). We also examined the function of Sso1p-ΔNRD in vivo using a haploid plasmid shuffle strain that contains a genomic deletion in both and genes (JMY128). The viability of this strain is maintained by a low-copy plasmid expressing Sso1p under the control of its endogenous promoter (pJM198). The Sso1p-ΔNRD plasmid, driven by the galactose-inducible promoter, was transformed into the plasmid shuffle strain and maintained on glucose. Sso1p-ΔNRD expression was induced by shifting to growth on 2% galactose and was plated onto synthetic complete media lacking histidine with the drug 5-fluoroorotic acid (5-FOA). This drug is metabolized to a toxic intermediate in cells that express the gene product, thereby counterselecting for this marker. Because the wild-type Sso1p on the plasmid is required for viability, only coexpressing plasmids that produce a functional Sso1p will survive in the presence of 5-FOA. shows threefold serial dilutions of cells spotted on 5-FOA containing media. Wild-type Sso1p-HA driven by the promoter was also included as a positive control. These data show that Sso1p-ΔNRD cannot support the required function of Sso1p in vivo, similar to previous results with different length NH-terminal truncations (). The validity of this negative result was bolstered by the conformation of Sso1p-ΔNRD protein expression () and, more importantly, by appropriate localization to the yeast plasma membrane (). What is the role of the NRD in vivo? Several possibilities can be envisioned. The NRD of Sso1p could interact with regulatory proteins that control fusion independently of t-SNARE complex formation, thereby serving as a scaffold for recruiting other proteins to the site of fusion. Alternatively, the NRD could serve as an intramolecular chaperone preventing inappropriate associations with the Sso1p SNARE “core” domain (also known as the H3 domain; ). This function would also control access of Sec9p to the H3 region of Sso1p, thereby serving as a kinetic barrier regulating t-SNARE complex formation. We can begin to discriminate between these possibilities by modulating the ability of Sso1p and Sec9p to form t-SNARE complexes in vivo. If the NRD is regulating access by chaperoning the H3 domain, then increasing the local concentration of Sec9p, which is normally present at levels 5–10-fold less than Sso1p in wild-type yeast (see ; ), may allow yeast to survive without the NRD of Ssop. However, if the NRD is a scaffold, it will be required regardless of the extent of t-SNARE complex formation. An increase in overall t-SNARE complex formation could be achieved in a variety of ways, including an increase in global expression of soluble Sec9p by plasmid-based overexpression. However, this may not substantially increase Sec9p levels that have intimate contact with Sso1p or Sso2p. The ultimate means to achieve a stoichiometric level of Sec9p at the precise place where Ssop is located would be to covalently attach Sec9p to Sso1p, resulting in all three helices of the t-SNARE complex being expressed as a single protein. Such a chimeric protein would allow an intramolecular t-SNARE complex to form and remove any kinetic barrier. To this end, we created a series of chimeric proteins termed tandem t-SNAREs that physically link the SNAP25 homologous region of Sec9p to Sso1p. illustrates a schematic representation of these proteins. We appended Sec9c (amino acids 401–651) and an additional copy of the Sec9p interhelical region (IHR; amino acids 499–588) in between Sec9c and Sso1p to the NH terminus of full-length Sso1p (tandem t-SNARE). The additional Sec9p IHR was added to produce a chimera that contains sufficient conformational freedom to form the parallel three-helix bundle t-SNARE complex. Crystallographic analysis () determined that the NH terminal end of the H3 domain begins at residue 185. The next chimera removes the NRD and linker sequences, leaving the H3 domain, the juxtamembrane region, and the transmembrane domain (amino acids 179–290; tandem t-SNARE–ΔNRD). These chimeric proteins were expressed under the control of the inducible promoter from a high-copy number vector; however, expression levels were within about one- to twofold of episomal Sso1p in the haploid plasmid shuffle strain and endogenous Sso1p and Sso2p in a wild-type strain ( , , and not depicted). Expression was examined by immunoblotting with an anti-Ssop and anti-Sec9p antibody ( and ; see ). ). The functionality of the tandem t-SNARE plasmids was tested in the haploid plasmid shuffle strain (JMY128). shows threefold serial dilutions of cells spotted on 5-FOA containing media. Wild-type Sso1p was also included as a positive control. These data show that the full-length tandem t-SNARE is able to grow in the presence of 5-FOA nearly to the same degree as wild-type Sso1p. Cells carrying only the parental vector were unable to grow as expected. This result demonstrates that Sso1p function is not inhibited when Sec9c is physically linked to the NH terminus of Sso1p. If the tandem t-SNARE is capable of providing sole Ssop function as our drug selection suggests, then the drug-selected strain, or “postshuffle” strain, should only contain the tandem t-SNARE plasmid under the control of the –inducible promoter as the sole source of Sso protein. Growth of this strain is completely dependent on galactose as a carbon source, suggesting that the tandem t-SNARE is the only source of functional Sso1p (unpublished data). To examine the expression levels of all sources of Sso1p, we made total yeast extracts from cells harvested before and after selection on 5-FOA. These extracts were probed with polyclonal antibodies directed against both isoforms of Ssop to detect the tandem t-SNARE and wild-type Sso1p. demonstrates that the tandem t-SNARE is the only source of Sso1p in the postshuffle strain and that it is only modestly expressed (2.2-fold above episomal Sso1p) in the plasmid shuffle strain. Proper protein localization to the plasma membrane was also confirmed by immunofluorescence microscopy (). We also measured the growth rate of the tandem t-SNARE strains before and after selection. The doubling time for cells expressing both wild-type Sso1p and the tandem t-SNARE were substantially longer than those with only wild-type Sso1p (257 vs. 497 min). Growth is restored to approximately normal rates (224 min for wild type vs. 237 min) when the tandem t-SNARE is the only source of Sso1p in the postshuffle strain. This suggests that the tandem t-SNARE behaves as a slightly dominant-negative protein when both forms of Sso1p are present and that the dominant-negative effect of the tandem t-SNARE is caused by an interaction with wild-type Sso1p. Evidence of the tandem t-SNARE dominant-negative effect can also be seen in the BY1 genetic background used for analysis of function (see ). In accordance with the near wild-type growth rate, the extent of secretion measured by external invertase or α factor was similar in cells expressing Sso1p-HA or the tandem t-SNARE (Fig. S1, available at ). To examine the contribution of the NRD, we next analyzed the tandem t-SNARE lacking the NRD portion of Sso1p (tandem t-SNARE–ΔNRD) in a similar manner. Cells expressing the tandem t-SNARE–ΔNRD were selected on 5-FOA () in parallel with a wild-type Sso1p control plasmid and the NRD-deleted Sso1p without the addition of Sec9c. Similar to the full-length tandem t-SNARE, yeast expressing both wild-type Sso1p and the tandem t-SNARE–ΔNRD were slow growing, indicating a dominant-negative phenotype (doubling time of ∼541 min). However, growth was restored to near normal rates after selection against the wild-type plasmid in the postshuffle strain (224 min for wild type vs. 241 min). Additionally, secretion of invertase in the postshuffle strains does not appear to be impaired (Fig. S1). These results suggest that the tandem t-SNARE–ΔNRD is fully functional, similar to the full-length tandem t-SNARE. Protein expression () and localization () were also confirmed. This experiment clearly demonstrates that the Sso1p NRD is no longer required for cell viability when Sec9c is covalently attached. In addition to counterselection on 5-FOA, the functionality of the tandem t-SNARE–ΔNRD was also examined by tetrad dissection (unpublished data). Although sporulation and germination are very poor (; ; ), haploid spores could be isolated that were disrupted for both chromosomal copies of Sso1p and Sso2p but contained the tandem t-SNARE–ΔNRD as the only source of Sso protein (). The expression level of the tandem t-SNARE–ΔNRD was reduced relative to the full-length tandem t-SNARE. Although the tandem t-SNARE–ΔNRD is expressed from a 2-μm based multicopy plasmid under transcriptional control of the strong promoter, the expression level of the chimeric protein is slightly less than (∼95%) what Sso1p expressed under the control of the SSO1 promoter from a centromeric plasmid (). The protein at approximately the same molecular weight as Sso1p in the postshuffle and postdissection extracts (, asterisk) is a proteolytic fragment derived from the tandem t-SNARE–ΔNRD detected by both the Ssop and Sec9c antibodies (). Next, we examined the possibility that any sequence appended to the NH terminus of ΔNRD-Sso1p might maintain Sso1p function. It is possible, although unlikely, that Sec9c is simply masking the Sso1p H3 domain and not forming a functional t-SNARE complex, and any sequence that interacts with the H3 domain will suffice. To test this, we added the NRD (amino acids 1–182) of the homologous neuronal syntaxin rat syntaxin1A. Although Sso1p and syntaxin1A are only 27% identical (40% similar) in amino acid sequence, both fold into a very comparable closed conformation. When the structure of Sso1p was compared with syntaxin1A in the syntaxin–n–Sec1 complex (a stabilized closed conformation), the root mean square deviation was only 1.1 Å for 111 Cα pairs (). Although the syntaxin1A NRD is structurally similar to the endogenous Sso1p NRD, the rSyn1A-Sso1p chimera was unable to provide Ssop function even though it was well expressed and localized to the plasma membrane (). Although we are confident that the covalently attached Sec9c and Sso1p are working in tandem as an intramolecular t-SNARE complex, we sought to confirm this by creating a specific point mutation in Sec9c that should abolish function. We mutated a conserved glutamine residue in helix A of Sec9c (Q468 in the full-length Sec9p) to an arginine. This residue is one of four residues, one from each SNARE core helix, that compose the so-called “zero layer” or ionic layer in the SNARE coil (). In the prototypical neuronal SNARE structure, as well as in the yeast plasma membrane counterpart, these four residues consist of three glutamines and one arginine (3Q:1R). One glutamine is contributed by the SNARE core domain of Sso1p, the second is from helix A of Sec9p, and the third is contributed by helix B of Sec9p. The single arginine in the ionic layer is provided by Snc1/2p. These residues have a defined role in SNARE complex formation and are thought to set the “register” of the SNARE domain coiled-coil. A previous study has shown that a single mutation of Q468R renders Sec9p nonfunctional in vivo and reduces SNARE complex assembly in vitro (). We reasoned that the Q468R mutation in the context of the tandem t-SNARE–ΔNRD would also produce a nonfunctional t-SNARE complex by this very specific change. shows that the Q468R tandem t-SNARE–ΔNRD is unable to provide Sso1p function. This experiment demonstrates that a single residue change in Sec9c, known to disrupt Sec9p function, abolishes the function of the attached Sso1p, strongly supporting our assertion that the Sec9c–Sso1p–ΔNRD chimera is functioning as an intramolecular t-SNARE complex. Our functional analysis of the tandem t-SNARE chimeras also included examining Sec9p function. The tandem t-SNARE, under transcriptional control of the promoter, was transformed into a wild-type yeast strain as well as several temperature-sensitive mutant alleles. These strains were spotted onto synthetic media containing glucose and grown at permissive temperature (25°C) or restrictive temperature (37°C). shows that the tandem t-SNARE on both low- (CEN) and high- (2 μm) copy plasmids permits efficient growth of , , and , confirming that tandem t-SNAREs also provide Sec9p function, likely by intragenic complementation. Interestingly, the high- and low-copy vectors have differential effects with the various alleles. Higher expression of the tandem t-SNARE more efficiently complements and , whereas the low-copy vector is more effective with and possibly (). The tandem t-SNARE also minimally complements the allele under both conditions (). However, the tandem t-SNARE does not function as the only source of , as it was unable to complement a -null allele (not depicted). The expression level of the tandem t-SNAREs and the expression relative to total endogenous Sso1p and Sso2p as well as endogenous Sec9p is measured in . The centromeric tandem t-SNARE is ∼5% of endogenous Sso1/2p, whereas the 2-μm tandem t-SNARE is ∼13% (, left). The amount of tandem t-SNARE from the CEN vector is very similar to endogenous levels of Sec9p (, right). alleles at the nonpermissive temperature is not likely a result of suppression by overexpression of the Sso1p component of the tandem t-SNARE because low levels of the protein are expressed compared with endogenous Sso1p. However, some suppression of and is observed when Sso1p alone is expressed from its endogenous promoter on a centromeric plasmid. Conversely, no growth at 37°C is seen in the and strains when Sso1p alone was expressed from its own promoter on a centromeric plasmid () or overexpressed with the alcohol dehydrogenase promoter on a 2-μm vector (not depicted). alleles with the tandem t-SNARE lacking the NRD, most likely because of significantly lower expression of these proteins and the added complication of all tandem t-SNAREs behaving as dominant-negative mutants when wild-type Sso1/2p is present ( and not depicted). An intramolecular t-SNARE complex also affords us the opportunity to examine other late-acting sec mutants thought to play a role in regulated SNARE assembly. We transformed four mutants with the tandem t-SNARE in high ( ) and low copy (). We examined the SNARE regulator Sec1p, the Rab-GTPase family member Sec4p, the exocyst component Sec6p, and the general SNARE chaperone Sec18p. The tandem t-SNARE was unable to suppress mutations in these genes, suggesting that they function in roles other than, or in addition to, t-SNARE complex formation. In this study, we show that Sso1p lacking the NRD is fully functional for fusion in vitro () but cannot provide Sso1p function in vivo (, , and ), suggesting that the NRD is important for regulating SNARE complex formation. Replacement of the Sso1p NRD by the homologous sequence from the neuronal plasma membrane syntaxin, syntaxin1A, did not restore function (). However, we show that plasma membrane syntaxin function can be provided by a chimeric protein composed of Sso1p covalently linked to the t-SNARE light chains provided by Sec9c (). We also demonstrate that yeast can dispense with the NRD of Sso1p in the context of a tandem t-SNARE, illustrating that this otherwise essential portion of Sso1p is no longer required if the other t-SNARE component is physically connected (). Much to our surprise, the tandem t-SNARE could not provide full Sec9p function. alleles to some degree, we could not complement a null with a tandem t-SNARE plasmid ( and not depicted). alleles does indicate that the tandem t-SNARE provides Sec9p function because the various mutations map to different sites of the protein. These observations suggest the intriguing possibility that Sec9p has roles in addition to exocytosis. The nature of the tandem t-SNARE chimera is such that we permanently affix Sec9c to the plasma membrane, preventing it from cycling through the cytoplasm. The mechanism of wild-type Sec9p membrane association is largely unknown, but is unlikely to be entirely through its interaction with Sso1/2p. If Sec9p provides another required function not localized to the plasma membrane, our chimera would not support this role. Recent results with the neuronal isoform of Sec9p, SNAP25, suggest that an additional function may be on internal membranes because SNAP25 was found to drive membrane fusion when complexed with syntaxin 13 on an endosomal membrane (). Another interpretation of our results is that Sec9c is simply replacing the chaperone function of the Sso1p NRD rather than functioning as an active participant in a t-SNARE complex. In this case, endogenous Sec9p would be required for viability. Although this formally remains a possibility as a result of our inability to cover a Sec9p-null allele, our intragenic complementation of multiple temperature-sensitive SEC9 alleles makes this possibility unlikely. Furthermore, this interpretation does not change our primary conclusion that the NRD cannot be required as a scaffold given that Sso1p without an NRD is still functional. Additionally, careful consideration of the amount of endogenous Sec9p and its localization relative to the tandem t-SNARE further limits the possibility that the physically attached Sec9c is chaperoning access of the Sso1p H3 domain for endogenous Sec9p. In our experiments examining Sso1p function, the tandem t-SNARE is expressed at >10 times the concentration of total endogenous Sec9p, which is present both in the cytoplasm and on membranes. For endogenous Sec9p to function with the truncated Sso1p in the context of the tandem t-SNARE, it must first attach to the plasma membrane if it is present in the soluble pool or, minimally, locate the Sso1p portion of the tandem t-SNARE in the plane of the bilayer. Next, it must be present in a sufficiently high concentration that it can displace, or minimally compete with, a covalently attached version of itself that is essentially at an infinite concentration because it is permanently attached. Such a concentration of Sec9p on the membrane in the vicinity of Sso1p seems unachievable. Additionally, we have shown that a single point mutation in the Sec9c portion of the tandem t-SNARE–ΔNRD eliminates Sso1p function (). In the context of the Q468R mutant, the interaction of the Sec9c portion of the tandem t-SNARE is likely reduced; however, this construct is nonfunctional in a situation where endogenous Sec9p should have greater access to the H3 domain of Sso1p. Finally, the addition of the homologous NRD from syntaxin 1A does not produce a functional chimera, a domain that has a reasonable possibility of providing a chaperone function for the Sso1p H3 domain. We also found that both the full-length and NH-terminally truncated tandem t-SNAREs were dominant interfering when a wild-type copy of Sso1p was present in the same cells. This result suggests that wild-type Sso1p may interact with the tandem t-SNARE. One possible explanation for this observation is that a four-helix complex composed of the core H3 domain of endogenous Sso1p and the three-helix bundle of the tandem t-SNARE form in the same membrane. This dead-ended cis-SNARE complex would likely be toxic and difficult for the SNARE recycling machinery Sec17p and Sec18p to resolve and recycle. Clearly, other interpretations are also possible. Much of what we know about the in vivo function of the Sso1p NRD was derived from work that generated mutants of Sso1p inefficient at forming a “closed” conformation. found that targeted mutations “opened” the Sso1p structure and virtually eliminated the strong kinetic impediment to t-SNARE complex formation with Sec9c in vitro. They also determined that these open mutants could provide Sso1p function in vivo. These findings suggest that although deletion of the NH terminus was fatal, significantly changing the rate of t-SNARE complex formation was tolerated substantially. One interpretation was that additional machinery may use the Sso1p NH terminus as a scaffold for their function during fusion. On the surface, our results are seemingly at odds with this interpretation; however, a closer evaluation of the results with the open mutants suggests that this may not be the case. Although mutations in the noncore domain in Sso1p strongly affect its ability to form the closed conformation, it does not eliminate it. The observed in vitro enhancement of t-SNARE complex formation is large (1,100–1,300×), but not as large as removing the NH terminus entirely (∼3,000×). As noted, there is still a factor of three “inhibition” with the open mutants, and all of these measurements are kinetic effects in vitro. Additionally, Sso1p is in large excess (∼10×) relative to its partner Sec9 (). These results clearly make room for the possibility that the mutant Sso1p can still be closed, although not nearly as well. We would argue that the open mutants can still spend a significant portion of their lifetime in a closed conformation in vivo, allowing for the interpretation that the NRD is still influencing t-SNARE complex formation. Our data strongly suggest that the NRD of Sso1p and, by extension, all plasma membrane syntaxins are involved in facilitating t-SNARE complex formation or preventing inappropriate associations with the H3 portion of Sso1p. However, our data effectively eliminate a scaffold function of the NRD to recruit other required components to the site of SNARE action. All lipids were purchased from Avanti Polar Lipids, Inc.; detergents were obtained from Calbiochem (n-octyl β-glucopyranoside) and Fisher Scientific (Triton X-100); 5-FOA was purchased from Zymo Research; and G418 sulfate was obtained from Research Product International Corp. Bacterial and yeast media components, including yeast peptone dextrose (YPD), synthetic complete media, raffinose, and amino acid and nucleotide supplements, were obtained from Qbiogene; yeast nitrogen base was purchased from Difco; bacto agar was obtained from BD Biosciences; and the carbon sources glucose and galactose were purchased from Fisher Scientific. Restriction endonucleases were purchased from New England Biolabs, Inc. polymerase was obtained from Roche, and oligonucleotides were purchased from Integrated DNA Technologies. The monoclonal anti-HA 16B12 antibody was purchased from Covance, and secondary antibodies were purchased from the following companies: goat anti–mouse IgG HRP from Rockland Immunochemicals; goat anti–rabbit IgG Fc HRP from Pierce Chemical Co.; and AlexaFluor488 goat anti–mouse IgG from Invitrogen. All strains made for this study are in the W303 background. JMY120 ( ) is a haploid strain with the loci disrupted with the kanamycin resistance gene. The genetic disruption was produced by creating a kanMX2 fragment of 1,450 bp from pFA6 () using oligonucleotides 36 and 37 ( ). The PCR-amplified product was transformed into W3031A, plated onto YPD plates, and grown overnight at 30°C. The samples were then replica plated onto YPD plates with 200 μg/ml G-418 sulfate. Confirmation of deletions was performed by PCR analysis. H404 (obtained from H. Ronne, Swedish University of Agricultural Sciences, Uppsala, Sweden; ) is a haploid strain with the loci disrupted with the gene. JMY123 is the diploid from the cross of JMY120 with H404. JMY128 was made by sporulating JMY123 transformed with pJM198 followed by tetrad dissection to obtain the double-deletion plasmid shuffle strain. JMY298 was made by sporulating JMY123 transformed with pJM311 followed by tetrad dissection to obtain the double-deletion strain with the tandem t-SNARE–ΔNRD. All plasmids were propagated in the strain DH5α, and standard DNA manipulation techniques were used. All PCR procedures were performed with polymerase. All other DNA modifying enzymes were obtained from New England Biolabs, Inc. pJM198 ( ) is a yeast expression vector coding for Sso1p under the transcriptional control of its own promoter. A 1,615-bp PCR fragment containing the SSO1 promoter, ORF, and terminator was generated from genomic DNA and the oligonucleotides 42 and 43. This fragment was cut with EcoRI and XbaI and ligated into pRS316 () cut with the same enzymes. pJM222 is a yeast expression vector coding for the Sso1p-ΔNRD under the transcriptional control of the promoter. A 356-bp fragment coding for amino acids 179–290 was generated by PCR using pJM88 () as template DNA and the oligonucleotides 41 and 44. This fragment was cut with EcoRI and MluI and ligated into pYX223 (Invitrogen) cut with the same enzymes. pJM290 was previously described (). pJM291 is a yeast expression vector coding for the tandem t-SNARE under the transcriptional control of the promoter. It was generated by PCR with oligonucleotides 113 and 179 using pJM89 as a template. The 794-bp PCR product was cut with EcoRI and BamHI, and the 794-bp fragment was ligated into pJM290 cut with the same enzymes. pJM293 is a yeast expression vector coding for the tandem t-SNARE with IHR under the transcriptional control of the promoter. It was generated with PCR with oligonucleotides 180 and 181 and pJM89 as a template. The 294-bp fragment of the Sec9 IHR was digested with ClaI and BspEI and ligated into pJM291 cut with the same enzymes. pJM311 is a yeast expression vector coding for the tandem t-SNARE–ΔNRD under transcriptional control of the promoter. It was generated with PCR using the oligonucleotides 211 and 41 with pJM88 as a template. The 358-bp PCR product was digested with BspEI and MluI and ligated into pJM293 cut with the same enzymes. pJM334 is a yeast expression vector coding for the tandem t-SNARE with IHR under the transcriptional control of the SSO1 promoter. The ∼1,977-bp fragment was generated by digesting pJM293 with EcoRI and XhoI. The ∼496-bp fragment containing the 5′ untranslated region of was PCR amplified from genomic DNA as a template with oligonucleotides 216 and 217. The ∼310-bp fragment containing the 3′ untranslated region of was PCR amplified from genomic DNA as a template with oligonucleotides 218 and 219. All of these fragments were assembled into pRS316, an plasmid (). pJM414 is a yeast expression vector coding for the rat syntaxin1A NRD-appended Sso1p-ΔNRD under transcriptional control of the promoter. A 565-bp fragment coding for amino acids 1–178 of rat syntaxin1A was generated by PCR using pTW20 () as template DNA and oligonucleotides 301 and 302. This fragment was cut with EcoRI and ClaI and ligated into pJM311 cut with the same enzymes. pJM427 is a high-copy yeast expression vector coding for the tandem t-SNARE with IHR under the transcriptional control of the SSO1 promoter. The ∼2,783-bp fragment tandem t-SNARE with promoter and terminator was generated by digesting pJM334 with KpnI and SacI and ligated into pRS426 cut with the same enzymes. pJM429 is a yeast expression vector coding for the tandem t-SNARE–ΔNRD with a Q468R point mutation in the Sec9p helix A and is under the transcriptional control of the promoter. PCR sewing was used to make the specific point mutation. The first round of PCR used oligonucleotides 113 and 326 with pJM311 as the template to generate a 222-bp fragment and used oligonucleotides 41 and 325 with pJM311 as the template to generate a 1,177-bp fragment. The second round of PCR used oligonucleotides 113 and 41, with the two PCR products from the first round as template DNA to generate a ∼1,380-bp fragment. This fragment was cut with EcoRI and BspEI and ligated into pJM311 cut with the same enzymes. pJM367 (GST-Sso1p-ΔNRD) was a gift from Y.-K. Shin (Iowa State University, Aimes, IO) and has been previously described (). W3031A transformed with the indicated plasmids were grown at 30°C in synthetic complete media and analyzed by indirect immunofluorescence microscopy by standard methods () with an anti- HA mAb (16B12; 1:1,000; Covance) followed by fluorescent secondary antibodies (AlexaFluor488 goat anti–mouse IgG (Invitrogen) at 1:1,000. The cells were mounted on slides using the ProLong antifade kit (Invitrogen). Fluorescent images were taken and analyzed with a microscope (Axioplan 2; Carl Zeiss MicroImaging, Inc.) and a plan-Neofluar 100× NA 1.3 oil immersion objective (Carl Zeiss MicroImaging, Inc.) using filter sets for fluorescein (FITC, excitation 480 nm, emission 535 nm, and dichroic Q505LP; Chroma Technology Corp.). Images were captured at room temperature using a digital camera (CoolSNAP HQ; Roper Scientific) and MetaMorph Imaging software (version 6.1; Universal Imaging Corp.). The images were deconvolved using the no-neighbors algorithm and digitally magnified before assembly with Adobe Photoshop version 8.0. Several different isolates of the strain were examined to confirm the reproducibility of the observed localization of the indicated Sso1p. The transformed JMY128 shuffle strain was grown in synthetic complete media minus histidine and uracil with 2% galactose at 30°C for 2 d followed by a subsequent back dilution and another overnight growth. 10-OD cells were spun down and resuspended in 1 ml sterile water. A threefold serial dilution with 20-μl spots were plated onto synthetic complete minus histidine with 2% galactose and 1 g/liter 5-FOA plates and incubated for 3 d at 30°C. Total cell extracts were made by glass bead lysis of TCA-killed cells. The amount of total cell extract indicated in the figure legends was resolved by SDS-PAGE and probed with anti-HA or anti-Ssop antibodies. Primary antibodies were at 1:1,000 (anti-HA and anti-Ssop) or 1:2,000 (anti-Sec9p) dilutions. The HRP-conjugated secondary antibodies were at a 1:10,000 dilution. Immunoblots were developed using ECL detection (GE Healthcare). BY1, BY41 (), BY70 (), BY392 (), BY200 (), BY445 (), BY29 (), BY33 (), BY37 () (obtained from P. Brennwald, University of North Carolina, Chapel Hill, NC), and RSY271 () () were transformed with pJM334, pJM407, and pJM427. , and are unpublished alleles. Clustered charge-to-alanine mutagenesis (; ) was used to generate , which contains two mutations, K434A and K437A, and , which contains E591A and E593A. The third allele, , was made by random mutagenesis. The strains were grown in synthetic complete media minus uracil with 2% glucose at 25°C overnight. 5-OD cells were spun down and resuspended in 1 ml sterile water. Threefold serial dilutions with 20-μl spots were plated onto synthetic complete minus uracil plates with 2% glucose. The plates were grown at 25 or 37°C for ∼2.5 d. GST-Sso1p-ΔNRD (185–290; C266A) was expressed in 12 liters of super broth media and induced with 1 mM IPTG at 30°C for 4 h. Protein was purified by GST affinity chromatography as described previously () except that cells were lysed in 1% Triton X-100, and the protein was eluted with 1% n-octyl β-glucopyranoside. His-Sso1p (pJM88), Snc1p-His (pJM90), and GST-Sec9c (BB442) were expressed and purified as described previously (). All proteoliposomes were formed and used standard fusion assays as previously described (). The polyclonal anti-Sec9c antibodies (RC62 and RC63) were generated by Cocalico Biologicals, Inc. in rabbits immunized with recombinant Sec9c. Initial injections consisted of 200 μg/rabbit followed by 100 μg/rabbit boost injections. Antisera at a 1:2,000 dilution in TBS with 1% Tween-20 (TBS-T) was used for the detection of Sec9p in conjunction with a 1:10,000 dilution in TBS-T of secondary antibody goat anti–rabbit HRP. RC62 antisera was used throughout this study. The polyclonal anti-Ssop was previously described (). Fig. S1 shows the secretion in strains expressing Sso1p or tandem t-SNAREs. Secretion is quantitatively measured by analyzing externalized invertase activity and is qualitatively examined by mating factor secretion with a halo assay. Online supplemental material is available at .

Anthrax toxin, one of the two major virulence factors produced by , is composed of three independent polypeptide chains: the protective antigen (PA), which is involved in target cell binding; the edema factor (EF), a calmodulin-dependent adenylate cyclase; and the lethal factor (LF), a zinc-dependent metalloprotease (for reviews see ; ; ). Only PA is able to bind to target cells; thus, EF and LF must always act in binary combination with PA to be transported to the target cell cytosol, where they exert their activities. The two identified PA receptors, tumor endothelial marker 8 (TEM8) and capillary morphogenesis gene 2 (CMG2), are type I transmembrane proteins sharing ∼60% homology in their extracellular von Willebrand factor A domains and 68% identity in the first 145 residues of their cytoplasmic tails (). A relatively clear view of the mode of action of anthrax toxin has emerged over recent years (for reviews see ; ; ). PA is produced as an 83-kD protein that is unable to interact with EF and LF. At the target cell surface, proteolytic processing of PA83 leads to PA63, which remains receptor bound and can polymerize into a heptameric (PA) ring called the prepore. This prepore is able to bind up to three molecules of EF and/or LF, thus leading to a large hetero-oligomeric complex containing EF–LF–PA and receptors (for reviews see ; ). Once formed, this complex is rapidly internalized via a pathway that depends both on lipid rafts and clathrin (). The complex is then delivered to early endosomes, where it associates with intraluminal vesicles (). At the low pH of endosomes, the prepore undergoes a conformational change that leads to its membrane insertion and pore formation. Interestingly, the pH sensitivity is determined by the receptor, and a lower pH is required when PA is bound to CMG2 when compared with TEM8 (; ). Low pH also triggers partial unfolding of EF and LF, which can translocate across the PA channel (). Because channel formation appears to occur preferentially in the intraluminal vesicles of the multivesicular endosomes, EF and LF end up in the lumen of these vesicles (). Final release of EF and LF from endosomes to the cytoplasm requires back fusion events between intraluminal vesicles and the limiting membrane at the level of late endosomes (). Because EF and LF are unable to bind to cells or cross membranes on their own, binding to PA is an absolute prerequisite for internalization and intoxication. Therefore, PA receptors must remain at the cell surface until heptamerization and binding of the enzymatic subunits have taken place. In fact, PA83 is poorly internalized in comparison with PA63 (for reviews see ; ; ) as a result of a differential localization on the plasma membrane (). Whereas PA83 localizes to the glycerophospholipidic region, PA63 (in particular the heptameric form) is found in specialized cholesterol-rich domains called lipid rafts. This raft association is essential for subsequent internalization of the toxin (). Therefore, although it is receptor mediated, anthrax toxin endocytosis is actually toxin driven. To understand the molecular mechanisms that govern this well-orchestrated behavior of anthrax toxin receptors at the cell surface, we investigated the roles of posttranslational modifications of the receptor cytoplasmic tails. We focused on two modifications: S-palmitoylation and ubiquitination. S-palmitoylation is a reversible lipid modification involving the addition of a saturated 16-carbon palmitate moiety to specific cysteines via a thioester linkage. This modification is used by several hydrophilic proteins such as Ras to associate with membranes but is also found in transmembrane proteins such as the transferrin receptor (), LAT (linker protein for activation of T cells; ), influenza HA (), or the G protein of the vesicular stomatitis virus (). The exact function of this modification in membrane proteins is mostly unclear, but it might modulate the interaction of these proteins with membranes or membrane domains () as well as with other proteins. Ubiquitination is the addition of a ubiquitin (Ub) moiety to cytoplasmic lysines by E3 Ub ligases, the third enzymes in the ubiquitination pathway (). This added Ub itself may or may not be subsequently ubiquitinated on Lys or on Lys leading to polyubiquitin chains (). Although polyubiquitination on Lys is essentially involved in the degradation of proteins by the proteasome, monoubiquitination (the addition of a single Ub moiety to one or multiple lysines) as well as polyubiquitination on Lys have been shown to be involved in endocytosis of plasma membrane proteins (G-coupled receptors, growth factors, and transporters; ) and sorting of internalized receptors into multivesicular bodies (). We report that TEM8 and CMG2 are targets for both palmitoylation and ubiquitination and that these two modifications, via counteracting effects, control receptor endocytosis. TEM8 and CMG2 both exist as four isoforms, two of which, in each case, act as anthrax toxin receptors: isoforms 1 and 2 of TEM8 (TEM8/1 and TEM8/2) and isoforms 1 and 4 of CMG2 (CMG2/1 and CMG2/4; for review see ). These isoforms differ only in their cytoplasmic tails ( ); TEM8/2 has a far shorter cytoplasmic tail than TEM8/1 and CMG2/1, and CMG2/4 differs only in the last 12–13 residues. Even between TEM8 and CMG2, the long tails have a very high degree of conservation (). Therefore, we have focused this study mainly on TEM8/1 but repeated key experiments with the short TEM8/2 isoform as well as with CMG2/4. TEM8/1 was tagged with an HA epitope at the COOH terminus and was either transiently transfected or stably expressed in receptor-deficient CHO cells (CHO; ). Expression levels were overall higher upon transient transfection () but were still readily detectable in the stably TEM8/1-HA–expressing cells (). Most of the expressed receptor was present at the cell surface as indicated by its sensitivity to cell surface trypsinization (). TEM8-HA always migrated as a doublet with an upper smeared band (, m) and a lower well-defined band (, p), the intensity of which varied greatly from experiment to experiment. We investigated whether this migration pattern was caused by glycosylation because the extracellular domain of TEM8 has three predicted -glycosylation sites. Treatment of cells with tunicamycin, an inhibitor of -glycosylation, led to the appearance of a third, lower mobility band (u; unglycosilated) with a concomitant decrease in the intensity of bands m and p (). -glycosidase F treatment of control cell extracts led to the complete disappearance of bands m and p to the benefit of band u (), showing that m and p both correspond to glycosylated TEM8. Band m was endoglycosidase H (Endo H) resistant, and band p was Endo H sensitive (). This shows that TEM8 in band m (mature) had acquired complex Golgi-modified sugars, whereas that in band p (precursor) contained incompletely modified sugars. CMG2/4 was expressed with a COOH-terminal V5 tag (). As for TEM8, CMG2/4-V5 migrated as a smear plus a lower molecular weight band (). Both bands were sensitive to -glycosidase F treatment, and, as for TEM8, only the lower band was sensitive to Endo H (). Altogether, this indicates that CMG2/4 is also glycosylated, as predicted by the presence of two -glycosylation sites in the ectodomain. TEM8/1 and /2 as well as CMG2/1 and /4 all contain cysteine residues in their cytoplasmic tails, which are potential sites for S-palmitoylation (). More specifically, all four proteins have two conserved juxtamembranous cysteines (; Cys-346 and Cys-347 in TEM8), a third cysteine is conserved between TEM8/1 (Cys-481) and CMG2, and a fourth unconserved cysteine is found in TEM8/1 (Cys-521) and CMG2/1 (Cys482). To test whether palmitoylation plays a role in anthrax toxin endocytosis, we analyzed whether the palmitoylation inhibitor 2-bromopalmitate () would affect the raft association of PA63 in BHK cells, a cell line that expresses transmembrane isoforms of both TEM8 and CMG2 (unpublished data). We first verified that PA binding was not affected by the treatment ( ). Raft association was then monitored by following the association with detergent-resistant membranes (DRMs; ). PA63 was associated with DRMs in control cells as previously observed () but shifted to the detergent-soluble fractions in drug-treated cells (). This was not caused by a general disruption of lipid rafts because glycosyl-phosphatidylinositol–anchored proteins, followed here using the glycosyl-phosphatidylinositol–specific bacterial toxin aerolysin () as well as caveolin-1, remained primarily in the DRM fractions. Although caveolin-1 is palmitoylated on three cysteines, the modification is not required for DRM association (). The inhibitory effect of 2-bromopalmitate on PA63-raft association was also accompanied by the inhibition of endocytosis as reflected by a lack of SDS-resistant PA pore and the absence of LF-mediated cleavage of one of the LF substrates, the MAPK kinase MEK1 (; for review see ). To test whether 2-bromopalmitate inhibited the heptamerization process itself, we designed an assay to detect cell surface–formed prepores. These are SDS sensitive and, therefore, are not detected by SDS-PAGE. However, they can be converted to the SDS-resistant phenotype by submitting cell extracts to low pH (pH 4.5) before SDS-PAGE analysis, as illustrated in (middle left) for control cells. The prepore was undetectable for 2-bromopalmitate–treated cells (). The aforementioned experiments show that palmitoylation events are important for anthrax toxin raft association and heptamerization. To determine whether the receptors themselves are palmitoylated, TEM8/1-HA and CMG2/4-V5 were immunoprecipitated from lysates of H-palmitic acid–labeled CHO cells transiently transfected with the respective constructs. Radiolabeled bands with motilities similar to that of TEM8/1-HA ( ) and CMG2/4-V5 (), respectively, were detected. This band was sensitive to in vitro hydroxylamine treatment () or cellular treatment with 2-bromopalmitate () as shown for TEM8/1-HA, indicating that palmitate addition occurred via a thioester bond. The short isoform of TEM8 containing only two cysteines was also palmitoylated (Fig. S1, available at ), as was CMG2/1 (not depicted). To investigate when during the life cycle of TEM8/1 palmitoylation occurred, H-palmitate labeling was performed either in the presence of cycloheximide to inhibit protein synthesis or in the presence of brefeldin A to inhibit transport from the endoplasmic reticulum to the Golgi. Neither drug significantly affected the amount of immunoprecipitated receptor (, anti-HA Western blot). However, both drugs led to an ∼50% reduction in H-palmitate incorporation, as quantified by densitometry, but not to full inhibition (), suggesting that palmitoylation occurs both in the early secretory pathway and later in the life cycle of the protein. A pulse-chase performed after H-palmitate incorporation showed that the palmitate groups were lost within an hour (). This loss was caused by the reversible nature of the modification and not by degradation of the protein because a pulse-chase experiment using [S]cysteine/methionine showed that the half-life of TEM8/1-HA in these experiments well exceeded 5 h (; same S curve as in ). To investigate which of the four TEM8/1 cytoplasmic cysteines can be palmitoylated, we first concentrated on the two juxtamembrane cysteines because TEM8/2, which contains only these two cysteines, is palmitoylated (Fig. S1). Moreover, palmitoylation sites adjacent to the transmembrane region have been previously reported for several proteins such as CD4 () or members of the SNARE family of membrane fusion proteins (). Cysteines at positions 346 and 347 in TEM8/1 were changed to alanine in single (mutants AC and CA) and double (mutant AA) mutants. All three mutants were expressed to lower levels than wild type (WT; ). This was not a result of lower transfection efficiencies because equivalent TEM8/1 expression levels were observed when treating cells with the proteasome inhibitor MG132 (). To compare cells that express similar amounts of receptor and have expression levels that allow the detection of H- palmitate, incorporation was performed on MG132-treated cells. Only when the second cysteine was modified were lower levels of H-palmitate incorporation observed (), indicating that Cys-347 is palmitoylated in the WT protein. The observation that the double mutant still incorporated significant amounts of H-palmitate, however, indicated that other cysteines were modified. Thus, Cys-481 and Cys-521 in TEM8/1 were also changed to alanine in single to quadruple mutants. Once more, all mutants were expressed at lower levels than the WT receptor (), an effect that could similarly be overcome by treating cells with MG132 (). These mutants were expressed at the cell surface as indicated by their ability to bind PA (). The quadruple mutant AAAA and the AAAC mutant did not incorporate H-palmitate (). However, the fact that the CCCA mutant in repeated experiments had a somewhat lower incorporation than the WT TEM8/1 suggests that Cys-521 can be palmitoylated, possibly in a subpopulation of receptors, but only when other palmitoylation sites are present. The AACA mutant was always significantly modified, as were all of the mutants with a cysteine at position 481. Thus, repeated palmitoylation experiments showed that mutations of Cys-347 and Cys-481 always led to a drastic decrease in H-palmitate incorporation, whereas the mutation of Cys-521 had a milder but significant effect. It has previously been observed that the mutation of palmitoylation sites leads to the aberrant palmitoylation of remaining cysteines, especially in double-cysteine motifs (; ; ). Therefore, we focused further experiments on the quadruple AAAA mutant in comparison with the WT (CCCC) receptor. The lower expression levels of all palmitoylation mutants when compared with WT suggested that palmitoylation affects the half-life of the receptor. Therefore, pulse-chase experiments using [S]cysteine/methionine labeling were performed in transiently transfected CHO cells. The initial level of synthesis was very similar for the WT and AAAA mutant receptors ( ). However, 50% of the AAAA mutant was lost after ∼130 min (), whereas >60% of the WT receptors were still present after 5 h. The 30% loss in AAAA TEM8/1 during the first hour of chase () suggests that palmitoylation might somewhat affect folding/trafficking through the early secretory pathway. Because this cannot account for the drastic reduction in receptor half-life, we investigated whether defective palmitoylation could cause premature targeting of TEM8/1 to lysosomes. We first generated stable cell lines expressing AAAA TEM8/1 and found, as expected, lower steady-state expression levels (). Cells were then fed with an inhibitor of lysosomal enzymes, leupeptin (inhibitor of serine and cysteine proteases). This treatment led to some protection of the WT receptor (at the low exposure shown in , CCCC TEM8/1 was only detected in leupeptin-treated cells). However, the effect was far more pronounced for AAAA TEM8/1 (). Intracellular accumulation was confirmed by immunofluorescence (). Although in the absence of treatment, AAAA TEM8/ 1-HA was undetectable by fluorescence microscopy, leupeptin feeding led to the appearance of punctate perinuclear structures (), which were presumably late endosomes/lysosomes. This was in contrast with the effect of MG132, which also led to a massive increase in staining but primarily at the plasma membrane (). Thus, palmitoylation of TEM8/1 appears to be crucial in preventing premature targeting to lysosomes. We next investigated whether the palmitoylation of TEM8/1 is involved in regulating its association with DRMs (). Stably expressed WT TEM8/1 was found in detergent-sensitive fractions ( , left) as previously described (). In contrast, AAAA TEM8/1 was almost entirely associated with DRMs (, right). These observations indicate that palmitoylation acts as a negative regulator of TEM8/1 DRM association. The drastic difference between the AAAA mutant and the WT receptor suggests that the bulk of the WT receptor is palmitoylated at steady state. These observations also raise the interesting possibility, which is not addressed in this study, that depalmitoylation by a protein-thioesterase activity could be a mechanism for the regulation of raft association. Regulation of protein localization by palmitoylation/depalmitoylation has been proposed for soluble proteins in particular (; ). The observation that palmitoylation-deficient TEM8/1 (AAAA) is exclusively found in DRMs () is in apparent contradiction with the fact that 2-bromopalmitate inhibited DRM association of PA63 (). Therefore, we tested whether the drug would also affect the DRM association of AAAA TEM8/1. The mature Endo H–resistant form of the mutant receptor entirely relocalized to the detergent-soluble fractions (), indicating the involvement of a palmitoylation event in the association of AAAA TEM8/1 with DRMs, the substrate of which must be a protein other than the receptor itself. We have previously shown that heptamerization of PA not only triggers the redistribution of the toxin–receptor complex to lipid rafts but also triggers its rapid endocytosis (). We investigated whether this toxin-induced uptake could be caused by a second posttranslational modification of the receptor cytoplasmic tail that would be raft dependent. We focused our attention on ubiquitination (). CHO cells transiently transfected with TEM8/1-HA or CMG2/4-V5 were treated with PA for various times. After immunoprecipitation of the receptors, Western blots were performed with anti-tag and anti-Ub antibodies. The addition of the toxin clearly led to the appearance of a smeared Ub-positive band both for TEM8/1 ( ) and for CMG2/4 (). The absence of ladder was suggestive of monoubiquitination rather than polyubiquitination with long chains, as observed for Ub ubiquitination and proteasomal degradation (). Experiments performed with the shorter TEM8/2 isoform similarly led to the detection of a ubiquitinated band (Fig. S2, available at ). The ubiquitinated band detected for TEM8/2 was smaller than that detected for TEM8/1, suggesting that the receptor itself was the modified protein rather than an interacting partner. To confirm this, lysine mutagenesis was performed. Of the 16 lysines in TEM8/1 and 14 in CMG2/4, we first changed Lys-352 to arginine in TEM8 and the corresponding Lys-350 in CMG2/4 because it is the only lysine common to TEM8/1 and TEM8/2; it is conserved between TEM8 and CMG2; and suggested that the palmitoylation of Tlg1 prevents access of juxtamembranous lysines by E3 ligases. Although the ubiquitinated band could still be detected after the addition of PA to K352R TEM8/1 expression cells, ubiquitination was greatly diminished (). A stronger effect was obtained when mutating 6 of the 16 lysines to arginine (KR mutant in which all lysines labeled with an asterisk in were changed to arginine, including Lys-352). These experiments confirm that TEM8/1 is the substrate of the ubiquitination reaction and that Lys-352 is one of the modified sites but that additional lysines might be modified. The effect of mutating the first conserved juxtamembranous lysine to arginine was even more drastic in CMG2/4. No Ub-positive band could be detected in PA-treated K350R CMG2/4–transfected cells (). To investigate whether ubiquitination of the receptor is important for endocytosis of the anthrax toxin, we monitored the appearance of the SDS-resistant PA pore in lysine mutant-expressing cells. As shown in , the appearance of the SDS-resistant pore was either delayed or strongly diminished in the TEM8/1 and CMG2/4 lysine mutant–expressing cells, demonstrating that ubiquitination of the receptor is important for efficient endocytosis. Because endocytosis of TEM8 requires both raft association () and ubiquitination (), we investigated whether these two events were linked. DRMs were isolated from transiently transfected CHO cells. The bulk of TEM8/1 was found in detergent-soluble fractions ( , left), as also observed in . In marked contrast, ubiquinated TEM8/1 was detected exclusively in DRMs (, left). When cells were treated with the toxin before Triton X-100 solubilization, PA63 was associated with DRMs (as in ) and led to the recruitment of toxin-bound TEM8/1 to this fraction (, right). Concomitantly, the ubiquitinated form of TEM8/1 was increased in the same fraction (). Raft impairment by cholesterol extraction using the sequestering agent β-methylcyclodextrin (ßMCD) led to a strong inhibition of PA-induced TEM8/1 ubiquitination (). This observation suggests that microdomain association precedes and is required for this posttranslational modification. Cbl is an E3 Ub ligase that can interact with lipid rafts (; ). To test for the involvement of Cbl in anthrax toxin endocytosis, we decided to perform RNA silencing. HeLa cells were used because of their human origin (the sequence of hamster Cbl is not available) and their high transfection efficiencies. These cells express TEM8 as indicated by the pH sensitivity of PA channel formation (). As shown in , Cbl could be efficiently silenced by this method. The absence of Cbl did not affect binding of the toxin as indicated by the unaltered presence of PA83/PA63. However, appearance of the PA pore was drastically inhibited (). To investigate the effect of Cbl RNA interference on the ubiquitination of TEM8 itself, RNA interference–treated cells were transfected with TEM8/1-HA, and toxin-induced ubiquitination after immunoprecipitation of the receptors was measured. As shown , the ubiquitinated form of TEM8/1 could not longer be detected. Theses experiments show that Cbl is responsible for the ubiquitination of TEM8/1 and its subsequent internalization. We found that AAAA TEM8/1 is constitutively associated with DRMs and that TEM8 ubiquitination is a raft-dependent modification. Therefore, we wondered whether AAAA TEM8/1 would be constitutively ubiquitinated. , the steady-state ubiquitination level of AAAA TEM8/1 was markedly higher than that of WT CCCC TEM8/1 (especially when comparing the levels of Ub vs. HA) both in stable cell lines and upon transient transfection (note that after immunoprecipitation of TEM8/1-HA from transiently transfected cells, the levels of expressed receptors seem to be similar even though analysis of total extracts shows a lower abundance of the AAAA mutant). Ubiquitination of AAAA TEM8/1 was dependent on the integrity of lipid rafts because the removal of cholesterol using ßMCD led to a drastic reduction in the level of AAAA TEM8/1 ubiquitination. This later observation also indicates that increased ubiquitination of AAAA TEM8/1 is not a consequence of misfolding of the cytoplasmic tail as a consequence of mutagenesis, because such an event would have been insensitive to acute cholesterol extraction from the plasma membrane. Constitutive ubiquitination of the AAAA TEM8/1 and, thus, its constitutive endocytosis are likely to affect its ability to act as an anthrax toxin receptor. To address this issue directly, we monitored the cleavage kinetics of the LF target MEK1. Whereas MEK1 underwent LF-dependent cleavage in WT TEM8/1–expressing cells, the MAPK kinase remained intact in the AAAA TEM8/1–expressing cells during the time course of the experiment (). To test whether this lack of cleavage was only a result of the reduced number of surface-expressed receptors (levels of TEM-HA), we treated AAAA TEM8/1–expressing cells with a higher concentration of PA to reach similar amounts of bound PA as on WT TEM8/1–expressing cells (Fig. S3, available at ). Interestingly, even under these conditions, MEK1 cleavage in AAAA TEM8/1–expressing cells was minimal (Fig. S3), indicating that reduced PA binding does fully account for the reduced MEK1 cleavage in these cells. Because AAAA TEM8/1 undergoes significant constitutive endocytosis, we tested whether it would mediate the internalization of PA83, which is an event that does not occur with the WT receptor and for which PA heptamerization is required (; ). In this study, we made use of a mutant PA (PA; ) that is modified in the furin consensus cleavage site and, thus, remains in the PA83 form. As expected (), PA was not internalized by WT receptors and was sensitive to surface trypsinization (). In contrast, all cell-bound PA became trypsin resistant in AAAA TEM8/ 1–expressing cells, indicating that it had been completely endocytosed (, right). #text Anthrax toxin subunits (; ) and aerolysin () were purified as described previously. PA63 was generated by trypsin cleavage of PA83 (). Antibodies against PA () and aerolysin () were polyclonals developed in our laboratories. Proteins and antibodies were obtained from the following companies: Anti NH-terminal MEK1 antibodies from Upstate Biotechnology; anti–caveolin-1 from Transduction Laboratories; anti-Ub (sc-8017) as well as antibodies and siRNA against human Cbl from Santa Cruz Biotechnology, Inc.; anti-HA coupled to beads, labeled, or unlabeled with HRP and anti-GFP from Roche; anti-V5 labeled or unlabeled with HRP from Invitrogen; HRP secondary antibodies from Pierce Chemical Co.; and FITC-conjugated secondary antibodies from Invitrogen. BHK, HeLa, and anthrax toxin receptor–deficient CHO (here designated as CHO) cells were grown as described previously (, ; ). The human CMG2 (isoform 4) gene tagged with a V5 epitope and cloned in the pcDNA3.1/ V5-HIS-TOPO expression vector was provided by J. Martignetti (Mount Sinai School of Medicine, New York, NY; ). TEM8 isoforms 1 and 2 tagged with GFP and cloned in the pHS001-EGFP expression vector was provided by J. Young (Salk Institute, San Diego, CA; ). Isoform 1 of human TEM8 gene tagged with an HA epitope was in the pIREShyg2 vector (). Cysteine to alanine mutant constructs were generated using the QuikChange Site-Directed Mutagenesis Kit (Stratagene) and were transfected into CHO (1 μg cDNA/9.6-cm plate) using Fugene (Roche). Stable lines were selected by two rounds of hygromycin resistance. Colonies were isolated by limited dilution. To silence Cbl, HeLa cells were transfected with 200 pmol/9.2-cm2 dish of siRNA using OligofectAMINE (Invitrogen) transfection reagent. Confluent cells were incubated in incubation medium (IM; Glasgow minimal essential medium buffered with 10 mM Hepes, pH 7.4) at 4°C for 1 h with various combinations of proaerolysin, PA, and LF, washed, and placed at 37°C in IM for different times. Cells were lysed by incubation for 30 min at 4°C with radioimmunoprecipitation buffer (1% NP-40, 50 mM Tris-HCl, pH 7.4, 0.25% sodium deoxycholate, 150 mm NaCl, 1 mM EDTA, and a cocktail of protease inhibitors; Roche). Protein concentrations of extracts were determined with bicinchoninic acid (Pierce Chemical Co.). To convert surface PA to an SDS-resistant form, cell extracts were incubated at room temperature for 10 min with 145 mM NaCl and 20 mM MES-Tris, pH 4.5. SDS-PAGE was performed using 4–20% gels (NOVEX) under nonreducing conditions. Gels were transferred onto nitrocellulose as described previously (). To inhibit -glycosylation, cells were treated with 10 μg/ml tunicamycin (Sigma-Aldrich) during the last 16 h of growth. For -deglycosylation, cell extracts were boiled for 5 min with 1% SDS and 1% β-mercaptoethanol, diluted fivefold in 40 mM phosphate buffer, pH 7.0, containing 10 mM EDTA, 1% Triton X-100, 2.5 mM PMSF, and 1% β-mercaptoethanol, and were incubated for 16 h at 37°C with 10 U/ml -glycosidase F. Endo H treatment was performed according to the manufacturer's instructions (New England Biolabs, Inc.). Palmitoylation was inhibited by pretreating cells with 100 μM 2-bromopalmitate (Sigma-Aldrich) for 1 h at 37°C . Chemical removal of S-palmitoylation was performed by treating cell extracts for 1 h at room temperature with 1 M hydroxylamine hydrochloride, pH 7.2. Protein synthesis was blocked by a 30-min treatment with 10 μg/ml cycloheximide at 37°C. Endoplasmic reticulum–to-Golgi transport was blocked by 20 μg/ml brefeldin A pretreatment for 30 min at 37°C. The proteasome inhibitor MG132 (Sigma-Aldrich) was used at 10 μM during the 16 h in culture medium. To block lysosomal enzymes, cells were fed for 16 h with 250 μg/ml leupeptin. To extract cholesterol, cells were treated with 10 mM βMCD (Sigma-Aldrich) in IM for 30 min at 37°C, leading to a 59.3 ± 3.8% decrease in total cholesterol as measured by thin layer chromatography (). DRMs were prepared using OptiPrep gradients as described previously (). Six fractions were collected from the top, and the total protein content of each fraction was precipitated with trichloroacetic acid (). For immunoprecipitations of TEM8 or CMG2, cells were lysed for 30 min at 4°C in immunoprecipitation buffer (0.5% NP-40, 500 mM Tris-HCl, pH 7.4, 20 mM EDTA, 10 mM NaF, 2 mM benzamidine, 1 mM -ethyl-maleimide, and a cocktail of protease inhibitors), centrifuged for 3 min at 2,000 , and supernatants were incubated for 2 h at 4°C with either HA-coupled agarose beads (Roche) or protein G–coupled beads (GE Healthcare) with 2 μg monoclonal antibody against V5. After washing of the beads, samples were boiled for 5 min under reducing conditions. To follow palmitoylation, TEM8/CMG2-expressing cells were incubated for 2 h at 37°C in IM with 200 μCI /ml H-palmitic acid (9,10-H(N); American Radiolabeled Chemicals, Inc), were washed, and submitted to immunoprecipitation. Beads were incubated for 30 min at 60°C in nonreducing sample buffer before SDS-PAGE. After fixation (25% isopropanol, 65% HO, and 10% acetic acid), gels were incubated for 30 min in enhancer Amplify NAMP100 (GE Healthcare), dried, and exposed to a Hyperfilm Multipurpose (GE Healthcare). For metabolic labeling, CHO cells were transiently transfected for 30 h with TEM8/1-HA cDNAs, washed with methionine/cysteine-free medium, incubated for a 30-min pulse at 37°C with 50 μCi/ml [S]methionine/cysteine (Hartman Analytics), washed, and further incubated for different times at 37°C in complete medium with a 10-fold excess of nonradioactive methionine and cysteine. Receptors were immunoprecipitated and analyzed by SDS-PAGE. CHO cells were fixed with 3% formaldehyde, permeabilized with 0.1% Triton X-100, and labeled with anti-HA monoclonal antibodies followed by fluorescein isothiocyanate–conjugated goat anti–mouse IgG. Images were acquired using a 100× lens on a microscope (Axiophot; Carl Zeiss MicroImaging, Inc.) equipped with a cooled camera (model Orca; Hamamatsu) using the Openlab acquisition software (Improvision). Three supplemental figures are provided. Fig. S1 shows that isoform 2 of TEM8, which has a very short cytoplasmic tail (), is palmitolyted. Fig. S2 shows that this isoform is also ubiquitinated. Finally, Fig. S3 shows than when toxin concentrations are adapted so that CCCC TEM8/1– and AAAA TEM8/1–expressing cells bind similar amounts of PA, MEK1 cleavage is observed in the former cells within an hour but not in the latter cells. Online supplemental material is available at .

Distant metastases are the main cause of death among breast cancer patients (). Clinical and pathological risk factors, such as patient age, tumor size, and steroid receptor status, are commonly used to assess the likelihood of metastasis development. When metastasis is likely, aggressive adjuvant therapy can be prescribed which has led to significant decreases in breast cancer mortality rates (). However, for the majority of patients with intermediate-risk breast cancer, the traditional factors are not strongly predictive (). Accordingly, approximately 70–80% of lymph node-negative patients may undergo adjuvant chemotherapy when it is in fact unnecessary (). Moreover, it is believed that many of the current risk factors are likely to be secondary manifestations rather than primary mechanisms of disease. An ongoing challenge is to identify new prognostic markers that are more directly related to disease and that can more accurately predict the risk of metastasis in individual patients. In the recent years, an increasing number of disease markers have been identified through analysis of genome-wide expression profiles (; ; ; ). Marker sets are selected by scoring each individual gene for how well its expression pattern can discriminate between different classes of disease. In breast cancer, two large-scale expression studies by and each identified a set of ∼70 gene markers that were 60–70% accurate for prediction of metastasis, rivaling the performance of clinical criteria. Strangely, however, these marker sets shared only three genes in common, with the first set of markers predicting metastasis less successfully when scoring patients from the second study, and vice versa (). One possible explanation for the different marker sets is that changes in expression of the relatively few genes governing metastatic potential may be subtle compared to those of the downstream effectors, which may vary considerably from patient to patient (; ; ). Due to these types of difficulties, many groups have hypothesized that a more effective means of marker identification may be to combine gene expression measurements over groups of genes that fall within common pathways. Several approaches have been proposed to score known pathways by the coherency of expression changes among their member genes (, ; ; ; ; ; ). Known pathways are drawn from sources such as the Gene Ontology (GO) () and KEGG () databases. Recently, pathway-based analysis has been extended to perform classification of expression profiles and applied to discriminate irradiated from non-irradiated yeast cells (). However, a remaining hurdle to pathway-based analysis is that the majority of human genes have not yet been assigned to a definitive pathway. The recent availability of large protein networks provides one means to at least partially address these challenges. Using protein–protein interaction networks derived from literature, the yeast two-hybrid system, or mass spectrometry (reviewed by ), a number of approaches have been demonstrated for extracting relevant subnetworks based on coherent expression patterns of their genes (; ) or on conservation of subnetworks across multiple species (). Each subnetwork is suggestive of a distinct functional pathway or complex, yielding many known and novel pathway hypotheses in organisms for which sufficient protein interaction data have been measured. Large protein networks have only recently become available for human (; ; ; ), enabling new opportunities for elucidating pathways involved in major diseases and pathologies (). Here, we pursue a protein-network-based approach for identifying markers of metastasis within gene expression profiles, which can be used to identify genetic alterations and to predict the likelihood of metastasis in unknown samples. The markers in question are not encoded as individual genes or proteins, but as subnetworks of interacting proteins within a larger human protein–protein interaction network. We find that the network-based method has several advantages over previous analyses of differential expression. First, the resulting subnetworks provide models of the molecular mechanisms underlying metastasis. Second, although genes with known breast cancer mutations are typically not detected through analysis of differential expression, such as P53, KRAS, HRAS, HER-2/neu, and PIK3CA, they play a central role in the protein network by interconnecting many expression-responsive genes. Third, the identified subnetworks are significantly more reproducible between different breast cancer cohorts than individual marker genes selected without network information. Finally, network-based classification achieves higher accuracy in prediction, as ascertained by selecting markers from one data set and applying them to a second independent validation data set. We applied a protein-network-based approach to analyze the expression profiles of the two cohorts of breast cancer patients previously reported by and . Both sets of expression profiles had been obtained from primary breast tumors but hybridized to different microarray platforms (Agilent oligonucleotide Hu25K microarrays and Affymetrix HG-U133a GeneChips, respectively). We restricted our analysis to the 8141 genes present in both data sets. For 78 patients in and 106 in , metastasis had been detected during follow-up visits within 5 years of surgery. Profiles for these patients were assigned to the class ‘metastatic,' whereas profiles for the remaining 217 and 180 patients were labeled ‘non-metastatic.' To obtain a corresponding human protein–protein interaction network, we assembled a pooled data set comprising 57 235 interactions among 11 203 proteins, integrated from yeast two-hybrid experiments (; ), predicted interactions via orthology and co-citation (), and curation of the literature (; ; ). To integrate the expression and network data sets, we overlaid the expression values of each gene on its corresponding protein in the network and searched for subnetworks whose activities across the patients were highly discriminative of metastasis. This process involved several scoring and search steps, as illustrated in and described further in Materials and methods. Briefly, a candidate subnetwork was first scored to assess its activity in each patient, defined by averaging its normalized gene expression values. This step yielded 295 and 286 activity scores per subnetwork, corresponding to the number of breast cancer patients in the two data sets, respectively. Second, the discriminative potential of a candidate subnetwork was computed based on the mutual information between its activity score and the metastatic/non-metastatic disease status over all patients. Significantly discriminative subnetworks were identified by comparing their discriminative potentials to those of random networks. A total of 149 and 243 discriminative subnetworks were identified in and data sets (consisting of 618 and 906 genes, respectively, and based on a panel of three separate tests for statistical significance—see Materials and methods). A compendium including all of these subnetworks is available online via the Cell Circuits database () (), which provides each subnetwork in graphical (GIF) and machine-readable (SIF) formats. Each significant subnetwork may be viewed as a putative marker for breast cancer metastasis, which is not based on a single gene but rather on the aggregate behavior of genes connected in a functional network. This feature is a significant departure from conventional expression-alone analysis, which does not provide functional insight into the identified markers. In all, 47.3% () and 65.4% () of the discriminative subnetworks were enriched for proteins functioning in a common biological process as annotated by GO (hypergeometric test with a false discovery rate of 5%). To test whether this functional enrichment might be solely due to network topology, we extracted 1000 random subnetworks of the same size as the identified discriminative subnetworks, but without regard to the expression profiles. In the two sets of random subnetworks, 25.4±0.6 and 26.5±0.1% (mean±s.d.) were enriched for proteins with a common biological process. Our higher rate suggests that integrating protein networks with cancer expression profiles is able to identify proteins coordinately functioning in pathways. Among the discriminative subnetworks, 66 identified from and 153 identified from corresponded to signaling of cell growth and survival, cell proliferation and replication, apoptosis, cell and tissue remodeling, circulation and coagulation, or metabolism (see for some example subnetworks; see CellCircuits database for all functional annotations). Together, these processes contribute to the major events that have been implicated in the progression of cancer (). Many extracellular matrix and inflammatory proteins related to tumor aggression, such as matrix metallopeptidase 9 (MMP9 in ) and interleukins (), were also included in the identified subnetworks. Approximately 88% of the 149 subnetworks identified from had higher activity levels in metastatic breast tumors than in non-metastatic ones, whereas the 243 subnetworks identified from were split roughly equally in their direction of activity change (124 versus 119). Next, we examined the agreement between markers identified from the two breast cancer cohorts using our network-based approach. As shown in , the subnetwork markers were significantly more reproducible between data sets than were individual marker genes selected without network information (12.7 versus 1.3%). In terms of biological function, extracellular signal-regulated kinase 1 (MAPK3) was reproducible as a central node in subnetworks identified from both data sets ( versus 2D). illustrate two other subnetworks that were discriminative in both data sets, although there was less consistency in the expression levels of genes comprising these subnetworks. For instance, PKMYT1 is significantly differentially expressed in but not in (; diamond versus circle), whereas CD44 is significantly differentially expressed in but not in (). However, by aggregating the expression ratios of these genes with their network neighbors, the subnetworks containing these genes are found to be significant in both data sets. One concern is that the increased overlap between subnetwork markers might be expected, given that the number of all possible subnetworks is smaller than the number of gene sets (selected irrespective of the network). However, the observed overlap between subnetworks was also significantly greater than that achieved among 1000 same-size sets of connected subnetworks chosen at random (<0.002). Another question is why, even using subnetworks, the percentage overlap is not larger. One reason may be the difference in clinical design of the two data sets. While all of patients in had lymph node-negative breast cancer, approximately half of the patients in were lymph node-positive and underwent adjuvant therapy before expression profiling. Another explanation may be the difference in microarray platforms or the incompleteness of the protein–protein interaction network, which covered only ∼40% of the gene expression levels measured in either study. We next tested the predictive performance of subnetwork markers during classification of a new expression profile as metastatic or non-metastatic. To use the subnetworks for classification, the expression levels of the genes in each subnetwork were averaged to compute a subnetwork activity score, in the same way the activity score was computed in identifying the subnetwork markers originally (see above). These activity scores were then used as feature values by a classifier based on logistic regression. At a fixed sensitivity of 90%, the subnetwork markers achieved 70.1% () and 72.2% () accuracy, measured as the percentage of correct classifications using the technique of five-fold cross-validation within each data set. This accuracy compares favorably with those reported in the original studies (; ) (62 and 63%; see ). In the above five-fold cross-validation, one-fifth of the samples were designated as ‘test' data and withheld during classifier training (in which the relative weights of each subnetwork feature are determined). However, the subnetwork features themselves were identified using all microarray samples before classification, which introduces possible circularity into the validation procedure. To achieve an unbiased evaluation of subnetwork performance, we further tested the subnetwork markers selected from one cohort of breast cancer patients as predictors of metastasis on the other cohort. This same cross-data set analysis was also run using individual marker genes according to the conventional method (controlled for size by providing the classifier with the set of 618 or 906 top discriminative genes in or , respectively, which is the same number of genes covered by the subnetwork markers; see Materials and methods). Similar to the procedure for the subnetwork markers, five-fold cross-validation was performed on one data set using the genes selected from the other data set. At 90% sensitivity, the subnetwork markers from achieved 48.8% accuracy in classifying samples in , and 55.8% accuracy for the reciprocal test. The single-gene markers achieved 45.3 and 41.5% accuracies, respectively. Although all marker sets have decreased performance in predicting metastasis in an independent data set, the accuracies remain significantly higher than random guesses (31.2 and 39.7%, respectively). To show that the better performance was not dependent on the chosen classification algorithm, we evaluated the markers by support vector machines (SVM) (), which led to the same trends (). To capture performance over the entire range of sensitivity/specificity values, we also analyzed the classifiers using the AUC metric (area under ROC curve). As shown in and , the subnetwork markers significantly outperformed the single-gene markers in both data sets. Subnetwork classification performance was also higher than classifiers built on random subnetworks (=0.046 and 0.012 against 1000 sets of same-sized random subnetworks on and , respectively); strangely, performance of the conventional classifiers was not (=0.124 and 0.174, respectively). Finally, we compared the classification performance of the subnetwork markers with markers based on predefined groups of functionally related genes (). These included 1446 sets of functionally related genes extracted from GO and 522 from the Molecular Signatures Database (MSigDB) (v1.0). Neither of these functionally related groupings performed as well as either the subnetwork markers or individual genes. This finding might indicate that some of the functional groupings relevant to breast cancer metastasis have not yet been curated in the current pathway databases. Beyond achieving better performance, the discriminative subnetworks lend insight into the biological basis for why samples are classified as metastatic or non-metastatic. For instance, a single cell cycle-related subnetwork was identified from , which could be used to predict the metastatic outcome of ∼60% of patients in (). Thioredoxin (TXN), which was not differentially expressed, mediated interconnections among many cell mobility and DNA replication proteins that were differentially expressed in , forming subnetworks that were informative for metastasis in (see for the TXN core motif shared in multiple subnetworks). Conversely, several subnetworks identified from , such as the RAD54L-related proteasome () and a Ras-related subnetwork (RAB1A and RAB11A; ), were predictive for patients in . Unlike conventional expression clustering or classification methods, network-based analyses can implicate proteins with low discriminative potential (e.g., those that are not differentially expressed), if such proteins participate in a subnetwork whose overall activity is discriminative. Such proteins can arise within a significant subnetwork if they are essential for maintaining its integrity, that is they are required to interconnect many higher scoring proteins. This property is important for the discovery of disease-causing genes, because the phenotypic changes most indicative of breast cancer metastasis need not be regulated at the level of expression (). Overall, 85.9 and 96.7% of the significant subnetworks contained at least one protein that was not significantly differentially expressed in metastasis (>0.05 from a two-tailed -test). Many well-established prognostic markers of breast cancer disease outcome, such as HER-2/neu (ERBB2), Myc, and cyclin D1, were not present in gene signatures from conventional expression-alone analysis (), but played a central role in the discriminative subnetworks by interconnecting many expression-responsive genes (see for examples and for all). Other examples are the SMAD family and the phosphoinositide-3-kinase catalytic subunit (PIK3CA) (): changes in SMAD phosphorylation have been linked to breast cancer metastasis (), and somatic mutations in PIK3CA are associated with constitutive upregulation of kinase activity in ∼30% of breast cancers (; ). To evaluate the power of a network-based method to uncover disease genes, we assembled a list of 60 breast cancer susceptibility genes that had been reported as such in previous literature and were also represented in our expression data sets (; ; ) (the complete list is provided in ). We found that 32 out of 149 discriminative subnetworks from and 27 out of 243 from contained at least one known cancer susceptibility gene (seven and five subnetworks, respectively, contained two or more known susceptibility genes). Some notable examples are RAD51 and TP53 shown in ; ESR1 and TP53 in ; ERBB2 in ; BRCA1 in ; ESR1, BRCA1, and CYP1A1 in ; PIK3CA and HRAS in ; GSTT1 in and KRAS and PIK3CA in . We compared these levels of enrichment to a conventional expression-alone analysis, which did not incorporate information on pathway structure. As shown in , subnetworks were significantly enriched with cancer susceptibility genes, in contrast to genes identified by a conventional analysis. Disease genes that can be only detected using network information include TP53, KRAS, HRAS, ERBB2, and PIK3CA. Finally, we also examined the enrichment of the discriminative subnetworks for a recently published list of 122 genes with somatic mutations associated with breast cancer () (71 of these were represented in the expression data sets we examined). Genes in this list were determined by DNA sequencing to have mutations in at least 1 of the 11 breast cancer cell lines, with no cancer cell line having more than six mutant genes in common with any other cancer. A total of 11 mutations mapped to proteins in the discriminative subnetworks (see for examples). Although still higher than the conventional method in (), this enrichment was not significant by either approach (=0.434 for subnetwork markers and 0.914 for single-gene markers). One explanation could be that the cancer cell lines capture a different disease state than that found in the population of patients surveyed by microarray profiling. Only two genes (p53 and BRCA1) reported in the sequencing study were linked with breast cancer in , perhaps because the newly discovered mutations are rare or not genetically transmissible. #text A subnetwork is defined as a gene set that induces a single connected component in the protein–protein interaction network. Given a particular subnetwork M, let represent its vector of activity scores over the tumor samples, and let represent the corresponding vector of class labels (metastatic or non-metastatic). To derive , expression values are normalized to z-transformed scores which for each gene has mean μ=0 and s.d. σ=1 over all samples (). The individual of each member gene in the subnetwork are averaged into a combined -score, which is designated the activity . Many types of statistic, such as the or Wilcoxon score, could be used to score the relationship between and . where and enumerate values of and , respectively, (, ) is the joint probability density function (pdf) of ′ and , and () and () are the marginal pdf's of ′ and . To derive ′ from , activity levels are discretized into ⌊log 2(# of samples) + 1⌋ = 9 equally spaced bins (). A rationale for using MI in cancer classification is to capture potential heterogeneity of expression in cancer patients (), that is, differences not only in the mean but in the variance of expression. For examples of the computation of MI see . The particular gene set maximizing (M) is regarded as optimal for classification. Given the discriminative score function , a greedy search is performed to identify subnetworks within the protein–protein interaction network for which the scores are locally maximal. Candidate subnetworks are seeded with a single protein and iteratively expanded. At each iteration, the search considers addition of a protein from the neighbors of proteins in the current subnetwork and within a specified network distance from the seed. The addition that yields the maximal score increase is adopted; the search stops when no addition increases the score over a specified improvement rate . Given that the median distance between any two proteins in the human protein–protein interaction network is five, we set =2 to provide a sufficient number of neighbors while keeping the search local. The parameter is chosen as 0.05 to avoid over-fitting to the expression data used. The majority of searches terminate due to the constraint on ; increasing the value of has only marginal effect on the results (data not shown). To assess the significance of the identified subnetworks, three tests of significance are performed. For the first test, we perform the same search procedure over 100 random trials in which the expression vectors of individual genes are randomly permuted on the network. Such permutation disrupts the correlation between expression and interaction. The score of each real subnetwork is indexed on the ‘global' null distribution of all random subnetwork scores. The second test indexes each real subnetwork score on a ‘local' null distribution, estimated from the scores of 100 random subnetworks initialized from the same seed protein as the real subnetwork (the distribution is assumed to be gamma-distributed; ). Third, we test whether the mutual information with the disease class is stronger than that obtained with random assignments of classes to patients (). For the random model, these assignments are permuted in 20 000 trials, yielding a null distribution of mutual information scores for each trial; the real score of each subnetwork is indexed on this null distribution. In this study, significant subnetworks are selected that satisfy all three tests with <0.05, <0.05, and <0.00005, according to the three different null distributions of . Logistic regression models () are trained on the subnetwork activity matrix (significant subnetworks versus patient samples) and the original gene expression matrix (i.e., conventional classification). Subnetwork markers or individual gene markers are selected using the whole first data set () and then tested on the second data set (); or vice versa. To measure unbiased classification performance, the patient samples in the second data set are divided into five subsets of equal size: three subsets are used as the training set to build the classifier using markers from the first data set, one subset is used as the validation set and the other subset is used as the test set. The -value of discriminative power to classify training samples () is used to rank markers (subnetworks or genes), after which the logistic regression model is built by adding markers sequentially in increasing order of -value. The number of markers used in the classifier is optimized by evaluating its area under ROC curve (AUC; see for details) on the validation set. The final classification performance is reported as the AUC on the test set using the optimized classifier. Each of the five patient subsets in the second data set is evaluated in turn as the test set, with the other four sets providing training and validation. The averaged AUC values among the five test sets are reported as a final classification performance for each marker set.

It is the premise of systems biology that biological processes are studied as integrated systems consisting of multiple interacting elements and that the basis for the system's properties is the contextual information of the elements interactions. Operationally, biological systems are frequently represented as networks and their properties are studied by iterative cycles of targeted network perturbation followed by quantitative measurement of all the system's elements (). Networks typically studied are transcriptional networks analyzed by gene expression arrays (; ) and CHIP on chip assays (; ), protein interaction networks analyzed by the yeast two-hybrid systems (; ; ) or mass spectrometry of purified protein complexes (; ; ; ) and genetic interactions analyzed by synthetic lethal screens (). Protein phosphorylation, a network of protein kinases and phosphatases and their respective cellular substrates, is a universal regulatory mechanism and plays a pivotal role in the control of most cellular process. Thus, the understanding of protein phosphorylation networks and their dynamic changes is of fundamental importance for systems biology (). Recently, phosphoproteomics has become a robust technique for the analysis of protein phosphorylation networks. Typically, (phospho)protein samples are digested with a protease, and the peptides are analyzed by liquid-chromatography tandem mass spectrometry (LC-MS/MS) (). As after the digestion of a proteome phosphopeptides are present at a low concentration, it is necessary to specifically enrich them before analysis (; ). Recently, several phosphopeptide enrichment methods have been described and their performance has been compared (). They include affinity chromatography and phosphoramidate chemistry-based purification. The most commonly used affinity-based methods are immobilized metal affinity chromatography (IMAC) () and titanium dioxide (TiO) (; ). As an alternative phosphoramidate chemistry (PAC), in which the phosphopeptides are covalently captured on an amino-modified solid phase (e.g. a dendrimer () or glass beads (; )) and are released by acid hydrolysis of the phosphoramidate bond (; ; , ) can be used. Using the technologies described above, several large scale data sets on protein phosphorylation have recently been published (; ; ; ). However, a number of factors limit the usefulness of these data for systems biology research. First, the data sets are far from being complete. Second, false-positive and false-negative error rates are frequently unknown and spectra may not be accessible to independently assess the quality of peptide identification and assigned site of phosphorylation. Third, the data are mostly presented as lists of identified phosphopeptides, limiting their use for further experimentation or meta-analysis. In this report, we describe PhosphoPep, a database for phosphopeptides and phosphoproteins from Kc167 cells and a suite of associated software tools as a resource for systems biology research in The small genome size, short generation time, the highly developed genetic tools that can be easily combined with biochemical analysis () and the high degree of conservation of signaling pathways between the fly and humans () make an ideal, but as yet largely unexplored species for systems biology. PhosphoPep contains over 10 000 high-confidence phosphorylation sites from 3472 gene models and 4583 distinct phosphoproteins, and therefore, is the as yet most completely mapped phosphoproteome of any single source. To support further experimentation and analysis of the phosphorylation data, we added to the PhosphoPep database a number of software tools. First, we implemented a search function to detect the sites of phosphorylation on individual proteins and to place phosphoproteins within cellular pathways as defined by the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway database (). Such pathways, along with the identified phosphoproteins can be interrogated by a pathway viewer and exported to Cytoscape (), a software tool, which supports the integration of the data from PhosphoPep and other databases. Second, we added utilities for the use of the phosphopeptide data for targeted proteomics experiments. In a typical experiment of this type, the known phosphorylation sites of a protein or set of proteins are detected and quantified in extracts representing different cellular conditions via targeted mass spectrometry experiments such as MRM (; ; ; ; ). Third, we made the data in PhosphoPep searchable by spectral matching through SpectraST (). Specifically, for each distinct phosphopeptide ion identified in this study, all corresponding MS2 spectra were collapsed into a single consensus spectrum. Unknown query spectra can then be identified by spectral searching against the library of phosphopeptide consensus spectra. Collectively, PhosphoPep and the associated software tools and data mining utilities support the use of the data for diverse types of studies, from the analysis of the state of phosphorylation of a single protein to the detection of quantitative changes in the state of phosphorylation of whole signaling pathways at different cellular states and has been designed to enable the iterative cycles of experimentation and analysis that are typical for systems biology research. To generate an extensive phosphopeptide map of KC 167 cells, we first performed a large-scale phosphorylation site mapping project as described in the and . Briefly, as the phosphoproteome strongly depends on the cellular state, we performed tryptic digestion of protein extracts from Kc167 cells grown under various conditions: nutrient-rich medium; nutrient-depleted medium; medium supplemented with insulin (a growth inducer); medium supplemented with rapamycin (a growth inhibitor); and medium containing Calyculin A, an inhibitor of protein phosphatase 1 and protein phosphatase 2A. The combined peptide sample was separated by peptide isoelectric focusing (IEF) in a free-flow electrophoresis (FFE) instrument (). From each fraction phosphopeptides were isolated using three different phosphopeptide isolation methods (IMAC, TiO2 and PAC) to maximize coverage of the phosphoproteome (). Each phosphopeptide fraction was then subjected to LC-MS/MS using a high mass accuracy tandem mass spectrometer. The generated LC-MS/MS data were searched against a protein (decoy) database and the identified phosphorylation sites were validated using the PeptideProphet software tool () or the target-decoy search strategy (). The resulting combined data set consisting of 10 118 high-confidence phosphorylation sites from 3472 gene models and 4583 distinct phosphoproteins was incorporated into the PhosphoPep database. The fragment ion spectra obtained in this study were assigned to (phospho)peptide sequences using the sequence database search tool Sequest () and were investigated for two forms of errors in the data set: first, the miss-assignment of the fragment ion spectrum to a peptide sequence (; ) and second, the miss-assignment of the phospho-amino acid in an otherwise correctly identified phosphopeptide (). When assessing the first type of error using the statistical tool PeptideProphet () or a decoy database (DD) (), we found that at a PeptideProphet probability score cut off value of 0.8 approximately 2.6% (1.8% DD), at a cut off of 0.9 1.5% (0.8% DD) and at a cut off of 0.99 approximately 0.2% (0% DD) of all identifications were false-positive assignments. Based on these results, we decided to upload all phosphopeptides with a PeptideProphet probability score greater than 0.8 into PhosphoPep. To assess the second type of error, the miss-assignment of the phospho-amino acid in a correctly identified phosphopeptide we used the dCn score computed by Sequest () as described in the and . We found that a dCn value greater than 0.1 corresponds to >90% certainty in phosphorylation site assignment. Overall, the application of a dCn threshold of 0.1 yielded 10 118 distinct phosphorylation sites (PeptideProphet probability score >0.9) or 12 756 phosphorylation sites (PeptideProphet probability score >0.8). Without any dCn filter PhosphoPep contains 12 596 (PeptideProphet probability score >0.9) or 16 608 phosphorylation sites (PeptideProphet probability score >0.8). We next analyzed the structural and functional properties, namely the distribution and number of phosphorylated residues per phosphopeptide, the molecular functions, and the biological processes and the pathways that are associated with the identified phosphoproteins along with their predicted abundance. To increase the utility of the phosphopeptide data set described above, we organized the data in a publicly accessible relational database, PhosphoPep, and added functions supporting data mining and meta-analysis. The following sections describe the database and the added functions. There is no ‘gold standard' phosphoproteome data set that could be used to assess the extent to which the Kc167 phosphoproteome has been mapped out. To further investigate the achieved phosphoproteome coverage, we compared the phosphorylation sites from our data set that matched the highly conserved (; ) and clinically relevant insulin/TOR pathway with the already known sites in . The results are shown in . Of the 15 pathway members, 6 (dAKT1, CHICO, dFOXO, dTSC2, dS6K and d4E-BP) have been known to be phosphorylated in . In our data set, we found all 15 members to be phosphorylated. Furthermore, for the proteins for which phosphorylation sites have been published previously, we were able to identify multiple new sites. The most prominent example is the insulin receptor substrate, CHICO, for which the number of known phosphorylation sites increased from 2 to 20. For dFOXO and d4E-BP, we identified all, and for dS6K, we identified one already known phosphorylation sites. For dAKT1, CHICO and dTSC2, the already known sites were not found in our experiments, indicating that in spite of the high number of sites identified in this study the KC167 phosphoproteome is likely not complete at this time (see ). This example shows that we have reached a depth in phosphoproteome coverage that is suitable for systems biology signaling research in and, due to a myriad of orthologous sites (), also in other species. l c h e m i c a l s , i f n o t o t h e r w i s e m e n t i o n e d , w e r e b o u g h t w i t h t h e h i g h e s t a v a i l a b l e p u r i t y f r o m S i g m a - A l d r i c h , T a u f k i r c h e n , G e r m a n y .

Long-distance intracellular trafficking is driven by kinesin and dynein motors that carry cargoes along microtubule (MT) tracks. Steady advances have been made in our understanding of the structure and mechanics of motor proteins (; ). However, less is known about how motor proteins bind to the proper cargo, become activated for transport, and deliver that cargo to the correct cellular locale. Several aspects of motor protein function are likely to be regulated in cells, most notably regulation of the motor–cargo and motor–MT interactions (; ). Recent work has begun to elucidate how motors attach to the appropriate cargoes (for review see ; ). A general picture is emerging whereby kinesin family members use adaptor/scaffolding proteins to link to their cargoes, although examples of direct interactions with transmembrane proteins exist (for review see ). In the case of Kinesin-1 (formerly conventional kinesin or Kif5), the kinesin light chain (KLC) subunit binds directly to JNK-interacting protein 1 (JIP1), JIP2, and JIP3/Syd. As scaffolding proteins, the JIP proteins function to organize JNK signaling as well as to link Kinesin-1 to vesicular cargoes (for review see ; ). For regulation of the motor–MT interaction, most of our understanding comes from work on Kinesin-1 (). In the absence of cargo, Kinesin-1 is thought to be inactive as a result of a folded conformation that enables autoinhibition of the N-terminal motor domain by C-terminal tail domains. Autoinhibition leads to a simple prediction for how Kinesin-1 is activated: cargo binding to the Kinesin-1 tail frees the motor domains for ATP-driven motility. Although recombinant kinesin heavy chain (KHC) constructs can be activated in vitro by binding artificial cargoes such as glass slides or beads (; ), activation by cellular binding partners remains to be demonstrated. Alternatively, cargo binding may not be sufficient to activate Kinesin-1, and subsequent events may be required (). We set out to test these models for Kinesin-1 activation. We show that binding of the JIP1 cargo protein to Kinesin-1 is not sufficient for activation. Thus, secondary mechanisms must contribute to regulation of the motor–MT interaction. We identify fasciculation and elongation protein ζ (FEZ) as a binding partner for the KHC tail. We then show that JIP1 and FEZ1 cooperate to activate Kinesin-1 for MT binding and motility. To test the model that cargo binding activates the motor, we investigated whether the binding of JIP1 activates Kinesin-1. Coexpression of myc-KHC + HA-KLC in mammalian cells () results in a complete Kinesin-1 holoenzyme that can be immunoprecipitated with antibodies to the myc (, lane 5) or HA tags (). When coexpressed, Flag-JIP1 bound to the inactive Kinesin-1 molecule (myc-KHC + HA-KLC), as shown by coimmunoprecipitation with antibodies to the myc or Flag tags (, lanes 11 and 12). No proteins were precipitated in the absence of specific antibodies (, lanes 1, 4, 7, and 10). We then used a MT-binding assay to test whether the binding of Flag-JIP1 to Kinesin-1 activates the motor. Myc-KHC expressed alone is not autoinhibited and can be cosedimented with MTs in the presence of 5′-adenylylimidodiphosphate (AMPPNP), a nonhydrolyzable analogue of ATP (, lane 3). Coexpression of myc-KHC + HA-KLC recreates the autoinhibited Kinesin-1 holoenzyme (, lane 6; ). Interaction with Flag-JIP1 was not sufficient for activation, as the myc-KHC–HA-KLC–Flag-JIP1 complex did not cosediment with MTs (, lane 12). Control experiments demonstrated that the sedimentation of KHC is dependent on the presence of MTs (, lanes 1, 4, 7, and 10) and that the hydrolysis of ATP allows the motor to release from the MT (, lanes 2, 5, 8, and 11). These results suggest that an additional event is required to activate Kinesin-1. One possibility is that phosphorylation or some other posttranslational modification is involved (). However, we have been unable to detect any role for phosphorylation in the activation of Kinesin-1 (unpublished data). As the complete autoinhibition of Kinesin-1 requires both the KHC inhibitory tail and the KLC subunit (; ), we hypothesized that the autoinhibitory effects of both of these regions must be relieved for activation. To identify potential cargoes and/or regulators of the KHC tail, we performed a two-hybrid screen of a human brain library using the stalk/tail regions of rat KHC (KHC(750–955); ) as a bait. 27 of the positive clones contained sequences encoding either FEZ1 and FEZ2 (), which is consistent with an interaction between the FEZ1 homologue, UNC-76, and KHC (). The interaction between mammalian FEZ1 and KHC was verified in vitro (Fig. S1 B, available at ). FEZ/UNC-76 likely plays an important if ill-defined role in axonal transport. First, the loss of function results in defects in locomotion and axon outgrowth that are similar to Kinesin-1 mutants (; ). Second, genetic interactions have been demonstrated between and both and (). Third, FEZ1/2 can facilitate neurite outgrowth in cultured cells (; ; ; ). Finally, FEZ1 can be cosedimented with MTs (). Two important domains have been identified within the KHC tail: the highly conserved coiled tail and the inhibitory globular tail (). Within the KHC inhibitory tail, the folding site interacts with the KHC motor/neck regions to generate the folded conformation, thus positioning the IAK site for inhibition of MT-stimulated ADP release (; ; ; ; ; ; ). To determine which of the KHC tail regions are required for interaction with FEZ1, we used a directed two- hybrid assay. Truncation of the KHC inhibitory tail or mutation of the folding site in the inhibitory tail abolished the interaction with FEZ1 (). Interestingly, mutation of the C-terminal residues of KHC, which are critical for the proper functioning of KHC (), also abolished the interaction with FEZ1 (). Importantly, FEZ is the first protein identified that binds to the inhibitory globular tail of KHC. These data indicate that FEZ1 is not likely to be strictly a cargo of Kinesin-1. This is supported by the diffuse localization of FEZ1 in neuronal cells (Fig. S1 E; ) and the lack of UNC-76 accumulation in axonal jams (). Collectively, these results suggest that FEZ1 binding to the KHC folding site could play a critical role in Kinesin-1 activation, perhaps by relieving the folded conformation. To test whether FEZ1 binding is sufficient to activate Kinesin-1, COS cell lysates expressing tagged versions of KHC, KLC, and FEZ1 () were used in coimmunoprecipitation and MT-binding experiments. Immunoprecipitation of KHC brought down a complex of myc-KHC, HA-KLC, and FEZ1-hsv (), demonstrating that FEZ1 interacts with inactive Kinesin-1. This interaction is not sufficient for the activation of Kinesin-1, as the myc-KHC–HA-KLC–FEZ1-hsv complex did not cosediment with MTs in the presence of AMPPNP (). We next tested whether the binding of both FEZ1 and JIP1 to their respective autoinhibitory regions, the KHC inhibitory tail and the KLC subunit, is required to relieve autoinhibition. Inactive Kinesin-1 motors (myc-KHC + HA-KLC) were mixed with tagged versions of both FEZ1 and JIP1 (). An interaction between Kinesin-1 and both Flag-JIP1 and FEZ1-hsv was then demonstrated by coimmunoprecipitation () and a protease protection assay (Fig. S1 F). Control experiments demonstrated that no interaction between Flag-JIP1 and FEZ1-hsv could be detected in the absence of Kinesin-1 (unpublished data). Importantly, the interaction of JIP1 and FEZ1 with Kinesin-1 was sufficient to activate Kinesin-1 for MT binding (). To test whether the activation of Kinesin-1 by JIP1 and FEZ1 resulted in MT-based motility, we generated fluorescent protein (FP)–tagged versions of Kinesin-1. Monomeric versions of the FPs enhanced CFP (mCFP) and citrine (mCit; a brighter variant of enhanced YFP; ) were used to minimize dimerization artifacts. To increase the FP–Kinesin-1 signal over the cellular autofluorescence and to ensure that observed motility was caused by assembled Kinesin-1 holoenzymes rather than active KHC subunits, the nonmotor KLC subunit was labeled with three tandem copies of mCit (3xmCit-KLC). Extraction of FP–Kinesin-1 from COS cells resulted in the movement of fluorescent spots along MTs in an in vitro motility assay (; ). When lysates expressing myc-KHC + 3xmCit-KLC were mixed with lysates of mock-transfected cells, very few motile events were observed (, left). However, when the same lysates were mixed with lysates expressing Flag-JIP1 and FEZ1-hsv, a dramatic increase in the number of motile events was observed (, right). These activated motors moved with a speed (0.71 ± 0.31 μm/s; ) similar to recombinant KHC motors in the same assay, although the processivity of our mammalian-expressed Kinesin-1 was lower (0.46 ± 0.5 μm/run; ). These results demonstrate that JIP1 and FEZ1 cooperate to activate Kinesin-1 for MT-based motility. To test whether JIP1 and FEZ1 cooperate to activate Kinesin-1 in vivo, bicistronic plasmids were generated for the coexpression of KHC-mCit, KLC-mECFP, FEZ1, and JIP1 (Fig. S2, available at ). To reveal the amount of Kinesin-1 that is in an active state in vivo, we developed an assay to trap active FP–Kinesin-1 molecules on MTs by the addition of AMPPNP to live cells transiently permeabilized with streptolysin O (Fig. S3 A). Exposure to AMPPNP resulted in the rapid accumulation of KHC-mCit on MTs ( and Video 1), whereas KHC-mCit + KLC-mCFP remained cytosolic ( and Video 2), indicating that the Kinesin-1 holoenzyme is inactive in vivo. Coexpression with either JIP1 or FEZ1 alone was not sufficient to activate Kinesin-1 (Fig. S3 B and Videos 3 and 4). In contrast, coexpression with both JIP1 and FEZ1 resulted in the rapid accumulation of Kinesin-1 on MTs (, C and D; and Video 5). A truncated version of FEZ1 (FEZ1(1–308)) could not activate Kinesin-1 even in the presence of JIP1 (). Importantly, both JIP1 and FEZ1 accumulated on MTs with Kinesin-1 upon AMPPNP addition (). For the first time, these results demonstrate the activation of a MT-based motor by cellular binding partners. Binding of both JIP1 to the KLC subunit and FEZ1 to the KHC tail likely relieves autoinhibition, freeing the motor domains for MT-based motility (Fig. S3 C). A role for FEZ1 in the activation of Kinesin-1 suggests that in the absence of FEZ1, Kinesin-1 should remain inactive. Indeed, the loss of UNC-76 function results in defects in axonal transport (; ). To test whether the loss of FEZ1 function specifically disrupts Kinesin-1 transport, we analyzed the targeted delivery of JIP1 to the tips of neurites in differentiated CAD cells. As we have been unable to knock down FEZ1 protein expression by RNAi (not depicted), we used the truncated FEZ1 protein (FEZ1(1–308)) that cannot activate Kinesin-1 (). The overexpression of myc-FEZ(1–308) caused a significant decrease (P < 0.001) in JIP1 localization to the tips of neurites () that was similar to the loss of JIP1 transport upon the expression of a Kinesin-1 dominant-negative protein (). The overexpression of full-length myc-FEZ1 or truncated myc-FEZ1(1–230) had less of an effect (). Our results show that FEZ plays an important role in the activation of Kinesin-1. Other FEZ partner proteins such as PKCζ, Disrupted-in-Schizophrenia-1, necdin, and E4B (; ; ; ) may contribute to the regulation of Kinesin-1. Alternatively, FEZ may function as both a regulator of Kinesin-1 and a cargo-linker molecule for the Kinesin-1 transport of FEZ-binding partners to the growth cone (), which is reminiscent of the multiple roles dynactin plays in cytoplasmic dynein transport processes (). Autoinhibition of Kinesin-1 in the absence of cargo prevents futile ATP hydrolysis and allows the rapid and specific control of motor activity both temporally and spatially. Two intramolecular interactions contribute to the autoinhibition of Kinesin-1. First, the KHC inhibitory tail blocks the KHC motor–MT interaction, and, second, the KLC tetritricopeptide repeat motifs push the two KHC motor domains apart (). Here, we demonstrate that binding partners of both the KHC tail and KLC subunit (FEZ1 and JIP1, respectively) are required for the full activation of Kinesin-1 for MT binding and motility (Fig. S4 C, available at ). Cargo binding may be a general mechanism for activating molecular motors. For example, autoinhibition of the kinesin-3 family member KIF1A is likely relieved by the cargo-induced localized dimerization of weak monomers (; ). Autoinhibition and cargo-dependent activation are also likely to play a role in the regulation of actin-based myosin motors (). Myc-tagged rat KHC, HA-tagged rat KLC, and Flag-tagged human JIP1 have been described previously (, ). Mutation of loop 12 (Δloop12 = H275, R279, and K282 to Ala) and all other deletions or mutations were made by QuikChange (Stratagene) or overlapping PCR and were verified by sequencing. A full-length clone of FEZ1 was isolated from a Marathon-Ready human brain cDNA library (CLONTECH Laboratories, Inc.) and inserted into the pLP-CMV-myc vector (CLONTECH Laboratories, Inc.) or pCDNA3 with an hsv-HIS tag. FP-tagged KHC and KLC were created in the vectors mCit/mECFP-N1 and mCit/mECFP-C1 (CLONTECH Laboratories, Inc.). A four-aa linker (SGAG) was inserted between KHC and the FP, whereas a five-aa linker (GPVAT) was inserted between KLC and the FP. For bicistronic vectors, a synthetic intron and the internal ribosome entry site (IRES) from encephalomyocarditus virus (pIRES-puro3; CLONTECH Laboratories, Inc.) were inserted immediately upstream of the KHC-mCit or HA-KLC-mECFP start codons, which were shifted to a more optimal position for translation (i.e., to the 11th AUG from the oligopyrimidine tract; ). Full-length FEZ1 or JIP1 sequences were then inserted immediately upstream of the synthetic intron/IRES cassette. Antibodies used were obtained as follows: myc and HA (Santa Cruz Biotechnology, Inc.), Flag (Sigma-Aldrich), KHC (1614 [Chemicon] and 13 []), and JIP1 (152 []). Polyclonal antibodies to FEZ1 were generated against peptides comprising aa 1–19 (MEAPLVSLDEEFEDLRPSC; 429) or 342–362 (CLNTVIPYEKKASPPSVEDLQ; 432) as described previously (). Fluorescein- and Rhodamine red-X–labeled secondary antibodies were obtained from Jackson ImmunoResearch Laboratories. Plasmid pGBKT7-KHC(750–955) was expressed in yeast strain AH109. A Matchmaker pretransformed human fetal brain library (CLONTECH Laboratories, Inc.) in strain Y187 was screened by yeast mating. 46 of the positive clones contained sequences encoding KLC as expected. 22 clones containing fragments of FEZ1 and five clones containing fragments of FEZ2 were isolated from 7.5 × 10 transformants. For directed two-hybrid analysis, cDNAs were cloned into plasmids pGBKT7 and pACT2 and expressed in yeast strains AH109 and Y187, respectively. A positive interaction was noted by growth on plates lacking histidine and plates lacking adenine. COS and CAD cells were cultured and transfected as described previously except that TransIT-LT1 (Mirus) was used for transfection (). Cell lysis, immunoprecipitation, and MT binding were performed as described previously (). Cells were processed for immunofluorescence () and mounted in 50% glycerol and 0.5% n-propyl gallate in PBS. Images of fixed cells were taken on a microscope (BX51; Olympus) with a UplanFl 60× NA 1.25 oil immersion objective (Olympus) and CCD camera (DP70; Olympus) or on a microscope (Axioplan; Carl Zeiss MicroImaging, Inc.) with a plan-Apochromat 63× NA 1.4 oil immersion objective (Carl Zeiss MicroImaging, Inc.) and color camera (Axiocam; Carl Zeiss MicroImaging, Inc.). ImageJ software (National Institutes of Health [NIH]) was used to quantify JIP1 fluorescence using the Free Hand option to select neurite tips and cell bodies. The mean cell body fluorescence was subtracted from the mean neurite tip fluorescence to provide a measure of JIP1 transport in each cell. Higher fluorescence in tips relative to the cell body represents positive JIP1 transport. All cells in each experimental condition were pooled for statistical analysis ( test; Microsoft Excel). Images were prepared with Photoshop (Adobe). A coverglass was assembled in a Leiden's chamber and maintained at 37°C in Ringer's buffer (10 mM Hepes, 155 mM NaCl, 5 mM KCl, 2 mM CaCl, 1 mM MgCl, 2 mM NaHPO, and 10 mM glucose, pH 7.2). 0.1 μg/ml streptolysin O with 10 mg/ml BSA was added for 30 s, and the cells were rapidly washed three times with buffer I (25 mM Hepes, 5 mM MgCl, 115 mM KOAc, 5 mM NaOAc, 0.5 mM EGTA, pH 7.2, and 10 mg/ml BSA) and placed in 1 mM AMPPNP in buffer I. Images were acquired using an inverted microscope (Eclipse TE-300; Nikon) with a planApo 60× NA 1.4 oil-immersion objective (Nikon) and a cooled CCD camera (Photometrics Quantix; Roper Scientific) controlled by MetaMorph 6.2r6 software (Universal Imaging Corp.). Excitation and emission wavelengths were selected using a filter set (model 86006; Chroma Technology Corp.) and filter wheel controller (Lambda 10-2; Sutter Instrument Co.). Relocation index was calculated on a frame by frame basis. In ImageJ, the MT pattern in the last frame was masked as MT Mask. The rest of the cell, excluding the nucleus and any aggregated motor in the extreme cell periphery, was masked as Other Region Mask. These masks were applied to each frame to measure the total fluorescence pixel intensities (Sum and Sum). The ratio (R) of Sum over Sum was calculated for each frame. The Relocation index was defined as the percent change of R as compared with before the addition of AMPPNP. A two-tailed test was used for statistical analysis. A 25 × 25-mm #1.5 coverglass and a microscope slide were assembled into a flow chamber with double-sided tape (chamber volume of ∼30 μl). Cy5-labeled MTs in P12 buffer (12 mM Pipes/KOH, 1 mM EGTA, and 2 mM MgCl, pH 6.8) with 10 μM taxol were flowed into the chamber and incubated at room temperature for 2 min. 15 mg/ml BSA (in P12 buffer with 10 μM taxol) was then flowed in and incubated for 10 min. 50 μl of oxygen scavenger buffer (1 mM DTT, 1 mM MgCl, 2 mM ATP, 10 mM glucose, 0.1 mg/ml glucose oxidase, 0.08 mg/ml catalase, 10 mg/ml BSA, and 10 μM taxol in P12) containing 10 μl total of COS lysate was flowed in, and the chamber was sealed with wax. Objective-type total internal reflection fluorescence microscopy was performed on a custom-modified microscope (Axiovert 135TV; Carl Zeiss MicroImaging, Inc.) equipped with an α-plan Fluar NA 1.45 objective, 2.5× optovar, 505DCXR dichroic and HQ540/70M emission filters (Chroma Technology Corp.), and a back-illuminated EMCCD camera (Cascade 512B; Roper Scientific). The 488-nm line of a tunable, single-mode, fiber-coupled argon ion laser with Littrow prism (Schäfter und Kirchhoff; Melles Griot) at an incident power of 0.55 mW was used for capturing video sequences at 10 Hz. Videos and images were prepared with ImageJ, Photoshop, and Illustrator (Adobe). Fig. S1 shows the characterization of FEZ1 and its interaction with KHC. Fig. S2 shows the coexpression of FP–Kinesin-1 with FEZ1 and JIP1 via bicistronic plasmids. Fig. S3 shows a live cell MT-binding assay. Videos 1 and 2 show the MT-binding activity of KHC-mCit (Video 1) and KHC-mCit + KLC-mECFP (Video 2) in live COS cells. Video 3 shows the MT-binding activity of KHC-mCit + KLC-mECFP expressed with JIP1 in live COS cells. Video 4 shows the MT-binding activity of KHC-mCit + KLC-mECFP expressed with FEZ1 in live COS cells. Video 5 shows the MT-binding activity of KHC-mCit + KLC-mECFP expressed with JIP1 and FEZ1 in live COS cells. Online supplemental material is available at .

Estrogen is vital for normal postpubertal mammary development, as well as for growth of the majority of breast cancers (; ). The relationship between mammary stem cells and hormone receptor–expressing cells is therefore a fundamental issue in breast biology. It is known that in both the developing and the adult mammary gland only a subset of cells express the estrogen receptor (ER). However, there are conflicting views as to the role of these cells. It has been proposed that ER-positive cells form a stem cell compartment that is directly stimulated by circulating hormones (; ). Alternatively, ER-positive cells may stimulate proliferation of a separate stem cell compartment in a paracrine manner (). Evidence that ER-positive cells form a stem cell compartment has come from studies of cell cycling times and proliferation in both human and rodent tissues. ER- and progesterone receptor–positive cells in the mouse mammary epithelium have been identified as cells that retain BrdU (label-retaining cells) in pulse–chase experiments (, ; ; ), suggesting that they form a slowly cycling cell compartment. Slow in vivo cycling time is thought to be a property of stem cells (; ). Consistent with the slow cycling times is the observation that, in the normal human adult tissue, ER-positive cells do not express markers of proliferation (). It has been proposed that ER down-regulation occurs in these cells before the proliferative response, as stimulation with estrogen led to a decrease in ER expression within 4 h in mice (). The paracrine stimulation theory is supported by observations that ER-null mammary cells can reconstitute the mammary epithelial network in cleared fat pad transplantation experiments only if cotransplanted with mammary cells from ER wild-type mice (). Resolution of this issue requires prospective isolation and functional analysis of the ER-expressing cellular compartment. We have used such an approach to directly demonstrate that the mammary epithelium contains separate hormone-sensing and stem/progenitor compartments. The mouse mammary epithelium consists of two cell compartments, an outer layer of basal epithelial cells, the majority of which are functionally specialized myoepithelial cells, and an inner layer of luminal epithelial cells. These cells can be distinguished by expression of cell type–specific cytoskeletal markers (Fig. S1 A, available at ; ). We have previously described CD24 as a marker that allows the separation and isolation of the luminal (CD24) and basal (CD24) compartments of the mouse mammary epithelium (). To prospectively isolate further subpopulations of the mammary epithelium, we costained mouse mammary cell preparations with CD24-FITC and two cell-surface markers of mouse hematopoietic stem cells, Sca- and prominin-1 (). Costaining with CD24 and Sca-1 defined a CD24/Sca-1 nonepithelial population and a CD24/Sca-1 luminal epithelial population (). Costaining with CD24 and prominin-1 defined only a CD24/prominin-1 population. No cells from the nonepithelial compartment stained with this marker. Simultaneous staining with CD24, Sca-1, and prominin-1 revealed an almost complete overlap between CD24/Sca-1 and CD24/prominin-1 cell populations (). To confirm the luminal epithelial nature of the CD24/prominin-1 and CD24/prominin-1 populations, cells from both these compartments and the CD24 population were sorted onto slides and stained for expression of the basal epithelial cell marker cytokeratin 14 (CK14) and the luminal epithelial cell marker, CK8/18 (CK18). The results () confirmed that the majority of CD24 cells were CK14 basal cells and that the CD24/prominin-1 and CD24/prominin-1 populations were CK18 luminal cells. Unexpectedly, the CD24/prominin-1 cells had more intense CK18 expression than the CD24/prominin-1 cells, suggesting that these were functionally distinct populations. Staining of the epithelial populations defined by CD24 and Sca-1 staining gave essentially identical results (unpublished data). To confirm that the CD24 population consisted predominantly of basal myoepithelial cells, CD24 cells were sorted onto slides and double stained with antibodies against CK14 and α-isoform smooth muscle actin (SMA). As expected, the majority (>90%) were indeed CK14/SMA myoepithelial cells (Fig. S1 B). Keratin-negative cells were also present, and a small number of CK14/SMA-negative cells were observed. To characterize the potential functional roles of the compartments defined by CD24 and prominin-1 staining, including the expression of hormone receptors, expression of several genes (Table S1, available at ) relative to total mammary epithelial cell expression was analyzed by quantitative real time PCR (qPCR). In the CD24 population, as expected, there was increased expression of the myoepithelial/basal cytoskeletal markers (CK14), (CK5), and and , which are two genes involved in actin–myosin activity, were also up-regulated in this population, reflecting the contractile function of myoepithelial cells. In both CD24 populations, expression of luminal cytoskeletal markers was elevated. CD24/prominin-1 cells had significantly elevated levels of (CK18), whereas in CD24/prominin-1 cells, expression of both and (CK19) was significantly increased (). Analysis of expression of five genes involved in sensing systemic hormones, (ERα), (progesterone receptor), (prolactin receptor), , and , demonstrated that all of these genes were significantly up-regulated in CD24/prominin-1 cells (). Four out of the five genes were also significantly down-regulated in the CD24 and the CD24/prominin-1 populations. Cited1 is involved in transcriptional coactivation together with ER (), and S100A6 is a calcium-binding protein that is likely to have several cellular functions, including binding the prolactin receptor (). The pattern of expression of these five genes strongly suggests that the CD24/prominin-1 cell population forms a specific hormone-sensing compartment. As the ultimate purpose of the mammary epithelium is milk production, the expression of four genes for milk proteins (, , , and ) was examined (). Despite the fact that the cells assayed were harvested from virgin animals, both and were significantly more highly expressed in the CD24/prominin-1 population, compared with total mammary epithelium. There was also elevated expression of and in these populations, but this did not achieve significance. These analyses therefore confirmed the basal/myoepithelial and luminal identities of the CD24 and CD24 populations, respectively, and suggested that prominin-1 expression is a marker of two different functional luminal epithelial populations, a hormone-sensing compartment (prominin-1), and a compartment containing cells involved in milk production (prominin-1). As expected, the CD24/prominin-1 hormone receptor–expressing population had very strongly up-regulated levels of both () and (; ). expression was significantly increased in this population, and it significantly decreased in the CD24 population. , which is a marker of basal cells () and breast cancer stem cells (), was significantly decreased in CD24/prominin-1 luminal cells. Consistent with the pattern of expression of cytoskeletal genes, CD24 cells were found to have increased expression of , which is a secreted matrix-binding protein that was previously described as a human breast myoepithelial cell marker and a marker of poor prognosis in breast cancers (). The CD24/prominin-1 cell population also showed a modest, but significantly higher, level of expression of , which is the gene for breast cancer–resistance protein 1. Abcg2 is an ABC transmembrane protein pump that is the main molecular determinant of the side population phenomenon (), and that is also localized to the terminally differentiated milk-producing cells of the alveolar epithelium during lactation (). Analysis of gene expression within the epithelial populations defined by CD24 and Sca-1 staining gave essentially identical results (unpublished data). To confirm that prominin-1 cells in the mouse mammary epithelium were ERα, frozen sections of adult mouse mammary gland were costained with antibodies against these proteins. The results () showed that prominin-1 is apically localized on a subset of mouse mammary luminal epithelial cells, and also confirmed that the majority of luminal epithelial cells with nuclear ERα staining were prominin-1. To provide quantification, cells were sorted onto slides, stained for nuclear ERα, and counted. The data () confirmed that the majority (>80%) of CD24/prominin-1 cells were ERα. To investigate the in vitro progenitor abilities of the CD24, CD24/prominin-1, and CD24/prominin-1 populations, colony forming assays were performed using freshly isolated single cells sorted into individual wells of 96-well plates to determine the relative proportions of mammary colony-forming cells (CFCs) within the populations. The results () demonstrated that the CD24/prominin-1 population contained the highest proportion of CFCs, with >40% of cells capable of forming colonies in vitro. The CD24/prominin-1 population contained significantly fewer CFCs (15%). The CD24 population showed very low levels of CFC activity. These data support an in vitro progenitor function for the CD24/prominin-1 cells. To derive qualitative information on the colonies formed in the CFC assays, freshly isolated single cells sorted from the different populations were cultured on coverslips for 10 d, and the colonies generated were stained to assess their CK14 and CK18 expression pattern. Colonies derived from CD24 CFCs were small and consisted of only CK14 cells, which is consistent with a basal/myoepithelial origin (). CFCs from both the CD24/prominin-1 and CD24/prominin-1 populations formed large colonies that contained cells with a heterogeneous staining pattern consisting of both single-stained CK14 and CK18 and double-stained CK14/CK18 cells, which is consistent with a luminal epithelial cell origin (; ). These data demonstrate that the mouse mammary epithelium contains at least two different CFCs—basal and luminal. To determine whether the differences in in vitro colony-forming ability were reflected by the proliferative status of the three populations, the cell cycle profile of freshly isolated CD24, CD24/prominin-1, and CD24/prominin-1 cells was determined. The results (Fig. S1 C) showed that while the majority of cells in all three populations were in G/G, the percentage of cells in S phase in both the CD24 compartments was significantly increased compared with the CD24 cells. The aforementioned data establish the existence of three distinct cell populations in the mammary epithelium; basal/myoepithelial cells, defined by the CD24 phenotype, and two distinct CD24 luminal epithelial populations, one of which is specialized for detecting systemic hormonal signals, and the other demonstrating elevated expression of milk protein genes and high in vitro colony-forming activity. The key in vivo functional assay for stem/progenitor cell activity in mammary epithelium is the cleared fat pad transplant, in which cells are transplanted into a mammary fat pad of a 3-wk-old mouse from which the endogenous epithelium has been surgically removed (; ). Therefore, to determine the in vivo stem cell activity of cells from the different epithelial populations, limiting dilution cleared fat pad transplantation of prospectively isolated basal epithelial, ER luminal epithelial, and ER luminal epithelial cells was performed. Both prominin-1 and Sca-1 staining were used to separate the ER and ER luminal populations in different transplant experiments. In some experiments, cells were sorted and then immediately resorted to achieve very high purity before transplantation (Fig. S1, D–H). Consistent with our previous data (), CD24 basal epithelial cells () were the most highly enriched for mammary epithelial stem/progenitor activity. The high in vivo growth potential of this population is in contrast to their low in vitro cloning efficiency. Some previous investigators have demonstrated a similar disparity (), whereas others show good correlation between in vitro and in vivo growth (). The reasons for these differences are not clear, but may be caused by differences in sorting strategies, culture conditions, or the ages of mice used to isolate cells. Both luminal ER and luminal ER populations had rates of in vivo epithelial outgrowth formation much lower than the CD24 basal cells, with the luminal ER populations showing the lowest transplantation activity. Double sorting for very high purity did not prevent these rare outgrowths from occurring. Histological examination showed that outgrowths derived from all populations contained SMA-positive myoepithelial cells, ER luminal epithelial cells, and ER luminal epithelial cells (). Thus, the in vivo differentiation potential of both the transplantable cells present at high frequency in the CD24 population and the rare transplantable cells in the luminal populations was similar. Mouse mammary epithelial cell populations highly enriched for stem cells (mammary-repopulating units [MRUs]), for in vitro colony-forming cells (mammary colony-forming cells [MaCFCs]), and for myoepithelial cells (MYOs) have recently been isolated on the basis of expression of CD24 and CD49f (). To relate these populations to the cell compartments we have described in this study and in a previous work (), we identified cells corresponding to the MRUs, MaCFCs, and MYOs in mammary epithelial cell preparations stained for CD24 and CD49f expression (Fig. S2 A, available at ). The identity of the MaCFC and MYO populations was confirmed by in vitro colony-forming assays, qPCR, and CK staining of cells sorted onto slides (unpublished data). Cleared fat pad transplantation activity was used to demonstrate that we had correctly identified the MRUs (). Mammary cell preparations were simultaneously stained with multiple antibodies to identify the MaCFCs, MYOs, and MRUs within the CD24, CD24, and CD24 populations. The results confirmed that the MRUs fell within the CD24 population, as did the MYOs. The MaCFCs corresponded to the CD24 luminal population (Fig. S2, A and B). In conclusion, we have for the first time prospectively isolated mouse mammary luminal ER and ER cells and directly analyzed their cleared fat pad repopulation activity (stem/progenitor cell activity). Our results indicate that the majority of stem/progenitor cell activity in the adult virgin mouse mammary epithelium is located in the basal compartment, confirming and extending previous observations (; ; ). In contrast, the ER luminal compartment contains little in vivo stem/progenitor cell activity, indicating that the hormone-sensing and in vivo stem/progenitor activities of the mammary epithelium are properties of distinct, separate cell populations. Gene expression analysis of the ER and ER luminal epithelial compartments revealed evidence of further distinct functional specialization. These results will provide a basis for elucidating the nature of the interaction between ER and ER luminal epithelial cells and basal stem cells in the mammary gland. Mammary epithelial organoids were harvested from fourth mammary fat pads of mature, virgin, female 10–12-wk-old FVB mice and processed to single cells, as previously described (). Cells were stained as previously described () with 0.5 μg/ml anti–CD24-FITC (clone M1/69; BD Biosciences), 0.25 μg/ml anti–CD45-phycoerythrin (PE)-Cy5 (clone 30-F11; BD Biosciences), 0.1 μg/ml anti–prominin-1-PE (clone 13A4; Insight Biotechnology), and 0.1 μg/ml anti–Sca-1-PE and/or anti–Sca-1-allophycocyanin (clone D7; Cambridge Bioscience). 0.01% DAPI or TO-PRO-3 were used to detect dead cells. Nonspecific IgG controls were used for compensation and to set sort gates. Analysis and exclusion of dead cells, CD45 cells, and nonsingle cells was performed as previously described (). For double sorting of populations, cells were isolated after the first round of sorting, pelleted, resuspended in fresh 0.01% TO-PRO-3, and resorted using the same gates. To investigate whether different fluorochrome conjugates affect the staining profile of CD24, cells at a density of 10 cells/ml were stained with 0.5 μg/ml anti–CD24-FITC (clone M1/69), anti–CD24-PE (clone M1/69; tested at a range of concentrations from 0.2 μg/ml up to 10 μg/ml), 0.25 μg/ml anti–CD24-PE-Cy5 (clone M1/69), or 0.5 μg/ml unconjugated anti-CD24 (clone M1/69), followed by anti–rat–Alexa Fluor 633 (A-21094; 20 μg/ml) in a two-step procedure. To confirm that populations identified with different fluorochrome conjugates were identical, cells at a density of 10 cells/ml were stained simultaneously with 0.5 μg/ml anti–CD24-FITC (clone M1/69) and 0.25 μg/ml anti–CD24-PE-Cy5 (Fig. S2, B and C) To identify previously described () MaCFCs, MYOs, and MRUs (stem cells) and locate them within the three-region CD24 profile described both in this study and in a previous work (), cells were stained with anti–CD24-PE (clone M1/69; 1.5 μg/ml; BD Biosciences), anti–CD49f-FITC (clone GoH3; 1:50 vol/vol dilution; BD Biosciences), and 0.25 μg/ml anti–CD45-PE-Cy7 (clone 30-F11; BD Biosciences) together with 0.25 μg/ml anti–CD24-PE-Cy5 (clone M1/69; Insight Biotechnology). 0.01% TO-PRO-3 was used to detect dead cells. Nonspecific IgG controls were used for compensation and to set sort gates. Analysis and exclusion of dead cells, CD45 cells and nonsingle cells was performed as previously described (). MaCFC, MYO, and MRU cells were identified, and their location within the three regions defined by anti–CD24-PE-Cy5 was determined by staining (Fig. S2 A). For prominin-1 and ERα staining of frozen sections of mouse mammary gland, small pieces (∼5 mm) of 10-wk-old mouse mammary fat pads were fixed in 4% paraformaldehyde in PBS for 1 h, infiltrated with 1 M sucrose overnight at 4°C as a cryoprotectant, and snap frozen in isopentane cooled in liquid nitrogen. 10-μm frozen sections were cut and stored at −80°C. Before use, sections were thawed at RT for 15–30 min and stained essentially as previously described () with 5 μg/ml anti–prominin-1 (rat monoclonal clone 13A4; Insight Biotechnology) and 6 μg/ml anti-ERα (mouse monoclonal clone 1D5; Insight Biotechnology). Sections were counterstained with TO-PRO-3 (0.01% in PBS) and mounted in Vectashield H1000 mounting medium (Vector Laboratories). For cytoskeletal marker staining, tissue was snap frozen with no prior fixation. Frozen sections were fixed in 1:1 methanol acetone at −20°C for 5 min and stained with antibodies against 2.1 μg/ml CK14 (mouse IgG3 clone LL002; Lab Vision) and 2 μg/ml CK8/18 (CK18; mouse IgG1 clone 5D3; Vision Biosystems) in addition to DAPI. Secondary antibodies were isotype-specific goat anti–mouse antibodies (A21127 and A21157; Invitrogen) conjugated to Alexa Fluor 488 or 555 fluorochromes. Sections were mounted in Vectashield. Lack of nonspecific staining by secondary antibodies was confirmed using isotype-matched control primary antibodies (Cambridge Bioscience). Lack of cross-reactivity was confirmed with controls incubated with single primary antibodies and both secondary antibodies. Immunophenotyping of cell populations sorted onto slides was performed as previously described () using the anti-CK14 (LLOO2) and anti-CK18 (5D3) primary antibodies, secondary antibodies, and DAPI, as described in the previous paragraph. Samples were mounted in Vectashield. For immunostaining of clones on coverslips, mouse mammary cells were cultured on coverslips with a feeder layer of irradiated 3T3-L1 preadipocytes in 1:1 DME/F12 (Sigma-Aldrich) with 10% fetal calf serum (Invitrogen), 5 μg/ml insulin (Sigma-Aldrich), 10 ng/ml epidermal growth factor (Sigma-Aldrich), and 10 ng/ml cholera toxin (Sigma-Aldrich). Cultures were maintained at 37°C in a 5% vol/vol CO/5% vol/vol O atmosphere for 8–10 d before fixation in cold (−20°C) 1:1 methanol/acetone and staining (). Clones were stained with anti-CK14 (LLOO2) and anti-CK18 (5D3) primary antibodies, secondary antibodies, and DAPI, as described in the previous paragraph. Samples were mounted in Vectashield. Analysis of the in vitro colony-forming potential of mammary cell subpopulations was performed in a 96-well plate format, as previously described (). The feeder layers, growth media, and conditions were identical to those described in this section for coverslip culture. Cells were pelleted and resuspended in ice-cold 70% ethanol and maintained at 4°C for at least 15 min. They were then pelleted and resuspended in PBS containing 40 μg/ml propidium iodide and 100 μg/ml RNase A. Samples were incubated at 37°C for 20 min and analyzed by standard protocols. qPCR reactions were performed as previously described () to determine fold changes in expression of a selection of genes (Table S1) in mammary epithelial cell subpopulations, compared with a leukocyte-depleted, bulk mammary epithelial cell (CD45/CD24) comparator sample. Significant deviation of the mean value of the three data points from a fold difference of 1 (no change compared with the comparator sample) was tested using a test on log10-transformed data. Freshly isolated cells were sorted and transplanted (). There was no intervening culture period before transplantation. All animal work was approved by the Local Ethics Committee and performed under Home Office approval. 8 wk after transplantation, fat pads were wholemounted and analyzed, as previously described (). Failed clears were excluded from the analysis. Wholemounts were examined on a binocular microscope (MZ12.5; Leica) with a Plan 1× lens and a Cold Light Source (Leica). Images were captured with a camera (DFC500; Leica) and IM50 image acquisition software (with auto white balance and auto exposure activated). A proportion of successful transplants had a region of epithelial outgrowth dissected out under the microscope for paraffin embedding by routine methods and routine immunocytochemistry to detect SMA (clone 1A4; Sigma-Aldrich) and ERα (clone 1D5). Images of stained sections were captured on a microscope (DM RA2; Leica) using a 63× oil Plan Apo lens (N/A 1.32), a camera (DFC320; Leica), and IM50 image acquisition software (with auto white balance and auto exposure activated). Photo montages were generated using Photoshop, but were not further processed. To confirm that differential survival of cells before transplantation did not confound assessment of their relative engraftment potentials, viability assays of CD24, CD24/prominin-1, and CD24/prominin-1 cells were performed by TO-PRO-3 staining and flow cytometric analysis immediately after separation, and again after 3.5 h on ice. The analysis showed that CD24 cells, which were the most potent at repopulating cleared fat pads (), had a viability immediately after sorting of >70%, dropping to >65% after 3.5 h. CD24/prominin-1 cells had a viability of >90% immediately after sorting, and this was maintained after 3.5 h on ice. CD24/prominin-1 cells had a viability of >75% immediately after sorting, and this was also maintained at >75% after 3.5 h on ice. Therefore, differential sensitivity to the sorting procedure or to being maintained on ice during transplantation could not explain the differences in engraftment potential. Table S1 lists the genes examined by qPCR analysis. Fig. S1 shows the characterization of mouse mammary epithelial cell subpopulations. Fig. S2 shows the identification of MaCFC, Myo, and MRU regions. Online supplemental material is available at .

Cells show dynamic reorganization of cytoskeletons to migrate and change their shape. Recent studies have revealed similarities in signaling pathways and cytoskeletal reorganization involved in cell shape changes during epithelial tissue morphogenesis and wound healing (for review see ). Nonmuscle myosin II is an actin-based motor protein and is presumed to contract the actin cytoskeleton. In higher eukaryotes, nonmuscle myosin II (hereafter referred to as myosin) assembly and motor activity are controlled by the phosphorylation of myosin regulatory light chain (MLC) at Thr18/Ser19 (). Myosin is required for epithelial morphogenesis in (; ; ; ) and in (). It is not clear, however, whether myosin is required as a motor to change cell shape, a regulator of cell–cell and cell–matrix adhesion, or a combination of both. Myosin also plays an important role during epithelial wound healing in which cell migration is coordinated with the purse string–like contraction of a thick cable of actin and myosin in the leading edge of marginal cells encompassing a large wound, drawing the wound edges together (; ; ). A purse string–like cable is also formed around dead cell remnants to extrude apoptosed cells (). How the actomyosin contractile apparatus is formed and how it drives cytoskeletal rearrangement to cause closure of epithelial sheets are important questions. To address these questions, we examined the sequence of steps involved during wound closure in MDCK cell sheets by using live-cell time-lapse imaging of the cell shape, myosin, and other associated proteins. Our observations identify a regulatory mechanism for localized assembly and activation of a myosin ring at the tight junction in cells surrounding the wound. Ring contraction can integrate neighboring cell shape changes and appears to cooperate with another myosin complex at the extending basal membrane. To study both cell and protein dynamics during different stages of wound closure of an epithelial sheet, we reproducibly formed small, circular wounds in MDCK cell monolayers by laser ablation () and performed live-cell time-lapse imaging during wound closure. After wounding, dead cell remnants were extruded from the wound site (Fig. S1 A, available at ), as reported previously (). When we wounded a monolayer of MDCK cells expressing E-cadherin fused to red fluorescent protein (Ecad-RFP), Ecad-RFP fluorescence decreased at the plasma membrane between the wounded cell and surrounding cells 5–10 min after laser ablation, and Ecad-RFP at contacts between surrounding cells extended toward the center of the wound and eventually coalesced into a vertex between multiple cells (). These results show rearrangement of cell–cell adhesions occurring concomitantly with cell shape change during wound closure. In addition to remodeling cell–cell adhesions, cell shape change involved dynamic rearrangement of the actin cytoskeleton. To define the sequence of steps that were involved in actomyosin assembly and movement to close the wound, we followed the dynamics of myosin localization using MLC fused to EGFP (MLC-EGFP). In confluent MDCK cell monolayers, MLC-EGFP was diffusely distributed in the cytoplasm and colocalized with F-actin fibers at the base of cells, but it was not localized to the boundary between cells either at the base of the cells or at lateral membranes (). After laser ablation, MLC-EGFP assembled into a ring in cells surrounding the wound, consistent with previous studies showing myosin localization around a wound in cell sheets and embryos (; ). This MLC ring contracted over time during wound closure (), as did an EGFP-actin ring assembled around the wound (Fig. S1 B; ; ). Next, we analyzed MLC-EGFP distribution along the apical–basal axis during wound closure in time-lapse images of XZ sections (). About 5 min after laser ablation, MLC-EGFP fluorescence began to accumulate at the borders between the apical and lateral membranes and between the basal and lateral membranes (referred to as the apical–lateral border and basal–lateral border, respectively) of cells facing the wound. Both apical and basal MLC-EGFP moved toward the wound center. However, apical MLC-EGFP moved toward the basal cell membrane and eventually coalesced with basal MLC-EGFP at the base of cells (). Note that the shape of cells viewed in XZ sections changed from rectangular to triangular as the apical MLC-EGFP drew the upper edge inward and down to the base (). Thus, myosin movement tracked the closure of the wound, thinning adjacent cells to cover the wound. In cells fixed ∼5 min after ablation, MLC-EGFP was concentrated in a ring around the wound, 3.5 μm above the base of the cells (). This MLC-EGFP ring corresponded to the accumulation of MLC-EGFP at the apical tip of the lateral membrane in XZ images (). The MLC-EGFP ring colocalized with F-actin identified by Alexa Fluor 568 phalloidin (). Basal XY sections of fixed cells also showed accumulation of MLC-EGFP fibers and F-actin around the wound (), and this MLC-EGFP accumulation appeared to correspond to the MLC-EGFP at the basal–lateral border in live cells (). An antibody specific for phospho-MLC (Thr18/Ser19) recognized both MLC at the apical–lateral and basal–lateral borders of cells surrounding the wound (). Thus, both actin and active myosin are present at the apical–lateral and basal–lateral borders, indicating a possible role of actomyosin contraction in wound closure. These results indicate that MLC accumulated in two spatially distinct complexes: one at the basal–lateral border adjacent to cell extensions into the wound and the other in a ring at the apical–lateral border of cells around the wound. The ring structure descended from its initial apical position to a basal position as it contracted and constricted the lateral membrane of cells surrounding the wound, thereby expelling remnants of dead cells into the medium as the continuity of the monolayer was restored (see ). We tested whether myosin activity was required to contract the ring and close the wound. In the presence of blebbistatin, a nonmuscle myosin II ATPase inhibitor, MLC-EGFP accumulated at the apical–lateral and basal–lateral borders in cells surrounding the wound (). However, the MLC-EGFP ring did not contract, although the ring shape changed as cells extended into the wound from the basal membrane ( b). Upon removal of blebbistatin, the MLC-EGFP ring began to contract and closed the wound within 5 min (). These results indicate that myosin is the motor providing the primary driving force to contract the MLC ring at the apical–lateral border but not to extend basal membranes into the wound and that ring contraction is required to constrict the lateral membrane of cells so that the cells closing the wound coalesce into a vertex. MLC phosphorylation is critical for actomyosin assembly and contractility. Among kinases known to phosphorylate MLC, both myosin light chain kinase (MLCK) and Rho-associated kinase (Rho-kinase) have been linked to wound closure and extrusion of apoptotic cells (; ). To examine their involvement in MLC accumulation at each site, we observed wound closure in the presence of their inhibitors. Y27632, a Rho-kinase inhibitor, inhibited formation of the MLC-EGFP ring at the apical–lateral border and subsequent contraction of the lateral membrane (). However, in the presence of Y27632, MLC-EGFP still accumulated at the basal–lateral border of cells around the wound even though actin stress fibers containing MLC-EGFP appeared mostly disrupted, and very thin basal cell membrane protrusions extended into the wound, partially closing it. However, at the level of the apical–lateral border, the wound space remained open (). ML-7, an MLCK inhibitor, had little or no effect on either MLC-EGFP accumulation or the ensuing cell shape changes in our system (Fig. S2, available at ). To determine the structure of the basal membrane protrusions in the XY plane, we analyzed cells expressing EGFP-actin. We found that cells extended lamellipodial protrusions into the wound site. However, in the presence of Y27632, lamellipodial protrusions were replaced by filopodia (). These results indicate that Rho-kinase is involved in actomyosin organization at both sites. At the apical–lateral border, Rho-kinase is required for assembling the MLC ring and the following changes in cell shape during ring contraction. At the basal–lateral border, Rho-kinase is required to activate lamellipodial protrusion and maintain its morphology. The mechanism for MLC accumulation at the basal–lateral border is unknown. Because MLC accumulation was inhibited at both sites in cells expressing C3 toxin (unpublished data), both accumulations appear to be Rho dependent. Because Rho-kinase is required for wound closure, we examined its localization during this process. Previous studies have shown that Rho-kinase is distributed in the cytoplasm () but is localized specifically at the cleavage furrow during cytokinesis () and along actin stress fibers () and vimentin intermediate filaments (). EGFP–Rho-kinase, coexpressed with MLC-RFP, distributed diffusely in the cytoplasm and was not found with MLC-RFP at the base of cells within the intact monolayer ( and not depicted). Expression of EGFP–Rho-kinase did not affect the localization of MLC-RFP. After laser ablation, EGFP–Rho-kinase accumulated simultaneously with MLC-RFP at the apical–lateral border facing the wound and then moved with MLC-RFP toward the base of the cell; meanwhile, coaccumulation of EGFP–Rho-kinase and MLC-RFP was not observed in the basal–lateral border (). It is unknown why the accumulation of Rho-kinase was observed only at the apical–lateral border and not at the basal–lateral border, whereas Rho-kinase activity was involved at both sites during wound closure. Because Y27632 inhibited MLC ring accumulation (), it is possible that Rho-kinase locally activates actomyosin ring assembly at that site by phosphorylating coaccumulated MLC. That Rho-kinase colocalized and moved with the actomyosin ring during wound closure suggests that Rho-kinase also plays a role in maintaining contractile activity of the actomyosin ring. These results might be correlated with the observation of a concentric Rho active zone around wounds in oocytes and its movement inward in concert with wound closure (). Our results show that the actomyosin ring and its activator Rho-kinase colocalize at the apical–lateral border of cells. This region of the plasma membrane of polarized epithelial cells contains many junctional complexes that constitute the tight junction and adherens junction. A previous report demonstrated that the tight junction scaffolding protein ZO-1 colocalized with actin filaments in the ring 1 h after wounding (). To examine at which sites the actomyosin ring was organized and the dynamics of its localization, we compared the distribution of junctional proteins with the distribution of MLC in the apical–lateral region. In a cell expressing both MLC-RFP and ZO-1–EGFP, MLC-RFP started to accumulate at the site of ZO-1–EGFP concentration adjacent to the ablated cells within 5 min of ablation, and the two comigrated thereafter (). As shown in , MLC accumulated slightly above the tip of E-cadherin or α-catenin signal but below claudin-1 and did not precisely colocalize with either of them, yet MLC accumulation colocalized with l-afadin at the apical–lateral border. Note that in MDCK cells the adherens junction is poorly developed and the major component of adherens junctions, E-cadherin, is localized over the entire surface of the lateral membrane and is not focused in an adherens junction (). Significantly, addition of Y27632 caused ZO-1–EGFP to remain at its initial position, and as shown previously, there was little or no accumulation of MLC or change in cell shape for 20 min after ablation (). These data suggest that a complex including ZO-1 and l-afadin localized at the cytoplasmic surface of tight junction strands serves as a scaffold for assembly and localization of the myosin ring adjacent to the wound. Although the molecular linkages involved are not fully understood, ZO-1 and l-afadin have been reported to link the actin cytoskeleton to tight junction strands and nectin, respectively (; ); therefore, ZO-1/claudin and l-afadin/nectin are good candidates to anchor the actomyosin cables to cell–cell adhesion sites around the wound to form a continuous ring. For the MLC-EGFP construct, chicken cDNA (; a gift from M. Ikebe, University of Massachusetts Medical School, Worcester, MA) was amplified by PCR and subcloned into pEGFP-N3 (pEGFP-N3-MLC). For the MLC-RFP construct, cDNA was amplified by PCR and cloned in place of cDNA in pEGFP-N3-MLC. The EGFP–Rho-kinase construct was provided by K. Kaibuchi (Nagoya University Graduate School of Medicine, Nagoya, Japan). pCYFP-ZO-1 and pNYFP-Cld1 () were provided by S. Tsukita (deceased). For ZO-1–EGFP and EGFP-Claudin1, and cDNA were cloned into pCAG-CGFP and -NGFP, respectively. The EGFP–l-afadin construct was provided by Y. Takai (Osaka University Graduate School of Medicine, Suita, Japan; ). The EYFP-C3 toxin construct was provided by E. Lemichez (Institut National de la Santé et de la Recherche Médicale, Nice, France). Blebbistatin, Y27632, and ML-7 were purchased from Calbiochem. MDCKII cells expressing Ecad-RFP, EGFP–α-catenin (), or EGFP-actin () were maintained in DME supplemented with 10% fetal bovine serum. MDCKII cells expressing MLC were maintained in DME supplemented with 10% calf serum. For live-cell imaging, we used complete medium buffered with 20 mM Hepes, pH 7.4. Plasmid transfection was performed with Lipofectamine 2000 (Invitrogen) or Nucleofector (Amaxa Biosystems) according to the manufacturer's protocol. To isolate MDCKII cells stably expressing MLC-EGFP, MLC-RFP, or Ecad-RFP, cells were transfected with respective plasmid and clones were selected using G-418 (Invitrogen). To ablate cells, we used an ultraviolet laser as described previously (), a N laser (model VSL-337ND; Laser Science, Inc.) that produced 3-ns pulses of 6 μJ (at the source) of a 337.1-nm UV light mounted on a microscope (IX70; Olympus). The laser beam was directed onto a dichroic mirror (380DCLP; Chroma Technology Corp.) and passed through the objective (UApo/340, 40×; Olympus) onto the specimen. The dose of UV light delivered was adjusted by counting the number of pulses at 2 Hz. The laser was focused onto the cell to be ablated and activated while under bright field observation. One to three cells were ablated in each experiment. Confluent monolayers of cells were grown on glass coverslips coated with collagen (type I; Sigma-Aldrich) for 24–48 h and laser ablated. Images of cells were collected at 37°C with a microscope (40×/1.35 NA, oil-immersion objective; Olympus) through a cooled charge-coupled device camera (CoolSNAP HQ; Roper Scientific, Inc.) using Simple PCI software (Compix, Inc.) or with a confocal microscope (Fluoview 300; PlanApo; 60×/1.4 NA; oil-immersion objective; Olympus) configured on a microscope with Fluoview 2.1 software for 15–60 min at 30–60-s intervals. XZ images were collected at 30-s intervals. Image analysis was done with the open-source program ImageJ (NIH). After laser ablation, cells were maintained at 37°C for the indicated time and fixed in 3.7% (vol/vol) formaldehyde in PBS. Fixed cells were permeabilized with 0.2% Triton X-100 in PBS for 10 min and incubated with anti–phospho-MLC (Thr18/Ser19) antibody (Cell Signaling) followed by detection with Alexa Fluor 568 anti-rabbit secondary antibody (Invitrogen). Cells were incubated with Alexa Fluor 568 phalloidin (Invitrogen) to visualize F-actin. Fig. S1 shows the extrusion of laser-ablated cells by surrounding cells and F-actin ring around the ablated cell. Fig. S2 shows that ML-7 had little or no effect on either the MLC-EGFP accumulation or the ensuing cell shape changes. Video 1 is a time-lapse video of Ecad-RFP. Video 2 is a time-lapse video of MLC-EGFP. Video 3 is a time-lapse video of MLC-EGFP. Video 4 is a time-lapse video of MLC-EGFP in XZ sections showing cells at both sides of the wound. Video 5 is a time-lapse video of MLC-EGFP with Y27632. Video 6 is a time-lapse video of EGFP–Rho-kinase. Video 7 is a time-lapse video of MLC-RFP. Video 8 is a time-lapse video of ZO-1–EGFP. Video 9 is a time-lapse video of MLC-RFP.

xref #text To characterize the role of Smurf1 in MDAMB-231 breast cancer cells, we reduced its expression by using siRNAs or short hairpin RNAs (shRNAs) targeting different sequences in the gene (). siRNA () or shRNA () silencing was accompanied by cell rounding, a reduction in size, and a loss of membrane protrusions. Accordingly, silenced cells displayed cortical actin staining, but the typical actin-rich lamellipodial extensions seen in control cells disappeared. In addition, some cells displayed bleblike structures typical of high Rho activity () with no effect on cell viability (unpublished data). These results confirm that Smurf1 expression is required for the formation of cellular protrusions; more precisely, we show in carcinoma cells that Smurf1 is required for the extension of lamellipodia. Although no substantial change in the overall accumulation of total RhoA or active GTP-bound RhoA could be observed in the Smurf1-silenced cells (), immunostaining studies showed a localized accumulation of RhoA in small areas at the cell periphery, particularly in some blebs (). The absence of a detectable change in the overall levels of RhoA might be explained by the presence of a large intracellular pool of RhoA unaltered by Smurf1 silencing. To test whether the pool of RhoA accumulated at the cell periphery is active, and therefore able to affect downstream signaling, we used an in situ probe for evaluation of the spatial Rho activity (). Although the majority of control cells only showed an intracellular cytoplasmic active Rho localization (C3 treatment to inactive endogenous Rho proteins in control cells revealed that the staining of the nucleus and perinuclear region was nonspecific; unpublished data), most of the Smurf1-silenced cells displayed numerous dense patches of active Rho at the cell periphery, which is consistent with the accumulation of RhoA protein (). Furthermore, we observed that the activation of Rho was accompanied with a recruitment of ROCK1 at the cell periphery, together with an accumulation of phosphorylated, active MLC2 (). In an extension of the work of , we conclude that Smurf1 down-regulates RhoA activity and the downstream ROCK–MLC2 signaling at the cell periphery. To test whether this was responsible for the phenotype observed in the Smurf1-silenced cells, we first used a cell-permeable C3 toxin (tat-C3) known to ribosylate and inhibit Rho protein activity. As expected, lamellipodia were restored, and blebs and cortical actin disappeared in cells treated with a low dose of C3 (25 nM; ). In both control and silenced cells, a high dose of C3 (>100 nM) resulted in the collapse of the actin cytoskeleton, which is consistent with a total inhibition of Rho activity and the formation of abnormal protrusions, as previously described (). RhoA is polyubiquitylated by Smurf1 on its conserved lysine 6 (), which is also present in RhoC. To test whether the phenotype of Smurf1-silenced cells is caused by a regulation of RhoA levels, but not of RhoC, we used siRNAs specifically targeting RhoA. We showed that the rounded phenotype and the appearance of pMLC-rich blebs are specifically antagonized by the RhoA siRNAs (); interestingly, we did not observe the more dramatic phenotype observed with 100 nM C3 treatment, suggesting that at high doses, C3 inhibits Rho GTPases in addition to RhoA. Similar experiments using siRNAs against RhoC did not produce any reversion of the Smurf1-silencing phenotype (unpublished data). Inhibition of ROCK using the synthetic inhibitor Y-27632 resulted in a phenotype similar to that observed with high doses of C3 (). Because MLC can also be activated by kinases other than ROCK (e.g., MLC kinase and MRCK), we showed that C3 and Y-27632 treatments also abolished the phosphorylation of MLC2 seen in Smurf1-silenced cells, indicating that it was Rho- and ROCK- dependent (). We next showed that blebbistatin, which is a nonmuscle myosin ATPase inhibitor, also rescued the effect of Smurf1 silencing (), suggesting that the absence of protrusion, the rounding, and the appearance of blebs are at least partially caused by an increase in peripheral actomyosin contractility. Importantly, we further confirmed the central role of Rho signaling by showing that Smurf1 silencing in tumor cells does not affect the other ubiquitylation targets of Smurf1, such as the SMAD proteins involved in the transcriptional response to TGFβ (Fig. S1 a, available at ; ) and the MEKK2–JNK signaling pathway (Fig. S1 b; ). Altogether, our work demonstrates that the activation of RhoA–ROCK–MLC2 signaling at the cell periphery observed in the Smurf1-silenced cells, and the consequent increase in actomyosin contractility, is responsible for the loss of lamellipodia and the formation of blebs. To evaluate the consequence of these morphological modifications on the invasive phenotype, we first monitored the random motility of MDAMB-231 cells upon Smurf1 silencing. Cell migration was reduced by ∼80%; it was partially restored C3 and Y-27632 treatments (). These results, together with the regulation of Rho signaling, were reproduced in BE and HT1080 tumor cells (unpublished data). Consistently, cell migration was also inhibited by overexpression of two Smurf1-interfering mutants: Smurf1-C699A, which is deficient in its ubiquitin ligase activity (), and Smurf1-ΔC2, which fails to localize at the plasma membrane (; ). As for Smurf1 silencing, both mutants induced cell rounding, loss of membrane protrusions, blebbing, and accumulation of RhoA at the cell periphery (). Notably, C3 and Y-27632 treatments in control cells did not significantly affect cell migration. One possibility is that in MDAMB-231 cells, the actomyosin contractility necessary for cell movement is also generated by the Cdc42–MRCK pathway (). These results indicate that RhoA–ROCK–MLC2 signaling is not essential for MDAMB-231 cell movement in 2D, and, more importantly, that it needs to be specifically inhibited by Smurf1 at the cell periphery to allow the formation of protrusions and tumor cell migration. The lowering of myosin-dependent contractile signals at the cell edge is probably required to diminish the local tension and adhesive forces, allowing the Rac-driven extension of the lamellipodium. The localized nature of Smurf1 activity permits it to maintain the intracellular Rho–ROCK–MLC2 activity, which is likely to be required for contraction of the cell body and retraction of the rear (Fig. S2, available at ). In contrast to the situation in 2D tissue culture, there are several mechanisms of tumor cell migration in vivo and in 3D models of invasion (; ). The protease-dependent mesenchymal mode of movement needs low activities of RhoA and ROCK to allow, as seen in 2D, the extension of Rac-dependent protrusions at the front (). On the contrary, the second main mode of motility, called amoeboid, is protease independent, but depends on high activities of Rho, ROCK, and MLC2 to generate cortical contractile forces used for matrix deformation; it does not extend protrusions and it is associated with a more rounded morphology (; ). Importantly, cancer cells can switch from one mode of movement to another, depending on the 3D microenvironment. Rho signaling is pivotal because its blockade induces the mesenchymal–amoeboid transition and, conversely, its constitutive activation leads to a conversion to the amoeboid mode of movement. To analyze the role of Smurf1 in invasion, we first studied cell motility in 3D Matrigel matrices in vitro. We used the previously described mesenchymal BE colon carcinoma cells, which display low RhoA activity, and the amoeboid A375m2 melanoma cells with high RhoA activity (; ). When Smurf1 was silenced in A375m2 cells, the morphology and the invasion of the 3D matrix were unaltered, as most of the cells kept the rounded phenotype (). In contrast, silencing of Smurf1 in BE cells resulted in a dramatic transition from the mesenchymal to the amoeboid morphology and mode of invasion, which was partially reversed by treatment with Y-27632. These results suggest that in cancer cells, Smurf1 expression, through the local down-regulation of RhoA–ROCK activity and the formation of cellular protrusions, favors the mesenchymal mode of invasion in 3D. We next asked if Smurf1 would also be required for the mesenchymal morphology in an in vivo tumor environment and what would be its impact on invasion. To trigger stable and long-term inhibition of Smurf1 in vivo, we generated BE tumor cell lines stably expressing GFP and shRNAs against Smurf1 (). As seen with BE and MDAMB-231 cells transiently expressing the siRNAs against Smurf1, these clones showed a reduced cell motility in 2D and displayed the amoeboid phenotype in Matrigel (unpublished data). BE clones were grown as subcutaneous tumors in nude mice, and multiphoton intravital microscopy was used on living animals to analyze tumor cell morphology and movement. In vivo, BE cells expressing the control shRNA had an overall elongated morphology, whereas cells expressing the Smurf1 shRNAs were mostly rounded, confirming our in vitro observations (). In agreement, in the control tumors, time-lapse intravital imaging showed that the moving cells used the mesenchymal mode of motility by extending protrusions at the front. Interestingly, only a minority of cells was migrating, but even the static cells presented the elongated morphology (; Video 1), indicating that additional factors intrinsic to the tumor microenvironment are required to promote cell movement. In the Smurf1-silenced tumors, all cells were rounded, particularly those migrating, thus confirming that they were using the amoeboid motility (; Video 2). These data demonstrate that in tumors, Smurf1 favors the mesenchymal mode of invasion, and that its inhibition induces the mesenchymal–amoeboid transition in agreement with the activation of Rho-dependent contractile signals in 2D. Interestingly, although previous studies showed that the different modes of invasion and the plasticity between those were characterized and regulated by the general level of RhoA–ROCK–MLC2 activity in the cells (; ), our work shows that subtle changes at the cell periphery induced by Smurf1, which were undetectable at the biochemical level, are sufficient to induce the transition in the invasion strategies. Factors that can regulate either Smurf1 expression or its activity in the process of tumor development are therefore potential key elements in the plasticity of cancer cell invasion. This is important because the plasticity between the different modes of movement allows tumor cells to be highly adaptable to the dynamic microenvironment in cancer, and it might provide an escape mechanism to antiinvasive treatments (). Finally, we sought to analyze the impact of Smurf1 silencing on the efficiency of cell movement, invasion, and metastasis. Intravital analyses revealed that upon Smurf1 silencing there were approximately three times as many cells migrating, indicating that the transition to the amoeboid mode movement significantly increases cell motility in tumors (). Strikingly, although control cells were never seen in tumor vessels (blood or lymphatic), Smurf1-silenced amoeboid BE cells were often imaged in the vessels or intravasating (; Videos 3 and 4). The number of blood vessels was unaffected by Smurf1 silencing (Fig. S3). However, the growth of Smurf1-silenced or control subcutaneous tumors never resulted in the formation of visible distant metastases, indicating that in this model additional factors are needed for BE cells to efficiently metastasize. Nevertheless, to test whether Smurf1-silencing could also affect the later stages of cancer dissemination, we injected tumor cells directly into the tail vein of immunodeficient mice for measurement of lung metastases. As for the subcutaneous model, we found that BE cells did not readily metastasize and that Smurf1-silencing did not increase metastasis (unpublished data). Collectively, these data show that Smurf1 knockdown and the mesenchymal–amoeboid transition in tumors increases tumor cell motility and favors intravasation, but it is not sufficient to promote metastasis after cells have entered the vessels. Our results are in agreement with previous work (), showing that mesenchymal cells often fragment when they enter the blood stream and that, instead, highly metastatic carcinoma cells efficiently crawl into the blood vessels by using the amoeboid locomotion. It is worth noting that the efficiency of tumor cell motility in 2D therefore does not necessarily reflect the invasive potential in vivo. In conclusion, our work demonstrates that Smurf1, through the regulation of peripheral RhoA–ROCK–MLC2 signaling, is a key element of tumor cell motility and invasion in vitro, as well as in vivo. Recent data have suggested that Smurf1 activity can be activated by TGFβ in normal epithelial cells to induce the degradation of RhoA at the tight junctions, leading to the dissolution of these cell–cell junctions and the subsequent TGFβ-dependent epithelial–mesenchymal transition (). We hypothesize that Smurf1 could be implicated in two crucial aspects of tumor progression in a Rho-dependent manner; in the first stages of the invasive progression, Smurf1 might be activated by TGFβ to disrupt normal epithelial organization (). In carcinomas that have lost their epithelial organization, such as those analyzed in this study, down-regulation of Smurf1 expression or activity would then increase motility. Investigation of the “Oncomine” resource () shows that overall Smurf1 levels do not change greatly with tumor grade or prognosis; however, this is entirely consistent with our results, which indicate that reduction of Smurf1 facilitates cell motility and intravasation specifically, and not the entire metastatic process. MDAMB-231 breast cancer cells, BE and LS174T colon carcinoma cells, A375m2 melanoma cells, and HT1080 fibrosarcoma cells were gifts from C.J. Marshall (Institute of Cancer Research, London, UK). Stable GFP-expressing clones were selected after transfection of pEGFP-N1 (CLONTECH Laboratories, Inc.). Cells were routinely maintained in DME supplemented with 10% FCS. cDNAs for the human Smurf1-interfering mutants Smurf1-C699A and -ΔC2 were gifts from J. Wrana (University of Toronto, Toronto, Canada). SBE4 reporter construct was a gift from B. Vogelstein (Johns Hopkins Oncology Center, Baltimore, MD). The truncated activated ROCK1 mutant ROCKΔ3 was described elsewhere (). Transfections were carried out using Lipofectamine 2000 (Invitrogen) according to the manufacturer's recommendations. Cell-permeable C3 (Tat-C3) was prepared as previously described (). Y-27632 was obtained from Tocris. Human TGFβ1 (R&D Systems) was a gift from E. van Obberghen-Schilling (Institute of Signaling Developmental Biology and Cancer, Nice, France) and was used at 1 ng/ml. Anisomycin was used at 0.5 μg/ml, and blebbistatin was used at 25 μM, for 60 min. Antibodies used were as follows: Smurf1 (Santa Cruz Biotechnology, Inc.); RhoA (Santa Cruz Biotechnology, Inc.); β-actin (Sigma-Aldrich); MYC (Cell Signaling Technology); phospho-JNK1/2 (Promega); phospho(Ser3)-cofilin (Cell Signaling Technology); ROCK1 (Santa Cruz Biotechnology, Inc.). Antibodies against phospho(Ser19)-MLC2 (Cell Signaling Technology), phospho(Thr18/Ser19)-MLC2 (Cell Signaling Technology), and phospho-LIMK (Cell Signaling Technology) were gifts from M. Olson (Cancer Research UK Beatson Institute for Cancer Research, Glasgow, UK). Secondary antibodies for immunofluorescence were obtained from Promega. The RBD(rhotekin)-GFP probe was a gift from R. Grosse (University of Heidelberg, Heidelberg, Germany) and was used as previously described (). Oligonucleotides were purchased from Dharmacon and Eurogentec. The siRNA sequences targeting human used were as follows: sense, CCGACACUGUGAAAAACACdTdT; antisense, GUGUUUUUCACAGUGUCGGdTdT. The RhoA siRNA sequences used were as follows: sense, GAACUAUGUGGCAGAUAUCUUdTdT; antisense, AAGAUAUCUGCCACAUAGUUCdTdT. The control oligonucleotides targeting the α gene were gifts from E. Berra (Institute of Signaling Developmental Biology and Cancer, Nice, France) and were previously described (). siRNAs were transfected using the Oligofectamine reagent (Invitrogen) according to the manufacturer's recommendations. For construction of the shRNA vectors targeting Smurf1, the targeted sequences in the human gene were 5′-TACGTCCGGTTGTATGTAA-3′ (shRNA-1) and 5′-TGAAGGAACGGTGTATGAA-3′ (shRNA-2). The corresponding oligonucleotides (sequences available upon request) were annealed and cloned into the shRNA-expressing vector pTER (). These vectors and a vector expressing a control shRNA (gift from E. Berra) were transfected into MDAMB-231 cells for transient expression of the shRNAs, or into BE cells for the generation of stable cell lines constitutively expressing the shRNAs. All the shRNA transfections were done using Lipofectamine 2000. Whole-cell extracts were harvested in 1.5× Laemmli sample buffer, and immunoblotting was performed using standard procedures. RhoA pull-down assays were performed using GST-rhotekin, as previously described (); levels of total and active RhoA were revealed using a RhoA-specific antibody (Santa Cruz Biotechnology, Inc.). For immunofluorescence, cells were washed with PBS, fixed with 4% paraformaldehyde in PBS, and permeabilized with 0.2% Triton X-100 in PBS. After several PBS washes, cells were either stained for the actin cytoskeleton with Alexa Fluor 568–phalloidin (Invitrogen) or primary antibodies were added for 2 h; after several PBS washes, appropriate secondary antibodies conjugated to Alexa Fluor 488, 568, or 647 were added for 1 h; cells were mounted after several additional PBS washes and viewed using a microscope (Axiovert 200) with Plan Apochromat 63×/1.4 NA oil or Plan Neofluor 40×/1.3 NA oil objectives, a camera (HAL100), and Immersol medium with N = 1,518 (all from Carl Zeiss MicroImaging, Inc.). For analysis of luciferase activity, transfected cells were lysed in 25 mM Tris-phosphate, pH 7.8, 2 mM DTT, 2 mM 1,2-diaminocyclohexane-,,′,′-tetraacetic acid, 10% glycerol, and 1% Triton X-100. Light emission was quantified using a luminometer (1450 Microbeta; Wallac) after addition of the luciferase substrate (20 mM Tricine, 1.07 mM (MgCO)Mg(OH).5HO, 2.67 mM MgSO, 0.1 mM EDTA, 33.3 mM DTT, 0.27 mM coenzyme A, 0.47 mM luciferin, and 0.53 mM ATP). As a control, β-galactosidase activity was quantified in the lysate using the Galacton Substrate kit (Tropix; Applied Biosystems). The 2D motility of cells was analyzed by recording phase-contrast or GFP fluorescence images with a multichannel time-lapse digital video microscope for several hours (Axiovert 200, 10×/0.25 NA plan Ph/VAR dry objective, HAL100 camera; incubation chamber at 37°C, 5% CO). Cell speed was determined by tracking cells with MetaMorph Software (Universal Imaging Corp.). The 3D morphology was analyzed by embedding tumor cells within a thick Matrigel matrix. After an 18-h incubation, cells were fixed in 4% paraformaldehyde and imaged by capturing at least 30 confocal sections of GFP fluorescence (SPI TCSNT confocal microscope [Leica]; Plan Apochromat 40×/1.3 NA oil objective; immersion oil #11513859 [Leica] with N = 1,518). These were subsequently reconstructed in 3D using Volocity software (Improvision). The 3D invasion assays were performed as previously described (). In brief, cells were allowed to attach to the underside (bottom) of the growth factor–depleted Matrigel-coated polycarbonate chambers (Transwells; 8-μm pore-size filters). The cells were then chemoattracted (10% FCS) across the filter and through the Matrigel above it. Cells were fixed in 4% paraformaldehyde, and GFP fluorescence was analyzed in z sections (one section every 4 μm) from the bottom of the filter, using a confocal microscope. 3D reconstructions of the GFP-expressing cells into the Matrigel were done using Volocity computer software. Cells were scored as amoeboid when the polarity index (long axis/short axis ratio) was less than two, with no apparent cellular protrusions. 10 tumor cells were injected subcutaneously into the flank of young adult immune-compromised mice. Once tumors were established (diam 4–7 mm), intravital imaging of tumor cell morphology and movement was performed on living anesthetized animals using two-photon microscopy, as described elsewhere (). The animal study protocols were conducted according to approved institutional guidelines for animal use. Tumors were fixed in 4% paraformaldehyde after imaging and stained for endomucin (details available on request) to reveal the presence of blood vessels. Fig. S1 shows that Smurf1 does not interfere with SMAD or JNK activity in carcinoma cells. Fig. S2 shows the overall activity of downstream effectors of the Rho–ROCK pathway upon Smurf1 silencing. Fig. S3 shows that Smurf1 silencing does not affect the vessel density in tumors. Video 1 shows a mesenchymal BE tumor cell moving in the tumor by extending a protrusion at the front. Video 2 shows rounded Smurf1-silenced BE tumor cells moving in the tumor with no protrusions, indicating that they are using the amoeboid mode of locomotion. Video 3 shows that mesenchymal BE tumor cells are not highly motile and that no tumor cells are seen in the vessels. Video 4 shows that amoeboid Smurf1-silenced BE tumor cells are highly motile and that many cells are visible in the vessels. Online supplemental material is available at .

Focal adhesions are integrin-mediated cell matrix junctions connecting the ECM to the actin cytoskeleton. The ECM proteins bind to the extracellular domains of integrin heterodimers, whereas the actin stress fibers link to integrin cytoplasmic tails via large molecular complexes. These complexes comprise actin-binding/modulating proteins, protein kinases, phosphatases, GTPases, and adaptor proteins () and are targets of regulatory signals that control focal adhesions' function, including cell adhesion, migration, proliferation, differentiation, and gene expression (; ). Dysregulation of these components is associated with diseases such as cancer (). Tensin is a gene family with four members (tensin1, tensin2, tensin3, and cten), and their encoding proteins are localized to the cytoplasmic side of focal adhesions. Tensin1, the prototype of the family, interacts with actin filaments in multiple ways () and contains an Src homology 2 (SH2) domain that binds to phosphotyrosine-containing proteins (; ), followed by a phosphotyrosine binding (PTB) domain that interacts with the NPXY motif on the β integrin cytoplasmic tails (). Tensin2 and -3 have domain structures that are very similar to those of tensin1, although the central regions are diverse (). On the other hand, cten (C-terminal tensin like) is a distant member of the family with smaller molecular mass, and the only sequence homologous region is the SH2 and PTB domains. The cten gene localizes to chromosome 17q21, a region frequently deleted in prostate cancer (; ; ), and its expression is reduced or absent in prostate cancer (), suggesting a role of cten as a tumor suppressor. However, the potential mechanism has not been well understood. In this study, we have identified deleted in liver cancer 1 (DLC-1) as one of the binding partners of cten, mapped the binding sites on cten and DLC-1, and demonstrated the biological relevance of this interaction. Our results provide new insight into how cten may be involved in preventing tumor formation. To understand cten's biological function and the potential mechanism involved, we set up experiments to identify cten-associated proteins by yeast two-hybrid assay, mass spectrometry analysis, and candidate screenings. One of the molecules identified is DLC-1, which is a tumor suppressor that regulates actin stress fibers and cell adhesion and inhibits tumor cell growth and migration (, ; ; ; ). Its rat homologue, p122RhoGAP (RhoGTPase-activating protein), is isolated as a phospholipase Cδ1–interacting protein () and is localized to caveolae () and focal adhesions (). To demonstrate the relation between DLC-1 and cten, an expression vector encoding GFP or GFP–DLC-1 was transfected into cten-expressing A549 cells and molecules associated with GFP–DLC-1 or GFP were immunoprecipitated with anti-GFP antibodies. The immunoblot analysis indicated that endogenous cten was present in the GFP–DLC-1–associated complexes, but not in the GFP control (). The reciprocal experiment also detected GFP–DLC-1 in the cten immunocomplexes (unpublished data). The interaction was further examined by a luciferase reporter–based mammalian two-hybrid assay. The positive interaction shown by a fourfold enhancement of luciferase activity was detected when DLC-1 and cten were cotransfected into NIH 3T3 cells (). Finally, to test the interaction between endogenous DLC-1 and cten, we screened numerous cell lines and found that MLC-SV40 (immortalized normal prostate epithelialcell line) expressed both cten and a low level of DLC-1. This cell line was used for coimmunoprecipitation assay, and the results demonstrated that cten interacted with endogenous DLC-1 (). To demonstrate the direct interaction and map the regions responsible for the binding, we have applied yeast two-hybrid assay. As expected, the full-length DLC-1 binds to intact cten (). With truncated constructs, the interaction regions were initially mapped to the N-terminal half (1–800) of DLC-1 and the C-terminal region (327–715) of cten, which contains the SH2 and PTB domains. We generated and examined constructs containing only the SH2 or PTB domain. Surprisingly, it was the SH2 domain that interacted with DLC-1. Because all tensin members contain the highly conserved SH2 domains (), we predicted that they were likely to bind to DLC-1 as well. Indeed, DLC-1 interacted with SH2 domains of tensin1, -2, and -3 in the yeast two-hybrid assay (). Furthermore, when the arginine residue at the critical position, βB5, in the SH2 domain of cten was mutated into alanine (R474A), it abolished the interaction. Therefore, we have confirmed that the SH2 domain of cten binds to DLC-1. By a similar approach, we have defined the binding region on DLC-1 (1–535; ). The SH2 domain is known as a binding motif for phosphotyrosine-containing peptides. However, yeast cells contain a very low level of, if any, phosphotyrosine. To test that this SH2–DLC-1 interaction is truly independent of tyrosine phosphorylation and to further map the binding region, because shorter fragments of N-terminal DLC-1 displayed self-activation activity in the yeast two-hybrid system, we performed a pull-down assay using recombinant GST-SH2 and Xpress–DLC-1 fragments expressed in bacteria, which contain no tyrosine kinase at all. The result showed that recombinant SH2 remained bound to DLC-1 (113–535; ). Together with our results using synthetic peptides for binding (see the following paragraph), we have confirmed that the interaction is phosphotyrosine independent. It has been shown that the interaction of the SH2 domain of SAP (also named SH2D1A), the gene product mutated in X-linked lymphoproliferative syndrome, to lymphocyte coreceptor SLAM is independent of tyrosine phosphorylation (). In fact, the SH2 domain of SAP interacts with non–tyrosine-phosphorylated peptides containing S/TIYxxI/V (), and we found that there was one such site, SIYDNV, in DLC-1. Coincidentally, this site resides in the SH2 binding region (113–535). When either S440 or Y442 was mutated (S440A or Y442F), the interaction was abolished in both mammalian and yeast two-hybrid assays ( and ), demonstrating that this is indeed the essential site on DLC-1 for binding to the SH2 domain of cten. Interestingly, although the SAP SH2 domain binds to a similar motif, SAP SH2 domain does not interact with DLC-1 (). We further tested whether DLC-1 might be able to interact with other SH2 domain containing proteins, such as Src and p85, by a pull-down assay using the DLC-1 peptide (CSRLSIYDNVPG)–conjugated beads. As shown in , only cten SH2 domain could interact with the DLC-1 peptide. In addition, cten SH2 domain did not bind to an EGFR peptide (CSVQNPVYHNQP) regardless of whether Y1086 was phosphorylated (). These results demonstrated the binding specificity between the DLC-1 and cten SH2 domain. Furthermore, we tested whether synthetic tyrosine-phosphorylated peptide (CSRLSIpYDNVPG) interacted with the cten SH2 domain and found that phosphorylation on Y442 slightly reduced the interaction (). Because no report had documented the tyrosine phosphorylation of DLC-1 and we did not detect tyrosine phosphorylation of DLC-1 (113–535) when incubated with recombinant Src (unpublished data), the biological relevance of this reduced binding is currently unknown. Because both cten and DLC-1 localize to focal adhesions and a previous study found that p122RhoGAP (117–533; ), corresponding to DLC-1 (125–541), contained the focal adhesion targeting site, which overlapped with the SH2 binding site identified in this study, we speculated that the DLC-1 and cten interaction might be responsible for recruiting DLC-1 to focal adhesions. If this is the case, S440A and Y442F DLC-1 mutants would not be able to localize to focal adhesions. In contrast to the colocalization of cten and GFP–DLC-1 (1–535) at focal adhesion sites, the GFP–DLC-1 (1–535) or GFP–DLC-1 (1–535) was diffusely distributed in the cytoplasm (), indicating that the SH2 binding site is essential for DLC-1's focal adhesion localization. The protein expressions of these constructs were confirmed by immunoblotting (). To further demonstrate that the cten SH2 domain is crucial for recruiting DLC-1 to a subcellular compartment, we generated a DsRed-cten SH2-SKL construct so that the DsRed-cten SH2 domain would be fused with the peroxisomal targeting peptide, SKL (), at the C terminus when expressed and be targeted to peroxisomes. As shown in , although the DsRed was distributed in the cytoplasm, the DsRed-cten SH2-SKL proteins were accumulated at peroxisomes. When these constructs were cotransfected with GFP–DLC-1 (1–535), the DsRed-cten SH2-SKL was able to recruit some GFP–DLC-1 (1–535) to peroxisomes, demonstrating that cten SH2 alone is sufficient for the interaction and recruitment of DLC-1. DLC-1 was identified as a candidate tumor suppressor, and its expression was lost or down-regulated in various cancers, including liver, breast, lung, brain, stomach, colon, and prostate, because of either genomic deletion or aberrant DNA methylation (, ; ; ; ; ; ). It has been reported that reexpression of DLC-1 in liver, breast, and lung cancer cell lines inhibits cancer cell growth (, ; ; ; ), supporting its role as a tumor suppressor. DLC-1 contains three conserved domains: the sterile α motif (SAM), RhoGAP, and steroidogenic acute regulatory-related lipid transfer (START) domains (). SAM domains have been implicated in protein–protein interactions and are highly versatile in their binding partners. Some SAM domains may bind to each other to form homodimers or polymers, whereas others can interact with other proteins, or even RNA and DNA (). START domains are predicted to contain a binding pocket for lipids, and modifications in the pocket may determine ligand binding specificity and function (). RhoGAP domains convert the active GTP-bound Rho proteins to the inactive GDP-bound state and function as negative regulators of RhoGTPases, which are involved in actin cytoskeleton organization, focal adhesion assembly, and cell proliferation (), and dysregulation of Rho activity has been implicated in tumorigenesis (). A recent study demonstrated that the RhoGAP and START domains of DLC-1 are required for its tumor suppression activity (). However, these two domains are not sufficient because expression of the RhoGAP and START domains alone does not inhibit tumor cell growth (). In fact, the shortest fragment with the suppression activity contains, in addition to the RhoGAP and START domains, a region overlapping with the SH2 binding site, which is critical for focal adhesion localization. We hypothesized that the appropriate focal adhesion localization is essential for DLC-1's functions, including tumor cell suppression activity. To test this, we performed the colony formation assay using MDA-MB-468 breast cancer cell line, in which the growth was suppressed by DLC-1 overexpression (). Consistent with a previous report, wild-type DLC-1 was able to suppress MDA-MB-468 cell growth. However, neither GFP–DLC-1 nor GFP–DLC-1 could inhibit MDA-MB-468 cell growth (). In agreement with these results, the growth curve of MDA-MB-468 cells was significantly slower with wild-type DLC-1 (). Thus, the SH2 binding site is not only essential for DLC-1's focal adhesion localization but also critical for its tumor suppression activity. To further address the importance of the focal adhesion localization of DLC-1 to its tumor suppression activity, we fused wild-type and mutant DLC-1 with the N-terminal focal adhesion binding (FAB) site (aa 65–360) of chicken tensin1 (), which is not conserved in cten. This FAB fusion forced the focal adhesion localization of GFP–FAB–DLC-1, GFP–FAB–DLC-1, and GFP–FAB–DLC-1 (). From the colony formation assay, the constitutive focal adhesion localizations of these molecules significantly enhanced the suppression activities of DLC-1 and DLC-1 mutants (). Therefore, the focal adhesion localization of DLC-1 is essential for its tumor suppression activity. Nonetheless, the suppression activities of DLC-1 and DLC-1 were not fully restored by linking to FAB. It is possible that the cten–DLC-1 interaction is not just for recruiting DLC-1 to focal adhesions but also for regulating its activity. In addition, because we fused DLC-1 with the N-terminal FAB site of tensin1 and DLC-1 normally binds to the C-terminal SH2 domains of tensins, these FAB fusion mutant proteins were not targeted to the precise position within the focal adhesion complexes. This spatial discrepancy may also contribute to the weaker suppression activities observed in GFP–FAB–DLC-1 and GFP–FAB–DLC-1. In this study, we have demonstrated that the tumor suppressor DLC-1 interacts with the SH2 domains of cten and other tensins as well. Although the SH2 binding site on DLC-1 also contains a critical tyrosine residue (Y442), the interaction does not rely on the phosphorylation of Y442. However, the phosphorylation of Y442 does reduce the interaction. This is a novel binding feature of tensins' SH2 domains. Furthermore, this interaction is highly specific for the SH2 domain of tensin family, as the SH2 domains of SAP, Src, and p85 all fail to bind to DLC-1. The biological significance of the cten–DLC-1 interaction is illustrated by mislocalization and the loss of tumor suppression activities of DLC-1 and DLC-1 mutants. Furthermore, the suppression activities of these mutants could be rescued by tagging with FAB sequence. Therefore, in addition to genomic deletion and promoter hypermethylation, mislocalization of the DLC-1 protein may be another mechanism for acquiring tumorigenicity involving DLC-1 dysregulation. In this regard, further investigations on the DLC-1 protein localization in cancer samples with “normal” DLC-1 expression level are highly warranted. Based on these findings, we propose that DLC-1 is recruited to focal adhesion sites by one or more tensin members, depending on cell types and tissues. At the focal adhesion site, the RhoGAP domain of DLC-1 negatively regulates Rho small GTPase, which organizes actin stress fibers, and focal adhesion turnover, in turn, mediates cell migration and proliferation. When the expression and/or localization of DLC-1 are compromised, the cells are more susceptible for transformation. The fact that DLC-1 is able to bind to all tensins through their SH2 domains may explain why DLC-1 relies on tensin members for its normal localization and function, yet DLC-1–knockout mice () displayed a more severe phenotype than tensin1 or -3 single-knockout mice (; ). It may require double or even triple tensin knockout to observe the defect results from mislocalization of DLC-1. On the other hand, recruiting DLC-1 to focal adhesion sites may not be the only function for cten. It is known that activated caspase3 cleaves cten at the DSTDS, site generating two cten fragments: 1–570 and 571–715 (). The later contains the PTB domain alone, which by itself is able to reduce cell growth by inducing apoptosis (). In this case, the loss of cten expression may lead to uncontrolled cell growth and result in cell transformation. Together with our current findings, cten may function as a tumor suppressor in multiple ways. The full-length coding sequence of the DLC-1 gene was amplified from human kidney cDNA. The full-length and truncated fragments of DLC-1 were subcloned in frame into mammalian expression vector pEGFP-C2 (CLONTECH Laboratories, Inc.), yeast expression vector pGBKT7 (CLONTECH Laboratories, Inc.), and mammalian two-hybrid vector pCMV-BD (Stratagene). The full-length and truncated fragments of cten were constructed into pGADT7 (CLONTECH Laboratories, Inc.) and pCMV-AD (Stratagene) for yeast and mammalian two-hybrid analyses, respectively. The cDNA encoding the SH2 domain of SAP was amplified from human thymus cDNA and subcloned into pGADT7. The corresponding coding regions of the SH2 domains of tensin1, tensin2, tensin3, and cten were subcloned into pGADT7 and bacterial expression vector pGEX-5X-1 (GE Healthcare). The region encoding DLC-1 residues 113–535 was inserted into pTrcHis containing His and Xpress tags (Invitrogen). The mutations in DLC-1 (S440A and Y442F) and cten SH2 domain (R474A) were generated by site-directed mutagenesis. The corresponding coding regions of the SH2 domains of Src and p85 were subcloned into pGEX-5X-1. To construct pDsRed1-cten SH2-SKL, SKL residues were introduced into the C terminus of cten SH2 fragment by PCR. The resulting amplified PCR products were then ligated into pDsRed1-C1 (CLONTECH Laboratories, Inc.). The N-terminal FAB site of chicken tensin1 (residues 65–360; ) was used to construct fusions to the N terminus of DLC-1 in pEGFP–DLC-1, pEGFP–DLC-1, or GFP–DLC-1 plasmids. All constructs were verified by DNA sequencing. MLC-SV40 cells, a gift from J. Rhim (Uniform Services University, Bethesda, MD), were cultured in keratinocyte serum-free medium supplemented with antibiotics, 5 ng/ml human recombinant EGF, and 0.05 mg/ml bovine pituitary extract (Invitrogen). A549, NIH3T3, and MDA-MB-468 cells purchased from American Type Culture Collection were cultured in DME supplemented with antibiotics and 10% fetal bovine serum. A549 cells were transfected using Lipofectamine 2000 (Invitrogen), whereas NIH 3T3 and MDA-MB-468 cells were transfected using SuperFect transfection reagent (QIAGEN) according to the manufacturer's instructions. Transiently transfected A549 or MDA-MB-468 cells with GFP fusion constructs were lysed in immunoprecipitation buffer (1% Triton X-100, 50 mM Tris-HCl, pH 8.0, 150 mM NaCl, 5 mM EDTA, 10 μg/ml aprotinin, 10 μg/ml leupeptin, 1 μg/ml pepstatine, and 1 μM PMSF) and cleared by centrifugation at 14,000 for 15 min at 4°C. The clarified cell lysates were incubated with 1 μg of an anti-GFP goat polyclonal antibody (Rockland) by rotating at 4°C for 1 h, followed by the addition of 30 μl of 50% of protein A–Sepharose slurry (GE Healthcare) for 1 h. The protein A beads were collected by centrifugation and washed with immunoprecipitation buffer. Samples were then boiled in protein loading buffer and subjected to immunoblotting analyses using anti-GFP rabbit polyclonal antibodies (Santa Cruz Biotechnology, Inc.). For coimmunoprecipitation, 24 h after transfection, A549 cells expressing the GFP or the GFP–DLC-1 constructs were lysed in coimmunoprecipitation buffer (0.1% Triton X-100, 25 mM Tris-HCl, pH 8.0, 50 mM NaCl, 0.2 mM EDTA, 10 μg/ml aprotinin, 10 μg/ml leupeptin, 1 μg/ml pepstatine, and 1 μM PMSF). Cell lysates were then sheared by passing through a syringe needle, and the cell debris was removed by centrifugation at 14,000 for 15 min at 4°C. 1.2 mg of the clarified cell lysates were incubated with 2 μg of an anti-GFP goat polyclonal antibody (Rockland) by rotating at 4°C for 4 h, followed by the addition of protein A–Sepharose. Samples were subjected to immunoblotting analyses using anti-cten () and anti-GFP rabbit polyclonal antibodies (Santa Cruz Biotechnology, Inc.). Plasmids (1 μg of each mammalian two-hybrid construct and 0.5 μg of pFR-Luc reporter) were transfected into NIH 3T3 cells using SuperFect. Cells were harvested 24 h after transfection. Firefly luciferase and activities in the cell extracts were determined by the procedure using Luciferase Assay System (Promega) and measured by luminometry. To assay the interaction between DLC-1 and cten, the strain AH109 was transformed with combinations of cten fragments in the activation domain (AD) plasmid, pGADT7, together with each of the DLC-1 fragments in the DNA-binding domain (DNA-BD) plasmid, pGBKT7. In brief, 5 ml of overnight culture of AH109 in YPD medium was diluted 50-fold and allowed to grow for another 4 h at 30°C. The yeast cells were harvested by centrifugation at 2,000 for 15 min at room temperature and washed twice with 25 ml of sterile water. The cells were resuspended in 0.5 ml of 0.1 M lithium acetate (LiAc). 100 μl of competent cells were mixed with 600 μl TE-LiAc-PEG (1× TE, 0.1 M LiAc, and 40% polyethylene glycol [mol wt 3,350]), 10 μl of salmon sperm DNA, and 1 μg of each plasmid. After incubation at 30°C for 30 min, 70 μl of DMSO was added to the cells and heat shocked at 42°C for 15 min. The transformation mixture was centrifuged and washed with 1 ml of sterile water. The cell pellets were subsequently resuspended in 1× TE buffer and plated on nutritional selection agar lacking leucine and tryptophan. The resulting colonies were then restreaked on quadruple dropout plates lacking Ade, His, Leu, and Trp. For GST pull-down assay, the cDNAs encoding the SH2 domain of cten were subcloned into pGEX-5X-1 to generate GST fusion proteins (GST-cten SH2). The corresponding coding region of 113–535 amino acids of human DLC-1 was ligated into pTrcHis to express an Xpress-tagged protein, Xpress–DLC-1 (113–535). GST-cten SH2 proteins were expressed in and purified from using glutathione-agarose (Sigma-Aldrich). 20 μg of GST or GST-cten SH2 on glutathione-agarose beads was mixed with 2 mg of bacterial lysates expressing Xpress–DLC-1 (113–535) in extraction buffer (0.1% Triton X-100, 50 mM Tris-HCl, pH 8.0, 150 mM NaCl, and 0.2 mM EDTA). After incubation on a rotator for 3 h at 4°C, the slurry was pelleted by centrifugation and washed five times with ice-cold extraction buffer. The pellet was resuspended in protein loading buffer and subjected to immunoblotting analyses using an anti-Xpress mouse monoclonal antibody (Invitrogen). A549 cells grown on glass coverslips were transfected and incubated at 37°C in 5% CO for 10–16 h before microscopic imaging. Cells were fixed with methanol at −20°C. After rinsing with PBS, cells were incubated with 1:25 anti-cten rabbit polyclonal antibody for 2 h. Samples were then incubated with 1:800 Alexa Fluor 594–conjugated secondary antibody (Invitrogen) for 1 h and visualized with a confocal microscope (LSM 510; Carl Zeiss MicroImaging, Inc.). MDA-MB-468 cells were seeded in a 6-well plate at 2 × 10 cells per well. 16 h later, 2 μg of pEGFPC2 vector or various DLC-1 constructs (pEGFP–DLC-1, pEGFP–DLC-1, pEGFP–DLC-1, pEGFP–FAB–DLC-1, pEGFP–FAB–DLC-1, or pEGFP–FAB–DLC-1) were transfected into the cells. After 48 h, the cells were seeded in a 6-well plate at a density of 10 cells per well and selected by 0.8 mg/ml G418 (Geneticin; Invitrogen) for 2 wk. Colony formation efficiency was determined by counting the G418-resistant colonies stained with crystal violet solution (0.25% crystal violet and 3.7% formaldehyde in 80% methanol). MDA-MB-468 cells were seeded in a 6-well plate at 2 × 10 cells per well. 16 h later, 2 μg of pEGFPC2 vector or various DLC-1 constructs (pEGFP–DLC-1, pEGFP–DLC-1, or pEGFP–DLC-1) were transfected into the cells. After 48 h, the cells were seeded in 100-mm dishes and selected by 0.8 mg/ml G418 for 6 d. Cells were then collected and seeded in triplicate in 60-mm dishes at 2 × 10 cell density. Cells were harvested at 24-h intervals for 4 d, and the numbers of viable cells were counted by trypan blue exclusion assay with a hematocytometer.

Long-distance intracellular transport is driven by kinesin and dynein motor proteins that ferry cargoes along microtubule tracks. A large body of work has revealed the chemomechanical mechanisms of motor proteins (; ). However, key aspects of motor-driven transport, such as cargo loading/unloading, regulation of motor activity, and coordination of bidirectional movement, remain unknown. It has been recognized for some time that motor protein activity must be tightly regulated in cells to prevent the futile hydrolysis of ATP. Current models suggest that only a fraction of Kinesin-1 inside cells is carrying cargo at any time (; ). How is Kinesin-1 kept inactive in the absence of cargo? In most species, Kinesin-1 is a heterotetramer of two kinesin heavy chain (KHC) and two kinesin light chain (KLC) polypeptides. Purified Kinesin-1 can exist in two conformations in vitro, depending on ionic concentration, that correlate with the activity of Kinesin-1–folded molecules are thought to be inactive for microtubule-based motility, whereas extended molecules are thought to be active. The folded conformation enables an autoinhibitory interaction of the KHC C-terminal tail with the KHC N-terminal motor (for reviews see ; ). The actin-based motor myosin Va undergoes a similar conformational change that correlates with activity (; ). Thus, autoinhibition may be a general mechanism for the regulation of cytoskeletal motor proteins. Several lines of evidence suggest that regulation of Kinesin-1 in vivo may be more complex. For example, the motor activity of native, purified Kinesin-1 is typically 10 times less than that of recombinant protein (; ). In addition, the intramolecular association between the N- and C-terminal domains of Kinesin-1 comprises low-affinity interactions in vitro (). Thus, the mechanisms of Kinesin-1 autoinhibition and activation in vivo remain unclear. To build a structural model of the spatial and molecular relationships within Kinesin-1 in intact cells, we used a quantitative fluorescence resonance energy transfer (FRET) approach (). We show that FRET stoichiometry (), which is a method for measuring both FRET efficiency and the fraction of interacting fluorescent protein (FP) molecules, reveals conformational states of Kinesin-1 in living cells. Specifically, we show that two conformational changes occur when Kinesin-1 is activated for interaction with microtubule tracks. First, we show for the first time in intact cells that inactive Kinesin-1 is folded in an autoinhibitory conformation, whereas active Kinesin-1 molecules are in a more extended conformation. Second, we describe a novel conformational change in which the two motor domains are pushed apart in the inactive molecule and brought closer together for productive interactions with the microtubules. This novel local conformational change contrasts with what was predicted based on the crystal structure of dimeric Kinesin-1 motors (). To analyze the structure of Kinesin-1 in living cells by FRET, donor (monomeric ECFP [mECFP]) and acceptor (monomeric Citrine [mCit]) FPs were fused to the N and/or C termini of both KHC and KLC (). COS cells were chosen for their flat morphology and because their low levels of endogenous Kinesin-1 are unlikely to interfere with formation of donor–acceptor FP complexes (Fig. S1 A, available at ; unpublished data). Only cells expressing low-to-medium levels of FP proteins were chosen for data analysis to avoid artifacts caused by protein aggregation and ATP-independent microtubule interactions (Fig. S1 B). FRET stoichiometry (which is discussed in the following two paragraphs) and coimmunoprecipitation (Fig. S1 A) experiments verified that FP fusions to KHC and KLC did not alter their interactions. FP-KHC and -KLC expressed in COS cells demonstrated similar localization patterns to those described previously for other tagged Kinesin-1 motors (Fig. S1 B). Because steady-state fluorescence patterns do not indicate the activity of kinesin motors, we developed an assay to delineate between active and inactive motors in vivo. To do this, we took advantage of the ability of the nonhydrolyzable ATP analogue AMPPNP to block the release of active kinesin motors from microtubules (). Live cells were transiently permeabilized with low levels of the bacterial toxin streptolysin O (SLO), and active FP-Kinesin-1 motors were trapped on microtubules by the addition of AMPPNP. FP-Kinesin-1 (e.g., mCit-KHC + HA-KLC; , column 1) did not become trapped on microtubules, but remained diffuse and cytosolic upon addition of AMPPNP, indicating that the Kinesin-1 holoenzyme is in an inactive state in vivo. In contrast, in cells expressing FP-KHC alone (e.g., mCit-KHC; , column 2), the motor rapidly accumulated on microtubules after exposure to AMPPNP, indicating that the KHC subunit exists in an active state in vivo. Several lines of evidence verify that FP-KHC alone is capable of ATP-dependent microtubule motility, and thus represents the Kinesin-1 active state. First, single molecule motility assays demonstrate that FP-KHC molecules are capable of microtubule-based motility in vitro (Fig. S1 C). Second, removal of the cryptic ATP-independent microtubule-binding site in the KHC tail (KHC[1–891]) resulted in a KHC molecule that retained ATP-dependent microtubule binding (, mCit-KHC[1–891]). Third, this microtubule localization was caused by direct interaction between the KHC motor domain and the microtubules because FRET between mCit-KHC(1–891) and mECFP-tubulin increased after addition of AMPPNP (Fig. S4). Fourth, mutation of the microtubule-binding site in the KHC motor domain (Δloop12 mutation; ) abolished the ability of FP-KHC to be locked in a microtubule- bound state after addition of AMPPNP (, mCit-KHC[1–891]/Δloop12). Collectively, these results indicate that KHC homodimers are active for microtubule binding and motility, whereas the complete Kinesin-1 holoenzyme (KHC + KLC) remains inactive and predominantly in the cytosol. In addition, these results validate the use of fluorophore-tagged subunits to study Kinesin-1 structure and function in vivo. For FRET stoichiometry of Kinesin-1, various combinations of KHC and KLC FRET pairs were cotransfected into COS cells, and 24 h later the data were collected on a wide-field fluorescence microscope calibrated for FRET stoichiometry. FRET stoichiometry uses three fluorescence images from a calibrated microscope to calculate three parameters that describe each pixel (Fig. S2, available at ; ; ): (a) R, the mole ratio of acceptor- to donor-labeled proteins, (b) , the apparent acceptor FRET efficiency (FRET efficiency × fraction of acceptor molecules in complex), and (c) , the apparent donor FRET efficiency (FRET efficiency × fraction of donor molecules in complex). and range between 0 and 100%, where 100% indicates all acceptor and donor molecules in the FRET complex and with complete energy transfer. Because protein expression levels influence the fraction of donor or acceptor molecules in FRET complex for nonlinked molecules, we analyzed cells with R close to 1.0 and we calculated an average FRET efficiency, = ( + )/2, which is less sensitive to expression ratio (). ≈ 37% (). should reflect structural changes in the Kinesin-1 molecule. Modeling the spatial arrangements between the FP and the KHC motor domain based on crystal structures supports this assumption, as the short linker sequences (4 or 5 aa) limit the flexibility of the FP (Fig. S1 D). To verify that FRET stoichiometry can detect conformational changes in Kinesin-1 in live cells, we obtained FRET efficiencies under ion concentrations known to induce Kinesin-1 conformational changes in vitro (). To monitor Kinesin-1 motor-to-tail FRET, FRET pairs were placed on the N and C termini of the same KHC polypeptide (mCit-KHC-mECFP; ). Coexpression with Myc-KHC was required to prevent aggregation of the four FPs in the KHC homodimer (Fig. S3, available at ). COS cells expressing mCit-KHC-mECFP + Myc-KHC + HA-KLC () were transiently permeabilized with SLO under physiological salt conditions (I ≈ 0.15). After 5 min, the cells were exchanged into high ionic strength buffer (I ≈ 0.8). High motor-to-tail FRET efficiencies were observed before permeabilization ( = 11.5 ± 1.9%; , ), indicating a close association of the KHC motor and tail regions. FRET remained high during permeabilization at physiological ionic strength (, B [ ] and C); however, high ionic strength buffer resulted in a rapid and significant decrease in FRET efficiency ( = 4.0 ± 1.1%; , B [ ] and C). COS cells expressing the mCit-16aa-mECFP calibration molecule exposed to the same conditions showed no significant change in FRET efficiency ( = 37.8 ± 2.1% at physiological ionic strength and 37.4 ± 1.8% at high ionic strength; , C and D [ ]). Note that R remained constant in both cases, indicating negligible differences in acceptor and donor photobleaching. These results indicate that Kinesin-1 is folded (high motor-to-tail FRET) at physiological ionic strength, but is more extended (low motor-to-tail FRET) under high ionic strength conditions. Thus, FRET stoichiometry can detect conformational changes in Kinesin-1 in living cells. To probe the overall structure of inactive Kinesin-1 in vivo, FP-labeled KHC and KLC were coexpressed in COS cells. We first measured FRET efficiencies for FRET pairs located on the KHC subunit (, ). For the KHC motor-to-tail relationship, higher FRET efficiencies were obtained for FRET pairs on the same KHC polypeptide ( = 12.4 ± 1.0% for mCit-KHC-mECFP + Myc-KHC + HA-KLC; , ) than for FRET pairs on separate KHC polypeptides ( = 4.8 ± 0.5% for mCit-KHC + KHC-mECFP + HA-KLC; , ). Although these data cannot distinguish the relationship between each motor and its tail domain because of differences in fraction of FP protein in complex and orientation of the FPs, these FRET efficiencies demonstrate that inactive Kinesin-1 molecules are in a folded conformation in vivo. For KHC motor-to-motor measurements, the FRET efficiency was low ( = 2.4 ± 0.5%; , ), suggesting that the KHC N-terminal motor domains are separated in the inactive molecule. In contrast, for KHC tail-to-tail measurements, the FRET efficiency was higher ( = 8.6 ± 0.9%; , ) indicating that the KHC C-terminal tail domains are relatively close together in vivo. We next measured FRET efficiencies within inactive Kinesin-1 molecules for FRET pairs located on the KLC subunit (, ). Little to no FRET was detected between the N and C termini of KLC ( = 0.2 ± 0.1%; , ), indicating that the KLC subunit is in an extended conformation. Low FRET efficiencies obtained for the C termini of KLC indicate that these regions are separated ( = 2.3 ± 0.4%; , ), whereas the higher FRET efficiencies obtained for the N termini of KLC indicate that these regions are in close proximity ( = 11.2 ± 1.7%; , ), presumably because of dimerization via the heptad repeats. Finally, we measured FRET efficiencies within inactive Kinesin-1 molecules for FRET pairs located on both the KHC and KLC subunits (, ). Moderate FRET efficiencies between the C terminus of KLC and either the N terminus of KHC ( = 5.8 ± 0.5%; , ) or the C terminus of KHC ( = 6.4 ± 0.4%; , 9) suggest that the KLC C terminus is in close proximity to both the KHC motor and tail domains. In contrast, negligible FRET efficiencies were observed between the N terminus of KLC and either the N terminus of KHC ( = 0.4 ± 0.1%; , 10) or the C terminus of KHC ( = 0.6 ± 0.3%; , 11). This suggests that the N terminus of the KLC subunit is close to the region in the KHC stalk that allows folding. These data also indicate that the KLC subunits lie in a direction parallel to the KHC subunits (N′ to N′ and C′ to C′; ). Collectively, these results support the overall structure of Kinesin-1 gleaned from various in vitro experiments () and demonstrate that inactive Kinesin-1 molecules are in a folded conformation in intact cells. To probe the structure of active Kinesin-1 in vivo, we measured FRET efficiencies from combinations of FP-KHCs expressed in COS cells (). FRET efficiencies from KHC molecules accumulated at the cell periphery in highly expressed cells (, ) were very high ( > 20%), regardless of FP position, and correlated with fluorescence intensities (Fig. S2 C), indicating that intermolecular FRET occurs between crowded KHC molecules accumulated at the plus ends of the microtubules. FRET efficiencies for FP-KHC molecules localized in the rest of the cell (, ) remained constant despite variations in fluorescence intensity (Fig. S2 C), suggesting that these FRET measurements represent only intramolecular FRET. Thus, we only collect data from these regions or from cells with low-to-medium expressions to avoid artifacts caused by KHC accumulation. For KHC motor-to-tail measurements, moderate FRET efficiencies were obtained for FRET pairs on the same KHC polypeptide ( = 7.9 ± 1.5%; , ) and on separate KHC polypeptides ( = 4.7 ± 1.0%; , ), indicating that the motor and tail domains of KHC remain in relatively close proximity upon activation. Moderate FRET efficiencies were also obtained for KHC motor-to-motor FRET pairs ( = 6.1 ± 1.2%; , ) indicating that the two motor domains are in close proximity, as expected for active Kinesin-1. FRET efficiencies obtained for KHC tail-to-tail FRET pairs ( = 8.3 ± 2.8%; , ) indicate that the KHC tails are also in close proximity. To compare the structure of KHC motor domains engaged with microtubules with those in cytosol, FRET efficiencies were measured for KHC molecules forced on or off the microtubules. FP-KHC(1–891) was forced to remain on the microtubule by addition of AMPPNP (, 9) or was prevented from binding to microtubules by mutation of the microtubule-binding site in the motor domain (Δloop12 mutation; , 10; ). Similar motor-to-motor FRET efficiencies were obtained for microtubule-bound and unbound motors ( = 6.7 ± 1.6% and 6.4 ± 0.6%, respectively). These results indicate that KHC motor domains in active molecules likely stay in close proximity regardless of whether they are on or off the microtubules. To identify conformational changes within Kinesin-1 upon activation, we compared the FRET efficiencies of inactive (KHC + KLC; ) and active (KHC alone; ) molecules. KHC motor-to-tail FRET pairs on the same KHC polypeptide had higher FRET efficiency in the presence ( = 12.4 ± 0.1%) than in the absence ( = 7.9 ± 1.5%) of KLC. This difference is statistically significant (P < 0.001; ) and indicates a smaller distance between the KHC motor and tail domains in the inactive state. This global conformational change (, green arrows) likely displaces the KHC tail from the KHC motor domains for Kinesin-1 activation. For KHC motor-to-motor FRET pairs, a lower FRET efficiency was observed in the presence ( = 2.4 ± 0.5%) than the absence ( = 6.1 ± 1.2%) of KLC. This difference is statistically significant (P < 0.001; ) and indicates a larger distance between the two KHC motor domains in the inactive state. That the two KHC motor domains are pushed apart in the inactive holoenzyme was surprising because crystallography and 3D cryoelectron microscopy suggested that the motor domains of truncated KHC molecules are closer together when free in solution than when engaged with a microtubule (). A local conformational change (, blue arrows) upon activation is, thus, likely required to position the motor domains for processive motility. To confirm that the two KHC motor domains are pushed apart in the inactive state, we tested biochemically whether the KHC neck coiled-coil segments are closer together in the active state (absence of KLC) than in the inactive state (presence of KLC). A Cys residue was introduced into the neck coiled coil of a Cys-lite version of KHC (KHC[Cys344]; ) at a position accessible to cross-linker, but demonstrated to have no effect on the motile properties of the truncated KHC (). When COS cell lysates expressing KHC(Cys344) in the absence of KLC (i.e., active Kinesin-1) were treated with the cross-linker 3-carboxy-4-nitrophenyl disulfide 6,6′-dinitro-3,3′-dithiodibenzoic acid Bis(3-carboxy-4-nitrophenyl) disulfide (DTNB), nearly all of the KHC(Cys344) was rapidly cross-linked as indicated by a shift to a slower mobility form (, lanes 2–5). In contrast, in the presence of KLC (i.e., inactive Kinesin-1), little to no cross-linking of KHC(Cys344) was observed (, lanes 7–9). Incubation in the presence of DTNB for long periods of time resulted in cross-linking of KHC(Cys344) + KLC (, lane 10), presumably caused by “breathing” of the Kinesin-1 holoenzyme. These results confirm that the KHC neck coiled coil is more separated in the inactive state than in the active state. The KHC globular tail domain has been implicated in contributing to both the folded conformation and autoinhibition of KHC in vitro (; ; ). In particular, a conserved stretch of residues in the KHC tail domain (the IAK region) is critical for autoinhibition of motor activity in vitro (). To determine whether the KHC tail and/or the IAK region play a role in autoinhibition or conformational changes in the Kinesin-1 holoenzyme in vivo, we expressed truncated (KHC[1–891]) and mutated (KHC[ΔIAK]) versions of KHC in COS cells. KHC(1–891) + KLC did not localize to microtubules or accumulate at the cell periphery at steady state, but became locked on microtubules upon exposure of permeabilized cells to AMPPNP (, A [left] and B). The microtubule-bound state of KHC(1–891) + KLC reflects a direct interaction between the motor domain of KHC(1–891) and the microtubule because FRET efficiency between the mCit-KHC(1–891) motor domain and mECFP-tubulin significantly increased upon AMPPNP addition (Fig. S4, available at ). Like the tail-truncated molecules, Myc-KHC(ΔIAK) + mCit-KLC molecules were capable of microtubule binding (locked on microtubules with AMPPNP; Fig. S5, D and E), but not processive motility (did not accumulate at ends of microtubules; ). These results indicate that removal of the KHC tail, or mutation of the IAK region, results in a Kinesin-1 holoenzyme that is active for microtubule binding, in contrast to KHC + KLC (, A [right] and B). Thus, the IAK segment of the KHC tail plays an important role in autoinhibition in vivo, specifically in preventing the microtubule association of Kinesin-1. To test whether activation of Kinesin-1 by mutation of the IAK segment results in a global conformational change in Kinesin-1, we measured motor-to-tail FRET for FRET pairs on the same KHC(ΔIAK) polypeptide (). High FRET efficiencies were obtained for KHC(ΔIAK) + KLC molecules in the absence of AMPPNP ( = 14.4 ± 1.8%, ), which is similar to wild-type Kinesin-1 (KHC + KLC) molecules with the same labeling ( = 12.4 ± 1.0%; , ), and no change (P > 0.5) was detected upon addition of AMPPNP ( = 14.5 ± 2.2%; ). Thus, despite a statistically significant difference (0.001< P <0.01) in the Relocation Index of KHC(ΔIAK) + KLC (, red), no difference in the motor-to-tail spatial relationship (, black) was detected (0.1< P <0.5), even after 30 min of exposure to AMPPNP. We next looked for a local conformational change in active KHC(ΔIAK) + KLC molecules by measuring motor-to-motor FRET (). Low FRET efficiencies were obtained in the absence of AMPPNP ( = 2.5 ± 1.5%; ), similar to the values obtained for wild-type Kinesin-1 ( = 2.4 ± 0.5%; , ), and no change (P > 0.5) was detected after 10 min of AMPPNP exposure ( = 3.1 ± 0.1%; ). Interestingly, if left in the presence of AMPPNP for 30 min, KHC(ΔIAK) + KLC molecules showed a statistically significant (P < 0.001) increase in motor-to-motor FRET ( = 5.7 ± 0.7%; , black). This may reflect the ability of Kinesin-1 motors to exist in single- and double-headed binding states in the presence of AMPPNP, with the double-headed state predominating at low load in vitro () and after prolonged incubation in vivo. The KLC subunit contributes to both the folded conformation of Kinesin-1 and to the separation of the KHC motor domains (). To determine the regions of KLC that contribute to autoinhibition in vivo, we used a truncated version of KLC (KLC[1–176]) that lacks the tetratricopeptide repeat (TPR) motifs required for cargo binding, but retains the heptad repeats required for association with KHC (). FP-KHC + FP-KLC(1–176) localized to the cytosol at steady-state and after exposure of cells to AMPPNP, similar to wild-type Kinesin-1 (Fig. S5, D and E), indicating that the heptad repeat region of KLC is sufficient for autoinhibition. This is likely caused by the ability of the KLC heptad repeats to maintain the folded conformation (, top), as no statistically significant difference was seen (0.1 < P < 0.5) in the motor-to-tail FRET efficiency of mCit-KHC-mECFP + KLC(1–176) ( = 12.1 ± 0.7%; , ) when compared with that of wild-type Kinesin-1 (mCit-KHC-mECFP + KLC, = 12.4 ± 1.0%; , ). We next looked for a local conformational change in KHC + KLC(1–176) molecules by measuring KHC motor-to-motor FRET (, ). Significantly higher FRET efficiencies (P < 0.001) were obtained for FP-KHC + KLC(1–176) ( = 5.3 ± 0.5%; , ) than for wild-type Kinesin-1 ( = 2.4 ± 0.5%; , ), indicating that the KHC motor domains are closer together when the KLC TPR motifs are removed (, bottom). To test whether the KHC tail domains also play a role in separating the motor domains, we compared the motor-to-motor relationships of molecules containing truncations of KLC, KHC, or both. KHC motors that are incapable of binding to microtubules (KHC[1–891]/Δloop 12 mutant) were used to eliminate potential effects that microtubule binding may have on motor-to-motor distances. Truncation of the KHC tail domain (KHC[1–891]/Δloop12 + KLC) caused no significant change (P > 0.5) in motor-to-motor FRET when compared with wild-type Kinesin-1 molecules (KHC + KLC; = 2.3 ± 0.8%; , , vs. = 2.4 ± 0.5%; , , respectively) and a small increase (0.01 < P < 0.02) in motor-to-motor FRET when compared with KLC-truncated Kinesin-1 molecules ( = 6.2 ± 1.9%; , vs. = 5.3 ± 0.5%; , ). In contrast, truncation of KLC caused a significant change (P < 0.001) in motor-to-motor FRET when compared with either wild-type Kinesin-1 molecules ( = 5.3 ± 0.5%; , vs. = 2.4 ± 0.5%; , , respectively) or to KHC-truncated Kinesin-1 molecules ( = 6.2 ± 1.9%; , vs. = 2.3 ± 0.8%; , , respectively). These results indicate that the major contribution for separation of the KHC motor domains in the inactive conformation is provided by the KLC TPR motifs. o w l e d g e o f t h e s t r u c t u r a l o r g a n i z a t i o n o f K i n e s i n - 1 i s e s s e n t i a l t o u n d e r s t a n d i n g h o w i t i n t e r a c t s w i t h c a r g o c o m p o n e n t s , h o w i t g e n e r a t e s m o t i l e f o r c e , a n d h o w m o t i l i t y i s c o n t r o l l e d . I n t h i s r e p o r t , w e d e m o n s t r a t e t h e f e a s i b i l i t y o f u s i n g F R E T s t o i c h i o m e t r y t o d e t e r m i n e t h e d o m a i n o r i e n t a t i o n s a n d s p a t i a l r e l a t i o n s h i p s o f K i n e s i n - 1 i n l i v i n g c e l l s . T h i s a p p r o a c h i s p a r t i c u l a r l y b e n e f i c i a l f o r s t u d y i n g p r o t e i n c o m p l e x e s w h e r e r e c o m b i n a n t e x p r e s s i o n m a y n o t r e f l e c t t h e t r u e f o l d i n g , s u b u n i t i n t e r a c t i o n s , a n d / o r p o s t t r a n s l a t i o n a l m o d i f i c a t i o n s o f a m u l t i p r o t e i n c o m p l e x . Myc-tagged rat KHC and KHC(1–891), as well as HA-tagged rat KLC and KLC(1–176), have been previously described (). Monomeric versions of the FPs ECFP (mECFP) and mCit (an EYFP variant that is a superior acceptor for FRET; ; ) were used to minimize dimerization artifacts. All FP-KHC and FP-KLC fusion proteins were created in the mECFP-C1, mECFP-N1, mCit-C1, and mCit-N1 vectors (CLONTECH Laboratories, Inc.) by PCR using primers with appropriate restriction sites and verified by DNA sequencing. A 4-aa linker (SGAG) was inserted between the FP and KHC or KLC in the FP-C1 vectors (e.g., mCit-4aa-KHC). A 5-aa linker (GPVAT) was inserted between KHC or KLC and the FP in FP-N1 vectors (e.g., KHC-5aa-mECFP), whereas a 10-aa linker (GAGTGGGGGT) was used for KHC constructs tagged with both mCit and mECFP molecules (e.g., mCit-4aa-KHC-10aa-mECFP). The KLC-mECFP and KLC-mCit constructs also contain an HA tag at the N terminus (e.g., HA-KLC-mECFP) that is not indicated in the text or figures for clarity (Fig. S1). The HA tag was not used for linking the FP and KLC, but rather for ease of cloning. ECFP-tubulin was from CLONTECH Laboratories, Inc. The linked mCit-16aa-mECFP calibration molecule has been previously described (). Mutation of loop 12 (Δloop12: H275, R279, and K282 to Ala; ) and the IAK region (ΔIAK: Q920, I921, A922, K923, P924, I925, R926, P927, and G928 to Ala) within the full-length rat KHC sequence was done by overlapping PCR. KHC(Cys344) was created by first fusing a Cys-lite version of human KHC(1–560) containing the introduced Cys344 (HK560[Cys344], a gift from R. Vale [University of California, San Francisco, San Francisco, CA]; ) to rat KHC(556–955) by overlapping PCR. Cys628 in rat KHC was then mutated to Ser by PCR. COS cells were cultured as previously described (), except that TransIT-LTI (Mirus) was used for transfection. A coverglass with transfected cells was assembled in a Leiden's chamber and maintained at 37°C in Ringer's buffer (10 mM Hepes, 155 mM NaCl, 5 mM KCl, 2 mM CaCl, 1 mM MgCl, 2 mM NaHPO, and 10 mM glucose, pH 7.2). 0.1 μg/ml SLO (in Ringer's buffer with 10 mg/ml of BSA) was added for 30 s, the cells were rapidly washed three times with Buffer I (25 mM Hepes, 5 mM MgCl, 115 mM KOAc, 5 mM NaOAc, 0.5 mM EGTA, pH 7.2, and 10 mg/ml of BSA), and 1 mM of AMPPNP or 625 mM NaCl was added. COS cells expressing KHC(Cys344) in the absence and presence of KLC were lysed in lysis buffer (25 mM Hepes/KOH, 5 mM MgCl, 115 mM KOAc, 5 mM NaOAc, and 0.5 mM EGTA, pH 7.2) at 4°C and cleared by centrifugation. Lysates were or were not incubated with 200 μM DTNB at 4°C for various times. The reaction was stopped by addition of SDS sample buffer. Proteins were separated on 6% SDS-PAGE gel and immunoblotted with a polyclonal antibody to the FP (Invitrogen). Fluorescence images were acquired using an inverted microscope (Eclipse TE-300; Nikon) with a 60×, NA 1.4, oil-immersion PlanApo objective lens (Nikon) and Lambda LS Xenon arc lamp (Sutter Instruments). Image acquisition was controlled by Metamorph 6.2r6 (Universal Imaging). Fluorescence excitation and emission wavelengths were selected using an 86006 filter set (Chroma Technology) and Lambda 10–2 filter wheel controller (Sutter Instruments) equipped with a shutter for epifluorescence illumination control. Images were acquired by a Photometrics Quantix cooled charge-coupled device camera (Roper Scientific) with exposure times of 100–800 ms. All microscopy was carried out at 37°C. All images used for FRET microscopy were corrected for illumination shading and bias offset by collecting shade images from a mixture of mECFP and mCit sandwiched between two coverslips, and by collecting bias images with the excitation light blocked. The FRET microscope was calibrated to obtain the parameters , , , and from COS7 cells expressing mCit (α), mECFP (β), or a mECFP-mCit molecule linked by 16 aa ( and ) whose FRET efficiency () was measured by fluorescence lifetime spectroscopy (). , , , and R images was performed using the corrected fluorescence images and FRET parameters as previously described (). These equations are identical to those described previously (), except that the ratio of γ/ξ was replaced with simply . and R images. For highly expressing cells only, a measurement region that excluded the nucleus and FP-molecules accumulated at the cell periphery (Fig. and R images. values. The Relocation Index is described in Fig. S5. For comparing time-lapse FRET or Relocation Index changes, a paired two-tailed test was used. Fig. S1 shows that FP fusions to KHC and KLC do not affect the interactions or activities of the Kinesin-1 subunits. Fig. S2 shows the equations and methods of FRET stoichiometry. Fig. S3 shows measurements of KHC motor-to-same tail FRET requires coexpression of mCit-KHC-mECFP with Myc-KHC. Fig. S4 shows microtubule localization of active Kinesin-1 molecules results in FRET between Kinesin-1 and tubulin. Fig. S5 shows that the KLC heptad repeats are sufficient for autoinhibition of Kinesin-1. Online supplemental material is available at .

Meiosis in mammalian oocytes is driven by changes in the activity of maturation-promoting factor (MPF). MPF activity is attributed entirely to the activity of the universal mitotic kinase, cdk1–cyclin B (; ; ). In mammals, oocytes are arrested in prophase of meiosis I with low levels of MPF (; ). An increase in MPF activity stimulates entry into M phase of meiosis I, the first sign of which is germinal vesicle (GV) breakdown (GVBD), which takes place ∼90 min after release from the follicle. Continued cyclin B synthesis and increasing levels of MPF drive the oocyte to metaphase of the first meiotic division (; ), which is followed by polar body extrusion 7–9 h after GVBD. After MI, the oocyte proceeds immediately to meiosis II, where it arrests with high levels of MPF activity that are stabilized by cytostatic factor (CSF; ; ; ; ). The MI→MII transition is characterized by a transient decrease in MPF activity brought about by the destruction of cyclin B (; ; ; ). Cyclin degradation is stimulated in meiosis and mitosis by the anaphase-promoting complex (APC), an E3 ubiquitin ligase for which several mitotic proteins, including cyclin B, are substrates. The APC-catalyzed formation of a ubiquitin chain on cyclin B marks it for destruction by the 26S proteasome, and the inactivation of MPF rapidly follows (; ; ; ). Regulation of cdk1 by cyclin availability requires that the APC is tightly regulated so as to destroy cyclin B (and its other substrates) in a timely manner (). Not surprisingly, therefore, the APC is subject to a variety of regulatory mechanisms. APC activity requires one of two positive regulators, Cdc20 and Cdh1, that are required for APC activity in mitosis and late mitosis/G1, respectively (; ; ; ). It is also positively regulated by cdk1–cyclin B–dependent phosphorylation, thereby providing a feedback loop that ensures a rapid exit from metaphase (Felix et al., 1990; ). Inactivation of the APC is mediated by several proteins that make up the spindle assembly checkpoint (; ). The proteins MAD2 and BubR1 complex with Cdc20 to ensure that the APC remains inactive until all chromosomes are bioriented on the spindle (; ; ). A more recently discovered APC inhibitor is early mitotic inhibitor 1 (Emi1; ,; ). Depletion of Emi1 from mammalian cells results in an S-phase arrest as a result of the failure to accumulate S-phase activators, including cyclin A. In cycling extracts, the depletion of Emi1 arrests the extract before entry into mitosis as a result of a failure to accumulate cyclin B (; ). In prometaphase, Polo-like kinase–dependent phosphorylation of Emi1 targets it for destruction through the Skp1–Cul1–F-box (SCF) E3 ubiquitin ligase pathway (; ). The F-box protein βTrCP is responsible for recruiting Emi1 to its SCF partners (SCF–βTrCP), triggering Emi1 degradation (; ). The inactivation of Emi1 in prometaphase is necessary to allow APC activation during mitosis, which is necessary to coordinate the destruction of securin and cyclin B. Experiments conducted thus far address the role of Emi1 in meiosis II but have not focused on a role in meiosis I (; ; ; ). However, there is evidence to indicate that Emi1 may play a role during meiosis I. A recent study shows that APC activity was found to be important for preventing the entry of prophase I–arrested mouse oocytes into the first meiotic division (). This effect may be mediated by suppressing the levels of meiotic activators such as cyclin B1. This rather surprising finding provides a new potential role for APC regulation by Emi1 in meiosis I. In this study, we investigate the role of Emi1 in meiosis I using exogenous GFP-hEmi1 and by depleting Emi1 from prophase I–arrested oocytes using morpholino (MO) oligonucleotides. We show that Emi1 is present in mouse oocytes and that its destruction is initiated immediately after GVBD. Microinjection of excess Emi1 accelerates entry into the first meiosis and arrests the oocyte at MI. The depletion of Emi1 delays entry into meiosis I and prevents normal MI spindle formation. Finally, we show that the effects of Emi1 depletion require Cdh1, thereby demonstrating that Emi1-dependent regulation of APC is essential for regulating prophase I arrest and progression through meiosis I. We first examined whether Emi1 was expressed in mouse oocytes. RT-PCR of mRNA extracted from 15 GV-stage oocytes generated a product of the expected molecular weight (218 bp). Control samples lacking reverse transcriptase failed to generate a PCR product (). In situ hybridization analysis of mouse ovaries reveals the wide-spread expression of Emi1, with the highest signal being present in oocytes and granulosa cells (). Negative controls show no signal arising from oocytes or granulosa cells (unpublished data). Western blotting analysis using an anti-Emi1 antibody revealed a band at the expected molecular mass of 49 kD (). To identify Emi1 in our Western blotting experiments, we used an Emi1 MO to deplete the protein. Emi1 MO and control MO were injected into GV-stage oocytes. Injected oocytes were maintained in 3-isobutyl-1-methylxanthine (IBMX)–mediated prophase I arrest for 24 h before being subjected to analysis by Western blotting (). Emi1 MO but not control MO resulted in the down-regulation of a 49-kD Emi1 band to ∼17% of control levels (). The depletion of Emi1 appeared to be specific, as several cross-reactive bands did not show any change, and reprobing the blot for actin revealed that loading was similar in the two lanes. Furthermore, using hEmi1 as an antigen block, we were able to eliminate the Emi1 band on the blot (). This Emi1 band was also present in interphase pronucleate-stage embryos but was reduced in one-cell embryos arrested in mitosis using nocodazole (). This is consistent with the known cell cycle–dependent destruction of Emi1 in prometaphase (; ). This was confirmed in 3T3 cells in the same conditions used for oocytes and embryos (). The levels of Emi1 protein present during oocyte maturation were examined using Western analysis. At the GV stage, an immunoreactive protein was present. During maturation, the intensity of the band decreased to barely detectable levels in MI- and MII-stage oocytes (). This indicates that Emi1 is destroyed between GVBD and MI. Emi1 destruction may be necessary during maturation to permit an increase in APC activity and, thereby, the destruction of securin and cyclin B during the MI→MII transition. To test the requirement of Emi1 destruction, we used two independent approaches. First, we expressed GFP-hEmi1 during meiosis so as to maintain elevated levels of Emi1 during meiosis, and, second, we inhibited Emi1 destruction using βTrCPΔF. βTrCPΔF is a dominant-negative form of βTrCP, which binds to substrates such as Emi1 but not to the SCF ligase (). In both experimental paradigms, the phenotype is the same: arrest at meiosis I with an intact spindle and condensed, aligned chromosomes (). Thus, Emi1 destruction is necessary during MI, presumably to allow the APC to become active once all other checkpoint requirements have been met. The loss of Emi1 immunoreactivity between the GV stage and MI suggests that Emi1 stability changes during oocyte maturation. In mitotic cells, Emi1 destruction is initiated in prophase and requires phosphorylation by Polo-like kinase 1 (; ). To relate the timing of Emi1 destruction to meiotic progression, we asked whether Emi1 stability changed around the time of GVBD as oocytes make the transition from prophase to M phase. Oocytes were coinjected with a 70-kD rhodamine-dextran to provide an accurate timing of GVBD and with complementary RNA (cRNA) for GFP-hEmi1 to provide a measure of Emi1 stability. Imaging experiments revealed that before GVBD, Emi1 was stable and showed only a minor decrease in fluorescence (). The onset of GVBD was marked by an increase in rhodamine fluorescence in the region of the GV, and, within 5 min, an increase in the destruction of GFP-hEmi1 was detectable (). The time course of destruction of endogenous Emi1 was investigated using Western analysis of oocytes at the GV stage after GVBD and progressively through meiosis I (3, 6, and 9 h after release from IBMX). The data show that Emi1 is high in GV-stage oocytes and is reduced by 3 h after release from meiotic arrest (). Together, these data show that GVBD is tightly correlated with and may initiate the onset of Emi1 destruction. To verify that Emi1 was being destroyed in oocytes by the established SCF–βTrCP-dependent pathway, GFP-hEmi1 stability was monitored in oocytes coinjected with the dominant-negative form of βTrCP (βTrCPΔF). The coinjection of βTrCPΔF resulted in a three- to fourfold decrease in the rate of Emi1 destruction after GVBD (). Because Emi1 is an APC inhibitor, it is likely to influence the levels of cyclin, which may alter the timing of progression through meiosis I. In experiments described in , we observed an apparent acceleration of GVBD in GFP-hEmi1–injected oocytes. This was confirmed by designing experiments in which GVBD was scored every 10–15 min in control and GFP-hEmi1–injected oocytes. GFP-hEmi1–injected oocytes underwent GVBD earlier than control GFP cRNA-injected oocytes (). Thus, after release from IBMX, 50% of GFP-hEmi1–injected oocytes had undergone GVBD at 45 min compared with 60 min in the control. Maximal rates of GVBD were achieved at 60 min in GFP-hEmi1–injected oocytes compared with 75 min in the controls (). We next examined whether the expression of GFP-hEmi1 could override cAMP-mediated prophase arrest. Oocytes were incubated in 30 μM IBMX, which allows ∼10% of GFP cRNA-injected oocytes to escape meiotic arrest during incubation for 20 h (). In these conditions, the rate of GVBD after the injection of GFP-hEmi1 was increased to 70% (). These data show that Emi1 has a dramatic effect on entry into the first meiotic division, presumably as a result of its ability to inhibit the APC. Next, we examined whether Emi1 regulates APC activity in GV-stage oocytes by monitoring the rate of destruction of the APC substrate cyclin B1–GFP in IBMX-arrested oocytes injected with a cRNA for myc-hEmi1. The injected cyclin B1–GFP protein was degraded slowly in GV-stage oocytes, with ∼10% of the signal being depleted over 3 h (). In oocytes previously injected with myc-Emi1, the rate of cyclin B1–GFP destruction was attenuated such that only 3% of the cyclin B1–GFP was destroyed in the same period. These experiments show that the APC is active in prophase-arrested oocytes and that its inhibition by Emi1 is sufficient to accelerate the resumption of meiosis. To understand the physiological role of Emi1 in meiosis I, we next examined the timing of GVBD in Emi1 MO–injected oocytes. As shown in , Emi1 MO–injected oocytes are depleted of ∼83% of the endogenous Emi1 after 24 h of culture. If Emi1-dependent regulation of the APC plays a role in the timing of meiosis entry, the depletion of Emi1 would be expected to increase APC activity and maintain meiotic arrest. The results in show that consistent with this hypothesis, the depletion of Emi1 delays GVBD such that 50% of control oocytes undergo GVBD at 65 min after release from IBMX compared with 110 min in Emi1 MO–treated oocytes (). This delay was rescued by the coinjection of cRNA encoding GFP-hEmi1, suggesting that effects are specific to the depletion of Emi1. The depletion of Emi1 may prevent the accumulation of APC substrates such as cyclin B1. This would be expected to result in a delay in MPF activation. To test this possibility, we measured H1 kinase activity in Emi1 MO–injected oocytes (). H1 kinase activity in control oocytes increased 2.5-fold in the first hour and fourfold over 3 h. Emi1 MO treatment delayed H1 kinase activity such that no increase was apparent after 1 h, reaching a maximum twofold increase at 2 h (). This increase was apparently sufficient to induce GVBD, but with much delayed timing. To confirm that Emi1 MO caused cyclin B1 instability in GV-stage oocytes, we monitored the rate of cyclin B1–GFP accumulation after microinjection with cyclin B1 cRNA in oocytes injected with control and Emi1 MOs. Emi1 MO–injected oocytes accumulated cyclin B1–GFP at a rate three- to fourfold slower than oocytes injected with control MO (). The decrease in cyclin B1–GFP fluorescence was caused by destruction because nondestructible cyclin B1-GFP was accumulated at a rate twofold greater than that of controls (). This further illustrates that cyclin B1–GFP is undergoing destruction in GV-stage oocytes. Continued cyclin synthesis and increasing MPF activity are required for progression through MI (; ). The decrease in MPF activity described in Emi1-depleted oocytes raised the possibility that Emi1 may be needed for the cyclin B accumulation necessary for progression through the first meiotic division. We tested this hypothesis by examining Emi1-depleted oocytes for normal MI spindle formation and the ability to extrude the first polar body. Emi1-depleted and control oocytes were matured and fixed, and their spindles and chromosomes were examined using confocal microscopy. Control MO–injected oocytes progress through MI and extrude the first polar body, so, for the purpose of comparison, an image of a control oocyte at MI is shown 7–8 h after release from IBMX (, I). In contrast to the normal spindles that form in controls, Emi1 MO–treated oocytes showed only the most rudimentary of spindle structures even after prolonged incubation for 12–24 h (). The extent of microtubule polymerization was limited and was often restricted to two small spindle poles (). Staining of chromatin in the Emi1-depleted oocytes revealed that the chromatin formed a condensed clump associated with the putative spindle structure. Individual chromosomes could not be resolved (). This phenotype is also observed after the expression of βTrCP in oocytes, which was used as an indirect way to verify that the Emi1 MO effect is specific to Emi1 depletion (). This effect was further confirmed to be specific for the effects of Emi1 depletion because it was partially reversed using GFP-hEmi1 (). Coinjection of GFP-hEmi1 cRNA and Emi1 MO resulted in the presence of spindlelike structures and an increased microtubular area after release from cAMP-mediated arrest compared with Emi1-depleted oocytes (). We confirmed in a separate series of experiments that Emi1 MO results in inhibition of the MI→MII transition (). This effect was not reversed by exogenous GFP-hEmi1, but this is not surprising given that excess Emi1 also causes arrest at MI (). To test the hypothesis that the effects of Emi1 on GVBD and polar body extrusion are mediated by APC, we coinjected Emi1 MO and Cdh1 MO. If Cdh1 is the downstream effector of Emi1, the depletion of Emi1 would be expected to be without effect in Cdh1-depleted oocytes. Oocytes were injected with control, Emi1, or a previously characterized Cdh1 MO () or were coinjected with Emi1 and Cdh1 MO. The oocytes were then incubated in 30 μM IBMX, and GVBD and polar body extrusion were examined at 24 and 48 h. As expected from previous experiments, the control and Emi1 MO–injected oocytes remained arrested at the GV stage throughout the 48-h incubation (). Consistent with a previous study (), oocytes injected with Cdh1 MO alone underwent GVBD such that ∼40 and 80% of oocytes underwent GVBD at 24 and 48 h, respectively (). Oocytes coinjected with Emi1 MO and Cdh1 MO underwent GVBD at rates similar to those receiving Cdh1 MO alone. This indicates that Emi1 acts in the same pathway as Cdh1 in regulating meiotic arrest. It was noted that most oocytes coinjected with Emi1 MO and Cdh1 MO progressed beyond GVBD and went on to extrude the first polar body by the 48-h time point (). This is consistent with the observation that several of these oocytes displayed apparently normal MI spindles with aligned chromosomes at the 24-h time point (unpublished data). To confirm that controls progress to MI and that oocytes injected with Emi1 MO alone display prometaphase arrest, as shown in , a proportion of these oocytes was released from IBMX after 24 h and scored for polar body extrusion at 48 h (). The results confirm that Emi1MO–injected oocytes arrest in prometaphase and that this arrest is reversed in the absence of Cdh1. sup #text GV-stage oocytes were retrieved from the ovaries of 21–24-d-old female MF1 mice 48 h after the administration of 7 IU of pregnant male serum gonadotrophin (PMSG; Intervet) by intraperitoneal injection. Ovaries were released into warmed M2 (Sigma-Aldrich) supplemented with 200 μM IBMX (Sigma-Aldrich) to prevent GVBD and were maintained at 37°C. For oocyte maturation experiments, GV-stage oocytes were washed three times and transferred to M16 at 37°C in an atmosphere of 5% CO in air. To recover mature (MII) oocytes, human chorionic gonadotrophin (hCG; Intervet) was administered 48–54 h after PMSG. Oviducts were removed 14–16 h after hCG. Cumulus masses were released into M2, and cumulus cells were removed by a brief incubation in M2 containing 300 μg/ml hyalauronidase (embryo-tested grade; Sigma-Aldrich). For the recovery of pronucleate embryos, female mice were mated with males at the time of hCG administration. The embryos were recovered from the oviduct in Hepes-buffered KSOM () 27–28 h after hCG and mating. Embryos that were not immediately used were transferred to KSOM at 37°C in an atmosphere of 5% CO in air. Oocytes were pressure injected using a micropipette and Narishige manipulators mounted on an inverted microscope (DM IRB; Leica). Oocytes were placed in a drop of M2 containing IBMX covered with mineral oil to prevent evaporation. Cells were immobilized with a holding pipette while the injection pipette was pushed through the zona pellucida until making contact with the oocyte plasma membrane. A brief overcompensation of negative capacitance caused the pipette tip to penetrate the cell. Microinjection was performed using a fixed pressure pulse through a picopump (World Precision Instruments). Injection volumes were estimated at 2–5% of total cell volume by cytoplasmic displacement. The oocyte volume is ∼250 pl. After microinjection, the oocytes were removed in fresh drops of M2 + IBMX under oil and allowed to recover for a few minutes before any further manipulation. Human Emi1 was cloned into pCS2 + myc5 as described previously () and pCS2-eGFP-c1 (CLONTECH Laboratories, Inc.). pCS2-GFP was a gift from M.W. Klymkowsky (Colorado University, Boulder, CO). βTrCP and βTrCPΔF were cloned in pCS2 + HA (). pMDL2–cyclin B1–GFP and pMDL2–cyclin B1-GFP were gifts from M. Herbert (University of Newcastle, Newcastle, United Kingdom; ). cRNAs encoding each of these constructs were made in vitro using the mMESSAGE mMACHINE kit (Ambion). The cRNAs were polyadenylated, purified, and dissolved in nuclease-free water to a concentration of ∼1 μg/μl before microinjection into GV-stage oocytes. For Emi1 knockdown in mouse oocytes, mEmi1 MO 5′-CGGGACAAGAAAGACAATGTTACTT-3′ (Gene Tools, LLC) was used at a concentration of 1.5 mM. For Cdh1 knockdown, we used an already characterized cdh1 MO (5′-CCTTCGCTCATAGTCCTGGTCCATG-3′; ). To control for possible nonspecific effects of the MOs, a control MO was also used (5′-CCTCTTACCTCATTACAATTTATA-3′). 15 GV-stage oocytes were collected in 9 μl of nuclease-free water. After incubation with 2 U DNase I at 37°C for 30 min, the samples were subjected to oligonucleotide (dT)-primed first-strand cDNA synthesis in 20 μl by using the RETROscript kit (Ambion). The entire reaction was then used for PCR amplification with Deep Vent DNA polymerase (New England Biolabs, Inc.) for 35 cycles. The primers used for amplifying a 218-bp-long mouse Emi1 sequence were 5′-GTGGAGGTGGCAAAGACATT-3′ and 5′-GGCAAAGGACCCACTTTAC-3′. The reaction was accompanied by a negative (minus RT enzyme) control, and the experiment was repeated three times. Ovaries of PMSG-primed mice were fixed in 4% PFA for 6 h followed by incubation in 0.5 M sucrose in PBS overnight at 4°C. The ovaries were embedded in optimal cutting temperature (Tissue-Tek), sectioned at 10 μm, and mounted on Superfrost slides (Fisher Scientific). The mouse Emi1 cDNA was subcloned into the pGEM3zf vector, linearized, and transcribed to synthesize S-labeled RNA probes. Hybridization mixtures with antisense and sense RNA probes were added to the slide and incubated overnight at 50°C. Post hybridization washes consisted of RNaseA treatment and decreasing concentrations of SSC washes. Hybridized slides were then dehydrated and dried. Slides were dipped into NTB2 emulsion (Kodak), exposed for 2 d, developed photographically, and counterstained with Gill's hematoxylin and eosin Y (0.5% wt/vol in ethanol). After counterstaining, tissues were cleared with xylene, mounted with Permount, visualized, and photographed with a camera (AxioCam; Carl Zeiss MicroImaging, Inc.). 3T3 cells were grown in DME supplemented with 10% FBS (Invitrogen). DMSO or 10 μM nocodazole/DMSO was added in subconfluent monolayers that were harvested 48 h later, lysed (50 mM Hepes, pH 7.5, 75 mM NaCl, 10 mM glycerophosphate, 2 mM EGTA, 15 mM MgCl, 0.1 mM sodium orthovanadate, 1 mM DTT, 0.5% Triton X-100, and protease inhibitor cocktail [Sigma-Aldrich]), and incubated for 10 min on ice. Lysates corresponding to interphase (DMSO) or mitosis (nocodazole/DMSO) were centrifuged for 10 min at 10,000 and 4°C. Protein concentration was determined using a protein assay kit (Bio-Rad Laboratories) according to the manufacturer's instructions. 5 μg of lysate was loaded onto gels for Western blotting. Oocytes (200 oocytes/sample) were washed in PBS/polyvinyl alcohol (PVA) and frozen in SDS sample buffer (). Proteins were separated on 4–12% NuPAGE gels (Invitrogen) and transferred to polyvinylidene fluoride Immobilon-P membranes (Millipore) using the XL II Blot Module (Invitrogen). The membranes were saturated with 5% nonfat dry milk in PBS containing 0.1% Tween 20 for 1 h at room temperature and were incubated with the primary antibodies overnight at 4°C. Three antibodies were used for Western blot analysis of Emi1. 1 μg/ml rabbit polyclonal (Gentaur), a mouse monoclonal (Zymed Laboratories), and 1 μg/ml of an affinity-purified rabbit polyclonal antibody were raised against a bacterially produced myelin basic protein (MBP)−mEmi1 fusion protein. Rabbit polyclonal antibodies were affinity purified with mEmi1 fused to GST. All antibodies provide similar results. For clarity, we have presented data obtained using the Gentaur antibody. Antigen block was performed using the Gentaur antibody by preincubation for 1 h at room temperature with a threefold molar excess of human Emi1-MBP. For the detection of actin, we used 1 μg/ml of a mouse monoclonal antiactin antibody (Chemicon). Secondary IgGs conjugated to 0.05 μM HRP (goat anti–rabbit IgG or mouse IgG; Sigma-Aldrich) were incubated with the membranes for 30 min at room temperature. Immunostained bands were detected by chemiluminescence (Pierce Chemical Co.). CDK1–cyclin B activity and MAPK activity were measured by their ability to phosphorylate histone H1 and MBP in vitro (H1 and MBP kinase assay), respectively, as previously described in detail (). Eight oocytes in 2 μl M2 were transferred in 3 μl of storing solution and immediately frozen on dry ice. The samples were diluted twice by the addition of concentrated kinase buffer (). The samples were then incubated at 37°C for 30 min and were analyzed by SDS-PAGE (as for Western blotting) followed by autoradiography. The autoradiographs were imaged using the phosphorimager system (Bas-100; Fuji) and analyzed with TINA 2.0 software. A baculovirus-based expression system was used to obtain recombinant human cyclin B1–GFP at a concentration of 2 mg/ml as described previously (). Sf9 insect cells infected with the baculovirus encoding His cyclin B1–GFP were a gift from J. Pines (Gurdon Institute, University of Cambridge, Cambridge, United Kingdom). Oocytes were fixed in freshly prepared 4% PFA (in PBS, pH 7.4, 1 mg/ml PVA) for 20 min, washed in PBS/PVA, and permeabilized in 0.1% Triton X-100 for 15 min. The cells were then incubated in blocking buffer (PBS, 3% BSA, and 10% normal goat serum) for 2 h at RT and in 5 μg/ml of a mouse anti–α-tubulin antibody (Abcam) in blocking buffer at 4°C overnight. The cells were incubated with 5 μg/ml of an AlexaFluor555-conjugated goat polyclonal anti–mouse IgG secondary antibody (Invitrogen) for 2 h at RT followed by incubation with 2 μM Hoechst 33342 to label the DNA. The immunostaining was visualized using a confocal microscope (LSM510 META; Carl Zeiss MicroImaging, Inc.). To ensure that comparisons of immunofluorescence could be made between treatment groups or developmental stages, the different samples were scanned and viewed with identical settings.

Mitochondria and chloroplasts contain β-barrel proteins in their outer membranes (; ; ). The only other biological membrane known to harbor β-barrel proteins is the outer membrane of Gram-negative bacteria (; ). This situation is believed to reflect the evolutionary origin of mitochondria and chloroplasts from endosymbionts that belong to the class of Gram-negative bacteria. Little is known about how newly synthesized β-barrel proteins are sorted in the eukaryotic cell, integrated into lipid bilayers, and assembled into oligomeric structures (; ; ; ). In the case of mitochondria, the precursors are initially recognized by the receptor components of the translocase of the outer mitochondrial membrane (TOM) complex, Tom20 and Tom70. They are then translocated through the import pore of the TOM complex (; ; ; ; ). From the TOM complex, β-barrel precursors are relayed to another complex in the outer membrane, the topogenesis of mitochondrial outer membrane β-barrel proteins (TOB) complex, also called the sorting and assembly machinery (SAM) complex (; ; ). On their way from the TOM to the TOB complex, β-barrel precursors are exposed to the intermembrane space (IMS), where they were reported to interact with small Tim components (; ; ). The major component of the TOB complex is Tob55 (also named Sam50/Omp85). Tob55 was found to be essential for viability of yeast cells and to promote the insertion of β-barrel proteins into the mitochondrial outer membrane (; ; ). Homologues of Tob55 appear to be present in virtually all eukaryotes (; ; ). Tob55 belongs to a family of β-barrel–shaped transporters, which includes the bacterial Omp85/YaeT (; ), alr2269 (), and the plastidic Toc75 (). Two further proteins, Tob38 (Tom38/Sam35) and Mas37 (Tom37/Sam37), were identified as subunits of the TOB complex (; ; ; ). The essential protein Tob38 is peripherally associated with Tob55 on the cytosolic surface of the outer membrane and, together with Tob55, forms the TOB core complex. Mas37 plays a, so far, undefined role. Another outer membrane component, Mdm10, a β-barrel protein, was also suggested to be a member of the TOB/SAM complex and to promote the assembly of Tom40 precursor (). The role of Mdm10 remains to be further clarified, as it was originally identified as a protein with a role in determining morphology and inheritance of mitochondria (). Despite some progress in our understanding of the structure of the TOB complex, the mechanism by which precursors of β-barrel proteins are transferred from the TOM to the TOB complex is still unknown. Moreover, many questions as to the functions of its subunits and their domains remain to be answered. These answers are crucial for obtaining a comprehensive view on the biogenesis of β-barrel membrane proteins. Here, we report on the contribution of the N-terminal domain of Tob55 to the function of the TOB complex. It has receptor-like function in recognizing precursors of β-barrel proteins in the IMS. The N-terminal domain of Tob55, comprising ∼100 amino acid residues, was suggested not to be part of the β-barrel structure () and to be exposed to the IMS (). Therefore, precursors of β-barrel membrane proteins that are on their way from the TOM to the TOB complex could interact with this domain. Because the location of the N-terminal domain is an essential element in such a working model, we decided to analyze the topology in detail. We used a Tob55 variant that carried a His tag at the N terminus () and an assay that monitors the formation of a characteristic fragment upon treatment of mitochondria with protease. As observed before, Tob55 is cleaved by proteinase K (PK) at a single position, resulting in an N-terminal fragment of ∼30 kD and a C-terminal fragment of ∼25 kD (; ; ). Treatment of mitochondria carrying His-tagged Tob55 with PK resulted, as expected, in the formation of a 30-kD fragment that could be immunodecorated with antibodies against the His tag (). Formation of this fragment was abolished when the outer membrane was ruptured by either osmotic shock or solubilizing the membrane with detergent (). Hence, these results imply that the N terminus of Tob55 is exposed to the IMS. To investigate the function of the N-terminal domain of Tob55, we created constructs in which 50, 80, or 102 of the N-terminal amino acid residues were deleted resulting in Tob55Δ50, Tob55Δ80, and Tob55Δ102, respectively. First, we asked whether the N-terminal domain is required for targeting of Tob55 precursor to mitochondria and its subsequent insertion into the outer membrane. To this end, we cloned these variants into a yeast expression vector and transformed wild-type cells with the resulting plasmids. Upon subcellular fractionation, all Tob55 variants were found in the mitochondrial fraction, like the mitochondrial marker proteins Tom20 (). Thus, all Tob55 variants are targeted to mitochondria in vivo. Mitochondrial targeting and membrane integration of Tob55 variants were further studied using an in vitro import assay with radiolabeled precursor proteins and isolated mitochondria. The read out of the assay was the formation of characteristic proteolytic fragments upon PK treatment (; ). Correct membrane insertion in vitro of the N-terminal truncated variants was expected to result in smaller N-terminal fragments, whereas the fragments representing the C-terminal part of the protein should remain unchanged. Indeed, upon incubation of radiolabeled Tob55 truncated variants with isolated mitochondria, the expected proteolytic fragments were formed (). Therefore, all variants appear to be targeted to mitochondria and become inserted into the outer membrane to reach the native topology. The topogenesis of newly inserted Tob55 molecules requires the import receptor Tom20, the translocation pore of the TOM complex, and the TOB complex (). We analyzed whether the truncated Tob55 variants follow this pathway. First, all variants displayed reduced efficiency of import into mitochondria deficient in Tom20 (Fig. S1, A and B, available at ). Second, blocking the TOM channel before import with a large excess of the recombinant matrix–destined preprotein, pSu9(1–69)–dihydrofolate reductase (DHFR), strongly inhibited the import and membrane insertion of the variant Tob55 proteins (Fig. S1 C). Third, to check for the involvement of the TOB complex, we compared the insertion of the Tob55 variants into wild-type mitochondria to insertion into mitochondria depleted of Tob55. All Tob55 radiolabeled variants were inserted with strongly reduced efficiency into the Tob55-depleted mitochondria (Fig. S1 D). Similar low efficiencies of insertion were observed for all Tob55 variants when import was performed with mitochondria depleted of Tob38 (unpublished data). Thus, preexisting TOB complexes are essential for the membrane insertion of all Tob55 variants. Collectively, Tob55 precursors follow the insertion pathway of β-barrel proteins even when lacking their N-terminal domain. Does the N-terminal domain of newly synthesized Tob55 precursors have a function in the integration into TOB complexes? Radiolabeled Tob55 variants were incubated with mitochondria, and the import reactions were analyzed by blue native gel electrophoresis (BNGE). The radiolabeled Tob55 precursor migrated upon BNGE as several species (; ; ; ). Initially, the radiolabeled molecule of Tob55 migrated preferentially with the uppermost one (species I). Upon prolonged incubation to allow for assembly into the preexisting TOB complexes, the precursor molecules migrated mostly as species II. This species migrated at an apparent molecular mass smaller than that of an intermediate of Tom40 precursor associated to the TOB complex; thus, it most likely represents an endogenous TOB complex. The three truncated variants assembled into the same complexes as the full-length protein (). Next, S-labeled Tob55 variants were incubated with mitochondria isolated from either wild type or a strain containing an HA-tagged version of Tob38 (Tob38; ). The TOB complex from the latter mitochondria migrates slightly slower upon BNGE than the wild-type complex. The radioactive Tob55 species imported into the Tob38 mitochondria showed the same reduced electrophoretic mobility, and they were in a complex that was recognized by antibodies against the HA tag (). This provides further evidence that all Tob55 variants studied become assembled into preexisting TOB complexes. To obtain additional independent support for this conclusion, further antibody shift experiments were performed (; ). We used an antibody raised against a peptide comprising amino acid residues 1–15 of Tob55, which does not recognize the two truncated variants, Tob55Δ80 and Tob55Δ102. Mitochondria containing imported S-labeled Tob55 variants were lysed, incubated with antibodies against the N-terminal peptide of Tob55, and analyzed by BNGE. The addition of this antibody, but not of a control antibody against Tim23, resulted in a shift of the upper S-labeled band to higher apparent molecular mass, obviously by formation of a supercomplex with the antibody (). Collectively, the absence of the N-terminal domain of Tob55 precursor does not impair its ability to become inserted into the outer membrane and assembled into preexisting TOB complexes. Is the N-terminal domain important for the function of Tob55? Because Tob55 is an essential protein, we had to use the “plasmid shuffling” method to test the ability of the truncated variants to complement the deletion of the wild-type protein (see Materials and methods). Strains that harbored a plasmid encoding full-length or truncated forms of Tob55 were tested for their ability to grow on glycerol- and glucose-containing medium at various temperatures (). Growing the cells at 30°C resulted in only minor differences in the growth rates of the various cells. In contrast, incubating the cells at 37 or 24°C resulted in a slower growth in the case of cells expressing Tob55Δ80, and even more so in cells harboring Tob55Δ102. As expected, the growth phenotype was more conspicuous on the nonfermentable carbon source, where yeast cells are dependent on mitochondria for energy production (). Thus, already the first 80 amino acid residues of Tob55 are required for optimal function of Tob55 and thus for normal growth of yeast cells. The growth phenotype of cells harboring deletions in Tob55 led us to investigate whether those cells contain normal levels of mitochondrial proteins. To that end, we isolated mitochondria from cells harboring plasmid-encoded full-length Tob55 or its truncated variants and controlled the amounts of expressed proteins by immunodecoration. The levels of the β-barrel proteins Tom40, Mdm10, and porin were reduced in mitochondria containing the truncated versions. Similarly, the levels of the truncated variants of Tob55 (β-barrel proteins themselves) and the other two components of the TOB complex, Tob38 and Mas37, were also reduced as compared with mitochondria containing full-length Tob55 (). In contrast, other proteins of the various mitochondrial subcompartments were present at roughly control levels (). Thus, the N-terminal domain of Tob55 appears to have an important role in the biogenesis of β-barrel proteins. We further investigated the assembly state of the TOB complex in the various mitochondria by analyzing them with BNGE, a method that usually results in several observed species of TOB complex (; ; ). As we observed that mitochondria harboring the truncated versions of Tob55 contain reduced levels of this protein, we analyzed a larger amount of these mitochondria. The TOB complex from mitochondria harboring the truncated versions migrated mainly as the higher molecular species of the TOB complex. Of note, all the Tob38 and Mas37 molecules in these mitochondria were assembled with Tob55 ( and not depicted), excluding the possibility that because of the reduced levels of Tob55, Tob38 and Mas37 build partial complexes, which exert dominant-negative effect. This conclusion is further supported by our previous observations that lower levels of Tob55 result in reduced biogenesis of both Tob38 and Mas37 (; ). In contrast to wild-type mitochondria or mitochondria harboring the truncated variants of Tob55, a substantial portion of the plasmid-expressed full-length Tob55 was found as low molecular weight unassembled species (). This behavior probably resulted from the fact that, like the other Tob55 variants, it was expressed from an overexpression plasmid, whereas the interacting partners, Tob38 and Mas37, are not overexpressed. To provide further support for the involvement of the N-terminal domain in biogenesis of β-barrel proteins, we performed in vitro protein import experiments with isolated mitochondria. In accordance with the in vivo results, the import efficiencies of newly synthesized β-barrel precursors like Tom40 and porin were substantially reduced in mitochondria containing the truncated versions (). Other precursor proteins, such as the inner membrane protein Tim23 and the matrix-destined pSu9-DHFR, were only moderately affected (). This latter reduction is probably due to the reduced level of Tom40 in the mitochondria harboring the truncated variants. Next, we investigated the importance of the N-terminal domain in preexisting Tob55 for the insertion of newly synthesized Tob55 precursor molecules and for the association of Mas37 precursors with mitochondria. The topogenesis of both proteins requires functional TOB complex (). A moderate reduction in the association of Mas37 was observed upon incubation with mitochondria harboring the truncated variants of Tob55 (). A stronger reduction was observed upon import of Tob55 precursor (). We propose that although the association of Mas37 with mitochondria is probably reduced because of the lower levels of Tob55, the insertion of Tob55 precursor (a β-barrel protein itself) is affected by both the reduced levels of Tob55 and the absence of the N-terminal domain. To exclude the possibility that the effect on the insertion of β-barrel proteins observed for mitochondria harboring truncated variants of Tob55 resulted only from reduced endogenous levels of both Tob55 and Tom40 in those organelles, we performed control in vitro import experiments. We incubated the radiolabeled precursor proteins with 50 μg of mitochondria harboring plasmid-encoded full-length Tob55, 150 μg of mitochondria harboring Tob55Δ80, or 100 μg of wild-type mitochondria. Under these conditions, comparable amounts of TOB and TOM complexes were present in import reactions with the two former types of mitochondria (). The matrix-destined protein, pSu9-DHFR, was imported under these conditions with similar efficiency in all reactions (). In contrast, the import of the β-barrel precursor porin into mitochondria carrying the truncated Tob55 variant was still impaired. Of note, although the samples with wild-type mitochondria contain far less Tob55 molecules in comparison with those containing Tob55Δ80, the efficiency of porin import into the former mitochondria was substantially higher (). Furthermore, the assembly of two other β-barrel proteins, Mdm10 and Tom40, as analyzed by BNGE, was dramatically reduced in mitochondria harboring the truncated versions of Tob55 (). Thus, the reduced import of β-barrel precursors into mitochondria harboring truncated Tob55 variants is caused mainly by impaired function of the corresponding Tob55 molecules and is not solely due to a reduced level of Tob55. Collectively, these experiments suggest that the N-terminal domain of Tob55 is playing a central role in the biogenesis of β-barrel proteins. To study the function of the N-terminal domain of Tob55, a fusion protein consisting of the N-terminal 120 amino acid residues and maltose binding protein (MBP) was expressed in and purified from . As controls, MBP alone and MBP fused to the cytosolic domain of the mitochondrial outer membrane protein Fis1 (MBP-Fis1) were expressed and purified in parallel (). All three proteins were analyzed for their ability to bind precursors of β-barrel proteins, porin and Mdm10. Background binding to the matrix was observed in the case of porin precursor. However, in the case of MBP-Tob55, we observed on top of these background levels specific binding that was severalfold higher than that observed with the control proteins (). Specific binding to MBP-Tob55 was observed also with the β-barrel protein, Mdm10. Only very low unspecific binding of a matrix-destined precursor, pSu9-DHFR, was observed with all proteins (). To further verify that the binding to the N-terminal domain is specific and saturable, we added increasing amounts of radiolabeled Mdm10 precursor to equal small amounts of either MBP-Tob55 or MBP alone as control. In each added amount, severalfold more Mdm10 molecules were bound to the MBP-Tob55 in comparison with the control protein. Furthermore, a saturation of the binding was observed when large amounts of precursor were used (). Thus, the first 120 amino acid residues of Tob55 appear to be sufficient to support specific interaction with β-barrel precursors. To obtain further support for this proposal, we investigated whether the N-terminal domain of Tob55 is able to compete out the import of porin and Mdm10. Radiolabeled precursors of porin and Mdm10 were incubated in the presence or absence of competing amounts of the purified MBP-Tob55 or control proteins, and isolated mitochondria were added. The presence of the N-terminal domain of Tob55 substantially impaired the import of both precursors, whereas the control proteins (MBP and MBP-Fis1) did not have a substantial effect (). Notably, the level of inhibition of porin insertion depended on the amount of added recombinant MBP-Tob55 (). This effect was only observed when the N-terminal domain was in a native state, as preincubation of the latter protein with urea impaired its ability to compete for the import of porin (). MBP-Tob55 did not compete out the translocation through the TOM pore of other precursor proteins, such as pSu9-DHFR and Tim23 ( and ). Hence, it is unlikely that this inhibitory effect is entirely due to an ability of residues 1–120 of Tob55 to cross the TOM pore and thereby to jam the import channel. Moreover, as shown in and , residues 1–102 are not required for import of Tob55 through the TOM complex. Along the same line, when radiolabeled Tob55(1–120) was synthesized in a cell-free system and incubated with isolated mitochondria, it did not become protected from degradation by added proteases (unpublished data). Thus, this domain is not competent for import across the outer membrane. It is also unlikely that the competence of MBP-Tob55 is due to an interaction of the N-terminal domain with the import receptors Tom20 and Tom70; MBP-Tob55 was also able to compete import into mitochondria lacking either Tom20 or Tom70 (unpublished data). To further substantiate the capacity of the N-terminal domain to bind β-barrel proteins, we investigated the binding of this domain to a water-soluble form of porin (). This water-soluble porin, isolated from detergent-purified porin from , has the properties of the precursor form of porin and can be imported into the mitochondrial outer membrane (; ). We first checked whether the precursor of porin can be imported into and assembled in the outer membrane of yeast mitochondria. Indeed, the orthologue was imported into yeast mitochondria in the pathway that involved the general insertion pore. Blocking this pathway with excess recombinant preprotein, pSu9-DHFR, inhibited membrane integration (; ; ; ). Furthermore, as was observed for yeast β-barrel precursors, the import of porin into yeast mitochondria was impaired in mitochondria lacking Tom20 or harboring reduced levels of Tob38 (unpublished data; ; ; ; ). Porin is known to form several oligomeric structures that can be observed by BNGE (; ; ). The radiolabeled porin was assembled upon its import into yeast mitochondria into the same oligomeric structures as the yeast protein (). We further verified that the water-soluble porin is import competent as was published before (; ). Indeed, water-soluble porin but not the control protein, MBP, was able to compete out the import of two other β-barrel precursors, Tom40 and Porin (). Collectively, we conclude that porin and its water-soluble form can use the yeast machinery for biogenesis of β-barrel proteins. To study the interaction of water-soluble porin with the N-terminal domain of Tob55, we used the seminative gel electrophoresis, which was used successfully to study the interaction of Omp85 with bacterial β-barrel proteins (). Under these conditions, MBP-Tob55 migrated as two dominant bands (). Various amounts of MBP-To b55 and two control proteins were subjected to seminative SDS-PAGE and transferred onto nitrocellulose membrane. Water-soluble porin was incubated with this membrane, and the bound porin was detected by immunodecoration. The N-terminal domain of Tob55 could bind specifically water-soluble porin in a concentration-dependent manner (). Next, we aimed to obtain more quantitative information on the interaction of water-soluble porin with the N-terminal domain of Tob55. To that end, water-soluble porin was radiolabeled with C-formaldehyde by reductive methylation (). To quantify the binding, increasing amounts of C-ws-porin were added to 2 μg of either MBP (as control) or MBP-Tob55 bound to amylose beads or were loaded directly on a gel to serve as loading standards. The beads were washed, and bound water-soluble porin was eluted and analyzed. Much higher levels of binding were observed with MBP-Tob55 as compared with the control protein (). Unfortunately, we could not demonstrate saturable binding for C-ws-porin because, under our experimental conditions, it tends to aggregate at higher concentrations. As shown in , the binding was concentration dependent at concentrations ranging from 1 to 152 nM. = 12 nM. Of note, this value is in the same range as the dissociation constant of the binding of presequence-containing precursor protein and the TOM complex (). Collectively, these results demonstrate that the N-terminal domain of Tob55 is able to interact with precursors of β-barrel proteins. Although β-barrel precursors were suggested to be exposed to the IMS on their transit from the TOM to the TOB complex (; ; , ; ; ), an open question is whether the TOB complex is required for the transfer of β-barrel precursors across the outer membrane. We observed a reduced import of β-barrel precursors into mitochondria containing truncated Tob55 variants (). Because we analyzed the import as protection against externally added protease, our observations suggest that functional Tob55 is required for translocation of β-barrel precursor across the outer membrane. Next, we wanted to investigate whether translocation of β-barrel precursors across the import pore of the TOM complex is required for efficient insertion of these precursor proteins into the outer membrane. To that end, we blocked the TOM channel with an excess of matrix-destined precursor to reduce both membrane insertion of β-barrel proteins and the import of matrix-destined precursor proteins (Fig. S1 and ; ; ; ; ). As was observed before, rupturing the outer membrane resulted in substantial reduction in the insertion efficiency of porin precursor (). This reduction is probably caused by the loss of the small Tim proteins, which were shown to be involved in the biogenesis of β-barrel proteins (; ; ). Surprisingly, blocking the TOM complex in mitochondria with ruptured outer membrane strongly impaired the insertion of β-barrel precursors (). This behavior is different from that of matrix-destined precursors, where rupturing the outer membrane can overcome such blockage of the TOM channel (). The residual insertion of porin precursor into the outer membrane of ruptured mitochondria did not result from the insertion capacity of subpopulation of intact mitochondria. We could not detect any DLD1 (a marker IMS protein) upon treatment of the ruptured mitochondria with external protease, suggesting that all mitochondria were ruptured (). Thus, we propose that the efficient recognition of β-barrel precursors by the TOB complex requires a preceding translocation across the import channel of the TOM complex. We present here evidence for the involvement of the N-terminal domain of Tob55 in recognition of β-barrel precursors and thus in the transfer of precursors from the IMS to the TOB complex. Such an involvement is in agreement with the location of this domain in the IMS. It is also in line with the suggestion that the N-terminal region of the bacterial Omp85/YaeT recognizes β-barrel precursors in the periplasmic space (). This region was named POTRA (polypeptide transport–associated domain; ). According to prediction programs, the POTRA domain of yeast Tob55 covers amino acid residues 29–108. Notably, chloroplast Toc75, another protein belonging to the β-barrel–type pores, also has a POTRA-like region at its N terminus. In contrast to Tob55, Toc75 is involved in translocation of precursor proteins with chloroplast targeting signals across the outer membrane. It is currently unclear whether this protein is also involved in the insertion of β-barrel precursors. The N-terminal domain of Toc75 was reported to play a role in the recognition of stroma-destined precursor proteins (). Collectively, a receptor-like function of a hydrophilic N-terminal domain might be a common feature of the β-barrel translocases of the extended Tob55/Toc75/Omp85 family. The growth behavior of cells with a Tob55 that lacks the N-terminal domain underscores the functional importance of this part of the protein. This domain appears to be required neither for targeting of Tob55 to mitochondria nor for its assembly into the TOB complex. There may be, however, a role of the N-terminal domain in the structural organization of the TOB complex, as we observed that the TOB complex containing the deletion variants of Tob55 has altered migration behavior in native gel system. We cannot exclude the possibility that part of the reduction in the biogenesis of β-barrel proteins in the strains containing the deletion variants is due to altered conformation of the TOB complex. However, our data strongly support our suggestion that this reduction results from the absence of the binding capacity of the N-terminal domain. Tob55 interacts with the other two components of the TOB complex, Tob38 and Mas37. The location of the N-terminal domain in the IMS makes it an unlikely candidate for such an interaction, as the two other subunits are attached to the cytosolic surface of the outer membrane (; ; ; ). What could be the signal in the β-barrel precursors that is being recognized by the N-terminal domain? Our current observations with Tob55 and previous results with the precursors of Tom40 and porin suggest that this signal does not reside in the N-terminal domain of the precursor proteins (; ). Currently, six putative β-barrel proteins in yeast mitochondria are known: two isoforms of porin, Tom40, Tob55 itself, and two proteins that seem to be involved in maintenance of mitochondrial morphology, Mdm10 and Mmm2. Despite their similar overall structure, these proteins show extensive divergence of their primary sequences. A linear consensus sequence as recognition and/or sorting signal is therefore unlikely. In the bacterial system, there are some hints as to signals in the precursors that are transported through the periplasm to the outer membrane (; ). It is currently unclear, however, whether these signals play a role in the insertion pathway mediated by Omp85/YaeT. Irrespective of which signals are being recognized by the N-terminal domain of Tob55, we can propose a working model for the biogenesis of mitochondrial β-barrel membrane proteins. Precursors of these proteins are translocated across the outer membrane to the IMS by the TOM complex. This way of delivery is required for the subsequent productive recognition of precursor molecules by the TOB complex and cannot be bypassed by opening the outer membrane. The N-terminal domain of Tob55 is playing an important role in the initial interaction of the TOB complex with the precursor proteins, most likely as soon as they emerge from the TOM complex. In the absence of this domain, the translocation across the TOM complex and subsequent membrane integration of β-barrel precursors are impaired. Thus, it appears that the translocation of β-barrel precursor proteins across the outer membrane and their recognition by the TOB complex are coupled processes. Currently, the molecular mechanism of this coupling is not clear. In conclusion, our data show that the N-terminal domain of Tob55, which contains the POTRA motif and is exposed in the IMS has a crucial role in guiding the precursor of β-barrel proteins from the IMS into the outer membrane. It will be very important to understand at the molecular level all the events that lead to the membrane integration of β-barrel precursors. Standard genetic techniques were used for the growth and manipulation of yeast strains (). The wild-type strain YPH499 (MAT ade2-101 his3-Δ200 leu2-Δ1 ura3-52 trp1-Δ63 lys2-801) was used. The -null strain YTJB64 and its corresponding parental strain YTJB4 were used (a gift from G. Schatz, Biozentrum der Universität Basel, Basel, Switzerland). Transformation of yeast was performed using the lithium-acetate method. Yeast cells were grown under aerobic conditions on YPD (1% [wt/vol] yeast extract, 2% [wt/vol] bactopeptone, and 2% glucose), YPG (1% [wt/vol] yeast extract, 2% [wt/vol] bactopeptone, and 3% glycerol), or synthetic medium. The gene was cloned by PCR from yeast genomic DNA using primers based on the published sequence. The PCR product was inserted into the yeast expression vector pVTU-102, which contains the selectable marker , and the resulting plasmid was transformed into the wild-type strain YPH499. The genomic open reading frame in this strain was replaced with the marker gene by homologous recombination. In the resulting HisUra strain, the complete coding sequence of the gene was deleted and a wild-type gene was present on a 2μ plasmid; this strain was termed YSH1. For growth tests on plates, cells were grown to log phase and diluted to an OD of 0.3. Cells were then diluted in 10-fold increments, and 5 μl of each was spotted onto the indicated solid media. The PCR products encoding full-length Tob55 or Tob55 variants with various deletions in the N-terminal domain were inserted into a centromeric yeast expression vector, pYX132xTPIp- (Invitrogen). To test for the ability of the Tob55 variants to complement the function of Tob55, these plasmids were transformed into the YSH1 strain, and UraTrp transformants were streaked onto 5-fluoro-orotic acid plates. This treatment resulted in the elimination of the wild-type copy of Tob55, which is encoded on the -containing plasmid and thus allowed to investigate whether the variant on the -containing vector could support growth. Native porin from was isolated by modification of a published procedure (). Shortly, 5 mg of outer membrane vesicles, which were isolated as described elsewhere (), were solubilized in 1 ml of buffer containing 50 mM Hepes-KOH, 1 mM PMSF, 10% glycerol, and 2% Triton X-100. After a clarifying spin (36,670 , 10 min, 2°C), the supernatant was loaded onto an anion-exchange column (ResQ; GE Healthcare). The flow-through that contains porin was collected. Further treatment to obtain water-soluble porin was done as described previously (). The labeling of water-soluble porin with C-formaldehyde was accomplished by reductive methylation in the presence of NaBHCN, as described previously (). For binding experiments, various amounts of radiolabeled water-soluble porin were added in binding buffer (100 mM KCl, 0.025% BSA, 10% glycerol, and 100 mM sodium phosphate, pH 6.8) to MBP or MBP-Tob55 prebound to amylose beads. We performed our experiments in low temperature (4°C) and in the presence of BSA and salt because it was reported that these conditions can reduce the tendency of water-soluble porin to adhere to surfaces and thus to cause unspecific binding (). After incubation at 4°C for 35 min, the beads were washed once with binding buffer, with binding buffer without BSA, and finally with buffer containing 100 mM NaCl and 90 mM Tris-base. Bound proteins were eluted with sample buffer and analyzed by SDS-PAGE and autoradiography. For quantification of the binding reactions, increasing amounts of radiolabeled water-soluble porin were analyzed directly by SDS-PAGE and autoradiography. Mitochondria were prepared by differential centrifugation as described previously (). Radiolabeled precursor proteins were synthesized in rabbit reticulocyte lysate in the presence of [S]methionine (MP Biomedicals) after in vitro transcription by SP6 polymerase from pGEM4 vectors containing the cDNA of interest. Import experiments were performed at 25°C in an import buffer containing 250 mM sucrose, 0.25 mg/ml BSA, 80 mM KCl, 5 mM MgCl, 10 mM MOPS-KOH, 2 mM NADH, and 2 mM ATP, pH 7.2. Mitochondria containing plasmid-encoded Tob55 were preincubated at 37°C for 10 min before the addition of the radiolabeled precursor proteins. The DNAs encoding either the N-terminal domain of Tob55 (amino acid residues 1–120) or the cytosolic domain of Fis1 (amino acid residues 1–98) were cloned into the pMalCRI plasmid (New England Biolabs, Inc.) and expressed in BL21 cells as soluble fusion proteins with MBP. Purification of the protein was performed according to the manufacturer's instructions. For in vitro binding assays, cells were lysed and proteins were applied to amylose resin. Unbound proteins were washed out with MBP-column buffer (20 mM Hepes, pH 7.4, 100 mM NaCl, 2 mM EDTA, and 1 mM PMSF). To minimize unspecific binding, the resin was further washed with import buffer containing 3% BSA and then incubated for 20 min with 50 μl reticulocyte lysate, which was not used for in vitro translation. The resin was washed again and incubated for 20 min at 4°C in import buffer with 50 μl reticulocyte lysate containing radiolabeled proteins. In the case of Mdm10, the binding was performed in the presence of 0.3% digitonin. The resin was then washed twice with import buffer, and bound proteins were eluted with 1 M NaCl. Seminative SDS-PAGE and far Western blotting were performed according to published procedures (). Mitochondria were lysed in 50 μl digitonin buffer (1% digitonin, 20 mM Tris-HCl, 0.1 mM EDTA, 50 mM NaCl, 10% glycerol, and 1 mM PMSF, pH 7.4). After incubation for 15 min at 4°C and a clarifying spin (36,670 , 15 min, 2°C), 5 μl sample buffer (5% [wt/vol] Coomassie brilliant blue G-250, 100 mM Bis-Tris, and 500 mM 6-aminocaproic acid, pH 7.0) were added, and the mixture was analyzed by electrophoresis in a 6–13% gradient blue native gel (). Antibody shift experiments were performed as described previously (). Fig. S1 includes results of experiments demonstrating that Tob55 precursors follow the insertion pathway of β-barrel proteins even when lacking their N-terminal domain. Specifically, their correct topogenesis requires the import receptor Tom20, the translocation pore of the TOM complex, and the TOB complex. Online supplemental material is available at .

WNK1 (with-no-K[Lys] kinase-1) was originally identified as an unusual kinase that lacked an invariant catalytic Lys residue in subdomain II of the catalytic domain that is crucial for binding of ATP (). WNK1 is a catalytically active kinase, and modeling (), as well as structural analysis, of the WNK1 catalytic domain () revealed that a Lys residue in subdomain I substitutes for the missing Lys residue in subdomain II. WNK1 is a widely expressed protein kinase comprising 2,382 residues. It possesses a kinase catalytic domain at its N terminus (residues 221–479), and apart from three putative coiled-coil domains, the remainder of the WNK1 polypeptides possess no obvious structural features (; ). Great interest in WNK1 was aroused after the finding that intronic deletions that increased WNK1 expression were observed in humans with an inherited hypertension and hyperkalemia (elevated plasma K) disorder termed Gordon's syndrome or pseudohypoaldosteronism type II (OMIM 145260; ). These findings indicated that overexpression of WNK1 may result in hypertension and, consistent with this, heterozygous WNK1 mice possess reduced blood pressure (). WNK1-knockout embryos fail to develop, indicating that WNK1 is also required for normal development. There are four isoforms of WNK (WNK1, -2, -3, and -4) in humans encoded by distinct genes (). Mutations in WNK4 have also been found in patients with Gordon's syndrome, but in contrast to WNK1, these comprise point mutations lying within noncatalytic regions of this enzyme (). It is not yet clear how mutations in WNK4 lead to Gordon's syndrome, but overexpression of a Gordon's syndrome mutant of WNK4, but not the wild-type enzyme, increased blood pressure in mice (). Most functional studies on WNK isoforms have focused on the overexpression of these enzymes in oocytes or epithelial cells and monitoring the effects that this has on the activity and membrane localization of coexpressed ion cotransporters or ion channels. These have thus far revealed that WNK isoforms have effects on the activity and/or membrane expression of the thiazide-sensitive Na:Cl cotransporter (NCC), the bumetanide-sensitive Na:K:2Cl cotransporter-1/2 (NKCC1/2), the K:Cl cotransporter-2, the Cl:HCO3 exchanger, the inwardly rectifying K channel, the epithelial Na channel, the tight junction claudin proteins, and the transient receptor potential vanilloid-4 Ca channel (for reviews see ; ; ). Recent findings indicate that the protein kinases WNK1 and -4 interact with high affinity with the protein kinases STE20/SPS1-related proline alanine–rich kinase (SPAK) and the oxidative stress response kinase-1 (OSR1; ; ; ). These observations were followed by the finding that WNK1 and -4 could phosphorylate and activate SPAK and OSR1 in vitro (; ; ). SPAK and OSR1 are phosphorylated by WNK1/WNK4 at a Thr residue located within the T-loop (Thr233-SPAK and Thr185-OSR1) as well as at a conserved noncatalytic Ser residue (Ser373-SPAK and Ser325-OSR1) lying within a region termed the S-motif (). Mutational analysis indicated that phosphorylation of the T-loop rather than the S-motif was required for the activation of SPAK and OSR1 by WNK1 (). SPAK and OSR1 were originally identified through their ability to interact, phosphorylate, and activate NKCC1 (; ) and may also regulate NCC (). SPAK and OSR1 are 68% identical in sequence and possess a highly similar kinase catalytic domain as well as a conserved C-terminal (CCT) domain, which interacts with RFXV/I motifs present in both WNK isoforms as well as NKCC1 (; ; ; ). The activity and phosphorylation of NKCC family cotransporters is stimulated by hyperosmotic stress (; ; ), conditions that have also been reported to enhance WNK1 activity (; ) and induce phosphorylation of NKCC1 at the sites targeted by SPAK/OSR1 in vitro (). In this study, we investigate the mechanism by which WNK1 as well as its substrates SPAK/OSR1 are regulated by hyperosmotic stress. It has been reported that hyperosmotic stress stimulates WNK1 activity (; ). As a prelude to investigating the mechanism by which WNK1 is regulated, we further assessed the activity of endogenous WNK1 in 293 cells. We found that WNK1 immunoprecipitated from sorbitol-treated cells phosphorylated OSR1 at an approximately threefold higher rate than WNK1 derived from nontreated cells (). Consistent with WNK1 mediating this phosphorylation, no phosphorylation of OSR1 was detected when a mutated form of OSR1 in which the T-loop (Thr185) and S-motif (Ser325) sites phosphorylated by WNK1 were changed to Ala or when preimmune antibody was used instead of the specific anti-WNK1 antibody in the immunoprecipitation step (). A dose-dependent increase in WNK1 activity was observed by treatment with 0.1–1.0 M sorbitol, resulting in up to a fivefold increase of WNK1 activity (). Activation of WNK1 by 0.5 M sorbitol was detected within 0.5 min, with maximal activation being reached within 1–2 min, which was sustained for 80 min (). Consistent with hyperosmotic stress activating WNK1, treatment of cells with increasing concentrations of NaCl also induced activation of WNK1, which was detectable at 50 mM and maximal at 0.15 M (). Similarly, treatment of cells with 0.5 M KCl led to a threefold increase of WNK1 activity (). We next explored whether WNK1 was activated by other stimuli, namely, serum, IGF-1, phorbol ester (TPA), oxidative stress (HO), hypotonic stress (medium diluted twofold in HO), an NKCC1/2 inhibitor (bumetanide), TNFα, a tyrosine phosphatase inhibitor (pervanadate), or a serine/threonine phosphatase inhibitor (calyculin A). These treatments failed to induce activation of WNK1 under conditions in which they triggered activation of other signaling pathways (). As 0.5 M sorbitol stimulated the extracellular signal–regulated kinase (ERK) 1/2, p38, and JNK protein kinases (), we investigated whether these enzymes might be involved in the activation of WNK1. Treatment of cells with the MEK inhibitor PD184352 () or U0126 (not depicted) abolished ERK activation by sorbitol, without affecting WNK1 activation. The p38 inhibitor SB203580 and high concentrations of BIRB0796 that inhibit all p38 and JNK isoforms () also failed to prevent activation of WNK1 by sorbitol (). Consistent with p38 and JNK not regulating WNK1, anisomycin that potently stimulates p38 and JNK failed to activate WNK1 (). Despite the large size of WNK1 (∼250 kD), we consistently observed a reduction in the electrophoretic mobility of endogenous WNK1 isolated from sorbitol-treated cells on a polyacrylamide gel (). To assess whether this was due to enhanced phosphorylation, 293 cells were labeled with P-orthophosphate and endogenous WNK1 was immunoprecipitated from control and sorbitol-treated cells. Electrophoresis on a polyacrylamide gel revealed a Coomassie-stained band at ∼250 kD that was identified as WNK1 by mass spectrometry (). Autoradiographic analysis of the gel revealed that phosphorylation of WNK1 was stimulated by sorbitol (). The P-labeled WNK1 from control and sorbitol-treated cells was digested with trypsin, and the resulting peptides were chromatographed on a C column. Sorbitol substantially increased the abundance of several P-labeled peptides (). All phosphopeptides were subjected to mass spectroscopy and some of them to solid phase Edman sequencing analysis, which enabled us to identify five sites of phosphorylation (). Phosphorylation of two sites, Ser1261 and Ser 2372, was stimulated by treatment of cells with sorbitol, whereas Ser2012, Ser2029, and Ser2032 were phosphorylated at similar levels in WNK1 isolated from control and sorbitol-stimulated cells. As several of the phosphopeptides observed in the P-cell–labeling experiments presented in were not present at sufficient levels to enable phosphorylation site assignment, we undertook a mass spectrometry phosphopeptide mapping analysis of a larger amount of endogenous WNK1 immunoprecipitated from unlabeled control and sorbitol-treated 293 cells (). The immunoprecipitated WNK1 was digested with trypsin, and the resulting peptides were subjected to liquid chromatography/mass spectrometry with precursor ion scanning on a Q-TRAP mass spectrometer, which enabled us to assign five residues (Ser15, Ser167, Ser382, Ser1261, and Thr1848) whose phosphorylation was markedly enhanced by treatment of cells with sorbitol (). One of these sites, Ser382, which lies within the T-loop of the WNK1 catalytic domain, has been reported to be an autophosphorylation site on WNK1 catalytic domain expressed in (). The location of all the phosphorylation sites identified within WNK1 and the amino acid sequences surrounding these sites are illustrated in . To verify whether the activity of WNK1 was influenced by its phosphorylation state, we incubated WNK1 isolated from control and sorbitol-treated 293 cells with the serine/threonine protein phosphatase-1γ (PP1γ) and assessed its activity. We observed that incubation with PP1γ substantially decreased both the basal and sorbitol-stimulated activity of WNK1 and also increased WNK1 electrophoretic mobility (). Microcystin-LR, a toxin that specifically inhibits serine/threonine phosphatases, prevented PP1γ-induced inactivation as well as the increase in WNK1 electrophoretic mobility (). We observed that an N-terminal fragment of WNK1 encompassing residues 1–667 retained the property of becoming activated after stimulation of 293 cells with sorbitol (). Mutation of Ser382 that lies within the T-loop of WNK1 to Ala prevented activation of WNK1 by sorbitol treatment (), consistent with the previous report that this mutation inhibited WNK1 activity expressed in (). Mutation of Ser382 to Glu enhanced the basal activity of WNK1[1–667] to a level that was over eightfold higher than that observed for wild-type WNK1[1–667] isolated from untreated cells (). The activity of the WNK1[S382E, 1–667] mutant was not enhanced by sorbitol stimulation. Sorbitol markedly increased recognition of WNK1[1–667] by a phospho-Ser382 antibody that we generated (). Consistent with this antibody being specific, mutation of Ser382 to Ala abolished its recognition of WNK1 (). We also observed that sorbitol treatment induced Ser382 phosphorylation of a catalytically inactive WNK1[D368A, 1–667] mutant to the same extent as the catalytically active WNK1[1–667] fragment (), indicating that phosphorylation of WNK1 at Ser382 was not mediated by intramolecular autophosphorylation. To determine whether WNK1 possesses the intrinsic ability to autophosphorylate on Ser382, we expressed wild-type WNK1[1–661] or kinase-inactive WNK1[D368A, 1–661] in . Consistent with previous work (), we observed that the wild-type WNK1[1–661] was heavily phosphorylated at Ser382 and also possessed a high specific activity of 5.2 U/mg (). The activity of expressed WNK1[1–661] was much higher than that of the WNK1[1–667] fragment isolated from sorbitol-stimulated 293 cells, which had a specific activity of only ∼0.13 U/mg (). We observed that the kinase-inactive WNK1[D368A, 1–661] mutant expressed in was not phosphorylated at Ser382 (). Moreover, the WNK1[S382A, 1–661], WNK1[S382E, 1–661], and WNK1[S382D, 1–661] mutants expressed in possessed low activities of 0.15, 0.66, and 0.26 U/mg, respectively (). Incubation of wild-type WNK1[1–661] with MgATP in vitro did not further increase phosphorylation of Ser382 (). As outlined in the introduction, WNK1 interacts with its substrates SPAK and OSR1 through RFXV/I motifs. Interestingly, the site of phosphorylation, Ser1261, is located adjacent to such a motif (), suggesting that phosphorylation of Ser1261 might influence the ability of WNK1 to bind SPAK and OSR1. To investigate this, we generated biotinylated peptides that encompass residues surrounding Ser1261 in its phosphorylated or nonphosphorylated form. These were conjugated to streptavidin–Sepharose and tested for ability to affinity purify endogenously expressed SPAK and OSR1 from 293 cell extracts. The nonphosphorylated Ser1261 peptide interacted with SPAK and OSR1 to a markedly greater extent than the phosphorylated peptide (). Using a surface plasmon resonance binding assay, we also observed that the nonphosphorylated Ser1261 peptide interacted with a dissociation constant of ∼40 nM with the isolated CCT domain of OSR1, whereas the phosphorylated Ser1261 peptide bound to OSR1 with a markedly lower affinity (). We next studied whether the downstream substrates of WNK1, SPAK and OSR1, were activated by hyperosmotic stress in HeLa and 293 cells. Endogenous SPAK and OSR1 were immunoprecipitated from control and sorbitol-treated cells with an antibody that immunoprecipitates both proteins and assayed using a peptide substrate termed CATCHtide (). The activity of SPAK/OSR1 was increased approximately fivefold by sorbitol stimulation of HeLa cells (, top) and approximately fourfold in 293 cells (Fig. S1, available at ). This is consistent with previous reports that these enzymes are activated by sorbitol (; ). We raised phosphospecific antibodies that recognize the T-loop and S-motif residues on SPAK and OSR1 phosphorylated by WNK1. The sequences surrounding these sites are almost identical in SPAK and OSR1 (), suggesting that the T-loop and S-motif phosphospecific antibodies should recognize the phosphorylated forms of both SPAK and OSR1. The specificity of the antibodies was confirmed by overexpressing OSR1 () or SPAK () in 293 cells and finding that mutation of the T-loop or S-motif residues to Ala abolished antibody recognition. Using these antibodies, we demonstrated that sorbitol treatment markedly stimulated the phosphorylation of endogenously expressed SPAK and OSR1 at both the T-loop and S-motif in HeLa cells () as well as in 293 cells (Fig. S1). Sorbitol also induced a decrease in the electrophoretic mobility of endogenous SPAK. Studying overexpressed OSR1 () and SPAK () in 293 cells, we observed that mutation of the T-loop, but not the S-motif, to Ala, inhibited sorbitol-induced activation of these enzymes. We also found that mutation of the T-loop residue to Glu markedly increased the basal activity of OSR1 and SPAK and the activity of these mutants was not further increased by sorbitol treatment of cells. Mutation of the S-motif to Glu moderately enhanced basal and sorbitol-stimulated OSR1 () and SPAK () activity. We next investigated the effect of depleting the levels of endogenous WNK1 using a siRNA duplex targeting WNK1 that reduces expression >90% (). Under these conditions, sorbitol-induced phosphorylation of the T-loop and S-motif of SPAK and OSR1 were markedly reduced. Depletion of WNK1 also inhibited the basal as well as sorbitol-induced SPAK/OSR1 activity by >50% without affecting sorbitol-induced phosphorylation of JNK isoforms at their T-loop residues (). We also studied the localization of GFP-WNK1 () that was stably expressed at a similar level to that of endogenous WNK1 () as well as the localization of endogenous WNK1 in 293 cells (). In unstimulated cells, WNK1 was diffusely localized throughout the cytosol ( and , panel 1), but after treatment with 0.2 M sorbitol for only 1 min, WNK1 was strikingly observed on discrete intracellular structures. Colocalization experiments revealed that both GFP-WNK1 and endogenous WNK1 colocalized with clathrin ( and , panel 2) as well as AP-1 ( and , panel 3), which is an adaptor for clathrin and is recruited to budding vesicles at the TGN and endosomes (). Partial colocalization between GFP-WNK1 and the TGN46 integral membrane protein that is predominantly localized to the TGN was also observed (, panel 4). In contrast, WNK1 did not colocalize with the early endosomal markers EEA1 () and Hrs () or the late endosome and lysosome marker LAMP1 (; and ). Nor did it colocalize with the AP-2 adaptor ( and ), which plays a central role in clathrin-mediated endocytosis by linking transmembrane receptors to be internalized to the clathrin lattice (). We next investigated the redistribution of stably expressed GFP-WNK1 in living 293 cells using time-lapse microscopy ( and Videos 1 and 2, available at ). The redistribution of GFP-WNK1 to intracellular vesicles after sorbitol (Video 1) or NaCl (Video 2) treatment was rapid and observed within ∼0.5 min, the earliest time point that we could monitor (). Frames of the video were taken every 2 s, and substantial movement of some of the GFP-WNK1 localized was observed within a 10–30-s time frame (Videos 1 and 2 and Figs. S2 and S3). This effect was reversible, as removal of sorbitol or NaCl resulted in WNK1 becoming diffusely localized in the cytosol within 2 min (). We also undertook an experiment of FRAP to monitor the dynamics of WNK1 movement within the cells. In sorbitol- or NaCl-treated cells (), WNK1 became redistributed to the photobleached area within 2 min. This was slower than in untreated cells, where the recovery occurred within 0.5 min (unpublished data). We observed that the C-terminal noncatalytic domain of WNK1 comprising residues 1504–2382 fused to GFP redistributed to intracellular vesicles after sorbitol stimulation, similar to full-length GFP-WNK1 (). In contrast, the N-terminal fragment (residues 1–667) comprising the catalytic domain remained diffusely localized throughout the cytoplasm after sorbitol stimulation. Our findings are consistent with previous reports (; ) suggesting that endogenously expressed WNK1 is specifically activated by hyperosmotic conditions (sorbitol, NaCl, and KCl) and not by other stresses (HO, anisomycin, and phosphatase inhibitors), growth factors, cytokines, phorbol esters, or the diuretic hypotensive agent bumetanide (). The activation of WNK1 by sorbitol is rapid and observed within 0.5 min, the earliest time point that we can practically investigate. Our data indicate that hyperosmotic stress activates WNK1 by stimulating its phosphorylation as incubation of WNK1 with PP1γ lead to a decrease in its activity. Phosphopeptide mapping resulted in the identification of six residues on endogenous WNK1 whose phosphorylation is stimulated with sorbitol (). Two of these sites are located N-terminal to the kinase domain (Ser15 and Ser167), one within the T-loop of the kinase domain (Ser382), and the other three sites are within the C-terminal noncatalytic region (Ser1261, Thr1848, and Ser2372). In addition, we have identified three other phosphorylation sites within the C-terminal region of WNK1 (Ser2012, Ser2029, and Ser2032) that are constitutively phosphorylated. We cannot rule out the possibility that there are additional phosphorylation sites on WNK1 that we have not been able to identify. Apart from Ser15, all the phosphorylation residues identified are conserved in mouse and rat WNK1. In WNK1, Ser382 and Ser2372 are conserved, whereas in WNK1, only Ser382 is conserved. The only phosphorylation site that is present in all human WNK isoforms is Ser382, and the residues surrounding this site are also identical in all WNK isoforms. Ser2372 is found in WNK2 and -3, but not WNK4. Alignment of the sequences surrounding the WNK1 phosphorylation sites () indicates that they are quite distinct, suggesting that different upstream kinases may be phosphorylating these residues in vivo. The only similarity between three of the phosphorylation sites (Ser1261, Ser2029, and Ser2032) is that they possess a proline residue following the site of phosphorylation. It would be interesting to investigate whether these residues were phosphorylated by a proline-directed kinase, such as an isoform of p38 or JNK. If this was the case, the phosphorylation of these sites may not be regulating WNK1 activity, as inhibitors of p38–JNK–ERK pathways did not prevent WNK1 activation (). Our results suggest that phosphorylation of Ser382 is required for sorbitol-induced activation of WNK1, as mutation of Ser382 to Ala prevented WNK1 activation by sorbitol, whereas mutation of Ser382 to Glu to mimic phosphorylation increased activity of WNK1 and prevented further activation by sorbitol (). The conservation of Ser382 in all species of WNK isoforms is consistent with the notion that phosphorylation of this residue plays a crucial role in controlling the activity of WNK isoforms. It was previously reported that the isolated catalytic domain of WNK1 encompassing residues 198–491, when expressed in , was phosphorylated at Ser378 and Ser382 (). This study showed that mutation of Ser382 to Ala inactivated the WNK1 enzyme, whereas mutation of Ser378 to Ala only moderately reduced activity (). In our peptide-mapping studies with endogenously expressed full-length WNK1 we have not been able to detect phosphorylation of Ser378. Our findings suggest that, in mammalian cells, phosphorylation of Ser382 may be mediated by a transautophosphorylation reaction. This conclusion is based on the finding that wild-type WNK1, but not kinase-inactive WNK1, when expressed in , is phosphorylated at Ser382 (). In contrast, we observe that a catalytically inactive WNK1 mutant is normally phosphorylated at Ser382 in response to sorbitol when expressed in 293 cells (). These observations could be explained by the ability of endogenous WNK1 to transphosphorylate the kinase-inactive WNK1 mutant at Ser382 (). However, our data do not rule out the possibility that there is another upstream kinase capable of phosphorylating Ser382 in response to hyperosmotic stress. Nor do our data rule out the possibility that hyperosmotic stress stimulates phosphorylation of Ser382 by inhibiting a protein phosphatase. Further work is required to establish the roles of the novel sites of phosphorylation on WNK1 that we have identified. Apart from Ser382 (), we have also analyzed the effects that mutating other sorbitol-stimulated phosphorylation sites (Ser167, Ser1261, Thr1848, and Ser2372) had on regulating the activity of WNK1 expressed in 293 cells. We observed that individual mutation of these residues to either Ala or Glu did not markedly affect basal or sorbitol-stimulated WNK1 activity (unpublished data). We suggest that phosphorylation of Ser1261 inhibits the interaction with the CCT domain of SPAK and OSR1 (). There are four RFXV/I motifs in the C-terminal region of WNK1, and a Ser or Thr residue follows all of these potential CCT binding sites. Interestingly, a Ser/Thr residue also follows several other RFXV/I motifs in other proposed SPAK/OSR1 binding proteins, including WNK4, NKCC2, and NCC. It is possible that phosphorylation of residues following RFXV/I motifs comprises a mechanism for dissociating WNK1 from the CCT domains of SPAK and OSR1. We have attempted to detect dissociation of SPAK/OSR1 from WNK1 after sorbitol stimulation of cells but have not observed a marked decrease in the association between SPAK/OSR1 and WNK1. It is possible that nonstoichiometric phosphorylation of Ser1261 and/or the presence of other RFQV/I SPAK/OSR1 binding sites in WNK1 masks detection of the dissociation of SPAK/OSR1 from WNK1 in such an experiment. We establish that hyperosmotic stress induces activation and phosphorylation of endogenous SPAK and OSR1 at their T-loop and S-motif, the sites phosphorylated by WNK1 in vitro (). Moreover, depletion of WNK1 by ∼90% using siRNA methodology markedly inhibited T-loop and S-motif phosphorylation of SPAK/OSR1 and repressed basal and sorbitol-induced activation of these enzymes. Other isoforms of WNK or the remaining low level of WNK1 in these cells may mediate the residual SPAK/OSR1 phosphorylation and activity. These data provide further evidence that SPAK/OSR1 are indeed regulated by WNK1 in vivo. While this study was under review, it was reported that siRNA knock down of WNK1 reduced OSR1 activity as well as its total phosphorylation in sorbitol-treated cells, but the sites of phosphorylation affected by WNK1 knockdown were not investigated in this study (). Our mutational analysis also confirms that phosphorylation of the T-loop of SPAK/OSR1 is required for sorbitol-induced activation of these enzymes (). Moreover, we find that individual mutation of the T-loop or S-motif does not affect phosphorylation of the other residue in sorbitol-stimulated cells. We observed that hyperosmotic stress (sorbitol and NaCl) induced a striking relocalization of WNK1 to intracellular vesicles that are highly mobile (Videos 1 and 2) and colocalize with clathrin, AP-1, and partially with TGN46, but not with the endosomal markers EEA1, Hrs, LAMP1, or AP-2, which colocalizes with plasma membrane–coated pits ( and ). These results are consistent with the notion that after hyperosmotic stress, a considerable pool of WNK1 is localized to TGN/recycling endosomes. We have not been able to demonstrate that endogenous clathrin and WNK1 coimmunoprecipitate with each other from sorbitol-stimulated cells (unpublished data), indicating that these proteins may not interact directly. The trafficking of several ion channels and cotransporters between intracellular vesicles and plasma membrane (see Introduction) is strongly influenced by the overexpression of WNK isoforms. It is therefore possible that relocalization of WNK1 to TGN/recycling endosomes may enable it to regulate trafficking and/or activity of certain ion channels/cotransporters. Our findings based on the overexpression of fragments of WNK1 in 293 cells indicate that a C-terminal noncatalytic region of WNK1 mediates this relocalization (). Overexpression of catalytically inactive WNK1 or -4 decreased the membrane expression of the renal outer medullary potassium channel (ROMK) through a clathrin-dependent endocytosis mechanism (; ). Overexpression of the C-terminal noncatalytic region of WNK3 decreased membrane expression of the potassium channel (), and the C-terminal portion of WNK4 coimmunoprecipitated with ROMK in 293 cells (). Overall, these data suggest that the C-terminal portion of the WNK isoforms plays an important role in influencing WNK cellular localization and function. In conclusion, our study defines the striking effects that exposure of cells to hyperosmotic conditions has on WNK1 phosphorylation, cellular localization, and catalytic activity as well as on its ability to interact with and activate its substrates SPAK and OSR1. In future studies, it would be important to determine whether the capacity of hyperosmotic stress to stimulate WNK1 and SPAK/OSR1 plays a role in controlling cell volume and blood pressure. It will also be interesting to establish whether the reported effects that WNK isoforms have on ion channels and other cotransporters are mediated through activation of SPAK/OSR1 and/or translocation of WNK1 to TGN/recycling endosomes. It would also be essential to address whether WNK2, -3, and -4 isoforms are regulated in a manner similar to WNK1. The PKB (total) antibody was raised in sheep against a peptide encompassing residues 466–480 of rat PKBα (RPMFPQFSYSASGTA). The following antibodies were purchased from Cell Signaling: ERK1/2 (total), ERK1/2 phospho-Thr202/Tyr204, p38α (total), p38α phospho-Thr180/Tyr182, and PKB phospho-Thr308. The JNK (total) and JNK phospho-Thr183/Tyr185 antibodies were obtained from Biosource International. The mouse monoclonal antibody recognizing the GST tag was purchased from Roche. Secondary antibodies coupled to horseradish peroxidase used for immunoblotting were obtained from Pierce Chemical Co. For Li-COR analysis, the IRDye 800-conjugated anti-sheep antibody was purchased from Rockland. Preimmune IgG used in control immunoprecipitation experiments were affinity purified from preimmune serum using protein G–Sepharose. The clathrin antibody was purchased from Abcam, the EEA1 and LAMP1 antibodies (raised in mouse) were purchased from BD Biosciences, and the Hrs antibody was a gift from H. Stenmark (The Norwegian Radium Hospital, Oslo, Norway) and has been described previously (). The AP-1 antibody (monoclonal anti–γ-adaptin AP6) was obtained from Sigma-Aldrich. The AP-2 antibody (monoclonal anti–α-adaptin) was obtained from Affinity BioReagents, Inc. The TGN46 antibody (polyclonal produced in sheep) was purchased from Serotec. Alexa Fluor 488 donkey anti–sheep, Alexa Fluor 595 donkey anti–rabbit, and Alexa Fluor 595 donkey anti–mouse secondary antibodies were obtained from Invitrogen. Anti-WNK1(CT) and preimmune IgG antibodies were covalently coupled to protein G–Sepharose (1 μg of antibody per 1 μl of beads) using a dimethyl pimelimidate cross-linking procedure. 0.5 mg of clarified cell lysate was incubated with 5 μg of anti-WNK1(CT) or preimmune IgG antibody conjugated to 5 μl of protein G–Sepharose and incubated for 1 h at 4°C with gentle agitation. The immunoprecipitates were washed twice with 1 ml of lysis buffer containing 0.5 M NaCl and twice with 1 ml of buffer A. The in vitro phosphorylation reaction mix contained a final volume of 25 μl in buffer A containing 5 μM OSR1[D164A], 0.1 mM [γP]ATP, and 10 mM magnesium acetate. The assays were performed for 20 min at 30°C, and the reactions were terminated by adding SDS sample buffer. The samples were electrophoresed on a polyacrylamide gel, which was stained with Coomassie blue, dried, and autoradiographed. The OSR1 Coomassie bands were excised, and incorporation of P-radioactivity was quantified by Cerenkov counting. 1 U of activity was defined as the amount of WNK1 that incorporated 1 nmol of P into OSR1[D164A]. For experiments in using overexpressed forms of GST-WNK1, WNK1 was affinity purified from 0.1 mg of cell lysate using 5 μl of glutathione–Sepharose. The beads were washed, and assays were undertaken as described above. 3 mg of clarified cell lysate was incubated with 5 μg of the SPAK/OSR1 (total) antibody conjugated to 5 μl of protein G–Sepharose and incubated for 1 h at 4°C with gentle agitation. The immunoprecipitates were washed twice with 1 ml of lysis buffer containing 0.5 M NaCl and twice with 1 ml of buffer A. The SPAK/OSR1 immunoprecipitates were either assayed with the CATCHtide peptide substrate (RRHYYYDTHTNTYYLRTFGHNTRR) that encompasses the SPAK/OSR1 phosphorylation sites on NKCC1 () or using the N-terminal fragment of NKCC1 encompassing residues 1–260 (NKCC1[1–260]; ). Assays were set up in a total volume of either 50 μl (CATCHtide assay) or 25 μl (NKCC1 assay) in buffer A containing 10 mM MgCl, 0.1 mM [γP]ATP (∼300 cpm/pmol) and 300 μM CATCHtide (RRHYYYDTHTNTYYLRTFGHNTRR), or 5 μM NKCC1[1–260]. After incubation for 10–60 min at 30°C, the reaction mixture was applied onto P81 phosphocellulose paper (for the CATCHtide assay), the papers were washed in phosphoric acid, and incorporation of P-radioactivity in CATCHtide was quantified. For the NKCC1 assay, the reaction was stopped by the addition of SDS sample buffer. The samples were electrophoresed on a polyacrylamide gel, which was stained with Coomassie blue, dried, and autoradiographed. The NKCC1[1–260] Coomassie bands were excised, and incorporation of P-radioactivity was quantified by Cerenkov counting. For experiments in using overexpressed forms of GST-OSR1 and -SPAK, the GST fusion proteins were affinity purified from 0.5 mg of cell lysate using 5 μl of glutathione–Sepharose. The beads were washed, and assays were undertaken as described above, using NKCC1[1–260] as a substrate. 293 cells stably expressing GFP-WNK1 at levels similar to that of endogenous WNK1 were described previously (). In experiments in which the cells were fixed before analysis, the cells were grown on coverslips (no. 1 1/2) and, after stimulation, were fixed for 5 min in freshly prepared 3% vol/vol formaldehyde in PBS. The cells were washed twice in PBS (5 min each wash) and once in PBS containing 0.2% wt/vol Triton X-100, incubated 10 min in PBS-TG (PBS containing 0.2% wt/vol Tween +3% wt/vol of fish skin gelatine), and incubated for 1 h with anti-clathrin/Hrs/AP-2/TGN46 antibodies (diluted 1:1,000 in PBS-TG), anti-EEA1 antibody (diluted 1:100 in PBS-TG), anti-LAMP1 antibody (diluted 1:50 in PBS-TG), or anti–AP-1 (diluted 1:2,500 in PBS-TG). Cells were washed three times in PBS-T (PBS containing 0.2% wt/vol Tween), incubated for 30 min with the secondary antibody diluted 1:500 in PBS-TG, washed three times in PBS-T and once in water, and mounted onto slides using hydromount (National Diagnostics). Images were collected using the α Plan 100× 1.45 NA Plan-Fluor objective on a confocal microscope (LSM 510 META; Carl Zeiss MicroImaging, Inc.). Fixed cells were imaged at room temperature (∼20°C) and live cells at 37°C. Images were acquired using LSM 510 acquisition software (Carl Zeiss MicroImaging, Inc.), and no further processing of images was performed apart from assembling montages in Photoshop/Illustrator (Adobe). Videos were edited in QuickTime Pro. For imaging of endogenous WNK1, untransfected 293 cells were treated as described above except that 5 μg/ml of anti-WNK1(Total) antibody was used. For live-cell imaging, 293 cells stably expressing GFP-WNK1 were grown on 35-mm-diameter glass-bottomed dishes (Willco). Cells were maintained at 37°C and 5% CO by the use of a microscopy incubator chamber (Carl Zeiss MicroImaging, Inc.). For each cell, optical sections of 0.5 μm were recorded at 2-s (for Videos 1 and 2 and Figs. S2 and S3) or 10-s () intervals. FRAP was performed using a bleached region of interest defined by the LSM 510 META software. Recovery of fluorescence was monitored by collecting images every second until recovery was complete. See the supplemental text (available at ) for further methodological details. The supplemental text contains additional methodological details on materials, buffers, and DNA constructs used, as well as protocols used for in vivo P-labeling, identification of phosphorylation sites by mass spectrometry, immunoprecipitation, immunoblotting, and siRNA knockdown protocol. Fig. S1 shows that hyperosmotic stress leads to activation and phosphorylation of SPAK and OSR1 in 293 cells. Figs. S2 and S3 show selected time frames of videos of cells treated with sorbitol (Fig. S2 and Video 1) or NaCl (Fig. S3 and Video 2) to illustrate movement of vesicles to which GFP-WNK1 is recruited. Online supplemental material is available at .

Guided axonal growth is essential for both the initial wiring of neuronal circuitry during development and the regeneration of synaptic connections in the adult nervous system after injury and diseases (; ; ; ). The directional motility of the growth cone at axonal tips is regulated by a variety of environmental factors that either promote/attract or inhibit/repel the axonal elongation (; ). Although many families of guidance ligands and receptors have been recently identified (; ; ), the intricate signaling cascades that control and regulate axonal growth and guidance remain to be fully understood. The second messenger, cAMP, represents an important intracellular signal that exhibits profound effects on growth cone motility and guidance. Previous studies have linked elevated cAMP signaling with enhanced elongation of growth cones (; ; ). The importance of cAMP regulation of axonal growth is further augmented by recent findings that manipulating the cAMP signaling pathway can overwrite the inhibitory/repulsive effects of some extracellular molecules on axonal growth, even converting them to attractive/positive responses (). For instance, an elevation of cAMP levels has been shown to convert myelin-associated glycoprotein (MAG)–induced growth cone repulsion to attraction in culture and promote axonal regeneration in vivo (; ; ). Therefore, the cAMP pathway could be a potential target for therapeutic intervention to promote nerve regeneration after injury and degeneration (; ). At present, the exact signaling mechanisms underlying cAMP effects on growth cones remain unclear. The existence of the intricate cross talk of cAMP to other signaling pathways has added more complexity to this issue. For example, both Ca and cAMP are key second messengers involved in growth cone guidance by several extracellular cues, and Ca-dependent turning responses can be modulated by the cAMP pathway: the elevation of cAMP levels dictates attraction, whereas the inhibition of PKA results in repulsion (). It has been proposed that cAMP signaling could affect the Ca signals elicited by extracellular cues through the modification of voltage-dependent Ca channels or Ca release from the intracellular Ca stores (; ; ). Our recent work suggests that PKA targets a downstream component in the Ca signaling pathway, protein phosphatase-1 (PP1), to allow the switching of repulsion to attraction (). It is conceivable that cAMP could act at multiple steps in the Ca signaling pathway to affect growth cone behaviors, but how it specifically targets distinct downstream effectors remains to be investigated. The cAMP molecule can diffuse over a long distance in the cytosol to activate a wide range of effectors (), and its major effector, PKA, is a multifunctional enzyme with a broad substrate specificity (). Therefore, the mechanisms for spatiotemporal selectivity and efficiency in cAMP/PKA signaling are of particular interest. Between two major subtypes of PKA, type II PKA is often localized to subcellular compartments for coupling to specific downstream targets through a large family of AKAPs (a kinase-anchoring proteins; ; for review see ). Such spatial targeting of PKA to specific cellular locations and signaling partners through the interaction of PKA regulatory subunits with AKAPs (; ) has been demonstrated to be crucial for many cellular functions (; ; for review see ), including PKA regulation of muscle contractibility () and synaptic plasticity (). Whether the spatial targeting of PKA is important for guidance signaling in growth cones is not clear. A recent study of axon guidance in suggests that the plexin A–binding protein Nervy functions as an AKAP to antagonize semaphorin 1A–plexin A-mediated repulsion by linking cAMP/PKA to plexin A receptor (). In the present study, we used cultured embryonic neurons to dissect the cAMP signaling mechanisms. We first investigated the subcellular distribution of the two major PKA subtypes in growth cones and found that type II, not type I, PKA was highly enriched in filopodia. Disruption of the filopodial localization of type II PKA abolished cAMP effects on growth cone guidance. Next, we identified a PP1 regulatory protein, inhibitor-1 (I-1), as the target of PKA in cAMP regulation of growth cone turning responses to several guidance molecules. Furthermore, we observed a colocalization of type II PKA, I-1, and PP1 in growth cone filopodia, indicating that the spatial coupling of these signaling components is important for cAMP regulation of growth cone responses. Finally, we found that I-1 and PP1 mediated growth cone repulsion induced by MAG. These findings indicate that distinct subcellular localization of type II PKA represents an important mechanism for specific cAMP/PKA signaling in growth cone guidance. To examine the spatial distribution of two major PKA subtypes, type I and II, in neurons, antibodies against their regulatory subunits (RI and RII) were used for immunofluorescent labeling. The reactivity and specificity of these antibodies to tissues were confirmed by Western blotting (). Double immunofluorescence shows that both subtypes of PKA were present in spinal neurons but with remarkably different spatial patterns (). Type I PKA was distributed evenly throughout the soma, the axonal shaft, and the central region of the growth cone (). In contrast, type II PKA was highly enriched in the peripheral region of the growth cone and, strikingly, in filopodia (). A close examination of the growth cones at a higher resolution shows that RII was predominantly concentrated in filopodia where RI was least abundant (). The distinct spatial patterns of PKA subtypes are clearly evidenced in color images merged from two channels (; red, RII; green, RI). If the RII antibody was premixed with saturating recombinant human RIIβ in otherwise the same immunostaining, the RII immunofluorescence in filopodia disappeared. However, RI staining was largely unaffected (), further confirming the specificity of the RII antibody in our immunostaining. To quantitatively demonstrate the filopodial enrichment of PKA RII, we measured the immunofluorescence intensity of RI and RII in three regions of the neuron: the filopodia, the growth cone (excluding filopodia), and the adjacent axonal shaft (see the schematic diagram in ). The measurements from the filopodia and the growth cone were normalized against that from the axonal shaft (). The results show that the filopodial RII immunofluorescence was approximately threefold stronger than that of the axonal shaft, whereas the RI immunofluorescence was the lowest in filopodia (). Similar spatial patterns of RII and RI were also observed in cultured rat hippocampal neurons (Fig. S1, available at ; and see ). Together, our findings indicate a spatial enrichment of type II PKA in growth cone filopodia that could play an important role in growth cone guidance. The type II PKA is often targeted to subcellular domains by AKAPs (; for review see ). Using a previously described PKA RII overlay approach (), we found several bands from neural tube tissues that were eliminated by Ht31 but not Ht31P peptides (Fig. S2, available at ). Ht31 encodes the PKA RII–binding sequence of a human thyroid AKAP () and is widely used for the competitive and specific disruption of PKA RII–AKAP binding, whereas Ht31P is the negative control derived from Ht31 with two amino acids mutated (). This result suggests the presence of AKAPs in neurons. To directly test the role of AKAPs in PKA localization in growth cones, we treated the neuronal culture with a membrane-permeable Ht31 peptide (stearated Ht31 [St-Ht31]) and found that the filopodial localization of RII was greatly reduced (), whereas the control peptide St-Ht31P was ineffective (). Quantitative analysis shows that St-Ht31, not St-Ht31P, largely reduced RII enrichment in the filopodia (). Importantly, the RI distribution remained unaffected under both St-Ht31 and St-Ht31P treatments (). It should be noted that Ht31 does not affect antibody binding to RII because Western blotting of endogenous RII was not affected by St-Ht31 (unpublished data). Therefore, these results support the notion that type II PKA is enriched in growth cone filopodia through AKAPs. Because filopodia are involved in directional responses of growth cones to guidance cues (; ; ; ), the localization of type II PKA in filopodia suggests a potential role in growth cone guidance. To test this notion, growth cone turning assays were performed in which a guidance gradient was established by pulsatile pressure ejection of a guidance solution from a micropipette (; ). Relatively large and motile growth cones of cultured spinal neurons on laminin were used for assessing turning responses to several extracellular gradients (; ). We first examined growth cone attraction induced by a gradient of membrane-permeant cAMP analogues (; ). A Sp-cAMPS gradient (20 mM in the ejection pipette and ∼20 μM reaching the growth cone) induced a marked attraction of growth cones within 30 min (). When 2 μM St-Ht31 was added to the bath, the attractive response to the cAMP gradient was abolished (), whereas 2 μM Ht31P had no effect (). We quantified the turning response of each growth cone by measuring its turning angle and net length of extension. A majority of the growth cones turned toward the Sp-cAMPS pipette (attraction) to bear a positive turning angle, which is depicted by the cumulative distribution of the turning angles (, top). In the presence of St-Ht31, growth cones did not exhibit any directional preference during the 30-min exposure to the Sp-cAMPS gradient (, top). The effective blockade of cAMP-induced attraction by St-Ht31 is also highlighted by the average turning angle from all of the growth cones examined (). For all of these manipulations, the growth cone extension remained similar (), indicating that only the directional response of growth cones was affected. We also examined growth cone attraction induced by a gradient of pituitary adenylate cyclase–activating polypeptide (PACAP), a neuropeptide that is known to activate G protein–coupled receptors to elevate cytosolic cAMP levels (). As we reported previously (), a PACAP gradient (1 μM in the pipette and ∼1 nM at the growth cone) induced a marked attraction of growth cones (, bottom). However, the attraction was completely attenuated by the bath application of St-Ht31 but not by St-Ht31P ( [bottom] and ). Similarly, the net growth cone extension under these manipulations did not show any significant difference (P > 0.05; ). To verify that the blockade of cAMP-induced turning was not a result of the nonspecific effect of Ht31, we examined growth cone attraction induced by a gradient of neurotrophin-3 (NT-3), which has been shown to be independent of cAMP signaling (; ). We found that St-Ht31 treatment had no influence on NT-3–induced attraction (). Collectively, our data indicate that growth cone attraction induced by local cAMP signaling depends on the spatial localization of type II PKA via AKAPs. One profound effect of the cAMP pathway on Ca-dependent growth cone guidance is to switch repulsion to attraction through PKA activation or vice versa (). We examined the role of type II PKA and its spatial localization in the switching of turning responses. growth cones plated on the laminin surface exhibited an attractive response to a gradient of brain-derived neurotrophic factor (BDNF; 50 μg/ml in the pipette and ∼50 ng/ml at the growth cone), which could be converted to repulsion by PKA inhibition through the bath application of 20 μM Rp-cAMPS (; ; ). We found that the bath application of 2 μM St-Ht31 led to growth cone repulsion in response to the same BDNF gradient, whereas St-Ht31P had no effect (). Considering that BDNF-induced attraction was similarly converted to repulsion by either PKA inhibition (Rp-cAMPS) or the blockade of PKA–AKAP interaction (Ht31), it is likely that type II PKA and its localized activity in filopodia are required to support the attractive growth cone response to BDNF. We next examined growth cone repulsion in response to a gradient of MAG, a key inhibitory molecule in axon regeneration after injury in the central nervous system (). In our cultures, a gradient of MAG effectively induced the repulsive turning of most growth cones to result in a negative average turning angle (). PKA activation by Sp-cAMPS was able to convert MAG repulsion to attraction, resulting in a positive average turning angle (; ). Bath application of either 2 μM St-Ht31 or St-Ht31P alone did not affect the repulsion induced by MAG gradients. However, St-Ht31 was able to abolish the switching of MAG repulsion to attraction by Sp-cAMPS (). Consistently, growth cone extension was not significantly affected by these treatments (P > 0.05; ). Finally, we found that NT-3– induced attraction was not affected by Ht31 treatment but was switched to repulsion by Rp-cGMPS (). These results are consistent with previous findings that the guidance effects of BDNF and MAG involve Ca signals and can be modulated by the cAMP pathway, whereas NT-3 is independent of Ca and can only be modulated by the cGMP pathway (). Our findings further indicate that the spatial localization of type II PKA (and its activity) in growth cone filopodia through AKAPs is required for Ca-dependent attractive responses. Localization of PKA via AKAPs is thought to couple PKA activity to specific downstream targets, a mechanism that is believed to control cAMP spatiotemporal signaling in the cell (; for review see ). Our previous study suggests that PKA regulates Ca-dependent growth cone turning through the inhibition of PP1 (). PP1 activity is known to be regulated by a regulatory protein, I-1, whose phosphorylation by PKA at Thr35 leads to PP1 inhibition, and dephosphorylation by calcineurin (CaN) relieves the inhibition (). Because PP1 inhibition appears to be important for cAMP switching of Ca-dependent repulsion to attraction (), we tested whether I-1 is a key substrate for spatially localized type II PKA in guidance regulation. Because our cultures typically contain a large population of myocytes, we used rat hippocampal cultures for biochemical analysis of I-1 phosphorylation and its dependence on AKAPs. At first, we performed immunofluorescent staining and found that PKA RII was also enriched in the growth cone filopodia of cultured hippocampal neurons (2–3 d in vitro; ), which is consistent with our observations on neurons. Using an antibody recognizing I-1, we found that I-1 was also enriched in filopodia and colocalized with PKA RII (). Western blot analysis using a phosphospecific I-1 antibody showed that the level of phosphorylated I-1 was very low at rest but was markedly increased upon forskolin treatment for cAMP elevation (20 μM for 20 min). The PKA phosphorylation of I-1 was largely abolished when the cells were incubated with St-Ht31 but not St-Ht31P for 30 min before forskolin stimulation (). Quantification of the data from three independent blotting analyses shows that the phosphorylation of I-1 by forskolin treatment was reduced by ∼90% upon Ht31 disruption of the RII–AKAP interaction (). These results indicate that spatially anchored type II PKA accounts for a large part of I-1 phosphorylation in neurons, particularly in the growth cone filopodia. We next examined the distribution of I-1 in growth cones and its role in guidance. Similar to hippocampal neurons, I-1 was found to be concentrated in filopodia and colocalized with PKA RII (). To understand the function of I-1 in growth cone guidance, we used I-1 morpholino (IM) antisense oligonucleotides to knock down the I-1 level in neurons through the microinjection of one-cell–stage embryos. As the control, a morpholino oligonucleotide of scrambled sequence (control morpholino [CM]) was used. We found that IM effectively knocked down the expression level of I-1 in injected embryos around stage 20, as indicated by Western blotting using the whole embryo lysates (). In contrast, CM injection did not affect the endogenous I-1 level. Importantly, IM did not change the expression levels of PKA RII and PP1γ (), demonstrating the specificity of I-1 knockdown by the morpholino approach. To test the functional role of I-1 in growth cone guidance, spinal neuron cultures were prepared from both IM- and CM-injected embryos, and cultured neurons exhibiting the fluorescence of coinjected Oregon green–dextran () were selected for turning assay. Immunostaining of I-1 in IM-injected cells did not detect a substantial level of I-1, supporting the Western blot result that I-1 was greatly reduced by IM (). When we exposed these I-1 knockdown growth cones to a gradient of Sp-cAMPS or PACAP, no preferential turning was induced (). On the other hand, growth cones of CM-injected neurons responded positively to the cAMP and PACAP gradients (). Again, the average lengths of growth cone extension were not significantly different among various groups (P > 0.05; ). Therefore, these results provide evidence that I-1 is a key molecule that mediates cAMP-induced growth cone attraction. Given that I-1 and RII are colocalized in filopodia, spatial targeting of PKA to I-1 in filopodia may be a crucial step in cAMP/PKA signaling that regulates growth cone turning. Many AKAPs not only bind PKA holoenzymes but also recruit other molecules to create signaling complexes, which is believed to be important for signal integration as well as for the efficiency and specificity of PKA activity (; for review see ). We hypothesized that PKA inhibition of PP1 through I-1 may also involve a similar mechanism to regulate growth cone responses to guidance molecules. We first performed coimmunoprecipitation (co-IP) using neural tube tissues to examine whether type II PKA and PP1 were present in a signaling complex. Using specific antibodies against PKA RII and PP1γ, we found that a small amount of PP1γ could be consistently pulled down by RII (but not by RI) and vice versa (). Importantly, the PP1γ signal decreased when co-IP was performed in the presence of St-Ht31 (). In support of the co-IP data, double immunofluorescence labeling also revealed the substantial colocalization of PP1γ with PKA RII in growth cone filopodia, although PP1γ fluorescence was also distributed in the axon (). Together with the aforementioned results, these observations suggest that spatially restricted formation of the signaling complex involving PKA, I-1, and PP1 in filopodia constitutes an important mechanism for the efficient and specific cAMP signaling that regulates growth cone turning. To further test the model that PKA exerts its regulation on growth cone turning primarily through the inhibition of PP1, we used the turning assay to examine the effects of the direct inhibition of PP1 using specific inhibitors. We first tested the notion that growth cone attraction induced by a gradient of cAMP involves a local (asymmetric) inhibition of PP1. Here, a gradient of the PP1 inhibitor tautomycin was generated by pulsatile ejection of 3 μM tautomycin solution from the pipette (estimated ∼3 nM at the growth cone). We found that asymmetric inhibition of PP1 by the tautomycin gradient elicited a marked attractive turning of growth cones (). As a crucial control, we also examined the attractive response to the tautomycin gradient in neurons containing morpholino oligonucleotides and found that neither IM nor CM affected the attraction (). Such results are expected because tautomycin directly inhibits PP1 without the involvement of I-1. To confirm that local PKA activation and PP1 inhibition are in the same signaling pathway to induce growth cone turning, we performed a series of cross-desensitization experiments. In the first set, we examined tautomycin-induced attraction in the presence of various cyclic nucleotides in bath. Tautomycin was found to lose its ability to attract growth cones only after Sp-cAMPS application (), supporting the notion that cAMP/PKA targets PP1 activity. In the second set of experiments, growth cone attraction induced by Sp-cAMPS or PACAP gradients was performed with or without tautomycin in bath. We found that the global inhibition of PP1 by tautomycin abolished the turning responses induced by both Sp-cAMPS and PACAP gradients (). In contrast, 1 nM of the PP2A inhibitor okadaic acid (OA) and 10 nM of the CaN inhibitor deltamethrin did not affect these turning responses (). Collectively, these findings support the idea that the asymmetric inhibition of PP1 at the growth cone by cAMP gradients mediates the attractive turning responses. MAG is a key inhibitory molecule in axon regeneration in the central nervous system (). MAG-induced growth cone repulsion has been shown to involve Ca and can be switched to attraction by cAMP (; ). In our cultures, a gradient of MAG consistently induced growth cone repulsion, and the repulsion could be converted to attraction by the bath application of Sp-cAMPS (, top). Importantly, MAG-induced growth cone repulsion was also effectively converted to attraction by PP1 inhibition through the bath application of 3 nM tautomycin. The low concentration of tautomycin we used (3 nM) is known to mainly inhibit PP1, although higher concentrations of tautomycin could inhibit another protein phosphatase, PP2A (IC = 10 nM). To confirm that the conversion of MAG repulsion to attraction by tautomycin specifically involved PP1 inhibition, we examined MAG-induced repulsion when PP2A was inhibited by 1 nM OA. Our data show that OA did not affect MAG-induced repulsion (, top). The average turning angles and lengths of growth cone extension in these groups are summarized in the bar graphs for comparison (). It is clear that either PKA activation by Sp-cAMPS or PP1 inhibition by tautomycin converted the MAG-induced repulsion to attraction in a similar way, suggesting that PP1 is a target of the cAMP regulation of MAG repulsion. Interestingly, when the endogenous I-1 in neurons was greatly reduced by IM, MAG-induced repulsion was completely abolished, whereas CM did not affect MAG-induced repulsion (). The complete blockade of MAG-induced repulsion together with data on the involvement of PP1 indicates that I-1 and PP1 are involved in signal transduction in MAG repulsion. The major cAMP pathway is believed to involve the activation of PKA, which, in turn, phosphorylates a variety of downstream substrates for distinct signaling cascades. It is interesting to note that cAMP molecules can diffuse over long distances in the cytosol (), and PKA does not possess a high degree of substrate selectivity (). Thus, the target specificity and phosphorylation efficiency of PKA in a distinct signaling pathway are believed to involve spatial coupling of PKA to its appropriate downstream effectors through a large family of AKAPs (; for review see ). It has been shown that type II PKA is the major subtype that associates with AKAPs and often exhibits localized subcellular distribution (; for review see ). In the present study, we present evidence that spatially distributed type II PKA is involved in growth cone guidance. The striking localization of type II PKA in growth cone filopodia indicates that it may be involved in regulation of the actin cytoskeleton dynamics underlying growth cone motility and guidance. Using the turning assay, we have provided direct evidence that the spatial localization of type II PKA in filopodia is required for cAMP-induced growth cone turning and switching of turning responses to guidance gradients. Furthermore, we have obtained evidence that PKA acts on PP1 through I-1 (all localized in filopodia) to regulate growth cone turning responses. These findings indicate an exciting model in which spatial targeting of PKA to its downstream targets in growth cone filopodia allows the profound cAMP regulation of growth cone responses to guidance cues. There is a large body of studies demonstrating that AKAPs are the major family of scaffolding proteins for the subcellular localization of PKA molecules (for review see ). The predominant localization of type II PKA in growth cone filopodia is likely mediated by AKAPs because it was diminished by the inhibitory peptide St-Ht31 (). Moreover, the peptide abolished cAMP-induced growth cone turning and switching of guidance responses ( and ). These results evince that AKAP-mediated PKA localization in filopodia is crucial for local cAMP signaling during growth cone guidance. In support of this notion, we found that a gradient of St-Ht31 induced repulsive turning in the presence of bath Sp-cAMPS (), indicating that local cAMP signaling can be achieved by the spatial regulation of PKA–AKAP interaction. It should be noted that a much higher concentration of St-Ht31 was used to diminish the localization of PKA RII in filopodia than that used to block turning responses. This is likely caused by the fact that Ht31 competes for PKA RII binding and that a higher concentration is required to remove PKA RII molecules that have already bound to AKAPs at the filopodia. On the other hand, cAMP regulation of growth cone turning may require dynamic PKA–AKAP interactions, which could be affected by low concentrations of the Ht31 peptides. This intriguing notion will be tested in future studies. Another interesting observation came from the immunostaining experiment in which anti-RII antibodies were premixed with recombinant RII proteins (). Although filopodial staining was eliminated, strong signals in the axonal shaft were detected, which is a reversal of the results from standard immunostaining using only the anti-RII antibody (). Because unoccupied AKAPs in fixed cells could bind RII in situ (), this observation suggests the availability of unoccupied RII-binding sites (potentially of AKAPs) in the axonal shaft but not in filopodia. At this moment, the mechanisms underlying this fascinating local enrichment of PKA RII in filopodia but not in other regions of the neuron remain unknown. It is plausible that the AKAPs in filopodia may have a much higher affinity for binding RII and/or that the affinity is spatially regulated (; for review see ). Alternatively, the expression level of PKA RII may be limited so that PKA RII is mostly recruited (or translocated) to filopodia. Currently, the exact AKAPs involved in PKA localization in growth cone filopodia are not known. Our RII overlay assay revealed several bands that might represent potential AKAPs (Fig. S2). Filopodia are actin-based motile membrane protrusions that are known to play an important role in growth cone motility and sensing of extracellular signals. The preferential enrichment of type II PKA in filopodia indicates that PKA may be associated with the actin cytoskeleton to regulate filopodial dynamics in a spatially restricted manner. A similar role of spatially anchored type II PKA has been implicated in chemotactic cells (). Moreover, our co-IP experiments suggest that type II PKA and PP1 interact (either directly or indirectly) in neurons. Thus, AKAPs that exhibit actin- and PP1-binding properties are, in principle, the good candidates (for review see ). Future identification of the AKAPs involved in filopodia localization of PKA RII will substantially advance the understanding of the mechanisms and functions of PKA spatial localization in growth cone guidance. The second major finding of this study is that the PKA inhibition of PP1 through I-1 acts as a key mechanism underlying cAMP regulation of growth cone guidance. Our previous work has shown that local activation of the CaN–PP1 pathway by Ca signals induces repulsion, and cAMP activation of PKA inhibits the CaN–PP1 repulsive pathway to allow the conversion of repulsion to attraction (). In this study, we provided evidence that I-1 is a target of PKA and CaN that oppositely controls PP1 activity for distinct turning responses (). Specifically, I-1 knockdown by morpholino abolished growth cone turning in response to local cAMP elevation (). However, I-1 knockdown did not affect growth cone turning induced by the direct local inhibition of PP1 by tautomycin (), confirming that I-1 acts upstream of PP1 in regulating growth cone turning. Dopamine- and cAMP-regulated phosphoprotein of 32 kD (DARPP-32) and I-1 are the two best-characterized PKA-activated inhibitors of PP1. I-1 is widely expressed in mammalian tissues, whereas DARPP-32 is mainly found in dopaminoceptive brain regions with very low expression levels in the spinal cord and hippocampus (). DARPP-32 is not expressed in the oocyte (), but it is not clear whether it is expressed in embryos around the developmental stages we studied. Although the antibody we used can recognize both proteins (∼30 kD), we believe that I-1 is the one that is expressed in neurons and is involved in growth cone guidance. This notion is supported by the observation that specific I-1 morpholino antisense almost completely eliminated the signals by Western blotting and immunofluorescence using the I-1/DARPP-32 antibody. MAG is an important component of inhibitory molecules in central nervous system axon regeneration, and one important insight into MAG signaling came from the observation that MAG-induced repulsion was blocked by I-1 knockdown. A previous study indicates that MAG-induced repulsive turning involves small local Ca elevations (), which likely act through CaN–PP1 (). Because CaN dephosphorylates I-1 at Thr35 (Thr34 for ) to remove its inhibition on PP1, the local Ca activation of CaN would lead to the local elevation of PP1 activity for repulsive turning. The fact that the direct inhibition of PP1 was able to convert the MAG effect to attraction provides direct support for Ca-CaN–PP1 in MAG signaling during repulsion. Morpholino knockdown of I-1 disrupts the signaling transduction from CaN to PP1, thus resulting in the blockade of repulsion. However, unlike PP1 inhibition or PKA activation, I-1 knockdown did not convert MAG-induced repulsion to attraction but instead abolished turning responses. Previous studies have indicated that a spatial balance of Ca-calmodulin–dependent kinase II (CaMKII) and PP1 activities control the bidirectional steering of the growth cones (; ; ). Importantly, Ca-dependent attraction involves the local activation of CaMKII, which can be dephosphorylated and inhibited by PP1 (; ). Therefore, the loss of I-1 by morpholino knockdown could result in elevated PP1 activity () to prevent the local CaMKII activation required for attraction (). Increasing evidence indicates that the spatiotemporal control of cascades of signaling reactions in confined subcellular locations is key for intricate signal transduction occurring in the living cells. This study has established that spatial targeting of PKA to filopodia allows the signal transduction through cAMP, type II PKA, I-1, and PP1 to regulate growth cone turning. Given the apparent localization of PKA RII, I-1, and some PP1 in filopodia, it is conceivable that interactions between type II PKA and PP1 may affect the activity of molecules that regulate the actin cytoskeleton and/or membrane-substrate adhesion. For instance, PP1 is found to directly dephosphorylate actin-depolymerizing factor/cofilin (), thereby promoting actin depolymerization. It is also possible that PP1 may modulate Ca signals through the negative regulation of Ca channels and stores () to influence Ca-dependent guidance. Our study has also provided additional support for the model that a spatial balance of CaMKII and CaN–PP1 activity controls bidirectional steering of the growth cone (). In principle, growth cone attraction can be induced by either the local elevation of CaMKII (e.g., by moderate Ca signals) or by the local inhibition of PP1 (e.g., by local cAMP/PKA activation or direct inhibition of PP1). In the latter case, baseline CaMKII activity would be required to generate a higher CaMKII/PP1 ratio on the near side of the growth cone for attraction. Indeed, global inhibition of CaMKII by KN93 abolished the attraction induced by tautomycin and cAMP gradients (unpublished data). In parallel, repulsion can be induced by either the local elevation of PP1 activity (e.g., by Ca-CaN activation) or by the local inhibition of CaMKII. Together, our findings indicate a model in which the spatial regulation of signal localization and activity balance controls the directional motility of growth cones in response to extracellular stimuli (). Antibodies against different antigens used in this study are as follows: PKA RIIβ (BD Biosciences); PKA RII and RI, PP1, and PP1γ (Santa Cruz Biotechnology, Inc.); I-1/DARPP-32 (Chemicon); and phospho–I-1/DARPP-32 (Novus Biologicals). Recombinant rat MAG-Fc was purchased from R&D Systems, and recombinant human BDNF was provided by Regeneron. PACAP was obtained from American Peptide. PKA RII proteins were provided by the laboratory of S. Taylor (University of California, San Diego, La Jolla, CA) as well as purchased from Biaffin GmbH and Co KG. St-Ht31 and St-Ht31P were purchased from Promega. Cyclic nucleotides (Sp-cAMPS, Rp-cAMPS, Sp-cGMPS, and Rp-cGMPS), tautomycin, deltamethrin, OA, and forskolin were all obtained from Calbiochem. A morpholino antisense oligonucleotide specific for I-1 was designed and synthesized by Gene Tools, LLC, with the sequence 5′-ATGGAGGCGAACAGTCCCAGGAAGA-3′. A control morpholino was used with the sequence 5′-CCTCTTACCTCAGTTACAATTTATA-3′. Embryonic cultures of spinal neurons were prepared from neural tube tissues of stage 20–22 embryos and plated on coverslips coated with poly--lysine and laminin (Sigma-Aldrich) in a serum-free medium as described previously (). The serum-free medium consisted of 50% (vol/vol) Leibovitz L-15 medium (Sigma-Aldrich), 50% (vol/vol) Ringer's solution (115 mM NaCl, 2 mM CaCl, 2.5 mM KCl, and 10 mM Hepes, pH 7.6), and 1% (wt/vol) BSA (Sigma-Aldrich). cultures were maintained at 20–22°C for 6–10 h before fixation or were used for the turning assay. In I-1 knockdown experiments, I-1 or control morpholino oligonucleotides together with a fixable Oregon green–dextran conjugate (Invitrogen) were injected into the animal pole of one-cell embryos (). The injected embryos were screened 24 h later for the presence of fluorescence and were used for cell culture. For turning assays, the cells exhibiting the green fluorescence of Oregon green–dextran were used. Primary hippocampal neurons were prepared from embryonic day 18 rat embryos (), plated on coverslips or into culture dishes coated with poly--lysine at a density of ∼600,000 cells/ml (for Western blotting) or ∼100,000 cells/ml (for immunofluorescence), and incubated at 37°C with 5% CO in MEM supplemented with 10% FBS (Invitrogen), 0.5% glucose, 1 mM sodium pyruvate, 25 μM glutamine, and 1× penicillin-streptomycin. The next day after plating, the medium was changed to the Neurobasal medium (Invitrogen) supplemented with 1× B27 supplement (Invitrogen). For immunofluorescence experiments, neurons were fixed after 2–3 d in vitro. For Western blotting, neurons were first treated with different drugs on day in vitro 5 and then were lysed in radioimmunoprecipitation buffer (50 mM Tris, 50 mM NaF, 2 mM EDTA, 150 mM NaCl, 1% Triton X-100, 0.25% sodium deoxycholate, 0.4 mM NaVO, 1 mM PMSF, 1 mg/ml leupeptin, 5 mg/ml chymostatin, 1 mg/ml pepstatin, and 5 mM E64). All studies involving vertebrate animals (frogs and rats) are performed in accordance with the National Institutes of Health (NIH) guidelines and were approved by the Institutional Animal Care and Use Committee of the University of Medicine and Dentistry of New Jersey. Cells on coverslips were fixed with 4% PFA in a cacodylate buffer () for 30 min at room temperature and were permeabilized with 0.5% Triton X-100 for 10 min. Samples were blocked overnight in 10% BSA in PBS followed by sequential incubation with primary and secondary antibodies. Coverslips were mounted on slides and visualized through a 60× NA 1.4 plan Fluor oil immersion objective (Nikon) on an inverted epifluorescence microscope (TE2000; Nikon) equipped with a CCD camera (SensiCam QE; Cooke Scientific). Fluorescence images were taken using IPLab software (version 3.7; BD Biosciences) and transferred to ImageJ software (NIH) for fluorescence intensity measurement and color coding and merging. The settings for imaging and processing were fixed throughout the experiments. Quantification of the fluorescence intensity was performed by first subtracting the background and then measuring the average intensity from the regions of interest that were hand traced using ImageJ. For the axonal shaft, we simply placed an oval region in the shaft for measurement of the intensity. Measurements from each growth cone were normalized, and results from multiple growth cones were then averaged. neural tube tissues were lysed in the radioimmunoprecipitation buffer, and the clarified lysates were incubated with antibodies against RII (with or without St-Ht31), RI, PP1γ, or normal IgG (Santa Cruz Biotechnology, Inc.) overnight at 4°C. Immune complexes were isolated by incubation with protein A/G PLUS agarose (Santa Cruz Biotechnology, Inc.) for 2 h at 4°C. Cell lysates or IP samples were separated on NuPAGE Novex 3–8% Tris-acetate gels (Invitrogen) or 10% Tris-glycine gels and were transferred to nitrocellulose membranes. After incubation with appropriate primary antibodies, the membranes were detected by HRP-conjugated secondary antibodies and ECL reagents (GE Healthcare). Growth cone turning induced by chemical gradients was performed with a modified Ringer's solution (115 mM NaCl, 2.6 mM KCl, 1 mM MgCl, 1 mM CaCl, and 10 mM Hepes, pH 7.6) as described previously (; ). Microscopic gradients of chemicals were produced by repetitive pressure ejection through a micropipette with an opening of 1 μm (pressure of 3 pounds per square inch, repetition of 2 Hz, and duration of 20 ms). Under these settings, the concentration of the chemical reaching the growth cone is estimated to be ∼1/1,000th of that in the pipette (; ). The original direction of growth cone extension at the beginning of the experiment was defined by the distal 20-μm segment of the neurite. The pipette tip was positioned 45° from the initial direction of extension and 100 μm away from the growth cone. Different reagents were added to the bath medium 20 min before the onset of turning assays. To quantify the turning responses, digital images of the growth cone at the onset and end of the 30-min turning assay were acquired and overlaid with pixel-to-pixel accuracy, and the trajectory of new neurite extension was traced using Photoshop (Adobe). The turning angle was defined by the angle between the original direction of neurite extension and a line connecting the positions of the growth cone at the onset and end of the experiment. A positive angle resulted from turning toward the pipette and vice versa. Neurite extension was quantified by measuring the trajectory of net neurite extension over the 30-min period. Only growth cones that extended 5 μm or more were scored and analyzed for turning responses. We used the nonparametric Mann-Whitney test to analyze turning angles (expressed as mean ± SEM) because they do not follow a normal distribution. Fig. S1 shows the filopodia localization of PKA RII in cultured hippocampal neurons. Fig. S2 shows the existence of potential AKAPs in neural tube tissues through an RII overlay assay. Online supplemental material is available at .

Presynaptic terminals release neurotransmitters in two modes: evoked release, which is induced by Ca flowing into the nerve terminal when stimulated by an action potential, and spontaneous release, which occurs in the absence of massive Ca influx (). Evoked release is clearly the more important form of release in terms of how the brain processes information, but many observations indicate that spontaneous release may also be physiologically important, and not just an “accident” of turbocharged presynaptic release machinery (for reviews see ; ). This evidence suggests that “spontaneous” synaptic vesicle exocytosis causing miniature postsynaptic currents (minis) may be mechanistically distinct from evoked exocytosis and independently regulated, and that minis may have a biological function. Mechanistically, minis appear to derive from a vesicle pool that differs from that which feeds evoked release (). Although evoked and spontaneous release both require SNARE proteins, deletions of the vesicular SNARE protein synaptobrevin/VAMP2 differentially alter evoked and spontaneous release (; ), and structure/function studies suggest that the sequences of synaptobrevin required for evoked and spontaneous release differ (). Moreover, spontaneous and evoked release appear to be differentially regulated by Ca (; ). Physiologically, minis may have a substantial function in regulating neural networks. In cultured hippocampal slices, blocking all release by botulinum toxin had a dramatic effect on spine morphology, whereas blocking only evoked release by tetrodotoxin did not (). Similarly, in cultured hippocampal neurons, spontaneous release was shown to stabilize synaptic function through tonic suppression of dendritic protein synthesis (). In addition, in small cerebellar interneurons, a single excitatory or inhibitory quantum, which is what is released by a mini event, can trigger or inhibit, respectively, the generation of an action potential (). In synapses of mouse cortical and hippocampal neurons, and in neuromuscular junctions, evoked synchronous neurotransmitter release is triggered by Ca binding to synaptotagmin-1 (; ; ; ; ), whereas synaptotagmin-2 performs a similar role in brainstem synapses (). In addition to mediating synchronous Ca-triggered release, synaptotagmin-1 and -2 both normally restrict spontaneous release (), suggesting that they are intrinsic components of the release machinery. We now find that another member of the synaptotagmin family, synaptotagmin-12, is colocalized with synaptotagmin-1 on synaptic vesicles, but is expressed much later in development. Synaptotagmin-12 was originally described as a thyroid hormone–inducible protein () that is homologous to synaptotagmin-1, but lacks its Ca-binding sequences, suggesting that it does not participate in Ca triggering of release. We demonstrate that expression of synaptotagmin-12 in cultured neurons at a time when no endogenous synaptotagmin-12 can be detected causes a dramatic and selective increase in spontaneous release. Moreover, we show that synaptotagmin-12 is phosphorylated by cAMP-dependent protein kinase A (PKA) at a single site, and that mutation of this site blocks the effect of synaptotagmin-12 on spontaneous release, suggesting that this phosphorylation activates its up-regulation of spontaneous release. Finally, we demonstrate that synaptotagmin-12 forms a tight constitutive complex with synaptotagmin-1 on synaptic vesicles, but regulates spontaneous release independently from synaptotagmin-1. Our data suggest a function for a Ca-independent synaptotagmin isoform in modulating spontaneous release. We generated an antibody against the linker sequence between the transmembrane region and C domains of synaptotagmin-12. As expected from the lack of sequence homology between synaptotagmin isoforms in the linker region, the synaptotagmin-12 antibody recognized only this isoform, but no other synaptotagmin (). Immunoblotting revealed that synaptotagmin-12 is expressed in the brain and adrenal medulla (), but not in other nonneuronal tissues tested (), and is also absent from various cell lines, including PC12 cells (). Analysis of different developmental time points showed that synaptotagmin-12 is undetectable during embryonic development (embryonic day 18) when synaptotagmin-1 is already robustly expressed, is present at very low levels postnatally, and becomes abundant after the first postnatal week (). Quantitation of the developmental time course of expression revealed that synaptotagmin-12 levels increase by ∼10-fold during the second postnatal week (). This expression profile is very different from that of synaptotagmin-1, the levels of which increase less than twofold over the same time period, but agrees well with the expression of the synaptotagmin-12 homologue that also exhibits low levels throughout embryonic development (). Immunoblotting of tissue homogenates isolated from various regions of adult brain () and immunolabeling of brain sections () revealed that synaptotagmin-12 is abundantly expressed throughout the brain, with the highest expression levels in cerebellum. This labeling was specific because preincubation with the antigen greatly reduced synaptotagmin-12 immunoreactivity in brain sections (). To establish the localization of native synaptotagmin-12, we fractionated the brain homogenates and analyzed the distribution of synaptotagmin-12 in different subcellular fractions. Synaptotagmin-12 was highly enriched in the fraction that contains synaptic vesicles (). Because subcellular fractionation is imprecise, we next tested the localization of synaptotagmin-12 by measuring its levels in the brains of synapsin 1/2 double knockout mice. In these brains, synaptic vesicles are selectively decreased in numbers, and thus the levels of all synaptic vesicle proteins are decreased, whereas the levels of other proteins, e.g., active zone proteins, plasma membrane proteins, or cytosolic proteins, are unchanged (). Indeed, we found that synaptotagmin-12 was reduced by ∼40% in the synapsin-deficient brains, which exactly corresponds to the decrease observed for other synaptic vesicle proteins (). The levels of synaptic proteins that are not localized on synaptic vesicles, such as the synaptotagmin isoforms 3, 6, and 7 or NMDA-receptors, were not decreased in synapsin double knockout mice (). To obtain a higher resolution localization, we cultured cortical neurons from newborn mice or rats. We first attempted to localize endogenous synaptotagmin-12 in the cultured neurons, but surprisingly, could not detect any synaptotagmin-12 in the neurons even after prolonged culture (). Quantitations of the levels of synaptotagmin-1 and -12 revealed that whereas the concentration of synaptotagmin-1 continuously increased in the neurons as a function of culture time, synaptotagmin-12 remained at levels below the sensitivity of our assay, even after 15 d in vitro (DIV; ). Thus, interestingly, synaptotagmin-12 is present in cultured neurons below the levels observed in the brains from which the neurons were obtained, suggesting that additional factors in brain that are absent under culture conditions must contribute to the regulation of synaptotagmin-12 expression. To be able to localize synaptotagmin-12 in cultured neurons by immunofluorescence labeling, we therefore expressed recombinant synaptotagmin-12 with a lentivirus. We found that recombinant synaptotagmin-12 was targeted to synapses where it precisely colocalized with synaptophysin (). Synaptic vesicle localization of native synaptotagmin-12 was supported by immunohistochemical analysis of brain sections from adult mice that showed that synaptotagmin-12 is distributed in a typically synaptic pattern (Fig. S1, available at ), and by immunoelectron microscopy that revealed synaptotagmin-12 immunoreactivity over the vesicle cluster in presynaptic terminals (). Together, these data show that synaptotagmin-12 is a synaptic vesicle protein. Vesicular targeting of synaptotagmin-1 is regulated by N-glycosylation (). To determine whether synaptotagmin-12 is also glycosylated, we tested the effects of various deglycosylating enzymes on the mobility of native synaptotagmin-12 during SDS-PAGE. In agreement with a previously published report (), removal of neuraminic acid and O- and N-linked sugars with sialidase, O-glycanase, PNGase, or different combinations of these enzymes resulted in a dramatic change in the mobility of synaptotagmin-1, but had no effect on the mobility of synaptotagmin-12, suggesting that synaptotagmin-12 is not glycosylated (). To test whether synaptotagmin-12 interacts with phospholipids, we performed phospholipid-binding assays with the recombinant CAB domain of synaptotagmin-12, using the CAB domain of synaptotagmin-1 as a positive control. Consistent with the primary sequence analysis that indicated that synaptotagmin-12 lacks Ca-binding sequences, no Ca-dependent phospholipid binding was observed when the CAB domain of synaptotagmin-12 was tested alone or in the presence of CAB domains of synaptotagmin-1 (). A screen of the phosphorylation of different synaptotagmins in synaptosomes that were incubated with [P]orthophosphate revealed that the strongest labeling with P under the conditions used was obtained for synaptotagmin-12 ( and not depicted). To identify which protein kinase may mediate synaptotagmin-12 phosphorylation, we immunoprecipitated synaptotagmin-12 from nonlabeled brain extracts and incubated the immunoprecipitated synaptotagmin-12 with brain lysate in the presence of 1 mM ATP and 10 μCi γ-[P]ATP alone or with addition of 0.1 mM cAMP, 0.1 mM cGMP, or 1 mM Ca. Without additions, no synaptotagmin-12 phosphorylation was detected, but cAMP caused strong phosphorylation (). cGMP induced marginal phosphorylation, possibly because cGMP at the high dose used (0.1 mM) can stimulate PKA, whereas Ca did not stimulate any synaptotagmin-12 phosphorylation (). We searched the synaptotagmin-12 sequence for potential PKA phosphorylation sites and identified two such sites in the synaptotagmin-12 linker region (T87 and S97; ). To determine whether one or both of these are being used, we first tested whether recombinant fragments of synaptotagmin-12, produced as GST-fusion proteins, were phosphorylated by PKA. Only the linker region, but not the C domains, were phosphorylated (). We next tested whether mutation of either T87 or S97 alters phosphorylation of the linker by PKA. We found that phosphorylation was completely abolished by substituting serine with alanine, whereas substituting threonine with alanine had no effect (), suggesting that synaptotagmin-12 is phosphorylated by PKA on serine. To determine whether the PKA phosphorylation site in synaptotagmin-12 is evolutionarily conserved, we compared the linker sequences from different species. Although no significant homology was found between fly and vertebrate synaptotagmin-12 linker sequences (unpublished data), serine with a canonical preceding lysine residue was found to be conserved in all vertebrate synaptotagmin-12 homologues from zebrafish to humans (). We next explored whether synaptotagmin-12 regulates synaptic transmission using cultured cortical neurons isolated from newborn mice or rats. Because the cultured neurons do not contain any detectable synaptagmin-12, even after prolonged culture (), the cultured neurons resemble a synaptotagmin-12 loss-of-function model. We infected the neurons with lentiviruses encoding full-length wild-type or S97A mutant synaptotagmin-12 and compared the expression levels of recombinant proteins after 15 DIV (and 10 d after lentivirus infection) with the levels of native synaptotagmin-12 in noninfected cultures and in brain (). We found that recombinant wild-type and S97A mutant synaptotagmin-12 were expressed in cultured neurons at equal levels, i.e., were properly translated, with an approximately sixfold higher concentration than that of synaptotagmin-12 in P15 brain (). Both wild-type and S97A mutant synaptotagmin-12 were targeted to synapses ( and not depicted). When compared with noninfected neurons, the neurons infected with synaptotagmin-12 lentiviruses did not display significant changes in general morphology, and the levels of endogenous synaptic proteins (, and not depicted). To test whether expressed synaptotagmin-12 is phosphorylated under physiological conditions, we analyzed noninfected neurons and neurons infected with wild-type and S97A mutant synaptotagmin-12 lentiviruses by immunoprecipitation. We labeled the neurons with [P]orthophosphate, lysed them, and measured P incorporation into immunoprecipitated synaptotagmin-12. Significant synaptotagmin-12 phosphorylation was only detected in neurons expressing wild-type synaptotagmin-12, whereas noninfected neurons and neurons expressing mutant synaptotagmin-12 exhibited no significant P labeling (, bottom). These data suggest that under the conditions used, synaptotagmin-12 is phosphorylated only on serine either because another constitutively active protein kinase phosphorylates this amino acid or because, in these cultured neurons, PKA is activated. To determine whether synaptotagmin-12 regulates evoked or spontaneous neurotransmitter release, we monitored inhibitory postsynaptic currents (IPSCs) in noninfected neurons or neurons expressing recombinant wild-type or S97A mutant synaptotagmin-12. We found that expression of wild-type synaptotagmin-12 increased the frequency of spontaneous release events threefold, but had no effect on their amplitude (). In contrast to spontaneous minis, synaptotagmin-12 had no effect on the sizes or kinetics of evoked responses, either when these were elicited by isolated single-action potentials or by high-frequency stimulus trains, indicating that both synchronous and asynchronous components or evoked release were unaffected (). Importantly, expression of S97A mutant synaptotagmin-12 had no effect on spontaneous release (), even though the mutant protein was expressed at the same level as wild-type synaptotagmin-12 (). Several previous studies indicated that spontaneous release is up-regulated by cAMP-dependent pathways (; ; ; ). To test the possibility that the effect of synaptotagmin-12 on spontaneous release—which depends on the PKA-substrate site at serine—intersects with the PKA-dependent regulation of spontaneous release, we examined the effect of forskolin (an activator of adenylate cyclase) on spontaneous release. We found that in control cultures, forskolin caused an ∼4.5-fold increase in the rate of spontaneous release (), but had no effect on evoked release (Fig. S2, available at ). In cultures expressing wild-type synaptotagmin-12, the increase in mini frequency induced by forskolin was much more potent than in control cultures, whereas in cultures expressing S97A mutant synaptotagmin-12, the effect of forskolin was identical to that observed in control cultures (). To obtain clues to the mechanism of action of synaptotagmin-12 in its selective effect on spontaneous release, we investigated whether synaptotagmin-12 might interact with SNARE proteins because binding to SNARE proteins is a salient property of other synaptotagmins (; ; ; ; ; ; ; ; ; Tang et al., 2005). We performed immunoprecipitations of detergent-solubilized brain proteins with antibodies to synaptotagmin-1 and -12, and analyzed the immunoprecipitates for the presence of SNARE proteins and synaptotagmins. These experiments failed to reveal binding of synaptotagmin-12 to SNARE proteins, but showed that synaptotagmin-1 was coimmunoprecipitated with synaptotagmin-12 (), and that synaptotagmin-12 was coimmunoprecipitated with synaptotagmin-1 (). SNARE proteins were selectively absent from the synaptotagmin-12 immunoprecipitates, but were present in the synaptotagmin-1 immunoprecipitates. Equal amounts of synaptotagmin-1 and -12 were coimmunoprecipitated in the presence of either 1 mM free Ca or EGTA, suggesting that the interaction between these proteins is Ca-independent (). Quantitations revealed that the equal amounts of synaptotagmin-1 and -12 were coimmunoprecipitated from nontreated brain homogenates and homogenates preincubated with PKA activator 8-Br-cAMP (1 mM), PKA inhibitor H-89 (5 μM), or PKC activator PDBu (1 μM), suggesting that interaction between synaptotagmin-1 and -12 is not regulated by phosphorylation (). To determine whether synaptotagmin-12 regulates spontaneous release through interaction with synaptotagmin-1, we infected cortical neurons isolated from newborn synaptotagmin-1–deficient mice () with lentiviruses expressing wild-type and S97A mutant synaptotagmin-12. We then monitored spontaneous release and release triggered by action potentials (). Because deletion of synaptotagmin-1 results in an increase in the rate of spontaneous release (; ,), we measured spontaneous release in a lower concentration of extracellular Ca that allows us to resolve individual miniature IPSC events. We found that similar to wild-type neurons, expression of synaptotagmin-12 in synaptotagmin-1–deficient neurons enhanced the rate of spontaneous release (), but did not alter evoked release triggered by action potentials (). Because evoked release is completely asynchronous in synaptotagmin-1– deficient neurons (), synaptotagmin- 12, thus, does not significantly alter asynchronous release. Although the increase in spontaneous release in synaptotagmin-1– deficient neurons expressing synaptotagmin-12 was not as dramatic as in wild-type cultures, the increase was still dependent on the phosphorylation of synaptotagmin-12 because the S97A mutant synaptotagmin-12 was unable to potentiate the mini IPSC (mIPSC) rate (). In these experiments, we could not determine the effects of forskolin on mIPSC frequency because the rate of spontaneous release was already too high. In this study, we demonstrate that synaptotagmin-12 is a synaptic vesicle protein that is widely expressed in the brain with a developmentally delayed onset (). Although synaptotagmin-12 is highly homologous to synaptotagmin-1 (which functions as the Ca sensor for fast synaptic vesicle exocytosis), and is colocalized with synaptotagmin-1 on synaptic vesicles ( and Fig. S1), synaptotagmin-12 differs from synaptotagmin-1 in that it does not bind Ca and phospholipids ( and not depicted). We also demonstrate that it is phosphorylated in a cAMP-dependent manner at a single residue, serine (). Because it is expressed in a developmentally delayed pattern, cultured neurons contain very low levels of endogenous synaptotagmin-12. Overexpression of recombinant synaptotagmin-12 in these neurons dramatically increased the rate of spontaneous release (), but had no effect on synchronous and asynchronous components of release triggered by action potentials (). At least two alternative hypotheses could potentially explain how synaptotagmin-12 controls spontaneous neurotransmitter release: synaptotagmin-12 regulates spontaneous release via its interaction with synaptotagmin-1, and synaptotagmin-12 acts as an independent modulator of spontaneous release. The first hypothesis is supported by the observation that synaptotagmin-12 forms a constitutive Ca-independent heterooligomeric complex with synaptotagmin-1 that is incompatible with the interaction of synaptotagmin-1 with SNARE complexes (). Previous studies indicate that synaptotagmin-1 is associated with SNARE complexes in a Ca-independent manner (; ), and that deletion of synaptotagmin-1 increases the rate of excitatory and inhibitory minis (; ), suggesting that freeing SNARE complexes from synaptotagmin-1 may disinhibit spontaneous release. Our finding that synaptotagmin-12 interacts with synaptotagmin-1 suggests that synaptotagmin-12 increases spontaneous neurotransmitter release by pulling synaptotagmin-1 off SNARE complexes, and thereby disinhibiting spontaneous exocytosis. The alternative hypothesis would be that that synaptotagmin-12 regulates spontaneous release by a synaptotagmin-1–independent (but phosphorylation-dependent) mechanism, and that the interaction of synaptotagmin-12 with synaptotagmin-1 performs an additional, as yet undetermined, role. To test the two hypotheses, we determined the effect of overexpression of synaptotagmin-12 in synaptotagmin-1–deficient neurons (). We found that in the absence of synaptotagmin-1, synaptotagmin-12 still produced an increase, albeit a moderate one, in the rate of spontaneous exocytosis. This result favors the second hypothesis, suggesting that the phosphorylation site–dependent modulation of mini release by synaptotagmin-12 does not require synaptotagmin-1. It is also unlikely that the effect of expression of synaptotagmin-12 on spontaneous release is caused by a massive change in synapse density or in the size of the readily releasable pool of synaptic vesicles because we observed no differences between neurons expressing or lacking wild-type or mutant synaptotagmin-12 in evoked synaptic responses; this applies both for responses triggered by single-action potentials or by trains of action potentials ( and ). A recent study has shown that spontaneous release is driven by an isolated pool of synaptic vesicles containing synaptotagmin-1 and controlled by synaptobrevin/VAMP2 (). It is conceivable that synaptotagmin-12 selectively regulates spontaneous release because it is only localized on this subpopulation of vesicles, but further biochemical analysis with an antibody against the luminal domain of synaptotagmin-12 will be required to test this possibility. Our findings have several implications for the understanding of synaptic transmission. First, the possible biological role of spontaneous release events is widely debated, with opinions ranging from considering such events as mere accidental byproducts of turbo-charged fusion machinery to biologically meaningful processes of synaptic communication (; ). Our data, by identifying a synaptic vesicle synaptotagmin isoform that selectively regulates spontaneous release in cultured neurons, lend credence to the notion that spontaneous release is a highly specific and regulated event. Notably, the rate of spontaneous release increases with age (), as do the expression levels of synaptotagmin-12. This finding suggests a connection, but several other factors such as developmental changes in connectivity and synapse maturation must also play a major role in this process (). Although the effect of synaptotagmin-12 in cultured neurons is specific for spontaneous as opposed to evoked release, this does not exclude the possibility that in an intact brain, synaptotagmin-12 could participate in PKA-dependent forms of plasticity. It is noticeable that synaptotagmin-12 was discovered as a thyroid hormone–inducible gene (; ) and comes on relatively late in development, suggesting that its role is under further regulation beyond the phosphorylation by PKA. Moreover, extensive studies indicate that cAMP-dependent phosphorylation regulates exocytosis in nonneuronal cells () and neurons (; ; ; ). Multiple pathways have been implicated in the cAMP-dependent modulation of synaptic strength (; ; ; ; ). The requirement of serine for the effect of synaptotagmin-12 on spontaneous release indicates that synaptotagmin-12 may be generally involved in the regulation of synaptic vesicle exocytosis by PKA-dependent phosphorylation. Finally, expression of synaptotagmin-12 was detected in adrenal glands (), suggesting that synaptotagmin-12 may also be localized on dense core vesicles and play a role in regulating calcium-independent secretion in nonneuronal cells. The following vectors were constructed for expression of various regions of rat synaptotagmin-12 as GST-fusion proteins: pGEX-KG-synaptotagmin-12-linker (aa 47–150); pGEX-KG-synaptotagmin-12-CA (aa 151–281); and pGEX-KG-synaptotagmin-12-CB (aa 282–421). For construction of synaptotagmin-12 lentivirus, full-length cDNA (aa 1–421) was subcloned into pFUGW shuttle vector. The point mutations were generated by PCR and verified by sequencing. To generate a polyclonal antibody against synaptotagmin-12, the linker region of rat synaptotagmin-12 (aa 47–150) was expressed in BL-21 strain as GST fusion protein and purified by affinity chromatography on glutathione–Agarose beads (GE Healthcare). The protein was eluted from the beads by 20 mM glutathione, dialyzed twice against PBS, and used for immunization at concentration of ∼1 mg/ml. The specificity of the serum was confirmed by immunoblotting of full-length recombinant synaptotagmins 1–13 expressed in COS7 cells. The antibody was then purified by sequential affinity chromatography on the columns containing GST and GST-synaptotagmin-12-linker covalently attached to NHS–Sepharose beads (Pierce Chemical Co.) Whole rat brains, including cerebellum and olfactory bulb, were homogenized in 0.32 M sucrose, 25 mM Hepes, pH 7.2, and centrifuged at 800 to remove the nuclear fraction. The postnuclear supernatant (LSP) was centrifuged at 12,000 to separate the soluble (synaptosomal supernatant) and membrane (synaptosomal pellet) fractions. The synaptosomal pellet was hypotonically lysed by incubation for 30 min in 5 mM Tris, pH 7.2, and recentrifuged at 27,000 to separate the crude synaptic plasma membranes and crude synaptic vesicles. The crude synaptic plasma membrane fraction was adjusted to 1.1 M sucrose, loaded to the bottom of the centrifuge tube, and layered by solutions of 0.86 and 0.32 M sucrose. The samples were then centrifuged at 19,000 RPM (SW41 Ti rotor; Beckman Coulter) to separate synaptic plasma membranes, myelin, and mitochondria. All fractions were adjusted to the total protein concentration of 4 mg/ml and analyzed by immunoblotting. Approximately 40 μg of total protein was loaded into each lane (). Rat brains were homogenized in 50 mM Hepes, 100 mM NaCl, 2 mM MgCl, 4 mM EGTA, pH 6.8, and solubilized for 2 h at 4°C in the same buffer containing 1% Triton X-100. The samples were then centrifuged for 1 h at 50,000 RPM (70 Ti rotor; Beckman Coulter) to remove insoluble material and incubated for 2 h at 4°C with protein A–or protein G–Sepharose beads (GE Healthcare) covered with polyclonal antibody to synaptotagmin-12, monoclonal antibody to synaptotagmin-1, corresponding preimmune serum, or control antibody. The protein complexes attached to the beads were washed five times with the extraction buffer, eluted with SDS sample buffer, and analyzed by SDS-page and immunoblotting. The input lanes were loaded with 1% of total protein extract used for immunoprecipitation. To determine the effects of activators or inhibitors of PKA-dependent phosphorylation, the brain homogenates were preincubated for 30 min with 1 mM 8-Br-cAMP, 5 μM H-89, or 1 μM PDBu (used as a negative control) before protein extraction, and the immunoprecipitations were carried in the presence of EGTA, as described above. The cortexes or hippocampi were dissected from the brains of embryonic day 18 embryos or newborn pups, dissociated by trypsin digestion, and plated on circle glass coverslips coated with Matrigel. The cultures were maintained in MEM medium (Invitrogen) supplemented with B-27 (Invitrogen), -glutamine, 0.5% glucose, 5% fetal bovine serum, and 2 mM Ara-C (Sigma-Aldrich). The cultures were used for experiments at 14–17 DIV. Neurons attached to the glass coverslips were rinsed once in PBS, fixed for 15 min on ice in 4% formaldehyde, 4% sucrose in PBS and permeabilized for 5 min at room temperature in 0.2% Triton X-100 (Roche) in PBS. After permeabilization, the neurons were incubated for 30 min in blocking solution containing 5% BSA (Sigma-Aldrich; fraction V) in PBS, followed by 1-h incubation with primary and rhodamine- and FITC-conjugated secondary antibodies diluted in blocking solution. The coverslips were then mounted on glass slides with Aqua-Poly/Mount medium (Polysciences, Inc.) and analyzed at room temperature using a confocal microscope (DMIRE2; Leica) and 63×/1.32–0.6 oil immersion objective. The images were collected using confocal software (Leica) and processed using Photoshop software (Adobe). The background fluorescence was digitally reduced by 25% using the “color balance” function in Photoshop. All digital manipulations were equally applied to the entire image. Brain sections from perfusion-fixed rats were permeabilized for 10 min in 0.5% Triton X-100 in PBS and incubated in blocking buffer containing 2% goat serum and 0.1% Triton X-100 in PBS. The sections were then incubated sequentially for 1 h in primary and secondary antibodies diluted in blocking buffer and developed in DAB substrate. For dehydration, the sections were incubated for 10 min in 70% ethanol, followed by 10-min incubation in 90% ethanol and 10-min incubation in 100% ethanol. The sections were then mounted on glass slides and analyzed by light microscopy. Adult mouse brains were perfusion fixed in 4% paraformaldehyde and 0.1% glutaraldehyde in PBS, pH 7.4, followed by overnight immersion fixation in 0.1% glutaraldehyde and 2% paraformaldehyde. The brain sections (120 μm) were permeabilized in 0.2% Triton X-100, blocked in 4% normal goat serum, and incubated overnight with primary antibody alone or primary antibody mixed with the antigen (∼1 mg/ml). The sections were then incubated with the secondary antibody conjugated with 1.4-nm gold particles (1:100 dilution; Nanoprobes) for 24 h, and immunogold signal was enhanced with the HQ silver enhancement kit (Nanoprobes). Sections were further fixed with 0.5% osmium tetroxide, dehydrated through a graded series of ethanol, and embedded in Poly/Bed 812 epoxy resin (Polysciences, Inc.). Ultrathin sections (65 nm) were stained with 5% uranyl acetate solution and examined under a transmission electron microscope (FEI Tecnai; FEI) at 120 kV accelerating voltage. Synaptosomes (∼1 mg of total protein) were prelabeled for 30 min at 37°C in phosphate-free aerated Krebs-Henseleit-Hepes buffer (118 mM NaCl, 3.5 mM KCl, 1.25 mM CaCl, 1.2 mM MgSO, 25 mM, NaHCO, 5 mM Hepes-NaOH, pH 7.4, and 115 mM glucose) containing 0.3 mCi of [P]orthophosphate and incubated for an additional 10 min with 1 μM of okadaic acid. P-labeled synaptosomes were precipitated by centrifugation at 14,000 rpm on a microcentrifuge (Eppendorf) and solubilized for 1 h at 4°C in Krebs-Henseleit-Hepes buffer containing 2 mM sodium orthovanadate, 10 mM sodium pyrophosphate, 50 mM NaF, 1 mM EGTA, 1 mM EDTA, 1% Triton X-100, 0.2% SDS, 1 mM PMSF, 5 μg/ml leupeptin, 2 μg/ml aprotinin, and 1 mM DTT. Solubilized proteins were diluted with equal volume of lysis buffer without detergents and immunoprecipitated with the antibody to synaptotagmin-12 or preimmune serum, as described in Immunoprecipitations. For in vitro phosphorylation of native synaptotagmin-12, brain proteins were extracted in 1% of Triton X-100 and immunoprecipitated with synaptotagmin-12 antibody or preimmune serum. Immunoprecipitates were extensively washed and incubated for 10 min at 37°C with 50 μl of rat brain cytosol (10 mg/ml in Tris buffer without protease inhibitors) mixed with 2.5 mM ATP and 20 μCi γ-[P]ATP alone with addition of 0.1 mM cAMP, 0.1 mM cGMP, or 1 mM Ca. For in vitro phosphorylation of recombinant proteins, 30 μg of each GST-fusion protein immobilized on glutathione–Sepharose beads were mixed with 500 μl of reaction mixture containing 50 mM Hepes, pH 7.2, 100 mM NaCl, 4 mM EGTA, 2 mM MgCl, 50 units of catalytic subunit of PKA (Sigma-Aldrich), 1 mM ATP, and 10 μCi γ-[P]ATP and incubated for 30 min at 30°C. The proteins attached to the beads were then washed three times with Hepes buffer (50 mM Hepes, pH 7.2, 100 mM NaCl, 4 mM EGTA, 2 mM MgCl, 2 mM sodium orthovanadate, 10 mM sodium pyrophosphate, and 50 mM NaF) and analyzed by SDS- PAGE. For detection, gels were dried and exposed to x-ray film for 24–48 h at −80°C. Cortical neurons were infected with lentiviruses encoding wild-type synaptotagmin-12, -12-S97A, or GFP and analyzed at 14–17 DIV (9–12 d after infection). Because efficiency of lentivaral infections exceeded 95% (as determined by GFP fluorescence and immunocytochemistry with antibody to synaptotagmin-12), we randomly selected the neurons for whole-cell recording assuming that most presynaptic inputs in infected cultures are formed by neurons expressing recombinant protein of interest. Inhibitory synaptic responses were triggered by 1-ms current injection (900 μA) through a local extracellular electrode (FHC, Inc.) and recorded in a whole-cell mode using Multiclamp 700A amplifier (Axon Instruments, Inc.). All experiments were performed at room temperature. The frequency, duration, and magnitude of extracellular stimulus were controlled with Model 2100 Isolated Pulse Stimulator (A-M Systems, Inc.). The whole-cell pipette solution contained 135 mM CsCl, 10 mM Hepes, 1 mM EGTA, 1 mM Na-GTP, 4 mM Mg-ATP, and 1 mM QX-314, pH 7.4. The bath solution contained 140 mM NaCl, 5 mM KCl, 0.5 mM or 2 mM CaCl, 0.8 mM MgCl, 10 mM Hepes, and 10 mM glucose, pH 7.4. Excitatory AMPA and NMDA currents were suppressed by the addition of 50 μM APV and 20 μM CNQX to the bath solution. Spontaneous mIPSCs were monitored in the presence of 1 μM tetrodotoxin to block action potentials. Forskolin (Sigma-Aldrich) was used at 50 μM. The currents were sampled at 10 kHz and analyzed off-line using pClamp9 (Axon Instruments, Inc.) and Origin7 (Microcal, Inc.) software. mIPSCs event detection was performed with pClamp template search function using a template constructed by averaging >100 manually collected individual mIPSCs. Statistical analysis was performed with test; ***, P ≤ 0.001. All data are shown as the mean ± the SEM. Production of recombinant lentiviruses, phospholipid-binding assays, deglycosylation, and immunoblotting were performed as previously described (; ; ). Fig. S1 shows localization of synaptotagmin-12 in brain slices. Fig. S2 shows the effect of Forskolin on evoked release in inhibitory synapses of cultured cortical neurons. Online supplemental material is available at .